Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
8f18cf13 | 6 | #include <linux/kernel.h> |
065631f6 | 7 | #include <linux/bio.h> |
39279cc3 | 8 | #include <linux/buffer_head.h> |
f2eb0a24 | 9 | #include <linux/file.h> |
39279cc3 CM |
10 | #include <linux/fs.h> |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/time.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/string.h> | |
39279cc3 | 16 | #include <linux/backing-dev.h> |
39279cc3 | 17 | #include <linux/writeback.h> |
39279cc3 | 18 | #include <linux/compat.h> |
5103e947 | 19 | #include <linux/xattr.h> |
33268eaf | 20 | #include <linux/posix_acl.h> |
d899e052 | 21 | #include <linux/falloc.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
7a36ddec | 23 | #include <linux/ratelimit.h> |
55e301fd | 24 | #include <linux/btrfs.h> |
53b381b3 | 25 | #include <linux/blkdev.h> |
f23b5a59 | 26 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 27 | #include <linux/uio.h> |
69fe2d75 | 28 | #include <linux/magic.h> |
ae5e165d | 29 | #include <linux/iversion.h> |
ed46ff3d | 30 | #include <linux/swap.h> |
b1c16ac9 | 31 | #include <linux/sched/mm.h> |
92d32170 | 32 | #include <asm/unaligned.h> |
39279cc3 CM |
33 | #include "ctree.h" |
34 | #include "disk-io.h" | |
35 | #include "transaction.h" | |
36 | #include "btrfs_inode.h" | |
39279cc3 | 37 | #include "print-tree.h" |
e6dcd2dc | 38 | #include "ordered-data.h" |
95819c05 | 39 | #include "xattr.h" |
e02119d5 | 40 | #include "tree-log.h" |
4a54c8c1 | 41 | #include "volumes.h" |
c8b97818 | 42 | #include "compression.h" |
b4ce94de | 43 | #include "locking.h" |
dc89e982 | 44 | #include "free-space-cache.h" |
581bb050 | 45 | #include "inode-map.h" |
38c227d8 | 46 | #include "backref.h" |
63541927 | 47 | #include "props.h" |
31193213 | 48 | #include "qgroup.h" |
dda3245e | 49 | #include "dedupe.h" |
39279cc3 CM |
50 | |
51 | struct btrfs_iget_args { | |
90d3e592 | 52 | struct btrfs_key *location; |
39279cc3 CM |
53 | struct btrfs_root *root; |
54 | }; | |
55 | ||
f28a4928 | 56 | struct btrfs_dio_data { |
f28a4928 FM |
57 | u64 reserve; |
58 | u64 unsubmitted_oe_range_start; | |
59 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 60 | int overwrite; |
f28a4928 FM |
61 | }; |
62 | ||
6e1d5dcc AD |
63 | static const struct inode_operations btrfs_dir_inode_operations; |
64 | static const struct inode_operations btrfs_symlink_inode_operations; | |
65 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
66 | static const struct inode_operations btrfs_special_inode_operations; | |
67 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b | 68 | static const struct address_space_operations btrfs_aops; |
828c0950 | 69 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 70 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
71 | |
72 | static struct kmem_cache *btrfs_inode_cachep; | |
73 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 74 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 75 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 | 76 | |
3972f260 | 77 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
213e8c55 | 78 | static int btrfs_truncate(struct inode *inode, bool skip_writeback); |
5fd02043 | 79 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
80 | static noinline int cow_file_range(struct inode *inode, |
81 | struct page *locked_page, | |
dda3245e WX |
82 | u64 start, u64 end, u64 delalloc_end, |
83 | int *page_started, unsigned long *nr_written, | |
84 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
85 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
86 | u64 orig_start, u64 block_start, | |
87 | u64 block_len, u64 orig_block_len, | |
88 | u64 ram_bytes, int compress_type, | |
89 | int type); | |
7b128766 | 90 | |
52427260 QW |
91 | static void __endio_write_update_ordered(struct inode *inode, |
92 | const u64 offset, const u64 bytes, | |
93 | const bool uptodate); | |
94 | ||
95 | /* | |
96 | * Cleanup all submitted ordered extents in specified range to handle errors | |
52042d8e | 97 | * from the btrfs_run_delalloc_range() callback. |
52427260 QW |
98 | * |
99 | * NOTE: caller must ensure that when an error happens, it can not call | |
100 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
101 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
102 | * to be released, which we want to happen only when finishing the ordered | |
d1051d6e | 103 | * extent (btrfs_finish_ordered_io()). |
52427260 QW |
104 | */ |
105 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
d1051d6e NB |
106 | struct page *locked_page, |
107 | u64 offset, u64 bytes) | |
52427260 | 108 | { |
63d71450 NA |
109 | unsigned long index = offset >> PAGE_SHIFT; |
110 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
d1051d6e NB |
111 | u64 page_start = page_offset(locked_page); |
112 | u64 page_end = page_start + PAGE_SIZE - 1; | |
113 | ||
63d71450 NA |
114 | struct page *page; |
115 | ||
116 | while (index <= end_index) { | |
117 | page = find_get_page(inode->i_mapping, index); | |
118 | index++; | |
119 | if (!page) | |
120 | continue; | |
121 | ClearPagePrivate2(page); | |
122 | put_page(page); | |
123 | } | |
d1051d6e NB |
124 | |
125 | /* | |
126 | * In case this page belongs to the delalloc range being instantiated | |
127 | * then skip it, since the first page of a range is going to be | |
128 | * properly cleaned up by the caller of run_delalloc_range | |
129 | */ | |
130 | if (page_start >= offset && page_end <= (offset + bytes - 1)) { | |
131 | offset += PAGE_SIZE; | |
132 | bytes -= PAGE_SIZE; | |
133 | } | |
134 | ||
135 | return __endio_write_update_ordered(inode, offset, bytes, false); | |
52427260 QW |
136 | } |
137 | ||
48a3b636 | 138 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 139 | |
6a3891c5 JB |
140 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
141 | void btrfs_test_inode_set_ops(struct inode *inode) | |
142 | { | |
143 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
144 | } | |
145 | #endif | |
146 | ||
f34f57a3 | 147 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
148 | struct inode *inode, struct inode *dir, |
149 | const struct qstr *qstr) | |
0279b4cd JO |
150 | { |
151 | int err; | |
152 | ||
f34f57a3 | 153 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 154 | if (!err) |
2a7dba39 | 155 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
156 | return err; |
157 | } | |
158 | ||
c8b97818 CM |
159 | /* |
160 | * this does all the hard work for inserting an inline extent into | |
161 | * the btree. The caller should have done a btrfs_drop_extents so that | |
162 | * no overlapping inline items exist in the btree | |
163 | */ | |
40f76580 | 164 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 165 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
166 | struct btrfs_root *root, struct inode *inode, |
167 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 168 | int compress_type, |
c8b97818 CM |
169 | struct page **compressed_pages) |
170 | { | |
c8b97818 CM |
171 | struct extent_buffer *leaf; |
172 | struct page *page = NULL; | |
173 | char *kaddr; | |
174 | unsigned long ptr; | |
175 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
176 | int ret; |
177 | size_t cur_size = size; | |
c8b97818 | 178 | unsigned long offset; |
c8b97818 | 179 | |
fe3f566c | 180 | if (compressed_size && compressed_pages) |
c8b97818 | 181 | cur_size = compressed_size; |
c8b97818 | 182 | |
1acae57b | 183 | inode_add_bytes(inode, size); |
c8b97818 | 184 | |
1acae57b FDBM |
185 | if (!extent_inserted) { |
186 | struct btrfs_key key; | |
187 | size_t datasize; | |
c8b97818 | 188 | |
4a0cc7ca | 189 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 190 | key.offset = start; |
962a298f | 191 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 192 | |
1acae57b FDBM |
193 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
194 | path->leave_spinning = 1; | |
195 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
196 | datasize); | |
79b4f4c6 | 197 | if (ret) |
1acae57b | 198 | goto fail; |
c8b97818 CM |
199 | } |
200 | leaf = path->nodes[0]; | |
201 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
202 | struct btrfs_file_extent_item); | |
203 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
204 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
205 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
206 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
207 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
208 | ptr = btrfs_file_extent_inline_start(ei); | |
209 | ||
261507a0 | 210 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
211 | struct page *cpage; |
212 | int i = 0; | |
d397712b | 213 | while (compressed_size > 0) { |
c8b97818 | 214 | cpage = compressed_pages[i]; |
5b050f04 | 215 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 216 | PAGE_SIZE); |
c8b97818 | 217 | |
7ac687d9 | 218 | kaddr = kmap_atomic(cpage); |
c8b97818 | 219 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 220 | kunmap_atomic(kaddr); |
c8b97818 CM |
221 | |
222 | i++; | |
223 | ptr += cur_size; | |
224 | compressed_size -= cur_size; | |
225 | } | |
226 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 227 | compress_type); |
c8b97818 CM |
228 | } else { |
229 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 230 | start >> PAGE_SHIFT); |
c8b97818 | 231 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 232 | kaddr = kmap_atomic(page); |
7073017a | 233 | offset = offset_in_page(start); |
c8b97818 | 234 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 235 | kunmap_atomic(kaddr); |
09cbfeaf | 236 | put_page(page); |
c8b97818 CM |
237 | } |
238 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 239 | btrfs_release_path(path); |
c8b97818 | 240 | |
c2167754 YZ |
241 | /* |
242 | * we're an inline extent, so nobody can | |
243 | * extend the file past i_size without locking | |
244 | * a page we already have locked. | |
245 | * | |
246 | * We must do any isize and inode updates | |
247 | * before we unlock the pages. Otherwise we | |
248 | * could end up racing with unlink. | |
249 | */ | |
c8b97818 | 250 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 251 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 252 | |
c8b97818 | 253 | fail: |
79b4f4c6 | 254 | return ret; |
c8b97818 CM |
255 | } |
256 | ||
257 | ||
258 | /* | |
259 | * conditionally insert an inline extent into the file. This | |
260 | * does the checks required to make sure the data is small enough | |
261 | * to fit as an inline extent. | |
262 | */ | |
d02c0e20 | 263 | static noinline int cow_file_range_inline(struct inode *inode, u64 start, |
00361589 JB |
264 | u64 end, size_t compressed_size, |
265 | int compress_type, | |
266 | struct page **compressed_pages) | |
c8b97818 | 267 | { |
d02c0e20 | 268 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0b246afa | 269 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 270 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
271 | u64 isize = i_size_read(inode); |
272 | u64 actual_end = min(end + 1, isize); | |
273 | u64 inline_len = actual_end - start; | |
0b246afa | 274 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
275 | u64 data_len = inline_len; |
276 | int ret; | |
1acae57b FDBM |
277 | struct btrfs_path *path; |
278 | int extent_inserted = 0; | |
279 | u32 extent_item_size; | |
c8b97818 CM |
280 | |
281 | if (compressed_size) | |
282 | data_len = compressed_size; | |
283 | ||
284 | if (start > 0 || | |
0b246afa JM |
285 | actual_end > fs_info->sectorsize || |
286 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 287 | (!compressed_size && |
0b246afa | 288 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 289 | end + 1 < isize || |
0b246afa | 290 | data_len > fs_info->max_inline) { |
c8b97818 CM |
291 | return 1; |
292 | } | |
293 | ||
1acae57b FDBM |
294 | path = btrfs_alloc_path(); |
295 | if (!path) | |
296 | return -ENOMEM; | |
297 | ||
00361589 | 298 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
299 | if (IS_ERR(trans)) { |
300 | btrfs_free_path(path); | |
00361589 | 301 | return PTR_ERR(trans); |
1acae57b | 302 | } |
69fe2d75 | 303 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
00361589 | 304 | |
1acae57b FDBM |
305 | if (compressed_size && compressed_pages) |
306 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
307 | compressed_size); | |
308 | else | |
309 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
310 | inline_len); | |
311 | ||
312 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
313 | start, aligned_end, NULL, | |
314 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 315 | if (ret) { |
66642832 | 316 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
317 | goto out; |
318 | } | |
c8b97818 CM |
319 | |
320 | if (isize > actual_end) | |
321 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
322 | ret = insert_inline_extent(trans, path, extent_inserted, |
323 | root, inode, start, | |
c8b97818 | 324 | inline_len, compressed_size, |
fe3f566c | 325 | compress_type, compressed_pages); |
2adcac1a | 326 | if (ret && ret != -ENOSPC) { |
66642832 | 327 | btrfs_abort_transaction(trans, ret); |
00361589 | 328 | goto out; |
2adcac1a | 329 | } else if (ret == -ENOSPC) { |
00361589 JB |
330 | ret = 1; |
331 | goto out; | |
79787eaa | 332 | } |
2adcac1a | 333 | |
bdc20e67 | 334 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
dcdbc059 | 335 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 336 | out: |
94ed938a QW |
337 | /* |
338 | * Don't forget to free the reserved space, as for inlined extent | |
339 | * it won't count as data extent, free them directly here. | |
340 | * And at reserve time, it's always aligned to page size, so | |
341 | * just free one page here. | |
342 | */ | |
bc42bda2 | 343 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 344 | btrfs_free_path(path); |
3a45bb20 | 345 | btrfs_end_transaction(trans); |
00361589 | 346 | return ret; |
c8b97818 CM |
347 | } |
348 | ||
771ed689 CM |
349 | struct async_extent { |
350 | u64 start; | |
351 | u64 ram_size; | |
352 | u64 compressed_size; | |
353 | struct page **pages; | |
354 | unsigned long nr_pages; | |
261507a0 | 355 | int compress_type; |
771ed689 CM |
356 | struct list_head list; |
357 | }; | |
358 | ||
97db1204 | 359 | struct async_chunk { |
771ed689 | 360 | struct inode *inode; |
771ed689 CM |
361 | struct page *locked_page; |
362 | u64 start; | |
363 | u64 end; | |
f82b7359 | 364 | unsigned int write_flags; |
771ed689 CM |
365 | struct list_head extents; |
366 | struct btrfs_work work; | |
97db1204 | 367 | atomic_t *pending; |
771ed689 CM |
368 | }; |
369 | ||
97db1204 NB |
370 | struct async_cow { |
371 | /* Number of chunks in flight; must be first in the structure */ | |
372 | atomic_t num_chunks; | |
373 | struct async_chunk chunks[]; | |
771ed689 CM |
374 | }; |
375 | ||
97db1204 | 376 | static noinline int add_async_extent(struct async_chunk *cow, |
771ed689 CM |
377 | u64 start, u64 ram_size, |
378 | u64 compressed_size, | |
379 | struct page **pages, | |
261507a0 LZ |
380 | unsigned long nr_pages, |
381 | int compress_type) | |
771ed689 CM |
382 | { |
383 | struct async_extent *async_extent; | |
384 | ||
385 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 386 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
387 | async_extent->start = start; |
388 | async_extent->ram_size = ram_size; | |
389 | async_extent->compressed_size = compressed_size; | |
390 | async_extent->pages = pages; | |
391 | async_extent->nr_pages = nr_pages; | |
261507a0 | 392 | async_extent->compress_type = compress_type; |
771ed689 CM |
393 | list_add_tail(&async_extent->list, &cow->extents); |
394 | return 0; | |
395 | } | |
396 | ||
c2fcdcdf | 397 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 398 | { |
0b246afa | 399 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
400 | |
401 | /* force compress */ | |
0b246afa | 402 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 403 | return 1; |
eec63c65 DS |
404 | /* defrag ioctl */ |
405 | if (BTRFS_I(inode)->defrag_compress) | |
406 | return 1; | |
f79707b0 WS |
407 | /* bad compression ratios */ |
408 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
409 | return 0; | |
0b246afa | 410 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 411 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 412 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 413 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
414 | return 0; |
415 | } | |
416 | ||
6158e1ce | 417 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
418 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
419 | { | |
420 | /* If this is a small write inside eof, kick off a defrag */ | |
421 | if (num_bytes < small_write && | |
6158e1ce | 422 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
423 | btrfs_add_inode_defrag(NULL, inode); |
424 | } | |
425 | ||
d352ac68 | 426 | /* |
771ed689 CM |
427 | * we create compressed extents in two phases. The first |
428 | * phase compresses a range of pages that have already been | |
429 | * locked (both pages and state bits are locked). | |
c8b97818 | 430 | * |
771ed689 CM |
431 | * This is done inside an ordered work queue, and the compression |
432 | * is spread across many cpus. The actual IO submission is step | |
433 | * two, and the ordered work queue takes care of making sure that | |
434 | * happens in the same order things were put onto the queue by | |
435 | * writepages and friends. | |
c8b97818 | 436 | * |
771ed689 CM |
437 | * If this code finds it can't get good compression, it puts an |
438 | * entry onto the work queue to write the uncompressed bytes. This | |
439 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
440 | * are written in the same order that the flusher thread sent them |
441 | * down. | |
d352ac68 | 442 | */ |
1368c6da NB |
443 | static noinline void compress_file_range(struct async_chunk *async_chunk, |
444 | int *num_added) | |
b888db2b | 445 | { |
1368c6da | 446 | struct inode *inode = async_chunk->inode; |
0b246afa | 447 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
0b246afa | 448 | u64 blocksize = fs_info->sectorsize; |
1368c6da NB |
449 | u64 start = async_chunk->start; |
450 | u64 end = async_chunk->end; | |
c8b97818 | 451 | u64 actual_end; |
e6dcd2dc | 452 | int ret = 0; |
c8b97818 CM |
453 | struct page **pages = NULL; |
454 | unsigned long nr_pages; | |
c8b97818 CM |
455 | unsigned long total_compressed = 0; |
456 | unsigned long total_in = 0; | |
c8b97818 CM |
457 | int i; |
458 | int will_compress; | |
0b246afa | 459 | int compress_type = fs_info->compress_type; |
4adaa611 | 460 | int redirty = 0; |
b888db2b | 461 | |
6158e1ce NB |
462 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
463 | SZ_16K); | |
4cb5300b | 464 | |
62b37622 | 465 | actual_end = min_t(u64, i_size_read(inode), end + 1); |
c8b97818 CM |
466 | again: |
467 | will_compress = 0; | |
09cbfeaf | 468 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
469 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
470 | nr_pages = min_t(unsigned long, nr_pages, | |
471 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 472 | |
f03d9301 CM |
473 | /* |
474 | * we don't want to send crud past the end of i_size through | |
475 | * compression, that's just a waste of CPU time. So, if the | |
476 | * end of the file is before the start of our current | |
477 | * requested range of bytes, we bail out to the uncompressed | |
478 | * cleanup code that can deal with all of this. | |
479 | * | |
480 | * It isn't really the fastest way to fix things, but this is a | |
481 | * very uncommon corner. | |
482 | */ | |
483 | if (actual_end <= start) | |
484 | goto cleanup_and_bail_uncompressed; | |
485 | ||
c8b97818 CM |
486 | total_compressed = actual_end - start; |
487 | ||
4bcbb332 SW |
488 | /* |
489 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 490 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
491 | */ |
492 | if (total_compressed <= blocksize && | |
493 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
494 | goto cleanup_and_bail_uncompressed; | |
495 | ||
069eac78 DS |
496 | total_compressed = min_t(unsigned long, total_compressed, |
497 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
498 | total_in = 0; |
499 | ret = 0; | |
db94535d | 500 | |
771ed689 CM |
501 | /* |
502 | * we do compression for mount -o compress and when the | |
503 | * inode has not been flagged as nocompress. This flag can | |
504 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 505 | */ |
c2fcdcdf | 506 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 507 | WARN_ON(pages); |
31e818fe | 508 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
509 | if (!pages) { |
510 | /* just bail out to the uncompressed code */ | |
3527a018 | 511 | nr_pages = 0; |
560f7d75 LZ |
512 | goto cont; |
513 | } | |
c8b97818 | 514 | |
eec63c65 DS |
515 | if (BTRFS_I(inode)->defrag_compress) |
516 | compress_type = BTRFS_I(inode)->defrag_compress; | |
517 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 518 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 519 | |
4adaa611 CM |
520 | /* |
521 | * we need to call clear_page_dirty_for_io on each | |
522 | * page in the range. Otherwise applications with the file | |
523 | * mmap'd can wander in and change the page contents while | |
524 | * we are compressing them. | |
525 | * | |
526 | * If the compression fails for any reason, we set the pages | |
527 | * dirty again later on. | |
e9679de3 TT |
528 | * |
529 | * Note that the remaining part is redirtied, the start pointer | |
530 | * has moved, the end is the original one. | |
4adaa611 | 531 | */ |
e9679de3 TT |
532 | if (!redirty) { |
533 | extent_range_clear_dirty_for_io(inode, start, end); | |
534 | redirty = 1; | |
535 | } | |
f51d2b59 DS |
536 | |
537 | /* Compression level is applied here and only here */ | |
538 | ret = btrfs_compress_pages( | |
539 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 540 | inode->i_mapping, start, |
38c31464 | 541 | pages, |
4d3a800e | 542 | &nr_pages, |
261507a0 | 543 | &total_in, |
e5d74902 | 544 | &total_compressed); |
c8b97818 CM |
545 | |
546 | if (!ret) { | |
7073017a | 547 | unsigned long offset = offset_in_page(total_compressed); |
4d3a800e | 548 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
549 | char *kaddr; |
550 | ||
551 | /* zero the tail end of the last page, we might be | |
552 | * sending it down to disk | |
553 | */ | |
554 | if (offset) { | |
7ac687d9 | 555 | kaddr = kmap_atomic(page); |
c8b97818 | 556 | memset(kaddr + offset, 0, |
09cbfeaf | 557 | PAGE_SIZE - offset); |
7ac687d9 | 558 | kunmap_atomic(kaddr); |
c8b97818 CM |
559 | } |
560 | will_compress = 1; | |
561 | } | |
562 | } | |
560f7d75 | 563 | cont: |
c8b97818 CM |
564 | if (start == 0) { |
565 | /* lets try to make an inline extent */ | |
6018ba0a | 566 | if (ret || total_in < actual_end) { |
c8b97818 | 567 | /* we didn't compress the entire range, try |
771ed689 | 568 | * to make an uncompressed inline extent. |
c8b97818 | 569 | */ |
d02c0e20 NB |
570 | ret = cow_file_range_inline(inode, start, end, 0, |
571 | BTRFS_COMPRESS_NONE, NULL); | |
c8b97818 | 572 | } else { |
771ed689 | 573 | /* try making a compressed inline extent */ |
d02c0e20 | 574 | ret = cow_file_range_inline(inode, start, end, |
fe3f566c LZ |
575 | total_compressed, |
576 | compress_type, pages); | |
c8b97818 | 577 | } |
79787eaa | 578 | if (ret <= 0) { |
151a41bc | 579 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
580 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
581 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
582 | unsigned long page_error_op; |
583 | ||
e6eb4314 | 584 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 585 | |
771ed689 | 586 | /* |
79787eaa JM |
587 | * inline extent creation worked or returned error, |
588 | * we don't need to create any more async work items. | |
589 | * Unlock and free up our temp pages. | |
8b62f87b JB |
590 | * |
591 | * We use DO_ACCOUNTING here because we need the | |
592 | * delalloc_release_metadata to be done _after_ we drop | |
593 | * our outstanding extent for clearing delalloc for this | |
594 | * range. | |
771ed689 | 595 | */ |
ba8b04c1 QW |
596 | extent_clear_unlock_delalloc(inode, start, end, end, |
597 | NULL, clear_flags, | |
598 | PAGE_UNLOCK | | |
c2790a2e JB |
599 | PAGE_CLEAR_DIRTY | |
600 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 601 | page_error_op | |
c2790a2e | 602 | PAGE_END_WRITEBACK); |
c8b97818 CM |
603 | goto free_pages_out; |
604 | } | |
605 | } | |
606 | ||
607 | if (will_compress) { | |
608 | /* | |
609 | * we aren't doing an inline extent round the compressed size | |
610 | * up to a block size boundary so the allocator does sane | |
611 | * things | |
612 | */ | |
fda2832f | 613 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
614 | |
615 | /* | |
616 | * one last check to make sure the compression is really a | |
170607eb TT |
617 | * win, compare the page count read with the blocks on disk, |
618 | * compression must free at least one sector size | |
c8b97818 | 619 | */ |
09cbfeaf | 620 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 621 | if (total_compressed + blocksize <= total_in) { |
c8bb0c8b AS |
622 | *num_added += 1; |
623 | ||
624 | /* | |
625 | * The async work queues will take care of doing actual | |
626 | * allocation on disk for these compressed pages, and | |
627 | * will submit them to the elevator. | |
628 | */ | |
b5326271 | 629 | add_async_extent(async_chunk, start, total_in, |
4d3a800e | 630 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
631 | compress_type); |
632 | ||
1170862d TT |
633 | if (start + total_in < end) { |
634 | start += total_in; | |
c8bb0c8b AS |
635 | pages = NULL; |
636 | cond_resched(); | |
637 | goto again; | |
638 | } | |
639 | return; | |
c8b97818 CM |
640 | } |
641 | } | |
c8bb0c8b | 642 | if (pages) { |
c8b97818 CM |
643 | /* |
644 | * the compression code ran but failed to make things smaller, | |
645 | * free any pages it allocated and our page pointer array | |
646 | */ | |
4d3a800e | 647 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 648 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 649 | put_page(pages[i]); |
c8b97818 CM |
650 | } |
651 | kfree(pages); | |
652 | pages = NULL; | |
653 | total_compressed = 0; | |
4d3a800e | 654 | nr_pages = 0; |
c8b97818 CM |
655 | |
656 | /* flag the file so we don't compress in the future */ | |
0b246afa | 657 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 658 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 659 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 660 | } |
c8b97818 | 661 | } |
f03d9301 | 662 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
663 | /* |
664 | * No compression, but we still need to write the pages in the file | |
665 | * we've been given so far. redirty the locked page if it corresponds | |
666 | * to our extent and set things up for the async work queue to run | |
667 | * cow_file_range to do the normal delalloc dance. | |
668 | */ | |
1368c6da NB |
669 | if (page_offset(async_chunk->locked_page) >= start && |
670 | page_offset(async_chunk->locked_page) <= end) | |
671 | __set_page_dirty_nobuffers(async_chunk->locked_page); | |
c8bb0c8b AS |
672 | /* unlocked later on in the async handlers */ |
673 | ||
674 | if (redirty) | |
675 | extent_range_redirty_for_io(inode, start, end); | |
b5326271 | 676 | add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0, |
c8bb0c8b AS |
677 | BTRFS_COMPRESS_NONE); |
678 | *num_added += 1; | |
3b951516 | 679 | |
c44f649e | 680 | return; |
771ed689 CM |
681 | |
682 | free_pages_out: | |
4d3a800e | 683 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 684 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 685 | put_page(pages[i]); |
771ed689 | 686 | } |
d397712b | 687 | kfree(pages); |
771ed689 | 688 | } |
771ed689 | 689 | |
40ae837b FM |
690 | static void free_async_extent_pages(struct async_extent *async_extent) |
691 | { | |
692 | int i; | |
693 | ||
694 | if (!async_extent->pages) | |
695 | return; | |
696 | ||
697 | for (i = 0; i < async_extent->nr_pages; i++) { | |
698 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 699 | put_page(async_extent->pages[i]); |
40ae837b FM |
700 | } |
701 | kfree(async_extent->pages); | |
702 | async_extent->nr_pages = 0; | |
703 | async_extent->pages = NULL; | |
771ed689 CM |
704 | } |
705 | ||
706 | /* | |
707 | * phase two of compressed writeback. This is the ordered portion | |
708 | * of the code, which only gets called in the order the work was | |
709 | * queued. We walk all the async extents created by compress_file_range | |
710 | * and send them down to the disk. | |
711 | */ | |
b5326271 | 712 | static noinline void submit_compressed_extents(struct async_chunk *async_chunk) |
771ed689 | 713 | { |
b5326271 | 714 | struct inode *inode = async_chunk->inode; |
0b246afa | 715 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
716 | struct async_extent *async_extent; |
717 | u64 alloc_hint = 0; | |
771ed689 CM |
718 | struct btrfs_key ins; |
719 | struct extent_map *em; | |
720 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4336650a | 721 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
f5a84ee3 | 722 | int ret = 0; |
771ed689 | 723 | |
3e04e7f1 | 724 | again: |
b5326271 NB |
725 | while (!list_empty(&async_chunk->extents)) { |
726 | async_extent = list_entry(async_chunk->extents.next, | |
771ed689 CM |
727 | struct async_extent, list); |
728 | list_del(&async_extent->list); | |
c8b97818 | 729 | |
f5a84ee3 | 730 | retry: |
7447555f NB |
731 | lock_extent(io_tree, async_extent->start, |
732 | async_extent->start + async_extent->ram_size - 1); | |
771ed689 CM |
733 | /* did the compression code fall back to uncompressed IO? */ |
734 | if (!async_extent->pages) { | |
735 | int page_started = 0; | |
736 | unsigned long nr_written = 0; | |
737 | ||
771ed689 | 738 | /* allocate blocks */ |
b5326271 | 739 | ret = cow_file_range(inode, async_chunk->locked_page, |
f5a84ee3 JB |
740 | async_extent->start, |
741 | async_extent->start + | |
742 | async_extent->ram_size - 1, | |
dda3245e WX |
743 | async_extent->start + |
744 | async_extent->ram_size - 1, | |
745 | &page_started, &nr_written, 0, | |
746 | NULL); | |
771ed689 | 747 | |
79787eaa JM |
748 | /* JDM XXX */ |
749 | ||
771ed689 CM |
750 | /* |
751 | * if page_started, cow_file_range inserted an | |
752 | * inline extent and took care of all the unlocking | |
753 | * and IO for us. Otherwise, we need to submit | |
754 | * all those pages down to the drive. | |
755 | */ | |
f5a84ee3 | 756 | if (!page_started && !ret) |
5e3ee236 NB |
757 | extent_write_locked_range(inode, |
758 | async_extent->start, | |
d397712b | 759 | async_extent->start + |
771ed689 | 760 | async_extent->ram_size - 1, |
771ed689 | 761 | WB_SYNC_ALL); |
3e04e7f1 | 762 | else if (ret) |
b5326271 | 763 | unlock_page(async_chunk->locked_page); |
771ed689 CM |
764 | kfree(async_extent); |
765 | cond_resched(); | |
766 | continue; | |
767 | } | |
768 | ||
18513091 | 769 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
770 | async_extent->compressed_size, |
771 | async_extent->compressed_size, | |
e570fd27 | 772 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 773 | if (ret) { |
40ae837b | 774 | free_async_extent_pages(async_extent); |
3e04e7f1 | 775 | |
fdf8e2ea JB |
776 | if (ret == -ENOSPC) { |
777 | unlock_extent(io_tree, async_extent->start, | |
778 | async_extent->start + | |
779 | async_extent->ram_size - 1); | |
ce62003f LB |
780 | |
781 | /* | |
782 | * we need to redirty the pages if we decide to | |
783 | * fallback to uncompressed IO, otherwise we | |
784 | * will not submit these pages down to lower | |
785 | * layers. | |
786 | */ | |
787 | extent_range_redirty_for_io(inode, | |
788 | async_extent->start, | |
789 | async_extent->start + | |
790 | async_extent->ram_size - 1); | |
791 | ||
79787eaa | 792 | goto retry; |
fdf8e2ea | 793 | } |
3e04e7f1 | 794 | goto out_free; |
f5a84ee3 | 795 | } |
c2167754 YZ |
796 | /* |
797 | * here we're doing allocation and writeback of the | |
798 | * compressed pages | |
799 | */ | |
6f9994db LB |
800 | em = create_io_em(inode, async_extent->start, |
801 | async_extent->ram_size, /* len */ | |
802 | async_extent->start, /* orig_start */ | |
803 | ins.objectid, /* block_start */ | |
804 | ins.offset, /* block_len */ | |
805 | ins.offset, /* orig_block_len */ | |
806 | async_extent->ram_size, /* ram_bytes */ | |
807 | async_extent->compress_type, | |
808 | BTRFS_ORDERED_COMPRESSED); | |
809 | if (IS_ERR(em)) | |
810 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 811 | goto out_free_reserve; |
6f9994db | 812 | free_extent_map(em); |
3e04e7f1 | 813 | |
261507a0 LZ |
814 | ret = btrfs_add_ordered_extent_compress(inode, |
815 | async_extent->start, | |
816 | ins.objectid, | |
817 | async_extent->ram_size, | |
818 | ins.offset, | |
819 | BTRFS_ORDERED_COMPRESSED, | |
820 | async_extent->compress_type); | |
d9f85963 | 821 | if (ret) { |
dcdbc059 NB |
822 | btrfs_drop_extent_cache(BTRFS_I(inode), |
823 | async_extent->start, | |
d9f85963 FM |
824 | async_extent->start + |
825 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 826 | goto out_free_reserve; |
d9f85963 | 827 | } |
0b246afa | 828 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 829 | |
771ed689 CM |
830 | /* |
831 | * clear dirty, set writeback and unlock the pages. | |
832 | */ | |
c2790a2e | 833 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
834 | async_extent->start + |
835 | async_extent->ram_size - 1, | |
a791e35e CM |
836 | async_extent->start + |
837 | async_extent->ram_size - 1, | |
151a41bc JB |
838 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
839 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 840 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 841 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
842 | async_extent->start, |
843 | async_extent->ram_size, | |
844 | ins.objectid, | |
845 | ins.offset, async_extent->pages, | |
f82b7359 | 846 | async_extent->nr_pages, |
b5326271 | 847 | async_chunk->write_flags)) { |
fce2a4e6 FM |
848 | struct page *p = async_extent->pages[0]; |
849 | const u64 start = async_extent->start; | |
850 | const u64 end = start + async_extent->ram_size - 1; | |
851 | ||
852 | p->mapping = inode->i_mapping; | |
c629732d | 853 | btrfs_writepage_endio_finish_ordered(p, start, end, 0); |
7087a9d8 | 854 | |
fce2a4e6 | 855 | p->mapping = NULL; |
ba8b04c1 QW |
856 | extent_clear_unlock_delalloc(inode, start, end, end, |
857 | NULL, 0, | |
fce2a4e6 FM |
858 | PAGE_END_WRITEBACK | |
859 | PAGE_SET_ERROR); | |
40ae837b | 860 | free_async_extent_pages(async_extent); |
fce2a4e6 | 861 | } |
771ed689 CM |
862 | alloc_hint = ins.objectid + ins.offset; |
863 | kfree(async_extent); | |
864 | cond_resched(); | |
865 | } | |
dec8f175 | 866 | return; |
3e04e7f1 | 867 | out_free_reserve: |
0b246afa | 868 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 869 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 870 | out_free: |
c2790a2e | 871 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
872 | async_extent->start + |
873 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
874 | async_extent->start + |
875 | async_extent->ram_size - 1, | |
c2790a2e | 876 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 877 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
878 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
879 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
880 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
881 | PAGE_SET_ERROR); | |
40ae837b | 882 | free_async_extent_pages(async_extent); |
79787eaa | 883 | kfree(async_extent); |
3e04e7f1 | 884 | goto again; |
771ed689 CM |
885 | } |
886 | ||
4b46fce2 JB |
887 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
888 | u64 num_bytes) | |
889 | { | |
890 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
891 | struct extent_map *em; | |
892 | u64 alloc_hint = 0; | |
893 | ||
894 | read_lock(&em_tree->lock); | |
895 | em = search_extent_mapping(em_tree, start, num_bytes); | |
896 | if (em) { | |
897 | /* | |
898 | * if block start isn't an actual block number then find the | |
899 | * first block in this inode and use that as a hint. If that | |
900 | * block is also bogus then just don't worry about it. | |
901 | */ | |
902 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
903 | free_extent_map(em); | |
904 | em = search_extent_mapping(em_tree, 0, 0); | |
905 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
906 | alloc_hint = em->block_start; | |
907 | if (em) | |
908 | free_extent_map(em); | |
909 | } else { | |
910 | alloc_hint = em->block_start; | |
911 | free_extent_map(em); | |
912 | } | |
913 | } | |
914 | read_unlock(&em_tree->lock); | |
915 | ||
916 | return alloc_hint; | |
917 | } | |
918 | ||
771ed689 CM |
919 | /* |
920 | * when extent_io.c finds a delayed allocation range in the file, | |
921 | * the call backs end up in this code. The basic idea is to | |
922 | * allocate extents on disk for the range, and create ordered data structs | |
923 | * in ram to track those extents. | |
924 | * | |
925 | * locked_page is the page that writepage had locked already. We use | |
926 | * it to make sure we don't do extra locks or unlocks. | |
927 | * | |
928 | * *page_started is set to one if we unlock locked_page and do everything | |
929 | * required to start IO on it. It may be clean and already done with | |
930 | * IO when we return. | |
931 | */ | |
00361589 JB |
932 | static noinline int cow_file_range(struct inode *inode, |
933 | struct page *locked_page, | |
dda3245e WX |
934 | u64 start, u64 end, u64 delalloc_end, |
935 | int *page_started, unsigned long *nr_written, | |
936 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 937 | { |
0b246afa | 938 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 939 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
940 | u64 alloc_hint = 0; |
941 | u64 num_bytes; | |
942 | unsigned long ram_size; | |
a315e68f | 943 | u64 cur_alloc_size = 0; |
0b246afa | 944 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
945 | struct btrfs_key ins; |
946 | struct extent_map *em; | |
a315e68f FM |
947 | unsigned clear_bits; |
948 | unsigned long page_ops; | |
949 | bool extent_reserved = false; | |
771ed689 CM |
950 | int ret = 0; |
951 | ||
70ddc553 | 952 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 953 | WARN_ON_ONCE(1); |
29bce2f3 JB |
954 | ret = -EINVAL; |
955 | goto out_unlock; | |
02ecd2c2 | 956 | } |
771ed689 | 957 | |
fda2832f | 958 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 | 959 | num_bytes = max(blocksize, num_bytes); |
566b1760 | 960 | ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 961 | |
6158e1ce | 962 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 963 | |
771ed689 CM |
964 | if (start == 0) { |
965 | /* lets try to make an inline extent */ | |
d02c0e20 NB |
966 | ret = cow_file_range_inline(inode, start, end, 0, |
967 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 968 | if (ret == 0) { |
8b62f87b JB |
969 | /* |
970 | * We use DO_ACCOUNTING here because we need the | |
971 | * delalloc_release_metadata to be run _after_ we drop | |
972 | * our outstanding extent for clearing delalloc for this | |
973 | * range. | |
974 | */ | |
ba8b04c1 QW |
975 | extent_clear_unlock_delalloc(inode, start, end, |
976 | delalloc_end, NULL, | |
c2790a2e | 977 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
978 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
979 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
980 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
981 | PAGE_END_WRITEBACK); | |
771ed689 | 982 | *nr_written = *nr_written + |
09cbfeaf | 983 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 984 | *page_started = 1; |
771ed689 | 985 | goto out; |
79787eaa | 986 | } else if (ret < 0) { |
79787eaa | 987 | goto out_unlock; |
771ed689 CM |
988 | } |
989 | } | |
990 | ||
4b46fce2 | 991 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
992 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
993 | start + num_bytes - 1, 0); | |
771ed689 | 994 | |
3752d22f AJ |
995 | while (num_bytes > 0) { |
996 | cur_alloc_size = num_bytes; | |
18513091 | 997 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 998 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 999 | &ins, 1, 1); |
00361589 | 1000 | if (ret < 0) |
79787eaa | 1001 | goto out_unlock; |
a315e68f FM |
1002 | cur_alloc_size = ins.offset; |
1003 | extent_reserved = true; | |
d397712b | 1004 | |
771ed689 | 1005 | ram_size = ins.offset; |
6f9994db LB |
1006 | em = create_io_em(inode, start, ins.offset, /* len */ |
1007 | start, /* orig_start */ | |
1008 | ins.objectid, /* block_start */ | |
1009 | ins.offset, /* block_len */ | |
1010 | ins.offset, /* orig_block_len */ | |
1011 | ram_size, /* ram_bytes */ | |
1012 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1013 | BTRFS_ORDERED_REGULAR /* type */); |
090a127a SY |
1014 | if (IS_ERR(em)) { |
1015 | ret = PTR_ERR(em); | |
ace68bac | 1016 | goto out_reserve; |
090a127a | 1017 | } |
6f9994db | 1018 | free_extent_map(em); |
e6dcd2dc | 1019 | |
e6dcd2dc | 1020 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1021 | ram_size, cur_alloc_size, 0); |
ace68bac | 1022 | if (ret) |
d9f85963 | 1023 | goto out_drop_extent_cache; |
c8b97818 | 1024 | |
17d217fe YZ |
1025 | if (root->root_key.objectid == |
1026 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1027 | ret = btrfs_reloc_clone_csums(inode, start, | |
1028 | cur_alloc_size); | |
4dbd80fb QW |
1029 | /* |
1030 | * Only drop cache here, and process as normal. | |
1031 | * | |
1032 | * We must not allow extent_clear_unlock_delalloc() | |
1033 | * at out_unlock label to free meta of this ordered | |
1034 | * extent, as its meta should be freed by | |
1035 | * btrfs_finish_ordered_io(). | |
1036 | * | |
1037 | * So we must continue until @start is increased to | |
1038 | * skip current ordered extent. | |
1039 | */ | |
00361589 | 1040 | if (ret) |
4dbd80fb QW |
1041 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1042 | start + ram_size - 1, 0); | |
17d217fe YZ |
1043 | } |
1044 | ||
0b246afa | 1045 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1046 | |
c8b97818 CM |
1047 | /* we're not doing compressed IO, don't unlock the first |
1048 | * page (which the caller expects to stay locked), don't | |
1049 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1050 | * |
1051 | * Do set the Private2 bit so we know this page was properly | |
1052 | * setup for writepage | |
c8b97818 | 1053 | */ |
a315e68f FM |
1054 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1055 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1056 | |
c2790a2e | 1057 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1058 | start + ram_size - 1, |
1059 | delalloc_end, locked_page, | |
c2790a2e | 1060 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1061 | page_ops); |
3752d22f AJ |
1062 | if (num_bytes < cur_alloc_size) |
1063 | num_bytes = 0; | |
4dbd80fb | 1064 | else |
3752d22f | 1065 | num_bytes -= cur_alloc_size; |
c59f8951 CM |
1066 | alloc_hint = ins.objectid + ins.offset; |
1067 | start += cur_alloc_size; | |
a315e68f | 1068 | extent_reserved = false; |
4dbd80fb QW |
1069 | |
1070 | /* | |
1071 | * btrfs_reloc_clone_csums() error, since start is increased | |
1072 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1073 | * free metadata of current ordered extent, we're OK to exit. | |
1074 | */ | |
1075 | if (ret) | |
1076 | goto out_unlock; | |
b888db2b | 1077 | } |
79787eaa | 1078 | out: |
be20aa9d | 1079 | return ret; |
b7d5b0a8 | 1080 | |
d9f85963 | 1081 | out_drop_extent_cache: |
dcdbc059 | 1082 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1083 | out_reserve: |
0b246afa | 1084 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1085 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1086 | out_unlock: |
a7e3b975 FM |
1087 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1088 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1089 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1090 | PAGE_END_WRITEBACK; | |
1091 | /* | |
1092 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1093 | * failed to create the respective ordered extent, then it means that | |
1094 | * when we reserved the extent we decremented the extent's size from | |
1095 | * the data space_info's bytes_may_use counter and incremented the | |
1096 | * space_info's bytes_reserved counter by the same amount. We must make | |
1097 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1098 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1099 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1100 | */ | |
1101 | if (extent_reserved) { | |
1102 | extent_clear_unlock_delalloc(inode, start, | |
1103 | start + cur_alloc_size, | |
1104 | start + cur_alloc_size, | |
1105 | locked_page, | |
1106 | clear_bits, | |
1107 | page_ops); | |
1108 | start += cur_alloc_size; | |
1109 | if (start >= end) | |
1110 | goto out; | |
1111 | } | |
ba8b04c1 QW |
1112 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1113 | locked_page, | |
a315e68f FM |
1114 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1115 | page_ops); | |
79787eaa | 1116 | goto out; |
771ed689 | 1117 | } |
c8b97818 | 1118 | |
771ed689 CM |
1119 | /* |
1120 | * work queue call back to started compression on a file and pages | |
1121 | */ | |
1122 | static noinline void async_cow_start(struct btrfs_work *work) | |
1123 | { | |
b5326271 | 1124 | struct async_chunk *async_chunk; |
771ed689 | 1125 | int num_added = 0; |
771ed689 | 1126 | |
b5326271 | 1127 | async_chunk = container_of(work, struct async_chunk, work); |
771ed689 | 1128 | |
1368c6da | 1129 | compress_file_range(async_chunk, &num_added); |
8180ef88 | 1130 | if (num_added == 0) { |
b5326271 NB |
1131 | btrfs_add_delayed_iput(async_chunk->inode); |
1132 | async_chunk->inode = NULL; | |
8180ef88 | 1133 | } |
771ed689 CM |
1134 | } |
1135 | ||
1136 | /* | |
1137 | * work queue call back to submit previously compressed pages | |
1138 | */ | |
1139 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1140 | { | |
c5a68aec NB |
1141 | struct async_chunk *async_chunk = container_of(work, struct async_chunk, |
1142 | work); | |
1143 | struct btrfs_fs_info *fs_info = btrfs_work_owner(work); | |
771ed689 CM |
1144 | unsigned long nr_pages; |
1145 | ||
b5326271 | 1146 | nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >> |
09cbfeaf | 1147 | PAGE_SHIFT; |
771ed689 | 1148 | |
093258e6 | 1149 | /* atomic_sub_return implies a barrier */ |
0b246afa | 1150 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
093258e6 DS |
1151 | 5 * SZ_1M) |
1152 | cond_wake_up_nomb(&fs_info->async_submit_wait); | |
771ed689 | 1153 | |
4546d178 | 1154 | /* |
b5326271 | 1155 | * ->inode could be NULL if async_chunk_start has failed to compress, |
4546d178 NB |
1156 | * in which case we don't have anything to submit, yet we need to |
1157 | * always adjust ->async_delalloc_pages as its paired with the init | |
1158 | * happening in cow_file_range_async | |
1159 | */ | |
b5326271 NB |
1160 | if (async_chunk->inode) |
1161 | submit_compressed_extents(async_chunk); | |
771ed689 | 1162 | } |
c8b97818 | 1163 | |
771ed689 CM |
1164 | static noinline void async_cow_free(struct btrfs_work *work) |
1165 | { | |
b5326271 | 1166 | struct async_chunk *async_chunk; |
97db1204 | 1167 | |
b5326271 NB |
1168 | async_chunk = container_of(work, struct async_chunk, work); |
1169 | if (async_chunk->inode) | |
1170 | btrfs_add_delayed_iput(async_chunk->inode); | |
97db1204 NB |
1171 | /* |
1172 | * Since the pointer to 'pending' is at the beginning of the array of | |
b5326271 | 1173 | * async_chunk's, freeing it ensures the whole array has been freed. |
97db1204 | 1174 | */ |
b5326271 | 1175 | if (atomic_dec_and_test(async_chunk->pending)) |
b1c16ac9 | 1176 | kvfree(async_chunk->pending); |
771ed689 CM |
1177 | } |
1178 | ||
1179 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1180 | u64 start, u64 end, int *page_started, | |
f82b7359 LB |
1181 | unsigned long *nr_written, |
1182 | unsigned int write_flags) | |
771ed689 | 1183 | { |
0b246afa | 1184 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
97db1204 NB |
1185 | struct async_cow *ctx; |
1186 | struct async_chunk *async_chunk; | |
771ed689 CM |
1187 | unsigned long nr_pages; |
1188 | u64 cur_end; | |
97db1204 NB |
1189 | u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K); |
1190 | int i; | |
1191 | bool should_compress; | |
b1c16ac9 | 1192 | unsigned nofs_flag; |
771ed689 | 1193 | |
69684c5a | 1194 | unlock_extent(&BTRFS_I(inode)->io_tree, start, end); |
97db1204 NB |
1195 | |
1196 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && | |
1197 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) { | |
1198 | num_chunks = 1; | |
1199 | should_compress = false; | |
1200 | } else { | |
1201 | should_compress = true; | |
1202 | } | |
1203 | ||
b1c16ac9 NB |
1204 | nofs_flag = memalloc_nofs_save(); |
1205 | ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL); | |
1206 | memalloc_nofs_restore(nofs_flag); | |
1207 | ||
97db1204 NB |
1208 | if (!ctx) { |
1209 | unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | | |
1210 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | | |
1211 | EXTENT_DO_ACCOUNTING; | |
1212 | unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1213 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | | |
1214 | PAGE_SET_ERROR; | |
1215 | ||
1216 | extent_clear_unlock_delalloc(inode, start, end, 0, locked_page, | |
1217 | clear_bits, page_ops); | |
1218 | return -ENOMEM; | |
1219 | } | |
1220 | ||
1221 | async_chunk = ctx->chunks; | |
1222 | atomic_set(&ctx->num_chunks, num_chunks); | |
1223 | ||
1224 | for (i = 0; i < num_chunks; i++) { | |
1225 | if (should_compress) | |
1226 | cur_end = min(end, start + SZ_512K - 1); | |
1227 | else | |
1228 | cur_end = end; | |
771ed689 | 1229 | |
bd4691a0 NB |
1230 | /* |
1231 | * igrab is called higher up in the call chain, take only the | |
1232 | * lightweight reference for the callback lifetime | |
1233 | */ | |
1234 | ihold(inode); | |
97db1204 NB |
1235 | async_chunk[i].pending = &ctx->num_chunks; |
1236 | async_chunk[i].inode = inode; | |
1237 | async_chunk[i].start = start; | |
1238 | async_chunk[i].end = cur_end; | |
97db1204 NB |
1239 | async_chunk[i].locked_page = locked_page; |
1240 | async_chunk[i].write_flags = write_flags; | |
1241 | INIT_LIST_HEAD(&async_chunk[i].extents); | |
1242 | ||
1243 | btrfs_init_work(&async_chunk[i].work, | |
9e0af237 LB |
1244 | btrfs_delalloc_helper, |
1245 | async_cow_start, async_cow_submit, | |
1246 | async_cow_free); | |
771ed689 | 1247 | |
97db1204 | 1248 | nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE); |
0b246afa | 1249 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1250 | |
97db1204 | 1251 | btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work); |
771ed689 | 1252 | |
771ed689 CM |
1253 | *nr_written += nr_pages; |
1254 | start = cur_end + 1; | |
1255 | } | |
1256 | *page_started = 1; | |
1257 | return 0; | |
be20aa9d CM |
1258 | } |
1259 | ||
2ff7e61e | 1260 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1261 | u64 bytenr, u64 num_bytes) |
1262 | { | |
1263 | int ret; | |
1264 | struct btrfs_ordered_sum *sums; | |
1265 | LIST_HEAD(list); | |
1266 | ||
0b246afa | 1267 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1268 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1269 | if (ret == 0 && list_empty(&list)) |
1270 | return 0; | |
1271 | ||
1272 | while (!list_empty(&list)) { | |
1273 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1274 | list_del(&sums->list); | |
1275 | kfree(sums); | |
1276 | } | |
58113753 LB |
1277 | if (ret < 0) |
1278 | return ret; | |
17d217fe YZ |
1279 | return 1; |
1280 | } | |
1281 | ||
d352ac68 CM |
1282 | /* |
1283 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1284 | * of the extents that exist in the file, and COWs the file as required. | |
1285 | * | |
1286 | * If no cow copies or snapshots exist, we write directly to the existing | |
1287 | * blocks on disk | |
1288 | */ | |
7f366cfe CM |
1289 | static noinline int run_delalloc_nocow(struct inode *inode, |
1290 | struct page *locked_page, | |
771ed689 CM |
1291 | u64 start, u64 end, int *page_started, int force, |
1292 | unsigned long *nr_written) | |
be20aa9d | 1293 | { |
0b246afa | 1294 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1295 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1296 | struct extent_buffer *leaf; | |
be20aa9d | 1297 | struct btrfs_path *path; |
80ff3856 | 1298 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1299 | struct btrfs_key found_key; |
6f9994db | 1300 | struct extent_map *em; |
80ff3856 YZ |
1301 | u64 cow_start; |
1302 | u64 cur_offset; | |
1303 | u64 extent_end; | |
5d4f98a2 | 1304 | u64 extent_offset; |
80ff3856 YZ |
1305 | u64 disk_bytenr; |
1306 | u64 num_bytes; | |
b4939680 | 1307 | u64 disk_num_bytes; |
cc95bef6 | 1308 | u64 ram_bytes; |
80ff3856 | 1309 | int extent_type; |
8ecebf4d | 1310 | int ret; |
d899e052 | 1311 | int type; |
80ff3856 YZ |
1312 | int nocow; |
1313 | int check_prev = 1; | |
82d5902d | 1314 | bool nolock; |
4a0cc7ca | 1315 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1316 | |
1317 | path = btrfs_alloc_path(); | |
17ca04af | 1318 | if (!path) { |
ba8b04c1 QW |
1319 | extent_clear_unlock_delalloc(inode, start, end, end, |
1320 | locked_page, | |
c2790a2e | 1321 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1322 | EXTENT_DO_ACCOUNTING | |
1323 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1324 | PAGE_CLEAR_DIRTY | |
1325 | PAGE_SET_WRITEBACK | | |
1326 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1327 | return -ENOMEM; |
17ca04af | 1328 | } |
82d5902d | 1329 | |
70ddc553 | 1330 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1331 | |
80ff3856 YZ |
1332 | cow_start = (u64)-1; |
1333 | cur_offset = start; | |
1334 | while (1) { | |
e4c3b2dc | 1335 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1336 | cur_offset, 0); |
d788a349 | 1337 | if (ret < 0) |
79787eaa | 1338 | goto error; |
80ff3856 YZ |
1339 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1340 | leaf = path->nodes[0]; | |
1341 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1342 | path->slots[0] - 1); | |
33345d01 | 1343 | if (found_key.objectid == ino && |
80ff3856 YZ |
1344 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1345 | path->slots[0]--; | |
1346 | } | |
1347 | check_prev = 0; | |
1348 | next_slot: | |
1349 | leaf = path->nodes[0]; | |
1350 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1351 | ret = btrfs_next_leaf(root, path); | |
e8916699 LB |
1352 | if (ret < 0) { |
1353 | if (cow_start != (u64)-1) | |
1354 | cur_offset = cow_start; | |
79787eaa | 1355 | goto error; |
e8916699 | 1356 | } |
80ff3856 YZ |
1357 | if (ret > 0) |
1358 | break; | |
1359 | leaf = path->nodes[0]; | |
1360 | } | |
be20aa9d | 1361 | |
80ff3856 YZ |
1362 | nocow = 0; |
1363 | disk_bytenr = 0; | |
17d217fe | 1364 | num_bytes = 0; |
80ff3856 YZ |
1365 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1366 | ||
1d512cb7 FM |
1367 | if (found_key.objectid > ino) |
1368 | break; | |
1369 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1370 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1371 | path->slots[0]++; | |
1372 | goto next_slot; | |
1373 | } | |
1374 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1375 | found_key.offset > end) |
1376 | break; | |
1377 | ||
1378 | if (found_key.offset > cur_offset) { | |
1379 | extent_end = found_key.offset; | |
e9061e21 | 1380 | extent_type = 0; |
80ff3856 YZ |
1381 | goto out_check; |
1382 | } | |
1383 | ||
1384 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1385 | struct btrfs_file_extent_item); | |
1386 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1387 | ||
cc95bef6 | 1388 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1389 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1390 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1391 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1392 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1393 | extent_end = found_key.offset + |
1394 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1395 | disk_num_bytes = |
1396 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1397 | if (extent_end <= start) { |
1398 | path->slots[0]++; | |
1399 | goto next_slot; | |
1400 | } | |
17d217fe YZ |
1401 | if (disk_bytenr == 0) |
1402 | goto out_check; | |
80ff3856 YZ |
1403 | if (btrfs_file_extent_compression(leaf, fi) || |
1404 | btrfs_file_extent_encryption(leaf, fi) || | |
1405 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1406 | goto out_check; | |
78d4295b EL |
1407 | /* |
1408 | * Do the same check as in btrfs_cross_ref_exist but | |
1409 | * without the unnecessary search. | |
1410 | */ | |
27a7ff55 LF |
1411 | if (!nolock && |
1412 | btrfs_file_extent_generation(leaf, fi) <= | |
78d4295b EL |
1413 | btrfs_root_last_snapshot(&root->root_item)) |
1414 | goto out_check; | |
d899e052 YZ |
1415 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1416 | goto out_check; | |
2ff7e61e | 1417 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1418 | goto out_check; |
58113753 LB |
1419 | ret = btrfs_cross_ref_exist(root, ino, |
1420 | found_key.offset - | |
1421 | extent_offset, disk_bytenr); | |
1422 | if (ret) { | |
1423 | /* | |
1424 | * ret could be -EIO if the above fails to read | |
1425 | * metadata. | |
1426 | */ | |
1427 | if (ret < 0) { | |
1428 | if (cow_start != (u64)-1) | |
1429 | cur_offset = cow_start; | |
1430 | goto error; | |
1431 | } | |
1432 | ||
1433 | WARN_ON_ONCE(nolock); | |
17d217fe | 1434 | goto out_check; |
58113753 | 1435 | } |
5d4f98a2 | 1436 | disk_bytenr += extent_offset; |
17d217fe YZ |
1437 | disk_bytenr += cur_offset - found_key.offset; |
1438 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1439 | /* |
1440 | * if there are pending snapshots for this root, | |
1441 | * we fall into common COW way. | |
1442 | */ | |
8ecebf4d RK |
1443 | if (!nolock && atomic_read(&root->snapshot_force_cow)) |
1444 | goto out_check; | |
17d217fe YZ |
1445 | /* |
1446 | * force cow if csum exists in the range. | |
1447 | * this ensure that csum for a given extent are | |
1448 | * either valid or do not exist. | |
1449 | */ | |
58113753 LB |
1450 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1451 | num_bytes); | |
1452 | if (ret) { | |
58113753 LB |
1453 | /* |
1454 | * ret could be -EIO if the above fails to read | |
1455 | * metadata. | |
1456 | */ | |
1457 | if (ret < 0) { | |
1458 | if (cow_start != (u64)-1) | |
1459 | cur_offset = cow_start; | |
1460 | goto error; | |
1461 | } | |
1462 | WARN_ON_ONCE(nolock); | |
17d217fe | 1463 | goto out_check; |
91e1f56a | 1464 | } |
8ecebf4d | 1465 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1466 | goto out_check; |
80ff3856 YZ |
1467 | nocow = 1; |
1468 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1469 | extent_end = found_key.offset + | |
e41ca589 | 1470 | btrfs_file_extent_ram_bytes(leaf, fi); |
da17066c | 1471 | extent_end = ALIGN(extent_end, |
0b246afa | 1472 | fs_info->sectorsize); |
80ff3856 | 1473 | } else { |
290342f6 | 1474 | BUG(); |
80ff3856 YZ |
1475 | } |
1476 | out_check: | |
1477 | if (extent_end <= start) { | |
1478 | path->slots[0]++; | |
f78c436c | 1479 | if (nocow) |
0b246afa | 1480 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1481 | goto next_slot; |
1482 | } | |
1483 | if (!nocow) { | |
1484 | if (cow_start == (u64)-1) | |
1485 | cow_start = cur_offset; | |
1486 | cur_offset = extent_end; | |
1487 | if (cur_offset > end) | |
1488 | break; | |
1489 | path->slots[0]++; | |
1490 | goto next_slot; | |
7ea394f1 YZ |
1491 | } |
1492 | ||
b3b4aa74 | 1493 | btrfs_release_path(path); |
80ff3856 | 1494 | if (cow_start != (u64)-1) { |
00361589 JB |
1495 | ret = cow_file_range(inode, locked_page, |
1496 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1497 | end, page_started, nr_written, 1, |
1498 | NULL); | |
e9894fd3 | 1499 | if (ret) { |
f78c436c | 1500 | if (nocow) |
0b246afa | 1501 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1502 | disk_bytenr); |
79787eaa | 1503 | goto error; |
e9894fd3 | 1504 | } |
80ff3856 | 1505 | cow_start = (u64)-1; |
7ea394f1 | 1506 | } |
80ff3856 | 1507 | |
d899e052 | 1508 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1509 | u64 orig_start = found_key.offset - extent_offset; |
1510 | ||
1511 | em = create_io_em(inode, cur_offset, num_bytes, | |
1512 | orig_start, | |
1513 | disk_bytenr, /* block_start */ | |
1514 | num_bytes, /* block_len */ | |
1515 | disk_num_bytes, /* orig_block_len */ | |
1516 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1517 | BTRFS_ORDERED_PREALLOC); | |
1518 | if (IS_ERR(em)) { | |
6f9994db LB |
1519 | if (nocow) |
1520 | btrfs_dec_nocow_writers(fs_info, | |
1521 | disk_bytenr); | |
1522 | ret = PTR_ERR(em); | |
1523 | goto error; | |
d899e052 | 1524 | } |
6f9994db LB |
1525 | free_extent_map(em); |
1526 | } | |
1527 | ||
1528 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1529 | type = BTRFS_ORDERED_PREALLOC; |
1530 | } else { | |
1531 | type = BTRFS_ORDERED_NOCOW; | |
1532 | } | |
80ff3856 YZ |
1533 | |
1534 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1535 | num_bytes, num_bytes, type); |
f78c436c | 1536 | if (nocow) |
0b246afa | 1537 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1538 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1539 | |
efa56464 | 1540 | if (root->root_key.objectid == |
4dbd80fb QW |
1541 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1542 | /* | |
1543 | * Error handled later, as we must prevent | |
1544 | * extent_clear_unlock_delalloc() in error handler | |
1545 | * from freeing metadata of created ordered extent. | |
1546 | */ | |
efa56464 YZ |
1547 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1548 | num_bytes); | |
efa56464 | 1549 | |
c2790a2e | 1550 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1551 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1552 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1553 | EXTENT_DELALLOC | |
1554 | EXTENT_CLEAR_DATA_RESV, | |
1555 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1556 | ||
80ff3856 | 1557 | cur_offset = extent_end; |
4dbd80fb QW |
1558 | |
1559 | /* | |
1560 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1561 | * handler, as metadata for created ordered extent will only | |
1562 | * be freed by btrfs_finish_ordered_io(). | |
1563 | */ | |
1564 | if (ret) | |
1565 | goto error; | |
80ff3856 YZ |
1566 | if (cur_offset > end) |
1567 | break; | |
be20aa9d | 1568 | } |
b3b4aa74 | 1569 | btrfs_release_path(path); |
80ff3856 | 1570 | |
506481b2 | 1571 | if (cur_offset <= end && cow_start == (u64)-1) |
80ff3856 | 1572 | cow_start = cur_offset; |
17ca04af | 1573 | |
80ff3856 | 1574 | if (cow_start != (u64)-1) { |
506481b2 | 1575 | cur_offset = end; |
dda3245e WX |
1576 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1577 | page_started, nr_written, 1, NULL); | |
d788a349 | 1578 | if (ret) |
79787eaa | 1579 | goto error; |
80ff3856 YZ |
1580 | } |
1581 | ||
79787eaa | 1582 | error: |
17ca04af | 1583 | if (ret && cur_offset < end) |
ba8b04c1 | 1584 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1585 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1586 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1587 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1588 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1589 | PAGE_SET_WRITEBACK | |
1590 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1591 | btrfs_free_path(path); |
79787eaa | 1592 | return ret; |
be20aa9d CM |
1593 | } |
1594 | ||
47059d93 WS |
1595 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1596 | { | |
1597 | ||
1598 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1599 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1600 | return 0; | |
1601 | ||
1602 | /* | |
1603 | * @defrag_bytes is a hint value, no spinlock held here, | |
1604 | * if is not zero, it means the file is defragging. | |
1605 | * Force cow if given extent needs to be defragged. | |
1606 | */ | |
1607 | if (BTRFS_I(inode)->defrag_bytes && | |
1608 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1609 | EXTENT_DEFRAG, 0, NULL)) | |
1610 | return 1; | |
1611 | ||
1612 | return 0; | |
1613 | } | |
1614 | ||
d352ac68 | 1615 | /* |
5eaad97a NB |
1616 | * Function to process delayed allocation (create CoW) for ranges which are |
1617 | * being touched for the first time. | |
d352ac68 | 1618 | */ |
bc9a8bf7 | 1619 | int btrfs_run_delalloc_range(struct inode *inode, struct page *locked_page, |
5eaad97a NB |
1620 | u64 start, u64 end, int *page_started, unsigned long *nr_written, |
1621 | struct writeback_control *wbc) | |
be20aa9d | 1622 | { |
be20aa9d | 1623 | int ret; |
47059d93 | 1624 | int force_cow = need_force_cow(inode, start, end); |
f82b7359 | 1625 | unsigned int write_flags = wbc_to_write_flags(wbc); |
a2135011 | 1626 | |
47059d93 | 1627 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1628 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1629 | page_started, 1, nr_written); |
47059d93 | 1630 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1631 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1632 | page_started, 0, nr_written); |
c2fcdcdf | 1633 | } else if (!inode_need_compress(inode, start, end)) { |
dda3245e WX |
1634 | ret = cow_file_range(inode, locked_page, start, end, end, |
1635 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1636 | } else { |
1637 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1638 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1639 | ret = cow_file_range_async(inode, locked_page, start, end, |
f82b7359 LB |
1640 | page_started, nr_written, |
1641 | write_flags); | |
7ddf5a42 | 1642 | } |
52427260 | 1643 | if (ret) |
d1051d6e NB |
1644 | btrfs_cleanup_ordered_extents(inode, locked_page, start, |
1645 | end - start + 1); | |
b888db2b CM |
1646 | return ret; |
1647 | } | |
1648 | ||
abbb55f4 NB |
1649 | void btrfs_split_delalloc_extent(struct inode *inode, |
1650 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1651 | { |
dcab6a3b JB |
1652 | u64 size; |
1653 | ||
0ca1f7ce | 1654 | /* not delalloc, ignore it */ |
9ed74f2d | 1655 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1656 | return; |
9ed74f2d | 1657 | |
dcab6a3b JB |
1658 | size = orig->end - orig->start + 1; |
1659 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1660 | u32 num_extents; |
dcab6a3b JB |
1661 | u64 new_size; |
1662 | ||
1663 | /* | |
5c848198 | 1664 | * See the explanation in btrfs_merge_delalloc_extent, the same |
ba117213 | 1665 | * applies here, just in reverse. |
dcab6a3b JB |
1666 | */ |
1667 | new_size = orig->end - split + 1; | |
823bb20a | 1668 | num_extents = count_max_extents(new_size); |
ba117213 | 1669 | new_size = split - orig->start; |
823bb20a DS |
1670 | num_extents += count_max_extents(new_size); |
1671 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1672 | return; |
1673 | } | |
1674 | ||
9e0baf60 | 1675 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1676 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1677 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1678 | } |
1679 | ||
1680 | /* | |
5c848198 NB |
1681 | * Handle merged delayed allocation extents so we can keep track of new extents |
1682 | * that are just merged onto old extents, such as when we are doing sequential | |
1683 | * writes, so we can properly account for the metadata space we'll need. | |
9ed74f2d | 1684 | */ |
5c848198 NB |
1685 | void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new, |
1686 | struct extent_state *other) | |
9ed74f2d | 1687 | { |
dcab6a3b | 1688 | u64 new_size, old_size; |
823bb20a | 1689 | u32 num_extents; |
dcab6a3b | 1690 | |
9ed74f2d JB |
1691 | /* not delalloc, ignore it */ |
1692 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1693 | return; |
9ed74f2d | 1694 | |
8461a3de JB |
1695 | if (new->start > other->start) |
1696 | new_size = new->end - other->start + 1; | |
1697 | else | |
1698 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1699 | |
1700 | /* we're not bigger than the max, unreserve the space and go */ | |
1701 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1702 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1703 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1704 | spin_unlock(&BTRFS_I(inode)->lock); |
1705 | return; | |
1706 | } | |
1707 | ||
1708 | /* | |
ba117213 JB |
1709 | * We have to add up either side to figure out how many extents were |
1710 | * accounted for before we merged into one big extent. If the number of | |
1711 | * extents we accounted for is <= the amount we need for the new range | |
1712 | * then we can return, otherwise drop. Think of it like this | |
1713 | * | |
1714 | * [ 4k][MAX_SIZE] | |
1715 | * | |
1716 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1717 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1718 | * we have 1 so they are == and we can return. But in this case | |
1719 | * | |
1720 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1721 | * | |
1722 | * Each range on their own accounts for 2 extents, but merged together | |
1723 | * they are only 3 extents worth of accounting, so we need to drop in | |
1724 | * this case. | |
dcab6a3b | 1725 | */ |
ba117213 | 1726 | old_size = other->end - other->start + 1; |
823bb20a | 1727 | num_extents = count_max_extents(old_size); |
ba117213 | 1728 | old_size = new->end - new->start + 1; |
823bb20a DS |
1729 | num_extents += count_max_extents(old_size); |
1730 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1731 | return; |
1732 | ||
9e0baf60 | 1733 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1734 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1735 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1736 | } |
1737 | ||
eb73c1b7 MX |
1738 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1739 | struct inode *inode) | |
1740 | { | |
0b246afa JM |
1741 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1742 | ||
eb73c1b7 MX |
1743 | spin_lock(&root->delalloc_lock); |
1744 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1745 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1746 | &root->delalloc_inodes); | |
1747 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1748 | &BTRFS_I(inode)->runtime_flags); | |
1749 | root->nr_delalloc_inodes++; | |
1750 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1751 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1752 | BUG_ON(!list_empty(&root->delalloc_root)); |
1753 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1754 | &fs_info->delalloc_roots); |
1755 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1756 | } |
1757 | } | |
1758 | spin_unlock(&root->delalloc_lock); | |
1759 | } | |
1760 | ||
2b877331 NB |
1761 | |
1762 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1763 | struct btrfs_inode *inode) | |
eb73c1b7 | 1764 | { |
3ffbd68c | 1765 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b246afa | 1766 | |
9e3e97f4 NB |
1767 | if (!list_empty(&inode->delalloc_inodes)) { |
1768 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1769 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1770 | &inode->runtime_flags); |
eb73c1b7 MX |
1771 | root->nr_delalloc_inodes--; |
1772 | if (!root->nr_delalloc_inodes) { | |
7c8a0d36 | 1773 | ASSERT(list_empty(&root->delalloc_inodes)); |
0b246afa | 1774 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1775 | BUG_ON(list_empty(&root->delalloc_root)); |
1776 | list_del_init(&root->delalloc_root); | |
0b246afa | 1777 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1778 | } |
1779 | } | |
2b877331 NB |
1780 | } |
1781 | ||
1782 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1783 | struct btrfs_inode *inode) | |
1784 | { | |
1785 | spin_lock(&root->delalloc_lock); | |
1786 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1787 | spin_unlock(&root->delalloc_lock); |
1788 | } | |
1789 | ||
d352ac68 | 1790 | /* |
e06a1fc9 NB |
1791 | * Properly track delayed allocation bytes in the inode and to maintain the |
1792 | * list of inodes that have pending delalloc work to be done. | |
d352ac68 | 1793 | */ |
e06a1fc9 NB |
1794 | void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state, |
1795 | unsigned *bits) | |
291d673e | 1796 | { |
0b246afa JM |
1797 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1798 | ||
47059d93 WS |
1799 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1800 | WARN_ON(1); | |
75eff68e CM |
1801 | /* |
1802 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1803 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1804 | * bit, which is only set or cleared with irqs on |
1805 | */ | |
0ca1f7ce | 1806 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1807 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1808 | u64 len = state->end + 1 - state->start; |
8b62f87b | 1809 | u32 num_extents = count_max_extents(len); |
70ddc553 | 1810 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1811 | |
8b62f87b JB |
1812 | spin_lock(&BTRFS_I(inode)->lock); |
1813 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
1814 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 1815 | |
6a3891c5 | 1816 | /* For sanity tests */ |
0b246afa | 1817 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1818 | return; |
1819 | ||
104b4e51 NB |
1820 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1821 | fs_info->delalloc_batch); | |
df0af1a5 | 1822 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1823 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1824 | if (*bits & EXTENT_DEFRAG) |
1825 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1826 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1827 | &BTRFS_I(inode)->runtime_flags)) |
1828 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1829 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1830 | } |
a7e3b975 FM |
1831 | |
1832 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1833 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1834 | spin_lock(&BTRFS_I(inode)->lock); | |
1835 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1836 | state->start; | |
1837 | spin_unlock(&BTRFS_I(inode)->lock); | |
1838 | } | |
291d673e CM |
1839 | } |
1840 | ||
d352ac68 | 1841 | /* |
a36bb5f9 NB |
1842 | * Once a range is no longer delalloc this function ensures that proper |
1843 | * accounting happens. | |
d352ac68 | 1844 | */ |
a36bb5f9 NB |
1845 | void btrfs_clear_delalloc_extent(struct inode *vfs_inode, |
1846 | struct extent_state *state, unsigned *bits) | |
291d673e | 1847 | { |
a36bb5f9 NB |
1848 | struct btrfs_inode *inode = BTRFS_I(vfs_inode); |
1849 | struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb); | |
47059d93 | 1850 | u64 len = state->end + 1 - state->start; |
823bb20a | 1851 | u32 num_extents = count_max_extents(len); |
47059d93 | 1852 | |
4a4b964f FM |
1853 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1854 | spin_lock(&inode->lock); | |
6fc0ef68 | 1855 | inode->defrag_bytes -= len; |
4a4b964f FM |
1856 | spin_unlock(&inode->lock); |
1857 | } | |
47059d93 | 1858 | |
75eff68e CM |
1859 | /* |
1860 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1861 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1862 | * bit, which is only set or cleared with irqs on |
1863 | */ | |
0ca1f7ce | 1864 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1865 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1866 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1867 | |
8b62f87b JB |
1868 | spin_lock(&inode->lock); |
1869 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
1870 | spin_unlock(&inode->lock); | |
0ca1f7ce | 1871 | |
b6d08f06 JB |
1872 | /* |
1873 | * We don't reserve metadata space for space cache inodes so we | |
52042d8e | 1874 | * don't need to call delalloc_release_metadata if there is an |
b6d08f06 JB |
1875 | * error. |
1876 | */ | |
a315e68f | 1877 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1878 | root != fs_info->tree_root) |
43b18595 | 1879 | btrfs_delalloc_release_metadata(inode, len, false); |
0ca1f7ce | 1880 | |
6a3891c5 | 1881 | /* For sanity tests. */ |
0b246afa | 1882 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1883 | return; |
1884 | ||
a315e68f FM |
1885 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1886 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1887 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1888 | btrfs_free_reserved_data_space_noquota( |
1889 | &inode->vfs_inode, | |
51773bec | 1890 | state->start, len); |
9ed74f2d | 1891 | |
104b4e51 NB |
1892 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1893 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1894 | spin_lock(&inode->lock); |
1895 | inode->delalloc_bytes -= len; | |
1896 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1897 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1898 | &inode->runtime_flags)) |
eb73c1b7 | 1899 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1900 | spin_unlock(&inode->lock); |
291d673e | 1901 | } |
a7e3b975 FM |
1902 | |
1903 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1904 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1905 | spin_lock(&inode->lock); | |
1906 | ASSERT(inode->new_delalloc_bytes >= len); | |
1907 | inode->new_delalloc_bytes -= len; | |
1908 | spin_unlock(&inode->lock); | |
1909 | } | |
291d673e CM |
1910 | } |
1911 | ||
d352ac68 | 1912 | /* |
da12fe54 NB |
1913 | * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit |
1914 | * in a chunk's stripe. This function ensures that bios do not span a | |
1915 | * stripe/chunk | |
6f034ece | 1916 | * |
da12fe54 NB |
1917 | * @page - The page we are about to add to the bio |
1918 | * @size - size we want to add to the bio | |
1919 | * @bio - bio we want to ensure is smaller than a stripe | |
1920 | * @bio_flags - flags of the bio | |
1921 | * | |
1922 | * return 1 if page cannot be added to the bio | |
1923 | * return 0 if page can be added to the bio | |
6f034ece | 1924 | * return error otherwise |
d352ac68 | 1925 | */ |
da12fe54 NB |
1926 | int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio, |
1927 | unsigned long bio_flags) | |
239b14b3 | 1928 | { |
0b246afa JM |
1929 | struct inode *inode = page->mapping->host; |
1930 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1931 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1932 | u64 length = 0; |
1933 | u64 map_length; | |
239b14b3 CM |
1934 | int ret; |
1935 | ||
771ed689 CM |
1936 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1937 | return 0; | |
1938 | ||
4f024f37 | 1939 | length = bio->bi_iter.bi_size; |
239b14b3 | 1940 | map_length = length; |
0b246afa JM |
1941 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1942 | NULL, 0); | |
6f034ece LB |
1943 | if (ret < 0) |
1944 | return ret; | |
d397712b | 1945 | if (map_length < length + size) |
239b14b3 | 1946 | return 1; |
3444a972 | 1947 | return 0; |
239b14b3 CM |
1948 | } |
1949 | ||
d352ac68 CM |
1950 | /* |
1951 | * in order to insert checksums into the metadata in large chunks, | |
1952 | * we wait until bio submission time. All the pages in the bio are | |
1953 | * checksummed and sums are attached onto the ordered extent record. | |
1954 | * | |
1955 | * At IO completion time the cums attached on the ordered extent record | |
1956 | * are inserted into the btree | |
1957 | */ | |
d0ee3934 | 1958 | static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio, |
eaf25d93 | 1959 | u64 bio_offset) |
065631f6 | 1960 | { |
c6100a4b | 1961 | struct inode *inode = private_data; |
4e4cbee9 | 1962 | blk_status_t ret = 0; |
e015640f | 1963 | |
2ff7e61e | 1964 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1965 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1966 | return 0; |
1967 | } | |
e015640f | 1968 | |
d352ac68 | 1969 | /* |
cad321ad | 1970 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
1971 | * on write, or reading the csums from the tree before a read. |
1972 | * | |
1973 | * Rules about async/sync submit, | |
1974 | * a) read: sync submit | |
1975 | * | |
1976 | * b) write without checksum: sync submit | |
1977 | * | |
1978 | * c) write with checksum: | |
1979 | * c-1) if bio is issued by fsync: sync submit | |
1980 | * (sync_writers != 0) | |
1981 | * | |
1982 | * c-2) if root is reloc root: sync submit | |
1983 | * (only in case of buffered IO) | |
1984 | * | |
1985 | * c-3) otherwise: async submit | |
d352ac68 | 1986 | */ |
a56b1c7b | 1987 | static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio, |
50489a57 NB |
1988 | int mirror_num, |
1989 | unsigned long bio_flags) | |
1990 | ||
44b8bd7e | 1991 | { |
0b246afa | 1992 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1993 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1994 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 1995 | blk_status_t ret = 0; |
19b9bdb0 | 1996 | int skip_sum; |
b812ce28 | 1997 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1998 | |
6cbff00f | 1999 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 2000 | |
70ddc553 | 2001 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2002 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2003 | |
37226b21 | 2004 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2005 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2006 | if (ret) |
61891923 | 2007 | goto out; |
5fd02043 | 2008 | |
d20f7043 | 2009 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2010 | ret = btrfs_submit_compressed_read(inode, bio, |
2011 | mirror_num, | |
2012 | bio_flags); | |
2013 | goto out; | |
c2db1073 | 2014 | } else if (!skip_sum) { |
2ff7e61e | 2015 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2016 | if (ret) |
61891923 | 2017 | goto out; |
c2db1073 | 2018 | } |
4d1b5fb4 | 2019 | goto mapit; |
b812ce28 | 2020 | } else if (async && !skip_sum) { |
17d217fe YZ |
2021 | /* csum items have already been cloned */ |
2022 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2023 | goto mapit; | |
19b9bdb0 | 2024 | /* we're doing a write, do the async checksumming */ |
c6100a4b | 2025 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
e7681167 | 2026 | 0, inode, btrfs_submit_bio_start); |
61891923 | 2027 | goto out; |
b812ce28 | 2028 | } else if (!skip_sum) { |
2ff7e61e | 2029 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2030 | if (ret) |
2031 | goto out; | |
19b9bdb0 CM |
2032 | } |
2033 | ||
0b86a832 | 2034 | mapit: |
2ff7e61e | 2035 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2036 | |
2037 | out: | |
4e4cbee9 CH |
2038 | if (ret) { |
2039 | bio->bi_status = ret; | |
4246a0b6 CH |
2040 | bio_endio(bio); |
2041 | } | |
61891923 | 2042 | return ret; |
065631f6 | 2043 | } |
6885f308 | 2044 | |
d352ac68 CM |
2045 | /* |
2046 | * given a list of ordered sums record them in the inode. This happens | |
2047 | * at IO completion time based on sums calculated at bio submission time. | |
2048 | */ | |
ba1da2f4 | 2049 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2050 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2051 | { |
e6dcd2dc | 2052 | struct btrfs_ordered_sum *sum; |
ac01f26a | 2053 | int ret; |
e6dcd2dc | 2054 | |
c6e30871 | 2055 | list_for_each_entry(sum, list, list) { |
7c2871a2 | 2056 | trans->adding_csums = true; |
ac01f26a | 2057 | ret = btrfs_csum_file_blocks(trans, |
d20f7043 | 2058 | BTRFS_I(inode)->root->fs_info->csum_root, sum); |
7c2871a2 | 2059 | trans->adding_csums = false; |
ac01f26a NB |
2060 | if (ret) |
2061 | return ret; | |
e6dcd2dc CM |
2062 | } |
2063 | return 0; | |
2064 | } | |
2065 | ||
2ac55d41 | 2066 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2067 | unsigned int extra_bits, |
ba8b04c1 | 2068 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2069 | { |
fdb1e121 | 2070 | WARN_ON(PAGE_ALIGNED(end)); |
ea8c2819 | 2071 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2072 | extra_bits, cached_state); |
ea8c2819 CM |
2073 | } |
2074 | ||
d352ac68 | 2075 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2076 | struct btrfs_writepage_fixup { |
2077 | struct page *page; | |
2078 | struct btrfs_work work; | |
2079 | }; | |
2080 | ||
b2950863 | 2081 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2082 | { |
2083 | struct btrfs_writepage_fixup *fixup; | |
2084 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2085 | struct extent_state *cached_state = NULL; |
364ecf36 | 2086 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2087 | struct page *page; |
2088 | struct inode *inode; | |
2089 | u64 page_start; | |
2090 | u64 page_end; | |
87826df0 | 2091 | int ret; |
247e743c CM |
2092 | |
2093 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2094 | page = fixup->page; | |
4a096752 | 2095 | again: |
247e743c CM |
2096 | lock_page(page); |
2097 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2098 | ClearPageChecked(page); | |
2099 | goto out_page; | |
2100 | } | |
2101 | ||
2102 | inode = page->mapping->host; | |
2103 | page_start = page_offset(page); | |
09cbfeaf | 2104 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2105 | |
ff13db41 | 2106 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2107 | &cached_state); |
4a096752 CM |
2108 | |
2109 | /* already ordered? We're done */ | |
8b62b72b | 2110 | if (PagePrivate2(page)) |
247e743c | 2111 | goto out; |
4a096752 | 2112 | |
a776c6fa | 2113 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2114 | PAGE_SIZE); |
4a096752 | 2115 | if (ordered) { |
2ac55d41 | 2116 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
e43bbe5e | 2117 | page_end, &cached_state); |
4a096752 CM |
2118 | unlock_page(page); |
2119 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2120 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2121 | goto again; |
2122 | } | |
247e743c | 2123 | |
364ecf36 | 2124 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2125 | PAGE_SIZE); |
87826df0 JM |
2126 | if (ret) { |
2127 | mapping_set_error(page->mapping, ret); | |
2128 | end_extent_writepage(page, ret, page_start, page_end); | |
2129 | ClearPageChecked(page); | |
2130 | goto out; | |
2131 | } | |
2132 | ||
f3038ee3 NB |
2133 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, |
2134 | &cached_state, 0); | |
2135 | if (ret) { | |
2136 | mapping_set_error(page->mapping, ret); | |
2137 | end_extent_writepage(page, ret, page_start, page_end); | |
2138 | ClearPageChecked(page); | |
2139 | goto out; | |
2140 | } | |
2141 | ||
247e743c | 2142 | ClearPageChecked(page); |
87826df0 | 2143 | set_page_dirty(page); |
43b18595 | 2144 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false); |
247e743c | 2145 | out: |
2ac55d41 | 2146 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
e43bbe5e | 2147 | &cached_state); |
247e743c CM |
2148 | out_page: |
2149 | unlock_page(page); | |
09cbfeaf | 2150 | put_page(page); |
b897abec | 2151 | kfree(fixup); |
364ecf36 | 2152 | extent_changeset_free(data_reserved); |
247e743c CM |
2153 | } |
2154 | ||
2155 | /* | |
2156 | * There are a few paths in the higher layers of the kernel that directly | |
2157 | * set the page dirty bit without asking the filesystem if it is a | |
2158 | * good idea. This causes problems because we want to make sure COW | |
2159 | * properly happens and the data=ordered rules are followed. | |
2160 | * | |
c8b97818 | 2161 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2162 | * hasn't been properly setup for IO. We kick off an async process |
2163 | * to fix it up. The async helper will wait for ordered extents, set | |
2164 | * the delalloc bit and make it safe to write the page. | |
2165 | */ | |
d75855b4 | 2166 | int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end) |
247e743c CM |
2167 | { |
2168 | struct inode *inode = page->mapping->host; | |
0b246afa | 2169 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2170 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2171 | |
8b62b72b CM |
2172 | /* this page is properly in the ordered list */ |
2173 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2174 | return 0; |
2175 | ||
2176 | if (PageChecked(page)) | |
2177 | return -EAGAIN; | |
2178 | ||
2179 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2180 | if (!fixup) | |
2181 | return -EAGAIN; | |
f421950f | 2182 | |
247e743c | 2183 | SetPageChecked(page); |
09cbfeaf | 2184 | get_page(page); |
9e0af237 LB |
2185 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2186 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2187 | fixup->page = page; |
0b246afa | 2188 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2189 | return -EBUSY; |
247e743c CM |
2190 | } |
2191 | ||
d899e052 YZ |
2192 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2193 | struct inode *inode, u64 file_pos, | |
2194 | u64 disk_bytenr, u64 disk_num_bytes, | |
2195 | u64 num_bytes, u64 ram_bytes, | |
2196 | u8 compression, u8 encryption, | |
2197 | u16 other_encoding, int extent_type) | |
2198 | { | |
2199 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2200 | struct btrfs_file_extent_item *fi; | |
2201 | struct btrfs_path *path; | |
2202 | struct extent_buffer *leaf; | |
2203 | struct btrfs_key ins; | |
a12b877b | 2204 | u64 qg_released; |
1acae57b | 2205 | int extent_inserted = 0; |
d899e052 YZ |
2206 | int ret; |
2207 | ||
2208 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2209 | if (!path) |
2210 | return -ENOMEM; | |
d899e052 | 2211 | |
a1ed835e CM |
2212 | /* |
2213 | * we may be replacing one extent in the tree with another. | |
2214 | * The new extent is pinned in the extent map, and we don't want | |
2215 | * to drop it from the cache until it is completely in the btree. | |
2216 | * | |
2217 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2218 | * the caller is expected to unpin it and allow it to be merged | |
2219 | * with the others. | |
2220 | */ | |
1acae57b FDBM |
2221 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2222 | file_pos + num_bytes, NULL, 0, | |
2223 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2224 | if (ret) |
2225 | goto out; | |
d899e052 | 2226 | |
1acae57b | 2227 | if (!extent_inserted) { |
4a0cc7ca | 2228 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2229 | ins.offset = file_pos; |
2230 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2231 | ||
2232 | path->leave_spinning = 1; | |
2233 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2234 | sizeof(*fi)); | |
2235 | if (ret) | |
2236 | goto out; | |
2237 | } | |
d899e052 YZ |
2238 | leaf = path->nodes[0]; |
2239 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2240 | struct btrfs_file_extent_item); | |
2241 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2242 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2243 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2244 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2245 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2246 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2247 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2248 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2249 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2250 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2251 | |
d899e052 | 2252 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2253 | btrfs_release_path(path); |
d899e052 YZ |
2254 | |
2255 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2256 | |
2257 | ins.objectid = disk_bytenr; | |
2258 | ins.offset = disk_num_bytes; | |
2259 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2260 | |
297d750b | 2261 | /* |
5846a3c2 QW |
2262 | * Release the reserved range from inode dirty range map, as it is |
2263 | * already moved into delayed_ref_head | |
297d750b | 2264 | */ |
a12b877b QW |
2265 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2266 | if (ret < 0) | |
2267 | goto out; | |
2268 | qg_released = ret; | |
84f7d8e6 JB |
2269 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2270 | btrfs_ino(BTRFS_I(inode)), | |
2271 | file_pos, qg_released, &ins); | |
79787eaa | 2272 | out: |
d899e052 | 2273 | btrfs_free_path(path); |
b9473439 | 2274 | |
79787eaa | 2275 | return ret; |
d899e052 YZ |
2276 | } |
2277 | ||
38c227d8 LB |
2278 | /* snapshot-aware defrag */ |
2279 | struct sa_defrag_extent_backref { | |
2280 | struct rb_node node; | |
2281 | struct old_sa_defrag_extent *old; | |
2282 | u64 root_id; | |
2283 | u64 inum; | |
2284 | u64 file_pos; | |
2285 | u64 extent_offset; | |
2286 | u64 num_bytes; | |
2287 | u64 generation; | |
2288 | }; | |
2289 | ||
2290 | struct old_sa_defrag_extent { | |
2291 | struct list_head list; | |
2292 | struct new_sa_defrag_extent *new; | |
2293 | ||
2294 | u64 extent_offset; | |
2295 | u64 bytenr; | |
2296 | u64 offset; | |
2297 | u64 len; | |
2298 | int count; | |
2299 | }; | |
2300 | ||
2301 | struct new_sa_defrag_extent { | |
2302 | struct rb_root root; | |
2303 | struct list_head head; | |
2304 | struct btrfs_path *path; | |
2305 | struct inode *inode; | |
2306 | u64 file_pos; | |
2307 | u64 len; | |
2308 | u64 bytenr; | |
2309 | u64 disk_len; | |
2310 | u8 compress_type; | |
2311 | }; | |
2312 | ||
2313 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2314 | struct sa_defrag_extent_backref *b2) | |
2315 | { | |
2316 | if (b1->root_id < b2->root_id) | |
2317 | return -1; | |
2318 | else if (b1->root_id > b2->root_id) | |
2319 | return 1; | |
2320 | ||
2321 | if (b1->inum < b2->inum) | |
2322 | return -1; | |
2323 | else if (b1->inum > b2->inum) | |
2324 | return 1; | |
2325 | ||
2326 | if (b1->file_pos < b2->file_pos) | |
2327 | return -1; | |
2328 | else if (b1->file_pos > b2->file_pos) | |
2329 | return 1; | |
2330 | ||
2331 | /* | |
2332 | * [------------------------------] ===> (a range of space) | |
2333 | * |<--->| |<---->| =============> (fs/file tree A) | |
2334 | * |<---------------------------->| ===> (fs/file tree B) | |
2335 | * | |
2336 | * A range of space can refer to two file extents in one tree while | |
2337 | * refer to only one file extent in another tree. | |
2338 | * | |
2339 | * So we may process a disk offset more than one time(two extents in A) | |
2340 | * and locate at the same extent(one extent in B), then insert two same | |
2341 | * backrefs(both refer to the extent in B). | |
2342 | */ | |
2343 | return 0; | |
2344 | } | |
2345 | ||
2346 | static void backref_insert(struct rb_root *root, | |
2347 | struct sa_defrag_extent_backref *backref) | |
2348 | { | |
2349 | struct rb_node **p = &root->rb_node; | |
2350 | struct rb_node *parent = NULL; | |
2351 | struct sa_defrag_extent_backref *entry; | |
2352 | int ret; | |
2353 | ||
2354 | while (*p) { | |
2355 | parent = *p; | |
2356 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2357 | ||
2358 | ret = backref_comp(backref, entry); | |
2359 | if (ret < 0) | |
2360 | p = &(*p)->rb_left; | |
2361 | else | |
2362 | p = &(*p)->rb_right; | |
2363 | } | |
2364 | ||
2365 | rb_link_node(&backref->node, parent, p); | |
2366 | rb_insert_color(&backref->node, root); | |
2367 | } | |
2368 | ||
2369 | /* | |
2370 | * Note the backref might has changed, and in this case we just return 0. | |
2371 | */ | |
2372 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2373 | void *ctx) | |
2374 | { | |
2375 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2376 | struct old_sa_defrag_extent *old = ctx; |
2377 | struct new_sa_defrag_extent *new = old->new; | |
2378 | struct btrfs_path *path = new->path; | |
2379 | struct btrfs_key key; | |
2380 | struct btrfs_root *root; | |
2381 | struct sa_defrag_extent_backref *backref; | |
2382 | struct extent_buffer *leaf; | |
2383 | struct inode *inode = new->inode; | |
0b246afa | 2384 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2385 | int slot; |
2386 | int ret; | |
2387 | u64 extent_offset; | |
2388 | u64 num_bytes; | |
2389 | ||
2390 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2391 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2392 | return 0; |
2393 | ||
2394 | key.objectid = root_id; | |
2395 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2396 | key.offset = (u64)-1; | |
2397 | ||
38c227d8 LB |
2398 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2399 | if (IS_ERR(root)) { | |
2400 | if (PTR_ERR(root) == -ENOENT) | |
2401 | return 0; | |
2402 | WARN_ON(1); | |
ab8d0fc4 | 2403 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2404 | inum, offset, root_id); |
2405 | return PTR_ERR(root); | |
2406 | } | |
2407 | ||
2408 | key.objectid = inum; | |
2409 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2410 | if (offset > (u64)-1 << 32) | |
2411 | key.offset = 0; | |
2412 | else | |
2413 | key.offset = offset; | |
2414 | ||
2415 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2416 | if (WARN_ON(ret < 0)) |
38c227d8 | 2417 | return ret; |
50f1319c | 2418 | ret = 0; |
38c227d8 LB |
2419 | |
2420 | while (1) { | |
2421 | cond_resched(); | |
2422 | ||
2423 | leaf = path->nodes[0]; | |
2424 | slot = path->slots[0]; | |
2425 | ||
2426 | if (slot >= btrfs_header_nritems(leaf)) { | |
2427 | ret = btrfs_next_leaf(root, path); | |
2428 | if (ret < 0) { | |
2429 | goto out; | |
2430 | } else if (ret > 0) { | |
2431 | ret = 0; | |
2432 | goto out; | |
2433 | } | |
2434 | continue; | |
2435 | } | |
2436 | ||
2437 | path->slots[0]++; | |
2438 | ||
2439 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2440 | ||
2441 | if (key.objectid > inum) | |
2442 | goto out; | |
2443 | ||
2444 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2445 | continue; | |
2446 | ||
2447 | extent = btrfs_item_ptr(leaf, slot, | |
2448 | struct btrfs_file_extent_item); | |
2449 | ||
2450 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2451 | continue; | |
2452 | ||
e68afa49 LB |
2453 | /* |
2454 | * 'offset' refers to the exact key.offset, | |
2455 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2456 | * (key.offset - extent_offset). | |
2457 | */ | |
2458 | if (key.offset != offset) | |
38c227d8 LB |
2459 | continue; |
2460 | ||
e68afa49 | 2461 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2462 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2463 | |
38c227d8 LB |
2464 | if (extent_offset >= old->extent_offset + old->offset + |
2465 | old->len || extent_offset + num_bytes <= | |
2466 | old->extent_offset + old->offset) | |
2467 | continue; | |
38c227d8 LB |
2468 | break; |
2469 | } | |
2470 | ||
2471 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2472 | if (!backref) { | |
2473 | ret = -ENOENT; | |
2474 | goto out; | |
2475 | } | |
2476 | ||
2477 | backref->root_id = root_id; | |
2478 | backref->inum = inum; | |
e68afa49 | 2479 | backref->file_pos = offset; |
38c227d8 LB |
2480 | backref->num_bytes = num_bytes; |
2481 | backref->extent_offset = extent_offset; | |
2482 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2483 | backref->old = old; | |
2484 | backref_insert(&new->root, backref); | |
2485 | old->count++; | |
2486 | out: | |
2487 | btrfs_release_path(path); | |
2488 | WARN_ON(ret); | |
2489 | return ret; | |
2490 | } | |
2491 | ||
2492 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2493 | struct new_sa_defrag_extent *new) | |
2494 | { | |
0b246afa | 2495 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2496 | struct old_sa_defrag_extent *old, *tmp; |
2497 | int ret; | |
2498 | ||
2499 | new->path = path; | |
2500 | ||
2501 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2502 | ret = iterate_inodes_from_logical(old->bytenr + |
2503 | old->extent_offset, fs_info, | |
38c227d8 | 2504 | path, record_one_backref, |
c995ab3c | 2505 | old, false); |
4724b106 JB |
2506 | if (ret < 0 && ret != -ENOENT) |
2507 | return false; | |
38c227d8 LB |
2508 | |
2509 | /* no backref to be processed for this extent */ | |
2510 | if (!old->count) { | |
2511 | list_del(&old->list); | |
2512 | kfree(old); | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | if (list_empty(&new->head)) | |
2517 | return false; | |
2518 | ||
2519 | return true; | |
2520 | } | |
2521 | ||
2522 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2523 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2524 | struct new_sa_defrag_extent *new) |
38c227d8 | 2525 | { |
116e0024 | 2526 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2527 | return 0; |
2528 | ||
2529 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2530 | return 0; | |
2531 | ||
116e0024 LB |
2532 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2533 | return 0; | |
2534 | ||
2535 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2536 | btrfs_file_extent_other_encoding(leaf, fi)) |
2537 | return 0; | |
2538 | ||
2539 | return 1; | |
2540 | } | |
2541 | ||
2542 | /* | |
2543 | * Note the backref might has changed, and in this case we just return 0. | |
2544 | */ | |
2545 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2546 | struct sa_defrag_extent_backref *prev, | |
2547 | struct sa_defrag_extent_backref *backref) | |
2548 | { | |
2549 | struct btrfs_file_extent_item *extent; | |
2550 | struct btrfs_file_extent_item *item; | |
2551 | struct btrfs_ordered_extent *ordered; | |
2552 | struct btrfs_trans_handle *trans; | |
82fa113f | 2553 | struct btrfs_ref ref = { 0 }; |
38c227d8 LB |
2554 | struct btrfs_root *root; |
2555 | struct btrfs_key key; | |
2556 | struct extent_buffer *leaf; | |
2557 | struct old_sa_defrag_extent *old = backref->old; | |
2558 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2559 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2560 | struct inode *inode; |
2561 | struct extent_state *cached = NULL; | |
2562 | int ret = 0; | |
2563 | u64 start; | |
2564 | u64 len; | |
2565 | u64 lock_start; | |
2566 | u64 lock_end; | |
2567 | bool merge = false; | |
2568 | int index; | |
2569 | ||
2570 | if (prev && prev->root_id == backref->root_id && | |
2571 | prev->inum == backref->inum && | |
2572 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2573 | merge = true; | |
2574 | ||
2575 | /* step 1: get root */ | |
2576 | key.objectid = backref->root_id; | |
2577 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2578 | key.offset = (u64)-1; | |
2579 | ||
38c227d8 LB |
2580 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2581 | ||
2582 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2583 | if (IS_ERR(root)) { | |
2584 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2585 | if (PTR_ERR(root) == -ENOENT) | |
2586 | return 0; | |
2587 | return PTR_ERR(root); | |
2588 | } | |
38c227d8 | 2589 | |
bcbba5e6 WS |
2590 | if (btrfs_root_readonly(root)) { |
2591 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2592 | return 0; | |
2593 | } | |
2594 | ||
38c227d8 LB |
2595 | /* step 2: get inode */ |
2596 | key.objectid = backref->inum; | |
2597 | key.type = BTRFS_INODE_ITEM_KEY; | |
2598 | key.offset = 0; | |
2599 | ||
2600 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2601 | if (IS_ERR(inode)) { | |
2602 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2603 | return 0; | |
2604 | } | |
2605 | ||
2606 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2607 | ||
2608 | /* step 3: relink backref */ | |
2609 | lock_start = backref->file_pos; | |
2610 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2611 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2612 | &cached); |
38c227d8 LB |
2613 | |
2614 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2615 | if (ordered) { | |
2616 | btrfs_put_ordered_extent(ordered); | |
2617 | goto out_unlock; | |
2618 | } | |
2619 | ||
2620 | trans = btrfs_join_transaction(root); | |
2621 | if (IS_ERR(trans)) { | |
2622 | ret = PTR_ERR(trans); | |
2623 | goto out_unlock; | |
2624 | } | |
2625 | ||
2626 | key.objectid = backref->inum; | |
2627 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2628 | key.offset = backref->file_pos; | |
2629 | ||
2630 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2631 | if (ret < 0) { | |
2632 | goto out_free_path; | |
2633 | } else if (ret > 0) { | |
2634 | ret = 0; | |
2635 | goto out_free_path; | |
2636 | } | |
2637 | ||
2638 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2639 | struct btrfs_file_extent_item); | |
2640 | ||
2641 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2642 | backref->generation) | |
2643 | goto out_free_path; | |
2644 | ||
2645 | btrfs_release_path(path); | |
2646 | ||
2647 | start = backref->file_pos; | |
2648 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2649 | start += old->extent_offset + old->offset - | |
2650 | backref->extent_offset; | |
2651 | ||
2652 | len = min(backref->extent_offset + backref->num_bytes, | |
2653 | old->extent_offset + old->offset + old->len); | |
2654 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2655 | ||
2656 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2657 | start + len, 1); | |
2658 | if (ret) | |
2659 | goto out_free_path; | |
2660 | again: | |
4a0cc7ca | 2661 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2662 | key.type = BTRFS_EXTENT_DATA_KEY; |
2663 | key.offset = start; | |
2664 | ||
a09a0a70 | 2665 | path->leave_spinning = 1; |
38c227d8 LB |
2666 | if (merge) { |
2667 | struct btrfs_file_extent_item *fi; | |
2668 | u64 extent_len; | |
2669 | struct btrfs_key found_key; | |
2670 | ||
3c9665df | 2671 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2672 | if (ret < 0) |
2673 | goto out_free_path; | |
2674 | ||
2675 | path->slots[0]--; | |
2676 | leaf = path->nodes[0]; | |
2677 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2678 | ||
2679 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2680 | struct btrfs_file_extent_item); | |
2681 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2682 | ||
116e0024 LB |
2683 | if (extent_len + found_key.offset == start && |
2684 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2685 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2686 | extent_len + len); | |
2687 | btrfs_mark_buffer_dirty(leaf); | |
2688 | inode_add_bytes(inode, len); | |
2689 | ||
2690 | ret = 1; | |
2691 | goto out_free_path; | |
2692 | } else { | |
2693 | merge = false; | |
2694 | btrfs_release_path(path); | |
2695 | goto again; | |
2696 | } | |
2697 | } | |
2698 | ||
2699 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2700 | sizeof(*extent)); | |
2701 | if (ret) { | |
66642832 | 2702 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2703 | goto out_free_path; |
2704 | } | |
2705 | ||
2706 | leaf = path->nodes[0]; | |
2707 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2708 | struct btrfs_file_extent_item); | |
2709 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2710 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2711 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2712 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2713 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2714 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2715 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2716 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2717 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2718 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2719 | ||
2720 | btrfs_mark_buffer_dirty(leaf); | |
2721 | inode_add_bytes(inode, len); | |
a09a0a70 | 2722 | btrfs_release_path(path); |
38c227d8 | 2723 | |
82fa113f QW |
2724 | btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new->bytenr, |
2725 | new->disk_len, 0); | |
2726 | btrfs_init_data_ref(&ref, backref->root_id, backref->inum, | |
2727 | new->file_pos); /* start - extent_offset */ | |
2728 | ret = btrfs_inc_extent_ref(trans, &ref); | |
38c227d8 | 2729 | if (ret) { |
66642832 | 2730 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2731 | goto out_free_path; |
2732 | } | |
2733 | ||
2734 | ret = 1; | |
2735 | out_free_path: | |
2736 | btrfs_release_path(path); | |
a09a0a70 | 2737 | path->leave_spinning = 0; |
3a45bb20 | 2738 | btrfs_end_transaction(trans); |
38c227d8 LB |
2739 | out_unlock: |
2740 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
e43bbe5e | 2741 | &cached); |
38c227d8 LB |
2742 | iput(inode); |
2743 | return ret; | |
2744 | } | |
2745 | ||
6f519564 LB |
2746 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2747 | { | |
2748 | struct old_sa_defrag_extent *old, *tmp; | |
2749 | ||
2750 | if (!new) | |
2751 | return; | |
2752 | ||
2753 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2754 | kfree(old); |
2755 | } | |
2756 | kfree(new); | |
2757 | } | |
2758 | ||
38c227d8 LB |
2759 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2760 | { | |
0b246afa | 2761 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2762 | struct btrfs_path *path; |
38c227d8 LB |
2763 | struct sa_defrag_extent_backref *backref; |
2764 | struct sa_defrag_extent_backref *prev = NULL; | |
38c227d8 LB |
2765 | struct rb_node *node; |
2766 | int ret; | |
2767 | ||
38c227d8 LB |
2768 | path = btrfs_alloc_path(); |
2769 | if (!path) | |
2770 | return; | |
2771 | ||
2772 | if (!record_extent_backrefs(path, new)) { | |
2773 | btrfs_free_path(path); | |
2774 | goto out; | |
2775 | } | |
2776 | btrfs_release_path(path); | |
2777 | ||
2778 | while (1) { | |
2779 | node = rb_first(&new->root); | |
2780 | if (!node) | |
2781 | break; | |
2782 | rb_erase(node, &new->root); | |
2783 | ||
2784 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2785 | ||
2786 | ret = relink_extent_backref(path, prev, backref); | |
2787 | WARN_ON(ret < 0); | |
2788 | ||
2789 | kfree(prev); | |
2790 | ||
2791 | if (ret == 1) | |
2792 | prev = backref; | |
2793 | else | |
2794 | prev = NULL; | |
2795 | cond_resched(); | |
2796 | } | |
2797 | kfree(prev); | |
2798 | ||
2799 | btrfs_free_path(path); | |
38c227d8 | 2800 | out: |
6f519564 LB |
2801 | free_sa_defrag_extent(new); |
2802 | ||
0b246afa JM |
2803 | atomic_dec(&fs_info->defrag_running); |
2804 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2805 | } |
2806 | ||
2807 | static struct new_sa_defrag_extent * | |
2808 | record_old_file_extents(struct inode *inode, | |
2809 | struct btrfs_ordered_extent *ordered) | |
2810 | { | |
0b246afa | 2811 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2812 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2813 | struct btrfs_path *path; | |
2814 | struct btrfs_key key; | |
6f519564 | 2815 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2816 | struct new_sa_defrag_extent *new; |
2817 | int ret; | |
2818 | ||
2819 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2820 | if (!new) | |
2821 | return NULL; | |
2822 | ||
2823 | new->inode = inode; | |
2824 | new->file_pos = ordered->file_offset; | |
2825 | new->len = ordered->len; | |
2826 | new->bytenr = ordered->start; | |
2827 | new->disk_len = ordered->disk_len; | |
2828 | new->compress_type = ordered->compress_type; | |
2829 | new->root = RB_ROOT; | |
2830 | INIT_LIST_HEAD(&new->head); | |
2831 | ||
2832 | path = btrfs_alloc_path(); | |
2833 | if (!path) | |
2834 | goto out_kfree; | |
2835 | ||
4a0cc7ca | 2836 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2837 | key.type = BTRFS_EXTENT_DATA_KEY; |
2838 | key.offset = new->file_pos; | |
2839 | ||
2840 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2841 | if (ret < 0) | |
2842 | goto out_free_path; | |
2843 | if (ret > 0 && path->slots[0] > 0) | |
2844 | path->slots[0]--; | |
2845 | ||
2846 | /* find out all the old extents for the file range */ | |
2847 | while (1) { | |
2848 | struct btrfs_file_extent_item *extent; | |
2849 | struct extent_buffer *l; | |
2850 | int slot; | |
2851 | u64 num_bytes; | |
2852 | u64 offset; | |
2853 | u64 end; | |
2854 | u64 disk_bytenr; | |
2855 | u64 extent_offset; | |
2856 | ||
2857 | l = path->nodes[0]; | |
2858 | slot = path->slots[0]; | |
2859 | ||
2860 | if (slot >= btrfs_header_nritems(l)) { | |
2861 | ret = btrfs_next_leaf(root, path); | |
2862 | if (ret < 0) | |
6f519564 | 2863 | goto out_free_path; |
38c227d8 LB |
2864 | else if (ret > 0) |
2865 | break; | |
2866 | continue; | |
2867 | } | |
2868 | ||
2869 | btrfs_item_key_to_cpu(l, &key, slot); | |
2870 | ||
4a0cc7ca | 2871 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2872 | break; |
2873 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2874 | break; | |
2875 | if (key.offset >= new->file_pos + new->len) | |
2876 | break; | |
2877 | ||
2878 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2879 | ||
2880 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2881 | if (key.offset + num_bytes < new->file_pos) | |
2882 | goto next; | |
2883 | ||
2884 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2885 | if (!disk_bytenr) | |
2886 | goto next; | |
2887 | ||
2888 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2889 | ||
2890 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2891 | if (!old) | |
6f519564 | 2892 | goto out_free_path; |
38c227d8 LB |
2893 | |
2894 | offset = max(new->file_pos, key.offset); | |
2895 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2896 | ||
2897 | old->bytenr = disk_bytenr; | |
2898 | old->extent_offset = extent_offset; | |
2899 | old->offset = offset - key.offset; | |
2900 | old->len = end - offset; | |
2901 | old->new = new; | |
2902 | old->count = 0; | |
2903 | list_add_tail(&old->list, &new->head); | |
2904 | next: | |
2905 | path->slots[0]++; | |
2906 | cond_resched(); | |
2907 | } | |
2908 | ||
2909 | btrfs_free_path(path); | |
0b246afa | 2910 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2911 | |
2912 | return new; | |
2913 | ||
38c227d8 LB |
2914 | out_free_path: |
2915 | btrfs_free_path(path); | |
2916 | out_kfree: | |
6f519564 | 2917 | free_sa_defrag_extent(new); |
38c227d8 LB |
2918 | return NULL; |
2919 | } | |
2920 | ||
2ff7e61e | 2921 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2922 | u64 start, u64 len) |
2923 | { | |
2924 | struct btrfs_block_group_cache *cache; | |
2925 | ||
0b246afa | 2926 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2927 | ASSERT(cache); |
2928 | ||
2929 | spin_lock(&cache->lock); | |
2930 | cache->delalloc_bytes -= len; | |
2931 | spin_unlock(&cache->lock); | |
2932 | ||
2933 | btrfs_put_block_group(cache); | |
2934 | } | |
2935 | ||
d352ac68 CM |
2936 | /* as ordered data IO finishes, this gets called so we can finish |
2937 | * an ordered extent if the range of bytes in the file it covers are | |
2938 | * fully written. | |
2939 | */ | |
5fd02043 | 2940 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2941 | { |
5fd02043 | 2942 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2943 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2944 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2945 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2946 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2947 | struct extent_state *cached_state = NULL; |
38c227d8 | 2948 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2949 | int compress_type = 0; |
77cef2ec JB |
2950 | int ret = 0; |
2951 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2952 | bool nolock; |
77cef2ec | 2953 | bool truncated = false; |
a7e3b975 FM |
2954 | bool range_locked = false; |
2955 | bool clear_new_delalloc_bytes = false; | |
49940bdd | 2956 | bool clear_reserved_extent = true; |
a7e3b975 FM |
2957 | |
2958 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2959 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2960 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2961 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2962 | |
70ddc553 | 2963 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2964 | |
5fd02043 JB |
2965 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2966 | ret = -EIO; | |
2967 | goto out; | |
2968 | } | |
2969 | ||
7ab7956e NB |
2970 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2971 | ordered_extent->file_offset, | |
2972 | ordered_extent->file_offset + | |
2973 | ordered_extent->len - 1); | |
f612496b | 2974 | |
77cef2ec JB |
2975 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2976 | truncated = true; | |
2977 | logical_len = ordered_extent->truncated_len; | |
2978 | /* Truncated the entire extent, don't bother adding */ | |
2979 | if (!logical_len) | |
2980 | goto out; | |
2981 | } | |
2982 | ||
c2167754 | 2983 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2984 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2985 | |
2986 | /* | |
2987 | * For mwrite(mmap + memset to write) case, we still reserve | |
2988 | * space for NOCOW range. | |
2989 | * As NOCOW won't cause a new delayed ref, just free the space | |
2990 | */ | |
bc42bda2 | 2991 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 2992 | ordered_extent->len); |
6c760c07 JB |
2993 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2994 | if (nolock) | |
2995 | trans = btrfs_join_transaction_nolock(root); | |
2996 | else | |
2997 | trans = btrfs_join_transaction(root); | |
2998 | if (IS_ERR(trans)) { | |
2999 | ret = PTR_ERR(trans); | |
3000 | trans = NULL; | |
3001 | goto out; | |
c2167754 | 3002 | } |
69fe2d75 | 3003 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
3004 | ret = btrfs_update_inode_fallback(trans, root, inode); |
3005 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 3006 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3007 | goto out; |
3008 | } | |
e6dcd2dc | 3009 | |
a7e3b975 | 3010 | range_locked = true; |
2ac55d41 JB |
3011 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
3012 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 3013 | &cached_state); |
e6dcd2dc | 3014 | |
38c227d8 LB |
3015 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
3016 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 3017 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
3018 | if (ret) { |
3019 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 3020 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
3021 | /* the inode is shared */ |
3022 | new = record_old_file_extents(inode, ordered_extent); | |
3023 | ||
3024 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
3025 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ae0f1625 | 3026 | EXTENT_DEFRAG, 0, 0, &cached_state); |
38c227d8 LB |
3027 | } |
3028 | ||
0cb59c99 | 3029 | if (nolock) |
7a7eaa40 | 3030 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3031 | else |
7a7eaa40 | 3032 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3033 | if (IS_ERR(trans)) { |
3034 | ret = PTR_ERR(trans); | |
3035 | trans = NULL; | |
a7e3b975 | 3036 | goto out; |
79787eaa | 3037 | } |
a79b7d4b | 3038 | |
69fe2d75 | 3039 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3040 | |
c8b97818 | 3041 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3042 | compress_type = ordered_extent->compress_type; |
d899e052 | 3043 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3044 | BUG_ON(compress_type); |
b430b775 JM |
3045 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3046 | ordered_extent->len); | |
7a6d7067 | 3047 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3048 | ordered_extent->file_offset, |
3049 | ordered_extent->file_offset + | |
77cef2ec | 3050 | logical_len); |
d899e052 | 3051 | } else { |
0b246afa | 3052 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3053 | ret = insert_reserved_file_extent(trans, inode, |
3054 | ordered_extent->file_offset, | |
3055 | ordered_extent->start, | |
3056 | ordered_extent->disk_len, | |
77cef2ec | 3057 | logical_len, logical_len, |
261507a0 | 3058 | compress_type, 0, 0, |
d899e052 | 3059 | BTRFS_FILE_EXTENT_REG); |
49940bdd JB |
3060 | if (!ret) { |
3061 | clear_reserved_extent = false; | |
2ff7e61e | 3062 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3063 | ordered_extent->start, |
3064 | ordered_extent->disk_len); | |
49940bdd | 3065 | } |
d899e052 | 3066 | } |
5dc562c5 JB |
3067 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3068 | ordered_extent->file_offset, ordered_extent->len, | |
3069 | trans->transid); | |
79787eaa | 3070 | if (ret < 0) { |
66642832 | 3071 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3072 | goto out; |
79787eaa | 3073 | } |
2ac55d41 | 3074 | |
ac01f26a NB |
3075 | ret = add_pending_csums(trans, inode, &ordered_extent->list); |
3076 | if (ret) { | |
3077 | btrfs_abort_transaction(trans, ret); | |
3078 | goto out; | |
3079 | } | |
e6dcd2dc | 3080 | |
6c760c07 JB |
3081 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3082 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3083 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3084 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3085 | goto out; |
1ef30be1 JB |
3086 | } |
3087 | ret = 0; | |
c2167754 | 3088 | out: |
a7e3b975 FM |
3089 | if (range_locked || clear_new_delalloc_bytes) { |
3090 | unsigned int clear_bits = 0; | |
3091 | ||
3092 | if (range_locked) | |
3093 | clear_bits |= EXTENT_LOCKED; | |
3094 | if (clear_new_delalloc_bytes) | |
3095 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3096 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3097 | ordered_extent->file_offset, | |
3098 | ordered_extent->file_offset + | |
3099 | ordered_extent->len - 1, | |
3100 | clear_bits, | |
3101 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
ae0f1625 | 3102 | 0, &cached_state); |
a7e3b975 FM |
3103 | } |
3104 | ||
a698d075 | 3105 | if (trans) |
3a45bb20 | 3106 | btrfs_end_transaction(trans); |
0cb59c99 | 3107 | |
77cef2ec JB |
3108 | if (ret || truncated) { |
3109 | u64 start, end; | |
3110 | ||
3111 | if (truncated) | |
3112 | start = ordered_extent->file_offset + logical_len; | |
3113 | else | |
3114 | start = ordered_extent->file_offset; | |
3115 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
f08dc36f | 3116 | clear_extent_uptodate(io_tree, start, end, NULL); |
77cef2ec JB |
3117 | |
3118 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3119 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3120 | |
0bec9ef5 JB |
3121 | /* |
3122 | * If the ordered extent had an IOERR or something else went | |
3123 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3124 | * back to the allocator. We only free the extent in the |
3125 | * truncated case if we didn't write out the extent at all. | |
49940bdd JB |
3126 | * |
3127 | * If we made it past insert_reserved_file_extent before we | |
3128 | * errored out then we don't need to do this as the accounting | |
3129 | * has already been done. | |
0bec9ef5 | 3130 | */ |
77cef2ec | 3131 | if ((ret || !logical_len) && |
49940bdd | 3132 | clear_reserved_extent && |
77cef2ec | 3133 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && |
0bec9ef5 | 3134 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3135 | btrfs_free_reserved_extent(fs_info, |
3136 | ordered_extent->start, | |
e570fd27 | 3137 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3138 | } |
3139 | ||
3140 | ||
5fd02043 | 3141 | /* |
8bad3c02 LB |
3142 | * This needs to be done to make sure anybody waiting knows we are done |
3143 | * updating everything for this ordered extent. | |
5fd02043 JB |
3144 | */ |
3145 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3146 | ||
38c227d8 | 3147 | /* for snapshot-aware defrag */ |
6f519564 LB |
3148 | if (new) { |
3149 | if (ret) { | |
3150 | free_sa_defrag_extent(new); | |
0b246afa | 3151 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3152 | } else { |
3153 | relink_file_extents(new); | |
3154 | } | |
3155 | } | |
38c227d8 | 3156 | |
e6dcd2dc CM |
3157 | /* once for us */ |
3158 | btrfs_put_ordered_extent(ordered_extent); | |
3159 | /* once for the tree */ | |
3160 | btrfs_put_ordered_extent(ordered_extent); | |
3161 | ||
5fd02043 JB |
3162 | return ret; |
3163 | } | |
3164 | ||
3165 | static void finish_ordered_fn(struct btrfs_work *work) | |
3166 | { | |
3167 | struct btrfs_ordered_extent *ordered_extent; | |
3168 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3169 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3170 | } |
3171 | ||
c629732d NB |
3172 | void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, |
3173 | u64 end, int uptodate) | |
211f90e6 | 3174 | { |
5fd02043 | 3175 | struct inode *inode = page->mapping->host; |
0b246afa | 3176 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3177 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3178 | struct btrfs_workqueue *wq; |
3179 | btrfs_work_func_t func; | |
5fd02043 | 3180 | |
1abe9b8a | 3181 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3182 | ||
8b62b72b | 3183 | ClearPagePrivate2(page); |
5fd02043 JB |
3184 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3185 | end - start + 1, uptodate)) | |
c3988d63 | 3186 | return; |
5fd02043 | 3187 | |
70ddc553 | 3188 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3189 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3190 | func = btrfs_freespace_write_helper; |
3191 | } else { | |
0b246afa | 3192 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3193 | func = btrfs_endio_write_helper; |
3194 | } | |
5fd02043 | 3195 | |
9e0af237 LB |
3196 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3197 | NULL); | |
3198 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3199 | } |
3200 | ||
dc380aea MX |
3201 | static int __readpage_endio_check(struct inode *inode, |
3202 | struct btrfs_io_bio *io_bio, | |
3203 | int icsum, struct page *page, | |
3204 | int pgoff, u64 start, size_t len) | |
3205 | { | |
3206 | char *kaddr; | |
3207 | u32 csum_expected; | |
3208 | u32 csum = ~(u32)0; | |
dc380aea MX |
3209 | |
3210 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3211 | ||
3212 | kaddr = kmap_atomic(page); | |
3213 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3214 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3215 | if (csum != csum_expected) |
3216 | goto zeroit; | |
3217 | ||
3218 | kunmap_atomic(kaddr); | |
3219 | return 0; | |
3220 | zeroit: | |
0970a22e | 3221 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3222 | io_bio->mirror_num); |
dc380aea MX |
3223 | memset(kaddr + pgoff, 1, len); |
3224 | flush_dcache_page(page); | |
3225 | kunmap_atomic(kaddr); | |
dc380aea MX |
3226 | return -EIO; |
3227 | } | |
3228 | ||
d352ac68 CM |
3229 | /* |
3230 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3231 | * if there's a match, we allow the bio to finish. If not, the code in |
3232 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3233 | */ |
facc8a22 MX |
3234 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3235 | u64 phy_offset, struct page *page, | |
3236 | u64 start, u64 end, int mirror) | |
07157aac | 3237 | { |
4eee4fa4 | 3238 | size_t offset = start - page_offset(page); |
07157aac | 3239 | struct inode *inode = page->mapping->host; |
d1310b2e | 3240 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3241 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3242 | |
d20f7043 CM |
3243 | if (PageChecked(page)) { |
3244 | ClearPageChecked(page); | |
dc380aea | 3245 | return 0; |
d20f7043 | 3246 | } |
6cbff00f CH |
3247 | |
3248 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3249 | return 0; |
17d217fe YZ |
3250 | |
3251 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3252 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3253 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3254 | return 0; |
17d217fe | 3255 | } |
d20f7043 | 3256 | |
facc8a22 | 3257 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3258 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3259 | start, (size_t)(end - start + 1)); | |
07157aac | 3260 | } |
b888db2b | 3261 | |
c1c3fac2 NB |
3262 | /* |
3263 | * btrfs_add_delayed_iput - perform a delayed iput on @inode | |
3264 | * | |
3265 | * @inode: The inode we want to perform iput on | |
3266 | * | |
3267 | * This function uses the generic vfs_inode::i_count to track whether we should | |
3268 | * just decrement it (in case it's > 1) or if this is the last iput then link | |
3269 | * the inode to the delayed iput machinery. Delayed iputs are processed at | |
3270 | * transaction commit time/superblock commit/cleaner kthread. | |
3271 | */ | |
24bbcf04 YZ |
3272 | void btrfs_add_delayed_iput(struct inode *inode) |
3273 | { | |
0b246afa | 3274 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3275 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3276 | |
3277 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3278 | return; | |
3279 | ||
034f784d | 3280 | atomic_inc(&fs_info->nr_delayed_iputs); |
24bbcf04 | 3281 | spin_lock(&fs_info->delayed_iput_lock); |
c1c3fac2 NB |
3282 | ASSERT(list_empty(&binode->delayed_iput)); |
3283 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
24bbcf04 | 3284 | spin_unlock(&fs_info->delayed_iput_lock); |
fd340d0f JB |
3285 | if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags)) |
3286 | wake_up_process(fs_info->cleaner_kthread); | |
24bbcf04 YZ |
3287 | } |
3288 | ||
2ff7e61e | 3289 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3290 | { |
24bbcf04 | 3291 | |
24bbcf04 | 3292 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3293 | while (!list_empty(&fs_info->delayed_iputs)) { |
3294 | struct btrfs_inode *inode; | |
3295 | ||
3296 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3297 | struct btrfs_inode, delayed_iput); | |
c1c3fac2 | 3298 | list_del_init(&inode->delayed_iput); |
8089fe62 DS |
3299 | spin_unlock(&fs_info->delayed_iput_lock); |
3300 | iput(&inode->vfs_inode); | |
034f784d JB |
3301 | if (atomic_dec_and_test(&fs_info->nr_delayed_iputs)) |
3302 | wake_up(&fs_info->delayed_iputs_wait); | |
8089fe62 | 3303 | spin_lock(&fs_info->delayed_iput_lock); |
24bbcf04 | 3304 | } |
8089fe62 | 3305 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3306 | } |
3307 | ||
034f784d JB |
3308 | /** |
3309 | * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running | |
3310 | * @fs_info - the fs_info for this fs | |
3311 | * @return - EINTR if we were killed, 0 if nothing's pending | |
3312 | * | |
3313 | * This will wait on any delayed iputs that are currently running with KILLABLE | |
3314 | * set. Once they are all done running we will return, unless we are killed in | |
3315 | * which case we return EINTR. This helps in user operations like fallocate etc | |
3316 | * that might get blocked on the iputs. | |
3317 | */ | |
3318 | int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info) | |
3319 | { | |
3320 | int ret = wait_event_killable(fs_info->delayed_iputs_wait, | |
3321 | atomic_read(&fs_info->nr_delayed_iputs) == 0); | |
3322 | if (ret) | |
3323 | return -EINTR; | |
3324 | return 0; | |
3325 | } | |
3326 | ||
7b128766 | 3327 | /* |
f7e9e8fc OS |
3328 | * This creates an orphan entry for the given inode in case something goes wrong |
3329 | * in the middle of an unlink. | |
7b128766 | 3330 | */ |
73f2e545 | 3331 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
27919067 | 3332 | struct btrfs_inode *inode) |
7b128766 | 3333 | { |
d68fc57b | 3334 | int ret; |
7b128766 | 3335 | |
27919067 OS |
3336 | ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode)); |
3337 | if (ret && ret != -EEXIST) { | |
3338 | btrfs_abort_transaction(trans, ret); | |
3339 | return ret; | |
d68fc57b YZ |
3340 | } |
3341 | ||
d68fc57b | 3342 | return 0; |
7b128766 JB |
3343 | } |
3344 | ||
3345 | /* | |
f7e9e8fc OS |
3346 | * We have done the delete so we can go ahead and remove the orphan item for |
3347 | * this particular inode. | |
7b128766 | 3348 | */ |
48a3b636 | 3349 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3350 | struct btrfs_inode *inode) |
7b128766 | 3351 | { |
27919067 | 3352 | return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode)); |
7b128766 JB |
3353 | } |
3354 | ||
3355 | /* | |
3356 | * this cleans up any orphans that may be left on the list from the last use | |
3357 | * of this root. | |
3358 | */ | |
66b4ffd1 | 3359 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3360 | { |
0b246afa | 3361 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3362 | struct btrfs_path *path; |
3363 | struct extent_buffer *leaf; | |
7b128766 JB |
3364 | struct btrfs_key key, found_key; |
3365 | struct btrfs_trans_handle *trans; | |
3366 | struct inode *inode; | |
8f6d7f4f | 3367 | u64 last_objectid = 0; |
f7e9e8fc | 3368 | int ret = 0, nr_unlink = 0; |
7b128766 | 3369 | |
d68fc57b | 3370 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3371 | return 0; |
c71bf099 YZ |
3372 | |
3373 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3374 | if (!path) { |
3375 | ret = -ENOMEM; | |
3376 | goto out; | |
3377 | } | |
e4058b54 | 3378 | path->reada = READA_BACK; |
7b128766 JB |
3379 | |
3380 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3381 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3382 | key.offset = (u64)-1; |
3383 | ||
7b128766 JB |
3384 | while (1) { |
3385 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3386 | if (ret < 0) |
3387 | goto out; | |
7b128766 JB |
3388 | |
3389 | /* | |
3390 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3391 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3392 | * find the key and see if we have stuff that matches |
3393 | */ | |
3394 | if (ret > 0) { | |
66b4ffd1 | 3395 | ret = 0; |
7b128766 JB |
3396 | if (path->slots[0] == 0) |
3397 | break; | |
3398 | path->slots[0]--; | |
3399 | } | |
3400 | ||
3401 | /* pull out the item */ | |
3402 | leaf = path->nodes[0]; | |
7b128766 JB |
3403 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3404 | ||
3405 | /* make sure the item matches what we want */ | |
3406 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3407 | break; | |
962a298f | 3408 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3409 | break; |
3410 | ||
3411 | /* release the path since we're done with it */ | |
b3b4aa74 | 3412 | btrfs_release_path(path); |
7b128766 JB |
3413 | |
3414 | /* | |
3415 | * this is where we are basically btrfs_lookup, without the | |
3416 | * crossing root thing. we store the inode number in the | |
3417 | * offset of the orphan item. | |
3418 | */ | |
8f6d7f4f JB |
3419 | |
3420 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3421 | btrfs_err(fs_info, |
3422 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3423 | ret = -EINVAL; |
3424 | goto out; | |
3425 | } | |
3426 | ||
3427 | last_objectid = found_key.offset; | |
3428 | ||
5d4f98a2 YZ |
3429 | found_key.objectid = found_key.offset; |
3430 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3431 | found_key.offset = 0; | |
0b246afa | 3432 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3433 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3434 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3435 | goto out; |
7b128766 | 3436 | |
0b246afa | 3437 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3438 | struct btrfs_root *dead_root; |
3439 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3440 | int is_dead_root = 0; | |
3441 | ||
3442 | /* | |
3443 | * this is an orphan in the tree root. Currently these | |
3444 | * could come from 2 sources: | |
3445 | * a) a snapshot deletion in progress | |
3446 | * b) a free space cache inode | |
3447 | * We need to distinguish those two, as the snapshot | |
3448 | * orphan must not get deleted. | |
3449 | * find_dead_roots already ran before us, so if this | |
3450 | * is a snapshot deletion, we should find the root | |
3451 | * in the dead_roots list | |
3452 | */ | |
3453 | spin_lock(&fs_info->trans_lock); | |
3454 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3455 | root_list) { | |
3456 | if (dead_root->root_key.objectid == | |
3457 | found_key.objectid) { | |
3458 | is_dead_root = 1; | |
3459 | break; | |
3460 | } | |
3461 | } | |
3462 | spin_unlock(&fs_info->trans_lock); | |
3463 | if (is_dead_root) { | |
3464 | /* prevent this orphan from being found again */ | |
3465 | key.offset = found_key.objectid - 1; | |
3466 | continue; | |
3467 | } | |
f7e9e8fc | 3468 | |
f8e9e0b0 | 3469 | } |
f7e9e8fc | 3470 | |
7b128766 | 3471 | /* |
f7e9e8fc OS |
3472 | * If we have an inode with links, there are a couple of |
3473 | * possibilities. Old kernels (before v3.12) used to create an | |
3474 | * orphan item for truncate indicating that there were possibly | |
3475 | * extent items past i_size that needed to be deleted. In v3.12, | |
3476 | * truncate was changed to update i_size in sync with the extent | |
3477 | * items, but the (useless) orphan item was still created. Since | |
3478 | * v4.18, we don't create the orphan item for truncate at all. | |
3479 | * | |
3480 | * So, this item could mean that we need to do a truncate, but | |
3481 | * only if this filesystem was last used on a pre-v3.12 kernel | |
3482 | * and was not cleanly unmounted. The odds of that are quite | |
3483 | * slim, and it's a pain to do the truncate now, so just delete | |
3484 | * the orphan item. | |
3485 | * | |
3486 | * It's also possible that this orphan item was supposed to be | |
3487 | * deleted but wasn't. The inode number may have been reused, | |
3488 | * but either way, we can delete the orphan item. | |
7b128766 | 3489 | */ |
f7e9e8fc OS |
3490 | if (ret == -ENOENT || inode->i_nlink) { |
3491 | if (!ret) | |
3492 | iput(inode); | |
a8c9e576 | 3493 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3494 | if (IS_ERR(trans)) { |
3495 | ret = PTR_ERR(trans); | |
3496 | goto out; | |
3497 | } | |
0b246afa JM |
3498 | btrfs_debug(fs_info, "auto deleting %Lu", |
3499 | found_key.objectid); | |
a8c9e576 JB |
3500 | ret = btrfs_del_orphan_item(trans, root, |
3501 | found_key.objectid); | |
3a45bb20 | 3502 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3503 | if (ret) |
3504 | goto out; | |
7b128766 JB |
3505 | continue; |
3506 | } | |
3507 | ||
f7e9e8fc | 3508 | nr_unlink++; |
7b128766 JB |
3509 | |
3510 | /* this will do delete_inode and everything for us */ | |
3511 | iput(inode); | |
3512 | } | |
3254c876 MX |
3513 | /* release the path since we're done with it */ |
3514 | btrfs_release_path(path); | |
3515 | ||
d68fc57b YZ |
3516 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3517 | ||
a575ceeb | 3518 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { |
7a7eaa40 | 3519 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3520 | if (!IS_ERR(trans)) |
3a45bb20 | 3521 | btrfs_end_transaction(trans); |
d68fc57b | 3522 | } |
7b128766 JB |
3523 | |
3524 | if (nr_unlink) | |
0b246afa | 3525 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
66b4ffd1 JB |
3526 | |
3527 | out: | |
3528 | if (ret) | |
0b246afa | 3529 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3530 | btrfs_free_path(path); |
3531 | return ret; | |
7b128766 JB |
3532 | } |
3533 | ||
46a53cca CM |
3534 | /* |
3535 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3536 | * don't find any xattrs, we know there can't be any acls. | |
3537 | * | |
3538 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3539 | */ | |
3540 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3541 | int slot, u64 objectid, |
3542 | int *first_xattr_slot) | |
46a53cca CM |
3543 | { |
3544 | u32 nritems = btrfs_header_nritems(leaf); | |
3545 | struct btrfs_key found_key; | |
f23b5a59 JB |
3546 | static u64 xattr_access = 0; |
3547 | static u64 xattr_default = 0; | |
46a53cca CM |
3548 | int scanned = 0; |
3549 | ||
f23b5a59 | 3550 | if (!xattr_access) { |
97d79299 AG |
3551 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3552 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3553 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3554 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3555 | } |
3556 | ||
46a53cca | 3557 | slot++; |
63541927 | 3558 | *first_xattr_slot = -1; |
46a53cca CM |
3559 | while (slot < nritems) { |
3560 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3561 | ||
3562 | /* we found a different objectid, there must not be acls */ | |
3563 | if (found_key.objectid != objectid) | |
3564 | return 0; | |
3565 | ||
3566 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3567 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3568 | if (*first_xattr_slot == -1) |
3569 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3570 | if (found_key.offset == xattr_access || |
3571 | found_key.offset == xattr_default) | |
3572 | return 1; | |
3573 | } | |
46a53cca CM |
3574 | |
3575 | /* | |
3576 | * we found a key greater than an xattr key, there can't | |
3577 | * be any acls later on | |
3578 | */ | |
3579 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3580 | return 0; | |
3581 | ||
3582 | slot++; | |
3583 | scanned++; | |
3584 | ||
3585 | /* | |
3586 | * it goes inode, inode backrefs, xattrs, extents, | |
3587 | * so if there are a ton of hard links to an inode there can | |
3588 | * be a lot of backrefs. Don't waste time searching too hard, | |
3589 | * this is just an optimization | |
3590 | */ | |
3591 | if (scanned >= 8) | |
3592 | break; | |
3593 | } | |
3594 | /* we hit the end of the leaf before we found an xattr or | |
3595 | * something larger than an xattr. We have to assume the inode | |
3596 | * has acls | |
3597 | */ | |
63541927 FDBM |
3598 | if (*first_xattr_slot == -1) |
3599 | *first_xattr_slot = slot; | |
46a53cca CM |
3600 | return 1; |
3601 | } | |
3602 | ||
d352ac68 CM |
3603 | /* |
3604 | * read an inode from the btree into the in-memory inode | |
3605 | */ | |
4222ea71 FM |
3606 | static int btrfs_read_locked_inode(struct inode *inode, |
3607 | struct btrfs_path *in_path) | |
39279cc3 | 3608 | { |
0b246afa | 3609 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4222ea71 | 3610 | struct btrfs_path *path = in_path; |
5f39d397 | 3611 | struct extent_buffer *leaf; |
39279cc3 CM |
3612 | struct btrfs_inode_item *inode_item; |
3613 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3614 | struct btrfs_key location; | |
67de1176 | 3615 | unsigned long ptr; |
46a53cca | 3616 | int maybe_acls; |
618e21d5 | 3617 | u32 rdev; |
39279cc3 | 3618 | int ret; |
2f7e33d4 | 3619 | bool filled = false; |
63541927 | 3620 | int first_xattr_slot; |
2f7e33d4 MX |
3621 | |
3622 | ret = btrfs_fill_inode(inode, &rdev); | |
3623 | if (!ret) | |
3624 | filled = true; | |
39279cc3 | 3625 | |
4222ea71 FM |
3626 | if (!path) { |
3627 | path = btrfs_alloc_path(); | |
3628 | if (!path) | |
3629 | return -ENOMEM; | |
3630 | } | |
1748f843 | 3631 | |
39279cc3 | 3632 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3633 | |
39279cc3 | 3634 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 | 3635 | if (ret) { |
4222ea71 FM |
3636 | if (path != in_path) |
3637 | btrfs_free_path(path); | |
f5b3a417 | 3638 | return ret; |
67710892 | 3639 | } |
39279cc3 | 3640 | |
5f39d397 | 3641 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3642 | |
3643 | if (filled) | |
67de1176 | 3644 | goto cache_index; |
2f7e33d4 | 3645 | |
5f39d397 CM |
3646 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3647 | struct btrfs_inode_item); | |
5f39d397 | 3648 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3649 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3650 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3651 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3652 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3653 | |
a937b979 DS |
3654 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3655 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3656 | |
a937b979 DS |
3657 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3658 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3659 | |
a937b979 DS |
3660 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3661 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3662 | |
9cc97d64 | 3663 | BTRFS_I(inode)->i_otime.tv_sec = |
3664 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3665 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3666 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3667 | |
a76a3cd4 | 3668 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3669 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3670 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3671 | ||
c7f88c4e JL |
3672 | inode_set_iversion_queried(inode, |
3673 | btrfs_inode_sequence(leaf, inode_item)); | |
6e17d30b YD |
3674 | inode->i_generation = BTRFS_I(inode)->generation; |
3675 | inode->i_rdev = 0; | |
3676 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3677 | ||
3678 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3679 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3680 | ||
3681 | cache_index: | |
5dc562c5 JB |
3682 | /* |
3683 | * If we were modified in the current generation and evicted from memory | |
3684 | * and then re-read we need to do a full sync since we don't have any | |
3685 | * idea about which extents were modified before we were evicted from | |
3686 | * cache. | |
6e17d30b YD |
3687 | * |
3688 | * This is required for both inode re-read from disk and delayed inode | |
3689 | * in delayed_nodes_tree. | |
5dc562c5 | 3690 | */ |
0b246afa | 3691 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3692 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3693 | &BTRFS_I(inode)->runtime_flags); | |
3694 | ||
bde6c242 FM |
3695 | /* |
3696 | * We don't persist the id of the transaction where an unlink operation | |
3697 | * against the inode was last made. So here we assume the inode might | |
3698 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3699 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3700 | * between the inode and its parent if the inode is fsync'ed and the log | |
3701 | * replayed. For example, in the scenario: | |
3702 | * | |
3703 | * touch mydir/foo | |
3704 | * ln mydir/foo mydir/bar | |
3705 | * sync | |
3706 | * unlink mydir/bar | |
3707 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3708 | * xfs_io -c fsync mydir/foo | |
3709 | * <power failure> | |
3710 | * mount fs, triggers fsync log replay | |
3711 | * | |
3712 | * We must make sure that when we fsync our inode foo we also log its | |
3713 | * parent inode, otherwise after log replay the parent still has the | |
3714 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3715 | * and doesn't have an inode ref with the name "bar" anymore. | |
3716 | * | |
3717 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3718 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3719 | * transaction commits on fsync if our inode is a directory, or if our |
3720 | * inode is not a directory, logging its parent unnecessarily. | |
3721 | */ | |
3722 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3723 | ||
67de1176 MX |
3724 | path->slots[0]++; |
3725 | if (inode->i_nlink != 1 || | |
3726 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3727 | goto cache_acl; | |
3728 | ||
3729 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3730 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3731 | goto cache_acl; |
3732 | ||
3733 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3734 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3735 | struct btrfs_inode_ref *ref; | |
3736 | ||
3737 | ref = (struct btrfs_inode_ref *)ptr; | |
3738 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3739 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3740 | struct btrfs_inode_extref *extref; | |
3741 | ||
3742 | extref = (struct btrfs_inode_extref *)ptr; | |
3743 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3744 | extref); | |
3745 | } | |
2f7e33d4 | 3746 | cache_acl: |
46a53cca CM |
3747 | /* |
3748 | * try to precache a NULL acl entry for files that don't have | |
3749 | * any xattrs or acls | |
3750 | */ | |
33345d01 | 3751 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3752 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3753 | if (first_xattr_slot != -1) { |
3754 | path->slots[0] = first_xattr_slot; | |
3755 | ret = btrfs_load_inode_props(inode, path); | |
3756 | if (ret) | |
0b246afa | 3757 | btrfs_err(fs_info, |
351fd353 | 3758 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3759 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3760 | root->root_key.objectid, ret); |
3761 | } | |
4222ea71 FM |
3762 | if (path != in_path) |
3763 | btrfs_free_path(path); | |
63541927 | 3764 | |
72c04902 AV |
3765 | if (!maybe_acls) |
3766 | cache_no_acl(inode); | |
46a53cca | 3767 | |
39279cc3 | 3768 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3769 | case S_IFREG: |
3770 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3771 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3772 | inode->i_fop = &btrfs_file_operations; |
3773 | inode->i_op = &btrfs_file_inode_operations; | |
3774 | break; | |
3775 | case S_IFDIR: | |
3776 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3777 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3778 | break; |
3779 | case S_IFLNK: | |
3780 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3781 | inode_nohighmem(inode); |
4779cc04 | 3782 | inode->i_mapping->a_ops = &btrfs_aops; |
39279cc3 | 3783 | break; |
618e21d5 | 3784 | default: |
0279b4cd | 3785 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3786 | init_special_inode(inode, inode->i_mode, rdev); |
3787 | break; | |
39279cc3 | 3788 | } |
6cbff00f | 3789 | |
7b6a221e | 3790 | btrfs_sync_inode_flags_to_i_flags(inode); |
67710892 | 3791 | return 0; |
39279cc3 CM |
3792 | } |
3793 | ||
d352ac68 CM |
3794 | /* |
3795 | * given a leaf and an inode, copy the inode fields into the leaf | |
3796 | */ | |
e02119d5 CM |
3797 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3798 | struct extent_buffer *leaf, | |
5f39d397 | 3799 | struct btrfs_inode_item *item, |
39279cc3 CM |
3800 | struct inode *inode) |
3801 | { | |
51fab693 LB |
3802 | struct btrfs_map_token token; |
3803 | ||
3804 | btrfs_init_map_token(&token); | |
5f39d397 | 3805 | |
51fab693 LB |
3806 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3807 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3808 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3809 | &token); | |
3810 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3811 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3812 | |
a937b979 | 3813 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3814 | inode->i_atime.tv_sec, &token); |
a937b979 | 3815 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3816 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3817 | |
a937b979 | 3818 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3819 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3820 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3821 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3822 | |
a937b979 | 3823 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3824 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3825 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3826 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3827 | |
9cc97d64 | 3828 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3829 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3830 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3831 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3832 | ||
51fab693 LB |
3833 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3834 | &token); | |
3835 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3836 | &token); | |
c7f88c4e JL |
3837 | btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode), |
3838 | &token); | |
51fab693 LB |
3839 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); |
3840 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3841 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3842 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3843 | } |
3844 | ||
d352ac68 CM |
3845 | /* |
3846 | * copy everything in the in-memory inode into the btree. | |
3847 | */ | |
2115133f | 3848 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3849 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3850 | { |
3851 | struct btrfs_inode_item *inode_item; | |
3852 | struct btrfs_path *path; | |
5f39d397 | 3853 | struct extent_buffer *leaf; |
39279cc3 CM |
3854 | int ret; |
3855 | ||
3856 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3857 | if (!path) |
3858 | return -ENOMEM; | |
3859 | ||
b9473439 | 3860 | path->leave_spinning = 1; |
16cdcec7 MX |
3861 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3862 | 1); | |
39279cc3 CM |
3863 | if (ret) { |
3864 | if (ret > 0) | |
3865 | ret = -ENOENT; | |
3866 | goto failed; | |
3867 | } | |
3868 | ||
5f39d397 CM |
3869 | leaf = path->nodes[0]; |
3870 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3871 | struct btrfs_inode_item); |
39279cc3 | 3872 | |
e02119d5 | 3873 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3874 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3875 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3876 | ret = 0; |
3877 | failed: | |
39279cc3 CM |
3878 | btrfs_free_path(path); |
3879 | return ret; | |
3880 | } | |
3881 | ||
2115133f CM |
3882 | /* |
3883 | * copy everything in the in-memory inode into the btree. | |
3884 | */ | |
3885 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3886 | struct btrfs_root *root, struct inode *inode) | |
3887 | { | |
0b246afa | 3888 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3889 | int ret; |
3890 | ||
3891 | /* | |
3892 | * If the inode is a free space inode, we can deadlock during commit | |
3893 | * if we put it into the delayed code. | |
3894 | * | |
3895 | * The data relocation inode should also be directly updated | |
3896 | * without delay | |
3897 | */ | |
70ddc553 | 3898 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 3899 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3900 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3901 | btrfs_update_root_times(trans, root); |
3902 | ||
2115133f CM |
3903 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3904 | if (!ret) | |
3905 | btrfs_set_inode_last_trans(trans, inode); | |
3906 | return ret; | |
3907 | } | |
3908 | ||
3909 | return btrfs_update_inode_item(trans, root, inode); | |
3910 | } | |
3911 | ||
be6aef60 JB |
3912 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3913 | struct btrfs_root *root, | |
3914 | struct inode *inode) | |
2115133f CM |
3915 | { |
3916 | int ret; | |
3917 | ||
3918 | ret = btrfs_update_inode(trans, root, inode); | |
3919 | if (ret == -ENOSPC) | |
3920 | return btrfs_update_inode_item(trans, root, inode); | |
3921 | return ret; | |
3922 | } | |
3923 | ||
d352ac68 CM |
3924 | /* |
3925 | * unlink helper that gets used here in inode.c and in the tree logging | |
3926 | * recovery code. It remove a link in a directory with a given name, and | |
3927 | * also drops the back refs in the inode to the directory | |
3928 | */ | |
92986796 AV |
3929 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3930 | struct btrfs_root *root, | |
4ec5934e NB |
3931 | struct btrfs_inode *dir, |
3932 | struct btrfs_inode *inode, | |
92986796 | 3933 | const char *name, int name_len) |
39279cc3 | 3934 | { |
0b246afa | 3935 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3936 | struct btrfs_path *path; |
39279cc3 | 3937 | int ret = 0; |
39279cc3 | 3938 | struct btrfs_dir_item *di; |
aec7477b | 3939 | u64 index; |
33345d01 LZ |
3940 | u64 ino = btrfs_ino(inode); |
3941 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3942 | |
3943 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3944 | if (!path) { |
3945 | ret = -ENOMEM; | |
554233a6 | 3946 | goto out; |
54aa1f4d CM |
3947 | } |
3948 | ||
b9473439 | 3949 | path->leave_spinning = 1; |
33345d01 | 3950 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 | 3951 | name, name_len, -1); |
3cf5068f LB |
3952 | if (IS_ERR_OR_NULL(di)) { |
3953 | ret = di ? PTR_ERR(di) : -ENOENT; | |
39279cc3 CM |
3954 | goto err; |
3955 | } | |
39279cc3 | 3956 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3957 | if (ret) |
3958 | goto err; | |
b3b4aa74 | 3959 | btrfs_release_path(path); |
39279cc3 | 3960 | |
67de1176 MX |
3961 | /* |
3962 | * If we don't have dir index, we have to get it by looking up | |
3963 | * the inode ref, since we get the inode ref, remove it directly, | |
3964 | * it is unnecessary to do delayed deletion. | |
3965 | * | |
3966 | * But if we have dir index, needn't search inode ref to get it. | |
3967 | * Since the inode ref is close to the inode item, it is better | |
3968 | * that we delay to delete it, and just do this deletion when | |
3969 | * we update the inode item. | |
3970 | */ | |
4ec5934e | 3971 | if (inode->dir_index) { |
67de1176 MX |
3972 | ret = btrfs_delayed_delete_inode_ref(inode); |
3973 | if (!ret) { | |
4ec5934e | 3974 | index = inode->dir_index; |
67de1176 MX |
3975 | goto skip_backref; |
3976 | } | |
3977 | } | |
3978 | ||
33345d01 LZ |
3979 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3980 | dir_ino, &index); | |
aec7477b | 3981 | if (ret) { |
0b246afa | 3982 | btrfs_info(fs_info, |
c2cf52eb | 3983 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 3984 | name_len, name, ino, dir_ino); |
66642832 | 3985 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
3986 | goto err; |
3987 | } | |
67de1176 | 3988 | skip_backref: |
9add2945 | 3989 | ret = btrfs_delete_delayed_dir_index(trans, dir, index); |
79787eaa | 3990 | if (ret) { |
66642832 | 3991 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 3992 | goto err; |
79787eaa | 3993 | } |
39279cc3 | 3994 | |
4ec5934e NB |
3995 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
3996 | dir_ino); | |
79787eaa | 3997 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 3998 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3999 | goto err; |
4000 | } | |
e02119d5 | 4001 | |
4ec5934e NB |
4002 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4003 | index); | |
6418c961 CM |
4004 | if (ret == -ENOENT) |
4005 | ret = 0; | |
d4e3991b | 4006 | else if (ret) |
66642832 | 4007 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4008 | err: |
4009 | btrfs_free_path(path); | |
e02119d5 CM |
4010 | if (ret) |
4011 | goto out; | |
4012 | ||
6ef06d27 | 4013 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4014 | inode_inc_iversion(&inode->vfs_inode); |
4015 | inode_inc_iversion(&dir->vfs_inode); | |
4016 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4017 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4018 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4019 | out: |
39279cc3 CM |
4020 | return ret; |
4021 | } | |
4022 | ||
92986796 AV |
4023 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4024 | struct btrfs_root *root, | |
4ec5934e | 4025 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4026 | const char *name, int name_len) |
4027 | { | |
4028 | int ret; | |
4029 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4030 | if (!ret) { | |
4ec5934e NB |
4031 | drop_nlink(&inode->vfs_inode); |
4032 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4033 | } |
4034 | return ret; | |
4035 | } | |
39279cc3 | 4036 | |
a22285a6 YZ |
4037 | /* |
4038 | * helper to start transaction for unlink and rmdir. | |
4039 | * | |
d52be818 JB |
4040 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4041 | * if we cannot make our reservations the normal way try and see if there is | |
4042 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4043 | * allow the unlink to occur. | |
a22285a6 | 4044 | */ |
d52be818 | 4045 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4046 | { |
a22285a6 | 4047 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4048 | |
e70bea5f JB |
4049 | /* |
4050 | * 1 for the possible orphan item | |
4051 | * 1 for the dir item | |
4052 | * 1 for the dir index | |
4053 | * 1 for the inode ref | |
e70bea5f JB |
4054 | * 1 for the inode |
4055 | */ | |
8eab77ff | 4056 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4057 | } |
4058 | ||
4059 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4060 | { | |
4061 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4062 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4063 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4064 | int ret; |
a22285a6 | 4065 | |
d52be818 | 4066 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4067 | if (IS_ERR(trans)) |
4068 | return PTR_ERR(trans); | |
5f39d397 | 4069 | |
4ec5934e NB |
4070 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4071 | 0); | |
12fcfd22 | 4072 | |
4ec5934e NB |
4073 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4074 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4075 | dentry->d_name.len); | |
b532402e TI |
4076 | if (ret) |
4077 | goto out; | |
7b128766 | 4078 | |
a22285a6 | 4079 | if (inode->i_nlink == 0) { |
73f2e545 | 4080 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4081 | if (ret) |
4082 | goto out; | |
a22285a6 | 4083 | } |
7b128766 | 4084 | |
b532402e | 4085 | out: |
3a45bb20 | 4086 | btrfs_end_transaction(trans); |
2ff7e61e | 4087 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4088 | return ret; |
4089 | } | |
4090 | ||
f60a2364 | 4091 | static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
401b3b19 LF |
4092 | struct inode *dir, u64 objectid, |
4093 | const char *name, int name_len) | |
4df27c4d | 4094 | { |
401b3b19 | 4095 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
4096 | struct btrfs_path *path; |
4097 | struct extent_buffer *leaf; | |
4098 | struct btrfs_dir_item *di; | |
4099 | struct btrfs_key key; | |
4100 | u64 index; | |
4101 | int ret; | |
4a0cc7ca | 4102 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4103 | |
4104 | path = btrfs_alloc_path(); | |
4105 | if (!path) | |
4106 | return -ENOMEM; | |
4107 | ||
33345d01 | 4108 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4109 | name, name_len, -1); |
79787eaa | 4110 | if (IS_ERR_OR_NULL(di)) { |
3cf5068f | 4111 | ret = di ? PTR_ERR(di) : -ENOENT; |
79787eaa JM |
4112 | goto out; |
4113 | } | |
4df27c4d YZ |
4114 | |
4115 | leaf = path->nodes[0]; | |
4116 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4117 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4118 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4119 | if (ret) { |
66642832 | 4120 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4121 | goto out; |
4122 | } | |
b3b4aa74 | 4123 | btrfs_release_path(path); |
4df27c4d | 4124 | |
3ee1c553 LF |
4125 | ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid, |
4126 | dir_ino, &index, name, name_len); | |
4df27c4d | 4127 | if (ret < 0) { |
79787eaa | 4128 | if (ret != -ENOENT) { |
66642832 | 4129 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4130 | goto out; |
4131 | } | |
33345d01 | 4132 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4133 | name, name_len); |
79787eaa JM |
4134 | if (IS_ERR_OR_NULL(di)) { |
4135 | if (!di) | |
4136 | ret = -ENOENT; | |
4137 | else | |
4138 | ret = PTR_ERR(di); | |
66642832 | 4139 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4140 | goto out; |
4141 | } | |
4df27c4d YZ |
4142 | |
4143 | leaf = path->nodes[0]; | |
4144 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4df27c4d YZ |
4145 | index = key.offset; |
4146 | } | |
945d8962 | 4147 | btrfs_release_path(path); |
4df27c4d | 4148 | |
9add2945 | 4149 | ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index); |
79787eaa | 4150 | if (ret) { |
66642832 | 4151 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4152 | goto out; |
4153 | } | |
4df27c4d | 4154 | |
6ef06d27 | 4155 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4156 | inode_inc_iversion(dir); |
c2050a45 | 4157 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4158 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4159 | if (ret) |
66642832 | 4160 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4161 | out: |
71d7aed0 | 4162 | btrfs_free_path(path); |
79787eaa | 4163 | return ret; |
4df27c4d YZ |
4164 | } |
4165 | ||
ec42f167 MT |
4166 | /* |
4167 | * Helper to check if the subvolume references other subvolumes or if it's | |
4168 | * default. | |
4169 | */ | |
f60a2364 | 4170 | static noinline int may_destroy_subvol(struct btrfs_root *root) |
ec42f167 MT |
4171 | { |
4172 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4173 | struct btrfs_path *path; | |
4174 | struct btrfs_dir_item *di; | |
4175 | struct btrfs_key key; | |
4176 | u64 dir_id; | |
4177 | int ret; | |
4178 | ||
4179 | path = btrfs_alloc_path(); | |
4180 | if (!path) | |
4181 | return -ENOMEM; | |
4182 | ||
4183 | /* Make sure this root isn't set as the default subvol */ | |
4184 | dir_id = btrfs_super_root_dir(fs_info->super_copy); | |
4185 | di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path, | |
4186 | dir_id, "default", 7, 0); | |
4187 | if (di && !IS_ERR(di)) { | |
4188 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
4189 | if (key.objectid == root->root_key.objectid) { | |
4190 | ret = -EPERM; | |
4191 | btrfs_err(fs_info, | |
4192 | "deleting default subvolume %llu is not allowed", | |
4193 | key.objectid); | |
4194 | goto out; | |
4195 | } | |
4196 | btrfs_release_path(path); | |
4197 | } | |
4198 | ||
4199 | key.objectid = root->root_key.objectid; | |
4200 | key.type = BTRFS_ROOT_REF_KEY; | |
4201 | key.offset = (u64)-1; | |
4202 | ||
4203 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
4204 | if (ret < 0) | |
4205 | goto out; | |
4206 | BUG_ON(ret == 0); | |
4207 | ||
4208 | ret = 0; | |
4209 | if (path->slots[0] > 0) { | |
4210 | path->slots[0]--; | |
4211 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4212 | if (key.objectid == root->root_key.objectid && | |
4213 | key.type == BTRFS_ROOT_REF_KEY) | |
4214 | ret = -ENOTEMPTY; | |
4215 | } | |
4216 | out: | |
4217 | btrfs_free_path(path); | |
4218 | return ret; | |
4219 | } | |
4220 | ||
20a68004 NB |
4221 | /* Delete all dentries for inodes belonging to the root */ |
4222 | static void btrfs_prune_dentries(struct btrfs_root *root) | |
4223 | { | |
4224 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4225 | struct rb_node *node; | |
4226 | struct rb_node *prev; | |
4227 | struct btrfs_inode *entry; | |
4228 | struct inode *inode; | |
4229 | u64 objectid = 0; | |
4230 | ||
4231 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) | |
4232 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4233 | ||
4234 | spin_lock(&root->inode_lock); | |
4235 | again: | |
4236 | node = root->inode_tree.rb_node; | |
4237 | prev = NULL; | |
4238 | while (node) { | |
4239 | prev = node; | |
4240 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4241 | ||
37508515 | 4242 | if (objectid < btrfs_ino(entry)) |
20a68004 | 4243 | node = node->rb_left; |
37508515 | 4244 | else if (objectid > btrfs_ino(entry)) |
20a68004 NB |
4245 | node = node->rb_right; |
4246 | else | |
4247 | break; | |
4248 | } | |
4249 | if (!node) { | |
4250 | while (prev) { | |
4251 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
37508515 | 4252 | if (objectid <= btrfs_ino(entry)) { |
20a68004 NB |
4253 | node = prev; |
4254 | break; | |
4255 | } | |
4256 | prev = rb_next(prev); | |
4257 | } | |
4258 | } | |
4259 | while (node) { | |
4260 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
37508515 | 4261 | objectid = btrfs_ino(entry) + 1; |
20a68004 NB |
4262 | inode = igrab(&entry->vfs_inode); |
4263 | if (inode) { | |
4264 | spin_unlock(&root->inode_lock); | |
4265 | if (atomic_read(&inode->i_count) > 1) | |
4266 | d_prune_aliases(inode); | |
4267 | /* | |
4268 | * btrfs_drop_inode will have it removed from the inode | |
4269 | * cache when its usage count hits zero. | |
4270 | */ | |
4271 | iput(inode); | |
4272 | cond_resched(); | |
4273 | spin_lock(&root->inode_lock); | |
4274 | goto again; | |
4275 | } | |
4276 | ||
4277 | if (cond_resched_lock(&root->inode_lock)) | |
4278 | goto again; | |
4279 | ||
4280 | node = rb_next(node); | |
4281 | } | |
4282 | spin_unlock(&root->inode_lock); | |
4283 | } | |
4284 | ||
f60a2364 MT |
4285 | int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry) |
4286 | { | |
4287 | struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb); | |
4288 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4289 | struct inode *inode = d_inode(dentry); | |
4290 | struct btrfs_root *dest = BTRFS_I(inode)->root; | |
4291 | struct btrfs_trans_handle *trans; | |
4292 | struct btrfs_block_rsv block_rsv; | |
4293 | u64 root_flags; | |
f60a2364 MT |
4294 | int ret; |
4295 | int err; | |
4296 | ||
4297 | /* | |
4298 | * Don't allow to delete a subvolume with send in progress. This is | |
4299 | * inside the inode lock so the error handling that has to drop the bit | |
4300 | * again is not run concurrently. | |
4301 | */ | |
4302 | spin_lock(&dest->root_item_lock); | |
a7176f74 | 4303 | if (dest->send_in_progress) { |
f60a2364 MT |
4304 | spin_unlock(&dest->root_item_lock); |
4305 | btrfs_warn(fs_info, | |
4306 | "attempt to delete subvolume %llu during send", | |
4307 | dest->root_key.objectid); | |
4308 | return -EPERM; | |
4309 | } | |
a7176f74 LF |
4310 | root_flags = btrfs_root_flags(&dest->root_item); |
4311 | btrfs_set_root_flags(&dest->root_item, | |
4312 | root_flags | BTRFS_ROOT_SUBVOL_DEAD); | |
4313 | spin_unlock(&dest->root_item_lock); | |
f60a2364 MT |
4314 | |
4315 | down_write(&fs_info->subvol_sem); | |
4316 | ||
4317 | err = may_destroy_subvol(dest); | |
4318 | if (err) | |
4319 | goto out_up_write; | |
4320 | ||
4321 | btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); | |
4322 | /* | |
4323 | * One for dir inode, | |
4324 | * two for dir entries, | |
4325 | * two for root ref/backref. | |
4326 | */ | |
c4c129db | 4327 | err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true); |
f60a2364 MT |
4328 | if (err) |
4329 | goto out_up_write; | |
4330 | ||
4331 | trans = btrfs_start_transaction(root, 0); | |
4332 | if (IS_ERR(trans)) { | |
4333 | err = PTR_ERR(trans); | |
4334 | goto out_release; | |
4335 | } | |
4336 | trans->block_rsv = &block_rsv; | |
4337 | trans->bytes_reserved = block_rsv.size; | |
4338 | ||
4339 | btrfs_record_snapshot_destroy(trans, BTRFS_I(dir)); | |
4340 | ||
401b3b19 LF |
4341 | ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid, |
4342 | dentry->d_name.name, dentry->d_name.len); | |
f60a2364 MT |
4343 | if (ret) { |
4344 | err = ret; | |
4345 | btrfs_abort_transaction(trans, ret); | |
4346 | goto out_end_trans; | |
4347 | } | |
4348 | ||
4349 | btrfs_record_root_in_trans(trans, dest); | |
4350 | ||
4351 | memset(&dest->root_item.drop_progress, 0, | |
4352 | sizeof(dest->root_item.drop_progress)); | |
4353 | dest->root_item.drop_level = 0; | |
4354 | btrfs_set_root_refs(&dest->root_item, 0); | |
4355 | ||
4356 | if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { | |
4357 | ret = btrfs_insert_orphan_item(trans, | |
4358 | fs_info->tree_root, | |
4359 | dest->root_key.objectid); | |
4360 | if (ret) { | |
4361 | btrfs_abort_transaction(trans, ret); | |
4362 | err = ret; | |
4363 | goto out_end_trans; | |
4364 | } | |
4365 | } | |
4366 | ||
d1957791 | 4367 | ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid, |
f60a2364 MT |
4368 | BTRFS_UUID_KEY_SUBVOL, |
4369 | dest->root_key.objectid); | |
4370 | if (ret && ret != -ENOENT) { | |
4371 | btrfs_abort_transaction(trans, ret); | |
4372 | err = ret; | |
4373 | goto out_end_trans; | |
4374 | } | |
4375 | if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { | |
d1957791 | 4376 | ret = btrfs_uuid_tree_remove(trans, |
f60a2364 MT |
4377 | dest->root_item.received_uuid, |
4378 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4379 | dest->root_key.objectid); | |
4380 | if (ret && ret != -ENOENT) { | |
4381 | btrfs_abort_transaction(trans, ret); | |
4382 | err = ret; | |
4383 | goto out_end_trans; | |
4384 | } | |
4385 | } | |
4386 | ||
4387 | out_end_trans: | |
4388 | trans->block_rsv = NULL; | |
4389 | trans->bytes_reserved = 0; | |
4390 | ret = btrfs_end_transaction(trans); | |
4391 | if (ret && !err) | |
4392 | err = ret; | |
4393 | inode->i_flags |= S_DEAD; | |
4394 | out_release: | |
4395 | btrfs_subvolume_release_metadata(fs_info, &block_rsv); | |
4396 | out_up_write: | |
4397 | up_write(&fs_info->subvol_sem); | |
4398 | if (err) { | |
4399 | spin_lock(&dest->root_item_lock); | |
4400 | root_flags = btrfs_root_flags(&dest->root_item); | |
4401 | btrfs_set_root_flags(&dest->root_item, | |
4402 | root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); | |
4403 | spin_unlock(&dest->root_item_lock); | |
4404 | } else { | |
4405 | d_invalidate(dentry); | |
20a68004 | 4406 | btrfs_prune_dentries(dest); |
f60a2364 MT |
4407 | ASSERT(dest->send_in_progress == 0); |
4408 | ||
4409 | /* the last ref */ | |
4410 | if (dest->ino_cache_inode) { | |
4411 | iput(dest->ino_cache_inode); | |
4412 | dest->ino_cache_inode = NULL; | |
4413 | } | |
4414 | } | |
4415 | ||
4416 | return err; | |
4417 | } | |
4418 | ||
39279cc3 CM |
4419 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4420 | { | |
2b0143b5 | 4421 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4422 | int err = 0; |
39279cc3 | 4423 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4424 | struct btrfs_trans_handle *trans; |
44f714da | 4425 | u64 last_unlink_trans; |
39279cc3 | 4426 | |
b3ae244e | 4427 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4428 | return -ENOTEMPTY; |
4a0cc7ca | 4429 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
a79a464d | 4430 | return btrfs_delete_subvolume(dir, dentry); |
134d4512 | 4431 | |
d52be818 | 4432 | trans = __unlink_start_trans(dir); |
a22285a6 | 4433 | if (IS_ERR(trans)) |
5df6a9f6 | 4434 | return PTR_ERR(trans); |
5df6a9f6 | 4435 | |
4a0cc7ca | 4436 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
401b3b19 | 4437 | err = btrfs_unlink_subvol(trans, dir, |
4df27c4d YZ |
4438 | BTRFS_I(inode)->location.objectid, |
4439 | dentry->d_name.name, | |
4440 | dentry->d_name.len); | |
4441 | goto out; | |
4442 | } | |
4443 | ||
73f2e545 | 4444 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4445 | if (err) |
4df27c4d | 4446 | goto out; |
7b128766 | 4447 | |
44f714da FM |
4448 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4449 | ||
39279cc3 | 4450 | /* now the directory is empty */ |
4ec5934e NB |
4451 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4452 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4453 | dentry->d_name.len); | |
44f714da | 4454 | if (!err) { |
6ef06d27 | 4455 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4456 | /* |
4457 | * Propagate the last_unlink_trans value of the deleted dir to | |
4458 | * its parent directory. This is to prevent an unrecoverable | |
4459 | * log tree in the case we do something like this: | |
4460 | * 1) create dir foo | |
4461 | * 2) create snapshot under dir foo | |
4462 | * 3) delete the snapshot | |
4463 | * 4) rmdir foo | |
4464 | * 5) mkdir foo | |
4465 | * 6) fsync foo or some file inside foo | |
4466 | */ | |
4467 | if (last_unlink_trans >= trans->transid) | |
4468 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4469 | } | |
4df27c4d | 4470 | out: |
3a45bb20 | 4471 | btrfs_end_transaction(trans); |
2ff7e61e | 4472 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4473 | |
39279cc3 CM |
4474 | return err; |
4475 | } | |
4476 | ||
ddfae63c JB |
4477 | /* |
4478 | * Return this if we need to call truncate_block for the last bit of the | |
4479 | * truncate. | |
4480 | */ | |
4481 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4482 | |
39279cc3 CM |
4483 | /* |
4484 | * this can truncate away extent items, csum items and directory items. | |
4485 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4486 | * any higher than new_size |
39279cc3 CM |
4487 | * |
4488 | * csum items that cross the new i_size are truncated to the new size | |
4489 | * as well. | |
7b128766 JB |
4490 | * |
4491 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4492 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4493 | */ |
8082510e YZ |
4494 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4495 | struct btrfs_root *root, | |
4496 | struct inode *inode, | |
4497 | u64 new_size, u32 min_type) | |
39279cc3 | 4498 | { |
0b246afa | 4499 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4500 | struct btrfs_path *path; |
5f39d397 | 4501 | struct extent_buffer *leaf; |
39279cc3 | 4502 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4503 | struct btrfs_key key; |
4504 | struct btrfs_key found_key; | |
39279cc3 | 4505 | u64 extent_start = 0; |
db94535d | 4506 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4507 | u64 extent_offset = 0; |
39279cc3 | 4508 | u64 item_end = 0; |
c1aa4575 | 4509 | u64 last_size = new_size; |
8082510e | 4510 | u32 found_type = (u8)-1; |
39279cc3 CM |
4511 | int found_extent; |
4512 | int del_item; | |
85e21bac CM |
4513 | int pending_del_nr = 0; |
4514 | int pending_del_slot = 0; | |
179e29e4 | 4515 | int extent_type = -1; |
8082510e | 4516 | int ret; |
4a0cc7ca | 4517 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4518 | u64 bytes_deleted = 0; |
897ca819 TM |
4519 | bool be_nice = false; |
4520 | bool should_throttle = false; | |
8082510e YZ |
4521 | |
4522 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4523 | |
28ed1345 CM |
4524 | /* |
4525 | * for non-free space inodes and ref cows, we want to back off from | |
4526 | * time to time | |
4527 | */ | |
70ddc553 | 4528 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4529 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4530 | be_nice = true; |
28ed1345 | 4531 | |
0eb0e19c MF |
4532 | path = btrfs_alloc_path(); |
4533 | if (!path) | |
4534 | return -ENOMEM; | |
e4058b54 | 4535 | path->reada = READA_BACK; |
0eb0e19c | 4536 | |
5dc562c5 JB |
4537 | /* |
4538 | * We want to drop from the next block forward in case this new size is | |
4539 | * not block aligned since we will be keeping the last block of the | |
4540 | * extent just the way it is. | |
4541 | */ | |
27cdeb70 | 4542 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4543 | root == fs_info->tree_root) |
dcdbc059 | 4544 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4545 | fs_info->sectorsize), |
da17066c | 4546 | (u64)-1, 0); |
8082510e | 4547 | |
16cdcec7 MX |
4548 | /* |
4549 | * This function is also used to drop the items in the log tree before | |
4550 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
52042d8e | 4551 | * it is used to drop the logged items. So we shouldn't kill the delayed |
16cdcec7 MX |
4552 | * items. |
4553 | */ | |
4554 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4555 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4556 | |
33345d01 | 4557 | key.objectid = ino; |
39279cc3 | 4558 | key.offset = (u64)-1; |
5f39d397 CM |
4559 | key.type = (u8)-1; |
4560 | ||
85e21bac | 4561 | search_again: |
28ed1345 CM |
4562 | /* |
4563 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4564 | * up a huge file in a single leaf. Most of the time that | |
4565 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4566 | */ | |
fd86a3a3 OS |
4567 | if (be_nice && bytes_deleted > SZ_32M && |
4568 | btrfs_should_end_transaction(trans)) { | |
4569 | ret = -EAGAIN; | |
4570 | goto out; | |
28ed1345 CM |
4571 | } |
4572 | ||
b9473439 | 4573 | path->leave_spinning = 1; |
85e21bac | 4574 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
fd86a3a3 | 4575 | if (ret < 0) |
8082510e | 4576 | goto out; |
d397712b | 4577 | |
85e21bac | 4578 | if (ret > 0) { |
fd86a3a3 | 4579 | ret = 0; |
e02119d5 CM |
4580 | /* there are no items in the tree for us to truncate, we're |
4581 | * done | |
4582 | */ | |
8082510e YZ |
4583 | if (path->slots[0] == 0) |
4584 | goto out; | |
85e21bac CM |
4585 | path->slots[0]--; |
4586 | } | |
4587 | ||
d397712b | 4588 | while (1) { |
39279cc3 | 4589 | fi = NULL; |
5f39d397 CM |
4590 | leaf = path->nodes[0]; |
4591 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4592 | found_type = found_key.type; |
39279cc3 | 4593 | |
33345d01 | 4594 | if (found_key.objectid != ino) |
39279cc3 | 4595 | break; |
5f39d397 | 4596 | |
85e21bac | 4597 | if (found_type < min_type) |
39279cc3 CM |
4598 | break; |
4599 | ||
5f39d397 | 4600 | item_end = found_key.offset; |
39279cc3 | 4601 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4602 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4603 | struct btrfs_file_extent_item); |
179e29e4 CM |
4604 | extent_type = btrfs_file_extent_type(leaf, fi); |
4605 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4606 | item_end += |
db94535d | 4607 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4608 | |
4609 | trace_btrfs_truncate_show_fi_regular( | |
4610 | BTRFS_I(inode), leaf, fi, | |
4611 | found_key.offset); | |
179e29e4 | 4612 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 QW |
4613 | item_end += btrfs_file_extent_ram_bytes(leaf, |
4614 | fi); | |
09ed2f16 LB |
4615 | |
4616 | trace_btrfs_truncate_show_fi_inline( | |
4617 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4618 | found_key.offset); | |
39279cc3 | 4619 | } |
008630c1 | 4620 | item_end--; |
39279cc3 | 4621 | } |
8082510e YZ |
4622 | if (found_type > min_type) { |
4623 | del_item = 1; | |
4624 | } else { | |
76b42abb | 4625 | if (item_end < new_size) |
b888db2b | 4626 | break; |
8082510e YZ |
4627 | if (found_key.offset >= new_size) |
4628 | del_item = 1; | |
4629 | else | |
4630 | del_item = 0; | |
39279cc3 | 4631 | } |
39279cc3 | 4632 | found_extent = 0; |
39279cc3 | 4633 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4634 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4635 | goto delete; | |
4636 | ||
4637 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4638 | u64 num_dec; |
db94535d | 4639 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4640 | if (!del_item) { |
db94535d CM |
4641 | u64 orig_num_bytes = |
4642 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4643 | extent_num_bytes = ALIGN(new_size - |
4644 | found_key.offset, | |
0b246afa | 4645 | fs_info->sectorsize); |
db94535d CM |
4646 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4647 | extent_num_bytes); | |
4648 | num_dec = (orig_num_bytes - | |
9069218d | 4649 | extent_num_bytes); |
27cdeb70 MX |
4650 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4651 | &root->state) && | |
4652 | extent_start != 0) | |
a76a3cd4 | 4653 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4654 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4655 | } else { |
db94535d CM |
4656 | extent_num_bytes = |
4657 | btrfs_file_extent_disk_num_bytes(leaf, | |
4658 | fi); | |
5d4f98a2 YZ |
4659 | extent_offset = found_key.offset - |
4660 | btrfs_file_extent_offset(leaf, fi); | |
4661 | ||
39279cc3 | 4662 | /* FIXME blocksize != 4096 */ |
9069218d | 4663 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4664 | if (extent_start != 0) { |
4665 | found_extent = 1; | |
27cdeb70 MX |
4666 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4667 | &root->state)) | |
a76a3cd4 | 4668 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4669 | } |
39279cc3 | 4670 | } |
9069218d | 4671 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4672 | /* |
4673 | * we can't truncate inline items that have had | |
4674 | * special encodings | |
4675 | */ | |
4676 | if (!del_item && | |
c8b97818 | 4677 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4678 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4679 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4680 | u32 size = (u32)(new_size - found_key.offset); | |
4681 | ||
4682 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4683 | size = btrfs_file_extent_calc_inline_size(size); | |
78ac4f9e | 4684 | btrfs_truncate_item(path, size, 1); |
ddfae63c | 4685 | } else if (!del_item) { |
514ac8ad | 4686 | /* |
ddfae63c JB |
4687 | * We have to bail so the last_size is set to |
4688 | * just before this extent. | |
514ac8ad | 4689 | */ |
fd86a3a3 | 4690 | ret = NEED_TRUNCATE_BLOCK; |
ddfae63c JB |
4691 | break; |
4692 | } | |
0305cd5f | 4693 | |
ddfae63c | 4694 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4695 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4696 | } |
179e29e4 | 4697 | delete: |
ddfae63c JB |
4698 | if (del_item) |
4699 | last_size = found_key.offset; | |
4700 | else | |
4701 | last_size = new_size; | |
39279cc3 | 4702 | if (del_item) { |
85e21bac CM |
4703 | if (!pending_del_nr) { |
4704 | /* no pending yet, add ourselves */ | |
4705 | pending_del_slot = path->slots[0]; | |
4706 | pending_del_nr = 1; | |
4707 | } else if (pending_del_nr && | |
4708 | path->slots[0] + 1 == pending_del_slot) { | |
4709 | /* hop on the pending chunk */ | |
4710 | pending_del_nr++; | |
4711 | pending_del_slot = path->slots[0]; | |
4712 | } else { | |
d397712b | 4713 | BUG(); |
85e21bac | 4714 | } |
39279cc3 CM |
4715 | } else { |
4716 | break; | |
4717 | } | |
897ca819 | 4718 | should_throttle = false; |
28f75a0e | 4719 | |
27cdeb70 MX |
4720 | if (found_extent && |
4721 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4722 | root == fs_info->tree_root)) { |
ffd4bb2a QW |
4723 | struct btrfs_ref ref = { 0 }; |
4724 | ||
b9473439 | 4725 | btrfs_set_path_blocking(path); |
28ed1345 | 4726 | bytes_deleted += extent_num_bytes; |
ffd4bb2a QW |
4727 | |
4728 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, | |
4729 | extent_start, extent_num_bytes, 0); | |
4730 | ref.real_root = root->root_key.objectid; | |
4731 | btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), | |
4732 | ino, extent_offset); | |
4733 | ret = btrfs_free_extent(trans, &ref); | |
05522109 OS |
4734 | if (ret) { |
4735 | btrfs_abort_transaction(trans, ret); | |
4736 | break; | |
4737 | } | |
28f75a0e | 4738 | if (be_nice) { |
7c861627 | 4739 | if (btrfs_should_throttle_delayed_refs(trans)) |
897ca819 | 4740 | should_throttle = true; |
28f75a0e | 4741 | } |
39279cc3 | 4742 | } |
85e21bac | 4743 | |
8082510e YZ |
4744 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4745 | break; | |
4746 | ||
4747 | if (path->slots[0] == 0 || | |
1262133b | 4748 | path->slots[0] != pending_del_slot || |
28bad212 | 4749 | should_throttle) { |
8082510e YZ |
4750 | if (pending_del_nr) { |
4751 | ret = btrfs_del_items(trans, root, path, | |
4752 | pending_del_slot, | |
4753 | pending_del_nr); | |
79787eaa | 4754 | if (ret) { |
66642832 | 4755 | btrfs_abort_transaction(trans, ret); |
fd86a3a3 | 4756 | break; |
79787eaa | 4757 | } |
8082510e YZ |
4758 | pending_del_nr = 0; |
4759 | } | |
b3b4aa74 | 4760 | btrfs_release_path(path); |
28bad212 | 4761 | |
28f75a0e | 4762 | /* |
28bad212 JB |
4763 | * We can generate a lot of delayed refs, so we need to |
4764 | * throttle every once and a while and make sure we're | |
4765 | * adding enough space to keep up with the work we are | |
4766 | * generating. Since we hold a transaction here we | |
4767 | * can't flush, and we don't want to FLUSH_LIMIT because | |
4768 | * we could have generated too many delayed refs to | |
4769 | * actually allocate, so just bail if we're short and | |
4770 | * let the normal reservation dance happen higher up. | |
28f75a0e | 4771 | */ |
28bad212 JB |
4772 | if (should_throttle) { |
4773 | ret = btrfs_delayed_refs_rsv_refill(fs_info, | |
4774 | BTRFS_RESERVE_NO_FLUSH); | |
4775 | if (ret) { | |
4776 | ret = -EAGAIN; | |
4777 | break; | |
4778 | } | |
28f75a0e | 4779 | } |
85e21bac | 4780 | goto search_again; |
8082510e YZ |
4781 | } else { |
4782 | path->slots[0]--; | |
85e21bac | 4783 | } |
39279cc3 | 4784 | } |
8082510e | 4785 | out: |
fd86a3a3 OS |
4786 | if (ret >= 0 && pending_del_nr) { |
4787 | int err; | |
4788 | ||
4789 | err = btrfs_del_items(trans, root, path, pending_del_slot, | |
85e21bac | 4790 | pending_del_nr); |
fd86a3a3 OS |
4791 | if (err) { |
4792 | btrfs_abort_transaction(trans, err); | |
4793 | ret = err; | |
4794 | } | |
85e21bac | 4795 | } |
76b42abb FM |
4796 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4797 | ASSERT(last_size >= new_size); | |
fd86a3a3 | 4798 | if (!ret && last_size > new_size) |
76b42abb | 4799 | last_size = new_size; |
7f4f6e0a | 4800 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4801 | } |
28ed1345 | 4802 | |
39279cc3 | 4803 | btrfs_free_path(path); |
fd86a3a3 | 4804 | return ret; |
39279cc3 CM |
4805 | } |
4806 | ||
4807 | /* | |
9703fefe | 4808 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4809 | * @inode - inode that we're zeroing |
4810 | * @from - the offset to start zeroing | |
4811 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4812 | * offset | |
4813 | * @front - zero up to the offset instead of from the offset on | |
4814 | * | |
9703fefe | 4815 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4816 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4817 | */ |
9703fefe | 4818 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4819 | int front) |
39279cc3 | 4820 | { |
0b246afa | 4821 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4822 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4823 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4824 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4825 | struct extent_state *cached_state = NULL; |
364ecf36 | 4826 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4827 | char *kaddr; |
0b246afa | 4828 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4829 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4830 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4831 | struct page *page; |
3b16a4e3 | 4832 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4833 | int ret = 0; |
9703fefe CR |
4834 | u64 block_start; |
4835 | u64 block_end; | |
39279cc3 | 4836 | |
b03ebd99 NB |
4837 | if (IS_ALIGNED(offset, blocksize) && |
4838 | (!len || IS_ALIGNED(len, blocksize))) | |
39279cc3 | 4839 | goto out; |
9703fefe | 4840 | |
8b62f87b JB |
4841 | block_start = round_down(from, blocksize); |
4842 | block_end = block_start + blocksize - 1; | |
4843 | ||
364ecf36 | 4844 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 4845 | block_start, blocksize); |
5d5e103a JB |
4846 | if (ret) |
4847 | goto out; | |
39279cc3 | 4848 | |
211c17f5 | 4849 | again: |
3b16a4e3 | 4850 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4851 | if (!page) { |
bc42bda2 | 4852 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
4853 | block_start, blocksize, true); |
4854 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true); | |
ac6a2b36 | 4855 | ret = -ENOMEM; |
39279cc3 | 4856 | goto out; |
5d5e103a | 4857 | } |
e6dcd2dc | 4858 | |
39279cc3 | 4859 | if (!PageUptodate(page)) { |
9ebefb18 | 4860 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4861 | lock_page(page); |
211c17f5 CM |
4862 | if (page->mapping != mapping) { |
4863 | unlock_page(page); | |
09cbfeaf | 4864 | put_page(page); |
211c17f5 CM |
4865 | goto again; |
4866 | } | |
39279cc3 CM |
4867 | if (!PageUptodate(page)) { |
4868 | ret = -EIO; | |
89642229 | 4869 | goto out_unlock; |
39279cc3 CM |
4870 | } |
4871 | } | |
211c17f5 | 4872 | wait_on_page_writeback(page); |
e6dcd2dc | 4873 | |
9703fefe | 4874 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4875 | set_page_extent_mapped(page); |
4876 | ||
9703fefe | 4877 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4878 | if (ordered) { |
9703fefe | 4879 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4880 | &cached_state); |
e6dcd2dc | 4881 | unlock_page(page); |
09cbfeaf | 4882 | put_page(page); |
eb84ae03 | 4883 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4884 | btrfs_put_ordered_extent(ordered); |
4885 | goto again; | |
4886 | } | |
4887 | ||
9703fefe | 4888 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4889 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4890 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 4891 | 0, 0, &cached_state); |
5d5e103a | 4892 | |
e3b8a485 | 4893 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
ba8b04c1 | 4894 | &cached_state, 0); |
9ed74f2d | 4895 | if (ret) { |
9703fefe | 4896 | unlock_extent_cached(io_tree, block_start, block_end, |
e43bbe5e | 4897 | &cached_state); |
9ed74f2d JB |
4898 | goto out_unlock; |
4899 | } | |
4900 | ||
9703fefe | 4901 | if (offset != blocksize) { |
2aaa6655 | 4902 | if (!len) |
9703fefe | 4903 | len = blocksize - offset; |
e6dcd2dc | 4904 | kaddr = kmap(page); |
2aaa6655 | 4905 | if (front) |
9703fefe CR |
4906 | memset(kaddr + (block_start - page_offset(page)), |
4907 | 0, offset); | |
2aaa6655 | 4908 | else |
9703fefe CR |
4909 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4910 | 0, len); | |
e6dcd2dc CM |
4911 | flush_dcache_page(page); |
4912 | kunmap(page); | |
4913 | } | |
247e743c | 4914 | ClearPageChecked(page); |
e6dcd2dc | 4915 | set_page_dirty(page); |
e43bbe5e | 4916 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state); |
39279cc3 | 4917 | |
89642229 | 4918 | out_unlock: |
5d5e103a | 4919 | if (ret) |
bc42bda2 | 4920 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
43b18595 QW |
4921 | blocksize, true); |
4922 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0)); | |
39279cc3 | 4923 | unlock_page(page); |
09cbfeaf | 4924 | put_page(page); |
39279cc3 | 4925 | out: |
364ecf36 | 4926 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4927 | return ret; |
4928 | } | |
4929 | ||
16e7549f JB |
4930 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4931 | u64 offset, u64 len) | |
4932 | { | |
0b246afa | 4933 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4934 | struct btrfs_trans_handle *trans; |
4935 | int ret; | |
4936 | ||
4937 | /* | |
4938 | * Still need to make sure the inode looks like it's been updated so | |
4939 | * that any holes get logged if we fsync. | |
4940 | */ | |
0b246afa JM |
4941 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4942 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4943 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4944 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4945 | return 0; | |
4946 | } | |
4947 | ||
4948 | /* | |
4949 | * 1 - for the one we're dropping | |
4950 | * 1 - for the one we're adding | |
4951 | * 1 - for updating the inode. | |
4952 | */ | |
4953 | trans = btrfs_start_transaction(root, 3); | |
4954 | if (IS_ERR(trans)) | |
4955 | return PTR_ERR(trans); | |
4956 | ||
4957 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4958 | if (ret) { | |
66642832 | 4959 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4960 | btrfs_end_transaction(trans); |
16e7549f JB |
4961 | return ret; |
4962 | } | |
4963 | ||
f85b7379 DS |
4964 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4965 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4966 | if (ret) |
66642832 | 4967 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4968 | else |
4969 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4970 | btrfs_end_transaction(trans); |
16e7549f JB |
4971 | return ret; |
4972 | } | |
4973 | ||
695a0d0d JB |
4974 | /* |
4975 | * This function puts in dummy file extents for the area we're creating a hole | |
4976 | * for. So if we are truncating this file to a larger size we need to insert | |
4977 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4978 | * the range between oldsize and size | |
4979 | */ | |
a41ad394 | 4980 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4981 | { |
0b246afa | 4982 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4983 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4984 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4985 | struct extent_map *em = NULL; |
2ac55d41 | 4986 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4987 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4988 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4989 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4990 | u64 last_byte; |
4991 | u64 cur_offset; | |
4992 | u64 hole_size; | |
9ed74f2d | 4993 | int err = 0; |
39279cc3 | 4994 | |
a71754fc | 4995 | /* |
9703fefe CR |
4996 | * If our size started in the middle of a block we need to zero out the |
4997 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4998 | * expose stale data. |
4999 | */ | |
9703fefe | 5000 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5001 | if (err) |
5002 | return err; | |
5003 | ||
9036c102 YZ |
5004 | if (size <= hole_start) |
5005 | return 0; | |
5006 | ||
23d31bd4 NB |
5007 | btrfs_lock_and_flush_ordered_range(io_tree, BTRFS_I(inode), hole_start, |
5008 | block_end - 1, &cached_state); | |
9036c102 YZ |
5009 | cur_offset = hole_start; |
5010 | while (1) { | |
fc4f21b1 | 5011 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 5012 | block_end - cur_offset, 0); |
79787eaa JM |
5013 | if (IS_ERR(em)) { |
5014 | err = PTR_ERR(em); | |
f2767956 | 5015 | em = NULL; |
79787eaa JM |
5016 | break; |
5017 | } | |
9036c102 | 5018 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5019 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 5020 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5021 | struct extent_map *hole_em; |
9036c102 | 5022 | hole_size = last_byte - cur_offset; |
9ed74f2d | 5023 | |
16e7549f JB |
5024 | err = maybe_insert_hole(root, inode, cur_offset, |
5025 | hole_size); | |
5026 | if (err) | |
3893e33b | 5027 | break; |
dcdbc059 | 5028 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
5029 | cur_offset + hole_size - 1, 0); |
5030 | hole_em = alloc_extent_map(); | |
5031 | if (!hole_em) { | |
5032 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
5033 | &BTRFS_I(inode)->runtime_flags); | |
5034 | goto next; | |
5035 | } | |
5036 | hole_em->start = cur_offset; | |
5037 | hole_em->len = hole_size; | |
5038 | hole_em->orig_start = cur_offset; | |
8082510e | 5039 | |
5dc562c5 JB |
5040 | hole_em->block_start = EXTENT_MAP_HOLE; |
5041 | hole_em->block_len = 0; | |
b4939680 | 5042 | hole_em->orig_block_len = 0; |
cc95bef6 | 5043 | hole_em->ram_bytes = hole_size; |
0b246afa | 5044 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 5045 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5046 | hole_em->generation = fs_info->generation; |
8082510e | 5047 | |
5dc562c5 JB |
5048 | while (1) { |
5049 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5050 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5051 | write_unlock(&em_tree->lock); |
5052 | if (err != -EEXIST) | |
5053 | break; | |
dcdbc059 NB |
5054 | btrfs_drop_extent_cache(BTRFS_I(inode), |
5055 | cur_offset, | |
5dc562c5 JB |
5056 | cur_offset + |
5057 | hole_size - 1, 0); | |
5058 | } | |
5059 | free_extent_map(hole_em); | |
9036c102 | 5060 | } |
16e7549f | 5061 | next: |
9036c102 | 5062 | free_extent_map(em); |
a22285a6 | 5063 | em = NULL; |
9036c102 | 5064 | cur_offset = last_byte; |
8082510e | 5065 | if (cur_offset >= block_end) |
9036c102 YZ |
5066 | break; |
5067 | } | |
a22285a6 | 5068 | free_extent_map(em); |
e43bbe5e | 5069 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state); |
9036c102 YZ |
5070 | return err; |
5071 | } | |
39279cc3 | 5072 | |
3972f260 | 5073 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5074 | { |
f4a2f4c5 MX |
5075 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5076 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5077 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5078 | loff_t newsize = attr->ia_size; |
5079 | int mask = attr->ia_valid; | |
8082510e YZ |
5080 | int ret; |
5081 | ||
3972f260 ES |
5082 | /* |
5083 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5084 | * special case where we need to update the times despite not having | |
5085 | * these flags set. For all other operations the VFS set these flags | |
5086 | * explicitly if it wants a timestamp update. | |
5087 | */ | |
dff6efc3 CH |
5088 | if (newsize != oldsize) { |
5089 | inode_inc_iversion(inode); | |
5090 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5091 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5092 | current_time(inode); |
dff6efc3 | 5093 | } |
3972f260 | 5094 | |
a41ad394 | 5095 | if (newsize > oldsize) { |
9ea24bbe | 5096 | /* |
ea14b57f | 5097 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5098 | * This is to ensure the snapshot captures a fully consistent |
5099 | * state of this file - if the snapshot captures this expanding | |
5100 | * truncation, it must capture all writes that happened before | |
5101 | * this truncation. | |
5102 | */ | |
0bc19f90 | 5103 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5104 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5105 | if (ret) { |
ea14b57f | 5106 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5107 | return ret; |
9ea24bbe | 5108 | } |
8082510e | 5109 | |
f4a2f4c5 | 5110 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5111 | if (IS_ERR(trans)) { |
ea14b57f | 5112 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5113 | return PTR_ERR(trans); |
9ea24bbe | 5114 | } |
f4a2f4c5 MX |
5115 | |
5116 | i_size_write(inode, newsize); | |
5117 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5118 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5119 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5120 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5121 | btrfs_end_transaction(trans); |
a41ad394 | 5122 | } else { |
8082510e | 5123 | |
a41ad394 JB |
5124 | /* |
5125 | * We're truncating a file that used to have good data down to | |
5126 | * zero. Make sure it gets into the ordered flush list so that | |
5127 | * any new writes get down to disk quickly. | |
5128 | */ | |
5129 | if (newsize == 0) | |
72ac3c0d JB |
5130 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5131 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5132 | |
a41ad394 | 5133 | truncate_setsize(inode, newsize); |
2e60a51e | 5134 | |
52042d8e | 5135 | /* Disable nonlocked read DIO to avoid the endless truncate */ |
abcefb1e | 5136 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5137 | inode_dio_wait(inode); |
0b581701 | 5138 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5139 | |
213e8c55 | 5140 | ret = btrfs_truncate(inode, newsize == oldsize); |
7f4f6e0a JB |
5141 | if (ret && inode->i_nlink) { |
5142 | int err; | |
5143 | ||
5144 | /* | |
f7e9e8fc OS |
5145 | * Truncate failed, so fix up the in-memory size. We |
5146 | * adjusted disk_i_size down as we removed extents, so | |
5147 | * wait for disk_i_size to be stable and then update the | |
5148 | * in-memory size to match. | |
7f4f6e0a | 5149 | */ |
f7e9e8fc | 5150 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); |
7f4f6e0a | 5151 | if (err) |
f7e9e8fc OS |
5152 | return err; |
5153 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
7f4f6e0a | 5154 | } |
8082510e YZ |
5155 | } |
5156 | ||
a41ad394 | 5157 | return ret; |
8082510e YZ |
5158 | } |
5159 | ||
9036c102 YZ |
5160 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5161 | { | |
2b0143b5 | 5162 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5163 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5164 | int err; |
39279cc3 | 5165 | |
b83cc969 LZ |
5166 | if (btrfs_root_readonly(root)) |
5167 | return -EROFS; | |
5168 | ||
31051c85 | 5169 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5170 | if (err) |
5171 | return err; | |
2bf5a725 | 5172 | |
5a3f23d5 | 5173 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5174 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5175 | if (err) |
5176 | return err; | |
39279cc3 | 5177 | } |
9036c102 | 5178 | |
1025774c CH |
5179 | if (attr->ia_valid) { |
5180 | setattr_copy(inode, attr); | |
0c4d2d95 | 5181 | inode_inc_iversion(inode); |
22c44fe6 | 5182 | err = btrfs_dirty_inode(inode); |
1025774c | 5183 | |
22c44fe6 | 5184 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5185 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5186 | } |
33268eaf | 5187 | |
39279cc3 CM |
5188 | return err; |
5189 | } | |
61295eb8 | 5190 | |
131e404a FDBM |
5191 | /* |
5192 | * While truncating the inode pages during eviction, we get the VFS calling | |
5193 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5194 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5195 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5196 | * extent_state structures over and over, wasting lots of time. | |
5197 | * | |
5198 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5199 | * those expensive operations on a per page basis and do only the ordered io | |
5200 | * finishing, while we release here the extent_map and extent_state structures, | |
5201 | * without the excessive merging and splitting. | |
5202 | */ | |
5203 | static void evict_inode_truncate_pages(struct inode *inode) | |
5204 | { | |
5205 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5206 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5207 | struct rb_node *node; | |
5208 | ||
5209 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5210 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5211 | |
5212 | write_lock(&map_tree->lock); | |
07e1ce09 | 5213 | while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) { |
131e404a FDBM |
5214 | struct extent_map *em; |
5215 | ||
07e1ce09 | 5216 | node = rb_first_cached(&map_tree->map); |
131e404a | 5217 | em = rb_entry(node, struct extent_map, rb_node); |
180589ef WS |
5218 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5219 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5220 | remove_extent_mapping(map_tree, em); |
5221 | free_extent_map(em); | |
7064dd5c FM |
5222 | if (need_resched()) { |
5223 | write_unlock(&map_tree->lock); | |
5224 | cond_resched(); | |
5225 | write_lock(&map_tree->lock); | |
5226 | } | |
131e404a FDBM |
5227 | } |
5228 | write_unlock(&map_tree->lock); | |
5229 | ||
6ca07097 FM |
5230 | /* |
5231 | * Keep looping until we have no more ranges in the io tree. | |
5232 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5233 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5234 | * still in progress (unlocked the pages in the bio but did not yet | |
5235 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5236 | * ranges can still be locked and eviction started because before |
5237 | * submitting those bios, which are executed by a separate task (work | |
5238 | * queue kthread), inode references (inode->i_count) were not taken | |
5239 | * (which would be dropped in the end io callback of each bio). | |
5240 | * Therefore here we effectively end up waiting for those bios and | |
5241 | * anyone else holding locked ranges without having bumped the inode's | |
5242 | * reference count - if we don't do it, when they access the inode's | |
5243 | * io_tree to unlock a range it may be too late, leading to an | |
5244 | * use-after-free issue. | |
5245 | */ | |
131e404a FDBM |
5246 | spin_lock(&io_tree->lock); |
5247 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5248 | struct extent_state *state; | |
5249 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5250 | u64 start; |
5251 | u64 end; | |
421f0922 | 5252 | unsigned state_flags; |
131e404a FDBM |
5253 | |
5254 | node = rb_first(&io_tree->state); | |
5255 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5256 | start = state->start; |
5257 | end = state->end; | |
421f0922 | 5258 | state_flags = state->state; |
131e404a FDBM |
5259 | spin_unlock(&io_tree->lock); |
5260 | ||
ff13db41 | 5261 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5262 | |
5263 | /* | |
5264 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5265 | * and its reserved space won't be freed by delayed_ref. | |
5266 | * So we need to free its reserved space here. | |
5267 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5268 | * | |
5269 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5270 | */ | |
421f0922 | 5271 | if (state_flags & EXTENT_DELALLOC) |
bc42bda2 | 5272 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5273 | |
6ca07097 | 5274 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5275 | EXTENT_LOCKED | EXTENT_DIRTY | |
5276 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
ae0f1625 | 5277 | EXTENT_DEFRAG, 1, 1, &cached_state); |
131e404a | 5278 | |
7064dd5c | 5279 | cond_resched(); |
131e404a FDBM |
5280 | spin_lock(&io_tree->lock); |
5281 | } | |
5282 | spin_unlock(&io_tree->lock); | |
5283 | } | |
5284 | ||
4b9d7b59 | 5285 | static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root, |
ad80cf50 | 5286 | struct btrfs_block_rsv *rsv) |
4b9d7b59 OS |
5287 | { |
5288 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5289 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
260e7702 | 5290 | u64 delayed_refs_extra = btrfs_calc_trans_metadata_size(fs_info, 1); |
4b9d7b59 OS |
5291 | int failures = 0; |
5292 | ||
5293 | for (;;) { | |
5294 | struct btrfs_trans_handle *trans; | |
5295 | int ret; | |
5296 | ||
260e7702 JB |
5297 | ret = btrfs_block_rsv_refill(root, rsv, |
5298 | rsv->size + delayed_refs_extra, | |
4b9d7b59 OS |
5299 | BTRFS_RESERVE_FLUSH_LIMIT); |
5300 | ||
5301 | if (ret && ++failures > 2) { | |
5302 | btrfs_warn(fs_info, | |
5303 | "could not allocate space for a delete; will truncate on mount"); | |
5304 | return ERR_PTR(-ENOSPC); | |
5305 | } | |
5306 | ||
260e7702 JB |
5307 | /* |
5308 | * Evict can generate a large amount of delayed refs without | |
5309 | * having a way to add space back since we exhaust our temporary | |
5310 | * block rsv. We aren't allowed to do FLUSH_ALL in this case | |
5311 | * because we could deadlock with so many things in the flushing | |
5312 | * code, so we have to try and hold some extra space to | |
5313 | * compensate for our delayed ref generation. If we can't get | |
5314 | * that space then we need see if we can steal our minimum from | |
5315 | * the global reserve. We will be ratelimited by the amount of | |
5316 | * space we have for the delayed refs rsv, so we'll end up | |
5317 | * committing and trying again. | |
5318 | */ | |
4b9d7b59 | 5319 | trans = btrfs_join_transaction(root); |
260e7702 JB |
5320 | if (IS_ERR(trans) || !ret) { |
5321 | if (!IS_ERR(trans)) { | |
5322 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5323 | trans->bytes_reserved = delayed_refs_extra; | |
5324 | btrfs_block_rsv_migrate(rsv, trans->block_rsv, | |
5325 | delayed_refs_extra, 1); | |
5326 | } | |
4b9d7b59 | 5327 | return trans; |
260e7702 | 5328 | } |
4b9d7b59 OS |
5329 | |
5330 | /* | |
5331 | * Try to steal from the global reserve if there is space for | |
5332 | * it. | |
5333 | */ | |
64403612 JB |
5334 | if (!btrfs_check_space_for_delayed_refs(fs_info) && |
5335 | !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) | |
4b9d7b59 OS |
5336 | return trans; |
5337 | ||
5338 | /* If not, commit and try again. */ | |
5339 | ret = btrfs_commit_transaction(trans); | |
5340 | if (ret) | |
5341 | return ERR_PTR(ret); | |
5342 | } | |
5343 | } | |
5344 | ||
bd555975 | 5345 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5346 | { |
0b246afa | 5347 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5348 | struct btrfs_trans_handle *trans; |
5349 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b9d7b59 | 5350 | struct btrfs_block_rsv *rsv; |
39279cc3 CM |
5351 | int ret; |
5352 | ||
1abe9b8a | 5353 | trace_btrfs_inode_evict(inode); |
5354 | ||
3d48d981 | 5355 | if (!root) { |
e8f1bc14 | 5356 | clear_inode(inode); |
3d48d981 NB |
5357 | return; |
5358 | } | |
5359 | ||
131e404a FDBM |
5360 | evict_inode_truncate_pages(inode); |
5361 | ||
69e9c6c6 SB |
5362 | if (inode->i_nlink && |
5363 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5364 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5365 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5366 | goto no_delete; |
5367 | ||
27919067 | 5368 | if (is_bad_inode(inode)) |
39279cc3 | 5369 | goto no_delete; |
5f39d397 | 5370 | |
7ab7956e | 5371 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5372 | |
7b40b695 | 5373 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
c71bf099 | 5374 | goto no_delete; |
c71bf099 | 5375 | |
76dda93c | 5376 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5377 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5378 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5379 | goto no_delete; |
5380 | } | |
5381 | ||
aa79021f | 5382 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
27919067 | 5383 | if (ret) |
0e8c36a9 | 5384 | goto no_delete; |
0e8c36a9 | 5385 | |
2ff7e61e | 5386 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
27919067 | 5387 | if (!rsv) |
4289a667 | 5388 | goto no_delete; |
ad80cf50 | 5389 | rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
ca7e70f5 | 5390 | rsv->failfast = 1; |
4289a667 | 5391 | |
6ef06d27 | 5392 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5393 | |
8082510e | 5394 | while (1) { |
ad80cf50 | 5395 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5396 | if (IS_ERR(trans)) |
5397 | goto free_rsv; | |
7b128766 | 5398 | |
4289a667 JB |
5399 | trans->block_rsv = rsv; |
5400 | ||
d68fc57b | 5401 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
27919067 OS |
5402 | trans->block_rsv = &fs_info->trans_block_rsv; |
5403 | btrfs_end_transaction(trans); | |
5404 | btrfs_btree_balance_dirty(fs_info); | |
5405 | if (ret && ret != -ENOSPC && ret != -EAGAIN) | |
5406 | goto free_rsv; | |
5407 | else if (!ret) | |
8082510e | 5408 | break; |
8082510e | 5409 | } |
5f39d397 | 5410 | |
4ef31a45 | 5411 | /* |
27919067 OS |
5412 | * Errors here aren't a big deal, it just means we leave orphan items in |
5413 | * the tree. They will be cleaned up on the next mount. If the inode | |
5414 | * number gets reused, cleanup deletes the orphan item without doing | |
5415 | * anything, and unlink reuses the existing orphan item. | |
5416 | * | |
5417 | * If it turns out that we are dropping too many of these, we might want | |
5418 | * to add a mechanism for retrying these after a commit. | |
4ef31a45 | 5419 | */ |
ad80cf50 | 5420 | trans = evict_refill_and_join(root, rsv); |
27919067 OS |
5421 | if (!IS_ERR(trans)) { |
5422 | trans->block_rsv = rsv; | |
5423 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
5424 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5425 | btrfs_end_transaction(trans); | |
5426 | } | |
54aa1f4d | 5427 | |
0b246afa | 5428 | if (!(root == fs_info->tree_root || |
581bb050 | 5429 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5430 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5431 | |
27919067 OS |
5432 | free_rsv: |
5433 | btrfs_free_block_rsv(fs_info, rsv); | |
39279cc3 | 5434 | no_delete: |
27919067 OS |
5435 | /* |
5436 | * If we didn't successfully delete, the orphan item will still be in | |
5437 | * the tree and we'll retry on the next mount. Again, we might also want | |
5438 | * to retry these periodically in the future. | |
5439 | */ | |
f48d1cf5 | 5440 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5441 | clear_inode(inode); |
39279cc3 CM |
5442 | } |
5443 | ||
5444 | /* | |
6bf9e4bd QW |
5445 | * Return the key found in the dir entry in the location pointer, fill @type |
5446 | * with BTRFS_FT_*, and return 0. | |
5447 | * | |
005d6712 SY |
5448 | * If no dir entries were found, returns -ENOENT. |
5449 | * If found a corrupted location in dir entry, returns -EUCLEAN. | |
39279cc3 CM |
5450 | */ |
5451 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
6bf9e4bd | 5452 | struct btrfs_key *location, u8 *type) |
39279cc3 CM |
5453 | { |
5454 | const char *name = dentry->d_name.name; | |
5455 | int namelen = dentry->d_name.len; | |
5456 | struct btrfs_dir_item *di; | |
5457 | struct btrfs_path *path; | |
5458 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5459 | int ret = 0; |
39279cc3 CM |
5460 | |
5461 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5462 | if (!path) |
5463 | return -ENOMEM; | |
3954401f | 5464 | |
f85b7379 DS |
5465 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5466 | name, namelen, 0); | |
3cf5068f LB |
5467 | if (IS_ERR_OR_NULL(di)) { |
5468 | ret = di ? PTR_ERR(di) : -ENOENT; | |
005d6712 SY |
5469 | goto out; |
5470 | } | |
d397712b | 5471 | |
5f39d397 | 5472 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5473 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5474 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
005d6712 | 5475 | ret = -EUCLEAN; |
56a0e706 LB |
5476 | btrfs_warn(root->fs_info, |
5477 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5478 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5479 | location->objectid, location->type, location->offset); | |
56a0e706 | 5480 | } |
6bf9e4bd QW |
5481 | if (!ret) |
5482 | *type = btrfs_dir_type(path->nodes[0], di); | |
39279cc3 | 5483 | out: |
39279cc3 CM |
5484 | btrfs_free_path(path); |
5485 | return ret; | |
5486 | } | |
5487 | ||
5488 | /* | |
5489 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5490 | * needs to be changed to reflect the root directory of the tree root. This | |
5491 | * is kind of like crossing a mount point. | |
5492 | */ | |
2ff7e61e | 5493 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5494 | struct inode *dir, |
5495 | struct dentry *dentry, | |
5496 | struct btrfs_key *location, | |
5497 | struct btrfs_root **sub_root) | |
39279cc3 | 5498 | { |
4df27c4d YZ |
5499 | struct btrfs_path *path; |
5500 | struct btrfs_root *new_root; | |
5501 | struct btrfs_root_ref *ref; | |
5502 | struct extent_buffer *leaf; | |
1d4c08e0 | 5503 | struct btrfs_key key; |
4df27c4d YZ |
5504 | int ret; |
5505 | int err = 0; | |
39279cc3 | 5506 | |
4df27c4d YZ |
5507 | path = btrfs_alloc_path(); |
5508 | if (!path) { | |
5509 | err = -ENOMEM; | |
5510 | goto out; | |
5511 | } | |
39279cc3 | 5512 | |
4df27c4d | 5513 | err = -ENOENT; |
1d4c08e0 DS |
5514 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5515 | key.type = BTRFS_ROOT_REF_KEY; | |
5516 | key.offset = location->objectid; | |
5517 | ||
0b246afa | 5518 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5519 | if (ret) { |
5520 | if (ret < 0) | |
5521 | err = ret; | |
5522 | goto out; | |
5523 | } | |
39279cc3 | 5524 | |
4df27c4d YZ |
5525 | leaf = path->nodes[0]; |
5526 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5527 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5528 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5529 | goto out; | |
39279cc3 | 5530 | |
4df27c4d YZ |
5531 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5532 | (unsigned long)(ref + 1), | |
5533 | dentry->d_name.len); | |
5534 | if (ret) | |
5535 | goto out; | |
5536 | ||
b3b4aa74 | 5537 | btrfs_release_path(path); |
4df27c4d | 5538 | |
0b246afa | 5539 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5540 | if (IS_ERR(new_root)) { |
5541 | err = PTR_ERR(new_root); | |
5542 | goto out; | |
5543 | } | |
5544 | ||
4df27c4d YZ |
5545 | *sub_root = new_root; |
5546 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5547 | location->type = BTRFS_INODE_ITEM_KEY; | |
5548 | location->offset = 0; | |
5549 | err = 0; | |
5550 | out: | |
5551 | btrfs_free_path(path); | |
5552 | return err; | |
39279cc3 CM |
5553 | } |
5554 | ||
5d4f98a2 YZ |
5555 | static void inode_tree_add(struct inode *inode) |
5556 | { | |
5557 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5558 | struct btrfs_inode *entry; | |
03e860bd FNP |
5559 | struct rb_node **p; |
5560 | struct rb_node *parent; | |
cef21937 | 5561 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5562 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5563 | |
1d3382cb | 5564 | if (inode_unhashed(inode)) |
76dda93c | 5565 | return; |
e1409cef | 5566 | parent = NULL; |
5d4f98a2 | 5567 | spin_lock(&root->inode_lock); |
e1409cef | 5568 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5569 | while (*p) { |
5570 | parent = *p; | |
5571 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5572 | ||
37508515 | 5573 | if (ino < btrfs_ino(entry)) |
03e860bd | 5574 | p = &parent->rb_left; |
37508515 | 5575 | else if (ino > btrfs_ino(entry)) |
03e860bd | 5576 | p = &parent->rb_right; |
5d4f98a2 YZ |
5577 | else { |
5578 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5579 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5580 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5581 | RB_CLEAR_NODE(parent); |
5582 | spin_unlock(&root->inode_lock); | |
cef21937 | 5583 | return; |
5d4f98a2 YZ |
5584 | } |
5585 | } | |
cef21937 FDBM |
5586 | rb_link_node(new, parent, p); |
5587 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5588 | spin_unlock(&root->inode_lock); |
5589 | } | |
5590 | ||
5591 | static void inode_tree_del(struct inode *inode) | |
5592 | { | |
0b246afa | 5593 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5594 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5595 | int empty = 0; |
5d4f98a2 | 5596 | |
03e860bd | 5597 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5598 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5599 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5600 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5601 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5602 | } |
03e860bd | 5603 | spin_unlock(&root->inode_lock); |
76dda93c | 5604 | |
69e9c6c6 | 5605 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5606 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5607 | spin_lock(&root->inode_lock); |
5608 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5609 | spin_unlock(&root->inode_lock); | |
5610 | if (empty) | |
5611 | btrfs_add_dead_root(root); | |
5612 | } | |
5613 | } | |
5614 | ||
5d4f98a2 | 5615 | |
e02119d5 CM |
5616 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5617 | { | |
5618 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5619 | inode->i_ino = args->location->objectid; |
5620 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5621 | sizeof(*args->location)); | |
e02119d5 | 5622 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5623 | return 0; |
5624 | } | |
5625 | ||
5626 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5627 | { | |
5628 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5629 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5630 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5631 | } |
5632 | ||
5d4f98a2 | 5633 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5634 | struct btrfs_key *location, |
5d4f98a2 | 5635 | struct btrfs_root *root) |
39279cc3 CM |
5636 | { |
5637 | struct inode *inode; | |
5638 | struct btrfs_iget_args args; | |
90d3e592 | 5639 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5640 | |
90d3e592 | 5641 | args.location = location; |
39279cc3 CM |
5642 | args.root = root; |
5643 | ||
778ba82b | 5644 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5645 | btrfs_init_locked_inode, |
5646 | (void *)&args); | |
5647 | return inode; | |
5648 | } | |
5649 | ||
1a54ef8c BR |
5650 | /* Get an inode object given its location and corresponding root. |
5651 | * Returns in *is_new if the inode was read from disk | |
5652 | */ | |
4222ea71 FM |
5653 | struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location, |
5654 | struct btrfs_root *root, int *new, | |
5655 | struct btrfs_path *path) | |
1a54ef8c BR |
5656 | { |
5657 | struct inode *inode; | |
5658 | ||
90d3e592 | 5659 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5660 | if (!inode) |
5d4f98a2 | 5661 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5662 | |
5663 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5664 | int ret; |
5665 | ||
4222ea71 | 5666 | ret = btrfs_read_locked_inode(inode, path); |
9bc2ceff | 5667 | if (!ret) { |
1748f843 MF |
5668 | inode_tree_add(inode); |
5669 | unlock_new_inode(inode); | |
5670 | if (new) | |
5671 | *new = 1; | |
5672 | } else { | |
f5b3a417 AV |
5673 | iget_failed(inode); |
5674 | /* | |
5675 | * ret > 0 can come from btrfs_search_slot called by | |
5676 | * btrfs_read_locked_inode, this means the inode item | |
5677 | * was not found. | |
5678 | */ | |
5679 | if (ret > 0) | |
5680 | ret = -ENOENT; | |
5681 | inode = ERR_PTR(ret); | |
1748f843 MF |
5682 | } |
5683 | } | |
5684 | ||
1a54ef8c BR |
5685 | return inode; |
5686 | } | |
5687 | ||
4222ea71 FM |
5688 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, |
5689 | struct btrfs_root *root, int *new) | |
5690 | { | |
5691 | return btrfs_iget_path(s, location, root, new, NULL); | |
5692 | } | |
5693 | ||
4df27c4d YZ |
5694 | static struct inode *new_simple_dir(struct super_block *s, |
5695 | struct btrfs_key *key, | |
5696 | struct btrfs_root *root) | |
5697 | { | |
5698 | struct inode *inode = new_inode(s); | |
5699 | ||
5700 | if (!inode) | |
5701 | return ERR_PTR(-ENOMEM); | |
5702 | ||
4df27c4d YZ |
5703 | BTRFS_I(inode)->root = root; |
5704 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5705 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5706 | |
5707 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5708 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5709 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5710 | inode->i_fop = &simple_dir_operations; |
5711 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5712 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5713 | inode->i_atime = inode->i_mtime; |
5714 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 5715 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
4df27c4d YZ |
5716 | |
5717 | return inode; | |
5718 | } | |
5719 | ||
6bf9e4bd QW |
5720 | static inline u8 btrfs_inode_type(struct inode *inode) |
5721 | { | |
5722 | /* | |
5723 | * Compile-time asserts that generic FT_* types still match | |
5724 | * BTRFS_FT_* types | |
5725 | */ | |
5726 | BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN); | |
5727 | BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE); | |
5728 | BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR); | |
5729 | BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV); | |
5730 | BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV); | |
5731 | BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO); | |
5732 | BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK); | |
5733 | BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK); | |
5734 | ||
5735 | return fs_umode_to_ftype(inode->i_mode); | |
5736 | } | |
5737 | ||
3de4586c | 5738 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5739 | { |
0b246afa | 5740 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5741 | struct inode *inode; |
4df27c4d | 5742 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5743 | struct btrfs_root *sub_root = root; |
5744 | struct btrfs_key location; | |
6bf9e4bd | 5745 | u8 di_type = 0; |
76dda93c | 5746 | int index; |
b4aff1f8 | 5747 | int ret = 0; |
39279cc3 CM |
5748 | |
5749 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5750 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5751 | |
6bf9e4bd | 5752 | ret = btrfs_inode_by_name(dir, dentry, &location, &di_type); |
39279cc3 CM |
5753 | if (ret < 0) |
5754 | return ERR_PTR(ret); | |
5f39d397 | 5755 | |
4df27c4d | 5756 | if (location.type == BTRFS_INODE_ITEM_KEY) { |
73f73415 | 5757 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
6bf9e4bd QW |
5758 | if (IS_ERR(inode)) |
5759 | return inode; | |
5760 | ||
5761 | /* Do extra check against inode mode with di_type */ | |
5762 | if (btrfs_inode_type(inode) != di_type) { | |
5763 | btrfs_crit(fs_info, | |
5764 | "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u", | |
5765 | inode->i_mode, btrfs_inode_type(inode), | |
5766 | di_type); | |
5767 | iput(inode); | |
5768 | return ERR_PTR(-EUCLEAN); | |
5769 | } | |
4df27c4d YZ |
5770 | return inode; |
5771 | } | |
5772 | ||
0b246afa | 5773 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5774 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5775 | &location, &sub_root); |
5776 | if (ret < 0) { | |
5777 | if (ret != -ENOENT) | |
5778 | inode = ERR_PTR(ret); | |
5779 | else | |
5780 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5781 | } else { | |
73f73415 | 5782 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5783 | } |
0b246afa | 5784 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5785 | |
34d19bad | 5786 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5787 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5788 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5789 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5790 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5791 | if (ret) { |
5792 | iput(inode); | |
66b4ffd1 | 5793 | inode = ERR_PTR(ret); |
01cd3367 | 5794 | } |
c71bf099 YZ |
5795 | } |
5796 | ||
3de4586c CM |
5797 | return inode; |
5798 | } | |
5799 | ||
fe15ce44 | 5800 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5801 | { |
5802 | struct btrfs_root *root; | |
2b0143b5 | 5803 | struct inode *inode = d_inode(dentry); |
76dda93c | 5804 | |
848cce0d | 5805 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5806 | inode = d_inode(dentry->d_parent); |
76dda93c | 5807 | |
848cce0d LZ |
5808 | if (inode) { |
5809 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5810 | if (btrfs_root_refs(&root->root_item) == 0) |
5811 | return 1; | |
848cce0d | 5812 | |
4a0cc7ca | 5813 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5814 | return 1; |
efefb143 | 5815 | } |
76dda93c YZ |
5816 | return 0; |
5817 | } | |
5818 | ||
3de4586c | 5819 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5820 | unsigned int flags) |
3de4586c | 5821 | { |
3837d208 | 5822 | struct inode *inode = btrfs_lookup_dentry(dir, dentry); |
5662344b | 5823 | |
3837d208 AV |
5824 | if (inode == ERR_PTR(-ENOENT)) |
5825 | inode = NULL; | |
41d28bca | 5826 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5827 | } |
5828 | ||
23b5ec74 JB |
5829 | /* |
5830 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5831 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5832 | * our information into that, and then dir_emit from the buffer. This is | |
5833 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5834 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5835 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5836 | * tree lock. | |
5837 | */ | |
5838 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5839 | { | |
5840 | struct btrfs_file_private *private; | |
5841 | ||
5842 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5843 | if (!private) | |
5844 | return -ENOMEM; | |
5845 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5846 | if (!private->filldir_buf) { | |
5847 | kfree(private); | |
5848 | return -ENOMEM; | |
5849 | } | |
5850 | file->private_data = private; | |
5851 | return 0; | |
5852 | } | |
5853 | ||
5854 | struct dir_entry { | |
5855 | u64 ino; | |
5856 | u64 offset; | |
5857 | unsigned type; | |
5858 | int name_len; | |
5859 | }; | |
5860 | ||
5861 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5862 | { | |
5863 | while (entries--) { | |
5864 | struct dir_entry *entry = addr; | |
5865 | char *name = (char *)(entry + 1); | |
5866 | ||
92d32170 DS |
5867 | ctx->pos = get_unaligned(&entry->offset); |
5868 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
5869 | get_unaligned(&entry->ino), | |
5870 | get_unaligned(&entry->type))) | |
23b5ec74 | 5871 | return 1; |
92d32170 DS |
5872 | addr += sizeof(struct dir_entry) + |
5873 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
5874 | ctx->pos++; |
5875 | } | |
5876 | return 0; | |
5877 | } | |
5878 | ||
9cdda8d3 | 5879 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5880 | { |
9cdda8d3 | 5881 | struct inode *inode = file_inode(file); |
39279cc3 | 5882 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5883 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5884 | struct btrfs_dir_item *di; |
5885 | struct btrfs_key key; | |
5f39d397 | 5886 | struct btrfs_key found_key; |
39279cc3 | 5887 | struct btrfs_path *path; |
23b5ec74 | 5888 | void *addr; |
16cdcec7 MX |
5889 | struct list_head ins_list; |
5890 | struct list_head del_list; | |
39279cc3 | 5891 | int ret; |
5f39d397 | 5892 | struct extent_buffer *leaf; |
39279cc3 | 5893 | int slot; |
5f39d397 CM |
5894 | char *name_ptr; |
5895 | int name_len; | |
23b5ec74 JB |
5896 | int entries = 0; |
5897 | int total_len = 0; | |
02dbfc99 | 5898 | bool put = false; |
c2951f32 | 5899 | struct btrfs_key location; |
5f39d397 | 5900 | |
9cdda8d3 AV |
5901 | if (!dir_emit_dots(file, ctx)) |
5902 | return 0; | |
5903 | ||
49593bfa | 5904 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5905 | if (!path) |
5906 | return -ENOMEM; | |
ff5714cc | 5907 | |
23b5ec74 | 5908 | addr = private->filldir_buf; |
e4058b54 | 5909 | path->reada = READA_FORWARD; |
49593bfa | 5910 | |
c2951f32 JM |
5911 | INIT_LIST_HEAD(&ins_list); |
5912 | INIT_LIST_HEAD(&del_list); | |
5913 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5914 | |
23b5ec74 | 5915 | again: |
c2951f32 | 5916 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5917 | key.offset = ctx->pos; |
4a0cc7ca | 5918 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5919 | |
39279cc3 CM |
5920 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5921 | if (ret < 0) | |
5922 | goto err; | |
49593bfa DW |
5923 | |
5924 | while (1) { | |
23b5ec74 JB |
5925 | struct dir_entry *entry; |
5926 | ||
5f39d397 | 5927 | leaf = path->nodes[0]; |
39279cc3 | 5928 | slot = path->slots[0]; |
b9e03af0 LZ |
5929 | if (slot >= btrfs_header_nritems(leaf)) { |
5930 | ret = btrfs_next_leaf(root, path); | |
5931 | if (ret < 0) | |
5932 | goto err; | |
5933 | else if (ret > 0) | |
5934 | break; | |
5935 | continue; | |
39279cc3 | 5936 | } |
3de4586c | 5937 | |
5f39d397 CM |
5938 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5939 | ||
5940 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5941 | break; |
c2951f32 | 5942 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5943 | break; |
9cdda8d3 | 5944 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5945 | goto next; |
c2951f32 | 5946 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5947 | goto next; |
39279cc3 | 5948 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
c2951f32 | 5949 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5950 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5951 | PAGE_SIZE) { | |
5952 | btrfs_release_path(path); | |
5953 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5954 | if (ret) | |
5955 | goto nopos; | |
5956 | addr = private->filldir_buf; | |
5957 | entries = 0; | |
5958 | total_len = 0; | |
5959 | goto again; | |
c2951f32 | 5960 | } |
23b5ec74 JB |
5961 | |
5962 | entry = addr; | |
92d32170 | 5963 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 5964 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
5965 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5966 | name_len); | |
7d157c3d | 5967 | put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)), |
92d32170 | 5968 | &entry->type); |
c2951f32 | 5969 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
92d32170 DS |
5970 | put_unaligned(location.objectid, &entry->ino); |
5971 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
5972 | entries++; |
5973 | addr += sizeof(struct dir_entry) + name_len; | |
5974 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
5975 | next: |
5976 | path->slots[0]++; | |
39279cc3 | 5977 | } |
23b5ec74 JB |
5978 | btrfs_release_path(path); |
5979 | ||
5980 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5981 | if (ret) | |
5982 | goto nopos; | |
49593bfa | 5983 | |
d2fbb2b5 | 5984 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5985 | if (ret) |
bc4ef759 DS |
5986 | goto nopos; |
5987 | ||
db62efbb ZB |
5988 | /* |
5989 | * Stop new entries from being returned after we return the last | |
5990 | * entry. | |
5991 | * | |
5992 | * New directory entries are assigned a strictly increasing | |
5993 | * offset. This means that new entries created during readdir | |
5994 | * are *guaranteed* to be seen in the future by that readdir. | |
5995 | * This has broken buggy programs which operate on names as | |
5996 | * they're returned by readdir. Until we re-use freed offsets | |
5997 | * we have this hack to stop new entries from being returned | |
5998 | * under the assumption that they'll never reach this huge | |
5999 | * offset. | |
6000 | * | |
6001 | * This is being careful not to overflow 32bit loff_t unless the | |
6002 | * last entry requires it because doing so has broken 32bit apps | |
6003 | * in the past. | |
6004 | */ | |
c2951f32 JM |
6005 | if (ctx->pos >= INT_MAX) |
6006 | ctx->pos = LLONG_MAX; | |
6007 | else | |
6008 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6009 | nopos: |
6010 | ret = 0; | |
6011 | err: | |
02dbfc99 OS |
6012 | if (put) |
6013 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6014 | btrfs_free_path(path); |
39279cc3 CM |
6015 | return ret; |
6016 | } | |
6017 | ||
39279cc3 | 6018 | /* |
54aa1f4d | 6019 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6020 | * inode changes. But, it is most likely to find the inode in cache. |
6021 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6022 | * to keep or drop this code. | |
6023 | */ | |
48a3b636 | 6024 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6025 | { |
2ff7e61e | 6026 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6027 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6028 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6029 | int ret; |
6030 | ||
72ac3c0d | 6031 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6032 | return 0; |
39279cc3 | 6033 | |
7a7eaa40 | 6034 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6035 | if (IS_ERR(trans)) |
6036 | return PTR_ERR(trans); | |
8929ecfa YZ |
6037 | |
6038 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6039 | if (ret && ret == -ENOSPC) { |
6040 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6041 | btrfs_end_transaction(trans); |
94b60442 | 6042 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6043 | if (IS_ERR(trans)) |
6044 | return PTR_ERR(trans); | |
8929ecfa | 6045 | |
94b60442 | 6046 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6047 | } |
3a45bb20 | 6048 | btrfs_end_transaction(trans); |
16cdcec7 | 6049 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6050 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6051 | |
6052 | return ret; | |
6053 | } | |
6054 | ||
6055 | /* | |
6056 | * This is a copy of file_update_time. We need this so we can return error on | |
6057 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6058 | */ | |
95582b00 | 6059 | static int btrfs_update_time(struct inode *inode, struct timespec64 *now, |
e41f941a | 6060 | int flags) |
22c44fe6 | 6061 | { |
2bc55652 | 6062 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3a8c7231 | 6063 | bool dirty = flags & ~S_VERSION; |
2bc55652 AB |
6064 | |
6065 | if (btrfs_root_readonly(root)) | |
6066 | return -EROFS; | |
6067 | ||
e41f941a | 6068 | if (flags & S_VERSION) |
3a8c7231 | 6069 | dirty |= inode_maybe_inc_iversion(inode, dirty); |
e41f941a JB |
6070 | if (flags & S_CTIME) |
6071 | inode->i_ctime = *now; | |
6072 | if (flags & S_MTIME) | |
6073 | inode->i_mtime = *now; | |
6074 | if (flags & S_ATIME) | |
6075 | inode->i_atime = *now; | |
3a8c7231 | 6076 | return dirty ? btrfs_dirty_inode(inode) : 0; |
39279cc3 CM |
6077 | } |
6078 | ||
d352ac68 CM |
6079 | /* |
6080 | * find the highest existing sequence number in a directory | |
6081 | * and then set the in-memory index_cnt variable to reflect | |
6082 | * free sequence numbers | |
6083 | */ | |
4c570655 | 6084 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6085 | { |
4c570655 | 6086 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6087 | struct btrfs_key key, found_key; |
6088 | struct btrfs_path *path; | |
6089 | struct extent_buffer *leaf; | |
6090 | int ret; | |
6091 | ||
4c570655 | 6092 | key.objectid = btrfs_ino(inode); |
962a298f | 6093 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6094 | key.offset = (u64)-1; |
6095 | ||
6096 | path = btrfs_alloc_path(); | |
6097 | if (!path) | |
6098 | return -ENOMEM; | |
6099 | ||
6100 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6101 | if (ret < 0) | |
6102 | goto out; | |
6103 | /* FIXME: we should be able to handle this */ | |
6104 | if (ret == 0) | |
6105 | goto out; | |
6106 | ret = 0; | |
6107 | ||
6108 | /* | |
6109 | * MAGIC NUMBER EXPLANATION: | |
6110 | * since we search a directory based on f_pos we have to start at 2 | |
6111 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6112 | * else has to start at 2 | |
6113 | */ | |
6114 | if (path->slots[0] == 0) { | |
4c570655 | 6115 | inode->index_cnt = 2; |
aec7477b JB |
6116 | goto out; |
6117 | } | |
6118 | ||
6119 | path->slots[0]--; | |
6120 | ||
6121 | leaf = path->nodes[0]; | |
6122 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6123 | ||
4c570655 | 6124 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6125 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6126 | inode->index_cnt = 2; |
aec7477b JB |
6127 | goto out; |
6128 | } | |
6129 | ||
4c570655 | 6130 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6131 | out: |
6132 | btrfs_free_path(path); | |
6133 | return ret; | |
6134 | } | |
6135 | ||
d352ac68 CM |
6136 | /* |
6137 | * helper to find a free sequence number in a given directory. This current | |
6138 | * code is very simple, later versions will do smarter things in the btree | |
6139 | */ | |
877574e2 | 6140 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6141 | { |
6142 | int ret = 0; | |
6143 | ||
877574e2 NB |
6144 | if (dir->index_cnt == (u64)-1) { |
6145 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6146 | if (ret) { |
6147 | ret = btrfs_set_inode_index_count(dir); | |
6148 | if (ret) | |
6149 | return ret; | |
6150 | } | |
aec7477b JB |
6151 | } |
6152 | ||
877574e2 NB |
6153 | *index = dir->index_cnt; |
6154 | dir->index_cnt++; | |
aec7477b JB |
6155 | |
6156 | return ret; | |
6157 | } | |
6158 | ||
b0d5d10f CM |
6159 | static int btrfs_insert_inode_locked(struct inode *inode) |
6160 | { | |
6161 | struct btrfs_iget_args args; | |
6162 | args.location = &BTRFS_I(inode)->location; | |
6163 | args.root = BTRFS_I(inode)->root; | |
6164 | ||
6165 | return insert_inode_locked4(inode, | |
6166 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6167 | btrfs_find_actor, &args); | |
6168 | } | |
6169 | ||
19aee8de AJ |
6170 | /* |
6171 | * Inherit flags from the parent inode. | |
6172 | * | |
6173 | * Currently only the compression flags and the cow flags are inherited. | |
6174 | */ | |
6175 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6176 | { | |
6177 | unsigned int flags; | |
6178 | ||
6179 | if (!dir) | |
6180 | return; | |
6181 | ||
6182 | flags = BTRFS_I(dir)->flags; | |
6183 | ||
6184 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6185 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6186 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6187 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6188 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6189 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6190 | } | |
6191 | ||
6192 | if (flags & BTRFS_INODE_NODATACOW) { | |
6193 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6194 | if (S_ISREG(inode->i_mode)) | |
6195 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6196 | } | |
6197 | ||
7b6a221e | 6198 | btrfs_sync_inode_flags_to_i_flags(inode); |
19aee8de AJ |
6199 | } |
6200 | ||
39279cc3 CM |
6201 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6202 | struct btrfs_root *root, | |
aec7477b | 6203 | struct inode *dir, |
9c58309d | 6204 | const char *name, int name_len, |
175a4eb7 AV |
6205 | u64 ref_objectid, u64 objectid, |
6206 | umode_t mode, u64 *index) | |
39279cc3 | 6207 | { |
0b246afa | 6208 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6209 | struct inode *inode; |
5f39d397 | 6210 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6211 | struct btrfs_key *location; |
5f39d397 | 6212 | struct btrfs_path *path; |
9c58309d CM |
6213 | struct btrfs_inode_ref *ref; |
6214 | struct btrfs_key key[2]; | |
6215 | u32 sizes[2]; | |
ef3b9af5 | 6216 | int nitems = name ? 2 : 1; |
9c58309d | 6217 | unsigned long ptr; |
39279cc3 | 6218 | int ret; |
39279cc3 | 6219 | |
5f39d397 | 6220 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6221 | if (!path) |
6222 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6223 | |
0b246afa | 6224 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6225 | if (!inode) { |
6226 | btrfs_free_path(path); | |
39279cc3 | 6227 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6228 | } |
39279cc3 | 6229 | |
5762b5c9 FM |
6230 | /* |
6231 | * O_TMPFILE, set link count to 0, so that after this point, | |
6232 | * we fill in an inode item with the correct link count. | |
6233 | */ | |
6234 | if (!name) | |
6235 | set_nlink(inode, 0); | |
6236 | ||
581bb050 LZ |
6237 | /* |
6238 | * we have to initialize this early, so we can reclaim the inode | |
6239 | * number if we fail afterwards in this function. | |
6240 | */ | |
6241 | inode->i_ino = objectid; | |
6242 | ||
ef3b9af5 | 6243 | if (dir && name) { |
1abe9b8a | 6244 | trace_btrfs_inode_request(dir); |
6245 | ||
877574e2 | 6246 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6247 | if (ret) { |
8fb27640 | 6248 | btrfs_free_path(path); |
09771430 | 6249 | iput(inode); |
aec7477b | 6250 | return ERR_PTR(ret); |
09771430 | 6251 | } |
ef3b9af5 FM |
6252 | } else if (dir) { |
6253 | *index = 0; | |
aec7477b JB |
6254 | } |
6255 | /* | |
6256 | * index_cnt is ignored for everything but a dir, | |
df6703e1 | 6257 | * btrfs_set_inode_index_count has an explanation for the magic |
aec7477b JB |
6258 | * number |
6259 | */ | |
6260 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6261 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6262 | BTRFS_I(inode)->root = root; |
e02119d5 | 6263 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6264 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6265 | |
5dc562c5 JB |
6266 | /* |
6267 | * We could have gotten an inode number from somebody who was fsynced | |
6268 | * and then removed in this same transaction, so let's just set full | |
6269 | * sync since it will be a full sync anyway and this will blow away the | |
6270 | * old info in the log. | |
6271 | */ | |
6272 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6273 | ||
9c58309d | 6274 | key[0].objectid = objectid; |
962a298f | 6275 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6276 | key[0].offset = 0; |
6277 | ||
9c58309d | 6278 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6279 | |
6280 | if (name) { | |
6281 | /* | |
6282 | * Start new inodes with an inode_ref. This is slightly more | |
6283 | * efficient for small numbers of hard links since they will | |
6284 | * be packed into one item. Extended refs will kick in if we | |
6285 | * add more hard links than can fit in the ref item. | |
6286 | */ | |
6287 | key[1].objectid = objectid; | |
962a298f | 6288 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6289 | key[1].offset = ref_objectid; |
6290 | ||
6291 | sizes[1] = name_len + sizeof(*ref); | |
6292 | } | |
9c58309d | 6293 | |
b0d5d10f CM |
6294 | location = &BTRFS_I(inode)->location; |
6295 | location->objectid = objectid; | |
6296 | location->offset = 0; | |
962a298f | 6297 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6298 | |
6299 | ret = btrfs_insert_inode_locked(inode); | |
32955c54 AV |
6300 | if (ret < 0) { |
6301 | iput(inode); | |
b0d5d10f | 6302 | goto fail; |
32955c54 | 6303 | } |
b0d5d10f | 6304 | |
b9473439 | 6305 | path->leave_spinning = 1; |
ef3b9af5 | 6306 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6307 | if (ret != 0) |
b0d5d10f | 6308 | goto fail_unlock; |
5f39d397 | 6309 | |
ecc11fab | 6310 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6311 | inode_set_bytes(inode, 0); |
9cc97d64 | 6312 | |
c2050a45 | 6313 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6314 | inode->i_atime = inode->i_mtime; |
6315 | inode->i_ctime = inode->i_mtime; | |
d3c6be6f | 6316 | BTRFS_I(inode)->i_otime = inode->i_mtime; |
9cc97d64 | 6317 | |
5f39d397 CM |
6318 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6319 | struct btrfs_inode_item); | |
b159fa28 | 6320 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6321 | sizeof(*inode_item)); |
e02119d5 | 6322 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6323 | |
ef3b9af5 FM |
6324 | if (name) { |
6325 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6326 | struct btrfs_inode_ref); | |
6327 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6328 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6329 | ptr = (unsigned long)(ref + 1); | |
6330 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6331 | } | |
9c58309d | 6332 | |
5f39d397 CM |
6333 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6334 | btrfs_free_path(path); | |
6335 | ||
6cbff00f CH |
6336 | btrfs_inherit_iflags(inode, dir); |
6337 | ||
569254b0 | 6338 | if (S_ISREG(mode)) { |
0b246afa | 6339 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6340 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6341 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6342 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6343 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6344 | } |
6345 | ||
5d4f98a2 | 6346 | inode_tree_add(inode); |
1abe9b8a | 6347 | |
6348 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6349 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6350 | |
8ea05e3a AB |
6351 | btrfs_update_root_times(trans, root); |
6352 | ||
63541927 FDBM |
6353 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6354 | if (ret) | |
0b246afa | 6355 | btrfs_err(fs_info, |
63541927 | 6356 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6357 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6358 | |
39279cc3 | 6359 | return inode; |
b0d5d10f CM |
6360 | |
6361 | fail_unlock: | |
32955c54 | 6362 | discard_new_inode(inode); |
5f39d397 | 6363 | fail: |
ef3b9af5 | 6364 | if (dir && name) |
aec7477b | 6365 | BTRFS_I(dir)->index_cnt--; |
5f39d397 CM |
6366 | btrfs_free_path(path); |
6367 | return ERR_PTR(ret); | |
39279cc3 CM |
6368 | } |
6369 | ||
d352ac68 CM |
6370 | /* |
6371 | * utility function to add 'inode' into 'parent_inode' with | |
6372 | * a give name and a given sequence number. | |
6373 | * if 'add_backref' is true, also insert a backref from the | |
6374 | * inode to the parent directory. | |
6375 | */ | |
e02119d5 | 6376 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6377 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6378 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6379 | { |
4df27c4d | 6380 | int ret = 0; |
39279cc3 | 6381 | struct btrfs_key key; |
db0a669f NB |
6382 | struct btrfs_root *root = parent_inode->root; |
6383 | u64 ino = btrfs_ino(inode); | |
6384 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6385 | |
33345d01 | 6386 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6387 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6388 | } else { |
33345d01 | 6389 | key.objectid = ino; |
962a298f | 6390 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6391 | key.offset = 0; |
6392 | } | |
6393 | ||
33345d01 | 6394 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
6025c19f | 6395 | ret = btrfs_add_root_ref(trans, key.objectid, |
0b246afa JM |
6396 | root->root_key.objectid, parent_ino, |
6397 | index, name, name_len); | |
4df27c4d | 6398 | } else if (add_backref) { |
33345d01 LZ |
6399 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6400 | parent_ino, index); | |
4df27c4d | 6401 | } |
39279cc3 | 6402 | |
79787eaa JM |
6403 | /* Nothing to clean up yet */ |
6404 | if (ret) | |
6405 | return ret; | |
4df27c4d | 6406 | |
684572df | 6407 | ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key, |
db0a669f | 6408 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6409 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6410 | goto fail_dir_item; |
6411 | else if (ret) { | |
66642832 | 6412 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6413 | return ret; |
39279cc3 | 6414 | } |
79787eaa | 6415 | |
db0a669f | 6416 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6417 | name_len * 2); |
db0a669f | 6418 | inode_inc_iversion(&parent_inode->vfs_inode); |
5338e43a FM |
6419 | /* |
6420 | * If we are replaying a log tree, we do not want to update the mtime | |
6421 | * and ctime of the parent directory with the current time, since the | |
6422 | * log replay procedure is responsible for setting them to their correct | |
6423 | * values (the ones it had when the fsync was done). | |
6424 | */ | |
6425 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) { | |
6426 | struct timespec64 now = current_time(&parent_inode->vfs_inode); | |
6427 | ||
6428 | parent_inode->vfs_inode.i_mtime = now; | |
6429 | parent_inode->vfs_inode.i_ctime = now; | |
6430 | } | |
db0a669f | 6431 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); |
79787eaa | 6432 | if (ret) |
66642832 | 6433 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6434 | return ret; |
fe66a05a CM |
6435 | |
6436 | fail_dir_item: | |
6437 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6438 | u64 local_index; | |
6439 | int err; | |
3ee1c553 | 6440 | err = btrfs_del_root_ref(trans, key.objectid, |
0b246afa JM |
6441 | root->root_key.objectid, parent_ino, |
6442 | &local_index, name, name_len); | |
1690dd41 JT |
6443 | if (err) |
6444 | btrfs_abort_transaction(trans, err); | |
fe66a05a CM |
6445 | } else if (add_backref) { |
6446 | u64 local_index; | |
6447 | int err; | |
6448 | ||
6449 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6450 | ino, parent_ino, &local_index); | |
1690dd41 JT |
6451 | if (err) |
6452 | btrfs_abort_transaction(trans, err); | |
fe66a05a | 6453 | } |
1690dd41 JT |
6454 | |
6455 | /* Return the original error code */ | |
fe66a05a | 6456 | return ret; |
39279cc3 CM |
6457 | } |
6458 | ||
6459 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6460 | struct btrfs_inode *dir, struct dentry *dentry, |
6461 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6462 | { |
a1b075d2 JB |
6463 | int err = btrfs_add_link(trans, dir, inode, |
6464 | dentry->d_name.name, dentry->d_name.len, | |
6465 | backref, index); | |
39279cc3 CM |
6466 | if (err > 0) |
6467 | err = -EEXIST; | |
6468 | return err; | |
6469 | } | |
6470 | ||
618e21d5 | 6471 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6472 | umode_t mode, dev_t rdev) |
618e21d5 | 6473 | { |
2ff7e61e | 6474 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6475 | struct btrfs_trans_handle *trans; |
6476 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6477 | struct inode *inode = NULL; |
618e21d5 | 6478 | int err; |
618e21d5 | 6479 | u64 objectid; |
00e4e6b3 | 6480 | u64 index = 0; |
618e21d5 | 6481 | |
9ed74f2d JB |
6482 | /* |
6483 | * 2 for inode item and ref | |
6484 | * 2 for dir items | |
6485 | * 1 for xattr if selinux is on | |
6486 | */ | |
a22285a6 YZ |
6487 | trans = btrfs_start_transaction(root, 5); |
6488 | if (IS_ERR(trans)) | |
6489 | return PTR_ERR(trans); | |
1832a6d5 | 6490 | |
581bb050 LZ |
6491 | err = btrfs_find_free_ino(root, &objectid); |
6492 | if (err) | |
6493 | goto out_unlock; | |
6494 | ||
aec7477b | 6495 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6496 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6497 | mode, &index); | |
7cf96da3 TI |
6498 | if (IS_ERR(inode)) { |
6499 | err = PTR_ERR(inode); | |
32955c54 | 6500 | inode = NULL; |
618e21d5 | 6501 | goto out_unlock; |
7cf96da3 | 6502 | } |
618e21d5 | 6503 | |
ad19db71 CS |
6504 | /* |
6505 | * If the active LSM wants to access the inode during | |
6506 | * d_instantiate it needs these. Smack checks to see | |
6507 | * if the filesystem supports xattrs by looking at the | |
6508 | * ops vector. | |
6509 | */ | |
ad19db71 | 6510 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6511 | init_special_inode(inode, inode->i_mode, rdev); |
6512 | ||
6513 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6514 | if (err) |
32955c54 | 6515 | goto out_unlock; |
b0d5d10f | 6516 | |
cef415af NB |
6517 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6518 | 0, index); | |
32955c54 AV |
6519 | if (err) |
6520 | goto out_unlock; | |
6521 | ||
6522 | btrfs_update_inode(trans, root, inode); | |
6523 | d_instantiate_new(dentry, inode); | |
b0d5d10f | 6524 | |
618e21d5 | 6525 | out_unlock: |
3a45bb20 | 6526 | btrfs_end_transaction(trans); |
2ff7e61e | 6527 | btrfs_btree_balance_dirty(fs_info); |
32955c54 | 6528 | if (err && inode) { |
618e21d5 | 6529 | inode_dec_link_count(inode); |
32955c54 | 6530 | discard_new_inode(inode); |
618e21d5 | 6531 | } |
618e21d5 JB |
6532 | return err; |
6533 | } | |
6534 | ||
39279cc3 | 6535 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6536 | umode_t mode, bool excl) |
39279cc3 | 6537 | { |
2ff7e61e | 6538 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6539 | struct btrfs_trans_handle *trans; |
6540 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6541 | struct inode *inode = NULL; |
a22285a6 | 6542 | int err; |
39279cc3 | 6543 | u64 objectid; |
00e4e6b3 | 6544 | u64 index = 0; |
39279cc3 | 6545 | |
9ed74f2d JB |
6546 | /* |
6547 | * 2 for inode item and ref | |
6548 | * 2 for dir items | |
6549 | * 1 for xattr if selinux is on | |
6550 | */ | |
a22285a6 YZ |
6551 | trans = btrfs_start_transaction(root, 5); |
6552 | if (IS_ERR(trans)) | |
6553 | return PTR_ERR(trans); | |
9ed74f2d | 6554 | |
581bb050 LZ |
6555 | err = btrfs_find_free_ino(root, &objectid); |
6556 | if (err) | |
6557 | goto out_unlock; | |
6558 | ||
aec7477b | 6559 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6560 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6561 | mode, &index); | |
7cf96da3 TI |
6562 | if (IS_ERR(inode)) { |
6563 | err = PTR_ERR(inode); | |
32955c54 | 6564 | inode = NULL; |
39279cc3 | 6565 | goto out_unlock; |
7cf96da3 | 6566 | } |
ad19db71 CS |
6567 | /* |
6568 | * If the active LSM wants to access the inode during | |
6569 | * d_instantiate it needs these. Smack checks to see | |
6570 | * if the filesystem supports xattrs by looking at the | |
6571 | * ops vector. | |
6572 | */ | |
6573 | inode->i_fop = &btrfs_file_operations; | |
6574 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6575 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6576 | |
6577 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6578 | if (err) | |
32955c54 | 6579 | goto out_unlock; |
b0d5d10f CM |
6580 | |
6581 | err = btrfs_update_inode(trans, root, inode); | |
6582 | if (err) | |
32955c54 | 6583 | goto out_unlock; |
ad19db71 | 6584 | |
cef415af NB |
6585 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6586 | 0, index); | |
39279cc3 | 6587 | if (err) |
32955c54 | 6588 | goto out_unlock; |
43baa579 | 6589 | |
43baa579 | 6590 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
1e2e547a | 6591 | d_instantiate_new(dentry, inode); |
43baa579 | 6592 | |
39279cc3 | 6593 | out_unlock: |
3a45bb20 | 6594 | btrfs_end_transaction(trans); |
32955c54 | 6595 | if (err && inode) { |
39279cc3 | 6596 | inode_dec_link_count(inode); |
32955c54 | 6597 | discard_new_inode(inode); |
39279cc3 | 6598 | } |
2ff7e61e | 6599 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6600 | return err; |
6601 | } | |
6602 | ||
6603 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6604 | struct dentry *dentry) | |
6605 | { | |
271dba45 | 6606 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6607 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6608 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6609 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6610 | u64 index; |
39279cc3 CM |
6611 | int err; |
6612 | int drop_inode = 0; | |
6613 | ||
4a8be425 | 6614 | /* do not allow sys_link's with other subvols of the same device */ |
4fd786e6 | 6615 | if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid) |
3ab3564f | 6616 | return -EXDEV; |
4a8be425 | 6617 | |
f186373f | 6618 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6619 | return -EMLINK; |
4a8be425 | 6620 | |
877574e2 | 6621 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6622 | if (err) |
6623 | goto fail; | |
6624 | ||
a22285a6 | 6625 | /* |
7e6b6465 | 6626 | * 2 items for inode and inode ref |
a22285a6 | 6627 | * 2 items for dir items |
7e6b6465 | 6628 | * 1 item for parent inode |
399b0bbf | 6629 | * 1 item for orphan item deletion if O_TMPFILE |
a22285a6 | 6630 | */ |
399b0bbf | 6631 | trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6); |
a22285a6 YZ |
6632 | if (IS_ERR(trans)) { |
6633 | err = PTR_ERR(trans); | |
271dba45 | 6634 | trans = NULL; |
a22285a6 YZ |
6635 | goto fail; |
6636 | } | |
5f39d397 | 6637 | |
67de1176 MX |
6638 | /* There are several dir indexes for this inode, clear the cache. */ |
6639 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6640 | inc_nlink(inode); |
0c4d2d95 | 6641 | inode_inc_iversion(inode); |
c2050a45 | 6642 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6643 | ihold(inode); |
e9976151 | 6644 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6645 | |
cef415af NB |
6646 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6647 | 1, index); | |
5f39d397 | 6648 | |
a5719521 | 6649 | if (err) { |
54aa1f4d | 6650 | drop_inode = 1; |
a5719521 | 6651 | } else { |
10d9f309 | 6652 | struct dentry *parent = dentry->d_parent; |
d4682ba0 FM |
6653 | int ret; |
6654 | ||
a5719521 | 6655 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6656 | if (err) |
6657 | goto fail; | |
ef3b9af5 FM |
6658 | if (inode->i_nlink == 1) { |
6659 | /* | |
6660 | * If new hard link count is 1, it's a file created | |
6661 | * with open(2) O_TMPFILE flag. | |
6662 | */ | |
3d6ae7bb | 6663 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6664 | if (err) |
6665 | goto fail; | |
6666 | } | |
08c422c2 | 6667 | d_instantiate(dentry, inode); |
d4682ba0 FM |
6668 | ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent, |
6669 | true, NULL); | |
6670 | if (ret == BTRFS_NEED_TRANS_COMMIT) { | |
6671 | err = btrfs_commit_transaction(trans); | |
6672 | trans = NULL; | |
6673 | } | |
a5719521 | 6674 | } |
39279cc3 | 6675 | |
1832a6d5 | 6676 | fail: |
271dba45 | 6677 | if (trans) |
3a45bb20 | 6678 | btrfs_end_transaction(trans); |
39279cc3 CM |
6679 | if (drop_inode) { |
6680 | inode_dec_link_count(inode); | |
6681 | iput(inode); | |
6682 | } | |
2ff7e61e | 6683 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6684 | return err; |
6685 | } | |
6686 | ||
18bb1db3 | 6687 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6688 | { |
2ff7e61e | 6689 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6690 | struct inode *inode = NULL; |
39279cc3 CM |
6691 | struct btrfs_trans_handle *trans; |
6692 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6693 | int err = 0; | |
b9d86667 | 6694 | u64 objectid = 0; |
00e4e6b3 | 6695 | u64 index = 0; |
39279cc3 | 6696 | |
9ed74f2d JB |
6697 | /* |
6698 | * 2 items for inode and ref | |
6699 | * 2 items for dir items | |
6700 | * 1 for xattr if selinux is on | |
6701 | */ | |
a22285a6 YZ |
6702 | trans = btrfs_start_transaction(root, 5); |
6703 | if (IS_ERR(trans)) | |
6704 | return PTR_ERR(trans); | |
39279cc3 | 6705 | |
581bb050 LZ |
6706 | err = btrfs_find_free_ino(root, &objectid); |
6707 | if (err) | |
6708 | goto out_fail; | |
6709 | ||
aec7477b | 6710 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6711 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6712 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6713 | if (IS_ERR(inode)) { |
6714 | err = PTR_ERR(inode); | |
32955c54 | 6715 | inode = NULL; |
39279cc3 CM |
6716 | goto out_fail; |
6717 | } | |
5f39d397 | 6718 | |
b0d5d10f CM |
6719 | /* these must be set before we unlock the inode */ |
6720 | inode->i_op = &btrfs_dir_inode_operations; | |
6721 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6722 | |
2a7dba39 | 6723 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6724 | if (err) |
32955c54 | 6725 | goto out_fail; |
39279cc3 | 6726 | |
6ef06d27 | 6727 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6728 | err = btrfs_update_inode(trans, root, inode); |
6729 | if (err) | |
32955c54 | 6730 | goto out_fail; |
5f39d397 | 6731 | |
db0a669f NB |
6732 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6733 | dentry->d_name.name, | |
6734 | dentry->d_name.len, 0, index); | |
39279cc3 | 6735 | if (err) |
32955c54 | 6736 | goto out_fail; |
5f39d397 | 6737 | |
1e2e547a | 6738 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
6739 | |
6740 | out_fail: | |
3a45bb20 | 6741 | btrfs_end_transaction(trans); |
32955c54 | 6742 | if (err && inode) { |
c7cfb8a5 | 6743 | inode_dec_link_count(inode); |
32955c54 | 6744 | discard_new_inode(inode); |
c7cfb8a5 | 6745 | } |
2ff7e61e | 6746 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6747 | return err; |
6748 | } | |
6749 | ||
c8b97818 | 6750 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6751 | struct page *page, |
c8b97818 CM |
6752 | size_t pg_offset, u64 extent_offset, |
6753 | struct btrfs_file_extent_item *item) | |
6754 | { | |
6755 | int ret; | |
6756 | struct extent_buffer *leaf = path->nodes[0]; | |
6757 | char *tmp; | |
6758 | size_t max_size; | |
6759 | unsigned long inline_size; | |
6760 | unsigned long ptr; | |
261507a0 | 6761 | int compress_type; |
c8b97818 CM |
6762 | |
6763 | WARN_ON(pg_offset != 0); | |
261507a0 | 6764 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6765 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6766 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6767 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6768 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6769 | if (!tmp) |
6770 | return -ENOMEM; | |
c8b97818 CM |
6771 | ptr = btrfs_file_extent_inline_start(item); |
6772 | ||
6773 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6774 | ||
09cbfeaf | 6775 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6776 | ret = btrfs_decompress(compress_type, tmp, page, |
6777 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6778 | |
6779 | /* | |
6780 | * decompression code contains a memset to fill in any space between the end | |
6781 | * of the uncompressed data and the end of max_size in case the decompressed | |
6782 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6783 | * the end of an inline extent and the beginning of the next block, so we | |
6784 | * cover that region here. | |
6785 | */ | |
6786 | ||
6787 | if (max_size + pg_offset < PAGE_SIZE) { | |
6788 | char *map = kmap(page); | |
6789 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6790 | kunmap(page); | |
6791 | } | |
c8b97818 | 6792 | kfree(tmp); |
166ae5a4 | 6793 | return ret; |
c8b97818 CM |
6794 | } |
6795 | ||
d352ac68 CM |
6796 | /* |
6797 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6798 | * the ugly parts come from merging extents from the disk with the in-ram |
6799 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6800 | * where the in-ram extents might be locked pending data=ordered completion. |
6801 | * | |
6802 | * This also copies inline extents directly into the page. | |
6803 | */ | |
fc4f21b1 | 6804 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
de2c6615 LB |
6805 | struct page *page, |
6806 | size_t pg_offset, u64 start, u64 len, | |
6807 | int create) | |
a52d9a80 | 6808 | { |
3ffbd68c | 6809 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a52d9a80 CM |
6810 | int ret; |
6811 | int err = 0; | |
a52d9a80 CM |
6812 | u64 extent_start = 0; |
6813 | u64 extent_end = 0; | |
fc4f21b1 | 6814 | u64 objectid = btrfs_ino(inode); |
7e74e235 | 6815 | int extent_type = -1; |
f421950f | 6816 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6817 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6818 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6819 | struct extent_buffer *leaf; |
6820 | struct btrfs_key found_key; | |
a52d9a80 | 6821 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6822 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6823 | struct extent_io_tree *io_tree = &inode->io_tree; | |
7ffbb598 | 6824 | const bool new_inline = !page || create; |
a52d9a80 | 6825 | |
890871be | 6826 | read_lock(&em_tree->lock); |
d1310b2e | 6827 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6828 | if (em) |
0b246afa | 6829 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6830 | read_unlock(&em_tree->lock); |
d1310b2e | 6831 | |
a52d9a80 | 6832 | if (em) { |
e1c4b745 CM |
6833 | if (em->start > start || em->start + em->len <= start) |
6834 | free_extent_map(em); | |
6835 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6836 | free_extent_map(em); |
6837 | else | |
6838 | goto out; | |
a52d9a80 | 6839 | } |
172ddd60 | 6840 | em = alloc_extent_map(); |
a52d9a80 | 6841 | if (!em) { |
d1310b2e CM |
6842 | err = -ENOMEM; |
6843 | goto out; | |
a52d9a80 | 6844 | } |
0b246afa | 6845 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6846 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6847 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6848 | em->len = (u64)-1; |
c8b97818 | 6849 | em->block_len = (u64)-1; |
f421950f | 6850 | |
bee6ec82 | 6851 | path = btrfs_alloc_path(); |
f421950f | 6852 | if (!path) { |
bee6ec82 LB |
6853 | err = -ENOMEM; |
6854 | goto out; | |
f421950f CM |
6855 | } |
6856 | ||
bee6ec82 LB |
6857 | /* Chances are we'll be called again, so go ahead and do readahead */ |
6858 | path->reada = READA_FORWARD; | |
6859 | ||
e49aabd9 LB |
6860 | /* |
6861 | * Unless we're going to uncompress the inline extent, no sleep would | |
6862 | * happen. | |
6863 | */ | |
6864 | path->leave_spinning = 1; | |
6865 | ||
5c9a702e | 6866 | ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0); |
a52d9a80 CM |
6867 | if (ret < 0) { |
6868 | err = ret; | |
6869 | goto out; | |
b8eeab7f | 6870 | } else if (ret > 0) { |
a52d9a80 CM |
6871 | if (path->slots[0] == 0) |
6872 | goto not_found; | |
6873 | path->slots[0]--; | |
6874 | } | |
6875 | ||
5f39d397 CM |
6876 | leaf = path->nodes[0]; |
6877 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6878 | struct btrfs_file_extent_item); |
5f39d397 | 6879 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5f39d397 | 6880 | if (found_key.objectid != objectid || |
694c12ed | 6881 | found_key.type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6882 | /* |
6883 | * If we backup past the first extent we want to move forward | |
6884 | * and see if there is an extent in front of us, otherwise we'll | |
6885 | * say there is a hole for our whole search range which can | |
6886 | * cause problems. | |
6887 | */ | |
6888 | extent_end = start; | |
6889 | goto next; | |
a52d9a80 CM |
6890 | } |
6891 | ||
694c12ed | 6892 | extent_type = btrfs_file_extent_type(leaf, item); |
5f39d397 | 6893 | extent_start = found_key.offset; |
694c12ed NB |
6894 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6895 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6bf9e4bd QW |
6896 | /* Only regular file could have regular/prealloc extent */ |
6897 | if (!S_ISREG(inode->vfs_inode.i_mode)) { | |
6898 | ret = -EUCLEAN; | |
6899 | btrfs_crit(fs_info, | |
6900 | "regular/prealloc extent found for non-regular inode %llu", | |
6901 | btrfs_ino(inode)); | |
6902 | goto out; | |
6903 | } | |
a52d9a80 | 6904 | extent_end = extent_start + |
db94535d | 6905 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
6906 | |
6907 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
6908 | extent_start); | |
694c12ed | 6909 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
9036c102 | 6910 | size_t size; |
e41ca589 QW |
6911 | |
6912 | size = btrfs_file_extent_ram_bytes(leaf, item); | |
da17066c | 6913 | extent_end = ALIGN(extent_start + size, |
0b246afa | 6914 | fs_info->sectorsize); |
09ed2f16 LB |
6915 | |
6916 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
6917 | path->slots[0], | |
6918 | extent_start); | |
9036c102 | 6919 | } |
25a50341 | 6920 | next: |
9036c102 YZ |
6921 | if (start >= extent_end) { |
6922 | path->slots[0]++; | |
6923 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6924 | ret = btrfs_next_leaf(root, path); | |
6925 | if (ret < 0) { | |
6926 | err = ret; | |
6927 | goto out; | |
b8eeab7f | 6928 | } else if (ret > 0) { |
9036c102 | 6929 | goto not_found; |
b8eeab7f | 6930 | } |
9036c102 | 6931 | leaf = path->nodes[0]; |
a52d9a80 | 6932 | } |
9036c102 YZ |
6933 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6934 | if (found_key.objectid != objectid || | |
6935 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6936 | goto not_found; | |
6937 | if (start + len <= found_key.offset) | |
6938 | goto not_found; | |
e2eca69d WS |
6939 | if (start > found_key.offset) |
6940 | goto next; | |
02a033df NB |
6941 | |
6942 | /* New extent overlaps with existing one */ | |
9036c102 | 6943 | em->start = start; |
70c8a91c | 6944 | em->orig_start = start; |
9036c102 | 6945 | em->len = found_key.offset - start; |
02a033df NB |
6946 | em->block_start = EXTENT_MAP_HOLE; |
6947 | goto insert; | |
9036c102 YZ |
6948 | } |
6949 | ||
fc4f21b1 | 6950 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 6951 | new_inline, em); |
7ffbb598 | 6952 | |
694c12ed NB |
6953 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
6954 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6955 | goto insert; |
694c12ed | 6956 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 6957 | unsigned long ptr; |
a52d9a80 | 6958 | char *map; |
3326d1b0 CM |
6959 | size_t size; |
6960 | size_t extent_offset; | |
6961 | size_t copy_size; | |
a52d9a80 | 6962 | |
7ffbb598 | 6963 | if (new_inline) |
689f9346 | 6964 | goto out; |
5f39d397 | 6965 | |
e41ca589 | 6966 | size = btrfs_file_extent_ram_bytes(leaf, item); |
9036c102 | 6967 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6968 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6969 | size - extent_offset); | |
3326d1b0 | 6970 | em->start = extent_start + extent_offset; |
0b246afa | 6971 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6972 | em->orig_block_len = em->len; |
70c8a91c | 6973 | em->orig_start = em->start; |
689f9346 | 6974 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
e49aabd9 LB |
6975 | |
6976 | btrfs_set_path_blocking(path); | |
bf46f52d | 6977 | if (!PageUptodate(page)) { |
261507a0 LZ |
6978 | if (btrfs_file_extent_compression(leaf, item) != |
6979 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6980 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6981 | extent_offset, item); |
166ae5a4 ZB |
6982 | if (ret) { |
6983 | err = ret; | |
6984 | goto out; | |
6985 | } | |
c8b97818 CM |
6986 | } else { |
6987 | map = kmap(page); | |
6988 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6989 | copy_size); | |
09cbfeaf | 6990 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6991 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6992 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6993 | copy_size); |
6994 | } | |
c8b97818 CM |
6995 | kunmap(page); |
6996 | } | |
179e29e4 | 6997 | flush_dcache_page(page); |
a52d9a80 | 6998 | } |
d1310b2e | 6999 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7000 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7001 | goto insert; |
a52d9a80 CM |
7002 | } |
7003 | not_found: | |
7004 | em->start = start; | |
70c8a91c | 7005 | em->orig_start = start; |
d1310b2e | 7006 | em->len = len; |
5f39d397 | 7007 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 | 7008 | insert: |
b3b4aa74 | 7009 | btrfs_release_path(path); |
d1310b2e | 7010 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7011 | btrfs_err(fs_info, |
5d163e0e JM |
7012 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7013 | em->start, em->len, start, len); | |
a52d9a80 CM |
7014 | err = -EIO; |
7015 | goto out; | |
7016 | } | |
d1310b2e CM |
7017 | |
7018 | err = 0; | |
890871be | 7019 | write_lock(&em_tree->lock); |
f46b24c9 | 7020 | err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len); |
890871be | 7021 | write_unlock(&em_tree->lock); |
a52d9a80 | 7022 | out: |
c6414280 | 7023 | btrfs_free_path(path); |
1abe9b8a | 7024 | |
fc4f21b1 | 7025 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7026 | |
a52d9a80 CM |
7027 | if (err) { |
7028 | free_extent_map(em); | |
a52d9a80 CM |
7029 | return ERR_PTR(err); |
7030 | } | |
79787eaa | 7031 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7032 | return em; |
7033 | } | |
7034 | ||
fc4f21b1 | 7035 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
4ab47a8d | 7036 | u64 start, u64 len) |
ec29ed5b CM |
7037 | { |
7038 | struct extent_map *em; | |
7039 | struct extent_map *hole_em = NULL; | |
f3714ef4 | 7040 | u64 delalloc_start = start; |
ec29ed5b | 7041 | u64 end; |
f3714ef4 NB |
7042 | u64 delalloc_len; |
7043 | u64 delalloc_end; | |
ec29ed5b CM |
7044 | int err = 0; |
7045 | ||
4ab47a8d | 7046 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
ec29ed5b CM |
7047 | if (IS_ERR(em)) |
7048 | return em; | |
9986277e DC |
7049 | /* |
7050 | * If our em maps to: | |
7051 | * - a hole or | |
7052 | * - a pre-alloc extent, | |
7053 | * there might actually be delalloc bytes behind it. | |
7054 | */ | |
7055 | if (em->block_start != EXTENT_MAP_HOLE && | |
7056 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7057 | return em; | |
7058 | else | |
7059 | hole_em = em; | |
ec29ed5b CM |
7060 | |
7061 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7062 | end = start + len; | |
7063 | if (end < start) | |
7064 | end = (u64)-1; | |
7065 | else | |
7066 | end -= 1; | |
7067 | ||
7068 | em = NULL; | |
7069 | ||
7070 | /* ok, we didn't find anything, lets look for delalloc */ | |
f3714ef4 | 7071 | delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start, |
ec29ed5b | 7072 | end, len, EXTENT_DELALLOC, 1); |
f3714ef4 NB |
7073 | delalloc_end = delalloc_start + delalloc_len; |
7074 | if (delalloc_end < delalloc_start) | |
7075 | delalloc_end = (u64)-1; | |
ec29ed5b CM |
7076 | |
7077 | /* | |
f3714ef4 NB |
7078 | * We didn't find anything useful, return the original results from |
7079 | * get_extent() | |
ec29ed5b | 7080 | */ |
f3714ef4 | 7081 | if (delalloc_start > end || delalloc_end <= start) { |
ec29ed5b CM |
7082 | em = hole_em; |
7083 | hole_em = NULL; | |
7084 | goto out; | |
7085 | } | |
7086 | ||
f3714ef4 NB |
7087 | /* |
7088 | * Adjust the delalloc_start to make sure it doesn't go backwards from | |
7089 | * the start they passed in | |
ec29ed5b | 7090 | */ |
f3714ef4 NB |
7091 | delalloc_start = max(start, delalloc_start); |
7092 | delalloc_len = delalloc_end - delalloc_start; | |
ec29ed5b | 7093 | |
f3714ef4 NB |
7094 | if (delalloc_len > 0) { |
7095 | u64 hole_start; | |
02950af4 | 7096 | u64 hole_len; |
f3714ef4 | 7097 | const u64 hole_end = extent_map_end(hole_em); |
ec29ed5b | 7098 | |
172ddd60 | 7099 | em = alloc_extent_map(); |
ec29ed5b CM |
7100 | if (!em) { |
7101 | err = -ENOMEM; | |
7102 | goto out; | |
7103 | } | |
f3714ef4 NB |
7104 | em->bdev = NULL; |
7105 | ||
7106 | ASSERT(hole_em); | |
ec29ed5b | 7107 | /* |
f3714ef4 NB |
7108 | * When btrfs_get_extent can't find anything it returns one |
7109 | * huge hole | |
ec29ed5b | 7110 | * |
f3714ef4 NB |
7111 | * Make sure what it found really fits our range, and adjust to |
7112 | * make sure it is based on the start from the caller | |
ec29ed5b | 7113 | */ |
f3714ef4 NB |
7114 | if (hole_end <= start || hole_em->start > end) { |
7115 | free_extent_map(hole_em); | |
7116 | hole_em = NULL; | |
7117 | } else { | |
7118 | hole_start = max(hole_em->start, start); | |
7119 | hole_len = hole_end - hole_start; | |
ec29ed5b | 7120 | } |
f3714ef4 NB |
7121 | |
7122 | if (hole_em && delalloc_start > hole_start) { | |
7123 | /* | |
7124 | * Our hole starts before our delalloc, so we have to | |
7125 | * return just the parts of the hole that go until the | |
7126 | * delalloc starts | |
ec29ed5b | 7127 | */ |
f3714ef4 | 7128 | em->len = min(hole_len, delalloc_start - hole_start); |
ec29ed5b CM |
7129 | em->start = hole_start; |
7130 | em->orig_start = hole_start; | |
7131 | /* | |
f3714ef4 NB |
7132 | * Don't adjust block start at all, it is fixed at |
7133 | * EXTENT_MAP_HOLE | |
ec29ed5b CM |
7134 | */ |
7135 | em->block_start = hole_em->block_start; | |
7136 | em->block_len = hole_len; | |
f9e4fb53 LB |
7137 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7138 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b | 7139 | } else { |
f3714ef4 NB |
7140 | /* |
7141 | * Hole is out of passed range or it starts after | |
7142 | * delalloc range | |
7143 | */ | |
7144 | em->start = delalloc_start; | |
7145 | em->len = delalloc_len; | |
7146 | em->orig_start = delalloc_start; | |
ec29ed5b | 7147 | em->block_start = EXTENT_MAP_DELALLOC; |
f3714ef4 | 7148 | em->block_len = delalloc_len; |
ec29ed5b | 7149 | } |
bf8d32b9 | 7150 | } else { |
ec29ed5b CM |
7151 | return hole_em; |
7152 | } | |
7153 | out: | |
7154 | ||
7155 | free_extent_map(hole_em); | |
7156 | if (err) { | |
7157 | free_extent_map(em); | |
7158 | return ERR_PTR(err); | |
7159 | } | |
7160 | return em; | |
7161 | } | |
7162 | ||
5f9a8a51 FM |
7163 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7164 | const u64 start, | |
7165 | const u64 len, | |
7166 | const u64 orig_start, | |
7167 | const u64 block_start, | |
7168 | const u64 block_len, | |
7169 | const u64 orig_block_len, | |
7170 | const u64 ram_bytes, | |
7171 | const int type) | |
7172 | { | |
7173 | struct extent_map *em = NULL; | |
7174 | int ret; | |
7175 | ||
5f9a8a51 | 7176 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7177 | em = create_io_em(inode, start, len, orig_start, |
7178 | block_start, block_len, orig_block_len, | |
7179 | ram_bytes, | |
7180 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7181 | type); | |
5f9a8a51 FM |
7182 | if (IS_ERR(em)) |
7183 | goto out; | |
7184 | } | |
7185 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7186 | len, block_len, type); | |
7187 | if (ret) { | |
7188 | if (em) { | |
7189 | free_extent_map(em); | |
dcdbc059 | 7190 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7191 | start + len - 1, 0); |
7192 | } | |
7193 | em = ERR_PTR(ret); | |
7194 | } | |
7195 | out: | |
5f9a8a51 FM |
7196 | |
7197 | return em; | |
7198 | } | |
7199 | ||
4b46fce2 JB |
7200 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7201 | u64 start, u64 len) | |
7202 | { | |
0b246afa | 7203 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7204 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7205 | struct extent_map *em; |
4b46fce2 JB |
7206 | struct btrfs_key ins; |
7207 | u64 alloc_hint; | |
7208 | int ret; | |
4b46fce2 | 7209 | |
4b46fce2 | 7210 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7211 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7212 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7213 | if (ret) |
7214 | return ERR_PTR(ret); | |
4b46fce2 | 7215 | |
5f9a8a51 FM |
7216 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7217 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7218 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7219 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7220 | if (IS_ERR(em)) |
2ff7e61e JM |
7221 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7222 | ins.offset, 1); | |
de0ee0ed | 7223 | |
4b46fce2 JB |
7224 | return em; |
7225 | } | |
7226 | ||
46bfbb5c CM |
7227 | /* |
7228 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7229 | * block must be cow'd | |
7230 | */ | |
00361589 | 7231 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7232 | u64 *orig_start, u64 *orig_block_len, |
7233 | u64 *ram_bytes) | |
46bfbb5c | 7234 | { |
2ff7e61e | 7235 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7236 | struct btrfs_path *path; |
7237 | int ret; | |
7238 | struct extent_buffer *leaf; | |
7239 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7240 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7241 | struct btrfs_file_extent_item *fi; |
7242 | struct btrfs_key key; | |
7243 | u64 disk_bytenr; | |
7244 | u64 backref_offset; | |
7245 | u64 extent_end; | |
7246 | u64 num_bytes; | |
7247 | int slot; | |
7248 | int found_type; | |
7ee9e440 | 7249 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7250 | |
46bfbb5c CM |
7251 | path = btrfs_alloc_path(); |
7252 | if (!path) | |
7253 | return -ENOMEM; | |
7254 | ||
f85b7379 DS |
7255 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7256 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7257 | if (ret < 0) |
7258 | goto out; | |
7259 | ||
7260 | slot = path->slots[0]; | |
7261 | if (ret == 1) { | |
7262 | if (slot == 0) { | |
7263 | /* can't find the item, must cow */ | |
7264 | ret = 0; | |
7265 | goto out; | |
7266 | } | |
7267 | slot--; | |
7268 | } | |
7269 | ret = 0; | |
7270 | leaf = path->nodes[0]; | |
7271 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7272 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7273 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7274 | /* not our file or wrong item type, must cow */ | |
7275 | goto out; | |
7276 | } | |
7277 | ||
7278 | if (key.offset > offset) { | |
7279 | /* Wrong offset, must cow */ | |
7280 | goto out; | |
7281 | } | |
7282 | ||
7283 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7284 | found_type = btrfs_file_extent_type(leaf, fi); | |
7285 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7286 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7287 | /* not a regular extent, must cow */ | |
7288 | goto out; | |
7289 | } | |
7ee9e440 JB |
7290 | |
7291 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7292 | goto out; | |
7293 | ||
e77751aa MX |
7294 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7295 | if (extent_end <= offset) | |
7296 | goto out; | |
7297 | ||
46bfbb5c | 7298 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7299 | if (disk_bytenr == 0) |
7300 | goto out; | |
7301 | ||
7302 | if (btrfs_file_extent_compression(leaf, fi) || | |
7303 | btrfs_file_extent_encryption(leaf, fi) || | |
7304 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7305 | goto out; | |
7306 | ||
78d4295b EL |
7307 | /* |
7308 | * Do the same check as in btrfs_cross_ref_exist but without the | |
7309 | * unnecessary search. | |
7310 | */ | |
7311 | if (btrfs_file_extent_generation(leaf, fi) <= | |
7312 | btrfs_root_last_snapshot(&root->root_item)) | |
7313 | goto out; | |
7314 | ||
46bfbb5c CM |
7315 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7316 | ||
7ee9e440 JB |
7317 | if (orig_start) { |
7318 | *orig_start = key.offset - backref_offset; | |
7319 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7320 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7321 | } | |
eb384b55 | 7322 | |
2ff7e61e | 7323 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7324 | goto out; |
7b2b7085 MX |
7325 | |
7326 | num_bytes = min(offset + *len, extent_end) - offset; | |
7327 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7328 | u64 range_end; | |
7329 | ||
da17066c JM |
7330 | range_end = round_up(offset + num_bytes, |
7331 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7332 | ret = test_range_bit(io_tree, offset, range_end, |
7333 | EXTENT_DELALLOC, 0, NULL); | |
7334 | if (ret) { | |
7335 | ret = -EAGAIN; | |
7336 | goto out; | |
7337 | } | |
7338 | } | |
7339 | ||
1bda19eb | 7340 | btrfs_release_path(path); |
46bfbb5c CM |
7341 | |
7342 | /* | |
7343 | * look for other files referencing this extent, if we | |
7344 | * find any we must cow | |
7345 | */ | |
00361589 | 7346 | |
e4c3b2dc | 7347 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7348 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7349 | if (ret) { |
7350 | ret = 0; | |
7351 | goto out; | |
7352 | } | |
46bfbb5c CM |
7353 | |
7354 | /* | |
7355 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7356 | * in this extent we are about to write. If there | |
7357 | * are any csums in that range we have to cow in order | |
7358 | * to keep the csums correct | |
7359 | */ | |
7360 | disk_bytenr += backref_offset; | |
7361 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7362 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7363 | goto out; | |
46bfbb5c CM |
7364 | /* |
7365 | * all of the above have passed, it is safe to overwrite this extent | |
7366 | * without cow | |
7367 | */ | |
eb384b55 | 7368 | *len = num_bytes; |
46bfbb5c CM |
7369 | ret = 1; |
7370 | out: | |
7371 | btrfs_free_path(path); | |
7372 | return ret; | |
7373 | } | |
7374 | ||
eb838e73 JB |
7375 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7376 | struct extent_state **cached_state, int writing) | |
7377 | { | |
7378 | struct btrfs_ordered_extent *ordered; | |
7379 | int ret = 0; | |
7380 | ||
7381 | while (1) { | |
7382 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7383 | cached_state); |
eb838e73 JB |
7384 | /* |
7385 | * We're concerned with the entire range that we're going to be | |
01327610 | 7386 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7387 | * extents in this range. |
7388 | */ | |
a776c6fa | 7389 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7390 | lockend - lockstart + 1); |
7391 | ||
7392 | /* | |
7393 | * We need to make sure there are no buffered pages in this | |
7394 | * range either, we could have raced between the invalidate in | |
7395 | * generic_file_direct_write and locking the extent. The | |
7396 | * invalidate needs to happen so that reads after a write do not | |
7397 | * get stale data. | |
7398 | */ | |
fc4adbff | 7399 | if (!ordered && |
051c98eb DS |
7400 | (!writing || !filemap_range_has_page(inode->i_mapping, |
7401 | lockstart, lockend))) | |
eb838e73 JB |
7402 | break; |
7403 | ||
7404 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
e43bbe5e | 7405 | cached_state); |
eb838e73 JB |
7406 | |
7407 | if (ordered) { | |
ade77029 FM |
7408 | /* |
7409 | * If we are doing a DIO read and the ordered extent we | |
7410 | * found is for a buffered write, we can not wait for it | |
7411 | * to complete and retry, because if we do so we can | |
7412 | * deadlock with concurrent buffered writes on page | |
7413 | * locks. This happens only if our DIO read covers more | |
7414 | * than one extent map, if at this point has already | |
7415 | * created an ordered extent for a previous extent map | |
7416 | * and locked its range in the inode's io tree, and a | |
7417 | * concurrent write against that previous extent map's | |
7418 | * range and this range started (we unlock the ranges | |
7419 | * in the io tree only when the bios complete and | |
7420 | * buffered writes always lock pages before attempting | |
7421 | * to lock range in the io tree). | |
7422 | */ | |
7423 | if (writing || | |
7424 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7425 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7426 | else | |
7427 | ret = -ENOTBLK; | |
eb838e73 JB |
7428 | btrfs_put_ordered_extent(ordered); |
7429 | } else { | |
eb838e73 | 7430 | /* |
b850ae14 FM |
7431 | * We could trigger writeback for this range (and wait |
7432 | * for it to complete) and then invalidate the pages for | |
7433 | * this range (through invalidate_inode_pages2_range()), | |
7434 | * but that can lead us to a deadlock with a concurrent | |
7435 | * call to readpages() (a buffered read or a defrag call | |
7436 | * triggered a readahead) on a page lock due to an | |
7437 | * ordered dio extent we created before but did not have | |
7438 | * yet a corresponding bio submitted (whence it can not | |
7439 | * complete), which makes readpages() wait for that | |
7440 | * ordered extent to complete while holding a lock on | |
7441 | * that page. | |
eb838e73 | 7442 | */ |
b850ae14 | 7443 | ret = -ENOTBLK; |
eb838e73 JB |
7444 | } |
7445 | ||
ade77029 FM |
7446 | if (ret) |
7447 | break; | |
7448 | ||
eb838e73 JB |
7449 | cond_resched(); |
7450 | } | |
7451 | ||
7452 | return ret; | |
7453 | } | |
7454 | ||
6f9994db LB |
7455 | /* The callers of this must take lock_extent() */ |
7456 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7457 | u64 orig_start, u64 block_start, | |
7458 | u64 block_len, u64 orig_block_len, | |
7459 | u64 ram_bytes, int compress_type, | |
7460 | int type) | |
69ffb543 JB |
7461 | { |
7462 | struct extent_map_tree *em_tree; | |
7463 | struct extent_map *em; | |
7464 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7465 | int ret; | |
7466 | ||
6f9994db LB |
7467 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7468 | type == BTRFS_ORDERED_COMPRESSED || | |
7469 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7470 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7471 | |
69ffb543 JB |
7472 | em_tree = &BTRFS_I(inode)->extent_tree; |
7473 | em = alloc_extent_map(); | |
7474 | if (!em) | |
7475 | return ERR_PTR(-ENOMEM); | |
7476 | ||
7477 | em->start = start; | |
7478 | em->orig_start = orig_start; | |
7479 | em->len = len; | |
7480 | em->block_len = block_len; | |
7481 | em->block_start = block_start; | |
7482 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7483 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7484 | em->ram_bytes = ram_bytes; |
70c8a91c | 7485 | em->generation = -1; |
69ffb543 | 7486 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7487 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7488 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7489 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7490 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7491 | em->compress_type = compress_type; | |
7492 | } | |
69ffb543 JB |
7493 | |
7494 | do { | |
dcdbc059 | 7495 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7496 | em->start + em->len - 1, 0); |
7497 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7498 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7499 | write_unlock(&em_tree->lock); |
6f9994db LB |
7500 | /* |
7501 | * The caller has taken lock_extent(), who could race with us | |
7502 | * to add em? | |
7503 | */ | |
69ffb543 JB |
7504 | } while (ret == -EEXIST); |
7505 | ||
7506 | if (ret) { | |
7507 | free_extent_map(em); | |
7508 | return ERR_PTR(ret); | |
7509 | } | |
7510 | ||
6f9994db | 7511 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7512 | return em; |
7513 | } | |
7514 | ||
1c8d0175 NB |
7515 | |
7516 | static int btrfs_get_blocks_direct_read(struct extent_map *em, | |
7517 | struct buffer_head *bh_result, | |
7518 | struct inode *inode, | |
7519 | u64 start, u64 len) | |
7520 | { | |
7521 | if (em->block_start == EXTENT_MAP_HOLE || | |
7522 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7523 | return -ENOENT; | |
7524 | ||
7525 | len = min(len, em->len - (start - em->start)); | |
7526 | ||
7527 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7528 | inode->i_blkbits; | |
7529 | bh_result->b_size = len; | |
7530 | bh_result->b_bdev = em->bdev; | |
7531 | set_buffer_mapped(bh_result); | |
7532 | ||
7533 | return 0; | |
7534 | } | |
7535 | ||
c5794e51 NB |
7536 | static int btrfs_get_blocks_direct_write(struct extent_map **map, |
7537 | struct buffer_head *bh_result, | |
7538 | struct inode *inode, | |
7539 | struct btrfs_dio_data *dio_data, | |
7540 | u64 start, u64 len) | |
7541 | { | |
7542 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
7543 | struct extent_map *em = *map; | |
7544 | int ret = 0; | |
7545 | ||
7546 | /* | |
7547 | * We don't allocate a new extent in the following cases | |
7548 | * | |
7549 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7550 | * existing extent. | |
7551 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7552 | * just use the extent. | |
7553 | * | |
7554 | */ | |
7555 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7556 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7557 | em->block_start != EXTENT_MAP_HOLE)) { | |
7558 | int type; | |
7559 | u64 block_start, orig_start, orig_block_len, ram_bytes; | |
7560 | ||
7561 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7562 | type = BTRFS_ORDERED_PREALLOC; | |
7563 | else | |
7564 | type = BTRFS_ORDERED_NOCOW; | |
7565 | len = min(len, em->len - (start - em->start)); | |
7566 | block_start = em->block_start + (start - em->start); | |
7567 | ||
7568 | if (can_nocow_extent(inode, start, &len, &orig_start, | |
7569 | &orig_block_len, &ram_bytes) == 1 && | |
7570 | btrfs_inc_nocow_writers(fs_info, block_start)) { | |
7571 | struct extent_map *em2; | |
7572 | ||
7573 | em2 = btrfs_create_dio_extent(inode, start, len, | |
7574 | orig_start, block_start, | |
7575 | len, orig_block_len, | |
7576 | ram_bytes, type); | |
7577 | btrfs_dec_nocow_writers(fs_info, block_start); | |
7578 | if (type == BTRFS_ORDERED_PREALLOC) { | |
7579 | free_extent_map(em); | |
7580 | *map = em = em2; | |
7581 | } | |
7582 | ||
7583 | if (em2 && IS_ERR(em2)) { | |
7584 | ret = PTR_ERR(em2); | |
7585 | goto out; | |
7586 | } | |
7587 | /* | |
7588 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7589 | * use the existing or preallocated extent, so does not | |
7590 | * need to adjust btrfs_space_info's bytes_may_use. | |
7591 | */ | |
7592 | btrfs_free_reserved_data_space_noquota(inode, start, | |
7593 | len); | |
7594 | goto skip_cow; | |
7595 | } | |
7596 | } | |
7597 | ||
7598 | /* this will cow the extent */ | |
7599 | len = bh_result->b_size; | |
7600 | free_extent_map(em); | |
7601 | *map = em = btrfs_new_extent_direct(inode, start, len); | |
7602 | if (IS_ERR(em)) { | |
7603 | ret = PTR_ERR(em); | |
7604 | goto out; | |
7605 | } | |
7606 | ||
7607 | len = min(len, em->len - (start - em->start)); | |
7608 | ||
7609 | skip_cow: | |
7610 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
7611 | inode->i_blkbits; | |
7612 | bh_result->b_size = len; | |
7613 | bh_result->b_bdev = em->bdev; | |
7614 | set_buffer_mapped(bh_result); | |
7615 | ||
7616 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7617 | set_buffer_new(bh_result); | |
7618 | ||
7619 | /* | |
7620 | * Need to update the i_size under the extent lock so buffered | |
7621 | * readers will get the updated i_size when we unlock. | |
7622 | */ | |
7623 | if (!dio_data->overwrite && start + len > i_size_read(inode)) | |
7624 | i_size_write(inode, start + len); | |
7625 | ||
7626 | WARN_ON(dio_data->reserve < len); | |
7627 | dio_data->reserve -= len; | |
7628 | dio_data->unsubmitted_oe_range_end = start + len; | |
7629 | current->journal_info = dio_data; | |
7630 | out: | |
7631 | return ret; | |
7632 | } | |
7633 | ||
4b46fce2 JB |
7634 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7635 | struct buffer_head *bh_result, int create) | |
7636 | { | |
0b246afa | 7637 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7638 | struct extent_map *em; |
eb838e73 | 7639 | struct extent_state *cached_state = NULL; |
50745b0a | 7640 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7641 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7642 | u64 lockstart, lockend; |
4b46fce2 | 7643 | u64 len = bh_result->b_size; |
eb838e73 | 7644 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7645 | int ret = 0; |
eb838e73 | 7646 | |
172a5049 | 7647 | if (create) |
3266789f | 7648 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7649 | else |
0b246afa | 7650 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7651 | |
c329861d JB |
7652 | lockstart = start; |
7653 | lockend = start + len - 1; | |
7654 | ||
e1cbbfa5 JB |
7655 | if (current->journal_info) { |
7656 | /* | |
7657 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7658 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7659 | * confused. |
7660 | */ | |
50745b0a | 7661 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7662 | current->journal_info = NULL; |
7663 | } | |
7664 | ||
eb838e73 JB |
7665 | /* |
7666 | * If this errors out it's because we couldn't invalidate pagecache for | |
7667 | * this range and we need to fallback to buffered. | |
7668 | */ | |
9c9464cc FM |
7669 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7670 | create)) { | |
7671 | ret = -ENOTBLK; | |
7672 | goto err; | |
7673 | } | |
eb838e73 | 7674 | |
fc4f21b1 | 7675 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7676 | if (IS_ERR(em)) { |
7677 | ret = PTR_ERR(em); | |
7678 | goto unlock_err; | |
7679 | } | |
4b46fce2 JB |
7680 | |
7681 | /* | |
7682 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7683 | * io. INLINE is special, and we could probably kludge it in here, but | |
7684 | * it's still buffered so for safety lets just fall back to the generic | |
7685 | * buffered path. | |
7686 | * | |
7687 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7688 | * decompress it, so there will be buffering required no matter what we | |
7689 | * do, so go ahead and fallback to buffered. | |
7690 | * | |
01327610 | 7691 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7692 | * to buffered IO. Don't blame me, this is the price we pay for using |
7693 | * the generic code. | |
7694 | */ | |
7695 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7696 | em->block_start == EXTENT_MAP_INLINE) { | |
7697 | free_extent_map(em); | |
eb838e73 JB |
7698 | ret = -ENOTBLK; |
7699 | goto unlock_err; | |
4b46fce2 JB |
7700 | } |
7701 | ||
c5794e51 NB |
7702 | if (create) { |
7703 | ret = btrfs_get_blocks_direct_write(&em, bh_result, inode, | |
7704 | dio_data, start, len); | |
7705 | if (ret < 0) | |
7706 | goto unlock_err; | |
7707 | ||
7708 | /* clear and unlock the entire range */ | |
7709 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7710 | unlock_bits, 1, 0, &cached_state); | |
7711 | } else { | |
1c8d0175 NB |
7712 | ret = btrfs_get_blocks_direct_read(em, bh_result, inode, |
7713 | start, len); | |
7714 | /* Can be negative only if we read from a hole */ | |
7715 | if (ret < 0) { | |
7716 | ret = 0; | |
7717 | free_extent_map(em); | |
7718 | goto unlock_err; | |
7719 | } | |
7720 | /* | |
7721 | * We need to unlock only the end area that we aren't using. | |
7722 | * The rest is going to be unlocked by the endio routine. | |
7723 | */ | |
7724 | lockstart = start + bh_result->b_size; | |
7725 | if (lockstart < lockend) { | |
7726 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
7727 | lockend, unlock_bits, 1, 0, | |
7728 | &cached_state); | |
7729 | } else { | |
7730 | free_extent_state(cached_state); | |
7731 | } | |
4b46fce2 JB |
7732 | } |
7733 | ||
4b46fce2 JB |
7734 | free_extent_map(em); |
7735 | ||
7736 | return 0; | |
eb838e73 JB |
7737 | |
7738 | unlock_err: | |
eb838e73 | 7739 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
ae0f1625 | 7740 | unlock_bits, 1, 0, &cached_state); |
9c9464cc | 7741 | err: |
50745b0a | 7742 | if (dio_data) |
7743 | current->journal_info = dio_data; | |
eb838e73 | 7744 | return ret; |
4b46fce2 JB |
7745 | } |
7746 | ||
58efbc9f OS |
7747 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7748 | struct bio *bio, | |
7749 | int mirror_num) | |
8b110e39 | 7750 | { |
2ff7e61e | 7751 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7752 | blk_status_t ret; |
8b110e39 | 7753 | |
37226b21 | 7754 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 | 7755 | |
2ff7e61e | 7756 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 | 7757 | if (ret) |
ea057f6d | 7758 | return ret; |
8b110e39 | 7759 | |
2ff7e61e | 7760 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
ea057f6d | 7761 | |
8b110e39 MX |
7762 | return ret; |
7763 | } | |
7764 | ||
7765 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7766 | struct bio *failed_bio, | |
7767 | struct io_failure_record *failrec, | |
7768 | int failed_mirror) | |
7769 | { | |
ab8d0fc4 | 7770 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7771 | int num_copies; |
7772 | ||
ab8d0fc4 | 7773 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7774 | if (num_copies == 1) { |
7775 | /* | |
7776 | * we only have a single copy of the data, so don't bother with | |
7777 | * all the retry and error correction code that follows. no | |
7778 | * matter what the error is, it is very likely to persist. | |
7779 | */ | |
ab8d0fc4 JM |
7780 | btrfs_debug(fs_info, |
7781 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7782 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7783 | return 0; |
7784 | } | |
7785 | ||
7786 | failrec->failed_mirror = failed_mirror; | |
7787 | failrec->this_mirror++; | |
7788 | if (failrec->this_mirror == failed_mirror) | |
7789 | failrec->this_mirror++; | |
7790 | ||
7791 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7792 | btrfs_debug(fs_info, |
7793 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7794 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7795 | return 0; |
7796 | } | |
7797 | ||
7798 | return 1; | |
7799 | } | |
7800 | ||
58efbc9f OS |
7801 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
7802 | struct page *page, unsigned int pgoff, | |
7803 | u64 start, u64 end, int failed_mirror, | |
7804 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7805 | { |
7806 | struct io_failure_record *failrec; | |
7870d082 JB |
7807 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7808 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
7809 | struct bio *bio; |
7810 | int isector; | |
f1c77c55 | 7811 | unsigned int read_mode = 0; |
17347cec | 7812 | int segs; |
8b110e39 | 7813 | int ret; |
58efbc9f | 7814 | blk_status_t status; |
c16a8ac3 | 7815 | struct bio_vec bvec; |
8b110e39 | 7816 | |
37226b21 | 7817 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7818 | |
7819 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7820 | if (ret) | |
58efbc9f | 7821 | return errno_to_blk_status(ret); |
8b110e39 MX |
7822 | |
7823 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7824 | failed_mirror); | |
7825 | if (!ret) { | |
7870d082 | 7826 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 7827 | return BLK_STS_IOERR; |
8b110e39 MX |
7828 | } |
7829 | ||
17347cec | 7830 | segs = bio_segments(failed_bio); |
c16a8ac3 | 7831 | bio_get_first_bvec(failed_bio, &bvec); |
17347cec | 7832 | if (segs > 1 || |
c16a8ac3 | 7833 | (bvec.bv_len > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7834 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7835 | |
7836 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7837 | isector >>= inode->i_sb->s_blocksize_bits; | |
7838 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7839 | pgoff, isector, repair_endio, repair_arg); |
ebcc3263 | 7840 | bio->bi_opf = REQ_OP_READ | read_mode; |
8b110e39 MX |
7841 | |
7842 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 7843 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
7844 | read_mode, failrec->this_mirror, failrec->in_validation); |
7845 | ||
58efbc9f OS |
7846 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
7847 | if (status) { | |
7870d082 | 7848 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
7849 | bio_put(bio); |
7850 | } | |
7851 | ||
58efbc9f | 7852 | return status; |
8b110e39 MX |
7853 | } |
7854 | ||
7855 | struct btrfs_retry_complete { | |
7856 | struct completion done; | |
7857 | struct inode *inode; | |
7858 | u64 start; | |
7859 | int uptodate; | |
7860 | }; | |
7861 | ||
4246a0b6 | 7862 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7863 | { |
7864 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 7865 | struct inode *inode = done->inode; |
8b110e39 | 7866 | struct bio_vec *bvec; |
7870d082 | 7867 | struct extent_io_tree *io_tree, *failure_tree; |
6dc4f100 | 7868 | struct bvec_iter_all iter_all; |
8b110e39 | 7869 | |
4e4cbee9 | 7870 | if (bio->bi_status) |
8b110e39 MX |
7871 | goto end; |
7872 | ||
2dabb324 | 7873 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
7874 | io_tree = &BTRFS_I(inode)->io_tree; |
7875 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
263663cd | 7876 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 7877 | |
8b110e39 | 7878 | done->uptodate = 1; |
c09abff8 | 7879 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 7880 | bio_for_each_segment_all(bvec, bio, iter_all) |
7870d082 JB |
7881 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
7882 | io_tree, done->start, bvec->bv_page, | |
7883 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
7884 | end: |
7885 | complete(&done->done); | |
7886 | bio_put(bio); | |
7887 | } | |
7888 | ||
58efbc9f OS |
7889 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
7890 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7891 | { |
2dabb324 | 7892 | struct btrfs_fs_info *fs_info; |
17347cec LB |
7893 | struct bio_vec bvec; |
7894 | struct bvec_iter iter; | |
8b110e39 | 7895 | struct btrfs_retry_complete done; |
4b46fce2 | 7896 | u64 start; |
2dabb324 CR |
7897 | unsigned int pgoff; |
7898 | u32 sectorsize; | |
7899 | int nr_sectors; | |
58efbc9f OS |
7900 | blk_status_t ret; |
7901 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 7902 | |
2dabb324 | 7903 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7904 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7905 | |
8b110e39 MX |
7906 | start = io_bio->logical; |
7907 | done.inode = inode; | |
17347cec | 7908 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 7909 | |
17347cec LB |
7910 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
7911 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
7912 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
7913 | |
7914 | next_block_or_try_again: | |
8b110e39 MX |
7915 | done.uptodate = 0; |
7916 | done.start = start; | |
7917 | init_completion(&done.done); | |
7918 | ||
17347cec | 7919 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
7920 | pgoff, start, start + sectorsize - 1, |
7921 | io_bio->mirror_num, | |
7922 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
7923 | if (ret) { |
7924 | err = ret; | |
7925 | goto next; | |
7926 | } | |
8b110e39 | 7927 | |
9c17f6cd | 7928 | wait_for_completion_io(&done.done); |
8b110e39 MX |
7929 | |
7930 | if (!done.uptodate) { | |
7931 | /* We might have another mirror, so try again */ | |
2dabb324 | 7932 | goto next_block_or_try_again; |
8b110e39 MX |
7933 | } |
7934 | ||
629ebf4f | 7935 | next: |
2dabb324 CR |
7936 | start += sectorsize; |
7937 | ||
97bf5a55 LB |
7938 | nr_sectors--; |
7939 | if (nr_sectors) { | |
2dabb324 | 7940 | pgoff += sectorsize; |
97bf5a55 | 7941 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
7942 | goto next_block_or_try_again; |
7943 | } | |
8b110e39 MX |
7944 | } |
7945 | ||
629ebf4f | 7946 | return err; |
8b110e39 MX |
7947 | } |
7948 | ||
4246a0b6 | 7949 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7950 | { |
7951 | struct btrfs_retry_complete *done = bio->bi_private; | |
7952 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
7953 | struct extent_io_tree *io_tree, *failure_tree; |
7954 | struct inode *inode = done->inode; | |
8b110e39 MX |
7955 | struct bio_vec *bvec; |
7956 | int uptodate; | |
7957 | int ret; | |
2b070cfe | 7958 | int i = 0; |
6dc4f100 | 7959 | struct bvec_iter_all iter_all; |
8b110e39 | 7960 | |
4e4cbee9 | 7961 | if (bio->bi_status) |
8b110e39 MX |
7962 | goto end; |
7963 | ||
7964 | uptodate = 1; | |
2dabb324 | 7965 | |
2dabb324 | 7966 | ASSERT(bio->bi_vcnt == 1); |
263663cd | 7967 | ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 7968 | |
7870d082 JB |
7969 | io_tree = &BTRFS_I(inode)->io_tree; |
7970 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
7971 | ||
c09abff8 | 7972 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 7973 | bio_for_each_segment_all(bvec, bio, iter_all) { |
7870d082 JB |
7974 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
7975 | bvec->bv_offset, done->start, | |
7976 | bvec->bv_len); | |
8b110e39 | 7977 | if (!ret) |
7870d082 JB |
7978 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
7979 | failure_tree, io_tree, done->start, | |
7980 | bvec->bv_page, | |
7981 | btrfs_ino(BTRFS_I(inode)), | |
7982 | bvec->bv_offset); | |
8b110e39 MX |
7983 | else |
7984 | uptodate = 0; | |
2b070cfe | 7985 | i++; |
8b110e39 MX |
7986 | } |
7987 | ||
7988 | done->uptodate = uptodate; | |
7989 | end: | |
7990 | complete(&done->done); | |
7991 | bio_put(bio); | |
7992 | } | |
7993 | ||
4e4cbee9 CH |
7994 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
7995 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 7996 | { |
2dabb324 | 7997 | struct btrfs_fs_info *fs_info; |
17347cec LB |
7998 | struct bio_vec bvec; |
7999 | struct bvec_iter iter; | |
8b110e39 MX |
8000 | struct btrfs_retry_complete done; |
8001 | u64 start; | |
8002 | u64 offset = 0; | |
2dabb324 CR |
8003 | u32 sectorsize; |
8004 | int nr_sectors; | |
8005 | unsigned int pgoff; | |
8006 | int csum_pos; | |
ef7cdac1 | 8007 | bool uptodate = (err == 0); |
8b110e39 | 8008 | int ret; |
58efbc9f | 8009 | blk_status_t status; |
dc380aea | 8010 | |
2dabb324 | 8011 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8012 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8013 | |
58efbc9f | 8014 | err = BLK_STS_OK; |
c1dc0896 | 8015 | start = io_bio->logical; |
8b110e39 | 8016 | done.inode = inode; |
17347cec | 8017 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8018 | |
17347cec LB |
8019 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8020 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8021 | |
17347cec | 8022 | pgoff = bvec.bv_offset; |
2dabb324 | 8023 | next_block: |
ef7cdac1 LB |
8024 | if (uptodate) { |
8025 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8026 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8027 | bvec.bv_page, pgoff, start, sectorsize); | |
8028 | if (likely(!ret)) | |
8029 | goto next; | |
8030 | } | |
8b110e39 MX |
8031 | try_again: |
8032 | done.uptodate = 0; | |
8033 | done.start = start; | |
8034 | init_completion(&done.done); | |
8035 | ||
58efbc9f OS |
8036 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8037 | pgoff, start, start + sectorsize - 1, | |
8038 | io_bio->mirror_num, btrfs_retry_endio, | |
8039 | &done); | |
8040 | if (status) { | |
8041 | err = status; | |
8b110e39 MX |
8042 | goto next; |
8043 | } | |
8044 | ||
9c17f6cd | 8045 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8046 | |
8047 | if (!done.uptodate) { | |
8048 | /* We might have another mirror, so try again */ | |
8049 | goto try_again; | |
8050 | } | |
8051 | next: | |
2dabb324 CR |
8052 | offset += sectorsize; |
8053 | start += sectorsize; | |
8054 | ||
8055 | ASSERT(nr_sectors); | |
8056 | ||
97bf5a55 LB |
8057 | nr_sectors--; |
8058 | if (nr_sectors) { | |
2dabb324 | 8059 | pgoff += sectorsize; |
97bf5a55 | 8060 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8061 | goto next_block; |
8062 | } | |
2c30c71b | 8063 | } |
c1dc0896 MX |
8064 | |
8065 | return err; | |
8066 | } | |
8067 | ||
4e4cbee9 CH |
8068 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8069 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8070 | { |
8071 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8072 | ||
8073 | if (skip_csum) { | |
8074 | if (unlikely(err)) | |
8075 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8076 | else | |
58efbc9f | 8077 | return BLK_STS_OK; |
8b110e39 MX |
8078 | } else { |
8079 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8080 | } | |
8081 | } | |
8082 | ||
4246a0b6 | 8083 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8084 | { |
8085 | struct btrfs_dio_private *dip = bio->bi_private; | |
8086 | struct inode *inode = dip->inode; | |
8087 | struct bio *dio_bio; | |
8088 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8089 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8090 | |
99c4e3b9 | 8091 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8092 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8093 | |
4b46fce2 | 8094 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8095 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8096 | dio_bio = dip->dio_bio; |
4b46fce2 | 8097 | |
4b46fce2 | 8098 | kfree(dip); |
c0da7aa1 | 8099 | |
99c4e3b9 | 8100 | dio_bio->bi_status = err; |
4055351c | 8101 | dio_end_io(dio_bio); |
b3a0dd50 | 8102 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8103 | bio_put(bio); |
4b46fce2 JB |
8104 | } |
8105 | ||
52427260 QW |
8106 | static void __endio_write_update_ordered(struct inode *inode, |
8107 | const u64 offset, const u64 bytes, | |
8108 | const bool uptodate) | |
4b46fce2 | 8109 | { |
0b246afa | 8110 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8111 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8112 | struct btrfs_workqueue *wq; |
8113 | btrfs_work_func_t func; | |
14543774 FM |
8114 | u64 ordered_offset = offset; |
8115 | u64 ordered_bytes = bytes; | |
67c003f9 | 8116 | u64 last_offset; |
4b46fce2 | 8117 | |
52427260 QW |
8118 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8119 | wq = fs_info->endio_freespace_worker; | |
8120 | func = btrfs_freespace_write_helper; | |
8121 | } else { | |
8122 | wq = fs_info->endio_write_workers; | |
8123 | func = btrfs_endio_write_helper; | |
8124 | } | |
8125 | ||
b25f0d00 NB |
8126 | while (ordered_offset < offset + bytes) { |
8127 | last_offset = ordered_offset; | |
8128 | if (btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8129 | &ordered_offset, | |
8130 | ordered_bytes, | |
8131 | uptodate)) { | |
8132 | btrfs_init_work(&ordered->work, func, | |
8133 | finish_ordered_fn, | |
8134 | NULL, NULL); | |
8135 | btrfs_queue_work(wq, &ordered->work); | |
8136 | } | |
8137 | /* | |
8138 | * If btrfs_dec_test_ordered_pending does not find any ordered | |
8139 | * extent in the range, we can exit. | |
8140 | */ | |
8141 | if (ordered_offset == last_offset) | |
8142 | return; | |
8143 | /* | |
8144 | * Our bio might span multiple ordered extents. In this case | |
52042d8e | 8145 | * we keep going until we have accounted the whole dio. |
b25f0d00 NB |
8146 | */ |
8147 | if (ordered_offset < offset + bytes) { | |
8148 | ordered_bytes = offset + bytes - ordered_offset; | |
8149 | ordered = NULL; | |
8150 | } | |
163cf09c | 8151 | } |
14543774 FM |
8152 | } |
8153 | ||
8154 | static void btrfs_endio_direct_write(struct bio *bio) | |
8155 | { | |
8156 | struct btrfs_dio_private *dip = bio->bi_private; | |
8157 | struct bio *dio_bio = dip->dio_bio; | |
8158 | ||
52427260 | 8159 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8160 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8161 | |
4b46fce2 | 8162 | kfree(dip); |
c0da7aa1 | 8163 | |
4e4cbee9 | 8164 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8165 | dio_end_io(dio_bio); |
9be3395b | 8166 | bio_put(bio); |
4b46fce2 JB |
8167 | } |
8168 | ||
d0ee3934 | 8169 | static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data, |
d0779291 | 8170 | struct bio *bio, u64 offset) |
eaf25d93 | 8171 | { |
c6100a4b | 8172 | struct inode *inode = private_data; |
4e4cbee9 | 8173 | blk_status_t ret; |
2ff7e61e | 8174 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8175 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8176 | return 0; |
8177 | } | |
8178 | ||
4246a0b6 | 8179 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8180 | { |
8181 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8182 | blk_status_t err = bio->bi_status; |
e65e1535 | 8183 | |
8b110e39 MX |
8184 | if (err) |
8185 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8186 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8187 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8188 | bio->bi_opf, | |
8b110e39 MX |
8189 | (unsigned long long)bio->bi_iter.bi_sector, |
8190 | bio->bi_iter.bi_size, err); | |
8191 | ||
8192 | if (dip->subio_endio) | |
8193 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8194 | |
8195 | if (err) { | |
e65e1535 | 8196 | /* |
de224b7c NB |
8197 | * We want to perceive the errors flag being set before |
8198 | * decrementing the reference count. We don't need a barrier | |
8199 | * since atomic operations with a return value are fully | |
8200 | * ordered as per atomic_t.txt | |
e65e1535 | 8201 | */ |
de224b7c | 8202 | dip->errors = 1; |
e65e1535 MX |
8203 | } |
8204 | ||
8205 | /* if there are more bios still pending for this dio, just exit */ | |
8206 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8207 | goto out; | |
8208 | ||
9be3395b | 8209 | if (dip->errors) { |
e65e1535 | 8210 | bio_io_error(dip->orig_bio); |
9be3395b | 8211 | } else { |
2dbe0c77 | 8212 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8213 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8214 | } |
8215 | out: | |
8216 | bio_put(bio); | |
8217 | } | |
8218 | ||
4e4cbee9 | 8219 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8220 | struct btrfs_dio_private *dip, |
8221 | struct bio *bio, | |
8222 | u64 file_offset) | |
8223 | { | |
8224 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8225 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8226 | blk_status_t ret; |
c1dc0896 MX |
8227 | |
8228 | /* | |
8229 | * We load all the csum data we need when we submit | |
8230 | * the first bio to reduce the csum tree search and | |
8231 | * contention. | |
8232 | */ | |
8233 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8234 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8235 | file_offset); |
8236 | if (ret) | |
8237 | return ret; | |
8238 | } | |
8239 | ||
8240 | if (bio == dip->orig_bio) | |
8241 | return 0; | |
8242 | ||
8243 | file_offset -= dip->logical_offset; | |
8244 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8245 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8246 | ||
8247 | return 0; | |
8248 | } | |
8249 | ||
d0ee3934 DS |
8250 | static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio, |
8251 | struct inode *inode, u64 file_offset, int async_submit) | |
e65e1535 | 8252 | { |
0b246afa | 8253 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8254 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8255 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8256 | blk_status_t ret; |
e65e1535 | 8257 | |
4c274bc6 | 8258 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8259 | if (async_submit) |
8260 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8261 | ||
5fd02043 | 8262 | if (!write) { |
0b246afa | 8263 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8264 | if (ret) |
8265 | goto err; | |
8266 | } | |
e65e1535 | 8267 | |
e6961cac | 8268 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8269 | goto map; |
8270 | ||
8271 | if (write && async_submit) { | |
c6100a4b JB |
8272 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8273 | file_offset, inode, | |
e288c080 | 8274 | btrfs_submit_bio_start_direct_io); |
e65e1535 | 8275 | goto err; |
1ae39938 JB |
8276 | } else if (write) { |
8277 | /* | |
8278 | * If we aren't doing async submit, calculate the csum of the | |
8279 | * bio now. | |
8280 | */ | |
2ff7e61e | 8281 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8282 | if (ret) |
8283 | goto err; | |
23ea8e5a | 8284 | } else { |
2ff7e61e | 8285 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8286 | file_offset); |
c2db1073 TI |
8287 | if (ret) |
8288 | goto err; | |
8289 | } | |
1ae39938 | 8290 | map: |
9b4a9b28 | 8291 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 | 8292 | err: |
e65e1535 MX |
8293 | return ret; |
8294 | } | |
8295 | ||
e6961cac | 8296 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8297 | { |
8298 | struct inode *inode = dip->inode; | |
0b246afa | 8299 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8300 | struct bio *bio; |
8301 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8302 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8303 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8304 | u64 map_length; |
1ae39938 | 8305 | int async_submit = 0; |
725130ba LB |
8306 | u64 submit_len; |
8307 | int clone_offset = 0; | |
8308 | int clone_len; | |
5f4dc8fc | 8309 | int ret; |
58efbc9f | 8310 | blk_status_t status; |
e65e1535 | 8311 | |
4f024f37 | 8312 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8313 | submit_len = map_length; |
0b246afa JM |
8314 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8315 | &map_length, NULL, 0); | |
7a5c3c9b | 8316 | if (ret) |
e65e1535 | 8317 | return -EIO; |
facc8a22 | 8318 | |
725130ba | 8319 | if (map_length >= submit_len) { |
02f57c7a | 8320 | bio = orig_bio; |
c1dc0896 | 8321 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8322 | goto submit; |
8323 | } | |
8324 | ||
53b381b3 | 8325 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8326 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8327 | async_submit = 0; |
8328 | else | |
8329 | async_submit = 1; | |
8330 | ||
725130ba LB |
8331 | /* bio split */ |
8332 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8333 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8334 | do { |
725130ba | 8335 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8336 | |
725130ba LB |
8337 | /* |
8338 | * This will never fail as it's passing GPF_NOFS and | |
8339 | * the allocation is backed by btrfs_bioset. | |
8340 | */ | |
e477094f | 8341 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8342 | clone_len); |
8343 | bio->bi_private = dip; | |
8344 | bio->bi_end_io = btrfs_end_dio_bio; | |
8345 | btrfs_io_bio(bio)->logical = file_offset; | |
8346 | ||
8347 | ASSERT(submit_len >= clone_len); | |
8348 | submit_len -= clone_len; | |
8349 | if (submit_len == 0) | |
8350 | break; | |
e65e1535 | 8351 | |
725130ba LB |
8352 | /* |
8353 | * Increase the count before we submit the bio so we know | |
8354 | * the end IO handler won't happen before we increase the | |
8355 | * count. Otherwise, the dip might get freed before we're | |
8356 | * done setting it up. | |
8357 | */ | |
8358 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8359 | |
d0ee3934 | 8360 | status = btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8361 | async_submit); |
8362 | if (status) { | |
725130ba LB |
8363 | bio_put(bio); |
8364 | atomic_dec(&dip->pending_bios); | |
8365 | goto out_err; | |
8366 | } | |
e65e1535 | 8367 | |
725130ba LB |
8368 | clone_offset += clone_len; |
8369 | start_sector += clone_len >> 9; | |
8370 | file_offset += clone_len; | |
5f4dc8fc | 8371 | |
725130ba LB |
8372 | map_length = submit_len; |
8373 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8374 | start_sector << 9, &map_length, NULL, 0); | |
8375 | if (ret) | |
8376 | goto out_err; | |
3c91ee69 | 8377 | } while (submit_len > 0); |
e65e1535 | 8378 | |
02f57c7a | 8379 | submit: |
d0ee3934 | 8380 | status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8381 | if (!status) |
e65e1535 MX |
8382 | return 0; |
8383 | ||
8384 | bio_put(bio); | |
8385 | out_err: | |
8386 | dip->errors = 1; | |
8387 | /* | |
de224b7c NB |
8388 | * Before atomic variable goto zero, we must make sure dip->errors is |
8389 | * perceived to be set. This ordering is ensured by the fact that an | |
8390 | * atomic operations with a return value are fully ordered as per | |
8391 | * atomic_t.txt | |
e65e1535 | 8392 | */ |
e65e1535 MX |
8393 | if (atomic_dec_and_test(&dip->pending_bios)) |
8394 | bio_io_error(dip->orig_bio); | |
8395 | ||
8396 | /* bio_end_io() will handle error, so we needn't return it */ | |
8397 | return 0; | |
8398 | } | |
8399 | ||
8a4c1e42 MC |
8400 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8401 | loff_t file_offset) | |
4b46fce2 | 8402 | { |
61de718f | 8403 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8404 | struct bio *bio = NULL; |
8405 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8406 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8407 | int ret = 0; |
8408 | ||
8b6c1d56 | 8409 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8410 | |
c1dc0896 | 8411 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8412 | if (!dip) { |
8413 | ret = -ENOMEM; | |
61de718f | 8414 | goto free_ordered; |
4b46fce2 | 8415 | } |
4b46fce2 | 8416 | |
9be3395b | 8417 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8418 | dip->inode = inode; |
8419 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8420 | dip->bytes = dio_bio->bi_iter.bi_size; |
8421 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8422 | bio->bi_private = dip; |
8423 | dip->orig_bio = bio; | |
9be3395b | 8424 | dip->dio_bio = dio_bio; |
e65e1535 | 8425 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8426 | io_bio = btrfs_io_bio(bio); |
8427 | io_bio->logical = file_offset; | |
4b46fce2 | 8428 | |
c1dc0896 | 8429 | if (write) { |
3892ac90 | 8430 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8431 | } else { |
3892ac90 | 8432 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8433 | dip->subio_endio = btrfs_subio_endio_read; |
8434 | } | |
4b46fce2 | 8435 | |
f28a4928 FM |
8436 | /* |
8437 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8438 | * even if we fail to submit a bio, because in such case we do the | |
8439 | * corresponding error handling below and it must not be done a second | |
8440 | * time by btrfs_direct_IO(). | |
8441 | */ | |
8442 | if (write) { | |
8443 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8444 | ||
8445 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8446 | dip->bytes; | |
8447 | dio_data->unsubmitted_oe_range_start = | |
8448 | dio_data->unsubmitted_oe_range_end; | |
8449 | } | |
8450 | ||
e6961cac | 8451 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8452 | if (!ret) |
eaf25d93 | 8453 | return; |
9be3395b | 8454 | |
b3a0dd50 | 8455 | btrfs_io_bio_free_csum(io_bio); |
9be3395b | 8456 | |
4b46fce2 JB |
8457 | free_ordered: |
8458 | /* | |
61de718f FM |
8459 | * If we arrived here it means either we failed to submit the dip |
8460 | * or we either failed to clone the dio_bio or failed to allocate the | |
8461 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8462 | * call bio_endio against our io_bio so that we get proper resource | |
8463 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8464 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8465 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8466 | */ |
3892ac90 | 8467 | if (bio && dip) { |
054ec2f6 | 8468 | bio_io_error(bio); |
61de718f | 8469 | /* |
3892ac90 | 8470 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8471 | * and all the cleanup and final put for dio_bio (through |
8472 | * dio_end_io()). | |
8473 | */ | |
8474 | dip = NULL; | |
3892ac90 | 8475 | bio = NULL; |
61de718f | 8476 | } else { |
14543774 | 8477 | if (write) |
52427260 | 8478 | __endio_write_update_ordered(inode, |
14543774 FM |
8479 | file_offset, |
8480 | dio_bio->bi_iter.bi_size, | |
52427260 | 8481 | false); |
14543774 | 8482 | else |
61de718f FM |
8483 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8484 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8485 | |
4e4cbee9 | 8486 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8487 | /* |
8488 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8489 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8490 | */ | |
4055351c | 8491 | dio_end_io(dio_bio); |
4b46fce2 | 8492 | } |
3892ac90 LB |
8493 | if (bio) |
8494 | bio_put(bio); | |
61de718f | 8495 | kfree(dip); |
4b46fce2 JB |
8496 | } |
8497 | ||
2ff7e61e | 8498 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8499 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8500 | { |
8501 | int seg; | |
a1b75f7d | 8502 | int i; |
0b246afa | 8503 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8504 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8505 | |
8506 | if (offset & blocksize_mask) | |
8507 | goto out; | |
8508 | ||
28060d5d AV |
8509 | if (iov_iter_alignment(iter) & blocksize_mask) |
8510 | goto out; | |
a1b75f7d | 8511 | |
28060d5d | 8512 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8513 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8514 | return 0; |
8515 | /* | |
8516 | * Check to make sure we don't have duplicate iov_base's in this | |
8517 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8518 | * when reading back. | |
8519 | */ | |
8520 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8521 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8522 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8523 | goto out; |
8524 | } | |
5a5f79b5 CM |
8525 | } |
8526 | retval = 0; | |
8527 | out: | |
8528 | return retval; | |
8529 | } | |
eb838e73 | 8530 | |
c8b8e32d | 8531 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8532 | { |
4b46fce2 JB |
8533 | struct file *file = iocb->ki_filp; |
8534 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8535 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8536 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8537 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8538 | loff_t offset = iocb->ki_pos; |
0934856d | 8539 | size_t count = 0; |
2e60a51e | 8540 | int flags = 0; |
38851cc1 MX |
8541 | bool wakeup = true; |
8542 | bool relock = false; | |
0934856d | 8543 | ssize_t ret; |
4b46fce2 | 8544 | |
8c70c9f8 | 8545 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8546 | return 0; |
3f7c579c | 8547 | |
fe0f07d0 | 8548 | inode_dio_begin(inode); |
38851cc1 | 8549 | |
0e267c44 | 8550 | /* |
41bd9ca4 MX |
8551 | * The generic stuff only does filemap_write_and_wait_range, which |
8552 | * isn't enough if we've written compressed pages to this area, so | |
8553 | * we need to flush the dirty pages again to make absolutely sure | |
8554 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8555 | */ |
a6cbcd4a | 8556 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8557 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8558 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8559 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8560 | offset + count - 1); | |
0e267c44 | 8561 | |
6f673763 | 8562 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8563 | /* |
8564 | * If the write DIO is beyond the EOF, we need update | |
8565 | * the isize, but it is protected by i_mutex. So we can | |
8566 | * not unlock the i_mutex at this case. | |
8567 | */ | |
8568 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8569 | dio_data.overwrite = 1; |
5955102c | 8570 | inode_unlock(inode); |
38851cc1 | 8571 | relock = true; |
edf064e7 GR |
8572 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8573 | ret = -EAGAIN; | |
8574 | goto out; | |
38851cc1 | 8575 | } |
364ecf36 QW |
8576 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8577 | offset, count); | |
0934856d | 8578 | if (ret) |
38851cc1 | 8579 | goto out; |
e1cbbfa5 JB |
8580 | |
8581 | /* | |
8582 | * We need to know how many extents we reserved so that we can | |
8583 | * do the accounting properly if we go over the number we | |
8584 | * originally calculated. Abuse current->journal_info for this. | |
8585 | */ | |
da17066c | 8586 | dio_data.reserve = round_up(count, |
0b246afa | 8587 | fs_info->sectorsize); |
f28a4928 FM |
8588 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8589 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8590 | current->journal_info = &dio_data; |
97dcdea0 | 8591 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8592 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8593 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8594 | inode_dio_end(inode); |
38851cc1 MX |
8595 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8596 | wakeup = false; | |
0934856d MX |
8597 | } |
8598 | ||
17f8c842 | 8599 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8600 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8601 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8602 | btrfs_submit_direct, flags); |
6f673763 | 8603 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8604 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8605 | current->journal_info = NULL; |
ddba1bfc | 8606 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8607 | if (dio_data.reserve) |
bc42bda2 | 8608 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 | 8609 | offset, dio_data.reserve, true); |
f28a4928 FM |
8610 | /* |
8611 | * On error we might have left some ordered extents | |
8612 | * without submitting corresponding bios for them, so | |
8613 | * cleanup them up to avoid other tasks getting them | |
8614 | * and waiting for them to complete forever. | |
8615 | */ | |
8616 | if (dio_data.unsubmitted_oe_range_start < | |
8617 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8618 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8619 | dio_data.unsubmitted_oe_range_start, |
8620 | dio_data.unsubmitted_oe_range_end - | |
8621 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8622 | false); |
ddba1bfc | 8623 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 | 8624 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8625 | offset, count - (size_t)ret, true); |
8626 | btrfs_delalloc_release_extents(BTRFS_I(inode), count, false); | |
0934856d | 8627 | } |
38851cc1 | 8628 | out: |
2e60a51e | 8629 | if (wakeup) |
fe0f07d0 | 8630 | inode_dio_end(inode); |
38851cc1 | 8631 | if (relock) |
5955102c | 8632 | inode_lock(inode); |
0934856d | 8633 | |
364ecf36 | 8634 | extent_changeset_free(data_reserved); |
0934856d | 8635 | return ret; |
16432985 CM |
8636 | } |
8637 | ||
05dadc09 TI |
8638 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8639 | ||
1506fcc8 YS |
8640 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8641 | __u64 start, __u64 len) | |
8642 | { | |
05dadc09 TI |
8643 | int ret; |
8644 | ||
8645 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8646 | if (ret) | |
8647 | return ret; | |
8648 | ||
2135fb9b | 8649 | return extent_fiemap(inode, fieinfo, start, len); |
1506fcc8 YS |
8650 | } |
8651 | ||
a52d9a80 | 8652 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8653 | { |
d1310b2e CM |
8654 | struct extent_io_tree *tree; |
8655 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8656 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8657 | } |
1832a6d5 | 8658 | |
a52d9a80 | 8659 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8660 | { |
be7bd730 JB |
8661 | struct inode *inode = page->mapping->host; |
8662 | int ret; | |
b888db2b CM |
8663 | |
8664 | if (current->flags & PF_MEMALLOC) { | |
8665 | redirty_page_for_writepage(wbc, page); | |
8666 | unlock_page(page); | |
8667 | return 0; | |
8668 | } | |
be7bd730 JB |
8669 | |
8670 | /* | |
8671 | * If we are under memory pressure we will call this directly from the | |
8672 | * VM, we need to make sure we have the inode referenced for the ordered | |
8673 | * extent. If not just return like we didn't do anything. | |
8674 | */ | |
8675 | if (!igrab(inode)) { | |
8676 | redirty_page_for_writepage(wbc, page); | |
8677 | return AOP_WRITEPAGE_ACTIVATE; | |
8678 | } | |
0a9b0e53 | 8679 | ret = extent_write_full_page(page, wbc); |
be7bd730 JB |
8680 | btrfs_add_delayed_iput(inode); |
8681 | return ret; | |
9ebefb18 CM |
8682 | } |
8683 | ||
48a3b636 ES |
8684 | static int btrfs_writepages(struct address_space *mapping, |
8685 | struct writeback_control *wbc) | |
b293f02e | 8686 | { |
8ae225a8 | 8687 | return extent_writepages(mapping, wbc); |
b293f02e CM |
8688 | } |
8689 | ||
3ab2fb5a CM |
8690 | static int |
8691 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8692 | struct list_head *pages, unsigned nr_pages) | |
8693 | { | |
2a3ff0ad | 8694 | return extent_readpages(mapping, pages, nr_pages); |
3ab2fb5a | 8695 | } |
2a3ff0ad | 8696 | |
e6dcd2dc | 8697 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8698 | { |
477a30ba | 8699 | int ret = try_release_extent_mapping(page, gfp_flags); |
a52d9a80 CM |
8700 | if (ret == 1) { |
8701 | ClearPagePrivate(page); | |
8702 | set_page_private(page, 0); | |
09cbfeaf | 8703 | put_page(page); |
39279cc3 | 8704 | } |
a52d9a80 | 8705 | return ret; |
39279cc3 CM |
8706 | } |
8707 | ||
e6dcd2dc CM |
8708 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8709 | { | |
98509cfc CM |
8710 | if (PageWriteback(page) || PageDirty(page)) |
8711 | return 0; | |
3ba7ab22 | 8712 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8713 | } |
8714 | ||
d47992f8 LC |
8715 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8716 | unsigned int length) | |
39279cc3 | 8717 | { |
5fd02043 | 8718 | struct inode *inode = page->mapping->host; |
d1310b2e | 8719 | struct extent_io_tree *tree; |
e6dcd2dc | 8720 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8721 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8722 | u64 page_start = page_offset(page); |
09cbfeaf | 8723 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8724 | u64 start; |
8725 | u64 end; | |
131e404a | 8726 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8727 | |
8b62b72b CM |
8728 | /* |
8729 | * we have the page locked, so new writeback can't start, | |
8730 | * and the dirty bit won't be cleared while we are here. | |
8731 | * | |
8732 | * Wait for IO on this page so that we can safely clear | |
8733 | * the PagePrivate2 bit and do ordered accounting | |
8734 | */ | |
e6dcd2dc | 8735 | wait_on_page_writeback(page); |
8b62b72b | 8736 | |
5fd02043 | 8737 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8738 | if (offset) { |
8739 | btrfs_releasepage(page, GFP_NOFS); | |
8740 | return; | |
8741 | } | |
131e404a FDBM |
8742 | |
8743 | if (!inode_evicting) | |
ff13db41 | 8744 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8745 | again: |
8746 | start = page_start; | |
a776c6fa | 8747 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8748 | page_end - start + 1); |
e6dcd2dc | 8749 | if (ordered) { |
dbfdb6d1 | 8750 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8751 | /* |
8752 | * IO on this page will never be started, so we need | |
8753 | * to account for any ordered extents now | |
8754 | */ | |
131e404a | 8755 | if (!inode_evicting) |
dbfdb6d1 | 8756 | clear_extent_bit(tree, start, end, |
131e404a | 8757 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8758 | EXTENT_DELALLOC_NEW | |
131e404a | 8759 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8760 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8761 | /* |
8762 | * whoever cleared the private bit is responsible | |
8763 | * for the finish_ordered_io | |
8764 | */ | |
77cef2ec JB |
8765 | if (TestClearPagePrivate2(page)) { |
8766 | struct btrfs_ordered_inode_tree *tree; | |
8767 | u64 new_len; | |
8768 | ||
8769 | tree = &BTRFS_I(inode)->ordered_tree; | |
8770 | ||
8771 | spin_lock_irq(&tree->lock); | |
8772 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8773 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8774 | if (new_len < ordered->truncated_len) |
8775 | ordered->truncated_len = new_len; | |
8776 | spin_unlock_irq(&tree->lock); | |
8777 | ||
8778 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8779 | start, |
8780 | end - start + 1, 1)) | |
77cef2ec | 8781 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8782 | } |
e6dcd2dc | 8783 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8784 | if (!inode_evicting) { |
8785 | cached_state = NULL; | |
dbfdb6d1 | 8786 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8787 | &cached_state); |
8788 | } | |
dbfdb6d1 CR |
8789 | |
8790 | start = end + 1; | |
8791 | if (start < page_end) | |
8792 | goto again; | |
131e404a FDBM |
8793 | } |
8794 | ||
b9d0b389 QW |
8795 | /* |
8796 | * Qgroup reserved space handler | |
8797 | * Page here will be either | |
8798 | * 1) Already written to disk | |
8799 | * In this case, its reserved space is released from data rsv map | |
8800 | * and will be freed by delayed_ref handler finally. | |
8801 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8802 | * space. | |
8803 | * 2) Not written to disk | |
0b34c261 GR |
8804 | * This means the reserved space should be freed here. However, |
8805 | * if a truncate invalidates the page (by clearing PageDirty) | |
8806 | * and the page is accounted for while allocating extent | |
8807 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8808 | * free the entire extent. | |
b9d0b389 | 8809 | */ |
0b34c261 | 8810 | if (PageDirty(page)) |
bc42bda2 | 8811 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
8812 | if (!inode_evicting) { |
8813 | clear_extent_bit(tree, page_start, page_end, | |
8814 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
8815 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
8816 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 8817 | &cached_state); |
131e404a FDBM |
8818 | |
8819 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8820 | } |
e6dcd2dc | 8821 | |
4a096752 | 8822 | ClearPageChecked(page); |
9ad6b7bc | 8823 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8824 | ClearPagePrivate(page); |
8825 | set_page_private(page, 0); | |
09cbfeaf | 8826 | put_page(page); |
9ad6b7bc | 8827 | } |
39279cc3 CM |
8828 | } |
8829 | ||
9ebefb18 CM |
8830 | /* |
8831 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8832 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8833 | * be careful to check for EOF conditions here. We set the page up correctly | |
8834 | * for a written page which means we get ENOSPC checking when writing into | |
8835 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8836 | * support these features. | |
8837 | * | |
8838 | * We are not allowed to take the i_mutex here so we have to play games to | |
8839 | * protect against truncate races as the page could now be beyond EOF. Because | |
d1342aad OS |
8840 | * truncate_setsize() writes the inode size before removing pages, once we have |
8841 | * the page lock we can determine safely if the page is beyond EOF. If it is not | |
9ebefb18 CM |
8842 | * beyond EOF, then the page is guaranteed safe against truncation until we |
8843 | * unlock the page. | |
8844 | */ | |
a528a241 | 8845 | vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8846 | { |
c2ec175c | 8847 | struct page *page = vmf->page; |
11bac800 | 8848 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8849 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8850 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8851 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8852 | struct extent_state *cached_state = NULL; |
364ecf36 | 8853 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
8854 | char *kaddr; |
8855 | unsigned long zero_start; | |
9ebefb18 | 8856 | loff_t size; |
a528a241 SJ |
8857 | vm_fault_t ret; |
8858 | int ret2; | |
9998eb70 | 8859 | int reserved = 0; |
d0b7da88 | 8860 | u64 reserved_space; |
a52d9a80 | 8861 | u64 page_start; |
e6dcd2dc | 8862 | u64 page_end; |
d0b7da88 CR |
8863 | u64 end; |
8864 | ||
09cbfeaf | 8865 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8866 | |
b2b5ef5c | 8867 | sb_start_pagefault(inode->i_sb); |
df480633 | 8868 | page_start = page_offset(page); |
09cbfeaf | 8869 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8870 | end = page_end; |
df480633 | 8871 | |
d0b7da88 CR |
8872 | /* |
8873 | * Reserving delalloc space after obtaining the page lock can lead to | |
8874 | * deadlock. For example, if a dirty page is locked by this function | |
8875 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8876 | * dirty page write out, then the btrfs_writepage() function could | |
8877 | * end up waiting indefinitely to get a lock on the page currently | |
8878 | * being processed by btrfs_page_mkwrite() function. | |
8879 | */ | |
a528a241 | 8880 | ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 8881 | reserved_space); |
a528a241 SJ |
8882 | if (!ret2) { |
8883 | ret2 = file_update_time(vmf->vma->vm_file); | |
9998eb70 CM |
8884 | reserved = 1; |
8885 | } | |
a528a241 SJ |
8886 | if (ret2) { |
8887 | ret = vmf_error(ret2); | |
9998eb70 CM |
8888 | if (reserved) |
8889 | goto out; | |
8890 | goto out_noreserve; | |
56a76f82 | 8891 | } |
1832a6d5 | 8892 | |
56a76f82 | 8893 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8894 | again: |
9ebefb18 | 8895 | lock_page(page); |
9ebefb18 | 8896 | size = i_size_read(inode); |
a52d9a80 | 8897 | |
9ebefb18 | 8898 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8899 | (page_start >= size)) { |
9ebefb18 CM |
8900 | /* page got truncated out from underneath us */ |
8901 | goto out_unlock; | |
8902 | } | |
e6dcd2dc CM |
8903 | wait_on_page_writeback(page); |
8904 | ||
ff13db41 | 8905 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8906 | set_page_extent_mapped(page); |
8907 | ||
eb84ae03 CM |
8908 | /* |
8909 | * we can't set the delalloc bits if there are pending ordered | |
8910 | * extents. Drop our locks and wait for them to finish | |
8911 | */ | |
a776c6fa NB |
8912 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8913 | PAGE_SIZE); | |
e6dcd2dc | 8914 | if (ordered) { |
2ac55d41 | 8915 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8916 | &cached_state); |
e6dcd2dc | 8917 | unlock_page(page); |
eb84ae03 | 8918 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8919 | btrfs_put_ordered_extent(ordered); |
8920 | goto again; | |
8921 | } | |
8922 | ||
09cbfeaf | 8923 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8924 | reserved_space = round_up(size - page_start, |
0b246afa | 8925 | fs_info->sectorsize); |
09cbfeaf | 8926 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 8927 | end = page_start + reserved_space - 1; |
bc42bda2 | 8928 | btrfs_delalloc_release_space(inode, data_reserved, |
43b18595 QW |
8929 | page_start, PAGE_SIZE - reserved_space, |
8930 | true); | |
d0b7da88 CR |
8931 | } |
8932 | } | |
8933 | ||
fbf19087 | 8934 | /* |
5416034f LB |
8935 | * page_mkwrite gets called when the page is firstly dirtied after it's |
8936 | * faulted in, but write(2) could also dirty a page and set delalloc | |
8937 | * bits, thus in this case for space account reason, we still need to | |
8938 | * clear any delalloc bits within this page range since we have to | |
8939 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 8940 | */ |
d0b7da88 | 8941 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
8942 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8943 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 8944 | 0, 0, &cached_state); |
fbf19087 | 8945 | |
a528a241 | 8946 | ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
ba8b04c1 | 8947 | &cached_state, 0); |
a528a241 | 8948 | if (ret2) { |
2ac55d41 | 8949 | unlock_extent_cached(io_tree, page_start, page_end, |
e43bbe5e | 8950 | &cached_state); |
9ed74f2d JB |
8951 | ret = VM_FAULT_SIGBUS; |
8952 | goto out_unlock; | |
8953 | } | |
a528a241 | 8954 | ret2 = 0; |
9ebefb18 CM |
8955 | |
8956 | /* page is wholly or partially inside EOF */ | |
09cbfeaf | 8957 | if (page_start + PAGE_SIZE > size) |
7073017a | 8958 | zero_start = offset_in_page(size); |
9ebefb18 | 8959 | else |
09cbfeaf | 8960 | zero_start = PAGE_SIZE; |
9ebefb18 | 8961 | |
09cbfeaf | 8962 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 8963 | kaddr = kmap(page); |
09cbfeaf | 8964 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
8965 | flush_dcache_page(page); |
8966 | kunmap(page); | |
8967 | } | |
247e743c | 8968 | ClearPageChecked(page); |
e6dcd2dc | 8969 | set_page_dirty(page); |
50a9b214 | 8970 | SetPageUptodate(page); |
5a3f23d5 | 8971 | |
0b246afa | 8972 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 8973 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 8974 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8975 | |
e43bbe5e | 8976 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state); |
9ebefb18 | 8977 | |
a528a241 | 8978 | if (!ret2) { |
43b18595 | 8979 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true); |
b2b5ef5c | 8980 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8981 | extent_changeset_free(data_reserved); |
50a9b214 | 8982 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8983 | } |
717beb96 CM |
8984 | |
8985 | out_unlock: | |
9ebefb18 | 8986 | unlock_page(page); |
1832a6d5 | 8987 | out: |
43b18595 | 8988 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0)); |
bc42bda2 | 8989 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
43b18595 | 8990 | reserved_space, (ret != 0)); |
9998eb70 | 8991 | out_noreserve: |
b2b5ef5c | 8992 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 8993 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
8994 | return ret; |
8995 | } | |
8996 | ||
213e8c55 | 8997 | static int btrfs_truncate(struct inode *inode, bool skip_writeback) |
39279cc3 | 8998 | { |
0b246afa | 8999 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9000 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9001 | struct btrfs_block_rsv *rsv; |
ad7e1a74 | 9002 | int ret; |
39279cc3 | 9003 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9004 | u64 mask = fs_info->sectorsize - 1; |
9005 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9006 | |
213e8c55 FM |
9007 | if (!skip_writeback) { |
9008 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), | |
9009 | (u64)-1); | |
9010 | if (ret) | |
9011 | return ret; | |
9012 | } | |
39279cc3 | 9013 | |
fcb80c2a | 9014 | /* |
f7e9e8fc OS |
9015 | * Yes ladies and gentlemen, this is indeed ugly. We have a couple of |
9016 | * things going on here: | |
fcb80c2a | 9017 | * |
f7e9e8fc | 9018 | * 1) We need to reserve space to update our inode. |
fcb80c2a | 9019 | * |
f7e9e8fc | 9020 | * 2) We need to have something to cache all the space that is going to |
fcb80c2a JB |
9021 | * be free'd up by the truncate operation, but also have some slack |
9022 | * space reserved in case it uses space during the truncate (thank you | |
9023 | * very much snapshotting). | |
9024 | * | |
f7e9e8fc | 9025 | * And we need these to be separate. The fact is we can use a lot of |
fcb80c2a | 9026 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9027 | * we will use, so we need the truncate reservation to be separate so it |
f7e9e8fc OS |
9028 | * doesn't end up using space reserved for updating the inode. We also |
9029 | * need to be able to stop the transaction and start a new one, which | |
9030 | * means we need to be able to update the inode several times, and we | |
9031 | * have no idea of knowing how many times that will be, so we can't just | |
9032 | * reserve 1 item for the entirety of the operation, so that has to be | |
9033 | * done separately as well. | |
fcb80c2a JB |
9034 | * |
9035 | * So that leaves us with | |
9036 | * | |
f7e9e8fc | 9037 | * 1) rsv - for the truncate reservation, which we will steal from the |
fcb80c2a | 9038 | * transaction reservation. |
f7e9e8fc | 9039 | * 2) fs_info->trans_block_rsv - this will have 1 items worth left for |
fcb80c2a JB |
9040 | * updating the inode. |
9041 | */ | |
2ff7e61e | 9042 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9043 | if (!rsv) |
9044 | return -ENOMEM; | |
4a338542 | 9045 | rsv->size = min_size; |
ca7e70f5 | 9046 | rsv->failfast = 1; |
f0cd846e | 9047 | |
907cbceb | 9048 | /* |
07127184 | 9049 | * 1 for the truncate slack space |
907cbceb JB |
9050 | * 1 for updating the inode. |
9051 | */ | |
f3fe820c | 9052 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a | 9053 | if (IS_ERR(trans)) { |
ad7e1a74 | 9054 | ret = PTR_ERR(trans); |
fcb80c2a JB |
9055 | goto out; |
9056 | } | |
f0cd846e | 9057 | |
907cbceb | 9058 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9059 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
3a584174 | 9060 | min_size, false); |
fcb80c2a | 9061 | BUG_ON(ret); |
f0cd846e | 9062 | |
5dc562c5 JB |
9063 | /* |
9064 | * So if we truncate and then write and fsync we normally would just | |
9065 | * write the extents that changed, which is a problem if we need to | |
9066 | * first truncate that entire inode. So set this flag so we write out | |
9067 | * all of the extents in the inode to the sync log so we're completely | |
9068 | * safe. | |
9069 | */ | |
9070 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9071 | trans->block_rsv = rsv; |
907cbceb | 9072 | |
8082510e YZ |
9073 | while (1) { |
9074 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9075 | inode->i_size, | |
9076 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9077 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 | 9078 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 9079 | break; |
39279cc3 | 9080 | |
8082510e | 9081 | ret = btrfs_update_inode(trans, root, inode); |
ad7e1a74 | 9082 | if (ret) |
3893e33b | 9083 | break; |
ca7e70f5 | 9084 | |
3a45bb20 | 9085 | btrfs_end_transaction(trans); |
2ff7e61e | 9086 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9087 | |
9088 | trans = btrfs_start_transaction(root, 2); | |
9089 | if (IS_ERR(trans)) { | |
ad7e1a74 | 9090 | ret = PTR_ERR(trans); |
ca7e70f5 JB |
9091 | trans = NULL; |
9092 | break; | |
9093 | } | |
9094 | ||
47b5d646 | 9095 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9096 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
3a584174 | 9097 | rsv, min_size, false); |
ca7e70f5 JB |
9098 | BUG_ON(ret); /* shouldn't happen */ |
9099 | trans->block_rsv = rsv; | |
8082510e YZ |
9100 | } |
9101 | ||
ddfae63c JB |
9102 | /* |
9103 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9104 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9105 | * we've truncated everything except the last little bit, and can do | |
9106 | * btrfs_truncate_block and then update the disk_i_size. | |
9107 | */ | |
9108 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9109 | btrfs_end_transaction(trans); | |
9110 | btrfs_btree_balance_dirty(fs_info); | |
9111 | ||
9112 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9113 | if (ret) | |
9114 | goto out; | |
9115 | trans = btrfs_start_transaction(root, 1); | |
9116 | if (IS_ERR(trans)) { | |
9117 | ret = PTR_ERR(trans); | |
9118 | goto out; | |
9119 | } | |
9120 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9121 | } | |
9122 | ||
917c16b2 | 9123 | if (trans) { |
ad7e1a74 OS |
9124 | int ret2; |
9125 | ||
0b246afa | 9126 | trans->block_rsv = &fs_info->trans_block_rsv; |
ad7e1a74 OS |
9127 | ret2 = btrfs_update_inode(trans, root, inode); |
9128 | if (ret2 && !ret) | |
9129 | ret = ret2; | |
7b128766 | 9130 | |
ad7e1a74 OS |
9131 | ret2 = btrfs_end_transaction(trans); |
9132 | if (ret2 && !ret) | |
9133 | ret = ret2; | |
2ff7e61e | 9134 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9135 | } |
fcb80c2a | 9136 | out: |
2ff7e61e | 9137 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9138 | |
ad7e1a74 | 9139 | return ret; |
39279cc3 CM |
9140 | } |
9141 | ||
d352ac68 CM |
9142 | /* |
9143 | * create a new subvolume directory/inode (helper for the ioctl). | |
9144 | */ | |
d2fb3437 | 9145 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9146 | struct btrfs_root *new_root, |
9147 | struct btrfs_root *parent_root, | |
9148 | u64 new_dirid) | |
39279cc3 | 9149 | { |
39279cc3 | 9150 | struct inode *inode; |
76dda93c | 9151 | int err; |
00e4e6b3 | 9152 | u64 index = 0; |
39279cc3 | 9153 | |
12fc9d09 FA |
9154 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9155 | new_dirid, new_dirid, | |
9156 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9157 | &index); | |
54aa1f4d | 9158 | if (IS_ERR(inode)) |
f46b5a66 | 9159 | return PTR_ERR(inode); |
39279cc3 CM |
9160 | inode->i_op = &btrfs_dir_inode_operations; |
9161 | inode->i_fop = &btrfs_dir_file_operations; | |
9162 | ||
bfe86848 | 9163 | set_nlink(inode, 1); |
6ef06d27 | 9164 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9165 | unlock_new_inode(inode); |
3b96362c | 9166 | |
63541927 FDBM |
9167 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9168 | if (err) | |
9169 | btrfs_err(new_root->fs_info, | |
351fd353 | 9170 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9171 | new_root->root_key.objectid, err); |
9172 | ||
76dda93c | 9173 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9174 | |
76dda93c | 9175 | iput(inode); |
ce598979 | 9176 | return err; |
39279cc3 CM |
9177 | } |
9178 | ||
39279cc3 CM |
9179 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9180 | { | |
69fe2d75 | 9181 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9182 | struct btrfs_inode *ei; |
2ead6ae7 | 9183 | struct inode *inode; |
39279cc3 | 9184 | |
712e36c5 | 9185 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL); |
39279cc3 CM |
9186 | if (!ei) |
9187 | return NULL; | |
2ead6ae7 YZ |
9188 | |
9189 | ei->root = NULL; | |
2ead6ae7 | 9190 | ei->generation = 0; |
15ee9bc7 | 9191 | ei->last_trans = 0; |
257c62e1 | 9192 | ei->last_sub_trans = 0; |
e02119d5 | 9193 | ei->logged_trans = 0; |
2ead6ae7 | 9194 | ei->delalloc_bytes = 0; |
a7e3b975 | 9195 | ei->new_delalloc_bytes = 0; |
47059d93 | 9196 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9197 | ei->disk_i_size = 0; |
9198 | ei->flags = 0; | |
7709cde3 | 9199 | ei->csum_bytes = 0; |
2ead6ae7 | 9200 | ei->index_cnt = (u64)-1; |
67de1176 | 9201 | ei->dir_index = 0; |
2ead6ae7 | 9202 | ei->last_unlink_trans = 0; |
46d8bc34 | 9203 | ei->last_log_commit = 0; |
2ead6ae7 | 9204 | |
9e0baf60 JB |
9205 | spin_lock_init(&ei->lock); |
9206 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9207 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9208 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9209 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9210 | ei->runtime_flags = 0; |
b52aa8c9 | 9211 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9212 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9213 | |
16cdcec7 MX |
9214 | ei->delayed_node = NULL; |
9215 | ||
9cc97d64 | 9216 | ei->i_otime.tv_sec = 0; |
9217 | ei->i_otime.tv_nsec = 0; | |
9218 | ||
2ead6ae7 | 9219 | inode = &ei->vfs_inode; |
a8067e02 | 9220 | extent_map_tree_init(&ei->extent_tree); |
43eb5f29 QW |
9221 | extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode); |
9222 | extent_io_tree_init(fs_info, &ei->io_failure_tree, | |
9223 | IO_TREE_INODE_IO_FAILURE, inode); | |
7b439738 DS |
9224 | ei->io_tree.track_uptodate = true; |
9225 | ei->io_failure_tree.track_uptodate = true; | |
b812ce28 | 9226 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9227 | mutex_init(&ei->log_mutex); |
f248679e | 9228 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9229 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9230 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9231 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9232 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9233 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9234 | |
9235 | return inode; | |
39279cc3 CM |
9236 | } |
9237 | ||
aaedb55b JB |
9238 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9239 | void btrfs_test_destroy_inode(struct inode *inode) | |
9240 | { | |
dcdbc059 | 9241 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9242 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9243 | } | |
9244 | #endif | |
9245 | ||
26602cab | 9246 | void btrfs_free_inode(struct inode *inode) |
fa0d7e3d | 9247 | { |
fa0d7e3d NP |
9248 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9249 | } | |
9250 | ||
39279cc3 CM |
9251 | void btrfs_destroy_inode(struct inode *inode) |
9252 | { | |
0b246afa | 9253 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9254 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9255 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9256 | ||
b3d9b7a3 | 9257 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9258 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9259 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9260 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9261 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9262 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9263 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9264 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9265 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9266 | |
a6dbd429 JB |
9267 | /* |
9268 | * This can happen where we create an inode, but somebody else also | |
9269 | * created the same inode and we need to destroy the one we already | |
9270 | * created. | |
9271 | */ | |
9272 | if (!root) | |
26602cab | 9273 | return; |
a6dbd429 | 9274 | |
d397712b | 9275 | while (1) { |
e6dcd2dc CM |
9276 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9277 | if (!ordered) | |
9278 | break; | |
9279 | else { | |
0b246afa | 9280 | btrfs_err(fs_info, |
5d163e0e JM |
9281 | "found ordered extent %llu %llu on inode cleanup", |
9282 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9283 | btrfs_remove_ordered_extent(inode, ordered); |
9284 | btrfs_put_ordered_extent(ordered); | |
9285 | btrfs_put_ordered_extent(ordered); | |
9286 | } | |
9287 | } | |
56fa9d07 | 9288 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9289 | inode_tree_del(inode); |
dcdbc059 | 9290 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
39279cc3 CM |
9291 | } |
9292 | ||
45321ac5 | 9293 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9294 | { |
9295 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9296 | |
6379ef9f NA |
9297 | if (root == NULL) |
9298 | return 1; | |
9299 | ||
fa6ac876 | 9300 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9301 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9302 | return 1; |
76dda93c | 9303 | else |
45321ac5 | 9304 | return generic_drop_inode(inode); |
76dda93c YZ |
9305 | } |
9306 | ||
0ee0fda0 | 9307 | static void init_once(void *foo) |
39279cc3 CM |
9308 | { |
9309 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9310 | ||
9311 | inode_init_once(&ei->vfs_inode); | |
9312 | } | |
9313 | ||
e67c718b | 9314 | void __cold btrfs_destroy_cachep(void) |
39279cc3 | 9315 | { |
8c0a8537 KS |
9316 | /* |
9317 | * Make sure all delayed rcu free inodes are flushed before we | |
9318 | * destroy cache. | |
9319 | */ | |
9320 | rcu_barrier(); | |
5598e900 KM |
9321 | kmem_cache_destroy(btrfs_inode_cachep); |
9322 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9323 | kmem_cache_destroy(btrfs_path_cachep); |
9324 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9325 | } |
9326 | ||
f5c29bd9 | 9327 | int __init btrfs_init_cachep(void) |
39279cc3 | 9328 | { |
837e1972 | 9329 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9330 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9331 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9332 | init_once); | |
39279cc3 CM |
9333 | if (!btrfs_inode_cachep) |
9334 | goto fail; | |
9601e3f6 | 9335 | |
837e1972 | 9336 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9337 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9338 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9339 | if (!btrfs_trans_handle_cachep) |
9340 | goto fail; | |
9601e3f6 | 9341 | |
837e1972 | 9342 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9343 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9344 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9345 | if (!btrfs_path_cachep) |
9346 | goto fail; | |
9601e3f6 | 9347 | |
837e1972 | 9348 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9349 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9350 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9351 | if (!btrfs_free_space_cachep) |
9352 | goto fail; | |
9353 | ||
39279cc3 CM |
9354 | return 0; |
9355 | fail: | |
9356 | btrfs_destroy_cachep(); | |
9357 | return -ENOMEM; | |
9358 | } | |
9359 | ||
a528d35e DH |
9360 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9361 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9362 | { |
df0af1a5 | 9363 | u64 delalloc_bytes; |
a528d35e | 9364 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9365 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9366 | u32 bi_flags = BTRFS_I(inode)->flags; |
9367 | ||
9368 | stat->result_mask |= STATX_BTIME; | |
9369 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9370 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9371 | if (bi_flags & BTRFS_INODE_APPEND) | |
9372 | stat->attributes |= STATX_ATTR_APPEND; | |
9373 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9374 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9375 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9376 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9377 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9378 | stat->attributes |= STATX_ATTR_NODUMP; | |
9379 | ||
9380 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9381 | STATX_ATTR_COMPRESSED | | |
9382 | STATX_ATTR_IMMUTABLE | | |
9383 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9384 | |
39279cc3 | 9385 | generic_fillattr(inode, stat); |
0ee5dc67 | 9386 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9387 | |
9388 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9389 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9390 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9391 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9392 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9393 | return 0; |
9394 | } | |
9395 | ||
cdd1fedf DF |
9396 | static int btrfs_rename_exchange(struct inode *old_dir, |
9397 | struct dentry *old_dentry, | |
9398 | struct inode *new_dir, | |
9399 | struct dentry *new_dentry) | |
9400 | { | |
0b246afa | 9401 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9402 | struct btrfs_trans_handle *trans; |
9403 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9404 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9405 | struct inode *new_inode = new_dentry->d_inode; | |
9406 | struct inode *old_inode = old_dentry->d_inode; | |
95582b00 | 9407 | struct timespec64 ctime = current_time(old_inode); |
cdd1fedf | 9408 | struct dentry *parent; |
4a0cc7ca NB |
9409 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9410 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9411 | u64 old_idx = 0; |
9412 | u64 new_idx = 0; | |
9413 | u64 root_objectid; | |
9414 | int ret; | |
86e8aa0e FM |
9415 | bool root_log_pinned = false; |
9416 | bool dest_log_pinned = false; | |
d4682ba0 FM |
9417 | struct btrfs_log_ctx ctx_root; |
9418 | struct btrfs_log_ctx ctx_dest; | |
9419 | bool sync_log_root = false; | |
9420 | bool sync_log_dest = false; | |
9421 | bool commit_transaction = false; | |
cdd1fedf DF |
9422 | |
9423 | /* we only allow rename subvolume link between subvolumes */ | |
9424 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9425 | return -EXDEV; | |
9426 | ||
d4682ba0 FM |
9427 | btrfs_init_log_ctx(&ctx_root, old_inode); |
9428 | btrfs_init_log_ctx(&ctx_dest, new_inode); | |
9429 | ||
cdd1fedf DF |
9430 | /* close the race window with snapshot create/destroy ioctl */ |
9431 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9432 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9433 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9434 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9435 | |
9436 | /* | |
9437 | * We want to reserve the absolute worst case amount of items. So if | |
9438 | * both inodes are subvols and we need to unlink them then that would | |
9439 | * require 4 item modifications, but if they are both normal inodes it | |
9440 | * would require 5 item modifications, so we'll assume their normal | |
9441 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9442 | * should cover the worst case number of items we'll modify. | |
9443 | */ | |
9444 | trans = btrfs_start_transaction(root, 12); | |
9445 | if (IS_ERR(trans)) { | |
9446 | ret = PTR_ERR(trans); | |
9447 | goto out_notrans; | |
9448 | } | |
9449 | ||
9450 | /* | |
9451 | * We need to find a free sequence number both in the source and | |
9452 | * in the destination directory for the exchange. | |
9453 | */ | |
877574e2 | 9454 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9455 | if (ret) |
9456 | goto out_fail; | |
877574e2 | 9457 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9458 | if (ret) |
9459 | goto out_fail; | |
9460 | ||
9461 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9462 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9463 | ||
9464 | /* Reference for the source. */ | |
9465 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9466 | /* force full log commit if subvolume involved. */ | |
90787766 | 9467 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9468 | } else { |
376e5a57 FM |
9469 | btrfs_pin_log_trans(root); |
9470 | root_log_pinned = true; | |
cdd1fedf DF |
9471 | ret = btrfs_insert_inode_ref(trans, dest, |
9472 | new_dentry->d_name.name, | |
9473 | new_dentry->d_name.len, | |
9474 | old_ino, | |
f85b7379 DS |
9475 | btrfs_ino(BTRFS_I(new_dir)), |
9476 | old_idx); | |
cdd1fedf DF |
9477 | if (ret) |
9478 | goto out_fail; | |
cdd1fedf DF |
9479 | } |
9480 | ||
9481 | /* And now for the dest. */ | |
9482 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9483 | /* force full log commit if subvolume involved. */ | |
90787766 | 9484 | btrfs_set_log_full_commit(trans); |
cdd1fedf | 9485 | } else { |
376e5a57 FM |
9486 | btrfs_pin_log_trans(dest); |
9487 | dest_log_pinned = true; | |
cdd1fedf DF |
9488 | ret = btrfs_insert_inode_ref(trans, root, |
9489 | old_dentry->d_name.name, | |
9490 | old_dentry->d_name.len, | |
9491 | new_ino, | |
f85b7379 DS |
9492 | btrfs_ino(BTRFS_I(old_dir)), |
9493 | new_idx); | |
cdd1fedf DF |
9494 | if (ret) |
9495 | goto out_fail; | |
cdd1fedf DF |
9496 | } |
9497 | ||
9498 | /* Update inode version and ctime/mtime. */ | |
9499 | inode_inc_iversion(old_dir); | |
9500 | inode_inc_iversion(new_dir); | |
9501 | inode_inc_iversion(old_inode); | |
9502 | inode_inc_iversion(new_inode); | |
9503 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9504 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9505 | old_inode->i_ctime = ctime; | |
9506 | new_inode->i_ctime = ctime; | |
9507 | ||
9508 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9509 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9510 | BTRFS_I(old_inode), 1); | |
9511 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9512 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9513 | } |
9514 | ||
9515 | /* src is a subvolume */ | |
9516 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9517 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
401b3b19 | 9518 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
cdd1fedf DF |
9519 | old_dentry->d_name.name, |
9520 | old_dentry->d_name.len); | |
9521 | } else { /* src is an inode */ | |
4ec5934e NB |
9522 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9523 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9524 | old_dentry->d_name.name, |
9525 | old_dentry->d_name.len); | |
9526 | if (!ret) | |
9527 | ret = btrfs_update_inode(trans, root, old_inode); | |
9528 | } | |
9529 | if (ret) { | |
66642832 | 9530 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9531 | goto out_fail; |
9532 | } | |
9533 | ||
9534 | /* dest is a subvolume */ | |
9535 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9536 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
401b3b19 | 9537 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
cdd1fedf DF |
9538 | new_dentry->d_name.name, |
9539 | new_dentry->d_name.len); | |
9540 | } else { /* dest is an inode */ | |
4ec5934e NB |
9541 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9542 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9543 | new_dentry->d_name.name, |
9544 | new_dentry->d_name.len); | |
9545 | if (!ret) | |
9546 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9547 | } | |
9548 | if (ret) { | |
66642832 | 9549 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9550 | goto out_fail; |
9551 | } | |
9552 | ||
db0a669f | 9553 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9554 | new_dentry->d_name.name, |
9555 | new_dentry->d_name.len, 0, old_idx); | |
9556 | if (ret) { | |
66642832 | 9557 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9558 | goto out_fail; |
9559 | } | |
9560 | ||
db0a669f | 9561 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9562 | old_dentry->d_name.name, |
9563 | old_dentry->d_name.len, 0, new_idx); | |
9564 | if (ret) { | |
66642832 | 9565 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9566 | goto out_fail; |
9567 | } | |
9568 | ||
9569 | if (old_inode->i_nlink == 1) | |
9570 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9571 | if (new_inode->i_nlink == 1) | |
9572 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9573 | ||
86e8aa0e | 9574 | if (root_log_pinned) { |
cdd1fedf | 9575 | parent = new_dentry->d_parent; |
d4682ba0 FM |
9576 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), |
9577 | BTRFS_I(old_dir), parent, | |
9578 | false, &ctx_root); | |
9579 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9580 | sync_log_root = true; | |
9581 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9582 | commit_transaction = true; | |
9583 | ret = 0; | |
cdd1fedf | 9584 | btrfs_end_log_trans(root); |
86e8aa0e | 9585 | root_log_pinned = false; |
cdd1fedf | 9586 | } |
86e8aa0e | 9587 | if (dest_log_pinned) { |
d4682ba0 FM |
9588 | if (!commit_transaction) { |
9589 | parent = old_dentry->d_parent; | |
9590 | ret = btrfs_log_new_name(trans, BTRFS_I(new_inode), | |
9591 | BTRFS_I(new_dir), parent, | |
9592 | false, &ctx_dest); | |
9593 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9594 | sync_log_dest = true; | |
9595 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9596 | commit_transaction = true; | |
9597 | ret = 0; | |
9598 | } | |
cdd1fedf | 9599 | btrfs_end_log_trans(dest); |
86e8aa0e | 9600 | dest_log_pinned = false; |
cdd1fedf DF |
9601 | } |
9602 | out_fail: | |
86e8aa0e FM |
9603 | /* |
9604 | * If we have pinned a log and an error happened, we unpin tasks | |
9605 | * trying to sync the log and force them to fallback to a transaction | |
9606 | * commit if the log currently contains any of the inodes involved in | |
9607 | * this rename operation (to ensure we do not persist a log with an | |
9608 | * inconsistent state for any of these inodes or leading to any | |
9609 | * inconsistencies when replayed). If the transaction was aborted, the | |
9610 | * abortion reason is propagated to userspace when attempting to commit | |
9611 | * the transaction. If the log does not contain any of these inodes, we | |
9612 | * allow the tasks to sync it. | |
9613 | */ | |
9614 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9615 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9616 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9617 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9618 | (new_inode && |
0f8939b8 | 9619 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9620 | btrfs_set_log_full_commit(trans); |
86e8aa0e FM |
9621 | |
9622 | if (root_log_pinned) { | |
9623 | btrfs_end_log_trans(root); | |
9624 | root_log_pinned = false; | |
9625 | } | |
9626 | if (dest_log_pinned) { | |
9627 | btrfs_end_log_trans(dest); | |
9628 | dest_log_pinned = false; | |
9629 | } | |
9630 | } | |
d4682ba0 FM |
9631 | if (!ret && sync_log_root && !commit_transaction) { |
9632 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, | |
9633 | &ctx_root); | |
9634 | if (ret) | |
9635 | commit_transaction = true; | |
9636 | } | |
9637 | if (!ret && sync_log_dest && !commit_transaction) { | |
9638 | ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root, | |
9639 | &ctx_dest); | |
9640 | if (ret) | |
9641 | commit_transaction = true; | |
9642 | } | |
9643 | if (commit_transaction) { | |
9644 | ret = btrfs_commit_transaction(trans); | |
9645 | } else { | |
9646 | int ret2; | |
9647 | ||
9648 | ret2 = btrfs_end_transaction(trans); | |
9649 | ret = ret ? ret : ret2; | |
9650 | } | |
cdd1fedf DF |
9651 | out_notrans: |
9652 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9653 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9654 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9655 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9656 | |
9657 | return ret; | |
9658 | } | |
9659 | ||
9660 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9661 | struct btrfs_root *root, | |
9662 | struct inode *dir, | |
9663 | struct dentry *dentry) | |
9664 | { | |
9665 | int ret; | |
9666 | struct inode *inode; | |
9667 | u64 objectid; | |
9668 | u64 index; | |
9669 | ||
9670 | ret = btrfs_find_free_ino(root, &objectid); | |
9671 | if (ret) | |
9672 | return ret; | |
9673 | ||
9674 | inode = btrfs_new_inode(trans, root, dir, | |
9675 | dentry->d_name.name, | |
9676 | dentry->d_name.len, | |
4a0cc7ca | 9677 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9678 | objectid, |
9679 | S_IFCHR | WHITEOUT_MODE, | |
9680 | &index); | |
9681 | ||
9682 | if (IS_ERR(inode)) { | |
9683 | ret = PTR_ERR(inode); | |
9684 | return ret; | |
9685 | } | |
9686 | ||
9687 | inode->i_op = &btrfs_special_inode_operations; | |
9688 | init_special_inode(inode, inode->i_mode, | |
9689 | WHITEOUT_DEV); | |
9690 | ||
9691 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9692 | &dentry->d_name); | |
9693 | if (ret) | |
c9901618 | 9694 | goto out; |
cdd1fedf | 9695 | |
cef415af NB |
9696 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9697 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9698 | if (ret) |
c9901618 | 9699 | goto out; |
cdd1fedf DF |
9700 | |
9701 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9702 | out: |
cdd1fedf | 9703 | unlock_new_inode(inode); |
c9901618 FM |
9704 | if (ret) |
9705 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9706 | iput(inode); |
9707 | ||
c9901618 | 9708 | return ret; |
cdd1fedf DF |
9709 | } |
9710 | ||
d397712b | 9711 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9712 | struct inode *new_dir, struct dentry *new_dentry, |
9713 | unsigned int flags) | |
39279cc3 | 9714 | { |
0b246afa | 9715 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9716 | struct btrfs_trans_handle *trans; |
5062af35 | 9717 | unsigned int trans_num_items; |
39279cc3 | 9718 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9719 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9720 | struct inode *new_inode = d_inode(new_dentry); |
9721 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9722 | u64 index = 0; |
4df27c4d | 9723 | u64 root_objectid; |
39279cc3 | 9724 | int ret; |
4a0cc7ca | 9725 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9726 | bool log_pinned = false; |
d4682ba0 FM |
9727 | struct btrfs_log_ctx ctx; |
9728 | bool sync_log = false; | |
9729 | bool commit_transaction = false; | |
39279cc3 | 9730 | |
4a0cc7ca | 9731 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9732 | return -EPERM; |
9733 | ||
4df27c4d | 9734 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9735 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9736 | return -EXDEV; |
9737 | ||
33345d01 | 9738 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9739 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9740 | return -ENOTEMPTY; |
5f39d397 | 9741 | |
4df27c4d YZ |
9742 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9743 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9744 | return -ENOTEMPTY; | |
9c52057c CM |
9745 | |
9746 | ||
9747 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9748 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9749 | new_dentry->d_name.name, |
9750 | new_dentry->d_name.len); | |
9751 | ||
9752 | if (ret) { | |
9753 | if (ret == -EEXIST) { | |
9754 | /* we shouldn't get | |
9755 | * eexist without a new_inode */ | |
fae7f21c | 9756 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9757 | return ret; |
9758 | } | |
9759 | } else { | |
9760 | /* maybe -EOVERFLOW */ | |
9761 | return ret; | |
9762 | } | |
9763 | } | |
9764 | ret = 0; | |
9765 | ||
5a3f23d5 | 9766 | /* |
8d875f95 CM |
9767 | * we're using rename to replace one file with another. Start IO on it |
9768 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9769 | */ |
8d875f95 | 9770 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9771 | filemap_flush(old_inode->i_mapping); |
9772 | ||
76dda93c | 9773 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9774 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9775 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9776 | /* |
9777 | * We want to reserve the absolute worst case amount of items. So if | |
9778 | * both inodes are subvols and we need to unlink them then that would | |
9779 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9780 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9781 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9782 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9783 | * If our rename has the whiteout flag, we need more 5 units for the |
9784 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9785 | * when selinux is enabled). | |
a22285a6 | 9786 | */ |
5062af35 FM |
9787 | trans_num_items = 11; |
9788 | if (flags & RENAME_WHITEOUT) | |
9789 | trans_num_items += 5; | |
9790 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9791 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9792 | ret = PTR_ERR(trans); |
9793 | goto out_notrans; | |
9794 | } | |
76dda93c | 9795 | |
4df27c4d YZ |
9796 | if (dest != root) |
9797 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9798 | |
877574e2 | 9799 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9800 | if (ret) |
9801 | goto out_fail; | |
5a3f23d5 | 9802 | |
67de1176 | 9803 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9804 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9805 | /* force full log commit if subvolume involved. */ |
90787766 | 9806 | btrfs_set_log_full_commit(trans); |
4df27c4d | 9807 | } else { |
c4aba954 FM |
9808 | btrfs_pin_log_trans(root); |
9809 | log_pinned = true; | |
a5719521 YZ |
9810 | ret = btrfs_insert_inode_ref(trans, dest, |
9811 | new_dentry->d_name.name, | |
9812 | new_dentry->d_name.len, | |
33345d01 | 9813 | old_ino, |
4a0cc7ca | 9814 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9815 | if (ret) |
9816 | goto out_fail; | |
4df27c4d | 9817 | } |
5a3f23d5 | 9818 | |
0c4d2d95 JB |
9819 | inode_inc_iversion(old_dir); |
9820 | inode_inc_iversion(new_dir); | |
9821 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9822 | old_dir->i_ctime = old_dir->i_mtime = |
9823 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9824 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9825 | |
12fcfd22 | 9826 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9827 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9828 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9829 | |
33345d01 | 9830 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9831 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
401b3b19 | 9832 | ret = btrfs_unlink_subvol(trans, old_dir, root_objectid, |
4df27c4d YZ |
9833 | old_dentry->d_name.name, |
9834 | old_dentry->d_name.len); | |
9835 | } else { | |
4ec5934e NB |
9836 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9837 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9838 | old_dentry->d_name.name, |
9839 | old_dentry->d_name.len); | |
9840 | if (!ret) | |
9841 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9842 | } |
79787eaa | 9843 | if (ret) { |
66642832 | 9844 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9845 | goto out_fail; |
9846 | } | |
39279cc3 CM |
9847 | |
9848 | if (new_inode) { | |
0c4d2d95 | 9849 | inode_inc_iversion(new_inode); |
c2050a45 | 9850 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9851 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9852 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9853 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
401b3b19 | 9854 | ret = btrfs_unlink_subvol(trans, new_dir, root_objectid, |
4df27c4d YZ |
9855 | new_dentry->d_name.name, |
9856 | new_dentry->d_name.len); | |
9857 | BUG_ON(new_inode->i_nlink == 0); | |
9858 | } else { | |
4ec5934e NB |
9859 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9860 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9861 | new_dentry->d_name.name, |
9862 | new_dentry->d_name.len); | |
9863 | } | |
4ef31a45 | 9864 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9865 | ret = btrfs_orphan_add(trans, |
9866 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9867 | if (ret) { |
66642832 | 9868 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9869 | goto out_fail; |
9870 | } | |
39279cc3 | 9871 | } |
aec7477b | 9872 | |
db0a669f | 9873 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9874 | new_dentry->d_name.name, |
a5719521 | 9875 | new_dentry->d_name.len, 0, index); |
79787eaa | 9876 | if (ret) { |
66642832 | 9877 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9878 | goto out_fail; |
9879 | } | |
39279cc3 | 9880 | |
67de1176 MX |
9881 | if (old_inode->i_nlink == 1) |
9882 | BTRFS_I(old_inode)->dir_index = index; | |
9883 | ||
3dc9e8f7 | 9884 | if (log_pinned) { |
10d9f309 | 9885 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9886 | |
d4682ba0 FM |
9887 | btrfs_init_log_ctx(&ctx, old_inode); |
9888 | ret = btrfs_log_new_name(trans, BTRFS_I(old_inode), | |
9889 | BTRFS_I(old_dir), parent, | |
9890 | false, &ctx); | |
9891 | if (ret == BTRFS_NEED_LOG_SYNC) | |
9892 | sync_log = true; | |
9893 | else if (ret == BTRFS_NEED_TRANS_COMMIT) | |
9894 | commit_transaction = true; | |
9895 | ret = 0; | |
4df27c4d | 9896 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9897 | log_pinned = false; |
4df27c4d | 9898 | } |
cdd1fedf DF |
9899 | |
9900 | if (flags & RENAME_WHITEOUT) { | |
9901 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9902 | old_dentry); | |
9903 | ||
9904 | if (ret) { | |
66642832 | 9905 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9906 | goto out_fail; |
9907 | } | |
4df27c4d | 9908 | } |
39279cc3 | 9909 | out_fail: |
3dc9e8f7 FM |
9910 | /* |
9911 | * If we have pinned the log and an error happened, we unpin tasks | |
9912 | * trying to sync the log and force them to fallback to a transaction | |
9913 | * commit if the log currently contains any of the inodes involved in | |
9914 | * this rename operation (to ensure we do not persist a log with an | |
9915 | * inconsistent state for any of these inodes or leading to any | |
9916 | * inconsistencies when replayed). If the transaction was aborted, the | |
9917 | * abortion reason is propagated to userspace when attempting to commit | |
9918 | * the transaction. If the log does not contain any of these inodes, we | |
9919 | * allow the tasks to sync it. | |
9920 | */ | |
9921 | if (ret && log_pinned) { | |
0f8939b8 NB |
9922 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9923 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9924 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9925 | (new_inode && |
0f8939b8 | 9926 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
90787766 | 9927 | btrfs_set_log_full_commit(trans); |
3dc9e8f7 FM |
9928 | |
9929 | btrfs_end_log_trans(root); | |
9930 | log_pinned = false; | |
9931 | } | |
d4682ba0 FM |
9932 | if (!ret && sync_log) { |
9933 | ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx); | |
9934 | if (ret) | |
9935 | commit_transaction = true; | |
9936 | } | |
9937 | if (commit_transaction) { | |
9938 | ret = btrfs_commit_transaction(trans); | |
9939 | } else { | |
9940 | int ret2; | |
9941 | ||
9942 | ret2 = btrfs_end_transaction(trans); | |
9943 | ret = ret ? ret : ret2; | |
9944 | } | |
b44c59a8 | 9945 | out_notrans: |
33345d01 | 9946 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9947 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9948 | |
39279cc3 CM |
9949 | return ret; |
9950 | } | |
9951 | ||
80ace85c MS |
9952 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9953 | struct inode *new_dir, struct dentry *new_dentry, | |
9954 | unsigned int flags) | |
9955 | { | |
cdd1fedf | 9956 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9957 | return -EINVAL; |
9958 | ||
cdd1fedf DF |
9959 | if (flags & RENAME_EXCHANGE) |
9960 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9961 | new_dentry); | |
9962 | ||
9963 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9964 | } |
9965 | ||
3a2f8c07 NB |
9966 | struct btrfs_delalloc_work { |
9967 | struct inode *inode; | |
9968 | struct completion completion; | |
9969 | struct list_head list; | |
9970 | struct btrfs_work work; | |
9971 | }; | |
9972 | ||
8ccf6f19 MX |
9973 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9974 | { | |
9975 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9976 | struct inode *inode; |
8ccf6f19 MX |
9977 | |
9978 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9979 | work); | |
9f23e289 | 9980 | inode = delalloc_work->inode; |
30424601 DS |
9981 | filemap_flush(inode->i_mapping); |
9982 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9983 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9984 | filemap_flush(inode->i_mapping); |
8ccf6f19 | 9985 | |
076da91c | 9986 | iput(inode); |
8ccf6f19 MX |
9987 | complete(&delalloc_work->completion); |
9988 | } | |
9989 | ||
3a2f8c07 | 9990 | static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode) |
8ccf6f19 MX |
9991 | { |
9992 | struct btrfs_delalloc_work *work; | |
9993 | ||
100d5702 | 9994 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9995 | if (!work) |
9996 | return NULL; | |
9997 | ||
9998 | init_completion(&work->completion); | |
9999 | INIT_LIST_HEAD(&work->list); | |
10000 | work->inode = inode; | |
9e0af237 LB |
10001 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, |
10002 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10003 | |
10004 | return work; | |
10005 | } | |
10006 | ||
d352ac68 CM |
10007 | /* |
10008 | * some fairly slow code that needs optimization. This walks the list | |
10009 | * of all the inodes with pending delalloc and forces them to disk. | |
10010 | */ | |
3cd24c69 | 10011 | static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot) |
ea8c2819 | 10012 | { |
ea8c2819 | 10013 | struct btrfs_inode *binode; |
5b21f2ed | 10014 | struct inode *inode; |
8ccf6f19 MX |
10015 | struct btrfs_delalloc_work *work, *next; |
10016 | struct list_head works; | |
1eafa6c7 | 10017 | struct list_head splice; |
8ccf6f19 | 10018 | int ret = 0; |
ea8c2819 | 10019 | |
8ccf6f19 | 10020 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10021 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10022 | |
573bfb72 | 10023 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10024 | spin_lock(&root->delalloc_lock); |
10025 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10026 | while (!list_empty(&splice)) { |
10027 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10028 | delalloc_inodes); |
1eafa6c7 | 10029 | |
eb73c1b7 MX |
10030 | list_move_tail(&binode->delalloc_inodes, |
10031 | &root->delalloc_inodes); | |
5b21f2ed | 10032 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10033 | if (!inode) { |
eb73c1b7 | 10034 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10035 | continue; |
df0af1a5 | 10036 | } |
eb73c1b7 | 10037 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10038 | |
3cd24c69 EL |
10039 | if (snapshot) |
10040 | set_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
10041 | &binode->runtime_flags); | |
076da91c | 10042 | work = btrfs_alloc_delalloc_work(inode); |
5d99a998 | 10043 | if (!work) { |
4fbb5147 | 10044 | iput(inode); |
1eafa6c7 | 10045 | ret = -ENOMEM; |
a1ecaabb | 10046 | goto out; |
5b21f2ed | 10047 | } |
1eafa6c7 | 10048 | list_add_tail(&work->list, &works); |
a44903ab QW |
10049 | btrfs_queue_work(root->fs_info->flush_workers, |
10050 | &work->work); | |
6c255e67 MX |
10051 | ret++; |
10052 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10053 | goto out; |
5b21f2ed | 10054 | cond_resched(); |
eb73c1b7 | 10055 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10056 | } |
eb73c1b7 | 10057 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10058 | |
a1ecaabb | 10059 | out: |
eb73c1b7 MX |
10060 | list_for_each_entry_safe(work, next, &works, list) { |
10061 | list_del_init(&work->list); | |
40012f96 NB |
10062 | wait_for_completion(&work->completion); |
10063 | kfree(work); | |
eb73c1b7 MX |
10064 | } |
10065 | ||
81f1d390 | 10066 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10067 | spin_lock(&root->delalloc_lock); |
10068 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10069 | spin_unlock(&root->delalloc_lock); | |
10070 | } | |
573bfb72 | 10071 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10072 | return ret; |
10073 | } | |
1eafa6c7 | 10074 | |
3cd24c69 | 10075 | int btrfs_start_delalloc_snapshot(struct btrfs_root *root) |
eb73c1b7 | 10076 | { |
0b246afa | 10077 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10078 | int ret; |
1eafa6c7 | 10079 | |
0b246afa | 10080 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10081 | return -EROFS; |
10082 | ||
3cd24c69 | 10083 | ret = start_delalloc_inodes(root, -1, true); |
6c255e67 MX |
10084 | if (ret > 0) |
10085 | ret = 0; | |
eb73c1b7 MX |
10086 | return ret; |
10087 | } | |
10088 | ||
82b3e53b | 10089 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr) |
eb73c1b7 MX |
10090 | { |
10091 | struct btrfs_root *root; | |
10092 | struct list_head splice; | |
10093 | int ret; | |
10094 | ||
2c21b4d7 | 10095 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10096 | return -EROFS; |
10097 | ||
10098 | INIT_LIST_HEAD(&splice); | |
10099 | ||
573bfb72 | 10100 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10101 | spin_lock(&fs_info->delalloc_root_lock); |
10102 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10103 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10104 | root = list_first_entry(&splice, struct btrfs_root, |
10105 | delalloc_root); | |
10106 | root = btrfs_grab_fs_root(root); | |
10107 | BUG_ON(!root); | |
10108 | list_move_tail(&root->delalloc_root, | |
10109 | &fs_info->delalloc_roots); | |
10110 | spin_unlock(&fs_info->delalloc_root_lock); | |
10111 | ||
3cd24c69 | 10112 | ret = start_delalloc_inodes(root, nr, false); |
eb73c1b7 | 10113 | btrfs_put_fs_root(root); |
6c255e67 | 10114 | if (ret < 0) |
eb73c1b7 MX |
10115 | goto out; |
10116 | ||
6c255e67 MX |
10117 | if (nr != -1) { |
10118 | nr -= ret; | |
10119 | WARN_ON(nr < 0); | |
10120 | } | |
eb73c1b7 | 10121 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10122 | } |
eb73c1b7 | 10123 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10124 | |
6c255e67 | 10125 | ret = 0; |
eb73c1b7 | 10126 | out: |
81f1d390 | 10127 | if (!list_empty(&splice)) { |
eb73c1b7 MX |
10128 | spin_lock(&fs_info->delalloc_root_lock); |
10129 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10130 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10131 | } |
573bfb72 | 10132 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10133 | return ret; |
ea8c2819 CM |
10134 | } |
10135 | ||
39279cc3 CM |
10136 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10137 | const char *symname) | |
10138 | { | |
0b246afa | 10139 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10140 | struct btrfs_trans_handle *trans; |
10141 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10142 | struct btrfs_path *path; | |
10143 | struct btrfs_key key; | |
1832a6d5 | 10144 | struct inode *inode = NULL; |
39279cc3 | 10145 | int err; |
39279cc3 | 10146 | u64 objectid; |
67871254 | 10147 | u64 index = 0; |
39279cc3 CM |
10148 | int name_len; |
10149 | int datasize; | |
5f39d397 | 10150 | unsigned long ptr; |
39279cc3 | 10151 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10152 | struct extent_buffer *leaf; |
39279cc3 | 10153 | |
f06becc4 | 10154 | name_len = strlen(symname); |
0b246afa | 10155 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10156 | return -ENAMETOOLONG; |
1832a6d5 | 10157 | |
9ed74f2d JB |
10158 | /* |
10159 | * 2 items for inode item and ref | |
10160 | * 2 items for dir items | |
9269d12b FM |
10161 | * 1 item for updating parent inode item |
10162 | * 1 item for the inline extent item | |
9ed74f2d JB |
10163 | * 1 item for xattr if selinux is on |
10164 | */ | |
9269d12b | 10165 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10166 | if (IS_ERR(trans)) |
10167 | return PTR_ERR(trans); | |
1832a6d5 | 10168 | |
581bb050 LZ |
10169 | err = btrfs_find_free_ino(root, &objectid); |
10170 | if (err) | |
10171 | goto out_unlock; | |
10172 | ||
aec7477b | 10173 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10174 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10175 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10176 | if (IS_ERR(inode)) { |
10177 | err = PTR_ERR(inode); | |
32955c54 | 10178 | inode = NULL; |
39279cc3 | 10179 | goto out_unlock; |
7cf96da3 | 10180 | } |
39279cc3 | 10181 | |
ad19db71 CS |
10182 | /* |
10183 | * If the active LSM wants to access the inode during | |
10184 | * d_instantiate it needs these. Smack checks to see | |
10185 | * if the filesystem supports xattrs by looking at the | |
10186 | * ops vector. | |
10187 | */ | |
10188 | inode->i_fop = &btrfs_file_operations; | |
10189 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10190 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10191 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10192 | ||
10193 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10194 | if (err) | |
32955c54 | 10195 | goto out_unlock; |
ad19db71 | 10196 | |
39279cc3 | 10197 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10198 | if (!path) { |
10199 | err = -ENOMEM; | |
32955c54 | 10200 | goto out_unlock; |
d8926bb3 | 10201 | } |
4a0cc7ca | 10202 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10203 | key.offset = 0; |
962a298f | 10204 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10205 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10206 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10207 | datasize); | |
54aa1f4d | 10208 | if (err) { |
b0839166 | 10209 | btrfs_free_path(path); |
32955c54 | 10210 | goto out_unlock; |
54aa1f4d | 10211 | } |
5f39d397 CM |
10212 | leaf = path->nodes[0]; |
10213 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10214 | struct btrfs_file_extent_item); | |
10215 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10216 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10217 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10218 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10219 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10220 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10221 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10222 | ||
39279cc3 | 10223 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10224 | write_extent_buffer(leaf, symname, ptr, name_len); |
10225 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10226 | btrfs_free_path(path); |
5f39d397 | 10227 | |
39279cc3 | 10228 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10229 | inode_nohighmem(inode); |
d899e052 | 10230 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10231 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10232 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10233 | /* |
10234 | * Last step, add directory indexes for our symlink inode. This is the | |
10235 | * last step to avoid extra cleanup of these indexes if an error happens | |
10236 | * elsewhere above. | |
10237 | */ | |
10238 | if (!err) | |
cef415af NB |
10239 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10240 | BTRFS_I(inode), 0, index); | |
32955c54 AV |
10241 | if (err) |
10242 | goto out_unlock; | |
b0d5d10f | 10243 | |
1e2e547a | 10244 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10245 | |
10246 | out_unlock: | |
3a45bb20 | 10247 | btrfs_end_transaction(trans); |
32955c54 | 10248 | if (err && inode) { |
39279cc3 | 10249 | inode_dec_link_count(inode); |
32955c54 | 10250 | discard_new_inode(inode); |
39279cc3 | 10251 | } |
2ff7e61e | 10252 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
10253 | return err; |
10254 | } | |
16432985 | 10255 | |
0af3d00b JB |
10256 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10257 | u64 start, u64 num_bytes, u64 min_size, | |
10258 | loff_t actual_len, u64 *alloc_hint, | |
10259 | struct btrfs_trans_handle *trans) | |
d899e052 | 10260 | { |
0b246afa | 10261 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10262 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10263 | struct extent_map *em; | |
d899e052 YZ |
10264 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10265 | struct btrfs_key ins; | |
d899e052 | 10266 | u64 cur_offset = start; |
55a61d1d | 10267 | u64 i_size; |
154ea289 | 10268 | u64 cur_bytes; |
0b670dc4 | 10269 | u64 last_alloc = (u64)-1; |
d899e052 | 10270 | int ret = 0; |
0af3d00b | 10271 | bool own_trans = true; |
18513091 | 10272 | u64 end = start + num_bytes - 1; |
d899e052 | 10273 | |
0af3d00b JB |
10274 | if (trans) |
10275 | own_trans = false; | |
d899e052 | 10276 | while (num_bytes > 0) { |
0af3d00b JB |
10277 | if (own_trans) { |
10278 | trans = btrfs_start_transaction(root, 3); | |
10279 | if (IS_ERR(trans)) { | |
10280 | ret = PTR_ERR(trans); | |
10281 | break; | |
10282 | } | |
5a303d5d YZ |
10283 | } |
10284 | ||
ee22184b | 10285 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10286 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10287 | /* |
10288 | * If we are severely fragmented we could end up with really | |
10289 | * small allocations, so if the allocator is returning small | |
10290 | * chunks lets make its job easier by only searching for those | |
10291 | * sized chunks. | |
10292 | */ | |
10293 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10294 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10295 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10296 | if (ret) { |
0af3d00b | 10297 | if (own_trans) |
3a45bb20 | 10298 | btrfs_end_transaction(trans); |
a22285a6 | 10299 | break; |
d899e052 | 10300 | } |
0b246afa | 10301 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10302 | |
0b670dc4 | 10303 | last_alloc = ins.offset; |
d899e052 YZ |
10304 | ret = insert_reserved_file_extent(trans, inode, |
10305 | cur_offset, ins.objectid, | |
10306 | ins.offset, ins.offset, | |
920bbbfb | 10307 | ins.offset, 0, 0, 0, |
d899e052 | 10308 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10309 | if (ret) { |
2ff7e61e | 10310 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10311 | ins.offset, 0); |
66642832 | 10312 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10313 | if (own_trans) |
3a45bb20 | 10314 | btrfs_end_transaction(trans); |
79787eaa JM |
10315 | break; |
10316 | } | |
31193213 | 10317 | |
dcdbc059 | 10318 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10319 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10320 | |
5dc562c5 JB |
10321 | em = alloc_extent_map(); |
10322 | if (!em) { | |
10323 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10324 | &BTRFS_I(inode)->runtime_flags); | |
10325 | goto next; | |
10326 | } | |
10327 | ||
10328 | em->start = cur_offset; | |
10329 | em->orig_start = cur_offset; | |
10330 | em->len = ins.offset; | |
10331 | em->block_start = ins.objectid; | |
10332 | em->block_len = ins.offset; | |
b4939680 | 10333 | em->orig_block_len = ins.offset; |
cc95bef6 | 10334 | em->ram_bytes = ins.offset; |
0b246afa | 10335 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10336 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10337 | em->generation = trans->transid; | |
10338 | ||
10339 | while (1) { | |
10340 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10341 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10342 | write_unlock(&em_tree->lock); |
10343 | if (ret != -EEXIST) | |
10344 | break; | |
dcdbc059 | 10345 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10346 | cur_offset + ins.offset - 1, |
10347 | 0); | |
10348 | } | |
10349 | free_extent_map(em); | |
10350 | next: | |
d899e052 YZ |
10351 | num_bytes -= ins.offset; |
10352 | cur_offset += ins.offset; | |
efa56464 | 10353 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10354 | |
0c4d2d95 | 10355 | inode_inc_iversion(inode); |
c2050a45 | 10356 | inode->i_ctime = current_time(inode); |
6cbff00f | 10357 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10358 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10359 | (actual_len > inode->i_size) && |
10360 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10361 | if (cur_offset > actual_len) |
55a61d1d | 10362 | i_size = actual_len; |
d1ea6a61 | 10363 | else |
55a61d1d JB |
10364 | i_size = cur_offset; |
10365 | i_size_write(inode, i_size); | |
10366 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10367 | } |
10368 | ||
d899e052 | 10369 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10370 | |
10371 | if (ret) { | |
66642832 | 10372 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10373 | if (own_trans) |
3a45bb20 | 10374 | btrfs_end_transaction(trans); |
79787eaa JM |
10375 | break; |
10376 | } | |
d899e052 | 10377 | |
0af3d00b | 10378 | if (own_trans) |
3a45bb20 | 10379 | btrfs_end_transaction(trans); |
5a303d5d | 10380 | } |
18513091 | 10381 | if (cur_offset < end) |
bc42bda2 | 10382 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10383 | end - cur_offset + 1); |
d899e052 YZ |
10384 | return ret; |
10385 | } | |
10386 | ||
0af3d00b JB |
10387 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10388 | u64 start, u64 num_bytes, u64 min_size, | |
10389 | loff_t actual_len, u64 *alloc_hint) | |
10390 | { | |
10391 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10392 | min_size, actual_len, alloc_hint, | |
10393 | NULL); | |
10394 | } | |
10395 | ||
10396 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10397 | struct btrfs_trans_handle *trans, int mode, | |
10398 | u64 start, u64 num_bytes, u64 min_size, | |
10399 | loff_t actual_len, u64 *alloc_hint) | |
10400 | { | |
10401 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10402 | min_size, actual_len, alloc_hint, trans); | |
10403 | } | |
10404 | ||
e6dcd2dc CM |
10405 | static int btrfs_set_page_dirty(struct page *page) |
10406 | { | |
e6dcd2dc CM |
10407 | return __set_page_dirty_nobuffers(page); |
10408 | } | |
10409 | ||
10556cb2 | 10410 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10411 | { |
b83cc969 | 10412 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10413 | umode_t mode = inode->i_mode; |
b83cc969 | 10414 | |
cb6db4e5 JM |
10415 | if (mask & MAY_WRITE && |
10416 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10417 | if (btrfs_root_readonly(root)) | |
10418 | return -EROFS; | |
10419 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10420 | return -EACCES; | |
10421 | } | |
2830ba7f | 10422 | return generic_permission(inode, mask); |
fdebe2bd | 10423 | } |
39279cc3 | 10424 | |
ef3b9af5 FM |
10425 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10426 | { | |
2ff7e61e | 10427 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10428 | struct btrfs_trans_handle *trans; |
10429 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10430 | struct inode *inode = NULL; | |
10431 | u64 objectid; | |
10432 | u64 index; | |
10433 | int ret = 0; | |
10434 | ||
10435 | /* | |
10436 | * 5 units required for adding orphan entry | |
10437 | */ | |
10438 | trans = btrfs_start_transaction(root, 5); | |
10439 | if (IS_ERR(trans)) | |
10440 | return PTR_ERR(trans); | |
10441 | ||
10442 | ret = btrfs_find_free_ino(root, &objectid); | |
10443 | if (ret) | |
10444 | goto out; | |
10445 | ||
10446 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10447 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10448 | if (IS_ERR(inode)) { |
10449 | ret = PTR_ERR(inode); | |
10450 | inode = NULL; | |
10451 | goto out; | |
10452 | } | |
10453 | ||
ef3b9af5 FM |
10454 | inode->i_fop = &btrfs_file_operations; |
10455 | inode->i_op = &btrfs_file_inode_operations; | |
10456 | ||
10457 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10458 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10459 | ||
b0d5d10f CM |
10460 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10461 | if (ret) | |
32955c54 | 10462 | goto out; |
b0d5d10f CM |
10463 | |
10464 | ret = btrfs_update_inode(trans, root, inode); | |
10465 | if (ret) | |
32955c54 | 10466 | goto out; |
73f2e545 | 10467 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10468 | if (ret) |
32955c54 | 10469 | goto out; |
ef3b9af5 | 10470 | |
5762b5c9 FM |
10471 | /* |
10472 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10473 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10474 | * through: | |
10475 | * | |
10476 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10477 | */ | |
10478 | set_nlink(inode, 1); | |
ef3b9af5 | 10479 | d_tmpfile(dentry, inode); |
32955c54 | 10480 | unlock_new_inode(inode); |
ef3b9af5 | 10481 | mark_inode_dirty(inode); |
ef3b9af5 | 10482 | out: |
3a45bb20 | 10483 | btrfs_end_transaction(trans); |
32955c54 AV |
10484 | if (ret && inode) |
10485 | discard_new_inode(inode); | |
2ff7e61e | 10486 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 FM |
10487 | return ret; |
10488 | } | |
10489 | ||
5cdc84bf | 10490 | void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end) |
c6100a4b | 10491 | { |
5cdc84bf | 10492 | struct inode *inode = tree->private_data; |
c6100a4b JB |
10493 | unsigned long index = start >> PAGE_SHIFT; |
10494 | unsigned long end_index = end >> PAGE_SHIFT; | |
10495 | struct page *page; | |
10496 | ||
10497 | while (index <= end_index) { | |
10498 | page = find_get_page(inode->i_mapping, index); | |
10499 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10500 | set_page_writeback(page); | |
10501 | put_page(page); | |
10502 | index++; | |
10503 | } | |
10504 | } | |
10505 | ||
ed46ff3d OS |
10506 | #ifdef CONFIG_SWAP |
10507 | /* | |
10508 | * Add an entry indicating a block group or device which is pinned by a | |
10509 | * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a | |
10510 | * negative errno on failure. | |
10511 | */ | |
10512 | static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr, | |
10513 | bool is_block_group) | |
10514 | { | |
10515 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10516 | struct btrfs_swapfile_pin *sp, *entry; | |
10517 | struct rb_node **p; | |
10518 | struct rb_node *parent = NULL; | |
10519 | ||
10520 | sp = kmalloc(sizeof(*sp), GFP_NOFS); | |
10521 | if (!sp) | |
10522 | return -ENOMEM; | |
10523 | sp->ptr = ptr; | |
10524 | sp->inode = inode; | |
10525 | sp->is_block_group = is_block_group; | |
10526 | ||
10527 | spin_lock(&fs_info->swapfile_pins_lock); | |
10528 | p = &fs_info->swapfile_pins.rb_node; | |
10529 | while (*p) { | |
10530 | parent = *p; | |
10531 | entry = rb_entry(parent, struct btrfs_swapfile_pin, node); | |
10532 | if (sp->ptr < entry->ptr || | |
10533 | (sp->ptr == entry->ptr && sp->inode < entry->inode)) { | |
10534 | p = &(*p)->rb_left; | |
10535 | } else if (sp->ptr > entry->ptr || | |
10536 | (sp->ptr == entry->ptr && sp->inode > entry->inode)) { | |
10537 | p = &(*p)->rb_right; | |
10538 | } else { | |
10539 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10540 | kfree(sp); | |
10541 | return 1; | |
10542 | } | |
10543 | } | |
10544 | rb_link_node(&sp->node, parent, p); | |
10545 | rb_insert_color(&sp->node, &fs_info->swapfile_pins); | |
10546 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10547 | return 0; | |
10548 | } | |
10549 | ||
10550 | /* Free all of the entries pinned by this swapfile. */ | |
10551 | static void btrfs_free_swapfile_pins(struct inode *inode) | |
10552 | { | |
10553 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10554 | struct btrfs_swapfile_pin *sp; | |
10555 | struct rb_node *node, *next; | |
10556 | ||
10557 | spin_lock(&fs_info->swapfile_pins_lock); | |
10558 | node = rb_first(&fs_info->swapfile_pins); | |
10559 | while (node) { | |
10560 | next = rb_next(node); | |
10561 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
10562 | if (sp->inode == inode) { | |
10563 | rb_erase(&sp->node, &fs_info->swapfile_pins); | |
10564 | if (sp->is_block_group) | |
10565 | btrfs_put_block_group(sp->ptr); | |
10566 | kfree(sp); | |
10567 | } | |
10568 | node = next; | |
10569 | } | |
10570 | spin_unlock(&fs_info->swapfile_pins_lock); | |
10571 | } | |
10572 | ||
10573 | struct btrfs_swap_info { | |
10574 | u64 start; | |
10575 | u64 block_start; | |
10576 | u64 block_len; | |
10577 | u64 lowest_ppage; | |
10578 | u64 highest_ppage; | |
10579 | unsigned long nr_pages; | |
10580 | int nr_extents; | |
10581 | }; | |
10582 | ||
10583 | static int btrfs_add_swap_extent(struct swap_info_struct *sis, | |
10584 | struct btrfs_swap_info *bsi) | |
10585 | { | |
10586 | unsigned long nr_pages; | |
10587 | u64 first_ppage, first_ppage_reported, next_ppage; | |
10588 | int ret; | |
10589 | ||
10590 | first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT; | |
10591 | next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len, | |
10592 | PAGE_SIZE) >> PAGE_SHIFT; | |
10593 | ||
10594 | if (first_ppage >= next_ppage) | |
10595 | return 0; | |
10596 | nr_pages = next_ppage - first_ppage; | |
10597 | ||
10598 | first_ppage_reported = first_ppage; | |
10599 | if (bsi->start == 0) | |
10600 | first_ppage_reported++; | |
10601 | if (bsi->lowest_ppage > first_ppage_reported) | |
10602 | bsi->lowest_ppage = first_ppage_reported; | |
10603 | if (bsi->highest_ppage < (next_ppage - 1)) | |
10604 | bsi->highest_ppage = next_ppage - 1; | |
10605 | ||
10606 | ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage); | |
10607 | if (ret < 0) | |
10608 | return ret; | |
10609 | bsi->nr_extents += ret; | |
10610 | bsi->nr_pages += nr_pages; | |
10611 | return 0; | |
10612 | } | |
10613 | ||
10614 | static void btrfs_swap_deactivate(struct file *file) | |
10615 | { | |
10616 | struct inode *inode = file_inode(file); | |
10617 | ||
10618 | btrfs_free_swapfile_pins(inode); | |
10619 | atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles); | |
10620 | } | |
10621 | ||
10622 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10623 | sector_t *span) | |
10624 | { | |
10625 | struct inode *inode = file_inode(file); | |
10626 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
10627 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
10628 | struct extent_state *cached_state = NULL; | |
10629 | struct extent_map *em = NULL; | |
10630 | struct btrfs_device *device = NULL; | |
10631 | struct btrfs_swap_info bsi = { | |
10632 | .lowest_ppage = (sector_t)-1ULL, | |
10633 | }; | |
10634 | int ret = 0; | |
10635 | u64 isize; | |
10636 | u64 start; | |
10637 | ||
10638 | /* | |
10639 | * If the swap file was just created, make sure delalloc is done. If the | |
10640 | * file changes again after this, the user is doing something stupid and | |
10641 | * we don't really care. | |
10642 | */ | |
10643 | ret = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
10644 | if (ret) | |
10645 | return ret; | |
10646 | ||
10647 | /* | |
10648 | * The inode is locked, so these flags won't change after we check them. | |
10649 | */ | |
10650 | if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) { | |
10651 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10652 | return -EINVAL; | |
10653 | } | |
10654 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) { | |
10655 | btrfs_warn(fs_info, "swapfile must not be copy-on-write"); | |
10656 | return -EINVAL; | |
10657 | } | |
10658 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
10659 | btrfs_warn(fs_info, "swapfile must not be checksummed"); | |
10660 | return -EINVAL; | |
10661 | } | |
10662 | ||
10663 | /* | |
10664 | * Balance or device remove/replace/resize can move stuff around from | |
10665 | * under us. The EXCL_OP flag makes sure they aren't running/won't run | |
10666 | * concurrently while we are mapping the swap extents, and | |
10667 | * fs_info->swapfile_pins prevents them from running while the swap file | |
10668 | * is active and moving the extents. Note that this also prevents a | |
10669 | * concurrent device add which isn't actually necessary, but it's not | |
10670 | * really worth the trouble to allow it. | |
10671 | */ | |
10672 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) { | |
10673 | btrfs_warn(fs_info, | |
10674 | "cannot activate swapfile while exclusive operation is running"); | |
10675 | return -EBUSY; | |
10676 | } | |
10677 | /* | |
10678 | * Snapshots can create extents which require COW even if NODATACOW is | |
10679 | * set. We use this counter to prevent snapshots. We must increment it | |
10680 | * before walking the extents because we don't want a concurrent | |
10681 | * snapshot to run after we've already checked the extents. | |
10682 | */ | |
10683 | atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles); | |
10684 | ||
10685 | isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize); | |
10686 | ||
10687 | lock_extent_bits(io_tree, 0, isize - 1, &cached_state); | |
10688 | start = 0; | |
10689 | while (start < isize) { | |
10690 | u64 logical_block_start, physical_block_start; | |
10691 | struct btrfs_block_group_cache *bg; | |
10692 | u64 len = isize - start; | |
10693 | ||
10694 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); | |
10695 | if (IS_ERR(em)) { | |
10696 | ret = PTR_ERR(em); | |
10697 | goto out; | |
10698 | } | |
10699 | ||
10700 | if (em->block_start == EXTENT_MAP_HOLE) { | |
10701 | btrfs_warn(fs_info, "swapfile must not have holes"); | |
10702 | ret = -EINVAL; | |
10703 | goto out; | |
10704 | } | |
10705 | if (em->block_start == EXTENT_MAP_INLINE) { | |
10706 | /* | |
10707 | * It's unlikely we'll ever actually find ourselves | |
10708 | * here, as a file small enough to fit inline won't be | |
10709 | * big enough to store more than the swap header, but in | |
10710 | * case something changes in the future, let's catch it | |
10711 | * here rather than later. | |
10712 | */ | |
10713 | btrfs_warn(fs_info, "swapfile must not be inline"); | |
10714 | ret = -EINVAL; | |
10715 | goto out; | |
10716 | } | |
10717 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
10718 | btrfs_warn(fs_info, "swapfile must not be compressed"); | |
10719 | ret = -EINVAL; | |
10720 | goto out; | |
10721 | } | |
10722 | ||
10723 | logical_block_start = em->block_start + (start - em->start); | |
10724 | len = min(len, em->len - (start - em->start)); | |
10725 | free_extent_map(em); | |
10726 | em = NULL; | |
10727 | ||
10728 | ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL); | |
10729 | if (ret < 0) { | |
10730 | goto out; | |
10731 | } else if (ret) { | |
10732 | ret = 0; | |
10733 | } else { | |
10734 | btrfs_warn(fs_info, | |
10735 | "swapfile must not be copy-on-write"); | |
10736 | ret = -EINVAL; | |
10737 | goto out; | |
10738 | } | |
10739 | ||
10740 | em = btrfs_get_chunk_map(fs_info, logical_block_start, len); | |
10741 | if (IS_ERR(em)) { | |
10742 | ret = PTR_ERR(em); | |
10743 | goto out; | |
10744 | } | |
10745 | ||
10746 | if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
10747 | btrfs_warn(fs_info, | |
10748 | "swapfile must have single data profile"); | |
10749 | ret = -EINVAL; | |
10750 | goto out; | |
10751 | } | |
10752 | ||
10753 | if (device == NULL) { | |
10754 | device = em->map_lookup->stripes[0].dev; | |
10755 | ret = btrfs_add_swapfile_pin(inode, device, false); | |
10756 | if (ret == 1) | |
10757 | ret = 0; | |
10758 | else if (ret) | |
10759 | goto out; | |
10760 | } else if (device != em->map_lookup->stripes[0].dev) { | |
10761 | btrfs_warn(fs_info, "swapfile must be on one device"); | |
10762 | ret = -EINVAL; | |
10763 | goto out; | |
10764 | } | |
10765 | ||
10766 | physical_block_start = (em->map_lookup->stripes[0].physical + | |
10767 | (logical_block_start - em->start)); | |
10768 | len = min(len, em->len - (logical_block_start - em->start)); | |
10769 | free_extent_map(em); | |
10770 | em = NULL; | |
10771 | ||
10772 | bg = btrfs_lookup_block_group(fs_info, logical_block_start); | |
10773 | if (!bg) { | |
10774 | btrfs_warn(fs_info, | |
10775 | "could not find block group containing swapfile"); | |
10776 | ret = -EINVAL; | |
10777 | goto out; | |
10778 | } | |
10779 | ||
10780 | ret = btrfs_add_swapfile_pin(inode, bg, true); | |
10781 | if (ret) { | |
10782 | btrfs_put_block_group(bg); | |
10783 | if (ret == 1) | |
10784 | ret = 0; | |
10785 | else | |
10786 | goto out; | |
10787 | } | |
10788 | ||
10789 | if (bsi.block_len && | |
10790 | bsi.block_start + bsi.block_len == physical_block_start) { | |
10791 | bsi.block_len += len; | |
10792 | } else { | |
10793 | if (bsi.block_len) { | |
10794 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10795 | if (ret) | |
10796 | goto out; | |
10797 | } | |
10798 | bsi.start = start; | |
10799 | bsi.block_start = physical_block_start; | |
10800 | bsi.block_len = len; | |
10801 | } | |
10802 | ||
10803 | start += len; | |
10804 | } | |
10805 | ||
10806 | if (bsi.block_len) | |
10807 | ret = btrfs_add_swap_extent(sis, &bsi); | |
10808 | ||
10809 | out: | |
10810 | if (!IS_ERR_OR_NULL(em)) | |
10811 | free_extent_map(em); | |
10812 | ||
10813 | unlock_extent_cached(io_tree, 0, isize - 1, &cached_state); | |
10814 | ||
10815 | if (ret) | |
10816 | btrfs_swap_deactivate(file); | |
10817 | ||
10818 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); | |
10819 | ||
10820 | if (ret) | |
10821 | return ret; | |
10822 | ||
10823 | if (device) | |
10824 | sis->bdev = device->bdev; | |
10825 | *span = bsi.highest_ppage - bsi.lowest_ppage + 1; | |
10826 | sis->max = bsi.nr_pages; | |
10827 | sis->pages = bsi.nr_pages - 1; | |
10828 | sis->highest_bit = bsi.nr_pages - 1; | |
10829 | return bsi.nr_extents; | |
10830 | } | |
10831 | #else | |
10832 | static void btrfs_swap_deactivate(struct file *file) | |
10833 | { | |
10834 | } | |
10835 | ||
10836 | static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
10837 | sector_t *span) | |
10838 | { | |
10839 | return -EOPNOTSUPP; | |
10840 | } | |
10841 | #endif | |
10842 | ||
6e1d5dcc | 10843 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10844 | .getattr = btrfs_getattr, |
39279cc3 CM |
10845 | .lookup = btrfs_lookup, |
10846 | .create = btrfs_create, | |
10847 | .unlink = btrfs_unlink, | |
10848 | .link = btrfs_link, | |
10849 | .mkdir = btrfs_mkdir, | |
10850 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10851 | .rename = btrfs_rename2, |
39279cc3 CM |
10852 | .symlink = btrfs_symlink, |
10853 | .setattr = btrfs_setattr, | |
618e21d5 | 10854 | .mknod = btrfs_mknod, |
5103e947 | 10855 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10856 | .permission = btrfs_permission, |
4e34e719 | 10857 | .get_acl = btrfs_get_acl, |
996a710d | 10858 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10859 | .update_time = btrfs_update_time, |
ef3b9af5 | 10860 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10861 | }; |
6e1d5dcc | 10862 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10863 | .lookup = btrfs_lookup, |
fdebe2bd | 10864 | .permission = btrfs_permission, |
93fd63c2 | 10865 | .update_time = btrfs_update_time, |
39279cc3 | 10866 | }; |
76dda93c | 10867 | |
828c0950 | 10868 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10869 | .llseek = generic_file_llseek, |
10870 | .read = generic_read_dir, | |
02dbfc99 | 10871 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10872 | .open = btrfs_opendir, |
34287aa3 | 10873 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10874 | #ifdef CONFIG_COMPAT |
4c63c245 | 10875 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10876 | #endif |
6bf13c0c | 10877 | .release = btrfs_release_file, |
e02119d5 | 10878 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10879 | }; |
10880 | ||
20e5506b | 10881 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10882 | /* mandatory callbacks */ |
065631f6 | 10883 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac CM |
10884 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
10885 | }; | |
10886 | ||
35054394 CM |
10887 | /* |
10888 | * btrfs doesn't support the bmap operation because swapfiles | |
10889 | * use bmap to make a mapping of extents in the file. They assume | |
10890 | * these extents won't change over the life of the file and they | |
10891 | * use the bmap result to do IO directly to the drive. | |
10892 | * | |
10893 | * the btrfs bmap call would return logical addresses that aren't | |
10894 | * suitable for IO and they also will change frequently as COW | |
10895 | * operations happen. So, swapfile + btrfs == corruption. | |
10896 | * | |
10897 | * For now we're avoiding this by dropping bmap. | |
10898 | */ | |
7f09410b | 10899 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10900 | .readpage = btrfs_readpage, |
10901 | .writepage = btrfs_writepage, | |
b293f02e | 10902 | .writepages = btrfs_writepages, |
3ab2fb5a | 10903 | .readpages = btrfs_readpages, |
16432985 | 10904 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10905 | .invalidatepage = btrfs_invalidatepage, |
10906 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10907 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10908 | .error_remove_page = generic_error_remove_page, |
ed46ff3d OS |
10909 | .swap_activate = btrfs_swap_activate, |
10910 | .swap_deactivate = btrfs_swap_deactivate, | |
39279cc3 CM |
10911 | }; |
10912 | ||
6e1d5dcc | 10913 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10914 | .getattr = btrfs_getattr, |
10915 | .setattr = btrfs_setattr, | |
5103e947 | 10916 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10917 | .permission = btrfs_permission, |
1506fcc8 | 10918 | .fiemap = btrfs_fiemap, |
4e34e719 | 10919 | .get_acl = btrfs_get_acl, |
996a710d | 10920 | .set_acl = btrfs_set_acl, |
e41f941a | 10921 | .update_time = btrfs_update_time, |
39279cc3 | 10922 | }; |
6e1d5dcc | 10923 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10924 | .getattr = btrfs_getattr, |
10925 | .setattr = btrfs_setattr, | |
fdebe2bd | 10926 | .permission = btrfs_permission, |
33268eaf | 10927 | .listxattr = btrfs_listxattr, |
4e34e719 | 10928 | .get_acl = btrfs_get_acl, |
996a710d | 10929 | .set_acl = btrfs_set_acl, |
e41f941a | 10930 | .update_time = btrfs_update_time, |
618e21d5 | 10931 | }; |
6e1d5dcc | 10932 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10933 | .get_link = page_get_link, |
f209561a | 10934 | .getattr = btrfs_getattr, |
22c44fe6 | 10935 | .setattr = btrfs_setattr, |
fdebe2bd | 10936 | .permission = btrfs_permission, |
0279b4cd | 10937 | .listxattr = btrfs_listxattr, |
e41f941a | 10938 | .update_time = btrfs_update_time, |
39279cc3 | 10939 | }; |
76dda93c | 10940 | |
82d339d9 | 10941 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
10942 | .d_delete = btrfs_dentry_delete, |
10943 | }; |