Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
a27bb332 | 35 | #include <linux/aio.h> |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
5103e947 | 37 | #include <linux/xattr.h> |
33268eaf | 38 | #include <linux/posix_acl.h> |
d899e052 | 39 | #include <linux/falloc.h> |
5a0e3ad6 | 40 | #include <linux/slab.h> |
7a36ddec | 41 | #include <linux/ratelimit.h> |
22c44fe6 | 42 | #include <linux/mount.h> |
55e301fd | 43 | #include <linux/btrfs.h> |
53b381b3 | 44 | #include <linux/blkdev.h> |
f23b5a59 | 45 | #include <linux/posix_acl_xattr.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
39279cc3 CM |
62 | |
63 | struct btrfs_iget_args { | |
90d3e592 | 64 | struct btrfs_key *location; |
39279cc3 CM |
65 | struct btrfs_root *root; |
66 | }; | |
67 | ||
6e1d5dcc AD |
68 | static const struct inode_operations btrfs_dir_inode_operations; |
69 | static const struct inode_operations btrfs_symlink_inode_operations; | |
70 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
71 | static const struct inode_operations btrfs_special_inode_operations; | |
72 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
73 | static const struct address_space_operations btrfs_aops; |
74 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 75 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 76 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
77 | |
78 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 79 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
80 | struct kmem_cache *btrfs_trans_handle_cachep; |
81 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 82 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 83 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
84 | |
85 | #define S_SHIFT 12 | |
86 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
87 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
88 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
89 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
90 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
91 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
92 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
93 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
94 | }; | |
95 | ||
3972f260 | 96 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 97 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 98 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
99 | static noinline int cow_file_range(struct inode *inode, |
100 | struct page *locked_page, | |
101 | u64 start, u64 end, int *page_started, | |
102 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
103 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
104 | u64 len, u64 orig_start, | |
105 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
106 | u64 orig_block_len, u64 ram_bytes, |
107 | int type); | |
7b128766 | 108 | |
48a3b636 | 109 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 110 | |
f34f57a3 | 111 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
112 | struct inode *inode, struct inode *dir, |
113 | const struct qstr *qstr) | |
0279b4cd JO |
114 | { |
115 | int err; | |
116 | ||
f34f57a3 | 117 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 118 | if (!err) |
2a7dba39 | 119 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
120 | return err; |
121 | } | |
122 | ||
c8b97818 CM |
123 | /* |
124 | * this does all the hard work for inserting an inline extent into | |
125 | * the btree. The caller should have done a btrfs_drop_extents so that | |
126 | * no overlapping inline items exist in the btree | |
127 | */ | |
40f76580 | 128 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 129 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
130 | struct btrfs_root *root, struct inode *inode, |
131 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 132 | int compress_type, |
c8b97818 CM |
133 | struct page **compressed_pages) |
134 | { | |
c8b97818 CM |
135 | struct extent_buffer *leaf; |
136 | struct page *page = NULL; | |
137 | char *kaddr; | |
138 | unsigned long ptr; | |
139 | struct btrfs_file_extent_item *ei; | |
140 | int err = 0; | |
141 | int ret; | |
142 | size_t cur_size = size; | |
c8b97818 | 143 | unsigned long offset; |
c8b97818 | 144 | |
fe3f566c | 145 | if (compressed_size && compressed_pages) |
c8b97818 | 146 | cur_size = compressed_size; |
c8b97818 | 147 | |
1acae57b | 148 | inode_add_bytes(inode, size); |
c8b97818 | 149 | |
1acae57b FDBM |
150 | if (!extent_inserted) { |
151 | struct btrfs_key key; | |
152 | size_t datasize; | |
c8b97818 | 153 | |
1acae57b FDBM |
154 | key.objectid = btrfs_ino(inode); |
155 | key.offset = start; | |
962a298f | 156 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 157 | |
1acae57b FDBM |
158 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
159 | path->leave_spinning = 1; | |
160 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
161 | datasize); | |
162 | if (ret) { | |
163 | err = ret; | |
164 | goto fail; | |
165 | } | |
c8b97818 CM |
166 | } |
167 | leaf = path->nodes[0]; | |
168 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
169 | struct btrfs_file_extent_item); | |
170 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
171 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
172 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
173 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
174 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
175 | ptr = btrfs_file_extent_inline_start(ei); | |
176 | ||
261507a0 | 177 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
178 | struct page *cpage; |
179 | int i = 0; | |
d397712b | 180 | while (compressed_size > 0) { |
c8b97818 | 181 | cpage = compressed_pages[i]; |
5b050f04 | 182 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
183 | PAGE_CACHE_SIZE); |
184 | ||
7ac687d9 | 185 | kaddr = kmap_atomic(cpage); |
c8b97818 | 186 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 187 | kunmap_atomic(kaddr); |
c8b97818 CM |
188 | |
189 | i++; | |
190 | ptr += cur_size; | |
191 | compressed_size -= cur_size; | |
192 | } | |
193 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 194 | compress_type); |
c8b97818 CM |
195 | } else { |
196 | page = find_get_page(inode->i_mapping, | |
197 | start >> PAGE_CACHE_SHIFT); | |
198 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 199 | kaddr = kmap_atomic(page); |
c8b97818 CM |
200 | offset = start & (PAGE_CACHE_SIZE - 1); |
201 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 202 | kunmap_atomic(kaddr); |
c8b97818 CM |
203 | page_cache_release(page); |
204 | } | |
205 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 206 | btrfs_release_path(path); |
c8b97818 | 207 | |
c2167754 YZ |
208 | /* |
209 | * we're an inline extent, so nobody can | |
210 | * extend the file past i_size without locking | |
211 | * a page we already have locked. | |
212 | * | |
213 | * We must do any isize and inode updates | |
214 | * before we unlock the pages. Otherwise we | |
215 | * could end up racing with unlink. | |
216 | */ | |
c8b97818 | 217 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 218 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 219 | |
79787eaa | 220 | return ret; |
c8b97818 | 221 | fail: |
c8b97818 CM |
222 | return err; |
223 | } | |
224 | ||
225 | ||
226 | /* | |
227 | * conditionally insert an inline extent into the file. This | |
228 | * does the checks required to make sure the data is small enough | |
229 | * to fit as an inline extent. | |
230 | */ | |
00361589 JB |
231 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
232 | struct inode *inode, u64 start, | |
233 | u64 end, size_t compressed_size, | |
234 | int compress_type, | |
235 | struct page **compressed_pages) | |
c8b97818 | 236 | { |
00361589 | 237 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
238 | u64 isize = i_size_read(inode); |
239 | u64 actual_end = min(end + 1, isize); | |
240 | u64 inline_len = actual_end - start; | |
fda2832f | 241 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
242 | u64 data_len = inline_len; |
243 | int ret; | |
1acae57b FDBM |
244 | struct btrfs_path *path; |
245 | int extent_inserted = 0; | |
246 | u32 extent_item_size; | |
c8b97818 CM |
247 | |
248 | if (compressed_size) | |
249 | data_len = compressed_size; | |
250 | ||
251 | if (start > 0 || | |
354877be WS |
252 | actual_end > PAGE_CACHE_SIZE || |
253 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) || | |
c8b97818 CM |
254 | (!compressed_size && |
255 | (actual_end & (root->sectorsize - 1)) == 0) || | |
256 | end + 1 < isize || | |
257 | data_len > root->fs_info->max_inline) { | |
258 | return 1; | |
259 | } | |
260 | ||
1acae57b FDBM |
261 | path = btrfs_alloc_path(); |
262 | if (!path) | |
263 | return -ENOMEM; | |
264 | ||
00361589 | 265 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
266 | if (IS_ERR(trans)) { |
267 | btrfs_free_path(path); | |
00361589 | 268 | return PTR_ERR(trans); |
1acae57b | 269 | } |
00361589 JB |
270 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
271 | ||
1acae57b FDBM |
272 | if (compressed_size && compressed_pages) |
273 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
274 | compressed_size); | |
275 | else | |
276 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
277 | inline_len); | |
278 | ||
279 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
280 | start, aligned_end, NULL, | |
281 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
282 | if (ret) { |
283 | btrfs_abort_transaction(trans, root, ret); | |
284 | goto out; | |
285 | } | |
c8b97818 CM |
286 | |
287 | if (isize > actual_end) | |
288 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
289 | ret = insert_inline_extent(trans, path, extent_inserted, |
290 | root, inode, start, | |
c8b97818 | 291 | inline_len, compressed_size, |
fe3f566c | 292 | compress_type, compressed_pages); |
2adcac1a | 293 | if (ret && ret != -ENOSPC) { |
79787eaa | 294 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 295 | goto out; |
2adcac1a | 296 | } else if (ret == -ENOSPC) { |
00361589 JB |
297 | ret = 1; |
298 | goto out; | |
79787eaa | 299 | } |
2adcac1a | 300 | |
bdc20e67 | 301 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 302 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 303 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 304 | out: |
1acae57b | 305 | btrfs_free_path(path); |
00361589 JB |
306 | btrfs_end_transaction(trans, root); |
307 | return ret; | |
c8b97818 CM |
308 | } |
309 | ||
771ed689 CM |
310 | struct async_extent { |
311 | u64 start; | |
312 | u64 ram_size; | |
313 | u64 compressed_size; | |
314 | struct page **pages; | |
315 | unsigned long nr_pages; | |
261507a0 | 316 | int compress_type; |
771ed689 CM |
317 | struct list_head list; |
318 | }; | |
319 | ||
320 | struct async_cow { | |
321 | struct inode *inode; | |
322 | struct btrfs_root *root; | |
323 | struct page *locked_page; | |
324 | u64 start; | |
325 | u64 end; | |
326 | struct list_head extents; | |
327 | struct btrfs_work work; | |
328 | }; | |
329 | ||
330 | static noinline int add_async_extent(struct async_cow *cow, | |
331 | u64 start, u64 ram_size, | |
332 | u64 compressed_size, | |
333 | struct page **pages, | |
261507a0 LZ |
334 | unsigned long nr_pages, |
335 | int compress_type) | |
771ed689 CM |
336 | { |
337 | struct async_extent *async_extent; | |
338 | ||
339 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 340 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
341 | async_extent->start = start; |
342 | async_extent->ram_size = ram_size; | |
343 | async_extent->compressed_size = compressed_size; | |
344 | async_extent->pages = pages; | |
345 | async_extent->nr_pages = nr_pages; | |
261507a0 | 346 | async_extent->compress_type = compress_type; |
771ed689 CM |
347 | list_add_tail(&async_extent->list, &cow->extents); |
348 | return 0; | |
349 | } | |
350 | ||
f79707b0 WS |
351 | static inline int inode_need_compress(struct inode *inode) |
352 | { | |
353 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
354 | ||
355 | /* force compress */ | |
356 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
357 | return 1; | |
358 | /* bad compression ratios */ | |
359 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
360 | return 0; | |
361 | if (btrfs_test_opt(root, COMPRESS) || | |
362 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
363 | BTRFS_I(inode)->force_compress) | |
364 | return 1; | |
365 | return 0; | |
366 | } | |
367 | ||
d352ac68 | 368 | /* |
771ed689 CM |
369 | * we create compressed extents in two phases. The first |
370 | * phase compresses a range of pages that have already been | |
371 | * locked (both pages and state bits are locked). | |
c8b97818 | 372 | * |
771ed689 CM |
373 | * This is done inside an ordered work queue, and the compression |
374 | * is spread across many cpus. The actual IO submission is step | |
375 | * two, and the ordered work queue takes care of making sure that | |
376 | * happens in the same order things were put onto the queue by | |
377 | * writepages and friends. | |
c8b97818 | 378 | * |
771ed689 CM |
379 | * If this code finds it can't get good compression, it puts an |
380 | * entry onto the work queue to write the uncompressed bytes. This | |
381 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
382 | * are written in the same order that the flusher thread sent them |
383 | * down. | |
d352ac68 | 384 | */ |
c44f649e | 385 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
386 | struct page *locked_page, |
387 | u64 start, u64 end, | |
388 | struct async_cow *async_cow, | |
389 | int *num_added) | |
b888db2b CM |
390 | { |
391 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 392 | u64 num_bytes; |
db94535d | 393 | u64 blocksize = root->sectorsize; |
c8b97818 | 394 | u64 actual_end; |
42dc7bab | 395 | u64 isize = i_size_read(inode); |
e6dcd2dc | 396 | int ret = 0; |
c8b97818 CM |
397 | struct page **pages = NULL; |
398 | unsigned long nr_pages; | |
399 | unsigned long nr_pages_ret = 0; | |
400 | unsigned long total_compressed = 0; | |
401 | unsigned long total_in = 0; | |
402 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 403 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
404 | int i; |
405 | int will_compress; | |
261507a0 | 406 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 407 | int redirty = 0; |
b888db2b | 408 | |
4cb13e5d LB |
409 | /* if this is a small write inside eof, kick off a defrag */ |
410 | if ((end - start + 1) < 16 * 1024 && | |
411 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
412 | btrfs_add_inode_defrag(NULL, inode); |
413 | ||
42dc7bab | 414 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
415 | again: |
416 | will_compress = 0; | |
417 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
418 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 419 | |
f03d9301 CM |
420 | /* |
421 | * we don't want to send crud past the end of i_size through | |
422 | * compression, that's just a waste of CPU time. So, if the | |
423 | * end of the file is before the start of our current | |
424 | * requested range of bytes, we bail out to the uncompressed | |
425 | * cleanup code that can deal with all of this. | |
426 | * | |
427 | * It isn't really the fastest way to fix things, but this is a | |
428 | * very uncommon corner. | |
429 | */ | |
430 | if (actual_end <= start) | |
431 | goto cleanup_and_bail_uncompressed; | |
432 | ||
c8b97818 CM |
433 | total_compressed = actual_end - start; |
434 | ||
4bcbb332 SW |
435 | /* |
436 | * skip compression for a small file range(<=blocksize) that | |
437 | * isn't an inline extent, since it dosen't save disk space at all. | |
438 | */ | |
439 | if (total_compressed <= blocksize && | |
440 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
441 | goto cleanup_and_bail_uncompressed; | |
442 | ||
c8b97818 CM |
443 | /* we want to make sure that amount of ram required to uncompress |
444 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
445 | * of a compressed extent to 128k. This is a crucial number |
446 | * because it also controls how easily we can spread reads across | |
447 | * cpus for decompression. | |
448 | * | |
449 | * We also want to make sure the amount of IO required to do | |
450 | * a random read is reasonably small, so we limit the size of | |
451 | * a compressed extent to 128k. | |
c8b97818 CM |
452 | */ |
453 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 454 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 455 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
456 | total_in = 0; |
457 | ret = 0; | |
db94535d | 458 | |
771ed689 CM |
459 | /* |
460 | * we do compression for mount -o compress and when the | |
461 | * inode has not been flagged as nocompress. This flag can | |
462 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 463 | */ |
f79707b0 | 464 | if (inode_need_compress(inode)) { |
c8b97818 | 465 | WARN_ON(pages); |
cfbc246e | 466 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
467 | if (!pages) { |
468 | /* just bail out to the uncompressed code */ | |
469 | goto cont; | |
470 | } | |
c8b97818 | 471 | |
261507a0 LZ |
472 | if (BTRFS_I(inode)->force_compress) |
473 | compress_type = BTRFS_I(inode)->force_compress; | |
474 | ||
4adaa611 CM |
475 | /* |
476 | * we need to call clear_page_dirty_for_io on each | |
477 | * page in the range. Otherwise applications with the file | |
478 | * mmap'd can wander in and change the page contents while | |
479 | * we are compressing them. | |
480 | * | |
481 | * If the compression fails for any reason, we set the pages | |
482 | * dirty again later on. | |
483 | */ | |
484 | extent_range_clear_dirty_for_io(inode, start, end); | |
485 | redirty = 1; | |
261507a0 LZ |
486 | ret = btrfs_compress_pages(compress_type, |
487 | inode->i_mapping, start, | |
488 | total_compressed, pages, | |
489 | nr_pages, &nr_pages_ret, | |
490 | &total_in, | |
491 | &total_compressed, | |
492 | max_compressed); | |
c8b97818 CM |
493 | |
494 | if (!ret) { | |
495 | unsigned long offset = total_compressed & | |
496 | (PAGE_CACHE_SIZE - 1); | |
497 | struct page *page = pages[nr_pages_ret - 1]; | |
498 | char *kaddr; | |
499 | ||
500 | /* zero the tail end of the last page, we might be | |
501 | * sending it down to disk | |
502 | */ | |
503 | if (offset) { | |
7ac687d9 | 504 | kaddr = kmap_atomic(page); |
c8b97818 CM |
505 | memset(kaddr + offset, 0, |
506 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 507 | kunmap_atomic(kaddr); |
c8b97818 CM |
508 | } |
509 | will_compress = 1; | |
510 | } | |
511 | } | |
560f7d75 | 512 | cont: |
c8b97818 CM |
513 | if (start == 0) { |
514 | /* lets try to make an inline extent */ | |
771ed689 | 515 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 516 | /* we didn't compress the entire range, try |
771ed689 | 517 | * to make an uncompressed inline extent. |
c8b97818 | 518 | */ |
00361589 JB |
519 | ret = cow_file_range_inline(root, inode, start, end, |
520 | 0, 0, NULL); | |
c8b97818 | 521 | } else { |
771ed689 | 522 | /* try making a compressed inline extent */ |
00361589 | 523 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
524 | total_compressed, |
525 | compress_type, pages); | |
c8b97818 | 526 | } |
79787eaa | 527 | if (ret <= 0) { |
151a41bc JB |
528 | unsigned long clear_flags = EXTENT_DELALLOC | |
529 | EXTENT_DEFRAG; | |
e6eb4314 FM |
530 | unsigned long page_error_op; |
531 | ||
151a41bc | 532 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 533 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 534 | |
771ed689 | 535 | /* |
79787eaa JM |
536 | * inline extent creation worked or returned error, |
537 | * we don't need to create any more async work items. | |
538 | * Unlock and free up our temp pages. | |
771ed689 | 539 | */ |
c2790a2e | 540 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 541 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
542 | PAGE_CLEAR_DIRTY | |
543 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 544 | page_error_op | |
c2790a2e | 545 | PAGE_END_WRITEBACK); |
c8b97818 CM |
546 | goto free_pages_out; |
547 | } | |
548 | } | |
549 | ||
550 | if (will_compress) { | |
551 | /* | |
552 | * we aren't doing an inline extent round the compressed size | |
553 | * up to a block size boundary so the allocator does sane | |
554 | * things | |
555 | */ | |
fda2832f | 556 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
557 | |
558 | /* | |
559 | * one last check to make sure the compression is really a | |
560 | * win, compare the page count read with the blocks on disk | |
561 | */ | |
fda2832f | 562 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
563 | if (total_compressed >= total_in) { |
564 | will_compress = 0; | |
565 | } else { | |
c8b97818 CM |
566 | num_bytes = total_in; |
567 | } | |
568 | } | |
569 | if (!will_compress && pages) { | |
570 | /* | |
571 | * the compression code ran but failed to make things smaller, | |
572 | * free any pages it allocated and our page pointer array | |
573 | */ | |
574 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 575 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
576 | page_cache_release(pages[i]); |
577 | } | |
578 | kfree(pages); | |
579 | pages = NULL; | |
580 | total_compressed = 0; | |
581 | nr_pages_ret = 0; | |
582 | ||
583 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
584 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
585 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 586 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 587 | } |
c8b97818 | 588 | } |
771ed689 CM |
589 | if (will_compress) { |
590 | *num_added += 1; | |
c8b97818 | 591 | |
771ed689 CM |
592 | /* the async work queues will take care of doing actual |
593 | * allocation on disk for these compressed pages, | |
594 | * and will submit them to the elevator. | |
595 | */ | |
596 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
597 | total_compressed, pages, nr_pages_ret, |
598 | compress_type); | |
179e29e4 | 599 | |
24ae6365 | 600 | if (start + num_bytes < end) { |
771ed689 CM |
601 | start += num_bytes; |
602 | pages = NULL; | |
603 | cond_resched(); | |
604 | goto again; | |
605 | } | |
606 | } else { | |
f03d9301 | 607 | cleanup_and_bail_uncompressed: |
771ed689 CM |
608 | /* |
609 | * No compression, but we still need to write the pages in | |
610 | * the file we've been given so far. redirty the locked | |
611 | * page if it corresponds to our extent and set things up | |
612 | * for the async work queue to run cow_file_range to do | |
613 | * the normal delalloc dance | |
614 | */ | |
615 | if (page_offset(locked_page) >= start && | |
616 | page_offset(locked_page) <= end) { | |
617 | __set_page_dirty_nobuffers(locked_page); | |
618 | /* unlocked later on in the async handlers */ | |
619 | } | |
4adaa611 CM |
620 | if (redirty) |
621 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
622 | add_async_extent(async_cow, start, end - start + 1, |
623 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
624 | *num_added += 1; |
625 | } | |
3b951516 | 626 | |
c44f649e | 627 | return; |
771ed689 CM |
628 | |
629 | free_pages_out: | |
630 | for (i = 0; i < nr_pages_ret; i++) { | |
631 | WARN_ON(pages[i]->mapping); | |
632 | page_cache_release(pages[i]); | |
633 | } | |
d397712b | 634 | kfree(pages); |
771ed689 | 635 | } |
771ed689 | 636 | |
40ae837b FM |
637 | static void free_async_extent_pages(struct async_extent *async_extent) |
638 | { | |
639 | int i; | |
640 | ||
641 | if (!async_extent->pages) | |
642 | return; | |
643 | ||
644 | for (i = 0; i < async_extent->nr_pages; i++) { | |
645 | WARN_ON(async_extent->pages[i]->mapping); | |
646 | page_cache_release(async_extent->pages[i]); | |
647 | } | |
648 | kfree(async_extent->pages); | |
649 | async_extent->nr_pages = 0; | |
650 | async_extent->pages = NULL; | |
771ed689 CM |
651 | } |
652 | ||
653 | /* | |
654 | * phase two of compressed writeback. This is the ordered portion | |
655 | * of the code, which only gets called in the order the work was | |
656 | * queued. We walk all the async extents created by compress_file_range | |
657 | * and send them down to the disk. | |
658 | */ | |
dec8f175 | 659 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
660 | struct async_cow *async_cow) |
661 | { | |
662 | struct async_extent *async_extent; | |
663 | u64 alloc_hint = 0; | |
771ed689 CM |
664 | struct btrfs_key ins; |
665 | struct extent_map *em; | |
666 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
667 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
668 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 669 | int ret = 0; |
771ed689 | 670 | |
3e04e7f1 | 671 | again: |
d397712b | 672 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
673 | async_extent = list_entry(async_cow->extents.next, |
674 | struct async_extent, list); | |
675 | list_del(&async_extent->list); | |
c8b97818 | 676 | |
771ed689 CM |
677 | io_tree = &BTRFS_I(inode)->io_tree; |
678 | ||
f5a84ee3 | 679 | retry: |
771ed689 CM |
680 | /* did the compression code fall back to uncompressed IO? */ |
681 | if (!async_extent->pages) { | |
682 | int page_started = 0; | |
683 | unsigned long nr_written = 0; | |
684 | ||
685 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 686 | async_extent->start + |
d0082371 | 687 | async_extent->ram_size - 1); |
771ed689 CM |
688 | |
689 | /* allocate blocks */ | |
f5a84ee3 JB |
690 | ret = cow_file_range(inode, async_cow->locked_page, |
691 | async_extent->start, | |
692 | async_extent->start + | |
693 | async_extent->ram_size - 1, | |
694 | &page_started, &nr_written, 0); | |
771ed689 | 695 | |
79787eaa JM |
696 | /* JDM XXX */ |
697 | ||
771ed689 CM |
698 | /* |
699 | * if page_started, cow_file_range inserted an | |
700 | * inline extent and took care of all the unlocking | |
701 | * and IO for us. Otherwise, we need to submit | |
702 | * all those pages down to the drive. | |
703 | */ | |
f5a84ee3 | 704 | if (!page_started && !ret) |
771ed689 CM |
705 | extent_write_locked_range(io_tree, |
706 | inode, async_extent->start, | |
d397712b | 707 | async_extent->start + |
771ed689 CM |
708 | async_extent->ram_size - 1, |
709 | btrfs_get_extent, | |
710 | WB_SYNC_ALL); | |
3e04e7f1 JB |
711 | else if (ret) |
712 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
713 | kfree(async_extent); |
714 | cond_resched(); | |
715 | continue; | |
716 | } | |
717 | ||
718 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 719 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 720 | |
00361589 | 721 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
722 | async_extent->compressed_size, |
723 | async_extent->compressed_size, | |
e570fd27 | 724 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 725 | if (ret) { |
40ae837b | 726 | free_async_extent_pages(async_extent); |
3e04e7f1 | 727 | |
fdf8e2ea JB |
728 | if (ret == -ENOSPC) { |
729 | unlock_extent(io_tree, async_extent->start, | |
730 | async_extent->start + | |
731 | async_extent->ram_size - 1); | |
ce62003f LB |
732 | |
733 | /* | |
734 | * we need to redirty the pages if we decide to | |
735 | * fallback to uncompressed IO, otherwise we | |
736 | * will not submit these pages down to lower | |
737 | * layers. | |
738 | */ | |
739 | extent_range_redirty_for_io(inode, | |
740 | async_extent->start, | |
741 | async_extent->start + | |
742 | async_extent->ram_size - 1); | |
743 | ||
79787eaa | 744 | goto retry; |
fdf8e2ea | 745 | } |
3e04e7f1 | 746 | goto out_free; |
f5a84ee3 JB |
747 | } |
748 | ||
c2167754 YZ |
749 | /* |
750 | * here we're doing allocation and writeback of the | |
751 | * compressed pages | |
752 | */ | |
753 | btrfs_drop_extent_cache(inode, async_extent->start, | |
754 | async_extent->start + | |
755 | async_extent->ram_size - 1, 0); | |
756 | ||
172ddd60 | 757 | em = alloc_extent_map(); |
b9aa55be LB |
758 | if (!em) { |
759 | ret = -ENOMEM; | |
3e04e7f1 | 760 | goto out_free_reserve; |
b9aa55be | 761 | } |
771ed689 CM |
762 | em->start = async_extent->start; |
763 | em->len = async_extent->ram_size; | |
445a6944 | 764 | em->orig_start = em->start; |
2ab28f32 JB |
765 | em->mod_start = em->start; |
766 | em->mod_len = em->len; | |
c8b97818 | 767 | |
771ed689 CM |
768 | em->block_start = ins.objectid; |
769 | em->block_len = ins.offset; | |
b4939680 | 770 | em->orig_block_len = ins.offset; |
cc95bef6 | 771 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 772 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 773 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
774 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
775 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 776 | em->generation = -1; |
771ed689 | 777 | |
d397712b | 778 | while (1) { |
890871be | 779 | write_lock(&em_tree->lock); |
09a2a8f9 | 780 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 781 | write_unlock(&em_tree->lock); |
771ed689 CM |
782 | if (ret != -EEXIST) { |
783 | free_extent_map(em); | |
784 | break; | |
785 | } | |
786 | btrfs_drop_extent_cache(inode, async_extent->start, | |
787 | async_extent->start + | |
788 | async_extent->ram_size - 1, 0); | |
789 | } | |
790 | ||
3e04e7f1 JB |
791 | if (ret) |
792 | goto out_free_reserve; | |
793 | ||
261507a0 LZ |
794 | ret = btrfs_add_ordered_extent_compress(inode, |
795 | async_extent->start, | |
796 | ins.objectid, | |
797 | async_extent->ram_size, | |
798 | ins.offset, | |
799 | BTRFS_ORDERED_COMPRESSED, | |
800 | async_extent->compress_type); | |
d9f85963 FM |
801 | if (ret) { |
802 | btrfs_drop_extent_cache(inode, async_extent->start, | |
803 | async_extent->start + | |
804 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 805 | goto out_free_reserve; |
d9f85963 | 806 | } |
771ed689 | 807 | |
771ed689 CM |
808 | /* |
809 | * clear dirty, set writeback and unlock the pages. | |
810 | */ | |
c2790a2e | 811 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
812 | async_extent->start + |
813 | async_extent->ram_size - 1, | |
151a41bc JB |
814 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
815 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 816 | PAGE_SET_WRITEBACK); |
771ed689 | 817 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
818 | async_extent->start, |
819 | async_extent->ram_size, | |
820 | ins.objectid, | |
821 | ins.offset, async_extent->pages, | |
822 | async_extent->nr_pages); | |
fce2a4e6 FM |
823 | if (ret) { |
824 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
825 | struct page *p = async_extent->pages[0]; | |
826 | const u64 start = async_extent->start; | |
827 | const u64 end = start + async_extent->ram_size - 1; | |
828 | ||
829 | p->mapping = inode->i_mapping; | |
830 | tree->ops->writepage_end_io_hook(p, start, end, | |
831 | NULL, 0); | |
832 | p->mapping = NULL; | |
833 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
834 | PAGE_END_WRITEBACK | | |
835 | PAGE_SET_ERROR); | |
40ae837b | 836 | free_async_extent_pages(async_extent); |
fce2a4e6 | 837 | } |
771ed689 CM |
838 | alloc_hint = ins.objectid + ins.offset; |
839 | kfree(async_extent); | |
840 | cond_resched(); | |
841 | } | |
dec8f175 | 842 | return; |
3e04e7f1 | 843 | out_free_reserve: |
e570fd27 | 844 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 845 | out_free: |
c2790a2e | 846 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
847 | async_extent->start + |
848 | async_extent->ram_size - 1, | |
c2790a2e | 849 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
850 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
851 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
852 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
853 | PAGE_SET_ERROR); | |
40ae837b | 854 | free_async_extent_pages(async_extent); |
79787eaa | 855 | kfree(async_extent); |
3e04e7f1 | 856 | goto again; |
771ed689 CM |
857 | } |
858 | ||
4b46fce2 JB |
859 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
860 | u64 num_bytes) | |
861 | { | |
862 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
863 | struct extent_map *em; | |
864 | u64 alloc_hint = 0; | |
865 | ||
866 | read_lock(&em_tree->lock); | |
867 | em = search_extent_mapping(em_tree, start, num_bytes); | |
868 | if (em) { | |
869 | /* | |
870 | * if block start isn't an actual block number then find the | |
871 | * first block in this inode and use that as a hint. If that | |
872 | * block is also bogus then just don't worry about it. | |
873 | */ | |
874 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
875 | free_extent_map(em); | |
876 | em = search_extent_mapping(em_tree, 0, 0); | |
877 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
878 | alloc_hint = em->block_start; | |
879 | if (em) | |
880 | free_extent_map(em); | |
881 | } else { | |
882 | alloc_hint = em->block_start; | |
883 | free_extent_map(em); | |
884 | } | |
885 | } | |
886 | read_unlock(&em_tree->lock); | |
887 | ||
888 | return alloc_hint; | |
889 | } | |
890 | ||
771ed689 CM |
891 | /* |
892 | * when extent_io.c finds a delayed allocation range in the file, | |
893 | * the call backs end up in this code. The basic idea is to | |
894 | * allocate extents on disk for the range, and create ordered data structs | |
895 | * in ram to track those extents. | |
896 | * | |
897 | * locked_page is the page that writepage had locked already. We use | |
898 | * it to make sure we don't do extra locks or unlocks. | |
899 | * | |
900 | * *page_started is set to one if we unlock locked_page and do everything | |
901 | * required to start IO on it. It may be clean and already done with | |
902 | * IO when we return. | |
903 | */ | |
00361589 JB |
904 | static noinline int cow_file_range(struct inode *inode, |
905 | struct page *locked_page, | |
906 | u64 start, u64 end, int *page_started, | |
907 | unsigned long *nr_written, | |
908 | int unlock) | |
771ed689 | 909 | { |
00361589 | 910 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
911 | u64 alloc_hint = 0; |
912 | u64 num_bytes; | |
913 | unsigned long ram_size; | |
914 | u64 disk_num_bytes; | |
915 | u64 cur_alloc_size; | |
916 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
917 | struct btrfs_key ins; |
918 | struct extent_map *em; | |
919 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
920 | int ret = 0; | |
921 | ||
02ecd2c2 JB |
922 | if (btrfs_is_free_space_inode(inode)) { |
923 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
924 | ret = -EINVAL; |
925 | goto out_unlock; | |
02ecd2c2 | 926 | } |
771ed689 | 927 | |
fda2832f | 928 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
929 | num_bytes = max(blocksize, num_bytes); |
930 | disk_num_bytes = num_bytes; | |
771ed689 | 931 | |
4cb5300b | 932 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
933 | if (num_bytes < 64 * 1024 && |
934 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 935 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 936 | |
771ed689 CM |
937 | if (start == 0) { |
938 | /* lets try to make an inline extent */ | |
00361589 JB |
939 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
940 | NULL); | |
771ed689 | 941 | if (ret == 0) { |
c2790a2e JB |
942 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
943 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 944 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
945 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
946 | PAGE_END_WRITEBACK); | |
c2167754 | 947 | |
771ed689 CM |
948 | *nr_written = *nr_written + |
949 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
950 | *page_started = 1; | |
771ed689 | 951 | goto out; |
79787eaa | 952 | } else if (ret < 0) { |
79787eaa | 953 | goto out_unlock; |
771ed689 CM |
954 | } |
955 | } | |
956 | ||
957 | BUG_ON(disk_num_bytes > | |
6c41761f | 958 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 959 | |
4b46fce2 | 960 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
961 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
962 | ||
d397712b | 963 | while (disk_num_bytes > 0) { |
a791e35e CM |
964 | unsigned long op; |
965 | ||
287a0ab9 | 966 | cur_alloc_size = disk_num_bytes; |
00361589 | 967 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 968 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 969 | &ins, 1, 1); |
00361589 | 970 | if (ret < 0) |
79787eaa | 971 | goto out_unlock; |
d397712b | 972 | |
172ddd60 | 973 | em = alloc_extent_map(); |
b9aa55be LB |
974 | if (!em) { |
975 | ret = -ENOMEM; | |
ace68bac | 976 | goto out_reserve; |
b9aa55be | 977 | } |
e6dcd2dc | 978 | em->start = start; |
445a6944 | 979 | em->orig_start = em->start; |
771ed689 CM |
980 | ram_size = ins.offset; |
981 | em->len = ins.offset; | |
2ab28f32 JB |
982 | em->mod_start = em->start; |
983 | em->mod_len = em->len; | |
c8b97818 | 984 | |
e6dcd2dc | 985 | em->block_start = ins.objectid; |
c8b97818 | 986 | em->block_len = ins.offset; |
b4939680 | 987 | em->orig_block_len = ins.offset; |
cc95bef6 | 988 | em->ram_bytes = ram_size; |
e6dcd2dc | 989 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 990 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 991 | em->generation = -1; |
c8b97818 | 992 | |
d397712b | 993 | while (1) { |
890871be | 994 | write_lock(&em_tree->lock); |
09a2a8f9 | 995 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 996 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
997 | if (ret != -EEXIST) { |
998 | free_extent_map(em); | |
999 | break; | |
1000 | } | |
1001 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1002 | start + ram_size - 1, 0); |
e6dcd2dc | 1003 | } |
ace68bac LB |
1004 | if (ret) |
1005 | goto out_reserve; | |
e6dcd2dc | 1006 | |
98d20f67 | 1007 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1008 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1009 | ram_size, cur_alloc_size, 0); |
ace68bac | 1010 | if (ret) |
d9f85963 | 1011 | goto out_drop_extent_cache; |
c8b97818 | 1012 | |
17d217fe YZ |
1013 | if (root->root_key.objectid == |
1014 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1015 | ret = btrfs_reloc_clone_csums(inode, start, | |
1016 | cur_alloc_size); | |
00361589 | 1017 | if (ret) |
d9f85963 | 1018 | goto out_drop_extent_cache; |
17d217fe YZ |
1019 | } |
1020 | ||
d397712b | 1021 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1022 | break; |
d397712b | 1023 | |
c8b97818 CM |
1024 | /* we're not doing compressed IO, don't unlock the first |
1025 | * page (which the caller expects to stay locked), don't | |
1026 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1027 | * |
1028 | * Do set the Private2 bit so we know this page was properly | |
1029 | * setup for writepage | |
c8b97818 | 1030 | */ |
c2790a2e JB |
1031 | op = unlock ? PAGE_UNLOCK : 0; |
1032 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1033 | |
c2790a2e JB |
1034 | extent_clear_unlock_delalloc(inode, start, |
1035 | start + ram_size - 1, locked_page, | |
1036 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1037 | op); | |
c8b97818 | 1038 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1039 | num_bytes -= cur_alloc_size; |
1040 | alloc_hint = ins.objectid + ins.offset; | |
1041 | start += cur_alloc_size; | |
b888db2b | 1042 | } |
79787eaa | 1043 | out: |
be20aa9d | 1044 | return ret; |
b7d5b0a8 | 1045 | |
d9f85963 FM |
1046 | out_drop_extent_cache: |
1047 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1048 | out_reserve: |
e570fd27 | 1049 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1050 | out_unlock: |
c2790a2e | 1051 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1052 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1053 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1054 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1055 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1056 | goto out; |
771ed689 | 1057 | } |
c8b97818 | 1058 | |
771ed689 CM |
1059 | /* |
1060 | * work queue call back to started compression on a file and pages | |
1061 | */ | |
1062 | static noinline void async_cow_start(struct btrfs_work *work) | |
1063 | { | |
1064 | struct async_cow *async_cow; | |
1065 | int num_added = 0; | |
1066 | async_cow = container_of(work, struct async_cow, work); | |
1067 | ||
1068 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1069 | async_cow->start, async_cow->end, async_cow, | |
1070 | &num_added); | |
8180ef88 | 1071 | if (num_added == 0) { |
cb77fcd8 | 1072 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1073 | async_cow->inode = NULL; |
8180ef88 | 1074 | } |
771ed689 CM |
1075 | } |
1076 | ||
1077 | /* | |
1078 | * work queue call back to submit previously compressed pages | |
1079 | */ | |
1080 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1081 | { | |
1082 | struct async_cow *async_cow; | |
1083 | struct btrfs_root *root; | |
1084 | unsigned long nr_pages; | |
1085 | ||
1086 | async_cow = container_of(work, struct async_cow, work); | |
1087 | ||
1088 | root = async_cow->root; | |
1089 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1090 | PAGE_CACHE_SHIFT; | |
1091 | ||
66657b31 | 1092 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1093 | 5 * 1024 * 1024 && |
771ed689 CM |
1094 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1095 | wake_up(&root->fs_info->async_submit_wait); | |
1096 | ||
d397712b | 1097 | if (async_cow->inode) |
771ed689 | 1098 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1099 | } |
c8b97818 | 1100 | |
771ed689 CM |
1101 | static noinline void async_cow_free(struct btrfs_work *work) |
1102 | { | |
1103 | struct async_cow *async_cow; | |
1104 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1105 | if (async_cow->inode) |
cb77fcd8 | 1106 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1107 | kfree(async_cow); |
1108 | } | |
1109 | ||
1110 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1111 | u64 start, u64 end, int *page_started, | |
1112 | unsigned long *nr_written) | |
1113 | { | |
1114 | struct async_cow *async_cow; | |
1115 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1116 | unsigned long nr_pages; | |
1117 | u64 cur_end; | |
287082b0 | 1118 | int limit = 10 * 1024 * 1024; |
771ed689 | 1119 | |
a3429ab7 CM |
1120 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1121 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1122 | while (start < end) { |
771ed689 | 1123 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1124 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1125 | async_cow->inode = igrab(inode); |
771ed689 CM |
1126 | async_cow->root = root; |
1127 | async_cow->locked_page = locked_page; | |
1128 | async_cow->start = start; | |
1129 | ||
f79707b0 WS |
1130 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1131 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1132 | cur_end = end; |
1133 | else | |
1134 | cur_end = min(end, start + 512 * 1024 - 1); | |
1135 | ||
1136 | async_cow->end = cur_end; | |
1137 | INIT_LIST_HEAD(&async_cow->extents); | |
1138 | ||
9e0af237 LB |
1139 | btrfs_init_work(&async_cow->work, |
1140 | btrfs_delalloc_helper, | |
1141 | async_cow_start, async_cow_submit, | |
1142 | async_cow_free); | |
771ed689 | 1143 | |
771ed689 CM |
1144 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1145 | PAGE_CACHE_SHIFT; | |
1146 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1147 | ||
afe3d242 QW |
1148 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1149 | &async_cow->work); | |
771ed689 CM |
1150 | |
1151 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1152 | wait_event(root->fs_info->async_submit_wait, | |
1153 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1154 | limit)); | |
1155 | } | |
1156 | ||
d397712b | 1157 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1158 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1159 | wait_event(root->fs_info->async_submit_wait, | |
1160 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1161 | 0)); | |
1162 | } | |
1163 | ||
1164 | *nr_written += nr_pages; | |
1165 | start = cur_end + 1; | |
1166 | } | |
1167 | *page_started = 1; | |
1168 | return 0; | |
be20aa9d CM |
1169 | } |
1170 | ||
d397712b | 1171 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1172 | u64 bytenr, u64 num_bytes) |
1173 | { | |
1174 | int ret; | |
1175 | struct btrfs_ordered_sum *sums; | |
1176 | LIST_HEAD(list); | |
1177 | ||
07d400a6 | 1178 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1179 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1180 | if (ret == 0 && list_empty(&list)) |
1181 | return 0; | |
1182 | ||
1183 | while (!list_empty(&list)) { | |
1184 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1185 | list_del(&sums->list); | |
1186 | kfree(sums); | |
1187 | } | |
1188 | return 1; | |
1189 | } | |
1190 | ||
d352ac68 CM |
1191 | /* |
1192 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1193 | * of the extents that exist in the file, and COWs the file as required. | |
1194 | * | |
1195 | * If no cow copies or snapshots exist, we write directly to the existing | |
1196 | * blocks on disk | |
1197 | */ | |
7f366cfe CM |
1198 | static noinline int run_delalloc_nocow(struct inode *inode, |
1199 | struct page *locked_page, | |
771ed689 CM |
1200 | u64 start, u64 end, int *page_started, int force, |
1201 | unsigned long *nr_written) | |
be20aa9d | 1202 | { |
be20aa9d | 1203 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1204 | struct btrfs_trans_handle *trans; |
be20aa9d | 1205 | struct extent_buffer *leaf; |
be20aa9d | 1206 | struct btrfs_path *path; |
80ff3856 | 1207 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1208 | struct btrfs_key found_key; |
80ff3856 YZ |
1209 | u64 cow_start; |
1210 | u64 cur_offset; | |
1211 | u64 extent_end; | |
5d4f98a2 | 1212 | u64 extent_offset; |
80ff3856 YZ |
1213 | u64 disk_bytenr; |
1214 | u64 num_bytes; | |
b4939680 | 1215 | u64 disk_num_bytes; |
cc95bef6 | 1216 | u64 ram_bytes; |
80ff3856 | 1217 | int extent_type; |
79787eaa | 1218 | int ret, err; |
d899e052 | 1219 | int type; |
80ff3856 YZ |
1220 | int nocow; |
1221 | int check_prev = 1; | |
82d5902d | 1222 | bool nolock; |
33345d01 | 1223 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1224 | |
1225 | path = btrfs_alloc_path(); | |
17ca04af | 1226 | if (!path) { |
c2790a2e JB |
1227 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1228 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1229 | EXTENT_DO_ACCOUNTING | |
1230 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1231 | PAGE_CLEAR_DIRTY | |
1232 | PAGE_SET_WRITEBACK | | |
1233 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1234 | return -ENOMEM; |
17ca04af | 1235 | } |
82d5902d | 1236 | |
83eea1f1 | 1237 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1238 | |
1239 | if (nolock) | |
7a7eaa40 | 1240 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1241 | else |
7a7eaa40 | 1242 | trans = btrfs_join_transaction(root); |
ff5714cc | 1243 | |
79787eaa | 1244 | if (IS_ERR(trans)) { |
c2790a2e JB |
1245 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1246 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1247 | EXTENT_DO_ACCOUNTING | |
1248 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1249 | PAGE_CLEAR_DIRTY | |
1250 | PAGE_SET_WRITEBACK | | |
1251 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1252 | btrfs_free_path(path); |
1253 | return PTR_ERR(trans); | |
1254 | } | |
1255 | ||
74b21075 | 1256 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1257 | |
80ff3856 YZ |
1258 | cow_start = (u64)-1; |
1259 | cur_offset = start; | |
1260 | while (1) { | |
33345d01 | 1261 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1262 | cur_offset, 0); |
d788a349 | 1263 | if (ret < 0) |
79787eaa | 1264 | goto error; |
80ff3856 YZ |
1265 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1266 | leaf = path->nodes[0]; | |
1267 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1268 | path->slots[0] - 1); | |
33345d01 | 1269 | if (found_key.objectid == ino && |
80ff3856 YZ |
1270 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1271 | path->slots[0]--; | |
1272 | } | |
1273 | check_prev = 0; | |
1274 | next_slot: | |
1275 | leaf = path->nodes[0]; | |
1276 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1277 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1278 | if (ret < 0) |
79787eaa | 1279 | goto error; |
80ff3856 YZ |
1280 | if (ret > 0) |
1281 | break; | |
1282 | leaf = path->nodes[0]; | |
1283 | } | |
be20aa9d | 1284 | |
80ff3856 YZ |
1285 | nocow = 0; |
1286 | disk_bytenr = 0; | |
17d217fe | 1287 | num_bytes = 0; |
80ff3856 YZ |
1288 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1289 | ||
33345d01 | 1290 | if (found_key.objectid > ino || |
80ff3856 YZ |
1291 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1292 | found_key.offset > end) | |
1293 | break; | |
1294 | ||
1295 | if (found_key.offset > cur_offset) { | |
1296 | extent_end = found_key.offset; | |
e9061e21 | 1297 | extent_type = 0; |
80ff3856 YZ |
1298 | goto out_check; |
1299 | } | |
1300 | ||
1301 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1302 | struct btrfs_file_extent_item); | |
1303 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1304 | ||
cc95bef6 | 1305 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1306 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1307 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1308 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1309 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1310 | extent_end = found_key.offset + |
1311 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1312 | disk_num_bytes = |
1313 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1314 | if (extent_end <= start) { |
1315 | path->slots[0]++; | |
1316 | goto next_slot; | |
1317 | } | |
17d217fe YZ |
1318 | if (disk_bytenr == 0) |
1319 | goto out_check; | |
80ff3856 YZ |
1320 | if (btrfs_file_extent_compression(leaf, fi) || |
1321 | btrfs_file_extent_encryption(leaf, fi) || | |
1322 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1323 | goto out_check; | |
d899e052 YZ |
1324 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1325 | goto out_check; | |
d2fb3437 | 1326 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1327 | goto out_check; |
33345d01 | 1328 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1329 | found_key.offset - |
1330 | extent_offset, disk_bytenr)) | |
17d217fe | 1331 | goto out_check; |
5d4f98a2 | 1332 | disk_bytenr += extent_offset; |
17d217fe YZ |
1333 | disk_bytenr += cur_offset - found_key.offset; |
1334 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1335 | /* |
1336 | * if there are pending snapshots for this root, | |
1337 | * we fall into common COW way. | |
1338 | */ | |
1339 | if (!nolock) { | |
9ea24bbe | 1340 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1341 | if (!err) |
1342 | goto out_check; | |
1343 | } | |
17d217fe YZ |
1344 | /* |
1345 | * force cow if csum exists in the range. | |
1346 | * this ensure that csum for a given extent are | |
1347 | * either valid or do not exist. | |
1348 | */ | |
1349 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1350 | goto out_check; | |
80ff3856 YZ |
1351 | nocow = 1; |
1352 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1353 | extent_end = found_key.offset + | |
514ac8ad CM |
1354 | btrfs_file_extent_inline_len(leaf, |
1355 | path->slots[0], fi); | |
80ff3856 YZ |
1356 | extent_end = ALIGN(extent_end, root->sectorsize); |
1357 | } else { | |
1358 | BUG_ON(1); | |
1359 | } | |
1360 | out_check: | |
1361 | if (extent_end <= start) { | |
1362 | path->slots[0]++; | |
e9894fd3 | 1363 | if (!nolock && nocow) |
9ea24bbe | 1364 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1365 | goto next_slot; |
1366 | } | |
1367 | if (!nocow) { | |
1368 | if (cow_start == (u64)-1) | |
1369 | cow_start = cur_offset; | |
1370 | cur_offset = extent_end; | |
1371 | if (cur_offset > end) | |
1372 | break; | |
1373 | path->slots[0]++; | |
1374 | goto next_slot; | |
7ea394f1 YZ |
1375 | } |
1376 | ||
b3b4aa74 | 1377 | btrfs_release_path(path); |
80ff3856 | 1378 | if (cow_start != (u64)-1) { |
00361589 JB |
1379 | ret = cow_file_range(inode, locked_page, |
1380 | cow_start, found_key.offset - 1, | |
1381 | page_started, nr_written, 1); | |
e9894fd3 WS |
1382 | if (ret) { |
1383 | if (!nolock && nocow) | |
9ea24bbe | 1384 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1385 | goto error; |
e9894fd3 | 1386 | } |
80ff3856 | 1387 | cow_start = (u64)-1; |
7ea394f1 | 1388 | } |
80ff3856 | 1389 | |
d899e052 YZ |
1390 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1391 | struct extent_map *em; | |
1392 | struct extent_map_tree *em_tree; | |
1393 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1394 | em = alloc_extent_map(); |
79787eaa | 1395 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1396 | em->start = cur_offset; |
70c8a91c | 1397 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1398 | em->len = num_bytes; |
1399 | em->block_len = num_bytes; | |
1400 | em->block_start = disk_bytenr; | |
b4939680 | 1401 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1402 | em->ram_bytes = ram_bytes; |
d899e052 | 1403 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1404 | em->mod_start = em->start; |
1405 | em->mod_len = em->len; | |
d899e052 | 1406 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1407 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1408 | em->generation = -1; |
d899e052 | 1409 | while (1) { |
890871be | 1410 | write_lock(&em_tree->lock); |
09a2a8f9 | 1411 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1412 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1413 | if (ret != -EEXIST) { |
1414 | free_extent_map(em); | |
1415 | break; | |
1416 | } | |
1417 | btrfs_drop_extent_cache(inode, em->start, | |
1418 | em->start + em->len - 1, 0); | |
1419 | } | |
1420 | type = BTRFS_ORDERED_PREALLOC; | |
1421 | } else { | |
1422 | type = BTRFS_ORDERED_NOCOW; | |
1423 | } | |
80ff3856 YZ |
1424 | |
1425 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1426 | num_bytes, num_bytes, type); |
79787eaa | 1427 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1428 | |
efa56464 YZ |
1429 | if (root->root_key.objectid == |
1430 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1431 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1432 | num_bytes); | |
e9894fd3 WS |
1433 | if (ret) { |
1434 | if (!nolock && nocow) | |
9ea24bbe | 1435 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1436 | goto error; |
e9894fd3 | 1437 | } |
efa56464 YZ |
1438 | } |
1439 | ||
c2790a2e JB |
1440 | extent_clear_unlock_delalloc(inode, cur_offset, |
1441 | cur_offset + num_bytes - 1, | |
1442 | locked_page, EXTENT_LOCKED | | |
1443 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1444 | PAGE_SET_PRIVATE2); | |
e9894fd3 | 1445 | if (!nolock && nocow) |
9ea24bbe | 1446 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1447 | cur_offset = extent_end; |
1448 | if (cur_offset > end) | |
1449 | break; | |
be20aa9d | 1450 | } |
b3b4aa74 | 1451 | btrfs_release_path(path); |
80ff3856 | 1452 | |
17ca04af | 1453 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1454 | cow_start = cur_offset; |
17ca04af JB |
1455 | cur_offset = end; |
1456 | } | |
1457 | ||
80ff3856 | 1458 | if (cow_start != (u64)-1) { |
00361589 JB |
1459 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1460 | page_started, nr_written, 1); | |
d788a349 | 1461 | if (ret) |
79787eaa | 1462 | goto error; |
80ff3856 YZ |
1463 | } |
1464 | ||
79787eaa | 1465 | error: |
a698d075 | 1466 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1467 | if (!ret) |
1468 | ret = err; | |
1469 | ||
17ca04af | 1470 | if (ret && cur_offset < end) |
c2790a2e JB |
1471 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1472 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1473 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1474 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1475 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1476 | PAGE_SET_WRITEBACK | |
1477 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1478 | btrfs_free_path(path); |
79787eaa | 1479 | return ret; |
be20aa9d CM |
1480 | } |
1481 | ||
47059d93 WS |
1482 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1483 | { | |
1484 | ||
1485 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1486 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1487 | return 0; | |
1488 | ||
1489 | /* | |
1490 | * @defrag_bytes is a hint value, no spinlock held here, | |
1491 | * if is not zero, it means the file is defragging. | |
1492 | * Force cow if given extent needs to be defragged. | |
1493 | */ | |
1494 | if (BTRFS_I(inode)->defrag_bytes && | |
1495 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1496 | EXTENT_DEFRAG, 0, NULL)) | |
1497 | return 1; | |
1498 | ||
1499 | return 0; | |
1500 | } | |
1501 | ||
d352ac68 CM |
1502 | /* |
1503 | * extent_io.c call back to do delayed allocation processing | |
1504 | */ | |
c8b97818 | 1505 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1506 | u64 start, u64 end, int *page_started, |
1507 | unsigned long *nr_written) | |
be20aa9d | 1508 | { |
be20aa9d | 1509 | int ret; |
47059d93 | 1510 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1511 | |
47059d93 | 1512 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1513 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1514 | page_started, 1, nr_written); |
47059d93 | 1515 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1516 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1517 | page_started, 0, nr_written); |
7816030e | 1518 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1519 | ret = cow_file_range(inode, locked_page, start, end, |
1520 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1521 | } else { |
1522 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1523 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1524 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1525 | page_started, nr_written); |
7ddf5a42 | 1526 | } |
b888db2b CM |
1527 | return ret; |
1528 | } | |
1529 | ||
1bf85046 JM |
1530 | static void btrfs_split_extent_hook(struct inode *inode, |
1531 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1532 | { |
dcab6a3b JB |
1533 | u64 size; |
1534 | ||
0ca1f7ce | 1535 | /* not delalloc, ignore it */ |
9ed74f2d | 1536 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1537 | return; |
9ed74f2d | 1538 | |
dcab6a3b JB |
1539 | size = orig->end - orig->start + 1; |
1540 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
1541 | u64 num_extents; | |
1542 | u64 new_size; | |
1543 | ||
1544 | /* | |
1545 | * We need the largest size of the remaining extent to see if we | |
1546 | * need to add a new outstanding extent. Think of the following | |
1547 | * case | |
1548 | * | |
1549 | * [MEAX_EXTENT_SIZEx2 - 4k][4k] | |
1550 | * | |
1551 | * The new_size would just be 4k and we'd think we had enough | |
1552 | * outstanding extents for this if we only took one side of the | |
1553 | * split, same goes for the other direction. We need to see if | |
1554 | * the larger size still is the same amount of extents as the | |
1555 | * original size, because if it is we need to add a new | |
1556 | * outstanding extent. But if we split up and the larger size | |
1557 | * is less than the original then we are good to go since we've | |
1558 | * already accounted for the extra extent in our original | |
1559 | * accounting. | |
1560 | */ | |
1561 | new_size = orig->end - split + 1; | |
1562 | if ((split - orig->start) > new_size) | |
1563 | new_size = split - orig->start; | |
1564 | ||
1565 | num_extents = div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1566 | BTRFS_MAX_EXTENT_SIZE); | |
1567 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1568 | BTRFS_MAX_EXTENT_SIZE) < num_extents) | |
1569 | return; | |
1570 | } | |
1571 | ||
9e0baf60 JB |
1572 | spin_lock(&BTRFS_I(inode)->lock); |
1573 | BTRFS_I(inode)->outstanding_extents++; | |
1574 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1575 | } |
1576 | ||
1577 | /* | |
1578 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1579 | * extents so we can keep track of new extents that are just merged onto old | |
1580 | * extents, such as when we are doing sequential writes, so we can properly | |
1581 | * account for the metadata space we'll need. | |
1582 | */ | |
1bf85046 JM |
1583 | static void btrfs_merge_extent_hook(struct inode *inode, |
1584 | struct extent_state *new, | |
1585 | struct extent_state *other) | |
9ed74f2d | 1586 | { |
dcab6a3b JB |
1587 | u64 new_size, old_size; |
1588 | u64 num_extents; | |
1589 | ||
9ed74f2d JB |
1590 | /* not delalloc, ignore it */ |
1591 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1592 | return; |
9ed74f2d | 1593 | |
dcab6a3b | 1594 | old_size = other->end - other->start + 1; |
8461a3de JB |
1595 | if (old_size < (new->end - new->start + 1)) |
1596 | old_size = (new->end - new->start + 1); | |
1597 | if (new->start > other->start) | |
1598 | new_size = new->end - other->start + 1; | |
1599 | else | |
1600 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1601 | |
1602 | /* we're not bigger than the max, unreserve the space and go */ | |
1603 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1604 | spin_lock(&BTRFS_I(inode)->lock); | |
1605 | BTRFS_I(inode)->outstanding_extents--; | |
1606 | spin_unlock(&BTRFS_I(inode)->lock); | |
1607 | return; | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * If we grew by another max_extent, just return, we want to keep that | |
1612 | * reserved amount. | |
1613 | */ | |
1614 | num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1615 | BTRFS_MAX_EXTENT_SIZE); | |
1616 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1617 | BTRFS_MAX_EXTENT_SIZE) > num_extents) | |
1618 | return; | |
1619 | ||
9e0baf60 JB |
1620 | spin_lock(&BTRFS_I(inode)->lock); |
1621 | BTRFS_I(inode)->outstanding_extents--; | |
1622 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1623 | } |
1624 | ||
eb73c1b7 MX |
1625 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1626 | struct inode *inode) | |
1627 | { | |
1628 | spin_lock(&root->delalloc_lock); | |
1629 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1630 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1631 | &root->delalloc_inodes); | |
1632 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1633 | &BTRFS_I(inode)->runtime_flags); | |
1634 | root->nr_delalloc_inodes++; | |
1635 | if (root->nr_delalloc_inodes == 1) { | |
1636 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1637 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1638 | list_add_tail(&root->delalloc_root, | |
1639 | &root->fs_info->delalloc_roots); | |
1640 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1641 | } | |
1642 | } | |
1643 | spin_unlock(&root->delalloc_lock); | |
1644 | } | |
1645 | ||
1646 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1647 | struct inode *inode) | |
1648 | { | |
1649 | spin_lock(&root->delalloc_lock); | |
1650 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1651 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1652 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1653 | &BTRFS_I(inode)->runtime_flags); | |
1654 | root->nr_delalloc_inodes--; | |
1655 | if (!root->nr_delalloc_inodes) { | |
1656 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1657 | BUG_ON(list_empty(&root->delalloc_root)); | |
1658 | list_del_init(&root->delalloc_root); | |
1659 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1660 | } | |
1661 | } | |
1662 | spin_unlock(&root->delalloc_lock); | |
1663 | } | |
1664 | ||
d352ac68 CM |
1665 | /* |
1666 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1667 | * bytes in this file, and to maintain the list of inodes that | |
1668 | * have pending delalloc work to be done. | |
1669 | */ | |
1bf85046 | 1670 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1671 | struct extent_state *state, unsigned *bits) |
291d673e | 1672 | { |
9ed74f2d | 1673 | |
47059d93 WS |
1674 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1675 | WARN_ON(1); | |
75eff68e CM |
1676 | /* |
1677 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1678 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1679 | * bit, which is only set or cleared with irqs on |
1680 | */ | |
0ca1f7ce | 1681 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1682 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1683 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1684 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1685 | |
9e0baf60 | 1686 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1687 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1688 | } else { |
1689 | spin_lock(&BTRFS_I(inode)->lock); | |
1690 | BTRFS_I(inode)->outstanding_extents++; | |
1691 | spin_unlock(&BTRFS_I(inode)->lock); | |
1692 | } | |
287a0ab9 | 1693 | |
963d678b MX |
1694 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1695 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1696 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1697 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1698 | if (*bits & EXTENT_DEFRAG) |
1699 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1700 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1701 | &BTRFS_I(inode)->runtime_flags)) |
1702 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1703 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1704 | } |
291d673e CM |
1705 | } |
1706 | ||
d352ac68 CM |
1707 | /* |
1708 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1709 | */ | |
1bf85046 | 1710 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1711 | struct extent_state *state, |
9ee49a04 | 1712 | unsigned *bits) |
291d673e | 1713 | { |
47059d93 | 1714 | u64 len = state->end + 1 - state->start; |
dcab6a3b JB |
1715 | u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1, |
1716 | BTRFS_MAX_EXTENT_SIZE); | |
47059d93 WS |
1717 | |
1718 | spin_lock(&BTRFS_I(inode)->lock); | |
1719 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1720 | BTRFS_I(inode)->defrag_bytes -= len; | |
1721 | spin_unlock(&BTRFS_I(inode)->lock); | |
1722 | ||
75eff68e CM |
1723 | /* |
1724 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1725 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1726 | * bit, which is only set or cleared with irqs on |
1727 | */ | |
0ca1f7ce | 1728 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1729 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1730 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1731 | |
9e0baf60 | 1732 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1733 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1734 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1735 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1736 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1737 | spin_unlock(&BTRFS_I(inode)->lock); |
1738 | } | |
0ca1f7ce | 1739 | |
b6d08f06 JB |
1740 | /* |
1741 | * We don't reserve metadata space for space cache inodes so we | |
1742 | * don't need to call dellalloc_release_metadata if there is an | |
1743 | * error. | |
1744 | */ | |
1745 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1746 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1747 | btrfs_delalloc_release_metadata(inode, len); |
1748 | ||
0cb59c99 | 1749 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1750 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1751 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1752 | |
963d678b MX |
1753 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1754 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1755 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1756 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1757 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1758 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1759 | &BTRFS_I(inode)->runtime_flags)) |
1760 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1761 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1762 | } |
291d673e CM |
1763 | } |
1764 | ||
d352ac68 CM |
1765 | /* |
1766 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1767 | * we don't create bios that span stripes or chunks | |
1768 | */ | |
64a16701 | 1769 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1770 | size_t size, struct bio *bio, |
1771 | unsigned long bio_flags) | |
239b14b3 CM |
1772 | { |
1773 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1774 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1775 | u64 length = 0; |
1776 | u64 map_length; | |
239b14b3 CM |
1777 | int ret; |
1778 | ||
771ed689 CM |
1779 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1780 | return 0; | |
1781 | ||
4f024f37 | 1782 | length = bio->bi_iter.bi_size; |
239b14b3 | 1783 | map_length = length; |
64a16701 | 1784 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1785 | &map_length, NULL, 0); |
3ec706c8 | 1786 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1787 | BUG_ON(ret < 0); |
d397712b | 1788 | if (map_length < length + size) |
239b14b3 | 1789 | return 1; |
3444a972 | 1790 | return 0; |
239b14b3 CM |
1791 | } |
1792 | ||
d352ac68 CM |
1793 | /* |
1794 | * in order to insert checksums into the metadata in large chunks, | |
1795 | * we wait until bio submission time. All the pages in the bio are | |
1796 | * checksummed and sums are attached onto the ordered extent record. | |
1797 | * | |
1798 | * At IO completion time the cums attached on the ordered extent record | |
1799 | * are inserted into the btree | |
1800 | */ | |
d397712b CM |
1801 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1802 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1803 | unsigned long bio_flags, |
1804 | u64 bio_offset) | |
065631f6 | 1805 | { |
065631f6 | 1806 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1807 | int ret = 0; |
e015640f | 1808 | |
d20f7043 | 1809 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1810 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1811 | return 0; |
1812 | } | |
e015640f | 1813 | |
4a69a410 CM |
1814 | /* |
1815 | * in order to insert checksums into the metadata in large chunks, | |
1816 | * we wait until bio submission time. All the pages in the bio are | |
1817 | * checksummed and sums are attached onto the ordered extent record. | |
1818 | * | |
1819 | * At IO completion time the cums attached on the ordered extent record | |
1820 | * are inserted into the btree | |
1821 | */ | |
b2950863 | 1822 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1823 | int mirror_num, unsigned long bio_flags, |
1824 | u64 bio_offset) | |
4a69a410 CM |
1825 | { |
1826 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1827 | int ret; |
1828 | ||
1829 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1830 | if (ret) | |
1831 | bio_endio(bio, ret); | |
1832 | return ret; | |
44b8bd7e CM |
1833 | } |
1834 | ||
d352ac68 | 1835 | /* |
cad321ad CM |
1836 | * extent_io.c submission hook. This does the right thing for csum calculation |
1837 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1838 | */ |
b2950863 | 1839 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1840 | int mirror_num, unsigned long bio_flags, |
1841 | u64 bio_offset) | |
44b8bd7e CM |
1842 | { |
1843 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1844 | int ret = 0; | |
19b9bdb0 | 1845 | int skip_sum; |
0417341e | 1846 | int metadata = 0; |
b812ce28 | 1847 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1848 | |
6cbff00f | 1849 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1850 | |
83eea1f1 | 1851 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1852 | metadata = 2; |
1853 | ||
7b6d91da | 1854 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1855 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1856 | if (ret) | |
61891923 | 1857 | goto out; |
5fd02043 | 1858 | |
d20f7043 | 1859 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1860 | ret = btrfs_submit_compressed_read(inode, bio, |
1861 | mirror_num, | |
1862 | bio_flags); | |
1863 | goto out; | |
c2db1073 TI |
1864 | } else if (!skip_sum) { |
1865 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1866 | if (ret) | |
61891923 | 1867 | goto out; |
c2db1073 | 1868 | } |
4d1b5fb4 | 1869 | goto mapit; |
b812ce28 | 1870 | } else if (async && !skip_sum) { |
17d217fe YZ |
1871 | /* csum items have already been cloned */ |
1872 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1873 | goto mapit; | |
19b9bdb0 | 1874 | /* we're doing a write, do the async checksumming */ |
61891923 | 1875 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1876 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1877 | bio_flags, bio_offset, |
1878 | __btrfs_submit_bio_start, | |
4a69a410 | 1879 | __btrfs_submit_bio_done); |
61891923 | 1880 | goto out; |
b812ce28 JB |
1881 | } else if (!skip_sum) { |
1882 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1883 | if (ret) | |
1884 | goto out; | |
19b9bdb0 CM |
1885 | } |
1886 | ||
0b86a832 | 1887 | mapit: |
61891923 SB |
1888 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1889 | ||
1890 | out: | |
1891 | if (ret < 0) | |
1892 | bio_endio(bio, ret); | |
1893 | return ret; | |
065631f6 | 1894 | } |
6885f308 | 1895 | |
d352ac68 CM |
1896 | /* |
1897 | * given a list of ordered sums record them in the inode. This happens | |
1898 | * at IO completion time based on sums calculated at bio submission time. | |
1899 | */ | |
ba1da2f4 | 1900 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1901 | struct inode *inode, u64 file_offset, |
1902 | struct list_head *list) | |
1903 | { | |
e6dcd2dc CM |
1904 | struct btrfs_ordered_sum *sum; |
1905 | ||
c6e30871 | 1906 | list_for_each_entry(sum, list, list) { |
39847c4d | 1907 | trans->adding_csums = 1; |
d20f7043 CM |
1908 | btrfs_csum_file_blocks(trans, |
1909 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1910 | trans->adding_csums = 0; |
e6dcd2dc CM |
1911 | } |
1912 | return 0; | |
1913 | } | |
1914 | ||
2ac55d41 JB |
1915 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1916 | struct extent_state **cached_state) | |
ea8c2819 | 1917 | { |
6c1500f2 | 1918 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1919 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1920 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1921 | } |
1922 | ||
d352ac68 | 1923 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1924 | struct btrfs_writepage_fixup { |
1925 | struct page *page; | |
1926 | struct btrfs_work work; | |
1927 | }; | |
1928 | ||
b2950863 | 1929 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1930 | { |
1931 | struct btrfs_writepage_fixup *fixup; | |
1932 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1933 | struct extent_state *cached_state = NULL; |
247e743c CM |
1934 | struct page *page; |
1935 | struct inode *inode; | |
1936 | u64 page_start; | |
1937 | u64 page_end; | |
87826df0 | 1938 | int ret; |
247e743c CM |
1939 | |
1940 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1941 | page = fixup->page; | |
4a096752 | 1942 | again: |
247e743c CM |
1943 | lock_page(page); |
1944 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1945 | ClearPageChecked(page); | |
1946 | goto out_page; | |
1947 | } | |
1948 | ||
1949 | inode = page->mapping->host; | |
1950 | page_start = page_offset(page); | |
1951 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1952 | ||
2ac55d41 | 1953 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1954 | &cached_state); |
4a096752 CM |
1955 | |
1956 | /* already ordered? We're done */ | |
8b62b72b | 1957 | if (PagePrivate2(page)) |
247e743c | 1958 | goto out; |
4a096752 CM |
1959 | |
1960 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1961 | if (ordered) { | |
2ac55d41 JB |
1962 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1963 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1964 | unlock_page(page); |
1965 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1966 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1967 | goto again; |
1968 | } | |
247e743c | 1969 | |
87826df0 JM |
1970 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1971 | if (ret) { | |
1972 | mapping_set_error(page->mapping, ret); | |
1973 | end_extent_writepage(page, ret, page_start, page_end); | |
1974 | ClearPageChecked(page); | |
1975 | goto out; | |
1976 | } | |
1977 | ||
2ac55d41 | 1978 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1979 | ClearPageChecked(page); |
87826df0 | 1980 | set_page_dirty(page); |
247e743c | 1981 | out: |
2ac55d41 JB |
1982 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1983 | &cached_state, GFP_NOFS); | |
247e743c CM |
1984 | out_page: |
1985 | unlock_page(page); | |
1986 | page_cache_release(page); | |
b897abec | 1987 | kfree(fixup); |
247e743c CM |
1988 | } |
1989 | ||
1990 | /* | |
1991 | * There are a few paths in the higher layers of the kernel that directly | |
1992 | * set the page dirty bit without asking the filesystem if it is a | |
1993 | * good idea. This causes problems because we want to make sure COW | |
1994 | * properly happens and the data=ordered rules are followed. | |
1995 | * | |
c8b97818 | 1996 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1997 | * hasn't been properly setup for IO. We kick off an async process |
1998 | * to fix it up. The async helper will wait for ordered extents, set | |
1999 | * the delalloc bit and make it safe to write the page. | |
2000 | */ | |
b2950863 | 2001 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2002 | { |
2003 | struct inode *inode = page->mapping->host; | |
2004 | struct btrfs_writepage_fixup *fixup; | |
2005 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 2006 | |
8b62b72b CM |
2007 | /* this page is properly in the ordered list */ |
2008 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2009 | return 0; |
2010 | ||
2011 | if (PageChecked(page)) | |
2012 | return -EAGAIN; | |
2013 | ||
2014 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2015 | if (!fixup) | |
2016 | return -EAGAIN; | |
f421950f | 2017 | |
247e743c CM |
2018 | SetPageChecked(page); |
2019 | page_cache_get(page); | |
9e0af237 LB |
2020 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2021 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2022 | fixup->page = page; |
dc6e3209 | 2023 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 2024 | return -EBUSY; |
247e743c CM |
2025 | } |
2026 | ||
d899e052 YZ |
2027 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2028 | struct inode *inode, u64 file_pos, | |
2029 | u64 disk_bytenr, u64 disk_num_bytes, | |
2030 | u64 num_bytes, u64 ram_bytes, | |
2031 | u8 compression, u8 encryption, | |
2032 | u16 other_encoding, int extent_type) | |
2033 | { | |
2034 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2035 | struct btrfs_file_extent_item *fi; | |
2036 | struct btrfs_path *path; | |
2037 | struct extent_buffer *leaf; | |
2038 | struct btrfs_key ins; | |
1acae57b | 2039 | int extent_inserted = 0; |
d899e052 YZ |
2040 | int ret; |
2041 | ||
2042 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2043 | if (!path) |
2044 | return -ENOMEM; | |
d899e052 | 2045 | |
a1ed835e CM |
2046 | /* |
2047 | * we may be replacing one extent in the tree with another. | |
2048 | * The new extent is pinned in the extent map, and we don't want | |
2049 | * to drop it from the cache until it is completely in the btree. | |
2050 | * | |
2051 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2052 | * the caller is expected to unpin it and allow it to be merged | |
2053 | * with the others. | |
2054 | */ | |
1acae57b FDBM |
2055 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2056 | file_pos + num_bytes, NULL, 0, | |
2057 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2058 | if (ret) |
2059 | goto out; | |
d899e052 | 2060 | |
1acae57b FDBM |
2061 | if (!extent_inserted) { |
2062 | ins.objectid = btrfs_ino(inode); | |
2063 | ins.offset = file_pos; | |
2064 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2065 | ||
2066 | path->leave_spinning = 1; | |
2067 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2068 | sizeof(*fi)); | |
2069 | if (ret) | |
2070 | goto out; | |
2071 | } | |
d899e052 YZ |
2072 | leaf = path->nodes[0]; |
2073 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2074 | struct btrfs_file_extent_item); | |
2075 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2076 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2077 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2078 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2079 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2080 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2081 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2082 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2083 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2084 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2085 | |
d899e052 | 2086 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2087 | btrfs_release_path(path); |
d899e052 YZ |
2088 | |
2089 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2090 | |
2091 | ins.objectid = disk_bytenr; | |
2092 | ins.offset = disk_num_bytes; | |
2093 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2094 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2095 | root->root_key.objectid, | |
33345d01 | 2096 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 2097 | out: |
d899e052 | 2098 | btrfs_free_path(path); |
b9473439 | 2099 | |
79787eaa | 2100 | return ret; |
d899e052 YZ |
2101 | } |
2102 | ||
38c227d8 LB |
2103 | /* snapshot-aware defrag */ |
2104 | struct sa_defrag_extent_backref { | |
2105 | struct rb_node node; | |
2106 | struct old_sa_defrag_extent *old; | |
2107 | u64 root_id; | |
2108 | u64 inum; | |
2109 | u64 file_pos; | |
2110 | u64 extent_offset; | |
2111 | u64 num_bytes; | |
2112 | u64 generation; | |
2113 | }; | |
2114 | ||
2115 | struct old_sa_defrag_extent { | |
2116 | struct list_head list; | |
2117 | struct new_sa_defrag_extent *new; | |
2118 | ||
2119 | u64 extent_offset; | |
2120 | u64 bytenr; | |
2121 | u64 offset; | |
2122 | u64 len; | |
2123 | int count; | |
2124 | }; | |
2125 | ||
2126 | struct new_sa_defrag_extent { | |
2127 | struct rb_root root; | |
2128 | struct list_head head; | |
2129 | struct btrfs_path *path; | |
2130 | struct inode *inode; | |
2131 | u64 file_pos; | |
2132 | u64 len; | |
2133 | u64 bytenr; | |
2134 | u64 disk_len; | |
2135 | u8 compress_type; | |
2136 | }; | |
2137 | ||
2138 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2139 | struct sa_defrag_extent_backref *b2) | |
2140 | { | |
2141 | if (b1->root_id < b2->root_id) | |
2142 | return -1; | |
2143 | else if (b1->root_id > b2->root_id) | |
2144 | return 1; | |
2145 | ||
2146 | if (b1->inum < b2->inum) | |
2147 | return -1; | |
2148 | else if (b1->inum > b2->inum) | |
2149 | return 1; | |
2150 | ||
2151 | if (b1->file_pos < b2->file_pos) | |
2152 | return -1; | |
2153 | else if (b1->file_pos > b2->file_pos) | |
2154 | return 1; | |
2155 | ||
2156 | /* | |
2157 | * [------------------------------] ===> (a range of space) | |
2158 | * |<--->| |<---->| =============> (fs/file tree A) | |
2159 | * |<---------------------------->| ===> (fs/file tree B) | |
2160 | * | |
2161 | * A range of space can refer to two file extents in one tree while | |
2162 | * refer to only one file extent in another tree. | |
2163 | * | |
2164 | * So we may process a disk offset more than one time(two extents in A) | |
2165 | * and locate at the same extent(one extent in B), then insert two same | |
2166 | * backrefs(both refer to the extent in B). | |
2167 | */ | |
2168 | return 0; | |
2169 | } | |
2170 | ||
2171 | static void backref_insert(struct rb_root *root, | |
2172 | struct sa_defrag_extent_backref *backref) | |
2173 | { | |
2174 | struct rb_node **p = &root->rb_node; | |
2175 | struct rb_node *parent = NULL; | |
2176 | struct sa_defrag_extent_backref *entry; | |
2177 | int ret; | |
2178 | ||
2179 | while (*p) { | |
2180 | parent = *p; | |
2181 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2182 | ||
2183 | ret = backref_comp(backref, entry); | |
2184 | if (ret < 0) | |
2185 | p = &(*p)->rb_left; | |
2186 | else | |
2187 | p = &(*p)->rb_right; | |
2188 | } | |
2189 | ||
2190 | rb_link_node(&backref->node, parent, p); | |
2191 | rb_insert_color(&backref->node, root); | |
2192 | } | |
2193 | ||
2194 | /* | |
2195 | * Note the backref might has changed, and in this case we just return 0. | |
2196 | */ | |
2197 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2198 | void *ctx) | |
2199 | { | |
2200 | struct btrfs_file_extent_item *extent; | |
2201 | struct btrfs_fs_info *fs_info; | |
2202 | struct old_sa_defrag_extent *old = ctx; | |
2203 | struct new_sa_defrag_extent *new = old->new; | |
2204 | struct btrfs_path *path = new->path; | |
2205 | struct btrfs_key key; | |
2206 | struct btrfs_root *root; | |
2207 | struct sa_defrag_extent_backref *backref; | |
2208 | struct extent_buffer *leaf; | |
2209 | struct inode *inode = new->inode; | |
2210 | int slot; | |
2211 | int ret; | |
2212 | u64 extent_offset; | |
2213 | u64 num_bytes; | |
2214 | ||
2215 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2216 | inum == btrfs_ino(inode)) | |
2217 | return 0; | |
2218 | ||
2219 | key.objectid = root_id; | |
2220 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2221 | key.offset = (u64)-1; | |
2222 | ||
2223 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2224 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2225 | if (IS_ERR(root)) { | |
2226 | if (PTR_ERR(root) == -ENOENT) | |
2227 | return 0; | |
2228 | WARN_ON(1); | |
2229 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2230 | inum, offset, root_id); | |
2231 | return PTR_ERR(root); | |
2232 | } | |
2233 | ||
2234 | key.objectid = inum; | |
2235 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2236 | if (offset > (u64)-1 << 32) | |
2237 | key.offset = 0; | |
2238 | else | |
2239 | key.offset = offset; | |
2240 | ||
2241 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2242 | if (WARN_ON(ret < 0)) |
38c227d8 | 2243 | return ret; |
50f1319c | 2244 | ret = 0; |
38c227d8 LB |
2245 | |
2246 | while (1) { | |
2247 | cond_resched(); | |
2248 | ||
2249 | leaf = path->nodes[0]; | |
2250 | slot = path->slots[0]; | |
2251 | ||
2252 | if (slot >= btrfs_header_nritems(leaf)) { | |
2253 | ret = btrfs_next_leaf(root, path); | |
2254 | if (ret < 0) { | |
2255 | goto out; | |
2256 | } else if (ret > 0) { | |
2257 | ret = 0; | |
2258 | goto out; | |
2259 | } | |
2260 | continue; | |
2261 | } | |
2262 | ||
2263 | path->slots[0]++; | |
2264 | ||
2265 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2266 | ||
2267 | if (key.objectid > inum) | |
2268 | goto out; | |
2269 | ||
2270 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2271 | continue; | |
2272 | ||
2273 | extent = btrfs_item_ptr(leaf, slot, | |
2274 | struct btrfs_file_extent_item); | |
2275 | ||
2276 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2277 | continue; | |
2278 | ||
e68afa49 LB |
2279 | /* |
2280 | * 'offset' refers to the exact key.offset, | |
2281 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2282 | * (key.offset - extent_offset). | |
2283 | */ | |
2284 | if (key.offset != offset) | |
38c227d8 LB |
2285 | continue; |
2286 | ||
e68afa49 | 2287 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2288 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2289 | |
38c227d8 LB |
2290 | if (extent_offset >= old->extent_offset + old->offset + |
2291 | old->len || extent_offset + num_bytes <= | |
2292 | old->extent_offset + old->offset) | |
2293 | continue; | |
38c227d8 LB |
2294 | break; |
2295 | } | |
2296 | ||
2297 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2298 | if (!backref) { | |
2299 | ret = -ENOENT; | |
2300 | goto out; | |
2301 | } | |
2302 | ||
2303 | backref->root_id = root_id; | |
2304 | backref->inum = inum; | |
e68afa49 | 2305 | backref->file_pos = offset; |
38c227d8 LB |
2306 | backref->num_bytes = num_bytes; |
2307 | backref->extent_offset = extent_offset; | |
2308 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2309 | backref->old = old; | |
2310 | backref_insert(&new->root, backref); | |
2311 | old->count++; | |
2312 | out: | |
2313 | btrfs_release_path(path); | |
2314 | WARN_ON(ret); | |
2315 | return ret; | |
2316 | } | |
2317 | ||
2318 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2319 | struct new_sa_defrag_extent *new) | |
2320 | { | |
2321 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2322 | struct old_sa_defrag_extent *old, *tmp; | |
2323 | int ret; | |
2324 | ||
2325 | new->path = path; | |
2326 | ||
2327 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2328 | ret = iterate_inodes_from_logical(old->bytenr + |
2329 | old->extent_offset, fs_info, | |
38c227d8 LB |
2330 | path, record_one_backref, |
2331 | old); | |
4724b106 JB |
2332 | if (ret < 0 && ret != -ENOENT) |
2333 | return false; | |
38c227d8 LB |
2334 | |
2335 | /* no backref to be processed for this extent */ | |
2336 | if (!old->count) { | |
2337 | list_del(&old->list); | |
2338 | kfree(old); | |
2339 | } | |
2340 | } | |
2341 | ||
2342 | if (list_empty(&new->head)) | |
2343 | return false; | |
2344 | ||
2345 | return true; | |
2346 | } | |
2347 | ||
2348 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2349 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2350 | struct new_sa_defrag_extent *new) |
38c227d8 | 2351 | { |
116e0024 | 2352 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2353 | return 0; |
2354 | ||
2355 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2356 | return 0; | |
2357 | ||
116e0024 LB |
2358 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2359 | return 0; | |
2360 | ||
2361 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2362 | btrfs_file_extent_other_encoding(leaf, fi)) |
2363 | return 0; | |
2364 | ||
2365 | return 1; | |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * Note the backref might has changed, and in this case we just return 0. | |
2370 | */ | |
2371 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2372 | struct sa_defrag_extent_backref *prev, | |
2373 | struct sa_defrag_extent_backref *backref) | |
2374 | { | |
2375 | struct btrfs_file_extent_item *extent; | |
2376 | struct btrfs_file_extent_item *item; | |
2377 | struct btrfs_ordered_extent *ordered; | |
2378 | struct btrfs_trans_handle *trans; | |
2379 | struct btrfs_fs_info *fs_info; | |
2380 | struct btrfs_root *root; | |
2381 | struct btrfs_key key; | |
2382 | struct extent_buffer *leaf; | |
2383 | struct old_sa_defrag_extent *old = backref->old; | |
2384 | struct new_sa_defrag_extent *new = old->new; | |
2385 | struct inode *src_inode = new->inode; | |
2386 | struct inode *inode; | |
2387 | struct extent_state *cached = NULL; | |
2388 | int ret = 0; | |
2389 | u64 start; | |
2390 | u64 len; | |
2391 | u64 lock_start; | |
2392 | u64 lock_end; | |
2393 | bool merge = false; | |
2394 | int index; | |
2395 | ||
2396 | if (prev && prev->root_id == backref->root_id && | |
2397 | prev->inum == backref->inum && | |
2398 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2399 | merge = true; | |
2400 | ||
2401 | /* step 1: get root */ | |
2402 | key.objectid = backref->root_id; | |
2403 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2404 | key.offset = (u64)-1; | |
2405 | ||
2406 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2407 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2408 | ||
2409 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2410 | if (IS_ERR(root)) { | |
2411 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2412 | if (PTR_ERR(root) == -ENOENT) | |
2413 | return 0; | |
2414 | return PTR_ERR(root); | |
2415 | } | |
38c227d8 | 2416 | |
bcbba5e6 WS |
2417 | if (btrfs_root_readonly(root)) { |
2418 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2419 | return 0; | |
2420 | } | |
2421 | ||
38c227d8 LB |
2422 | /* step 2: get inode */ |
2423 | key.objectid = backref->inum; | |
2424 | key.type = BTRFS_INODE_ITEM_KEY; | |
2425 | key.offset = 0; | |
2426 | ||
2427 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2428 | if (IS_ERR(inode)) { | |
2429 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2430 | return 0; | |
2431 | } | |
2432 | ||
2433 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2434 | ||
2435 | /* step 3: relink backref */ | |
2436 | lock_start = backref->file_pos; | |
2437 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2438 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2439 | 0, &cached); | |
2440 | ||
2441 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2442 | if (ordered) { | |
2443 | btrfs_put_ordered_extent(ordered); | |
2444 | goto out_unlock; | |
2445 | } | |
2446 | ||
2447 | trans = btrfs_join_transaction(root); | |
2448 | if (IS_ERR(trans)) { | |
2449 | ret = PTR_ERR(trans); | |
2450 | goto out_unlock; | |
2451 | } | |
2452 | ||
2453 | key.objectid = backref->inum; | |
2454 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2455 | key.offset = backref->file_pos; | |
2456 | ||
2457 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2458 | if (ret < 0) { | |
2459 | goto out_free_path; | |
2460 | } else if (ret > 0) { | |
2461 | ret = 0; | |
2462 | goto out_free_path; | |
2463 | } | |
2464 | ||
2465 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2466 | struct btrfs_file_extent_item); | |
2467 | ||
2468 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2469 | backref->generation) | |
2470 | goto out_free_path; | |
2471 | ||
2472 | btrfs_release_path(path); | |
2473 | ||
2474 | start = backref->file_pos; | |
2475 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2476 | start += old->extent_offset + old->offset - | |
2477 | backref->extent_offset; | |
2478 | ||
2479 | len = min(backref->extent_offset + backref->num_bytes, | |
2480 | old->extent_offset + old->offset + old->len); | |
2481 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2482 | ||
2483 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2484 | start + len, 1); | |
2485 | if (ret) | |
2486 | goto out_free_path; | |
2487 | again: | |
2488 | key.objectid = btrfs_ino(inode); | |
2489 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2490 | key.offset = start; | |
2491 | ||
a09a0a70 | 2492 | path->leave_spinning = 1; |
38c227d8 LB |
2493 | if (merge) { |
2494 | struct btrfs_file_extent_item *fi; | |
2495 | u64 extent_len; | |
2496 | struct btrfs_key found_key; | |
2497 | ||
3c9665df | 2498 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2499 | if (ret < 0) |
2500 | goto out_free_path; | |
2501 | ||
2502 | path->slots[0]--; | |
2503 | leaf = path->nodes[0]; | |
2504 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2505 | ||
2506 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2507 | struct btrfs_file_extent_item); | |
2508 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2509 | ||
116e0024 LB |
2510 | if (extent_len + found_key.offset == start && |
2511 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2512 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2513 | extent_len + len); | |
2514 | btrfs_mark_buffer_dirty(leaf); | |
2515 | inode_add_bytes(inode, len); | |
2516 | ||
2517 | ret = 1; | |
2518 | goto out_free_path; | |
2519 | } else { | |
2520 | merge = false; | |
2521 | btrfs_release_path(path); | |
2522 | goto again; | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2527 | sizeof(*extent)); | |
2528 | if (ret) { | |
2529 | btrfs_abort_transaction(trans, root, ret); | |
2530 | goto out_free_path; | |
2531 | } | |
2532 | ||
2533 | leaf = path->nodes[0]; | |
2534 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2535 | struct btrfs_file_extent_item); | |
2536 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2537 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2538 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2539 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2540 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2541 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2542 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2543 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2544 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2545 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2546 | ||
2547 | btrfs_mark_buffer_dirty(leaf); | |
2548 | inode_add_bytes(inode, len); | |
a09a0a70 | 2549 | btrfs_release_path(path); |
38c227d8 LB |
2550 | |
2551 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2552 | new->disk_len, 0, | |
2553 | backref->root_id, backref->inum, | |
2554 | new->file_pos, 0); /* start - extent_offset */ | |
2555 | if (ret) { | |
2556 | btrfs_abort_transaction(trans, root, ret); | |
2557 | goto out_free_path; | |
2558 | } | |
2559 | ||
2560 | ret = 1; | |
2561 | out_free_path: | |
2562 | btrfs_release_path(path); | |
a09a0a70 | 2563 | path->leave_spinning = 0; |
38c227d8 LB |
2564 | btrfs_end_transaction(trans, root); |
2565 | out_unlock: | |
2566 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2567 | &cached, GFP_NOFS); | |
2568 | iput(inode); | |
2569 | return ret; | |
2570 | } | |
2571 | ||
6f519564 LB |
2572 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2573 | { | |
2574 | struct old_sa_defrag_extent *old, *tmp; | |
2575 | ||
2576 | if (!new) | |
2577 | return; | |
2578 | ||
2579 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2580 | list_del(&old->list); | |
2581 | kfree(old); | |
2582 | } | |
2583 | kfree(new); | |
2584 | } | |
2585 | ||
38c227d8 LB |
2586 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2587 | { | |
2588 | struct btrfs_path *path; | |
38c227d8 LB |
2589 | struct sa_defrag_extent_backref *backref; |
2590 | struct sa_defrag_extent_backref *prev = NULL; | |
2591 | struct inode *inode; | |
2592 | struct btrfs_root *root; | |
2593 | struct rb_node *node; | |
2594 | int ret; | |
2595 | ||
2596 | inode = new->inode; | |
2597 | root = BTRFS_I(inode)->root; | |
2598 | ||
2599 | path = btrfs_alloc_path(); | |
2600 | if (!path) | |
2601 | return; | |
2602 | ||
2603 | if (!record_extent_backrefs(path, new)) { | |
2604 | btrfs_free_path(path); | |
2605 | goto out; | |
2606 | } | |
2607 | btrfs_release_path(path); | |
2608 | ||
2609 | while (1) { | |
2610 | node = rb_first(&new->root); | |
2611 | if (!node) | |
2612 | break; | |
2613 | rb_erase(node, &new->root); | |
2614 | ||
2615 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2616 | ||
2617 | ret = relink_extent_backref(path, prev, backref); | |
2618 | WARN_ON(ret < 0); | |
2619 | ||
2620 | kfree(prev); | |
2621 | ||
2622 | if (ret == 1) | |
2623 | prev = backref; | |
2624 | else | |
2625 | prev = NULL; | |
2626 | cond_resched(); | |
2627 | } | |
2628 | kfree(prev); | |
2629 | ||
2630 | btrfs_free_path(path); | |
38c227d8 | 2631 | out: |
6f519564 LB |
2632 | free_sa_defrag_extent(new); |
2633 | ||
38c227d8 LB |
2634 | atomic_dec(&root->fs_info->defrag_running); |
2635 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2636 | } |
2637 | ||
2638 | static struct new_sa_defrag_extent * | |
2639 | record_old_file_extents(struct inode *inode, | |
2640 | struct btrfs_ordered_extent *ordered) | |
2641 | { | |
2642 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2643 | struct btrfs_path *path; | |
2644 | struct btrfs_key key; | |
6f519564 | 2645 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2646 | struct new_sa_defrag_extent *new; |
2647 | int ret; | |
2648 | ||
2649 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2650 | if (!new) | |
2651 | return NULL; | |
2652 | ||
2653 | new->inode = inode; | |
2654 | new->file_pos = ordered->file_offset; | |
2655 | new->len = ordered->len; | |
2656 | new->bytenr = ordered->start; | |
2657 | new->disk_len = ordered->disk_len; | |
2658 | new->compress_type = ordered->compress_type; | |
2659 | new->root = RB_ROOT; | |
2660 | INIT_LIST_HEAD(&new->head); | |
2661 | ||
2662 | path = btrfs_alloc_path(); | |
2663 | if (!path) | |
2664 | goto out_kfree; | |
2665 | ||
2666 | key.objectid = btrfs_ino(inode); | |
2667 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2668 | key.offset = new->file_pos; | |
2669 | ||
2670 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2671 | if (ret < 0) | |
2672 | goto out_free_path; | |
2673 | if (ret > 0 && path->slots[0] > 0) | |
2674 | path->slots[0]--; | |
2675 | ||
2676 | /* find out all the old extents for the file range */ | |
2677 | while (1) { | |
2678 | struct btrfs_file_extent_item *extent; | |
2679 | struct extent_buffer *l; | |
2680 | int slot; | |
2681 | u64 num_bytes; | |
2682 | u64 offset; | |
2683 | u64 end; | |
2684 | u64 disk_bytenr; | |
2685 | u64 extent_offset; | |
2686 | ||
2687 | l = path->nodes[0]; | |
2688 | slot = path->slots[0]; | |
2689 | ||
2690 | if (slot >= btrfs_header_nritems(l)) { | |
2691 | ret = btrfs_next_leaf(root, path); | |
2692 | if (ret < 0) | |
6f519564 | 2693 | goto out_free_path; |
38c227d8 LB |
2694 | else if (ret > 0) |
2695 | break; | |
2696 | continue; | |
2697 | } | |
2698 | ||
2699 | btrfs_item_key_to_cpu(l, &key, slot); | |
2700 | ||
2701 | if (key.objectid != btrfs_ino(inode)) | |
2702 | break; | |
2703 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2704 | break; | |
2705 | if (key.offset >= new->file_pos + new->len) | |
2706 | break; | |
2707 | ||
2708 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2709 | ||
2710 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2711 | if (key.offset + num_bytes < new->file_pos) | |
2712 | goto next; | |
2713 | ||
2714 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2715 | if (!disk_bytenr) | |
2716 | goto next; | |
2717 | ||
2718 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2719 | ||
2720 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2721 | if (!old) | |
6f519564 | 2722 | goto out_free_path; |
38c227d8 LB |
2723 | |
2724 | offset = max(new->file_pos, key.offset); | |
2725 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2726 | ||
2727 | old->bytenr = disk_bytenr; | |
2728 | old->extent_offset = extent_offset; | |
2729 | old->offset = offset - key.offset; | |
2730 | old->len = end - offset; | |
2731 | old->new = new; | |
2732 | old->count = 0; | |
2733 | list_add_tail(&old->list, &new->head); | |
2734 | next: | |
2735 | path->slots[0]++; | |
2736 | cond_resched(); | |
2737 | } | |
2738 | ||
2739 | btrfs_free_path(path); | |
2740 | atomic_inc(&root->fs_info->defrag_running); | |
2741 | ||
2742 | return new; | |
2743 | ||
38c227d8 LB |
2744 | out_free_path: |
2745 | btrfs_free_path(path); | |
2746 | out_kfree: | |
6f519564 | 2747 | free_sa_defrag_extent(new); |
38c227d8 LB |
2748 | return NULL; |
2749 | } | |
2750 | ||
e570fd27 MX |
2751 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2752 | u64 start, u64 len) | |
2753 | { | |
2754 | struct btrfs_block_group_cache *cache; | |
2755 | ||
2756 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2757 | ASSERT(cache); | |
2758 | ||
2759 | spin_lock(&cache->lock); | |
2760 | cache->delalloc_bytes -= len; | |
2761 | spin_unlock(&cache->lock); | |
2762 | ||
2763 | btrfs_put_block_group(cache); | |
2764 | } | |
2765 | ||
d352ac68 CM |
2766 | /* as ordered data IO finishes, this gets called so we can finish |
2767 | * an ordered extent if the range of bytes in the file it covers are | |
2768 | * fully written. | |
2769 | */ | |
5fd02043 | 2770 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2771 | { |
5fd02043 | 2772 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2773 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2774 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2775 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2776 | struct extent_state *cached_state = NULL; |
38c227d8 | 2777 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2778 | int compress_type = 0; |
77cef2ec JB |
2779 | int ret = 0; |
2780 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2781 | bool nolock; |
77cef2ec | 2782 | bool truncated = false; |
e6dcd2dc | 2783 | |
83eea1f1 | 2784 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2785 | |
5fd02043 JB |
2786 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2787 | ret = -EIO; | |
2788 | goto out; | |
2789 | } | |
2790 | ||
f612496b MX |
2791 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2792 | ordered_extent->file_offset + | |
2793 | ordered_extent->len - 1); | |
2794 | ||
77cef2ec JB |
2795 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2796 | truncated = true; | |
2797 | logical_len = ordered_extent->truncated_len; | |
2798 | /* Truncated the entire extent, don't bother adding */ | |
2799 | if (!logical_len) | |
2800 | goto out; | |
2801 | } | |
2802 | ||
c2167754 | 2803 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2804 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2805 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2806 | if (nolock) | |
2807 | trans = btrfs_join_transaction_nolock(root); | |
2808 | else | |
2809 | trans = btrfs_join_transaction(root); | |
2810 | if (IS_ERR(trans)) { | |
2811 | ret = PTR_ERR(trans); | |
2812 | trans = NULL; | |
2813 | goto out; | |
c2167754 | 2814 | } |
6c760c07 JB |
2815 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2816 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2817 | if (ret) /* -ENOMEM or corruption */ | |
2818 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2819 | goto out; |
2820 | } | |
e6dcd2dc | 2821 | |
2ac55d41 JB |
2822 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2823 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2824 | 0, &cached_state); |
e6dcd2dc | 2825 | |
38c227d8 LB |
2826 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2827 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2828 | EXTENT_DEFRAG, 1, cached_state); | |
2829 | if (ret) { | |
2830 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2831 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2832 | /* the inode is shared */ |
2833 | new = record_old_file_extents(inode, ordered_extent); | |
2834 | ||
2835 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2836 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2837 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2838 | } | |
2839 | ||
0cb59c99 | 2840 | if (nolock) |
7a7eaa40 | 2841 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2842 | else |
7a7eaa40 | 2843 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2844 | if (IS_ERR(trans)) { |
2845 | ret = PTR_ERR(trans); | |
2846 | trans = NULL; | |
2847 | goto out_unlock; | |
2848 | } | |
a79b7d4b | 2849 | |
0ca1f7ce | 2850 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2851 | |
c8b97818 | 2852 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2853 | compress_type = ordered_extent->compress_type; |
d899e052 | 2854 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2855 | BUG_ON(compress_type); |
920bbbfb | 2856 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2857 | ordered_extent->file_offset, |
2858 | ordered_extent->file_offset + | |
77cef2ec | 2859 | logical_len); |
d899e052 | 2860 | } else { |
0af3d00b | 2861 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2862 | ret = insert_reserved_file_extent(trans, inode, |
2863 | ordered_extent->file_offset, | |
2864 | ordered_extent->start, | |
2865 | ordered_extent->disk_len, | |
77cef2ec | 2866 | logical_len, logical_len, |
261507a0 | 2867 | compress_type, 0, 0, |
d899e052 | 2868 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2869 | if (!ret) |
2870 | btrfs_release_delalloc_bytes(root, | |
2871 | ordered_extent->start, | |
2872 | ordered_extent->disk_len); | |
d899e052 | 2873 | } |
5dc562c5 JB |
2874 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2875 | ordered_extent->file_offset, ordered_extent->len, | |
2876 | trans->transid); | |
79787eaa JM |
2877 | if (ret < 0) { |
2878 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2879 | goto out_unlock; |
79787eaa | 2880 | } |
2ac55d41 | 2881 | |
e6dcd2dc CM |
2882 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2883 | &ordered_extent->list); | |
2884 | ||
6c760c07 JB |
2885 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2886 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2887 | if (ret) { /* -ENOMEM or corruption */ | |
2888 | btrfs_abort_transaction(trans, root, ret); | |
2889 | goto out_unlock; | |
1ef30be1 JB |
2890 | } |
2891 | ret = 0; | |
5fd02043 JB |
2892 | out_unlock: |
2893 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2894 | ordered_extent->file_offset + | |
2895 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2896 | out: |
5b0e95bf | 2897 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2898 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2899 | if (trans) |
2900 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2901 | |
77cef2ec JB |
2902 | if (ret || truncated) { |
2903 | u64 start, end; | |
2904 | ||
2905 | if (truncated) | |
2906 | start = ordered_extent->file_offset + logical_len; | |
2907 | else | |
2908 | start = ordered_extent->file_offset; | |
2909 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2910 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2911 | ||
2912 | /* Drop the cache for the part of the extent we didn't write. */ | |
2913 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2914 | |
0bec9ef5 JB |
2915 | /* |
2916 | * If the ordered extent had an IOERR or something else went | |
2917 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2918 | * back to the allocator. We only free the extent in the |
2919 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2920 | */ |
77cef2ec JB |
2921 | if ((ret || !logical_len) && |
2922 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2923 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2924 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 2925 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2926 | } |
2927 | ||
2928 | ||
5fd02043 | 2929 | /* |
8bad3c02 LB |
2930 | * This needs to be done to make sure anybody waiting knows we are done |
2931 | * updating everything for this ordered extent. | |
5fd02043 JB |
2932 | */ |
2933 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2934 | ||
38c227d8 | 2935 | /* for snapshot-aware defrag */ |
6f519564 LB |
2936 | if (new) { |
2937 | if (ret) { | |
2938 | free_sa_defrag_extent(new); | |
2939 | atomic_dec(&root->fs_info->defrag_running); | |
2940 | } else { | |
2941 | relink_file_extents(new); | |
2942 | } | |
2943 | } | |
38c227d8 | 2944 | |
e6dcd2dc CM |
2945 | /* once for us */ |
2946 | btrfs_put_ordered_extent(ordered_extent); | |
2947 | /* once for the tree */ | |
2948 | btrfs_put_ordered_extent(ordered_extent); | |
2949 | ||
5fd02043 JB |
2950 | return ret; |
2951 | } | |
2952 | ||
2953 | static void finish_ordered_fn(struct btrfs_work *work) | |
2954 | { | |
2955 | struct btrfs_ordered_extent *ordered_extent; | |
2956 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2957 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2958 | } |
2959 | ||
b2950863 | 2960 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2961 | struct extent_state *state, int uptodate) |
2962 | { | |
5fd02043 JB |
2963 | struct inode *inode = page->mapping->host; |
2964 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2965 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
2966 | struct btrfs_workqueue *wq; |
2967 | btrfs_work_func_t func; | |
5fd02043 | 2968 | |
1abe9b8a | 2969 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2970 | ||
8b62b72b | 2971 | ClearPagePrivate2(page); |
5fd02043 JB |
2972 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2973 | end - start + 1, uptodate)) | |
2974 | return 0; | |
2975 | ||
9e0af237 LB |
2976 | if (btrfs_is_free_space_inode(inode)) { |
2977 | wq = root->fs_info->endio_freespace_worker; | |
2978 | func = btrfs_freespace_write_helper; | |
2979 | } else { | |
2980 | wq = root->fs_info->endio_write_workers; | |
2981 | func = btrfs_endio_write_helper; | |
2982 | } | |
5fd02043 | 2983 | |
9e0af237 LB |
2984 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
2985 | NULL); | |
2986 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
2987 | |
2988 | return 0; | |
211f90e6 CM |
2989 | } |
2990 | ||
dc380aea MX |
2991 | static int __readpage_endio_check(struct inode *inode, |
2992 | struct btrfs_io_bio *io_bio, | |
2993 | int icsum, struct page *page, | |
2994 | int pgoff, u64 start, size_t len) | |
2995 | { | |
2996 | char *kaddr; | |
2997 | u32 csum_expected; | |
2998 | u32 csum = ~(u32)0; | |
2999 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3000 | DEFAULT_RATELIMIT_BURST); | |
3001 | ||
3002 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3003 | ||
3004 | kaddr = kmap_atomic(page); | |
3005 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
3006 | btrfs_csum_final(csum, (char *)&csum); | |
3007 | if (csum != csum_expected) | |
3008 | goto zeroit; | |
3009 | ||
3010 | kunmap_atomic(kaddr); | |
3011 | return 0; | |
3012 | zeroit: | |
3013 | if (__ratelimit(&_rs)) | |
f0954c66 | 3014 | btrfs_warn(BTRFS_I(inode)->root->fs_info, |
dc380aea MX |
3015 | "csum failed ino %llu off %llu csum %u expected csum %u", |
3016 | btrfs_ino(inode), start, csum, csum_expected); | |
3017 | memset(kaddr + pgoff, 1, len); | |
3018 | flush_dcache_page(page); | |
3019 | kunmap_atomic(kaddr); | |
3020 | if (csum_expected == 0) | |
3021 | return 0; | |
3022 | return -EIO; | |
3023 | } | |
3024 | ||
d352ac68 CM |
3025 | /* |
3026 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3027 | * if there's a match, we allow the bio to finish. If not, the code in |
3028 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3029 | */ |
facc8a22 MX |
3030 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3031 | u64 phy_offset, struct page *page, | |
3032 | u64 start, u64 end, int mirror) | |
07157aac | 3033 | { |
4eee4fa4 | 3034 | size_t offset = start - page_offset(page); |
07157aac | 3035 | struct inode *inode = page->mapping->host; |
d1310b2e | 3036 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3037 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3038 | |
d20f7043 CM |
3039 | if (PageChecked(page)) { |
3040 | ClearPageChecked(page); | |
dc380aea | 3041 | return 0; |
d20f7043 | 3042 | } |
6cbff00f CH |
3043 | |
3044 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3045 | return 0; |
17d217fe YZ |
3046 | |
3047 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3048 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
3049 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
3050 | GFP_NOFS); | |
b6cda9bc | 3051 | return 0; |
17d217fe | 3052 | } |
d20f7043 | 3053 | |
facc8a22 | 3054 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3055 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3056 | start, (size_t)(end - start + 1)); | |
07157aac | 3057 | } |
b888db2b | 3058 | |
24bbcf04 YZ |
3059 | struct delayed_iput { |
3060 | struct list_head list; | |
3061 | struct inode *inode; | |
3062 | }; | |
3063 | ||
79787eaa JM |
3064 | /* JDM: If this is fs-wide, why can't we add a pointer to |
3065 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
3066 | void btrfs_add_delayed_iput(struct inode *inode) |
3067 | { | |
3068 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
3069 | struct delayed_iput *delayed; | |
3070 | ||
3071 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3072 | return; | |
3073 | ||
3074 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
3075 | delayed->inode = inode; | |
3076 | ||
3077 | spin_lock(&fs_info->delayed_iput_lock); | |
3078 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
3079 | spin_unlock(&fs_info->delayed_iput_lock); | |
3080 | } | |
3081 | ||
3082 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3083 | { | |
3084 | LIST_HEAD(list); | |
3085 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3086 | struct delayed_iput *delayed; | |
3087 | int empty; | |
3088 | ||
3089 | spin_lock(&fs_info->delayed_iput_lock); | |
3090 | empty = list_empty(&fs_info->delayed_iputs); | |
3091 | spin_unlock(&fs_info->delayed_iput_lock); | |
3092 | if (empty) | |
3093 | return; | |
3094 | ||
24bbcf04 YZ |
3095 | spin_lock(&fs_info->delayed_iput_lock); |
3096 | list_splice_init(&fs_info->delayed_iputs, &list); | |
3097 | spin_unlock(&fs_info->delayed_iput_lock); | |
3098 | ||
3099 | while (!list_empty(&list)) { | |
3100 | delayed = list_entry(list.next, struct delayed_iput, list); | |
3101 | list_del(&delayed->list); | |
3102 | iput(delayed->inode); | |
3103 | kfree(delayed); | |
3104 | } | |
24bbcf04 YZ |
3105 | } |
3106 | ||
d68fc57b | 3107 | /* |
42b2aa86 | 3108 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3109 | * files in the subvolume, it removes orphan item and frees block_rsv |
3110 | * structure. | |
3111 | */ | |
3112 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3113 | struct btrfs_root *root) | |
3114 | { | |
90290e19 | 3115 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3116 | int ret; |
3117 | ||
8a35d95f | 3118 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3119 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3120 | return; | |
3121 | ||
90290e19 | 3122 | spin_lock(&root->orphan_lock); |
8a35d95f | 3123 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3124 | spin_unlock(&root->orphan_lock); |
3125 | return; | |
3126 | } | |
3127 | ||
3128 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3129 | spin_unlock(&root->orphan_lock); | |
3130 | return; | |
3131 | } | |
3132 | ||
3133 | block_rsv = root->orphan_block_rsv; | |
3134 | root->orphan_block_rsv = NULL; | |
3135 | spin_unlock(&root->orphan_lock); | |
3136 | ||
27cdeb70 | 3137 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3138 | btrfs_root_refs(&root->root_item) > 0) { |
3139 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3140 | root->root_key.objectid); | |
4ef31a45 JB |
3141 | if (ret) |
3142 | btrfs_abort_transaction(trans, root, ret); | |
3143 | else | |
27cdeb70 MX |
3144 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3145 | &root->state); | |
d68fc57b YZ |
3146 | } |
3147 | ||
90290e19 JB |
3148 | if (block_rsv) { |
3149 | WARN_ON(block_rsv->size > 0); | |
3150 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3151 | } |
3152 | } | |
3153 | ||
7b128766 JB |
3154 | /* |
3155 | * This creates an orphan entry for the given inode in case something goes | |
3156 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3157 | * |
3158 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3159 | * this function. | |
7b128766 JB |
3160 | */ |
3161 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3162 | { | |
3163 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3164 | struct btrfs_block_rsv *block_rsv = NULL; |
3165 | int reserve = 0; | |
3166 | int insert = 0; | |
3167 | int ret; | |
7b128766 | 3168 | |
d68fc57b | 3169 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3170 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3171 | if (!block_rsv) |
3172 | return -ENOMEM; | |
d68fc57b | 3173 | } |
7b128766 | 3174 | |
d68fc57b YZ |
3175 | spin_lock(&root->orphan_lock); |
3176 | if (!root->orphan_block_rsv) { | |
3177 | root->orphan_block_rsv = block_rsv; | |
3178 | } else if (block_rsv) { | |
3179 | btrfs_free_block_rsv(root, block_rsv); | |
3180 | block_rsv = NULL; | |
7b128766 | 3181 | } |
7b128766 | 3182 | |
8a35d95f JB |
3183 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3184 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3185 | #if 0 |
3186 | /* | |
3187 | * For proper ENOSPC handling, we should do orphan | |
3188 | * cleanup when mounting. But this introduces backward | |
3189 | * compatibility issue. | |
3190 | */ | |
3191 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3192 | insert = 2; | |
3193 | else | |
3194 | insert = 1; | |
3195 | #endif | |
3196 | insert = 1; | |
321f0e70 | 3197 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3198 | } |
3199 | ||
72ac3c0d JB |
3200 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3201 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3202 | reserve = 1; |
d68fc57b | 3203 | spin_unlock(&root->orphan_lock); |
7b128766 | 3204 | |
d68fc57b YZ |
3205 | /* grab metadata reservation from transaction handle */ |
3206 | if (reserve) { | |
3207 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3208 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3209 | } |
7b128766 | 3210 | |
d68fc57b YZ |
3211 | /* insert an orphan item to track this unlinked/truncated file */ |
3212 | if (insert >= 1) { | |
33345d01 | 3213 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3214 | if (ret) { |
703c88e0 | 3215 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3216 | if (reserve) { |
3217 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3218 | &BTRFS_I(inode)->runtime_flags); | |
3219 | btrfs_orphan_release_metadata(inode); | |
3220 | } | |
3221 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3222 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3223 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3224 | btrfs_abort_transaction(trans, root, ret); |
3225 | return ret; | |
3226 | } | |
79787eaa JM |
3227 | } |
3228 | ret = 0; | |
d68fc57b YZ |
3229 | } |
3230 | ||
3231 | /* insert an orphan item to track subvolume contains orphan files */ | |
3232 | if (insert >= 2) { | |
3233 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3234 | root->root_key.objectid); | |
79787eaa JM |
3235 | if (ret && ret != -EEXIST) { |
3236 | btrfs_abort_transaction(trans, root, ret); | |
3237 | return ret; | |
3238 | } | |
d68fc57b YZ |
3239 | } |
3240 | return 0; | |
7b128766 JB |
3241 | } |
3242 | ||
3243 | /* | |
3244 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3245 | * item for this particular inode. | |
3246 | */ | |
48a3b636 ES |
3247 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3248 | struct inode *inode) | |
7b128766 JB |
3249 | { |
3250 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3251 | int delete_item = 0; |
3252 | int release_rsv = 0; | |
7b128766 JB |
3253 | int ret = 0; |
3254 | ||
d68fc57b | 3255 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3256 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3257 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3258 | delete_item = 1; |
7b128766 | 3259 | |
72ac3c0d JB |
3260 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3261 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3262 | release_rsv = 1; |
d68fc57b | 3263 | spin_unlock(&root->orphan_lock); |
7b128766 | 3264 | |
703c88e0 | 3265 | if (delete_item) { |
8a35d95f | 3266 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3267 | if (trans) |
3268 | ret = btrfs_del_orphan_item(trans, root, | |
3269 | btrfs_ino(inode)); | |
8a35d95f | 3270 | } |
7b128766 | 3271 | |
703c88e0 FDBM |
3272 | if (release_rsv) |
3273 | btrfs_orphan_release_metadata(inode); | |
3274 | ||
4ef31a45 | 3275 | return ret; |
7b128766 JB |
3276 | } |
3277 | ||
3278 | /* | |
3279 | * this cleans up any orphans that may be left on the list from the last use | |
3280 | * of this root. | |
3281 | */ | |
66b4ffd1 | 3282 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3283 | { |
3284 | struct btrfs_path *path; | |
3285 | struct extent_buffer *leaf; | |
7b128766 JB |
3286 | struct btrfs_key key, found_key; |
3287 | struct btrfs_trans_handle *trans; | |
3288 | struct inode *inode; | |
8f6d7f4f | 3289 | u64 last_objectid = 0; |
7b128766 JB |
3290 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3291 | ||
d68fc57b | 3292 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3293 | return 0; |
c71bf099 YZ |
3294 | |
3295 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3296 | if (!path) { |
3297 | ret = -ENOMEM; | |
3298 | goto out; | |
3299 | } | |
7b128766 JB |
3300 | path->reada = -1; |
3301 | ||
3302 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3303 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3304 | key.offset = (u64)-1; |
3305 | ||
7b128766 JB |
3306 | while (1) { |
3307 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3308 | if (ret < 0) |
3309 | goto out; | |
7b128766 JB |
3310 | |
3311 | /* | |
3312 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3313 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3314 | * find the key and see if we have stuff that matches |
3315 | */ | |
3316 | if (ret > 0) { | |
66b4ffd1 | 3317 | ret = 0; |
7b128766 JB |
3318 | if (path->slots[0] == 0) |
3319 | break; | |
3320 | path->slots[0]--; | |
3321 | } | |
3322 | ||
3323 | /* pull out the item */ | |
3324 | leaf = path->nodes[0]; | |
7b128766 JB |
3325 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3326 | ||
3327 | /* make sure the item matches what we want */ | |
3328 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3329 | break; | |
962a298f | 3330 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3331 | break; |
3332 | ||
3333 | /* release the path since we're done with it */ | |
b3b4aa74 | 3334 | btrfs_release_path(path); |
7b128766 JB |
3335 | |
3336 | /* | |
3337 | * this is where we are basically btrfs_lookup, without the | |
3338 | * crossing root thing. we store the inode number in the | |
3339 | * offset of the orphan item. | |
3340 | */ | |
8f6d7f4f JB |
3341 | |
3342 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3343 | btrfs_err(root->fs_info, |
3344 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3345 | ret = -EINVAL; |
3346 | goto out; | |
3347 | } | |
3348 | ||
3349 | last_objectid = found_key.offset; | |
3350 | ||
5d4f98a2 YZ |
3351 | found_key.objectid = found_key.offset; |
3352 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3353 | found_key.offset = 0; | |
73f73415 | 3354 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3355 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3356 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3357 | goto out; |
7b128766 | 3358 | |
f8e9e0b0 AJ |
3359 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3360 | struct btrfs_root *dead_root; | |
3361 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3362 | int is_dead_root = 0; | |
3363 | ||
3364 | /* | |
3365 | * this is an orphan in the tree root. Currently these | |
3366 | * could come from 2 sources: | |
3367 | * a) a snapshot deletion in progress | |
3368 | * b) a free space cache inode | |
3369 | * We need to distinguish those two, as the snapshot | |
3370 | * orphan must not get deleted. | |
3371 | * find_dead_roots already ran before us, so if this | |
3372 | * is a snapshot deletion, we should find the root | |
3373 | * in the dead_roots list | |
3374 | */ | |
3375 | spin_lock(&fs_info->trans_lock); | |
3376 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3377 | root_list) { | |
3378 | if (dead_root->root_key.objectid == | |
3379 | found_key.objectid) { | |
3380 | is_dead_root = 1; | |
3381 | break; | |
3382 | } | |
3383 | } | |
3384 | spin_unlock(&fs_info->trans_lock); | |
3385 | if (is_dead_root) { | |
3386 | /* prevent this orphan from being found again */ | |
3387 | key.offset = found_key.objectid - 1; | |
3388 | continue; | |
3389 | } | |
3390 | } | |
7b128766 | 3391 | /* |
a8c9e576 JB |
3392 | * Inode is already gone but the orphan item is still there, |
3393 | * kill the orphan item. | |
7b128766 | 3394 | */ |
a8c9e576 JB |
3395 | if (ret == -ESTALE) { |
3396 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3397 | if (IS_ERR(trans)) { |
3398 | ret = PTR_ERR(trans); | |
3399 | goto out; | |
3400 | } | |
c2cf52eb SK |
3401 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3402 | found_key.objectid); | |
a8c9e576 JB |
3403 | ret = btrfs_del_orphan_item(trans, root, |
3404 | found_key.objectid); | |
5b21f2ed | 3405 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3406 | if (ret) |
3407 | goto out; | |
7b128766 JB |
3408 | continue; |
3409 | } | |
3410 | ||
a8c9e576 JB |
3411 | /* |
3412 | * add this inode to the orphan list so btrfs_orphan_del does | |
3413 | * the proper thing when we hit it | |
3414 | */ | |
8a35d95f JB |
3415 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3416 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3417 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3418 | |
7b128766 JB |
3419 | /* if we have links, this was a truncate, lets do that */ |
3420 | if (inode->i_nlink) { | |
fae7f21c | 3421 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3422 | iput(inode); |
3423 | continue; | |
3424 | } | |
7b128766 | 3425 | nr_truncate++; |
f3fe820c JB |
3426 | |
3427 | /* 1 for the orphan item deletion. */ | |
3428 | trans = btrfs_start_transaction(root, 1); | |
3429 | if (IS_ERR(trans)) { | |
c69b26b0 | 3430 | iput(inode); |
f3fe820c JB |
3431 | ret = PTR_ERR(trans); |
3432 | goto out; | |
3433 | } | |
3434 | ret = btrfs_orphan_add(trans, inode); | |
3435 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3436 | if (ret) { |
3437 | iput(inode); | |
f3fe820c | 3438 | goto out; |
c69b26b0 | 3439 | } |
f3fe820c | 3440 | |
66b4ffd1 | 3441 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3442 | if (ret) |
3443 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3444 | } else { |
3445 | nr_unlink++; | |
3446 | } | |
3447 | ||
3448 | /* this will do delete_inode and everything for us */ | |
3449 | iput(inode); | |
66b4ffd1 JB |
3450 | if (ret) |
3451 | goto out; | |
7b128766 | 3452 | } |
3254c876 MX |
3453 | /* release the path since we're done with it */ |
3454 | btrfs_release_path(path); | |
3455 | ||
d68fc57b YZ |
3456 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3457 | ||
3458 | if (root->orphan_block_rsv) | |
3459 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3460 | (u64)-1); | |
3461 | ||
27cdeb70 MX |
3462 | if (root->orphan_block_rsv || |
3463 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3464 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3465 | if (!IS_ERR(trans)) |
3466 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3467 | } |
7b128766 JB |
3468 | |
3469 | if (nr_unlink) | |
4884b476 | 3470 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3471 | if (nr_truncate) |
4884b476 | 3472 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3473 | |
3474 | out: | |
3475 | if (ret) | |
68b663d1 | 3476 | btrfs_err(root->fs_info, |
c2cf52eb | 3477 | "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3478 | btrfs_free_path(path); |
3479 | return ret; | |
7b128766 JB |
3480 | } |
3481 | ||
46a53cca CM |
3482 | /* |
3483 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3484 | * don't find any xattrs, we know there can't be any acls. | |
3485 | * | |
3486 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3487 | */ | |
3488 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3489 | int slot, u64 objectid, |
3490 | int *first_xattr_slot) | |
46a53cca CM |
3491 | { |
3492 | u32 nritems = btrfs_header_nritems(leaf); | |
3493 | struct btrfs_key found_key; | |
f23b5a59 JB |
3494 | static u64 xattr_access = 0; |
3495 | static u64 xattr_default = 0; | |
46a53cca CM |
3496 | int scanned = 0; |
3497 | ||
f23b5a59 JB |
3498 | if (!xattr_access) { |
3499 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3500 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3501 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3502 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3503 | } | |
3504 | ||
46a53cca | 3505 | slot++; |
63541927 | 3506 | *first_xattr_slot = -1; |
46a53cca CM |
3507 | while (slot < nritems) { |
3508 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3509 | ||
3510 | /* we found a different objectid, there must not be acls */ | |
3511 | if (found_key.objectid != objectid) | |
3512 | return 0; | |
3513 | ||
3514 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3515 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3516 | if (*first_xattr_slot == -1) |
3517 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3518 | if (found_key.offset == xattr_access || |
3519 | found_key.offset == xattr_default) | |
3520 | return 1; | |
3521 | } | |
46a53cca CM |
3522 | |
3523 | /* | |
3524 | * we found a key greater than an xattr key, there can't | |
3525 | * be any acls later on | |
3526 | */ | |
3527 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3528 | return 0; | |
3529 | ||
3530 | slot++; | |
3531 | scanned++; | |
3532 | ||
3533 | /* | |
3534 | * it goes inode, inode backrefs, xattrs, extents, | |
3535 | * so if there are a ton of hard links to an inode there can | |
3536 | * be a lot of backrefs. Don't waste time searching too hard, | |
3537 | * this is just an optimization | |
3538 | */ | |
3539 | if (scanned >= 8) | |
3540 | break; | |
3541 | } | |
3542 | /* we hit the end of the leaf before we found an xattr or | |
3543 | * something larger than an xattr. We have to assume the inode | |
3544 | * has acls | |
3545 | */ | |
63541927 FDBM |
3546 | if (*first_xattr_slot == -1) |
3547 | *first_xattr_slot = slot; | |
46a53cca CM |
3548 | return 1; |
3549 | } | |
3550 | ||
d352ac68 CM |
3551 | /* |
3552 | * read an inode from the btree into the in-memory inode | |
3553 | */ | |
5d4f98a2 | 3554 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3555 | { |
3556 | struct btrfs_path *path; | |
5f39d397 | 3557 | struct extent_buffer *leaf; |
39279cc3 CM |
3558 | struct btrfs_inode_item *inode_item; |
3559 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3560 | struct btrfs_key location; | |
67de1176 | 3561 | unsigned long ptr; |
46a53cca | 3562 | int maybe_acls; |
618e21d5 | 3563 | u32 rdev; |
39279cc3 | 3564 | int ret; |
2f7e33d4 | 3565 | bool filled = false; |
63541927 | 3566 | int first_xattr_slot; |
2f7e33d4 MX |
3567 | |
3568 | ret = btrfs_fill_inode(inode, &rdev); | |
3569 | if (!ret) | |
3570 | filled = true; | |
39279cc3 CM |
3571 | |
3572 | path = btrfs_alloc_path(); | |
1748f843 MF |
3573 | if (!path) |
3574 | goto make_bad; | |
3575 | ||
39279cc3 | 3576 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3577 | |
39279cc3 | 3578 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3579 | if (ret) |
39279cc3 | 3580 | goto make_bad; |
39279cc3 | 3581 | |
5f39d397 | 3582 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3583 | |
3584 | if (filled) | |
67de1176 | 3585 | goto cache_index; |
2f7e33d4 | 3586 | |
5f39d397 CM |
3587 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3588 | struct btrfs_inode_item); | |
5f39d397 | 3589 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3590 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3591 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3592 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3593 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3594 | |
a937b979 DS |
3595 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3596 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3597 | |
a937b979 DS |
3598 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3599 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3600 | |
a937b979 DS |
3601 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3602 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3603 | |
9cc97d64 | 3604 | BTRFS_I(inode)->i_otime.tv_sec = |
3605 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3606 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3607 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
3608 | ||
a76a3cd4 | 3609 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3610 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3611 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3612 | ||
3613 | /* | |
3614 | * If we were modified in the current generation and evicted from memory | |
3615 | * and then re-read we need to do a full sync since we don't have any | |
3616 | * idea about which extents were modified before we were evicted from | |
3617 | * cache. | |
3618 | */ | |
3619 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3620 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3621 | &BTRFS_I(inode)->runtime_flags); | |
3622 | ||
0c4d2d95 | 3623 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3624 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3625 | inode->i_rdev = 0; |
5f39d397 CM |
3626 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3627 | ||
aec7477b | 3628 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3629 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
67de1176 MX |
3630 | |
3631 | cache_index: | |
3632 | path->slots[0]++; | |
3633 | if (inode->i_nlink != 1 || | |
3634 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3635 | goto cache_acl; | |
3636 | ||
3637 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3638 | if (location.objectid != btrfs_ino(inode)) | |
3639 | goto cache_acl; | |
3640 | ||
3641 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3642 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3643 | struct btrfs_inode_ref *ref; | |
3644 | ||
3645 | ref = (struct btrfs_inode_ref *)ptr; | |
3646 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3647 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3648 | struct btrfs_inode_extref *extref; | |
3649 | ||
3650 | extref = (struct btrfs_inode_extref *)ptr; | |
3651 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3652 | extref); | |
3653 | } | |
2f7e33d4 | 3654 | cache_acl: |
46a53cca CM |
3655 | /* |
3656 | * try to precache a NULL acl entry for files that don't have | |
3657 | * any xattrs or acls | |
3658 | */ | |
33345d01 | 3659 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3660 | btrfs_ino(inode), &first_xattr_slot); |
3661 | if (first_xattr_slot != -1) { | |
3662 | path->slots[0] = first_xattr_slot; | |
3663 | ret = btrfs_load_inode_props(inode, path); | |
3664 | if (ret) | |
3665 | btrfs_err(root->fs_info, | |
351fd353 | 3666 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3667 | btrfs_ino(inode), |
3668 | root->root_key.objectid, ret); | |
3669 | } | |
3670 | btrfs_free_path(path); | |
3671 | ||
72c04902 AV |
3672 | if (!maybe_acls) |
3673 | cache_no_acl(inode); | |
46a53cca | 3674 | |
39279cc3 | 3675 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3676 | case S_IFREG: |
3677 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3678 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3679 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3680 | inode->i_fop = &btrfs_file_operations; |
3681 | inode->i_op = &btrfs_file_inode_operations; | |
3682 | break; | |
3683 | case S_IFDIR: | |
3684 | inode->i_fop = &btrfs_dir_file_operations; | |
3685 | if (root == root->fs_info->tree_root) | |
3686 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3687 | else | |
3688 | inode->i_op = &btrfs_dir_inode_operations; | |
3689 | break; | |
3690 | case S_IFLNK: | |
3691 | inode->i_op = &btrfs_symlink_inode_operations; | |
3692 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3693 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3694 | break; |
618e21d5 | 3695 | default: |
0279b4cd | 3696 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3697 | init_special_inode(inode, inode->i_mode, rdev); |
3698 | break; | |
39279cc3 | 3699 | } |
6cbff00f CH |
3700 | |
3701 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3702 | return; |
3703 | ||
3704 | make_bad: | |
39279cc3 | 3705 | btrfs_free_path(path); |
39279cc3 CM |
3706 | make_bad_inode(inode); |
3707 | } | |
3708 | ||
d352ac68 CM |
3709 | /* |
3710 | * given a leaf and an inode, copy the inode fields into the leaf | |
3711 | */ | |
e02119d5 CM |
3712 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3713 | struct extent_buffer *leaf, | |
5f39d397 | 3714 | struct btrfs_inode_item *item, |
39279cc3 CM |
3715 | struct inode *inode) |
3716 | { | |
51fab693 LB |
3717 | struct btrfs_map_token token; |
3718 | ||
3719 | btrfs_init_map_token(&token); | |
5f39d397 | 3720 | |
51fab693 LB |
3721 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3722 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3723 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3724 | &token); | |
3725 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3726 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3727 | |
a937b979 | 3728 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3729 | inode->i_atime.tv_sec, &token); |
a937b979 | 3730 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3731 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3732 | |
a937b979 | 3733 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3734 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3735 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3736 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3737 | |
a937b979 | 3738 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3739 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3740 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3741 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3742 | |
9cc97d64 | 3743 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3744 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3745 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3746 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3747 | ||
51fab693 LB |
3748 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3749 | &token); | |
3750 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3751 | &token); | |
3752 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3753 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3754 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3755 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3756 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3757 | } |
3758 | ||
d352ac68 CM |
3759 | /* |
3760 | * copy everything in the in-memory inode into the btree. | |
3761 | */ | |
2115133f | 3762 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3763 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3764 | { |
3765 | struct btrfs_inode_item *inode_item; | |
3766 | struct btrfs_path *path; | |
5f39d397 | 3767 | struct extent_buffer *leaf; |
39279cc3 CM |
3768 | int ret; |
3769 | ||
3770 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3771 | if (!path) |
3772 | return -ENOMEM; | |
3773 | ||
b9473439 | 3774 | path->leave_spinning = 1; |
16cdcec7 MX |
3775 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3776 | 1); | |
39279cc3 CM |
3777 | if (ret) { |
3778 | if (ret > 0) | |
3779 | ret = -ENOENT; | |
3780 | goto failed; | |
3781 | } | |
3782 | ||
5f39d397 CM |
3783 | leaf = path->nodes[0]; |
3784 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3785 | struct btrfs_inode_item); |
39279cc3 | 3786 | |
e02119d5 | 3787 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3788 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3789 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3790 | ret = 0; |
3791 | failed: | |
39279cc3 CM |
3792 | btrfs_free_path(path); |
3793 | return ret; | |
3794 | } | |
3795 | ||
2115133f CM |
3796 | /* |
3797 | * copy everything in the in-memory inode into the btree. | |
3798 | */ | |
3799 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3800 | struct btrfs_root *root, struct inode *inode) | |
3801 | { | |
3802 | int ret; | |
3803 | ||
3804 | /* | |
3805 | * If the inode is a free space inode, we can deadlock during commit | |
3806 | * if we put it into the delayed code. | |
3807 | * | |
3808 | * The data relocation inode should also be directly updated | |
3809 | * without delay | |
3810 | */ | |
83eea1f1 | 3811 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3812 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3813 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3814 | btrfs_update_root_times(trans, root); |
3815 | ||
2115133f CM |
3816 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3817 | if (!ret) | |
3818 | btrfs_set_inode_last_trans(trans, inode); | |
3819 | return ret; | |
3820 | } | |
3821 | ||
3822 | return btrfs_update_inode_item(trans, root, inode); | |
3823 | } | |
3824 | ||
be6aef60 JB |
3825 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3826 | struct btrfs_root *root, | |
3827 | struct inode *inode) | |
2115133f CM |
3828 | { |
3829 | int ret; | |
3830 | ||
3831 | ret = btrfs_update_inode(trans, root, inode); | |
3832 | if (ret == -ENOSPC) | |
3833 | return btrfs_update_inode_item(trans, root, inode); | |
3834 | return ret; | |
3835 | } | |
3836 | ||
d352ac68 CM |
3837 | /* |
3838 | * unlink helper that gets used here in inode.c and in the tree logging | |
3839 | * recovery code. It remove a link in a directory with a given name, and | |
3840 | * also drops the back refs in the inode to the directory | |
3841 | */ | |
92986796 AV |
3842 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3843 | struct btrfs_root *root, | |
3844 | struct inode *dir, struct inode *inode, | |
3845 | const char *name, int name_len) | |
39279cc3 CM |
3846 | { |
3847 | struct btrfs_path *path; | |
39279cc3 | 3848 | int ret = 0; |
5f39d397 | 3849 | struct extent_buffer *leaf; |
39279cc3 | 3850 | struct btrfs_dir_item *di; |
5f39d397 | 3851 | struct btrfs_key key; |
aec7477b | 3852 | u64 index; |
33345d01 LZ |
3853 | u64 ino = btrfs_ino(inode); |
3854 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3855 | |
3856 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3857 | if (!path) { |
3858 | ret = -ENOMEM; | |
554233a6 | 3859 | goto out; |
54aa1f4d CM |
3860 | } |
3861 | ||
b9473439 | 3862 | path->leave_spinning = 1; |
33345d01 | 3863 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3864 | name, name_len, -1); |
3865 | if (IS_ERR(di)) { | |
3866 | ret = PTR_ERR(di); | |
3867 | goto err; | |
3868 | } | |
3869 | if (!di) { | |
3870 | ret = -ENOENT; | |
3871 | goto err; | |
3872 | } | |
5f39d397 CM |
3873 | leaf = path->nodes[0]; |
3874 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3875 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3876 | if (ret) |
3877 | goto err; | |
b3b4aa74 | 3878 | btrfs_release_path(path); |
39279cc3 | 3879 | |
67de1176 MX |
3880 | /* |
3881 | * If we don't have dir index, we have to get it by looking up | |
3882 | * the inode ref, since we get the inode ref, remove it directly, | |
3883 | * it is unnecessary to do delayed deletion. | |
3884 | * | |
3885 | * But if we have dir index, needn't search inode ref to get it. | |
3886 | * Since the inode ref is close to the inode item, it is better | |
3887 | * that we delay to delete it, and just do this deletion when | |
3888 | * we update the inode item. | |
3889 | */ | |
3890 | if (BTRFS_I(inode)->dir_index) { | |
3891 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3892 | if (!ret) { | |
3893 | index = BTRFS_I(inode)->dir_index; | |
3894 | goto skip_backref; | |
3895 | } | |
3896 | } | |
3897 | ||
33345d01 LZ |
3898 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3899 | dir_ino, &index); | |
aec7477b | 3900 | if (ret) { |
c2cf52eb SK |
3901 | btrfs_info(root->fs_info, |
3902 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3903 | name_len, name, ino, dir_ino); |
79787eaa | 3904 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3905 | goto err; |
3906 | } | |
67de1176 | 3907 | skip_backref: |
16cdcec7 | 3908 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3909 | if (ret) { |
3910 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3911 | goto err; |
79787eaa | 3912 | } |
39279cc3 | 3913 | |
e02119d5 | 3914 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3915 | inode, dir_ino); |
79787eaa JM |
3916 | if (ret != 0 && ret != -ENOENT) { |
3917 | btrfs_abort_transaction(trans, root, ret); | |
3918 | goto err; | |
3919 | } | |
e02119d5 CM |
3920 | |
3921 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3922 | dir, index); | |
6418c961 CM |
3923 | if (ret == -ENOENT) |
3924 | ret = 0; | |
d4e3991b ZB |
3925 | else if (ret) |
3926 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3927 | err: |
3928 | btrfs_free_path(path); | |
e02119d5 CM |
3929 | if (ret) |
3930 | goto out; | |
3931 | ||
3932 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3933 | inode_inc_iversion(inode); |
3934 | inode_inc_iversion(dir); | |
e02119d5 | 3935 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3936 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3937 | out: |
39279cc3 CM |
3938 | return ret; |
3939 | } | |
3940 | ||
92986796 AV |
3941 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3942 | struct btrfs_root *root, | |
3943 | struct inode *dir, struct inode *inode, | |
3944 | const char *name, int name_len) | |
3945 | { | |
3946 | int ret; | |
3947 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3948 | if (!ret) { | |
8b558c5f | 3949 | drop_nlink(inode); |
92986796 AV |
3950 | ret = btrfs_update_inode(trans, root, inode); |
3951 | } | |
3952 | return ret; | |
3953 | } | |
39279cc3 | 3954 | |
a22285a6 YZ |
3955 | /* |
3956 | * helper to start transaction for unlink and rmdir. | |
3957 | * | |
d52be818 JB |
3958 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3959 | * if we cannot make our reservations the normal way try and see if there is | |
3960 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3961 | * allow the unlink to occur. | |
a22285a6 | 3962 | */ |
d52be818 | 3963 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3964 | { |
39279cc3 | 3965 | struct btrfs_trans_handle *trans; |
a22285a6 | 3966 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3967 | int ret; |
3968 | ||
e70bea5f JB |
3969 | /* |
3970 | * 1 for the possible orphan item | |
3971 | * 1 for the dir item | |
3972 | * 1 for the dir index | |
3973 | * 1 for the inode ref | |
e70bea5f JB |
3974 | * 1 for the inode |
3975 | */ | |
6e137ed3 | 3976 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3977 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3978 | return trans; | |
4df27c4d | 3979 | |
d52be818 JB |
3980 | if (PTR_ERR(trans) == -ENOSPC) { |
3981 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3982 | |
d52be818 JB |
3983 | trans = btrfs_start_transaction(root, 0); |
3984 | if (IS_ERR(trans)) | |
3985 | return trans; | |
3986 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3987 | &root->fs_info->trans_block_rsv, | |
3988 | num_bytes, 5); | |
3989 | if (ret) { | |
3990 | btrfs_end_transaction(trans, root); | |
3991 | return ERR_PTR(ret); | |
a22285a6 | 3992 | } |
5a77d76c | 3993 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3994 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3995 | } |
d52be818 | 3996 | return trans; |
a22285a6 YZ |
3997 | } |
3998 | ||
3999 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4000 | { | |
4001 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4002 | struct btrfs_trans_handle *trans; | |
4003 | struct inode *inode = dentry->d_inode; | |
4004 | int ret; | |
a22285a6 | 4005 | |
d52be818 | 4006 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4007 | if (IS_ERR(trans)) |
4008 | return PTR_ERR(trans); | |
5f39d397 | 4009 | |
12fcfd22 CM |
4010 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
4011 | ||
e02119d5 CM |
4012 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
4013 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
4014 | if (ret) |
4015 | goto out; | |
7b128766 | 4016 | |
a22285a6 | 4017 | if (inode->i_nlink == 0) { |
7b128766 | 4018 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4019 | if (ret) |
4020 | goto out; | |
a22285a6 | 4021 | } |
7b128766 | 4022 | |
b532402e | 4023 | out: |
d52be818 | 4024 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4025 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
4026 | return ret; |
4027 | } | |
4028 | ||
4df27c4d YZ |
4029 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4030 | struct btrfs_root *root, | |
4031 | struct inode *dir, u64 objectid, | |
4032 | const char *name, int name_len) | |
4033 | { | |
4034 | struct btrfs_path *path; | |
4035 | struct extent_buffer *leaf; | |
4036 | struct btrfs_dir_item *di; | |
4037 | struct btrfs_key key; | |
4038 | u64 index; | |
4039 | int ret; | |
33345d01 | 4040 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
4041 | |
4042 | path = btrfs_alloc_path(); | |
4043 | if (!path) | |
4044 | return -ENOMEM; | |
4045 | ||
33345d01 | 4046 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4047 | name, name_len, -1); |
79787eaa JM |
4048 | if (IS_ERR_OR_NULL(di)) { |
4049 | if (!di) | |
4050 | ret = -ENOENT; | |
4051 | else | |
4052 | ret = PTR_ERR(di); | |
4053 | goto out; | |
4054 | } | |
4df27c4d YZ |
4055 | |
4056 | leaf = path->nodes[0]; | |
4057 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4058 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4059 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
4060 | if (ret) { |
4061 | btrfs_abort_transaction(trans, root, ret); | |
4062 | goto out; | |
4063 | } | |
b3b4aa74 | 4064 | btrfs_release_path(path); |
4df27c4d YZ |
4065 | |
4066 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4067 | objectid, root->root_key.objectid, | |
33345d01 | 4068 | dir_ino, &index, name, name_len); |
4df27c4d | 4069 | if (ret < 0) { |
79787eaa JM |
4070 | if (ret != -ENOENT) { |
4071 | btrfs_abort_transaction(trans, root, ret); | |
4072 | goto out; | |
4073 | } | |
33345d01 | 4074 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4075 | name, name_len); |
79787eaa JM |
4076 | if (IS_ERR_OR_NULL(di)) { |
4077 | if (!di) | |
4078 | ret = -ENOENT; | |
4079 | else | |
4080 | ret = PTR_ERR(di); | |
4081 | btrfs_abort_transaction(trans, root, ret); | |
4082 | goto out; | |
4083 | } | |
4df27c4d YZ |
4084 | |
4085 | leaf = path->nodes[0]; | |
4086 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4087 | btrfs_release_path(path); |
4df27c4d YZ |
4088 | index = key.offset; |
4089 | } | |
945d8962 | 4090 | btrfs_release_path(path); |
4df27c4d | 4091 | |
16cdcec7 | 4092 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4093 | if (ret) { |
4094 | btrfs_abort_transaction(trans, root, ret); | |
4095 | goto out; | |
4096 | } | |
4df27c4d YZ |
4097 | |
4098 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4099 | inode_inc_iversion(dir); |
4df27c4d | 4100 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 4101 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4102 | if (ret) |
4103 | btrfs_abort_transaction(trans, root, ret); | |
4104 | out: | |
71d7aed0 | 4105 | btrfs_free_path(path); |
79787eaa | 4106 | return ret; |
4df27c4d YZ |
4107 | } |
4108 | ||
39279cc3 CM |
4109 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4110 | { | |
4111 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 4112 | int err = 0; |
39279cc3 | 4113 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4114 | struct btrfs_trans_handle *trans; |
39279cc3 | 4115 | |
b3ae244e | 4116 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4117 | return -ENOTEMPTY; |
b3ae244e DS |
4118 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4119 | return -EPERM; | |
134d4512 | 4120 | |
d52be818 | 4121 | trans = __unlink_start_trans(dir); |
a22285a6 | 4122 | if (IS_ERR(trans)) |
5df6a9f6 | 4123 | return PTR_ERR(trans); |
5df6a9f6 | 4124 | |
33345d01 | 4125 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4126 | err = btrfs_unlink_subvol(trans, root, dir, |
4127 | BTRFS_I(inode)->location.objectid, | |
4128 | dentry->d_name.name, | |
4129 | dentry->d_name.len); | |
4130 | goto out; | |
4131 | } | |
4132 | ||
7b128766 JB |
4133 | err = btrfs_orphan_add(trans, inode); |
4134 | if (err) | |
4df27c4d | 4135 | goto out; |
7b128766 | 4136 | |
39279cc3 | 4137 | /* now the directory is empty */ |
e02119d5 CM |
4138 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
4139 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 4140 | if (!err) |
dbe674a9 | 4141 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4142 | out: |
d52be818 | 4143 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4144 | btrfs_btree_balance_dirty(root); |
3954401f | 4145 | |
39279cc3 CM |
4146 | return err; |
4147 | } | |
4148 | ||
39279cc3 CM |
4149 | /* |
4150 | * this can truncate away extent items, csum items and directory items. | |
4151 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4152 | * any higher than new_size |
39279cc3 CM |
4153 | * |
4154 | * csum items that cross the new i_size are truncated to the new size | |
4155 | * as well. | |
7b128766 JB |
4156 | * |
4157 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4158 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4159 | */ |
8082510e YZ |
4160 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4161 | struct btrfs_root *root, | |
4162 | struct inode *inode, | |
4163 | u64 new_size, u32 min_type) | |
39279cc3 | 4164 | { |
39279cc3 | 4165 | struct btrfs_path *path; |
5f39d397 | 4166 | struct extent_buffer *leaf; |
39279cc3 | 4167 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4168 | struct btrfs_key key; |
4169 | struct btrfs_key found_key; | |
39279cc3 | 4170 | u64 extent_start = 0; |
db94535d | 4171 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4172 | u64 extent_offset = 0; |
39279cc3 | 4173 | u64 item_end = 0; |
7f4f6e0a | 4174 | u64 last_size = (u64)-1; |
8082510e | 4175 | u32 found_type = (u8)-1; |
39279cc3 CM |
4176 | int found_extent; |
4177 | int del_item; | |
85e21bac CM |
4178 | int pending_del_nr = 0; |
4179 | int pending_del_slot = 0; | |
179e29e4 | 4180 | int extent_type = -1; |
8082510e YZ |
4181 | int ret; |
4182 | int err = 0; | |
33345d01 | 4183 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
4184 | |
4185 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4186 | |
0eb0e19c MF |
4187 | path = btrfs_alloc_path(); |
4188 | if (!path) | |
4189 | return -ENOMEM; | |
4190 | path->reada = -1; | |
4191 | ||
5dc562c5 JB |
4192 | /* |
4193 | * We want to drop from the next block forward in case this new size is | |
4194 | * not block aligned since we will be keeping the last block of the | |
4195 | * extent just the way it is. | |
4196 | */ | |
27cdeb70 MX |
4197 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4198 | root == root->fs_info->tree_root) | |
fda2832f QW |
4199 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4200 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4201 | |
16cdcec7 MX |
4202 | /* |
4203 | * This function is also used to drop the items in the log tree before | |
4204 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4205 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4206 | * items. | |
4207 | */ | |
4208 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4209 | btrfs_kill_delayed_inode_items(inode); | |
4210 | ||
33345d01 | 4211 | key.objectid = ino; |
39279cc3 | 4212 | key.offset = (u64)-1; |
5f39d397 CM |
4213 | key.type = (u8)-1; |
4214 | ||
85e21bac | 4215 | search_again: |
b9473439 | 4216 | path->leave_spinning = 1; |
85e21bac | 4217 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4218 | if (ret < 0) { |
4219 | err = ret; | |
4220 | goto out; | |
4221 | } | |
d397712b | 4222 | |
85e21bac | 4223 | if (ret > 0) { |
e02119d5 CM |
4224 | /* there are no items in the tree for us to truncate, we're |
4225 | * done | |
4226 | */ | |
8082510e YZ |
4227 | if (path->slots[0] == 0) |
4228 | goto out; | |
85e21bac CM |
4229 | path->slots[0]--; |
4230 | } | |
4231 | ||
d397712b | 4232 | while (1) { |
39279cc3 | 4233 | fi = NULL; |
5f39d397 CM |
4234 | leaf = path->nodes[0]; |
4235 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4236 | found_type = found_key.type; |
39279cc3 | 4237 | |
33345d01 | 4238 | if (found_key.objectid != ino) |
39279cc3 | 4239 | break; |
5f39d397 | 4240 | |
85e21bac | 4241 | if (found_type < min_type) |
39279cc3 CM |
4242 | break; |
4243 | ||
5f39d397 | 4244 | item_end = found_key.offset; |
39279cc3 | 4245 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4246 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4247 | struct btrfs_file_extent_item); |
179e29e4 CM |
4248 | extent_type = btrfs_file_extent_type(leaf, fi); |
4249 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4250 | item_end += |
db94535d | 4251 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4252 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4253 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4254 | path->slots[0], fi); |
39279cc3 | 4255 | } |
008630c1 | 4256 | item_end--; |
39279cc3 | 4257 | } |
8082510e YZ |
4258 | if (found_type > min_type) { |
4259 | del_item = 1; | |
4260 | } else { | |
4261 | if (item_end < new_size) | |
b888db2b | 4262 | break; |
8082510e YZ |
4263 | if (found_key.offset >= new_size) |
4264 | del_item = 1; | |
4265 | else | |
4266 | del_item = 0; | |
39279cc3 | 4267 | } |
39279cc3 | 4268 | found_extent = 0; |
39279cc3 | 4269 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4270 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4271 | goto delete; | |
4272 | ||
7f4f6e0a JB |
4273 | if (del_item) |
4274 | last_size = found_key.offset; | |
4275 | else | |
4276 | last_size = new_size; | |
4277 | ||
179e29e4 | 4278 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4279 | u64 num_dec; |
db94535d | 4280 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4281 | if (!del_item) { |
db94535d CM |
4282 | u64 orig_num_bytes = |
4283 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4284 | extent_num_bytes = ALIGN(new_size - |
4285 | found_key.offset, | |
4286 | root->sectorsize); | |
db94535d CM |
4287 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4288 | extent_num_bytes); | |
4289 | num_dec = (orig_num_bytes - | |
9069218d | 4290 | extent_num_bytes); |
27cdeb70 MX |
4291 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4292 | &root->state) && | |
4293 | extent_start != 0) | |
a76a3cd4 | 4294 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4295 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4296 | } else { |
db94535d CM |
4297 | extent_num_bytes = |
4298 | btrfs_file_extent_disk_num_bytes(leaf, | |
4299 | fi); | |
5d4f98a2 YZ |
4300 | extent_offset = found_key.offset - |
4301 | btrfs_file_extent_offset(leaf, fi); | |
4302 | ||
39279cc3 | 4303 | /* FIXME blocksize != 4096 */ |
9069218d | 4304 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4305 | if (extent_start != 0) { |
4306 | found_extent = 1; | |
27cdeb70 MX |
4307 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4308 | &root->state)) | |
a76a3cd4 | 4309 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4310 | } |
39279cc3 | 4311 | } |
9069218d | 4312 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4313 | /* |
4314 | * we can't truncate inline items that have had | |
4315 | * special encodings | |
4316 | */ | |
4317 | if (!del_item && | |
4318 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4319 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4320 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4321 | u32 size = new_size - found_key.offset; |
4322 | ||
27cdeb70 | 4323 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
a76a3cd4 YZ |
4324 | inode_sub_bytes(inode, item_end + 1 - |
4325 | new_size); | |
514ac8ad CM |
4326 | |
4327 | /* | |
4328 | * update the ram bytes to properly reflect | |
4329 | * the new size of our item | |
4330 | */ | |
4331 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
e02119d5 CM |
4332 | size = |
4333 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4334 | btrfs_truncate_item(root, path, size, 1); |
27cdeb70 MX |
4335 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4336 | &root->state)) { | |
a76a3cd4 YZ |
4337 | inode_sub_bytes(inode, item_end + 1 - |
4338 | found_key.offset); | |
9069218d | 4339 | } |
39279cc3 | 4340 | } |
179e29e4 | 4341 | delete: |
39279cc3 | 4342 | if (del_item) { |
85e21bac CM |
4343 | if (!pending_del_nr) { |
4344 | /* no pending yet, add ourselves */ | |
4345 | pending_del_slot = path->slots[0]; | |
4346 | pending_del_nr = 1; | |
4347 | } else if (pending_del_nr && | |
4348 | path->slots[0] + 1 == pending_del_slot) { | |
4349 | /* hop on the pending chunk */ | |
4350 | pending_del_nr++; | |
4351 | pending_del_slot = path->slots[0]; | |
4352 | } else { | |
d397712b | 4353 | BUG(); |
85e21bac | 4354 | } |
39279cc3 CM |
4355 | } else { |
4356 | break; | |
4357 | } | |
27cdeb70 MX |
4358 | if (found_extent && |
4359 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4360 | root == root->fs_info->tree_root)) { | |
b9473439 | 4361 | btrfs_set_path_blocking(path); |
39279cc3 | 4362 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4363 | extent_num_bytes, 0, |
4364 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4365 | ino, extent_offset, 0); |
39279cc3 CM |
4366 | BUG_ON(ret); |
4367 | } | |
85e21bac | 4368 | |
8082510e YZ |
4369 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4370 | break; | |
4371 | ||
4372 | if (path->slots[0] == 0 || | |
4373 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4374 | if (pending_del_nr) { |
4375 | ret = btrfs_del_items(trans, root, path, | |
4376 | pending_del_slot, | |
4377 | pending_del_nr); | |
79787eaa JM |
4378 | if (ret) { |
4379 | btrfs_abort_transaction(trans, | |
4380 | root, ret); | |
4381 | goto error; | |
4382 | } | |
8082510e YZ |
4383 | pending_del_nr = 0; |
4384 | } | |
b3b4aa74 | 4385 | btrfs_release_path(path); |
85e21bac | 4386 | goto search_again; |
8082510e YZ |
4387 | } else { |
4388 | path->slots[0]--; | |
85e21bac | 4389 | } |
39279cc3 | 4390 | } |
8082510e | 4391 | out: |
85e21bac CM |
4392 | if (pending_del_nr) { |
4393 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4394 | pending_del_nr); | |
79787eaa JM |
4395 | if (ret) |
4396 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4397 | } |
79787eaa | 4398 | error: |
dac5705c FM |
4399 | if (last_size != (u64)-1 && |
4400 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
7f4f6e0a | 4401 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
39279cc3 | 4402 | btrfs_free_path(path); |
8082510e | 4403 | return err; |
39279cc3 CM |
4404 | } |
4405 | ||
4406 | /* | |
2aaa6655 JB |
4407 | * btrfs_truncate_page - read, zero a chunk and write a page |
4408 | * @inode - inode that we're zeroing | |
4409 | * @from - the offset to start zeroing | |
4410 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4411 | * offset | |
4412 | * @front - zero up to the offset instead of from the offset on | |
4413 | * | |
4414 | * This will find the page for the "from" offset and cow the page and zero the | |
4415 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4416 | */ |
2aaa6655 JB |
4417 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4418 | int front) | |
39279cc3 | 4419 | { |
2aaa6655 | 4420 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4421 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4422 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4423 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4424 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4425 | char *kaddr; |
db94535d | 4426 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4427 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4428 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4429 | struct page *page; | |
3b16a4e3 | 4430 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4431 | int ret = 0; |
a52d9a80 | 4432 | u64 page_start; |
e6dcd2dc | 4433 | u64 page_end; |
39279cc3 | 4434 | |
2aaa6655 JB |
4435 | if ((offset & (blocksize - 1)) == 0 && |
4436 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4437 | goto out; |
0ca1f7ce | 4438 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4439 | if (ret) |
4440 | goto out; | |
39279cc3 | 4441 | |
211c17f5 | 4442 | again: |
3b16a4e3 | 4443 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4444 | if (!page) { |
0ca1f7ce | 4445 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4446 | ret = -ENOMEM; |
39279cc3 | 4447 | goto out; |
5d5e103a | 4448 | } |
e6dcd2dc CM |
4449 | |
4450 | page_start = page_offset(page); | |
4451 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4452 | ||
39279cc3 | 4453 | if (!PageUptodate(page)) { |
9ebefb18 | 4454 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4455 | lock_page(page); |
211c17f5 CM |
4456 | if (page->mapping != mapping) { |
4457 | unlock_page(page); | |
4458 | page_cache_release(page); | |
4459 | goto again; | |
4460 | } | |
39279cc3 CM |
4461 | if (!PageUptodate(page)) { |
4462 | ret = -EIO; | |
89642229 | 4463 | goto out_unlock; |
39279cc3 CM |
4464 | } |
4465 | } | |
211c17f5 | 4466 | wait_on_page_writeback(page); |
e6dcd2dc | 4467 | |
d0082371 | 4468 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4469 | set_page_extent_mapped(page); |
4470 | ||
4471 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4472 | if (ordered) { | |
2ac55d41 JB |
4473 | unlock_extent_cached(io_tree, page_start, page_end, |
4474 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4475 | unlock_page(page); |
4476 | page_cache_release(page); | |
eb84ae03 | 4477 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4478 | btrfs_put_ordered_extent(ordered); |
4479 | goto again; | |
4480 | } | |
4481 | ||
2ac55d41 | 4482 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4483 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4484 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4485 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4486 | |
2ac55d41 JB |
4487 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4488 | &cached_state); | |
9ed74f2d | 4489 | if (ret) { |
2ac55d41 JB |
4490 | unlock_extent_cached(io_tree, page_start, page_end, |
4491 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4492 | goto out_unlock; |
4493 | } | |
4494 | ||
e6dcd2dc | 4495 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4496 | if (!len) |
4497 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4498 | kaddr = kmap(page); |
2aaa6655 JB |
4499 | if (front) |
4500 | memset(kaddr, 0, offset); | |
4501 | else | |
4502 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4503 | flush_dcache_page(page); |
4504 | kunmap(page); | |
4505 | } | |
247e743c | 4506 | ClearPageChecked(page); |
e6dcd2dc | 4507 | set_page_dirty(page); |
2ac55d41 JB |
4508 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4509 | GFP_NOFS); | |
39279cc3 | 4510 | |
89642229 | 4511 | out_unlock: |
5d5e103a | 4512 | if (ret) |
0ca1f7ce | 4513 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4514 | unlock_page(page); |
4515 | page_cache_release(page); | |
4516 | out: | |
4517 | return ret; | |
4518 | } | |
4519 | ||
16e7549f JB |
4520 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4521 | u64 offset, u64 len) | |
4522 | { | |
4523 | struct btrfs_trans_handle *trans; | |
4524 | int ret; | |
4525 | ||
4526 | /* | |
4527 | * Still need to make sure the inode looks like it's been updated so | |
4528 | * that any holes get logged if we fsync. | |
4529 | */ | |
4530 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4531 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4532 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4533 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4534 | return 0; | |
4535 | } | |
4536 | ||
4537 | /* | |
4538 | * 1 - for the one we're dropping | |
4539 | * 1 - for the one we're adding | |
4540 | * 1 - for updating the inode. | |
4541 | */ | |
4542 | trans = btrfs_start_transaction(root, 3); | |
4543 | if (IS_ERR(trans)) | |
4544 | return PTR_ERR(trans); | |
4545 | ||
4546 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4547 | if (ret) { | |
4548 | btrfs_abort_transaction(trans, root, ret); | |
4549 | btrfs_end_transaction(trans, root); | |
4550 | return ret; | |
4551 | } | |
4552 | ||
4553 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4554 | 0, 0, len, 0, len, 0, 0, 0); | |
4555 | if (ret) | |
4556 | btrfs_abort_transaction(trans, root, ret); | |
4557 | else | |
4558 | btrfs_update_inode(trans, root, inode); | |
4559 | btrfs_end_transaction(trans, root); | |
4560 | return ret; | |
4561 | } | |
4562 | ||
695a0d0d JB |
4563 | /* |
4564 | * This function puts in dummy file extents for the area we're creating a hole | |
4565 | * for. So if we are truncating this file to a larger size we need to insert | |
4566 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4567 | * the range between oldsize and size | |
4568 | */ | |
a41ad394 | 4569 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4570 | { |
9036c102 YZ |
4571 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4572 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4573 | struct extent_map *em = NULL; |
2ac55d41 | 4574 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4575 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4576 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4577 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4578 | u64 last_byte; |
4579 | u64 cur_offset; | |
4580 | u64 hole_size; | |
9ed74f2d | 4581 | int err = 0; |
39279cc3 | 4582 | |
a71754fc JB |
4583 | /* |
4584 | * If our size started in the middle of a page we need to zero out the | |
4585 | * rest of the page before we expand the i_size, otherwise we could | |
4586 | * expose stale data. | |
4587 | */ | |
4588 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4589 | if (err) | |
4590 | return err; | |
4591 | ||
9036c102 YZ |
4592 | if (size <= hole_start) |
4593 | return 0; | |
4594 | ||
9036c102 YZ |
4595 | while (1) { |
4596 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4597 | |
2ac55d41 | 4598 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4599 | &cached_state); |
fa7c1494 MX |
4600 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4601 | block_end - hole_start); | |
9036c102 YZ |
4602 | if (!ordered) |
4603 | break; | |
2ac55d41 JB |
4604 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4605 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4606 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4607 | btrfs_put_ordered_extent(ordered); |
4608 | } | |
39279cc3 | 4609 | |
9036c102 YZ |
4610 | cur_offset = hole_start; |
4611 | while (1) { | |
4612 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4613 | block_end - cur_offset, 0); | |
79787eaa JM |
4614 | if (IS_ERR(em)) { |
4615 | err = PTR_ERR(em); | |
f2767956 | 4616 | em = NULL; |
79787eaa JM |
4617 | break; |
4618 | } | |
9036c102 | 4619 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4620 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4621 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4622 | struct extent_map *hole_em; |
9036c102 | 4623 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4624 | |
16e7549f JB |
4625 | err = maybe_insert_hole(root, inode, cur_offset, |
4626 | hole_size); | |
4627 | if (err) | |
3893e33b | 4628 | break; |
5dc562c5 JB |
4629 | btrfs_drop_extent_cache(inode, cur_offset, |
4630 | cur_offset + hole_size - 1, 0); | |
4631 | hole_em = alloc_extent_map(); | |
4632 | if (!hole_em) { | |
4633 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4634 | &BTRFS_I(inode)->runtime_flags); | |
4635 | goto next; | |
4636 | } | |
4637 | hole_em->start = cur_offset; | |
4638 | hole_em->len = hole_size; | |
4639 | hole_em->orig_start = cur_offset; | |
8082510e | 4640 | |
5dc562c5 JB |
4641 | hole_em->block_start = EXTENT_MAP_HOLE; |
4642 | hole_em->block_len = 0; | |
b4939680 | 4643 | hole_em->orig_block_len = 0; |
cc95bef6 | 4644 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4645 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4646 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4647 | hole_em->generation = root->fs_info->generation; |
8082510e | 4648 | |
5dc562c5 JB |
4649 | while (1) { |
4650 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4651 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4652 | write_unlock(&em_tree->lock); |
4653 | if (err != -EEXIST) | |
4654 | break; | |
4655 | btrfs_drop_extent_cache(inode, cur_offset, | |
4656 | cur_offset + | |
4657 | hole_size - 1, 0); | |
4658 | } | |
4659 | free_extent_map(hole_em); | |
9036c102 | 4660 | } |
16e7549f | 4661 | next: |
9036c102 | 4662 | free_extent_map(em); |
a22285a6 | 4663 | em = NULL; |
9036c102 | 4664 | cur_offset = last_byte; |
8082510e | 4665 | if (cur_offset >= block_end) |
9036c102 YZ |
4666 | break; |
4667 | } | |
a22285a6 | 4668 | free_extent_map(em); |
2ac55d41 JB |
4669 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4670 | GFP_NOFS); | |
9036c102 YZ |
4671 | return err; |
4672 | } | |
39279cc3 | 4673 | |
9ea24bbe FM |
4674 | static int wait_snapshoting_atomic_t(atomic_t *a) |
4675 | { | |
4676 | schedule(); | |
4677 | return 0; | |
4678 | } | |
4679 | ||
4680 | static void wait_for_snapshot_creation(struct btrfs_root *root) | |
4681 | { | |
4682 | while (true) { | |
4683 | int ret; | |
4684 | ||
4685 | ret = btrfs_start_write_no_snapshoting(root); | |
4686 | if (ret) | |
4687 | break; | |
4688 | wait_on_atomic_t(&root->will_be_snapshoted, | |
4689 | wait_snapshoting_atomic_t, | |
4690 | TASK_UNINTERRUPTIBLE); | |
4691 | } | |
4692 | } | |
4693 | ||
3972f260 | 4694 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4695 | { |
f4a2f4c5 MX |
4696 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4697 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4698 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4699 | loff_t newsize = attr->ia_size; |
4700 | int mask = attr->ia_valid; | |
8082510e YZ |
4701 | int ret; |
4702 | ||
3972f260 ES |
4703 | /* |
4704 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4705 | * special case where we need to update the times despite not having | |
4706 | * these flags set. For all other operations the VFS set these flags | |
4707 | * explicitly if it wants a timestamp update. | |
4708 | */ | |
dff6efc3 CH |
4709 | if (newsize != oldsize) { |
4710 | inode_inc_iversion(inode); | |
4711 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4712 | inode->i_ctime = inode->i_mtime = | |
4713 | current_fs_time(inode->i_sb); | |
4714 | } | |
3972f260 | 4715 | |
a41ad394 | 4716 | if (newsize > oldsize) { |
7caef267 | 4717 | truncate_pagecache(inode, newsize); |
9ea24bbe FM |
4718 | /* |
4719 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4720 | * This is to ensure the snapshot captures a fully consistent | |
4721 | * state of this file - if the snapshot captures this expanding | |
4722 | * truncation, it must capture all writes that happened before | |
4723 | * this truncation. | |
4724 | */ | |
4725 | wait_for_snapshot_creation(root); | |
a41ad394 | 4726 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4727 | if (ret) { |
4728 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4729 | return ret; |
9ea24bbe | 4730 | } |
8082510e | 4731 | |
f4a2f4c5 | 4732 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4733 | if (IS_ERR(trans)) { |
4734 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4735 | return PTR_ERR(trans); |
9ea24bbe | 4736 | } |
f4a2f4c5 MX |
4737 | |
4738 | i_size_write(inode, newsize); | |
4739 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4740 | ret = btrfs_update_inode(trans, root, inode); | |
9ea24bbe | 4741 | btrfs_end_write_no_snapshoting(root); |
7ad85bb7 | 4742 | btrfs_end_transaction(trans, root); |
a41ad394 | 4743 | } else { |
8082510e | 4744 | |
a41ad394 JB |
4745 | /* |
4746 | * We're truncating a file that used to have good data down to | |
4747 | * zero. Make sure it gets into the ordered flush list so that | |
4748 | * any new writes get down to disk quickly. | |
4749 | */ | |
4750 | if (newsize == 0) | |
72ac3c0d JB |
4751 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4752 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4753 | |
f3fe820c JB |
4754 | /* |
4755 | * 1 for the orphan item we're going to add | |
4756 | * 1 for the orphan item deletion. | |
4757 | */ | |
4758 | trans = btrfs_start_transaction(root, 2); | |
4759 | if (IS_ERR(trans)) | |
4760 | return PTR_ERR(trans); | |
4761 | ||
4762 | /* | |
4763 | * We need to do this in case we fail at _any_ point during the | |
4764 | * actual truncate. Once we do the truncate_setsize we could | |
4765 | * invalidate pages which forces any outstanding ordered io to | |
4766 | * be instantly completed which will give us extents that need | |
4767 | * to be truncated. If we fail to get an orphan inode down we | |
4768 | * could have left over extents that were never meant to live, | |
4769 | * so we need to garuntee from this point on that everything | |
4770 | * will be consistent. | |
4771 | */ | |
4772 | ret = btrfs_orphan_add(trans, inode); | |
4773 | btrfs_end_transaction(trans, root); | |
4774 | if (ret) | |
4775 | return ret; | |
4776 | ||
a41ad394 JB |
4777 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4778 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4779 | |
4780 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4781 | btrfs_inode_block_unlocked_dio(inode); | |
4782 | inode_dio_wait(inode); | |
4783 | btrfs_inode_resume_unlocked_dio(inode); | |
4784 | ||
a41ad394 | 4785 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4786 | if (ret && inode->i_nlink) { |
4787 | int err; | |
4788 | ||
4789 | /* | |
4790 | * failed to truncate, disk_i_size is only adjusted down | |
4791 | * as we remove extents, so it should represent the true | |
4792 | * size of the inode, so reset the in memory size and | |
4793 | * delete our orphan entry. | |
4794 | */ | |
4795 | trans = btrfs_join_transaction(root); | |
4796 | if (IS_ERR(trans)) { | |
4797 | btrfs_orphan_del(NULL, inode); | |
4798 | return ret; | |
4799 | } | |
4800 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4801 | err = btrfs_orphan_del(trans, inode); | |
4802 | if (err) | |
4803 | btrfs_abort_transaction(trans, root, err); | |
4804 | btrfs_end_transaction(trans, root); | |
4805 | } | |
8082510e YZ |
4806 | } |
4807 | ||
a41ad394 | 4808 | return ret; |
8082510e YZ |
4809 | } |
4810 | ||
9036c102 YZ |
4811 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4812 | { | |
4813 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4814 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4815 | int err; |
39279cc3 | 4816 | |
b83cc969 LZ |
4817 | if (btrfs_root_readonly(root)) |
4818 | return -EROFS; | |
4819 | ||
9036c102 YZ |
4820 | err = inode_change_ok(inode, attr); |
4821 | if (err) | |
4822 | return err; | |
2bf5a725 | 4823 | |
5a3f23d5 | 4824 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4825 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4826 | if (err) |
4827 | return err; | |
39279cc3 | 4828 | } |
9036c102 | 4829 | |
1025774c CH |
4830 | if (attr->ia_valid) { |
4831 | setattr_copy(inode, attr); | |
0c4d2d95 | 4832 | inode_inc_iversion(inode); |
22c44fe6 | 4833 | err = btrfs_dirty_inode(inode); |
1025774c | 4834 | |
22c44fe6 | 4835 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 4836 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 4837 | } |
33268eaf | 4838 | |
39279cc3 CM |
4839 | return err; |
4840 | } | |
61295eb8 | 4841 | |
131e404a FDBM |
4842 | /* |
4843 | * While truncating the inode pages during eviction, we get the VFS calling | |
4844 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
4845 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
4846 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
4847 | * extent_state structures over and over, wasting lots of time. | |
4848 | * | |
4849 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
4850 | * those expensive operations on a per page basis and do only the ordered io | |
4851 | * finishing, while we release here the extent_map and extent_state structures, | |
4852 | * without the excessive merging and splitting. | |
4853 | */ | |
4854 | static void evict_inode_truncate_pages(struct inode *inode) | |
4855 | { | |
4856 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
4857 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
4858 | struct rb_node *node; | |
4859 | ||
4860 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 4861 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
4862 | |
4863 | write_lock(&map_tree->lock); | |
4864 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
4865 | struct extent_map *em; | |
4866 | ||
4867 | node = rb_first(&map_tree->map); | |
4868 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
4869 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
4870 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
4871 | remove_extent_mapping(map_tree, em); |
4872 | free_extent_map(em); | |
7064dd5c FM |
4873 | if (need_resched()) { |
4874 | write_unlock(&map_tree->lock); | |
4875 | cond_resched(); | |
4876 | write_lock(&map_tree->lock); | |
4877 | } | |
131e404a FDBM |
4878 | } |
4879 | write_unlock(&map_tree->lock); | |
4880 | ||
4881 | spin_lock(&io_tree->lock); | |
4882 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
4883 | struct extent_state *state; | |
4884 | struct extent_state *cached_state = NULL; | |
4885 | ||
4886 | node = rb_first(&io_tree->state); | |
4887 | state = rb_entry(node, struct extent_state, rb_node); | |
4888 | atomic_inc(&state->refs); | |
4889 | spin_unlock(&io_tree->lock); | |
4890 | ||
4891 | lock_extent_bits(io_tree, state->start, state->end, | |
4892 | 0, &cached_state); | |
4893 | clear_extent_bit(io_tree, state->start, state->end, | |
4894 | EXTENT_LOCKED | EXTENT_DIRTY | | |
4895 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
4896 | EXTENT_DEFRAG, 1, 1, | |
4897 | &cached_state, GFP_NOFS); | |
4898 | free_extent_state(state); | |
4899 | ||
7064dd5c | 4900 | cond_resched(); |
131e404a FDBM |
4901 | spin_lock(&io_tree->lock); |
4902 | } | |
4903 | spin_unlock(&io_tree->lock); | |
4904 | } | |
4905 | ||
bd555975 | 4906 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4907 | { |
4908 | struct btrfs_trans_handle *trans; | |
4909 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4910 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4911 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4912 | int ret; |
4913 | ||
1abe9b8a | 4914 | trace_btrfs_inode_evict(inode); |
4915 | ||
131e404a FDBM |
4916 | evict_inode_truncate_pages(inode); |
4917 | ||
69e9c6c6 SB |
4918 | if (inode->i_nlink && |
4919 | ((btrfs_root_refs(&root->root_item) != 0 && | |
4920 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
4921 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
4922 | goto no_delete; |
4923 | ||
39279cc3 | 4924 | if (is_bad_inode(inode)) { |
7b128766 | 4925 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4926 | goto no_delete; |
4927 | } | |
bd555975 | 4928 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4929 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4930 | |
f612496b MX |
4931 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
4932 | ||
c71bf099 | 4933 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4934 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4935 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4936 | goto no_delete; |
4937 | } | |
4938 | ||
76dda93c | 4939 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
4940 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
4941 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
4942 | goto no_delete; |
4943 | } | |
4944 | ||
0e8c36a9 MX |
4945 | ret = btrfs_commit_inode_delayed_inode(inode); |
4946 | if (ret) { | |
4947 | btrfs_orphan_del(NULL, inode); | |
4948 | goto no_delete; | |
4949 | } | |
4950 | ||
66d8f3dd | 4951 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4952 | if (!rsv) { |
4953 | btrfs_orphan_del(NULL, inode); | |
4954 | goto no_delete; | |
4955 | } | |
4a338542 | 4956 | rsv->size = min_size; |
ca7e70f5 | 4957 | rsv->failfast = 1; |
726c35fa | 4958 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4959 | |
dbe674a9 | 4960 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4961 | |
4289a667 | 4962 | /* |
8407aa46 MX |
4963 | * This is a bit simpler than btrfs_truncate since we've already |
4964 | * reserved our space for our orphan item in the unlink, so we just | |
4965 | * need to reserve some slack space in case we add bytes and update | |
4966 | * inode item when doing the truncate. | |
4289a667 | 4967 | */ |
8082510e | 4968 | while (1) { |
08e007d2 MX |
4969 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4970 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4971 | |
4972 | /* | |
4973 | * Try and steal from the global reserve since we will | |
4974 | * likely not use this space anyway, we want to try as | |
4975 | * hard as possible to get this to work. | |
4976 | */ | |
4977 | if (ret) | |
4978 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4979 | |
d68fc57b | 4980 | if (ret) { |
c2cf52eb SK |
4981 | btrfs_warn(root->fs_info, |
4982 | "Could not get space for a delete, will truncate on mount %d", | |
4983 | ret); | |
4289a667 JB |
4984 | btrfs_orphan_del(NULL, inode); |
4985 | btrfs_free_block_rsv(root, rsv); | |
4986 | goto no_delete; | |
d68fc57b | 4987 | } |
7b128766 | 4988 | |
0e8c36a9 | 4989 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4990 | if (IS_ERR(trans)) { |
4991 | btrfs_orphan_del(NULL, inode); | |
4992 | btrfs_free_block_rsv(root, rsv); | |
4993 | goto no_delete; | |
d68fc57b | 4994 | } |
7b128766 | 4995 | |
4289a667 JB |
4996 | trans->block_rsv = rsv; |
4997 | ||
d68fc57b | 4998 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4999 | if (ret != -ENOSPC) |
8082510e | 5000 | break; |
85e21bac | 5001 | |
8407aa46 | 5002 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
5003 | btrfs_end_transaction(trans, root); |
5004 | trans = NULL; | |
b53d3f5d | 5005 | btrfs_btree_balance_dirty(root); |
8082510e | 5006 | } |
5f39d397 | 5007 | |
4289a667 JB |
5008 | btrfs_free_block_rsv(root, rsv); |
5009 | ||
4ef31a45 JB |
5010 | /* |
5011 | * Errors here aren't a big deal, it just means we leave orphan items | |
5012 | * in the tree. They will be cleaned up on the next mount. | |
5013 | */ | |
8082510e | 5014 | if (ret == 0) { |
4289a667 | 5015 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5016 | btrfs_orphan_del(trans, inode); |
5017 | } else { | |
5018 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5019 | } |
54aa1f4d | 5020 | |
4289a667 | 5021 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
5022 | if (!(root == root->fs_info->tree_root || |
5023 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 5024 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 5025 | |
54aa1f4d | 5026 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5027 | btrfs_btree_balance_dirty(root); |
39279cc3 | 5028 | no_delete: |
89042e5a | 5029 | btrfs_remove_delayed_node(inode); |
dbd5768f | 5030 | clear_inode(inode); |
8082510e | 5031 | return; |
39279cc3 CM |
5032 | } |
5033 | ||
5034 | /* | |
5035 | * this returns the key found in the dir entry in the location pointer. | |
5036 | * If no dir entries were found, location->objectid is 0. | |
5037 | */ | |
5038 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5039 | struct btrfs_key *location) | |
5040 | { | |
5041 | const char *name = dentry->d_name.name; | |
5042 | int namelen = dentry->d_name.len; | |
5043 | struct btrfs_dir_item *di; | |
5044 | struct btrfs_path *path; | |
5045 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5046 | int ret = 0; |
39279cc3 CM |
5047 | |
5048 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5049 | if (!path) |
5050 | return -ENOMEM; | |
3954401f | 5051 | |
33345d01 | 5052 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 5053 | namelen, 0); |
0d9f7f3e Y |
5054 | if (IS_ERR(di)) |
5055 | ret = PTR_ERR(di); | |
d397712b | 5056 | |
c704005d | 5057 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5058 | goto out_err; |
d397712b | 5059 | |
5f39d397 | 5060 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5061 | out: |
39279cc3 CM |
5062 | btrfs_free_path(path); |
5063 | return ret; | |
3954401f CM |
5064 | out_err: |
5065 | location->objectid = 0; | |
5066 | goto out; | |
39279cc3 CM |
5067 | } |
5068 | ||
5069 | /* | |
5070 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5071 | * needs to be changed to reflect the root directory of the tree root. This | |
5072 | * is kind of like crossing a mount point. | |
5073 | */ | |
5074 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
5075 | struct inode *dir, |
5076 | struct dentry *dentry, | |
5077 | struct btrfs_key *location, | |
5078 | struct btrfs_root **sub_root) | |
39279cc3 | 5079 | { |
4df27c4d YZ |
5080 | struct btrfs_path *path; |
5081 | struct btrfs_root *new_root; | |
5082 | struct btrfs_root_ref *ref; | |
5083 | struct extent_buffer *leaf; | |
1d4c08e0 | 5084 | struct btrfs_key key; |
4df27c4d YZ |
5085 | int ret; |
5086 | int err = 0; | |
39279cc3 | 5087 | |
4df27c4d YZ |
5088 | path = btrfs_alloc_path(); |
5089 | if (!path) { | |
5090 | err = -ENOMEM; | |
5091 | goto out; | |
5092 | } | |
39279cc3 | 5093 | |
4df27c4d | 5094 | err = -ENOENT; |
1d4c08e0 DS |
5095 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5096 | key.type = BTRFS_ROOT_REF_KEY; | |
5097 | key.offset = location->objectid; | |
5098 | ||
5099 | ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path, | |
5100 | 0, 0); | |
4df27c4d YZ |
5101 | if (ret) { |
5102 | if (ret < 0) | |
5103 | err = ret; | |
5104 | goto out; | |
5105 | } | |
39279cc3 | 5106 | |
4df27c4d YZ |
5107 | leaf = path->nodes[0]; |
5108 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 5109 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
5110 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5111 | goto out; | |
39279cc3 | 5112 | |
4df27c4d YZ |
5113 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5114 | (unsigned long)(ref + 1), | |
5115 | dentry->d_name.len); | |
5116 | if (ret) | |
5117 | goto out; | |
5118 | ||
b3b4aa74 | 5119 | btrfs_release_path(path); |
4df27c4d YZ |
5120 | |
5121 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5122 | if (IS_ERR(new_root)) { | |
5123 | err = PTR_ERR(new_root); | |
5124 | goto out; | |
5125 | } | |
5126 | ||
4df27c4d YZ |
5127 | *sub_root = new_root; |
5128 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5129 | location->type = BTRFS_INODE_ITEM_KEY; | |
5130 | location->offset = 0; | |
5131 | err = 0; | |
5132 | out: | |
5133 | btrfs_free_path(path); | |
5134 | return err; | |
39279cc3 CM |
5135 | } |
5136 | ||
5d4f98a2 YZ |
5137 | static void inode_tree_add(struct inode *inode) |
5138 | { | |
5139 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5140 | struct btrfs_inode *entry; | |
03e860bd FNP |
5141 | struct rb_node **p; |
5142 | struct rb_node *parent; | |
cef21937 | 5143 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5144 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5145 | |
1d3382cb | 5146 | if (inode_unhashed(inode)) |
76dda93c | 5147 | return; |
e1409cef | 5148 | parent = NULL; |
5d4f98a2 | 5149 | spin_lock(&root->inode_lock); |
e1409cef | 5150 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5151 | while (*p) { |
5152 | parent = *p; | |
5153 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5154 | ||
33345d01 | 5155 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5156 | p = &parent->rb_left; |
33345d01 | 5157 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5158 | p = &parent->rb_right; |
5d4f98a2 YZ |
5159 | else { |
5160 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5161 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5162 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5163 | RB_CLEAR_NODE(parent); |
5164 | spin_unlock(&root->inode_lock); | |
cef21937 | 5165 | return; |
5d4f98a2 YZ |
5166 | } |
5167 | } | |
cef21937 FDBM |
5168 | rb_link_node(new, parent, p); |
5169 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5170 | spin_unlock(&root->inode_lock); |
5171 | } | |
5172 | ||
5173 | static void inode_tree_del(struct inode *inode) | |
5174 | { | |
5175 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5176 | int empty = 0; |
5d4f98a2 | 5177 | |
03e860bd | 5178 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5179 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5180 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5181 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5182 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5183 | } |
03e860bd | 5184 | spin_unlock(&root->inode_lock); |
76dda93c | 5185 | |
69e9c6c6 | 5186 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5187 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5188 | spin_lock(&root->inode_lock); | |
5189 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5190 | spin_unlock(&root->inode_lock); | |
5191 | if (empty) | |
5192 | btrfs_add_dead_root(root); | |
5193 | } | |
5194 | } | |
5195 | ||
143bede5 | 5196 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5197 | { |
5198 | struct rb_node *node; | |
5199 | struct rb_node *prev; | |
5200 | struct btrfs_inode *entry; | |
5201 | struct inode *inode; | |
5202 | u64 objectid = 0; | |
5203 | ||
7813b3db LB |
5204 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5205 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5206 | |
5207 | spin_lock(&root->inode_lock); | |
5208 | again: | |
5209 | node = root->inode_tree.rb_node; | |
5210 | prev = NULL; | |
5211 | while (node) { | |
5212 | prev = node; | |
5213 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5214 | ||
33345d01 | 5215 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5216 | node = node->rb_left; |
33345d01 | 5217 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5218 | node = node->rb_right; |
5219 | else | |
5220 | break; | |
5221 | } | |
5222 | if (!node) { | |
5223 | while (prev) { | |
5224 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5225 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5226 | node = prev; |
5227 | break; | |
5228 | } | |
5229 | prev = rb_next(prev); | |
5230 | } | |
5231 | } | |
5232 | while (node) { | |
5233 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5234 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5235 | inode = igrab(&entry->vfs_inode); |
5236 | if (inode) { | |
5237 | spin_unlock(&root->inode_lock); | |
5238 | if (atomic_read(&inode->i_count) > 1) | |
5239 | d_prune_aliases(inode); | |
5240 | /* | |
45321ac5 | 5241 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5242 | * the inode cache when its usage count |
5243 | * hits zero. | |
5244 | */ | |
5245 | iput(inode); | |
5246 | cond_resched(); | |
5247 | spin_lock(&root->inode_lock); | |
5248 | goto again; | |
5249 | } | |
5250 | ||
5251 | if (cond_resched_lock(&root->inode_lock)) | |
5252 | goto again; | |
5253 | ||
5254 | node = rb_next(node); | |
5255 | } | |
5256 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5257 | } |
5258 | ||
e02119d5 CM |
5259 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5260 | { | |
5261 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5262 | inode->i_ino = args->location->objectid; |
5263 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5264 | sizeof(*args->location)); | |
e02119d5 | 5265 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5266 | return 0; |
5267 | } | |
5268 | ||
5269 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5270 | { | |
5271 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5272 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5273 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5274 | } |
5275 | ||
5d4f98a2 | 5276 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5277 | struct btrfs_key *location, |
5d4f98a2 | 5278 | struct btrfs_root *root) |
39279cc3 CM |
5279 | { |
5280 | struct inode *inode; | |
5281 | struct btrfs_iget_args args; | |
90d3e592 | 5282 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5283 | |
90d3e592 | 5284 | args.location = location; |
39279cc3 CM |
5285 | args.root = root; |
5286 | ||
778ba82b | 5287 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5288 | btrfs_init_locked_inode, |
5289 | (void *)&args); | |
5290 | return inode; | |
5291 | } | |
5292 | ||
1a54ef8c BR |
5293 | /* Get an inode object given its location and corresponding root. |
5294 | * Returns in *is_new if the inode was read from disk | |
5295 | */ | |
5296 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5297 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5298 | { |
5299 | struct inode *inode; | |
5300 | ||
90d3e592 | 5301 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5302 | if (!inode) |
5d4f98a2 | 5303 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5304 | |
5305 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5306 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5307 | if (!is_bad_inode(inode)) { |
5308 | inode_tree_add(inode); | |
5309 | unlock_new_inode(inode); | |
5310 | if (new) | |
5311 | *new = 1; | |
5312 | } else { | |
e0b6d65b ST |
5313 | unlock_new_inode(inode); |
5314 | iput(inode); | |
5315 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5316 | } |
5317 | } | |
5318 | ||
1a54ef8c BR |
5319 | return inode; |
5320 | } | |
5321 | ||
4df27c4d YZ |
5322 | static struct inode *new_simple_dir(struct super_block *s, |
5323 | struct btrfs_key *key, | |
5324 | struct btrfs_root *root) | |
5325 | { | |
5326 | struct inode *inode = new_inode(s); | |
5327 | ||
5328 | if (!inode) | |
5329 | return ERR_PTR(-ENOMEM); | |
5330 | ||
4df27c4d YZ |
5331 | BTRFS_I(inode)->root = root; |
5332 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5333 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5334 | |
5335 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5336 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5337 | inode->i_fop = &simple_dir_operations; |
5338 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
9cc97d64 | 5339 | inode->i_mtime = CURRENT_TIME; |
5340 | inode->i_atime = inode->i_mtime; | |
5341 | inode->i_ctime = inode->i_mtime; | |
5342 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5343 | |
5344 | return inode; | |
5345 | } | |
5346 | ||
3de4586c | 5347 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5348 | { |
d397712b | 5349 | struct inode *inode; |
4df27c4d | 5350 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5351 | struct btrfs_root *sub_root = root; |
5352 | struct btrfs_key location; | |
76dda93c | 5353 | int index; |
b4aff1f8 | 5354 | int ret = 0; |
39279cc3 CM |
5355 | |
5356 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5357 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5358 | |
39e3c955 | 5359 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5360 | if (ret < 0) |
5361 | return ERR_PTR(ret); | |
5f39d397 | 5362 | |
4df27c4d | 5363 | if (location.objectid == 0) |
5662344b | 5364 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5365 | |
5366 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5367 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5368 | return inode; |
5369 | } | |
5370 | ||
5371 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5372 | ||
76dda93c | 5373 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5374 | ret = fixup_tree_root_location(root, dir, dentry, |
5375 | &location, &sub_root); | |
5376 | if (ret < 0) { | |
5377 | if (ret != -ENOENT) | |
5378 | inode = ERR_PTR(ret); | |
5379 | else | |
5380 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5381 | } else { | |
73f73415 | 5382 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5383 | } |
76dda93c YZ |
5384 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5385 | ||
34d19bad | 5386 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5387 | down_read(&root->fs_info->cleanup_work_sem); |
5388 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5389 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5390 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5391 | if (ret) { |
5392 | iput(inode); | |
66b4ffd1 | 5393 | inode = ERR_PTR(ret); |
01cd3367 | 5394 | } |
c71bf099 YZ |
5395 | } |
5396 | ||
3de4586c CM |
5397 | return inode; |
5398 | } | |
5399 | ||
fe15ce44 | 5400 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5401 | { |
5402 | struct btrfs_root *root; | |
848cce0d | 5403 | struct inode *inode = dentry->d_inode; |
76dda93c | 5404 | |
848cce0d LZ |
5405 | if (!inode && !IS_ROOT(dentry)) |
5406 | inode = dentry->d_parent->d_inode; | |
76dda93c | 5407 | |
848cce0d LZ |
5408 | if (inode) { |
5409 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5410 | if (btrfs_root_refs(&root->root_item) == 0) |
5411 | return 1; | |
848cce0d LZ |
5412 | |
5413 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5414 | return 1; | |
efefb143 | 5415 | } |
76dda93c YZ |
5416 | return 0; |
5417 | } | |
5418 | ||
b4aff1f8 JB |
5419 | static void btrfs_dentry_release(struct dentry *dentry) |
5420 | { | |
944a4515 | 5421 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5422 | } |
5423 | ||
3de4586c | 5424 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5425 | unsigned int flags) |
3de4586c | 5426 | { |
5662344b | 5427 | struct inode *inode; |
a66e7cc6 | 5428 | |
5662344b TI |
5429 | inode = btrfs_lookup_dentry(dir, dentry); |
5430 | if (IS_ERR(inode)) { | |
5431 | if (PTR_ERR(inode) == -ENOENT) | |
5432 | inode = NULL; | |
5433 | else | |
5434 | return ERR_CAST(inode); | |
5435 | } | |
5436 | ||
41d28bca | 5437 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5438 | } |
5439 | ||
16cdcec7 | 5440 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5441 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5442 | }; | |
5443 | ||
9cdda8d3 | 5444 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5445 | { |
9cdda8d3 | 5446 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5447 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5448 | struct btrfs_item *item; | |
5449 | struct btrfs_dir_item *di; | |
5450 | struct btrfs_key key; | |
5f39d397 | 5451 | struct btrfs_key found_key; |
39279cc3 | 5452 | struct btrfs_path *path; |
16cdcec7 MX |
5453 | struct list_head ins_list; |
5454 | struct list_head del_list; | |
39279cc3 | 5455 | int ret; |
5f39d397 | 5456 | struct extent_buffer *leaf; |
39279cc3 | 5457 | int slot; |
39279cc3 CM |
5458 | unsigned char d_type; |
5459 | int over = 0; | |
5460 | u32 di_cur; | |
5461 | u32 di_total; | |
5462 | u32 di_len; | |
5463 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5464 | char tmp_name[32]; |
5465 | char *name_ptr; | |
5466 | int name_len; | |
9cdda8d3 | 5467 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5468 | |
5469 | /* FIXME, use a real flag for deciding about the key type */ | |
5470 | if (root->fs_info->tree_root == root) | |
5471 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5472 | |
9cdda8d3 AV |
5473 | if (!dir_emit_dots(file, ctx)) |
5474 | return 0; | |
5475 | ||
49593bfa | 5476 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5477 | if (!path) |
5478 | return -ENOMEM; | |
ff5714cc | 5479 | |
026fd317 | 5480 | path->reada = 1; |
49593bfa | 5481 | |
16cdcec7 MX |
5482 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5483 | INIT_LIST_HEAD(&ins_list); | |
5484 | INIT_LIST_HEAD(&del_list); | |
5485 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5486 | } | |
5487 | ||
962a298f | 5488 | key.type = key_type; |
9cdda8d3 | 5489 | key.offset = ctx->pos; |
33345d01 | 5490 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5491 | |
39279cc3 CM |
5492 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5493 | if (ret < 0) | |
5494 | goto err; | |
49593bfa DW |
5495 | |
5496 | while (1) { | |
5f39d397 | 5497 | leaf = path->nodes[0]; |
39279cc3 | 5498 | slot = path->slots[0]; |
b9e03af0 LZ |
5499 | if (slot >= btrfs_header_nritems(leaf)) { |
5500 | ret = btrfs_next_leaf(root, path); | |
5501 | if (ret < 0) | |
5502 | goto err; | |
5503 | else if (ret > 0) | |
5504 | break; | |
5505 | continue; | |
39279cc3 | 5506 | } |
3de4586c | 5507 | |
dd3cc16b | 5508 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5509 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5510 | ||
5511 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5512 | break; |
962a298f | 5513 | if (found_key.type != key_type) |
39279cc3 | 5514 | break; |
9cdda8d3 | 5515 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5516 | goto next; |
16cdcec7 MX |
5517 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5518 | btrfs_should_delete_dir_index(&del_list, | |
5519 | found_key.offset)) | |
5520 | goto next; | |
5f39d397 | 5521 | |
9cdda8d3 | 5522 | ctx->pos = found_key.offset; |
16cdcec7 | 5523 | is_curr = 1; |
49593bfa | 5524 | |
39279cc3 CM |
5525 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5526 | di_cur = 0; | |
5f39d397 | 5527 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5528 | |
5529 | while (di_cur < di_total) { | |
5f39d397 CM |
5530 | struct btrfs_key location; |
5531 | ||
22a94d44 JB |
5532 | if (verify_dir_item(root, leaf, di)) |
5533 | break; | |
5534 | ||
5f39d397 | 5535 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5536 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5537 | name_ptr = tmp_name; |
5538 | } else { | |
5539 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5540 | if (!name_ptr) { |
5541 | ret = -ENOMEM; | |
5542 | goto err; | |
5543 | } | |
5f39d397 CM |
5544 | } |
5545 | read_extent_buffer(leaf, name_ptr, | |
5546 | (unsigned long)(di + 1), name_len); | |
5547 | ||
5548 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5549 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5550 | |
fede766f | 5551 | |
3de4586c | 5552 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5553 | * skip it. |
5554 | * | |
5555 | * In contrast to old kernels, we insert the snapshot's | |
5556 | * dir item and dir index after it has been created, so | |
5557 | * we won't find a reference to our own snapshot. We | |
5558 | * still keep the following code for backward | |
5559 | * compatibility. | |
3de4586c CM |
5560 | */ |
5561 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5562 | location.objectid == root->root_key.objectid) { | |
5563 | over = 0; | |
5564 | goto skip; | |
5565 | } | |
9cdda8d3 AV |
5566 | over = !dir_emit(ctx, name_ptr, name_len, |
5567 | location.objectid, d_type); | |
5f39d397 | 5568 | |
3de4586c | 5569 | skip: |
5f39d397 CM |
5570 | if (name_ptr != tmp_name) |
5571 | kfree(name_ptr); | |
5572 | ||
39279cc3 CM |
5573 | if (over) |
5574 | goto nopos; | |
5103e947 | 5575 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5576 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5577 | di_cur += di_len; |
5578 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5579 | } | |
b9e03af0 LZ |
5580 | next: |
5581 | path->slots[0]++; | |
39279cc3 | 5582 | } |
49593bfa | 5583 | |
16cdcec7 MX |
5584 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5585 | if (is_curr) | |
9cdda8d3 AV |
5586 | ctx->pos++; |
5587 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5588 | if (ret) |
5589 | goto nopos; | |
5590 | } | |
5591 | ||
49593bfa | 5592 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5593 | ctx->pos++; |
5594 | ||
5595 | /* | |
5596 | * Stop new entries from being returned after we return the last | |
5597 | * entry. | |
5598 | * | |
5599 | * New directory entries are assigned a strictly increasing | |
5600 | * offset. This means that new entries created during readdir | |
5601 | * are *guaranteed* to be seen in the future by that readdir. | |
5602 | * This has broken buggy programs which operate on names as | |
5603 | * they're returned by readdir. Until we re-use freed offsets | |
5604 | * we have this hack to stop new entries from being returned | |
5605 | * under the assumption that they'll never reach this huge | |
5606 | * offset. | |
5607 | * | |
5608 | * This is being careful not to overflow 32bit loff_t unless the | |
5609 | * last entry requires it because doing so has broken 32bit apps | |
5610 | * in the past. | |
5611 | */ | |
5612 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5613 | if (ctx->pos >= INT_MAX) | |
5614 | ctx->pos = LLONG_MAX; | |
5615 | else | |
5616 | ctx->pos = INT_MAX; | |
5617 | } | |
39279cc3 CM |
5618 | nopos: |
5619 | ret = 0; | |
5620 | err: | |
16cdcec7 MX |
5621 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5622 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5623 | btrfs_free_path(path); |
39279cc3 CM |
5624 | return ret; |
5625 | } | |
5626 | ||
a9185b41 | 5627 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5628 | { |
5629 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5630 | struct btrfs_trans_handle *trans; | |
5631 | int ret = 0; | |
0af3d00b | 5632 | bool nolock = false; |
39279cc3 | 5633 | |
72ac3c0d | 5634 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5635 | return 0; |
5636 | ||
83eea1f1 | 5637 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5638 | nolock = true; |
0af3d00b | 5639 | |
a9185b41 | 5640 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5641 | if (nolock) |
7a7eaa40 | 5642 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5643 | else |
7a7eaa40 | 5644 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5645 | if (IS_ERR(trans)) |
5646 | return PTR_ERR(trans); | |
a698d075 | 5647 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5648 | } |
5649 | return ret; | |
5650 | } | |
5651 | ||
5652 | /* | |
54aa1f4d | 5653 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5654 | * inode changes. But, it is most likely to find the inode in cache. |
5655 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5656 | * to keep or drop this code. | |
5657 | */ | |
48a3b636 | 5658 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5659 | { |
5660 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5661 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5662 | int ret; |
5663 | ||
72ac3c0d | 5664 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5665 | return 0; |
39279cc3 | 5666 | |
7a7eaa40 | 5667 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5668 | if (IS_ERR(trans)) |
5669 | return PTR_ERR(trans); | |
8929ecfa YZ |
5670 | |
5671 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5672 | if (ret && ret == -ENOSPC) { |
5673 | /* whoops, lets try again with the full transaction */ | |
5674 | btrfs_end_transaction(trans, root); | |
5675 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5676 | if (IS_ERR(trans)) |
5677 | return PTR_ERR(trans); | |
8929ecfa | 5678 | |
94b60442 | 5679 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5680 | } |
39279cc3 | 5681 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5682 | if (BTRFS_I(inode)->delayed_node) |
5683 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5684 | |
5685 | return ret; | |
5686 | } | |
5687 | ||
5688 | /* | |
5689 | * This is a copy of file_update_time. We need this so we can return error on | |
5690 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5691 | */ | |
e41f941a JB |
5692 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5693 | int flags) | |
22c44fe6 | 5694 | { |
2bc55652 AB |
5695 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5696 | ||
5697 | if (btrfs_root_readonly(root)) | |
5698 | return -EROFS; | |
5699 | ||
e41f941a | 5700 | if (flags & S_VERSION) |
22c44fe6 | 5701 | inode_inc_iversion(inode); |
e41f941a JB |
5702 | if (flags & S_CTIME) |
5703 | inode->i_ctime = *now; | |
5704 | if (flags & S_MTIME) | |
5705 | inode->i_mtime = *now; | |
5706 | if (flags & S_ATIME) | |
5707 | inode->i_atime = *now; | |
5708 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5709 | } |
5710 | ||
d352ac68 CM |
5711 | /* |
5712 | * find the highest existing sequence number in a directory | |
5713 | * and then set the in-memory index_cnt variable to reflect | |
5714 | * free sequence numbers | |
5715 | */ | |
aec7477b JB |
5716 | static int btrfs_set_inode_index_count(struct inode *inode) |
5717 | { | |
5718 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5719 | struct btrfs_key key, found_key; | |
5720 | struct btrfs_path *path; | |
5721 | struct extent_buffer *leaf; | |
5722 | int ret; | |
5723 | ||
33345d01 | 5724 | key.objectid = btrfs_ino(inode); |
962a298f | 5725 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5726 | key.offset = (u64)-1; |
5727 | ||
5728 | path = btrfs_alloc_path(); | |
5729 | if (!path) | |
5730 | return -ENOMEM; | |
5731 | ||
5732 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5733 | if (ret < 0) | |
5734 | goto out; | |
5735 | /* FIXME: we should be able to handle this */ | |
5736 | if (ret == 0) | |
5737 | goto out; | |
5738 | ret = 0; | |
5739 | ||
5740 | /* | |
5741 | * MAGIC NUMBER EXPLANATION: | |
5742 | * since we search a directory based on f_pos we have to start at 2 | |
5743 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5744 | * else has to start at 2 | |
5745 | */ | |
5746 | if (path->slots[0] == 0) { | |
5747 | BTRFS_I(inode)->index_cnt = 2; | |
5748 | goto out; | |
5749 | } | |
5750 | ||
5751 | path->slots[0]--; | |
5752 | ||
5753 | leaf = path->nodes[0]; | |
5754 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5755 | ||
33345d01 | 5756 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 5757 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
5758 | BTRFS_I(inode)->index_cnt = 2; |
5759 | goto out; | |
5760 | } | |
5761 | ||
5762 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5763 | out: | |
5764 | btrfs_free_path(path); | |
5765 | return ret; | |
5766 | } | |
5767 | ||
d352ac68 CM |
5768 | /* |
5769 | * helper to find a free sequence number in a given directory. This current | |
5770 | * code is very simple, later versions will do smarter things in the btree | |
5771 | */ | |
3de4586c | 5772 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5773 | { |
5774 | int ret = 0; | |
5775 | ||
5776 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5777 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5778 | if (ret) { | |
5779 | ret = btrfs_set_inode_index_count(dir); | |
5780 | if (ret) | |
5781 | return ret; | |
5782 | } | |
aec7477b JB |
5783 | } |
5784 | ||
00e4e6b3 | 5785 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5786 | BTRFS_I(dir)->index_cnt++; |
5787 | ||
5788 | return ret; | |
5789 | } | |
5790 | ||
b0d5d10f CM |
5791 | static int btrfs_insert_inode_locked(struct inode *inode) |
5792 | { | |
5793 | struct btrfs_iget_args args; | |
5794 | args.location = &BTRFS_I(inode)->location; | |
5795 | args.root = BTRFS_I(inode)->root; | |
5796 | ||
5797 | return insert_inode_locked4(inode, | |
5798 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
5799 | btrfs_find_actor, &args); | |
5800 | } | |
5801 | ||
39279cc3 CM |
5802 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5803 | struct btrfs_root *root, | |
aec7477b | 5804 | struct inode *dir, |
9c58309d | 5805 | const char *name, int name_len, |
175a4eb7 AV |
5806 | u64 ref_objectid, u64 objectid, |
5807 | umode_t mode, u64 *index) | |
39279cc3 CM |
5808 | { |
5809 | struct inode *inode; | |
5f39d397 | 5810 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5811 | struct btrfs_key *location; |
5f39d397 | 5812 | struct btrfs_path *path; |
9c58309d CM |
5813 | struct btrfs_inode_ref *ref; |
5814 | struct btrfs_key key[2]; | |
5815 | u32 sizes[2]; | |
ef3b9af5 | 5816 | int nitems = name ? 2 : 1; |
9c58309d | 5817 | unsigned long ptr; |
39279cc3 | 5818 | int ret; |
39279cc3 | 5819 | |
5f39d397 | 5820 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5821 | if (!path) |
5822 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5823 | |
39279cc3 | 5824 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5825 | if (!inode) { |
5826 | btrfs_free_path(path); | |
39279cc3 | 5827 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5828 | } |
39279cc3 | 5829 | |
5762b5c9 FM |
5830 | /* |
5831 | * O_TMPFILE, set link count to 0, so that after this point, | |
5832 | * we fill in an inode item with the correct link count. | |
5833 | */ | |
5834 | if (!name) | |
5835 | set_nlink(inode, 0); | |
5836 | ||
581bb050 LZ |
5837 | /* |
5838 | * we have to initialize this early, so we can reclaim the inode | |
5839 | * number if we fail afterwards in this function. | |
5840 | */ | |
5841 | inode->i_ino = objectid; | |
5842 | ||
ef3b9af5 | 5843 | if (dir && name) { |
1abe9b8a | 5844 | trace_btrfs_inode_request(dir); |
5845 | ||
3de4586c | 5846 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5847 | if (ret) { |
8fb27640 | 5848 | btrfs_free_path(path); |
09771430 | 5849 | iput(inode); |
aec7477b | 5850 | return ERR_PTR(ret); |
09771430 | 5851 | } |
ef3b9af5 FM |
5852 | } else if (dir) { |
5853 | *index = 0; | |
aec7477b JB |
5854 | } |
5855 | /* | |
5856 | * index_cnt is ignored for everything but a dir, | |
5857 | * btrfs_get_inode_index_count has an explanation for the magic | |
5858 | * number | |
5859 | */ | |
5860 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 5861 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 5862 | BTRFS_I(inode)->root = root; |
e02119d5 | 5863 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5864 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5865 | |
5dc562c5 JB |
5866 | /* |
5867 | * We could have gotten an inode number from somebody who was fsynced | |
5868 | * and then removed in this same transaction, so let's just set full | |
5869 | * sync since it will be a full sync anyway and this will blow away the | |
5870 | * old info in the log. | |
5871 | */ | |
5872 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5873 | ||
9c58309d | 5874 | key[0].objectid = objectid; |
962a298f | 5875 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
5876 | key[0].offset = 0; |
5877 | ||
9c58309d | 5878 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
5879 | |
5880 | if (name) { | |
5881 | /* | |
5882 | * Start new inodes with an inode_ref. This is slightly more | |
5883 | * efficient for small numbers of hard links since they will | |
5884 | * be packed into one item. Extended refs will kick in if we | |
5885 | * add more hard links than can fit in the ref item. | |
5886 | */ | |
5887 | key[1].objectid = objectid; | |
962a298f | 5888 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
5889 | key[1].offset = ref_objectid; |
5890 | ||
5891 | sizes[1] = name_len + sizeof(*ref); | |
5892 | } | |
9c58309d | 5893 | |
b0d5d10f CM |
5894 | location = &BTRFS_I(inode)->location; |
5895 | location->objectid = objectid; | |
5896 | location->offset = 0; | |
962a298f | 5897 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
5898 | |
5899 | ret = btrfs_insert_inode_locked(inode); | |
5900 | if (ret < 0) | |
5901 | goto fail; | |
5902 | ||
b9473439 | 5903 | path->leave_spinning = 1; |
ef3b9af5 | 5904 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 5905 | if (ret != 0) |
b0d5d10f | 5906 | goto fail_unlock; |
5f39d397 | 5907 | |
ecc11fab | 5908 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5909 | inode_set_bytes(inode, 0); |
9cc97d64 | 5910 | |
5911 | inode->i_mtime = CURRENT_TIME; | |
5912 | inode->i_atime = inode->i_mtime; | |
5913 | inode->i_ctime = inode->i_mtime; | |
5914 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
5915 | ||
5f39d397 CM |
5916 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5917 | struct btrfs_inode_item); | |
293f7e07 LZ |
5918 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5919 | sizeof(*inode_item)); | |
e02119d5 | 5920 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 5921 | |
ef3b9af5 FM |
5922 | if (name) { |
5923 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5924 | struct btrfs_inode_ref); | |
5925 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
5926 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
5927 | ptr = (unsigned long)(ref + 1); | |
5928 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5929 | } | |
9c58309d | 5930 | |
5f39d397 CM |
5931 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5932 | btrfs_free_path(path); | |
5933 | ||
6cbff00f CH |
5934 | btrfs_inherit_iflags(inode, dir); |
5935 | ||
569254b0 | 5936 | if (S_ISREG(mode)) { |
94272164 CM |
5937 | if (btrfs_test_opt(root, NODATASUM)) |
5938 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5939 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5940 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5941 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5942 | } |
5943 | ||
5d4f98a2 | 5944 | inode_tree_add(inode); |
1abe9b8a | 5945 | |
5946 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5947 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5948 | |
8ea05e3a AB |
5949 | btrfs_update_root_times(trans, root); |
5950 | ||
63541927 FDBM |
5951 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
5952 | if (ret) | |
5953 | btrfs_err(root->fs_info, | |
5954 | "error inheriting props for ino %llu (root %llu): %d", | |
5955 | btrfs_ino(inode), root->root_key.objectid, ret); | |
5956 | ||
39279cc3 | 5957 | return inode; |
b0d5d10f CM |
5958 | |
5959 | fail_unlock: | |
5960 | unlock_new_inode(inode); | |
5f39d397 | 5961 | fail: |
ef3b9af5 | 5962 | if (dir && name) |
aec7477b | 5963 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 5964 | btrfs_free_path(path); |
09771430 | 5965 | iput(inode); |
5f39d397 | 5966 | return ERR_PTR(ret); |
39279cc3 CM |
5967 | } |
5968 | ||
5969 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5970 | { | |
5971 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5972 | } | |
5973 | ||
d352ac68 CM |
5974 | /* |
5975 | * utility function to add 'inode' into 'parent_inode' with | |
5976 | * a give name and a given sequence number. | |
5977 | * if 'add_backref' is true, also insert a backref from the | |
5978 | * inode to the parent directory. | |
5979 | */ | |
e02119d5 CM |
5980 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5981 | struct inode *parent_inode, struct inode *inode, | |
5982 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5983 | { |
4df27c4d | 5984 | int ret = 0; |
39279cc3 | 5985 | struct btrfs_key key; |
e02119d5 | 5986 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5987 | u64 ino = btrfs_ino(inode); |
5988 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5989 | |
33345d01 | 5990 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5991 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5992 | } else { | |
33345d01 | 5993 | key.objectid = ino; |
962a298f | 5994 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
5995 | key.offset = 0; |
5996 | } | |
5997 | ||
33345d01 | 5998 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5999 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
6000 | key.objectid, root->root_key.objectid, | |
33345d01 | 6001 | parent_ino, index, name, name_len); |
4df27c4d | 6002 | } else if (add_backref) { |
33345d01 LZ |
6003 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6004 | parent_ino, index); | |
4df27c4d | 6005 | } |
39279cc3 | 6006 | |
79787eaa JM |
6007 | /* Nothing to clean up yet */ |
6008 | if (ret) | |
6009 | return ret; | |
4df27c4d | 6010 | |
79787eaa JM |
6011 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6012 | parent_inode, &key, | |
6013 | btrfs_inode_type(inode), index); | |
9c52057c | 6014 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6015 | goto fail_dir_item; |
6016 | else if (ret) { | |
6017 | btrfs_abort_transaction(trans, root, ret); | |
6018 | return ret; | |
39279cc3 | 6019 | } |
79787eaa JM |
6020 | |
6021 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6022 | name_len * 2); | |
0c4d2d95 | 6023 | inode_inc_iversion(parent_inode); |
79787eaa JM |
6024 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
6025 | ret = btrfs_update_inode(trans, root, parent_inode); | |
6026 | if (ret) | |
6027 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 6028 | return ret; |
fe66a05a CM |
6029 | |
6030 | fail_dir_item: | |
6031 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6032 | u64 local_index; | |
6033 | int err; | |
6034 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
6035 | key.objectid, root->root_key.objectid, | |
6036 | parent_ino, &local_index, name, name_len); | |
6037 | ||
6038 | } else if (add_backref) { | |
6039 | u64 local_index; | |
6040 | int err; | |
6041 | ||
6042 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6043 | ino, parent_ino, &local_index); | |
6044 | } | |
6045 | return ret; | |
39279cc3 CM |
6046 | } |
6047 | ||
6048 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6049 | struct inode *dir, struct dentry *dentry, |
6050 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6051 | { |
a1b075d2 JB |
6052 | int err = btrfs_add_link(trans, dir, inode, |
6053 | dentry->d_name.name, dentry->d_name.len, | |
6054 | backref, index); | |
39279cc3 CM |
6055 | if (err > 0) |
6056 | err = -EEXIST; | |
6057 | return err; | |
6058 | } | |
6059 | ||
618e21d5 | 6060 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6061 | umode_t mode, dev_t rdev) |
618e21d5 JB |
6062 | { |
6063 | struct btrfs_trans_handle *trans; | |
6064 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6065 | struct inode *inode = NULL; |
618e21d5 JB |
6066 | int err; |
6067 | int drop_inode = 0; | |
6068 | u64 objectid; | |
00e4e6b3 | 6069 | u64 index = 0; |
618e21d5 JB |
6070 | |
6071 | if (!new_valid_dev(rdev)) | |
6072 | return -EINVAL; | |
6073 | ||
9ed74f2d JB |
6074 | /* |
6075 | * 2 for inode item and ref | |
6076 | * 2 for dir items | |
6077 | * 1 for xattr if selinux is on | |
6078 | */ | |
a22285a6 YZ |
6079 | trans = btrfs_start_transaction(root, 5); |
6080 | if (IS_ERR(trans)) | |
6081 | return PTR_ERR(trans); | |
1832a6d5 | 6082 | |
581bb050 LZ |
6083 | err = btrfs_find_free_ino(root, &objectid); |
6084 | if (err) | |
6085 | goto out_unlock; | |
6086 | ||
aec7477b | 6087 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6088 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6089 | mode, &index); |
7cf96da3 TI |
6090 | if (IS_ERR(inode)) { |
6091 | err = PTR_ERR(inode); | |
618e21d5 | 6092 | goto out_unlock; |
7cf96da3 | 6093 | } |
618e21d5 | 6094 | |
ad19db71 CS |
6095 | /* |
6096 | * If the active LSM wants to access the inode during | |
6097 | * d_instantiate it needs these. Smack checks to see | |
6098 | * if the filesystem supports xattrs by looking at the | |
6099 | * ops vector. | |
6100 | */ | |
ad19db71 | 6101 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6102 | init_special_inode(inode, inode->i_mode, rdev); |
6103 | ||
6104 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6105 | if (err) |
b0d5d10f CM |
6106 | goto out_unlock_inode; |
6107 | ||
6108 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6109 | if (err) { | |
6110 | goto out_unlock_inode; | |
6111 | } else { | |
1b4ab1bb | 6112 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6113 | unlock_new_inode(inode); |
08c422c2 | 6114 | d_instantiate(dentry, inode); |
618e21d5 | 6115 | } |
b0d5d10f | 6116 | |
618e21d5 | 6117 | out_unlock: |
7ad85bb7 | 6118 | btrfs_end_transaction(trans, root); |
c581afc8 | 6119 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6120 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6121 | if (drop_inode) { |
6122 | inode_dec_link_count(inode); | |
6123 | iput(inode); | |
6124 | } | |
618e21d5 | 6125 | return err; |
b0d5d10f CM |
6126 | |
6127 | out_unlock_inode: | |
6128 | drop_inode = 1; | |
6129 | unlock_new_inode(inode); | |
6130 | goto out_unlock; | |
6131 | ||
618e21d5 JB |
6132 | } |
6133 | ||
39279cc3 | 6134 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6135 | umode_t mode, bool excl) |
39279cc3 CM |
6136 | { |
6137 | struct btrfs_trans_handle *trans; | |
6138 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6139 | struct inode *inode = NULL; |
43baa579 | 6140 | int drop_inode_on_err = 0; |
a22285a6 | 6141 | int err; |
39279cc3 | 6142 | u64 objectid; |
00e4e6b3 | 6143 | u64 index = 0; |
39279cc3 | 6144 | |
9ed74f2d JB |
6145 | /* |
6146 | * 2 for inode item and ref | |
6147 | * 2 for dir items | |
6148 | * 1 for xattr if selinux is on | |
6149 | */ | |
a22285a6 YZ |
6150 | trans = btrfs_start_transaction(root, 5); |
6151 | if (IS_ERR(trans)) | |
6152 | return PTR_ERR(trans); | |
9ed74f2d | 6153 | |
581bb050 LZ |
6154 | err = btrfs_find_free_ino(root, &objectid); |
6155 | if (err) | |
6156 | goto out_unlock; | |
6157 | ||
aec7477b | 6158 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6159 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6160 | mode, &index); |
7cf96da3 TI |
6161 | if (IS_ERR(inode)) { |
6162 | err = PTR_ERR(inode); | |
39279cc3 | 6163 | goto out_unlock; |
7cf96da3 | 6164 | } |
43baa579 | 6165 | drop_inode_on_err = 1; |
ad19db71 CS |
6166 | /* |
6167 | * If the active LSM wants to access the inode during | |
6168 | * d_instantiate it needs these. Smack checks to see | |
6169 | * if the filesystem supports xattrs by looking at the | |
6170 | * ops vector. | |
6171 | */ | |
6172 | inode->i_fop = &btrfs_file_operations; | |
6173 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
6174 | inode->i_mapping->a_ops = &btrfs_aops; |
6175 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
6176 | ||
6177 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6178 | if (err) | |
6179 | goto out_unlock_inode; | |
6180 | ||
6181 | err = btrfs_update_inode(trans, root, inode); | |
6182 | if (err) | |
6183 | goto out_unlock_inode; | |
ad19db71 | 6184 | |
a1b075d2 | 6185 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6186 | if (err) |
b0d5d10f | 6187 | goto out_unlock_inode; |
43baa579 | 6188 | |
43baa579 | 6189 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6190 | unlock_new_inode(inode); |
43baa579 FB |
6191 | d_instantiate(dentry, inode); |
6192 | ||
39279cc3 | 6193 | out_unlock: |
7ad85bb7 | 6194 | btrfs_end_transaction(trans, root); |
43baa579 | 6195 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6196 | inode_dec_link_count(inode); |
6197 | iput(inode); | |
6198 | } | |
c581afc8 | 6199 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6200 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6201 | return err; |
b0d5d10f CM |
6202 | |
6203 | out_unlock_inode: | |
6204 | unlock_new_inode(inode); | |
6205 | goto out_unlock; | |
6206 | ||
39279cc3 CM |
6207 | } |
6208 | ||
6209 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6210 | struct dentry *dentry) | |
6211 | { | |
6212 | struct btrfs_trans_handle *trans; | |
6213 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6214 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 6215 | u64 index; |
39279cc3 CM |
6216 | int err; |
6217 | int drop_inode = 0; | |
6218 | ||
4a8be425 TH |
6219 | /* do not allow sys_link's with other subvols of the same device */ |
6220 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6221 | return -EXDEV; |
4a8be425 | 6222 | |
f186373f | 6223 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6224 | return -EMLINK; |
4a8be425 | 6225 | |
3de4586c | 6226 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6227 | if (err) |
6228 | goto fail; | |
6229 | ||
a22285a6 | 6230 | /* |
7e6b6465 | 6231 | * 2 items for inode and inode ref |
a22285a6 | 6232 | * 2 items for dir items |
7e6b6465 | 6233 | * 1 item for parent inode |
a22285a6 | 6234 | */ |
7e6b6465 | 6235 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6236 | if (IS_ERR(trans)) { |
6237 | err = PTR_ERR(trans); | |
6238 | goto fail; | |
6239 | } | |
5f39d397 | 6240 | |
67de1176 MX |
6241 | /* There are several dir indexes for this inode, clear the cache. */ |
6242 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6243 | inc_nlink(inode); |
0c4d2d95 | 6244 | inode_inc_iversion(inode); |
3153495d | 6245 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 6246 | ihold(inode); |
e9976151 | 6247 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6248 | |
a1b075d2 | 6249 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6250 | |
a5719521 | 6251 | if (err) { |
54aa1f4d | 6252 | drop_inode = 1; |
a5719521 | 6253 | } else { |
10d9f309 | 6254 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6255 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6256 | if (err) |
6257 | goto fail; | |
ef3b9af5 FM |
6258 | if (inode->i_nlink == 1) { |
6259 | /* | |
6260 | * If new hard link count is 1, it's a file created | |
6261 | * with open(2) O_TMPFILE flag. | |
6262 | */ | |
6263 | err = btrfs_orphan_del(trans, inode); | |
6264 | if (err) | |
6265 | goto fail; | |
6266 | } | |
08c422c2 | 6267 | d_instantiate(dentry, inode); |
6a912213 | 6268 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6269 | } |
39279cc3 | 6270 | |
7ad85bb7 | 6271 | btrfs_end_transaction(trans, root); |
c581afc8 | 6272 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6273 | fail: |
39279cc3 CM |
6274 | if (drop_inode) { |
6275 | inode_dec_link_count(inode); | |
6276 | iput(inode); | |
6277 | } | |
b53d3f5d | 6278 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6279 | return err; |
6280 | } | |
6281 | ||
18bb1db3 | 6282 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6283 | { |
b9d86667 | 6284 | struct inode *inode = NULL; |
39279cc3 CM |
6285 | struct btrfs_trans_handle *trans; |
6286 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6287 | int err = 0; | |
6288 | int drop_on_err = 0; | |
b9d86667 | 6289 | u64 objectid = 0; |
00e4e6b3 | 6290 | u64 index = 0; |
39279cc3 | 6291 | |
9ed74f2d JB |
6292 | /* |
6293 | * 2 items for inode and ref | |
6294 | * 2 items for dir items | |
6295 | * 1 for xattr if selinux is on | |
6296 | */ | |
a22285a6 YZ |
6297 | trans = btrfs_start_transaction(root, 5); |
6298 | if (IS_ERR(trans)) | |
6299 | return PTR_ERR(trans); | |
39279cc3 | 6300 | |
581bb050 LZ |
6301 | err = btrfs_find_free_ino(root, &objectid); |
6302 | if (err) | |
6303 | goto out_fail; | |
6304 | ||
aec7477b | 6305 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6306 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6307 | S_IFDIR | mode, &index); |
39279cc3 CM |
6308 | if (IS_ERR(inode)) { |
6309 | err = PTR_ERR(inode); | |
6310 | goto out_fail; | |
6311 | } | |
5f39d397 | 6312 | |
39279cc3 | 6313 | drop_on_err = 1; |
b0d5d10f CM |
6314 | /* these must be set before we unlock the inode */ |
6315 | inode->i_op = &btrfs_dir_inode_operations; | |
6316 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6317 | |
2a7dba39 | 6318 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6319 | if (err) |
b0d5d10f | 6320 | goto out_fail_inode; |
39279cc3 | 6321 | |
dbe674a9 | 6322 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6323 | err = btrfs_update_inode(trans, root, inode); |
6324 | if (err) | |
b0d5d10f | 6325 | goto out_fail_inode; |
5f39d397 | 6326 | |
a1b075d2 JB |
6327 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6328 | dentry->d_name.len, 0, index); | |
39279cc3 | 6329 | if (err) |
b0d5d10f | 6330 | goto out_fail_inode; |
5f39d397 | 6331 | |
39279cc3 | 6332 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6333 | /* |
6334 | * mkdir is special. We're unlocking after we call d_instantiate | |
6335 | * to avoid a race with nfsd calling d_instantiate. | |
6336 | */ | |
6337 | unlock_new_inode(inode); | |
39279cc3 | 6338 | drop_on_err = 0; |
39279cc3 CM |
6339 | |
6340 | out_fail: | |
7ad85bb7 | 6341 | btrfs_end_transaction(trans, root); |
c7cfb8a5 WS |
6342 | if (drop_on_err) { |
6343 | inode_dec_link_count(inode); | |
39279cc3 | 6344 | iput(inode); |
c7cfb8a5 | 6345 | } |
c581afc8 | 6346 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6347 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6348 | return err; |
b0d5d10f CM |
6349 | |
6350 | out_fail_inode: | |
6351 | unlock_new_inode(inode); | |
6352 | goto out_fail; | |
39279cc3 CM |
6353 | } |
6354 | ||
e6c4efd8 QW |
6355 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6356 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6357 | { | |
6358 | struct rb_node *next; | |
6359 | ||
6360 | next = rb_next(&em->rb_node); | |
6361 | if (!next) | |
6362 | return NULL; | |
6363 | return container_of(next, struct extent_map, rb_node); | |
6364 | } | |
6365 | ||
6366 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6367 | { | |
6368 | struct rb_node *prev; | |
6369 | ||
6370 | prev = rb_prev(&em->rb_node); | |
6371 | if (!prev) | |
6372 | return NULL; | |
6373 | return container_of(prev, struct extent_map, rb_node); | |
6374 | } | |
6375 | ||
d352ac68 | 6376 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6377 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6378 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6379 | * the best fitted new extent into the tree. |
d352ac68 | 6380 | */ |
3b951516 CM |
6381 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6382 | struct extent_map *existing, | |
e6dcd2dc | 6383 | struct extent_map *em, |
51f395ad | 6384 | u64 map_start) |
3b951516 | 6385 | { |
e6c4efd8 QW |
6386 | struct extent_map *prev; |
6387 | struct extent_map *next; | |
6388 | u64 start; | |
6389 | u64 end; | |
3b951516 | 6390 | u64 start_diff; |
3b951516 | 6391 | |
e6dcd2dc | 6392 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6393 | |
6394 | if (existing->start > map_start) { | |
6395 | next = existing; | |
6396 | prev = prev_extent_map(next); | |
6397 | } else { | |
6398 | prev = existing; | |
6399 | next = next_extent_map(prev); | |
6400 | } | |
6401 | ||
6402 | start = prev ? extent_map_end(prev) : em->start; | |
6403 | start = max_t(u64, start, em->start); | |
6404 | end = next ? next->start : extent_map_end(em); | |
6405 | end = min_t(u64, end, extent_map_end(em)); | |
6406 | start_diff = start - em->start; | |
6407 | em->start = start; | |
6408 | em->len = end - start; | |
c8b97818 CM |
6409 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6410 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6411 | em->block_start += start_diff; |
c8b97818 CM |
6412 | em->block_len -= start_diff; |
6413 | } | |
09a2a8f9 | 6414 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6415 | } |
6416 | ||
c8b97818 CM |
6417 | static noinline int uncompress_inline(struct btrfs_path *path, |
6418 | struct inode *inode, struct page *page, | |
6419 | size_t pg_offset, u64 extent_offset, | |
6420 | struct btrfs_file_extent_item *item) | |
6421 | { | |
6422 | int ret; | |
6423 | struct extent_buffer *leaf = path->nodes[0]; | |
6424 | char *tmp; | |
6425 | size_t max_size; | |
6426 | unsigned long inline_size; | |
6427 | unsigned long ptr; | |
261507a0 | 6428 | int compress_type; |
c8b97818 CM |
6429 | |
6430 | WARN_ON(pg_offset != 0); | |
261507a0 | 6431 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6432 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6433 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6434 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6435 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6436 | if (!tmp) |
6437 | return -ENOMEM; | |
c8b97818 CM |
6438 | ptr = btrfs_file_extent_inline_start(item); |
6439 | ||
6440 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6441 | ||
5b050f04 | 6442 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
6443 | ret = btrfs_decompress(compress_type, tmp, page, |
6444 | extent_offset, inline_size, max_size); | |
c8b97818 | 6445 | kfree(tmp); |
166ae5a4 | 6446 | return ret; |
c8b97818 CM |
6447 | } |
6448 | ||
d352ac68 CM |
6449 | /* |
6450 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6451 | * the ugly parts come from merging extents from the disk with the in-ram |
6452 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6453 | * where the in-ram extents might be locked pending data=ordered completion. |
6454 | * | |
6455 | * This also copies inline extents directly into the page. | |
6456 | */ | |
d397712b | 6457 | |
a52d9a80 | 6458 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6459 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6460 | int create) |
6461 | { | |
6462 | int ret; | |
6463 | int err = 0; | |
a52d9a80 CM |
6464 | u64 extent_start = 0; |
6465 | u64 extent_end = 0; | |
33345d01 | 6466 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6467 | u32 found_type; |
f421950f | 6468 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6469 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6470 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6471 | struct extent_buffer *leaf; |
6472 | struct btrfs_key found_key; | |
a52d9a80 CM |
6473 | struct extent_map *em = NULL; |
6474 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6475 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6476 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6477 | const bool new_inline = !page || create; |
a52d9a80 | 6478 | |
a52d9a80 | 6479 | again: |
890871be | 6480 | read_lock(&em_tree->lock); |
d1310b2e | 6481 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6482 | if (em) |
6483 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6484 | read_unlock(&em_tree->lock); |
d1310b2e | 6485 | |
a52d9a80 | 6486 | if (em) { |
e1c4b745 CM |
6487 | if (em->start > start || em->start + em->len <= start) |
6488 | free_extent_map(em); | |
6489 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6490 | free_extent_map(em); |
6491 | else | |
6492 | goto out; | |
a52d9a80 | 6493 | } |
172ddd60 | 6494 | em = alloc_extent_map(); |
a52d9a80 | 6495 | if (!em) { |
d1310b2e CM |
6496 | err = -ENOMEM; |
6497 | goto out; | |
a52d9a80 | 6498 | } |
e6dcd2dc | 6499 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6500 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6501 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6502 | em->len = (u64)-1; |
c8b97818 | 6503 | em->block_len = (u64)-1; |
f421950f CM |
6504 | |
6505 | if (!path) { | |
6506 | path = btrfs_alloc_path(); | |
026fd317 JB |
6507 | if (!path) { |
6508 | err = -ENOMEM; | |
6509 | goto out; | |
6510 | } | |
6511 | /* | |
6512 | * Chances are we'll be called again, so go ahead and do | |
6513 | * readahead | |
6514 | */ | |
6515 | path->reada = 1; | |
f421950f CM |
6516 | } |
6517 | ||
179e29e4 CM |
6518 | ret = btrfs_lookup_file_extent(trans, root, path, |
6519 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6520 | if (ret < 0) { |
6521 | err = ret; | |
6522 | goto out; | |
6523 | } | |
6524 | ||
6525 | if (ret != 0) { | |
6526 | if (path->slots[0] == 0) | |
6527 | goto not_found; | |
6528 | path->slots[0]--; | |
6529 | } | |
6530 | ||
5f39d397 CM |
6531 | leaf = path->nodes[0]; |
6532 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6533 | struct btrfs_file_extent_item); |
a52d9a80 | 6534 | /* are we inside the extent that was found? */ |
5f39d397 | 6535 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6536 | found_type = found_key.type; |
5f39d397 | 6537 | if (found_key.objectid != objectid || |
a52d9a80 | 6538 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6539 | /* |
6540 | * If we backup past the first extent we want to move forward | |
6541 | * and see if there is an extent in front of us, otherwise we'll | |
6542 | * say there is a hole for our whole search range which can | |
6543 | * cause problems. | |
6544 | */ | |
6545 | extent_end = start; | |
6546 | goto next; | |
a52d9a80 CM |
6547 | } |
6548 | ||
5f39d397 CM |
6549 | found_type = btrfs_file_extent_type(leaf, item); |
6550 | extent_start = found_key.offset; | |
d899e052 YZ |
6551 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6552 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6553 | extent_end = extent_start + |
db94535d | 6554 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6555 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6556 | size_t size; | |
514ac8ad | 6557 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6558 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6559 | } |
25a50341 | 6560 | next: |
9036c102 YZ |
6561 | if (start >= extent_end) { |
6562 | path->slots[0]++; | |
6563 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6564 | ret = btrfs_next_leaf(root, path); | |
6565 | if (ret < 0) { | |
6566 | err = ret; | |
6567 | goto out; | |
a52d9a80 | 6568 | } |
9036c102 YZ |
6569 | if (ret > 0) |
6570 | goto not_found; | |
6571 | leaf = path->nodes[0]; | |
a52d9a80 | 6572 | } |
9036c102 YZ |
6573 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6574 | if (found_key.objectid != objectid || | |
6575 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6576 | goto not_found; | |
6577 | if (start + len <= found_key.offset) | |
6578 | goto not_found; | |
e2eca69d WS |
6579 | if (start > found_key.offset) |
6580 | goto next; | |
9036c102 | 6581 | em->start = start; |
70c8a91c | 6582 | em->orig_start = start; |
9036c102 YZ |
6583 | em->len = found_key.offset - start; |
6584 | goto not_found_em; | |
6585 | } | |
6586 | ||
7ffbb598 FM |
6587 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6588 | ||
d899e052 YZ |
6589 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6590 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6591 | goto insert; |
6592 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6593 | unsigned long ptr; |
a52d9a80 | 6594 | char *map; |
3326d1b0 CM |
6595 | size_t size; |
6596 | size_t extent_offset; | |
6597 | size_t copy_size; | |
a52d9a80 | 6598 | |
7ffbb598 | 6599 | if (new_inline) |
689f9346 | 6600 | goto out; |
5f39d397 | 6601 | |
514ac8ad | 6602 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6603 | extent_offset = page_offset(page) + pg_offset - extent_start; |
70dec807 | 6604 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6605 | size - extent_offset); |
3326d1b0 | 6606 | em->start = extent_start + extent_offset; |
fda2832f | 6607 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6608 | em->orig_block_len = em->len; |
70c8a91c | 6609 | em->orig_start = em->start; |
689f9346 | 6610 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6611 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6612 | if (btrfs_file_extent_compression(leaf, item) != |
6613 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6614 | ret = uncompress_inline(path, inode, page, |
6615 | pg_offset, | |
6616 | extent_offset, item); | |
166ae5a4 ZB |
6617 | if (ret) { |
6618 | err = ret; | |
6619 | goto out; | |
6620 | } | |
c8b97818 CM |
6621 | } else { |
6622 | map = kmap(page); | |
6623 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6624 | copy_size); | |
93c82d57 CM |
6625 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6626 | memset(map + pg_offset + copy_size, 0, | |
6627 | PAGE_CACHE_SIZE - pg_offset - | |
6628 | copy_size); | |
6629 | } | |
c8b97818 CM |
6630 | kunmap(page); |
6631 | } | |
179e29e4 CM |
6632 | flush_dcache_page(page); |
6633 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6634 | BUG(); |
179e29e4 CM |
6635 | if (!trans) { |
6636 | kunmap(page); | |
6637 | free_extent_map(em); | |
6638 | em = NULL; | |
ff5714cc | 6639 | |
b3b4aa74 | 6640 | btrfs_release_path(path); |
7a7eaa40 | 6641 | trans = btrfs_join_transaction(root); |
ff5714cc | 6642 | |
3612b495 TI |
6643 | if (IS_ERR(trans)) |
6644 | return ERR_CAST(trans); | |
179e29e4 CM |
6645 | goto again; |
6646 | } | |
c8b97818 | 6647 | map = kmap(page); |
70dec807 | 6648 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6649 | copy_size); |
c8b97818 | 6650 | kunmap(page); |
179e29e4 | 6651 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6652 | } |
d1310b2e | 6653 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6654 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6655 | goto insert; |
a52d9a80 CM |
6656 | } |
6657 | not_found: | |
6658 | em->start = start; | |
70c8a91c | 6659 | em->orig_start = start; |
d1310b2e | 6660 | em->len = len; |
a52d9a80 | 6661 | not_found_em: |
5f39d397 | 6662 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6663 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6664 | insert: |
b3b4aa74 | 6665 | btrfs_release_path(path); |
d1310b2e | 6666 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6667 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6668 | em->start, em->len, start, len); |
a52d9a80 CM |
6669 | err = -EIO; |
6670 | goto out; | |
6671 | } | |
d1310b2e CM |
6672 | |
6673 | err = 0; | |
890871be | 6674 | write_lock(&em_tree->lock); |
09a2a8f9 | 6675 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6676 | /* it is possible that someone inserted the extent into the tree |
6677 | * while we had the lock dropped. It is also possible that | |
6678 | * an overlapping map exists in the tree | |
6679 | */ | |
a52d9a80 | 6680 | if (ret == -EEXIST) { |
3b951516 | 6681 | struct extent_map *existing; |
e6dcd2dc CM |
6682 | |
6683 | ret = 0; | |
6684 | ||
e6c4efd8 QW |
6685 | existing = search_extent_mapping(em_tree, start, len); |
6686 | /* | |
6687 | * existing will always be non-NULL, since there must be | |
6688 | * extent causing the -EEXIST. | |
6689 | */ | |
6690 | if (start >= extent_map_end(existing) || | |
32be3a1a | 6691 | start <= existing->start) { |
e6c4efd8 QW |
6692 | /* |
6693 | * The existing extent map is the one nearest to | |
6694 | * the [start, start + len) range which overlaps | |
6695 | */ | |
6696 | err = merge_extent_mapping(em_tree, existing, | |
6697 | em, start); | |
e1c4b745 | 6698 | free_extent_map(existing); |
e6c4efd8 | 6699 | if (err) { |
3b951516 CM |
6700 | free_extent_map(em); |
6701 | em = NULL; | |
6702 | } | |
6703 | } else { | |
6704 | free_extent_map(em); | |
6705 | em = existing; | |
e6dcd2dc | 6706 | err = 0; |
a52d9a80 | 6707 | } |
a52d9a80 | 6708 | } |
890871be | 6709 | write_unlock(&em_tree->lock); |
a52d9a80 | 6710 | out: |
1abe9b8a | 6711 | |
4cd8587c | 6712 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6713 | |
f421950f CM |
6714 | if (path) |
6715 | btrfs_free_path(path); | |
a52d9a80 CM |
6716 | if (trans) { |
6717 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6718 | if (!err) |
a52d9a80 CM |
6719 | err = ret; |
6720 | } | |
a52d9a80 CM |
6721 | if (err) { |
6722 | free_extent_map(em); | |
a52d9a80 CM |
6723 | return ERR_PTR(err); |
6724 | } | |
79787eaa | 6725 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6726 | return em; |
6727 | } | |
6728 | ||
ec29ed5b CM |
6729 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6730 | size_t pg_offset, u64 start, u64 len, | |
6731 | int create) | |
6732 | { | |
6733 | struct extent_map *em; | |
6734 | struct extent_map *hole_em = NULL; | |
6735 | u64 range_start = start; | |
6736 | u64 end; | |
6737 | u64 found; | |
6738 | u64 found_end; | |
6739 | int err = 0; | |
6740 | ||
6741 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6742 | if (IS_ERR(em)) | |
6743 | return em; | |
6744 | if (em) { | |
6745 | /* | |
f9e4fb53 LB |
6746 | * if our em maps to |
6747 | * - a hole or | |
6748 | * - a pre-alloc extent, | |
6749 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6750 | */ |
f9e4fb53 LB |
6751 | if (em->block_start != EXTENT_MAP_HOLE && |
6752 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6753 | return em; |
6754 | else | |
6755 | hole_em = em; | |
6756 | } | |
6757 | ||
6758 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6759 | end = start + len; | |
6760 | if (end < start) | |
6761 | end = (u64)-1; | |
6762 | else | |
6763 | end -= 1; | |
6764 | ||
6765 | em = NULL; | |
6766 | ||
6767 | /* ok, we didn't find anything, lets look for delalloc */ | |
6768 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6769 | end, len, EXTENT_DELALLOC, 1); | |
6770 | found_end = range_start + found; | |
6771 | if (found_end < range_start) | |
6772 | found_end = (u64)-1; | |
6773 | ||
6774 | /* | |
6775 | * we didn't find anything useful, return | |
6776 | * the original results from get_extent() | |
6777 | */ | |
6778 | if (range_start > end || found_end <= start) { | |
6779 | em = hole_em; | |
6780 | hole_em = NULL; | |
6781 | goto out; | |
6782 | } | |
6783 | ||
6784 | /* adjust the range_start to make sure it doesn't | |
6785 | * go backwards from the start they passed in | |
6786 | */ | |
67871254 | 6787 | range_start = max(start, range_start); |
ec29ed5b CM |
6788 | found = found_end - range_start; |
6789 | ||
6790 | if (found > 0) { | |
6791 | u64 hole_start = start; | |
6792 | u64 hole_len = len; | |
6793 | ||
172ddd60 | 6794 | em = alloc_extent_map(); |
ec29ed5b CM |
6795 | if (!em) { |
6796 | err = -ENOMEM; | |
6797 | goto out; | |
6798 | } | |
6799 | /* | |
6800 | * when btrfs_get_extent can't find anything it | |
6801 | * returns one huge hole | |
6802 | * | |
6803 | * make sure what it found really fits our range, and | |
6804 | * adjust to make sure it is based on the start from | |
6805 | * the caller | |
6806 | */ | |
6807 | if (hole_em) { | |
6808 | u64 calc_end = extent_map_end(hole_em); | |
6809 | ||
6810 | if (calc_end <= start || (hole_em->start > end)) { | |
6811 | free_extent_map(hole_em); | |
6812 | hole_em = NULL; | |
6813 | } else { | |
6814 | hole_start = max(hole_em->start, start); | |
6815 | hole_len = calc_end - hole_start; | |
6816 | } | |
6817 | } | |
6818 | em->bdev = NULL; | |
6819 | if (hole_em && range_start > hole_start) { | |
6820 | /* our hole starts before our delalloc, so we | |
6821 | * have to return just the parts of the hole | |
6822 | * that go until the delalloc starts | |
6823 | */ | |
6824 | em->len = min(hole_len, | |
6825 | range_start - hole_start); | |
6826 | em->start = hole_start; | |
6827 | em->orig_start = hole_start; | |
6828 | /* | |
6829 | * don't adjust block start at all, | |
6830 | * it is fixed at EXTENT_MAP_HOLE | |
6831 | */ | |
6832 | em->block_start = hole_em->block_start; | |
6833 | em->block_len = hole_len; | |
f9e4fb53 LB |
6834 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6835 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6836 | } else { |
6837 | em->start = range_start; | |
6838 | em->len = found; | |
6839 | em->orig_start = range_start; | |
6840 | em->block_start = EXTENT_MAP_DELALLOC; | |
6841 | em->block_len = found; | |
6842 | } | |
6843 | } else if (hole_em) { | |
6844 | return hole_em; | |
6845 | } | |
6846 | out: | |
6847 | ||
6848 | free_extent_map(hole_em); | |
6849 | if (err) { | |
6850 | free_extent_map(em); | |
6851 | return ERR_PTR(err); | |
6852 | } | |
6853 | return em; | |
6854 | } | |
6855 | ||
4b46fce2 JB |
6856 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6857 | u64 start, u64 len) | |
6858 | { | |
6859 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 6860 | struct extent_map *em; |
4b46fce2 JB |
6861 | struct btrfs_key ins; |
6862 | u64 alloc_hint; | |
6863 | int ret; | |
4b46fce2 | 6864 | |
4b46fce2 | 6865 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 6866 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 6867 | alloc_hint, &ins, 1, 1); |
00361589 JB |
6868 | if (ret) |
6869 | return ERR_PTR(ret); | |
4b46fce2 | 6870 | |
70c8a91c | 6871 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6872 | ins.offset, ins.offset, ins.offset, 0); |
00361589 | 6873 | if (IS_ERR(em)) { |
e570fd27 | 6874 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6875 | return em; |
6876 | } | |
4b46fce2 JB |
6877 | |
6878 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6879 | ins.offset, ins.offset, 0); | |
6880 | if (ret) { | |
e570fd27 | 6881 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
6882 | free_extent_map(em); |
6883 | return ERR_PTR(ret); | |
4b46fce2 | 6884 | } |
00361589 | 6885 | |
4b46fce2 JB |
6886 | return em; |
6887 | } | |
6888 | ||
46bfbb5c CM |
6889 | /* |
6890 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6891 | * block must be cow'd | |
6892 | */ | |
00361589 | 6893 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6894 | u64 *orig_start, u64 *orig_block_len, |
6895 | u64 *ram_bytes) | |
46bfbb5c | 6896 | { |
00361589 | 6897 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
6898 | struct btrfs_path *path; |
6899 | int ret; | |
6900 | struct extent_buffer *leaf; | |
6901 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 6902 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
6903 | struct btrfs_file_extent_item *fi; |
6904 | struct btrfs_key key; | |
6905 | u64 disk_bytenr; | |
6906 | u64 backref_offset; | |
6907 | u64 extent_end; | |
6908 | u64 num_bytes; | |
6909 | int slot; | |
6910 | int found_type; | |
7ee9e440 | 6911 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 6912 | |
46bfbb5c CM |
6913 | path = btrfs_alloc_path(); |
6914 | if (!path) | |
6915 | return -ENOMEM; | |
6916 | ||
00361589 | 6917 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6918 | offset, 0); |
6919 | if (ret < 0) | |
6920 | goto out; | |
6921 | ||
6922 | slot = path->slots[0]; | |
6923 | if (ret == 1) { | |
6924 | if (slot == 0) { | |
6925 | /* can't find the item, must cow */ | |
6926 | ret = 0; | |
6927 | goto out; | |
6928 | } | |
6929 | slot--; | |
6930 | } | |
6931 | ret = 0; | |
6932 | leaf = path->nodes[0]; | |
6933 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6934 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6935 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6936 | /* not our file or wrong item type, must cow */ | |
6937 | goto out; | |
6938 | } | |
6939 | ||
6940 | if (key.offset > offset) { | |
6941 | /* Wrong offset, must cow */ | |
6942 | goto out; | |
6943 | } | |
6944 | ||
6945 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6946 | found_type = btrfs_file_extent_type(leaf, fi); | |
6947 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6948 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6949 | /* not a regular extent, must cow */ | |
6950 | goto out; | |
6951 | } | |
7ee9e440 JB |
6952 | |
6953 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6954 | goto out; | |
6955 | ||
e77751aa MX |
6956 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
6957 | if (extent_end <= offset) | |
6958 | goto out; | |
6959 | ||
46bfbb5c | 6960 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6961 | if (disk_bytenr == 0) |
6962 | goto out; | |
6963 | ||
6964 | if (btrfs_file_extent_compression(leaf, fi) || | |
6965 | btrfs_file_extent_encryption(leaf, fi) || | |
6966 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6967 | goto out; | |
6968 | ||
46bfbb5c CM |
6969 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6970 | ||
7ee9e440 JB |
6971 | if (orig_start) { |
6972 | *orig_start = key.offset - backref_offset; | |
6973 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6974 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6975 | } | |
eb384b55 | 6976 | |
46bfbb5c CM |
6977 | if (btrfs_extent_readonly(root, disk_bytenr)) |
6978 | goto out; | |
7b2b7085 MX |
6979 | |
6980 | num_bytes = min(offset + *len, extent_end) - offset; | |
6981 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6982 | u64 range_end; | |
6983 | ||
6984 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
6985 | ret = test_range_bit(io_tree, offset, range_end, | |
6986 | EXTENT_DELALLOC, 0, NULL); | |
6987 | if (ret) { | |
6988 | ret = -EAGAIN; | |
6989 | goto out; | |
6990 | } | |
6991 | } | |
6992 | ||
1bda19eb | 6993 | btrfs_release_path(path); |
46bfbb5c CM |
6994 | |
6995 | /* | |
6996 | * look for other files referencing this extent, if we | |
6997 | * find any we must cow | |
6998 | */ | |
00361589 JB |
6999 | trans = btrfs_join_transaction(root); |
7000 | if (IS_ERR(trans)) { | |
7001 | ret = 0; | |
46bfbb5c | 7002 | goto out; |
00361589 JB |
7003 | } |
7004 | ||
7005 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
7006 | key.offset - backref_offset, disk_bytenr); | |
7007 | btrfs_end_transaction(trans, root); | |
7008 | if (ret) { | |
7009 | ret = 0; | |
7010 | goto out; | |
7011 | } | |
46bfbb5c CM |
7012 | |
7013 | /* | |
7014 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7015 | * in this extent we are about to write. If there | |
7016 | * are any csums in that range we have to cow in order | |
7017 | * to keep the csums correct | |
7018 | */ | |
7019 | disk_bytenr += backref_offset; | |
7020 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
7021 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
7022 | goto out; | |
7023 | /* | |
7024 | * all of the above have passed, it is safe to overwrite this extent | |
7025 | * without cow | |
7026 | */ | |
eb384b55 | 7027 | *len = num_bytes; |
46bfbb5c CM |
7028 | ret = 1; |
7029 | out: | |
7030 | btrfs_free_path(path); | |
7031 | return ret; | |
7032 | } | |
7033 | ||
fc4adbff AG |
7034 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7035 | { | |
7036 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7037 | int found = false; | |
7038 | void **pagep = NULL; | |
7039 | struct page *page = NULL; | |
7040 | int start_idx; | |
7041 | int end_idx; | |
7042 | ||
7043 | start_idx = start >> PAGE_CACHE_SHIFT; | |
7044 | ||
7045 | /* | |
7046 | * end is the last byte in the last page. end == start is legal | |
7047 | */ | |
7048 | end_idx = end >> PAGE_CACHE_SHIFT; | |
7049 | ||
7050 | rcu_read_lock(); | |
7051 | ||
7052 | /* Most of the code in this while loop is lifted from | |
7053 | * find_get_page. It's been modified to begin searching from a | |
7054 | * page and return just the first page found in that range. If the | |
7055 | * found idx is less than or equal to the end idx then we know that | |
7056 | * a page exists. If no pages are found or if those pages are | |
7057 | * outside of the range then we're fine (yay!) */ | |
7058 | while (page == NULL && | |
7059 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7060 | page = radix_tree_deref_slot(pagep); | |
7061 | if (unlikely(!page)) | |
7062 | break; | |
7063 | ||
7064 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7065 | if (radix_tree_deref_retry(page)) { |
7066 | page = NULL; | |
fc4adbff | 7067 | continue; |
809f9016 | 7068 | } |
fc4adbff AG |
7069 | /* |
7070 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7071 | * here as an exceptional entry: so return it without | |
7072 | * attempting to raise page count. | |
7073 | */ | |
6fdef6d4 | 7074 | page = NULL; |
fc4adbff AG |
7075 | break; /* TODO: Is this relevant for this use case? */ |
7076 | } | |
7077 | ||
91405151 FM |
7078 | if (!page_cache_get_speculative(page)) { |
7079 | page = NULL; | |
fc4adbff | 7080 | continue; |
91405151 | 7081 | } |
fc4adbff AG |
7082 | |
7083 | /* | |
7084 | * Has the page moved? | |
7085 | * This is part of the lockless pagecache protocol. See | |
7086 | * include/linux/pagemap.h for details. | |
7087 | */ | |
7088 | if (unlikely(page != *pagep)) { | |
7089 | page_cache_release(page); | |
7090 | page = NULL; | |
7091 | } | |
7092 | } | |
7093 | ||
7094 | if (page) { | |
7095 | if (page->index <= end_idx) | |
7096 | found = true; | |
7097 | page_cache_release(page); | |
7098 | } | |
7099 | ||
7100 | rcu_read_unlock(); | |
7101 | return found; | |
7102 | } | |
7103 | ||
eb838e73 JB |
7104 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7105 | struct extent_state **cached_state, int writing) | |
7106 | { | |
7107 | struct btrfs_ordered_extent *ordered; | |
7108 | int ret = 0; | |
7109 | ||
7110 | while (1) { | |
7111 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7112 | 0, cached_state); | |
7113 | /* | |
7114 | * We're concerned with the entire range that we're going to be | |
7115 | * doing DIO to, so we need to make sure theres no ordered | |
7116 | * extents in this range. | |
7117 | */ | |
7118 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7119 | lockend - lockstart + 1); | |
7120 | ||
7121 | /* | |
7122 | * We need to make sure there are no buffered pages in this | |
7123 | * range either, we could have raced between the invalidate in | |
7124 | * generic_file_direct_write and locking the extent. The | |
7125 | * invalidate needs to happen so that reads after a write do not | |
7126 | * get stale data. | |
7127 | */ | |
fc4adbff AG |
7128 | if (!ordered && |
7129 | (!writing || | |
7130 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7131 | break; |
7132 | ||
7133 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7134 | cached_state, GFP_NOFS); | |
7135 | ||
7136 | if (ordered) { | |
7137 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7138 | btrfs_put_ordered_extent(ordered); | |
7139 | } else { | |
7140 | /* Screw you mmap */ | |
728404da | 7141 | ret = btrfs_fdatawrite_range(inode, lockstart, lockend); |
075bdbdb FM |
7142 | if (ret) |
7143 | break; | |
7144 | ret = filemap_fdatawait_range(inode->i_mapping, | |
7145 | lockstart, | |
7146 | lockend); | |
eb838e73 JB |
7147 | if (ret) |
7148 | break; | |
7149 | ||
7150 | /* | |
7151 | * If we found a page that couldn't be invalidated just | |
7152 | * fall back to buffered. | |
7153 | */ | |
7154 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
7155 | lockstart >> PAGE_CACHE_SHIFT, | |
7156 | lockend >> PAGE_CACHE_SHIFT); | |
7157 | if (ret) | |
7158 | break; | |
7159 | } | |
7160 | ||
7161 | cond_resched(); | |
7162 | } | |
7163 | ||
7164 | return ret; | |
7165 | } | |
7166 | ||
69ffb543 JB |
7167 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7168 | u64 len, u64 orig_start, | |
7169 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7170 | u64 orig_block_len, u64 ram_bytes, |
7171 | int type) | |
69ffb543 JB |
7172 | { |
7173 | struct extent_map_tree *em_tree; | |
7174 | struct extent_map *em; | |
7175 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7176 | int ret; | |
7177 | ||
7178 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7179 | em = alloc_extent_map(); | |
7180 | if (!em) | |
7181 | return ERR_PTR(-ENOMEM); | |
7182 | ||
7183 | em->start = start; | |
7184 | em->orig_start = orig_start; | |
2ab28f32 JB |
7185 | em->mod_start = start; |
7186 | em->mod_len = len; | |
69ffb543 JB |
7187 | em->len = len; |
7188 | em->block_len = block_len; | |
7189 | em->block_start = block_start; | |
7190 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7191 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7192 | em->ram_bytes = ram_bytes; |
70c8a91c | 7193 | em->generation = -1; |
69ffb543 JB |
7194 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7195 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7196 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7197 | |
7198 | do { | |
7199 | btrfs_drop_extent_cache(inode, em->start, | |
7200 | em->start + em->len - 1, 0); | |
7201 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7202 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7203 | write_unlock(&em_tree->lock); |
7204 | } while (ret == -EEXIST); | |
7205 | ||
7206 | if (ret) { | |
7207 | free_extent_map(em); | |
7208 | return ERR_PTR(ret); | |
7209 | } | |
7210 | ||
7211 | return em; | |
7212 | } | |
7213 | ||
7214 | ||
4b46fce2 JB |
7215 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7216 | struct buffer_head *bh_result, int create) | |
7217 | { | |
7218 | struct extent_map *em; | |
7219 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7220 | struct extent_state *cached_state = NULL; |
4b46fce2 | 7221 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7222 | u64 lockstart, lockend; |
4b46fce2 | 7223 | u64 len = bh_result->b_size; |
3e05bde8 | 7224 | u64 orig_len = len; |
eb838e73 | 7225 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7226 | int ret = 0; |
eb838e73 | 7227 | |
172a5049 | 7228 | if (create) |
3266789f | 7229 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7230 | else |
c329861d | 7231 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7232 | |
c329861d JB |
7233 | lockstart = start; |
7234 | lockend = start + len - 1; | |
7235 | ||
eb838e73 JB |
7236 | /* |
7237 | * If this errors out it's because we couldn't invalidate pagecache for | |
7238 | * this range and we need to fallback to buffered. | |
7239 | */ | |
7240 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
7241 | return -ENOTBLK; | |
7242 | ||
4b46fce2 | 7243 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7244 | if (IS_ERR(em)) { |
7245 | ret = PTR_ERR(em); | |
7246 | goto unlock_err; | |
7247 | } | |
4b46fce2 JB |
7248 | |
7249 | /* | |
7250 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7251 | * io. INLINE is special, and we could probably kludge it in here, but | |
7252 | * it's still buffered so for safety lets just fall back to the generic | |
7253 | * buffered path. | |
7254 | * | |
7255 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7256 | * decompress it, so there will be buffering required no matter what we | |
7257 | * do, so go ahead and fallback to buffered. | |
7258 | * | |
7259 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
7260 | * to buffered IO. Don't blame me, this is the price we pay for using | |
7261 | * the generic code. | |
7262 | */ | |
7263 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7264 | em->block_start == EXTENT_MAP_INLINE) { | |
7265 | free_extent_map(em); | |
eb838e73 JB |
7266 | ret = -ENOTBLK; |
7267 | goto unlock_err; | |
4b46fce2 JB |
7268 | } |
7269 | ||
7270 | /* Just a good old fashioned hole, return */ | |
7271 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7272 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7273 | free_extent_map(em); | |
eb838e73 | 7274 | goto unlock_err; |
4b46fce2 JB |
7275 | } |
7276 | ||
7277 | /* | |
7278 | * We don't allocate a new extent in the following cases | |
7279 | * | |
7280 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7281 | * existing extent. | |
7282 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7283 | * just use the extent. | |
7284 | * | |
7285 | */ | |
46bfbb5c | 7286 | if (!create) { |
eb838e73 JB |
7287 | len = min(len, em->len - (start - em->start)); |
7288 | lockstart = start + len; | |
7289 | goto unlock; | |
46bfbb5c | 7290 | } |
4b46fce2 JB |
7291 | |
7292 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7293 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7294 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7295 | int type; |
eb384b55 | 7296 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7297 | |
7298 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7299 | type = BTRFS_ORDERED_PREALLOC; | |
7300 | else | |
7301 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7302 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7303 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7304 | |
00361589 | 7305 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 7306 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
7307 | if (type == BTRFS_ORDERED_PREALLOC) { |
7308 | free_extent_map(em); | |
7309 | em = create_pinned_em(inode, start, len, | |
7310 | orig_start, | |
b4939680 | 7311 | block_start, len, |
cc95bef6 JB |
7312 | orig_block_len, |
7313 | ram_bytes, type); | |
555e1286 FM |
7314 | if (IS_ERR(em)) { |
7315 | ret = PTR_ERR(em); | |
69ffb543 | 7316 | goto unlock_err; |
555e1286 | 7317 | } |
69ffb543 JB |
7318 | } |
7319 | ||
46bfbb5c CM |
7320 | ret = btrfs_add_ordered_extent_dio(inode, start, |
7321 | block_start, len, len, type); | |
46bfbb5c CM |
7322 | if (ret) { |
7323 | free_extent_map(em); | |
eb838e73 | 7324 | goto unlock_err; |
46bfbb5c CM |
7325 | } |
7326 | goto unlock; | |
4b46fce2 | 7327 | } |
4b46fce2 | 7328 | } |
00361589 | 7329 | |
46bfbb5c CM |
7330 | /* |
7331 | * this will cow the extent, reset the len in case we changed | |
7332 | * it above | |
7333 | */ | |
7334 | len = bh_result->b_size; | |
70c8a91c JB |
7335 | free_extent_map(em); |
7336 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7337 | if (IS_ERR(em)) { |
7338 | ret = PTR_ERR(em); | |
7339 | goto unlock_err; | |
7340 | } | |
46bfbb5c CM |
7341 | len = min(len, em->len - (start - em->start)); |
7342 | unlock: | |
4b46fce2 JB |
7343 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7344 | inode->i_blkbits; | |
46bfbb5c | 7345 | bh_result->b_size = len; |
4b46fce2 JB |
7346 | bh_result->b_bdev = em->bdev; |
7347 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7348 | if (create) { |
7349 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7350 | set_buffer_new(bh_result); | |
7351 | ||
7352 | /* | |
7353 | * Need to update the i_size under the extent lock so buffered | |
7354 | * readers will get the updated i_size when we unlock. | |
7355 | */ | |
7356 | if (start + len > i_size_read(inode)) | |
7357 | i_size_write(inode, start + len); | |
0934856d | 7358 | |
3e05bde8 JB |
7359 | if (len < orig_len) { |
7360 | spin_lock(&BTRFS_I(inode)->lock); | |
7361 | BTRFS_I(inode)->outstanding_extents++; | |
7362 | spin_unlock(&BTRFS_I(inode)->lock); | |
7363 | } | |
3266789f | 7364 | btrfs_free_reserved_data_space(inode, len); |
c3473e83 | 7365 | } |
4b46fce2 | 7366 | |
eb838e73 JB |
7367 | /* |
7368 | * In the case of write we need to clear and unlock the entire range, | |
7369 | * in the case of read we need to unlock only the end area that we | |
7370 | * aren't using if there is any left over space. | |
7371 | */ | |
24c03fa5 | 7372 | if (lockstart < lockend) { |
0934856d MX |
7373 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7374 | lockend, unlock_bits, 1, 0, | |
7375 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7376 | } else { |
eb838e73 | 7377 | free_extent_state(cached_state); |
24c03fa5 | 7378 | } |
eb838e73 | 7379 | |
4b46fce2 JB |
7380 | free_extent_map(em); |
7381 | ||
7382 | return 0; | |
eb838e73 JB |
7383 | |
7384 | unlock_err: | |
eb838e73 JB |
7385 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7386 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
7387 | return ret; | |
4b46fce2 JB |
7388 | } |
7389 | ||
8b110e39 MX |
7390 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7391 | int rw, int mirror_num) | |
7392 | { | |
7393 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7394 | int ret; | |
7395 | ||
7396 | BUG_ON(rw & REQ_WRITE); | |
7397 | ||
7398 | bio_get(bio); | |
7399 | ||
7400 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7401 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7402 | if (ret) | |
7403 | goto err; | |
7404 | ||
7405 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7406 | err: | |
7407 | bio_put(bio); | |
7408 | return ret; | |
7409 | } | |
7410 | ||
7411 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7412 | struct bio *failed_bio, | |
7413 | struct io_failure_record *failrec, | |
7414 | int failed_mirror) | |
7415 | { | |
7416 | int num_copies; | |
7417 | ||
7418 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7419 | failrec->logical, failrec->len); | |
7420 | if (num_copies == 1) { | |
7421 | /* | |
7422 | * we only have a single copy of the data, so don't bother with | |
7423 | * all the retry and error correction code that follows. no | |
7424 | * matter what the error is, it is very likely to persist. | |
7425 | */ | |
7426 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7427 | num_copies, failrec->this_mirror, failed_mirror); | |
7428 | return 0; | |
7429 | } | |
7430 | ||
7431 | failrec->failed_mirror = failed_mirror; | |
7432 | failrec->this_mirror++; | |
7433 | if (failrec->this_mirror == failed_mirror) | |
7434 | failrec->this_mirror++; | |
7435 | ||
7436 | if (failrec->this_mirror > num_copies) { | |
7437 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7438 | num_copies, failrec->this_mirror, failed_mirror); | |
7439 | return 0; | |
7440 | } | |
7441 | ||
7442 | return 1; | |
7443 | } | |
7444 | ||
7445 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
7446 | struct page *page, u64 start, u64 end, | |
7447 | int failed_mirror, bio_end_io_t *repair_endio, | |
7448 | void *repair_arg) | |
7449 | { | |
7450 | struct io_failure_record *failrec; | |
7451 | struct bio *bio; | |
7452 | int isector; | |
7453 | int read_mode; | |
7454 | int ret; | |
7455 | ||
7456 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7457 | ||
7458 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7459 | if (ret) | |
7460 | return ret; | |
7461 | ||
7462 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7463 | failed_mirror); | |
7464 | if (!ret) { | |
7465 | free_io_failure(inode, failrec); | |
7466 | return -EIO; | |
7467 | } | |
7468 | ||
7469 | if (failed_bio->bi_vcnt > 1) | |
7470 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
7471 | else | |
7472 | read_mode = READ_SYNC; | |
7473 | ||
7474 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7475 | isector >>= inode->i_sb->s_blocksize_bits; | |
7476 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
7477 | 0, isector, repair_endio, repair_arg); | |
7478 | if (!bio) { | |
7479 | free_io_failure(inode, failrec); | |
7480 | return -EIO; | |
7481 | } | |
7482 | ||
7483 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7484 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7485 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7486 | ||
7487 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7488 | failrec->this_mirror); | |
7489 | if (ret) { | |
7490 | free_io_failure(inode, failrec); | |
7491 | bio_put(bio); | |
7492 | } | |
7493 | ||
7494 | return ret; | |
7495 | } | |
7496 | ||
7497 | struct btrfs_retry_complete { | |
7498 | struct completion done; | |
7499 | struct inode *inode; | |
7500 | u64 start; | |
7501 | int uptodate; | |
7502 | }; | |
7503 | ||
7504 | static void btrfs_retry_endio_nocsum(struct bio *bio, int err) | |
7505 | { | |
7506 | struct btrfs_retry_complete *done = bio->bi_private; | |
7507 | struct bio_vec *bvec; | |
7508 | int i; | |
7509 | ||
7510 | if (err) | |
7511 | goto end; | |
7512 | ||
7513 | done->uptodate = 1; | |
7514 | bio_for_each_segment_all(bvec, bio, i) | |
7515 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7516 | end: | |
7517 | complete(&done->done); | |
7518 | bio_put(bio); | |
7519 | } | |
7520 | ||
7521 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7522 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7523 | { |
2c30c71b | 7524 | struct bio_vec *bvec; |
8b110e39 | 7525 | struct btrfs_retry_complete done; |
4b46fce2 | 7526 | u64 start; |
2c30c71b | 7527 | int i; |
c1dc0896 | 7528 | int ret; |
4b46fce2 | 7529 | |
8b110e39 MX |
7530 | start = io_bio->logical; |
7531 | done.inode = inode; | |
7532 | ||
7533 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
7534 | try_again: | |
7535 | done.uptodate = 0; | |
7536 | done.start = start; | |
7537 | init_completion(&done.done); | |
7538 | ||
7539 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7540 | start + bvec->bv_len - 1, | |
7541 | io_bio->mirror_num, | |
7542 | btrfs_retry_endio_nocsum, &done); | |
7543 | if (ret) | |
7544 | return ret; | |
7545 | ||
7546 | wait_for_completion(&done.done); | |
7547 | ||
7548 | if (!done.uptodate) { | |
7549 | /* We might have another mirror, so try again */ | |
7550 | goto try_again; | |
7551 | } | |
7552 | ||
7553 | start += bvec->bv_len; | |
7554 | } | |
7555 | ||
7556 | return 0; | |
7557 | } | |
7558 | ||
7559 | static void btrfs_retry_endio(struct bio *bio, int err) | |
7560 | { | |
7561 | struct btrfs_retry_complete *done = bio->bi_private; | |
7562 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7563 | struct bio_vec *bvec; | |
7564 | int uptodate; | |
7565 | int ret; | |
7566 | int i; | |
7567 | ||
7568 | if (err) | |
7569 | goto end; | |
7570 | ||
7571 | uptodate = 1; | |
7572 | bio_for_each_segment_all(bvec, bio, i) { | |
7573 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
7574 | bvec->bv_page, 0, | |
7575 | done->start, bvec->bv_len); | |
7576 | if (!ret) | |
7577 | clean_io_failure(done->inode, done->start, | |
7578 | bvec->bv_page, 0); | |
7579 | else | |
7580 | uptodate = 0; | |
7581 | } | |
7582 | ||
7583 | done->uptodate = uptodate; | |
7584 | end: | |
7585 | complete(&done->done); | |
7586 | bio_put(bio); | |
7587 | } | |
7588 | ||
7589 | static int __btrfs_subio_endio_read(struct inode *inode, | |
7590 | struct btrfs_io_bio *io_bio, int err) | |
7591 | { | |
7592 | struct bio_vec *bvec; | |
7593 | struct btrfs_retry_complete done; | |
7594 | u64 start; | |
7595 | u64 offset = 0; | |
7596 | int i; | |
7597 | int ret; | |
dc380aea | 7598 | |
8b110e39 | 7599 | err = 0; |
c1dc0896 | 7600 | start = io_bio->logical; |
8b110e39 MX |
7601 | done.inode = inode; |
7602 | ||
c1dc0896 | 7603 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
dc380aea MX |
7604 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
7605 | 0, start, bvec->bv_len); | |
8b110e39 MX |
7606 | if (likely(!ret)) |
7607 | goto next; | |
7608 | try_again: | |
7609 | done.uptodate = 0; | |
7610 | done.start = start; | |
7611 | init_completion(&done.done); | |
7612 | ||
7613 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7614 | start + bvec->bv_len - 1, | |
7615 | io_bio->mirror_num, | |
7616 | btrfs_retry_endio, &done); | |
7617 | if (ret) { | |
7618 | err = ret; | |
7619 | goto next; | |
7620 | } | |
7621 | ||
7622 | wait_for_completion(&done.done); | |
7623 | ||
7624 | if (!done.uptodate) { | |
7625 | /* We might have another mirror, so try again */ | |
7626 | goto try_again; | |
7627 | } | |
7628 | next: | |
7629 | offset += bvec->bv_len; | |
4b46fce2 | 7630 | start += bvec->bv_len; |
2c30c71b | 7631 | } |
c1dc0896 MX |
7632 | |
7633 | return err; | |
7634 | } | |
7635 | ||
8b110e39 MX |
7636 | static int btrfs_subio_endio_read(struct inode *inode, |
7637 | struct btrfs_io_bio *io_bio, int err) | |
7638 | { | |
7639 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7640 | ||
7641 | if (skip_csum) { | |
7642 | if (unlikely(err)) | |
7643 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
7644 | else | |
7645 | return 0; | |
7646 | } else { | |
7647 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
7648 | } | |
7649 | } | |
7650 | ||
c1dc0896 MX |
7651 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
7652 | { | |
7653 | struct btrfs_dio_private *dip = bio->bi_private; | |
7654 | struct inode *inode = dip->inode; | |
7655 | struct bio *dio_bio; | |
7656 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7657 | ||
8b110e39 MX |
7658 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
7659 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 7660 | |
4b46fce2 | 7661 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 7662 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 7663 | dio_bio = dip->dio_bio; |
4b46fce2 | 7664 | |
4b46fce2 | 7665 | kfree(dip); |
c0da7aa1 JB |
7666 | |
7667 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
7668 | if (err) | |
9be3395b CM |
7669 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7670 | dio_end_io(dio_bio, err); | |
23ea8e5a MX |
7671 | |
7672 | if (io_bio->end_io) | |
7673 | io_bio->end_io(io_bio, err); | |
9be3395b | 7674 | bio_put(bio); |
4b46fce2 JB |
7675 | } |
7676 | ||
7677 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
7678 | { | |
7679 | struct btrfs_dio_private *dip = bio->bi_private; | |
7680 | struct inode *inode = dip->inode; | |
7681 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 7682 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
7683 | u64 ordered_offset = dip->logical_offset; |
7684 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 7685 | struct bio *dio_bio; |
4b46fce2 JB |
7686 | int ret; |
7687 | ||
7688 | if (err) | |
7689 | goto out_done; | |
163cf09c CM |
7690 | again: |
7691 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
7692 | &ordered_offset, | |
5fd02043 | 7693 | ordered_bytes, !err); |
4b46fce2 | 7694 | if (!ret) |
163cf09c | 7695 | goto out_test; |
4b46fce2 | 7696 | |
9e0af237 LB |
7697 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
7698 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
7699 | btrfs_queue_work(root->fs_info->endio_write_workers, |
7700 | &ordered->work); | |
163cf09c CM |
7701 | out_test: |
7702 | /* | |
7703 | * our bio might span multiple ordered extents. If we haven't | |
7704 | * completed the accounting for the whole dio, go back and try again | |
7705 | */ | |
7706 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
7707 | ordered_bytes = dip->logical_offset + dip->bytes - | |
7708 | ordered_offset; | |
5fd02043 | 7709 | ordered = NULL; |
163cf09c CM |
7710 | goto again; |
7711 | } | |
4b46fce2 | 7712 | out_done: |
9be3395b | 7713 | dio_bio = dip->dio_bio; |
4b46fce2 | 7714 | |
4b46fce2 | 7715 | kfree(dip); |
c0da7aa1 JB |
7716 | |
7717 | /* If we had an error make sure to clear the uptodate flag */ | |
7718 | if (err) | |
9be3395b CM |
7719 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
7720 | dio_end_io(dio_bio, err); | |
7721 | bio_put(bio); | |
4b46fce2 JB |
7722 | } |
7723 | ||
eaf25d93 CM |
7724 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
7725 | struct bio *bio, int mirror_num, | |
7726 | unsigned long bio_flags, u64 offset) | |
7727 | { | |
7728 | int ret; | |
7729 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7730 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 7731 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
7732 | return 0; |
7733 | } | |
7734 | ||
e65e1535 MX |
7735 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
7736 | { | |
7737 | struct btrfs_dio_private *dip = bio->bi_private; | |
7738 | ||
8b110e39 MX |
7739 | if (err) |
7740 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
7741 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
7742 | btrfs_ino(dip->inode), bio->bi_rw, | |
7743 | (unsigned long long)bio->bi_iter.bi_sector, | |
7744 | bio->bi_iter.bi_size, err); | |
7745 | ||
7746 | if (dip->subio_endio) | |
7747 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
7748 | |
7749 | if (err) { | |
e65e1535 MX |
7750 | dip->errors = 1; |
7751 | ||
7752 | /* | |
7753 | * before atomic variable goto zero, we must make sure | |
7754 | * dip->errors is perceived to be set. | |
7755 | */ | |
4e857c58 | 7756 | smp_mb__before_atomic(); |
e65e1535 MX |
7757 | } |
7758 | ||
7759 | /* if there are more bios still pending for this dio, just exit */ | |
7760 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
7761 | goto out; | |
7762 | ||
9be3395b | 7763 | if (dip->errors) { |
e65e1535 | 7764 | bio_io_error(dip->orig_bio); |
9be3395b CM |
7765 | } else { |
7766 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
7767 | bio_endio(dip->orig_bio, 0); |
7768 | } | |
7769 | out: | |
7770 | bio_put(bio); | |
7771 | } | |
7772 | ||
7773 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
7774 | u64 first_sector, gfp_t gfp_flags) | |
7775 | { | |
7776 | int nr_vecs = bio_get_nr_vecs(bdev); | |
7777 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
7778 | } | |
7779 | ||
c1dc0896 MX |
7780 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
7781 | struct inode *inode, | |
7782 | struct btrfs_dio_private *dip, | |
7783 | struct bio *bio, | |
7784 | u64 file_offset) | |
7785 | { | |
7786 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7787 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
7788 | int ret; | |
7789 | ||
7790 | /* | |
7791 | * We load all the csum data we need when we submit | |
7792 | * the first bio to reduce the csum tree search and | |
7793 | * contention. | |
7794 | */ | |
7795 | if (dip->logical_offset == file_offset) { | |
7796 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
7797 | file_offset); | |
7798 | if (ret) | |
7799 | return ret; | |
7800 | } | |
7801 | ||
7802 | if (bio == dip->orig_bio) | |
7803 | return 0; | |
7804 | ||
7805 | file_offset -= dip->logical_offset; | |
7806 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
7807 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
7808 | ||
7809 | return 0; | |
7810 | } | |
7811 | ||
e65e1535 MX |
7812 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
7813 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 7814 | int async_submit) |
e65e1535 | 7815 | { |
facc8a22 | 7816 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
7817 | int write = rw & REQ_WRITE; |
7818 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7819 | int ret; | |
7820 | ||
b812ce28 JB |
7821 | if (async_submit) |
7822 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
7823 | ||
e65e1535 | 7824 | bio_get(bio); |
5fd02043 JB |
7825 | |
7826 | if (!write) { | |
bfebd8b5 DS |
7827 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
7828 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
7829 | if (ret) |
7830 | goto err; | |
7831 | } | |
e65e1535 | 7832 | |
1ae39938 JB |
7833 | if (skip_sum) |
7834 | goto map; | |
7835 | ||
7836 | if (write && async_submit) { | |
e65e1535 MX |
7837 | ret = btrfs_wq_submit_bio(root->fs_info, |
7838 | inode, rw, bio, 0, 0, | |
7839 | file_offset, | |
7840 | __btrfs_submit_bio_start_direct_io, | |
7841 | __btrfs_submit_bio_done); | |
7842 | goto err; | |
1ae39938 JB |
7843 | } else if (write) { |
7844 | /* | |
7845 | * If we aren't doing async submit, calculate the csum of the | |
7846 | * bio now. | |
7847 | */ | |
7848 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
7849 | if (ret) | |
7850 | goto err; | |
23ea8e5a | 7851 | } else { |
c1dc0896 MX |
7852 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
7853 | file_offset); | |
c2db1073 TI |
7854 | if (ret) |
7855 | goto err; | |
7856 | } | |
1ae39938 JB |
7857 | map: |
7858 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
7859 | err: |
7860 | bio_put(bio); | |
7861 | return ret; | |
7862 | } | |
7863 | ||
7864 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
7865 | int skip_sum) | |
7866 | { | |
7867 | struct inode *inode = dip->inode; | |
7868 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7869 | struct bio *bio; |
7870 | struct bio *orig_bio = dip->orig_bio; | |
7871 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 7872 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
7873 | u64 file_offset = dip->logical_offset; |
7874 | u64 submit_len = 0; | |
7875 | u64 map_length; | |
7876 | int nr_pages = 0; | |
23ea8e5a | 7877 | int ret; |
1ae39938 | 7878 | int async_submit = 0; |
e65e1535 | 7879 | |
4f024f37 | 7880 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7881 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 7882 | &map_length, NULL, 0); |
7a5c3c9b | 7883 | if (ret) |
e65e1535 | 7884 | return -EIO; |
facc8a22 | 7885 | |
4f024f37 | 7886 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 7887 | bio = orig_bio; |
c1dc0896 | 7888 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
7889 | goto submit; |
7890 | } | |
7891 | ||
53b381b3 | 7892 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 7893 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
7894 | async_submit = 0; |
7895 | else | |
7896 | async_submit = 1; | |
7897 | ||
02f57c7a JB |
7898 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7899 | if (!bio) | |
7900 | return -ENOMEM; | |
7a5c3c9b | 7901 | |
02f57c7a JB |
7902 | bio->bi_private = dip; |
7903 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 7904 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
7905 | atomic_inc(&dip->pending_bios); |
7906 | ||
e65e1535 | 7907 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
ee39b432 | 7908 | if (map_length < submit_len + bvec->bv_len || |
e65e1535 | 7909 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, |
ee39b432 | 7910 | bvec->bv_offset) < bvec->bv_len) { |
e65e1535 MX |
7911 | /* |
7912 | * inc the count before we submit the bio so | |
7913 | * we know the end IO handler won't happen before | |
7914 | * we inc the count. Otherwise, the dip might get freed | |
7915 | * before we're done setting it up | |
7916 | */ | |
7917 | atomic_inc(&dip->pending_bios); | |
7918 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7919 | file_offset, skip_sum, | |
c329861d | 7920 | async_submit); |
e65e1535 MX |
7921 | if (ret) { |
7922 | bio_put(bio); | |
7923 | atomic_dec(&dip->pending_bios); | |
7924 | goto out_err; | |
7925 | } | |
7926 | ||
e65e1535 MX |
7927 | start_sector += submit_len >> 9; |
7928 | file_offset += submit_len; | |
7929 | ||
7930 | submit_len = 0; | |
7931 | nr_pages = 0; | |
7932 | ||
7933 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7934 | start_sector, GFP_NOFS); | |
7935 | if (!bio) | |
7936 | goto out_err; | |
7937 | bio->bi_private = dip; | |
7938 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 7939 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 7940 | |
4f024f37 | 7941 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 7942 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7943 | start_sector << 9, |
e65e1535 MX |
7944 | &map_length, NULL, 0); |
7945 | if (ret) { | |
7946 | bio_put(bio); | |
7947 | goto out_err; | |
7948 | } | |
7949 | } else { | |
7950 | submit_len += bvec->bv_len; | |
67871254 | 7951 | nr_pages++; |
e65e1535 MX |
7952 | bvec++; |
7953 | } | |
7954 | } | |
7955 | ||
02f57c7a | 7956 | submit: |
e65e1535 | 7957 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7958 | async_submit); |
e65e1535 MX |
7959 | if (!ret) |
7960 | return 0; | |
7961 | ||
7962 | bio_put(bio); | |
7963 | out_err: | |
7964 | dip->errors = 1; | |
7965 | /* | |
7966 | * before atomic variable goto zero, we must | |
7967 | * make sure dip->errors is perceived to be set. | |
7968 | */ | |
4e857c58 | 7969 | smp_mb__before_atomic(); |
e65e1535 MX |
7970 | if (atomic_dec_and_test(&dip->pending_bios)) |
7971 | bio_io_error(dip->orig_bio); | |
7972 | ||
7973 | /* bio_end_io() will handle error, so we needn't return it */ | |
7974 | return 0; | |
7975 | } | |
7976 | ||
9be3395b CM |
7977 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7978 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7979 | { |
7980 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7981 | struct btrfs_dio_private *dip; | |
9be3395b | 7982 | struct bio *io_bio; |
23ea8e5a | 7983 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 7984 | int skip_sum; |
7b6d91da | 7985 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
7986 | int ret = 0; |
7987 | ||
7988 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7989 | ||
9be3395b | 7990 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7991 | if (!io_bio) { |
7992 | ret = -ENOMEM; | |
7993 | goto free_ordered; | |
7994 | } | |
7995 | ||
c1dc0896 | 7996 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
7997 | if (!dip) { |
7998 | ret = -ENOMEM; | |
9be3395b | 7999 | goto free_io_bio; |
4b46fce2 | 8000 | } |
4b46fce2 | 8001 | |
9be3395b | 8002 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8003 | dip->inode = inode; |
8004 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8005 | dip->bytes = dio_bio->bi_iter.bi_size; |
8006 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8007 | io_bio->bi_private = dip; |
9be3395b CM |
8008 | dip->orig_bio = io_bio; |
8009 | dip->dio_bio = dio_bio; | |
e65e1535 | 8010 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8011 | btrfs_bio = btrfs_io_bio(io_bio); |
8012 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8013 | |
c1dc0896 | 8014 | if (write) { |
9be3395b | 8015 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8016 | } else { |
9be3395b | 8017 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8018 | dip->subio_endio = btrfs_subio_endio_read; |
8019 | } | |
4b46fce2 | 8020 | |
e65e1535 MX |
8021 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
8022 | if (!ret) | |
eaf25d93 | 8023 | return; |
9be3395b | 8024 | |
23ea8e5a MX |
8025 | if (btrfs_bio->end_io) |
8026 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b CM |
8027 | free_io_bio: |
8028 | bio_put(io_bio); | |
8029 | ||
4b46fce2 JB |
8030 | free_ordered: |
8031 | /* | |
8032 | * If this is a write, we need to clean up the reserved space and kill | |
8033 | * the ordered extent. | |
8034 | */ | |
8035 | if (write) { | |
8036 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 8037 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
8038 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
8039 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
8040 | btrfs_free_reserved_extent(root, ordered->start, | |
e570fd27 | 8041 | ordered->disk_len, 1); |
4b46fce2 JB |
8042 | btrfs_put_ordered_extent(ordered); |
8043 | btrfs_put_ordered_extent(ordered); | |
8044 | } | |
9be3395b | 8045 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
8046 | } |
8047 | ||
5a5f79b5 | 8048 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
28060d5d | 8049 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8050 | { |
8051 | int seg; | |
a1b75f7d | 8052 | int i; |
5a5f79b5 CM |
8053 | unsigned blocksize_mask = root->sectorsize - 1; |
8054 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
8055 | |
8056 | if (offset & blocksize_mask) | |
8057 | goto out; | |
8058 | ||
28060d5d AV |
8059 | if (iov_iter_alignment(iter) & blocksize_mask) |
8060 | goto out; | |
a1b75f7d | 8061 | |
28060d5d AV |
8062 | /* If this is a write we don't need to check anymore */ |
8063 | if (rw & WRITE) | |
8064 | return 0; | |
8065 | /* | |
8066 | * Check to make sure we don't have duplicate iov_base's in this | |
8067 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8068 | * when reading back. | |
8069 | */ | |
8070 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8071 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8072 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8073 | goto out; |
8074 | } | |
5a5f79b5 CM |
8075 | } |
8076 | retval = 0; | |
8077 | out: | |
8078 | return retval; | |
8079 | } | |
eb838e73 | 8080 | |
16432985 | 8081 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
d8d3d94b | 8082 | struct iov_iter *iter, loff_t offset) |
16432985 | 8083 | { |
4b46fce2 JB |
8084 | struct file *file = iocb->ki_filp; |
8085 | struct inode *inode = file->f_mapping->host; | |
0934856d | 8086 | size_t count = 0; |
2e60a51e | 8087 | int flags = 0; |
38851cc1 MX |
8088 | bool wakeup = true; |
8089 | bool relock = false; | |
0934856d | 8090 | ssize_t ret; |
4b46fce2 | 8091 | |
28060d5d | 8092 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset)) |
5a5f79b5 | 8093 | return 0; |
3f7c579c | 8094 | |
38851cc1 | 8095 | atomic_inc(&inode->i_dio_count); |
4e857c58 | 8096 | smp_mb__after_atomic(); |
38851cc1 | 8097 | |
0e267c44 | 8098 | /* |
41bd9ca4 MX |
8099 | * The generic stuff only does filemap_write_and_wait_range, which |
8100 | * isn't enough if we've written compressed pages to this area, so | |
8101 | * we need to flush the dirty pages again to make absolutely sure | |
8102 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8103 | */ |
a6cbcd4a | 8104 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8105 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8106 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8107 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8108 | offset + count - 1); | |
0e267c44 | 8109 | |
0934856d | 8110 | if (rw & WRITE) { |
38851cc1 MX |
8111 | /* |
8112 | * If the write DIO is beyond the EOF, we need update | |
8113 | * the isize, but it is protected by i_mutex. So we can | |
8114 | * not unlock the i_mutex at this case. | |
8115 | */ | |
8116 | if (offset + count <= inode->i_size) { | |
8117 | mutex_unlock(&inode->i_mutex); | |
8118 | relock = true; | |
8119 | } | |
0934856d MX |
8120 | ret = btrfs_delalloc_reserve_space(inode, count); |
8121 | if (ret) | |
38851cc1 | 8122 | goto out; |
ee39b432 DS |
8123 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8124 | &BTRFS_I(inode)->runtime_flags)) { | |
38851cc1 MX |
8125 | inode_dio_done(inode); |
8126 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
8127 | wakeup = false; | |
0934856d MX |
8128 | } |
8129 | ||
8130 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
8131 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
31b14039 | 8132 | iter, offset, btrfs_get_blocks_direct, NULL, |
2e60a51e | 8133 | btrfs_submit_direct, flags); |
0934856d MX |
8134 | if (rw & WRITE) { |
8135 | if (ret < 0 && ret != -EIOCBQUEUED) | |
8136 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 8137 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
8138 | btrfs_delalloc_release_space(inode, |
8139 | count - (size_t)ret); | |
0934856d | 8140 | } |
38851cc1 | 8141 | out: |
2e60a51e MX |
8142 | if (wakeup) |
8143 | inode_dio_done(inode); | |
38851cc1 MX |
8144 | if (relock) |
8145 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
8146 | |
8147 | return ret; | |
16432985 CM |
8148 | } |
8149 | ||
05dadc09 TI |
8150 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8151 | ||
1506fcc8 YS |
8152 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8153 | __u64 start, __u64 len) | |
8154 | { | |
05dadc09 TI |
8155 | int ret; |
8156 | ||
8157 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8158 | if (ret) | |
8159 | return ret; | |
8160 | ||
ec29ed5b | 8161 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8162 | } |
8163 | ||
a52d9a80 | 8164 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8165 | { |
d1310b2e CM |
8166 | struct extent_io_tree *tree; |
8167 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8168 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8169 | } |
1832a6d5 | 8170 | |
a52d9a80 | 8171 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8172 | { |
d1310b2e | 8173 | struct extent_io_tree *tree; |
b888db2b CM |
8174 | |
8175 | ||
8176 | if (current->flags & PF_MEMALLOC) { | |
8177 | redirty_page_for_writepage(wbc, page); | |
8178 | unlock_page(page); | |
8179 | return 0; | |
8180 | } | |
d1310b2e | 8181 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 8182 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
8183 | } |
8184 | ||
48a3b636 ES |
8185 | static int btrfs_writepages(struct address_space *mapping, |
8186 | struct writeback_control *wbc) | |
b293f02e | 8187 | { |
d1310b2e | 8188 | struct extent_io_tree *tree; |
771ed689 | 8189 | |
d1310b2e | 8190 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8191 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8192 | } | |
8193 | ||
3ab2fb5a CM |
8194 | static int |
8195 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8196 | struct list_head *pages, unsigned nr_pages) | |
8197 | { | |
d1310b2e CM |
8198 | struct extent_io_tree *tree; |
8199 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8200 | return extent_readpages(tree, mapping, pages, nr_pages, |
8201 | btrfs_get_extent); | |
8202 | } | |
e6dcd2dc | 8203 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8204 | { |
d1310b2e CM |
8205 | struct extent_io_tree *tree; |
8206 | struct extent_map_tree *map; | |
a52d9a80 | 8207 | int ret; |
8c2383c3 | 8208 | |
d1310b2e CM |
8209 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8210 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8211 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8212 | if (ret == 1) { |
8213 | ClearPagePrivate(page); | |
8214 | set_page_private(page, 0); | |
8215 | page_cache_release(page); | |
39279cc3 | 8216 | } |
a52d9a80 | 8217 | return ret; |
39279cc3 CM |
8218 | } |
8219 | ||
e6dcd2dc CM |
8220 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8221 | { | |
98509cfc CM |
8222 | if (PageWriteback(page) || PageDirty(page)) |
8223 | return 0; | |
b335b003 | 8224 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8225 | } |
8226 | ||
d47992f8 LC |
8227 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8228 | unsigned int length) | |
39279cc3 | 8229 | { |
5fd02043 | 8230 | struct inode *inode = page->mapping->host; |
d1310b2e | 8231 | struct extent_io_tree *tree; |
e6dcd2dc | 8232 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8233 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8234 | u64 page_start = page_offset(page); |
8235 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
131e404a | 8236 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8237 | |
8b62b72b CM |
8238 | /* |
8239 | * we have the page locked, so new writeback can't start, | |
8240 | * and the dirty bit won't be cleared while we are here. | |
8241 | * | |
8242 | * Wait for IO on this page so that we can safely clear | |
8243 | * the PagePrivate2 bit and do ordered accounting | |
8244 | */ | |
e6dcd2dc | 8245 | wait_on_page_writeback(page); |
8b62b72b | 8246 | |
5fd02043 | 8247 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8248 | if (offset) { |
8249 | btrfs_releasepage(page, GFP_NOFS); | |
8250 | return; | |
8251 | } | |
131e404a FDBM |
8252 | |
8253 | if (!inode_evicting) | |
8254 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); | |
8255 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
e6dcd2dc | 8256 | if (ordered) { |
eb84ae03 CM |
8257 | /* |
8258 | * IO on this page will never be started, so we need | |
8259 | * to account for any ordered extents now | |
8260 | */ | |
131e404a FDBM |
8261 | if (!inode_evicting) |
8262 | clear_extent_bit(tree, page_start, page_end, | |
8263 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
8264 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8265 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8266 | GFP_NOFS); | |
8b62b72b CM |
8267 | /* |
8268 | * whoever cleared the private bit is responsible | |
8269 | * for the finish_ordered_io | |
8270 | */ | |
77cef2ec JB |
8271 | if (TestClearPagePrivate2(page)) { |
8272 | struct btrfs_ordered_inode_tree *tree; | |
8273 | u64 new_len; | |
8274 | ||
8275 | tree = &BTRFS_I(inode)->ordered_tree; | |
8276 | ||
8277 | spin_lock_irq(&tree->lock); | |
8278 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8279 | new_len = page_start - ordered->file_offset; | |
8280 | if (new_len < ordered->truncated_len) | |
8281 | ordered->truncated_len = new_len; | |
8282 | spin_unlock_irq(&tree->lock); | |
8283 | ||
8284 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
8285 | page_start, | |
8286 | PAGE_CACHE_SIZE, 1)) | |
8287 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 8288 | } |
e6dcd2dc | 8289 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8290 | if (!inode_evicting) { |
8291 | cached_state = NULL; | |
8292 | lock_extent_bits(tree, page_start, page_end, 0, | |
8293 | &cached_state); | |
8294 | } | |
8295 | } | |
8296 | ||
8297 | if (!inode_evicting) { | |
8298 | clear_extent_bit(tree, page_start, page_end, | |
8299 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8300 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8301 | EXTENT_DEFRAG, 1, 1, | |
8302 | &cached_state, GFP_NOFS); | |
8303 | ||
8304 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8305 | } |
e6dcd2dc | 8306 | |
4a096752 | 8307 | ClearPageChecked(page); |
9ad6b7bc | 8308 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8309 | ClearPagePrivate(page); |
8310 | set_page_private(page, 0); | |
8311 | page_cache_release(page); | |
8312 | } | |
39279cc3 CM |
8313 | } |
8314 | ||
9ebefb18 CM |
8315 | /* |
8316 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8317 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8318 | * be careful to check for EOF conditions here. We set the page up correctly | |
8319 | * for a written page which means we get ENOSPC checking when writing into | |
8320 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8321 | * support these features. | |
8322 | * | |
8323 | * We are not allowed to take the i_mutex here so we have to play games to | |
8324 | * protect against truncate races as the page could now be beyond EOF. Because | |
8325 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8326 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8327 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8328 | * unlock the page. | |
8329 | */ | |
c2ec175c | 8330 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8331 | { |
c2ec175c | 8332 | struct page *page = vmf->page; |
496ad9aa | 8333 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8334 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8335 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8336 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8337 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8338 | char *kaddr; |
8339 | unsigned long zero_start; | |
9ebefb18 | 8340 | loff_t size; |
1832a6d5 | 8341 | int ret; |
9998eb70 | 8342 | int reserved = 0; |
a52d9a80 | 8343 | u64 page_start; |
e6dcd2dc | 8344 | u64 page_end; |
9ebefb18 | 8345 | |
b2b5ef5c | 8346 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 8347 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8348 | if (!ret) { |
e41f941a | 8349 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8350 | reserved = 1; |
8351 | } | |
56a76f82 NP |
8352 | if (ret) { |
8353 | if (ret == -ENOMEM) | |
8354 | ret = VM_FAULT_OOM; | |
8355 | else /* -ENOSPC, -EIO, etc */ | |
8356 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8357 | if (reserved) |
8358 | goto out; | |
8359 | goto out_noreserve; | |
56a76f82 | 8360 | } |
1832a6d5 | 8361 | |
56a76f82 | 8362 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8363 | again: |
9ebefb18 | 8364 | lock_page(page); |
9ebefb18 | 8365 | size = i_size_read(inode); |
e6dcd2dc CM |
8366 | page_start = page_offset(page); |
8367 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 8368 | |
9ebefb18 | 8369 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8370 | (page_start >= size)) { |
9ebefb18 CM |
8371 | /* page got truncated out from underneath us */ |
8372 | goto out_unlock; | |
8373 | } | |
e6dcd2dc CM |
8374 | wait_on_page_writeback(page); |
8375 | ||
d0082371 | 8376 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
8377 | set_page_extent_mapped(page); |
8378 | ||
eb84ae03 CM |
8379 | /* |
8380 | * we can't set the delalloc bits if there are pending ordered | |
8381 | * extents. Drop our locks and wait for them to finish | |
8382 | */ | |
e6dcd2dc CM |
8383 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
8384 | if (ordered) { | |
2ac55d41 JB |
8385 | unlock_extent_cached(io_tree, page_start, page_end, |
8386 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8387 | unlock_page(page); |
eb84ae03 | 8388 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8389 | btrfs_put_ordered_extent(ordered); |
8390 | goto again; | |
8391 | } | |
8392 | ||
fbf19087 JB |
8393 | /* |
8394 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8395 | * if it was already dirty, so for space accounting reasons we need to | |
8396 | * clear any delalloc bits for the range we are fixing to save. There | |
8397 | * is probably a better way to do this, but for now keep consistent with | |
8398 | * prepare_pages in the normal write path. | |
8399 | */ | |
2ac55d41 | 8400 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
8401 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8402 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 8403 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 8404 | |
2ac55d41 JB |
8405 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
8406 | &cached_state); | |
9ed74f2d | 8407 | if (ret) { |
2ac55d41 JB |
8408 | unlock_extent_cached(io_tree, page_start, page_end, |
8409 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
8410 | ret = VM_FAULT_SIGBUS; |
8411 | goto out_unlock; | |
8412 | } | |
e6dcd2dc | 8413 | ret = 0; |
9ebefb18 CM |
8414 | |
8415 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 8416 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 8417 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 8418 | else |
e6dcd2dc | 8419 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 8420 | |
e6dcd2dc CM |
8421 | if (zero_start != PAGE_CACHE_SIZE) { |
8422 | kaddr = kmap(page); | |
8423 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
8424 | flush_dcache_page(page); | |
8425 | kunmap(page); | |
8426 | } | |
247e743c | 8427 | ClearPageChecked(page); |
e6dcd2dc | 8428 | set_page_dirty(page); |
50a9b214 | 8429 | SetPageUptodate(page); |
5a3f23d5 | 8430 | |
257c62e1 CM |
8431 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
8432 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 8433 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8434 | |
2ac55d41 | 8435 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
8436 | |
8437 | out_unlock: | |
b2b5ef5c JK |
8438 | if (!ret) { |
8439 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 8440 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8441 | } |
9ebefb18 | 8442 | unlock_page(page); |
1832a6d5 | 8443 | out: |
ec39e180 | 8444 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8445 | out_noreserve: |
b2b5ef5c | 8446 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
8447 | return ret; |
8448 | } | |
8449 | ||
a41ad394 | 8450 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
8451 | { |
8452 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 8453 | struct btrfs_block_rsv *rsv; |
a71754fc | 8454 | int ret = 0; |
3893e33b | 8455 | int err = 0; |
39279cc3 | 8456 | struct btrfs_trans_handle *trans; |
dbe674a9 | 8457 | u64 mask = root->sectorsize - 1; |
07127184 | 8458 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 8459 | |
0ef8b726 JB |
8460 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
8461 | (u64)-1); | |
8462 | if (ret) | |
8463 | return ret; | |
39279cc3 | 8464 | |
fcb80c2a JB |
8465 | /* |
8466 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
8467 | * 3 things going on here | |
8468 | * | |
8469 | * 1) We need to reserve space for our orphan item and the space to | |
8470 | * delete our orphan item. Lord knows we don't want to have a dangling | |
8471 | * orphan item because we didn't reserve space to remove it. | |
8472 | * | |
8473 | * 2) We need to reserve space to update our inode. | |
8474 | * | |
8475 | * 3) We need to have something to cache all the space that is going to | |
8476 | * be free'd up by the truncate operation, but also have some slack | |
8477 | * space reserved in case it uses space during the truncate (thank you | |
8478 | * very much snapshotting). | |
8479 | * | |
8480 | * And we need these to all be seperate. The fact is we can use alot of | |
8481 | * space doing the truncate, and we have no earthly idea how much space | |
8482 | * we will use, so we need the truncate reservation to be seperate so it | |
8483 | * doesn't end up using space reserved for updating the inode or | |
8484 | * removing the orphan item. We also need to be able to stop the | |
8485 | * transaction and start a new one, which means we need to be able to | |
8486 | * update the inode several times, and we have no idea of knowing how | |
8487 | * many times that will be, so we can't just reserve 1 item for the | |
8488 | * entirety of the opration, so that has to be done seperately as well. | |
8489 | * Then there is the orphan item, which does indeed need to be held on | |
8490 | * to for the whole operation, and we need nobody to touch this reserved | |
8491 | * space except the orphan code. | |
8492 | * | |
8493 | * So that leaves us with | |
8494 | * | |
8495 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
8496 | * 2) rsv - for the truncate reservation, which we will steal from the | |
8497 | * transaction reservation. | |
8498 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
8499 | * updating the inode. | |
8500 | */ | |
66d8f3dd | 8501 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8502 | if (!rsv) |
8503 | return -ENOMEM; | |
4a338542 | 8504 | rsv->size = min_size; |
ca7e70f5 | 8505 | rsv->failfast = 1; |
f0cd846e | 8506 | |
907cbceb | 8507 | /* |
07127184 | 8508 | * 1 for the truncate slack space |
907cbceb JB |
8509 | * 1 for updating the inode. |
8510 | */ | |
f3fe820c | 8511 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
8512 | if (IS_ERR(trans)) { |
8513 | err = PTR_ERR(trans); | |
8514 | goto out; | |
8515 | } | |
f0cd846e | 8516 | |
907cbceb JB |
8517 | /* Migrate the slack space for the truncate to our reserve */ |
8518 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
8519 | min_size); | |
fcb80c2a | 8520 | BUG_ON(ret); |
f0cd846e | 8521 | |
5dc562c5 JB |
8522 | /* |
8523 | * So if we truncate and then write and fsync we normally would just | |
8524 | * write the extents that changed, which is a problem if we need to | |
8525 | * first truncate that entire inode. So set this flag so we write out | |
8526 | * all of the extents in the inode to the sync log so we're completely | |
8527 | * safe. | |
8528 | */ | |
8529 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8530 | trans->block_rsv = rsv; |
907cbceb | 8531 | |
8082510e YZ |
8532 | while (1) { |
8533 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
8534 | inode->i_size, | |
8535 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 8536 | if (ret != -ENOSPC) { |
3893e33b | 8537 | err = ret; |
8082510e | 8538 | break; |
3893e33b | 8539 | } |
39279cc3 | 8540 | |
fcb80c2a | 8541 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 8542 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
8543 | if (ret) { |
8544 | err = ret; | |
8545 | break; | |
8546 | } | |
ca7e70f5 | 8547 | |
8082510e | 8548 | btrfs_end_transaction(trans, root); |
b53d3f5d | 8549 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
8550 | |
8551 | trans = btrfs_start_transaction(root, 2); | |
8552 | if (IS_ERR(trans)) { | |
8553 | ret = err = PTR_ERR(trans); | |
8554 | trans = NULL; | |
8555 | break; | |
8556 | } | |
8557 | ||
8558 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
8559 | rsv, min_size); | |
8560 | BUG_ON(ret); /* shouldn't happen */ | |
8561 | trans->block_rsv = rsv; | |
8082510e YZ |
8562 | } |
8563 | ||
8564 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 8565 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 8566 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
8567 | if (ret) |
8568 | err = ret; | |
8082510e YZ |
8569 | } |
8570 | ||
917c16b2 CM |
8571 | if (trans) { |
8572 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
8573 | ret = btrfs_update_inode(trans, root, inode); | |
8574 | if (ret && !err) | |
8575 | err = ret; | |
7b128766 | 8576 | |
7ad85bb7 | 8577 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 8578 | btrfs_btree_balance_dirty(root); |
917c16b2 | 8579 | } |
fcb80c2a JB |
8580 | |
8581 | out: | |
8582 | btrfs_free_block_rsv(root, rsv); | |
8583 | ||
3893e33b JB |
8584 | if (ret && !err) |
8585 | err = ret; | |
a41ad394 | 8586 | |
3893e33b | 8587 | return err; |
39279cc3 CM |
8588 | } |
8589 | ||
d352ac68 CM |
8590 | /* |
8591 | * create a new subvolume directory/inode (helper for the ioctl). | |
8592 | */ | |
d2fb3437 | 8593 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8594 | struct btrfs_root *new_root, |
8595 | struct btrfs_root *parent_root, | |
8596 | u64 new_dirid) | |
39279cc3 | 8597 | { |
39279cc3 | 8598 | struct inode *inode; |
76dda93c | 8599 | int err; |
00e4e6b3 | 8600 | u64 index = 0; |
39279cc3 | 8601 | |
12fc9d09 FA |
8602 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8603 | new_dirid, new_dirid, | |
8604 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8605 | &index); | |
54aa1f4d | 8606 | if (IS_ERR(inode)) |
f46b5a66 | 8607 | return PTR_ERR(inode); |
39279cc3 CM |
8608 | inode->i_op = &btrfs_dir_inode_operations; |
8609 | inode->i_fop = &btrfs_dir_file_operations; | |
8610 | ||
bfe86848 | 8611 | set_nlink(inode, 1); |
dbe674a9 | 8612 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 8613 | unlock_new_inode(inode); |
3b96362c | 8614 | |
63541927 FDBM |
8615 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8616 | if (err) | |
8617 | btrfs_err(new_root->fs_info, | |
351fd353 | 8618 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8619 | new_root->root_key.objectid, err); |
8620 | ||
76dda93c | 8621 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 8622 | |
76dda93c | 8623 | iput(inode); |
ce598979 | 8624 | return err; |
39279cc3 CM |
8625 | } |
8626 | ||
39279cc3 CM |
8627 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8628 | { | |
8629 | struct btrfs_inode *ei; | |
2ead6ae7 | 8630 | struct inode *inode; |
39279cc3 CM |
8631 | |
8632 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
8633 | if (!ei) | |
8634 | return NULL; | |
2ead6ae7 YZ |
8635 | |
8636 | ei->root = NULL; | |
2ead6ae7 | 8637 | ei->generation = 0; |
15ee9bc7 | 8638 | ei->last_trans = 0; |
257c62e1 | 8639 | ei->last_sub_trans = 0; |
e02119d5 | 8640 | ei->logged_trans = 0; |
2ead6ae7 | 8641 | ei->delalloc_bytes = 0; |
47059d93 | 8642 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8643 | ei->disk_i_size = 0; |
8644 | ei->flags = 0; | |
7709cde3 | 8645 | ei->csum_bytes = 0; |
2ead6ae7 | 8646 | ei->index_cnt = (u64)-1; |
67de1176 | 8647 | ei->dir_index = 0; |
2ead6ae7 | 8648 | ei->last_unlink_trans = 0; |
46d8bc34 | 8649 | ei->last_log_commit = 0; |
2ead6ae7 | 8650 | |
9e0baf60 JB |
8651 | spin_lock_init(&ei->lock); |
8652 | ei->outstanding_extents = 0; | |
8653 | ei->reserved_extents = 0; | |
2ead6ae7 | 8654 | |
72ac3c0d | 8655 | ei->runtime_flags = 0; |
261507a0 | 8656 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8657 | |
16cdcec7 MX |
8658 | ei->delayed_node = NULL; |
8659 | ||
9cc97d64 | 8660 | ei->i_otime.tv_sec = 0; |
8661 | ei->i_otime.tv_nsec = 0; | |
8662 | ||
2ead6ae7 | 8663 | inode = &ei->vfs_inode; |
a8067e02 | 8664 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
8665 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
8666 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
8667 | ei->io_tree.track_uptodate = 1; |
8668 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 8669 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8670 | mutex_init(&ei->log_mutex); |
f248679e | 8671 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 8672 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8673 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
2ead6ae7 YZ |
8674 | RB_CLEAR_NODE(&ei->rb_node); |
8675 | ||
8676 | return inode; | |
39279cc3 CM |
8677 | } |
8678 | ||
aaedb55b JB |
8679 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
8680 | void btrfs_test_destroy_inode(struct inode *inode) | |
8681 | { | |
8682 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
8683 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
8684 | } | |
8685 | #endif | |
8686 | ||
fa0d7e3d NP |
8687 | static void btrfs_i_callback(struct rcu_head *head) |
8688 | { | |
8689 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
8690 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
8691 | } | |
8692 | ||
39279cc3 CM |
8693 | void btrfs_destroy_inode(struct inode *inode) |
8694 | { | |
e6dcd2dc | 8695 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
8696 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8697 | ||
b3d9b7a3 | 8698 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 8699 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
8700 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
8701 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
8702 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
8703 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 8704 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 8705 | |
a6dbd429 JB |
8706 | /* |
8707 | * This can happen where we create an inode, but somebody else also | |
8708 | * created the same inode and we need to destroy the one we already | |
8709 | * created. | |
8710 | */ | |
8711 | if (!root) | |
8712 | goto free; | |
8713 | ||
8a35d95f JB |
8714 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8715 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 8716 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 8717 | btrfs_ino(inode)); |
8a35d95f | 8718 | atomic_dec(&root->orphan_inodes); |
7b128766 | 8719 | } |
7b128766 | 8720 | |
d397712b | 8721 | while (1) { |
e6dcd2dc CM |
8722 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
8723 | if (!ordered) | |
8724 | break; | |
8725 | else { | |
c2cf52eb | 8726 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 8727 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
8728 | btrfs_remove_ordered_extent(inode, ordered); |
8729 | btrfs_put_ordered_extent(ordered); | |
8730 | btrfs_put_ordered_extent(ordered); | |
8731 | } | |
8732 | } | |
5d4f98a2 | 8733 | inode_tree_del(inode); |
5b21f2ed | 8734 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 8735 | free: |
fa0d7e3d | 8736 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
8737 | } |
8738 | ||
45321ac5 | 8739 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
8740 | { |
8741 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 8742 | |
6379ef9f NA |
8743 | if (root == NULL) |
8744 | return 1; | |
8745 | ||
fa6ac876 | 8746 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 8747 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 8748 | return 1; |
76dda93c | 8749 | else |
45321ac5 | 8750 | return generic_drop_inode(inode); |
76dda93c YZ |
8751 | } |
8752 | ||
0ee0fda0 | 8753 | static void init_once(void *foo) |
39279cc3 CM |
8754 | { |
8755 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
8756 | ||
8757 | inode_init_once(&ei->vfs_inode); | |
8758 | } | |
8759 | ||
8760 | void btrfs_destroy_cachep(void) | |
8761 | { | |
8c0a8537 KS |
8762 | /* |
8763 | * Make sure all delayed rcu free inodes are flushed before we | |
8764 | * destroy cache. | |
8765 | */ | |
8766 | rcu_barrier(); | |
39279cc3 CM |
8767 | if (btrfs_inode_cachep) |
8768 | kmem_cache_destroy(btrfs_inode_cachep); | |
8769 | if (btrfs_trans_handle_cachep) | |
8770 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
8771 | if (btrfs_transaction_cachep) | |
8772 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
8773 | if (btrfs_path_cachep) |
8774 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
8775 | if (btrfs_free_space_cachep) |
8776 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
8777 | if (btrfs_delalloc_work_cachep) |
8778 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
8779 | } |
8780 | ||
8781 | int btrfs_init_cachep(void) | |
8782 | { | |
837e1972 | 8783 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
8784 | sizeof(struct btrfs_inode), 0, |
8785 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
8786 | if (!btrfs_inode_cachep) |
8787 | goto fail; | |
9601e3f6 | 8788 | |
837e1972 | 8789 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
8790 | sizeof(struct btrfs_trans_handle), 0, |
8791 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8792 | if (!btrfs_trans_handle_cachep) |
8793 | goto fail; | |
9601e3f6 | 8794 | |
837e1972 | 8795 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
8796 | sizeof(struct btrfs_transaction), 0, |
8797 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8798 | if (!btrfs_transaction_cachep) |
8799 | goto fail; | |
9601e3f6 | 8800 | |
837e1972 | 8801 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
8802 | sizeof(struct btrfs_path), 0, |
8803 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
8804 | if (!btrfs_path_cachep) |
8805 | goto fail; | |
9601e3f6 | 8806 | |
837e1972 | 8807 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
8808 | sizeof(struct btrfs_free_space), 0, |
8809 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
8810 | if (!btrfs_free_space_cachep) | |
8811 | goto fail; | |
8812 | ||
8ccf6f19 MX |
8813 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
8814 | sizeof(struct btrfs_delalloc_work), 0, | |
8815 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
8816 | NULL); | |
8817 | if (!btrfs_delalloc_work_cachep) | |
8818 | goto fail; | |
8819 | ||
39279cc3 CM |
8820 | return 0; |
8821 | fail: | |
8822 | btrfs_destroy_cachep(); | |
8823 | return -ENOMEM; | |
8824 | } | |
8825 | ||
8826 | static int btrfs_getattr(struct vfsmount *mnt, | |
8827 | struct dentry *dentry, struct kstat *stat) | |
8828 | { | |
df0af1a5 | 8829 | u64 delalloc_bytes; |
39279cc3 | 8830 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
8831 | u32 blocksize = inode->i_sb->s_blocksize; |
8832 | ||
39279cc3 | 8833 | generic_fillattr(inode, stat); |
0ee5dc67 | 8834 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 8835 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
8836 | |
8837 | spin_lock(&BTRFS_I(inode)->lock); | |
8838 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
8839 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 8840 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 8841 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
8842 | return 0; |
8843 | } | |
8844 | ||
d397712b CM |
8845 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
8846 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
8847 | { |
8848 | struct btrfs_trans_handle *trans; | |
8849 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 8850 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
8851 | struct inode *new_inode = new_dentry->d_inode; |
8852 | struct inode *old_inode = old_dentry->d_inode; | |
8853 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 8854 | u64 index = 0; |
4df27c4d | 8855 | u64 root_objectid; |
39279cc3 | 8856 | int ret; |
33345d01 | 8857 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 8858 | |
33345d01 | 8859 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
8860 | return -EPERM; |
8861 | ||
4df27c4d | 8862 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 8863 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
8864 | return -EXDEV; |
8865 | ||
33345d01 LZ |
8866 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
8867 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 8868 | return -ENOTEMPTY; |
5f39d397 | 8869 | |
4df27c4d YZ |
8870 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
8871 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
8872 | return -ENOTEMPTY; | |
9c52057c CM |
8873 | |
8874 | ||
8875 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 8876 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
8877 | new_dentry->d_name.name, |
8878 | new_dentry->d_name.len); | |
8879 | ||
8880 | if (ret) { | |
8881 | if (ret == -EEXIST) { | |
8882 | /* we shouldn't get | |
8883 | * eexist without a new_inode */ | |
fae7f21c | 8884 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
8885 | return ret; |
8886 | } | |
8887 | } else { | |
8888 | /* maybe -EOVERFLOW */ | |
8889 | return ret; | |
8890 | } | |
8891 | } | |
8892 | ret = 0; | |
8893 | ||
5a3f23d5 | 8894 | /* |
8d875f95 CM |
8895 | * we're using rename to replace one file with another. Start IO on it |
8896 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 8897 | */ |
8d875f95 | 8898 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
8899 | filemap_flush(old_inode->i_mapping); |
8900 | ||
76dda93c | 8901 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8902 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8903 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8904 | /* |
8905 | * We want to reserve the absolute worst case amount of items. So if | |
8906 | * both inodes are subvols and we need to unlink them then that would | |
8907 | * require 4 item modifications, but if they are both normal inodes it | |
8908 | * would require 5 item modifications, so we'll assume their normal | |
8909 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8910 | * should cover the worst case number of items we'll modify. | |
8911 | */ | |
6e137ed3 | 8912 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8913 | if (IS_ERR(trans)) { |
8914 | ret = PTR_ERR(trans); | |
8915 | goto out_notrans; | |
8916 | } | |
76dda93c | 8917 | |
4df27c4d YZ |
8918 | if (dest != root) |
8919 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8920 | |
a5719521 YZ |
8921 | ret = btrfs_set_inode_index(new_dir, &index); |
8922 | if (ret) | |
8923 | goto out_fail; | |
5a3f23d5 | 8924 | |
67de1176 | 8925 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 8926 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 8927 | /* force full log commit if subvolume involved. */ |
995946dd | 8928 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 8929 | } else { |
a5719521 YZ |
8930 | ret = btrfs_insert_inode_ref(trans, dest, |
8931 | new_dentry->d_name.name, | |
8932 | new_dentry->d_name.len, | |
33345d01 LZ |
8933 | old_ino, |
8934 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8935 | if (ret) |
8936 | goto out_fail; | |
4df27c4d YZ |
8937 | /* |
8938 | * this is an ugly little race, but the rename is required | |
8939 | * to make sure that if we crash, the inode is either at the | |
8940 | * old name or the new one. pinning the log transaction lets | |
8941 | * us make sure we don't allow a log commit to come in after | |
8942 | * we unlink the name but before we add the new name back in. | |
8943 | */ | |
8944 | btrfs_pin_log_trans(root); | |
8945 | } | |
5a3f23d5 | 8946 | |
0c4d2d95 JB |
8947 | inode_inc_iversion(old_dir); |
8948 | inode_inc_iversion(new_dir); | |
8949 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8950 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8951 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8952 | old_inode->i_ctime = ctime; | |
5f39d397 | 8953 | |
12fcfd22 CM |
8954 | if (old_dentry->d_parent != new_dentry->d_parent) |
8955 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8956 | ||
33345d01 | 8957 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8958 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8959 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8960 | old_dentry->d_name.name, | |
8961 | old_dentry->d_name.len); | |
8962 | } else { | |
92986796 AV |
8963 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8964 | old_dentry->d_inode, | |
8965 | old_dentry->d_name.name, | |
8966 | old_dentry->d_name.len); | |
8967 | if (!ret) | |
8968 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8969 | } |
79787eaa JM |
8970 | if (ret) { |
8971 | btrfs_abort_transaction(trans, root, ret); | |
8972 | goto out_fail; | |
8973 | } | |
39279cc3 CM |
8974 | |
8975 | if (new_inode) { | |
0c4d2d95 | 8976 | inode_inc_iversion(new_inode); |
39279cc3 | 8977 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8978 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8979 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8980 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8981 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8982 | root_objectid, | |
8983 | new_dentry->d_name.name, | |
8984 | new_dentry->d_name.len); | |
8985 | BUG_ON(new_inode->i_nlink == 0); | |
8986 | } else { | |
8987 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8988 | new_dentry->d_inode, | |
8989 | new_dentry->d_name.name, | |
8990 | new_dentry->d_name.len); | |
8991 | } | |
4ef31a45 | 8992 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8993 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8994 | if (ret) { |
8995 | btrfs_abort_transaction(trans, root, ret); | |
8996 | goto out_fail; | |
8997 | } | |
39279cc3 | 8998 | } |
aec7477b | 8999 | |
4df27c4d YZ |
9000 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9001 | new_dentry->d_name.name, | |
a5719521 | 9002 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
9003 | if (ret) { |
9004 | btrfs_abort_transaction(trans, root, ret); | |
9005 | goto out_fail; | |
9006 | } | |
39279cc3 | 9007 | |
67de1176 MX |
9008 | if (old_inode->i_nlink == 1) |
9009 | BTRFS_I(old_inode)->dir_index = index; | |
9010 | ||
33345d01 | 9011 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 9012 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 9013 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
9014 | btrfs_end_log_trans(root); |
9015 | } | |
39279cc3 | 9016 | out_fail: |
7ad85bb7 | 9017 | btrfs_end_transaction(trans, root); |
b44c59a8 | 9018 | out_notrans: |
33345d01 | 9019 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9020 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 9021 | |
39279cc3 CM |
9022 | return ret; |
9023 | } | |
9024 | ||
80ace85c MS |
9025 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9026 | struct inode *new_dir, struct dentry *new_dentry, | |
9027 | unsigned int flags) | |
9028 | { | |
9029 | if (flags & ~RENAME_NOREPLACE) | |
9030 | return -EINVAL; | |
9031 | ||
9032 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry); | |
9033 | } | |
9034 | ||
8ccf6f19 MX |
9035 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9036 | { | |
9037 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9038 | struct inode *inode; |
8ccf6f19 MX |
9039 | |
9040 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9041 | work); | |
9f23e289 JB |
9042 | inode = delalloc_work->inode; |
9043 | if (delalloc_work->wait) { | |
9044 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
9045 | } else { | |
9046 | filemap_flush(inode->i_mapping); | |
9047 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9048 | &BTRFS_I(inode)->runtime_flags)) | |
9049 | filemap_flush(inode->i_mapping); | |
9050 | } | |
8ccf6f19 MX |
9051 | |
9052 | if (delalloc_work->delay_iput) | |
9f23e289 | 9053 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9054 | else |
9f23e289 | 9055 | iput(inode); |
8ccf6f19 MX |
9056 | complete(&delalloc_work->completion); |
9057 | } | |
9058 | ||
9059 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
9060 | int wait, int delay_iput) | |
9061 | { | |
9062 | struct btrfs_delalloc_work *work; | |
9063 | ||
9064 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
9065 | if (!work) | |
9066 | return NULL; | |
9067 | ||
9068 | init_completion(&work->completion); | |
9069 | INIT_LIST_HEAD(&work->list); | |
9070 | work->inode = inode; | |
9071 | work->wait = wait; | |
9072 | work->delay_iput = delay_iput; | |
9e0af237 LB |
9073 | WARN_ON_ONCE(!inode); |
9074 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9075 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9076 | |
9077 | return work; | |
9078 | } | |
9079 | ||
9080 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9081 | { | |
9082 | wait_for_completion(&work->completion); | |
9083 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
9084 | } | |
9085 | ||
d352ac68 CM |
9086 | /* |
9087 | * some fairly slow code that needs optimization. This walks the list | |
9088 | * of all the inodes with pending delalloc and forces them to disk. | |
9089 | */ | |
6c255e67 MX |
9090 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
9091 | int nr) | |
ea8c2819 | 9092 | { |
ea8c2819 | 9093 | struct btrfs_inode *binode; |
5b21f2ed | 9094 | struct inode *inode; |
8ccf6f19 MX |
9095 | struct btrfs_delalloc_work *work, *next; |
9096 | struct list_head works; | |
1eafa6c7 | 9097 | struct list_head splice; |
8ccf6f19 | 9098 | int ret = 0; |
ea8c2819 | 9099 | |
8ccf6f19 | 9100 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9101 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9102 | |
573bfb72 | 9103 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9104 | spin_lock(&root->delalloc_lock); |
9105 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9106 | while (!list_empty(&splice)) { |
9107 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9108 | delalloc_inodes); |
1eafa6c7 | 9109 | |
eb73c1b7 MX |
9110 | list_move_tail(&binode->delalloc_inodes, |
9111 | &root->delalloc_inodes); | |
5b21f2ed | 9112 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9113 | if (!inode) { |
eb73c1b7 | 9114 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9115 | continue; |
df0af1a5 | 9116 | } |
eb73c1b7 | 9117 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
9118 | |
9119 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
5d99a998 | 9120 | if (!work) { |
f4ab9ea7 JB |
9121 | if (delay_iput) |
9122 | btrfs_add_delayed_iput(inode); | |
9123 | else | |
9124 | iput(inode); | |
1eafa6c7 | 9125 | ret = -ENOMEM; |
a1ecaabb | 9126 | goto out; |
5b21f2ed | 9127 | } |
1eafa6c7 | 9128 | list_add_tail(&work->list, &works); |
a44903ab QW |
9129 | btrfs_queue_work(root->fs_info->flush_workers, |
9130 | &work->work); | |
6c255e67 MX |
9131 | ret++; |
9132 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 9133 | goto out; |
5b21f2ed | 9134 | cond_resched(); |
eb73c1b7 | 9135 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9136 | } |
eb73c1b7 | 9137 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9138 | |
a1ecaabb | 9139 | out: |
eb73c1b7 MX |
9140 | list_for_each_entry_safe(work, next, &works, list) { |
9141 | list_del_init(&work->list); | |
9142 | btrfs_wait_and_free_delalloc_work(work); | |
9143 | } | |
9144 | ||
9145 | if (!list_empty_careful(&splice)) { | |
9146 | spin_lock(&root->delalloc_lock); | |
9147 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9148 | spin_unlock(&root->delalloc_lock); | |
9149 | } | |
573bfb72 | 9150 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9151 | return ret; |
9152 | } | |
1eafa6c7 | 9153 | |
eb73c1b7 MX |
9154 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
9155 | { | |
9156 | int ret; | |
1eafa6c7 | 9157 | |
2c21b4d7 | 9158 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
9159 | return -EROFS; |
9160 | ||
6c255e67 MX |
9161 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
9162 | if (ret > 0) | |
9163 | ret = 0; | |
eb73c1b7 MX |
9164 | /* |
9165 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
9166 | * we have to make sure the IO is actually started and that |
9167 | * ordered extents get created before we return | |
9168 | */ | |
9169 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 9170 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 9171 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 9172 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
9173 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
9174 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
9175 | } |
9176 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
9177 | return ret; |
9178 | } | |
9179 | ||
6c255e67 MX |
9180 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
9181 | int nr) | |
eb73c1b7 MX |
9182 | { |
9183 | struct btrfs_root *root; | |
9184 | struct list_head splice; | |
9185 | int ret; | |
9186 | ||
2c21b4d7 | 9187 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9188 | return -EROFS; |
9189 | ||
9190 | INIT_LIST_HEAD(&splice); | |
9191 | ||
573bfb72 | 9192 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9193 | spin_lock(&fs_info->delalloc_root_lock); |
9194 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9195 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
9196 | root = list_first_entry(&splice, struct btrfs_root, |
9197 | delalloc_root); | |
9198 | root = btrfs_grab_fs_root(root); | |
9199 | BUG_ON(!root); | |
9200 | list_move_tail(&root->delalloc_root, | |
9201 | &fs_info->delalloc_roots); | |
9202 | spin_unlock(&fs_info->delalloc_root_lock); | |
9203 | ||
6c255e67 | 9204 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 9205 | btrfs_put_fs_root(root); |
6c255e67 | 9206 | if (ret < 0) |
eb73c1b7 MX |
9207 | goto out; |
9208 | ||
6c255e67 MX |
9209 | if (nr != -1) { |
9210 | nr -= ret; | |
9211 | WARN_ON(nr < 0); | |
9212 | } | |
eb73c1b7 | 9213 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9214 | } |
eb73c1b7 | 9215 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9216 | |
6c255e67 | 9217 | ret = 0; |
eb73c1b7 MX |
9218 | atomic_inc(&fs_info->async_submit_draining); |
9219 | while (atomic_read(&fs_info->nr_async_submits) || | |
9220 | atomic_read(&fs_info->async_delalloc_pages)) { | |
9221 | wait_event(fs_info->async_submit_wait, | |
9222 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
9223 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
9224 | } | |
9225 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 9226 | out: |
1eafa6c7 | 9227 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
9228 | spin_lock(&fs_info->delalloc_root_lock); |
9229 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9230 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9231 | } |
573bfb72 | 9232 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9233 | return ret; |
ea8c2819 CM |
9234 | } |
9235 | ||
39279cc3 CM |
9236 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9237 | const char *symname) | |
9238 | { | |
9239 | struct btrfs_trans_handle *trans; | |
9240 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9241 | struct btrfs_path *path; | |
9242 | struct btrfs_key key; | |
1832a6d5 | 9243 | struct inode *inode = NULL; |
39279cc3 CM |
9244 | int err; |
9245 | int drop_inode = 0; | |
9246 | u64 objectid; | |
67871254 | 9247 | u64 index = 0; |
39279cc3 CM |
9248 | int name_len; |
9249 | int datasize; | |
5f39d397 | 9250 | unsigned long ptr; |
39279cc3 | 9251 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9252 | struct extent_buffer *leaf; |
39279cc3 | 9253 | |
f06becc4 | 9254 | name_len = strlen(symname); |
39279cc3 CM |
9255 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
9256 | return -ENAMETOOLONG; | |
1832a6d5 | 9257 | |
9ed74f2d JB |
9258 | /* |
9259 | * 2 items for inode item and ref | |
9260 | * 2 items for dir items | |
9261 | * 1 item for xattr if selinux is on | |
9262 | */ | |
a22285a6 YZ |
9263 | trans = btrfs_start_transaction(root, 5); |
9264 | if (IS_ERR(trans)) | |
9265 | return PTR_ERR(trans); | |
1832a6d5 | 9266 | |
581bb050 LZ |
9267 | err = btrfs_find_free_ino(root, &objectid); |
9268 | if (err) | |
9269 | goto out_unlock; | |
9270 | ||
aec7477b | 9271 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 9272 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 9273 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
9274 | if (IS_ERR(inode)) { |
9275 | err = PTR_ERR(inode); | |
39279cc3 | 9276 | goto out_unlock; |
7cf96da3 | 9277 | } |
39279cc3 | 9278 | |
ad19db71 CS |
9279 | /* |
9280 | * If the active LSM wants to access the inode during | |
9281 | * d_instantiate it needs these. Smack checks to see | |
9282 | * if the filesystem supports xattrs by looking at the | |
9283 | * ops vector. | |
9284 | */ | |
9285 | inode->i_fop = &btrfs_file_operations; | |
9286 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f CM |
9287 | inode->i_mapping->a_ops = &btrfs_aops; |
9288 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
9289 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
9290 | ||
9291 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9292 | if (err) | |
9293 | goto out_unlock_inode; | |
ad19db71 | 9294 | |
a1b075d2 | 9295 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 9296 | if (err) |
b0d5d10f | 9297 | goto out_unlock_inode; |
39279cc3 CM |
9298 | |
9299 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
9300 | if (!path) { |
9301 | err = -ENOMEM; | |
b0d5d10f | 9302 | goto out_unlock_inode; |
d8926bb3 | 9303 | } |
33345d01 | 9304 | key.objectid = btrfs_ino(inode); |
39279cc3 | 9305 | key.offset = 0; |
962a298f | 9306 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9307 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9308 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9309 | datasize); | |
54aa1f4d | 9310 | if (err) { |
b0839166 | 9311 | btrfs_free_path(path); |
b0d5d10f | 9312 | goto out_unlock_inode; |
54aa1f4d | 9313 | } |
5f39d397 CM |
9314 | leaf = path->nodes[0]; |
9315 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9316 | struct btrfs_file_extent_item); | |
9317 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9318 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9319 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9320 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9321 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9322 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9323 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9324 | ||
39279cc3 | 9325 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9326 | write_extent_buffer(leaf, symname, ptr, name_len); |
9327 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9328 | btrfs_free_path(path); |
5f39d397 | 9329 | |
39279cc3 CM |
9330 | inode->i_op = &btrfs_symlink_inode_operations; |
9331 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 9332 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 9333 | inode_set_bytes(inode, name_len); |
f06becc4 | 9334 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 9335 | err = btrfs_update_inode(trans, root, inode); |
b0d5d10f | 9336 | if (err) { |
54aa1f4d | 9337 | drop_inode = 1; |
b0d5d10f CM |
9338 | goto out_unlock_inode; |
9339 | } | |
9340 | ||
9341 | unlock_new_inode(inode); | |
9342 | d_instantiate(dentry, inode); | |
39279cc3 CM |
9343 | |
9344 | out_unlock: | |
7ad85bb7 | 9345 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
9346 | if (drop_inode) { |
9347 | inode_dec_link_count(inode); | |
9348 | iput(inode); | |
9349 | } | |
b53d3f5d | 9350 | btrfs_btree_balance_dirty(root); |
39279cc3 | 9351 | return err; |
b0d5d10f CM |
9352 | |
9353 | out_unlock_inode: | |
9354 | drop_inode = 1; | |
9355 | unlock_new_inode(inode); | |
9356 | goto out_unlock; | |
39279cc3 | 9357 | } |
16432985 | 9358 | |
0af3d00b JB |
9359 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9360 | u64 start, u64 num_bytes, u64 min_size, | |
9361 | loff_t actual_len, u64 *alloc_hint, | |
9362 | struct btrfs_trans_handle *trans) | |
d899e052 | 9363 | { |
5dc562c5 JB |
9364 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9365 | struct extent_map *em; | |
d899e052 YZ |
9366 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9367 | struct btrfs_key ins; | |
d899e052 | 9368 | u64 cur_offset = start; |
55a61d1d | 9369 | u64 i_size; |
154ea289 | 9370 | u64 cur_bytes; |
d899e052 | 9371 | int ret = 0; |
0af3d00b | 9372 | bool own_trans = true; |
d899e052 | 9373 | |
0af3d00b JB |
9374 | if (trans) |
9375 | own_trans = false; | |
d899e052 | 9376 | while (num_bytes > 0) { |
0af3d00b JB |
9377 | if (own_trans) { |
9378 | trans = btrfs_start_transaction(root, 3); | |
9379 | if (IS_ERR(trans)) { | |
9380 | ret = PTR_ERR(trans); | |
9381 | break; | |
9382 | } | |
5a303d5d YZ |
9383 | } |
9384 | ||
154ea289 CM |
9385 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
9386 | cur_bytes = max(cur_bytes, min_size); | |
00361589 | 9387 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 9388 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 9389 | if (ret) { |
0af3d00b JB |
9390 | if (own_trans) |
9391 | btrfs_end_transaction(trans, root); | |
a22285a6 | 9392 | break; |
d899e052 | 9393 | } |
5a303d5d | 9394 | |
d899e052 YZ |
9395 | ret = insert_reserved_file_extent(trans, inode, |
9396 | cur_offset, ins.objectid, | |
9397 | ins.offset, ins.offset, | |
920bbbfb | 9398 | ins.offset, 0, 0, 0, |
d899e052 | 9399 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 9400 | if (ret) { |
857cc2fc | 9401 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 9402 | ins.offset, 0); |
79787eaa JM |
9403 | btrfs_abort_transaction(trans, root, ret); |
9404 | if (own_trans) | |
9405 | btrfs_end_transaction(trans, root); | |
9406 | break; | |
9407 | } | |
a1ed835e CM |
9408 | btrfs_drop_extent_cache(inode, cur_offset, |
9409 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 9410 | |
5dc562c5 JB |
9411 | em = alloc_extent_map(); |
9412 | if (!em) { | |
9413 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9414 | &BTRFS_I(inode)->runtime_flags); | |
9415 | goto next; | |
9416 | } | |
9417 | ||
9418 | em->start = cur_offset; | |
9419 | em->orig_start = cur_offset; | |
9420 | em->len = ins.offset; | |
9421 | em->block_start = ins.objectid; | |
9422 | em->block_len = ins.offset; | |
b4939680 | 9423 | em->orig_block_len = ins.offset; |
cc95bef6 | 9424 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9425 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
9426 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
9427 | em->generation = trans->transid; | |
9428 | ||
9429 | while (1) { | |
9430 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9431 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9432 | write_unlock(&em_tree->lock); |
9433 | if (ret != -EEXIST) | |
9434 | break; | |
9435 | btrfs_drop_extent_cache(inode, cur_offset, | |
9436 | cur_offset + ins.offset - 1, | |
9437 | 0); | |
9438 | } | |
9439 | free_extent_map(em); | |
9440 | next: | |
d899e052 YZ |
9441 | num_bytes -= ins.offset; |
9442 | cur_offset += ins.offset; | |
efa56464 | 9443 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9444 | |
0c4d2d95 | 9445 | inode_inc_iversion(inode); |
d899e052 | 9446 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 9447 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9448 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9449 | (actual_len > inode->i_size) && |
9450 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9451 | if (cur_offset > actual_len) |
55a61d1d | 9452 | i_size = actual_len; |
d1ea6a61 | 9453 | else |
55a61d1d JB |
9454 | i_size = cur_offset; |
9455 | i_size_write(inode, i_size); | |
9456 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
9457 | } |
9458 | ||
d899e052 | 9459 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
9460 | |
9461 | if (ret) { | |
9462 | btrfs_abort_transaction(trans, root, ret); | |
9463 | if (own_trans) | |
9464 | btrfs_end_transaction(trans, root); | |
9465 | break; | |
9466 | } | |
d899e052 | 9467 | |
0af3d00b JB |
9468 | if (own_trans) |
9469 | btrfs_end_transaction(trans, root); | |
5a303d5d | 9470 | } |
d899e052 YZ |
9471 | return ret; |
9472 | } | |
9473 | ||
0af3d00b JB |
9474 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9475 | u64 start, u64 num_bytes, u64 min_size, | |
9476 | loff_t actual_len, u64 *alloc_hint) | |
9477 | { | |
9478 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9479 | min_size, actual_len, alloc_hint, | |
9480 | NULL); | |
9481 | } | |
9482 | ||
9483 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9484 | struct btrfs_trans_handle *trans, int mode, | |
9485 | u64 start, u64 num_bytes, u64 min_size, | |
9486 | loff_t actual_len, u64 *alloc_hint) | |
9487 | { | |
9488 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9489 | min_size, actual_len, alloc_hint, trans); | |
9490 | } | |
9491 | ||
e6dcd2dc CM |
9492 | static int btrfs_set_page_dirty(struct page *page) |
9493 | { | |
e6dcd2dc CM |
9494 | return __set_page_dirty_nobuffers(page); |
9495 | } | |
9496 | ||
10556cb2 | 9497 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9498 | { |
b83cc969 | 9499 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9500 | umode_t mode = inode->i_mode; |
b83cc969 | 9501 | |
cb6db4e5 JM |
9502 | if (mask & MAY_WRITE && |
9503 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9504 | if (btrfs_root_readonly(root)) | |
9505 | return -EROFS; | |
9506 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9507 | return -EACCES; | |
9508 | } | |
2830ba7f | 9509 | return generic_permission(inode, mask); |
fdebe2bd | 9510 | } |
39279cc3 | 9511 | |
ef3b9af5 FM |
9512 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9513 | { | |
9514 | struct btrfs_trans_handle *trans; | |
9515 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9516 | struct inode *inode = NULL; | |
9517 | u64 objectid; | |
9518 | u64 index; | |
9519 | int ret = 0; | |
9520 | ||
9521 | /* | |
9522 | * 5 units required for adding orphan entry | |
9523 | */ | |
9524 | trans = btrfs_start_transaction(root, 5); | |
9525 | if (IS_ERR(trans)) | |
9526 | return PTR_ERR(trans); | |
9527 | ||
9528 | ret = btrfs_find_free_ino(root, &objectid); | |
9529 | if (ret) | |
9530 | goto out; | |
9531 | ||
9532 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
9533 | btrfs_ino(dir), objectid, mode, &index); | |
9534 | if (IS_ERR(inode)) { | |
9535 | ret = PTR_ERR(inode); | |
9536 | inode = NULL; | |
9537 | goto out; | |
9538 | } | |
9539 | ||
ef3b9af5 FM |
9540 | inode->i_fop = &btrfs_file_operations; |
9541 | inode->i_op = &btrfs_file_inode_operations; | |
9542 | ||
9543 | inode->i_mapping->a_ops = &btrfs_aops; | |
9544 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
9545 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
9546 | ||
b0d5d10f CM |
9547 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9548 | if (ret) | |
9549 | goto out_inode; | |
9550 | ||
9551 | ret = btrfs_update_inode(trans, root, inode); | |
9552 | if (ret) | |
9553 | goto out_inode; | |
ef3b9af5 FM |
9554 | ret = btrfs_orphan_add(trans, inode); |
9555 | if (ret) | |
b0d5d10f | 9556 | goto out_inode; |
ef3b9af5 | 9557 | |
5762b5c9 FM |
9558 | /* |
9559 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9560 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9561 | * through: | |
9562 | * | |
9563 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9564 | */ | |
9565 | set_nlink(inode, 1); | |
b0d5d10f | 9566 | unlock_new_inode(inode); |
ef3b9af5 FM |
9567 | d_tmpfile(dentry, inode); |
9568 | mark_inode_dirty(inode); | |
9569 | ||
9570 | out: | |
9571 | btrfs_end_transaction(trans, root); | |
9572 | if (ret) | |
9573 | iput(inode); | |
9574 | btrfs_balance_delayed_items(root); | |
9575 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 9576 | return ret; |
b0d5d10f CM |
9577 | |
9578 | out_inode: | |
9579 | unlock_new_inode(inode); | |
9580 | goto out; | |
9581 | ||
ef3b9af5 FM |
9582 | } |
9583 | ||
b38ef71c FM |
9584 | /* Inspired by filemap_check_errors() */ |
9585 | int btrfs_inode_check_errors(struct inode *inode) | |
9586 | { | |
9587 | int ret = 0; | |
9588 | ||
9589 | if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) && | |
9590 | test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags)) | |
9591 | ret = -ENOSPC; | |
9592 | if (test_bit(AS_EIO, &inode->i_mapping->flags) && | |
9593 | test_and_clear_bit(AS_EIO, &inode->i_mapping->flags)) | |
9594 | ret = -EIO; | |
9595 | ||
9596 | return ret; | |
9597 | } | |
9598 | ||
6e1d5dcc | 9599 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 9600 | .getattr = btrfs_getattr, |
39279cc3 CM |
9601 | .lookup = btrfs_lookup, |
9602 | .create = btrfs_create, | |
9603 | .unlink = btrfs_unlink, | |
9604 | .link = btrfs_link, | |
9605 | .mkdir = btrfs_mkdir, | |
9606 | .rmdir = btrfs_rmdir, | |
80ace85c | 9607 | .rename2 = btrfs_rename2, |
39279cc3 CM |
9608 | .symlink = btrfs_symlink, |
9609 | .setattr = btrfs_setattr, | |
618e21d5 | 9610 | .mknod = btrfs_mknod, |
95819c05 CH |
9611 | .setxattr = btrfs_setxattr, |
9612 | .getxattr = btrfs_getxattr, | |
5103e947 | 9613 | .listxattr = btrfs_listxattr, |
95819c05 | 9614 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9615 | .permission = btrfs_permission, |
4e34e719 | 9616 | .get_acl = btrfs_get_acl, |
996a710d | 9617 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9618 | .update_time = btrfs_update_time, |
ef3b9af5 | 9619 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 9620 | }; |
6e1d5dcc | 9621 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 9622 | .lookup = btrfs_lookup, |
fdebe2bd | 9623 | .permission = btrfs_permission, |
4e34e719 | 9624 | .get_acl = btrfs_get_acl, |
996a710d | 9625 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9626 | .update_time = btrfs_update_time, |
39279cc3 | 9627 | }; |
76dda93c | 9628 | |
828c0950 | 9629 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
9630 | .llseek = generic_file_llseek, |
9631 | .read = generic_read_dir, | |
9cdda8d3 | 9632 | .iterate = btrfs_real_readdir, |
34287aa3 | 9633 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 9634 | #ifdef CONFIG_COMPAT |
34287aa3 | 9635 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 9636 | #endif |
6bf13c0c | 9637 | .release = btrfs_release_file, |
e02119d5 | 9638 | .fsync = btrfs_sync_file, |
39279cc3 CM |
9639 | }; |
9640 | ||
d1310b2e | 9641 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 9642 | .fill_delalloc = run_delalloc_range, |
065631f6 | 9643 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 9644 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 9645 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 9646 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 9647 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
9648 | .set_bit_hook = btrfs_set_bit_hook, |
9649 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
9650 | .merge_extent_hook = btrfs_merge_extent_hook, |
9651 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
9652 | }; |
9653 | ||
35054394 CM |
9654 | /* |
9655 | * btrfs doesn't support the bmap operation because swapfiles | |
9656 | * use bmap to make a mapping of extents in the file. They assume | |
9657 | * these extents won't change over the life of the file and they | |
9658 | * use the bmap result to do IO directly to the drive. | |
9659 | * | |
9660 | * the btrfs bmap call would return logical addresses that aren't | |
9661 | * suitable for IO and they also will change frequently as COW | |
9662 | * operations happen. So, swapfile + btrfs == corruption. | |
9663 | * | |
9664 | * For now we're avoiding this by dropping bmap. | |
9665 | */ | |
7f09410b | 9666 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
9667 | .readpage = btrfs_readpage, |
9668 | .writepage = btrfs_writepage, | |
b293f02e | 9669 | .writepages = btrfs_writepages, |
3ab2fb5a | 9670 | .readpages = btrfs_readpages, |
16432985 | 9671 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
9672 | .invalidatepage = btrfs_invalidatepage, |
9673 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 9674 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 9675 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
9676 | }; |
9677 | ||
7f09410b | 9678 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
9679 | .readpage = btrfs_readpage, |
9680 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
9681 | .invalidatepage = btrfs_invalidatepage, |
9682 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
9683 | }; |
9684 | ||
6e1d5dcc | 9685 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
9686 | .getattr = btrfs_getattr, |
9687 | .setattr = btrfs_setattr, | |
95819c05 CH |
9688 | .setxattr = btrfs_setxattr, |
9689 | .getxattr = btrfs_getxattr, | |
5103e947 | 9690 | .listxattr = btrfs_listxattr, |
95819c05 | 9691 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9692 | .permission = btrfs_permission, |
1506fcc8 | 9693 | .fiemap = btrfs_fiemap, |
4e34e719 | 9694 | .get_acl = btrfs_get_acl, |
996a710d | 9695 | .set_acl = btrfs_set_acl, |
e41f941a | 9696 | .update_time = btrfs_update_time, |
39279cc3 | 9697 | }; |
6e1d5dcc | 9698 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
9699 | .getattr = btrfs_getattr, |
9700 | .setattr = btrfs_setattr, | |
fdebe2bd | 9701 | .permission = btrfs_permission, |
95819c05 CH |
9702 | .setxattr = btrfs_setxattr, |
9703 | .getxattr = btrfs_getxattr, | |
33268eaf | 9704 | .listxattr = btrfs_listxattr, |
95819c05 | 9705 | .removexattr = btrfs_removexattr, |
4e34e719 | 9706 | .get_acl = btrfs_get_acl, |
996a710d | 9707 | .set_acl = btrfs_set_acl, |
e41f941a | 9708 | .update_time = btrfs_update_time, |
618e21d5 | 9709 | }; |
6e1d5dcc | 9710 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
9711 | .readlink = generic_readlink, |
9712 | .follow_link = page_follow_link_light, | |
9713 | .put_link = page_put_link, | |
f209561a | 9714 | .getattr = btrfs_getattr, |
22c44fe6 | 9715 | .setattr = btrfs_setattr, |
fdebe2bd | 9716 | .permission = btrfs_permission, |
0279b4cd JO |
9717 | .setxattr = btrfs_setxattr, |
9718 | .getxattr = btrfs_getxattr, | |
9719 | .listxattr = btrfs_listxattr, | |
9720 | .removexattr = btrfs_removexattr, | |
e41f941a | 9721 | .update_time = btrfs_update_time, |
39279cc3 | 9722 | }; |
76dda93c | 9723 | |
82d339d9 | 9724 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 9725 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 9726 | .d_release = btrfs_dentry_release, |
76dda93c | 9727 | }; |