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> |
4b4e25f2 | 46 | #include "compat.h" |
39279cc3 CM |
47 | #include "ctree.h" |
48 | #include "disk-io.h" | |
49 | #include "transaction.h" | |
50 | #include "btrfs_inode.h" | |
39279cc3 | 51 | #include "print-tree.h" |
e6dcd2dc | 52 | #include "ordered-data.h" |
95819c05 | 53 | #include "xattr.h" |
e02119d5 | 54 | #include "tree-log.h" |
4a54c8c1 | 55 | #include "volumes.h" |
c8b97818 | 56 | #include "compression.h" |
b4ce94de | 57 | #include "locking.h" |
dc89e982 | 58 | #include "free-space-cache.h" |
581bb050 | 59 | #include "inode-map.h" |
38c227d8 | 60 | #include "backref.h" |
f23b5a59 | 61 | #include "hash.h" |
39279cc3 CM |
62 | |
63 | struct btrfs_iget_args { | |
64 | u64 ino; | |
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 | */ | |
d397712b | 128 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
129 | struct btrfs_root *root, struct inode *inode, |
130 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 131 | int compress_type, |
c8b97818 CM |
132 | struct page **compressed_pages) |
133 | { | |
134 | struct btrfs_key key; | |
135 | struct btrfs_path *path; | |
136 | struct extent_buffer *leaf; | |
137 | struct page *page = NULL; | |
138 | char *kaddr; | |
139 | unsigned long ptr; | |
140 | struct btrfs_file_extent_item *ei; | |
141 | int err = 0; | |
142 | int ret; | |
143 | size_t cur_size = size; | |
144 | size_t datasize; | |
145 | unsigned long offset; | |
c8b97818 | 146 | |
fe3f566c | 147 | if (compressed_size && compressed_pages) |
c8b97818 | 148 | cur_size = compressed_size; |
c8b97818 | 149 | |
d397712b CM |
150 | path = btrfs_alloc_path(); |
151 | if (!path) | |
c8b97818 CM |
152 | return -ENOMEM; |
153 | ||
b9473439 | 154 | path->leave_spinning = 1; |
c8b97818 | 155 | |
33345d01 | 156 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
157 | key.offset = start; |
158 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
159 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
160 | ||
161 | inode_add_bytes(inode, size); | |
162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
163 | datasize); | |
c8b97818 CM |
164 | if (ret) { |
165 | err = ret; | |
c8b97818 CM |
166 | goto fail; |
167 | } | |
168 | leaf = path->nodes[0]; | |
169 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
170 | struct btrfs_file_extent_item); | |
171 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
172 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
173 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
174 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
175 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
176 | ptr = btrfs_file_extent_inline_start(ei); | |
177 | ||
261507a0 | 178 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
179 | struct page *cpage; |
180 | int i = 0; | |
d397712b | 181 | while (compressed_size > 0) { |
c8b97818 | 182 | cpage = compressed_pages[i]; |
5b050f04 | 183 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
184 | PAGE_CACHE_SIZE); |
185 | ||
7ac687d9 | 186 | kaddr = kmap_atomic(cpage); |
c8b97818 | 187 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 188 | kunmap_atomic(kaddr); |
c8b97818 CM |
189 | |
190 | i++; | |
191 | ptr += cur_size; | |
192 | compressed_size -= cur_size; | |
193 | } | |
194 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 195 | compress_type); |
c8b97818 CM |
196 | } else { |
197 | page = find_get_page(inode->i_mapping, | |
198 | start >> PAGE_CACHE_SHIFT); | |
199 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 200 | kaddr = kmap_atomic(page); |
c8b97818 CM |
201 | offset = start & (PAGE_CACHE_SIZE - 1); |
202 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 203 | kunmap_atomic(kaddr); |
c8b97818 CM |
204 | page_cache_release(page); |
205 | } | |
206 | btrfs_mark_buffer_dirty(leaf); | |
207 | btrfs_free_path(path); | |
208 | ||
c2167754 YZ |
209 | /* |
210 | * we're an inline extent, so nobody can | |
211 | * extend the file past i_size without locking | |
212 | * a page we already have locked. | |
213 | * | |
214 | * We must do any isize and inode updates | |
215 | * before we unlock the pages. Otherwise we | |
216 | * could end up racing with unlink. | |
217 | */ | |
c8b97818 | 218 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 219 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 220 | |
79787eaa | 221 | return ret; |
c8b97818 CM |
222 | fail: |
223 | btrfs_free_path(path); | |
224 | return err; | |
225 | } | |
226 | ||
227 | ||
228 | /* | |
229 | * conditionally insert an inline extent into the file. This | |
230 | * does the checks required to make sure the data is small enough | |
231 | * to fit as an inline extent. | |
232 | */ | |
00361589 JB |
233 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
234 | struct inode *inode, u64 start, | |
235 | u64 end, size_t compressed_size, | |
236 | int compress_type, | |
237 | struct page **compressed_pages) | |
c8b97818 | 238 | { |
00361589 | 239 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
240 | u64 isize = i_size_read(inode); |
241 | u64 actual_end = min(end + 1, isize); | |
242 | u64 inline_len = actual_end - start; | |
fda2832f | 243 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
244 | u64 data_len = inline_len; |
245 | int ret; | |
246 | ||
247 | if (compressed_size) | |
248 | data_len = compressed_size; | |
249 | ||
250 | if (start > 0 || | |
70b99e69 | 251 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
252 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
253 | (!compressed_size && | |
254 | (actual_end & (root->sectorsize - 1)) == 0) || | |
255 | end + 1 < isize || | |
256 | data_len > root->fs_info->max_inline) { | |
257 | return 1; | |
258 | } | |
259 | ||
00361589 JB |
260 | trans = btrfs_join_transaction(root); |
261 | if (IS_ERR(trans)) | |
262 | return PTR_ERR(trans); | |
263 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
264 | ||
2671485d | 265 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
00361589 JB |
266 | if (ret) { |
267 | btrfs_abort_transaction(trans, root, ret); | |
268 | goto out; | |
269 | } | |
c8b97818 CM |
270 | |
271 | if (isize > actual_end) | |
272 | inline_len = min_t(u64, isize, actual_end); | |
273 | ret = insert_inline_extent(trans, root, inode, start, | |
274 | inline_len, compressed_size, | |
fe3f566c | 275 | compress_type, compressed_pages); |
2adcac1a | 276 | if (ret && ret != -ENOSPC) { |
79787eaa | 277 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 278 | goto out; |
2adcac1a | 279 | } else if (ret == -ENOSPC) { |
00361589 JB |
280 | ret = 1; |
281 | goto out; | |
79787eaa | 282 | } |
2adcac1a | 283 | |
bdc20e67 | 284 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 285 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 286 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 JB |
287 | out: |
288 | btrfs_end_transaction(trans, root); | |
289 | return ret; | |
c8b97818 CM |
290 | } |
291 | ||
771ed689 CM |
292 | struct async_extent { |
293 | u64 start; | |
294 | u64 ram_size; | |
295 | u64 compressed_size; | |
296 | struct page **pages; | |
297 | unsigned long nr_pages; | |
261507a0 | 298 | int compress_type; |
771ed689 CM |
299 | struct list_head list; |
300 | }; | |
301 | ||
302 | struct async_cow { | |
303 | struct inode *inode; | |
304 | struct btrfs_root *root; | |
305 | struct page *locked_page; | |
306 | u64 start; | |
307 | u64 end; | |
308 | struct list_head extents; | |
309 | struct btrfs_work work; | |
310 | }; | |
311 | ||
312 | static noinline int add_async_extent(struct async_cow *cow, | |
313 | u64 start, u64 ram_size, | |
314 | u64 compressed_size, | |
315 | struct page **pages, | |
261507a0 LZ |
316 | unsigned long nr_pages, |
317 | int compress_type) | |
771ed689 CM |
318 | { |
319 | struct async_extent *async_extent; | |
320 | ||
321 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 322 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
323 | async_extent->start = start; |
324 | async_extent->ram_size = ram_size; | |
325 | async_extent->compressed_size = compressed_size; | |
326 | async_extent->pages = pages; | |
327 | async_extent->nr_pages = nr_pages; | |
261507a0 | 328 | async_extent->compress_type = compress_type; |
771ed689 CM |
329 | list_add_tail(&async_extent->list, &cow->extents); |
330 | return 0; | |
331 | } | |
332 | ||
d352ac68 | 333 | /* |
771ed689 CM |
334 | * we create compressed extents in two phases. The first |
335 | * phase compresses a range of pages that have already been | |
336 | * locked (both pages and state bits are locked). | |
c8b97818 | 337 | * |
771ed689 CM |
338 | * This is done inside an ordered work queue, and the compression |
339 | * is spread across many cpus. The actual IO submission is step | |
340 | * two, and the ordered work queue takes care of making sure that | |
341 | * happens in the same order things were put onto the queue by | |
342 | * writepages and friends. | |
c8b97818 | 343 | * |
771ed689 CM |
344 | * If this code finds it can't get good compression, it puts an |
345 | * entry onto the work queue to write the uncompressed bytes. This | |
346 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
347 | * are written in the same order that the flusher thread sent them |
348 | * down. | |
d352ac68 | 349 | */ |
771ed689 CM |
350 | static noinline int compress_file_range(struct inode *inode, |
351 | struct page *locked_page, | |
352 | u64 start, u64 end, | |
353 | struct async_cow *async_cow, | |
354 | int *num_added) | |
b888db2b CM |
355 | { |
356 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 357 | u64 num_bytes; |
db94535d | 358 | u64 blocksize = root->sectorsize; |
c8b97818 | 359 | u64 actual_end; |
42dc7bab | 360 | u64 isize = i_size_read(inode); |
e6dcd2dc | 361 | int ret = 0; |
c8b97818 CM |
362 | struct page **pages = NULL; |
363 | unsigned long nr_pages; | |
364 | unsigned long nr_pages_ret = 0; | |
365 | unsigned long total_compressed = 0; | |
366 | unsigned long total_in = 0; | |
367 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 368 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
369 | int i; |
370 | int will_compress; | |
261507a0 | 371 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 372 | int redirty = 0; |
b888db2b | 373 | |
4cb13e5d LB |
374 | /* if this is a small write inside eof, kick off a defrag */ |
375 | if ((end - start + 1) < 16 * 1024 && | |
376 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
377 | btrfs_add_inode_defrag(NULL, inode); |
378 | ||
42dc7bab | 379 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
380 | again: |
381 | will_compress = 0; | |
382 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
383 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 384 | |
f03d9301 CM |
385 | /* |
386 | * we don't want to send crud past the end of i_size through | |
387 | * compression, that's just a waste of CPU time. So, if the | |
388 | * end of the file is before the start of our current | |
389 | * requested range of bytes, we bail out to the uncompressed | |
390 | * cleanup code that can deal with all of this. | |
391 | * | |
392 | * It isn't really the fastest way to fix things, but this is a | |
393 | * very uncommon corner. | |
394 | */ | |
395 | if (actual_end <= start) | |
396 | goto cleanup_and_bail_uncompressed; | |
397 | ||
c8b97818 CM |
398 | total_compressed = actual_end - start; |
399 | ||
400 | /* we want to make sure that amount of ram required to uncompress | |
401 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
402 | * of a compressed extent to 128k. This is a crucial number |
403 | * because it also controls how easily we can spread reads across | |
404 | * cpus for decompression. | |
405 | * | |
406 | * We also want to make sure the amount of IO required to do | |
407 | * a random read is reasonably small, so we limit the size of | |
408 | * a compressed extent to 128k. | |
c8b97818 CM |
409 | */ |
410 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 411 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 412 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
413 | total_in = 0; |
414 | ret = 0; | |
db94535d | 415 | |
771ed689 CM |
416 | /* |
417 | * we do compression for mount -o compress and when the | |
418 | * inode has not been flagged as nocompress. This flag can | |
419 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 420 | */ |
6cbff00f | 421 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 422 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
423 | (BTRFS_I(inode)->force_compress) || |
424 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 425 | WARN_ON(pages); |
cfbc246e | 426 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
427 | if (!pages) { |
428 | /* just bail out to the uncompressed code */ | |
429 | goto cont; | |
430 | } | |
c8b97818 | 431 | |
261507a0 LZ |
432 | if (BTRFS_I(inode)->force_compress) |
433 | compress_type = BTRFS_I(inode)->force_compress; | |
434 | ||
4adaa611 CM |
435 | /* |
436 | * we need to call clear_page_dirty_for_io on each | |
437 | * page in the range. Otherwise applications with the file | |
438 | * mmap'd can wander in and change the page contents while | |
439 | * we are compressing them. | |
440 | * | |
441 | * If the compression fails for any reason, we set the pages | |
442 | * dirty again later on. | |
443 | */ | |
444 | extent_range_clear_dirty_for_io(inode, start, end); | |
445 | redirty = 1; | |
261507a0 LZ |
446 | ret = btrfs_compress_pages(compress_type, |
447 | inode->i_mapping, start, | |
448 | total_compressed, pages, | |
449 | nr_pages, &nr_pages_ret, | |
450 | &total_in, | |
451 | &total_compressed, | |
452 | max_compressed); | |
c8b97818 CM |
453 | |
454 | if (!ret) { | |
455 | unsigned long offset = total_compressed & | |
456 | (PAGE_CACHE_SIZE - 1); | |
457 | struct page *page = pages[nr_pages_ret - 1]; | |
458 | char *kaddr; | |
459 | ||
460 | /* zero the tail end of the last page, we might be | |
461 | * sending it down to disk | |
462 | */ | |
463 | if (offset) { | |
7ac687d9 | 464 | kaddr = kmap_atomic(page); |
c8b97818 CM |
465 | memset(kaddr + offset, 0, |
466 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 467 | kunmap_atomic(kaddr); |
c8b97818 CM |
468 | } |
469 | will_compress = 1; | |
470 | } | |
471 | } | |
560f7d75 | 472 | cont: |
c8b97818 CM |
473 | if (start == 0) { |
474 | /* lets try to make an inline extent */ | |
771ed689 | 475 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 476 | /* we didn't compress the entire range, try |
771ed689 | 477 | * to make an uncompressed inline extent. |
c8b97818 | 478 | */ |
00361589 JB |
479 | ret = cow_file_range_inline(root, inode, start, end, |
480 | 0, 0, NULL); | |
c8b97818 | 481 | } else { |
771ed689 | 482 | /* try making a compressed inline extent */ |
00361589 | 483 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
484 | total_compressed, |
485 | compress_type, pages); | |
c8b97818 | 486 | } |
79787eaa | 487 | if (ret <= 0) { |
151a41bc JB |
488 | unsigned long clear_flags = EXTENT_DELALLOC | |
489 | EXTENT_DEFRAG; | |
490 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; | |
491 | ||
771ed689 | 492 | /* |
79787eaa JM |
493 | * inline extent creation worked or returned error, |
494 | * we don't need to create any more async work items. | |
495 | * Unlock and free up our temp pages. | |
771ed689 | 496 | */ |
c2790a2e | 497 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 498 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
499 | PAGE_CLEAR_DIRTY | |
500 | PAGE_SET_WRITEBACK | | |
501 | PAGE_END_WRITEBACK); | |
c8b97818 CM |
502 | goto free_pages_out; |
503 | } | |
504 | } | |
505 | ||
506 | if (will_compress) { | |
507 | /* | |
508 | * we aren't doing an inline extent round the compressed size | |
509 | * up to a block size boundary so the allocator does sane | |
510 | * things | |
511 | */ | |
fda2832f | 512 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
513 | |
514 | /* | |
515 | * one last check to make sure the compression is really a | |
516 | * win, compare the page count read with the blocks on disk | |
517 | */ | |
fda2832f | 518 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
519 | if (total_compressed >= total_in) { |
520 | will_compress = 0; | |
521 | } else { | |
c8b97818 CM |
522 | num_bytes = total_in; |
523 | } | |
524 | } | |
525 | if (!will_compress && pages) { | |
526 | /* | |
527 | * the compression code ran but failed to make things smaller, | |
528 | * free any pages it allocated and our page pointer array | |
529 | */ | |
530 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 531 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
532 | page_cache_release(pages[i]); |
533 | } | |
534 | kfree(pages); | |
535 | pages = NULL; | |
536 | total_compressed = 0; | |
537 | nr_pages_ret = 0; | |
538 | ||
539 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
540 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
541 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 542 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 543 | } |
c8b97818 | 544 | } |
771ed689 CM |
545 | if (will_compress) { |
546 | *num_added += 1; | |
c8b97818 | 547 | |
771ed689 CM |
548 | /* the async work queues will take care of doing actual |
549 | * allocation on disk for these compressed pages, | |
550 | * and will submit them to the elevator. | |
551 | */ | |
552 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
553 | total_compressed, pages, nr_pages_ret, |
554 | compress_type); | |
179e29e4 | 555 | |
24ae6365 | 556 | if (start + num_bytes < end) { |
771ed689 CM |
557 | start += num_bytes; |
558 | pages = NULL; | |
559 | cond_resched(); | |
560 | goto again; | |
561 | } | |
562 | } else { | |
f03d9301 | 563 | cleanup_and_bail_uncompressed: |
771ed689 CM |
564 | /* |
565 | * No compression, but we still need to write the pages in | |
566 | * the file we've been given so far. redirty the locked | |
567 | * page if it corresponds to our extent and set things up | |
568 | * for the async work queue to run cow_file_range to do | |
569 | * the normal delalloc dance | |
570 | */ | |
571 | if (page_offset(locked_page) >= start && | |
572 | page_offset(locked_page) <= end) { | |
573 | __set_page_dirty_nobuffers(locked_page); | |
574 | /* unlocked later on in the async handlers */ | |
575 | } | |
4adaa611 CM |
576 | if (redirty) |
577 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
578 | add_async_extent(async_cow, start, end - start + 1, |
579 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
580 | *num_added += 1; |
581 | } | |
3b951516 | 582 | |
771ed689 | 583 | out: |
79787eaa | 584 | return ret; |
771ed689 CM |
585 | |
586 | free_pages_out: | |
587 | for (i = 0; i < nr_pages_ret; i++) { | |
588 | WARN_ON(pages[i]->mapping); | |
589 | page_cache_release(pages[i]); | |
590 | } | |
d397712b | 591 | kfree(pages); |
771ed689 CM |
592 | |
593 | goto out; | |
594 | } | |
595 | ||
596 | /* | |
597 | * phase two of compressed writeback. This is the ordered portion | |
598 | * of the code, which only gets called in the order the work was | |
599 | * queued. We walk all the async extents created by compress_file_range | |
600 | * and send them down to the disk. | |
601 | */ | |
602 | static noinline int submit_compressed_extents(struct inode *inode, | |
603 | struct async_cow *async_cow) | |
604 | { | |
605 | struct async_extent *async_extent; | |
606 | u64 alloc_hint = 0; | |
771ed689 CM |
607 | struct btrfs_key ins; |
608 | struct extent_map *em; | |
609 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
610 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
611 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 612 | int ret = 0; |
771ed689 CM |
613 | |
614 | if (list_empty(&async_cow->extents)) | |
615 | return 0; | |
616 | ||
3e04e7f1 | 617 | again: |
d397712b | 618 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
619 | async_extent = list_entry(async_cow->extents.next, |
620 | struct async_extent, list); | |
621 | list_del(&async_extent->list); | |
c8b97818 | 622 | |
771ed689 CM |
623 | io_tree = &BTRFS_I(inode)->io_tree; |
624 | ||
f5a84ee3 | 625 | retry: |
771ed689 CM |
626 | /* did the compression code fall back to uncompressed IO? */ |
627 | if (!async_extent->pages) { | |
628 | int page_started = 0; | |
629 | unsigned long nr_written = 0; | |
630 | ||
631 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 632 | async_extent->start + |
d0082371 | 633 | async_extent->ram_size - 1); |
771ed689 CM |
634 | |
635 | /* allocate blocks */ | |
f5a84ee3 JB |
636 | ret = cow_file_range(inode, async_cow->locked_page, |
637 | async_extent->start, | |
638 | async_extent->start + | |
639 | async_extent->ram_size - 1, | |
640 | &page_started, &nr_written, 0); | |
771ed689 | 641 | |
79787eaa JM |
642 | /* JDM XXX */ |
643 | ||
771ed689 CM |
644 | /* |
645 | * if page_started, cow_file_range inserted an | |
646 | * inline extent and took care of all the unlocking | |
647 | * and IO for us. Otherwise, we need to submit | |
648 | * all those pages down to the drive. | |
649 | */ | |
f5a84ee3 | 650 | if (!page_started && !ret) |
771ed689 CM |
651 | extent_write_locked_range(io_tree, |
652 | inode, async_extent->start, | |
d397712b | 653 | async_extent->start + |
771ed689 CM |
654 | async_extent->ram_size - 1, |
655 | btrfs_get_extent, | |
656 | WB_SYNC_ALL); | |
3e04e7f1 JB |
657 | else if (ret) |
658 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
659 | kfree(async_extent); |
660 | cond_resched(); | |
661 | continue; | |
662 | } | |
663 | ||
664 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 665 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 666 | |
00361589 | 667 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
668 | async_extent->compressed_size, |
669 | async_extent->compressed_size, | |
81c9ad23 | 670 | 0, alloc_hint, &ins, 1); |
f5a84ee3 JB |
671 | if (ret) { |
672 | int i; | |
3e04e7f1 | 673 | |
f5a84ee3 JB |
674 | for (i = 0; i < async_extent->nr_pages; i++) { |
675 | WARN_ON(async_extent->pages[i]->mapping); | |
676 | page_cache_release(async_extent->pages[i]); | |
677 | } | |
678 | kfree(async_extent->pages); | |
679 | async_extent->nr_pages = 0; | |
680 | async_extent->pages = NULL; | |
3e04e7f1 | 681 | |
fdf8e2ea JB |
682 | if (ret == -ENOSPC) { |
683 | unlock_extent(io_tree, async_extent->start, | |
684 | async_extent->start + | |
685 | async_extent->ram_size - 1); | |
79787eaa | 686 | goto retry; |
fdf8e2ea | 687 | } |
3e04e7f1 | 688 | goto out_free; |
f5a84ee3 JB |
689 | } |
690 | ||
c2167754 YZ |
691 | /* |
692 | * here we're doing allocation and writeback of the | |
693 | * compressed pages | |
694 | */ | |
695 | btrfs_drop_extent_cache(inode, async_extent->start, | |
696 | async_extent->start + | |
697 | async_extent->ram_size - 1, 0); | |
698 | ||
172ddd60 | 699 | em = alloc_extent_map(); |
b9aa55be LB |
700 | if (!em) { |
701 | ret = -ENOMEM; | |
3e04e7f1 | 702 | goto out_free_reserve; |
b9aa55be | 703 | } |
771ed689 CM |
704 | em->start = async_extent->start; |
705 | em->len = async_extent->ram_size; | |
445a6944 | 706 | em->orig_start = em->start; |
2ab28f32 JB |
707 | em->mod_start = em->start; |
708 | em->mod_len = em->len; | |
c8b97818 | 709 | |
771ed689 CM |
710 | em->block_start = ins.objectid; |
711 | em->block_len = ins.offset; | |
b4939680 | 712 | em->orig_block_len = ins.offset; |
cc95bef6 | 713 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 714 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 715 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
716 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
717 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 718 | em->generation = -1; |
771ed689 | 719 | |
d397712b | 720 | while (1) { |
890871be | 721 | write_lock(&em_tree->lock); |
09a2a8f9 | 722 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 723 | write_unlock(&em_tree->lock); |
771ed689 CM |
724 | if (ret != -EEXIST) { |
725 | free_extent_map(em); | |
726 | break; | |
727 | } | |
728 | btrfs_drop_extent_cache(inode, async_extent->start, | |
729 | async_extent->start + | |
730 | async_extent->ram_size - 1, 0); | |
731 | } | |
732 | ||
3e04e7f1 JB |
733 | if (ret) |
734 | goto out_free_reserve; | |
735 | ||
261507a0 LZ |
736 | ret = btrfs_add_ordered_extent_compress(inode, |
737 | async_extent->start, | |
738 | ins.objectid, | |
739 | async_extent->ram_size, | |
740 | ins.offset, | |
741 | BTRFS_ORDERED_COMPRESSED, | |
742 | async_extent->compress_type); | |
3e04e7f1 JB |
743 | if (ret) |
744 | goto out_free_reserve; | |
771ed689 | 745 | |
771ed689 CM |
746 | /* |
747 | * clear dirty, set writeback and unlock the pages. | |
748 | */ | |
c2790a2e | 749 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
750 | async_extent->start + |
751 | async_extent->ram_size - 1, | |
151a41bc JB |
752 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
753 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 754 | PAGE_SET_WRITEBACK); |
771ed689 | 755 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
756 | async_extent->start, |
757 | async_extent->ram_size, | |
758 | ins.objectid, | |
759 | ins.offset, async_extent->pages, | |
760 | async_extent->nr_pages); | |
771ed689 CM |
761 | alloc_hint = ins.objectid + ins.offset; |
762 | kfree(async_extent); | |
3e04e7f1 JB |
763 | if (ret) |
764 | goto out; | |
771ed689 CM |
765 | cond_resched(); |
766 | } | |
79787eaa JM |
767 | ret = 0; |
768 | out: | |
769 | return ret; | |
3e04e7f1 JB |
770 | out_free_reserve: |
771 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 772 | out_free: |
c2790a2e | 773 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
774 | async_extent->start + |
775 | async_extent->ram_size - 1, | |
c2790a2e | 776 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
777 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
778 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
779 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 780 | kfree(async_extent); |
3e04e7f1 | 781 | goto again; |
771ed689 CM |
782 | } |
783 | ||
4b46fce2 JB |
784 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
785 | u64 num_bytes) | |
786 | { | |
787 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
788 | struct extent_map *em; | |
789 | u64 alloc_hint = 0; | |
790 | ||
791 | read_lock(&em_tree->lock); | |
792 | em = search_extent_mapping(em_tree, start, num_bytes); | |
793 | if (em) { | |
794 | /* | |
795 | * if block start isn't an actual block number then find the | |
796 | * first block in this inode and use that as a hint. If that | |
797 | * block is also bogus then just don't worry about it. | |
798 | */ | |
799 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
800 | free_extent_map(em); | |
801 | em = search_extent_mapping(em_tree, 0, 0); | |
802 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
803 | alloc_hint = em->block_start; | |
804 | if (em) | |
805 | free_extent_map(em); | |
806 | } else { | |
807 | alloc_hint = em->block_start; | |
808 | free_extent_map(em); | |
809 | } | |
810 | } | |
811 | read_unlock(&em_tree->lock); | |
812 | ||
813 | return alloc_hint; | |
814 | } | |
815 | ||
771ed689 CM |
816 | /* |
817 | * when extent_io.c finds a delayed allocation range in the file, | |
818 | * the call backs end up in this code. The basic idea is to | |
819 | * allocate extents on disk for the range, and create ordered data structs | |
820 | * in ram to track those extents. | |
821 | * | |
822 | * locked_page is the page that writepage had locked already. We use | |
823 | * it to make sure we don't do extra locks or unlocks. | |
824 | * | |
825 | * *page_started is set to one if we unlock locked_page and do everything | |
826 | * required to start IO on it. It may be clean and already done with | |
827 | * IO when we return. | |
828 | */ | |
00361589 JB |
829 | static noinline int cow_file_range(struct inode *inode, |
830 | struct page *locked_page, | |
831 | u64 start, u64 end, int *page_started, | |
832 | unsigned long *nr_written, | |
833 | int unlock) | |
771ed689 | 834 | { |
00361589 | 835 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
836 | u64 alloc_hint = 0; |
837 | u64 num_bytes; | |
838 | unsigned long ram_size; | |
839 | u64 disk_num_bytes; | |
840 | u64 cur_alloc_size; | |
841 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
842 | struct btrfs_key ins; |
843 | struct extent_map *em; | |
844 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
845 | int ret = 0; | |
846 | ||
83eea1f1 | 847 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 848 | |
fda2832f | 849 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
850 | num_bytes = max(blocksize, num_bytes); |
851 | disk_num_bytes = num_bytes; | |
771ed689 | 852 | |
4cb5300b | 853 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
854 | if (num_bytes < 64 * 1024 && |
855 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 856 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 857 | |
771ed689 CM |
858 | if (start == 0) { |
859 | /* lets try to make an inline extent */ | |
00361589 JB |
860 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
861 | NULL); | |
771ed689 | 862 | if (ret == 0) { |
c2790a2e JB |
863 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
864 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 865 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
866 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
867 | PAGE_END_WRITEBACK); | |
c2167754 | 868 | |
771ed689 CM |
869 | *nr_written = *nr_written + |
870 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
871 | *page_started = 1; | |
771ed689 | 872 | goto out; |
79787eaa | 873 | } else if (ret < 0) { |
79787eaa | 874 | goto out_unlock; |
771ed689 CM |
875 | } |
876 | } | |
877 | ||
878 | BUG_ON(disk_num_bytes > | |
6c41761f | 879 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 880 | |
4b46fce2 | 881 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
882 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
883 | ||
d397712b | 884 | while (disk_num_bytes > 0) { |
a791e35e CM |
885 | unsigned long op; |
886 | ||
287a0ab9 | 887 | cur_alloc_size = disk_num_bytes; |
00361589 | 888 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 889 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 890 | &ins, 1); |
00361589 | 891 | if (ret < 0) |
79787eaa | 892 | goto out_unlock; |
d397712b | 893 | |
172ddd60 | 894 | em = alloc_extent_map(); |
b9aa55be LB |
895 | if (!em) { |
896 | ret = -ENOMEM; | |
ace68bac | 897 | goto out_reserve; |
b9aa55be | 898 | } |
e6dcd2dc | 899 | em->start = start; |
445a6944 | 900 | em->orig_start = em->start; |
771ed689 CM |
901 | ram_size = ins.offset; |
902 | em->len = ins.offset; | |
2ab28f32 JB |
903 | em->mod_start = em->start; |
904 | em->mod_len = em->len; | |
c8b97818 | 905 | |
e6dcd2dc | 906 | em->block_start = ins.objectid; |
c8b97818 | 907 | em->block_len = ins.offset; |
b4939680 | 908 | em->orig_block_len = ins.offset; |
cc95bef6 | 909 | em->ram_bytes = ram_size; |
e6dcd2dc | 910 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 911 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 912 | em->generation = -1; |
c8b97818 | 913 | |
d397712b | 914 | while (1) { |
890871be | 915 | write_lock(&em_tree->lock); |
09a2a8f9 | 916 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 917 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
918 | if (ret != -EEXIST) { |
919 | free_extent_map(em); | |
920 | break; | |
921 | } | |
922 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 923 | start + ram_size - 1, 0); |
e6dcd2dc | 924 | } |
ace68bac LB |
925 | if (ret) |
926 | goto out_reserve; | |
e6dcd2dc | 927 | |
98d20f67 | 928 | cur_alloc_size = ins.offset; |
e6dcd2dc | 929 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 930 | ram_size, cur_alloc_size, 0); |
ace68bac LB |
931 | if (ret) |
932 | goto out_reserve; | |
c8b97818 | 933 | |
17d217fe YZ |
934 | if (root->root_key.objectid == |
935 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
936 | ret = btrfs_reloc_clone_csums(inode, start, | |
937 | cur_alloc_size); | |
00361589 | 938 | if (ret) |
ace68bac | 939 | goto out_reserve; |
17d217fe YZ |
940 | } |
941 | ||
d397712b | 942 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 943 | break; |
d397712b | 944 | |
c8b97818 CM |
945 | /* we're not doing compressed IO, don't unlock the first |
946 | * page (which the caller expects to stay locked), don't | |
947 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
948 | * |
949 | * Do set the Private2 bit so we know this page was properly | |
950 | * setup for writepage | |
c8b97818 | 951 | */ |
c2790a2e JB |
952 | op = unlock ? PAGE_UNLOCK : 0; |
953 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 954 | |
c2790a2e JB |
955 | extent_clear_unlock_delalloc(inode, start, |
956 | start + ram_size - 1, locked_page, | |
957 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
958 | op); | |
c8b97818 | 959 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
960 | num_bytes -= cur_alloc_size; |
961 | alloc_hint = ins.objectid + ins.offset; | |
962 | start += cur_alloc_size; | |
b888db2b | 963 | } |
79787eaa | 964 | out: |
be20aa9d | 965 | return ret; |
b7d5b0a8 | 966 | |
ace68bac LB |
967 | out_reserve: |
968 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 969 | out_unlock: |
c2790a2e | 970 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
971 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
972 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
973 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
974 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 975 | goto out; |
771ed689 | 976 | } |
c8b97818 | 977 | |
771ed689 CM |
978 | /* |
979 | * work queue call back to started compression on a file and pages | |
980 | */ | |
981 | static noinline void async_cow_start(struct btrfs_work *work) | |
982 | { | |
983 | struct async_cow *async_cow; | |
984 | int num_added = 0; | |
985 | async_cow = container_of(work, struct async_cow, work); | |
986 | ||
987 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
988 | async_cow->start, async_cow->end, async_cow, | |
989 | &num_added); | |
8180ef88 | 990 | if (num_added == 0) { |
cb77fcd8 | 991 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 992 | async_cow->inode = NULL; |
8180ef88 | 993 | } |
771ed689 CM |
994 | } |
995 | ||
996 | /* | |
997 | * work queue call back to submit previously compressed pages | |
998 | */ | |
999 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1000 | { | |
1001 | struct async_cow *async_cow; | |
1002 | struct btrfs_root *root; | |
1003 | unsigned long nr_pages; | |
1004 | ||
1005 | async_cow = container_of(work, struct async_cow, work); | |
1006 | ||
1007 | root = async_cow->root; | |
1008 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1009 | PAGE_CACHE_SHIFT; | |
1010 | ||
66657b31 | 1011 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1012 | 5 * 1024 * 1024 && |
771ed689 CM |
1013 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1014 | wake_up(&root->fs_info->async_submit_wait); | |
1015 | ||
d397712b | 1016 | if (async_cow->inode) |
771ed689 | 1017 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1018 | } |
c8b97818 | 1019 | |
771ed689 CM |
1020 | static noinline void async_cow_free(struct btrfs_work *work) |
1021 | { | |
1022 | struct async_cow *async_cow; | |
1023 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1024 | if (async_cow->inode) |
cb77fcd8 | 1025 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1026 | kfree(async_cow); |
1027 | } | |
1028 | ||
1029 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1030 | u64 start, u64 end, int *page_started, | |
1031 | unsigned long *nr_written) | |
1032 | { | |
1033 | struct async_cow *async_cow; | |
1034 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1035 | unsigned long nr_pages; | |
1036 | u64 cur_end; | |
287082b0 | 1037 | int limit = 10 * 1024 * 1024; |
771ed689 | 1038 | |
a3429ab7 CM |
1039 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1040 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1041 | while (start < end) { |
771ed689 | 1042 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1043 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1044 | async_cow->inode = igrab(inode); |
771ed689 CM |
1045 | async_cow->root = root; |
1046 | async_cow->locked_page = locked_page; | |
1047 | async_cow->start = start; | |
1048 | ||
6cbff00f | 1049 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1050 | cur_end = end; |
1051 | else | |
1052 | cur_end = min(end, start + 512 * 1024 - 1); | |
1053 | ||
1054 | async_cow->end = cur_end; | |
1055 | INIT_LIST_HEAD(&async_cow->extents); | |
1056 | ||
1057 | async_cow->work.func = async_cow_start; | |
1058 | async_cow->work.ordered_func = async_cow_submit; | |
1059 | async_cow->work.ordered_free = async_cow_free; | |
1060 | async_cow->work.flags = 0; | |
1061 | ||
771ed689 CM |
1062 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1063 | PAGE_CACHE_SHIFT; | |
1064 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1065 | ||
1066 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1067 | &async_cow->work); | |
1068 | ||
1069 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1070 | wait_event(root->fs_info->async_submit_wait, | |
1071 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1072 | limit)); | |
1073 | } | |
1074 | ||
d397712b | 1075 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1076 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1077 | wait_event(root->fs_info->async_submit_wait, | |
1078 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1079 | 0)); | |
1080 | } | |
1081 | ||
1082 | *nr_written += nr_pages; | |
1083 | start = cur_end + 1; | |
1084 | } | |
1085 | *page_started = 1; | |
1086 | return 0; | |
be20aa9d CM |
1087 | } |
1088 | ||
d397712b | 1089 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1090 | u64 bytenr, u64 num_bytes) |
1091 | { | |
1092 | int ret; | |
1093 | struct btrfs_ordered_sum *sums; | |
1094 | LIST_HEAD(list); | |
1095 | ||
07d400a6 | 1096 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1097 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1098 | if (ret == 0 && list_empty(&list)) |
1099 | return 0; | |
1100 | ||
1101 | while (!list_empty(&list)) { | |
1102 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1103 | list_del(&sums->list); | |
1104 | kfree(sums); | |
1105 | } | |
1106 | return 1; | |
1107 | } | |
1108 | ||
d352ac68 CM |
1109 | /* |
1110 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1111 | * of the extents that exist in the file, and COWs the file as required. | |
1112 | * | |
1113 | * If no cow copies or snapshots exist, we write directly to the existing | |
1114 | * blocks on disk | |
1115 | */ | |
7f366cfe CM |
1116 | static noinline int run_delalloc_nocow(struct inode *inode, |
1117 | struct page *locked_page, | |
771ed689 CM |
1118 | u64 start, u64 end, int *page_started, int force, |
1119 | unsigned long *nr_written) | |
be20aa9d | 1120 | { |
be20aa9d | 1121 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1122 | struct btrfs_trans_handle *trans; |
be20aa9d | 1123 | struct extent_buffer *leaf; |
be20aa9d | 1124 | struct btrfs_path *path; |
80ff3856 | 1125 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1126 | struct btrfs_key found_key; |
80ff3856 YZ |
1127 | u64 cow_start; |
1128 | u64 cur_offset; | |
1129 | u64 extent_end; | |
5d4f98a2 | 1130 | u64 extent_offset; |
80ff3856 YZ |
1131 | u64 disk_bytenr; |
1132 | u64 num_bytes; | |
b4939680 | 1133 | u64 disk_num_bytes; |
cc95bef6 | 1134 | u64 ram_bytes; |
80ff3856 | 1135 | int extent_type; |
79787eaa | 1136 | int ret, err; |
d899e052 | 1137 | int type; |
80ff3856 YZ |
1138 | int nocow; |
1139 | int check_prev = 1; | |
82d5902d | 1140 | bool nolock; |
33345d01 | 1141 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1142 | |
1143 | path = btrfs_alloc_path(); | |
17ca04af | 1144 | if (!path) { |
c2790a2e JB |
1145 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1146 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1147 | EXTENT_DO_ACCOUNTING | |
1148 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1149 | PAGE_CLEAR_DIRTY | |
1150 | PAGE_SET_WRITEBACK | | |
1151 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1152 | return -ENOMEM; |
17ca04af | 1153 | } |
82d5902d | 1154 | |
83eea1f1 | 1155 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1156 | |
1157 | if (nolock) | |
7a7eaa40 | 1158 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1159 | else |
7a7eaa40 | 1160 | trans = btrfs_join_transaction(root); |
ff5714cc | 1161 | |
79787eaa | 1162 | if (IS_ERR(trans)) { |
c2790a2e JB |
1163 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1164 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1165 | EXTENT_DO_ACCOUNTING | |
1166 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1167 | PAGE_CLEAR_DIRTY | |
1168 | PAGE_SET_WRITEBACK | | |
1169 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1170 | btrfs_free_path(path); |
1171 | return PTR_ERR(trans); | |
1172 | } | |
1173 | ||
74b21075 | 1174 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1175 | |
80ff3856 YZ |
1176 | cow_start = (u64)-1; |
1177 | cur_offset = start; | |
1178 | while (1) { | |
33345d01 | 1179 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1180 | cur_offset, 0); |
79787eaa JM |
1181 | if (ret < 0) { |
1182 | btrfs_abort_transaction(trans, root, ret); | |
1183 | goto error; | |
1184 | } | |
80ff3856 YZ |
1185 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1186 | leaf = path->nodes[0]; | |
1187 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1188 | path->slots[0] - 1); | |
33345d01 | 1189 | if (found_key.objectid == ino && |
80ff3856 YZ |
1190 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1191 | path->slots[0]--; | |
1192 | } | |
1193 | check_prev = 0; | |
1194 | next_slot: | |
1195 | leaf = path->nodes[0]; | |
1196 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1197 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1198 | if (ret < 0) { |
1199 | btrfs_abort_transaction(trans, root, ret); | |
1200 | goto error; | |
1201 | } | |
80ff3856 YZ |
1202 | if (ret > 0) |
1203 | break; | |
1204 | leaf = path->nodes[0]; | |
1205 | } | |
be20aa9d | 1206 | |
80ff3856 YZ |
1207 | nocow = 0; |
1208 | disk_bytenr = 0; | |
17d217fe | 1209 | num_bytes = 0; |
80ff3856 YZ |
1210 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1211 | ||
33345d01 | 1212 | if (found_key.objectid > ino || |
80ff3856 YZ |
1213 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1214 | found_key.offset > end) | |
1215 | break; | |
1216 | ||
1217 | if (found_key.offset > cur_offset) { | |
1218 | extent_end = found_key.offset; | |
e9061e21 | 1219 | extent_type = 0; |
80ff3856 YZ |
1220 | goto out_check; |
1221 | } | |
1222 | ||
1223 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1224 | struct btrfs_file_extent_item); | |
1225 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1226 | ||
cc95bef6 | 1227 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1228 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1229 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1230 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1231 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1232 | extent_end = found_key.offset + |
1233 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1234 | disk_num_bytes = |
1235 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1236 | if (extent_end <= start) { |
1237 | path->slots[0]++; | |
1238 | goto next_slot; | |
1239 | } | |
17d217fe YZ |
1240 | if (disk_bytenr == 0) |
1241 | goto out_check; | |
80ff3856 YZ |
1242 | if (btrfs_file_extent_compression(leaf, fi) || |
1243 | btrfs_file_extent_encryption(leaf, fi) || | |
1244 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1245 | goto out_check; | |
d899e052 YZ |
1246 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1247 | goto out_check; | |
d2fb3437 | 1248 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1249 | goto out_check; |
33345d01 | 1250 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1251 | found_key.offset - |
1252 | extent_offset, disk_bytenr)) | |
17d217fe | 1253 | goto out_check; |
5d4f98a2 | 1254 | disk_bytenr += extent_offset; |
17d217fe YZ |
1255 | disk_bytenr += cur_offset - found_key.offset; |
1256 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1257 | /* | |
1258 | * force cow if csum exists in the range. | |
1259 | * this ensure that csum for a given extent are | |
1260 | * either valid or do not exist. | |
1261 | */ | |
1262 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1263 | goto out_check; | |
80ff3856 YZ |
1264 | nocow = 1; |
1265 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1266 | extent_end = found_key.offset + | |
1267 | btrfs_file_extent_inline_len(leaf, fi); | |
1268 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1269 | } else { | |
1270 | BUG_ON(1); | |
1271 | } | |
1272 | out_check: | |
1273 | if (extent_end <= start) { | |
1274 | path->slots[0]++; | |
1275 | goto next_slot; | |
1276 | } | |
1277 | if (!nocow) { | |
1278 | if (cow_start == (u64)-1) | |
1279 | cow_start = cur_offset; | |
1280 | cur_offset = extent_end; | |
1281 | if (cur_offset > end) | |
1282 | break; | |
1283 | path->slots[0]++; | |
1284 | goto next_slot; | |
7ea394f1 YZ |
1285 | } |
1286 | ||
b3b4aa74 | 1287 | btrfs_release_path(path); |
80ff3856 | 1288 | if (cow_start != (u64)-1) { |
00361589 JB |
1289 | ret = cow_file_range(inode, locked_page, |
1290 | cow_start, found_key.offset - 1, | |
1291 | page_started, nr_written, 1); | |
79787eaa JM |
1292 | if (ret) { |
1293 | btrfs_abort_transaction(trans, root, ret); | |
1294 | goto error; | |
1295 | } | |
80ff3856 | 1296 | cow_start = (u64)-1; |
7ea394f1 | 1297 | } |
80ff3856 | 1298 | |
d899e052 YZ |
1299 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1300 | struct extent_map *em; | |
1301 | struct extent_map_tree *em_tree; | |
1302 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1303 | em = alloc_extent_map(); |
79787eaa | 1304 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1305 | em->start = cur_offset; |
70c8a91c | 1306 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1307 | em->len = num_bytes; |
1308 | em->block_len = num_bytes; | |
1309 | em->block_start = disk_bytenr; | |
b4939680 | 1310 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1311 | em->ram_bytes = ram_bytes; |
d899e052 | 1312 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1313 | em->mod_start = em->start; |
1314 | em->mod_len = em->len; | |
d899e052 | 1315 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1316 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1317 | em->generation = -1; |
d899e052 | 1318 | while (1) { |
890871be | 1319 | write_lock(&em_tree->lock); |
09a2a8f9 | 1320 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1321 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1322 | if (ret != -EEXIST) { |
1323 | free_extent_map(em); | |
1324 | break; | |
1325 | } | |
1326 | btrfs_drop_extent_cache(inode, em->start, | |
1327 | em->start + em->len - 1, 0); | |
1328 | } | |
1329 | type = BTRFS_ORDERED_PREALLOC; | |
1330 | } else { | |
1331 | type = BTRFS_ORDERED_NOCOW; | |
1332 | } | |
80ff3856 YZ |
1333 | |
1334 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1335 | num_bytes, num_bytes, type); |
79787eaa | 1336 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1337 | |
efa56464 YZ |
1338 | if (root->root_key.objectid == |
1339 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1340 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1341 | num_bytes); | |
79787eaa JM |
1342 | if (ret) { |
1343 | btrfs_abort_transaction(trans, root, ret); | |
1344 | goto error; | |
1345 | } | |
efa56464 YZ |
1346 | } |
1347 | ||
c2790a2e JB |
1348 | extent_clear_unlock_delalloc(inode, cur_offset, |
1349 | cur_offset + num_bytes - 1, | |
1350 | locked_page, EXTENT_LOCKED | | |
1351 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1352 | PAGE_SET_PRIVATE2); | |
80ff3856 YZ |
1353 | cur_offset = extent_end; |
1354 | if (cur_offset > end) | |
1355 | break; | |
be20aa9d | 1356 | } |
b3b4aa74 | 1357 | btrfs_release_path(path); |
80ff3856 | 1358 | |
17ca04af | 1359 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1360 | cow_start = cur_offset; |
17ca04af JB |
1361 | cur_offset = end; |
1362 | } | |
1363 | ||
80ff3856 | 1364 | if (cow_start != (u64)-1) { |
00361589 JB |
1365 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1366 | page_started, nr_written, 1); | |
79787eaa JM |
1367 | if (ret) { |
1368 | btrfs_abort_transaction(trans, root, ret); | |
1369 | goto error; | |
1370 | } | |
80ff3856 YZ |
1371 | } |
1372 | ||
79787eaa | 1373 | error: |
a698d075 | 1374 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1375 | if (!ret) |
1376 | ret = err; | |
1377 | ||
17ca04af | 1378 | if (ret && cur_offset < end) |
c2790a2e JB |
1379 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1380 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1381 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1382 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1383 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1384 | PAGE_SET_WRITEBACK | |
1385 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1386 | btrfs_free_path(path); |
79787eaa | 1387 | return ret; |
be20aa9d CM |
1388 | } |
1389 | ||
d352ac68 CM |
1390 | /* |
1391 | * extent_io.c call back to do delayed allocation processing | |
1392 | */ | |
c8b97818 | 1393 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1394 | u64 start, u64 end, int *page_started, |
1395 | unsigned long *nr_written) | |
be20aa9d | 1396 | { |
be20aa9d | 1397 | int ret; |
7f366cfe | 1398 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1399 | |
7ddf5a42 | 1400 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1401 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1402 | page_started, 1, nr_written); |
7ddf5a42 | 1403 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1404 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1405 | page_started, 0, nr_written); |
7ddf5a42 JB |
1406 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1407 | !(BTRFS_I(inode)->force_compress) && | |
1408 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1409 | ret = cow_file_range(inode, locked_page, start, end, |
1410 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1411 | } else { |
1412 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1413 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1414 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1415 | page_started, nr_written); |
7ddf5a42 | 1416 | } |
b888db2b CM |
1417 | return ret; |
1418 | } | |
1419 | ||
1bf85046 JM |
1420 | static void btrfs_split_extent_hook(struct inode *inode, |
1421 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1422 | { |
0ca1f7ce | 1423 | /* not delalloc, ignore it */ |
9ed74f2d | 1424 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1425 | return; |
9ed74f2d | 1426 | |
9e0baf60 JB |
1427 | spin_lock(&BTRFS_I(inode)->lock); |
1428 | BTRFS_I(inode)->outstanding_extents++; | |
1429 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1430 | } |
1431 | ||
1432 | /* | |
1433 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1434 | * extents so we can keep track of new extents that are just merged onto old | |
1435 | * extents, such as when we are doing sequential writes, so we can properly | |
1436 | * account for the metadata space we'll need. | |
1437 | */ | |
1bf85046 JM |
1438 | static void btrfs_merge_extent_hook(struct inode *inode, |
1439 | struct extent_state *new, | |
1440 | struct extent_state *other) | |
9ed74f2d | 1441 | { |
9ed74f2d JB |
1442 | /* not delalloc, ignore it */ |
1443 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1444 | return; |
9ed74f2d | 1445 | |
9e0baf60 JB |
1446 | spin_lock(&BTRFS_I(inode)->lock); |
1447 | BTRFS_I(inode)->outstanding_extents--; | |
1448 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1449 | } |
1450 | ||
eb73c1b7 MX |
1451 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1452 | struct inode *inode) | |
1453 | { | |
1454 | spin_lock(&root->delalloc_lock); | |
1455 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1456 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1457 | &root->delalloc_inodes); | |
1458 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1459 | &BTRFS_I(inode)->runtime_flags); | |
1460 | root->nr_delalloc_inodes++; | |
1461 | if (root->nr_delalloc_inodes == 1) { | |
1462 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1463 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1464 | list_add_tail(&root->delalloc_root, | |
1465 | &root->fs_info->delalloc_roots); | |
1466 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1467 | } | |
1468 | } | |
1469 | spin_unlock(&root->delalloc_lock); | |
1470 | } | |
1471 | ||
1472 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1473 | struct inode *inode) | |
1474 | { | |
1475 | spin_lock(&root->delalloc_lock); | |
1476 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1477 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1478 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1479 | &BTRFS_I(inode)->runtime_flags); | |
1480 | root->nr_delalloc_inodes--; | |
1481 | if (!root->nr_delalloc_inodes) { | |
1482 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1483 | BUG_ON(list_empty(&root->delalloc_root)); | |
1484 | list_del_init(&root->delalloc_root); | |
1485 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1486 | } | |
1487 | } | |
1488 | spin_unlock(&root->delalloc_lock); | |
1489 | } | |
1490 | ||
d352ac68 CM |
1491 | /* |
1492 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1493 | * bytes in this file, and to maintain the list of inodes that | |
1494 | * have pending delalloc work to be done. | |
1495 | */ | |
1bf85046 | 1496 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1497 | struct extent_state *state, unsigned long *bits) |
291d673e | 1498 | { |
9ed74f2d | 1499 | |
75eff68e CM |
1500 | /* |
1501 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1502 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1503 | * bit, which is only set or cleared with irqs on |
1504 | */ | |
0ca1f7ce | 1505 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1506 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1507 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1508 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1509 | |
9e0baf60 | 1510 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1511 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1512 | } else { |
1513 | spin_lock(&BTRFS_I(inode)->lock); | |
1514 | BTRFS_I(inode)->outstanding_extents++; | |
1515 | spin_unlock(&BTRFS_I(inode)->lock); | |
1516 | } | |
287a0ab9 | 1517 | |
963d678b MX |
1518 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1519 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1520 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1521 | BTRFS_I(inode)->delalloc_bytes += len; |
df0af1a5 | 1522 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1523 | &BTRFS_I(inode)->runtime_flags)) |
1524 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1525 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1526 | } |
291d673e CM |
1527 | } |
1528 | ||
d352ac68 CM |
1529 | /* |
1530 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1531 | */ | |
1bf85046 | 1532 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1533 | struct extent_state *state, |
1534 | unsigned long *bits) | |
291d673e | 1535 | { |
75eff68e CM |
1536 | /* |
1537 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1538 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1539 | * bit, which is only set or cleared with irqs on |
1540 | */ | |
0ca1f7ce | 1541 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1542 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1543 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1544 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1545 | |
9e0baf60 | 1546 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1547 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1548 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1549 | spin_lock(&BTRFS_I(inode)->lock); | |
1550 | BTRFS_I(inode)->outstanding_extents--; | |
1551 | spin_unlock(&BTRFS_I(inode)->lock); | |
1552 | } | |
0ca1f7ce YZ |
1553 | |
1554 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1555 | btrfs_delalloc_release_metadata(inode, len); | |
1556 | ||
0cb59c99 | 1557 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1558 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1559 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1560 | |
963d678b MX |
1561 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1562 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1563 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1564 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1565 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1566 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1567 | &BTRFS_I(inode)->runtime_flags)) |
1568 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1569 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1570 | } |
291d673e CM |
1571 | } |
1572 | ||
d352ac68 CM |
1573 | /* |
1574 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1575 | * we don't create bios that span stripes or chunks | |
1576 | */ | |
64a16701 | 1577 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1578 | size_t size, struct bio *bio, |
1579 | unsigned long bio_flags) | |
239b14b3 CM |
1580 | { |
1581 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1582 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1583 | u64 length = 0; |
1584 | u64 map_length; | |
239b14b3 CM |
1585 | int ret; |
1586 | ||
771ed689 CM |
1587 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1588 | return 0; | |
1589 | ||
f2d8d74d | 1590 | length = bio->bi_size; |
239b14b3 | 1591 | map_length = length; |
64a16701 | 1592 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1593 | &map_length, NULL, 0); |
3ec706c8 | 1594 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1595 | BUG_ON(ret < 0); |
d397712b | 1596 | if (map_length < length + size) |
239b14b3 | 1597 | return 1; |
3444a972 | 1598 | return 0; |
239b14b3 CM |
1599 | } |
1600 | ||
d352ac68 CM |
1601 | /* |
1602 | * in order to insert checksums into the metadata in large chunks, | |
1603 | * we wait until bio submission time. All the pages in the bio are | |
1604 | * checksummed and sums are attached onto the ordered extent record. | |
1605 | * | |
1606 | * At IO completion time the cums attached on the ordered extent record | |
1607 | * are inserted into the btree | |
1608 | */ | |
d397712b CM |
1609 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1610 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1611 | unsigned long bio_flags, |
1612 | u64 bio_offset) | |
065631f6 | 1613 | { |
065631f6 | 1614 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1615 | int ret = 0; |
e015640f | 1616 | |
d20f7043 | 1617 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1618 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1619 | return 0; |
1620 | } | |
e015640f | 1621 | |
4a69a410 CM |
1622 | /* |
1623 | * in order to insert checksums into the metadata in large chunks, | |
1624 | * we wait until bio submission time. All the pages in the bio are | |
1625 | * checksummed and sums are attached onto the ordered extent record. | |
1626 | * | |
1627 | * At IO completion time the cums attached on the ordered extent record | |
1628 | * are inserted into the btree | |
1629 | */ | |
b2950863 | 1630 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1631 | int mirror_num, unsigned long bio_flags, |
1632 | u64 bio_offset) | |
4a69a410 CM |
1633 | { |
1634 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1635 | int ret; |
1636 | ||
1637 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1638 | if (ret) | |
1639 | bio_endio(bio, ret); | |
1640 | return ret; | |
44b8bd7e CM |
1641 | } |
1642 | ||
d352ac68 | 1643 | /* |
cad321ad CM |
1644 | * extent_io.c submission hook. This does the right thing for csum calculation |
1645 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1646 | */ |
b2950863 | 1647 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1648 | int mirror_num, unsigned long bio_flags, |
1649 | u64 bio_offset) | |
44b8bd7e CM |
1650 | { |
1651 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1652 | int ret = 0; | |
19b9bdb0 | 1653 | int skip_sum; |
0417341e | 1654 | int metadata = 0; |
b812ce28 | 1655 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1656 | |
6cbff00f | 1657 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1658 | |
83eea1f1 | 1659 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1660 | metadata = 2; |
1661 | ||
7b6d91da | 1662 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1663 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1664 | if (ret) | |
61891923 | 1665 | goto out; |
5fd02043 | 1666 | |
d20f7043 | 1667 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1668 | ret = btrfs_submit_compressed_read(inode, bio, |
1669 | mirror_num, | |
1670 | bio_flags); | |
1671 | goto out; | |
c2db1073 TI |
1672 | } else if (!skip_sum) { |
1673 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1674 | if (ret) | |
61891923 | 1675 | goto out; |
c2db1073 | 1676 | } |
4d1b5fb4 | 1677 | goto mapit; |
b812ce28 | 1678 | } else if (async && !skip_sum) { |
17d217fe YZ |
1679 | /* csum items have already been cloned */ |
1680 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1681 | goto mapit; | |
19b9bdb0 | 1682 | /* we're doing a write, do the async checksumming */ |
61891923 | 1683 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1684 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1685 | bio_flags, bio_offset, |
1686 | __btrfs_submit_bio_start, | |
4a69a410 | 1687 | __btrfs_submit_bio_done); |
61891923 | 1688 | goto out; |
b812ce28 JB |
1689 | } else if (!skip_sum) { |
1690 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1691 | if (ret) | |
1692 | goto out; | |
19b9bdb0 CM |
1693 | } |
1694 | ||
0b86a832 | 1695 | mapit: |
61891923 SB |
1696 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1697 | ||
1698 | out: | |
1699 | if (ret < 0) | |
1700 | bio_endio(bio, ret); | |
1701 | return ret; | |
065631f6 | 1702 | } |
6885f308 | 1703 | |
d352ac68 CM |
1704 | /* |
1705 | * given a list of ordered sums record them in the inode. This happens | |
1706 | * at IO completion time based on sums calculated at bio submission time. | |
1707 | */ | |
ba1da2f4 | 1708 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1709 | struct inode *inode, u64 file_offset, |
1710 | struct list_head *list) | |
1711 | { | |
e6dcd2dc CM |
1712 | struct btrfs_ordered_sum *sum; |
1713 | ||
c6e30871 | 1714 | list_for_each_entry(sum, list, list) { |
39847c4d | 1715 | trans->adding_csums = 1; |
d20f7043 CM |
1716 | btrfs_csum_file_blocks(trans, |
1717 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1718 | trans->adding_csums = 0; |
e6dcd2dc CM |
1719 | } |
1720 | return 0; | |
1721 | } | |
1722 | ||
2ac55d41 JB |
1723 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1724 | struct extent_state **cached_state) | |
ea8c2819 | 1725 | { |
6c1500f2 | 1726 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1727 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1728 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1729 | } |
1730 | ||
d352ac68 | 1731 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1732 | struct btrfs_writepage_fixup { |
1733 | struct page *page; | |
1734 | struct btrfs_work work; | |
1735 | }; | |
1736 | ||
b2950863 | 1737 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1738 | { |
1739 | struct btrfs_writepage_fixup *fixup; | |
1740 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1741 | struct extent_state *cached_state = NULL; |
247e743c CM |
1742 | struct page *page; |
1743 | struct inode *inode; | |
1744 | u64 page_start; | |
1745 | u64 page_end; | |
87826df0 | 1746 | int ret; |
247e743c CM |
1747 | |
1748 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1749 | page = fixup->page; | |
4a096752 | 1750 | again: |
247e743c CM |
1751 | lock_page(page); |
1752 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1753 | ClearPageChecked(page); | |
1754 | goto out_page; | |
1755 | } | |
1756 | ||
1757 | inode = page->mapping->host; | |
1758 | page_start = page_offset(page); | |
1759 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1760 | ||
2ac55d41 | 1761 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1762 | &cached_state); |
4a096752 CM |
1763 | |
1764 | /* already ordered? We're done */ | |
8b62b72b | 1765 | if (PagePrivate2(page)) |
247e743c | 1766 | goto out; |
4a096752 CM |
1767 | |
1768 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1769 | if (ordered) { | |
2ac55d41 JB |
1770 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1771 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1772 | unlock_page(page); |
1773 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1774 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1775 | goto again; |
1776 | } | |
247e743c | 1777 | |
87826df0 JM |
1778 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1779 | if (ret) { | |
1780 | mapping_set_error(page->mapping, ret); | |
1781 | end_extent_writepage(page, ret, page_start, page_end); | |
1782 | ClearPageChecked(page); | |
1783 | goto out; | |
1784 | } | |
1785 | ||
2ac55d41 | 1786 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1787 | ClearPageChecked(page); |
87826df0 | 1788 | set_page_dirty(page); |
247e743c | 1789 | out: |
2ac55d41 JB |
1790 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1791 | &cached_state, GFP_NOFS); | |
247e743c CM |
1792 | out_page: |
1793 | unlock_page(page); | |
1794 | page_cache_release(page); | |
b897abec | 1795 | kfree(fixup); |
247e743c CM |
1796 | } |
1797 | ||
1798 | /* | |
1799 | * There are a few paths in the higher layers of the kernel that directly | |
1800 | * set the page dirty bit without asking the filesystem if it is a | |
1801 | * good idea. This causes problems because we want to make sure COW | |
1802 | * properly happens and the data=ordered rules are followed. | |
1803 | * | |
c8b97818 | 1804 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1805 | * hasn't been properly setup for IO. We kick off an async process |
1806 | * to fix it up. The async helper will wait for ordered extents, set | |
1807 | * the delalloc bit and make it safe to write the page. | |
1808 | */ | |
b2950863 | 1809 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1810 | { |
1811 | struct inode *inode = page->mapping->host; | |
1812 | struct btrfs_writepage_fixup *fixup; | |
1813 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1814 | |
8b62b72b CM |
1815 | /* this page is properly in the ordered list */ |
1816 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1817 | return 0; |
1818 | ||
1819 | if (PageChecked(page)) | |
1820 | return -EAGAIN; | |
1821 | ||
1822 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1823 | if (!fixup) | |
1824 | return -EAGAIN; | |
f421950f | 1825 | |
247e743c CM |
1826 | SetPageChecked(page); |
1827 | page_cache_get(page); | |
1828 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1829 | fixup->page = page; | |
1830 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1831 | return -EBUSY; |
247e743c CM |
1832 | } |
1833 | ||
d899e052 YZ |
1834 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1835 | struct inode *inode, u64 file_pos, | |
1836 | u64 disk_bytenr, u64 disk_num_bytes, | |
1837 | u64 num_bytes, u64 ram_bytes, | |
1838 | u8 compression, u8 encryption, | |
1839 | u16 other_encoding, int extent_type) | |
1840 | { | |
1841 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1842 | struct btrfs_file_extent_item *fi; | |
1843 | struct btrfs_path *path; | |
1844 | struct extent_buffer *leaf; | |
1845 | struct btrfs_key ins; | |
d899e052 YZ |
1846 | int ret; |
1847 | ||
1848 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1849 | if (!path) |
1850 | return -ENOMEM; | |
d899e052 | 1851 | |
b9473439 | 1852 | path->leave_spinning = 1; |
a1ed835e CM |
1853 | |
1854 | /* | |
1855 | * we may be replacing one extent in the tree with another. | |
1856 | * The new extent is pinned in the extent map, and we don't want | |
1857 | * to drop it from the cache until it is completely in the btree. | |
1858 | * | |
1859 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1860 | * the caller is expected to unpin it and allow it to be merged | |
1861 | * with the others. | |
1862 | */ | |
5dc562c5 | 1863 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1864 | file_pos + num_bytes, 0); |
79787eaa JM |
1865 | if (ret) |
1866 | goto out; | |
d899e052 | 1867 | |
33345d01 | 1868 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1869 | ins.offset = file_pos; |
1870 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1871 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1872 | if (ret) |
1873 | goto out; | |
d899e052 YZ |
1874 | leaf = path->nodes[0]; |
1875 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1876 | struct btrfs_file_extent_item); | |
1877 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1878 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1879 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1880 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1881 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1882 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1883 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1884 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1885 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1886 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1887 | |
d899e052 | 1888 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1889 | btrfs_release_path(path); |
d899e052 YZ |
1890 | |
1891 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1892 | |
1893 | ins.objectid = disk_bytenr; | |
1894 | ins.offset = disk_num_bytes; | |
1895 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1896 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1897 | root->root_key.objectid, | |
33345d01 | 1898 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1899 | out: |
d899e052 | 1900 | btrfs_free_path(path); |
b9473439 | 1901 | |
79787eaa | 1902 | return ret; |
d899e052 YZ |
1903 | } |
1904 | ||
38c227d8 LB |
1905 | /* snapshot-aware defrag */ |
1906 | struct sa_defrag_extent_backref { | |
1907 | struct rb_node node; | |
1908 | struct old_sa_defrag_extent *old; | |
1909 | u64 root_id; | |
1910 | u64 inum; | |
1911 | u64 file_pos; | |
1912 | u64 extent_offset; | |
1913 | u64 num_bytes; | |
1914 | u64 generation; | |
1915 | }; | |
1916 | ||
1917 | struct old_sa_defrag_extent { | |
1918 | struct list_head list; | |
1919 | struct new_sa_defrag_extent *new; | |
1920 | ||
1921 | u64 extent_offset; | |
1922 | u64 bytenr; | |
1923 | u64 offset; | |
1924 | u64 len; | |
1925 | int count; | |
1926 | }; | |
1927 | ||
1928 | struct new_sa_defrag_extent { | |
1929 | struct rb_root root; | |
1930 | struct list_head head; | |
1931 | struct btrfs_path *path; | |
1932 | struct inode *inode; | |
1933 | u64 file_pos; | |
1934 | u64 len; | |
1935 | u64 bytenr; | |
1936 | u64 disk_len; | |
1937 | u8 compress_type; | |
1938 | }; | |
1939 | ||
1940 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
1941 | struct sa_defrag_extent_backref *b2) | |
1942 | { | |
1943 | if (b1->root_id < b2->root_id) | |
1944 | return -1; | |
1945 | else if (b1->root_id > b2->root_id) | |
1946 | return 1; | |
1947 | ||
1948 | if (b1->inum < b2->inum) | |
1949 | return -1; | |
1950 | else if (b1->inum > b2->inum) | |
1951 | return 1; | |
1952 | ||
1953 | if (b1->file_pos < b2->file_pos) | |
1954 | return -1; | |
1955 | else if (b1->file_pos > b2->file_pos) | |
1956 | return 1; | |
1957 | ||
1958 | /* | |
1959 | * [------------------------------] ===> (a range of space) | |
1960 | * |<--->| |<---->| =============> (fs/file tree A) | |
1961 | * |<---------------------------->| ===> (fs/file tree B) | |
1962 | * | |
1963 | * A range of space can refer to two file extents in one tree while | |
1964 | * refer to only one file extent in another tree. | |
1965 | * | |
1966 | * So we may process a disk offset more than one time(two extents in A) | |
1967 | * and locate at the same extent(one extent in B), then insert two same | |
1968 | * backrefs(both refer to the extent in B). | |
1969 | */ | |
1970 | return 0; | |
1971 | } | |
1972 | ||
1973 | static void backref_insert(struct rb_root *root, | |
1974 | struct sa_defrag_extent_backref *backref) | |
1975 | { | |
1976 | struct rb_node **p = &root->rb_node; | |
1977 | struct rb_node *parent = NULL; | |
1978 | struct sa_defrag_extent_backref *entry; | |
1979 | int ret; | |
1980 | ||
1981 | while (*p) { | |
1982 | parent = *p; | |
1983 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
1984 | ||
1985 | ret = backref_comp(backref, entry); | |
1986 | if (ret < 0) | |
1987 | p = &(*p)->rb_left; | |
1988 | else | |
1989 | p = &(*p)->rb_right; | |
1990 | } | |
1991 | ||
1992 | rb_link_node(&backref->node, parent, p); | |
1993 | rb_insert_color(&backref->node, root); | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * Note the backref might has changed, and in this case we just return 0. | |
1998 | */ | |
1999 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2000 | void *ctx) | |
2001 | { | |
2002 | struct btrfs_file_extent_item *extent; | |
2003 | struct btrfs_fs_info *fs_info; | |
2004 | struct old_sa_defrag_extent *old = ctx; | |
2005 | struct new_sa_defrag_extent *new = old->new; | |
2006 | struct btrfs_path *path = new->path; | |
2007 | struct btrfs_key key; | |
2008 | struct btrfs_root *root; | |
2009 | struct sa_defrag_extent_backref *backref; | |
2010 | struct extent_buffer *leaf; | |
2011 | struct inode *inode = new->inode; | |
2012 | int slot; | |
2013 | int ret; | |
2014 | u64 extent_offset; | |
2015 | u64 num_bytes; | |
2016 | ||
2017 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2018 | inum == btrfs_ino(inode)) | |
2019 | return 0; | |
2020 | ||
2021 | key.objectid = root_id; | |
2022 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2023 | key.offset = (u64)-1; | |
2024 | ||
2025 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2026 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2027 | if (IS_ERR(root)) { | |
2028 | if (PTR_ERR(root) == -ENOENT) | |
2029 | return 0; | |
2030 | WARN_ON(1); | |
2031 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2032 | inum, offset, root_id); | |
2033 | return PTR_ERR(root); | |
2034 | } | |
2035 | ||
2036 | key.objectid = inum; | |
2037 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2038 | if (offset > (u64)-1 << 32) | |
2039 | key.offset = 0; | |
2040 | else | |
2041 | key.offset = offset; | |
2042 | ||
2043 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2044 | if (ret < 0) { | |
2045 | WARN_ON(1); | |
2046 | return ret; | |
2047 | } | |
50f1319c | 2048 | ret = 0; |
38c227d8 LB |
2049 | |
2050 | while (1) { | |
2051 | cond_resched(); | |
2052 | ||
2053 | leaf = path->nodes[0]; | |
2054 | slot = path->slots[0]; | |
2055 | ||
2056 | if (slot >= btrfs_header_nritems(leaf)) { | |
2057 | ret = btrfs_next_leaf(root, path); | |
2058 | if (ret < 0) { | |
2059 | goto out; | |
2060 | } else if (ret > 0) { | |
2061 | ret = 0; | |
2062 | goto out; | |
2063 | } | |
2064 | continue; | |
2065 | } | |
2066 | ||
2067 | path->slots[0]++; | |
2068 | ||
2069 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2070 | ||
2071 | if (key.objectid > inum) | |
2072 | goto out; | |
2073 | ||
2074 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2075 | continue; | |
2076 | ||
2077 | extent = btrfs_item_ptr(leaf, slot, | |
2078 | struct btrfs_file_extent_item); | |
2079 | ||
2080 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2081 | continue; | |
2082 | ||
e68afa49 LB |
2083 | /* |
2084 | * 'offset' refers to the exact key.offset, | |
2085 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2086 | * (key.offset - extent_offset). | |
2087 | */ | |
2088 | if (key.offset != offset) | |
38c227d8 LB |
2089 | continue; |
2090 | ||
e68afa49 | 2091 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2092 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2093 | |
38c227d8 LB |
2094 | if (extent_offset >= old->extent_offset + old->offset + |
2095 | old->len || extent_offset + num_bytes <= | |
2096 | old->extent_offset + old->offset) | |
2097 | continue; | |
38c227d8 LB |
2098 | break; |
2099 | } | |
2100 | ||
2101 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2102 | if (!backref) { | |
2103 | ret = -ENOENT; | |
2104 | goto out; | |
2105 | } | |
2106 | ||
2107 | backref->root_id = root_id; | |
2108 | backref->inum = inum; | |
e68afa49 | 2109 | backref->file_pos = offset; |
38c227d8 LB |
2110 | backref->num_bytes = num_bytes; |
2111 | backref->extent_offset = extent_offset; | |
2112 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2113 | backref->old = old; | |
2114 | backref_insert(&new->root, backref); | |
2115 | old->count++; | |
2116 | out: | |
2117 | btrfs_release_path(path); | |
2118 | WARN_ON(ret); | |
2119 | return ret; | |
2120 | } | |
2121 | ||
2122 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2123 | struct new_sa_defrag_extent *new) | |
2124 | { | |
2125 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2126 | struct old_sa_defrag_extent *old, *tmp; | |
2127 | int ret; | |
2128 | ||
2129 | new->path = path; | |
2130 | ||
2131 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2132 | ret = iterate_inodes_from_logical(old->bytenr + |
2133 | old->extent_offset, fs_info, | |
38c227d8 LB |
2134 | path, record_one_backref, |
2135 | old); | |
2136 | BUG_ON(ret < 0 && ret != -ENOENT); | |
2137 | ||
2138 | /* no backref to be processed for this extent */ | |
2139 | if (!old->count) { | |
2140 | list_del(&old->list); | |
2141 | kfree(old); | |
2142 | } | |
2143 | } | |
2144 | ||
2145 | if (list_empty(&new->head)) | |
2146 | return false; | |
2147 | ||
2148 | return true; | |
2149 | } | |
2150 | ||
2151 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2152 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2153 | struct new_sa_defrag_extent *new) |
38c227d8 | 2154 | { |
116e0024 | 2155 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2156 | return 0; |
2157 | ||
2158 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2159 | return 0; | |
2160 | ||
116e0024 LB |
2161 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2162 | return 0; | |
2163 | ||
2164 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2165 | btrfs_file_extent_other_encoding(leaf, fi)) |
2166 | return 0; | |
2167 | ||
2168 | return 1; | |
2169 | } | |
2170 | ||
2171 | /* | |
2172 | * Note the backref might has changed, and in this case we just return 0. | |
2173 | */ | |
2174 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2175 | struct sa_defrag_extent_backref *prev, | |
2176 | struct sa_defrag_extent_backref *backref) | |
2177 | { | |
2178 | struct btrfs_file_extent_item *extent; | |
2179 | struct btrfs_file_extent_item *item; | |
2180 | struct btrfs_ordered_extent *ordered; | |
2181 | struct btrfs_trans_handle *trans; | |
2182 | struct btrfs_fs_info *fs_info; | |
2183 | struct btrfs_root *root; | |
2184 | struct btrfs_key key; | |
2185 | struct extent_buffer *leaf; | |
2186 | struct old_sa_defrag_extent *old = backref->old; | |
2187 | struct new_sa_defrag_extent *new = old->new; | |
2188 | struct inode *src_inode = new->inode; | |
2189 | struct inode *inode; | |
2190 | struct extent_state *cached = NULL; | |
2191 | int ret = 0; | |
2192 | u64 start; | |
2193 | u64 len; | |
2194 | u64 lock_start; | |
2195 | u64 lock_end; | |
2196 | bool merge = false; | |
2197 | int index; | |
2198 | ||
2199 | if (prev && prev->root_id == backref->root_id && | |
2200 | prev->inum == backref->inum && | |
2201 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2202 | merge = true; | |
2203 | ||
2204 | /* step 1: get root */ | |
2205 | key.objectid = backref->root_id; | |
2206 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2207 | key.offset = (u64)-1; | |
2208 | ||
2209 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2210 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2211 | ||
2212 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2213 | if (IS_ERR(root)) { | |
2214 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2215 | if (PTR_ERR(root) == -ENOENT) | |
2216 | return 0; | |
2217 | return PTR_ERR(root); | |
2218 | } | |
38c227d8 LB |
2219 | |
2220 | /* step 2: get inode */ | |
2221 | key.objectid = backref->inum; | |
2222 | key.type = BTRFS_INODE_ITEM_KEY; | |
2223 | key.offset = 0; | |
2224 | ||
2225 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2226 | if (IS_ERR(inode)) { | |
2227 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2228 | return 0; | |
2229 | } | |
2230 | ||
2231 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2232 | ||
2233 | /* step 3: relink backref */ | |
2234 | lock_start = backref->file_pos; | |
2235 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2236 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2237 | 0, &cached); | |
2238 | ||
2239 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2240 | if (ordered) { | |
2241 | btrfs_put_ordered_extent(ordered); | |
2242 | goto out_unlock; | |
2243 | } | |
2244 | ||
2245 | trans = btrfs_join_transaction(root); | |
2246 | if (IS_ERR(trans)) { | |
2247 | ret = PTR_ERR(trans); | |
2248 | goto out_unlock; | |
2249 | } | |
2250 | ||
2251 | key.objectid = backref->inum; | |
2252 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2253 | key.offset = backref->file_pos; | |
2254 | ||
2255 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2256 | if (ret < 0) { | |
2257 | goto out_free_path; | |
2258 | } else if (ret > 0) { | |
2259 | ret = 0; | |
2260 | goto out_free_path; | |
2261 | } | |
2262 | ||
2263 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2264 | struct btrfs_file_extent_item); | |
2265 | ||
2266 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2267 | backref->generation) | |
2268 | goto out_free_path; | |
2269 | ||
2270 | btrfs_release_path(path); | |
2271 | ||
2272 | start = backref->file_pos; | |
2273 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2274 | start += old->extent_offset + old->offset - | |
2275 | backref->extent_offset; | |
2276 | ||
2277 | len = min(backref->extent_offset + backref->num_bytes, | |
2278 | old->extent_offset + old->offset + old->len); | |
2279 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2280 | ||
2281 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2282 | start + len, 1); | |
2283 | if (ret) | |
2284 | goto out_free_path; | |
2285 | again: | |
2286 | key.objectid = btrfs_ino(inode); | |
2287 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2288 | key.offset = start; | |
2289 | ||
a09a0a70 | 2290 | path->leave_spinning = 1; |
38c227d8 LB |
2291 | if (merge) { |
2292 | struct btrfs_file_extent_item *fi; | |
2293 | u64 extent_len; | |
2294 | struct btrfs_key found_key; | |
2295 | ||
2296 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | |
2297 | if (ret < 0) | |
2298 | goto out_free_path; | |
2299 | ||
2300 | path->slots[0]--; | |
2301 | leaf = path->nodes[0]; | |
2302 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2303 | ||
2304 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2305 | struct btrfs_file_extent_item); | |
2306 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2307 | ||
116e0024 LB |
2308 | if (extent_len + found_key.offset == start && |
2309 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2310 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2311 | extent_len + len); | |
2312 | btrfs_mark_buffer_dirty(leaf); | |
2313 | inode_add_bytes(inode, len); | |
2314 | ||
2315 | ret = 1; | |
2316 | goto out_free_path; | |
2317 | } else { | |
2318 | merge = false; | |
2319 | btrfs_release_path(path); | |
2320 | goto again; | |
2321 | } | |
2322 | } | |
2323 | ||
2324 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2325 | sizeof(*extent)); | |
2326 | if (ret) { | |
2327 | btrfs_abort_transaction(trans, root, ret); | |
2328 | goto out_free_path; | |
2329 | } | |
2330 | ||
2331 | leaf = path->nodes[0]; | |
2332 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2333 | struct btrfs_file_extent_item); | |
2334 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2335 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2336 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2337 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2338 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2339 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2340 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2341 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2342 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2343 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2344 | ||
2345 | btrfs_mark_buffer_dirty(leaf); | |
2346 | inode_add_bytes(inode, len); | |
a09a0a70 | 2347 | btrfs_release_path(path); |
38c227d8 LB |
2348 | |
2349 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2350 | new->disk_len, 0, | |
2351 | backref->root_id, backref->inum, | |
2352 | new->file_pos, 0); /* start - extent_offset */ | |
2353 | if (ret) { | |
2354 | btrfs_abort_transaction(trans, root, ret); | |
2355 | goto out_free_path; | |
2356 | } | |
2357 | ||
2358 | ret = 1; | |
2359 | out_free_path: | |
2360 | btrfs_release_path(path); | |
a09a0a70 | 2361 | path->leave_spinning = 0; |
38c227d8 LB |
2362 | btrfs_end_transaction(trans, root); |
2363 | out_unlock: | |
2364 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2365 | &cached, GFP_NOFS); | |
2366 | iput(inode); | |
2367 | return ret; | |
2368 | } | |
2369 | ||
2370 | static void relink_file_extents(struct new_sa_defrag_extent *new) | |
2371 | { | |
2372 | struct btrfs_path *path; | |
2373 | struct old_sa_defrag_extent *old, *tmp; | |
2374 | struct sa_defrag_extent_backref *backref; | |
2375 | struct sa_defrag_extent_backref *prev = NULL; | |
2376 | struct inode *inode; | |
2377 | struct btrfs_root *root; | |
2378 | struct rb_node *node; | |
2379 | int ret; | |
2380 | ||
2381 | inode = new->inode; | |
2382 | root = BTRFS_I(inode)->root; | |
2383 | ||
2384 | path = btrfs_alloc_path(); | |
2385 | if (!path) | |
2386 | return; | |
2387 | ||
2388 | if (!record_extent_backrefs(path, new)) { | |
2389 | btrfs_free_path(path); | |
2390 | goto out; | |
2391 | } | |
2392 | btrfs_release_path(path); | |
2393 | ||
2394 | while (1) { | |
2395 | node = rb_first(&new->root); | |
2396 | if (!node) | |
2397 | break; | |
2398 | rb_erase(node, &new->root); | |
2399 | ||
2400 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2401 | ||
2402 | ret = relink_extent_backref(path, prev, backref); | |
2403 | WARN_ON(ret < 0); | |
2404 | ||
2405 | kfree(prev); | |
2406 | ||
2407 | if (ret == 1) | |
2408 | prev = backref; | |
2409 | else | |
2410 | prev = NULL; | |
2411 | cond_resched(); | |
2412 | } | |
2413 | kfree(prev); | |
2414 | ||
2415 | btrfs_free_path(path); | |
2416 | ||
2417 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2418 | list_del(&old->list); | |
2419 | kfree(old); | |
2420 | } | |
2421 | out: | |
2422 | atomic_dec(&root->fs_info->defrag_running); | |
2423 | wake_up(&root->fs_info->transaction_wait); | |
2424 | ||
2425 | kfree(new); | |
2426 | } | |
2427 | ||
2428 | static struct new_sa_defrag_extent * | |
2429 | record_old_file_extents(struct inode *inode, | |
2430 | struct btrfs_ordered_extent *ordered) | |
2431 | { | |
2432 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2433 | struct btrfs_path *path; | |
2434 | struct btrfs_key key; | |
2435 | struct old_sa_defrag_extent *old, *tmp; | |
2436 | struct new_sa_defrag_extent *new; | |
2437 | int ret; | |
2438 | ||
2439 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2440 | if (!new) | |
2441 | return NULL; | |
2442 | ||
2443 | new->inode = inode; | |
2444 | new->file_pos = ordered->file_offset; | |
2445 | new->len = ordered->len; | |
2446 | new->bytenr = ordered->start; | |
2447 | new->disk_len = ordered->disk_len; | |
2448 | new->compress_type = ordered->compress_type; | |
2449 | new->root = RB_ROOT; | |
2450 | INIT_LIST_HEAD(&new->head); | |
2451 | ||
2452 | path = btrfs_alloc_path(); | |
2453 | if (!path) | |
2454 | goto out_kfree; | |
2455 | ||
2456 | key.objectid = btrfs_ino(inode); | |
2457 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2458 | key.offset = new->file_pos; | |
2459 | ||
2460 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2461 | if (ret < 0) | |
2462 | goto out_free_path; | |
2463 | if (ret > 0 && path->slots[0] > 0) | |
2464 | path->slots[0]--; | |
2465 | ||
2466 | /* find out all the old extents for the file range */ | |
2467 | while (1) { | |
2468 | struct btrfs_file_extent_item *extent; | |
2469 | struct extent_buffer *l; | |
2470 | int slot; | |
2471 | u64 num_bytes; | |
2472 | u64 offset; | |
2473 | u64 end; | |
2474 | u64 disk_bytenr; | |
2475 | u64 extent_offset; | |
2476 | ||
2477 | l = path->nodes[0]; | |
2478 | slot = path->slots[0]; | |
2479 | ||
2480 | if (slot >= btrfs_header_nritems(l)) { | |
2481 | ret = btrfs_next_leaf(root, path); | |
2482 | if (ret < 0) | |
2483 | goto out_free_list; | |
2484 | else if (ret > 0) | |
2485 | break; | |
2486 | continue; | |
2487 | } | |
2488 | ||
2489 | btrfs_item_key_to_cpu(l, &key, slot); | |
2490 | ||
2491 | if (key.objectid != btrfs_ino(inode)) | |
2492 | break; | |
2493 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2494 | break; | |
2495 | if (key.offset >= new->file_pos + new->len) | |
2496 | break; | |
2497 | ||
2498 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2499 | ||
2500 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2501 | if (key.offset + num_bytes < new->file_pos) | |
2502 | goto next; | |
2503 | ||
2504 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2505 | if (!disk_bytenr) | |
2506 | goto next; | |
2507 | ||
2508 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2509 | ||
2510 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2511 | if (!old) | |
2512 | goto out_free_list; | |
2513 | ||
2514 | offset = max(new->file_pos, key.offset); | |
2515 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2516 | ||
2517 | old->bytenr = disk_bytenr; | |
2518 | old->extent_offset = extent_offset; | |
2519 | old->offset = offset - key.offset; | |
2520 | old->len = end - offset; | |
2521 | old->new = new; | |
2522 | old->count = 0; | |
2523 | list_add_tail(&old->list, &new->head); | |
2524 | next: | |
2525 | path->slots[0]++; | |
2526 | cond_resched(); | |
2527 | } | |
2528 | ||
2529 | btrfs_free_path(path); | |
2530 | atomic_inc(&root->fs_info->defrag_running); | |
2531 | ||
2532 | return new; | |
2533 | ||
2534 | out_free_list: | |
2535 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2536 | list_del(&old->list); | |
2537 | kfree(old); | |
2538 | } | |
2539 | out_free_path: | |
2540 | btrfs_free_path(path); | |
2541 | out_kfree: | |
2542 | kfree(new); | |
2543 | return NULL; | |
2544 | } | |
2545 | ||
5d13a98f CM |
2546 | /* |
2547 | * helper function for btrfs_finish_ordered_io, this | |
2548 | * just reads in some of the csum leaves to prime them into ram | |
2549 | * before we start the transaction. It limits the amount of btree | |
2550 | * reads required while inside the transaction. | |
2551 | */ | |
d352ac68 CM |
2552 | /* as ordered data IO finishes, this gets called so we can finish |
2553 | * an ordered extent if the range of bytes in the file it covers are | |
2554 | * fully written. | |
2555 | */ | |
5fd02043 | 2556 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2557 | { |
5fd02043 | 2558 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2559 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2560 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2561 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2562 | struct extent_state *cached_state = NULL; |
38c227d8 | 2563 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2564 | int compress_type = 0; |
e6dcd2dc | 2565 | int ret; |
82d5902d | 2566 | bool nolock; |
e6dcd2dc | 2567 | |
83eea1f1 | 2568 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2569 | |
5fd02043 JB |
2570 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2571 | ret = -EIO; | |
2572 | goto out; | |
2573 | } | |
2574 | ||
c2167754 | 2575 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2576 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2577 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2578 | if (nolock) | |
2579 | trans = btrfs_join_transaction_nolock(root); | |
2580 | else | |
2581 | trans = btrfs_join_transaction(root); | |
2582 | if (IS_ERR(trans)) { | |
2583 | ret = PTR_ERR(trans); | |
2584 | trans = NULL; | |
2585 | goto out; | |
c2167754 | 2586 | } |
6c760c07 JB |
2587 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2588 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2589 | if (ret) /* -ENOMEM or corruption */ | |
2590 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2591 | goto out; |
2592 | } | |
e6dcd2dc | 2593 | |
2ac55d41 JB |
2594 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2595 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2596 | 0, &cached_state); |
e6dcd2dc | 2597 | |
38c227d8 LB |
2598 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2599 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2600 | EXTENT_DEFRAG, 1, cached_state); | |
2601 | if (ret) { | |
2602 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
2603 | if (last_snapshot >= BTRFS_I(inode)->generation) | |
2604 | /* the inode is shared */ | |
2605 | new = record_old_file_extents(inode, ordered_extent); | |
2606 | ||
2607 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2608 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2609 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2610 | } | |
2611 | ||
0cb59c99 | 2612 | if (nolock) |
7a7eaa40 | 2613 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2614 | else |
7a7eaa40 | 2615 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2616 | if (IS_ERR(trans)) { |
2617 | ret = PTR_ERR(trans); | |
2618 | trans = NULL; | |
2619 | goto out_unlock; | |
2620 | } | |
0ca1f7ce | 2621 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2622 | |
c8b97818 | 2623 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2624 | compress_type = ordered_extent->compress_type; |
d899e052 | 2625 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2626 | BUG_ON(compress_type); |
920bbbfb | 2627 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2628 | ordered_extent->file_offset, |
2629 | ordered_extent->file_offset + | |
2630 | ordered_extent->len); | |
d899e052 | 2631 | } else { |
0af3d00b | 2632 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2633 | ret = insert_reserved_file_extent(trans, inode, |
2634 | ordered_extent->file_offset, | |
2635 | ordered_extent->start, | |
2636 | ordered_extent->disk_len, | |
2637 | ordered_extent->len, | |
2638 | ordered_extent->len, | |
261507a0 | 2639 | compress_type, 0, 0, |
d899e052 | 2640 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 2641 | } |
5dc562c5 JB |
2642 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2643 | ordered_extent->file_offset, ordered_extent->len, | |
2644 | trans->transid); | |
79787eaa JM |
2645 | if (ret < 0) { |
2646 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2647 | goto out_unlock; |
79787eaa | 2648 | } |
2ac55d41 | 2649 | |
e6dcd2dc CM |
2650 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2651 | &ordered_extent->list); | |
2652 | ||
6c760c07 JB |
2653 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2654 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2655 | if (ret) { /* -ENOMEM or corruption */ | |
2656 | btrfs_abort_transaction(trans, root, ret); | |
2657 | goto out_unlock; | |
1ef30be1 JB |
2658 | } |
2659 | ret = 0; | |
5fd02043 JB |
2660 | out_unlock: |
2661 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2662 | ordered_extent->file_offset + | |
2663 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2664 | out: |
5b0e95bf | 2665 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2666 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2667 | if (trans) |
2668 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2669 | |
0bec9ef5 | 2670 | if (ret) { |
5fd02043 JB |
2671 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, |
2672 | ordered_extent->file_offset + | |
2673 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
2674 | ||
0bec9ef5 JB |
2675 | /* |
2676 | * If the ordered extent had an IOERR or something else went | |
2677 | * wrong we need to return the space for this ordered extent | |
2678 | * back to the allocator. | |
2679 | */ | |
2680 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2681 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) | |
2682 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2683 | ordered_extent->disk_len); | |
2684 | } | |
2685 | ||
2686 | ||
5fd02043 | 2687 | /* |
8bad3c02 LB |
2688 | * This needs to be done to make sure anybody waiting knows we are done |
2689 | * updating everything for this ordered extent. | |
5fd02043 JB |
2690 | */ |
2691 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2692 | ||
38c227d8 LB |
2693 | /* for snapshot-aware defrag */ |
2694 | if (new) | |
2695 | relink_file_extents(new); | |
2696 | ||
e6dcd2dc CM |
2697 | /* once for us */ |
2698 | btrfs_put_ordered_extent(ordered_extent); | |
2699 | /* once for the tree */ | |
2700 | btrfs_put_ordered_extent(ordered_extent); | |
2701 | ||
5fd02043 JB |
2702 | return ret; |
2703 | } | |
2704 | ||
2705 | static void finish_ordered_fn(struct btrfs_work *work) | |
2706 | { | |
2707 | struct btrfs_ordered_extent *ordered_extent; | |
2708 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2709 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2710 | } |
2711 | ||
b2950863 | 2712 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2713 | struct extent_state *state, int uptodate) |
2714 | { | |
5fd02043 JB |
2715 | struct inode *inode = page->mapping->host; |
2716 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2717 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2718 | struct btrfs_workers *workers; | |
2719 | ||
1abe9b8a | 2720 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2721 | ||
8b62b72b | 2722 | ClearPagePrivate2(page); |
5fd02043 JB |
2723 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2724 | end - start + 1, uptodate)) | |
2725 | return 0; | |
2726 | ||
2727 | ordered_extent->work.func = finish_ordered_fn; | |
2728 | ordered_extent->work.flags = 0; | |
2729 | ||
83eea1f1 | 2730 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2731 | workers = &root->fs_info->endio_freespace_worker; |
2732 | else | |
2733 | workers = &root->fs_info->endio_write_workers; | |
2734 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2735 | ||
2736 | return 0; | |
211f90e6 CM |
2737 | } |
2738 | ||
d352ac68 CM |
2739 | /* |
2740 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2741 | * if there's a match, we allow the bio to finish. If not, the code in |
2742 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2743 | */ |
facc8a22 MX |
2744 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
2745 | u64 phy_offset, struct page *page, | |
2746 | u64 start, u64 end, int mirror) | |
07157aac | 2747 | { |
4eee4fa4 | 2748 | size_t offset = start - page_offset(page); |
07157aac | 2749 | struct inode *inode = page->mapping->host; |
d1310b2e | 2750 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2751 | char *kaddr; |
ff79f819 | 2752 | struct btrfs_root *root = BTRFS_I(inode)->root; |
facc8a22 | 2753 | u32 csum_expected; |
ff79f819 | 2754 | u32 csum = ~(u32)0; |
c2cf52eb SK |
2755 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
2756 | DEFAULT_RATELIMIT_BURST); | |
d1310b2e | 2757 | |
d20f7043 CM |
2758 | if (PageChecked(page)) { |
2759 | ClearPageChecked(page); | |
2760 | goto good; | |
2761 | } | |
6cbff00f CH |
2762 | |
2763 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2764 | goto good; |
17d217fe YZ |
2765 | |
2766 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2767 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2768 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2769 | GFP_NOFS); | |
b6cda9bc | 2770 | return 0; |
17d217fe | 2771 | } |
d20f7043 | 2772 | |
facc8a22 MX |
2773 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
2774 | csum_expected = *(((u32 *)io_bio->csum) + phy_offset); | |
d397712b | 2775 | |
facc8a22 | 2776 | kaddr = kmap_atomic(page); |
b0496686 | 2777 | csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1); |
ff79f819 | 2778 | btrfs_csum_final(csum, (char *)&csum); |
facc8a22 | 2779 | if (csum != csum_expected) |
07157aac | 2780 | goto zeroit; |
d397712b | 2781 | |
7ac687d9 | 2782 | kunmap_atomic(kaddr); |
d20f7043 | 2783 | good: |
07157aac CM |
2784 | return 0; |
2785 | ||
2786 | zeroit: | |
c2cf52eb | 2787 | if (__ratelimit(&_rs)) |
facc8a22 | 2788 | btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", |
c2cf52eb | 2789 | (unsigned long long)btrfs_ino(page->mapping->host), |
facc8a22 | 2790 | (unsigned long long)start, csum, csum_expected); |
db94535d CM |
2791 | memset(kaddr + offset, 1, end - start + 1); |
2792 | flush_dcache_page(page); | |
7ac687d9 | 2793 | kunmap_atomic(kaddr); |
facc8a22 | 2794 | if (csum_expected == 0) |
3b951516 | 2795 | return 0; |
7e38326f | 2796 | return -EIO; |
07157aac | 2797 | } |
b888db2b | 2798 | |
24bbcf04 YZ |
2799 | struct delayed_iput { |
2800 | struct list_head list; | |
2801 | struct inode *inode; | |
2802 | }; | |
2803 | ||
79787eaa JM |
2804 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2805 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2806 | void btrfs_add_delayed_iput(struct inode *inode) |
2807 | { | |
2808 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2809 | struct delayed_iput *delayed; | |
2810 | ||
2811 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2812 | return; | |
2813 | ||
2814 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2815 | delayed->inode = inode; | |
2816 | ||
2817 | spin_lock(&fs_info->delayed_iput_lock); | |
2818 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2819 | spin_unlock(&fs_info->delayed_iput_lock); | |
2820 | } | |
2821 | ||
2822 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2823 | { | |
2824 | LIST_HEAD(list); | |
2825 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2826 | struct delayed_iput *delayed; | |
2827 | int empty; | |
2828 | ||
2829 | spin_lock(&fs_info->delayed_iput_lock); | |
2830 | empty = list_empty(&fs_info->delayed_iputs); | |
2831 | spin_unlock(&fs_info->delayed_iput_lock); | |
2832 | if (empty) | |
2833 | return; | |
2834 | ||
24bbcf04 YZ |
2835 | spin_lock(&fs_info->delayed_iput_lock); |
2836 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2837 | spin_unlock(&fs_info->delayed_iput_lock); | |
2838 | ||
2839 | while (!list_empty(&list)) { | |
2840 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2841 | list_del(&delayed->list); | |
2842 | iput(delayed->inode); | |
2843 | kfree(delayed); | |
2844 | } | |
24bbcf04 YZ |
2845 | } |
2846 | ||
d68fc57b | 2847 | /* |
42b2aa86 | 2848 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2849 | * files in the subvolume, it removes orphan item and frees block_rsv |
2850 | * structure. | |
2851 | */ | |
2852 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2853 | struct btrfs_root *root) | |
2854 | { | |
90290e19 | 2855 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2856 | int ret; |
2857 | ||
8a35d95f | 2858 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2859 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2860 | return; | |
2861 | ||
90290e19 | 2862 | spin_lock(&root->orphan_lock); |
8a35d95f | 2863 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2864 | spin_unlock(&root->orphan_lock); |
2865 | return; | |
2866 | } | |
2867 | ||
2868 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2869 | spin_unlock(&root->orphan_lock); | |
2870 | return; | |
2871 | } | |
2872 | ||
2873 | block_rsv = root->orphan_block_rsv; | |
2874 | root->orphan_block_rsv = NULL; | |
2875 | spin_unlock(&root->orphan_lock); | |
2876 | ||
d68fc57b YZ |
2877 | if (root->orphan_item_inserted && |
2878 | btrfs_root_refs(&root->root_item) > 0) { | |
2879 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2880 | root->root_key.objectid); | |
4ef31a45 JB |
2881 | if (ret) |
2882 | btrfs_abort_transaction(trans, root, ret); | |
2883 | else | |
2884 | root->orphan_item_inserted = 0; | |
d68fc57b YZ |
2885 | } |
2886 | ||
90290e19 JB |
2887 | if (block_rsv) { |
2888 | WARN_ON(block_rsv->size > 0); | |
2889 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2890 | } |
2891 | } | |
2892 | ||
7b128766 JB |
2893 | /* |
2894 | * This creates an orphan entry for the given inode in case something goes | |
2895 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2896 | * |
2897 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2898 | * this function. | |
7b128766 JB |
2899 | */ |
2900 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2901 | { | |
2902 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2903 | struct btrfs_block_rsv *block_rsv = NULL; |
2904 | int reserve = 0; | |
2905 | int insert = 0; | |
2906 | int ret; | |
7b128766 | 2907 | |
d68fc57b | 2908 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2909 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2910 | if (!block_rsv) |
2911 | return -ENOMEM; | |
d68fc57b | 2912 | } |
7b128766 | 2913 | |
d68fc57b YZ |
2914 | spin_lock(&root->orphan_lock); |
2915 | if (!root->orphan_block_rsv) { | |
2916 | root->orphan_block_rsv = block_rsv; | |
2917 | } else if (block_rsv) { | |
2918 | btrfs_free_block_rsv(root, block_rsv); | |
2919 | block_rsv = NULL; | |
7b128766 | 2920 | } |
7b128766 | 2921 | |
8a35d95f JB |
2922 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2923 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2924 | #if 0 |
2925 | /* | |
2926 | * For proper ENOSPC handling, we should do orphan | |
2927 | * cleanup when mounting. But this introduces backward | |
2928 | * compatibility issue. | |
2929 | */ | |
2930 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2931 | insert = 2; | |
2932 | else | |
2933 | insert = 1; | |
2934 | #endif | |
2935 | insert = 1; | |
321f0e70 | 2936 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2937 | } |
2938 | ||
72ac3c0d JB |
2939 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2940 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2941 | reserve = 1; |
d68fc57b | 2942 | spin_unlock(&root->orphan_lock); |
7b128766 | 2943 | |
d68fc57b YZ |
2944 | /* grab metadata reservation from transaction handle */ |
2945 | if (reserve) { | |
2946 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2947 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2948 | } |
7b128766 | 2949 | |
d68fc57b YZ |
2950 | /* insert an orphan item to track this unlinked/truncated file */ |
2951 | if (insert >= 1) { | |
33345d01 | 2952 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 2953 | if (ret) { |
8a35d95f JB |
2954 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2955 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
2956 | if (reserve) { |
2957 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
2958 | &BTRFS_I(inode)->runtime_flags); | |
2959 | btrfs_orphan_release_metadata(inode); | |
2960 | } | |
2961 | if (ret != -EEXIST) { | |
2962 | btrfs_abort_transaction(trans, root, ret); | |
2963 | return ret; | |
2964 | } | |
79787eaa JM |
2965 | } |
2966 | ret = 0; | |
d68fc57b YZ |
2967 | } |
2968 | ||
2969 | /* insert an orphan item to track subvolume contains orphan files */ | |
2970 | if (insert >= 2) { | |
2971 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2972 | root->root_key.objectid); | |
79787eaa JM |
2973 | if (ret && ret != -EEXIST) { |
2974 | btrfs_abort_transaction(trans, root, ret); | |
2975 | return ret; | |
2976 | } | |
d68fc57b YZ |
2977 | } |
2978 | return 0; | |
7b128766 JB |
2979 | } |
2980 | ||
2981 | /* | |
2982 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2983 | * item for this particular inode. | |
2984 | */ | |
48a3b636 ES |
2985 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
2986 | struct inode *inode) | |
7b128766 JB |
2987 | { |
2988 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2989 | int delete_item = 0; |
2990 | int release_rsv = 0; | |
7b128766 JB |
2991 | int ret = 0; |
2992 | ||
d68fc57b | 2993 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2994 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2995 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2996 | delete_item = 1; |
7b128766 | 2997 | |
72ac3c0d JB |
2998 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2999 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3000 | release_rsv = 1; |
d68fc57b | 3001 | spin_unlock(&root->orphan_lock); |
7b128766 | 3002 | |
4ef31a45 | 3003 | if (trans && delete_item) |
33345d01 | 3004 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
7b128766 | 3005 | |
8a35d95f | 3006 | if (release_rsv) { |
d68fc57b | 3007 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
3008 | atomic_dec(&root->orphan_inodes); |
3009 | } | |
7b128766 | 3010 | |
4ef31a45 | 3011 | return ret; |
7b128766 JB |
3012 | } |
3013 | ||
3014 | /* | |
3015 | * this cleans up any orphans that may be left on the list from the last use | |
3016 | * of this root. | |
3017 | */ | |
66b4ffd1 | 3018 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3019 | { |
3020 | struct btrfs_path *path; | |
3021 | struct extent_buffer *leaf; | |
7b128766 JB |
3022 | struct btrfs_key key, found_key; |
3023 | struct btrfs_trans_handle *trans; | |
3024 | struct inode *inode; | |
8f6d7f4f | 3025 | u64 last_objectid = 0; |
7b128766 JB |
3026 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3027 | ||
d68fc57b | 3028 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3029 | return 0; |
c71bf099 YZ |
3030 | |
3031 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3032 | if (!path) { |
3033 | ret = -ENOMEM; | |
3034 | goto out; | |
3035 | } | |
7b128766 JB |
3036 | path->reada = -1; |
3037 | ||
3038 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
3039 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
3040 | key.offset = (u64)-1; | |
3041 | ||
7b128766 JB |
3042 | while (1) { |
3043 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3044 | if (ret < 0) |
3045 | goto out; | |
7b128766 JB |
3046 | |
3047 | /* | |
3048 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3049 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3050 | * find the key and see if we have stuff that matches |
3051 | */ | |
3052 | if (ret > 0) { | |
66b4ffd1 | 3053 | ret = 0; |
7b128766 JB |
3054 | if (path->slots[0] == 0) |
3055 | break; | |
3056 | path->slots[0]--; | |
3057 | } | |
3058 | ||
3059 | /* pull out the item */ | |
3060 | leaf = path->nodes[0]; | |
7b128766 JB |
3061 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3062 | ||
3063 | /* make sure the item matches what we want */ | |
3064 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3065 | break; | |
3066 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
3067 | break; | |
3068 | ||
3069 | /* release the path since we're done with it */ | |
b3b4aa74 | 3070 | btrfs_release_path(path); |
7b128766 JB |
3071 | |
3072 | /* | |
3073 | * this is where we are basically btrfs_lookup, without the | |
3074 | * crossing root thing. we store the inode number in the | |
3075 | * offset of the orphan item. | |
3076 | */ | |
8f6d7f4f JB |
3077 | |
3078 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3079 | btrfs_err(root->fs_info, |
3080 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3081 | ret = -EINVAL; |
3082 | goto out; | |
3083 | } | |
3084 | ||
3085 | last_objectid = found_key.offset; | |
3086 | ||
5d4f98a2 YZ |
3087 | found_key.objectid = found_key.offset; |
3088 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3089 | found_key.offset = 0; | |
73f73415 | 3090 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
3091 | ret = PTR_RET(inode); |
3092 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 3093 | goto out; |
7b128766 | 3094 | |
f8e9e0b0 AJ |
3095 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3096 | struct btrfs_root *dead_root; | |
3097 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3098 | int is_dead_root = 0; | |
3099 | ||
3100 | /* | |
3101 | * this is an orphan in the tree root. Currently these | |
3102 | * could come from 2 sources: | |
3103 | * a) a snapshot deletion in progress | |
3104 | * b) a free space cache inode | |
3105 | * We need to distinguish those two, as the snapshot | |
3106 | * orphan must not get deleted. | |
3107 | * find_dead_roots already ran before us, so if this | |
3108 | * is a snapshot deletion, we should find the root | |
3109 | * in the dead_roots list | |
3110 | */ | |
3111 | spin_lock(&fs_info->trans_lock); | |
3112 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3113 | root_list) { | |
3114 | if (dead_root->root_key.objectid == | |
3115 | found_key.objectid) { | |
3116 | is_dead_root = 1; | |
3117 | break; | |
3118 | } | |
3119 | } | |
3120 | spin_unlock(&fs_info->trans_lock); | |
3121 | if (is_dead_root) { | |
3122 | /* prevent this orphan from being found again */ | |
3123 | key.offset = found_key.objectid - 1; | |
3124 | continue; | |
3125 | } | |
3126 | } | |
7b128766 | 3127 | /* |
a8c9e576 JB |
3128 | * Inode is already gone but the orphan item is still there, |
3129 | * kill the orphan item. | |
7b128766 | 3130 | */ |
a8c9e576 JB |
3131 | if (ret == -ESTALE) { |
3132 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3133 | if (IS_ERR(trans)) { |
3134 | ret = PTR_ERR(trans); | |
3135 | goto out; | |
3136 | } | |
c2cf52eb SK |
3137 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3138 | found_key.objectid); | |
a8c9e576 JB |
3139 | ret = btrfs_del_orphan_item(trans, root, |
3140 | found_key.objectid); | |
5b21f2ed | 3141 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3142 | if (ret) |
3143 | goto out; | |
7b128766 JB |
3144 | continue; |
3145 | } | |
3146 | ||
a8c9e576 JB |
3147 | /* |
3148 | * add this inode to the orphan list so btrfs_orphan_del does | |
3149 | * the proper thing when we hit it | |
3150 | */ | |
8a35d95f JB |
3151 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3152 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3153 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3154 | |
7b128766 JB |
3155 | /* if we have links, this was a truncate, lets do that */ |
3156 | if (inode->i_nlink) { | |
a41ad394 JB |
3157 | if (!S_ISREG(inode->i_mode)) { |
3158 | WARN_ON(1); | |
3159 | iput(inode); | |
3160 | continue; | |
3161 | } | |
7b128766 | 3162 | nr_truncate++; |
f3fe820c JB |
3163 | |
3164 | /* 1 for the orphan item deletion. */ | |
3165 | trans = btrfs_start_transaction(root, 1); | |
3166 | if (IS_ERR(trans)) { | |
c69b26b0 | 3167 | iput(inode); |
f3fe820c JB |
3168 | ret = PTR_ERR(trans); |
3169 | goto out; | |
3170 | } | |
3171 | ret = btrfs_orphan_add(trans, inode); | |
3172 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3173 | if (ret) { |
3174 | iput(inode); | |
f3fe820c | 3175 | goto out; |
c69b26b0 | 3176 | } |
f3fe820c | 3177 | |
66b4ffd1 | 3178 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3179 | if (ret) |
3180 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3181 | } else { |
3182 | nr_unlink++; | |
3183 | } | |
3184 | ||
3185 | /* this will do delete_inode and everything for us */ | |
3186 | iput(inode); | |
66b4ffd1 JB |
3187 | if (ret) |
3188 | goto out; | |
7b128766 | 3189 | } |
3254c876 MX |
3190 | /* release the path since we're done with it */ |
3191 | btrfs_release_path(path); | |
3192 | ||
d68fc57b YZ |
3193 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3194 | ||
3195 | if (root->orphan_block_rsv) | |
3196 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3197 | (u64)-1); | |
3198 | ||
3199 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 3200 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3201 | if (!IS_ERR(trans)) |
3202 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3203 | } |
7b128766 JB |
3204 | |
3205 | if (nr_unlink) | |
4884b476 | 3206 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3207 | if (nr_truncate) |
4884b476 | 3208 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3209 | |
3210 | out: | |
3211 | if (ret) | |
c2cf52eb SK |
3212 | btrfs_crit(root->fs_info, |
3213 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3214 | btrfs_free_path(path); |
3215 | return ret; | |
7b128766 JB |
3216 | } |
3217 | ||
46a53cca CM |
3218 | /* |
3219 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3220 | * don't find any xattrs, we know there can't be any acls. | |
3221 | * | |
3222 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3223 | */ | |
3224 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
3225 | int slot, u64 objectid) | |
3226 | { | |
3227 | u32 nritems = btrfs_header_nritems(leaf); | |
3228 | struct btrfs_key found_key; | |
f23b5a59 JB |
3229 | static u64 xattr_access = 0; |
3230 | static u64 xattr_default = 0; | |
46a53cca CM |
3231 | int scanned = 0; |
3232 | ||
f23b5a59 JB |
3233 | if (!xattr_access) { |
3234 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3235 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3236 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3237 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3238 | } | |
3239 | ||
46a53cca CM |
3240 | slot++; |
3241 | while (slot < nritems) { | |
3242 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3243 | ||
3244 | /* we found a different objectid, there must not be acls */ | |
3245 | if (found_key.objectid != objectid) | |
3246 | return 0; | |
3247 | ||
3248 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 JB |
3249 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
3250 | if (found_key.offset == xattr_access || | |
3251 | found_key.offset == xattr_default) | |
3252 | return 1; | |
3253 | } | |
46a53cca CM |
3254 | |
3255 | /* | |
3256 | * we found a key greater than an xattr key, there can't | |
3257 | * be any acls later on | |
3258 | */ | |
3259 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3260 | return 0; | |
3261 | ||
3262 | slot++; | |
3263 | scanned++; | |
3264 | ||
3265 | /* | |
3266 | * it goes inode, inode backrefs, xattrs, extents, | |
3267 | * so if there are a ton of hard links to an inode there can | |
3268 | * be a lot of backrefs. Don't waste time searching too hard, | |
3269 | * this is just an optimization | |
3270 | */ | |
3271 | if (scanned >= 8) | |
3272 | break; | |
3273 | } | |
3274 | /* we hit the end of the leaf before we found an xattr or | |
3275 | * something larger than an xattr. We have to assume the inode | |
3276 | * has acls | |
3277 | */ | |
3278 | return 1; | |
3279 | } | |
3280 | ||
d352ac68 CM |
3281 | /* |
3282 | * read an inode from the btree into the in-memory inode | |
3283 | */ | |
5d4f98a2 | 3284 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3285 | { |
3286 | struct btrfs_path *path; | |
5f39d397 | 3287 | struct extent_buffer *leaf; |
39279cc3 | 3288 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3289 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3290 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3291 | struct btrfs_key location; | |
46a53cca | 3292 | int maybe_acls; |
618e21d5 | 3293 | u32 rdev; |
39279cc3 | 3294 | int ret; |
2f7e33d4 MX |
3295 | bool filled = false; |
3296 | ||
3297 | ret = btrfs_fill_inode(inode, &rdev); | |
3298 | if (!ret) | |
3299 | filled = true; | |
39279cc3 CM |
3300 | |
3301 | path = btrfs_alloc_path(); | |
1748f843 MF |
3302 | if (!path) |
3303 | goto make_bad; | |
3304 | ||
d90c7321 | 3305 | path->leave_spinning = 1; |
39279cc3 | 3306 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3307 | |
39279cc3 | 3308 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3309 | if (ret) |
39279cc3 | 3310 | goto make_bad; |
39279cc3 | 3311 | |
5f39d397 | 3312 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3313 | |
3314 | if (filled) | |
3315 | goto cache_acl; | |
3316 | ||
5f39d397 CM |
3317 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3318 | struct btrfs_inode_item); | |
5f39d397 | 3319 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3320 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3321 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3322 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3323 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3324 | |
3325 | tspec = btrfs_inode_atime(inode_item); | |
3326 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3327 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3328 | ||
3329 | tspec = btrfs_inode_mtime(inode_item); | |
3330 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3331 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3332 | ||
3333 | tspec = btrfs_inode_ctime(inode_item); | |
3334 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3335 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3336 | ||
a76a3cd4 | 3337 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3338 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3339 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3340 | ||
3341 | /* | |
3342 | * If we were modified in the current generation and evicted from memory | |
3343 | * and then re-read we need to do a full sync since we don't have any | |
3344 | * idea about which extents were modified before we were evicted from | |
3345 | * cache. | |
3346 | */ | |
3347 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3348 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3349 | &BTRFS_I(inode)->runtime_flags); | |
3350 | ||
0c4d2d95 | 3351 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3352 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3353 | inode->i_rdev = 0; |
5f39d397 CM |
3354 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3355 | ||
aec7477b | 3356 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3357 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 3358 | cache_acl: |
46a53cca CM |
3359 | /* |
3360 | * try to precache a NULL acl entry for files that don't have | |
3361 | * any xattrs or acls | |
3362 | */ | |
33345d01 LZ |
3363 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
3364 | btrfs_ino(inode)); | |
72c04902 AV |
3365 | if (!maybe_acls) |
3366 | cache_no_acl(inode); | |
46a53cca | 3367 | |
39279cc3 | 3368 | btrfs_free_path(path); |
39279cc3 | 3369 | |
39279cc3 | 3370 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3371 | case S_IFREG: |
3372 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3373 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3374 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3375 | inode->i_fop = &btrfs_file_operations; |
3376 | inode->i_op = &btrfs_file_inode_operations; | |
3377 | break; | |
3378 | case S_IFDIR: | |
3379 | inode->i_fop = &btrfs_dir_file_operations; | |
3380 | if (root == root->fs_info->tree_root) | |
3381 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3382 | else | |
3383 | inode->i_op = &btrfs_dir_inode_operations; | |
3384 | break; | |
3385 | case S_IFLNK: | |
3386 | inode->i_op = &btrfs_symlink_inode_operations; | |
3387 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3388 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3389 | break; |
618e21d5 | 3390 | default: |
0279b4cd | 3391 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3392 | init_special_inode(inode, inode->i_mode, rdev); |
3393 | break; | |
39279cc3 | 3394 | } |
6cbff00f CH |
3395 | |
3396 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3397 | return; |
3398 | ||
3399 | make_bad: | |
39279cc3 | 3400 | btrfs_free_path(path); |
39279cc3 CM |
3401 | make_bad_inode(inode); |
3402 | } | |
3403 | ||
d352ac68 CM |
3404 | /* |
3405 | * given a leaf and an inode, copy the inode fields into the leaf | |
3406 | */ | |
e02119d5 CM |
3407 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3408 | struct extent_buffer *leaf, | |
5f39d397 | 3409 | struct btrfs_inode_item *item, |
39279cc3 CM |
3410 | struct inode *inode) |
3411 | { | |
51fab693 LB |
3412 | struct btrfs_map_token token; |
3413 | ||
3414 | btrfs_init_map_token(&token); | |
5f39d397 | 3415 | |
51fab693 LB |
3416 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3417 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3418 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3419 | &token); | |
3420 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3421 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3422 | |
51fab693 LB |
3423 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3424 | inode->i_atime.tv_sec, &token); | |
3425 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3426 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3427 | |
51fab693 LB |
3428 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3429 | inode->i_mtime.tv_sec, &token); | |
3430 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3431 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3432 | |
51fab693 LB |
3433 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3434 | inode->i_ctime.tv_sec, &token); | |
3435 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3436 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3437 | |
51fab693 LB |
3438 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3439 | &token); | |
3440 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3441 | &token); | |
3442 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3443 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3444 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3445 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3446 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3447 | } |
3448 | ||
d352ac68 CM |
3449 | /* |
3450 | * copy everything in the in-memory inode into the btree. | |
3451 | */ | |
2115133f | 3452 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3453 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3454 | { |
3455 | struct btrfs_inode_item *inode_item; | |
3456 | struct btrfs_path *path; | |
5f39d397 | 3457 | struct extent_buffer *leaf; |
39279cc3 CM |
3458 | int ret; |
3459 | ||
3460 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3461 | if (!path) |
3462 | return -ENOMEM; | |
3463 | ||
b9473439 | 3464 | path->leave_spinning = 1; |
16cdcec7 MX |
3465 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3466 | 1); | |
39279cc3 CM |
3467 | if (ret) { |
3468 | if (ret > 0) | |
3469 | ret = -ENOENT; | |
3470 | goto failed; | |
3471 | } | |
3472 | ||
b4ce94de | 3473 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
3474 | leaf = path->nodes[0]; |
3475 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3476 | struct btrfs_inode_item); |
39279cc3 | 3477 | |
e02119d5 | 3478 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3479 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3480 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3481 | ret = 0; |
3482 | failed: | |
39279cc3 CM |
3483 | btrfs_free_path(path); |
3484 | return ret; | |
3485 | } | |
3486 | ||
2115133f CM |
3487 | /* |
3488 | * copy everything in the in-memory inode into the btree. | |
3489 | */ | |
3490 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3491 | struct btrfs_root *root, struct inode *inode) | |
3492 | { | |
3493 | int ret; | |
3494 | ||
3495 | /* | |
3496 | * If the inode is a free space inode, we can deadlock during commit | |
3497 | * if we put it into the delayed code. | |
3498 | * | |
3499 | * The data relocation inode should also be directly updated | |
3500 | * without delay | |
3501 | */ | |
83eea1f1 | 3502 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 3503 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
3504 | btrfs_update_root_times(trans, root); |
3505 | ||
2115133f CM |
3506 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3507 | if (!ret) | |
3508 | btrfs_set_inode_last_trans(trans, inode); | |
3509 | return ret; | |
3510 | } | |
3511 | ||
3512 | return btrfs_update_inode_item(trans, root, inode); | |
3513 | } | |
3514 | ||
be6aef60 JB |
3515 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3516 | struct btrfs_root *root, | |
3517 | struct inode *inode) | |
2115133f CM |
3518 | { |
3519 | int ret; | |
3520 | ||
3521 | ret = btrfs_update_inode(trans, root, inode); | |
3522 | if (ret == -ENOSPC) | |
3523 | return btrfs_update_inode_item(trans, root, inode); | |
3524 | return ret; | |
3525 | } | |
3526 | ||
d352ac68 CM |
3527 | /* |
3528 | * unlink helper that gets used here in inode.c and in the tree logging | |
3529 | * recovery code. It remove a link in a directory with a given name, and | |
3530 | * also drops the back refs in the inode to the directory | |
3531 | */ | |
92986796 AV |
3532 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3533 | struct btrfs_root *root, | |
3534 | struct inode *dir, struct inode *inode, | |
3535 | const char *name, int name_len) | |
39279cc3 CM |
3536 | { |
3537 | struct btrfs_path *path; | |
39279cc3 | 3538 | int ret = 0; |
5f39d397 | 3539 | struct extent_buffer *leaf; |
39279cc3 | 3540 | struct btrfs_dir_item *di; |
5f39d397 | 3541 | struct btrfs_key key; |
aec7477b | 3542 | u64 index; |
33345d01 LZ |
3543 | u64 ino = btrfs_ino(inode); |
3544 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3545 | |
3546 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3547 | if (!path) { |
3548 | ret = -ENOMEM; | |
554233a6 | 3549 | goto out; |
54aa1f4d CM |
3550 | } |
3551 | ||
b9473439 | 3552 | path->leave_spinning = 1; |
33345d01 | 3553 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3554 | name, name_len, -1); |
3555 | if (IS_ERR(di)) { | |
3556 | ret = PTR_ERR(di); | |
3557 | goto err; | |
3558 | } | |
3559 | if (!di) { | |
3560 | ret = -ENOENT; | |
3561 | goto err; | |
3562 | } | |
5f39d397 CM |
3563 | leaf = path->nodes[0]; |
3564 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3565 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3566 | if (ret) |
3567 | goto err; | |
b3b4aa74 | 3568 | btrfs_release_path(path); |
39279cc3 | 3569 | |
33345d01 LZ |
3570 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3571 | dir_ino, &index); | |
aec7477b | 3572 | if (ret) { |
c2cf52eb SK |
3573 | btrfs_info(root->fs_info, |
3574 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
3575 | name_len, name, | |
3576 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 3577 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3578 | goto err; |
3579 | } | |
3580 | ||
16cdcec7 | 3581 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3582 | if (ret) { |
3583 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3584 | goto err; |
79787eaa | 3585 | } |
39279cc3 | 3586 | |
e02119d5 | 3587 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3588 | inode, dir_ino); |
79787eaa JM |
3589 | if (ret != 0 && ret != -ENOENT) { |
3590 | btrfs_abort_transaction(trans, root, ret); | |
3591 | goto err; | |
3592 | } | |
e02119d5 CM |
3593 | |
3594 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3595 | dir, index); | |
6418c961 CM |
3596 | if (ret == -ENOENT) |
3597 | ret = 0; | |
d4e3991b ZB |
3598 | else if (ret) |
3599 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3600 | err: |
3601 | btrfs_free_path(path); | |
e02119d5 CM |
3602 | if (ret) |
3603 | goto out; | |
3604 | ||
3605 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3606 | inode_inc_iversion(inode); |
3607 | inode_inc_iversion(dir); | |
e02119d5 | 3608 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3609 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3610 | out: |
39279cc3 CM |
3611 | return ret; |
3612 | } | |
3613 | ||
92986796 AV |
3614 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3615 | struct btrfs_root *root, | |
3616 | struct inode *dir, struct inode *inode, | |
3617 | const char *name, int name_len) | |
3618 | { | |
3619 | int ret; | |
3620 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3621 | if (!ret) { | |
3622 | btrfs_drop_nlink(inode); | |
3623 | ret = btrfs_update_inode(trans, root, inode); | |
3624 | } | |
3625 | return ret; | |
3626 | } | |
39279cc3 | 3627 | |
a22285a6 YZ |
3628 | /* |
3629 | * helper to start transaction for unlink and rmdir. | |
3630 | * | |
d52be818 JB |
3631 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3632 | * if we cannot make our reservations the normal way try and see if there is | |
3633 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3634 | * allow the unlink to occur. | |
a22285a6 | 3635 | */ |
d52be818 | 3636 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3637 | { |
39279cc3 | 3638 | struct btrfs_trans_handle *trans; |
a22285a6 | 3639 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3640 | int ret; |
3641 | ||
e70bea5f JB |
3642 | /* |
3643 | * 1 for the possible orphan item | |
3644 | * 1 for the dir item | |
3645 | * 1 for the dir index | |
3646 | * 1 for the inode ref | |
e70bea5f JB |
3647 | * 1 for the inode |
3648 | */ | |
6e137ed3 | 3649 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3650 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3651 | return trans; | |
4df27c4d | 3652 | |
d52be818 JB |
3653 | if (PTR_ERR(trans) == -ENOSPC) { |
3654 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 3655 | |
d52be818 JB |
3656 | trans = btrfs_start_transaction(root, 0); |
3657 | if (IS_ERR(trans)) | |
3658 | return trans; | |
3659 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3660 | &root->fs_info->trans_block_rsv, | |
3661 | num_bytes, 5); | |
3662 | if (ret) { | |
3663 | btrfs_end_transaction(trans, root); | |
3664 | return ERR_PTR(ret); | |
a22285a6 | 3665 | } |
5a77d76c | 3666 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3667 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3668 | } |
d52be818 | 3669 | return trans; |
a22285a6 YZ |
3670 | } |
3671 | ||
3672 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3673 | { | |
3674 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3675 | struct btrfs_trans_handle *trans; | |
3676 | struct inode *inode = dentry->d_inode; | |
3677 | int ret; | |
a22285a6 | 3678 | |
d52be818 | 3679 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3680 | if (IS_ERR(trans)) |
3681 | return PTR_ERR(trans); | |
5f39d397 | 3682 | |
12fcfd22 CM |
3683 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3684 | ||
e02119d5 CM |
3685 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3686 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3687 | if (ret) |
3688 | goto out; | |
7b128766 | 3689 | |
a22285a6 | 3690 | if (inode->i_nlink == 0) { |
7b128766 | 3691 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3692 | if (ret) |
3693 | goto out; | |
a22285a6 | 3694 | } |
7b128766 | 3695 | |
b532402e | 3696 | out: |
d52be818 | 3697 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3698 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3699 | return ret; |
3700 | } | |
3701 | ||
4df27c4d YZ |
3702 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3703 | struct btrfs_root *root, | |
3704 | struct inode *dir, u64 objectid, | |
3705 | const char *name, int name_len) | |
3706 | { | |
3707 | struct btrfs_path *path; | |
3708 | struct extent_buffer *leaf; | |
3709 | struct btrfs_dir_item *di; | |
3710 | struct btrfs_key key; | |
3711 | u64 index; | |
3712 | int ret; | |
33345d01 | 3713 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3714 | |
3715 | path = btrfs_alloc_path(); | |
3716 | if (!path) | |
3717 | return -ENOMEM; | |
3718 | ||
33345d01 | 3719 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3720 | name, name_len, -1); |
79787eaa JM |
3721 | if (IS_ERR_OR_NULL(di)) { |
3722 | if (!di) | |
3723 | ret = -ENOENT; | |
3724 | else | |
3725 | ret = PTR_ERR(di); | |
3726 | goto out; | |
3727 | } | |
4df27c4d YZ |
3728 | |
3729 | leaf = path->nodes[0]; | |
3730 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3731 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3732 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3733 | if (ret) { |
3734 | btrfs_abort_transaction(trans, root, ret); | |
3735 | goto out; | |
3736 | } | |
b3b4aa74 | 3737 | btrfs_release_path(path); |
4df27c4d YZ |
3738 | |
3739 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3740 | objectid, root->root_key.objectid, | |
33345d01 | 3741 | dir_ino, &index, name, name_len); |
4df27c4d | 3742 | if (ret < 0) { |
79787eaa JM |
3743 | if (ret != -ENOENT) { |
3744 | btrfs_abort_transaction(trans, root, ret); | |
3745 | goto out; | |
3746 | } | |
33345d01 | 3747 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3748 | name, name_len); |
79787eaa JM |
3749 | if (IS_ERR_OR_NULL(di)) { |
3750 | if (!di) | |
3751 | ret = -ENOENT; | |
3752 | else | |
3753 | ret = PTR_ERR(di); | |
3754 | btrfs_abort_transaction(trans, root, ret); | |
3755 | goto out; | |
3756 | } | |
4df27c4d YZ |
3757 | |
3758 | leaf = path->nodes[0]; | |
3759 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3760 | btrfs_release_path(path); |
4df27c4d YZ |
3761 | index = key.offset; |
3762 | } | |
945d8962 | 3763 | btrfs_release_path(path); |
4df27c4d | 3764 | |
16cdcec7 | 3765 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3766 | if (ret) { |
3767 | btrfs_abort_transaction(trans, root, ret); | |
3768 | goto out; | |
3769 | } | |
4df27c4d YZ |
3770 | |
3771 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3772 | inode_inc_iversion(dir); |
4df27c4d | 3773 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3774 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3775 | if (ret) |
3776 | btrfs_abort_transaction(trans, root, ret); | |
3777 | out: | |
71d7aed0 | 3778 | btrfs_free_path(path); |
79787eaa | 3779 | return ret; |
4df27c4d YZ |
3780 | } |
3781 | ||
39279cc3 CM |
3782 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3783 | { | |
3784 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3785 | int err = 0; |
39279cc3 | 3786 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3787 | struct btrfs_trans_handle *trans; |
39279cc3 | 3788 | |
b3ae244e | 3789 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3790 | return -ENOTEMPTY; |
b3ae244e DS |
3791 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3792 | return -EPERM; | |
134d4512 | 3793 | |
d52be818 | 3794 | trans = __unlink_start_trans(dir); |
a22285a6 | 3795 | if (IS_ERR(trans)) |
5df6a9f6 | 3796 | return PTR_ERR(trans); |
5df6a9f6 | 3797 | |
33345d01 | 3798 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3799 | err = btrfs_unlink_subvol(trans, root, dir, |
3800 | BTRFS_I(inode)->location.objectid, | |
3801 | dentry->d_name.name, | |
3802 | dentry->d_name.len); | |
3803 | goto out; | |
3804 | } | |
3805 | ||
7b128766 JB |
3806 | err = btrfs_orphan_add(trans, inode); |
3807 | if (err) | |
4df27c4d | 3808 | goto out; |
7b128766 | 3809 | |
39279cc3 | 3810 | /* now the directory is empty */ |
e02119d5 CM |
3811 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3812 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3813 | if (!err) |
dbe674a9 | 3814 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3815 | out: |
d52be818 | 3816 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3817 | btrfs_btree_balance_dirty(root); |
3954401f | 3818 | |
39279cc3 CM |
3819 | return err; |
3820 | } | |
3821 | ||
39279cc3 CM |
3822 | /* |
3823 | * this can truncate away extent items, csum items and directory items. | |
3824 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3825 | * any higher than new_size |
39279cc3 CM |
3826 | * |
3827 | * csum items that cross the new i_size are truncated to the new size | |
3828 | * as well. | |
7b128766 JB |
3829 | * |
3830 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3831 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3832 | */ |
8082510e YZ |
3833 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3834 | struct btrfs_root *root, | |
3835 | struct inode *inode, | |
3836 | u64 new_size, u32 min_type) | |
39279cc3 | 3837 | { |
39279cc3 | 3838 | struct btrfs_path *path; |
5f39d397 | 3839 | struct extent_buffer *leaf; |
39279cc3 | 3840 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3841 | struct btrfs_key key; |
3842 | struct btrfs_key found_key; | |
39279cc3 | 3843 | u64 extent_start = 0; |
db94535d | 3844 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3845 | u64 extent_offset = 0; |
39279cc3 | 3846 | u64 item_end = 0; |
8082510e | 3847 | u32 found_type = (u8)-1; |
39279cc3 CM |
3848 | int found_extent; |
3849 | int del_item; | |
85e21bac CM |
3850 | int pending_del_nr = 0; |
3851 | int pending_del_slot = 0; | |
179e29e4 | 3852 | int extent_type = -1; |
8082510e YZ |
3853 | int ret; |
3854 | int err = 0; | |
33345d01 | 3855 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3856 | |
3857 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3858 | |
0eb0e19c MF |
3859 | path = btrfs_alloc_path(); |
3860 | if (!path) | |
3861 | return -ENOMEM; | |
3862 | path->reada = -1; | |
3863 | ||
5dc562c5 JB |
3864 | /* |
3865 | * We want to drop from the next block forward in case this new size is | |
3866 | * not block aligned since we will be keeping the last block of the | |
3867 | * extent just the way it is. | |
3868 | */ | |
0af3d00b | 3869 | if (root->ref_cows || root == root->fs_info->tree_root) |
fda2832f QW |
3870 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
3871 | root->sectorsize), (u64)-1, 0); | |
8082510e | 3872 | |
16cdcec7 MX |
3873 | /* |
3874 | * This function is also used to drop the items in the log tree before | |
3875 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3876 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3877 | * items. | |
3878 | */ | |
3879 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3880 | btrfs_kill_delayed_inode_items(inode); | |
3881 | ||
33345d01 | 3882 | key.objectid = ino; |
39279cc3 | 3883 | key.offset = (u64)-1; |
5f39d397 CM |
3884 | key.type = (u8)-1; |
3885 | ||
85e21bac | 3886 | search_again: |
b9473439 | 3887 | path->leave_spinning = 1; |
85e21bac | 3888 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3889 | if (ret < 0) { |
3890 | err = ret; | |
3891 | goto out; | |
3892 | } | |
d397712b | 3893 | |
85e21bac | 3894 | if (ret > 0) { |
e02119d5 CM |
3895 | /* there are no items in the tree for us to truncate, we're |
3896 | * done | |
3897 | */ | |
8082510e YZ |
3898 | if (path->slots[0] == 0) |
3899 | goto out; | |
85e21bac CM |
3900 | path->slots[0]--; |
3901 | } | |
3902 | ||
d397712b | 3903 | while (1) { |
39279cc3 | 3904 | fi = NULL; |
5f39d397 CM |
3905 | leaf = path->nodes[0]; |
3906 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3907 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3908 | |
33345d01 | 3909 | if (found_key.objectid != ino) |
39279cc3 | 3910 | break; |
5f39d397 | 3911 | |
85e21bac | 3912 | if (found_type < min_type) |
39279cc3 CM |
3913 | break; |
3914 | ||
5f39d397 | 3915 | item_end = found_key.offset; |
39279cc3 | 3916 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3917 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3918 | struct btrfs_file_extent_item); |
179e29e4 CM |
3919 | extent_type = btrfs_file_extent_type(leaf, fi); |
3920 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3921 | item_end += |
db94535d | 3922 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3923 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3924 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3925 | fi); |
39279cc3 | 3926 | } |
008630c1 | 3927 | item_end--; |
39279cc3 | 3928 | } |
8082510e YZ |
3929 | if (found_type > min_type) { |
3930 | del_item = 1; | |
3931 | } else { | |
3932 | if (item_end < new_size) | |
b888db2b | 3933 | break; |
8082510e YZ |
3934 | if (found_key.offset >= new_size) |
3935 | del_item = 1; | |
3936 | else | |
3937 | del_item = 0; | |
39279cc3 | 3938 | } |
39279cc3 | 3939 | found_extent = 0; |
39279cc3 | 3940 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3941 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3942 | goto delete; | |
3943 | ||
3944 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3945 | u64 num_dec; |
db94535d | 3946 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3947 | if (!del_item) { |
db94535d CM |
3948 | u64 orig_num_bytes = |
3949 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
3950 | extent_num_bytes = ALIGN(new_size - |
3951 | found_key.offset, | |
3952 | root->sectorsize); | |
db94535d CM |
3953 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3954 | extent_num_bytes); | |
3955 | num_dec = (orig_num_bytes - | |
9069218d | 3956 | extent_num_bytes); |
e02119d5 | 3957 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3958 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3959 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3960 | } else { |
db94535d CM |
3961 | extent_num_bytes = |
3962 | btrfs_file_extent_disk_num_bytes(leaf, | |
3963 | fi); | |
5d4f98a2 YZ |
3964 | extent_offset = found_key.offset - |
3965 | btrfs_file_extent_offset(leaf, fi); | |
3966 | ||
39279cc3 | 3967 | /* FIXME blocksize != 4096 */ |
9069218d | 3968 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3969 | if (extent_start != 0) { |
3970 | found_extent = 1; | |
e02119d5 | 3971 | if (root->ref_cows) |
a76a3cd4 | 3972 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3973 | } |
39279cc3 | 3974 | } |
9069218d | 3975 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3976 | /* |
3977 | * we can't truncate inline items that have had | |
3978 | * special encodings | |
3979 | */ | |
3980 | if (!del_item && | |
3981 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3982 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3983 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3984 | u32 size = new_size - found_key.offset; |
3985 | ||
3986 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3987 | inode_sub_bytes(inode, item_end + 1 - |
3988 | new_size); | |
e02119d5 CM |
3989 | } |
3990 | size = | |
3991 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 3992 | btrfs_truncate_item(root, path, size, 1); |
e02119d5 | 3993 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3994 | inode_sub_bytes(inode, item_end + 1 - |
3995 | found_key.offset); | |
9069218d | 3996 | } |
39279cc3 | 3997 | } |
179e29e4 | 3998 | delete: |
39279cc3 | 3999 | if (del_item) { |
85e21bac CM |
4000 | if (!pending_del_nr) { |
4001 | /* no pending yet, add ourselves */ | |
4002 | pending_del_slot = path->slots[0]; | |
4003 | pending_del_nr = 1; | |
4004 | } else if (pending_del_nr && | |
4005 | path->slots[0] + 1 == pending_del_slot) { | |
4006 | /* hop on the pending chunk */ | |
4007 | pending_del_nr++; | |
4008 | pending_del_slot = path->slots[0]; | |
4009 | } else { | |
d397712b | 4010 | BUG(); |
85e21bac | 4011 | } |
39279cc3 CM |
4012 | } else { |
4013 | break; | |
4014 | } | |
0af3d00b JB |
4015 | if (found_extent && (root->ref_cows || |
4016 | root == root->fs_info->tree_root)) { | |
b9473439 | 4017 | btrfs_set_path_blocking(path); |
39279cc3 | 4018 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4019 | extent_num_bytes, 0, |
4020 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4021 | ino, extent_offset, 0); |
39279cc3 CM |
4022 | BUG_ON(ret); |
4023 | } | |
85e21bac | 4024 | |
8082510e YZ |
4025 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4026 | break; | |
4027 | ||
4028 | if (path->slots[0] == 0 || | |
4029 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4030 | if (pending_del_nr) { |
4031 | ret = btrfs_del_items(trans, root, path, | |
4032 | pending_del_slot, | |
4033 | pending_del_nr); | |
79787eaa JM |
4034 | if (ret) { |
4035 | btrfs_abort_transaction(trans, | |
4036 | root, ret); | |
4037 | goto error; | |
4038 | } | |
8082510e YZ |
4039 | pending_del_nr = 0; |
4040 | } | |
b3b4aa74 | 4041 | btrfs_release_path(path); |
85e21bac | 4042 | goto search_again; |
8082510e YZ |
4043 | } else { |
4044 | path->slots[0]--; | |
85e21bac | 4045 | } |
39279cc3 | 4046 | } |
8082510e | 4047 | out: |
85e21bac CM |
4048 | if (pending_del_nr) { |
4049 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4050 | pending_del_nr); | |
79787eaa JM |
4051 | if (ret) |
4052 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4053 | } |
79787eaa | 4054 | error: |
39279cc3 | 4055 | btrfs_free_path(path); |
8082510e | 4056 | return err; |
39279cc3 CM |
4057 | } |
4058 | ||
4059 | /* | |
2aaa6655 JB |
4060 | * btrfs_truncate_page - read, zero a chunk and write a page |
4061 | * @inode - inode that we're zeroing | |
4062 | * @from - the offset to start zeroing | |
4063 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4064 | * offset | |
4065 | * @front - zero up to the offset instead of from the offset on | |
4066 | * | |
4067 | * This will find the page for the "from" offset and cow the page and zero the | |
4068 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4069 | */ |
2aaa6655 JB |
4070 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4071 | int front) | |
39279cc3 | 4072 | { |
2aaa6655 | 4073 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4074 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4075 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4076 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4077 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4078 | char *kaddr; |
db94535d | 4079 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4080 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4081 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4082 | struct page *page; | |
3b16a4e3 | 4083 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4084 | int ret = 0; |
a52d9a80 | 4085 | u64 page_start; |
e6dcd2dc | 4086 | u64 page_end; |
39279cc3 | 4087 | |
2aaa6655 JB |
4088 | if ((offset & (blocksize - 1)) == 0 && |
4089 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4090 | goto out; |
0ca1f7ce | 4091 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4092 | if (ret) |
4093 | goto out; | |
39279cc3 | 4094 | |
211c17f5 | 4095 | again: |
3b16a4e3 | 4096 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4097 | if (!page) { |
0ca1f7ce | 4098 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4099 | ret = -ENOMEM; |
39279cc3 | 4100 | goto out; |
5d5e103a | 4101 | } |
e6dcd2dc CM |
4102 | |
4103 | page_start = page_offset(page); | |
4104 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4105 | ||
39279cc3 | 4106 | if (!PageUptodate(page)) { |
9ebefb18 | 4107 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4108 | lock_page(page); |
211c17f5 CM |
4109 | if (page->mapping != mapping) { |
4110 | unlock_page(page); | |
4111 | page_cache_release(page); | |
4112 | goto again; | |
4113 | } | |
39279cc3 CM |
4114 | if (!PageUptodate(page)) { |
4115 | ret = -EIO; | |
89642229 | 4116 | goto out_unlock; |
39279cc3 CM |
4117 | } |
4118 | } | |
211c17f5 | 4119 | wait_on_page_writeback(page); |
e6dcd2dc | 4120 | |
d0082371 | 4121 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4122 | set_page_extent_mapped(page); |
4123 | ||
4124 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4125 | if (ordered) { | |
2ac55d41 JB |
4126 | unlock_extent_cached(io_tree, page_start, page_end, |
4127 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4128 | unlock_page(page); |
4129 | page_cache_release(page); | |
eb84ae03 | 4130 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4131 | btrfs_put_ordered_extent(ordered); |
4132 | goto again; | |
4133 | } | |
4134 | ||
2ac55d41 | 4135 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4136 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4137 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4138 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4139 | |
2ac55d41 JB |
4140 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4141 | &cached_state); | |
9ed74f2d | 4142 | if (ret) { |
2ac55d41 JB |
4143 | unlock_extent_cached(io_tree, page_start, page_end, |
4144 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4145 | goto out_unlock; |
4146 | } | |
4147 | ||
e6dcd2dc | 4148 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4149 | if (!len) |
4150 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4151 | kaddr = kmap(page); |
2aaa6655 JB |
4152 | if (front) |
4153 | memset(kaddr, 0, offset); | |
4154 | else | |
4155 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4156 | flush_dcache_page(page); |
4157 | kunmap(page); | |
4158 | } | |
247e743c | 4159 | ClearPageChecked(page); |
e6dcd2dc | 4160 | set_page_dirty(page); |
2ac55d41 JB |
4161 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4162 | GFP_NOFS); | |
39279cc3 | 4163 | |
89642229 | 4164 | out_unlock: |
5d5e103a | 4165 | if (ret) |
0ca1f7ce | 4166 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4167 | unlock_page(page); |
4168 | page_cache_release(page); | |
4169 | out: | |
4170 | return ret; | |
4171 | } | |
4172 | ||
695a0d0d JB |
4173 | /* |
4174 | * This function puts in dummy file extents for the area we're creating a hole | |
4175 | * for. So if we are truncating this file to a larger size we need to insert | |
4176 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4177 | * the range between oldsize and size | |
4178 | */ | |
a41ad394 | 4179 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4180 | { |
9036c102 YZ |
4181 | struct btrfs_trans_handle *trans; |
4182 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4183 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4184 | struct extent_map *em = NULL; |
2ac55d41 | 4185 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4186 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4187 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4188 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4189 | u64 last_byte; |
4190 | u64 cur_offset; | |
4191 | u64 hole_size; | |
9ed74f2d | 4192 | int err = 0; |
39279cc3 | 4193 | |
a71754fc JB |
4194 | /* |
4195 | * If our size started in the middle of a page we need to zero out the | |
4196 | * rest of the page before we expand the i_size, otherwise we could | |
4197 | * expose stale data. | |
4198 | */ | |
4199 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4200 | if (err) | |
4201 | return err; | |
4202 | ||
9036c102 YZ |
4203 | if (size <= hole_start) |
4204 | return 0; | |
4205 | ||
9036c102 YZ |
4206 | while (1) { |
4207 | struct btrfs_ordered_extent *ordered; | |
4208 | btrfs_wait_ordered_range(inode, hole_start, | |
4209 | block_end - hole_start); | |
2ac55d41 | 4210 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4211 | &cached_state); |
9036c102 YZ |
4212 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
4213 | if (!ordered) | |
4214 | break; | |
2ac55d41 JB |
4215 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4216 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
4217 | btrfs_put_ordered_extent(ordered); |
4218 | } | |
39279cc3 | 4219 | |
9036c102 YZ |
4220 | cur_offset = hole_start; |
4221 | while (1) { | |
4222 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4223 | block_end - cur_offset, 0); | |
79787eaa JM |
4224 | if (IS_ERR(em)) { |
4225 | err = PTR_ERR(em); | |
f2767956 | 4226 | em = NULL; |
79787eaa JM |
4227 | break; |
4228 | } | |
9036c102 | 4229 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4230 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4231 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4232 | struct extent_map *hole_em; |
9036c102 | 4233 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4234 | |
3642320e | 4235 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
4236 | if (IS_ERR(trans)) { |
4237 | err = PTR_ERR(trans); | |
9ed74f2d | 4238 | break; |
a22285a6 | 4239 | } |
8082510e | 4240 | |
5dc562c5 JB |
4241 | err = btrfs_drop_extents(trans, root, inode, |
4242 | cur_offset, | |
2671485d | 4243 | cur_offset + hole_size, 1); |
5b397377 | 4244 | if (err) { |
79787eaa | 4245 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4246 | btrfs_end_transaction(trans, root); |
3893e33b | 4247 | break; |
5b397377 | 4248 | } |
8082510e | 4249 | |
9036c102 | 4250 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 4251 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
4252 | 0, hole_size, 0, hole_size, |
4253 | 0, 0, 0); | |
5b397377 | 4254 | if (err) { |
79787eaa | 4255 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4256 | btrfs_end_transaction(trans, root); |
3893e33b | 4257 | break; |
5b397377 | 4258 | } |
8082510e | 4259 | |
5dc562c5 JB |
4260 | btrfs_drop_extent_cache(inode, cur_offset, |
4261 | cur_offset + hole_size - 1, 0); | |
4262 | hole_em = alloc_extent_map(); | |
4263 | if (!hole_em) { | |
4264 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4265 | &BTRFS_I(inode)->runtime_flags); | |
4266 | goto next; | |
4267 | } | |
4268 | hole_em->start = cur_offset; | |
4269 | hole_em->len = hole_size; | |
4270 | hole_em->orig_start = cur_offset; | |
8082510e | 4271 | |
5dc562c5 JB |
4272 | hole_em->block_start = EXTENT_MAP_HOLE; |
4273 | hole_em->block_len = 0; | |
b4939680 | 4274 | hole_em->orig_block_len = 0; |
cc95bef6 | 4275 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4276 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4277 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
4278 | hole_em->generation = trans->transid; | |
8082510e | 4279 | |
5dc562c5 JB |
4280 | while (1) { |
4281 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4282 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4283 | write_unlock(&em_tree->lock); |
4284 | if (err != -EEXIST) | |
4285 | break; | |
4286 | btrfs_drop_extent_cache(inode, cur_offset, | |
4287 | cur_offset + | |
4288 | hole_size - 1, 0); | |
4289 | } | |
4290 | free_extent_map(hole_em); | |
4291 | next: | |
3642320e | 4292 | btrfs_update_inode(trans, root, inode); |
8082510e | 4293 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
4294 | } |
4295 | free_extent_map(em); | |
a22285a6 | 4296 | em = NULL; |
9036c102 | 4297 | cur_offset = last_byte; |
8082510e | 4298 | if (cur_offset >= block_end) |
9036c102 YZ |
4299 | break; |
4300 | } | |
1832a6d5 | 4301 | |
a22285a6 | 4302 | free_extent_map(em); |
2ac55d41 JB |
4303 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4304 | GFP_NOFS); | |
9036c102 YZ |
4305 | return err; |
4306 | } | |
39279cc3 | 4307 | |
3972f260 | 4308 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4309 | { |
f4a2f4c5 MX |
4310 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4311 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4312 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4313 | loff_t newsize = attr->ia_size; |
4314 | int mask = attr->ia_valid; | |
8082510e YZ |
4315 | int ret; |
4316 | ||
3972f260 ES |
4317 | /* |
4318 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4319 | * special case where we need to update the times despite not having | |
4320 | * these flags set. For all other operations the VFS set these flags | |
4321 | * explicitly if it wants a timestamp update. | |
4322 | */ | |
4323 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
4324 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
4325 | ||
a41ad394 | 4326 | if (newsize > oldsize) { |
a41ad394 JB |
4327 | truncate_pagecache(inode, oldsize, newsize); |
4328 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 4329 | if (ret) |
8082510e | 4330 | return ret; |
8082510e | 4331 | |
f4a2f4c5 MX |
4332 | trans = btrfs_start_transaction(root, 1); |
4333 | if (IS_ERR(trans)) | |
4334 | return PTR_ERR(trans); | |
4335 | ||
4336 | i_size_write(inode, newsize); | |
4337 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4338 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4339 | btrfs_end_transaction(trans, root); |
a41ad394 | 4340 | } else { |
8082510e | 4341 | |
a41ad394 JB |
4342 | /* |
4343 | * We're truncating a file that used to have good data down to | |
4344 | * zero. Make sure it gets into the ordered flush list so that | |
4345 | * any new writes get down to disk quickly. | |
4346 | */ | |
4347 | if (newsize == 0) | |
72ac3c0d JB |
4348 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4349 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4350 | |
f3fe820c JB |
4351 | /* |
4352 | * 1 for the orphan item we're going to add | |
4353 | * 1 for the orphan item deletion. | |
4354 | */ | |
4355 | trans = btrfs_start_transaction(root, 2); | |
4356 | if (IS_ERR(trans)) | |
4357 | return PTR_ERR(trans); | |
4358 | ||
4359 | /* | |
4360 | * We need to do this in case we fail at _any_ point during the | |
4361 | * actual truncate. Once we do the truncate_setsize we could | |
4362 | * invalidate pages which forces any outstanding ordered io to | |
4363 | * be instantly completed which will give us extents that need | |
4364 | * to be truncated. If we fail to get an orphan inode down we | |
4365 | * could have left over extents that were never meant to live, | |
4366 | * so we need to garuntee from this point on that everything | |
4367 | * will be consistent. | |
4368 | */ | |
4369 | ret = btrfs_orphan_add(trans, inode); | |
4370 | btrfs_end_transaction(trans, root); | |
4371 | if (ret) | |
4372 | return ret; | |
4373 | ||
a41ad394 JB |
4374 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4375 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4376 | |
4377 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4378 | btrfs_inode_block_unlocked_dio(inode); | |
4379 | inode_dio_wait(inode); | |
4380 | btrfs_inode_resume_unlocked_dio(inode); | |
4381 | ||
a41ad394 | 4382 | ret = btrfs_truncate(inode); |
f3fe820c JB |
4383 | if (ret && inode->i_nlink) |
4384 | btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
4385 | } |
4386 | ||
a41ad394 | 4387 | return ret; |
8082510e YZ |
4388 | } |
4389 | ||
9036c102 YZ |
4390 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4391 | { | |
4392 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4393 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4394 | int err; |
39279cc3 | 4395 | |
b83cc969 LZ |
4396 | if (btrfs_root_readonly(root)) |
4397 | return -EROFS; | |
4398 | ||
9036c102 YZ |
4399 | err = inode_change_ok(inode, attr); |
4400 | if (err) | |
4401 | return err; | |
2bf5a725 | 4402 | |
5a3f23d5 | 4403 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4404 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4405 | if (err) |
4406 | return err; | |
39279cc3 | 4407 | } |
9036c102 | 4408 | |
1025774c CH |
4409 | if (attr->ia_valid) { |
4410 | setattr_copy(inode, attr); | |
0c4d2d95 | 4411 | inode_inc_iversion(inode); |
22c44fe6 | 4412 | err = btrfs_dirty_inode(inode); |
1025774c | 4413 | |
22c44fe6 | 4414 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
4415 | err = btrfs_acl_chmod(inode); |
4416 | } | |
33268eaf | 4417 | |
39279cc3 CM |
4418 | return err; |
4419 | } | |
61295eb8 | 4420 | |
bd555975 | 4421 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4422 | { |
4423 | struct btrfs_trans_handle *trans; | |
4424 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4425 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4426 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4427 | int ret; |
4428 | ||
1abe9b8a | 4429 | trace_btrfs_inode_evict(inode); |
4430 | ||
39279cc3 | 4431 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 4432 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 4433 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
4434 | goto no_delete; |
4435 | ||
39279cc3 | 4436 | if (is_bad_inode(inode)) { |
7b128766 | 4437 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4438 | goto no_delete; |
4439 | } | |
bd555975 | 4440 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4441 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4442 | |
c71bf099 | 4443 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4444 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4445 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4446 | goto no_delete; |
4447 | } | |
4448 | ||
76dda93c YZ |
4449 | if (inode->i_nlink > 0) { |
4450 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
4451 | goto no_delete; | |
4452 | } | |
4453 | ||
0e8c36a9 MX |
4454 | ret = btrfs_commit_inode_delayed_inode(inode); |
4455 | if (ret) { | |
4456 | btrfs_orphan_del(NULL, inode); | |
4457 | goto no_delete; | |
4458 | } | |
4459 | ||
66d8f3dd | 4460 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4461 | if (!rsv) { |
4462 | btrfs_orphan_del(NULL, inode); | |
4463 | goto no_delete; | |
4464 | } | |
4a338542 | 4465 | rsv->size = min_size; |
ca7e70f5 | 4466 | rsv->failfast = 1; |
726c35fa | 4467 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4468 | |
dbe674a9 | 4469 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4470 | |
4289a667 | 4471 | /* |
8407aa46 MX |
4472 | * This is a bit simpler than btrfs_truncate since we've already |
4473 | * reserved our space for our orphan item in the unlink, so we just | |
4474 | * need to reserve some slack space in case we add bytes and update | |
4475 | * inode item when doing the truncate. | |
4289a667 | 4476 | */ |
8082510e | 4477 | while (1) { |
08e007d2 MX |
4478 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4479 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4480 | |
4481 | /* | |
4482 | * Try and steal from the global reserve since we will | |
4483 | * likely not use this space anyway, we want to try as | |
4484 | * hard as possible to get this to work. | |
4485 | */ | |
4486 | if (ret) | |
4487 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4488 | |
d68fc57b | 4489 | if (ret) { |
c2cf52eb SK |
4490 | btrfs_warn(root->fs_info, |
4491 | "Could not get space for a delete, will truncate on mount %d", | |
4492 | ret); | |
4289a667 JB |
4493 | btrfs_orphan_del(NULL, inode); |
4494 | btrfs_free_block_rsv(root, rsv); | |
4495 | goto no_delete; | |
d68fc57b | 4496 | } |
7b128766 | 4497 | |
0e8c36a9 | 4498 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4499 | if (IS_ERR(trans)) { |
4500 | btrfs_orphan_del(NULL, inode); | |
4501 | btrfs_free_block_rsv(root, rsv); | |
4502 | goto no_delete; | |
d68fc57b | 4503 | } |
7b128766 | 4504 | |
4289a667 JB |
4505 | trans->block_rsv = rsv; |
4506 | ||
d68fc57b | 4507 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4508 | if (ret != -ENOSPC) |
8082510e | 4509 | break; |
85e21bac | 4510 | |
8407aa46 | 4511 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4512 | btrfs_end_transaction(trans, root); |
4513 | trans = NULL; | |
b53d3f5d | 4514 | btrfs_btree_balance_dirty(root); |
8082510e | 4515 | } |
5f39d397 | 4516 | |
4289a667 JB |
4517 | btrfs_free_block_rsv(root, rsv); |
4518 | ||
4ef31a45 JB |
4519 | /* |
4520 | * Errors here aren't a big deal, it just means we leave orphan items | |
4521 | * in the tree. They will be cleaned up on the next mount. | |
4522 | */ | |
8082510e | 4523 | if (ret == 0) { |
4289a667 | 4524 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
4525 | btrfs_orphan_del(trans, inode); |
4526 | } else { | |
4527 | btrfs_orphan_del(NULL, inode); | |
8082510e | 4528 | } |
54aa1f4d | 4529 | |
4289a667 | 4530 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4531 | if (!(root == root->fs_info->tree_root || |
4532 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4533 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4534 | |
54aa1f4d | 4535 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4536 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4537 | no_delete: |
89042e5a | 4538 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4539 | clear_inode(inode); |
8082510e | 4540 | return; |
39279cc3 CM |
4541 | } |
4542 | ||
4543 | /* | |
4544 | * this returns the key found in the dir entry in the location pointer. | |
4545 | * If no dir entries were found, location->objectid is 0. | |
4546 | */ | |
4547 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4548 | struct btrfs_key *location) | |
4549 | { | |
4550 | const char *name = dentry->d_name.name; | |
4551 | int namelen = dentry->d_name.len; | |
4552 | struct btrfs_dir_item *di; | |
4553 | struct btrfs_path *path; | |
4554 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4555 | int ret = 0; |
39279cc3 CM |
4556 | |
4557 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4558 | if (!path) |
4559 | return -ENOMEM; | |
3954401f | 4560 | |
33345d01 | 4561 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4562 | namelen, 0); |
0d9f7f3e Y |
4563 | if (IS_ERR(di)) |
4564 | ret = PTR_ERR(di); | |
d397712b | 4565 | |
c704005d | 4566 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4567 | goto out_err; |
d397712b | 4568 | |
5f39d397 | 4569 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4570 | out: |
39279cc3 CM |
4571 | btrfs_free_path(path); |
4572 | return ret; | |
3954401f CM |
4573 | out_err: |
4574 | location->objectid = 0; | |
4575 | goto out; | |
39279cc3 CM |
4576 | } |
4577 | ||
4578 | /* | |
4579 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4580 | * needs to be changed to reflect the root directory of the tree root. This | |
4581 | * is kind of like crossing a mount point. | |
4582 | */ | |
4583 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4584 | struct inode *dir, |
4585 | struct dentry *dentry, | |
4586 | struct btrfs_key *location, | |
4587 | struct btrfs_root **sub_root) | |
39279cc3 | 4588 | { |
4df27c4d YZ |
4589 | struct btrfs_path *path; |
4590 | struct btrfs_root *new_root; | |
4591 | struct btrfs_root_ref *ref; | |
4592 | struct extent_buffer *leaf; | |
4593 | int ret; | |
4594 | int err = 0; | |
39279cc3 | 4595 | |
4df27c4d YZ |
4596 | path = btrfs_alloc_path(); |
4597 | if (!path) { | |
4598 | err = -ENOMEM; | |
4599 | goto out; | |
4600 | } | |
39279cc3 | 4601 | |
4df27c4d YZ |
4602 | err = -ENOENT; |
4603 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
4604 | BTRFS_I(dir)->root->root_key.objectid, | |
4605 | location->objectid); | |
4606 | if (ret) { | |
4607 | if (ret < 0) | |
4608 | err = ret; | |
4609 | goto out; | |
4610 | } | |
39279cc3 | 4611 | |
4df27c4d YZ |
4612 | leaf = path->nodes[0]; |
4613 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4614 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4615 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4616 | goto out; | |
39279cc3 | 4617 | |
4df27c4d YZ |
4618 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4619 | (unsigned long)(ref + 1), | |
4620 | dentry->d_name.len); | |
4621 | if (ret) | |
4622 | goto out; | |
4623 | ||
b3b4aa74 | 4624 | btrfs_release_path(path); |
4df27c4d YZ |
4625 | |
4626 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4627 | if (IS_ERR(new_root)) { | |
4628 | err = PTR_ERR(new_root); | |
4629 | goto out; | |
4630 | } | |
4631 | ||
4df27c4d YZ |
4632 | *sub_root = new_root; |
4633 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4634 | location->type = BTRFS_INODE_ITEM_KEY; | |
4635 | location->offset = 0; | |
4636 | err = 0; | |
4637 | out: | |
4638 | btrfs_free_path(path); | |
4639 | return err; | |
39279cc3 CM |
4640 | } |
4641 | ||
5d4f98a2 YZ |
4642 | static void inode_tree_add(struct inode *inode) |
4643 | { | |
4644 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4645 | struct btrfs_inode *entry; | |
03e860bd FNP |
4646 | struct rb_node **p; |
4647 | struct rb_node *parent; | |
33345d01 | 4648 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 4649 | |
1d3382cb | 4650 | if (inode_unhashed(inode)) |
76dda93c | 4651 | return; |
e1409cef MX |
4652 | again: |
4653 | parent = NULL; | |
5d4f98a2 | 4654 | spin_lock(&root->inode_lock); |
e1409cef | 4655 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
4656 | while (*p) { |
4657 | parent = *p; | |
4658 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4659 | ||
33345d01 | 4660 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4661 | p = &parent->rb_left; |
33345d01 | 4662 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4663 | p = &parent->rb_right; |
5d4f98a2 YZ |
4664 | else { |
4665 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4666 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4667 | rb_erase(parent, &root->inode_tree); |
4668 | RB_CLEAR_NODE(parent); | |
4669 | spin_unlock(&root->inode_lock); | |
4670 | goto again; | |
5d4f98a2 YZ |
4671 | } |
4672 | } | |
4673 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4674 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4675 | spin_unlock(&root->inode_lock); | |
4676 | } | |
4677 | ||
4678 | static void inode_tree_del(struct inode *inode) | |
4679 | { | |
4680 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4681 | int empty = 0; |
5d4f98a2 | 4682 | |
03e860bd | 4683 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4684 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4685 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4686 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4687 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4688 | } |
03e860bd | 4689 | spin_unlock(&root->inode_lock); |
76dda93c | 4690 | |
0af3d00b JB |
4691 | /* |
4692 | * Free space cache has inodes in the tree root, but the tree root has a | |
4693 | * root_refs of 0, so this could end up dropping the tree root as a | |
4694 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4695 | * make sure we don't drop it. | |
4696 | */ | |
4697 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4698 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4699 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4700 | spin_lock(&root->inode_lock); | |
4701 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4702 | spin_unlock(&root->inode_lock); | |
4703 | if (empty) | |
4704 | btrfs_add_dead_root(root); | |
4705 | } | |
4706 | } | |
4707 | ||
143bede5 | 4708 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4709 | { |
4710 | struct rb_node *node; | |
4711 | struct rb_node *prev; | |
4712 | struct btrfs_inode *entry; | |
4713 | struct inode *inode; | |
4714 | u64 objectid = 0; | |
4715 | ||
4716 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4717 | ||
4718 | spin_lock(&root->inode_lock); | |
4719 | again: | |
4720 | node = root->inode_tree.rb_node; | |
4721 | prev = NULL; | |
4722 | while (node) { | |
4723 | prev = node; | |
4724 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4725 | ||
33345d01 | 4726 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4727 | node = node->rb_left; |
33345d01 | 4728 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4729 | node = node->rb_right; |
4730 | else | |
4731 | break; | |
4732 | } | |
4733 | if (!node) { | |
4734 | while (prev) { | |
4735 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4736 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4737 | node = prev; |
4738 | break; | |
4739 | } | |
4740 | prev = rb_next(prev); | |
4741 | } | |
4742 | } | |
4743 | while (node) { | |
4744 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4745 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4746 | inode = igrab(&entry->vfs_inode); |
4747 | if (inode) { | |
4748 | spin_unlock(&root->inode_lock); | |
4749 | if (atomic_read(&inode->i_count) > 1) | |
4750 | d_prune_aliases(inode); | |
4751 | /* | |
45321ac5 | 4752 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4753 | * the inode cache when its usage count |
4754 | * hits zero. | |
4755 | */ | |
4756 | iput(inode); | |
4757 | cond_resched(); | |
4758 | spin_lock(&root->inode_lock); | |
4759 | goto again; | |
4760 | } | |
4761 | ||
4762 | if (cond_resched_lock(&root->inode_lock)) | |
4763 | goto again; | |
4764 | ||
4765 | node = rb_next(node); | |
4766 | } | |
4767 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4768 | } |
4769 | ||
e02119d5 CM |
4770 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4771 | { | |
4772 | struct btrfs_iget_args *args = p; | |
4773 | inode->i_ino = args->ino; | |
e02119d5 | 4774 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4775 | return 0; |
4776 | } | |
4777 | ||
4778 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4779 | { | |
4780 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4781 | return args->ino == btrfs_ino(inode) && |
d397712b | 4782 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4783 | } |
4784 | ||
5d4f98a2 YZ |
4785 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4786 | u64 objectid, | |
4787 | struct btrfs_root *root) | |
39279cc3 CM |
4788 | { |
4789 | struct inode *inode; | |
4790 | struct btrfs_iget_args args; | |
4791 | args.ino = objectid; | |
4792 | args.root = root; | |
4793 | ||
4794 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4795 | btrfs_init_locked_inode, | |
4796 | (void *)&args); | |
4797 | return inode; | |
4798 | } | |
4799 | ||
1a54ef8c BR |
4800 | /* Get an inode object given its location and corresponding root. |
4801 | * Returns in *is_new if the inode was read from disk | |
4802 | */ | |
4803 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4804 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4805 | { |
4806 | struct inode *inode; | |
4807 | ||
4808 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4809 | if (!inode) | |
5d4f98a2 | 4810 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4811 | |
4812 | if (inode->i_state & I_NEW) { | |
4813 | BTRFS_I(inode)->root = root; | |
4814 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4815 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4816 | if (!is_bad_inode(inode)) { |
4817 | inode_tree_add(inode); | |
4818 | unlock_new_inode(inode); | |
4819 | if (new) | |
4820 | *new = 1; | |
4821 | } else { | |
e0b6d65b ST |
4822 | unlock_new_inode(inode); |
4823 | iput(inode); | |
4824 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4825 | } |
4826 | } | |
4827 | ||
1a54ef8c BR |
4828 | return inode; |
4829 | } | |
4830 | ||
4df27c4d YZ |
4831 | static struct inode *new_simple_dir(struct super_block *s, |
4832 | struct btrfs_key *key, | |
4833 | struct btrfs_root *root) | |
4834 | { | |
4835 | struct inode *inode = new_inode(s); | |
4836 | ||
4837 | if (!inode) | |
4838 | return ERR_PTR(-ENOMEM); | |
4839 | ||
4df27c4d YZ |
4840 | BTRFS_I(inode)->root = root; |
4841 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4842 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4843 | |
4844 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4845 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4846 | inode->i_fop = &simple_dir_operations; |
4847 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4848 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4849 | ||
4850 | return inode; | |
4851 | } | |
4852 | ||
3de4586c | 4853 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4854 | { |
d397712b | 4855 | struct inode *inode; |
4df27c4d | 4856 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4857 | struct btrfs_root *sub_root = root; |
4858 | struct btrfs_key location; | |
76dda93c | 4859 | int index; |
b4aff1f8 | 4860 | int ret = 0; |
39279cc3 CM |
4861 | |
4862 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4863 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4864 | |
39e3c955 | 4865 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
4866 | if (ret < 0) |
4867 | return ERR_PTR(ret); | |
5f39d397 | 4868 | |
4df27c4d YZ |
4869 | if (location.objectid == 0) |
4870 | return NULL; | |
4871 | ||
4872 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4873 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4874 | return inode; |
4875 | } | |
4876 | ||
4877 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4878 | ||
76dda93c | 4879 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4880 | ret = fixup_tree_root_location(root, dir, dentry, |
4881 | &location, &sub_root); | |
4882 | if (ret < 0) { | |
4883 | if (ret != -ENOENT) | |
4884 | inode = ERR_PTR(ret); | |
4885 | else | |
4886 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4887 | } else { | |
73f73415 | 4888 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4889 | } |
76dda93c YZ |
4890 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4891 | ||
34d19bad | 4892 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4893 | down_read(&root->fs_info->cleanup_work_sem); |
4894 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4895 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4896 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
4897 | if (ret) { |
4898 | iput(inode); | |
66b4ffd1 | 4899 | inode = ERR_PTR(ret); |
01cd3367 | 4900 | } |
c71bf099 YZ |
4901 | } |
4902 | ||
3de4586c CM |
4903 | return inode; |
4904 | } | |
4905 | ||
fe15ce44 | 4906 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4907 | { |
4908 | struct btrfs_root *root; | |
848cce0d | 4909 | struct inode *inode = dentry->d_inode; |
76dda93c | 4910 | |
848cce0d LZ |
4911 | if (!inode && !IS_ROOT(dentry)) |
4912 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4913 | |
848cce0d LZ |
4914 | if (inode) { |
4915 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4916 | if (btrfs_root_refs(&root->root_item) == 0) |
4917 | return 1; | |
848cce0d LZ |
4918 | |
4919 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4920 | return 1; | |
efefb143 | 4921 | } |
76dda93c YZ |
4922 | return 0; |
4923 | } | |
4924 | ||
b4aff1f8 JB |
4925 | static void btrfs_dentry_release(struct dentry *dentry) |
4926 | { | |
4927 | if (dentry->d_fsdata) | |
4928 | kfree(dentry->d_fsdata); | |
4929 | } | |
4930 | ||
3de4586c | 4931 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4932 | unsigned int flags) |
3de4586c | 4933 | { |
a66e7cc6 JB |
4934 | struct dentry *ret; |
4935 | ||
4936 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
a66e7cc6 | 4937 | return ret; |
39279cc3 CM |
4938 | } |
4939 | ||
16cdcec7 | 4940 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4941 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4942 | }; | |
4943 | ||
9cdda8d3 | 4944 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 4945 | { |
9cdda8d3 | 4946 | struct inode *inode = file_inode(file); |
39279cc3 CM |
4947 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4948 | struct btrfs_item *item; | |
4949 | struct btrfs_dir_item *di; | |
4950 | struct btrfs_key key; | |
5f39d397 | 4951 | struct btrfs_key found_key; |
39279cc3 | 4952 | struct btrfs_path *path; |
16cdcec7 MX |
4953 | struct list_head ins_list; |
4954 | struct list_head del_list; | |
39279cc3 | 4955 | int ret; |
5f39d397 | 4956 | struct extent_buffer *leaf; |
39279cc3 | 4957 | int slot; |
39279cc3 CM |
4958 | unsigned char d_type; |
4959 | int over = 0; | |
4960 | u32 di_cur; | |
4961 | u32 di_total; | |
4962 | u32 di_len; | |
4963 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4964 | char tmp_name[32]; |
4965 | char *name_ptr; | |
4966 | int name_len; | |
9cdda8d3 | 4967 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
4968 | |
4969 | /* FIXME, use a real flag for deciding about the key type */ | |
4970 | if (root->fs_info->tree_root == root) | |
4971 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4972 | |
9cdda8d3 AV |
4973 | if (!dir_emit_dots(file, ctx)) |
4974 | return 0; | |
4975 | ||
49593bfa | 4976 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4977 | if (!path) |
4978 | return -ENOMEM; | |
ff5714cc | 4979 | |
026fd317 | 4980 | path->reada = 1; |
49593bfa | 4981 | |
16cdcec7 MX |
4982 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4983 | INIT_LIST_HEAD(&ins_list); | |
4984 | INIT_LIST_HEAD(&del_list); | |
4985 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4986 | } | |
4987 | ||
39279cc3 | 4988 | btrfs_set_key_type(&key, key_type); |
9cdda8d3 | 4989 | key.offset = ctx->pos; |
33345d01 | 4990 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4991 | |
39279cc3 CM |
4992 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4993 | if (ret < 0) | |
4994 | goto err; | |
49593bfa DW |
4995 | |
4996 | while (1) { | |
5f39d397 | 4997 | leaf = path->nodes[0]; |
39279cc3 | 4998 | slot = path->slots[0]; |
b9e03af0 LZ |
4999 | if (slot >= btrfs_header_nritems(leaf)) { |
5000 | ret = btrfs_next_leaf(root, path); | |
5001 | if (ret < 0) | |
5002 | goto err; | |
5003 | else if (ret > 0) | |
5004 | break; | |
5005 | continue; | |
39279cc3 | 5006 | } |
3de4586c | 5007 | |
5f39d397 CM |
5008 | item = btrfs_item_nr(leaf, slot); |
5009 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5010 | ||
5011 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5012 | break; |
5f39d397 | 5013 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 5014 | break; |
9cdda8d3 | 5015 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5016 | goto next; |
16cdcec7 MX |
5017 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5018 | btrfs_should_delete_dir_index(&del_list, | |
5019 | found_key.offset)) | |
5020 | goto next; | |
5f39d397 | 5021 | |
9cdda8d3 | 5022 | ctx->pos = found_key.offset; |
16cdcec7 | 5023 | is_curr = 1; |
49593bfa | 5024 | |
39279cc3 CM |
5025 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5026 | di_cur = 0; | |
5f39d397 | 5027 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5028 | |
5029 | while (di_cur < di_total) { | |
5f39d397 CM |
5030 | struct btrfs_key location; |
5031 | ||
22a94d44 JB |
5032 | if (verify_dir_item(root, leaf, di)) |
5033 | break; | |
5034 | ||
5f39d397 | 5035 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5036 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5037 | name_ptr = tmp_name; |
5038 | } else { | |
5039 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5040 | if (!name_ptr) { |
5041 | ret = -ENOMEM; | |
5042 | goto err; | |
5043 | } | |
5f39d397 CM |
5044 | } |
5045 | read_extent_buffer(leaf, name_ptr, | |
5046 | (unsigned long)(di + 1), name_len); | |
5047 | ||
5048 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5049 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5050 | |
fede766f | 5051 | |
3de4586c | 5052 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5053 | * skip it. |
5054 | * | |
5055 | * In contrast to old kernels, we insert the snapshot's | |
5056 | * dir item and dir index after it has been created, so | |
5057 | * we won't find a reference to our own snapshot. We | |
5058 | * still keep the following code for backward | |
5059 | * compatibility. | |
3de4586c CM |
5060 | */ |
5061 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5062 | location.objectid == root->root_key.objectid) { | |
5063 | over = 0; | |
5064 | goto skip; | |
5065 | } | |
9cdda8d3 AV |
5066 | over = !dir_emit(ctx, name_ptr, name_len, |
5067 | location.objectid, d_type); | |
5f39d397 | 5068 | |
3de4586c | 5069 | skip: |
5f39d397 CM |
5070 | if (name_ptr != tmp_name) |
5071 | kfree(name_ptr); | |
5072 | ||
39279cc3 CM |
5073 | if (over) |
5074 | goto nopos; | |
5103e947 | 5075 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5076 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5077 | di_cur += di_len; |
5078 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5079 | } | |
b9e03af0 LZ |
5080 | next: |
5081 | path->slots[0]++; | |
39279cc3 | 5082 | } |
49593bfa | 5083 | |
16cdcec7 MX |
5084 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5085 | if (is_curr) | |
9cdda8d3 AV |
5086 | ctx->pos++; |
5087 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5088 | if (ret) |
5089 | goto nopos; | |
5090 | } | |
5091 | ||
49593bfa | 5092 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5093 | ctx->pos++; |
5094 | ||
5095 | /* | |
5096 | * Stop new entries from being returned after we return the last | |
5097 | * entry. | |
5098 | * | |
5099 | * New directory entries are assigned a strictly increasing | |
5100 | * offset. This means that new entries created during readdir | |
5101 | * are *guaranteed* to be seen in the future by that readdir. | |
5102 | * This has broken buggy programs which operate on names as | |
5103 | * they're returned by readdir. Until we re-use freed offsets | |
5104 | * we have this hack to stop new entries from being returned | |
5105 | * under the assumption that they'll never reach this huge | |
5106 | * offset. | |
5107 | * | |
5108 | * This is being careful not to overflow 32bit loff_t unless the | |
5109 | * last entry requires it because doing so has broken 32bit apps | |
5110 | * in the past. | |
5111 | */ | |
5112 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5113 | if (ctx->pos >= INT_MAX) | |
5114 | ctx->pos = LLONG_MAX; | |
5115 | else | |
5116 | ctx->pos = INT_MAX; | |
5117 | } | |
39279cc3 CM |
5118 | nopos: |
5119 | ret = 0; | |
5120 | err: | |
16cdcec7 MX |
5121 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5122 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5123 | btrfs_free_path(path); |
39279cc3 CM |
5124 | return ret; |
5125 | } | |
5126 | ||
a9185b41 | 5127 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5128 | { |
5129 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5130 | struct btrfs_trans_handle *trans; | |
5131 | int ret = 0; | |
0af3d00b | 5132 | bool nolock = false; |
39279cc3 | 5133 | |
72ac3c0d | 5134 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5135 | return 0; |
5136 | ||
83eea1f1 | 5137 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5138 | nolock = true; |
0af3d00b | 5139 | |
a9185b41 | 5140 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5141 | if (nolock) |
7a7eaa40 | 5142 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5143 | else |
7a7eaa40 | 5144 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5145 | if (IS_ERR(trans)) |
5146 | return PTR_ERR(trans); | |
a698d075 | 5147 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5148 | } |
5149 | return ret; | |
5150 | } | |
5151 | ||
5152 | /* | |
54aa1f4d | 5153 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5154 | * inode changes. But, it is most likely to find the inode in cache. |
5155 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5156 | * to keep or drop this code. | |
5157 | */ | |
48a3b636 | 5158 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5159 | { |
5160 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5161 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5162 | int ret; |
5163 | ||
72ac3c0d | 5164 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5165 | return 0; |
39279cc3 | 5166 | |
7a7eaa40 | 5167 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5168 | if (IS_ERR(trans)) |
5169 | return PTR_ERR(trans); | |
8929ecfa YZ |
5170 | |
5171 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5172 | if (ret && ret == -ENOSPC) { |
5173 | /* whoops, lets try again with the full transaction */ | |
5174 | btrfs_end_transaction(trans, root); | |
5175 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5176 | if (IS_ERR(trans)) |
5177 | return PTR_ERR(trans); | |
8929ecfa | 5178 | |
94b60442 | 5179 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5180 | } |
39279cc3 | 5181 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5182 | if (BTRFS_I(inode)->delayed_node) |
5183 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5184 | |
5185 | return ret; | |
5186 | } | |
5187 | ||
5188 | /* | |
5189 | * This is a copy of file_update_time. We need this so we can return error on | |
5190 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5191 | */ | |
e41f941a JB |
5192 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5193 | int flags) | |
22c44fe6 | 5194 | { |
2bc55652 AB |
5195 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5196 | ||
5197 | if (btrfs_root_readonly(root)) | |
5198 | return -EROFS; | |
5199 | ||
e41f941a | 5200 | if (flags & S_VERSION) |
22c44fe6 | 5201 | inode_inc_iversion(inode); |
e41f941a JB |
5202 | if (flags & S_CTIME) |
5203 | inode->i_ctime = *now; | |
5204 | if (flags & S_MTIME) | |
5205 | inode->i_mtime = *now; | |
5206 | if (flags & S_ATIME) | |
5207 | inode->i_atime = *now; | |
5208 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5209 | } |
5210 | ||
d352ac68 CM |
5211 | /* |
5212 | * find the highest existing sequence number in a directory | |
5213 | * and then set the in-memory index_cnt variable to reflect | |
5214 | * free sequence numbers | |
5215 | */ | |
aec7477b JB |
5216 | static int btrfs_set_inode_index_count(struct inode *inode) |
5217 | { | |
5218 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5219 | struct btrfs_key key, found_key; | |
5220 | struct btrfs_path *path; | |
5221 | struct extent_buffer *leaf; | |
5222 | int ret; | |
5223 | ||
33345d01 | 5224 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
5225 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
5226 | key.offset = (u64)-1; | |
5227 | ||
5228 | path = btrfs_alloc_path(); | |
5229 | if (!path) | |
5230 | return -ENOMEM; | |
5231 | ||
5232 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5233 | if (ret < 0) | |
5234 | goto out; | |
5235 | /* FIXME: we should be able to handle this */ | |
5236 | if (ret == 0) | |
5237 | goto out; | |
5238 | ret = 0; | |
5239 | ||
5240 | /* | |
5241 | * MAGIC NUMBER EXPLANATION: | |
5242 | * since we search a directory based on f_pos we have to start at 2 | |
5243 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5244 | * else has to start at 2 | |
5245 | */ | |
5246 | if (path->slots[0] == 0) { | |
5247 | BTRFS_I(inode)->index_cnt = 2; | |
5248 | goto out; | |
5249 | } | |
5250 | ||
5251 | path->slots[0]--; | |
5252 | ||
5253 | leaf = path->nodes[0]; | |
5254 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5255 | ||
33345d01 | 5256 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
5257 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
5258 | BTRFS_I(inode)->index_cnt = 2; | |
5259 | goto out; | |
5260 | } | |
5261 | ||
5262 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5263 | out: | |
5264 | btrfs_free_path(path); | |
5265 | return ret; | |
5266 | } | |
5267 | ||
d352ac68 CM |
5268 | /* |
5269 | * helper to find a free sequence number in a given directory. This current | |
5270 | * code is very simple, later versions will do smarter things in the btree | |
5271 | */ | |
3de4586c | 5272 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5273 | { |
5274 | int ret = 0; | |
5275 | ||
5276 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5277 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5278 | if (ret) { | |
5279 | ret = btrfs_set_inode_index_count(dir); | |
5280 | if (ret) | |
5281 | return ret; | |
5282 | } | |
aec7477b JB |
5283 | } |
5284 | ||
00e4e6b3 | 5285 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5286 | BTRFS_I(dir)->index_cnt++; |
5287 | ||
5288 | return ret; | |
5289 | } | |
5290 | ||
39279cc3 CM |
5291 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5292 | struct btrfs_root *root, | |
aec7477b | 5293 | struct inode *dir, |
9c58309d | 5294 | const char *name, int name_len, |
175a4eb7 AV |
5295 | u64 ref_objectid, u64 objectid, |
5296 | umode_t mode, u64 *index) | |
39279cc3 CM |
5297 | { |
5298 | struct inode *inode; | |
5f39d397 | 5299 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5300 | struct btrfs_key *location; |
5f39d397 | 5301 | struct btrfs_path *path; |
9c58309d CM |
5302 | struct btrfs_inode_ref *ref; |
5303 | struct btrfs_key key[2]; | |
5304 | u32 sizes[2]; | |
5305 | unsigned long ptr; | |
39279cc3 CM |
5306 | int ret; |
5307 | int owner; | |
5308 | ||
5f39d397 | 5309 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5310 | if (!path) |
5311 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5312 | |
39279cc3 | 5313 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5314 | if (!inode) { |
5315 | btrfs_free_path(path); | |
39279cc3 | 5316 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5317 | } |
39279cc3 | 5318 | |
581bb050 LZ |
5319 | /* |
5320 | * we have to initialize this early, so we can reclaim the inode | |
5321 | * number if we fail afterwards in this function. | |
5322 | */ | |
5323 | inode->i_ino = objectid; | |
5324 | ||
aec7477b | 5325 | if (dir) { |
1abe9b8a | 5326 | trace_btrfs_inode_request(dir); |
5327 | ||
3de4586c | 5328 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5329 | if (ret) { |
8fb27640 | 5330 | btrfs_free_path(path); |
09771430 | 5331 | iput(inode); |
aec7477b | 5332 | return ERR_PTR(ret); |
09771430 | 5333 | } |
aec7477b JB |
5334 | } |
5335 | /* | |
5336 | * index_cnt is ignored for everything but a dir, | |
5337 | * btrfs_get_inode_index_count has an explanation for the magic | |
5338 | * number | |
5339 | */ | |
5340 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 5341 | BTRFS_I(inode)->root = root; |
e02119d5 | 5342 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5343 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5344 | |
5dc562c5 JB |
5345 | /* |
5346 | * We could have gotten an inode number from somebody who was fsynced | |
5347 | * and then removed in this same transaction, so let's just set full | |
5348 | * sync since it will be a full sync anyway and this will blow away the | |
5349 | * old info in the log. | |
5350 | */ | |
5351 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5352 | ||
569254b0 | 5353 | if (S_ISDIR(mode)) |
39279cc3 CM |
5354 | owner = 0; |
5355 | else | |
5356 | owner = 1; | |
9c58309d CM |
5357 | |
5358 | key[0].objectid = objectid; | |
5359 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
5360 | key[0].offset = 0; | |
5361 | ||
f186373f MF |
5362 | /* |
5363 | * Start new inodes with an inode_ref. This is slightly more | |
5364 | * efficient for small numbers of hard links since they will | |
5365 | * be packed into one item. Extended refs will kick in if we | |
5366 | * add more hard links than can fit in the ref item. | |
5367 | */ | |
9c58309d CM |
5368 | key[1].objectid = objectid; |
5369 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
5370 | key[1].offset = ref_objectid; | |
5371 | ||
5372 | sizes[0] = sizeof(struct btrfs_inode_item); | |
5373 | sizes[1] = name_len + sizeof(*ref); | |
5374 | ||
b9473439 | 5375 | path->leave_spinning = 1; |
9c58309d CM |
5376 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
5377 | if (ret != 0) | |
5f39d397 CM |
5378 | goto fail; |
5379 | ||
ecc11fab | 5380 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5381 | inode_set_bytes(inode, 0); |
39279cc3 | 5382 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5383 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5384 | struct btrfs_inode_item); | |
293f7e07 LZ |
5385 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5386 | sizeof(*inode_item)); | |
e02119d5 | 5387 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
5388 | |
5389 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5390 | struct btrfs_inode_ref); | |
5391 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 5392 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
5393 | ptr = (unsigned long)(ref + 1); |
5394 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5395 | ||
5f39d397 CM |
5396 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5397 | btrfs_free_path(path); | |
5398 | ||
39279cc3 CM |
5399 | location = &BTRFS_I(inode)->location; |
5400 | location->objectid = objectid; | |
39279cc3 CM |
5401 | location->offset = 0; |
5402 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
5403 | ||
6cbff00f CH |
5404 | btrfs_inherit_iflags(inode, dir); |
5405 | ||
569254b0 | 5406 | if (S_ISREG(mode)) { |
94272164 CM |
5407 | if (btrfs_test_opt(root, NODATASUM)) |
5408 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5409 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5410 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5411 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5412 | } |
5413 | ||
39279cc3 | 5414 | insert_inode_hash(inode); |
5d4f98a2 | 5415 | inode_tree_add(inode); |
1abe9b8a | 5416 | |
5417 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5418 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5419 | |
8ea05e3a AB |
5420 | btrfs_update_root_times(trans, root); |
5421 | ||
39279cc3 | 5422 | return inode; |
5f39d397 | 5423 | fail: |
aec7477b JB |
5424 | if (dir) |
5425 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 5426 | btrfs_free_path(path); |
09771430 | 5427 | iput(inode); |
5f39d397 | 5428 | return ERR_PTR(ret); |
39279cc3 CM |
5429 | } |
5430 | ||
5431 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5432 | { | |
5433 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5434 | } | |
5435 | ||
d352ac68 CM |
5436 | /* |
5437 | * utility function to add 'inode' into 'parent_inode' with | |
5438 | * a give name and a given sequence number. | |
5439 | * if 'add_backref' is true, also insert a backref from the | |
5440 | * inode to the parent directory. | |
5441 | */ | |
e02119d5 CM |
5442 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5443 | struct inode *parent_inode, struct inode *inode, | |
5444 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5445 | { |
4df27c4d | 5446 | int ret = 0; |
39279cc3 | 5447 | struct btrfs_key key; |
e02119d5 | 5448 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5449 | u64 ino = btrfs_ino(inode); |
5450 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5451 | |
33345d01 | 5452 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5453 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5454 | } else { | |
33345d01 | 5455 | key.objectid = ino; |
4df27c4d YZ |
5456 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
5457 | key.offset = 0; | |
5458 | } | |
5459 | ||
33345d01 | 5460 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5461 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5462 | key.objectid, root->root_key.objectid, | |
33345d01 | 5463 | parent_ino, index, name, name_len); |
4df27c4d | 5464 | } else if (add_backref) { |
33345d01 LZ |
5465 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5466 | parent_ino, index); | |
4df27c4d | 5467 | } |
39279cc3 | 5468 | |
79787eaa JM |
5469 | /* Nothing to clean up yet */ |
5470 | if (ret) | |
5471 | return ret; | |
4df27c4d | 5472 | |
79787eaa JM |
5473 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5474 | parent_inode, &key, | |
5475 | btrfs_inode_type(inode), index); | |
9c52057c | 5476 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5477 | goto fail_dir_item; |
5478 | else if (ret) { | |
5479 | btrfs_abort_transaction(trans, root, ret); | |
5480 | return ret; | |
39279cc3 | 5481 | } |
79787eaa JM |
5482 | |
5483 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5484 | name_len * 2); | |
0c4d2d95 | 5485 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5486 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5487 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5488 | if (ret) | |
5489 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5490 | return ret; |
fe66a05a CM |
5491 | |
5492 | fail_dir_item: | |
5493 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5494 | u64 local_index; | |
5495 | int err; | |
5496 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5497 | key.objectid, root->root_key.objectid, | |
5498 | parent_ino, &local_index, name, name_len); | |
5499 | ||
5500 | } else if (add_backref) { | |
5501 | u64 local_index; | |
5502 | int err; | |
5503 | ||
5504 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5505 | ino, parent_ino, &local_index); | |
5506 | } | |
5507 | return ret; | |
39279cc3 CM |
5508 | } |
5509 | ||
5510 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5511 | struct inode *dir, struct dentry *dentry, |
5512 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5513 | { |
a1b075d2 JB |
5514 | int err = btrfs_add_link(trans, dir, inode, |
5515 | dentry->d_name.name, dentry->d_name.len, | |
5516 | backref, index); | |
39279cc3 CM |
5517 | if (err > 0) |
5518 | err = -EEXIST; | |
5519 | return err; | |
5520 | } | |
5521 | ||
618e21d5 | 5522 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5523 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5524 | { |
5525 | struct btrfs_trans_handle *trans; | |
5526 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5527 | struct inode *inode = NULL; |
618e21d5 JB |
5528 | int err; |
5529 | int drop_inode = 0; | |
5530 | u64 objectid; | |
00e4e6b3 | 5531 | u64 index = 0; |
618e21d5 JB |
5532 | |
5533 | if (!new_valid_dev(rdev)) | |
5534 | return -EINVAL; | |
5535 | ||
9ed74f2d JB |
5536 | /* |
5537 | * 2 for inode item and ref | |
5538 | * 2 for dir items | |
5539 | * 1 for xattr if selinux is on | |
5540 | */ | |
a22285a6 YZ |
5541 | trans = btrfs_start_transaction(root, 5); |
5542 | if (IS_ERR(trans)) | |
5543 | return PTR_ERR(trans); | |
1832a6d5 | 5544 | |
581bb050 LZ |
5545 | err = btrfs_find_free_ino(root, &objectid); |
5546 | if (err) | |
5547 | goto out_unlock; | |
5548 | ||
aec7477b | 5549 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5550 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5551 | mode, &index); |
7cf96da3 TI |
5552 | if (IS_ERR(inode)) { |
5553 | err = PTR_ERR(inode); | |
618e21d5 | 5554 | goto out_unlock; |
7cf96da3 | 5555 | } |
618e21d5 | 5556 | |
2a7dba39 | 5557 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5558 | if (err) { |
5559 | drop_inode = 1; | |
5560 | goto out_unlock; | |
5561 | } | |
5562 | ||
ad19db71 CS |
5563 | /* |
5564 | * If the active LSM wants to access the inode during | |
5565 | * d_instantiate it needs these. Smack checks to see | |
5566 | * if the filesystem supports xattrs by looking at the | |
5567 | * ops vector. | |
5568 | */ | |
5569 | ||
5570 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5571 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5572 | if (err) |
5573 | drop_inode = 1; | |
5574 | else { | |
618e21d5 | 5575 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5576 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5577 | d_instantiate(dentry, inode); |
618e21d5 | 5578 | } |
618e21d5 | 5579 | out_unlock: |
7ad85bb7 | 5580 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5581 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5582 | if (drop_inode) { |
5583 | inode_dec_link_count(inode); | |
5584 | iput(inode); | |
5585 | } | |
618e21d5 JB |
5586 | return err; |
5587 | } | |
5588 | ||
39279cc3 | 5589 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5590 | umode_t mode, bool excl) |
39279cc3 CM |
5591 | { |
5592 | struct btrfs_trans_handle *trans; | |
5593 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5594 | struct inode *inode = NULL; |
43baa579 | 5595 | int drop_inode_on_err = 0; |
a22285a6 | 5596 | int err; |
39279cc3 | 5597 | u64 objectid; |
00e4e6b3 | 5598 | u64 index = 0; |
39279cc3 | 5599 | |
9ed74f2d JB |
5600 | /* |
5601 | * 2 for inode item and ref | |
5602 | * 2 for dir items | |
5603 | * 1 for xattr if selinux is on | |
5604 | */ | |
a22285a6 YZ |
5605 | trans = btrfs_start_transaction(root, 5); |
5606 | if (IS_ERR(trans)) | |
5607 | return PTR_ERR(trans); | |
9ed74f2d | 5608 | |
581bb050 LZ |
5609 | err = btrfs_find_free_ino(root, &objectid); |
5610 | if (err) | |
5611 | goto out_unlock; | |
5612 | ||
aec7477b | 5613 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5614 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5615 | mode, &index); |
7cf96da3 TI |
5616 | if (IS_ERR(inode)) { |
5617 | err = PTR_ERR(inode); | |
39279cc3 | 5618 | goto out_unlock; |
7cf96da3 | 5619 | } |
43baa579 | 5620 | drop_inode_on_err = 1; |
39279cc3 | 5621 | |
2a7dba39 | 5622 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5623 | if (err) |
33268eaf | 5624 | goto out_unlock; |
33268eaf | 5625 | |
9185aa58 FB |
5626 | err = btrfs_update_inode(trans, root, inode); |
5627 | if (err) | |
5628 | goto out_unlock; | |
5629 | ||
ad19db71 CS |
5630 | /* |
5631 | * If the active LSM wants to access the inode during | |
5632 | * d_instantiate it needs these. Smack checks to see | |
5633 | * if the filesystem supports xattrs by looking at the | |
5634 | * ops vector. | |
5635 | */ | |
5636 | inode->i_fop = &btrfs_file_operations; | |
5637 | inode->i_op = &btrfs_file_inode_operations; | |
5638 | ||
a1b075d2 | 5639 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5640 | if (err) |
43baa579 FB |
5641 | goto out_unlock; |
5642 | ||
5643 | inode->i_mapping->a_ops = &btrfs_aops; | |
5644 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5645 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5646 | d_instantiate(dentry, inode); | |
5647 | ||
39279cc3 | 5648 | out_unlock: |
7ad85bb7 | 5649 | btrfs_end_transaction(trans, root); |
43baa579 | 5650 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5651 | inode_dec_link_count(inode); |
5652 | iput(inode); | |
5653 | } | |
b53d3f5d | 5654 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5655 | return err; |
5656 | } | |
5657 | ||
5658 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5659 | struct dentry *dentry) | |
5660 | { | |
5661 | struct btrfs_trans_handle *trans; | |
5662 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5663 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5664 | u64 index; |
39279cc3 CM |
5665 | int err; |
5666 | int drop_inode = 0; | |
5667 | ||
4a8be425 TH |
5668 | /* do not allow sys_link's with other subvols of the same device */ |
5669 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5670 | return -EXDEV; |
4a8be425 | 5671 | |
f186373f | 5672 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5673 | return -EMLINK; |
4a8be425 | 5674 | |
3de4586c | 5675 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5676 | if (err) |
5677 | goto fail; | |
5678 | ||
a22285a6 | 5679 | /* |
7e6b6465 | 5680 | * 2 items for inode and inode ref |
a22285a6 | 5681 | * 2 items for dir items |
7e6b6465 | 5682 | * 1 item for parent inode |
a22285a6 | 5683 | */ |
7e6b6465 | 5684 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5685 | if (IS_ERR(trans)) { |
5686 | err = PTR_ERR(trans); | |
5687 | goto fail; | |
5688 | } | |
5f39d397 | 5689 | |
3153495d | 5690 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5691 | inode_inc_iversion(inode); |
3153495d | 5692 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5693 | ihold(inode); |
e9976151 | 5694 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5695 | |
a1b075d2 | 5696 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5697 | |
a5719521 | 5698 | if (err) { |
54aa1f4d | 5699 | drop_inode = 1; |
a5719521 | 5700 | } else { |
10d9f309 | 5701 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5702 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5703 | if (err) |
5704 | goto fail; | |
08c422c2 | 5705 | d_instantiate(dentry, inode); |
6a912213 | 5706 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5707 | } |
39279cc3 | 5708 | |
7ad85bb7 | 5709 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5710 | fail: |
39279cc3 CM |
5711 | if (drop_inode) { |
5712 | inode_dec_link_count(inode); | |
5713 | iput(inode); | |
5714 | } | |
b53d3f5d | 5715 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5716 | return err; |
5717 | } | |
5718 | ||
18bb1db3 | 5719 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5720 | { |
b9d86667 | 5721 | struct inode *inode = NULL; |
39279cc3 CM |
5722 | struct btrfs_trans_handle *trans; |
5723 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5724 | int err = 0; | |
5725 | int drop_on_err = 0; | |
b9d86667 | 5726 | u64 objectid = 0; |
00e4e6b3 | 5727 | u64 index = 0; |
39279cc3 | 5728 | |
9ed74f2d JB |
5729 | /* |
5730 | * 2 items for inode and ref | |
5731 | * 2 items for dir items | |
5732 | * 1 for xattr if selinux is on | |
5733 | */ | |
a22285a6 YZ |
5734 | trans = btrfs_start_transaction(root, 5); |
5735 | if (IS_ERR(trans)) | |
5736 | return PTR_ERR(trans); | |
39279cc3 | 5737 | |
581bb050 LZ |
5738 | err = btrfs_find_free_ino(root, &objectid); |
5739 | if (err) | |
5740 | goto out_fail; | |
5741 | ||
aec7477b | 5742 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5743 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5744 | S_IFDIR | mode, &index); |
39279cc3 CM |
5745 | if (IS_ERR(inode)) { |
5746 | err = PTR_ERR(inode); | |
5747 | goto out_fail; | |
5748 | } | |
5f39d397 | 5749 | |
39279cc3 | 5750 | drop_on_err = 1; |
33268eaf | 5751 | |
2a7dba39 | 5752 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5753 | if (err) |
5754 | goto out_fail; | |
5755 | ||
39279cc3 CM |
5756 | inode->i_op = &btrfs_dir_inode_operations; |
5757 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5758 | |
dbe674a9 | 5759 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5760 | err = btrfs_update_inode(trans, root, inode); |
5761 | if (err) | |
5762 | goto out_fail; | |
5f39d397 | 5763 | |
a1b075d2 JB |
5764 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5765 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5766 | if (err) |
5767 | goto out_fail; | |
5f39d397 | 5768 | |
39279cc3 CM |
5769 | d_instantiate(dentry, inode); |
5770 | drop_on_err = 0; | |
39279cc3 CM |
5771 | |
5772 | out_fail: | |
7ad85bb7 | 5773 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5774 | if (drop_on_err) |
5775 | iput(inode); | |
b53d3f5d | 5776 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5777 | return err; |
5778 | } | |
5779 | ||
d352ac68 CM |
5780 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5781 | * and an extent that you want to insert, deal with overlap and insert | |
5782 | * the new extent into the tree. | |
5783 | */ | |
3b951516 CM |
5784 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5785 | struct extent_map *existing, | |
e6dcd2dc CM |
5786 | struct extent_map *em, |
5787 | u64 map_start, u64 map_len) | |
3b951516 CM |
5788 | { |
5789 | u64 start_diff; | |
3b951516 | 5790 | |
e6dcd2dc CM |
5791 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5792 | start_diff = map_start - em->start; | |
5793 | em->start = map_start; | |
5794 | em->len = map_len; | |
c8b97818 CM |
5795 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5796 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5797 | em->block_start += start_diff; |
c8b97818 CM |
5798 | em->block_len -= start_diff; |
5799 | } | |
09a2a8f9 | 5800 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
5801 | } |
5802 | ||
c8b97818 CM |
5803 | static noinline int uncompress_inline(struct btrfs_path *path, |
5804 | struct inode *inode, struct page *page, | |
5805 | size_t pg_offset, u64 extent_offset, | |
5806 | struct btrfs_file_extent_item *item) | |
5807 | { | |
5808 | int ret; | |
5809 | struct extent_buffer *leaf = path->nodes[0]; | |
5810 | char *tmp; | |
5811 | size_t max_size; | |
5812 | unsigned long inline_size; | |
5813 | unsigned long ptr; | |
261507a0 | 5814 | int compress_type; |
c8b97818 CM |
5815 | |
5816 | WARN_ON(pg_offset != 0); | |
261507a0 | 5817 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5818 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5819 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5820 | btrfs_item_nr(leaf, path->slots[0])); | |
5821 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5822 | if (!tmp) |
5823 | return -ENOMEM; | |
c8b97818 CM |
5824 | ptr = btrfs_file_extent_inline_start(item); |
5825 | ||
5826 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5827 | ||
5b050f04 | 5828 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5829 | ret = btrfs_decompress(compress_type, tmp, page, |
5830 | extent_offset, inline_size, max_size); | |
c8b97818 | 5831 | if (ret) { |
7ac687d9 | 5832 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5833 | unsigned long copy_size = min_t(u64, |
5834 | PAGE_CACHE_SIZE - pg_offset, | |
5835 | max_size - extent_offset); | |
5836 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5837 | kunmap_atomic(kaddr); |
c8b97818 CM |
5838 | } |
5839 | kfree(tmp); | |
5840 | return 0; | |
5841 | } | |
5842 | ||
d352ac68 CM |
5843 | /* |
5844 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5845 | * the ugly parts come from merging extents from the disk with the in-ram |
5846 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5847 | * where the in-ram extents might be locked pending data=ordered completion. |
5848 | * | |
5849 | * This also copies inline extents directly into the page. | |
5850 | */ | |
d397712b | 5851 | |
a52d9a80 | 5852 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5853 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5854 | int create) |
5855 | { | |
5856 | int ret; | |
5857 | int err = 0; | |
db94535d | 5858 | u64 bytenr; |
a52d9a80 CM |
5859 | u64 extent_start = 0; |
5860 | u64 extent_end = 0; | |
33345d01 | 5861 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5862 | u32 found_type; |
f421950f | 5863 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5864 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5865 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5866 | struct extent_buffer *leaf; |
5867 | struct btrfs_key found_key; | |
a52d9a80 CM |
5868 | struct extent_map *em = NULL; |
5869 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5870 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5871 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5872 | int compress_type; |
a52d9a80 | 5873 | |
a52d9a80 | 5874 | again: |
890871be | 5875 | read_lock(&em_tree->lock); |
d1310b2e | 5876 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5877 | if (em) |
5878 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5879 | read_unlock(&em_tree->lock); |
d1310b2e | 5880 | |
a52d9a80 | 5881 | if (em) { |
e1c4b745 CM |
5882 | if (em->start > start || em->start + em->len <= start) |
5883 | free_extent_map(em); | |
5884 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5885 | free_extent_map(em); |
5886 | else | |
5887 | goto out; | |
a52d9a80 | 5888 | } |
172ddd60 | 5889 | em = alloc_extent_map(); |
a52d9a80 | 5890 | if (!em) { |
d1310b2e CM |
5891 | err = -ENOMEM; |
5892 | goto out; | |
a52d9a80 | 5893 | } |
e6dcd2dc | 5894 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5895 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5896 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5897 | em->len = (u64)-1; |
c8b97818 | 5898 | em->block_len = (u64)-1; |
f421950f CM |
5899 | |
5900 | if (!path) { | |
5901 | path = btrfs_alloc_path(); | |
026fd317 JB |
5902 | if (!path) { |
5903 | err = -ENOMEM; | |
5904 | goto out; | |
5905 | } | |
5906 | /* | |
5907 | * Chances are we'll be called again, so go ahead and do | |
5908 | * readahead | |
5909 | */ | |
5910 | path->reada = 1; | |
f421950f CM |
5911 | } |
5912 | ||
179e29e4 CM |
5913 | ret = btrfs_lookup_file_extent(trans, root, path, |
5914 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5915 | if (ret < 0) { |
5916 | err = ret; | |
5917 | goto out; | |
5918 | } | |
5919 | ||
5920 | if (ret != 0) { | |
5921 | if (path->slots[0] == 0) | |
5922 | goto not_found; | |
5923 | path->slots[0]--; | |
5924 | } | |
5925 | ||
5f39d397 CM |
5926 | leaf = path->nodes[0]; |
5927 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5928 | struct btrfs_file_extent_item); |
a52d9a80 | 5929 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5930 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5931 | found_type = btrfs_key_type(&found_key); | |
5932 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5933 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5934 | goto not_found; | |
5935 | } | |
5936 | ||
5f39d397 CM |
5937 | found_type = btrfs_file_extent_type(leaf, item); |
5938 | extent_start = found_key.offset; | |
261507a0 | 5939 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5940 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5941 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5942 | extent_end = extent_start + |
db94535d | 5943 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5944 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5945 | size_t size; | |
5946 | size = btrfs_file_extent_inline_len(leaf, item); | |
fda2832f | 5947 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 YZ |
5948 | } |
5949 | ||
5950 | if (start >= extent_end) { | |
5951 | path->slots[0]++; | |
5952 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5953 | ret = btrfs_next_leaf(root, path); | |
5954 | if (ret < 0) { | |
5955 | err = ret; | |
5956 | goto out; | |
a52d9a80 | 5957 | } |
9036c102 YZ |
5958 | if (ret > 0) |
5959 | goto not_found; | |
5960 | leaf = path->nodes[0]; | |
a52d9a80 | 5961 | } |
9036c102 YZ |
5962 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5963 | if (found_key.objectid != objectid || | |
5964 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5965 | goto not_found; | |
5966 | if (start + len <= found_key.offset) | |
5967 | goto not_found; | |
5968 | em->start = start; | |
70c8a91c | 5969 | em->orig_start = start; |
9036c102 YZ |
5970 | em->len = found_key.offset - start; |
5971 | goto not_found_em; | |
5972 | } | |
5973 | ||
cc95bef6 | 5974 | em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); |
d899e052 YZ |
5975 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5976 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5977 | em->start = extent_start; |
5978 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5979 | em->orig_start = extent_start - |
5980 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
5981 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
5982 | item); | |
db94535d CM |
5983 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5984 | if (bytenr == 0) { | |
5f39d397 | 5985 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5986 | goto insert; |
5987 | } | |
261507a0 | 5988 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5989 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5990 | em->compress_type = compress_type; |
c8b97818 | 5991 | em->block_start = bytenr; |
b4939680 | 5992 | em->block_len = em->orig_block_len; |
c8b97818 CM |
5993 | } else { |
5994 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5995 | em->block_start = bytenr; | |
5996 | em->block_len = em->len; | |
d899e052 YZ |
5997 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5998 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5999 | } |
a52d9a80 CM |
6000 | goto insert; |
6001 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6002 | unsigned long ptr; |
a52d9a80 | 6003 | char *map; |
3326d1b0 CM |
6004 | size_t size; |
6005 | size_t extent_offset; | |
6006 | size_t copy_size; | |
a52d9a80 | 6007 | |
689f9346 | 6008 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 6009 | if (!page || create) { |
689f9346 | 6010 | em->start = extent_start; |
9036c102 | 6011 | em->len = extent_end - extent_start; |
689f9346 Y |
6012 | goto out; |
6013 | } | |
5f39d397 | 6014 | |
9036c102 YZ |
6015 | size = btrfs_file_extent_inline_len(leaf, item); |
6016 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 6017 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6018 | size - extent_offset); |
3326d1b0 | 6019 | em->start = extent_start + extent_offset; |
fda2832f | 6020 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6021 | em->orig_block_len = em->len; |
70c8a91c | 6022 | em->orig_start = em->start; |
261507a0 | 6023 | if (compress_type) { |
c8b97818 | 6024 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
6025 | em->compress_type = compress_type; |
6026 | } | |
689f9346 | 6027 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6028 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6029 | if (btrfs_file_extent_compression(leaf, item) != |
6030 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6031 | ret = uncompress_inline(path, inode, page, |
6032 | pg_offset, | |
6033 | extent_offset, item); | |
79787eaa | 6034 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
6035 | } else { |
6036 | map = kmap(page); | |
6037 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6038 | copy_size); | |
93c82d57 CM |
6039 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6040 | memset(map + pg_offset + copy_size, 0, | |
6041 | PAGE_CACHE_SIZE - pg_offset - | |
6042 | copy_size); | |
6043 | } | |
c8b97818 CM |
6044 | kunmap(page); |
6045 | } | |
179e29e4 CM |
6046 | flush_dcache_page(page); |
6047 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6048 | BUG(); |
179e29e4 CM |
6049 | if (!trans) { |
6050 | kunmap(page); | |
6051 | free_extent_map(em); | |
6052 | em = NULL; | |
ff5714cc | 6053 | |
b3b4aa74 | 6054 | btrfs_release_path(path); |
7a7eaa40 | 6055 | trans = btrfs_join_transaction(root); |
ff5714cc | 6056 | |
3612b495 TI |
6057 | if (IS_ERR(trans)) |
6058 | return ERR_CAST(trans); | |
179e29e4 CM |
6059 | goto again; |
6060 | } | |
c8b97818 | 6061 | map = kmap(page); |
70dec807 | 6062 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6063 | copy_size); |
c8b97818 | 6064 | kunmap(page); |
179e29e4 | 6065 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6066 | } |
d1310b2e | 6067 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6068 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
6069 | goto insert; |
6070 | } else { | |
31b1a2bd | 6071 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
6072 | } |
6073 | not_found: | |
6074 | em->start = start; | |
70c8a91c | 6075 | em->orig_start = start; |
d1310b2e | 6076 | em->len = len; |
a52d9a80 | 6077 | not_found_em: |
5f39d397 | 6078 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6079 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6080 | insert: |
b3b4aa74 | 6081 | btrfs_release_path(path); |
d1310b2e | 6082 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb SK |
6083 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6084 | (unsigned long long)em->start, | |
6085 | (unsigned long long)em->len, | |
6086 | (unsigned long long)start, | |
6087 | (unsigned long long)len); | |
a52d9a80 CM |
6088 | err = -EIO; |
6089 | goto out; | |
6090 | } | |
d1310b2e CM |
6091 | |
6092 | err = 0; | |
890871be | 6093 | write_lock(&em_tree->lock); |
09a2a8f9 | 6094 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6095 | /* it is possible that someone inserted the extent into the tree |
6096 | * while we had the lock dropped. It is also possible that | |
6097 | * an overlapping map exists in the tree | |
6098 | */ | |
a52d9a80 | 6099 | if (ret == -EEXIST) { |
3b951516 | 6100 | struct extent_map *existing; |
e6dcd2dc CM |
6101 | |
6102 | ret = 0; | |
6103 | ||
3b951516 | 6104 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
6105 | if (existing && (existing->start > start || |
6106 | existing->start + existing->len <= start)) { | |
6107 | free_extent_map(existing); | |
6108 | existing = NULL; | |
6109 | } | |
3b951516 CM |
6110 | if (!existing) { |
6111 | existing = lookup_extent_mapping(em_tree, em->start, | |
6112 | em->len); | |
6113 | if (existing) { | |
6114 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
6115 | em, start, |
6116 | root->sectorsize); | |
3b951516 CM |
6117 | free_extent_map(existing); |
6118 | if (err) { | |
6119 | free_extent_map(em); | |
6120 | em = NULL; | |
6121 | } | |
6122 | } else { | |
6123 | err = -EIO; | |
3b951516 CM |
6124 | free_extent_map(em); |
6125 | em = NULL; | |
6126 | } | |
6127 | } else { | |
6128 | free_extent_map(em); | |
6129 | em = existing; | |
e6dcd2dc | 6130 | err = 0; |
a52d9a80 | 6131 | } |
a52d9a80 | 6132 | } |
890871be | 6133 | write_unlock(&em_tree->lock); |
a52d9a80 | 6134 | out: |
1abe9b8a | 6135 | |
f0bd95ea TI |
6136 | if (em) |
6137 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 6138 | |
f421950f CM |
6139 | if (path) |
6140 | btrfs_free_path(path); | |
a52d9a80 CM |
6141 | if (trans) { |
6142 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6143 | if (!err) |
a52d9a80 CM |
6144 | err = ret; |
6145 | } | |
a52d9a80 CM |
6146 | if (err) { |
6147 | free_extent_map(em); | |
a52d9a80 CM |
6148 | return ERR_PTR(err); |
6149 | } | |
79787eaa | 6150 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6151 | return em; |
6152 | } | |
6153 | ||
ec29ed5b CM |
6154 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6155 | size_t pg_offset, u64 start, u64 len, | |
6156 | int create) | |
6157 | { | |
6158 | struct extent_map *em; | |
6159 | struct extent_map *hole_em = NULL; | |
6160 | u64 range_start = start; | |
6161 | u64 end; | |
6162 | u64 found; | |
6163 | u64 found_end; | |
6164 | int err = 0; | |
6165 | ||
6166 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6167 | if (IS_ERR(em)) | |
6168 | return em; | |
6169 | if (em) { | |
6170 | /* | |
f9e4fb53 LB |
6171 | * if our em maps to |
6172 | * - a hole or | |
6173 | * - a pre-alloc extent, | |
6174 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6175 | */ |
f9e4fb53 LB |
6176 | if (em->block_start != EXTENT_MAP_HOLE && |
6177 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6178 | return em; |
6179 | else | |
6180 | hole_em = em; | |
6181 | } | |
6182 | ||
6183 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6184 | end = start + len; | |
6185 | if (end < start) | |
6186 | end = (u64)-1; | |
6187 | else | |
6188 | end -= 1; | |
6189 | ||
6190 | em = NULL; | |
6191 | ||
6192 | /* ok, we didn't find anything, lets look for delalloc */ | |
6193 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6194 | end, len, EXTENT_DELALLOC, 1); | |
6195 | found_end = range_start + found; | |
6196 | if (found_end < range_start) | |
6197 | found_end = (u64)-1; | |
6198 | ||
6199 | /* | |
6200 | * we didn't find anything useful, return | |
6201 | * the original results from get_extent() | |
6202 | */ | |
6203 | if (range_start > end || found_end <= start) { | |
6204 | em = hole_em; | |
6205 | hole_em = NULL; | |
6206 | goto out; | |
6207 | } | |
6208 | ||
6209 | /* adjust the range_start to make sure it doesn't | |
6210 | * go backwards from the start they passed in | |
6211 | */ | |
6212 | range_start = max(start,range_start); | |
6213 | found = found_end - range_start; | |
6214 | ||
6215 | if (found > 0) { | |
6216 | u64 hole_start = start; | |
6217 | u64 hole_len = len; | |
6218 | ||
172ddd60 | 6219 | em = alloc_extent_map(); |
ec29ed5b CM |
6220 | if (!em) { |
6221 | err = -ENOMEM; | |
6222 | goto out; | |
6223 | } | |
6224 | /* | |
6225 | * when btrfs_get_extent can't find anything it | |
6226 | * returns one huge hole | |
6227 | * | |
6228 | * make sure what it found really fits our range, and | |
6229 | * adjust to make sure it is based on the start from | |
6230 | * the caller | |
6231 | */ | |
6232 | if (hole_em) { | |
6233 | u64 calc_end = extent_map_end(hole_em); | |
6234 | ||
6235 | if (calc_end <= start || (hole_em->start > end)) { | |
6236 | free_extent_map(hole_em); | |
6237 | hole_em = NULL; | |
6238 | } else { | |
6239 | hole_start = max(hole_em->start, start); | |
6240 | hole_len = calc_end - hole_start; | |
6241 | } | |
6242 | } | |
6243 | em->bdev = NULL; | |
6244 | if (hole_em && range_start > hole_start) { | |
6245 | /* our hole starts before our delalloc, so we | |
6246 | * have to return just the parts of the hole | |
6247 | * that go until the delalloc starts | |
6248 | */ | |
6249 | em->len = min(hole_len, | |
6250 | range_start - hole_start); | |
6251 | em->start = hole_start; | |
6252 | em->orig_start = hole_start; | |
6253 | /* | |
6254 | * don't adjust block start at all, | |
6255 | * it is fixed at EXTENT_MAP_HOLE | |
6256 | */ | |
6257 | em->block_start = hole_em->block_start; | |
6258 | em->block_len = hole_len; | |
f9e4fb53 LB |
6259 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6260 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6261 | } else { |
6262 | em->start = range_start; | |
6263 | em->len = found; | |
6264 | em->orig_start = range_start; | |
6265 | em->block_start = EXTENT_MAP_DELALLOC; | |
6266 | em->block_len = found; | |
6267 | } | |
6268 | } else if (hole_em) { | |
6269 | return hole_em; | |
6270 | } | |
6271 | out: | |
6272 | ||
6273 | free_extent_map(hole_em); | |
6274 | if (err) { | |
6275 | free_extent_map(em); | |
6276 | return ERR_PTR(err); | |
6277 | } | |
6278 | return em; | |
6279 | } | |
6280 | ||
4b46fce2 JB |
6281 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6282 | u64 start, u64 len) | |
6283 | { | |
6284 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 6285 | struct extent_map *em; |
4b46fce2 JB |
6286 | struct btrfs_key ins; |
6287 | u64 alloc_hint; | |
6288 | int ret; | |
4b46fce2 | 6289 | |
4b46fce2 | 6290 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 6291 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
81c9ad23 | 6292 | alloc_hint, &ins, 1); |
00361589 JB |
6293 | if (ret) |
6294 | return ERR_PTR(ret); | |
4b46fce2 | 6295 | |
70c8a91c | 6296 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6297 | ins.offset, ins.offset, ins.offset, 0); |
00361589 JB |
6298 | if (IS_ERR(em)) { |
6299 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
6300 | return em; | |
6301 | } | |
4b46fce2 JB |
6302 | |
6303 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6304 | ins.offset, ins.offset, 0); | |
6305 | if (ret) { | |
6306 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
00361589 JB |
6307 | free_extent_map(em); |
6308 | return ERR_PTR(ret); | |
4b46fce2 | 6309 | } |
00361589 | 6310 | |
4b46fce2 JB |
6311 | return em; |
6312 | } | |
6313 | ||
46bfbb5c CM |
6314 | /* |
6315 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6316 | * block must be cow'd | |
6317 | */ | |
00361589 | 6318 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
6319 | u64 *orig_start, u64 *orig_block_len, |
6320 | u64 *ram_bytes) | |
46bfbb5c | 6321 | { |
00361589 | 6322 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
6323 | struct btrfs_path *path; |
6324 | int ret; | |
6325 | struct extent_buffer *leaf; | |
6326 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6327 | struct btrfs_file_extent_item *fi; | |
6328 | struct btrfs_key key; | |
6329 | u64 disk_bytenr; | |
6330 | u64 backref_offset; | |
6331 | u64 extent_end; | |
6332 | u64 num_bytes; | |
6333 | int slot; | |
6334 | int found_type; | |
7ee9e440 | 6335 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
46bfbb5c CM |
6336 | path = btrfs_alloc_path(); |
6337 | if (!path) | |
6338 | return -ENOMEM; | |
6339 | ||
00361589 | 6340 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6341 | offset, 0); |
6342 | if (ret < 0) | |
6343 | goto out; | |
6344 | ||
6345 | slot = path->slots[0]; | |
6346 | if (ret == 1) { | |
6347 | if (slot == 0) { | |
6348 | /* can't find the item, must cow */ | |
6349 | ret = 0; | |
6350 | goto out; | |
6351 | } | |
6352 | slot--; | |
6353 | } | |
6354 | ret = 0; | |
6355 | leaf = path->nodes[0]; | |
6356 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6357 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6358 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6359 | /* not our file or wrong item type, must cow */ | |
6360 | goto out; | |
6361 | } | |
6362 | ||
6363 | if (key.offset > offset) { | |
6364 | /* Wrong offset, must cow */ | |
6365 | goto out; | |
6366 | } | |
6367 | ||
6368 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6369 | found_type = btrfs_file_extent_type(leaf, fi); | |
6370 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6371 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6372 | /* not a regular extent, must cow */ | |
6373 | goto out; | |
6374 | } | |
7ee9e440 JB |
6375 | |
6376 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
6377 | goto out; | |
6378 | ||
46bfbb5c | 6379 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
6380 | if (disk_bytenr == 0) |
6381 | goto out; | |
6382 | ||
6383 | if (btrfs_file_extent_compression(leaf, fi) || | |
6384 | btrfs_file_extent_encryption(leaf, fi) || | |
6385 | btrfs_file_extent_other_encoding(leaf, fi)) | |
6386 | goto out; | |
6387 | ||
46bfbb5c CM |
6388 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
6389 | ||
7ee9e440 JB |
6390 | if (orig_start) { |
6391 | *orig_start = key.offset - backref_offset; | |
6392 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6393 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6394 | } | |
eb384b55 | 6395 | |
46bfbb5c | 6396 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
46bfbb5c CM |
6397 | |
6398 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
6399 | goto out; | |
6400 | ||
6401 | /* | |
6402 | * look for other files referencing this extent, if we | |
6403 | * find any we must cow | |
6404 | */ | |
00361589 JB |
6405 | trans = btrfs_join_transaction(root); |
6406 | if (IS_ERR(trans)) { | |
6407 | ret = 0; | |
46bfbb5c | 6408 | goto out; |
00361589 JB |
6409 | } |
6410 | ||
6411 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
6412 | key.offset - backref_offset, disk_bytenr); | |
6413 | btrfs_end_transaction(trans, root); | |
6414 | if (ret) { | |
6415 | ret = 0; | |
6416 | goto out; | |
6417 | } | |
46bfbb5c CM |
6418 | |
6419 | /* | |
6420 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6421 | * in this extent we are about to write. If there | |
6422 | * are any csums in that range we have to cow in order | |
6423 | * to keep the csums correct | |
6424 | */ | |
6425 | disk_bytenr += backref_offset; | |
6426 | disk_bytenr += offset - key.offset; | |
eb384b55 | 6427 | num_bytes = min(offset + *len, extent_end) - offset; |
46bfbb5c CM |
6428 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6429 | goto out; | |
6430 | /* | |
6431 | * all of the above have passed, it is safe to overwrite this extent | |
6432 | * without cow | |
6433 | */ | |
eb384b55 | 6434 | *len = num_bytes; |
46bfbb5c CM |
6435 | ret = 1; |
6436 | out: | |
6437 | btrfs_free_path(path); | |
6438 | return ret; | |
6439 | } | |
6440 | ||
eb838e73 JB |
6441 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6442 | struct extent_state **cached_state, int writing) | |
6443 | { | |
6444 | struct btrfs_ordered_extent *ordered; | |
6445 | int ret = 0; | |
6446 | ||
6447 | while (1) { | |
6448 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6449 | 0, cached_state); | |
6450 | /* | |
6451 | * We're concerned with the entire range that we're going to be | |
6452 | * doing DIO to, so we need to make sure theres no ordered | |
6453 | * extents in this range. | |
6454 | */ | |
6455 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6456 | lockend - lockstart + 1); | |
6457 | ||
6458 | /* | |
6459 | * We need to make sure there are no buffered pages in this | |
6460 | * range either, we could have raced between the invalidate in | |
6461 | * generic_file_direct_write and locking the extent. The | |
6462 | * invalidate needs to happen so that reads after a write do not | |
6463 | * get stale data. | |
6464 | */ | |
6465 | if (!ordered && (!writing || | |
6466 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
6467 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
6468 | *cached_state))) | |
6469 | break; | |
6470 | ||
6471 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6472 | cached_state, GFP_NOFS); | |
6473 | ||
6474 | if (ordered) { | |
6475 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6476 | btrfs_put_ordered_extent(ordered); | |
6477 | } else { | |
6478 | /* Screw you mmap */ | |
6479 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
6480 | lockstart, | |
6481 | lockend); | |
6482 | if (ret) | |
6483 | break; | |
6484 | ||
6485 | /* | |
6486 | * If we found a page that couldn't be invalidated just | |
6487 | * fall back to buffered. | |
6488 | */ | |
6489 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6490 | lockstart >> PAGE_CACHE_SHIFT, | |
6491 | lockend >> PAGE_CACHE_SHIFT); | |
6492 | if (ret) | |
6493 | break; | |
6494 | } | |
6495 | ||
6496 | cond_resched(); | |
6497 | } | |
6498 | ||
6499 | return ret; | |
6500 | } | |
6501 | ||
69ffb543 JB |
6502 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
6503 | u64 len, u64 orig_start, | |
6504 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
6505 | u64 orig_block_len, u64 ram_bytes, |
6506 | int type) | |
69ffb543 JB |
6507 | { |
6508 | struct extent_map_tree *em_tree; | |
6509 | struct extent_map *em; | |
6510 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6511 | int ret; | |
6512 | ||
6513 | em_tree = &BTRFS_I(inode)->extent_tree; | |
6514 | em = alloc_extent_map(); | |
6515 | if (!em) | |
6516 | return ERR_PTR(-ENOMEM); | |
6517 | ||
6518 | em->start = start; | |
6519 | em->orig_start = orig_start; | |
2ab28f32 JB |
6520 | em->mod_start = start; |
6521 | em->mod_len = len; | |
69ffb543 JB |
6522 | em->len = len; |
6523 | em->block_len = block_len; | |
6524 | em->block_start = block_start; | |
6525 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6526 | em->orig_block_len = orig_block_len; |
cc95bef6 | 6527 | em->ram_bytes = ram_bytes; |
70c8a91c | 6528 | em->generation = -1; |
69ffb543 JB |
6529 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6530 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6531 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6532 | |
6533 | do { | |
6534 | btrfs_drop_extent_cache(inode, em->start, | |
6535 | em->start + em->len - 1, 0); | |
6536 | write_lock(&em_tree->lock); | |
09a2a8f9 | 6537 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
6538 | write_unlock(&em_tree->lock); |
6539 | } while (ret == -EEXIST); | |
6540 | ||
6541 | if (ret) { | |
6542 | free_extent_map(em); | |
6543 | return ERR_PTR(ret); | |
6544 | } | |
6545 | ||
6546 | return em; | |
6547 | } | |
6548 | ||
6549 | ||
4b46fce2 JB |
6550 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6551 | struct buffer_head *bh_result, int create) | |
6552 | { | |
6553 | struct extent_map *em; | |
6554 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6555 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6556 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6557 | u64 lockstart, lockend; |
4b46fce2 | 6558 | u64 len = bh_result->b_size; |
eb838e73 | 6559 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 6560 | int ret = 0; |
eb838e73 | 6561 | |
172a5049 | 6562 | if (create) |
eb838e73 | 6563 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 6564 | else |
c329861d | 6565 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 6566 | |
c329861d JB |
6567 | lockstart = start; |
6568 | lockend = start + len - 1; | |
6569 | ||
eb838e73 JB |
6570 | /* |
6571 | * If this errors out it's because we couldn't invalidate pagecache for | |
6572 | * this range and we need to fallback to buffered. | |
6573 | */ | |
6574 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6575 | return -ENOTBLK; | |
6576 | ||
4b46fce2 | 6577 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
6578 | if (IS_ERR(em)) { |
6579 | ret = PTR_ERR(em); | |
6580 | goto unlock_err; | |
6581 | } | |
4b46fce2 JB |
6582 | |
6583 | /* | |
6584 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6585 | * io. INLINE is special, and we could probably kludge it in here, but | |
6586 | * it's still buffered so for safety lets just fall back to the generic | |
6587 | * buffered path. | |
6588 | * | |
6589 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6590 | * decompress it, so there will be buffering required no matter what we | |
6591 | * do, so go ahead and fallback to buffered. | |
6592 | * | |
6593 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6594 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6595 | * the generic code. | |
6596 | */ | |
6597 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6598 | em->block_start == EXTENT_MAP_INLINE) { | |
6599 | free_extent_map(em); | |
eb838e73 JB |
6600 | ret = -ENOTBLK; |
6601 | goto unlock_err; | |
4b46fce2 JB |
6602 | } |
6603 | ||
6604 | /* Just a good old fashioned hole, return */ | |
6605 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6606 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6607 | free_extent_map(em); | |
eb838e73 | 6608 | goto unlock_err; |
4b46fce2 JB |
6609 | } |
6610 | ||
6611 | /* | |
6612 | * We don't allocate a new extent in the following cases | |
6613 | * | |
6614 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6615 | * existing extent. | |
6616 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6617 | * just use the extent. | |
6618 | * | |
6619 | */ | |
46bfbb5c | 6620 | if (!create) { |
eb838e73 JB |
6621 | len = min(len, em->len - (start - em->start)); |
6622 | lockstart = start + len; | |
6623 | goto unlock; | |
46bfbb5c | 6624 | } |
4b46fce2 JB |
6625 | |
6626 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6627 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6628 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6629 | int type; |
6630 | int ret; | |
eb384b55 | 6631 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
6632 | |
6633 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6634 | type = BTRFS_ORDERED_PREALLOC; | |
6635 | else | |
6636 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6637 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6638 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 6639 | |
00361589 | 6640 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 6641 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
6642 | if (type == BTRFS_ORDERED_PREALLOC) { |
6643 | free_extent_map(em); | |
6644 | em = create_pinned_em(inode, start, len, | |
6645 | orig_start, | |
b4939680 | 6646 | block_start, len, |
cc95bef6 JB |
6647 | orig_block_len, |
6648 | ram_bytes, type); | |
00361589 | 6649 | if (IS_ERR(em)) |
69ffb543 | 6650 | goto unlock_err; |
69ffb543 JB |
6651 | } |
6652 | ||
46bfbb5c CM |
6653 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6654 | block_start, len, len, type); | |
46bfbb5c CM |
6655 | if (ret) { |
6656 | free_extent_map(em); | |
eb838e73 | 6657 | goto unlock_err; |
46bfbb5c CM |
6658 | } |
6659 | goto unlock; | |
4b46fce2 | 6660 | } |
4b46fce2 | 6661 | } |
00361589 | 6662 | |
46bfbb5c CM |
6663 | /* |
6664 | * this will cow the extent, reset the len in case we changed | |
6665 | * it above | |
6666 | */ | |
6667 | len = bh_result->b_size; | |
70c8a91c JB |
6668 | free_extent_map(em); |
6669 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6670 | if (IS_ERR(em)) { |
6671 | ret = PTR_ERR(em); | |
6672 | goto unlock_err; | |
6673 | } | |
46bfbb5c CM |
6674 | len = min(len, em->len - (start - em->start)); |
6675 | unlock: | |
4b46fce2 JB |
6676 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6677 | inode->i_blkbits; | |
46bfbb5c | 6678 | bh_result->b_size = len; |
4b46fce2 JB |
6679 | bh_result->b_bdev = em->bdev; |
6680 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6681 | if (create) { |
6682 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6683 | set_buffer_new(bh_result); | |
6684 | ||
6685 | /* | |
6686 | * Need to update the i_size under the extent lock so buffered | |
6687 | * readers will get the updated i_size when we unlock. | |
6688 | */ | |
6689 | if (start + len > i_size_read(inode)) | |
6690 | i_size_write(inode, start + len); | |
0934856d | 6691 | |
172a5049 MX |
6692 | spin_lock(&BTRFS_I(inode)->lock); |
6693 | BTRFS_I(inode)->outstanding_extents++; | |
6694 | spin_unlock(&BTRFS_I(inode)->lock); | |
6695 | ||
0934856d MX |
6696 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6697 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
6698 | &cached_state, GFP_NOFS); | |
6699 | BUG_ON(ret); | |
c3473e83 | 6700 | } |
4b46fce2 | 6701 | |
eb838e73 JB |
6702 | /* |
6703 | * In the case of write we need to clear and unlock the entire range, | |
6704 | * in the case of read we need to unlock only the end area that we | |
6705 | * aren't using if there is any left over space. | |
6706 | */ | |
24c03fa5 | 6707 | if (lockstart < lockend) { |
0934856d MX |
6708 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6709 | lockend, unlock_bits, 1, 0, | |
6710 | &cached_state, GFP_NOFS); | |
24c03fa5 | 6711 | } else { |
eb838e73 | 6712 | free_extent_state(cached_state); |
24c03fa5 | 6713 | } |
eb838e73 | 6714 | |
4b46fce2 JB |
6715 | free_extent_map(em); |
6716 | ||
6717 | return 0; | |
eb838e73 JB |
6718 | |
6719 | unlock_err: | |
eb838e73 JB |
6720 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6721 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6722 | return ret; | |
4b46fce2 JB |
6723 | } |
6724 | ||
4b46fce2 JB |
6725 | static void btrfs_endio_direct_read(struct bio *bio, int err) |
6726 | { | |
e65e1535 | 6727 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6728 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6729 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6730 | struct inode *inode = dip->inode; |
6731 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9be3395b | 6732 | struct bio *dio_bio; |
facc8a22 MX |
6733 | u32 *csums = (u32 *)dip->csum; |
6734 | int index = 0; | |
4b46fce2 | 6735 | u64 start; |
4b46fce2 JB |
6736 | |
6737 | start = dip->logical_offset; | |
6738 | do { | |
6739 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6740 | struct page *page = bvec->bv_page; | |
6741 | char *kaddr; | |
6742 | u32 csum = ~(u32)0; | |
6743 | unsigned long flags; | |
6744 | ||
6745 | local_irq_save(flags); | |
7ac687d9 | 6746 | kaddr = kmap_atomic(page); |
b0496686 | 6747 | csum = btrfs_csum_data(kaddr + bvec->bv_offset, |
4b46fce2 JB |
6748 | csum, bvec->bv_len); |
6749 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6750 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6751 | local_irq_restore(flags); |
6752 | ||
6753 | flush_dcache_page(bvec->bv_page); | |
facc8a22 MX |
6754 | if (csum != csums[index]) { |
6755 | btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u", | |
6756 | (unsigned long long)btrfs_ino(inode), | |
6757 | (unsigned long long)start, | |
6758 | csum, csums[index]); | |
4b46fce2 JB |
6759 | err = -EIO; |
6760 | } | |
6761 | } | |
6762 | ||
6763 | start += bvec->bv_len; | |
4b46fce2 | 6764 | bvec++; |
facc8a22 | 6765 | index++; |
4b46fce2 JB |
6766 | } while (bvec <= bvec_end); |
6767 | ||
6768 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6769 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 6770 | dio_bio = dip->dio_bio; |
4b46fce2 | 6771 | |
4b46fce2 | 6772 | kfree(dip); |
c0da7aa1 JB |
6773 | |
6774 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6775 | if (err) | |
9be3395b CM |
6776 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6777 | dio_end_io(dio_bio, err); | |
6778 | bio_put(bio); | |
4b46fce2 JB |
6779 | } |
6780 | ||
6781 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6782 | { | |
6783 | struct btrfs_dio_private *dip = bio->bi_private; | |
6784 | struct inode *inode = dip->inode; | |
6785 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6786 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6787 | u64 ordered_offset = dip->logical_offset; |
6788 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 6789 | struct bio *dio_bio; |
4b46fce2 JB |
6790 | int ret; |
6791 | ||
6792 | if (err) | |
6793 | goto out_done; | |
163cf09c CM |
6794 | again: |
6795 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6796 | &ordered_offset, | |
5fd02043 | 6797 | ordered_bytes, !err); |
4b46fce2 | 6798 | if (!ret) |
163cf09c | 6799 | goto out_test; |
4b46fce2 | 6800 | |
5fd02043 JB |
6801 | ordered->work.func = finish_ordered_fn; |
6802 | ordered->work.flags = 0; | |
6803 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6804 | &ordered->work); | |
163cf09c CM |
6805 | out_test: |
6806 | /* | |
6807 | * our bio might span multiple ordered extents. If we haven't | |
6808 | * completed the accounting for the whole dio, go back and try again | |
6809 | */ | |
6810 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6811 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6812 | ordered_offset; | |
5fd02043 | 6813 | ordered = NULL; |
163cf09c CM |
6814 | goto again; |
6815 | } | |
4b46fce2 | 6816 | out_done: |
9be3395b | 6817 | dio_bio = dip->dio_bio; |
4b46fce2 | 6818 | |
4b46fce2 | 6819 | kfree(dip); |
c0da7aa1 JB |
6820 | |
6821 | /* If we had an error make sure to clear the uptodate flag */ | |
6822 | if (err) | |
9be3395b CM |
6823 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6824 | dio_end_io(dio_bio, err); | |
6825 | bio_put(bio); | |
4b46fce2 JB |
6826 | } |
6827 | ||
eaf25d93 CM |
6828 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6829 | struct bio *bio, int mirror_num, | |
6830 | unsigned long bio_flags, u64 offset) | |
6831 | { | |
6832 | int ret; | |
6833 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6834 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6835 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6836 | return 0; |
6837 | } | |
6838 | ||
e65e1535 MX |
6839 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6840 | { | |
6841 | struct btrfs_dio_private *dip = bio->bi_private; | |
6842 | ||
6843 | if (err) { | |
33345d01 | 6844 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6845 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6846 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6847 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6848 | dip->errors = 1; |
6849 | ||
6850 | /* | |
6851 | * before atomic variable goto zero, we must make sure | |
6852 | * dip->errors is perceived to be set. | |
6853 | */ | |
6854 | smp_mb__before_atomic_dec(); | |
6855 | } | |
6856 | ||
6857 | /* if there are more bios still pending for this dio, just exit */ | |
6858 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6859 | goto out; | |
6860 | ||
9be3395b | 6861 | if (dip->errors) { |
e65e1535 | 6862 | bio_io_error(dip->orig_bio); |
9be3395b CM |
6863 | } else { |
6864 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
6865 | bio_endio(dip->orig_bio, 0); |
6866 | } | |
6867 | out: | |
6868 | bio_put(bio); | |
6869 | } | |
6870 | ||
6871 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6872 | u64 first_sector, gfp_t gfp_flags) | |
6873 | { | |
6874 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6875 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6876 | } | |
6877 | ||
6878 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6879 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6880 | int async_submit) |
e65e1535 | 6881 | { |
facc8a22 | 6882 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
6883 | int write = rw & REQ_WRITE; |
6884 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6885 | int ret; | |
6886 | ||
b812ce28 JB |
6887 | if (async_submit) |
6888 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
6889 | ||
e65e1535 | 6890 | bio_get(bio); |
5fd02043 JB |
6891 | |
6892 | if (!write) { | |
6893 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6894 | if (ret) | |
6895 | goto err; | |
6896 | } | |
e65e1535 | 6897 | |
1ae39938 JB |
6898 | if (skip_sum) |
6899 | goto map; | |
6900 | ||
6901 | if (write && async_submit) { | |
e65e1535 MX |
6902 | ret = btrfs_wq_submit_bio(root->fs_info, |
6903 | inode, rw, bio, 0, 0, | |
6904 | file_offset, | |
6905 | __btrfs_submit_bio_start_direct_io, | |
6906 | __btrfs_submit_bio_done); | |
6907 | goto err; | |
1ae39938 JB |
6908 | } else if (write) { |
6909 | /* | |
6910 | * If we aren't doing async submit, calculate the csum of the | |
6911 | * bio now. | |
6912 | */ | |
6913 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6914 | if (ret) | |
6915 | goto err; | |
c2db1073 | 6916 | } else if (!skip_sum) { |
facc8a22 MX |
6917 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio, |
6918 | file_offset); | |
c2db1073 TI |
6919 | if (ret) |
6920 | goto err; | |
6921 | } | |
e65e1535 | 6922 | |
1ae39938 JB |
6923 | map: |
6924 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6925 | err: |
6926 | bio_put(bio); | |
6927 | return ret; | |
6928 | } | |
6929 | ||
6930 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6931 | int skip_sum) | |
6932 | { | |
6933 | struct inode *inode = dip->inode; | |
6934 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
6935 | struct bio *bio; |
6936 | struct bio *orig_bio = dip->orig_bio; | |
6937 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6938 | u64 start_sector = orig_bio->bi_sector; | |
6939 | u64 file_offset = dip->logical_offset; | |
6940 | u64 submit_len = 0; | |
6941 | u64 map_length; | |
6942 | int nr_pages = 0; | |
e65e1535 | 6943 | int ret = 0; |
1ae39938 | 6944 | int async_submit = 0; |
e65e1535 | 6945 | |
e65e1535 | 6946 | map_length = orig_bio->bi_size; |
53b381b3 | 6947 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 MX |
6948 | &map_length, NULL, 0); |
6949 | if (ret) { | |
64728bbb | 6950 | bio_put(orig_bio); |
e65e1535 MX |
6951 | return -EIO; |
6952 | } | |
facc8a22 | 6953 | |
02f57c7a JB |
6954 | if (map_length >= orig_bio->bi_size) { |
6955 | bio = orig_bio; | |
6956 | goto submit; | |
6957 | } | |
6958 | ||
53b381b3 DW |
6959 | /* async crcs make it difficult to collect full stripe writes. */ |
6960 | if (btrfs_get_alloc_profile(root, 1) & | |
6961 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
6962 | async_submit = 0; | |
6963 | else | |
6964 | async_submit = 1; | |
6965 | ||
02f57c7a JB |
6966 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6967 | if (!bio) | |
6968 | return -ENOMEM; | |
6969 | bio->bi_private = dip; | |
6970 | bio->bi_end_io = btrfs_end_dio_bio; | |
6971 | atomic_inc(&dip->pending_bios); | |
6972 | ||
e65e1535 MX |
6973 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6974 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6975 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6976 | bvec->bv_offset) < bvec->bv_len)) { | |
6977 | /* | |
6978 | * inc the count before we submit the bio so | |
6979 | * we know the end IO handler won't happen before | |
6980 | * we inc the count. Otherwise, the dip might get freed | |
6981 | * before we're done setting it up | |
6982 | */ | |
6983 | atomic_inc(&dip->pending_bios); | |
6984 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6985 | file_offset, skip_sum, | |
c329861d | 6986 | async_submit); |
e65e1535 MX |
6987 | if (ret) { |
6988 | bio_put(bio); | |
6989 | atomic_dec(&dip->pending_bios); | |
6990 | goto out_err; | |
6991 | } | |
6992 | ||
e65e1535 MX |
6993 | start_sector += submit_len >> 9; |
6994 | file_offset += submit_len; | |
6995 | ||
6996 | submit_len = 0; | |
6997 | nr_pages = 0; | |
6998 | ||
6999 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7000 | start_sector, GFP_NOFS); | |
7001 | if (!bio) | |
7002 | goto out_err; | |
7003 | bio->bi_private = dip; | |
7004 | bio->bi_end_io = btrfs_end_dio_bio; | |
7005 | ||
7006 | map_length = orig_bio->bi_size; | |
53b381b3 | 7007 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7008 | start_sector << 9, |
e65e1535 MX |
7009 | &map_length, NULL, 0); |
7010 | if (ret) { | |
7011 | bio_put(bio); | |
7012 | goto out_err; | |
7013 | } | |
7014 | } else { | |
7015 | submit_len += bvec->bv_len; | |
7016 | nr_pages ++; | |
7017 | bvec++; | |
7018 | } | |
7019 | } | |
7020 | ||
02f57c7a | 7021 | submit: |
e65e1535 | 7022 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7023 | async_submit); |
e65e1535 MX |
7024 | if (!ret) |
7025 | return 0; | |
7026 | ||
7027 | bio_put(bio); | |
7028 | out_err: | |
7029 | dip->errors = 1; | |
7030 | /* | |
7031 | * before atomic variable goto zero, we must | |
7032 | * make sure dip->errors is perceived to be set. | |
7033 | */ | |
7034 | smp_mb__before_atomic_dec(); | |
7035 | if (atomic_dec_and_test(&dip->pending_bios)) | |
7036 | bio_io_error(dip->orig_bio); | |
7037 | ||
7038 | /* bio_end_io() will handle error, so we needn't return it */ | |
7039 | return 0; | |
7040 | } | |
7041 | ||
9be3395b CM |
7042 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7043 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7044 | { |
7045 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7046 | struct btrfs_dio_private *dip; | |
9be3395b | 7047 | struct bio *io_bio; |
4b46fce2 | 7048 | int skip_sum; |
facc8a22 | 7049 | int sum_len; |
7b6d91da | 7050 | int write = rw & REQ_WRITE; |
4b46fce2 | 7051 | int ret = 0; |
facc8a22 | 7052 | u16 csum_size; |
4b46fce2 JB |
7053 | |
7054 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7055 | ||
9be3395b | 7056 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
7057 | if (!io_bio) { |
7058 | ret = -ENOMEM; | |
7059 | goto free_ordered; | |
7060 | } | |
7061 | ||
facc8a22 MX |
7062 | if (!skip_sum && !write) { |
7063 | csum_size = btrfs_super_csum_size(root->fs_info->super_copy); | |
7064 | sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits; | |
7065 | sum_len *= csum_size; | |
7066 | } else { | |
7067 | sum_len = 0; | |
7068 | } | |
7069 | ||
7070 | dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS); | |
4b46fce2 JB |
7071 | if (!dip) { |
7072 | ret = -ENOMEM; | |
9be3395b | 7073 | goto free_io_bio; |
4b46fce2 | 7074 | } |
4b46fce2 | 7075 | |
9be3395b | 7076 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
7077 | dip->inode = inode; |
7078 | dip->logical_offset = file_offset; | |
e6da5d2e | 7079 | dip->bytes = dio_bio->bi_size; |
9be3395b CM |
7080 | dip->disk_bytenr = (u64)dio_bio->bi_sector << 9; |
7081 | io_bio->bi_private = dip; | |
e65e1535 | 7082 | dip->errors = 0; |
9be3395b CM |
7083 | dip->orig_bio = io_bio; |
7084 | dip->dio_bio = dio_bio; | |
e65e1535 | 7085 | atomic_set(&dip->pending_bios, 0); |
4b46fce2 JB |
7086 | |
7087 | if (write) | |
9be3395b | 7088 | io_bio->bi_end_io = btrfs_endio_direct_write; |
4b46fce2 | 7089 | else |
9be3395b | 7090 | io_bio->bi_end_io = btrfs_endio_direct_read; |
4b46fce2 | 7091 | |
e65e1535 MX |
7092 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7093 | if (!ret) | |
eaf25d93 | 7094 | return; |
9be3395b CM |
7095 | |
7096 | free_io_bio: | |
7097 | bio_put(io_bio); | |
7098 | ||
4b46fce2 JB |
7099 | free_ordered: |
7100 | /* | |
7101 | * If this is a write, we need to clean up the reserved space and kill | |
7102 | * the ordered extent. | |
7103 | */ | |
7104 | if (write) { | |
7105 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7106 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7107 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7108 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7109 | btrfs_free_reserved_extent(root, ordered->start, | |
7110 | ordered->disk_len); | |
7111 | btrfs_put_ordered_extent(ordered); | |
7112 | btrfs_put_ordered_extent(ordered); | |
7113 | } | |
9be3395b | 7114 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7115 | } |
7116 | ||
5a5f79b5 CM |
7117 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
7118 | const struct iovec *iov, loff_t offset, | |
7119 | unsigned long nr_segs) | |
7120 | { | |
7121 | int seg; | |
a1b75f7d | 7122 | int i; |
5a5f79b5 CM |
7123 | size_t size; |
7124 | unsigned long addr; | |
7125 | unsigned blocksize_mask = root->sectorsize - 1; | |
7126 | ssize_t retval = -EINVAL; | |
7127 | loff_t end = offset; | |
7128 | ||
7129 | if (offset & blocksize_mask) | |
7130 | goto out; | |
7131 | ||
7132 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
7133 | for (seg = 0; seg < nr_segs; seg++) { | |
7134 | addr = (unsigned long)iov[seg].iov_base; | |
7135 | size = iov[seg].iov_len; | |
7136 | end += size; | |
a1b75f7d | 7137 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 7138 | goto out; |
a1b75f7d JB |
7139 | |
7140 | /* If this is a write we don't need to check anymore */ | |
7141 | if (rw & WRITE) | |
7142 | continue; | |
7143 | ||
7144 | /* | |
7145 | * Check to make sure we don't have duplicate iov_base's in this | |
7146 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7147 | * when reading back. | |
7148 | */ | |
7149 | for (i = seg + 1; i < nr_segs; i++) { | |
7150 | if (iov[seg].iov_base == iov[i].iov_base) | |
7151 | goto out; | |
7152 | } | |
5a5f79b5 CM |
7153 | } |
7154 | retval = 0; | |
7155 | out: | |
7156 | return retval; | |
7157 | } | |
eb838e73 | 7158 | |
16432985 CM |
7159 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
7160 | const struct iovec *iov, loff_t offset, | |
7161 | unsigned long nr_segs) | |
7162 | { | |
4b46fce2 JB |
7163 | struct file *file = iocb->ki_filp; |
7164 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7165 | size_t count = 0; |
2e60a51e | 7166 | int flags = 0; |
38851cc1 MX |
7167 | bool wakeup = true; |
7168 | bool relock = false; | |
0934856d | 7169 | ssize_t ret; |
4b46fce2 | 7170 | |
5a5f79b5 | 7171 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 7172 | offset, nr_segs)) |
5a5f79b5 | 7173 | return 0; |
3f7c579c | 7174 | |
38851cc1 MX |
7175 | atomic_inc(&inode->i_dio_count); |
7176 | smp_mb__after_atomic_inc(); | |
7177 | ||
0e267c44 JB |
7178 | /* |
7179 | * The generic stuff only does filemap_write_and_wait_range, which isn't | |
7180 | * enough if we've written compressed pages to this area, so we need to | |
7181 | * call btrfs_wait_ordered_range to make absolutely sure that any | |
7182 | * outstanding dirty pages are on disk. | |
7183 | */ | |
7184 | count = iov_length(iov, nr_segs); | |
7185 | btrfs_wait_ordered_range(inode, offset, count); | |
7186 | ||
0934856d | 7187 | if (rw & WRITE) { |
38851cc1 MX |
7188 | /* |
7189 | * If the write DIO is beyond the EOF, we need update | |
7190 | * the isize, but it is protected by i_mutex. So we can | |
7191 | * not unlock the i_mutex at this case. | |
7192 | */ | |
7193 | if (offset + count <= inode->i_size) { | |
7194 | mutex_unlock(&inode->i_mutex); | |
7195 | relock = true; | |
7196 | } | |
0934856d MX |
7197 | ret = btrfs_delalloc_reserve_space(inode, count); |
7198 | if (ret) | |
38851cc1 MX |
7199 | goto out; |
7200 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
7201 | &BTRFS_I(inode)->runtime_flags))) { | |
7202 | inode_dio_done(inode); | |
7203 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7204 | wakeup = false; | |
0934856d MX |
7205 | } |
7206 | ||
7207 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
7208 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
7209 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
2e60a51e | 7210 | btrfs_submit_direct, flags); |
0934856d MX |
7211 | if (rw & WRITE) { |
7212 | if (ret < 0 && ret != -EIOCBQUEUED) | |
7213 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 7214 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
7215 | btrfs_delalloc_release_space(inode, |
7216 | count - (size_t)ret); | |
172a5049 MX |
7217 | else |
7218 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 7219 | } |
38851cc1 | 7220 | out: |
2e60a51e MX |
7221 | if (wakeup) |
7222 | inode_dio_done(inode); | |
38851cc1 MX |
7223 | if (relock) |
7224 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
7225 | |
7226 | return ret; | |
16432985 CM |
7227 | } |
7228 | ||
05dadc09 TI |
7229 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
7230 | ||
1506fcc8 YS |
7231 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
7232 | __u64 start, __u64 len) | |
7233 | { | |
05dadc09 TI |
7234 | int ret; |
7235 | ||
7236 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
7237 | if (ret) | |
7238 | return ret; | |
7239 | ||
ec29ed5b | 7240 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
7241 | } |
7242 | ||
a52d9a80 | 7243 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7244 | { |
d1310b2e CM |
7245 | struct extent_io_tree *tree; |
7246 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 7247 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 7248 | } |
1832a6d5 | 7249 | |
a52d9a80 | 7250 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7251 | { |
d1310b2e | 7252 | struct extent_io_tree *tree; |
b888db2b CM |
7253 | |
7254 | ||
7255 | if (current->flags & PF_MEMALLOC) { | |
7256 | redirty_page_for_writepage(wbc, page); | |
7257 | unlock_page(page); | |
7258 | return 0; | |
7259 | } | |
d1310b2e | 7260 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 7261 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
7262 | } |
7263 | ||
48a3b636 ES |
7264 | static int btrfs_writepages(struct address_space *mapping, |
7265 | struct writeback_control *wbc) | |
b293f02e | 7266 | { |
d1310b2e | 7267 | struct extent_io_tree *tree; |
771ed689 | 7268 | |
d1310b2e | 7269 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
7270 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
7271 | } | |
7272 | ||
3ab2fb5a CM |
7273 | static int |
7274 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
7275 | struct list_head *pages, unsigned nr_pages) | |
7276 | { | |
d1310b2e CM |
7277 | struct extent_io_tree *tree; |
7278 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
7279 | return extent_readpages(tree, mapping, pages, nr_pages, |
7280 | btrfs_get_extent); | |
7281 | } | |
e6dcd2dc | 7282 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 7283 | { |
d1310b2e CM |
7284 | struct extent_io_tree *tree; |
7285 | struct extent_map_tree *map; | |
a52d9a80 | 7286 | int ret; |
8c2383c3 | 7287 | |
d1310b2e CM |
7288 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
7289 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 7290 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
7291 | if (ret == 1) { |
7292 | ClearPagePrivate(page); | |
7293 | set_page_private(page, 0); | |
7294 | page_cache_release(page); | |
39279cc3 | 7295 | } |
a52d9a80 | 7296 | return ret; |
39279cc3 CM |
7297 | } |
7298 | ||
e6dcd2dc CM |
7299 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
7300 | { | |
98509cfc CM |
7301 | if (PageWriteback(page) || PageDirty(page)) |
7302 | return 0; | |
b335b003 | 7303 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
7304 | } |
7305 | ||
d47992f8 LC |
7306 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
7307 | unsigned int length) | |
39279cc3 | 7308 | { |
5fd02043 | 7309 | struct inode *inode = page->mapping->host; |
d1310b2e | 7310 | struct extent_io_tree *tree; |
e6dcd2dc | 7311 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 7312 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7313 | u64 page_start = page_offset(page); |
7314 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 7315 | |
8b62b72b CM |
7316 | /* |
7317 | * we have the page locked, so new writeback can't start, | |
7318 | * and the dirty bit won't be cleared while we are here. | |
7319 | * | |
7320 | * Wait for IO on this page so that we can safely clear | |
7321 | * the PagePrivate2 bit and do ordered accounting | |
7322 | */ | |
e6dcd2dc | 7323 | wait_on_page_writeback(page); |
8b62b72b | 7324 | |
5fd02043 | 7325 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
7326 | if (offset) { |
7327 | btrfs_releasepage(page, GFP_NOFS); | |
7328 | return; | |
7329 | } | |
d0082371 | 7330 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
4eee4fa4 | 7331 | ordered = btrfs_lookup_ordered_extent(inode, page_offset(page)); |
e6dcd2dc | 7332 | if (ordered) { |
eb84ae03 CM |
7333 | /* |
7334 | * IO on this page will never be started, so we need | |
7335 | * to account for any ordered extents now | |
7336 | */ | |
e6dcd2dc CM |
7337 | clear_extent_bit(tree, page_start, page_end, |
7338 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
7339 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
7340 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
7341 | /* |
7342 | * whoever cleared the private bit is responsible | |
7343 | * for the finish_ordered_io | |
7344 | */ | |
5fd02043 JB |
7345 | if (TestClearPagePrivate2(page) && |
7346 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
7347 | PAGE_CACHE_SIZE, 1)) { | |
7348 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 7349 | } |
e6dcd2dc | 7350 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 7351 | cached_state = NULL; |
d0082371 | 7352 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7353 | } |
7354 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 7355 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
7356 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
7357 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
7358 | __btrfs_releasepage(page, GFP_NOFS); |
7359 | ||
4a096752 | 7360 | ClearPageChecked(page); |
9ad6b7bc | 7361 | if (PagePrivate(page)) { |
9ad6b7bc CM |
7362 | ClearPagePrivate(page); |
7363 | set_page_private(page, 0); | |
7364 | page_cache_release(page); | |
7365 | } | |
39279cc3 CM |
7366 | } |
7367 | ||
9ebefb18 CM |
7368 | /* |
7369 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
7370 | * called from a page fault handler when a page is first dirtied. Hence we must | |
7371 | * be careful to check for EOF conditions here. We set the page up correctly | |
7372 | * for a written page which means we get ENOSPC checking when writing into | |
7373 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
7374 | * support these features. | |
7375 | * | |
7376 | * We are not allowed to take the i_mutex here so we have to play games to | |
7377 | * protect against truncate races as the page could now be beyond EOF. Because | |
7378 | * vmtruncate() writes the inode size before removing pages, once we have the | |
7379 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
7380 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
7381 | * unlock the page. | |
7382 | */ | |
c2ec175c | 7383 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 7384 | { |
c2ec175c | 7385 | struct page *page = vmf->page; |
496ad9aa | 7386 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 7387 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
7388 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7389 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 7390 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7391 | char *kaddr; |
7392 | unsigned long zero_start; | |
9ebefb18 | 7393 | loff_t size; |
1832a6d5 | 7394 | int ret; |
9998eb70 | 7395 | int reserved = 0; |
a52d9a80 | 7396 | u64 page_start; |
e6dcd2dc | 7397 | u64 page_end; |
9ebefb18 | 7398 | |
b2b5ef5c | 7399 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 7400 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7401 | if (!ret) { |
e41f941a | 7402 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
7403 | reserved = 1; |
7404 | } | |
56a76f82 NP |
7405 | if (ret) { |
7406 | if (ret == -ENOMEM) | |
7407 | ret = VM_FAULT_OOM; | |
7408 | else /* -ENOSPC, -EIO, etc */ | |
7409 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
7410 | if (reserved) |
7411 | goto out; | |
7412 | goto out_noreserve; | |
56a76f82 | 7413 | } |
1832a6d5 | 7414 | |
56a76f82 | 7415 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 7416 | again: |
9ebefb18 | 7417 | lock_page(page); |
9ebefb18 | 7418 | size = i_size_read(inode); |
e6dcd2dc CM |
7419 | page_start = page_offset(page); |
7420 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 7421 | |
9ebefb18 | 7422 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 7423 | (page_start >= size)) { |
9ebefb18 CM |
7424 | /* page got truncated out from underneath us */ |
7425 | goto out_unlock; | |
7426 | } | |
e6dcd2dc CM |
7427 | wait_on_page_writeback(page); |
7428 | ||
d0082371 | 7429 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7430 | set_page_extent_mapped(page); |
7431 | ||
eb84ae03 CM |
7432 | /* |
7433 | * we can't set the delalloc bits if there are pending ordered | |
7434 | * extents. Drop our locks and wait for them to finish | |
7435 | */ | |
e6dcd2dc CM |
7436 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
7437 | if (ordered) { | |
2ac55d41 JB |
7438 | unlock_extent_cached(io_tree, page_start, page_end, |
7439 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 7440 | unlock_page(page); |
eb84ae03 | 7441 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
7442 | btrfs_put_ordered_extent(ordered); |
7443 | goto again; | |
7444 | } | |
7445 | ||
fbf19087 JB |
7446 | /* |
7447 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
7448 | * if it was already dirty, so for space accounting reasons we need to | |
7449 | * clear any delalloc bits for the range we are fixing to save. There | |
7450 | * is probably a better way to do this, but for now keep consistent with | |
7451 | * prepare_pages in the normal write path. | |
7452 | */ | |
2ac55d41 | 7453 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
7454 | EXTENT_DIRTY | EXTENT_DELALLOC | |
7455 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 7456 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 7457 | |
2ac55d41 JB |
7458 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
7459 | &cached_state); | |
9ed74f2d | 7460 | if (ret) { |
2ac55d41 JB |
7461 | unlock_extent_cached(io_tree, page_start, page_end, |
7462 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
7463 | ret = VM_FAULT_SIGBUS; |
7464 | goto out_unlock; | |
7465 | } | |
e6dcd2dc | 7466 | ret = 0; |
9ebefb18 CM |
7467 | |
7468 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 7469 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 7470 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 7471 | else |
e6dcd2dc | 7472 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 7473 | |
e6dcd2dc CM |
7474 | if (zero_start != PAGE_CACHE_SIZE) { |
7475 | kaddr = kmap(page); | |
7476 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
7477 | flush_dcache_page(page); | |
7478 | kunmap(page); | |
7479 | } | |
247e743c | 7480 | ClearPageChecked(page); |
e6dcd2dc | 7481 | set_page_dirty(page); |
50a9b214 | 7482 | SetPageUptodate(page); |
5a3f23d5 | 7483 | |
257c62e1 CM |
7484 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
7485 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 7486 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 7487 | |
2ac55d41 | 7488 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
7489 | |
7490 | out_unlock: | |
b2b5ef5c JK |
7491 | if (!ret) { |
7492 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 7493 | return VM_FAULT_LOCKED; |
b2b5ef5c | 7494 | } |
9ebefb18 | 7495 | unlock_page(page); |
1832a6d5 | 7496 | out: |
ec39e180 | 7497 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7498 | out_noreserve: |
b2b5ef5c | 7499 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
7500 | return ret; |
7501 | } | |
7502 | ||
a41ad394 | 7503 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
7504 | { |
7505 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 7506 | struct btrfs_block_rsv *rsv; |
a71754fc | 7507 | int ret = 0; |
3893e33b | 7508 | int err = 0; |
39279cc3 | 7509 | struct btrfs_trans_handle *trans; |
dbe674a9 | 7510 | u64 mask = root->sectorsize - 1; |
07127184 | 7511 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 7512 | |
4a096752 | 7513 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 7514 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 7515 | |
fcb80c2a JB |
7516 | /* |
7517 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
7518 | * 3 things going on here | |
7519 | * | |
7520 | * 1) We need to reserve space for our orphan item and the space to | |
7521 | * delete our orphan item. Lord knows we don't want to have a dangling | |
7522 | * orphan item because we didn't reserve space to remove it. | |
7523 | * | |
7524 | * 2) We need to reserve space to update our inode. | |
7525 | * | |
7526 | * 3) We need to have something to cache all the space that is going to | |
7527 | * be free'd up by the truncate operation, but also have some slack | |
7528 | * space reserved in case it uses space during the truncate (thank you | |
7529 | * very much snapshotting). | |
7530 | * | |
7531 | * And we need these to all be seperate. The fact is we can use alot of | |
7532 | * space doing the truncate, and we have no earthly idea how much space | |
7533 | * we will use, so we need the truncate reservation to be seperate so it | |
7534 | * doesn't end up using space reserved for updating the inode or | |
7535 | * removing the orphan item. We also need to be able to stop the | |
7536 | * transaction and start a new one, which means we need to be able to | |
7537 | * update the inode several times, and we have no idea of knowing how | |
7538 | * many times that will be, so we can't just reserve 1 item for the | |
7539 | * entirety of the opration, so that has to be done seperately as well. | |
7540 | * Then there is the orphan item, which does indeed need to be held on | |
7541 | * to for the whole operation, and we need nobody to touch this reserved | |
7542 | * space except the orphan code. | |
7543 | * | |
7544 | * So that leaves us with | |
7545 | * | |
7546 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7547 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7548 | * transaction reservation. | |
7549 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7550 | * updating the inode. | |
7551 | */ | |
66d8f3dd | 7552 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7553 | if (!rsv) |
7554 | return -ENOMEM; | |
4a338542 | 7555 | rsv->size = min_size; |
ca7e70f5 | 7556 | rsv->failfast = 1; |
f0cd846e | 7557 | |
907cbceb | 7558 | /* |
07127184 | 7559 | * 1 for the truncate slack space |
907cbceb JB |
7560 | * 1 for updating the inode. |
7561 | */ | |
f3fe820c | 7562 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7563 | if (IS_ERR(trans)) { |
7564 | err = PTR_ERR(trans); | |
7565 | goto out; | |
7566 | } | |
f0cd846e | 7567 | |
907cbceb JB |
7568 | /* Migrate the slack space for the truncate to our reserve */ |
7569 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7570 | min_size); | |
fcb80c2a | 7571 | BUG_ON(ret); |
f0cd846e | 7572 | |
5a3f23d5 CM |
7573 | /* |
7574 | * setattr is responsible for setting the ordered_data_close flag, | |
7575 | * but that is only tested during the last file release. That | |
7576 | * could happen well after the next commit, leaving a great big | |
7577 | * window where new writes may get lost if someone chooses to write | |
7578 | * to this file after truncating to zero | |
7579 | * | |
7580 | * The inode doesn't have any dirty data here, and so if we commit | |
7581 | * this is a noop. If someone immediately starts writing to the inode | |
7582 | * it is very likely we'll catch some of their writes in this | |
7583 | * transaction, and the commit will find this file on the ordered | |
7584 | * data list with good things to send down. | |
7585 | * | |
7586 | * This is a best effort solution, there is still a window where | |
7587 | * using truncate to replace the contents of the file will | |
7588 | * end up with a zero length file after a crash. | |
7589 | */ | |
72ac3c0d JB |
7590 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7591 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7592 | btrfs_add_ordered_operation(trans, root, inode); |
7593 | ||
5dc562c5 JB |
7594 | /* |
7595 | * So if we truncate and then write and fsync we normally would just | |
7596 | * write the extents that changed, which is a problem if we need to | |
7597 | * first truncate that entire inode. So set this flag so we write out | |
7598 | * all of the extents in the inode to the sync log so we're completely | |
7599 | * safe. | |
7600 | */ | |
7601 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7602 | trans->block_rsv = rsv; |
907cbceb | 7603 | |
8082510e YZ |
7604 | while (1) { |
7605 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7606 | inode->i_size, | |
7607 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7608 | if (ret != -ENOSPC) { |
3893e33b | 7609 | err = ret; |
8082510e | 7610 | break; |
3893e33b | 7611 | } |
39279cc3 | 7612 | |
fcb80c2a | 7613 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7614 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7615 | if (ret) { |
7616 | err = ret; | |
7617 | break; | |
7618 | } | |
ca7e70f5 | 7619 | |
8082510e | 7620 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7621 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7622 | |
7623 | trans = btrfs_start_transaction(root, 2); | |
7624 | if (IS_ERR(trans)) { | |
7625 | ret = err = PTR_ERR(trans); | |
7626 | trans = NULL; | |
7627 | break; | |
7628 | } | |
7629 | ||
7630 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7631 | rsv, min_size); | |
7632 | BUG_ON(ret); /* shouldn't happen */ | |
7633 | trans->block_rsv = rsv; | |
8082510e YZ |
7634 | } |
7635 | ||
7636 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7637 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7638 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7639 | if (ret) |
7640 | err = ret; | |
8082510e YZ |
7641 | } |
7642 | ||
917c16b2 CM |
7643 | if (trans) { |
7644 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7645 | ret = btrfs_update_inode(trans, root, inode); | |
7646 | if (ret && !err) | |
7647 | err = ret; | |
7b128766 | 7648 | |
7ad85bb7 | 7649 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7650 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7651 | } |
fcb80c2a JB |
7652 | |
7653 | out: | |
7654 | btrfs_free_block_rsv(root, rsv); | |
7655 | ||
3893e33b JB |
7656 | if (ret && !err) |
7657 | err = ret; | |
a41ad394 | 7658 | |
3893e33b | 7659 | return err; |
39279cc3 CM |
7660 | } |
7661 | ||
d352ac68 CM |
7662 | /* |
7663 | * create a new subvolume directory/inode (helper for the ioctl). | |
7664 | */ | |
d2fb3437 | 7665 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7666 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7667 | { |
39279cc3 | 7668 | struct inode *inode; |
76dda93c | 7669 | int err; |
00e4e6b3 | 7670 | u64 index = 0; |
39279cc3 | 7671 | |
12fc9d09 FA |
7672 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7673 | new_dirid, new_dirid, | |
7674 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7675 | &index); | |
54aa1f4d | 7676 | if (IS_ERR(inode)) |
f46b5a66 | 7677 | return PTR_ERR(inode); |
39279cc3 CM |
7678 | inode->i_op = &btrfs_dir_inode_operations; |
7679 | inode->i_fop = &btrfs_dir_file_operations; | |
7680 | ||
bfe86848 | 7681 | set_nlink(inode, 1); |
dbe674a9 | 7682 | btrfs_i_size_write(inode, 0); |
3b96362c | 7683 | |
76dda93c | 7684 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7685 | |
76dda93c | 7686 | iput(inode); |
ce598979 | 7687 | return err; |
39279cc3 CM |
7688 | } |
7689 | ||
39279cc3 CM |
7690 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7691 | { | |
7692 | struct btrfs_inode *ei; | |
2ead6ae7 | 7693 | struct inode *inode; |
39279cc3 CM |
7694 | |
7695 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7696 | if (!ei) | |
7697 | return NULL; | |
2ead6ae7 YZ |
7698 | |
7699 | ei->root = NULL; | |
2ead6ae7 | 7700 | ei->generation = 0; |
15ee9bc7 | 7701 | ei->last_trans = 0; |
257c62e1 | 7702 | ei->last_sub_trans = 0; |
e02119d5 | 7703 | ei->logged_trans = 0; |
2ead6ae7 | 7704 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7705 | ei->disk_i_size = 0; |
7706 | ei->flags = 0; | |
7709cde3 | 7707 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7708 | ei->index_cnt = (u64)-1; |
7709 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7710 | ei->last_log_commit = 0; |
2ead6ae7 | 7711 | |
9e0baf60 JB |
7712 | spin_lock_init(&ei->lock); |
7713 | ei->outstanding_extents = 0; | |
7714 | ei->reserved_extents = 0; | |
2ead6ae7 | 7715 | |
72ac3c0d | 7716 | ei->runtime_flags = 0; |
261507a0 | 7717 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7718 | |
16cdcec7 MX |
7719 | ei->delayed_node = NULL; |
7720 | ||
2ead6ae7 | 7721 | inode = &ei->vfs_inode; |
a8067e02 | 7722 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7723 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7724 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7725 | ei->io_tree.track_uptodate = 1; |
7726 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7727 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7728 | mutex_init(&ei->log_mutex); |
f248679e | 7729 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7730 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7731 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7732 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7733 | RB_CLEAR_NODE(&ei->rb_node); |
7734 | ||
7735 | return inode; | |
39279cc3 CM |
7736 | } |
7737 | ||
fa0d7e3d NP |
7738 | static void btrfs_i_callback(struct rcu_head *head) |
7739 | { | |
7740 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7741 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7742 | } | |
7743 | ||
39279cc3 CM |
7744 | void btrfs_destroy_inode(struct inode *inode) |
7745 | { | |
e6dcd2dc | 7746 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7747 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7748 | ||
b3d9b7a3 | 7749 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7750 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7751 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7752 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7753 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7754 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7755 | |
a6dbd429 JB |
7756 | /* |
7757 | * This can happen where we create an inode, but somebody else also | |
7758 | * created the same inode and we need to destroy the one we already | |
7759 | * created. | |
7760 | */ | |
7761 | if (!root) | |
7762 | goto free; | |
7763 | ||
5a3f23d5 CM |
7764 | /* |
7765 | * Make sure we're properly removed from the ordered operation | |
7766 | * lists. | |
7767 | */ | |
7768 | smp_mb(); | |
7769 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
199c2a9c | 7770 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 7771 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
199c2a9c | 7772 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
7773 | } |
7774 | ||
8a35d95f JB |
7775 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7776 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb SK |
7777 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
7778 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7779 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7780 | } |
7b128766 | 7781 | |
d397712b | 7782 | while (1) { |
e6dcd2dc CM |
7783 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7784 | if (!ordered) | |
7785 | break; | |
7786 | else { | |
c2cf52eb SK |
7787 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
7788 | (unsigned long long)ordered->file_offset, | |
7789 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7790 | btrfs_remove_ordered_extent(inode, ordered); |
7791 | btrfs_put_ordered_extent(ordered); | |
7792 | btrfs_put_ordered_extent(ordered); | |
7793 | } | |
7794 | } | |
5d4f98a2 | 7795 | inode_tree_del(inode); |
5b21f2ed | 7796 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7797 | free: |
fa0d7e3d | 7798 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7799 | } |
7800 | ||
45321ac5 | 7801 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7802 | { |
7803 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7804 | |
6379ef9f NA |
7805 | if (root == NULL) |
7806 | return 1; | |
7807 | ||
fa6ac876 | 7808 | /* the snap/subvol tree is on deleting */ |
0af3d00b | 7809 | if (btrfs_root_refs(&root->root_item) == 0 && |
fa6ac876 | 7810 | root != root->fs_info->tree_root) |
45321ac5 | 7811 | return 1; |
76dda93c | 7812 | else |
45321ac5 | 7813 | return generic_drop_inode(inode); |
76dda93c YZ |
7814 | } |
7815 | ||
0ee0fda0 | 7816 | static void init_once(void *foo) |
39279cc3 CM |
7817 | { |
7818 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7819 | ||
7820 | inode_init_once(&ei->vfs_inode); | |
7821 | } | |
7822 | ||
7823 | void btrfs_destroy_cachep(void) | |
7824 | { | |
8c0a8537 KS |
7825 | /* |
7826 | * Make sure all delayed rcu free inodes are flushed before we | |
7827 | * destroy cache. | |
7828 | */ | |
7829 | rcu_barrier(); | |
39279cc3 CM |
7830 | if (btrfs_inode_cachep) |
7831 | kmem_cache_destroy(btrfs_inode_cachep); | |
7832 | if (btrfs_trans_handle_cachep) | |
7833 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7834 | if (btrfs_transaction_cachep) | |
7835 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7836 | if (btrfs_path_cachep) |
7837 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7838 | if (btrfs_free_space_cachep) |
7839 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7840 | if (btrfs_delalloc_work_cachep) |
7841 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7842 | } |
7843 | ||
7844 | int btrfs_init_cachep(void) | |
7845 | { | |
837e1972 | 7846 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7847 | sizeof(struct btrfs_inode), 0, |
7848 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7849 | if (!btrfs_inode_cachep) |
7850 | goto fail; | |
9601e3f6 | 7851 | |
837e1972 | 7852 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7853 | sizeof(struct btrfs_trans_handle), 0, |
7854 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7855 | if (!btrfs_trans_handle_cachep) |
7856 | goto fail; | |
9601e3f6 | 7857 | |
837e1972 | 7858 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7859 | sizeof(struct btrfs_transaction), 0, |
7860 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7861 | if (!btrfs_transaction_cachep) |
7862 | goto fail; | |
9601e3f6 | 7863 | |
837e1972 | 7864 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7865 | sizeof(struct btrfs_path), 0, |
7866 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7867 | if (!btrfs_path_cachep) |
7868 | goto fail; | |
9601e3f6 | 7869 | |
837e1972 | 7870 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7871 | sizeof(struct btrfs_free_space), 0, |
7872 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7873 | if (!btrfs_free_space_cachep) | |
7874 | goto fail; | |
7875 | ||
8ccf6f19 MX |
7876 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7877 | sizeof(struct btrfs_delalloc_work), 0, | |
7878 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7879 | NULL); | |
7880 | if (!btrfs_delalloc_work_cachep) | |
7881 | goto fail; | |
7882 | ||
39279cc3 CM |
7883 | return 0; |
7884 | fail: | |
7885 | btrfs_destroy_cachep(); | |
7886 | return -ENOMEM; | |
7887 | } | |
7888 | ||
7889 | static int btrfs_getattr(struct vfsmount *mnt, | |
7890 | struct dentry *dentry, struct kstat *stat) | |
7891 | { | |
df0af1a5 | 7892 | u64 delalloc_bytes; |
39279cc3 | 7893 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
7894 | u32 blocksize = inode->i_sb->s_blocksize; |
7895 | ||
39279cc3 | 7896 | generic_fillattr(inode, stat); |
0ee5dc67 | 7897 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7898 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
7899 | |
7900 | spin_lock(&BTRFS_I(inode)->lock); | |
7901 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
7902 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 7903 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 7904 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
7905 | return 0; |
7906 | } | |
7907 | ||
d397712b CM |
7908 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7909 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7910 | { |
7911 | struct btrfs_trans_handle *trans; | |
7912 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7913 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7914 | struct inode *new_inode = new_dentry->d_inode; |
7915 | struct inode *old_inode = old_dentry->d_inode; | |
7916 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7917 | u64 index = 0; |
4df27c4d | 7918 | u64 root_objectid; |
39279cc3 | 7919 | int ret; |
33345d01 | 7920 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7921 | |
33345d01 | 7922 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7923 | return -EPERM; |
7924 | ||
4df27c4d | 7925 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7926 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7927 | return -EXDEV; |
7928 | ||
33345d01 LZ |
7929 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7930 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7931 | return -ENOTEMPTY; |
5f39d397 | 7932 | |
4df27c4d YZ |
7933 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7934 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7935 | return -ENOTEMPTY; | |
9c52057c CM |
7936 | |
7937 | ||
7938 | /* check for collisions, even if the name isn't there */ | |
7939 | ret = btrfs_check_dir_item_collision(root, new_dir->i_ino, | |
7940 | new_dentry->d_name.name, | |
7941 | new_dentry->d_name.len); | |
7942 | ||
7943 | if (ret) { | |
7944 | if (ret == -EEXIST) { | |
7945 | /* we shouldn't get | |
7946 | * eexist without a new_inode */ | |
7947 | if (!new_inode) { | |
7948 | WARN_ON(1); | |
7949 | return ret; | |
7950 | } | |
7951 | } else { | |
7952 | /* maybe -EOVERFLOW */ | |
7953 | return ret; | |
7954 | } | |
7955 | } | |
7956 | ret = 0; | |
7957 | ||
5a3f23d5 CM |
7958 | /* |
7959 | * we're using rename to replace one file with another. | |
7960 | * and the replacement file is large. Start IO on it now so | |
7961 | * we don't add too much work to the end of the transaction | |
7962 | */ | |
4baf8c92 | 7963 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7964 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7965 | filemap_flush(old_inode->i_mapping); | |
7966 | ||
76dda93c | 7967 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7968 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7969 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7970 | /* |
7971 | * We want to reserve the absolute worst case amount of items. So if | |
7972 | * both inodes are subvols and we need to unlink them then that would | |
7973 | * require 4 item modifications, but if they are both normal inodes it | |
7974 | * would require 5 item modifications, so we'll assume their normal | |
7975 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7976 | * should cover the worst case number of items we'll modify. | |
7977 | */ | |
6e137ed3 | 7978 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
7979 | if (IS_ERR(trans)) { |
7980 | ret = PTR_ERR(trans); | |
7981 | goto out_notrans; | |
7982 | } | |
76dda93c | 7983 | |
4df27c4d YZ |
7984 | if (dest != root) |
7985 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7986 | |
a5719521 YZ |
7987 | ret = btrfs_set_inode_index(new_dir, &index); |
7988 | if (ret) | |
7989 | goto out_fail; | |
5a3f23d5 | 7990 | |
33345d01 | 7991 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7992 | /* force full log commit if subvolume involved. */ |
7993 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7994 | } else { | |
a5719521 YZ |
7995 | ret = btrfs_insert_inode_ref(trans, dest, |
7996 | new_dentry->d_name.name, | |
7997 | new_dentry->d_name.len, | |
33345d01 LZ |
7998 | old_ino, |
7999 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8000 | if (ret) |
8001 | goto out_fail; | |
4df27c4d YZ |
8002 | /* |
8003 | * this is an ugly little race, but the rename is required | |
8004 | * to make sure that if we crash, the inode is either at the | |
8005 | * old name or the new one. pinning the log transaction lets | |
8006 | * us make sure we don't allow a log commit to come in after | |
8007 | * we unlink the name but before we add the new name back in. | |
8008 | */ | |
8009 | btrfs_pin_log_trans(root); | |
8010 | } | |
5a3f23d5 CM |
8011 | /* |
8012 | * make sure the inode gets flushed if it is replacing | |
8013 | * something. | |
8014 | */ | |
33345d01 | 8015 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 8016 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 8017 | |
0c4d2d95 JB |
8018 | inode_inc_iversion(old_dir); |
8019 | inode_inc_iversion(new_dir); | |
8020 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8021 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8022 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8023 | old_inode->i_ctime = ctime; | |
5f39d397 | 8024 | |
12fcfd22 CM |
8025 | if (old_dentry->d_parent != new_dentry->d_parent) |
8026 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8027 | ||
33345d01 | 8028 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8029 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8030 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8031 | old_dentry->d_name.name, | |
8032 | old_dentry->d_name.len); | |
8033 | } else { | |
92986796 AV |
8034 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8035 | old_dentry->d_inode, | |
8036 | old_dentry->d_name.name, | |
8037 | old_dentry->d_name.len); | |
8038 | if (!ret) | |
8039 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8040 | } |
79787eaa JM |
8041 | if (ret) { |
8042 | btrfs_abort_transaction(trans, root, ret); | |
8043 | goto out_fail; | |
8044 | } | |
39279cc3 CM |
8045 | |
8046 | if (new_inode) { | |
0c4d2d95 | 8047 | inode_inc_iversion(new_inode); |
39279cc3 | 8048 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8049 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8050 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8051 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8052 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8053 | root_objectid, | |
8054 | new_dentry->d_name.name, | |
8055 | new_dentry->d_name.len); | |
8056 | BUG_ON(new_inode->i_nlink == 0); | |
8057 | } else { | |
8058 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8059 | new_dentry->d_inode, | |
8060 | new_dentry->d_name.name, | |
8061 | new_dentry->d_name.len); | |
8062 | } | |
4ef31a45 | 8063 | if (!ret && new_inode->i_nlink == 0) |
e02119d5 | 8064 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
79787eaa JM |
8065 | if (ret) { |
8066 | btrfs_abort_transaction(trans, root, ret); | |
8067 | goto out_fail; | |
8068 | } | |
39279cc3 | 8069 | } |
aec7477b | 8070 | |
4df27c4d YZ |
8071 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8072 | new_dentry->d_name.name, | |
a5719521 | 8073 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8074 | if (ret) { |
8075 | btrfs_abort_transaction(trans, root, ret); | |
8076 | goto out_fail; | |
8077 | } | |
39279cc3 | 8078 | |
33345d01 | 8079 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8080 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8081 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8082 | btrfs_end_log_trans(root); |
8083 | } | |
39279cc3 | 8084 | out_fail: |
7ad85bb7 | 8085 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8086 | out_notrans: |
33345d01 | 8087 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8088 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8089 | |
39279cc3 CM |
8090 | return ret; |
8091 | } | |
8092 | ||
8ccf6f19 MX |
8093 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8094 | { | |
8095 | struct btrfs_delalloc_work *delalloc_work; | |
8096 | ||
8097 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8098 | work); | |
8099 | if (delalloc_work->wait) | |
8100 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
8101 | else | |
8102 | filemap_flush(delalloc_work->inode->i_mapping); | |
8103 | ||
8104 | if (delalloc_work->delay_iput) | |
8105 | btrfs_add_delayed_iput(delalloc_work->inode); | |
8106 | else | |
8107 | iput(delalloc_work->inode); | |
8108 | complete(&delalloc_work->completion); | |
8109 | } | |
8110 | ||
8111 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8112 | int wait, int delay_iput) | |
8113 | { | |
8114 | struct btrfs_delalloc_work *work; | |
8115 | ||
8116 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8117 | if (!work) | |
8118 | return NULL; | |
8119 | ||
8120 | init_completion(&work->completion); | |
8121 | INIT_LIST_HEAD(&work->list); | |
8122 | work->inode = inode; | |
8123 | work->wait = wait; | |
8124 | work->delay_iput = delay_iput; | |
8125 | work->work.func = btrfs_run_delalloc_work; | |
8126 | ||
8127 | return work; | |
8128 | } | |
8129 | ||
8130 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8131 | { | |
8132 | wait_for_completion(&work->completion); | |
8133 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8134 | } | |
8135 | ||
d352ac68 CM |
8136 | /* |
8137 | * some fairly slow code that needs optimization. This walks the list | |
8138 | * of all the inodes with pending delalloc and forces them to disk. | |
8139 | */ | |
eb73c1b7 | 8140 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 8141 | { |
ea8c2819 | 8142 | struct btrfs_inode *binode; |
5b21f2ed | 8143 | struct inode *inode; |
8ccf6f19 MX |
8144 | struct btrfs_delalloc_work *work, *next; |
8145 | struct list_head works; | |
1eafa6c7 | 8146 | struct list_head splice; |
8ccf6f19 | 8147 | int ret = 0; |
ea8c2819 | 8148 | |
8ccf6f19 | 8149 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8150 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8151 | |
eb73c1b7 MX |
8152 | spin_lock(&root->delalloc_lock); |
8153 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8154 | while (!list_empty(&splice)) { |
8155 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8156 | delalloc_inodes); |
1eafa6c7 | 8157 | |
eb73c1b7 MX |
8158 | list_move_tail(&binode->delalloc_inodes, |
8159 | &root->delalloc_inodes); | |
5b21f2ed | 8160 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8161 | if (!inode) { |
eb73c1b7 | 8162 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8163 | continue; |
df0af1a5 | 8164 | } |
eb73c1b7 | 8165 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8166 | |
8167 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
8168 | if (unlikely(!work)) { | |
8169 | ret = -ENOMEM; | |
8170 | goto out; | |
5b21f2ed | 8171 | } |
1eafa6c7 MX |
8172 | list_add_tail(&work->list, &works); |
8173 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
8174 | &work->work); | |
8175 | ||
5b21f2ed | 8176 | cond_resched(); |
eb73c1b7 | 8177 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 8178 | } |
eb73c1b7 | 8179 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 8180 | |
1eafa6c7 MX |
8181 | list_for_each_entry_safe(work, next, &works, list) { |
8182 | list_del_init(&work->list); | |
8183 | btrfs_wait_and_free_delalloc_work(work); | |
8184 | } | |
eb73c1b7 MX |
8185 | return 0; |
8186 | out: | |
8187 | list_for_each_entry_safe(work, next, &works, list) { | |
8188 | list_del_init(&work->list); | |
8189 | btrfs_wait_and_free_delalloc_work(work); | |
8190 | } | |
8191 | ||
8192 | if (!list_empty_careful(&splice)) { | |
8193 | spin_lock(&root->delalloc_lock); | |
8194 | list_splice_tail(&splice, &root->delalloc_inodes); | |
8195 | spin_unlock(&root->delalloc_lock); | |
8196 | } | |
8197 | return ret; | |
8198 | } | |
1eafa6c7 | 8199 | |
eb73c1b7 MX |
8200 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
8201 | { | |
8202 | int ret; | |
1eafa6c7 | 8203 | |
eb73c1b7 MX |
8204 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
8205 | return -EROFS; | |
8206 | ||
8207 | ret = __start_delalloc_inodes(root, delay_iput); | |
8208 | /* | |
8209 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
8210 | * we have to make sure the IO is actually started and that |
8211 | * ordered extents get created before we return | |
8212 | */ | |
8213 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 8214 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 8215 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 8216 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
8217 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
8218 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
8219 | } |
8220 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
8221 | return ret; |
8222 | } | |
8223 | ||
8224 | int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info, | |
8225 | int delay_iput) | |
8226 | { | |
8227 | struct btrfs_root *root; | |
8228 | struct list_head splice; | |
8229 | int ret; | |
8230 | ||
8231 | if (fs_info->sb->s_flags & MS_RDONLY) | |
8232 | return -EROFS; | |
8233 | ||
8234 | INIT_LIST_HEAD(&splice); | |
8235 | ||
8236 | spin_lock(&fs_info->delalloc_root_lock); | |
8237 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
8238 | while (!list_empty(&splice)) { | |
8239 | root = list_first_entry(&splice, struct btrfs_root, | |
8240 | delalloc_root); | |
8241 | root = btrfs_grab_fs_root(root); | |
8242 | BUG_ON(!root); | |
8243 | list_move_tail(&root->delalloc_root, | |
8244 | &fs_info->delalloc_roots); | |
8245 | spin_unlock(&fs_info->delalloc_root_lock); | |
8246 | ||
8247 | ret = __start_delalloc_inodes(root, delay_iput); | |
8248 | btrfs_put_fs_root(root); | |
8249 | if (ret) | |
8250 | goto out; | |
8251 | ||
8252 | spin_lock(&fs_info->delalloc_root_lock); | |
8ccf6f19 | 8253 | } |
eb73c1b7 | 8254 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 8255 | |
eb73c1b7 MX |
8256 | atomic_inc(&fs_info->async_submit_draining); |
8257 | while (atomic_read(&fs_info->nr_async_submits) || | |
8258 | atomic_read(&fs_info->async_delalloc_pages)) { | |
8259 | wait_event(fs_info->async_submit_wait, | |
8260 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
8261 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
8262 | } | |
8263 | atomic_dec(&fs_info->async_submit_draining); | |
8264 | return 0; | |
8265 | out: | |
1eafa6c7 | 8266 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
8267 | spin_lock(&fs_info->delalloc_root_lock); |
8268 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
8269 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 8270 | } |
8ccf6f19 | 8271 | return ret; |
ea8c2819 CM |
8272 | } |
8273 | ||
39279cc3 CM |
8274 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
8275 | const char *symname) | |
8276 | { | |
8277 | struct btrfs_trans_handle *trans; | |
8278 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
8279 | struct btrfs_path *path; | |
8280 | struct btrfs_key key; | |
1832a6d5 | 8281 | struct inode *inode = NULL; |
39279cc3 CM |
8282 | int err; |
8283 | int drop_inode = 0; | |
8284 | u64 objectid; | |
00e4e6b3 | 8285 | u64 index = 0 ; |
39279cc3 CM |
8286 | int name_len; |
8287 | int datasize; | |
5f39d397 | 8288 | unsigned long ptr; |
39279cc3 | 8289 | struct btrfs_file_extent_item *ei; |
5f39d397 | 8290 | struct extent_buffer *leaf; |
39279cc3 CM |
8291 | |
8292 | name_len = strlen(symname) + 1; | |
8293 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
8294 | return -ENAMETOOLONG; | |
1832a6d5 | 8295 | |
9ed74f2d JB |
8296 | /* |
8297 | * 2 items for inode item and ref | |
8298 | * 2 items for dir items | |
8299 | * 1 item for xattr if selinux is on | |
8300 | */ | |
a22285a6 YZ |
8301 | trans = btrfs_start_transaction(root, 5); |
8302 | if (IS_ERR(trans)) | |
8303 | return PTR_ERR(trans); | |
1832a6d5 | 8304 | |
581bb050 LZ |
8305 | err = btrfs_find_free_ino(root, &objectid); |
8306 | if (err) | |
8307 | goto out_unlock; | |
8308 | ||
aec7477b | 8309 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 8310 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 8311 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
8312 | if (IS_ERR(inode)) { |
8313 | err = PTR_ERR(inode); | |
39279cc3 | 8314 | goto out_unlock; |
7cf96da3 | 8315 | } |
39279cc3 | 8316 | |
2a7dba39 | 8317 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
8318 | if (err) { |
8319 | drop_inode = 1; | |
8320 | goto out_unlock; | |
8321 | } | |
8322 | ||
ad19db71 CS |
8323 | /* |
8324 | * If the active LSM wants to access the inode during | |
8325 | * d_instantiate it needs these. Smack checks to see | |
8326 | * if the filesystem supports xattrs by looking at the | |
8327 | * ops vector. | |
8328 | */ | |
8329 | inode->i_fop = &btrfs_file_operations; | |
8330 | inode->i_op = &btrfs_file_inode_operations; | |
8331 | ||
a1b075d2 | 8332 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
8333 | if (err) |
8334 | drop_inode = 1; | |
8335 | else { | |
8336 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 8337 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 8338 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 8339 | } |
39279cc3 CM |
8340 | if (drop_inode) |
8341 | goto out_unlock; | |
8342 | ||
8343 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
8344 | if (!path) { |
8345 | err = -ENOMEM; | |
8346 | drop_inode = 1; | |
8347 | goto out_unlock; | |
8348 | } | |
33345d01 | 8349 | key.objectid = btrfs_ino(inode); |
39279cc3 | 8350 | key.offset = 0; |
39279cc3 CM |
8351 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
8352 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
8353 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
8354 | datasize); | |
54aa1f4d CM |
8355 | if (err) { |
8356 | drop_inode = 1; | |
b0839166 | 8357 | btrfs_free_path(path); |
54aa1f4d CM |
8358 | goto out_unlock; |
8359 | } | |
5f39d397 CM |
8360 | leaf = path->nodes[0]; |
8361 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
8362 | struct btrfs_file_extent_item); | |
8363 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
8364 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 8365 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
8366 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
8367 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
8368 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
8369 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
8370 | ||
39279cc3 | 8371 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
8372 | write_extent_buffer(leaf, symname, ptr, name_len); |
8373 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 8374 | btrfs_free_path(path); |
5f39d397 | 8375 | |
39279cc3 CM |
8376 | inode->i_op = &btrfs_symlink_inode_operations; |
8377 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 8378 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 8379 | inode_set_bytes(inode, name_len); |
dbe674a9 | 8380 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
8381 | err = btrfs_update_inode(trans, root, inode); |
8382 | if (err) | |
8383 | drop_inode = 1; | |
39279cc3 CM |
8384 | |
8385 | out_unlock: | |
08c422c2 AV |
8386 | if (!err) |
8387 | d_instantiate(dentry, inode); | |
7ad85bb7 | 8388 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
8389 | if (drop_inode) { |
8390 | inode_dec_link_count(inode); | |
8391 | iput(inode); | |
8392 | } | |
b53d3f5d | 8393 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
8394 | return err; |
8395 | } | |
16432985 | 8396 | |
0af3d00b JB |
8397 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
8398 | u64 start, u64 num_bytes, u64 min_size, | |
8399 | loff_t actual_len, u64 *alloc_hint, | |
8400 | struct btrfs_trans_handle *trans) | |
d899e052 | 8401 | { |
5dc562c5 JB |
8402 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
8403 | struct extent_map *em; | |
d899e052 YZ |
8404 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8405 | struct btrfs_key ins; | |
d899e052 | 8406 | u64 cur_offset = start; |
55a61d1d | 8407 | u64 i_size; |
154ea289 | 8408 | u64 cur_bytes; |
d899e052 | 8409 | int ret = 0; |
0af3d00b | 8410 | bool own_trans = true; |
d899e052 | 8411 | |
0af3d00b JB |
8412 | if (trans) |
8413 | own_trans = false; | |
d899e052 | 8414 | while (num_bytes > 0) { |
0af3d00b JB |
8415 | if (own_trans) { |
8416 | trans = btrfs_start_transaction(root, 3); | |
8417 | if (IS_ERR(trans)) { | |
8418 | ret = PTR_ERR(trans); | |
8419 | break; | |
8420 | } | |
5a303d5d YZ |
8421 | } |
8422 | ||
154ea289 CM |
8423 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
8424 | cur_bytes = max(cur_bytes, min_size); | |
00361589 JB |
8425 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
8426 | *alloc_hint, &ins, 1); | |
5a303d5d | 8427 | if (ret) { |
0af3d00b JB |
8428 | if (own_trans) |
8429 | btrfs_end_transaction(trans, root); | |
a22285a6 | 8430 | break; |
d899e052 | 8431 | } |
5a303d5d | 8432 | |
d899e052 YZ |
8433 | ret = insert_reserved_file_extent(trans, inode, |
8434 | cur_offset, ins.objectid, | |
8435 | ins.offset, ins.offset, | |
920bbbfb | 8436 | ins.offset, 0, 0, 0, |
d899e052 | 8437 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
8438 | if (ret) { |
8439 | btrfs_abort_transaction(trans, root, ret); | |
8440 | if (own_trans) | |
8441 | btrfs_end_transaction(trans, root); | |
8442 | break; | |
8443 | } | |
a1ed835e CM |
8444 | btrfs_drop_extent_cache(inode, cur_offset, |
8445 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 8446 | |
5dc562c5 JB |
8447 | em = alloc_extent_map(); |
8448 | if (!em) { | |
8449 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
8450 | &BTRFS_I(inode)->runtime_flags); | |
8451 | goto next; | |
8452 | } | |
8453 | ||
8454 | em->start = cur_offset; | |
8455 | em->orig_start = cur_offset; | |
8456 | em->len = ins.offset; | |
8457 | em->block_start = ins.objectid; | |
8458 | em->block_len = ins.offset; | |
b4939680 | 8459 | em->orig_block_len = ins.offset; |
cc95bef6 | 8460 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
8461 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
8462 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
8463 | em->generation = trans->transid; | |
8464 | ||
8465 | while (1) { | |
8466 | write_lock(&em_tree->lock); | |
09a2a8f9 | 8467 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
8468 | write_unlock(&em_tree->lock); |
8469 | if (ret != -EEXIST) | |
8470 | break; | |
8471 | btrfs_drop_extent_cache(inode, cur_offset, | |
8472 | cur_offset + ins.offset - 1, | |
8473 | 0); | |
8474 | } | |
8475 | free_extent_map(em); | |
8476 | next: | |
d899e052 YZ |
8477 | num_bytes -= ins.offset; |
8478 | cur_offset += ins.offset; | |
efa56464 | 8479 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 8480 | |
0c4d2d95 | 8481 | inode_inc_iversion(inode); |
d899e052 | 8482 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 8483 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 8484 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
8485 | (actual_len > inode->i_size) && |
8486 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 8487 | if (cur_offset > actual_len) |
55a61d1d | 8488 | i_size = actual_len; |
d1ea6a61 | 8489 | else |
55a61d1d JB |
8490 | i_size = cur_offset; |
8491 | i_size_write(inode, i_size); | |
8492 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
8493 | } |
8494 | ||
d899e052 | 8495 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
8496 | |
8497 | if (ret) { | |
8498 | btrfs_abort_transaction(trans, root, ret); | |
8499 | if (own_trans) | |
8500 | btrfs_end_transaction(trans, root); | |
8501 | break; | |
8502 | } | |
d899e052 | 8503 | |
0af3d00b JB |
8504 | if (own_trans) |
8505 | btrfs_end_transaction(trans, root); | |
5a303d5d | 8506 | } |
d899e052 YZ |
8507 | return ret; |
8508 | } | |
8509 | ||
0af3d00b JB |
8510 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
8511 | u64 start, u64 num_bytes, u64 min_size, | |
8512 | loff_t actual_len, u64 *alloc_hint) | |
8513 | { | |
8514 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8515 | min_size, actual_len, alloc_hint, | |
8516 | NULL); | |
8517 | } | |
8518 | ||
8519 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
8520 | struct btrfs_trans_handle *trans, int mode, | |
8521 | u64 start, u64 num_bytes, u64 min_size, | |
8522 | loff_t actual_len, u64 *alloc_hint) | |
8523 | { | |
8524 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8525 | min_size, actual_len, alloc_hint, trans); | |
8526 | } | |
8527 | ||
e6dcd2dc CM |
8528 | static int btrfs_set_page_dirty(struct page *page) |
8529 | { | |
e6dcd2dc CM |
8530 | return __set_page_dirty_nobuffers(page); |
8531 | } | |
8532 | ||
10556cb2 | 8533 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 8534 | { |
b83cc969 | 8535 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 8536 | umode_t mode = inode->i_mode; |
b83cc969 | 8537 | |
cb6db4e5 JM |
8538 | if (mask & MAY_WRITE && |
8539 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
8540 | if (btrfs_root_readonly(root)) | |
8541 | return -EROFS; | |
8542 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
8543 | return -EACCES; | |
8544 | } | |
2830ba7f | 8545 | return generic_permission(inode, mask); |
fdebe2bd | 8546 | } |
39279cc3 | 8547 | |
6e1d5dcc | 8548 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 8549 | .getattr = btrfs_getattr, |
39279cc3 CM |
8550 | .lookup = btrfs_lookup, |
8551 | .create = btrfs_create, | |
8552 | .unlink = btrfs_unlink, | |
8553 | .link = btrfs_link, | |
8554 | .mkdir = btrfs_mkdir, | |
8555 | .rmdir = btrfs_rmdir, | |
8556 | .rename = btrfs_rename, | |
8557 | .symlink = btrfs_symlink, | |
8558 | .setattr = btrfs_setattr, | |
618e21d5 | 8559 | .mknod = btrfs_mknod, |
95819c05 CH |
8560 | .setxattr = btrfs_setxattr, |
8561 | .getxattr = btrfs_getxattr, | |
5103e947 | 8562 | .listxattr = btrfs_listxattr, |
95819c05 | 8563 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8564 | .permission = btrfs_permission, |
4e34e719 | 8565 | .get_acl = btrfs_get_acl, |
39279cc3 | 8566 | }; |
6e1d5dcc | 8567 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8568 | .lookup = btrfs_lookup, |
fdebe2bd | 8569 | .permission = btrfs_permission, |
4e34e719 | 8570 | .get_acl = btrfs_get_acl, |
39279cc3 | 8571 | }; |
76dda93c | 8572 | |
828c0950 | 8573 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8574 | .llseek = generic_file_llseek, |
8575 | .read = generic_read_dir, | |
9cdda8d3 | 8576 | .iterate = btrfs_real_readdir, |
34287aa3 | 8577 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8578 | #ifdef CONFIG_COMPAT |
34287aa3 | 8579 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8580 | #endif |
6bf13c0c | 8581 | .release = btrfs_release_file, |
e02119d5 | 8582 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8583 | }; |
8584 | ||
d1310b2e | 8585 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8586 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8587 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8588 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8589 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8590 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8591 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8592 | .set_bit_hook = btrfs_set_bit_hook, |
8593 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8594 | .merge_extent_hook = btrfs_merge_extent_hook, |
8595 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8596 | }; |
8597 | ||
35054394 CM |
8598 | /* |
8599 | * btrfs doesn't support the bmap operation because swapfiles | |
8600 | * use bmap to make a mapping of extents in the file. They assume | |
8601 | * these extents won't change over the life of the file and they | |
8602 | * use the bmap result to do IO directly to the drive. | |
8603 | * | |
8604 | * the btrfs bmap call would return logical addresses that aren't | |
8605 | * suitable for IO and they also will change frequently as COW | |
8606 | * operations happen. So, swapfile + btrfs == corruption. | |
8607 | * | |
8608 | * For now we're avoiding this by dropping bmap. | |
8609 | */ | |
7f09410b | 8610 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8611 | .readpage = btrfs_readpage, |
8612 | .writepage = btrfs_writepage, | |
b293f02e | 8613 | .writepages = btrfs_writepages, |
3ab2fb5a | 8614 | .readpages = btrfs_readpages, |
16432985 | 8615 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8616 | .invalidatepage = btrfs_invalidatepage, |
8617 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8618 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8619 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8620 | }; |
8621 | ||
7f09410b | 8622 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8623 | .readpage = btrfs_readpage, |
8624 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8625 | .invalidatepage = btrfs_invalidatepage, |
8626 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8627 | }; |
8628 | ||
6e1d5dcc | 8629 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8630 | .getattr = btrfs_getattr, |
8631 | .setattr = btrfs_setattr, | |
95819c05 CH |
8632 | .setxattr = btrfs_setxattr, |
8633 | .getxattr = btrfs_getxattr, | |
5103e947 | 8634 | .listxattr = btrfs_listxattr, |
95819c05 | 8635 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8636 | .permission = btrfs_permission, |
1506fcc8 | 8637 | .fiemap = btrfs_fiemap, |
4e34e719 | 8638 | .get_acl = btrfs_get_acl, |
e41f941a | 8639 | .update_time = btrfs_update_time, |
39279cc3 | 8640 | }; |
6e1d5dcc | 8641 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8642 | .getattr = btrfs_getattr, |
8643 | .setattr = btrfs_setattr, | |
fdebe2bd | 8644 | .permission = btrfs_permission, |
95819c05 CH |
8645 | .setxattr = btrfs_setxattr, |
8646 | .getxattr = btrfs_getxattr, | |
33268eaf | 8647 | .listxattr = btrfs_listxattr, |
95819c05 | 8648 | .removexattr = btrfs_removexattr, |
4e34e719 | 8649 | .get_acl = btrfs_get_acl, |
e41f941a | 8650 | .update_time = btrfs_update_time, |
618e21d5 | 8651 | }; |
6e1d5dcc | 8652 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8653 | .readlink = generic_readlink, |
8654 | .follow_link = page_follow_link_light, | |
8655 | .put_link = page_put_link, | |
f209561a | 8656 | .getattr = btrfs_getattr, |
22c44fe6 | 8657 | .setattr = btrfs_setattr, |
fdebe2bd | 8658 | .permission = btrfs_permission, |
0279b4cd JO |
8659 | .setxattr = btrfs_setxattr, |
8660 | .getxattr = btrfs_getxattr, | |
8661 | .listxattr = btrfs_listxattr, | |
8662 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8663 | .get_acl = btrfs_get_acl, |
e41f941a | 8664 | .update_time = btrfs_update_time, |
39279cc3 | 8665 | }; |
76dda93c | 8666 | |
82d339d9 | 8667 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8668 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8669 | .d_release = btrfs_dentry_release, |
76dda93c | 8670 | }; |