]>
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> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
53b381b3 | 43 | #include <linux/blkdev.h> |
4b4e25f2 | 44 | #include "compat.h" |
39279cc3 CM |
45 | #include "ctree.h" |
46 | #include "disk-io.h" | |
47 | #include "transaction.h" | |
48 | #include "btrfs_inode.h" | |
39279cc3 | 49 | #include "print-tree.h" |
e6dcd2dc | 50 | #include "ordered-data.h" |
95819c05 | 51 | #include "xattr.h" |
e02119d5 | 52 | #include "tree-log.h" |
4a54c8c1 | 53 | #include "volumes.h" |
c8b97818 | 54 | #include "compression.h" |
b4ce94de | 55 | #include "locking.h" |
dc89e982 | 56 | #include "free-space-cache.h" |
581bb050 | 57 | #include "inode-map.h" |
38c227d8 | 58 | #include "backref.h" |
39279cc3 CM |
59 | |
60 | struct btrfs_iget_args { | |
61 | u64 ino; | |
62 | struct btrfs_root *root; | |
63 | }; | |
64 | ||
6e1d5dcc AD |
65 | static const struct inode_operations btrfs_dir_inode_operations; |
66 | static const struct inode_operations btrfs_symlink_inode_operations; | |
67 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
68 | static const struct inode_operations btrfs_special_inode_operations; | |
69 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
70 | static const struct address_space_operations btrfs_aops; |
71 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 72 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 73 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
74 | |
75 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 76 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
77 | struct kmem_cache *btrfs_trans_handle_cachep; |
78 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 79 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 80 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
81 | |
82 | #define S_SHIFT 12 | |
83 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
84 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
85 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
86 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
87 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
88 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
89 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
90 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
91 | }; | |
92 | ||
3972f260 | 93 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 94 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 95 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
96 | static noinline int cow_file_range(struct inode *inode, |
97 | struct page *locked_page, | |
98 | u64 start, u64 end, int *page_started, | |
99 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
100 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
101 | u64 len, u64 orig_start, | |
102 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
103 | u64 orig_block_len, u64 ram_bytes, |
104 | int type); | |
7b128766 | 105 | |
48a3b636 ES |
106 | static int btrfs_dirty_inode(struct inode *inode); |
107 | ||
f34f57a3 | 108 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
109 | struct inode *inode, struct inode *dir, |
110 | const struct qstr *qstr) | |
0279b4cd JO |
111 | { |
112 | int err; | |
113 | ||
f34f57a3 | 114 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 115 | if (!err) |
2a7dba39 | 116 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
117 | return err; |
118 | } | |
119 | ||
c8b97818 CM |
120 | /* |
121 | * this does all the hard work for inserting an inline extent into | |
122 | * the btree. The caller should have done a btrfs_drop_extents so that | |
123 | * no overlapping inline items exist in the btree | |
124 | */ | |
d397712b | 125 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
126 | struct btrfs_root *root, struct inode *inode, |
127 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 128 | int compress_type, |
c8b97818 CM |
129 | struct page **compressed_pages) |
130 | { | |
131 | struct btrfs_key key; | |
132 | struct btrfs_path *path; | |
133 | struct extent_buffer *leaf; | |
134 | struct page *page = NULL; | |
135 | char *kaddr; | |
136 | unsigned long ptr; | |
137 | struct btrfs_file_extent_item *ei; | |
138 | int err = 0; | |
139 | int ret; | |
140 | size_t cur_size = size; | |
141 | size_t datasize; | |
142 | unsigned long offset; | |
c8b97818 | 143 | |
fe3f566c | 144 | if (compressed_size && compressed_pages) |
c8b97818 | 145 | cur_size = compressed_size; |
c8b97818 | 146 | |
d397712b CM |
147 | path = btrfs_alloc_path(); |
148 | if (!path) | |
c8b97818 CM |
149 | return -ENOMEM; |
150 | ||
b9473439 | 151 | path->leave_spinning = 1; |
c8b97818 | 152 | |
33345d01 | 153 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
154 | key.offset = start; |
155 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
156 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
157 | ||
158 | inode_add_bytes(inode, size); | |
159 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
160 | datasize); | |
c8b97818 CM |
161 | if (ret) { |
162 | err = ret; | |
c8b97818 CM |
163 | goto fail; |
164 | } | |
165 | leaf = path->nodes[0]; | |
166 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
167 | struct btrfs_file_extent_item); | |
168 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
169 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
170 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
171 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
172 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
173 | ptr = btrfs_file_extent_inline_start(ei); | |
174 | ||
261507a0 | 175 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
176 | struct page *cpage; |
177 | int i = 0; | |
d397712b | 178 | while (compressed_size > 0) { |
c8b97818 | 179 | cpage = compressed_pages[i]; |
5b050f04 | 180 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
181 | PAGE_CACHE_SIZE); |
182 | ||
7ac687d9 | 183 | kaddr = kmap_atomic(cpage); |
c8b97818 | 184 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 185 | kunmap_atomic(kaddr); |
c8b97818 CM |
186 | |
187 | i++; | |
188 | ptr += cur_size; | |
189 | compressed_size -= cur_size; | |
190 | } | |
191 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 192 | compress_type); |
c8b97818 CM |
193 | } else { |
194 | page = find_get_page(inode->i_mapping, | |
195 | start >> PAGE_CACHE_SHIFT); | |
196 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 197 | kaddr = kmap_atomic(page); |
c8b97818 CM |
198 | offset = start & (PAGE_CACHE_SIZE - 1); |
199 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 200 | kunmap_atomic(kaddr); |
c8b97818 CM |
201 | page_cache_release(page); |
202 | } | |
203 | btrfs_mark_buffer_dirty(leaf); | |
204 | btrfs_free_path(path); | |
205 | ||
c2167754 YZ |
206 | /* |
207 | * we're an inline extent, so nobody can | |
208 | * extend the file past i_size without locking | |
209 | * a page we already have locked. | |
210 | * | |
211 | * We must do any isize and inode updates | |
212 | * before we unlock the pages. Otherwise we | |
213 | * could end up racing with unlink. | |
214 | */ | |
c8b97818 | 215 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 216 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 217 | |
79787eaa | 218 | return ret; |
c8b97818 CM |
219 | fail: |
220 | btrfs_free_path(path); | |
221 | return err; | |
222 | } | |
223 | ||
224 | ||
225 | /* | |
226 | * conditionally insert an inline extent into the file. This | |
227 | * does the checks required to make sure the data is small enough | |
228 | * to fit as an inline extent. | |
229 | */ | |
7f366cfe | 230 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
231 | struct btrfs_root *root, |
232 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 233 | size_t compressed_size, int compress_type, |
c8b97818 CM |
234 | struct page **compressed_pages) |
235 | { | |
236 | u64 isize = i_size_read(inode); | |
237 | u64 actual_end = min(end + 1, isize); | |
238 | u64 inline_len = actual_end - start; | |
fda2832f | 239 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
240 | u64 data_len = inline_len; |
241 | int ret; | |
242 | ||
243 | if (compressed_size) | |
244 | data_len = compressed_size; | |
245 | ||
246 | if (start > 0 || | |
70b99e69 | 247 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
248 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
249 | (!compressed_size && | |
250 | (actual_end & (root->sectorsize - 1)) == 0) || | |
251 | end + 1 < isize || | |
252 | data_len > root->fs_info->max_inline) { | |
253 | return 1; | |
254 | } | |
255 | ||
2671485d | 256 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
257 | if (ret) |
258 | return ret; | |
c8b97818 CM |
259 | |
260 | if (isize > actual_end) | |
261 | inline_len = min_t(u64, isize, actual_end); | |
262 | ret = insert_inline_extent(trans, root, inode, start, | |
263 | inline_len, compressed_size, | |
fe3f566c | 264 | compress_type, compressed_pages); |
2adcac1a | 265 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
266 | btrfs_abort_transaction(trans, root, ret); |
267 | return ret; | |
2adcac1a JB |
268 | } else if (ret == -ENOSPC) { |
269 | return 1; | |
79787eaa | 270 | } |
2adcac1a | 271 | |
bdc20e67 | 272 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 273 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 274 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
275 | return 0; |
276 | } | |
277 | ||
771ed689 CM |
278 | struct async_extent { |
279 | u64 start; | |
280 | u64 ram_size; | |
281 | u64 compressed_size; | |
282 | struct page **pages; | |
283 | unsigned long nr_pages; | |
261507a0 | 284 | int compress_type; |
771ed689 CM |
285 | struct list_head list; |
286 | }; | |
287 | ||
288 | struct async_cow { | |
289 | struct inode *inode; | |
290 | struct btrfs_root *root; | |
291 | struct page *locked_page; | |
292 | u64 start; | |
293 | u64 end; | |
294 | struct list_head extents; | |
295 | struct btrfs_work work; | |
296 | }; | |
297 | ||
298 | static noinline int add_async_extent(struct async_cow *cow, | |
299 | u64 start, u64 ram_size, | |
300 | u64 compressed_size, | |
301 | struct page **pages, | |
261507a0 LZ |
302 | unsigned long nr_pages, |
303 | int compress_type) | |
771ed689 CM |
304 | { |
305 | struct async_extent *async_extent; | |
306 | ||
307 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 308 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
309 | async_extent->start = start; |
310 | async_extent->ram_size = ram_size; | |
311 | async_extent->compressed_size = compressed_size; | |
312 | async_extent->pages = pages; | |
313 | async_extent->nr_pages = nr_pages; | |
261507a0 | 314 | async_extent->compress_type = compress_type; |
771ed689 CM |
315 | list_add_tail(&async_extent->list, &cow->extents); |
316 | return 0; | |
317 | } | |
318 | ||
d352ac68 | 319 | /* |
771ed689 CM |
320 | * we create compressed extents in two phases. The first |
321 | * phase compresses a range of pages that have already been | |
322 | * locked (both pages and state bits are locked). | |
c8b97818 | 323 | * |
771ed689 CM |
324 | * This is done inside an ordered work queue, and the compression |
325 | * is spread across many cpus. The actual IO submission is step | |
326 | * two, and the ordered work queue takes care of making sure that | |
327 | * happens in the same order things were put onto the queue by | |
328 | * writepages and friends. | |
c8b97818 | 329 | * |
771ed689 CM |
330 | * If this code finds it can't get good compression, it puts an |
331 | * entry onto the work queue to write the uncompressed bytes. This | |
332 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
333 | * are written in the same order that the flusher thread sent them |
334 | * down. | |
d352ac68 | 335 | */ |
771ed689 CM |
336 | static noinline int compress_file_range(struct inode *inode, |
337 | struct page *locked_page, | |
338 | u64 start, u64 end, | |
339 | struct async_cow *async_cow, | |
340 | int *num_added) | |
b888db2b CM |
341 | { |
342 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
343 | struct btrfs_trans_handle *trans; | |
db94535d | 344 | u64 num_bytes; |
db94535d | 345 | u64 blocksize = root->sectorsize; |
c8b97818 | 346 | u64 actual_end; |
42dc7bab | 347 | u64 isize = i_size_read(inode); |
e6dcd2dc | 348 | int ret = 0; |
c8b97818 CM |
349 | struct page **pages = NULL; |
350 | unsigned long nr_pages; | |
351 | unsigned long nr_pages_ret = 0; | |
352 | unsigned long total_compressed = 0; | |
353 | unsigned long total_in = 0; | |
354 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 355 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
356 | int i; |
357 | int will_compress; | |
261507a0 | 358 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 359 | int redirty = 0; |
b888db2b | 360 | |
4cb13e5d LB |
361 | /* if this is a small write inside eof, kick off a defrag */ |
362 | if ((end - start + 1) < 16 * 1024 && | |
363 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
364 | btrfs_add_inode_defrag(NULL, inode); |
365 | ||
42dc7bab | 366 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
367 | again: |
368 | will_compress = 0; | |
369 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
370 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 371 | |
f03d9301 CM |
372 | /* |
373 | * we don't want to send crud past the end of i_size through | |
374 | * compression, that's just a waste of CPU time. So, if the | |
375 | * end of the file is before the start of our current | |
376 | * requested range of bytes, we bail out to the uncompressed | |
377 | * cleanup code that can deal with all of this. | |
378 | * | |
379 | * It isn't really the fastest way to fix things, but this is a | |
380 | * very uncommon corner. | |
381 | */ | |
382 | if (actual_end <= start) | |
383 | goto cleanup_and_bail_uncompressed; | |
384 | ||
c8b97818 CM |
385 | total_compressed = actual_end - start; |
386 | ||
387 | /* we want to make sure that amount of ram required to uncompress | |
388 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
389 | * of a compressed extent to 128k. This is a crucial number |
390 | * because it also controls how easily we can spread reads across | |
391 | * cpus for decompression. | |
392 | * | |
393 | * We also want to make sure the amount of IO required to do | |
394 | * a random read is reasonably small, so we limit the size of | |
395 | * a compressed extent to 128k. | |
c8b97818 CM |
396 | */ |
397 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 398 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 399 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
400 | total_in = 0; |
401 | ret = 0; | |
db94535d | 402 | |
771ed689 CM |
403 | /* |
404 | * we do compression for mount -o compress and when the | |
405 | * inode has not been flagged as nocompress. This flag can | |
406 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 407 | */ |
6cbff00f | 408 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 409 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
410 | (BTRFS_I(inode)->force_compress) || |
411 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 412 | WARN_ON(pages); |
cfbc246e | 413 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
414 | if (!pages) { |
415 | /* just bail out to the uncompressed code */ | |
416 | goto cont; | |
417 | } | |
c8b97818 | 418 | |
261507a0 LZ |
419 | if (BTRFS_I(inode)->force_compress) |
420 | compress_type = BTRFS_I(inode)->force_compress; | |
421 | ||
4adaa611 CM |
422 | /* |
423 | * we need to call clear_page_dirty_for_io on each | |
424 | * page in the range. Otherwise applications with the file | |
425 | * mmap'd can wander in and change the page contents while | |
426 | * we are compressing them. | |
427 | * | |
428 | * If the compression fails for any reason, we set the pages | |
429 | * dirty again later on. | |
430 | */ | |
431 | extent_range_clear_dirty_for_io(inode, start, end); | |
432 | redirty = 1; | |
261507a0 LZ |
433 | ret = btrfs_compress_pages(compress_type, |
434 | inode->i_mapping, start, | |
435 | total_compressed, pages, | |
436 | nr_pages, &nr_pages_ret, | |
437 | &total_in, | |
438 | &total_compressed, | |
439 | max_compressed); | |
c8b97818 CM |
440 | |
441 | if (!ret) { | |
442 | unsigned long offset = total_compressed & | |
443 | (PAGE_CACHE_SIZE - 1); | |
444 | struct page *page = pages[nr_pages_ret - 1]; | |
445 | char *kaddr; | |
446 | ||
447 | /* zero the tail end of the last page, we might be | |
448 | * sending it down to disk | |
449 | */ | |
450 | if (offset) { | |
7ac687d9 | 451 | kaddr = kmap_atomic(page); |
c8b97818 CM |
452 | memset(kaddr + offset, 0, |
453 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 454 | kunmap_atomic(kaddr); |
c8b97818 CM |
455 | } |
456 | will_compress = 1; | |
457 | } | |
458 | } | |
560f7d75 | 459 | cont: |
c8b97818 | 460 | if (start == 0) { |
7a7eaa40 | 461 | trans = btrfs_join_transaction(root); |
79787eaa JM |
462 | if (IS_ERR(trans)) { |
463 | ret = PTR_ERR(trans); | |
464 | trans = NULL; | |
465 | goto cleanup_and_out; | |
466 | } | |
0ca1f7ce | 467 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 468 | |
c8b97818 | 469 | /* lets try to make an inline extent */ |
771ed689 | 470 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 471 | /* we didn't compress the entire range, try |
771ed689 | 472 | * to make an uncompressed inline extent. |
c8b97818 CM |
473 | */ |
474 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 475 | start, end, 0, 0, NULL); |
c8b97818 | 476 | } else { |
771ed689 | 477 | /* try making a compressed inline extent */ |
c8b97818 CM |
478 | ret = cow_file_range_inline(trans, root, inode, |
479 | start, end, | |
fe3f566c LZ |
480 | total_compressed, |
481 | compress_type, pages); | |
c8b97818 | 482 | } |
79787eaa | 483 | if (ret <= 0) { |
771ed689 | 484 | /* |
79787eaa JM |
485 | * inline extent creation worked or returned error, |
486 | * we don't need to create any more async work items. | |
487 | * Unlock and free up our temp pages. | |
771ed689 | 488 | */ |
c8b97818 | 489 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
490 | &BTRFS_I(inode)->io_tree, |
491 | start, end, NULL, | |
492 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 493 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 494 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
495 | |
496 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
497 | goto free_pages_out; |
498 | } | |
c2167754 | 499 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
500 | } |
501 | ||
502 | if (will_compress) { | |
503 | /* | |
504 | * we aren't doing an inline extent round the compressed size | |
505 | * up to a block size boundary so the allocator does sane | |
506 | * things | |
507 | */ | |
fda2832f | 508 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
509 | |
510 | /* | |
511 | * one last check to make sure the compression is really a | |
512 | * win, compare the page count read with the blocks on disk | |
513 | */ | |
fda2832f | 514 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
515 | if (total_compressed >= total_in) { |
516 | will_compress = 0; | |
517 | } else { | |
c8b97818 CM |
518 | num_bytes = total_in; |
519 | } | |
520 | } | |
521 | if (!will_compress && pages) { | |
522 | /* | |
523 | * the compression code ran but failed to make things smaller, | |
524 | * free any pages it allocated and our page pointer array | |
525 | */ | |
526 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 527 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
528 | page_cache_release(pages[i]); |
529 | } | |
530 | kfree(pages); | |
531 | pages = NULL; | |
532 | total_compressed = 0; | |
533 | nr_pages_ret = 0; | |
534 | ||
535 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
536 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
537 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 538 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 539 | } |
c8b97818 | 540 | } |
771ed689 CM |
541 | if (will_compress) { |
542 | *num_added += 1; | |
c8b97818 | 543 | |
771ed689 CM |
544 | /* the async work queues will take care of doing actual |
545 | * allocation on disk for these compressed pages, | |
546 | * and will submit them to the elevator. | |
547 | */ | |
548 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
549 | total_compressed, pages, nr_pages_ret, |
550 | compress_type); | |
179e29e4 | 551 | |
24ae6365 | 552 | if (start + num_bytes < end) { |
771ed689 CM |
553 | start += num_bytes; |
554 | pages = NULL; | |
555 | cond_resched(); | |
556 | goto again; | |
557 | } | |
558 | } else { | |
f03d9301 | 559 | cleanup_and_bail_uncompressed: |
771ed689 CM |
560 | /* |
561 | * No compression, but we still need to write the pages in | |
562 | * the file we've been given so far. redirty the locked | |
563 | * page if it corresponds to our extent and set things up | |
564 | * for the async work queue to run cow_file_range to do | |
565 | * the normal delalloc dance | |
566 | */ | |
567 | if (page_offset(locked_page) >= start && | |
568 | page_offset(locked_page) <= end) { | |
569 | __set_page_dirty_nobuffers(locked_page); | |
570 | /* unlocked later on in the async handlers */ | |
571 | } | |
4adaa611 CM |
572 | if (redirty) |
573 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
574 | add_async_extent(async_cow, start, end - start + 1, |
575 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
576 | *num_added += 1; |
577 | } | |
3b951516 | 578 | |
771ed689 | 579 | out: |
79787eaa | 580 | return ret; |
771ed689 CM |
581 | |
582 | free_pages_out: | |
583 | for (i = 0; i < nr_pages_ret; i++) { | |
584 | WARN_ON(pages[i]->mapping); | |
585 | page_cache_release(pages[i]); | |
586 | } | |
d397712b | 587 | kfree(pages); |
771ed689 CM |
588 | |
589 | goto out; | |
79787eaa JM |
590 | |
591 | cleanup_and_out: | |
592 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
593 | start, end, NULL, | |
594 | EXTENT_CLEAR_UNLOCK_PAGE | | |
595 | EXTENT_CLEAR_DIRTY | | |
596 | EXTENT_CLEAR_DELALLOC | | |
597 | EXTENT_SET_WRITEBACK | | |
598 | EXTENT_END_WRITEBACK); | |
599 | if (!trans || IS_ERR(trans)) | |
600 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
601 | else | |
602 | btrfs_abort_transaction(trans, root, ret); | |
603 | goto free_pages_out; | |
771ed689 CM |
604 | } |
605 | ||
606 | /* | |
607 | * phase two of compressed writeback. This is the ordered portion | |
608 | * of the code, which only gets called in the order the work was | |
609 | * queued. We walk all the async extents created by compress_file_range | |
610 | * and send them down to the disk. | |
611 | */ | |
612 | static noinline int submit_compressed_extents(struct inode *inode, | |
613 | struct async_cow *async_cow) | |
614 | { | |
615 | struct async_extent *async_extent; | |
616 | u64 alloc_hint = 0; | |
617 | struct btrfs_trans_handle *trans; | |
618 | struct btrfs_key ins; | |
619 | struct extent_map *em; | |
620 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
621 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
622 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 623 | int ret = 0; |
771ed689 CM |
624 | |
625 | if (list_empty(&async_cow->extents)) | |
626 | return 0; | |
627 | ||
3e04e7f1 | 628 | again: |
d397712b | 629 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
630 | async_extent = list_entry(async_cow->extents.next, |
631 | struct async_extent, list); | |
632 | list_del(&async_extent->list); | |
c8b97818 | 633 | |
771ed689 CM |
634 | io_tree = &BTRFS_I(inode)->io_tree; |
635 | ||
f5a84ee3 | 636 | retry: |
771ed689 CM |
637 | /* did the compression code fall back to uncompressed IO? */ |
638 | if (!async_extent->pages) { | |
639 | int page_started = 0; | |
640 | unsigned long nr_written = 0; | |
641 | ||
642 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 643 | async_extent->start + |
d0082371 | 644 | async_extent->ram_size - 1); |
771ed689 CM |
645 | |
646 | /* allocate blocks */ | |
f5a84ee3 JB |
647 | ret = cow_file_range(inode, async_cow->locked_page, |
648 | async_extent->start, | |
649 | async_extent->start + | |
650 | async_extent->ram_size - 1, | |
651 | &page_started, &nr_written, 0); | |
771ed689 | 652 | |
79787eaa JM |
653 | /* JDM XXX */ |
654 | ||
771ed689 CM |
655 | /* |
656 | * if page_started, cow_file_range inserted an | |
657 | * inline extent and took care of all the unlocking | |
658 | * and IO for us. Otherwise, we need to submit | |
659 | * all those pages down to the drive. | |
660 | */ | |
f5a84ee3 | 661 | if (!page_started && !ret) |
771ed689 CM |
662 | extent_write_locked_range(io_tree, |
663 | inode, async_extent->start, | |
d397712b | 664 | async_extent->start + |
771ed689 CM |
665 | async_extent->ram_size - 1, |
666 | btrfs_get_extent, | |
667 | WB_SYNC_ALL); | |
3e04e7f1 JB |
668 | else if (ret) |
669 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
670 | kfree(async_extent); |
671 | cond_resched(); | |
672 | continue; | |
673 | } | |
674 | ||
675 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 676 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 677 | |
7a7eaa40 | 678 | trans = btrfs_join_transaction(root); |
79787eaa JM |
679 | if (IS_ERR(trans)) { |
680 | ret = PTR_ERR(trans); | |
681 | } else { | |
682 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
683 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
684 | async_extent->compressed_size, |
685 | async_extent->compressed_size, | |
81c9ad23 | 686 | 0, alloc_hint, &ins, 1); |
962197ba | 687 | if (ret && ret != -ENOSPC) |
79787eaa JM |
688 | btrfs_abort_transaction(trans, root, ret); |
689 | btrfs_end_transaction(trans, root); | |
690 | } | |
c2167754 | 691 | |
f5a84ee3 JB |
692 | if (ret) { |
693 | int i; | |
3e04e7f1 | 694 | |
f5a84ee3 JB |
695 | for (i = 0; i < async_extent->nr_pages; i++) { |
696 | WARN_ON(async_extent->pages[i]->mapping); | |
697 | page_cache_release(async_extent->pages[i]); | |
698 | } | |
699 | kfree(async_extent->pages); | |
700 | async_extent->nr_pages = 0; | |
701 | async_extent->pages = NULL; | |
3e04e7f1 | 702 | |
79787eaa JM |
703 | if (ret == -ENOSPC) |
704 | goto retry; | |
3e04e7f1 | 705 | goto out_free; |
f5a84ee3 JB |
706 | } |
707 | ||
c2167754 YZ |
708 | /* |
709 | * here we're doing allocation and writeback of the | |
710 | * compressed pages | |
711 | */ | |
712 | btrfs_drop_extent_cache(inode, async_extent->start, | |
713 | async_extent->start + | |
714 | async_extent->ram_size - 1, 0); | |
715 | ||
172ddd60 | 716 | em = alloc_extent_map(); |
b9aa55be LB |
717 | if (!em) { |
718 | ret = -ENOMEM; | |
3e04e7f1 | 719 | goto out_free_reserve; |
b9aa55be | 720 | } |
771ed689 CM |
721 | em->start = async_extent->start; |
722 | em->len = async_extent->ram_size; | |
445a6944 | 723 | em->orig_start = em->start; |
2ab28f32 JB |
724 | em->mod_start = em->start; |
725 | em->mod_len = em->len; | |
c8b97818 | 726 | |
771ed689 CM |
727 | em->block_start = ins.objectid; |
728 | em->block_len = ins.offset; | |
b4939680 | 729 | em->orig_block_len = ins.offset; |
cc95bef6 | 730 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 731 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 732 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
733 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
734 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 735 | em->generation = -1; |
771ed689 | 736 | |
d397712b | 737 | while (1) { |
890871be | 738 | write_lock(&em_tree->lock); |
09a2a8f9 | 739 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 740 | write_unlock(&em_tree->lock); |
771ed689 CM |
741 | if (ret != -EEXIST) { |
742 | free_extent_map(em); | |
743 | break; | |
744 | } | |
745 | btrfs_drop_extent_cache(inode, async_extent->start, | |
746 | async_extent->start + | |
747 | async_extent->ram_size - 1, 0); | |
748 | } | |
749 | ||
3e04e7f1 JB |
750 | if (ret) |
751 | goto out_free_reserve; | |
752 | ||
261507a0 LZ |
753 | ret = btrfs_add_ordered_extent_compress(inode, |
754 | async_extent->start, | |
755 | ins.objectid, | |
756 | async_extent->ram_size, | |
757 | ins.offset, | |
758 | BTRFS_ORDERED_COMPRESSED, | |
759 | async_extent->compress_type); | |
3e04e7f1 JB |
760 | if (ret) |
761 | goto out_free_reserve; | |
771ed689 | 762 | |
771ed689 CM |
763 | /* |
764 | * clear dirty, set writeback and unlock the pages. | |
765 | */ | |
766 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
767 | &BTRFS_I(inode)->io_tree, |
768 | async_extent->start, | |
769 | async_extent->start + | |
770 | async_extent->ram_size - 1, | |
771 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
772 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 773 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 774 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
775 | |
776 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
777 | async_extent->start, |
778 | async_extent->ram_size, | |
779 | ins.objectid, | |
780 | ins.offset, async_extent->pages, | |
781 | async_extent->nr_pages); | |
771ed689 CM |
782 | alloc_hint = ins.objectid + ins.offset; |
783 | kfree(async_extent); | |
3e04e7f1 JB |
784 | if (ret) |
785 | goto out; | |
771ed689 CM |
786 | cond_resched(); |
787 | } | |
79787eaa JM |
788 | ret = 0; |
789 | out: | |
790 | return ret; | |
3e04e7f1 JB |
791 | out_free_reserve: |
792 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa | 793 | out_free: |
3e04e7f1 JB |
794 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
795 | async_extent->start, | |
796 | async_extent->start + | |
797 | async_extent->ram_size - 1, | |
798 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
799 | EXTENT_CLEAR_UNLOCK | | |
800 | EXTENT_CLEAR_DELALLOC | | |
801 | EXTENT_CLEAR_DIRTY | | |
802 | EXTENT_SET_WRITEBACK | | |
803 | EXTENT_END_WRITEBACK); | |
79787eaa | 804 | kfree(async_extent); |
3e04e7f1 | 805 | goto again; |
771ed689 CM |
806 | } |
807 | ||
4b46fce2 JB |
808 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
809 | u64 num_bytes) | |
810 | { | |
811 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
812 | struct extent_map *em; | |
813 | u64 alloc_hint = 0; | |
814 | ||
815 | read_lock(&em_tree->lock); | |
816 | em = search_extent_mapping(em_tree, start, num_bytes); | |
817 | if (em) { | |
818 | /* | |
819 | * if block start isn't an actual block number then find the | |
820 | * first block in this inode and use that as a hint. If that | |
821 | * block is also bogus then just don't worry about it. | |
822 | */ | |
823 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
824 | free_extent_map(em); | |
825 | em = search_extent_mapping(em_tree, 0, 0); | |
826 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
827 | alloc_hint = em->block_start; | |
828 | if (em) | |
829 | free_extent_map(em); | |
830 | } else { | |
831 | alloc_hint = em->block_start; | |
832 | free_extent_map(em); | |
833 | } | |
834 | } | |
835 | read_unlock(&em_tree->lock); | |
836 | ||
837 | return alloc_hint; | |
838 | } | |
839 | ||
771ed689 CM |
840 | /* |
841 | * when extent_io.c finds a delayed allocation range in the file, | |
842 | * the call backs end up in this code. The basic idea is to | |
843 | * allocate extents on disk for the range, and create ordered data structs | |
844 | * in ram to track those extents. | |
845 | * | |
846 | * locked_page is the page that writepage had locked already. We use | |
847 | * it to make sure we don't do extra locks or unlocks. | |
848 | * | |
849 | * *page_started is set to one if we unlock locked_page and do everything | |
850 | * required to start IO on it. It may be clean and already done with | |
851 | * IO when we return. | |
852 | */ | |
b7d5b0a8 MX |
853 | static noinline int __cow_file_range(struct btrfs_trans_handle *trans, |
854 | struct inode *inode, | |
855 | struct btrfs_root *root, | |
856 | struct page *locked_page, | |
857 | u64 start, u64 end, int *page_started, | |
858 | unsigned long *nr_written, | |
859 | int unlock) | |
771ed689 | 860 | { |
771ed689 CM |
861 | u64 alloc_hint = 0; |
862 | u64 num_bytes; | |
863 | unsigned long ram_size; | |
864 | u64 disk_num_bytes; | |
865 | u64 cur_alloc_size; | |
866 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
867 | struct btrfs_key ins; |
868 | struct extent_map *em; | |
869 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
870 | int ret = 0; | |
871 | ||
83eea1f1 | 872 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 873 | |
fda2832f | 874 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
875 | num_bytes = max(blocksize, num_bytes); |
876 | disk_num_bytes = num_bytes; | |
771ed689 | 877 | |
4cb5300b | 878 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
879 | if (num_bytes < 64 * 1024 && |
880 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
881 | btrfs_add_inode_defrag(trans, inode); |
882 | ||
771ed689 CM |
883 | if (start == 0) { |
884 | /* lets try to make an inline extent */ | |
885 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 886 | start, end, 0, 0, NULL); |
771ed689 CM |
887 | if (ret == 0) { |
888 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
889 | &BTRFS_I(inode)->io_tree, |
890 | start, end, NULL, | |
891 | EXTENT_CLEAR_UNLOCK_PAGE | | |
892 | EXTENT_CLEAR_UNLOCK | | |
893 | EXTENT_CLEAR_DELALLOC | | |
894 | EXTENT_CLEAR_DIRTY | | |
895 | EXTENT_SET_WRITEBACK | | |
896 | EXTENT_END_WRITEBACK); | |
c2167754 | 897 | |
771ed689 CM |
898 | *nr_written = *nr_written + |
899 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
900 | *page_started = 1; | |
771ed689 | 901 | goto out; |
79787eaa JM |
902 | } else if (ret < 0) { |
903 | btrfs_abort_transaction(trans, root, ret); | |
904 | goto out_unlock; | |
771ed689 CM |
905 | } |
906 | } | |
907 | ||
908 | BUG_ON(disk_num_bytes > | |
6c41761f | 909 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 910 | |
4b46fce2 | 911 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
912 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
913 | ||
d397712b | 914 | while (disk_num_bytes > 0) { |
a791e35e CM |
915 | unsigned long op; |
916 | ||
287a0ab9 | 917 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 918 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 919 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 920 | &ins, 1); |
79787eaa JM |
921 | if (ret < 0) { |
922 | btrfs_abort_transaction(trans, root, ret); | |
923 | goto out_unlock; | |
924 | } | |
d397712b | 925 | |
172ddd60 | 926 | em = alloc_extent_map(); |
b9aa55be LB |
927 | if (!em) { |
928 | ret = -ENOMEM; | |
ace68bac | 929 | goto out_reserve; |
b9aa55be | 930 | } |
e6dcd2dc | 931 | em->start = start; |
445a6944 | 932 | em->orig_start = em->start; |
771ed689 CM |
933 | ram_size = ins.offset; |
934 | em->len = ins.offset; | |
2ab28f32 JB |
935 | em->mod_start = em->start; |
936 | em->mod_len = em->len; | |
c8b97818 | 937 | |
e6dcd2dc | 938 | em->block_start = ins.objectid; |
c8b97818 | 939 | em->block_len = ins.offset; |
b4939680 | 940 | em->orig_block_len = ins.offset; |
cc95bef6 | 941 | em->ram_bytes = ram_size; |
e6dcd2dc | 942 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 943 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 944 | em->generation = -1; |
c8b97818 | 945 | |
d397712b | 946 | while (1) { |
890871be | 947 | write_lock(&em_tree->lock); |
09a2a8f9 | 948 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 949 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
950 | if (ret != -EEXIST) { |
951 | free_extent_map(em); | |
952 | break; | |
953 | } | |
954 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 955 | start + ram_size - 1, 0); |
e6dcd2dc | 956 | } |
ace68bac LB |
957 | if (ret) |
958 | goto out_reserve; | |
e6dcd2dc | 959 | |
98d20f67 | 960 | cur_alloc_size = ins.offset; |
e6dcd2dc | 961 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 962 | ram_size, cur_alloc_size, 0); |
ace68bac LB |
963 | if (ret) |
964 | goto out_reserve; | |
c8b97818 | 965 | |
17d217fe YZ |
966 | if (root->root_key.objectid == |
967 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
968 | ret = btrfs_reloc_clone_csums(inode, start, | |
969 | cur_alloc_size); | |
79787eaa JM |
970 | if (ret) { |
971 | btrfs_abort_transaction(trans, root, ret); | |
ace68bac | 972 | goto out_reserve; |
79787eaa | 973 | } |
17d217fe YZ |
974 | } |
975 | ||
d397712b | 976 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 977 | break; |
d397712b | 978 | |
c8b97818 CM |
979 | /* we're not doing compressed IO, don't unlock the first |
980 | * page (which the caller expects to stay locked), don't | |
981 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
982 | * |
983 | * Do set the Private2 bit so we know this page was properly | |
984 | * setup for writepage | |
c8b97818 | 985 | */ |
a791e35e CM |
986 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
987 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
988 | EXTENT_SET_PRIVATE2; | |
989 | ||
c8b97818 CM |
990 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
991 | start, start + ram_size - 1, | |
a791e35e | 992 | locked_page, op); |
c8b97818 | 993 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
994 | num_bytes -= cur_alloc_size; |
995 | alloc_hint = ins.objectid + ins.offset; | |
996 | start += cur_alloc_size; | |
b888db2b | 997 | } |
79787eaa | 998 | out: |
be20aa9d | 999 | return ret; |
b7d5b0a8 | 1000 | |
ace68bac LB |
1001 | out_reserve: |
1002 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
79787eaa JM |
1003 | out_unlock: |
1004 | extent_clear_unlock_delalloc(inode, | |
1005 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 1006 | start, end, locked_page, |
79787eaa JM |
1007 | EXTENT_CLEAR_UNLOCK_PAGE | |
1008 | EXTENT_CLEAR_UNLOCK | | |
1009 | EXTENT_CLEAR_DELALLOC | | |
1010 | EXTENT_CLEAR_DIRTY | | |
1011 | EXTENT_SET_WRITEBACK | | |
1012 | EXTENT_END_WRITEBACK); | |
1013 | ||
1014 | goto out; | |
771ed689 | 1015 | } |
c8b97818 | 1016 | |
b7d5b0a8 MX |
1017 | static noinline int cow_file_range(struct inode *inode, |
1018 | struct page *locked_page, | |
1019 | u64 start, u64 end, int *page_started, | |
1020 | unsigned long *nr_written, | |
1021 | int unlock) | |
1022 | { | |
1023 | struct btrfs_trans_handle *trans; | |
1024 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1025 | int ret; | |
1026 | ||
1027 | trans = btrfs_join_transaction(root); | |
1028 | if (IS_ERR(trans)) { | |
1029 | extent_clear_unlock_delalloc(inode, | |
1030 | &BTRFS_I(inode)->io_tree, | |
1031 | start, end, locked_page, | |
1032 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1033 | EXTENT_CLEAR_UNLOCK | | |
1034 | EXTENT_CLEAR_DELALLOC | | |
1035 | EXTENT_CLEAR_DIRTY | | |
1036 | EXTENT_SET_WRITEBACK | | |
1037 | EXTENT_END_WRITEBACK); | |
1038 | return PTR_ERR(trans); | |
1039 | } | |
1040 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
1041 | ||
1042 | ret = __cow_file_range(trans, inode, root, locked_page, start, end, | |
1043 | page_started, nr_written, unlock); | |
1044 | ||
1045 | btrfs_end_transaction(trans, root); | |
1046 | ||
1047 | return ret; | |
1048 | } | |
1049 | ||
771ed689 CM |
1050 | /* |
1051 | * work queue call back to started compression on a file and pages | |
1052 | */ | |
1053 | static noinline void async_cow_start(struct btrfs_work *work) | |
1054 | { | |
1055 | struct async_cow *async_cow; | |
1056 | int num_added = 0; | |
1057 | async_cow = container_of(work, struct async_cow, work); | |
1058 | ||
1059 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1060 | async_cow->start, async_cow->end, async_cow, | |
1061 | &num_added); | |
8180ef88 | 1062 | if (num_added == 0) { |
cb77fcd8 | 1063 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1064 | async_cow->inode = NULL; |
8180ef88 | 1065 | } |
771ed689 CM |
1066 | } |
1067 | ||
1068 | /* | |
1069 | * work queue call back to submit previously compressed pages | |
1070 | */ | |
1071 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1072 | { | |
1073 | struct async_cow *async_cow; | |
1074 | struct btrfs_root *root; | |
1075 | unsigned long nr_pages; | |
1076 | ||
1077 | async_cow = container_of(work, struct async_cow, work); | |
1078 | ||
1079 | root = async_cow->root; | |
1080 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1081 | PAGE_CACHE_SHIFT; | |
1082 | ||
66657b31 | 1083 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1084 | 5 * 1024 * 1024 && |
771ed689 CM |
1085 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1086 | wake_up(&root->fs_info->async_submit_wait); | |
1087 | ||
d397712b | 1088 | if (async_cow->inode) |
771ed689 | 1089 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1090 | } |
c8b97818 | 1091 | |
771ed689 CM |
1092 | static noinline void async_cow_free(struct btrfs_work *work) |
1093 | { | |
1094 | struct async_cow *async_cow; | |
1095 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1096 | if (async_cow->inode) |
cb77fcd8 | 1097 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1098 | kfree(async_cow); |
1099 | } | |
1100 | ||
1101 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1102 | u64 start, u64 end, int *page_started, | |
1103 | unsigned long *nr_written) | |
1104 | { | |
1105 | struct async_cow *async_cow; | |
1106 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1107 | unsigned long nr_pages; | |
1108 | u64 cur_end; | |
287082b0 | 1109 | int limit = 10 * 1024 * 1024; |
771ed689 | 1110 | |
a3429ab7 CM |
1111 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1112 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1113 | while (start < end) { |
771ed689 | 1114 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1115 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1116 | async_cow->inode = igrab(inode); |
771ed689 CM |
1117 | async_cow->root = root; |
1118 | async_cow->locked_page = locked_page; | |
1119 | async_cow->start = start; | |
1120 | ||
6cbff00f | 1121 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1122 | cur_end = end; |
1123 | else | |
1124 | cur_end = min(end, start + 512 * 1024 - 1); | |
1125 | ||
1126 | async_cow->end = cur_end; | |
1127 | INIT_LIST_HEAD(&async_cow->extents); | |
1128 | ||
1129 | async_cow->work.func = async_cow_start; | |
1130 | async_cow->work.ordered_func = async_cow_submit; | |
1131 | async_cow->work.ordered_free = async_cow_free; | |
1132 | async_cow->work.flags = 0; | |
1133 | ||
771ed689 CM |
1134 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1135 | PAGE_CACHE_SHIFT; | |
1136 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1137 | ||
1138 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1139 | &async_cow->work); | |
1140 | ||
1141 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1142 | wait_event(root->fs_info->async_submit_wait, | |
1143 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1144 | limit)); | |
1145 | } | |
1146 | ||
d397712b | 1147 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1148 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1149 | wait_event(root->fs_info->async_submit_wait, | |
1150 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1151 | 0)); | |
1152 | } | |
1153 | ||
1154 | *nr_written += nr_pages; | |
1155 | start = cur_end + 1; | |
1156 | } | |
1157 | *page_started = 1; | |
1158 | return 0; | |
be20aa9d CM |
1159 | } |
1160 | ||
d397712b | 1161 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1162 | u64 bytenr, u64 num_bytes) |
1163 | { | |
1164 | int ret; | |
1165 | struct btrfs_ordered_sum *sums; | |
1166 | LIST_HEAD(list); | |
1167 | ||
07d400a6 | 1168 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1169 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1170 | if (ret == 0 && list_empty(&list)) |
1171 | return 0; | |
1172 | ||
1173 | while (!list_empty(&list)) { | |
1174 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1175 | list_del(&sums->list); | |
1176 | kfree(sums); | |
1177 | } | |
1178 | return 1; | |
1179 | } | |
1180 | ||
d352ac68 CM |
1181 | /* |
1182 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1183 | * of the extents that exist in the file, and COWs the file as required. | |
1184 | * | |
1185 | * If no cow copies or snapshots exist, we write directly to the existing | |
1186 | * blocks on disk | |
1187 | */ | |
7f366cfe CM |
1188 | static noinline int run_delalloc_nocow(struct inode *inode, |
1189 | struct page *locked_page, | |
771ed689 CM |
1190 | u64 start, u64 end, int *page_started, int force, |
1191 | unsigned long *nr_written) | |
be20aa9d | 1192 | { |
be20aa9d | 1193 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1194 | struct btrfs_trans_handle *trans; |
be20aa9d | 1195 | struct extent_buffer *leaf; |
be20aa9d | 1196 | struct btrfs_path *path; |
80ff3856 | 1197 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1198 | struct btrfs_key found_key; |
80ff3856 YZ |
1199 | u64 cow_start; |
1200 | u64 cur_offset; | |
1201 | u64 extent_end; | |
5d4f98a2 | 1202 | u64 extent_offset; |
80ff3856 YZ |
1203 | u64 disk_bytenr; |
1204 | u64 num_bytes; | |
b4939680 | 1205 | u64 disk_num_bytes; |
cc95bef6 | 1206 | u64 ram_bytes; |
80ff3856 | 1207 | int extent_type; |
79787eaa | 1208 | int ret, err; |
d899e052 | 1209 | int type; |
80ff3856 YZ |
1210 | int nocow; |
1211 | int check_prev = 1; | |
82d5902d | 1212 | bool nolock; |
33345d01 | 1213 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1214 | |
1215 | path = btrfs_alloc_path(); | |
17ca04af JB |
1216 | if (!path) { |
1217 | extent_clear_unlock_delalloc(inode, | |
1218 | &BTRFS_I(inode)->io_tree, | |
1219 | start, end, locked_page, | |
1220 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1221 | EXTENT_CLEAR_UNLOCK | | |
1222 | EXTENT_CLEAR_DELALLOC | | |
1223 | EXTENT_CLEAR_DIRTY | | |
1224 | EXTENT_SET_WRITEBACK | | |
1225 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1226 | return -ENOMEM; |
17ca04af | 1227 | } |
82d5902d | 1228 | |
83eea1f1 | 1229 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1230 | |
1231 | if (nolock) | |
7a7eaa40 | 1232 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1233 | else |
7a7eaa40 | 1234 | trans = btrfs_join_transaction(root); |
ff5714cc | 1235 | |
79787eaa | 1236 | if (IS_ERR(trans)) { |
17ca04af JB |
1237 | extent_clear_unlock_delalloc(inode, |
1238 | &BTRFS_I(inode)->io_tree, | |
1239 | start, end, locked_page, | |
1240 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1241 | EXTENT_CLEAR_UNLOCK | | |
1242 | EXTENT_CLEAR_DELALLOC | | |
1243 | EXTENT_CLEAR_DIRTY | | |
1244 | EXTENT_SET_WRITEBACK | | |
1245 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1246 | btrfs_free_path(path); |
1247 | return PTR_ERR(trans); | |
1248 | } | |
1249 | ||
74b21075 | 1250 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1251 | |
80ff3856 YZ |
1252 | cow_start = (u64)-1; |
1253 | cur_offset = start; | |
1254 | while (1) { | |
33345d01 | 1255 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1256 | cur_offset, 0); |
79787eaa JM |
1257 | if (ret < 0) { |
1258 | btrfs_abort_transaction(trans, root, ret); | |
1259 | goto error; | |
1260 | } | |
80ff3856 YZ |
1261 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1262 | leaf = path->nodes[0]; | |
1263 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1264 | path->slots[0] - 1); | |
33345d01 | 1265 | if (found_key.objectid == ino && |
80ff3856 YZ |
1266 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1267 | path->slots[0]--; | |
1268 | } | |
1269 | check_prev = 0; | |
1270 | next_slot: | |
1271 | leaf = path->nodes[0]; | |
1272 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1273 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1274 | if (ret < 0) { |
1275 | btrfs_abort_transaction(trans, root, ret); | |
1276 | goto error; | |
1277 | } | |
80ff3856 YZ |
1278 | if (ret > 0) |
1279 | break; | |
1280 | leaf = path->nodes[0]; | |
1281 | } | |
be20aa9d | 1282 | |
80ff3856 YZ |
1283 | nocow = 0; |
1284 | disk_bytenr = 0; | |
17d217fe | 1285 | num_bytes = 0; |
80ff3856 YZ |
1286 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1287 | ||
33345d01 | 1288 | if (found_key.objectid > ino || |
80ff3856 YZ |
1289 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1290 | found_key.offset > end) | |
1291 | break; | |
1292 | ||
1293 | if (found_key.offset > cur_offset) { | |
1294 | extent_end = found_key.offset; | |
e9061e21 | 1295 | extent_type = 0; |
80ff3856 YZ |
1296 | goto out_check; |
1297 | } | |
1298 | ||
1299 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1300 | struct btrfs_file_extent_item); | |
1301 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1302 | ||
cc95bef6 | 1303 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1304 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1305 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1306 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1307 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1308 | extent_end = found_key.offset + |
1309 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1310 | disk_num_bytes = |
1311 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1312 | if (extent_end <= start) { |
1313 | path->slots[0]++; | |
1314 | goto next_slot; | |
1315 | } | |
17d217fe YZ |
1316 | if (disk_bytenr == 0) |
1317 | goto out_check; | |
80ff3856 YZ |
1318 | if (btrfs_file_extent_compression(leaf, fi) || |
1319 | btrfs_file_extent_encryption(leaf, fi) || | |
1320 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1321 | goto out_check; | |
d899e052 YZ |
1322 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1323 | goto out_check; | |
d2fb3437 | 1324 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1325 | goto out_check; |
33345d01 | 1326 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1327 | found_key.offset - |
1328 | extent_offset, disk_bytenr)) | |
17d217fe | 1329 | goto out_check; |
5d4f98a2 | 1330 | disk_bytenr += extent_offset; |
17d217fe YZ |
1331 | disk_bytenr += cur_offset - found_key.offset; |
1332 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1333 | /* | |
1334 | * force cow if csum exists in the range. | |
1335 | * this ensure that csum for a given extent are | |
1336 | * either valid or do not exist. | |
1337 | */ | |
1338 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1339 | goto out_check; | |
80ff3856 YZ |
1340 | nocow = 1; |
1341 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1342 | extent_end = found_key.offset + | |
1343 | btrfs_file_extent_inline_len(leaf, fi); | |
1344 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1345 | } else { | |
1346 | BUG_ON(1); | |
1347 | } | |
1348 | out_check: | |
1349 | if (extent_end <= start) { | |
1350 | path->slots[0]++; | |
1351 | goto next_slot; | |
1352 | } | |
1353 | if (!nocow) { | |
1354 | if (cow_start == (u64)-1) | |
1355 | cow_start = cur_offset; | |
1356 | cur_offset = extent_end; | |
1357 | if (cur_offset > end) | |
1358 | break; | |
1359 | path->slots[0]++; | |
1360 | goto next_slot; | |
7ea394f1 YZ |
1361 | } |
1362 | ||
b3b4aa74 | 1363 | btrfs_release_path(path); |
80ff3856 | 1364 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1365 | ret = __cow_file_range(trans, inode, root, locked_page, |
1366 | cow_start, found_key.offset - 1, | |
1367 | page_started, nr_written, 1); | |
79787eaa JM |
1368 | if (ret) { |
1369 | btrfs_abort_transaction(trans, root, ret); | |
1370 | goto error; | |
1371 | } | |
80ff3856 | 1372 | cow_start = (u64)-1; |
7ea394f1 | 1373 | } |
80ff3856 | 1374 | |
d899e052 YZ |
1375 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1376 | struct extent_map *em; | |
1377 | struct extent_map_tree *em_tree; | |
1378 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1379 | em = alloc_extent_map(); |
79787eaa | 1380 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1381 | em->start = cur_offset; |
70c8a91c | 1382 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1383 | em->len = num_bytes; |
1384 | em->block_len = num_bytes; | |
1385 | em->block_start = disk_bytenr; | |
b4939680 | 1386 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1387 | em->ram_bytes = ram_bytes; |
d899e052 | 1388 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1389 | em->mod_start = em->start; |
1390 | em->mod_len = em->len; | |
d899e052 | 1391 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1392 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1393 | em->generation = -1; |
d899e052 | 1394 | while (1) { |
890871be | 1395 | write_lock(&em_tree->lock); |
09a2a8f9 | 1396 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1397 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1398 | if (ret != -EEXIST) { |
1399 | free_extent_map(em); | |
1400 | break; | |
1401 | } | |
1402 | btrfs_drop_extent_cache(inode, em->start, | |
1403 | em->start + em->len - 1, 0); | |
1404 | } | |
1405 | type = BTRFS_ORDERED_PREALLOC; | |
1406 | } else { | |
1407 | type = BTRFS_ORDERED_NOCOW; | |
1408 | } | |
80ff3856 YZ |
1409 | |
1410 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1411 | num_bytes, num_bytes, type); |
79787eaa | 1412 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1413 | |
efa56464 YZ |
1414 | if (root->root_key.objectid == |
1415 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1416 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1417 | num_bytes); | |
79787eaa JM |
1418 | if (ret) { |
1419 | btrfs_abort_transaction(trans, root, ret); | |
1420 | goto error; | |
1421 | } | |
efa56464 YZ |
1422 | } |
1423 | ||
d899e052 | 1424 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1425 | cur_offset, cur_offset + num_bytes - 1, |
1426 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1427 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1428 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1429 | cur_offset = extent_end; |
1430 | if (cur_offset > end) | |
1431 | break; | |
be20aa9d | 1432 | } |
b3b4aa74 | 1433 | btrfs_release_path(path); |
80ff3856 | 1434 | |
17ca04af | 1435 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1436 | cow_start = cur_offset; |
17ca04af JB |
1437 | cur_offset = end; |
1438 | } | |
1439 | ||
80ff3856 | 1440 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1441 | ret = __cow_file_range(trans, inode, root, locked_page, |
1442 | cow_start, end, | |
1443 | page_started, nr_written, 1); | |
79787eaa JM |
1444 | if (ret) { |
1445 | btrfs_abort_transaction(trans, root, ret); | |
1446 | goto error; | |
1447 | } | |
80ff3856 YZ |
1448 | } |
1449 | ||
79787eaa | 1450 | error: |
a698d075 | 1451 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1452 | if (!ret) |
1453 | ret = err; | |
1454 | ||
17ca04af JB |
1455 | if (ret && cur_offset < end) |
1456 | extent_clear_unlock_delalloc(inode, | |
1457 | &BTRFS_I(inode)->io_tree, | |
1458 | cur_offset, end, locked_page, | |
1459 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1460 | EXTENT_CLEAR_UNLOCK | | |
1461 | EXTENT_CLEAR_DELALLOC | | |
1462 | EXTENT_CLEAR_DIRTY | | |
1463 | EXTENT_SET_WRITEBACK | | |
1464 | EXTENT_END_WRITEBACK); | |
1465 | ||
7ea394f1 | 1466 | btrfs_free_path(path); |
79787eaa | 1467 | return ret; |
be20aa9d CM |
1468 | } |
1469 | ||
d352ac68 CM |
1470 | /* |
1471 | * extent_io.c call back to do delayed allocation processing | |
1472 | */ | |
c8b97818 | 1473 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1474 | u64 start, u64 end, int *page_started, |
1475 | unsigned long *nr_written) | |
be20aa9d | 1476 | { |
be20aa9d | 1477 | int ret; |
7f366cfe | 1478 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1479 | |
7ddf5a42 | 1480 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1481 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1482 | page_started, 1, nr_written); |
7ddf5a42 | 1483 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1484 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1485 | page_started, 0, nr_written); |
7ddf5a42 JB |
1486 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1487 | !(BTRFS_I(inode)->force_compress) && | |
1488 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1489 | ret = cow_file_range(inode, locked_page, start, end, |
1490 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1491 | } else { |
1492 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1493 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1494 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1495 | page_started, nr_written); |
7ddf5a42 | 1496 | } |
b888db2b CM |
1497 | return ret; |
1498 | } | |
1499 | ||
1bf85046 JM |
1500 | static void btrfs_split_extent_hook(struct inode *inode, |
1501 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1502 | { |
0ca1f7ce | 1503 | /* not delalloc, ignore it */ |
9ed74f2d | 1504 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1505 | return; |
9ed74f2d | 1506 | |
9e0baf60 JB |
1507 | spin_lock(&BTRFS_I(inode)->lock); |
1508 | BTRFS_I(inode)->outstanding_extents++; | |
1509 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1510 | } |
1511 | ||
1512 | /* | |
1513 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1514 | * extents so we can keep track of new extents that are just merged onto old | |
1515 | * extents, such as when we are doing sequential writes, so we can properly | |
1516 | * account for the metadata space we'll need. | |
1517 | */ | |
1bf85046 JM |
1518 | static void btrfs_merge_extent_hook(struct inode *inode, |
1519 | struct extent_state *new, | |
1520 | struct extent_state *other) | |
9ed74f2d | 1521 | { |
9ed74f2d JB |
1522 | /* not delalloc, ignore it */ |
1523 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1524 | return; |
9ed74f2d | 1525 | |
9e0baf60 JB |
1526 | spin_lock(&BTRFS_I(inode)->lock); |
1527 | BTRFS_I(inode)->outstanding_extents--; | |
1528 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1529 | } |
1530 | ||
eb73c1b7 MX |
1531 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1532 | struct inode *inode) | |
1533 | { | |
1534 | spin_lock(&root->delalloc_lock); | |
1535 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1536 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1537 | &root->delalloc_inodes); | |
1538 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1539 | &BTRFS_I(inode)->runtime_flags); | |
1540 | root->nr_delalloc_inodes++; | |
1541 | if (root->nr_delalloc_inodes == 1) { | |
1542 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1543 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1544 | list_add_tail(&root->delalloc_root, | |
1545 | &root->fs_info->delalloc_roots); | |
1546 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1547 | } | |
1548 | } | |
1549 | spin_unlock(&root->delalloc_lock); | |
1550 | } | |
1551 | ||
1552 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1553 | struct inode *inode) | |
1554 | { | |
1555 | spin_lock(&root->delalloc_lock); | |
1556 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1557 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1558 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1559 | &BTRFS_I(inode)->runtime_flags); | |
1560 | root->nr_delalloc_inodes--; | |
1561 | if (!root->nr_delalloc_inodes) { | |
1562 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1563 | BUG_ON(list_empty(&root->delalloc_root)); | |
1564 | list_del_init(&root->delalloc_root); | |
1565 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1566 | } | |
1567 | } | |
1568 | spin_unlock(&root->delalloc_lock); | |
1569 | } | |
1570 | ||
d352ac68 CM |
1571 | /* |
1572 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1573 | * bytes in this file, and to maintain the list of inodes that | |
1574 | * have pending delalloc work to be done. | |
1575 | */ | |
1bf85046 | 1576 | static void btrfs_set_bit_hook(struct inode *inode, |
41074888 | 1577 | struct extent_state *state, unsigned long *bits) |
291d673e | 1578 | { |
9ed74f2d | 1579 | |
75eff68e CM |
1580 | /* |
1581 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1582 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1583 | * bit, which is only set or cleared with irqs on |
1584 | */ | |
0ca1f7ce | 1585 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1586 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1587 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1588 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1589 | |
9e0baf60 | 1590 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1591 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1592 | } else { |
1593 | spin_lock(&BTRFS_I(inode)->lock); | |
1594 | BTRFS_I(inode)->outstanding_extents++; | |
1595 | spin_unlock(&BTRFS_I(inode)->lock); | |
1596 | } | |
287a0ab9 | 1597 | |
963d678b MX |
1598 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1599 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1600 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1601 | BTRFS_I(inode)->delalloc_bytes += len; |
df0af1a5 | 1602 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1603 | &BTRFS_I(inode)->runtime_flags)) |
1604 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1605 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1606 | } |
291d673e CM |
1607 | } |
1608 | ||
d352ac68 CM |
1609 | /* |
1610 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1611 | */ | |
1bf85046 | 1612 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 DS |
1613 | struct extent_state *state, |
1614 | unsigned long *bits) | |
291d673e | 1615 | { |
75eff68e CM |
1616 | /* |
1617 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1618 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1619 | * bit, which is only set or cleared with irqs on |
1620 | */ | |
0ca1f7ce | 1621 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1622 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1623 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1624 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1625 | |
9e0baf60 | 1626 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1627 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1628 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1629 | spin_lock(&BTRFS_I(inode)->lock); | |
1630 | BTRFS_I(inode)->outstanding_extents--; | |
1631 | spin_unlock(&BTRFS_I(inode)->lock); | |
1632 | } | |
0ca1f7ce YZ |
1633 | |
1634 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1635 | btrfs_delalloc_release_metadata(inode, len); | |
1636 | ||
0cb59c99 JB |
1637 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1638 | && do_list) | |
0ca1f7ce | 1639 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1640 | |
963d678b MX |
1641 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1642 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1643 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1644 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1645 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1646 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1647 | &BTRFS_I(inode)->runtime_flags)) |
1648 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1649 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1650 | } |
291d673e CM |
1651 | } |
1652 | ||
d352ac68 CM |
1653 | /* |
1654 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1655 | * we don't create bios that span stripes or chunks | |
1656 | */ | |
64a16701 | 1657 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1658 | size_t size, struct bio *bio, |
1659 | unsigned long bio_flags) | |
239b14b3 CM |
1660 | { |
1661 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1662 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1663 | u64 length = 0; |
1664 | u64 map_length; | |
239b14b3 CM |
1665 | int ret; |
1666 | ||
771ed689 CM |
1667 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1668 | return 0; | |
1669 | ||
f2d8d74d | 1670 | length = bio->bi_size; |
239b14b3 | 1671 | map_length = length; |
64a16701 | 1672 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1673 | &map_length, NULL, 0); |
3ec706c8 | 1674 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1675 | BUG_ON(ret < 0); |
d397712b | 1676 | if (map_length < length + size) |
239b14b3 | 1677 | return 1; |
3444a972 | 1678 | return 0; |
239b14b3 CM |
1679 | } |
1680 | ||
d352ac68 CM |
1681 | /* |
1682 | * in order to insert checksums into the metadata in large chunks, | |
1683 | * we wait until bio submission time. All the pages in the bio are | |
1684 | * checksummed and sums are attached onto the ordered extent record. | |
1685 | * | |
1686 | * At IO completion time the cums attached on the ordered extent record | |
1687 | * are inserted into the btree | |
1688 | */ | |
d397712b CM |
1689 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1690 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1691 | unsigned long bio_flags, |
1692 | u64 bio_offset) | |
065631f6 | 1693 | { |
065631f6 | 1694 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1695 | int ret = 0; |
e015640f | 1696 | |
d20f7043 | 1697 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1698 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1699 | return 0; |
1700 | } | |
e015640f | 1701 | |
4a69a410 CM |
1702 | /* |
1703 | * in order to insert checksums into the metadata in large chunks, | |
1704 | * we wait until bio submission time. All the pages in the bio are | |
1705 | * checksummed and sums are attached onto the ordered extent record. | |
1706 | * | |
1707 | * At IO completion time the cums attached on the ordered extent record | |
1708 | * are inserted into the btree | |
1709 | */ | |
b2950863 | 1710 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1711 | int mirror_num, unsigned long bio_flags, |
1712 | u64 bio_offset) | |
4a69a410 CM |
1713 | { |
1714 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1715 | int ret; |
1716 | ||
1717 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1718 | if (ret) | |
1719 | bio_endio(bio, ret); | |
1720 | return ret; | |
44b8bd7e CM |
1721 | } |
1722 | ||
d352ac68 | 1723 | /* |
cad321ad CM |
1724 | * extent_io.c submission hook. This does the right thing for csum calculation |
1725 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1726 | */ |
b2950863 | 1727 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1728 | int mirror_num, unsigned long bio_flags, |
1729 | u64 bio_offset) | |
44b8bd7e CM |
1730 | { |
1731 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1732 | int ret = 0; | |
19b9bdb0 | 1733 | int skip_sum; |
0417341e | 1734 | int metadata = 0; |
b812ce28 | 1735 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1736 | |
6cbff00f | 1737 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1738 | |
83eea1f1 | 1739 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1740 | metadata = 2; |
1741 | ||
7b6d91da | 1742 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1743 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1744 | if (ret) | |
61891923 | 1745 | goto out; |
5fd02043 | 1746 | |
d20f7043 | 1747 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1748 | ret = btrfs_submit_compressed_read(inode, bio, |
1749 | mirror_num, | |
1750 | bio_flags); | |
1751 | goto out; | |
c2db1073 TI |
1752 | } else if (!skip_sum) { |
1753 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1754 | if (ret) | |
61891923 | 1755 | goto out; |
c2db1073 | 1756 | } |
4d1b5fb4 | 1757 | goto mapit; |
b812ce28 | 1758 | } else if (async && !skip_sum) { |
17d217fe YZ |
1759 | /* csum items have already been cloned */ |
1760 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1761 | goto mapit; | |
19b9bdb0 | 1762 | /* we're doing a write, do the async checksumming */ |
61891923 | 1763 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1764 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1765 | bio_flags, bio_offset, |
1766 | __btrfs_submit_bio_start, | |
4a69a410 | 1767 | __btrfs_submit_bio_done); |
61891923 | 1768 | goto out; |
b812ce28 JB |
1769 | } else if (!skip_sum) { |
1770 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1771 | if (ret) | |
1772 | goto out; | |
19b9bdb0 CM |
1773 | } |
1774 | ||
0b86a832 | 1775 | mapit: |
61891923 SB |
1776 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1777 | ||
1778 | out: | |
1779 | if (ret < 0) | |
1780 | bio_endio(bio, ret); | |
1781 | return ret; | |
065631f6 | 1782 | } |
6885f308 | 1783 | |
d352ac68 CM |
1784 | /* |
1785 | * given a list of ordered sums record them in the inode. This happens | |
1786 | * at IO completion time based on sums calculated at bio submission time. | |
1787 | */ | |
ba1da2f4 | 1788 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1789 | struct inode *inode, u64 file_offset, |
1790 | struct list_head *list) | |
1791 | { | |
e6dcd2dc CM |
1792 | struct btrfs_ordered_sum *sum; |
1793 | ||
c6e30871 | 1794 | list_for_each_entry(sum, list, list) { |
39847c4d | 1795 | trans->adding_csums = 1; |
d20f7043 CM |
1796 | btrfs_csum_file_blocks(trans, |
1797 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1798 | trans->adding_csums = 0; |
e6dcd2dc CM |
1799 | } |
1800 | return 0; | |
1801 | } | |
1802 | ||
2ac55d41 JB |
1803 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1804 | struct extent_state **cached_state) | |
ea8c2819 | 1805 | { |
6c1500f2 | 1806 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1807 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1808 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1809 | } |
1810 | ||
d352ac68 | 1811 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1812 | struct btrfs_writepage_fixup { |
1813 | struct page *page; | |
1814 | struct btrfs_work work; | |
1815 | }; | |
1816 | ||
b2950863 | 1817 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1818 | { |
1819 | struct btrfs_writepage_fixup *fixup; | |
1820 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1821 | struct extent_state *cached_state = NULL; |
247e743c CM |
1822 | struct page *page; |
1823 | struct inode *inode; | |
1824 | u64 page_start; | |
1825 | u64 page_end; | |
87826df0 | 1826 | int ret; |
247e743c CM |
1827 | |
1828 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1829 | page = fixup->page; | |
4a096752 | 1830 | again: |
247e743c CM |
1831 | lock_page(page); |
1832 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1833 | ClearPageChecked(page); | |
1834 | goto out_page; | |
1835 | } | |
1836 | ||
1837 | inode = page->mapping->host; | |
1838 | page_start = page_offset(page); | |
1839 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1840 | ||
2ac55d41 | 1841 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1842 | &cached_state); |
4a096752 CM |
1843 | |
1844 | /* already ordered? We're done */ | |
8b62b72b | 1845 | if (PagePrivate2(page)) |
247e743c | 1846 | goto out; |
4a096752 CM |
1847 | |
1848 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1849 | if (ordered) { | |
2ac55d41 JB |
1850 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1851 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1852 | unlock_page(page); |
1853 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1854 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1855 | goto again; |
1856 | } | |
247e743c | 1857 | |
87826df0 JM |
1858 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1859 | if (ret) { | |
1860 | mapping_set_error(page->mapping, ret); | |
1861 | end_extent_writepage(page, ret, page_start, page_end); | |
1862 | ClearPageChecked(page); | |
1863 | goto out; | |
1864 | } | |
1865 | ||
2ac55d41 | 1866 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1867 | ClearPageChecked(page); |
87826df0 | 1868 | set_page_dirty(page); |
247e743c | 1869 | out: |
2ac55d41 JB |
1870 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1871 | &cached_state, GFP_NOFS); | |
247e743c CM |
1872 | out_page: |
1873 | unlock_page(page); | |
1874 | page_cache_release(page); | |
b897abec | 1875 | kfree(fixup); |
247e743c CM |
1876 | } |
1877 | ||
1878 | /* | |
1879 | * There are a few paths in the higher layers of the kernel that directly | |
1880 | * set the page dirty bit without asking the filesystem if it is a | |
1881 | * good idea. This causes problems because we want to make sure COW | |
1882 | * properly happens and the data=ordered rules are followed. | |
1883 | * | |
c8b97818 | 1884 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1885 | * hasn't been properly setup for IO. We kick off an async process |
1886 | * to fix it up. The async helper will wait for ordered extents, set | |
1887 | * the delalloc bit and make it safe to write the page. | |
1888 | */ | |
b2950863 | 1889 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1890 | { |
1891 | struct inode *inode = page->mapping->host; | |
1892 | struct btrfs_writepage_fixup *fixup; | |
1893 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1894 | |
8b62b72b CM |
1895 | /* this page is properly in the ordered list */ |
1896 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1897 | return 0; |
1898 | ||
1899 | if (PageChecked(page)) | |
1900 | return -EAGAIN; | |
1901 | ||
1902 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1903 | if (!fixup) | |
1904 | return -EAGAIN; | |
f421950f | 1905 | |
247e743c CM |
1906 | SetPageChecked(page); |
1907 | page_cache_get(page); | |
1908 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1909 | fixup->page = page; | |
1910 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1911 | return -EBUSY; |
247e743c CM |
1912 | } |
1913 | ||
d899e052 YZ |
1914 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1915 | struct inode *inode, u64 file_pos, | |
1916 | u64 disk_bytenr, u64 disk_num_bytes, | |
1917 | u64 num_bytes, u64 ram_bytes, | |
1918 | u8 compression, u8 encryption, | |
1919 | u16 other_encoding, int extent_type) | |
1920 | { | |
1921 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1922 | struct btrfs_file_extent_item *fi; | |
1923 | struct btrfs_path *path; | |
1924 | struct extent_buffer *leaf; | |
1925 | struct btrfs_key ins; | |
d899e052 YZ |
1926 | int ret; |
1927 | ||
1928 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1929 | if (!path) |
1930 | return -ENOMEM; | |
d899e052 | 1931 | |
b9473439 | 1932 | path->leave_spinning = 1; |
a1ed835e CM |
1933 | |
1934 | /* | |
1935 | * we may be replacing one extent in the tree with another. | |
1936 | * The new extent is pinned in the extent map, and we don't want | |
1937 | * to drop it from the cache until it is completely in the btree. | |
1938 | * | |
1939 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1940 | * the caller is expected to unpin it and allow it to be merged | |
1941 | * with the others. | |
1942 | */ | |
5dc562c5 | 1943 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1944 | file_pos + num_bytes, 0); |
79787eaa JM |
1945 | if (ret) |
1946 | goto out; | |
d899e052 | 1947 | |
33345d01 | 1948 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1949 | ins.offset = file_pos; |
1950 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1951 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1952 | if (ret) |
1953 | goto out; | |
d899e052 YZ |
1954 | leaf = path->nodes[0]; |
1955 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1956 | struct btrfs_file_extent_item); | |
1957 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1958 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1959 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1960 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1961 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1962 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1963 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1964 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1965 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1966 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1967 | |
d899e052 | 1968 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1969 | btrfs_release_path(path); |
d899e052 YZ |
1970 | |
1971 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1972 | |
1973 | ins.objectid = disk_bytenr; | |
1974 | ins.offset = disk_num_bytes; | |
1975 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1976 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1977 | root->root_key.objectid, | |
33345d01 | 1978 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1979 | out: |
d899e052 | 1980 | btrfs_free_path(path); |
b9473439 | 1981 | |
79787eaa | 1982 | return ret; |
d899e052 YZ |
1983 | } |
1984 | ||
38c227d8 LB |
1985 | /* snapshot-aware defrag */ |
1986 | struct sa_defrag_extent_backref { | |
1987 | struct rb_node node; | |
1988 | struct old_sa_defrag_extent *old; | |
1989 | u64 root_id; | |
1990 | u64 inum; | |
1991 | u64 file_pos; | |
1992 | u64 extent_offset; | |
1993 | u64 num_bytes; | |
1994 | u64 generation; | |
1995 | }; | |
1996 | ||
1997 | struct old_sa_defrag_extent { | |
1998 | struct list_head list; | |
1999 | struct new_sa_defrag_extent *new; | |
2000 | ||
2001 | u64 extent_offset; | |
2002 | u64 bytenr; | |
2003 | u64 offset; | |
2004 | u64 len; | |
2005 | int count; | |
2006 | }; | |
2007 | ||
2008 | struct new_sa_defrag_extent { | |
2009 | struct rb_root root; | |
2010 | struct list_head head; | |
2011 | struct btrfs_path *path; | |
2012 | struct inode *inode; | |
2013 | u64 file_pos; | |
2014 | u64 len; | |
2015 | u64 bytenr; | |
2016 | u64 disk_len; | |
2017 | u8 compress_type; | |
2018 | }; | |
2019 | ||
2020 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2021 | struct sa_defrag_extent_backref *b2) | |
2022 | { | |
2023 | if (b1->root_id < b2->root_id) | |
2024 | return -1; | |
2025 | else if (b1->root_id > b2->root_id) | |
2026 | return 1; | |
2027 | ||
2028 | if (b1->inum < b2->inum) | |
2029 | return -1; | |
2030 | else if (b1->inum > b2->inum) | |
2031 | return 1; | |
2032 | ||
2033 | if (b1->file_pos < b2->file_pos) | |
2034 | return -1; | |
2035 | else if (b1->file_pos > b2->file_pos) | |
2036 | return 1; | |
2037 | ||
2038 | /* | |
2039 | * [------------------------------] ===> (a range of space) | |
2040 | * |<--->| |<---->| =============> (fs/file tree A) | |
2041 | * |<---------------------------->| ===> (fs/file tree B) | |
2042 | * | |
2043 | * A range of space can refer to two file extents in one tree while | |
2044 | * refer to only one file extent in another tree. | |
2045 | * | |
2046 | * So we may process a disk offset more than one time(two extents in A) | |
2047 | * and locate at the same extent(one extent in B), then insert two same | |
2048 | * backrefs(both refer to the extent in B). | |
2049 | */ | |
2050 | return 0; | |
2051 | } | |
2052 | ||
2053 | static void backref_insert(struct rb_root *root, | |
2054 | struct sa_defrag_extent_backref *backref) | |
2055 | { | |
2056 | struct rb_node **p = &root->rb_node; | |
2057 | struct rb_node *parent = NULL; | |
2058 | struct sa_defrag_extent_backref *entry; | |
2059 | int ret; | |
2060 | ||
2061 | while (*p) { | |
2062 | parent = *p; | |
2063 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2064 | ||
2065 | ret = backref_comp(backref, entry); | |
2066 | if (ret < 0) | |
2067 | p = &(*p)->rb_left; | |
2068 | else | |
2069 | p = &(*p)->rb_right; | |
2070 | } | |
2071 | ||
2072 | rb_link_node(&backref->node, parent, p); | |
2073 | rb_insert_color(&backref->node, root); | |
2074 | } | |
2075 | ||
2076 | /* | |
2077 | * Note the backref might has changed, and in this case we just return 0. | |
2078 | */ | |
2079 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2080 | void *ctx) | |
2081 | { | |
2082 | struct btrfs_file_extent_item *extent; | |
2083 | struct btrfs_fs_info *fs_info; | |
2084 | struct old_sa_defrag_extent *old = ctx; | |
2085 | struct new_sa_defrag_extent *new = old->new; | |
2086 | struct btrfs_path *path = new->path; | |
2087 | struct btrfs_key key; | |
2088 | struct btrfs_root *root; | |
2089 | struct sa_defrag_extent_backref *backref; | |
2090 | struct extent_buffer *leaf; | |
2091 | struct inode *inode = new->inode; | |
2092 | int slot; | |
2093 | int ret; | |
2094 | u64 extent_offset; | |
2095 | u64 num_bytes; | |
2096 | ||
2097 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2098 | inum == btrfs_ino(inode)) | |
2099 | return 0; | |
2100 | ||
2101 | key.objectid = root_id; | |
2102 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2103 | key.offset = (u64)-1; | |
2104 | ||
2105 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2106 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2107 | if (IS_ERR(root)) { | |
2108 | if (PTR_ERR(root) == -ENOENT) | |
2109 | return 0; | |
2110 | WARN_ON(1); | |
2111 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2112 | inum, offset, root_id); | |
2113 | return PTR_ERR(root); | |
2114 | } | |
2115 | ||
2116 | key.objectid = inum; | |
2117 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2118 | if (offset > (u64)-1 << 32) | |
2119 | key.offset = 0; | |
2120 | else | |
2121 | key.offset = offset; | |
2122 | ||
2123 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2124 | if (ret < 0) { | |
2125 | WARN_ON(1); | |
2126 | return ret; | |
2127 | } | |
2128 | ||
2129 | while (1) { | |
2130 | cond_resched(); | |
2131 | ||
2132 | leaf = path->nodes[0]; | |
2133 | slot = path->slots[0]; | |
2134 | ||
2135 | if (slot >= btrfs_header_nritems(leaf)) { | |
2136 | ret = btrfs_next_leaf(root, path); | |
2137 | if (ret < 0) { | |
2138 | goto out; | |
2139 | } else if (ret > 0) { | |
2140 | ret = 0; | |
2141 | goto out; | |
2142 | } | |
2143 | continue; | |
2144 | } | |
2145 | ||
2146 | path->slots[0]++; | |
2147 | ||
2148 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2149 | ||
2150 | if (key.objectid > inum) | |
2151 | goto out; | |
2152 | ||
2153 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2154 | continue; | |
2155 | ||
2156 | extent = btrfs_item_ptr(leaf, slot, | |
2157 | struct btrfs_file_extent_item); | |
2158 | ||
2159 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2160 | continue; | |
2161 | ||
2162 | extent_offset = btrfs_file_extent_offset(leaf, extent); | |
2163 | if (key.offset - extent_offset != offset) | |
2164 | continue; | |
2165 | ||
2166 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); | |
2167 | if (extent_offset >= old->extent_offset + old->offset + | |
2168 | old->len || extent_offset + num_bytes <= | |
2169 | old->extent_offset + old->offset) | |
2170 | continue; | |
2171 | ||
2172 | break; | |
2173 | } | |
2174 | ||
2175 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2176 | if (!backref) { | |
2177 | ret = -ENOENT; | |
2178 | goto out; | |
2179 | } | |
2180 | ||
2181 | backref->root_id = root_id; | |
2182 | backref->inum = inum; | |
2183 | backref->file_pos = offset + extent_offset; | |
2184 | backref->num_bytes = num_bytes; | |
2185 | backref->extent_offset = extent_offset; | |
2186 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2187 | backref->old = old; | |
2188 | backref_insert(&new->root, backref); | |
2189 | old->count++; | |
2190 | out: | |
2191 | btrfs_release_path(path); | |
2192 | WARN_ON(ret); | |
2193 | return ret; | |
2194 | } | |
2195 | ||
2196 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2197 | struct new_sa_defrag_extent *new) | |
2198 | { | |
2199 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2200 | struct old_sa_defrag_extent *old, *tmp; | |
2201 | int ret; | |
2202 | ||
2203 | new->path = path; | |
2204 | ||
2205 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2206 | ret = iterate_inodes_from_logical(old->bytenr, fs_info, | |
2207 | path, record_one_backref, | |
2208 | old); | |
2209 | BUG_ON(ret < 0 && ret != -ENOENT); | |
2210 | ||
2211 | /* no backref to be processed for this extent */ | |
2212 | if (!old->count) { | |
2213 | list_del(&old->list); | |
2214 | kfree(old); | |
2215 | } | |
2216 | } | |
2217 | ||
2218 | if (list_empty(&new->head)) | |
2219 | return false; | |
2220 | ||
2221 | return true; | |
2222 | } | |
2223 | ||
2224 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2225 | struct btrfs_file_extent_item *fi, | |
2226 | u64 disk_bytenr) | |
2227 | { | |
2228 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr) | |
2229 | return 0; | |
2230 | ||
2231 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2232 | return 0; | |
2233 | ||
2234 | if (btrfs_file_extent_compression(leaf, fi) || | |
2235 | btrfs_file_extent_encryption(leaf, fi) || | |
2236 | btrfs_file_extent_other_encoding(leaf, fi)) | |
2237 | return 0; | |
2238 | ||
2239 | return 1; | |
2240 | } | |
2241 | ||
2242 | /* | |
2243 | * Note the backref might has changed, and in this case we just return 0. | |
2244 | */ | |
2245 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2246 | struct sa_defrag_extent_backref *prev, | |
2247 | struct sa_defrag_extent_backref *backref) | |
2248 | { | |
2249 | struct btrfs_file_extent_item *extent; | |
2250 | struct btrfs_file_extent_item *item; | |
2251 | struct btrfs_ordered_extent *ordered; | |
2252 | struct btrfs_trans_handle *trans; | |
2253 | struct btrfs_fs_info *fs_info; | |
2254 | struct btrfs_root *root; | |
2255 | struct btrfs_key key; | |
2256 | struct extent_buffer *leaf; | |
2257 | struct old_sa_defrag_extent *old = backref->old; | |
2258 | struct new_sa_defrag_extent *new = old->new; | |
2259 | struct inode *src_inode = new->inode; | |
2260 | struct inode *inode; | |
2261 | struct extent_state *cached = NULL; | |
2262 | int ret = 0; | |
2263 | u64 start; | |
2264 | u64 len; | |
2265 | u64 lock_start; | |
2266 | u64 lock_end; | |
2267 | bool merge = false; | |
2268 | int index; | |
2269 | ||
2270 | if (prev && prev->root_id == backref->root_id && | |
2271 | prev->inum == backref->inum && | |
2272 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2273 | merge = true; | |
2274 | ||
2275 | /* step 1: get root */ | |
2276 | key.objectid = backref->root_id; | |
2277 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2278 | key.offset = (u64)-1; | |
2279 | ||
2280 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2281 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2282 | ||
2283 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2284 | if (IS_ERR(root)) { | |
2285 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2286 | if (PTR_ERR(root) == -ENOENT) | |
2287 | return 0; | |
2288 | return PTR_ERR(root); | |
2289 | } | |
38c227d8 LB |
2290 | |
2291 | /* step 2: get inode */ | |
2292 | key.objectid = backref->inum; | |
2293 | key.type = BTRFS_INODE_ITEM_KEY; | |
2294 | key.offset = 0; | |
2295 | ||
2296 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2297 | if (IS_ERR(inode)) { | |
2298 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2299 | return 0; | |
2300 | } | |
2301 | ||
2302 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2303 | ||
2304 | /* step 3: relink backref */ | |
2305 | lock_start = backref->file_pos; | |
2306 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2307 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2308 | 0, &cached); | |
2309 | ||
2310 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2311 | if (ordered) { | |
2312 | btrfs_put_ordered_extent(ordered); | |
2313 | goto out_unlock; | |
2314 | } | |
2315 | ||
2316 | trans = btrfs_join_transaction(root); | |
2317 | if (IS_ERR(trans)) { | |
2318 | ret = PTR_ERR(trans); | |
2319 | goto out_unlock; | |
2320 | } | |
2321 | ||
2322 | key.objectid = backref->inum; | |
2323 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2324 | key.offset = backref->file_pos; | |
2325 | ||
2326 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2327 | if (ret < 0) { | |
2328 | goto out_free_path; | |
2329 | } else if (ret > 0) { | |
2330 | ret = 0; | |
2331 | goto out_free_path; | |
2332 | } | |
2333 | ||
2334 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2335 | struct btrfs_file_extent_item); | |
2336 | ||
2337 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2338 | backref->generation) | |
2339 | goto out_free_path; | |
2340 | ||
2341 | btrfs_release_path(path); | |
2342 | ||
2343 | start = backref->file_pos; | |
2344 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2345 | start += old->extent_offset + old->offset - | |
2346 | backref->extent_offset; | |
2347 | ||
2348 | len = min(backref->extent_offset + backref->num_bytes, | |
2349 | old->extent_offset + old->offset + old->len); | |
2350 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2351 | ||
2352 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2353 | start + len, 1); | |
2354 | if (ret) | |
2355 | goto out_free_path; | |
2356 | again: | |
2357 | key.objectid = btrfs_ino(inode); | |
2358 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2359 | key.offset = start; | |
2360 | ||
a09a0a70 | 2361 | path->leave_spinning = 1; |
38c227d8 LB |
2362 | if (merge) { |
2363 | struct btrfs_file_extent_item *fi; | |
2364 | u64 extent_len; | |
2365 | struct btrfs_key found_key; | |
2366 | ||
2367 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | |
2368 | if (ret < 0) | |
2369 | goto out_free_path; | |
2370 | ||
2371 | path->slots[0]--; | |
2372 | leaf = path->nodes[0]; | |
2373 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2374 | ||
2375 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2376 | struct btrfs_file_extent_item); | |
2377 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2378 | ||
2379 | if (relink_is_mergable(leaf, fi, new->bytenr) && | |
2380 | extent_len + found_key.offset == start) { | |
2381 | btrfs_set_file_extent_num_bytes(leaf, fi, | |
2382 | extent_len + len); | |
2383 | btrfs_mark_buffer_dirty(leaf); | |
2384 | inode_add_bytes(inode, len); | |
2385 | ||
2386 | ret = 1; | |
2387 | goto out_free_path; | |
2388 | } else { | |
2389 | merge = false; | |
2390 | btrfs_release_path(path); | |
2391 | goto again; | |
2392 | } | |
2393 | } | |
2394 | ||
2395 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2396 | sizeof(*extent)); | |
2397 | if (ret) { | |
2398 | btrfs_abort_transaction(trans, root, ret); | |
2399 | goto out_free_path; | |
2400 | } | |
2401 | ||
2402 | leaf = path->nodes[0]; | |
2403 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2404 | struct btrfs_file_extent_item); | |
2405 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2406 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2407 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2408 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2409 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2410 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2411 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2412 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2413 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2414 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2415 | ||
2416 | btrfs_mark_buffer_dirty(leaf); | |
2417 | inode_add_bytes(inode, len); | |
a09a0a70 | 2418 | btrfs_release_path(path); |
38c227d8 LB |
2419 | |
2420 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2421 | new->disk_len, 0, | |
2422 | backref->root_id, backref->inum, | |
2423 | new->file_pos, 0); /* start - extent_offset */ | |
2424 | if (ret) { | |
2425 | btrfs_abort_transaction(trans, root, ret); | |
2426 | goto out_free_path; | |
2427 | } | |
2428 | ||
2429 | ret = 1; | |
2430 | out_free_path: | |
2431 | btrfs_release_path(path); | |
a09a0a70 | 2432 | path->leave_spinning = 0; |
38c227d8 LB |
2433 | btrfs_end_transaction(trans, root); |
2434 | out_unlock: | |
2435 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2436 | &cached, GFP_NOFS); | |
2437 | iput(inode); | |
2438 | return ret; | |
2439 | } | |
2440 | ||
2441 | static void relink_file_extents(struct new_sa_defrag_extent *new) | |
2442 | { | |
2443 | struct btrfs_path *path; | |
2444 | struct old_sa_defrag_extent *old, *tmp; | |
2445 | struct sa_defrag_extent_backref *backref; | |
2446 | struct sa_defrag_extent_backref *prev = NULL; | |
2447 | struct inode *inode; | |
2448 | struct btrfs_root *root; | |
2449 | struct rb_node *node; | |
2450 | int ret; | |
2451 | ||
2452 | inode = new->inode; | |
2453 | root = BTRFS_I(inode)->root; | |
2454 | ||
2455 | path = btrfs_alloc_path(); | |
2456 | if (!path) | |
2457 | return; | |
2458 | ||
2459 | if (!record_extent_backrefs(path, new)) { | |
2460 | btrfs_free_path(path); | |
2461 | goto out; | |
2462 | } | |
2463 | btrfs_release_path(path); | |
2464 | ||
2465 | while (1) { | |
2466 | node = rb_first(&new->root); | |
2467 | if (!node) | |
2468 | break; | |
2469 | rb_erase(node, &new->root); | |
2470 | ||
2471 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2472 | ||
2473 | ret = relink_extent_backref(path, prev, backref); | |
2474 | WARN_ON(ret < 0); | |
2475 | ||
2476 | kfree(prev); | |
2477 | ||
2478 | if (ret == 1) | |
2479 | prev = backref; | |
2480 | else | |
2481 | prev = NULL; | |
2482 | cond_resched(); | |
2483 | } | |
2484 | kfree(prev); | |
2485 | ||
2486 | btrfs_free_path(path); | |
2487 | ||
2488 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2489 | list_del(&old->list); | |
2490 | kfree(old); | |
2491 | } | |
2492 | out: | |
2493 | atomic_dec(&root->fs_info->defrag_running); | |
2494 | wake_up(&root->fs_info->transaction_wait); | |
2495 | ||
2496 | kfree(new); | |
2497 | } | |
2498 | ||
2499 | static struct new_sa_defrag_extent * | |
2500 | record_old_file_extents(struct inode *inode, | |
2501 | struct btrfs_ordered_extent *ordered) | |
2502 | { | |
2503 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2504 | struct btrfs_path *path; | |
2505 | struct btrfs_key key; | |
2506 | struct old_sa_defrag_extent *old, *tmp; | |
2507 | struct new_sa_defrag_extent *new; | |
2508 | int ret; | |
2509 | ||
2510 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2511 | if (!new) | |
2512 | return NULL; | |
2513 | ||
2514 | new->inode = inode; | |
2515 | new->file_pos = ordered->file_offset; | |
2516 | new->len = ordered->len; | |
2517 | new->bytenr = ordered->start; | |
2518 | new->disk_len = ordered->disk_len; | |
2519 | new->compress_type = ordered->compress_type; | |
2520 | new->root = RB_ROOT; | |
2521 | INIT_LIST_HEAD(&new->head); | |
2522 | ||
2523 | path = btrfs_alloc_path(); | |
2524 | if (!path) | |
2525 | goto out_kfree; | |
2526 | ||
2527 | key.objectid = btrfs_ino(inode); | |
2528 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2529 | key.offset = new->file_pos; | |
2530 | ||
2531 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2532 | if (ret < 0) | |
2533 | goto out_free_path; | |
2534 | if (ret > 0 && path->slots[0] > 0) | |
2535 | path->slots[0]--; | |
2536 | ||
2537 | /* find out all the old extents for the file range */ | |
2538 | while (1) { | |
2539 | struct btrfs_file_extent_item *extent; | |
2540 | struct extent_buffer *l; | |
2541 | int slot; | |
2542 | u64 num_bytes; | |
2543 | u64 offset; | |
2544 | u64 end; | |
2545 | u64 disk_bytenr; | |
2546 | u64 extent_offset; | |
2547 | ||
2548 | l = path->nodes[0]; | |
2549 | slot = path->slots[0]; | |
2550 | ||
2551 | if (slot >= btrfs_header_nritems(l)) { | |
2552 | ret = btrfs_next_leaf(root, path); | |
2553 | if (ret < 0) | |
2554 | goto out_free_list; | |
2555 | else if (ret > 0) | |
2556 | break; | |
2557 | continue; | |
2558 | } | |
2559 | ||
2560 | btrfs_item_key_to_cpu(l, &key, slot); | |
2561 | ||
2562 | if (key.objectid != btrfs_ino(inode)) | |
2563 | break; | |
2564 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2565 | break; | |
2566 | if (key.offset >= new->file_pos + new->len) | |
2567 | break; | |
2568 | ||
2569 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2570 | ||
2571 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2572 | if (key.offset + num_bytes < new->file_pos) | |
2573 | goto next; | |
2574 | ||
2575 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2576 | if (!disk_bytenr) | |
2577 | goto next; | |
2578 | ||
2579 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2580 | ||
2581 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2582 | if (!old) | |
2583 | goto out_free_list; | |
2584 | ||
2585 | offset = max(new->file_pos, key.offset); | |
2586 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2587 | ||
2588 | old->bytenr = disk_bytenr; | |
2589 | old->extent_offset = extent_offset; | |
2590 | old->offset = offset - key.offset; | |
2591 | old->len = end - offset; | |
2592 | old->new = new; | |
2593 | old->count = 0; | |
2594 | list_add_tail(&old->list, &new->head); | |
2595 | next: | |
2596 | path->slots[0]++; | |
2597 | cond_resched(); | |
2598 | } | |
2599 | ||
2600 | btrfs_free_path(path); | |
2601 | atomic_inc(&root->fs_info->defrag_running); | |
2602 | ||
2603 | return new; | |
2604 | ||
2605 | out_free_list: | |
2606 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
2607 | list_del(&old->list); | |
2608 | kfree(old); | |
2609 | } | |
2610 | out_free_path: | |
2611 | btrfs_free_path(path); | |
2612 | out_kfree: | |
2613 | kfree(new); | |
2614 | return NULL; | |
2615 | } | |
2616 | ||
5d13a98f CM |
2617 | /* |
2618 | * helper function for btrfs_finish_ordered_io, this | |
2619 | * just reads in some of the csum leaves to prime them into ram | |
2620 | * before we start the transaction. It limits the amount of btree | |
2621 | * reads required while inside the transaction. | |
2622 | */ | |
d352ac68 CM |
2623 | /* as ordered data IO finishes, this gets called so we can finish |
2624 | * an ordered extent if the range of bytes in the file it covers are | |
2625 | * fully written. | |
2626 | */ | |
5fd02043 | 2627 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2628 | { |
5fd02043 | 2629 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2630 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2631 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2632 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2633 | struct extent_state *cached_state = NULL; |
38c227d8 | 2634 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2635 | int compress_type = 0; |
e6dcd2dc | 2636 | int ret; |
82d5902d | 2637 | bool nolock; |
e6dcd2dc | 2638 | |
83eea1f1 | 2639 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2640 | |
5fd02043 JB |
2641 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2642 | ret = -EIO; | |
2643 | goto out; | |
2644 | } | |
2645 | ||
c2167754 | 2646 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2647 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2648 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2649 | if (nolock) | |
2650 | trans = btrfs_join_transaction_nolock(root); | |
2651 | else | |
2652 | trans = btrfs_join_transaction(root); | |
2653 | if (IS_ERR(trans)) { | |
2654 | ret = PTR_ERR(trans); | |
2655 | trans = NULL; | |
2656 | goto out; | |
c2167754 | 2657 | } |
6c760c07 JB |
2658 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2659 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2660 | if (ret) /* -ENOMEM or corruption */ | |
2661 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2662 | goto out; |
2663 | } | |
e6dcd2dc | 2664 | |
2ac55d41 JB |
2665 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2666 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2667 | 0, &cached_state); |
e6dcd2dc | 2668 | |
38c227d8 LB |
2669 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2670 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2671 | EXTENT_DEFRAG, 1, cached_state); | |
2672 | if (ret) { | |
2673 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
2674 | if (last_snapshot >= BTRFS_I(inode)->generation) | |
2675 | /* the inode is shared */ | |
2676 | new = record_old_file_extents(inode, ordered_extent); | |
2677 | ||
2678 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2679 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2680 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2681 | } | |
2682 | ||
0cb59c99 | 2683 | if (nolock) |
7a7eaa40 | 2684 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2685 | else |
7a7eaa40 | 2686 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2687 | if (IS_ERR(trans)) { |
2688 | ret = PTR_ERR(trans); | |
2689 | trans = NULL; | |
2690 | goto out_unlock; | |
2691 | } | |
0ca1f7ce | 2692 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2693 | |
c8b97818 | 2694 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2695 | compress_type = ordered_extent->compress_type; |
d899e052 | 2696 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2697 | BUG_ON(compress_type); |
920bbbfb | 2698 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2699 | ordered_extent->file_offset, |
2700 | ordered_extent->file_offset + | |
2701 | ordered_extent->len); | |
d899e052 | 2702 | } else { |
0af3d00b | 2703 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2704 | ret = insert_reserved_file_extent(trans, inode, |
2705 | ordered_extent->file_offset, | |
2706 | ordered_extent->start, | |
2707 | ordered_extent->disk_len, | |
2708 | ordered_extent->len, | |
2709 | ordered_extent->len, | |
261507a0 | 2710 | compress_type, 0, 0, |
d899e052 | 2711 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 2712 | } |
5dc562c5 JB |
2713 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2714 | ordered_extent->file_offset, ordered_extent->len, | |
2715 | trans->transid); | |
79787eaa JM |
2716 | if (ret < 0) { |
2717 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2718 | goto out_unlock; |
79787eaa | 2719 | } |
2ac55d41 | 2720 | |
e6dcd2dc CM |
2721 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2722 | &ordered_extent->list); | |
2723 | ||
6c760c07 JB |
2724 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2725 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2726 | if (ret) { /* -ENOMEM or corruption */ | |
2727 | btrfs_abort_transaction(trans, root, ret); | |
2728 | goto out_unlock; | |
1ef30be1 JB |
2729 | } |
2730 | ret = 0; | |
5fd02043 JB |
2731 | out_unlock: |
2732 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2733 | ordered_extent->file_offset + | |
2734 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2735 | out: |
5b0e95bf | 2736 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2737 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2738 | if (trans) |
2739 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2740 | |
0bec9ef5 | 2741 | if (ret) { |
5fd02043 JB |
2742 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, |
2743 | ordered_extent->file_offset + | |
2744 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
2745 | ||
0bec9ef5 JB |
2746 | /* |
2747 | * If the ordered extent had an IOERR or something else went | |
2748 | * wrong we need to return the space for this ordered extent | |
2749 | * back to the allocator. | |
2750 | */ | |
2751 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2752 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) | |
2753 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2754 | ordered_extent->disk_len); | |
2755 | } | |
2756 | ||
2757 | ||
5fd02043 | 2758 | /* |
8bad3c02 LB |
2759 | * This needs to be done to make sure anybody waiting knows we are done |
2760 | * updating everything for this ordered extent. | |
5fd02043 JB |
2761 | */ |
2762 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2763 | ||
38c227d8 LB |
2764 | /* for snapshot-aware defrag */ |
2765 | if (new) | |
2766 | relink_file_extents(new); | |
2767 | ||
e6dcd2dc CM |
2768 | /* once for us */ |
2769 | btrfs_put_ordered_extent(ordered_extent); | |
2770 | /* once for the tree */ | |
2771 | btrfs_put_ordered_extent(ordered_extent); | |
2772 | ||
5fd02043 JB |
2773 | return ret; |
2774 | } | |
2775 | ||
2776 | static void finish_ordered_fn(struct btrfs_work *work) | |
2777 | { | |
2778 | struct btrfs_ordered_extent *ordered_extent; | |
2779 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2780 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2781 | } |
2782 | ||
b2950863 | 2783 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2784 | struct extent_state *state, int uptodate) |
2785 | { | |
5fd02043 JB |
2786 | struct inode *inode = page->mapping->host; |
2787 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2788 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2789 | struct btrfs_workers *workers; | |
2790 | ||
1abe9b8a | 2791 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2792 | ||
8b62b72b | 2793 | ClearPagePrivate2(page); |
5fd02043 JB |
2794 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2795 | end - start + 1, uptodate)) | |
2796 | return 0; | |
2797 | ||
2798 | ordered_extent->work.func = finish_ordered_fn; | |
2799 | ordered_extent->work.flags = 0; | |
2800 | ||
83eea1f1 | 2801 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2802 | workers = &root->fs_info->endio_freespace_worker; |
2803 | else | |
2804 | workers = &root->fs_info->endio_write_workers; | |
2805 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2806 | ||
2807 | return 0; | |
211f90e6 CM |
2808 | } |
2809 | ||
d352ac68 CM |
2810 | /* |
2811 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2812 | * if there's a match, we allow the bio to finish. If not, the code in |
2813 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2814 | */ |
b2950863 | 2815 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2816 | struct extent_state *state, int mirror) |
07157aac | 2817 | { |
4eee4fa4 | 2818 | size_t offset = start - page_offset(page); |
07157aac | 2819 | struct inode *inode = page->mapping->host; |
d1310b2e | 2820 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2821 | char *kaddr; |
aadfeb6e | 2822 | u64 private = ~(u32)0; |
07157aac | 2823 | int ret; |
ff79f819 CM |
2824 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2825 | u32 csum = ~(u32)0; | |
c2cf52eb SK |
2826 | static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, |
2827 | DEFAULT_RATELIMIT_BURST); | |
d1310b2e | 2828 | |
d20f7043 CM |
2829 | if (PageChecked(page)) { |
2830 | ClearPageChecked(page); | |
2831 | goto good; | |
2832 | } | |
6cbff00f CH |
2833 | |
2834 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2835 | goto good; |
17d217fe YZ |
2836 | |
2837 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2838 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2839 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2840 | GFP_NOFS); | |
b6cda9bc | 2841 | return 0; |
17d217fe | 2842 | } |
d20f7043 | 2843 | |
c2e639f0 | 2844 | if (state && state->start == start) { |
70dec807 CM |
2845 | private = state->private; |
2846 | ret = 0; | |
2847 | } else { | |
2848 | ret = get_state_private(io_tree, start, &private); | |
2849 | } | |
7ac687d9 | 2850 | kaddr = kmap_atomic(page); |
d397712b | 2851 | if (ret) |
07157aac | 2852 | goto zeroit; |
d397712b | 2853 | |
b0496686 | 2854 | csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1); |
ff79f819 | 2855 | btrfs_csum_final(csum, (char *)&csum); |
d397712b | 2856 | if (csum != private) |
07157aac | 2857 | goto zeroit; |
d397712b | 2858 | |
7ac687d9 | 2859 | kunmap_atomic(kaddr); |
d20f7043 | 2860 | good: |
07157aac CM |
2861 | return 0; |
2862 | ||
2863 | zeroit: | |
c2cf52eb SK |
2864 | if (__ratelimit(&_rs)) |
2865 | btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu", | |
2866 | (unsigned long long)btrfs_ino(page->mapping->host), | |
2867 | (unsigned long long)start, csum, | |
2868 | (unsigned long long)private); | |
db94535d CM |
2869 | memset(kaddr + offset, 1, end - start + 1); |
2870 | flush_dcache_page(page); | |
7ac687d9 | 2871 | kunmap_atomic(kaddr); |
3b951516 CM |
2872 | if (private == 0) |
2873 | return 0; | |
7e38326f | 2874 | return -EIO; |
07157aac | 2875 | } |
b888db2b | 2876 | |
24bbcf04 YZ |
2877 | struct delayed_iput { |
2878 | struct list_head list; | |
2879 | struct inode *inode; | |
2880 | }; | |
2881 | ||
79787eaa JM |
2882 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2883 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2884 | void btrfs_add_delayed_iput(struct inode *inode) |
2885 | { | |
2886 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2887 | struct delayed_iput *delayed; | |
2888 | ||
2889 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2890 | return; | |
2891 | ||
2892 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2893 | delayed->inode = inode; | |
2894 | ||
2895 | spin_lock(&fs_info->delayed_iput_lock); | |
2896 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2897 | spin_unlock(&fs_info->delayed_iput_lock); | |
2898 | } | |
2899 | ||
2900 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2901 | { | |
2902 | LIST_HEAD(list); | |
2903 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2904 | struct delayed_iput *delayed; | |
2905 | int empty; | |
2906 | ||
2907 | spin_lock(&fs_info->delayed_iput_lock); | |
2908 | empty = list_empty(&fs_info->delayed_iputs); | |
2909 | spin_unlock(&fs_info->delayed_iput_lock); | |
2910 | if (empty) | |
2911 | return; | |
2912 | ||
24bbcf04 YZ |
2913 | spin_lock(&fs_info->delayed_iput_lock); |
2914 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2915 | spin_unlock(&fs_info->delayed_iput_lock); | |
2916 | ||
2917 | while (!list_empty(&list)) { | |
2918 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2919 | list_del(&delayed->list); | |
2920 | iput(delayed->inode); | |
2921 | kfree(delayed); | |
2922 | } | |
24bbcf04 YZ |
2923 | } |
2924 | ||
d68fc57b | 2925 | /* |
42b2aa86 | 2926 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2927 | * files in the subvolume, it removes orphan item and frees block_rsv |
2928 | * structure. | |
2929 | */ | |
2930 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2931 | struct btrfs_root *root) | |
2932 | { | |
90290e19 | 2933 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2934 | int ret; |
2935 | ||
8a35d95f | 2936 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2937 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2938 | return; | |
2939 | ||
90290e19 | 2940 | spin_lock(&root->orphan_lock); |
8a35d95f | 2941 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2942 | spin_unlock(&root->orphan_lock); |
2943 | return; | |
2944 | } | |
2945 | ||
2946 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2947 | spin_unlock(&root->orphan_lock); | |
2948 | return; | |
2949 | } | |
2950 | ||
2951 | block_rsv = root->orphan_block_rsv; | |
2952 | root->orphan_block_rsv = NULL; | |
2953 | spin_unlock(&root->orphan_lock); | |
2954 | ||
d68fc57b YZ |
2955 | if (root->orphan_item_inserted && |
2956 | btrfs_root_refs(&root->root_item) > 0) { | |
2957 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2958 | root->root_key.objectid); | |
2959 | BUG_ON(ret); | |
2960 | root->orphan_item_inserted = 0; | |
2961 | } | |
2962 | ||
90290e19 JB |
2963 | if (block_rsv) { |
2964 | WARN_ON(block_rsv->size > 0); | |
2965 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2966 | } |
2967 | } | |
2968 | ||
7b128766 JB |
2969 | /* |
2970 | * This creates an orphan entry for the given inode in case something goes | |
2971 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2972 | * |
2973 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2974 | * this function. | |
7b128766 JB |
2975 | */ |
2976 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2977 | { | |
2978 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2979 | struct btrfs_block_rsv *block_rsv = NULL; |
2980 | int reserve = 0; | |
2981 | int insert = 0; | |
2982 | int ret; | |
7b128766 | 2983 | |
d68fc57b | 2984 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2985 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2986 | if (!block_rsv) |
2987 | return -ENOMEM; | |
d68fc57b | 2988 | } |
7b128766 | 2989 | |
d68fc57b YZ |
2990 | spin_lock(&root->orphan_lock); |
2991 | if (!root->orphan_block_rsv) { | |
2992 | root->orphan_block_rsv = block_rsv; | |
2993 | } else if (block_rsv) { | |
2994 | btrfs_free_block_rsv(root, block_rsv); | |
2995 | block_rsv = NULL; | |
7b128766 | 2996 | } |
7b128766 | 2997 | |
8a35d95f JB |
2998 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2999 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3000 | #if 0 |
3001 | /* | |
3002 | * For proper ENOSPC handling, we should do orphan | |
3003 | * cleanup when mounting. But this introduces backward | |
3004 | * compatibility issue. | |
3005 | */ | |
3006 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3007 | insert = 2; | |
3008 | else | |
3009 | insert = 1; | |
3010 | #endif | |
3011 | insert = 1; | |
321f0e70 | 3012 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3013 | } |
3014 | ||
72ac3c0d JB |
3015 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3016 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3017 | reserve = 1; |
d68fc57b | 3018 | spin_unlock(&root->orphan_lock); |
7b128766 | 3019 | |
d68fc57b YZ |
3020 | /* grab metadata reservation from transaction handle */ |
3021 | if (reserve) { | |
3022 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3023 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3024 | } |
7b128766 | 3025 | |
d68fc57b YZ |
3026 | /* insert an orphan item to track this unlinked/truncated file */ |
3027 | if (insert >= 1) { | |
33345d01 | 3028 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 3029 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
3030 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3031 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
3032 | btrfs_abort_transaction(trans, root, ret); |
3033 | return ret; | |
3034 | } | |
3035 | ret = 0; | |
d68fc57b YZ |
3036 | } |
3037 | ||
3038 | /* insert an orphan item to track subvolume contains orphan files */ | |
3039 | if (insert >= 2) { | |
3040 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3041 | root->root_key.objectid); | |
79787eaa JM |
3042 | if (ret && ret != -EEXIST) { |
3043 | btrfs_abort_transaction(trans, root, ret); | |
3044 | return ret; | |
3045 | } | |
d68fc57b YZ |
3046 | } |
3047 | return 0; | |
7b128766 JB |
3048 | } |
3049 | ||
3050 | /* | |
3051 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3052 | * item for this particular inode. | |
3053 | */ | |
48a3b636 ES |
3054 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3055 | struct inode *inode) | |
7b128766 JB |
3056 | { |
3057 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3058 | int delete_item = 0; |
3059 | int release_rsv = 0; | |
7b128766 JB |
3060 | int ret = 0; |
3061 | ||
d68fc57b | 3062 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3063 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3064 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3065 | delete_item = 1; |
7b128766 | 3066 | |
72ac3c0d JB |
3067 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3068 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3069 | release_rsv = 1; |
d68fc57b | 3070 | spin_unlock(&root->orphan_lock); |
7b128766 | 3071 | |
d68fc57b | 3072 | if (trans && delete_item) { |
33345d01 | 3073 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 3074 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 3075 | } |
7b128766 | 3076 | |
8a35d95f | 3077 | if (release_rsv) { |
d68fc57b | 3078 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
3079 | atomic_dec(&root->orphan_inodes); |
3080 | } | |
7b128766 | 3081 | |
d68fc57b | 3082 | return 0; |
7b128766 JB |
3083 | } |
3084 | ||
3085 | /* | |
3086 | * this cleans up any orphans that may be left on the list from the last use | |
3087 | * of this root. | |
3088 | */ | |
66b4ffd1 | 3089 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3090 | { |
3091 | struct btrfs_path *path; | |
3092 | struct extent_buffer *leaf; | |
7b128766 JB |
3093 | struct btrfs_key key, found_key; |
3094 | struct btrfs_trans_handle *trans; | |
3095 | struct inode *inode; | |
8f6d7f4f | 3096 | u64 last_objectid = 0; |
7b128766 JB |
3097 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3098 | ||
d68fc57b | 3099 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3100 | return 0; |
c71bf099 YZ |
3101 | |
3102 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3103 | if (!path) { |
3104 | ret = -ENOMEM; | |
3105 | goto out; | |
3106 | } | |
7b128766 JB |
3107 | path->reada = -1; |
3108 | ||
3109 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
3110 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
3111 | key.offset = (u64)-1; | |
3112 | ||
7b128766 JB |
3113 | while (1) { |
3114 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3115 | if (ret < 0) |
3116 | goto out; | |
7b128766 JB |
3117 | |
3118 | /* | |
3119 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3120 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3121 | * find the key and see if we have stuff that matches |
3122 | */ | |
3123 | if (ret > 0) { | |
66b4ffd1 | 3124 | ret = 0; |
7b128766 JB |
3125 | if (path->slots[0] == 0) |
3126 | break; | |
3127 | path->slots[0]--; | |
3128 | } | |
3129 | ||
3130 | /* pull out the item */ | |
3131 | leaf = path->nodes[0]; | |
7b128766 JB |
3132 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3133 | ||
3134 | /* make sure the item matches what we want */ | |
3135 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3136 | break; | |
3137 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
3138 | break; | |
3139 | ||
3140 | /* release the path since we're done with it */ | |
b3b4aa74 | 3141 | btrfs_release_path(path); |
7b128766 JB |
3142 | |
3143 | /* | |
3144 | * this is where we are basically btrfs_lookup, without the | |
3145 | * crossing root thing. we store the inode number in the | |
3146 | * offset of the orphan item. | |
3147 | */ | |
8f6d7f4f JB |
3148 | |
3149 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3150 | btrfs_err(root->fs_info, |
3151 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3152 | ret = -EINVAL; |
3153 | goto out; | |
3154 | } | |
3155 | ||
3156 | last_objectid = found_key.offset; | |
3157 | ||
5d4f98a2 YZ |
3158 | found_key.objectid = found_key.offset; |
3159 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3160 | found_key.offset = 0; | |
73f73415 | 3161 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
3162 | ret = PTR_RET(inode); |
3163 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 3164 | goto out; |
7b128766 | 3165 | |
f8e9e0b0 AJ |
3166 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3167 | struct btrfs_root *dead_root; | |
3168 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3169 | int is_dead_root = 0; | |
3170 | ||
3171 | /* | |
3172 | * this is an orphan in the tree root. Currently these | |
3173 | * could come from 2 sources: | |
3174 | * a) a snapshot deletion in progress | |
3175 | * b) a free space cache inode | |
3176 | * We need to distinguish those two, as the snapshot | |
3177 | * orphan must not get deleted. | |
3178 | * find_dead_roots already ran before us, so if this | |
3179 | * is a snapshot deletion, we should find the root | |
3180 | * in the dead_roots list | |
3181 | */ | |
3182 | spin_lock(&fs_info->trans_lock); | |
3183 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3184 | root_list) { | |
3185 | if (dead_root->root_key.objectid == | |
3186 | found_key.objectid) { | |
3187 | is_dead_root = 1; | |
3188 | break; | |
3189 | } | |
3190 | } | |
3191 | spin_unlock(&fs_info->trans_lock); | |
3192 | if (is_dead_root) { | |
3193 | /* prevent this orphan from being found again */ | |
3194 | key.offset = found_key.objectid - 1; | |
3195 | continue; | |
3196 | } | |
3197 | } | |
7b128766 | 3198 | /* |
a8c9e576 JB |
3199 | * Inode is already gone but the orphan item is still there, |
3200 | * kill the orphan item. | |
7b128766 | 3201 | */ |
a8c9e576 JB |
3202 | if (ret == -ESTALE) { |
3203 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3204 | if (IS_ERR(trans)) { |
3205 | ret = PTR_ERR(trans); | |
3206 | goto out; | |
3207 | } | |
c2cf52eb SK |
3208 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3209 | found_key.objectid); | |
a8c9e576 JB |
3210 | ret = btrfs_del_orphan_item(trans, root, |
3211 | found_key.objectid); | |
79787eaa | 3212 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 3213 | btrfs_end_transaction(trans, root); |
7b128766 JB |
3214 | continue; |
3215 | } | |
3216 | ||
a8c9e576 JB |
3217 | /* |
3218 | * add this inode to the orphan list so btrfs_orphan_del does | |
3219 | * the proper thing when we hit it | |
3220 | */ | |
8a35d95f JB |
3221 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3222 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3223 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3224 | |
7b128766 JB |
3225 | /* if we have links, this was a truncate, lets do that */ |
3226 | if (inode->i_nlink) { | |
a41ad394 JB |
3227 | if (!S_ISREG(inode->i_mode)) { |
3228 | WARN_ON(1); | |
3229 | iput(inode); | |
3230 | continue; | |
3231 | } | |
7b128766 | 3232 | nr_truncate++; |
f3fe820c JB |
3233 | |
3234 | /* 1 for the orphan item deletion. */ | |
3235 | trans = btrfs_start_transaction(root, 1); | |
3236 | if (IS_ERR(trans)) { | |
c69b26b0 | 3237 | iput(inode); |
f3fe820c JB |
3238 | ret = PTR_ERR(trans); |
3239 | goto out; | |
3240 | } | |
3241 | ret = btrfs_orphan_add(trans, inode); | |
3242 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3243 | if (ret) { |
3244 | iput(inode); | |
f3fe820c | 3245 | goto out; |
c69b26b0 | 3246 | } |
f3fe820c | 3247 | |
66b4ffd1 | 3248 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3249 | if (ret) |
3250 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3251 | } else { |
3252 | nr_unlink++; | |
3253 | } | |
3254 | ||
3255 | /* this will do delete_inode and everything for us */ | |
3256 | iput(inode); | |
66b4ffd1 JB |
3257 | if (ret) |
3258 | goto out; | |
7b128766 | 3259 | } |
3254c876 MX |
3260 | /* release the path since we're done with it */ |
3261 | btrfs_release_path(path); | |
3262 | ||
d68fc57b YZ |
3263 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3264 | ||
3265 | if (root->orphan_block_rsv) | |
3266 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3267 | (u64)-1); | |
3268 | ||
3269 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 3270 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3271 | if (!IS_ERR(trans)) |
3272 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3273 | } |
7b128766 JB |
3274 | |
3275 | if (nr_unlink) | |
4884b476 | 3276 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3277 | if (nr_truncate) |
4884b476 | 3278 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3279 | |
3280 | out: | |
3281 | if (ret) | |
c2cf52eb SK |
3282 | btrfs_crit(root->fs_info, |
3283 | "could not do orphan cleanup %d", ret); | |
66b4ffd1 JB |
3284 | btrfs_free_path(path); |
3285 | return ret; | |
7b128766 JB |
3286 | } |
3287 | ||
46a53cca CM |
3288 | /* |
3289 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3290 | * don't find any xattrs, we know there can't be any acls. | |
3291 | * | |
3292 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3293 | */ | |
3294 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
3295 | int slot, u64 objectid) | |
3296 | { | |
3297 | u32 nritems = btrfs_header_nritems(leaf); | |
3298 | struct btrfs_key found_key; | |
3299 | int scanned = 0; | |
3300 | ||
3301 | slot++; | |
3302 | while (slot < nritems) { | |
3303 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3304 | ||
3305 | /* we found a different objectid, there must not be acls */ | |
3306 | if (found_key.objectid != objectid) | |
3307 | return 0; | |
3308 | ||
3309 | /* we found an xattr, assume we've got an acl */ | |
3310 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
3311 | return 1; | |
3312 | ||
3313 | /* | |
3314 | * we found a key greater than an xattr key, there can't | |
3315 | * be any acls later on | |
3316 | */ | |
3317 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3318 | return 0; | |
3319 | ||
3320 | slot++; | |
3321 | scanned++; | |
3322 | ||
3323 | /* | |
3324 | * it goes inode, inode backrefs, xattrs, extents, | |
3325 | * so if there are a ton of hard links to an inode there can | |
3326 | * be a lot of backrefs. Don't waste time searching too hard, | |
3327 | * this is just an optimization | |
3328 | */ | |
3329 | if (scanned >= 8) | |
3330 | break; | |
3331 | } | |
3332 | /* we hit the end of the leaf before we found an xattr or | |
3333 | * something larger than an xattr. We have to assume the inode | |
3334 | * has acls | |
3335 | */ | |
3336 | return 1; | |
3337 | } | |
3338 | ||
d352ac68 CM |
3339 | /* |
3340 | * read an inode from the btree into the in-memory inode | |
3341 | */ | |
5d4f98a2 | 3342 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3343 | { |
3344 | struct btrfs_path *path; | |
5f39d397 | 3345 | struct extent_buffer *leaf; |
39279cc3 | 3346 | struct btrfs_inode_item *inode_item; |
0b86a832 | 3347 | struct btrfs_timespec *tspec; |
39279cc3 CM |
3348 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3349 | struct btrfs_key location; | |
46a53cca | 3350 | int maybe_acls; |
618e21d5 | 3351 | u32 rdev; |
39279cc3 | 3352 | int ret; |
2f7e33d4 MX |
3353 | bool filled = false; |
3354 | ||
3355 | ret = btrfs_fill_inode(inode, &rdev); | |
3356 | if (!ret) | |
3357 | filled = true; | |
39279cc3 CM |
3358 | |
3359 | path = btrfs_alloc_path(); | |
1748f843 MF |
3360 | if (!path) |
3361 | goto make_bad; | |
3362 | ||
d90c7321 | 3363 | path->leave_spinning = 1; |
39279cc3 | 3364 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3365 | |
39279cc3 | 3366 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3367 | if (ret) |
39279cc3 | 3368 | goto make_bad; |
39279cc3 | 3369 | |
5f39d397 | 3370 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3371 | |
3372 | if (filled) | |
3373 | goto cache_acl; | |
3374 | ||
5f39d397 CM |
3375 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3376 | struct btrfs_inode_item); | |
5f39d397 | 3377 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3378 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3379 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3380 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3381 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
3382 | |
3383 | tspec = btrfs_inode_atime(inode_item); | |
3384 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3385 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3386 | ||
3387 | tspec = btrfs_inode_mtime(inode_item); | |
3388 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3389 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3390 | ||
3391 | tspec = btrfs_inode_ctime(inode_item); | |
3392 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
3393 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
3394 | ||
a76a3cd4 | 3395 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3396 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3397 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3398 | ||
3399 | /* | |
3400 | * If we were modified in the current generation and evicted from memory | |
3401 | * and then re-read we need to do a full sync since we don't have any | |
3402 | * idea about which extents were modified before we were evicted from | |
3403 | * cache. | |
3404 | */ | |
3405 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3406 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3407 | &BTRFS_I(inode)->runtime_flags); | |
3408 | ||
0c4d2d95 | 3409 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 3410 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 3411 | inode->i_rdev = 0; |
5f39d397 CM |
3412 | rdev = btrfs_inode_rdev(leaf, inode_item); |
3413 | ||
aec7477b | 3414 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 3415 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 3416 | cache_acl: |
46a53cca CM |
3417 | /* |
3418 | * try to precache a NULL acl entry for files that don't have | |
3419 | * any xattrs or acls | |
3420 | */ | |
33345d01 LZ |
3421 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
3422 | btrfs_ino(inode)); | |
72c04902 AV |
3423 | if (!maybe_acls) |
3424 | cache_no_acl(inode); | |
46a53cca | 3425 | |
39279cc3 | 3426 | btrfs_free_path(path); |
39279cc3 | 3427 | |
39279cc3 | 3428 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3429 | case S_IFREG: |
3430 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3431 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 3432 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3433 | inode->i_fop = &btrfs_file_operations; |
3434 | inode->i_op = &btrfs_file_inode_operations; | |
3435 | break; | |
3436 | case S_IFDIR: | |
3437 | inode->i_fop = &btrfs_dir_file_operations; | |
3438 | if (root == root->fs_info->tree_root) | |
3439 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3440 | else | |
3441 | inode->i_op = &btrfs_dir_inode_operations; | |
3442 | break; | |
3443 | case S_IFLNK: | |
3444 | inode->i_op = &btrfs_symlink_inode_operations; | |
3445 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 3446 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 3447 | break; |
618e21d5 | 3448 | default: |
0279b4cd | 3449 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3450 | init_special_inode(inode, inode->i_mode, rdev); |
3451 | break; | |
39279cc3 | 3452 | } |
6cbff00f CH |
3453 | |
3454 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3455 | return; |
3456 | ||
3457 | make_bad: | |
39279cc3 | 3458 | btrfs_free_path(path); |
39279cc3 CM |
3459 | make_bad_inode(inode); |
3460 | } | |
3461 | ||
d352ac68 CM |
3462 | /* |
3463 | * given a leaf and an inode, copy the inode fields into the leaf | |
3464 | */ | |
e02119d5 CM |
3465 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3466 | struct extent_buffer *leaf, | |
5f39d397 | 3467 | struct btrfs_inode_item *item, |
39279cc3 CM |
3468 | struct inode *inode) |
3469 | { | |
51fab693 LB |
3470 | struct btrfs_map_token token; |
3471 | ||
3472 | btrfs_init_map_token(&token); | |
5f39d397 | 3473 | |
51fab693 LB |
3474 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3475 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3476 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3477 | &token); | |
3478 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3479 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3480 | |
51fab693 LB |
3481 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), |
3482 | inode->i_atime.tv_sec, &token); | |
3483 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
3484 | inode->i_atime.tv_nsec, &token); | |
5f39d397 | 3485 | |
51fab693 LB |
3486 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), |
3487 | inode->i_mtime.tv_sec, &token); | |
3488 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
3489 | inode->i_mtime.tv_nsec, &token); | |
5f39d397 | 3490 | |
51fab693 LB |
3491 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), |
3492 | inode->i_ctime.tv_sec, &token); | |
3493 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
3494 | inode->i_ctime.tv_nsec, &token); | |
5f39d397 | 3495 | |
51fab693 LB |
3496 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3497 | &token); | |
3498 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3499 | &token); | |
3500 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3501 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3502 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3503 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3504 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3505 | } |
3506 | ||
d352ac68 CM |
3507 | /* |
3508 | * copy everything in the in-memory inode into the btree. | |
3509 | */ | |
2115133f | 3510 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3511 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3512 | { |
3513 | struct btrfs_inode_item *inode_item; | |
3514 | struct btrfs_path *path; | |
5f39d397 | 3515 | struct extent_buffer *leaf; |
39279cc3 CM |
3516 | int ret; |
3517 | ||
3518 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3519 | if (!path) |
3520 | return -ENOMEM; | |
3521 | ||
b9473439 | 3522 | path->leave_spinning = 1; |
16cdcec7 MX |
3523 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3524 | 1); | |
39279cc3 CM |
3525 | if (ret) { |
3526 | if (ret > 0) | |
3527 | ret = -ENOENT; | |
3528 | goto failed; | |
3529 | } | |
3530 | ||
b4ce94de | 3531 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
3532 | leaf = path->nodes[0]; |
3533 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3534 | struct btrfs_inode_item); |
39279cc3 | 3535 | |
e02119d5 | 3536 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3537 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3538 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3539 | ret = 0; |
3540 | failed: | |
39279cc3 CM |
3541 | btrfs_free_path(path); |
3542 | return ret; | |
3543 | } | |
3544 | ||
2115133f CM |
3545 | /* |
3546 | * copy everything in the in-memory inode into the btree. | |
3547 | */ | |
3548 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3549 | struct btrfs_root *root, struct inode *inode) | |
3550 | { | |
3551 | int ret; | |
3552 | ||
3553 | /* | |
3554 | * If the inode is a free space inode, we can deadlock during commit | |
3555 | * if we put it into the delayed code. | |
3556 | * | |
3557 | * The data relocation inode should also be directly updated | |
3558 | * without delay | |
3559 | */ | |
83eea1f1 | 3560 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 3561 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
3562 | btrfs_update_root_times(trans, root); |
3563 | ||
2115133f CM |
3564 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3565 | if (!ret) | |
3566 | btrfs_set_inode_last_trans(trans, inode); | |
3567 | return ret; | |
3568 | } | |
3569 | ||
3570 | return btrfs_update_inode_item(trans, root, inode); | |
3571 | } | |
3572 | ||
be6aef60 JB |
3573 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3574 | struct btrfs_root *root, | |
3575 | struct inode *inode) | |
2115133f CM |
3576 | { |
3577 | int ret; | |
3578 | ||
3579 | ret = btrfs_update_inode(trans, root, inode); | |
3580 | if (ret == -ENOSPC) | |
3581 | return btrfs_update_inode_item(trans, root, inode); | |
3582 | return ret; | |
3583 | } | |
3584 | ||
d352ac68 CM |
3585 | /* |
3586 | * unlink helper that gets used here in inode.c and in the tree logging | |
3587 | * recovery code. It remove a link in a directory with a given name, and | |
3588 | * also drops the back refs in the inode to the directory | |
3589 | */ | |
92986796 AV |
3590 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3591 | struct btrfs_root *root, | |
3592 | struct inode *dir, struct inode *inode, | |
3593 | const char *name, int name_len) | |
39279cc3 CM |
3594 | { |
3595 | struct btrfs_path *path; | |
39279cc3 | 3596 | int ret = 0; |
5f39d397 | 3597 | struct extent_buffer *leaf; |
39279cc3 | 3598 | struct btrfs_dir_item *di; |
5f39d397 | 3599 | struct btrfs_key key; |
aec7477b | 3600 | u64 index; |
33345d01 LZ |
3601 | u64 ino = btrfs_ino(inode); |
3602 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3603 | |
3604 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3605 | if (!path) { |
3606 | ret = -ENOMEM; | |
554233a6 | 3607 | goto out; |
54aa1f4d CM |
3608 | } |
3609 | ||
b9473439 | 3610 | path->leave_spinning = 1; |
33345d01 | 3611 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3612 | name, name_len, -1); |
3613 | if (IS_ERR(di)) { | |
3614 | ret = PTR_ERR(di); | |
3615 | goto err; | |
3616 | } | |
3617 | if (!di) { | |
3618 | ret = -ENOENT; | |
3619 | goto err; | |
3620 | } | |
5f39d397 CM |
3621 | leaf = path->nodes[0]; |
3622 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3623 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3624 | if (ret) |
3625 | goto err; | |
b3b4aa74 | 3626 | btrfs_release_path(path); |
39279cc3 | 3627 | |
33345d01 LZ |
3628 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3629 | dir_ino, &index); | |
aec7477b | 3630 | if (ret) { |
c2cf52eb SK |
3631 | btrfs_info(root->fs_info, |
3632 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
3633 | name_len, name, | |
3634 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 3635 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3636 | goto err; |
3637 | } | |
3638 | ||
16cdcec7 | 3639 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3640 | if (ret) { |
3641 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3642 | goto err; |
79787eaa | 3643 | } |
39279cc3 | 3644 | |
e02119d5 | 3645 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3646 | inode, dir_ino); |
79787eaa JM |
3647 | if (ret != 0 && ret != -ENOENT) { |
3648 | btrfs_abort_transaction(trans, root, ret); | |
3649 | goto err; | |
3650 | } | |
e02119d5 CM |
3651 | |
3652 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3653 | dir, index); | |
6418c961 CM |
3654 | if (ret == -ENOENT) |
3655 | ret = 0; | |
d4e3991b ZB |
3656 | else if (ret) |
3657 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3658 | err: |
3659 | btrfs_free_path(path); | |
e02119d5 CM |
3660 | if (ret) |
3661 | goto out; | |
3662 | ||
3663 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3664 | inode_inc_iversion(inode); |
3665 | inode_inc_iversion(dir); | |
e02119d5 | 3666 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3667 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3668 | out: |
39279cc3 CM |
3669 | return ret; |
3670 | } | |
3671 | ||
92986796 AV |
3672 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3673 | struct btrfs_root *root, | |
3674 | struct inode *dir, struct inode *inode, | |
3675 | const char *name, int name_len) | |
3676 | { | |
3677 | int ret; | |
3678 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
3679 | if (!ret) { | |
3680 | btrfs_drop_nlink(inode); | |
3681 | ret = btrfs_update_inode(trans, root, inode); | |
3682 | } | |
3683 | return ret; | |
3684 | } | |
39279cc3 | 3685 | |
a22285a6 YZ |
3686 | /* |
3687 | * helper to start transaction for unlink and rmdir. | |
3688 | * | |
d52be818 JB |
3689 | * unlink and rmdir are special in btrfs, they do not always free space, so |
3690 | * if we cannot make our reservations the normal way try and see if there is | |
3691 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
3692 | * allow the unlink to occur. | |
a22285a6 | 3693 | */ |
d52be818 | 3694 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 3695 | { |
39279cc3 | 3696 | struct btrfs_trans_handle *trans; |
a22285a6 | 3697 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
3698 | int ret; |
3699 | ||
e70bea5f JB |
3700 | /* |
3701 | * 1 for the possible orphan item | |
3702 | * 1 for the dir item | |
3703 | * 1 for the dir index | |
3704 | * 1 for the inode ref | |
e70bea5f JB |
3705 | * 1 for the inode |
3706 | */ | |
6e137ed3 | 3707 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
3708 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3709 | return trans; | |
4df27c4d | 3710 | |
d52be818 JB |
3711 | if (PTR_ERR(trans) == -ENOSPC) { |
3712 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
a22285a6 | 3713 | |
d52be818 JB |
3714 | trans = btrfs_start_transaction(root, 0); |
3715 | if (IS_ERR(trans)) | |
3716 | return trans; | |
3717 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
3718 | &root->fs_info->trans_block_rsv, | |
3719 | num_bytes, 5); | |
3720 | if (ret) { | |
3721 | btrfs_end_transaction(trans, root); | |
3722 | return ERR_PTR(ret); | |
a22285a6 | 3723 | } |
5a77d76c | 3724 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 3725 | trans->bytes_reserved = num_bytes; |
a22285a6 | 3726 | } |
d52be818 | 3727 | return trans; |
a22285a6 YZ |
3728 | } |
3729 | ||
3730 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3731 | { | |
3732 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3733 | struct btrfs_trans_handle *trans; | |
3734 | struct inode *inode = dentry->d_inode; | |
3735 | int ret; | |
a22285a6 | 3736 | |
d52be818 | 3737 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
3738 | if (IS_ERR(trans)) |
3739 | return PTR_ERR(trans); | |
5f39d397 | 3740 | |
12fcfd22 CM |
3741 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3742 | ||
e02119d5 CM |
3743 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3744 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3745 | if (ret) |
3746 | goto out; | |
7b128766 | 3747 | |
a22285a6 | 3748 | if (inode->i_nlink == 0) { |
7b128766 | 3749 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3750 | if (ret) |
3751 | goto out; | |
a22285a6 | 3752 | } |
7b128766 | 3753 | |
b532402e | 3754 | out: |
d52be818 | 3755 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3756 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3757 | return ret; |
3758 | } | |
3759 | ||
4df27c4d YZ |
3760 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3761 | struct btrfs_root *root, | |
3762 | struct inode *dir, u64 objectid, | |
3763 | const char *name, int name_len) | |
3764 | { | |
3765 | struct btrfs_path *path; | |
3766 | struct extent_buffer *leaf; | |
3767 | struct btrfs_dir_item *di; | |
3768 | struct btrfs_key key; | |
3769 | u64 index; | |
3770 | int ret; | |
33345d01 | 3771 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3772 | |
3773 | path = btrfs_alloc_path(); | |
3774 | if (!path) | |
3775 | return -ENOMEM; | |
3776 | ||
33345d01 | 3777 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3778 | name, name_len, -1); |
79787eaa JM |
3779 | if (IS_ERR_OR_NULL(di)) { |
3780 | if (!di) | |
3781 | ret = -ENOENT; | |
3782 | else | |
3783 | ret = PTR_ERR(di); | |
3784 | goto out; | |
3785 | } | |
4df27c4d YZ |
3786 | |
3787 | leaf = path->nodes[0]; | |
3788 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3789 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3790 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3791 | if (ret) { |
3792 | btrfs_abort_transaction(trans, root, ret); | |
3793 | goto out; | |
3794 | } | |
b3b4aa74 | 3795 | btrfs_release_path(path); |
4df27c4d YZ |
3796 | |
3797 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3798 | objectid, root->root_key.objectid, | |
33345d01 | 3799 | dir_ino, &index, name, name_len); |
4df27c4d | 3800 | if (ret < 0) { |
79787eaa JM |
3801 | if (ret != -ENOENT) { |
3802 | btrfs_abort_transaction(trans, root, ret); | |
3803 | goto out; | |
3804 | } | |
33345d01 | 3805 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3806 | name, name_len); |
79787eaa JM |
3807 | if (IS_ERR_OR_NULL(di)) { |
3808 | if (!di) | |
3809 | ret = -ENOENT; | |
3810 | else | |
3811 | ret = PTR_ERR(di); | |
3812 | btrfs_abort_transaction(trans, root, ret); | |
3813 | goto out; | |
3814 | } | |
4df27c4d YZ |
3815 | |
3816 | leaf = path->nodes[0]; | |
3817 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3818 | btrfs_release_path(path); |
4df27c4d YZ |
3819 | index = key.offset; |
3820 | } | |
945d8962 | 3821 | btrfs_release_path(path); |
4df27c4d | 3822 | |
16cdcec7 | 3823 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3824 | if (ret) { |
3825 | btrfs_abort_transaction(trans, root, ret); | |
3826 | goto out; | |
3827 | } | |
4df27c4d YZ |
3828 | |
3829 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3830 | inode_inc_iversion(dir); |
4df27c4d | 3831 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3832 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3833 | if (ret) |
3834 | btrfs_abort_transaction(trans, root, ret); | |
3835 | out: | |
71d7aed0 | 3836 | btrfs_free_path(path); |
79787eaa | 3837 | return ret; |
4df27c4d YZ |
3838 | } |
3839 | ||
39279cc3 CM |
3840 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3841 | { | |
3842 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3843 | int err = 0; |
39279cc3 | 3844 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3845 | struct btrfs_trans_handle *trans; |
39279cc3 | 3846 | |
b3ae244e | 3847 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3848 | return -ENOTEMPTY; |
b3ae244e DS |
3849 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3850 | return -EPERM; | |
134d4512 | 3851 | |
d52be818 | 3852 | trans = __unlink_start_trans(dir); |
a22285a6 | 3853 | if (IS_ERR(trans)) |
5df6a9f6 | 3854 | return PTR_ERR(trans); |
5df6a9f6 | 3855 | |
33345d01 | 3856 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3857 | err = btrfs_unlink_subvol(trans, root, dir, |
3858 | BTRFS_I(inode)->location.objectid, | |
3859 | dentry->d_name.name, | |
3860 | dentry->d_name.len); | |
3861 | goto out; | |
3862 | } | |
3863 | ||
7b128766 JB |
3864 | err = btrfs_orphan_add(trans, inode); |
3865 | if (err) | |
4df27c4d | 3866 | goto out; |
7b128766 | 3867 | |
39279cc3 | 3868 | /* now the directory is empty */ |
e02119d5 CM |
3869 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3870 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3871 | if (!err) |
dbe674a9 | 3872 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3873 | out: |
d52be818 | 3874 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3875 | btrfs_btree_balance_dirty(root); |
3954401f | 3876 | |
39279cc3 CM |
3877 | return err; |
3878 | } | |
3879 | ||
39279cc3 CM |
3880 | /* |
3881 | * this can truncate away extent items, csum items and directory items. | |
3882 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3883 | * any higher than new_size |
39279cc3 CM |
3884 | * |
3885 | * csum items that cross the new i_size are truncated to the new size | |
3886 | * as well. | |
7b128766 JB |
3887 | * |
3888 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3889 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3890 | */ |
8082510e YZ |
3891 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3892 | struct btrfs_root *root, | |
3893 | struct inode *inode, | |
3894 | u64 new_size, u32 min_type) | |
39279cc3 | 3895 | { |
39279cc3 | 3896 | struct btrfs_path *path; |
5f39d397 | 3897 | struct extent_buffer *leaf; |
39279cc3 | 3898 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3899 | struct btrfs_key key; |
3900 | struct btrfs_key found_key; | |
39279cc3 | 3901 | u64 extent_start = 0; |
db94535d | 3902 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3903 | u64 extent_offset = 0; |
39279cc3 | 3904 | u64 item_end = 0; |
8082510e | 3905 | u32 found_type = (u8)-1; |
39279cc3 CM |
3906 | int found_extent; |
3907 | int del_item; | |
85e21bac CM |
3908 | int pending_del_nr = 0; |
3909 | int pending_del_slot = 0; | |
179e29e4 | 3910 | int extent_type = -1; |
8082510e YZ |
3911 | int ret; |
3912 | int err = 0; | |
33345d01 | 3913 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3914 | |
3915 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3916 | |
0eb0e19c MF |
3917 | path = btrfs_alloc_path(); |
3918 | if (!path) | |
3919 | return -ENOMEM; | |
3920 | path->reada = -1; | |
3921 | ||
5dc562c5 JB |
3922 | /* |
3923 | * We want to drop from the next block forward in case this new size is | |
3924 | * not block aligned since we will be keeping the last block of the | |
3925 | * extent just the way it is. | |
3926 | */ | |
0af3d00b | 3927 | if (root->ref_cows || root == root->fs_info->tree_root) |
fda2832f QW |
3928 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
3929 | root->sectorsize), (u64)-1, 0); | |
8082510e | 3930 | |
16cdcec7 MX |
3931 | /* |
3932 | * This function is also used to drop the items in the log tree before | |
3933 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3934 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3935 | * items. | |
3936 | */ | |
3937 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3938 | btrfs_kill_delayed_inode_items(inode); | |
3939 | ||
33345d01 | 3940 | key.objectid = ino; |
39279cc3 | 3941 | key.offset = (u64)-1; |
5f39d397 CM |
3942 | key.type = (u8)-1; |
3943 | ||
85e21bac | 3944 | search_again: |
b9473439 | 3945 | path->leave_spinning = 1; |
85e21bac | 3946 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3947 | if (ret < 0) { |
3948 | err = ret; | |
3949 | goto out; | |
3950 | } | |
d397712b | 3951 | |
85e21bac | 3952 | if (ret > 0) { |
e02119d5 CM |
3953 | /* there are no items in the tree for us to truncate, we're |
3954 | * done | |
3955 | */ | |
8082510e YZ |
3956 | if (path->slots[0] == 0) |
3957 | goto out; | |
85e21bac CM |
3958 | path->slots[0]--; |
3959 | } | |
3960 | ||
d397712b | 3961 | while (1) { |
39279cc3 | 3962 | fi = NULL; |
5f39d397 CM |
3963 | leaf = path->nodes[0]; |
3964 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3965 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3966 | |
33345d01 | 3967 | if (found_key.objectid != ino) |
39279cc3 | 3968 | break; |
5f39d397 | 3969 | |
85e21bac | 3970 | if (found_type < min_type) |
39279cc3 CM |
3971 | break; |
3972 | ||
5f39d397 | 3973 | item_end = found_key.offset; |
39279cc3 | 3974 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3975 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3976 | struct btrfs_file_extent_item); |
179e29e4 CM |
3977 | extent_type = btrfs_file_extent_type(leaf, fi); |
3978 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3979 | item_end += |
db94535d | 3980 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3981 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3982 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3983 | fi); |
39279cc3 | 3984 | } |
008630c1 | 3985 | item_end--; |
39279cc3 | 3986 | } |
8082510e YZ |
3987 | if (found_type > min_type) { |
3988 | del_item = 1; | |
3989 | } else { | |
3990 | if (item_end < new_size) | |
b888db2b | 3991 | break; |
8082510e YZ |
3992 | if (found_key.offset >= new_size) |
3993 | del_item = 1; | |
3994 | else | |
3995 | del_item = 0; | |
39279cc3 | 3996 | } |
39279cc3 | 3997 | found_extent = 0; |
39279cc3 | 3998 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3999 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4000 | goto delete; | |
4001 | ||
4002 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4003 | u64 num_dec; |
db94535d | 4004 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4005 | if (!del_item) { |
db94535d CM |
4006 | u64 orig_num_bytes = |
4007 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4008 | extent_num_bytes = ALIGN(new_size - |
4009 | found_key.offset, | |
4010 | root->sectorsize); | |
db94535d CM |
4011 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4012 | extent_num_bytes); | |
4013 | num_dec = (orig_num_bytes - | |
9069218d | 4014 | extent_num_bytes); |
e02119d5 | 4015 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 4016 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4017 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4018 | } else { |
db94535d CM |
4019 | extent_num_bytes = |
4020 | btrfs_file_extent_disk_num_bytes(leaf, | |
4021 | fi); | |
5d4f98a2 YZ |
4022 | extent_offset = found_key.offset - |
4023 | btrfs_file_extent_offset(leaf, fi); | |
4024 | ||
39279cc3 | 4025 | /* FIXME blocksize != 4096 */ |
9069218d | 4026 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4027 | if (extent_start != 0) { |
4028 | found_extent = 1; | |
e02119d5 | 4029 | if (root->ref_cows) |
a76a3cd4 | 4030 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4031 | } |
39279cc3 | 4032 | } |
9069218d | 4033 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4034 | /* |
4035 | * we can't truncate inline items that have had | |
4036 | * special encodings | |
4037 | */ | |
4038 | if (!del_item && | |
4039 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
4040 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
4041 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
4042 | u32 size = new_size - found_key.offset; |
4043 | ||
4044 | if (root->ref_cows) { | |
a76a3cd4 YZ |
4045 | inode_sub_bytes(inode, item_end + 1 - |
4046 | new_size); | |
e02119d5 CM |
4047 | } |
4048 | size = | |
4049 | btrfs_file_extent_calc_inline_size(size); | |
afe5fea7 | 4050 | btrfs_truncate_item(root, path, size, 1); |
e02119d5 | 4051 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
4052 | inode_sub_bytes(inode, item_end + 1 - |
4053 | found_key.offset); | |
9069218d | 4054 | } |
39279cc3 | 4055 | } |
179e29e4 | 4056 | delete: |
39279cc3 | 4057 | if (del_item) { |
85e21bac CM |
4058 | if (!pending_del_nr) { |
4059 | /* no pending yet, add ourselves */ | |
4060 | pending_del_slot = path->slots[0]; | |
4061 | pending_del_nr = 1; | |
4062 | } else if (pending_del_nr && | |
4063 | path->slots[0] + 1 == pending_del_slot) { | |
4064 | /* hop on the pending chunk */ | |
4065 | pending_del_nr++; | |
4066 | pending_del_slot = path->slots[0]; | |
4067 | } else { | |
d397712b | 4068 | BUG(); |
85e21bac | 4069 | } |
39279cc3 CM |
4070 | } else { |
4071 | break; | |
4072 | } | |
0af3d00b JB |
4073 | if (found_extent && (root->ref_cows || |
4074 | root == root->fs_info->tree_root)) { | |
b9473439 | 4075 | btrfs_set_path_blocking(path); |
39279cc3 | 4076 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4077 | extent_num_bytes, 0, |
4078 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4079 | ino, extent_offset, 0); |
39279cc3 CM |
4080 | BUG_ON(ret); |
4081 | } | |
85e21bac | 4082 | |
8082510e YZ |
4083 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4084 | break; | |
4085 | ||
4086 | if (path->slots[0] == 0 || | |
4087 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
4088 | if (pending_del_nr) { |
4089 | ret = btrfs_del_items(trans, root, path, | |
4090 | pending_del_slot, | |
4091 | pending_del_nr); | |
79787eaa JM |
4092 | if (ret) { |
4093 | btrfs_abort_transaction(trans, | |
4094 | root, ret); | |
4095 | goto error; | |
4096 | } | |
8082510e YZ |
4097 | pending_del_nr = 0; |
4098 | } | |
b3b4aa74 | 4099 | btrfs_release_path(path); |
85e21bac | 4100 | goto search_again; |
8082510e YZ |
4101 | } else { |
4102 | path->slots[0]--; | |
85e21bac | 4103 | } |
39279cc3 | 4104 | } |
8082510e | 4105 | out: |
85e21bac CM |
4106 | if (pending_del_nr) { |
4107 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4108 | pending_del_nr); | |
79787eaa JM |
4109 | if (ret) |
4110 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4111 | } |
79787eaa | 4112 | error: |
39279cc3 | 4113 | btrfs_free_path(path); |
8082510e | 4114 | return err; |
39279cc3 CM |
4115 | } |
4116 | ||
4117 | /* | |
2aaa6655 JB |
4118 | * btrfs_truncate_page - read, zero a chunk and write a page |
4119 | * @inode - inode that we're zeroing | |
4120 | * @from - the offset to start zeroing | |
4121 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4122 | * offset | |
4123 | * @front - zero up to the offset instead of from the offset on | |
4124 | * | |
4125 | * This will find the page for the "from" offset and cow the page and zero the | |
4126 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4127 | */ |
2aaa6655 JB |
4128 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4129 | int front) | |
39279cc3 | 4130 | { |
2aaa6655 | 4131 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4132 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4133 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4134 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4135 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4136 | char *kaddr; |
db94535d | 4137 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4138 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4139 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4140 | struct page *page; | |
3b16a4e3 | 4141 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4142 | int ret = 0; |
a52d9a80 | 4143 | u64 page_start; |
e6dcd2dc | 4144 | u64 page_end; |
39279cc3 | 4145 | |
2aaa6655 JB |
4146 | if ((offset & (blocksize - 1)) == 0 && |
4147 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4148 | goto out; |
0ca1f7ce | 4149 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4150 | if (ret) |
4151 | goto out; | |
39279cc3 | 4152 | |
211c17f5 | 4153 | again: |
3b16a4e3 | 4154 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4155 | if (!page) { |
0ca1f7ce | 4156 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4157 | ret = -ENOMEM; |
39279cc3 | 4158 | goto out; |
5d5e103a | 4159 | } |
e6dcd2dc CM |
4160 | |
4161 | page_start = page_offset(page); | |
4162 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4163 | ||
39279cc3 | 4164 | if (!PageUptodate(page)) { |
9ebefb18 | 4165 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4166 | lock_page(page); |
211c17f5 CM |
4167 | if (page->mapping != mapping) { |
4168 | unlock_page(page); | |
4169 | page_cache_release(page); | |
4170 | goto again; | |
4171 | } | |
39279cc3 CM |
4172 | if (!PageUptodate(page)) { |
4173 | ret = -EIO; | |
89642229 | 4174 | goto out_unlock; |
39279cc3 CM |
4175 | } |
4176 | } | |
211c17f5 | 4177 | wait_on_page_writeback(page); |
e6dcd2dc | 4178 | |
d0082371 | 4179 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4180 | set_page_extent_mapped(page); |
4181 | ||
4182 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4183 | if (ordered) { | |
2ac55d41 JB |
4184 | unlock_extent_cached(io_tree, page_start, page_end, |
4185 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4186 | unlock_page(page); |
4187 | page_cache_release(page); | |
eb84ae03 | 4188 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4189 | btrfs_put_ordered_extent(ordered); |
4190 | goto again; | |
4191 | } | |
4192 | ||
2ac55d41 | 4193 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4194 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4195 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4196 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4197 | |
2ac55d41 JB |
4198 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4199 | &cached_state); | |
9ed74f2d | 4200 | if (ret) { |
2ac55d41 JB |
4201 | unlock_extent_cached(io_tree, page_start, page_end, |
4202 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4203 | goto out_unlock; |
4204 | } | |
4205 | ||
e6dcd2dc | 4206 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4207 | if (!len) |
4208 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4209 | kaddr = kmap(page); |
2aaa6655 JB |
4210 | if (front) |
4211 | memset(kaddr, 0, offset); | |
4212 | else | |
4213 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4214 | flush_dcache_page(page); |
4215 | kunmap(page); | |
4216 | } | |
247e743c | 4217 | ClearPageChecked(page); |
e6dcd2dc | 4218 | set_page_dirty(page); |
2ac55d41 JB |
4219 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4220 | GFP_NOFS); | |
39279cc3 | 4221 | |
89642229 | 4222 | out_unlock: |
5d5e103a | 4223 | if (ret) |
0ca1f7ce | 4224 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4225 | unlock_page(page); |
4226 | page_cache_release(page); | |
4227 | out: | |
4228 | return ret; | |
4229 | } | |
4230 | ||
695a0d0d JB |
4231 | /* |
4232 | * This function puts in dummy file extents for the area we're creating a hole | |
4233 | * for. So if we are truncating this file to a larger size we need to insert | |
4234 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4235 | * the range between oldsize and size | |
4236 | */ | |
a41ad394 | 4237 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4238 | { |
9036c102 YZ |
4239 | struct btrfs_trans_handle *trans; |
4240 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4241 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4242 | struct extent_map *em = NULL; |
2ac55d41 | 4243 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4244 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4245 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4246 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4247 | u64 last_byte; |
4248 | u64 cur_offset; | |
4249 | u64 hole_size; | |
9ed74f2d | 4250 | int err = 0; |
39279cc3 | 4251 | |
9036c102 YZ |
4252 | if (size <= hole_start) |
4253 | return 0; | |
4254 | ||
9036c102 YZ |
4255 | while (1) { |
4256 | struct btrfs_ordered_extent *ordered; | |
4257 | btrfs_wait_ordered_range(inode, hole_start, | |
4258 | block_end - hole_start); | |
2ac55d41 | 4259 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4260 | &cached_state); |
9036c102 YZ |
4261 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
4262 | if (!ordered) | |
4263 | break; | |
2ac55d41 JB |
4264 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4265 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
4266 | btrfs_put_ordered_extent(ordered); |
4267 | } | |
39279cc3 | 4268 | |
9036c102 YZ |
4269 | cur_offset = hole_start; |
4270 | while (1) { | |
4271 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4272 | block_end - cur_offset, 0); | |
79787eaa JM |
4273 | if (IS_ERR(em)) { |
4274 | err = PTR_ERR(em); | |
f2767956 | 4275 | em = NULL; |
79787eaa JM |
4276 | break; |
4277 | } | |
9036c102 | 4278 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4279 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4280 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4281 | struct extent_map *hole_em; |
9036c102 | 4282 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4283 | |
3642320e | 4284 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
4285 | if (IS_ERR(trans)) { |
4286 | err = PTR_ERR(trans); | |
9ed74f2d | 4287 | break; |
a22285a6 | 4288 | } |
8082510e | 4289 | |
5dc562c5 JB |
4290 | err = btrfs_drop_extents(trans, root, inode, |
4291 | cur_offset, | |
2671485d | 4292 | cur_offset + hole_size, 1); |
5b397377 | 4293 | if (err) { |
79787eaa | 4294 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4295 | btrfs_end_transaction(trans, root); |
3893e33b | 4296 | break; |
5b397377 | 4297 | } |
8082510e | 4298 | |
9036c102 | 4299 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 4300 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
4301 | 0, hole_size, 0, hole_size, |
4302 | 0, 0, 0); | |
5b397377 | 4303 | if (err) { |
79787eaa | 4304 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 4305 | btrfs_end_transaction(trans, root); |
3893e33b | 4306 | break; |
5b397377 | 4307 | } |
8082510e | 4308 | |
5dc562c5 JB |
4309 | btrfs_drop_extent_cache(inode, cur_offset, |
4310 | cur_offset + hole_size - 1, 0); | |
4311 | hole_em = alloc_extent_map(); | |
4312 | if (!hole_em) { | |
4313 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4314 | &BTRFS_I(inode)->runtime_flags); | |
4315 | goto next; | |
4316 | } | |
4317 | hole_em->start = cur_offset; | |
4318 | hole_em->len = hole_size; | |
4319 | hole_em->orig_start = cur_offset; | |
8082510e | 4320 | |
5dc562c5 JB |
4321 | hole_em->block_start = EXTENT_MAP_HOLE; |
4322 | hole_em->block_len = 0; | |
b4939680 | 4323 | hole_em->orig_block_len = 0; |
cc95bef6 | 4324 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4325 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4326 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
4327 | hole_em->generation = trans->transid; | |
8082510e | 4328 | |
5dc562c5 JB |
4329 | while (1) { |
4330 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4331 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4332 | write_unlock(&em_tree->lock); |
4333 | if (err != -EEXIST) | |
4334 | break; | |
4335 | btrfs_drop_extent_cache(inode, cur_offset, | |
4336 | cur_offset + | |
4337 | hole_size - 1, 0); | |
4338 | } | |
4339 | free_extent_map(hole_em); | |
4340 | next: | |
3642320e | 4341 | btrfs_update_inode(trans, root, inode); |
8082510e | 4342 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
4343 | } |
4344 | free_extent_map(em); | |
a22285a6 | 4345 | em = NULL; |
9036c102 | 4346 | cur_offset = last_byte; |
8082510e | 4347 | if (cur_offset >= block_end) |
9036c102 YZ |
4348 | break; |
4349 | } | |
1832a6d5 | 4350 | |
a22285a6 | 4351 | free_extent_map(em); |
2ac55d41 JB |
4352 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4353 | GFP_NOFS); | |
9036c102 YZ |
4354 | return err; |
4355 | } | |
39279cc3 | 4356 | |
3972f260 | 4357 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4358 | { |
f4a2f4c5 MX |
4359 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4360 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4361 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4362 | loff_t newsize = attr->ia_size; |
4363 | int mask = attr->ia_valid; | |
8082510e YZ |
4364 | int ret; |
4365 | ||
a41ad394 | 4366 | if (newsize == oldsize) |
8082510e YZ |
4367 | return 0; |
4368 | ||
3972f260 ES |
4369 | /* |
4370 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4371 | * special case where we need to update the times despite not having | |
4372 | * these flags set. For all other operations the VFS set these flags | |
4373 | * explicitly if it wants a timestamp update. | |
4374 | */ | |
4375 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
4376 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
4377 | ||
a41ad394 | 4378 | if (newsize > oldsize) { |
a41ad394 JB |
4379 | truncate_pagecache(inode, oldsize, newsize); |
4380 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 4381 | if (ret) |
8082510e | 4382 | return ret; |
8082510e | 4383 | |
f4a2f4c5 MX |
4384 | trans = btrfs_start_transaction(root, 1); |
4385 | if (IS_ERR(trans)) | |
4386 | return PTR_ERR(trans); | |
4387 | ||
4388 | i_size_write(inode, newsize); | |
4389 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4390 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 4391 | btrfs_end_transaction(trans, root); |
a41ad394 | 4392 | } else { |
8082510e | 4393 | |
a41ad394 JB |
4394 | /* |
4395 | * We're truncating a file that used to have good data down to | |
4396 | * zero. Make sure it gets into the ordered flush list so that | |
4397 | * any new writes get down to disk quickly. | |
4398 | */ | |
4399 | if (newsize == 0) | |
72ac3c0d JB |
4400 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4401 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4402 | |
f3fe820c JB |
4403 | /* |
4404 | * 1 for the orphan item we're going to add | |
4405 | * 1 for the orphan item deletion. | |
4406 | */ | |
4407 | trans = btrfs_start_transaction(root, 2); | |
4408 | if (IS_ERR(trans)) | |
4409 | return PTR_ERR(trans); | |
4410 | ||
4411 | /* | |
4412 | * We need to do this in case we fail at _any_ point during the | |
4413 | * actual truncate. Once we do the truncate_setsize we could | |
4414 | * invalidate pages which forces any outstanding ordered io to | |
4415 | * be instantly completed which will give us extents that need | |
4416 | * to be truncated. If we fail to get an orphan inode down we | |
4417 | * could have left over extents that were never meant to live, | |
4418 | * so we need to garuntee from this point on that everything | |
4419 | * will be consistent. | |
4420 | */ | |
4421 | ret = btrfs_orphan_add(trans, inode); | |
4422 | btrfs_end_transaction(trans, root); | |
4423 | if (ret) | |
4424 | return ret; | |
4425 | ||
a41ad394 JB |
4426 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4427 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4428 | |
4429 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4430 | btrfs_inode_block_unlocked_dio(inode); | |
4431 | inode_dio_wait(inode); | |
4432 | btrfs_inode_resume_unlocked_dio(inode); | |
4433 | ||
a41ad394 | 4434 | ret = btrfs_truncate(inode); |
f3fe820c JB |
4435 | if (ret && inode->i_nlink) |
4436 | btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
4437 | } |
4438 | ||
a41ad394 | 4439 | return ret; |
8082510e YZ |
4440 | } |
4441 | ||
9036c102 YZ |
4442 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
4443 | { | |
4444 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 4445 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 4446 | int err; |
39279cc3 | 4447 | |
b83cc969 LZ |
4448 | if (btrfs_root_readonly(root)) |
4449 | return -EROFS; | |
4450 | ||
9036c102 YZ |
4451 | err = inode_change_ok(inode, attr); |
4452 | if (err) | |
4453 | return err; | |
2bf5a725 | 4454 | |
5a3f23d5 | 4455 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 4456 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
4457 | if (err) |
4458 | return err; | |
39279cc3 | 4459 | } |
9036c102 | 4460 | |
1025774c CH |
4461 | if (attr->ia_valid) { |
4462 | setattr_copy(inode, attr); | |
0c4d2d95 | 4463 | inode_inc_iversion(inode); |
22c44fe6 | 4464 | err = btrfs_dirty_inode(inode); |
1025774c | 4465 | |
22c44fe6 | 4466 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
4467 | err = btrfs_acl_chmod(inode); |
4468 | } | |
33268eaf | 4469 | |
39279cc3 CM |
4470 | return err; |
4471 | } | |
61295eb8 | 4472 | |
bd555975 | 4473 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
4474 | { |
4475 | struct btrfs_trans_handle *trans; | |
4476 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 4477 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 4478 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
4479 | int ret; |
4480 | ||
1abe9b8a | 4481 | trace_btrfs_inode_evict(inode); |
4482 | ||
39279cc3 | 4483 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 4484 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 4485 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
4486 | goto no_delete; |
4487 | ||
39279cc3 | 4488 | if (is_bad_inode(inode)) { |
7b128766 | 4489 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
4490 | goto no_delete; |
4491 | } | |
bd555975 | 4492 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 4493 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 4494 | |
c71bf099 | 4495 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 4496 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 4497 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
4498 | goto no_delete; |
4499 | } | |
4500 | ||
76dda93c YZ |
4501 | if (inode->i_nlink > 0) { |
4502 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
4503 | goto no_delete; | |
4504 | } | |
4505 | ||
0e8c36a9 MX |
4506 | ret = btrfs_commit_inode_delayed_inode(inode); |
4507 | if (ret) { | |
4508 | btrfs_orphan_del(NULL, inode); | |
4509 | goto no_delete; | |
4510 | } | |
4511 | ||
66d8f3dd | 4512 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
4513 | if (!rsv) { |
4514 | btrfs_orphan_del(NULL, inode); | |
4515 | goto no_delete; | |
4516 | } | |
4a338542 | 4517 | rsv->size = min_size; |
ca7e70f5 | 4518 | rsv->failfast = 1; |
726c35fa | 4519 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 4520 | |
dbe674a9 | 4521 | btrfs_i_size_write(inode, 0); |
5f39d397 | 4522 | |
4289a667 | 4523 | /* |
8407aa46 MX |
4524 | * This is a bit simpler than btrfs_truncate since we've already |
4525 | * reserved our space for our orphan item in the unlink, so we just | |
4526 | * need to reserve some slack space in case we add bytes and update | |
4527 | * inode item when doing the truncate. | |
4289a667 | 4528 | */ |
8082510e | 4529 | while (1) { |
08e007d2 MX |
4530 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
4531 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
4532 | |
4533 | /* | |
4534 | * Try and steal from the global reserve since we will | |
4535 | * likely not use this space anyway, we want to try as | |
4536 | * hard as possible to get this to work. | |
4537 | */ | |
4538 | if (ret) | |
4539 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 4540 | |
d68fc57b | 4541 | if (ret) { |
c2cf52eb SK |
4542 | btrfs_warn(root->fs_info, |
4543 | "Could not get space for a delete, will truncate on mount %d", | |
4544 | ret); | |
4289a667 JB |
4545 | btrfs_orphan_del(NULL, inode); |
4546 | btrfs_free_block_rsv(root, rsv); | |
4547 | goto no_delete; | |
d68fc57b | 4548 | } |
7b128766 | 4549 | |
0e8c36a9 | 4550 | trans = btrfs_join_transaction(root); |
4289a667 JB |
4551 | if (IS_ERR(trans)) { |
4552 | btrfs_orphan_del(NULL, inode); | |
4553 | btrfs_free_block_rsv(root, rsv); | |
4554 | goto no_delete; | |
d68fc57b | 4555 | } |
7b128766 | 4556 | |
4289a667 JB |
4557 | trans->block_rsv = rsv; |
4558 | ||
d68fc57b | 4559 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4560 | if (ret != -ENOSPC) |
8082510e | 4561 | break; |
85e21bac | 4562 | |
8407aa46 | 4563 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4564 | btrfs_end_transaction(trans, root); |
4565 | trans = NULL; | |
b53d3f5d | 4566 | btrfs_btree_balance_dirty(root); |
8082510e | 4567 | } |
5f39d397 | 4568 | |
4289a667 JB |
4569 | btrfs_free_block_rsv(root, rsv); |
4570 | ||
8082510e | 4571 | if (ret == 0) { |
4289a667 | 4572 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
4573 | ret = btrfs_orphan_del(trans, inode); |
4574 | BUG_ON(ret); | |
4575 | } | |
54aa1f4d | 4576 | |
4289a667 | 4577 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4578 | if (!(root == root->fs_info->tree_root || |
4579 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4580 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4581 | |
54aa1f4d | 4582 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4583 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4584 | no_delete: |
89042e5a | 4585 | btrfs_remove_delayed_node(inode); |
dbd5768f | 4586 | clear_inode(inode); |
8082510e | 4587 | return; |
39279cc3 CM |
4588 | } |
4589 | ||
4590 | /* | |
4591 | * this returns the key found in the dir entry in the location pointer. | |
4592 | * If no dir entries were found, location->objectid is 0. | |
4593 | */ | |
4594 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4595 | struct btrfs_key *location) | |
4596 | { | |
4597 | const char *name = dentry->d_name.name; | |
4598 | int namelen = dentry->d_name.len; | |
4599 | struct btrfs_dir_item *di; | |
4600 | struct btrfs_path *path; | |
4601 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4602 | int ret = 0; |
39279cc3 CM |
4603 | |
4604 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4605 | if (!path) |
4606 | return -ENOMEM; | |
3954401f | 4607 | |
33345d01 | 4608 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4609 | namelen, 0); |
0d9f7f3e Y |
4610 | if (IS_ERR(di)) |
4611 | ret = PTR_ERR(di); | |
d397712b | 4612 | |
c704005d | 4613 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4614 | goto out_err; |
d397712b | 4615 | |
5f39d397 | 4616 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4617 | out: |
39279cc3 CM |
4618 | btrfs_free_path(path); |
4619 | return ret; | |
3954401f CM |
4620 | out_err: |
4621 | location->objectid = 0; | |
4622 | goto out; | |
39279cc3 CM |
4623 | } |
4624 | ||
4625 | /* | |
4626 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4627 | * needs to be changed to reflect the root directory of the tree root. This | |
4628 | * is kind of like crossing a mount point. | |
4629 | */ | |
4630 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4631 | struct inode *dir, |
4632 | struct dentry *dentry, | |
4633 | struct btrfs_key *location, | |
4634 | struct btrfs_root **sub_root) | |
39279cc3 | 4635 | { |
4df27c4d YZ |
4636 | struct btrfs_path *path; |
4637 | struct btrfs_root *new_root; | |
4638 | struct btrfs_root_ref *ref; | |
4639 | struct extent_buffer *leaf; | |
4640 | int ret; | |
4641 | int err = 0; | |
39279cc3 | 4642 | |
4df27c4d YZ |
4643 | path = btrfs_alloc_path(); |
4644 | if (!path) { | |
4645 | err = -ENOMEM; | |
4646 | goto out; | |
4647 | } | |
39279cc3 | 4648 | |
4df27c4d YZ |
4649 | err = -ENOENT; |
4650 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
4651 | BTRFS_I(dir)->root->root_key.objectid, | |
4652 | location->objectid); | |
4653 | if (ret) { | |
4654 | if (ret < 0) | |
4655 | err = ret; | |
4656 | goto out; | |
4657 | } | |
39279cc3 | 4658 | |
4df27c4d YZ |
4659 | leaf = path->nodes[0]; |
4660 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4661 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4662 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4663 | goto out; | |
39279cc3 | 4664 | |
4df27c4d YZ |
4665 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4666 | (unsigned long)(ref + 1), | |
4667 | dentry->d_name.len); | |
4668 | if (ret) | |
4669 | goto out; | |
4670 | ||
b3b4aa74 | 4671 | btrfs_release_path(path); |
4df27c4d YZ |
4672 | |
4673 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4674 | if (IS_ERR(new_root)) { | |
4675 | err = PTR_ERR(new_root); | |
4676 | goto out; | |
4677 | } | |
4678 | ||
4df27c4d YZ |
4679 | *sub_root = new_root; |
4680 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4681 | location->type = BTRFS_INODE_ITEM_KEY; | |
4682 | location->offset = 0; | |
4683 | err = 0; | |
4684 | out: | |
4685 | btrfs_free_path(path); | |
4686 | return err; | |
39279cc3 CM |
4687 | } |
4688 | ||
5d4f98a2 YZ |
4689 | static void inode_tree_add(struct inode *inode) |
4690 | { | |
4691 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4692 | struct btrfs_inode *entry; | |
03e860bd FNP |
4693 | struct rb_node **p; |
4694 | struct rb_node *parent; | |
33345d01 | 4695 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 4696 | |
1d3382cb | 4697 | if (inode_unhashed(inode)) |
76dda93c | 4698 | return; |
e1409cef MX |
4699 | again: |
4700 | parent = NULL; | |
5d4f98a2 | 4701 | spin_lock(&root->inode_lock); |
e1409cef | 4702 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
4703 | while (*p) { |
4704 | parent = *p; | |
4705 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4706 | ||
33345d01 | 4707 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4708 | p = &parent->rb_left; |
33345d01 | 4709 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4710 | p = &parent->rb_right; |
5d4f98a2 YZ |
4711 | else { |
4712 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4713 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4714 | rb_erase(parent, &root->inode_tree); |
4715 | RB_CLEAR_NODE(parent); | |
4716 | spin_unlock(&root->inode_lock); | |
4717 | goto again; | |
5d4f98a2 YZ |
4718 | } |
4719 | } | |
4720 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4721 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4722 | spin_unlock(&root->inode_lock); | |
4723 | } | |
4724 | ||
4725 | static void inode_tree_del(struct inode *inode) | |
4726 | { | |
4727 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4728 | int empty = 0; |
5d4f98a2 | 4729 | |
03e860bd | 4730 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4731 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4732 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4733 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4734 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4735 | } |
03e860bd | 4736 | spin_unlock(&root->inode_lock); |
76dda93c | 4737 | |
0af3d00b JB |
4738 | /* |
4739 | * Free space cache has inodes in the tree root, but the tree root has a | |
4740 | * root_refs of 0, so this could end up dropping the tree root as a | |
4741 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4742 | * make sure we don't drop it. | |
4743 | */ | |
4744 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4745 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4746 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4747 | spin_lock(&root->inode_lock); | |
4748 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4749 | spin_unlock(&root->inode_lock); | |
4750 | if (empty) | |
4751 | btrfs_add_dead_root(root); | |
4752 | } | |
4753 | } | |
4754 | ||
143bede5 | 4755 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4756 | { |
4757 | struct rb_node *node; | |
4758 | struct rb_node *prev; | |
4759 | struct btrfs_inode *entry; | |
4760 | struct inode *inode; | |
4761 | u64 objectid = 0; | |
4762 | ||
4763 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4764 | ||
4765 | spin_lock(&root->inode_lock); | |
4766 | again: | |
4767 | node = root->inode_tree.rb_node; | |
4768 | prev = NULL; | |
4769 | while (node) { | |
4770 | prev = node; | |
4771 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4772 | ||
33345d01 | 4773 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4774 | node = node->rb_left; |
33345d01 | 4775 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4776 | node = node->rb_right; |
4777 | else | |
4778 | break; | |
4779 | } | |
4780 | if (!node) { | |
4781 | while (prev) { | |
4782 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4783 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4784 | node = prev; |
4785 | break; | |
4786 | } | |
4787 | prev = rb_next(prev); | |
4788 | } | |
4789 | } | |
4790 | while (node) { | |
4791 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4792 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4793 | inode = igrab(&entry->vfs_inode); |
4794 | if (inode) { | |
4795 | spin_unlock(&root->inode_lock); | |
4796 | if (atomic_read(&inode->i_count) > 1) | |
4797 | d_prune_aliases(inode); | |
4798 | /* | |
45321ac5 | 4799 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4800 | * the inode cache when its usage count |
4801 | * hits zero. | |
4802 | */ | |
4803 | iput(inode); | |
4804 | cond_resched(); | |
4805 | spin_lock(&root->inode_lock); | |
4806 | goto again; | |
4807 | } | |
4808 | ||
4809 | if (cond_resched_lock(&root->inode_lock)) | |
4810 | goto again; | |
4811 | ||
4812 | node = rb_next(node); | |
4813 | } | |
4814 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4815 | } |
4816 | ||
e02119d5 CM |
4817 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4818 | { | |
4819 | struct btrfs_iget_args *args = p; | |
4820 | inode->i_ino = args->ino; | |
e02119d5 | 4821 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4822 | return 0; |
4823 | } | |
4824 | ||
4825 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4826 | { | |
4827 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4828 | return args->ino == btrfs_ino(inode) && |
d397712b | 4829 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4830 | } |
4831 | ||
5d4f98a2 YZ |
4832 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4833 | u64 objectid, | |
4834 | struct btrfs_root *root) | |
39279cc3 CM |
4835 | { |
4836 | struct inode *inode; | |
4837 | struct btrfs_iget_args args; | |
4838 | args.ino = objectid; | |
4839 | args.root = root; | |
4840 | ||
4841 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4842 | btrfs_init_locked_inode, | |
4843 | (void *)&args); | |
4844 | return inode; | |
4845 | } | |
4846 | ||
1a54ef8c BR |
4847 | /* Get an inode object given its location and corresponding root. |
4848 | * Returns in *is_new if the inode was read from disk | |
4849 | */ | |
4850 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4851 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4852 | { |
4853 | struct inode *inode; | |
4854 | ||
4855 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4856 | if (!inode) | |
5d4f98a2 | 4857 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4858 | |
4859 | if (inode->i_state & I_NEW) { | |
4860 | BTRFS_I(inode)->root = root; | |
4861 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4862 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4863 | if (!is_bad_inode(inode)) { |
4864 | inode_tree_add(inode); | |
4865 | unlock_new_inode(inode); | |
4866 | if (new) | |
4867 | *new = 1; | |
4868 | } else { | |
e0b6d65b ST |
4869 | unlock_new_inode(inode); |
4870 | iput(inode); | |
4871 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4872 | } |
4873 | } | |
4874 | ||
1a54ef8c BR |
4875 | return inode; |
4876 | } | |
4877 | ||
4df27c4d YZ |
4878 | static struct inode *new_simple_dir(struct super_block *s, |
4879 | struct btrfs_key *key, | |
4880 | struct btrfs_root *root) | |
4881 | { | |
4882 | struct inode *inode = new_inode(s); | |
4883 | ||
4884 | if (!inode) | |
4885 | return ERR_PTR(-ENOMEM); | |
4886 | ||
4df27c4d YZ |
4887 | BTRFS_I(inode)->root = root; |
4888 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4889 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4890 | |
4891 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4892 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4893 | inode->i_fop = &simple_dir_operations; |
4894 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4895 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4896 | ||
4897 | return inode; | |
4898 | } | |
4899 | ||
3de4586c | 4900 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4901 | { |
d397712b | 4902 | struct inode *inode; |
4df27c4d | 4903 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4904 | struct btrfs_root *sub_root = root; |
4905 | struct btrfs_key location; | |
76dda93c | 4906 | int index; |
b4aff1f8 | 4907 | int ret = 0; |
39279cc3 CM |
4908 | |
4909 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4910 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4911 | |
39e3c955 | 4912 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
4913 | if (ret < 0) |
4914 | return ERR_PTR(ret); | |
5f39d397 | 4915 | |
4df27c4d YZ |
4916 | if (location.objectid == 0) |
4917 | return NULL; | |
4918 | ||
4919 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4920 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4921 | return inode; |
4922 | } | |
4923 | ||
4924 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4925 | ||
76dda93c | 4926 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4927 | ret = fixup_tree_root_location(root, dir, dentry, |
4928 | &location, &sub_root); | |
4929 | if (ret < 0) { | |
4930 | if (ret != -ENOENT) | |
4931 | inode = ERR_PTR(ret); | |
4932 | else | |
4933 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4934 | } else { | |
73f73415 | 4935 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4936 | } |
76dda93c YZ |
4937 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4938 | ||
34d19bad | 4939 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4940 | down_read(&root->fs_info->cleanup_work_sem); |
4941 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4942 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4943 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
4944 | if (ret) { |
4945 | iput(inode); | |
66b4ffd1 | 4946 | inode = ERR_PTR(ret); |
01cd3367 | 4947 | } |
c71bf099 YZ |
4948 | } |
4949 | ||
3de4586c CM |
4950 | return inode; |
4951 | } | |
4952 | ||
fe15ce44 | 4953 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4954 | { |
4955 | struct btrfs_root *root; | |
848cce0d | 4956 | struct inode *inode = dentry->d_inode; |
76dda93c | 4957 | |
848cce0d LZ |
4958 | if (!inode && !IS_ROOT(dentry)) |
4959 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4960 | |
848cce0d LZ |
4961 | if (inode) { |
4962 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4963 | if (btrfs_root_refs(&root->root_item) == 0) |
4964 | return 1; | |
848cce0d LZ |
4965 | |
4966 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4967 | return 1; | |
efefb143 | 4968 | } |
76dda93c YZ |
4969 | return 0; |
4970 | } | |
4971 | ||
b4aff1f8 JB |
4972 | static void btrfs_dentry_release(struct dentry *dentry) |
4973 | { | |
4974 | if (dentry->d_fsdata) | |
4975 | kfree(dentry->d_fsdata); | |
4976 | } | |
4977 | ||
3de4586c | 4978 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4979 | unsigned int flags) |
3de4586c | 4980 | { |
a66e7cc6 JB |
4981 | struct dentry *ret; |
4982 | ||
4983 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
a66e7cc6 | 4984 | return ret; |
39279cc3 CM |
4985 | } |
4986 | ||
16cdcec7 | 4987 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4988 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4989 | }; | |
4990 | ||
cbdf5a24 DW |
4991 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4992 | filldir_t filldir) | |
39279cc3 | 4993 | { |
496ad9aa | 4994 | struct inode *inode = file_inode(filp); |
39279cc3 CM |
4995 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4996 | struct btrfs_item *item; | |
4997 | struct btrfs_dir_item *di; | |
4998 | struct btrfs_key key; | |
5f39d397 | 4999 | struct btrfs_key found_key; |
39279cc3 | 5000 | struct btrfs_path *path; |
16cdcec7 MX |
5001 | struct list_head ins_list; |
5002 | struct list_head del_list; | |
39279cc3 | 5003 | int ret; |
5f39d397 | 5004 | struct extent_buffer *leaf; |
39279cc3 | 5005 | int slot; |
39279cc3 CM |
5006 | unsigned char d_type; |
5007 | int over = 0; | |
5008 | u32 di_cur; | |
5009 | u32 di_total; | |
5010 | u32 di_len; | |
5011 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5012 | char tmp_name[32]; |
5013 | char *name_ptr; | |
5014 | int name_len; | |
16cdcec7 | 5015 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
5016 | |
5017 | /* FIXME, use a real flag for deciding about the key type */ | |
5018 | if (root->fs_info->tree_root == root) | |
5019 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5020 | |
3954401f CM |
5021 | /* special case for "." */ |
5022 | if (filp->f_pos == 0) { | |
3765fefa HS |
5023 | over = filldir(dirent, ".", 1, |
5024 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
5025 | if (over) |
5026 | return 0; | |
5027 | filp->f_pos = 1; | |
5028 | } | |
3954401f CM |
5029 | /* special case for .., just use the back ref */ |
5030 | if (filp->f_pos == 1) { | |
5ecc7e5d | 5031 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 5032 | over = filldir(dirent, "..", 2, |
3765fefa | 5033 | filp->f_pos, pino, DT_DIR); |
3954401f | 5034 | if (over) |
49593bfa | 5035 | return 0; |
3954401f CM |
5036 | filp->f_pos = 2; |
5037 | } | |
49593bfa | 5038 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5039 | if (!path) |
5040 | return -ENOMEM; | |
ff5714cc | 5041 | |
026fd317 | 5042 | path->reada = 1; |
49593bfa | 5043 | |
16cdcec7 MX |
5044 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5045 | INIT_LIST_HEAD(&ins_list); | |
5046 | INIT_LIST_HEAD(&del_list); | |
5047 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5048 | } | |
5049 | ||
39279cc3 CM |
5050 | btrfs_set_key_type(&key, key_type); |
5051 | key.offset = filp->f_pos; | |
33345d01 | 5052 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5053 | |
39279cc3 CM |
5054 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5055 | if (ret < 0) | |
5056 | goto err; | |
49593bfa DW |
5057 | |
5058 | while (1) { | |
5f39d397 | 5059 | leaf = path->nodes[0]; |
39279cc3 | 5060 | slot = path->slots[0]; |
b9e03af0 LZ |
5061 | if (slot >= btrfs_header_nritems(leaf)) { |
5062 | ret = btrfs_next_leaf(root, path); | |
5063 | if (ret < 0) | |
5064 | goto err; | |
5065 | else if (ret > 0) | |
5066 | break; | |
5067 | continue; | |
39279cc3 | 5068 | } |
3de4586c | 5069 | |
5f39d397 CM |
5070 | item = btrfs_item_nr(leaf, slot); |
5071 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
5072 | ||
5073 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5074 | break; |
5f39d397 | 5075 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 5076 | break; |
5f39d397 | 5077 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 5078 | goto next; |
16cdcec7 MX |
5079 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5080 | btrfs_should_delete_dir_index(&del_list, | |
5081 | found_key.offset)) | |
5082 | goto next; | |
5f39d397 CM |
5083 | |
5084 | filp->f_pos = found_key.offset; | |
16cdcec7 | 5085 | is_curr = 1; |
49593bfa | 5086 | |
39279cc3 CM |
5087 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5088 | di_cur = 0; | |
5f39d397 | 5089 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5090 | |
5091 | while (di_cur < di_total) { | |
5f39d397 CM |
5092 | struct btrfs_key location; |
5093 | ||
22a94d44 JB |
5094 | if (verify_dir_item(root, leaf, di)) |
5095 | break; | |
5096 | ||
5f39d397 | 5097 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5098 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5099 | name_ptr = tmp_name; |
5100 | } else { | |
5101 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5102 | if (!name_ptr) { |
5103 | ret = -ENOMEM; | |
5104 | goto err; | |
5105 | } | |
5f39d397 CM |
5106 | } |
5107 | read_extent_buffer(leaf, name_ptr, | |
5108 | (unsigned long)(di + 1), name_len); | |
5109 | ||
5110 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5111 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5112 | |
fede766f | 5113 | |
3de4586c | 5114 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5115 | * skip it. |
5116 | * | |
5117 | * In contrast to old kernels, we insert the snapshot's | |
5118 | * dir item and dir index after it has been created, so | |
5119 | * we won't find a reference to our own snapshot. We | |
5120 | * still keep the following code for backward | |
5121 | * compatibility. | |
3de4586c CM |
5122 | */ |
5123 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5124 | location.objectid == root->root_key.objectid) { | |
5125 | over = 0; | |
5126 | goto skip; | |
5127 | } | |
5f39d397 | 5128 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 5129 | found_key.offset, location.objectid, |
39279cc3 | 5130 | d_type); |
5f39d397 | 5131 | |
3de4586c | 5132 | skip: |
5f39d397 CM |
5133 | if (name_ptr != tmp_name) |
5134 | kfree(name_ptr); | |
5135 | ||
39279cc3 CM |
5136 | if (over) |
5137 | goto nopos; | |
5103e947 | 5138 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5139 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5140 | di_cur += di_len; |
5141 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5142 | } | |
b9e03af0 LZ |
5143 | next: |
5144 | path->slots[0]++; | |
39279cc3 | 5145 | } |
49593bfa | 5146 | |
16cdcec7 MX |
5147 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5148 | if (is_curr) | |
5149 | filp->f_pos++; | |
5150 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
5151 | &ins_list); | |
5152 | if (ret) | |
5153 | goto nopos; | |
5154 | } | |
5155 | ||
49593bfa | 5156 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 5157 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
5158 | /* |
5159 | * 32-bit glibc will use getdents64, but then strtol - | |
5160 | * so the last number we can serve is this. | |
5161 | */ | |
5162 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
5163 | else |
5164 | filp->f_pos++; | |
39279cc3 CM |
5165 | nopos: |
5166 | ret = 0; | |
5167 | err: | |
16cdcec7 MX |
5168 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5169 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5170 | btrfs_free_path(path); |
39279cc3 CM |
5171 | return ret; |
5172 | } | |
5173 | ||
a9185b41 | 5174 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5175 | { |
5176 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5177 | struct btrfs_trans_handle *trans; | |
5178 | int ret = 0; | |
0af3d00b | 5179 | bool nolock = false; |
39279cc3 | 5180 | |
72ac3c0d | 5181 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5182 | return 0; |
5183 | ||
83eea1f1 | 5184 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5185 | nolock = true; |
0af3d00b | 5186 | |
a9185b41 | 5187 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5188 | if (nolock) |
7a7eaa40 | 5189 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5190 | else |
7a7eaa40 | 5191 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5192 | if (IS_ERR(trans)) |
5193 | return PTR_ERR(trans); | |
a698d075 | 5194 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5195 | } |
5196 | return ret; | |
5197 | } | |
5198 | ||
5199 | /* | |
54aa1f4d | 5200 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5201 | * inode changes. But, it is most likely to find the inode in cache. |
5202 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5203 | * to keep or drop this code. | |
5204 | */ | |
48a3b636 | 5205 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5206 | { |
5207 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5208 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5209 | int ret; |
5210 | ||
72ac3c0d | 5211 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5212 | return 0; |
39279cc3 | 5213 | |
7a7eaa40 | 5214 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5215 | if (IS_ERR(trans)) |
5216 | return PTR_ERR(trans); | |
8929ecfa YZ |
5217 | |
5218 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5219 | if (ret && ret == -ENOSPC) { |
5220 | /* whoops, lets try again with the full transaction */ | |
5221 | btrfs_end_transaction(trans, root); | |
5222 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5223 | if (IS_ERR(trans)) |
5224 | return PTR_ERR(trans); | |
8929ecfa | 5225 | |
94b60442 | 5226 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5227 | } |
39279cc3 | 5228 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5229 | if (BTRFS_I(inode)->delayed_node) |
5230 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5231 | |
5232 | return ret; | |
5233 | } | |
5234 | ||
5235 | /* | |
5236 | * This is a copy of file_update_time. We need this so we can return error on | |
5237 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5238 | */ | |
e41f941a JB |
5239 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5240 | int flags) | |
22c44fe6 | 5241 | { |
2bc55652 AB |
5242 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5243 | ||
5244 | if (btrfs_root_readonly(root)) | |
5245 | return -EROFS; | |
5246 | ||
e41f941a | 5247 | if (flags & S_VERSION) |
22c44fe6 | 5248 | inode_inc_iversion(inode); |
e41f941a JB |
5249 | if (flags & S_CTIME) |
5250 | inode->i_ctime = *now; | |
5251 | if (flags & S_MTIME) | |
5252 | inode->i_mtime = *now; | |
5253 | if (flags & S_ATIME) | |
5254 | inode->i_atime = *now; | |
5255 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5256 | } |
5257 | ||
d352ac68 CM |
5258 | /* |
5259 | * find the highest existing sequence number in a directory | |
5260 | * and then set the in-memory index_cnt variable to reflect | |
5261 | * free sequence numbers | |
5262 | */ | |
aec7477b JB |
5263 | static int btrfs_set_inode_index_count(struct inode *inode) |
5264 | { | |
5265 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5266 | struct btrfs_key key, found_key; | |
5267 | struct btrfs_path *path; | |
5268 | struct extent_buffer *leaf; | |
5269 | int ret; | |
5270 | ||
33345d01 | 5271 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
5272 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
5273 | key.offset = (u64)-1; | |
5274 | ||
5275 | path = btrfs_alloc_path(); | |
5276 | if (!path) | |
5277 | return -ENOMEM; | |
5278 | ||
5279 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5280 | if (ret < 0) | |
5281 | goto out; | |
5282 | /* FIXME: we should be able to handle this */ | |
5283 | if (ret == 0) | |
5284 | goto out; | |
5285 | ret = 0; | |
5286 | ||
5287 | /* | |
5288 | * MAGIC NUMBER EXPLANATION: | |
5289 | * since we search a directory based on f_pos we have to start at 2 | |
5290 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5291 | * else has to start at 2 | |
5292 | */ | |
5293 | if (path->slots[0] == 0) { | |
5294 | BTRFS_I(inode)->index_cnt = 2; | |
5295 | goto out; | |
5296 | } | |
5297 | ||
5298 | path->slots[0]--; | |
5299 | ||
5300 | leaf = path->nodes[0]; | |
5301 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5302 | ||
33345d01 | 5303 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
5304 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
5305 | BTRFS_I(inode)->index_cnt = 2; | |
5306 | goto out; | |
5307 | } | |
5308 | ||
5309 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
5310 | out: | |
5311 | btrfs_free_path(path); | |
5312 | return ret; | |
5313 | } | |
5314 | ||
d352ac68 CM |
5315 | /* |
5316 | * helper to find a free sequence number in a given directory. This current | |
5317 | * code is very simple, later versions will do smarter things in the btree | |
5318 | */ | |
3de4586c | 5319 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
5320 | { |
5321 | int ret = 0; | |
5322 | ||
5323 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
5324 | ret = btrfs_inode_delayed_dir_index_count(dir); |
5325 | if (ret) { | |
5326 | ret = btrfs_set_inode_index_count(dir); | |
5327 | if (ret) | |
5328 | return ret; | |
5329 | } | |
aec7477b JB |
5330 | } |
5331 | ||
00e4e6b3 | 5332 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
5333 | BTRFS_I(dir)->index_cnt++; |
5334 | ||
5335 | return ret; | |
5336 | } | |
5337 | ||
39279cc3 CM |
5338 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
5339 | struct btrfs_root *root, | |
aec7477b | 5340 | struct inode *dir, |
9c58309d | 5341 | const char *name, int name_len, |
175a4eb7 AV |
5342 | u64 ref_objectid, u64 objectid, |
5343 | umode_t mode, u64 *index) | |
39279cc3 CM |
5344 | { |
5345 | struct inode *inode; | |
5f39d397 | 5346 | struct btrfs_inode_item *inode_item; |
39279cc3 | 5347 | struct btrfs_key *location; |
5f39d397 | 5348 | struct btrfs_path *path; |
9c58309d CM |
5349 | struct btrfs_inode_ref *ref; |
5350 | struct btrfs_key key[2]; | |
5351 | u32 sizes[2]; | |
5352 | unsigned long ptr; | |
39279cc3 CM |
5353 | int ret; |
5354 | int owner; | |
5355 | ||
5f39d397 | 5356 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5357 | if (!path) |
5358 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 5359 | |
39279cc3 | 5360 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
5361 | if (!inode) { |
5362 | btrfs_free_path(path); | |
39279cc3 | 5363 | return ERR_PTR(-ENOMEM); |
8fb27640 | 5364 | } |
39279cc3 | 5365 | |
581bb050 LZ |
5366 | /* |
5367 | * we have to initialize this early, so we can reclaim the inode | |
5368 | * number if we fail afterwards in this function. | |
5369 | */ | |
5370 | inode->i_ino = objectid; | |
5371 | ||
aec7477b | 5372 | if (dir) { |
1abe9b8a | 5373 | trace_btrfs_inode_request(dir); |
5374 | ||
3de4586c | 5375 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 5376 | if (ret) { |
8fb27640 | 5377 | btrfs_free_path(path); |
09771430 | 5378 | iput(inode); |
aec7477b | 5379 | return ERR_PTR(ret); |
09771430 | 5380 | } |
aec7477b JB |
5381 | } |
5382 | /* | |
5383 | * index_cnt is ignored for everything but a dir, | |
5384 | * btrfs_get_inode_index_count has an explanation for the magic | |
5385 | * number | |
5386 | */ | |
5387 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 5388 | BTRFS_I(inode)->root = root; |
e02119d5 | 5389 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 5390 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 5391 | |
5dc562c5 JB |
5392 | /* |
5393 | * We could have gotten an inode number from somebody who was fsynced | |
5394 | * and then removed in this same transaction, so let's just set full | |
5395 | * sync since it will be a full sync anyway and this will blow away the | |
5396 | * old info in the log. | |
5397 | */ | |
5398 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
5399 | ||
569254b0 | 5400 | if (S_ISDIR(mode)) |
39279cc3 CM |
5401 | owner = 0; |
5402 | else | |
5403 | owner = 1; | |
9c58309d CM |
5404 | |
5405 | key[0].objectid = objectid; | |
5406 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
5407 | key[0].offset = 0; | |
5408 | ||
f186373f MF |
5409 | /* |
5410 | * Start new inodes with an inode_ref. This is slightly more | |
5411 | * efficient for small numbers of hard links since they will | |
5412 | * be packed into one item. Extended refs will kick in if we | |
5413 | * add more hard links than can fit in the ref item. | |
5414 | */ | |
9c58309d CM |
5415 | key[1].objectid = objectid; |
5416 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
5417 | key[1].offset = ref_objectid; | |
5418 | ||
5419 | sizes[0] = sizeof(struct btrfs_inode_item); | |
5420 | sizes[1] = name_len + sizeof(*ref); | |
5421 | ||
b9473439 | 5422 | path->leave_spinning = 1; |
9c58309d CM |
5423 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
5424 | if (ret != 0) | |
5f39d397 CM |
5425 | goto fail; |
5426 | ||
ecc11fab | 5427 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 5428 | inode_set_bytes(inode, 0); |
39279cc3 | 5429 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
5430 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5431 | struct btrfs_inode_item); | |
293f7e07 LZ |
5432 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
5433 | sizeof(*inode_item)); | |
e02119d5 | 5434 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
5435 | |
5436 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
5437 | struct btrfs_inode_ref); | |
5438 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 5439 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
5440 | ptr = (unsigned long)(ref + 1); |
5441 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
5442 | ||
5f39d397 CM |
5443 | btrfs_mark_buffer_dirty(path->nodes[0]); |
5444 | btrfs_free_path(path); | |
5445 | ||
39279cc3 CM |
5446 | location = &BTRFS_I(inode)->location; |
5447 | location->objectid = objectid; | |
39279cc3 CM |
5448 | location->offset = 0; |
5449 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
5450 | ||
6cbff00f CH |
5451 | btrfs_inherit_iflags(inode, dir); |
5452 | ||
569254b0 | 5453 | if (S_ISREG(mode)) { |
94272164 CM |
5454 | if (btrfs_test_opt(root, NODATASUM)) |
5455 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 5456 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
5457 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
5458 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
5459 | } |
5460 | ||
39279cc3 | 5461 | insert_inode_hash(inode); |
5d4f98a2 | 5462 | inode_tree_add(inode); |
1abe9b8a | 5463 | |
5464 | trace_btrfs_inode_new(inode); | |
1973f0fa | 5465 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 5466 | |
8ea05e3a AB |
5467 | btrfs_update_root_times(trans, root); |
5468 | ||
39279cc3 | 5469 | return inode; |
5f39d397 | 5470 | fail: |
aec7477b JB |
5471 | if (dir) |
5472 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 5473 | btrfs_free_path(path); |
09771430 | 5474 | iput(inode); |
5f39d397 | 5475 | return ERR_PTR(ret); |
39279cc3 CM |
5476 | } |
5477 | ||
5478 | static inline u8 btrfs_inode_type(struct inode *inode) | |
5479 | { | |
5480 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
5481 | } | |
5482 | ||
d352ac68 CM |
5483 | /* |
5484 | * utility function to add 'inode' into 'parent_inode' with | |
5485 | * a give name and a given sequence number. | |
5486 | * if 'add_backref' is true, also insert a backref from the | |
5487 | * inode to the parent directory. | |
5488 | */ | |
e02119d5 CM |
5489 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
5490 | struct inode *parent_inode, struct inode *inode, | |
5491 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 5492 | { |
4df27c4d | 5493 | int ret = 0; |
39279cc3 | 5494 | struct btrfs_key key; |
e02119d5 | 5495 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
5496 | u64 ino = btrfs_ino(inode); |
5497 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 5498 | |
33345d01 | 5499 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5500 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
5501 | } else { | |
33345d01 | 5502 | key.objectid = ino; |
4df27c4d YZ |
5503 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
5504 | key.offset = 0; | |
5505 | } | |
5506 | ||
33345d01 | 5507 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5508 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
5509 | key.objectid, root->root_key.objectid, | |
33345d01 | 5510 | parent_ino, index, name, name_len); |
4df27c4d | 5511 | } else if (add_backref) { |
33345d01 LZ |
5512 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
5513 | parent_ino, index); | |
4df27c4d | 5514 | } |
39279cc3 | 5515 | |
79787eaa JM |
5516 | /* Nothing to clean up yet */ |
5517 | if (ret) | |
5518 | return ret; | |
4df27c4d | 5519 | |
79787eaa JM |
5520 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
5521 | parent_inode, &key, | |
5522 | btrfs_inode_type(inode), index); | |
9c52057c | 5523 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
5524 | goto fail_dir_item; |
5525 | else if (ret) { | |
5526 | btrfs_abort_transaction(trans, root, ret); | |
5527 | return ret; | |
39279cc3 | 5528 | } |
79787eaa JM |
5529 | |
5530 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
5531 | name_len * 2); | |
0c4d2d95 | 5532 | inode_inc_iversion(parent_inode); |
79787eaa JM |
5533 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
5534 | ret = btrfs_update_inode(trans, root, parent_inode); | |
5535 | if (ret) | |
5536 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 5537 | return ret; |
fe66a05a CM |
5538 | |
5539 | fail_dir_item: | |
5540 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
5541 | u64 local_index; | |
5542 | int err; | |
5543 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5544 | key.objectid, root->root_key.objectid, | |
5545 | parent_ino, &local_index, name, name_len); | |
5546 | ||
5547 | } else if (add_backref) { | |
5548 | u64 local_index; | |
5549 | int err; | |
5550 | ||
5551 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5552 | ino, parent_ino, &local_index); | |
5553 | } | |
5554 | return ret; | |
39279cc3 CM |
5555 | } |
5556 | ||
5557 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5558 | struct inode *dir, struct dentry *dentry, |
5559 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5560 | { |
a1b075d2 JB |
5561 | int err = btrfs_add_link(trans, dir, inode, |
5562 | dentry->d_name.name, dentry->d_name.len, | |
5563 | backref, index); | |
39279cc3 CM |
5564 | if (err > 0) |
5565 | err = -EEXIST; | |
5566 | return err; | |
5567 | } | |
5568 | ||
618e21d5 | 5569 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5570 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5571 | { |
5572 | struct btrfs_trans_handle *trans; | |
5573 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5574 | struct inode *inode = NULL; |
618e21d5 JB |
5575 | int err; |
5576 | int drop_inode = 0; | |
5577 | u64 objectid; | |
00e4e6b3 | 5578 | u64 index = 0; |
618e21d5 JB |
5579 | |
5580 | if (!new_valid_dev(rdev)) | |
5581 | return -EINVAL; | |
5582 | ||
9ed74f2d JB |
5583 | /* |
5584 | * 2 for inode item and ref | |
5585 | * 2 for dir items | |
5586 | * 1 for xattr if selinux is on | |
5587 | */ | |
a22285a6 YZ |
5588 | trans = btrfs_start_transaction(root, 5); |
5589 | if (IS_ERR(trans)) | |
5590 | return PTR_ERR(trans); | |
1832a6d5 | 5591 | |
581bb050 LZ |
5592 | err = btrfs_find_free_ino(root, &objectid); |
5593 | if (err) | |
5594 | goto out_unlock; | |
5595 | ||
aec7477b | 5596 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5597 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5598 | mode, &index); |
7cf96da3 TI |
5599 | if (IS_ERR(inode)) { |
5600 | err = PTR_ERR(inode); | |
618e21d5 | 5601 | goto out_unlock; |
7cf96da3 | 5602 | } |
618e21d5 | 5603 | |
2a7dba39 | 5604 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5605 | if (err) { |
5606 | drop_inode = 1; | |
5607 | goto out_unlock; | |
5608 | } | |
5609 | ||
ad19db71 CS |
5610 | /* |
5611 | * If the active LSM wants to access the inode during | |
5612 | * d_instantiate it needs these. Smack checks to see | |
5613 | * if the filesystem supports xattrs by looking at the | |
5614 | * ops vector. | |
5615 | */ | |
5616 | ||
5617 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5618 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5619 | if (err) |
5620 | drop_inode = 1; | |
5621 | else { | |
618e21d5 | 5622 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5623 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5624 | d_instantiate(dentry, inode); |
618e21d5 | 5625 | } |
618e21d5 | 5626 | out_unlock: |
7ad85bb7 | 5627 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5628 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5629 | if (drop_inode) { |
5630 | inode_dec_link_count(inode); | |
5631 | iput(inode); | |
5632 | } | |
618e21d5 JB |
5633 | return err; |
5634 | } | |
5635 | ||
39279cc3 | 5636 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5637 | umode_t mode, bool excl) |
39279cc3 CM |
5638 | { |
5639 | struct btrfs_trans_handle *trans; | |
5640 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5641 | struct inode *inode = NULL; |
43baa579 | 5642 | int drop_inode_on_err = 0; |
a22285a6 | 5643 | int err; |
39279cc3 | 5644 | u64 objectid; |
00e4e6b3 | 5645 | u64 index = 0; |
39279cc3 | 5646 | |
9ed74f2d JB |
5647 | /* |
5648 | * 2 for inode item and ref | |
5649 | * 2 for dir items | |
5650 | * 1 for xattr if selinux is on | |
5651 | */ | |
a22285a6 YZ |
5652 | trans = btrfs_start_transaction(root, 5); |
5653 | if (IS_ERR(trans)) | |
5654 | return PTR_ERR(trans); | |
9ed74f2d | 5655 | |
581bb050 LZ |
5656 | err = btrfs_find_free_ino(root, &objectid); |
5657 | if (err) | |
5658 | goto out_unlock; | |
5659 | ||
aec7477b | 5660 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5661 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5662 | mode, &index); |
7cf96da3 TI |
5663 | if (IS_ERR(inode)) { |
5664 | err = PTR_ERR(inode); | |
39279cc3 | 5665 | goto out_unlock; |
7cf96da3 | 5666 | } |
43baa579 | 5667 | drop_inode_on_err = 1; |
39279cc3 | 5668 | |
2a7dba39 | 5669 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5670 | if (err) |
33268eaf | 5671 | goto out_unlock; |
33268eaf | 5672 | |
9185aa58 FB |
5673 | err = btrfs_update_inode(trans, root, inode); |
5674 | if (err) | |
5675 | goto out_unlock; | |
5676 | ||
ad19db71 CS |
5677 | /* |
5678 | * If the active LSM wants to access the inode during | |
5679 | * d_instantiate it needs these. Smack checks to see | |
5680 | * if the filesystem supports xattrs by looking at the | |
5681 | * ops vector. | |
5682 | */ | |
5683 | inode->i_fop = &btrfs_file_operations; | |
5684 | inode->i_op = &btrfs_file_inode_operations; | |
5685 | ||
a1b075d2 | 5686 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5687 | if (err) |
43baa579 FB |
5688 | goto out_unlock; |
5689 | ||
5690 | inode->i_mapping->a_ops = &btrfs_aops; | |
5691 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5692 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5693 | d_instantiate(dentry, inode); | |
5694 | ||
39279cc3 | 5695 | out_unlock: |
7ad85bb7 | 5696 | btrfs_end_transaction(trans, root); |
43baa579 | 5697 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5698 | inode_dec_link_count(inode); |
5699 | iput(inode); | |
5700 | } | |
b53d3f5d | 5701 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5702 | return err; |
5703 | } | |
5704 | ||
5705 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5706 | struct dentry *dentry) | |
5707 | { | |
5708 | struct btrfs_trans_handle *trans; | |
5709 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5710 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5711 | u64 index; |
39279cc3 CM |
5712 | int err; |
5713 | int drop_inode = 0; | |
5714 | ||
4a8be425 TH |
5715 | /* do not allow sys_link's with other subvols of the same device */ |
5716 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5717 | return -EXDEV; |
4a8be425 | 5718 | |
f186373f | 5719 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5720 | return -EMLINK; |
4a8be425 | 5721 | |
3de4586c | 5722 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5723 | if (err) |
5724 | goto fail; | |
5725 | ||
a22285a6 | 5726 | /* |
7e6b6465 | 5727 | * 2 items for inode and inode ref |
a22285a6 | 5728 | * 2 items for dir items |
7e6b6465 | 5729 | * 1 item for parent inode |
a22285a6 | 5730 | */ |
7e6b6465 | 5731 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5732 | if (IS_ERR(trans)) { |
5733 | err = PTR_ERR(trans); | |
5734 | goto fail; | |
5735 | } | |
5f39d397 | 5736 | |
3153495d | 5737 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5738 | inode_inc_iversion(inode); |
3153495d | 5739 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5740 | ihold(inode); |
e9976151 | 5741 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5742 | |
a1b075d2 | 5743 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5744 | |
a5719521 | 5745 | if (err) { |
54aa1f4d | 5746 | drop_inode = 1; |
a5719521 | 5747 | } else { |
10d9f309 | 5748 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5749 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5750 | if (err) |
5751 | goto fail; | |
08c422c2 | 5752 | d_instantiate(dentry, inode); |
6a912213 | 5753 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5754 | } |
39279cc3 | 5755 | |
7ad85bb7 | 5756 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5757 | fail: |
39279cc3 CM |
5758 | if (drop_inode) { |
5759 | inode_dec_link_count(inode); | |
5760 | iput(inode); | |
5761 | } | |
b53d3f5d | 5762 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5763 | return err; |
5764 | } | |
5765 | ||
18bb1db3 | 5766 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5767 | { |
b9d86667 | 5768 | struct inode *inode = NULL; |
39279cc3 CM |
5769 | struct btrfs_trans_handle *trans; |
5770 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5771 | int err = 0; | |
5772 | int drop_on_err = 0; | |
b9d86667 | 5773 | u64 objectid = 0; |
00e4e6b3 | 5774 | u64 index = 0; |
39279cc3 | 5775 | |
9ed74f2d JB |
5776 | /* |
5777 | * 2 items for inode and ref | |
5778 | * 2 items for dir items | |
5779 | * 1 for xattr if selinux is on | |
5780 | */ | |
a22285a6 YZ |
5781 | trans = btrfs_start_transaction(root, 5); |
5782 | if (IS_ERR(trans)) | |
5783 | return PTR_ERR(trans); | |
39279cc3 | 5784 | |
581bb050 LZ |
5785 | err = btrfs_find_free_ino(root, &objectid); |
5786 | if (err) | |
5787 | goto out_fail; | |
5788 | ||
aec7477b | 5789 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5790 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5791 | S_IFDIR | mode, &index); |
39279cc3 CM |
5792 | if (IS_ERR(inode)) { |
5793 | err = PTR_ERR(inode); | |
5794 | goto out_fail; | |
5795 | } | |
5f39d397 | 5796 | |
39279cc3 | 5797 | drop_on_err = 1; |
33268eaf | 5798 | |
2a7dba39 | 5799 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5800 | if (err) |
5801 | goto out_fail; | |
5802 | ||
39279cc3 CM |
5803 | inode->i_op = &btrfs_dir_inode_operations; |
5804 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5805 | |
dbe674a9 | 5806 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5807 | err = btrfs_update_inode(trans, root, inode); |
5808 | if (err) | |
5809 | goto out_fail; | |
5f39d397 | 5810 | |
a1b075d2 JB |
5811 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5812 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5813 | if (err) |
5814 | goto out_fail; | |
5f39d397 | 5815 | |
39279cc3 CM |
5816 | d_instantiate(dentry, inode); |
5817 | drop_on_err = 0; | |
39279cc3 CM |
5818 | |
5819 | out_fail: | |
7ad85bb7 | 5820 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5821 | if (drop_on_err) |
5822 | iput(inode); | |
b53d3f5d | 5823 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5824 | return err; |
5825 | } | |
5826 | ||
d352ac68 CM |
5827 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5828 | * and an extent that you want to insert, deal with overlap and insert | |
5829 | * the new extent into the tree. | |
5830 | */ | |
3b951516 CM |
5831 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5832 | struct extent_map *existing, | |
e6dcd2dc CM |
5833 | struct extent_map *em, |
5834 | u64 map_start, u64 map_len) | |
3b951516 CM |
5835 | { |
5836 | u64 start_diff; | |
3b951516 | 5837 | |
e6dcd2dc CM |
5838 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5839 | start_diff = map_start - em->start; | |
5840 | em->start = map_start; | |
5841 | em->len = map_len; | |
c8b97818 CM |
5842 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5843 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5844 | em->block_start += start_diff; |
c8b97818 CM |
5845 | em->block_len -= start_diff; |
5846 | } | |
09a2a8f9 | 5847 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
5848 | } |
5849 | ||
c8b97818 CM |
5850 | static noinline int uncompress_inline(struct btrfs_path *path, |
5851 | struct inode *inode, struct page *page, | |
5852 | size_t pg_offset, u64 extent_offset, | |
5853 | struct btrfs_file_extent_item *item) | |
5854 | { | |
5855 | int ret; | |
5856 | struct extent_buffer *leaf = path->nodes[0]; | |
5857 | char *tmp; | |
5858 | size_t max_size; | |
5859 | unsigned long inline_size; | |
5860 | unsigned long ptr; | |
261507a0 | 5861 | int compress_type; |
c8b97818 CM |
5862 | |
5863 | WARN_ON(pg_offset != 0); | |
261507a0 | 5864 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5865 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5866 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5867 | btrfs_item_nr(leaf, path->slots[0])); | |
5868 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5869 | if (!tmp) |
5870 | return -ENOMEM; | |
c8b97818 CM |
5871 | ptr = btrfs_file_extent_inline_start(item); |
5872 | ||
5873 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5874 | ||
5b050f04 | 5875 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5876 | ret = btrfs_decompress(compress_type, tmp, page, |
5877 | extent_offset, inline_size, max_size); | |
c8b97818 | 5878 | if (ret) { |
7ac687d9 | 5879 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5880 | unsigned long copy_size = min_t(u64, |
5881 | PAGE_CACHE_SIZE - pg_offset, | |
5882 | max_size - extent_offset); | |
5883 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5884 | kunmap_atomic(kaddr); |
c8b97818 CM |
5885 | } |
5886 | kfree(tmp); | |
5887 | return 0; | |
5888 | } | |
5889 | ||
d352ac68 CM |
5890 | /* |
5891 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5892 | * the ugly parts come from merging extents from the disk with the in-ram |
5893 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5894 | * where the in-ram extents might be locked pending data=ordered completion. |
5895 | * | |
5896 | * This also copies inline extents directly into the page. | |
5897 | */ | |
d397712b | 5898 | |
a52d9a80 | 5899 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5900 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5901 | int create) |
5902 | { | |
5903 | int ret; | |
5904 | int err = 0; | |
db94535d | 5905 | u64 bytenr; |
a52d9a80 CM |
5906 | u64 extent_start = 0; |
5907 | u64 extent_end = 0; | |
33345d01 | 5908 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5909 | u32 found_type; |
f421950f | 5910 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5911 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5912 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5913 | struct extent_buffer *leaf; |
5914 | struct btrfs_key found_key; | |
a52d9a80 CM |
5915 | struct extent_map *em = NULL; |
5916 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5917 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5918 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5919 | int compress_type; |
a52d9a80 | 5920 | |
a52d9a80 | 5921 | again: |
890871be | 5922 | read_lock(&em_tree->lock); |
d1310b2e | 5923 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5924 | if (em) |
5925 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5926 | read_unlock(&em_tree->lock); |
d1310b2e | 5927 | |
a52d9a80 | 5928 | if (em) { |
e1c4b745 CM |
5929 | if (em->start > start || em->start + em->len <= start) |
5930 | free_extent_map(em); | |
5931 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5932 | free_extent_map(em); |
5933 | else | |
5934 | goto out; | |
a52d9a80 | 5935 | } |
172ddd60 | 5936 | em = alloc_extent_map(); |
a52d9a80 | 5937 | if (!em) { |
d1310b2e CM |
5938 | err = -ENOMEM; |
5939 | goto out; | |
a52d9a80 | 5940 | } |
e6dcd2dc | 5941 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5942 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5943 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5944 | em->len = (u64)-1; |
c8b97818 | 5945 | em->block_len = (u64)-1; |
f421950f CM |
5946 | |
5947 | if (!path) { | |
5948 | path = btrfs_alloc_path(); | |
026fd317 JB |
5949 | if (!path) { |
5950 | err = -ENOMEM; | |
5951 | goto out; | |
5952 | } | |
5953 | /* | |
5954 | * Chances are we'll be called again, so go ahead and do | |
5955 | * readahead | |
5956 | */ | |
5957 | path->reada = 1; | |
f421950f CM |
5958 | } |
5959 | ||
179e29e4 CM |
5960 | ret = btrfs_lookup_file_extent(trans, root, path, |
5961 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5962 | if (ret < 0) { |
5963 | err = ret; | |
5964 | goto out; | |
5965 | } | |
5966 | ||
5967 | if (ret != 0) { | |
5968 | if (path->slots[0] == 0) | |
5969 | goto not_found; | |
5970 | path->slots[0]--; | |
5971 | } | |
5972 | ||
5f39d397 CM |
5973 | leaf = path->nodes[0]; |
5974 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5975 | struct btrfs_file_extent_item); |
a52d9a80 | 5976 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5977 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5978 | found_type = btrfs_key_type(&found_key); | |
5979 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5980 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5981 | goto not_found; | |
5982 | } | |
5983 | ||
5f39d397 CM |
5984 | found_type = btrfs_file_extent_type(leaf, item); |
5985 | extent_start = found_key.offset; | |
261507a0 | 5986 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5987 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5988 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5989 | extent_end = extent_start + |
db94535d | 5990 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5991 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5992 | size_t size; | |
5993 | size = btrfs_file_extent_inline_len(leaf, item); | |
fda2832f | 5994 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 YZ |
5995 | } |
5996 | ||
5997 | if (start >= extent_end) { | |
5998 | path->slots[0]++; | |
5999 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6000 | ret = btrfs_next_leaf(root, path); | |
6001 | if (ret < 0) { | |
6002 | err = ret; | |
6003 | goto out; | |
a52d9a80 | 6004 | } |
9036c102 YZ |
6005 | if (ret > 0) |
6006 | goto not_found; | |
6007 | leaf = path->nodes[0]; | |
a52d9a80 | 6008 | } |
9036c102 YZ |
6009 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6010 | if (found_key.objectid != objectid || | |
6011 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6012 | goto not_found; | |
6013 | if (start + len <= found_key.offset) | |
6014 | goto not_found; | |
6015 | em->start = start; | |
70c8a91c | 6016 | em->orig_start = start; |
9036c102 YZ |
6017 | em->len = found_key.offset - start; |
6018 | goto not_found_em; | |
6019 | } | |
6020 | ||
cc95bef6 | 6021 | em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item); |
d899e052 YZ |
6022 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6023 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
6024 | em->start = extent_start; |
6025 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
6026 | em->orig_start = extent_start - |
6027 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
6028 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
6029 | item); | |
db94535d CM |
6030 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
6031 | if (bytenr == 0) { | |
5f39d397 | 6032 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
6033 | goto insert; |
6034 | } | |
261507a0 | 6035 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 6036 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 6037 | em->compress_type = compress_type; |
c8b97818 | 6038 | em->block_start = bytenr; |
b4939680 | 6039 | em->block_len = em->orig_block_len; |
c8b97818 CM |
6040 | } else { |
6041 | bytenr += btrfs_file_extent_offset(leaf, item); | |
6042 | em->block_start = bytenr; | |
6043 | em->block_len = em->len; | |
d899e052 YZ |
6044 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
6045 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 6046 | } |
a52d9a80 CM |
6047 | goto insert; |
6048 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6049 | unsigned long ptr; |
a52d9a80 | 6050 | char *map; |
3326d1b0 CM |
6051 | size_t size; |
6052 | size_t extent_offset; | |
6053 | size_t copy_size; | |
a52d9a80 | 6054 | |
689f9346 | 6055 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 6056 | if (!page || create) { |
689f9346 | 6057 | em->start = extent_start; |
9036c102 | 6058 | em->len = extent_end - extent_start; |
689f9346 Y |
6059 | goto out; |
6060 | } | |
5f39d397 | 6061 | |
9036c102 YZ |
6062 | size = btrfs_file_extent_inline_len(leaf, item); |
6063 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 6064 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6065 | size - extent_offset); |
3326d1b0 | 6066 | em->start = extent_start + extent_offset; |
fda2832f | 6067 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6068 | em->orig_block_len = em->len; |
70c8a91c | 6069 | em->orig_start = em->start; |
261507a0 | 6070 | if (compress_type) { |
c8b97818 | 6071 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
6072 | em->compress_type = compress_type; |
6073 | } | |
689f9346 | 6074 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6075 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6076 | if (btrfs_file_extent_compression(leaf, item) != |
6077 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6078 | ret = uncompress_inline(path, inode, page, |
6079 | pg_offset, | |
6080 | extent_offset, item); | |
79787eaa | 6081 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
6082 | } else { |
6083 | map = kmap(page); | |
6084 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6085 | copy_size); | |
93c82d57 CM |
6086 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6087 | memset(map + pg_offset + copy_size, 0, | |
6088 | PAGE_CACHE_SIZE - pg_offset - | |
6089 | copy_size); | |
6090 | } | |
c8b97818 CM |
6091 | kunmap(page); |
6092 | } | |
179e29e4 CM |
6093 | flush_dcache_page(page); |
6094 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6095 | BUG(); |
179e29e4 CM |
6096 | if (!trans) { |
6097 | kunmap(page); | |
6098 | free_extent_map(em); | |
6099 | em = NULL; | |
ff5714cc | 6100 | |
b3b4aa74 | 6101 | btrfs_release_path(path); |
7a7eaa40 | 6102 | trans = btrfs_join_transaction(root); |
ff5714cc | 6103 | |
3612b495 TI |
6104 | if (IS_ERR(trans)) |
6105 | return ERR_CAST(trans); | |
179e29e4 CM |
6106 | goto again; |
6107 | } | |
c8b97818 | 6108 | map = kmap(page); |
70dec807 | 6109 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6110 | copy_size); |
c8b97818 | 6111 | kunmap(page); |
179e29e4 | 6112 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6113 | } |
d1310b2e | 6114 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6115 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
6116 | goto insert; |
6117 | } else { | |
31b1a2bd | 6118 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
6119 | } |
6120 | not_found: | |
6121 | em->start = start; | |
70c8a91c | 6122 | em->orig_start = start; |
d1310b2e | 6123 | em->len = len; |
a52d9a80 | 6124 | not_found_em: |
5f39d397 | 6125 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6126 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6127 | insert: |
b3b4aa74 | 6128 | btrfs_release_path(path); |
d1310b2e | 6129 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb SK |
6130 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6131 | (unsigned long long)em->start, | |
6132 | (unsigned long long)em->len, | |
6133 | (unsigned long long)start, | |
6134 | (unsigned long long)len); | |
a52d9a80 CM |
6135 | err = -EIO; |
6136 | goto out; | |
6137 | } | |
d1310b2e CM |
6138 | |
6139 | err = 0; | |
890871be | 6140 | write_lock(&em_tree->lock); |
09a2a8f9 | 6141 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6142 | /* it is possible that someone inserted the extent into the tree |
6143 | * while we had the lock dropped. It is also possible that | |
6144 | * an overlapping map exists in the tree | |
6145 | */ | |
a52d9a80 | 6146 | if (ret == -EEXIST) { |
3b951516 | 6147 | struct extent_map *existing; |
e6dcd2dc CM |
6148 | |
6149 | ret = 0; | |
6150 | ||
3b951516 | 6151 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
6152 | if (existing && (existing->start > start || |
6153 | existing->start + existing->len <= start)) { | |
6154 | free_extent_map(existing); | |
6155 | existing = NULL; | |
6156 | } | |
3b951516 CM |
6157 | if (!existing) { |
6158 | existing = lookup_extent_mapping(em_tree, em->start, | |
6159 | em->len); | |
6160 | if (existing) { | |
6161 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
6162 | em, start, |
6163 | root->sectorsize); | |
3b951516 CM |
6164 | free_extent_map(existing); |
6165 | if (err) { | |
6166 | free_extent_map(em); | |
6167 | em = NULL; | |
6168 | } | |
6169 | } else { | |
6170 | err = -EIO; | |
3b951516 CM |
6171 | free_extent_map(em); |
6172 | em = NULL; | |
6173 | } | |
6174 | } else { | |
6175 | free_extent_map(em); | |
6176 | em = existing; | |
e6dcd2dc | 6177 | err = 0; |
a52d9a80 | 6178 | } |
a52d9a80 | 6179 | } |
890871be | 6180 | write_unlock(&em_tree->lock); |
a52d9a80 | 6181 | out: |
1abe9b8a | 6182 | |
f0bd95ea TI |
6183 | if (em) |
6184 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 6185 | |
f421950f CM |
6186 | if (path) |
6187 | btrfs_free_path(path); | |
a52d9a80 CM |
6188 | if (trans) { |
6189 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6190 | if (!err) |
a52d9a80 CM |
6191 | err = ret; |
6192 | } | |
a52d9a80 CM |
6193 | if (err) { |
6194 | free_extent_map(em); | |
a52d9a80 CM |
6195 | return ERR_PTR(err); |
6196 | } | |
79787eaa | 6197 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6198 | return em; |
6199 | } | |
6200 | ||
ec29ed5b CM |
6201 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6202 | size_t pg_offset, u64 start, u64 len, | |
6203 | int create) | |
6204 | { | |
6205 | struct extent_map *em; | |
6206 | struct extent_map *hole_em = NULL; | |
6207 | u64 range_start = start; | |
6208 | u64 end; | |
6209 | u64 found; | |
6210 | u64 found_end; | |
6211 | int err = 0; | |
6212 | ||
6213 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6214 | if (IS_ERR(em)) | |
6215 | return em; | |
6216 | if (em) { | |
6217 | /* | |
f9e4fb53 LB |
6218 | * if our em maps to |
6219 | * - a hole or | |
6220 | * - a pre-alloc extent, | |
6221 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 6222 | */ |
f9e4fb53 LB |
6223 | if (em->block_start != EXTENT_MAP_HOLE && |
6224 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
6225 | return em; |
6226 | else | |
6227 | hole_em = em; | |
6228 | } | |
6229 | ||
6230 | /* check to see if we've wrapped (len == -1 or similar) */ | |
6231 | end = start + len; | |
6232 | if (end < start) | |
6233 | end = (u64)-1; | |
6234 | else | |
6235 | end -= 1; | |
6236 | ||
6237 | em = NULL; | |
6238 | ||
6239 | /* ok, we didn't find anything, lets look for delalloc */ | |
6240 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
6241 | end, len, EXTENT_DELALLOC, 1); | |
6242 | found_end = range_start + found; | |
6243 | if (found_end < range_start) | |
6244 | found_end = (u64)-1; | |
6245 | ||
6246 | /* | |
6247 | * we didn't find anything useful, return | |
6248 | * the original results from get_extent() | |
6249 | */ | |
6250 | if (range_start > end || found_end <= start) { | |
6251 | em = hole_em; | |
6252 | hole_em = NULL; | |
6253 | goto out; | |
6254 | } | |
6255 | ||
6256 | /* adjust the range_start to make sure it doesn't | |
6257 | * go backwards from the start they passed in | |
6258 | */ | |
6259 | range_start = max(start,range_start); | |
6260 | found = found_end - range_start; | |
6261 | ||
6262 | if (found > 0) { | |
6263 | u64 hole_start = start; | |
6264 | u64 hole_len = len; | |
6265 | ||
172ddd60 | 6266 | em = alloc_extent_map(); |
ec29ed5b CM |
6267 | if (!em) { |
6268 | err = -ENOMEM; | |
6269 | goto out; | |
6270 | } | |
6271 | /* | |
6272 | * when btrfs_get_extent can't find anything it | |
6273 | * returns one huge hole | |
6274 | * | |
6275 | * make sure what it found really fits our range, and | |
6276 | * adjust to make sure it is based on the start from | |
6277 | * the caller | |
6278 | */ | |
6279 | if (hole_em) { | |
6280 | u64 calc_end = extent_map_end(hole_em); | |
6281 | ||
6282 | if (calc_end <= start || (hole_em->start > end)) { | |
6283 | free_extent_map(hole_em); | |
6284 | hole_em = NULL; | |
6285 | } else { | |
6286 | hole_start = max(hole_em->start, start); | |
6287 | hole_len = calc_end - hole_start; | |
6288 | } | |
6289 | } | |
6290 | em->bdev = NULL; | |
6291 | if (hole_em && range_start > hole_start) { | |
6292 | /* our hole starts before our delalloc, so we | |
6293 | * have to return just the parts of the hole | |
6294 | * that go until the delalloc starts | |
6295 | */ | |
6296 | em->len = min(hole_len, | |
6297 | range_start - hole_start); | |
6298 | em->start = hole_start; | |
6299 | em->orig_start = hole_start; | |
6300 | /* | |
6301 | * don't adjust block start at all, | |
6302 | * it is fixed at EXTENT_MAP_HOLE | |
6303 | */ | |
6304 | em->block_start = hole_em->block_start; | |
6305 | em->block_len = hole_len; | |
f9e4fb53 LB |
6306 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
6307 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
6308 | } else { |
6309 | em->start = range_start; | |
6310 | em->len = found; | |
6311 | em->orig_start = range_start; | |
6312 | em->block_start = EXTENT_MAP_DELALLOC; | |
6313 | em->block_len = found; | |
6314 | } | |
6315 | } else if (hole_em) { | |
6316 | return hole_em; | |
6317 | } | |
6318 | out: | |
6319 | ||
6320 | free_extent_map(hole_em); | |
6321 | if (err) { | |
6322 | free_extent_map(em); | |
6323 | return ERR_PTR(err); | |
6324 | } | |
6325 | return em; | |
6326 | } | |
6327 | ||
4b46fce2 JB |
6328 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
6329 | u64 start, u64 len) | |
6330 | { | |
6331 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6332 | struct btrfs_trans_handle *trans; | |
70c8a91c | 6333 | struct extent_map *em; |
4b46fce2 JB |
6334 | struct btrfs_key ins; |
6335 | u64 alloc_hint; | |
6336 | int ret; | |
4b46fce2 | 6337 | |
7a7eaa40 | 6338 | trans = btrfs_join_transaction(root); |
3612b495 TI |
6339 | if (IS_ERR(trans)) |
6340 | return ERR_CAST(trans); | |
4b46fce2 JB |
6341 | |
6342 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
6343 | ||
6344 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
6345 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 6346 | alloc_hint, &ins, 1); |
4b46fce2 JB |
6347 | if (ret) { |
6348 | em = ERR_PTR(ret); | |
6349 | goto out; | |
6350 | } | |
6351 | ||
70c8a91c | 6352 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 6353 | ins.offset, ins.offset, ins.offset, 0); |
70c8a91c JB |
6354 | if (IS_ERR(em)) |
6355 | goto out; | |
4b46fce2 JB |
6356 | |
6357 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
6358 | ins.offset, ins.offset, 0); | |
6359 | if (ret) { | |
6360 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
6361 | em = ERR_PTR(ret); | |
6362 | } | |
6363 | out: | |
6364 | btrfs_end_transaction(trans, root); | |
6365 | return em; | |
6366 | } | |
6367 | ||
46bfbb5c CM |
6368 | /* |
6369 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
6370 | * block must be cow'd | |
6371 | */ | |
6372 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
eb384b55 JB |
6373 | struct inode *inode, u64 offset, u64 *len, |
6374 | u64 *orig_start, u64 *orig_block_len, | |
6375 | u64 *ram_bytes) | |
46bfbb5c CM |
6376 | { |
6377 | struct btrfs_path *path; | |
6378 | int ret; | |
6379 | struct extent_buffer *leaf; | |
6380 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6381 | struct btrfs_file_extent_item *fi; | |
6382 | struct btrfs_key key; | |
6383 | u64 disk_bytenr; | |
6384 | u64 backref_offset; | |
6385 | u64 extent_end; | |
6386 | u64 num_bytes; | |
6387 | int slot; | |
6388 | int found_type; | |
6389 | ||
6390 | path = btrfs_alloc_path(); | |
6391 | if (!path) | |
6392 | return -ENOMEM; | |
6393 | ||
33345d01 | 6394 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
6395 | offset, 0); |
6396 | if (ret < 0) | |
6397 | goto out; | |
6398 | ||
6399 | slot = path->slots[0]; | |
6400 | if (ret == 1) { | |
6401 | if (slot == 0) { | |
6402 | /* can't find the item, must cow */ | |
6403 | ret = 0; | |
6404 | goto out; | |
6405 | } | |
6406 | slot--; | |
6407 | } | |
6408 | ret = 0; | |
6409 | leaf = path->nodes[0]; | |
6410 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 6411 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
6412 | key.type != BTRFS_EXTENT_DATA_KEY) { |
6413 | /* not our file or wrong item type, must cow */ | |
6414 | goto out; | |
6415 | } | |
6416 | ||
6417 | if (key.offset > offset) { | |
6418 | /* Wrong offset, must cow */ | |
6419 | goto out; | |
6420 | } | |
6421 | ||
6422 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
6423 | found_type = btrfs_file_extent_type(leaf, fi); | |
6424 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
6425 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
6426 | /* not a regular extent, must cow */ | |
6427 | goto out; | |
6428 | } | |
6429 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
6430 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
6431 | ||
eb384b55 JB |
6432 | *orig_start = key.offset - backref_offset; |
6433 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6434 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
6435 | ||
46bfbb5c | 6436 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
eb384b55 | 6437 | if (extent_end < offset + *len) { |
46bfbb5c CM |
6438 | /* extent doesn't include our full range, must cow */ |
6439 | goto out; | |
6440 | } | |
6441 | ||
6442 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
6443 | goto out; | |
6444 | ||
6445 | /* | |
6446 | * look for other files referencing this extent, if we | |
6447 | * find any we must cow | |
6448 | */ | |
33345d01 | 6449 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
6450 | key.offset - backref_offset, disk_bytenr)) |
6451 | goto out; | |
6452 | ||
6453 | /* | |
6454 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
6455 | * in this extent we are about to write. If there | |
6456 | * are any csums in that range we have to cow in order | |
6457 | * to keep the csums correct | |
6458 | */ | |
6459 | disk_bytenr += backref_offset; | |
6460 | disk_bytenr += offset - key.offset; | |
eb384b55 | 6461 | num_bytes = min(offset + *len, extent_end) - offset; |
46bfbb5c CM |
6462 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
6463 | goto out; | |
6464 | /* | |
6465 | * all of the above have passed, it is safe to overwrite this extent | |
6466 | * without cow | |
6467 | */ | |
eb384b55 | 6468 | *len = num_bytes; |
46bfbb5c CM |
6469 | ret = 1; |
6470 | out: | |
6471 | btrfs_free_path(path); | |
6472 | return ret; | |
6473 | } | |
6474 | ||
eb838e73 JB |
6475 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
6476 | struct extent_state **cached_state, int writing) | |
6477 | { | |
6478 | struct btrfs_ordered_extent *ordered; | |
6479 | int ret = 0; | |
6480 | ||
6481 | while (1) { | |
6482 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6483 | 0, cached_state); | |
6484 | /* | |
6485 | * We're concerned with the entire range that we're going to be | |
6486 | * doing DIO to, so we need to make sure theres no ordered | |
6487 | * extents in this range. | |
6488 | */ | |
6489 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6490 | lockend - lockstart + 1); | |
6491 | ||
6492 | /* | |
6493 | * We need to make sure there are no buffered pages in this | |
6494 | * range either, we could have raced between the invalidate in | |
6495 | * generic_file_direct_write and locking the extent. The | |
6496 | * invalidate needs to happen so that reads after a write do not | |
6497 | * get stale data. | |
6498 | */ | |
6499 | if (!ordered && (!writing || | |
6500 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
6501 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
6502 | *cached_state))) | |
6503 | break; | |
6504 | ||
6505 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6506 | cached_state, GFP_NOFS); | |
6507 | ||
6508 | if (ordered) { | |
6509 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6510 | btrfs_put_ordered_extent(ordered); | |
6511 | } else { | |
6512 | /* Screw you mmap */ | |
6513 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
6514 | lockstart, | |
6515 | lockend); | |
6516 | if (ret) | |
6517 | break; | |
6518 | ||
6519 | /* | |
6520 | * If we found a page that couldn't be invalidated just | |
6521 | * fall back to buffered. | |
6522 | */ | |
6523 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6524 | lockstart >> PAGE_CACHE_SHIFT, | |
6525 | lockend >> PAGE_CACHE_SHIFT); | |
6526 | if (ret) | |
6527 | break; | |
6528 | } | |
6529 | ||
6530 | cond_resched(); | |
6531 | } | |
6532 | ||
6533 | return ret; | |
6534 | } | |
6535 | ||
69ffb543 JB |
6536 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
6537 | u64 len, u64 orig_start, | |
6538 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
6539 | u64 orig_block_len, u64 ram_bytes, |
6540 | int type) | |
69ffb543 JB |
6541 | { |
6542 | struct extent_map_tree *em_tree; | |
6543 | struct extent_map *em; | |
6544 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6545 | int ret; | |
6546 | ||
6547 | em_tree = &BTRFS_I(inode)->extent_tree; | |
6548 | em = alloc_extent_map(); | |
6549 | if (!em) | |
6550 | return ERR_PTR(-ENOMEM); | |
6551 | ||
6552 | em->start = start; | |
6553 | em->orig_start = orig_start; | |
2ab28f32 JB |
6554 | em->mod_start = start; |
6555 | em->mod_len = len; | |
69ffb543 JB |
6556 | em->len = len; |
6557 | em->block_len = block_len; | |
6558 | em->block_start = block_start; | |
6559 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6560 | em->orig_block_len = orig_block_len; |
cc95bef6 | 6561 | em->ram_bytes = ram_bytes; |
70c8a91c | 6562 | em->generation = -1; |
69ffb543 JB |
6563 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6564 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6565 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6566 | |
6567 | do { | |
6568 | btrfs_drop_extent_cache(inode, em->start, | |
6569 | em->start + em->len - 1, 0); | |
6570 | write_lock(&em_tree->lock); | |
09a2a8f9 | 6571 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
6572 | write_unlock(&em_tree->lock); |
6573 | } while (ret == -EEXIST); | |
6574 | ||
6575 | if (ret) { | |
6576 | free_extent_map(em); | |
6577 | return ERR_PTR(ret); | |
6578 | } | |
6579 | ||
6580 | return em; | |
6581 | } | |
6582 | ||
6583 | ||
4b46fce2 JB |
6584 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6585 | struct buffer_head *bh_result, int create) | |
6586 | { | |
6587 | struct extent_map *em; | |
6588 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6589 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6590 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6591 | u64 lockstart, lockend; |
4b46fce2 | 6592 | u64 len = bh_result->b_size; |
46bfbb5c | 6593 | struct btrfs_trans_handle *trans; |
eb838e73 | 6594 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 6595 | int ret = 0; |
eb838e73 | 6596 | |
172a5049 | 6597 | if (create) |
eb838e73 | 6598 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; |
172a5049 | 6599 | else |
c329861d | 6600 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 6601 | |
c329861d JB |
6602 | lockstart = start; |
6603 | lockend = start + len - 1; | |
6604 | ||
eb838e73 JB |
6605 | /* |
6606 | * If this errors out it's because we couldn't invalidate pagecache for | |
6607 | * this range and we need to fallback to buffered. | |
6608 | */ | |
6609 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6610 | return -ENOTBLK; | |
6611 | ||
4b46fce2 | 6612 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
6613 | if (IS_ERR(em)) { |
6614 | ret = PTR_ERR(em); | |
6615 | goto unlock_err; | |
6616 | } | |
4b46fce2 JB |
6617 | |
6618 | /* | |
6619 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6620 | * io. INLINE is special, and we could probably kludge it in here, but | |
6621 | * it's still buffered so for safety lets just fall back to the generic | |
6622 | * buffered path. | |
6623 | * | |
6624 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6625 | * decompress it, so there will be buffering required no matter what we | |
6626 | * do, so go ahead and fallback to buffered. | |
6627 | * | |
6628 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6629 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6630 | * the generic code. | |
6631 | */ | |
6632 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6633 | em->block_start == EXTENT_MAP_INLINE) { | |
6634 | free_extent_map(em); | |
eb838e73 JB |
6635 | ret = -ENOTBLK; |
6636 | goto unlock_err; | |
4b46fce2 JB |
6637 | } |
6638 | ||
6639 | /* Just a good old fashioned hole, return */ | |
6640 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6641 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6642 | free_extent_map(em); | |
eb838e73 | 6643 | goto unlock_err; |
4b46fce2 JB |
6644 | } |
6645 | ||
6646 | /* | |
6647 | * We don't allocate a new extent in the following cases | |
6648 | * | |
6649 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6650 | * existing extent. | |
6651 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6652 | * just use the extent. | |
6653 | * | |
6654 | */ | |
46bfbb5c | 6655 | if (!create) { |
eb838e73 JB |
6656 | len = min(len, em->len - (start - em->start)); |
6657 | lockstart = start + len; | |
6658 | goto unlock; | |
46bfbb5c | 6659 | } |
4b46fce2 JB |
6660 | |
6661 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6662 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6663 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6664 | int type; |
6665 | int ret; | |
eb384b55 | 6666 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
6667 | |
6668 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6669 | type = BTRFS_ORDERED_PREALLOC; | |
6670 | else | |
6671 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6672 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6673 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6674 | |
6675 | /* | |
6676 | * we're not going to log anything, but we do need | |
6677 | * to make sure the current transaction stays open | |
6678 | * while we look for nocow cross refs | |
6679 | */ | |
7a7eaa40 | 6680 | trans = btrfs_join_transaction(root); |
3612b495 | 6681 | if (IS_ERR(trans)) |
46bfbb5c CM |
6682 | goto must_cow; |
6683 | ||
eb384b55 JB |
6684 | if (can_nocow_odirect(trans, inode, start, &len, &orig_start, |
6685 | &orig_block_len, &ram_bytes) == 1) { | |
69ffb543 JB |
6686 | if (type == BTRFS_ORDERED_PREALLOC) { |
6687 | free_extent_map(em); | |
6688 | em = create_pinned_em(inode, start, len, | |
6689 | orig_start, | |
b4939680 | 6690 | block_start, len, |
cc95bef6 JB |
6691 | orig_block_len, |
6692 | ram_bytes, type); | |
69ffb543 JB |
6693 | if (IS_ERR(em)) { |
6694 | btrfs_end_transaction(trans, root); | |
6695 | goto unlock_err; | |
6696 | } | |
6697 | } | |
6698 | ||
46bfbb5c CM |
6699 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6700 | block_start, len, len, type); | |
6701 | btrfs_end_transaction(trans, root); | |
6702 | if (ret) { | |
6703 | free_extent_map(em); | |
eb838e73 | 6704 | goto unlock_err; |
46bfbb5c CM |
6705 | } |
6706 | goto unlock; | |
4b46fce2 | 6707 | } |
46bfbb5c | 6708 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6709 | } |
46bfbb5c CM |
6710 | must_cow: |
6711 | /* | |
6712 | * this will cow the extent, reset the len in case we changed | |
6713 | * it above | |
6714 | */ | |
6715 | len = bh_result->b_size; | |
70c8a91c JB |
6716 | free_extent_map(em); |
6717 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6718 | if (IS_ERR(em)) { |
6719 | ret = PTR_ERR(em); | |
6720 | goto unlock_err; | |
6721 | } | |
46bfbb5c CM |
6722 | len = min(len, em->len - (start - em->start)); |
6723 | unlock: | |
4b46fce2 JB |
6724 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6725 | inode->i_blkbits; | |
46bfbb5c | 6726 | bh_result->b_size = len; |
4b46fce2 JB |
6727 | bh_result->b_bdev = em->bdev; |
6728 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6729 | if (create) { |
6730 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6731 | set_buffer_new(bh_result); | |
6732 | ||
6733 | /* | |
6734 | * Need to update the i_size under the extent lock so buffered | |
6735 | * readers will get the updated i_size when we unlock. | |
6736 | */ | |
6737 | if (start + len > i_size_read(inode)) | |
6738 | i_size_write(inode, start + len); | |
0934856d | 6739 | |
172a5049 MX |
6740 | spin_lock(&BTRFS_I(inode)->lock); |
6741 | BTRFS_I(inode)->outstanding_extents++; | |
6742 | spin_unlock(&BTRFS_I(inode)->lock); | |
6743 | ||
0934856d MX |
6744 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6745 | lockstart + len - 1, EXTENT_DELALLOC, NULL, | |
6746 | &cached_state, GFP_NOFS); | |
6747 | BUG_ON(ret); | |
c3473e83 | 6748 | } |
4b46fce2 | 6749 | |
eb838e73 JB |
6750 | /* |
6751 | * In the case of write we need to clear and unlock the entire range, | |
6752 | * in the case of read we need to unlock only the end area that we | |
6753 | * aren't using if there is any left over space. | |
6754 | */ | |
24c03fa5 | 6755 | if (lockstart < lockend) { |
0934856d MX |
6756 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
6757 | lockend, unlock_bits, 1, 0, | |
6758 | &cached_state, GFP_NOFS); | |
24c03fa5 | 6759 | } else { |
eb838e73 | 6760 | free_extent_state(cached_state); |
24c03fa5 | 6761 | } |
eb838e73 | 6762 | |
4b46fce2 JB |
6763 | free_extent_map(em); |
6764 | ||
6765 | return 0; | |
eb838e73 JB |
6766 | |
6767 | unlock_err: | |
eb838e73 JB |
6768 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6769 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6770 | return ret; | |
4b46fce2 JB |
6771 | } |
6772 | ||
6773 | struct btrfs_dio_private { | |
6774 | struct inode *inode; | |
6775 | u64 logical_offset; | |
6776 | u64 disk_bytenr; | |
6777 | u64 bytes; | |
4b46fce2 | 6778 | void *private; |
e65e1535 MX |
6779 | |
6780 | /* number of bios pending for this dio */ | |
6781 | atomic_t pending_bios; | |
6782 | ||
6783 | /* IO errors */ | |
6784 | int errors; | |
6785 | ||
9be3395b | 6786 | /* orig_bio is our btrfs_io_bio */ |
e65e1535 | 6787 | struct bio *orig_bio; |
9be3395b CM |
6788 | |
6789 | /* dio_bio came from fs/direct-io.c */ | |
6790 | struct bio *dio_bio; | |
4b46fce2 JB |
6791 | }; |
6792 | ||
6793 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6794 | { | |
e65e1535 | 6795 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6796 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6797 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6798 | struct inode *inode = dip->inode; |
c2cf52eb | 6799 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9be3395b | 6800 | struct bio *dio_bio; |
4b46fce2 | 6801 | u64 start; |
4b46fce2 JB |
6802 | |
6803 | start = dip->logical_offset; | |
6804 | do { | |
6805 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6806 | struct page *page = bvec->bv_page; | |
6807 | char *kaddr; | |
6808 | u32 csum = ~(u32)0; | |
c329861d | 6809 | u64 private = ~(u32)0; |
4b46fce2 JB |
6810 | unsigned long flags; |
6811 | ||
c329861d JB |
6812 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6813 | start, &private)) | |
6814 | goto failed; | |
4b46fce2 | 6815 | local_irq_save(flags); |
7ac687d9 | 6816 | kaddr = kmap_atomic(page); |
b0496686 | 6817 | csum = btrfs_csum_data(kaddr + bvec->bv_offset, |
4b46fce2 JB |
6818 | csum, bvec->bv_len); |
6819 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6820 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6821 | local_irq_restore(flags); |
6822 | ||
6823 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6824 | if (csum != private) { |
6825 | failed: | |
c2cf52eb SK |
6826 | btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u", |
6827 | (unsigned long long)btrfs_ino(inode), | |
6828 | (unsigned long long)start, | |
6829 | csum, (unsigned)private); | |
4b46fce2 JB |
6830 | err = -EIO; |
6831 | } | |
6832 | } | |
6833 | ||
6834 | start += bvec->bv_len; | |
4b46fce2 JB |
6835 | bvec++; |
6836 | } while (bvec <= bvec_end); | |
6837 | ||
6838 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6839 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 6840 | dio_bio = dip->dio_bio; |
4b46fce2 | 6841 | |
4b46fce2 | 6842 | kfree(dip); |
c0da7aa1 JB |
6843 | |
6844 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6845 | if (err) | |
9be3395b CM |
6846 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6847 | dio_end_io(dio_bio, err); | |
6848 | bio_put(bio); | |
4b46fce2 JB |
6849 | } |
6850 | ||
6851 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6852 | { | |
6853 | struct btrfs_dio_private *dip = bio->bi_private; | |
6854 | struct inode *inode = dip->inode; | |
6855 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6856 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6857 | u64 ordered_offset = dip->logical_offset; |
6858 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 6859 | struct bio *dio_bio; |
4b46fce2 JB |
6860 | int ret; |
6861 | ||
6862 | if (err) | |
6863 | goto out_done; | |
163cf09c CM |
6864 | again: |
6865 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6866 | &ordered_offset, | |
5fd02043 | 6867 | ordered_bytes, !err); |
4b46fce2 | 6868 | if (!ret) |
163cf09c | 6869 | goto out_test; |
4b46fce2 | 6870 | |
5fd02043 JB |
6871 | ordered->work.func = finish_ordered_fn; |
6872 | ordered->work.flags = 0; | |
6873 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6874 | &ordered->work); | |
163cf09c CM |
6875 | out_test: |
6876 | /* | |
6877 | * our bio might span multiple ordered extents. If we haven't | |
6878 | * completed the accounting for the whole dio, go back and try again | |
6879 | */ | |
6880 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6881 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6882 | ordered_offset; | |
5fd02043 | 6883 | ordered = NULL; |
163cf09c CM |
6884 | goto again; |
6885 | } | |
4b46fce2 | 6886 | out_done: |
9be3395b | 6887 | dio_bio = dip->dio_bio; |
4b46fce2 | 6888 | |
4b46fce2 | 6889 | kfree(dip); |
c0da7aa1 JB |
6890 | |
6891 | /* If we had an error make sure to clear the uptodate flag */ | |
6892 | if (err) | |
9be3395b CM |
6893 | clear_bit(BIO_UPTODATE, &dio_bio->bi_flags); |
6894 | dio_end_io(dio_bio, err); | |
6895 | bio_put(bio); | |
4b46fce2 JB |
6896 | } |
6897 | ||
eaf25d93 CM |
6898 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6899 | struct bio *bio, int mirror_num, | |
6900 | unsigned long bio_flags, u64 offset) | |
6901 | { | |
6902 | int ret; | |
6903 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6904 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6905 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6906 | return 0; |
6907 | } | |
6908 | ||
e65e1535 MX |
6909 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6910 | { | |
6911 | struct btrfs_dio_private *dip = bio->bi_private; | |
6912 | ||
6913 | if (err) { | |
33345d01 | 6914 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6915 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6916 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6917 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6918 | dip->errors = 1; |
6919 | ||
6920 | /* | |
6921 | * before atomic variable goto zero, we must make sure | |
6922 | * dip->errors is perceived to be set. | |
6923 | */ | |
6924 | smp_mb__before_atomic_dec(); | |
6925 | } | |
6926 | ||
6927 | /* if there are more bios still pending for this dio, just exit */ | |
6928 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6929 | goto out; | |
6930 | ||
9be3395b | 6931 | if (dip->errors) { |
e65e1535 | 6932 | bio_io_error(dip->orig_bio); |
9be3395b CM |
6933 | } else { |
6934 | set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags); | |
e65e1535 MX |
6935 | bio_endio(dip->orig_bio, 0); |
6936 | } | |
6937 | out: | |
6938 | bio_put(bio); | |
6939 | } | |
6940 | ||
6941 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6942 | u64 first_sector, gfp_t gfp_flags) | |
6943 | { | |
6944 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6945 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6946 | } | |
6947 | ||
6948 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6949 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6950 | int async_submit) |
e65e1535 MX |
6951 | { |
6952 | int write = rw & REQ_WRITE; | |
6953 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6954 | int ret; | |
6955 | ||
b812ce28 JB |
6956 | if (async_submit) |
6957 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
6958 | ||
e65e1535 | 6959 | bio_get(bio); |
5fd02043 JB |
6960 | |
6961 | if (!write) { | |
6962 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6963 | if (ret) | |
6964 | goto err; | |
6965 | } | |
e65e1535 | 6966 | |
1ae39938 JB |
6967 | if (skip_sum) |
6968 | goto map; | |
6969 | ||
6970 | if (write && async_submit) { | |
e65e1535 MX |
6971 | ret = btrfs_wq_submit_bio(root->fs_info, |
6972 | inode, rw, bio, 0, 0, | |
6973 | file_offset, | |
6974 | __btrfs_submit_bio_start_direct_io, | |
6975 | __btrfs_submit_bio_done); | |
6976 | goto err; | |
1ae39938 JB |
6977 | } else if (write) { |
6978 | /* | |
6979 | * If we aren't doing async submit, calculate the csum of the | |
6980 | * bio now. | |
6981 | */ | |
6982 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6983 | if (ret) | |
6984 | goto err; | |
c2db1073 | 6985 | } else if (!skip_sum) { |
c329861d | 6986 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6987 | if (ret) |
6988 | goto err; | |
6989 | } | |
e65e1535 | 6990 | |
1ae39938 JB |
6991 | map: |
6992 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6993 | err: |
6994 | bio_put(bio); | |
6995 | return ret; | |
6996 | } | |
6997 | ||
6998 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6999 | int skip_sum) | |
7000 | { | |
7001 | struct inode *inode = dip->inode; | |
7002 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
7003 | struct bio *bio; |
7004 | struct bio *orig_bio = dip->orig_bio; | |
7005 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
7006 | u64 start_sector = orig_bio->bi_sector; | |
7007 | u64 file_offset = dip->logical_offset; | |
7008 | u64 submit_len = 0; | |
7009 | u64 map_length; | |
7010 | int nr_pages = 0; | |
e65e1535 | 7011 | int ret = 0; |
1ae39938 | 7012 | int async_submit = 0; |
e65e1535 | 7013 | |
e65e1535 | 7014 | map_length = orig_bio->bi_size; |
53b381b3 | 7015 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 MX |
7016 | &map_length, NULL, 0); |
7017 | if (ret) { | |
64728bbb | 7018 | bio_put(orig_bio); |
e65e1535 MX |
7019 | return -EIO; |
7020 | } | |
02f57c7a JB |
7021 | if (map_length >= orig_bio->bi_size) { |
7022 | bio = orig_bio; | |
7023 | goto submit; | |
7024 | } | |
7025 | ||
53b381b3 DW |
7026 | /* async crcs make it difficult to collect full stripe writes. */ |
7027 | if (btrfs_get_alloc_profile(root, 1) & | |
7028 | (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) | |
7029 | async_submit = 0; | |
7030 | else | |
7031 | async_submit = 1; | |
7032 | ||
02f57c7a JB |
7033 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
7034 | if (!bio) | |
7035 | return -ENOMEM; | |
7036 | bio->bi_private = dip; | |
7037 | bio->bi_end_io = btrfs_end_dio_bio; | |
7038 | atomic_inc(&dip->pending_bios); | |
7039 | ||
e65e1535 MX |
7040 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
7041 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
7042 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
7043 | bvec->bv_offset) < bvec->bv_len)) { | |
7044 | /* | |
7045 | * inc the count before we submit the bio so | |
7046 | * we know the end IO handler won't happen before | |
7047 | * we inc the count. Otherwise, the dip might get freed | |
7048 | * before we're done setting it up | |
7049 | */ | |
7050 | atomic_inc(&dip->pending_bios); | |
7051 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
7052 | file_offset, skip_sum, | |
c329861d | 7053 | async_submit); |
e65e1535 MX |
7054 | if (ret) { |
7055 | bio_put(bio); | |
7056 | atomic_dec(&dip->pending_bios); | |
7057 | goto out_err; | |
7058 | } | |
7059 | ||
e65e1535 MX |
7060 | start_sector += submit_len >> 9; |
7061 | file_offset += submit_len; | |
7062 | ||
7063 | submit_len = 0; | |
7064 | nr_pages = 0; | |
7065 | ||
7066 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
7067 | start_sector, GFP_NOFS); | |
7068 | if (!bio) | |
7069 | goto out_err; | |
7070 | bio->bi_private = dip; | |
7071 | bio->bi_end_io = btrfs_end_dio_bio; | |
7072 | ||
7073 | map_length = orig_bio->bi_size; | |
53b381b3 | 7074 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 7075 | start_sector << 9, |
e65e1535 MX |
7076 | &map_length, NULL, 0); |
7077 | if (ret) { | |
7078 | bio_put(bio); | |
7079 | goto out_err; | |
7080 | } | |
7081 | } else { | |
7082 | submit_len += bvec->bv_len; | |
7083 | nr_pages ++; | |
7084 | bvec++; | |
7085 | } | |
7086 | } | |
7087 | ||
02f57c7a | 7088 | submit: |
e65e1535 | 7089 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 7090 | async_submit); |
e65e1535 MX |
7091 | if (!ret) |
7092 | return 0; | |
7093 | ||
7094 | bio_put(bio); | |
7095 | out_err: | |
7096 | dip->errors = 1; | |
7097 | /* | |
7098 | * before atomic variable goto zero, we must | |
7099 | * make sure dip->errors is perceived to be set. | |
7100 | */ | |
7101 | smp_mb__before_atomic_dec(); | |
7102 | if (atomic_dec_and_test(&dip->pending_bios)) | |
7103 | bio_io_error(dip->orig_bio); | |
7104 | ||
7105 | /* bio_end_io() will handle error, so we needn't return it */ | |
7106 | return 0; | |
7107 | } | |
7108 | ||
9be3395b CM |
7109 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
7110 | struct inode *inode, loff_t file_offset) | |
4b46fce2 JB |
7111 | { |
7112 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7113 | struct btrfs_dio_private *dip; | |
9be3395b CM |
7114 | struct bio_vec *bvec = dio_bio->bi_io_vec; |
7115 | struct bio *io_bio; | |
4b46fce2 | 7116 | int skip_sum; |
7b6d91da | 7117 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
7118 | int ret = 0; |
7119 | ||
7120 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7121 | ||
9be3395b CM |
7122 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
7123 | ||
7124 | if (!io_bio) { | |
7125 | ret = -ENOMEM; | |
7126 | goto free_ordered; | |
7127 | } | |
7128 | ||
4b46fce2 JB |
7129 | dip = kmalloc(sizeof(*dip), GFP_NOFS); |
7130 | if (!dip) { | |
7131 | ret = -ENOMEM; | |
9be3395b | 7132 | goto free_io_bio; |
4b46fce2 | 7133 | } |
4b46fce2 | 7134 | |
9be3395b CM |
7135 | dip->private = dio_bio->bi_private; |
7136 | io_bio->bi_private = dio_bio->bi_private; | |
4b46fce2 JB |
7137 | dip->inode = inode; |
7138 | dip->logical_offset = file_offset; | |
7139 | ||
4b46fce2 JB |
7140 | dip->bytes = 0; |
7141 | do { | |
7142 | dip->bytes += bvec->bv_len; | |
7143 | bvec++; | |
9be3395b | 7144 | } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1)); |
4b46fce2 | 7145 | |
9be3395b CM |
7146 | dip->disk_bytenr = (u64)dio_bio->bi_sector << 9; |
7147 | io_bio->bi_private = dip; | |
e65e1535 | 7148 | dip->errors = 0; |
9be3395b CM |
7149 | dip->orig_bio = io_bio; |
7150 | dip->dio_bio = dio_bio; | |
e65e1535 | 7151 | atomic_set(&dip->pending_bios, 0); |
4b46fce2 JB |
7152 | |
7153 | if (write) | |
9be3395b | 7154 | io_bio->bi_end_io = btrfs_endio_direct_write; |
4b46fce2 | 7155 | else |
9be3395b | 7156 | io_bio->bi_end_io = btrfs_endio_direct_read; |
4b46fce2 | 7157 | |
e65e1535 MX |
7158 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
7159 | if (!ret) | |
eaf25d93 | 7160 | return; |
9be3395b CM |
7161 | |
7162 | free_io_bio: | |
7163 | bio_put(io_bio); | |
7164 | ||
4b46fce2 JB |
7165 | free_ordered: |
7166 | /* | |
7167 | * If this is a write, we need to clean up the reserved space and kill | |
7168 | * the ordered extent. | |
7169 | */ | |
7170 | if (write) { | |
7171 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 7172 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
7173 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
7174 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
7175 | btrfs_free_reserved_extent(root, ordered->start, | |
7176 | ordered->disk_len); | |
7177 | btrfs_put_ordered_extent(ordered); | |
7178 | btrfs_put_ordered_extent(ordered); | |
7179 | } | |
9be3395b | 7180 | bio_endio(dio_bio, ret); |
4b46fce2 JB |
7181 | } |
7182 | ||
5a5f79b5 CM |
7183 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
7184 | const struct iovec *iov, loff_t offset, | |
7185 | unsigned long nr_segs) | |
7186 | { | |
7187 | int seg; | |
a1b75f7d | 7188 | int i; |
5a5f79b5 CM |
7189 | size_t size; |
7190 | unsigned long addr; | |
7191 | unsigned blocksize_mask = root->sectorsize - 1; | |
7192 | ssize_t retval = -EINVAL; | |
7193 | loff_t end = offset; | |
7194 | ||
7195 | if (offset & blocksize_mask) | |
7196 | goto out; | |
7197 | ||
7198 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
7199 | for (seg = 0; seg < nr_segs; seg++) { | |
7200 | addr = (unsigned long)iov[seg].iov_base; | |
7201 | size = iov[seg].iov_len; | |
7202 | end += size; | |
a1b75f7d | 7203 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 7204 | goto out; |
a1b75f7d JB |
7205 | |
7206 | /* If this is a write we don't need to check anymore */ | |
7207 | if (rw & WRITE) | |
7208 | continue; | |
7209 | ||
7210 | /* | |
7211 | * Check to make sure we don't have duplicate iov_base's in this | |
7212 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
7213 | * when reading back. | |
7214 | */ | |
7215 | for (i = seg + 1; i < nr_segs; i++) { | |
7216 | if (iov[seg].iov_base == iov[i].iov_base) | |
7217 | goto out; | |
7218 | } | |
5a5f79b5 CM |
7219 | } |
7220 | retval = 0; | |
7221 | out: | |
7222 | return retval; | |
7223 | } | |
eb838e73 | 7224 | |
16432985 CM |
7225 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
7226 | const struct iovec *iov, loff_t offset, | |
7227 | unsigned long nr_segs) | |
7228 | { | |
4b46fce2 JB |
7229 | struct file *file = iocb->ki_filp; |
7230 | struct inode *inode = file->f_mapping->host; | |
0934856d | 7231 | size_t count = 0; |
2e60a51e | 7232 | int flags = 0; |
38851cc1 MX |
7233 | bool wakeup = true; |
7234 | bool relock = false; | |
0934856d | 7235 | ssize_t ret; |
4b46fce2 | 7236 | |
5a5f79b5 | 7237 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 7238 | offset, nr_segs)) |
5a5f79b5 | 7239 | return 0; |
3f7c579c | 7240 | |
38851cc1 MX |
7241 | atomic_inc(&inode->i_dio_count); |
7242 | smp_mb__after_atomic_inc(); | |
7243 | ||
0934856d MX |
7244 | if (rw & WRITE) { |
7245 | count = iov_length(iov, nr_segs); | |
38851cc1 MX |
7246 | /* |
7247 | * If the write DIO is beyond the EOF, we need update | |
7248 | * the isize, but it is protected by i_mutex. So we can | |
7249 | * not unlock the i_mutex at this case. | |
7250 | */ | |
7251 | if (offset + count <= inode->i_size) { | |
7252 | mutex_unlock(&inode->i_mutex); | |
7253 | relock = true; | |
7254 | } | |
0934856d MX |
7255 | ret = btrfs_delalloc_reserve_space(inode, count); |
7256 | if (ret) | |
38851cc1 MX |
7257 | goto out; |
7258 | } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK, | |
7259 | &BTRFS_I(inode)->runtime_flags))) { | |
7260 | inode_dio_done(inode); | |
7261 | flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
7262 | wakeup = false; | |
0934856d MX |
7263 | } |
7264 | ||
7265 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
7266 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
7267 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
2e60a51e | 7268 | btrfs_submit_direct, flags); |
0934856d MX |
7269 | if (rw & WRITE) { |
7270 | if (ret < 0 && ret != -EIOCBQUEUED) | |
7271 | btrfs_delalloc_release_space(inode, count); | |
172a5049 | 7272 | else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
7273 | btrfs_delalloc_release_space(inode, |
7274 | count - (size_t)ret); | |
172a5049 MX |
7275 | else |
7276 | btrfs_delalloc_release_metadata(inode, 0); | |
0934856d | 7277 | } |
38851cc1 | 7278 | out: |
2e60a51e MX |
7279 | if (wakeup) |
7280 | inode_dio_done(inode); | |
38851cc1 MX |
7281 | if (relock) |
7282 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
7283 | |
7284 | return ret; | |
16432985 CM |
7285 | } |
7286 | ||
05dadc09 TI |
7287 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
7288 | ||
1506fcc8 YS |
7289 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
7290 | __u64 start, __u64 len) | |
7291 | { | |
05dadc09 TI |
7292 | int ret; |
7293 | ||
7294 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
7295 | if (ret) | |
7296 | return ret; | |
7297 | ||
ec29ed5b | 7298 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
7299 | } |
7300 | ||
a52d9a80 | 7301 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 7302 | { |
d1310b2e CM |
7303 | struct extent_io_tree *tree; |
7304 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 7305 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 7306 | } |
1832a6d5 | 7307 | |
a52d9a80 | 7308 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 7309 | { |
d1310b2e | 7310 | struct extent_io_tree *tree; |
b888db2b CM |
7311 | |
7312 | ||
7313 | if (current->flags & PF_MEMALLOC) { | |
7314 | redirty_page_for_writepage(wbc, page); | |
7315 | unlock_page(page); | |
7316 | return 0; | |
7317 | } | |
d1310b2e | 7318 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 7319 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
7320 | } |
7321 | ||
48a3b636 ES |
7322 | static int btrfs_writepages(struct address_space *mapping, |
7323 | struct writeback_control *wbc) | |
b293f02e | 7324 | { |
d1310b2e | 7325 | struct extent_io_tree *tree; |
771ed689 | 7326 | |
d1310b2e | 7327 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
7328 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
7329 | } | |
7330 | ||
3ab2fb5a CM |
7331 | static int |
7332 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
7333 | struct list_head *pages, unsigned nr_pages) | |
7334 | { | |
d1310b2e CM |
7335 | struct extent_io_tree *tree; |
7336 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
7337 | return extent_readpages(tree, mapping, pages, nr_pages, |
7338 | btrfs_get_extent); | |
7339 | } | |
e6dcd2dc | 7340 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 7341 | { |
d1310b2e CM |
7342 | struct extent_io_tree *tree; |
7343 | struct extent_map_tree *map; | |
a52d9a80 | 7344 | int ret; |
8c2383c3 | 7345 | |
d1310b2e CM |
7346 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
7347 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 7348 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
7349 | if (ret == 1) { |
7350 | ClearPagePrivate(page); | |
7351 | set_page_private(page, 0); | |
7352 | page_cache_release(page); | |
39279cc3 | 7353 | } |
a52d9a80 | 7354 | return ret; |
39279cc3 CM |
7355 | } |
7356 | ||
e6dcd2dc CM |
7357 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
7358 | { | |
98509cfc CM |
7359 | if (PageWriteback(page) || PageDirty(page)) |
7360 | return 0; | |
b335b003 | 7361 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
7362 | } |
7363 | ||
a52d9a80 | 7364 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 7365 | { |
5fd02043 | 7366 | struct inode *inode = page->mapping->host; |
d1310b2e | 7367 | struct extent_io_tree *tree; |
e6dcd2dc | 7368 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 7369 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7370 | u64 page_start = page_offset(page); |
7371 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 7372 | |
8b62b72b CM |
7373 | /* |
7374 | * we have the page locked, so new writeback can't start, | |
7375 | * and the dirty bit won't be cleared while we are here. | |
7376 | * | |
7377 | * Wait for IO on this page so that we can safely clear | |
7378 | * the PagePrivate2 bit and do ordered accounting | |
7379 | */ | |
e6dcd2dc | 7380 | wait_on_page_writeback(page); |
8b62b72b | 7381 | |
5fd02043 | 7382 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
7383 | if (offset) { |
7384 | btrfs_releasepage(page, GFP_NOFS); | |
7385 | return; | |
7386 | } | |
d0082371 | 7387 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
4eee4fa4 | 7388 | ordered = btrfs_lookup_ordered_extent(inode, page_offset(page)); |
e6dcd2dc | 7389 | if (ordered) { |
eb84ae03 CM |
7390 | /* |
7391 | * IO on this page will never be started, so we need | |
7392 | * to account for any ordered extents now | |
7393 | */ | |
e6dcd2dc CM |
7394 | clear_extent_bit(tree, page_start, page_end, |
7395 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
7396 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
7397 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
7398 | /* |
7399 | * whoever cleared the private bit is responsible | |
7400 | * for the finish_ordered_io | |
7401 | */ | |
5fd02043 JB |
7402 | if (TestClearPagePrivate2(page) && |
7403 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
7404 | PAGE_CACHE_SIZE, 1)) { | |
7405 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 7406 | } |
e6dcd2dc | 7407 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 7408 | cached_state = NULL; |
d0082371 | 7409 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7410 | } |
7411 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 7412 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
7413 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
7414 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
7415 | __btrfs_releasepage(page, GFP_NOFS); |
7416 | ||
4a096752 | 7417 | ClearPageChecked(page); |
9ad6b7bc | 7418 | if (PagePrivate(page)) { |
9ad6b7bc CM |
7419 | ClearPagePrivate(page); |
7420 | set_page_private(page, 0); | |
7421 | page_cache_release(page); | |
7422 | } | |
39279cc3 CM |
7423 | } |
7424 | ||
9ebefb18 CM |
7425 | /* |
7426 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
7427 | * called from a page fault handler when a page is first dirtied. Hence we must | |
7428 | * be careful to check for EOF conditions here. We set the page up correctly | |
7429 | * for a written page which means we get ENOSPC checking when writing into | |
7430 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
7431 | * support these features. | |
7432 | * | |
7433 | * We are not allowed to take the i_mutex here so we have to play games to | |
7434 | * protect against truncate races as the page could now be beyond EOF. Because | |
7435 | * vmtruncate() writes the inode size before removing pages, once we have the | |
7436 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
7437 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
7438 | * unlock the page. | |
7439 | */ | |
c2ec175c | 7440 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 7441 | { |
c2ec175c | 7442 | struct page *page = vmf->page; |
496ad9aa | 7443 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 7444 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
7445 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
7446 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 7447 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
7448 | char *kaddr; |
7449 | unsigned long zero_start; | |
9ebefb18 | 7450 | loff_t size; |
1832a6d5 | 7451 | int ret; |
9998eb70 | 7452 | int reserved = 0; |
a52d9a80 | 7453 | u64 page_start; |
e6dcd2dc | 7454 | u64 page_end; |
9ebefb18 | 7455 | |
b2b5ef5c | 7456 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 7457 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7458 | if (!ret) { |
e41f941a | 7459 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
7460 | reserved = 1; |
7461 | } | |
56a76f82 NP |
7462 | if (ret) { |
7463 | if (ret == -ENOMEM) | |
7464 | ret = VM_FAULT_OOM; | |
7465 | else /* -ENOSPC, -EIO, etc */ | |
7466 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
7467 | if (reserved) |
7468 | goto out; | |
7469 | goto out_noreserve; | |
56a76f82 | 7470 | } |
1832a6d5 | 7471 | |
56a76f82 | 7472 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 7473 | again: |
9ebefb18 | 7474 | lock_page(page); |
9ebefb18 | 7475 | size = i_size_read(inode); |
e6dcd2dc CM |
7476 | page_start = page_offset(page); |
7477 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 7478 | |
9ebefb18 | 7479 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 7480 | (page_start >= size)) { |
9ebefb18 CM |
7481 | /* page got truncated out from underneath us */ |
7482 | goto out_unlock; | |
7483 | } | |
e6dcd2dc CM |
7484 | wait_on_page_writeback(page); |
7485 | ||
d0082371 | 7486 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
7487 | set_page_extent_mapped(page); |
7488 | ||
eb84ae03 CM |
7489 | /* |
7490 | * we can't set the delalloc bits if there are pending ordered | |
7491 | * extents. Drop our locks and wait for them to finish | |
7492 | */ | |
e6dcd2dc CM |
7493 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
7494 | if (ordered) { | |
2ac55d41 JB |
7495 | unlock_extent_cached(io_tree, page_start, page_end, |
7496 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 7497 | unlock_page(page); |
eb84ae03 | 7498 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
7499 | btrfs_put_ordered_extent(ordered); |
7500 | goto again; | |
7501 | } | |
7502 | ||
fbf19087 JB |
7503 | /* |
7504 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
7505 | * if it was already dirty, so for space accounting reasons we need to | |
7506 | * clear any delalloc bits for the range we are fixing to save. There | |
7507 | * is probably a better way to do this, but for now keep consistent with | |
7508 | * prepare_pages in the normal write path. | |
7509 | */ | |
2ac55d41 | 7510 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
7511 | EXTENT_DIRTY | EXTENT_DELALLOC | |
7512 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 7513 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 7514 | |
2ac55d41 JB |
7515 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
7516 | &cached_state); | |
9ed74f2d | 7517 | if (ret) { |
2ac55d41 JB |
7518 | unlock_extent_cached(io_tree, page_start, page_end, |
7519 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
7520 | ret = VM_FAULT_SIGBUS; |
7521 | goto out_unlock; | |
7522 | } | |
e6dcd2dc | 7523 | ret = 0; |
9ebefb18 CM |
7524 | |
7525 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 7526 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 7527 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 7528 | else |
e6dcd2dc | 7529 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 7530 | |
e6dcd2dc CM |
7531 | if (zero_start != PAGE_CACHE_SIZE) { |
7532 | kaddr = kmap(page); | |
7533 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
7534 | flush_dcache_page(page); | |
7535 | kunmap(page); | |
7536 | } | |
247e743c | 7537 | ClearPageChecked(page); |
e6dcd2dc | 7538 | set_page_dirty(page); |
50a9b214 | 7539 | SetPageUptodate(page); |
5a3f23d5 | 7540 | |
257c62e1 CM |
7541 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
7542 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 7543 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 7544 | |
2ac55d41 | 7545 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
7546 | |
7547 | out_unlock: | |
b2b5ef5c JK |
7548 | if (!ret) { |
7549 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 7550 | return VM_FAULT_LOCKED; |
b2b5ef5c | 7551 | } |
9ebefb18 | 7552 | unlock_page(page); |
1832a6d5 | 7553 | out: |
ec39e180 | 7554 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 7555 | out_noreserve: |
b2b5ef5c | 7556 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
7557 | return ret; |
7558 | } | |
7559 | ||
a41ad394 | 7560 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
7561 | { |
7562 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 7563 | struct btrfs_block_rsv *rsv; |
39279cc3 | 7564 | int ret; |
3893e33b | 7565 | int err = 0; |
39279cc3 | 7566 | struct btrfs_trans_handle *trans; |
dbe674a9 | 7567 | u64 mask = root->sectorsize - 1; |
07127184 | 7568 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 7569 | |
2aaa6655 | 7570 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
5d5e103a | 7571 | if (ret) |
a41ad394 | 7572 | return ret; |
8082510e | 7573 | |
4a096752 | 7574 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 7575 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 7576 | |
fcb80c2a JB |
7577 | /* |
7578 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
7579 | * 3 things going on here | |
7580 | * | |
7581 | * 1) We need to reserve space for our orphan item and the space to | |
7582 | * delete our orphan item. Lord knows we don't want to have a dangling | |
7583 | * orphan item because we didn't reserve space to remove it. | |
7584 | * | |
7585 | * 2) We need to reserve space to update our inode. | |
7586 | * | |
7587 | * 3) We need to have something to cache all the space that is going to | |
7588 | * be free'd up by the truncate operation, but also have some slack | |
7589 | * space reserved in case it uses space during the truncate (thank you | |
7590 | * very much snapshotting). | |
7591 | * | |
7592 | * And we need these to all be seperate. The fact is we can use alot of | |
7593 | * space doing the truncate, and we have no earthly idea how much space | |
7594 | * we will use, so we need the truncate reservation to be seperate so it | |
7595 | * doesn't end up using space reserved for updating the inode or | |
7596 | * removing the orphan item. We also need to be able to stop the | |
7597 | * transaction and start a new one, which means we need to be able to | |
7598 | * update the inode several times, and we have no idea of knowing how | |
7599 | * many times that will be, so we can't just reserve 1 item for the | |
7600 | * entirety of the opration, so that has to be done seperately as well. | |
7601 | * Then there is the orphan item, which does indeed need to be held on | |
7602 | * to for the whole operation, and we need nobody to touch this reserved | |
7603 | * space except the orphan code. | |
7604 | * | |
7605 | * So that leaves us with | |
7606 | * | |
7607 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7608 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7609 | * transaction reservation. | |
7610 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7611 | * updating the inode. | |
7612 | */ | |
66d8f3dd | 7613 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7614 | if (!rsv) |
7615 | return -ENOMEM; | |
4a338542 | 7616 | rsv->size = min_size; |
ca7e70f5 | 7617 | rsv->failfast = 1; |
f0cd846e | 7618 | |
907cbceb | 7619 | /* |
07127184 | 7620 | * 1 for the truncate slack space |
907cbceb JB |
7621 | * 1 for updating the inode. |
7622 | */ | |
f3fe820c | 7623 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7624 | if (IS_ERR(trans)) { |
7625 | err = PTR_ERR(trans); | |
7626 | goto out; | |
7627 | } | |
f0cd846e | 7628 | |
907cbceb JB |
7629 | /* Migrate the slack space for the truncate to our reserve */ |
7630 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7631 | min_size); | |
fcb80c2a | 7632 | BUG_ON(ret); |
f0cd846e | 7633 | |
5a3f23d5 CM |
7634 | /* |
7635 | * setattr is responsible for setting the ordered_data_close flag, | |
7636 | * but that is only tested during the last file release. That | |
7637 | * could happen well after the next commit, leaving a great big | |
7638 | * window where new writes may get lost if someone chooses to write | |
7639 | * to this file after truncating to zero | |
7640 | * | |
7641 | * The inode doesn't have any dirty data here, and so if we commit | |
7642 | * this is a noop. If someone immediately starts writing to the inode | |
7643 | * it is very likely we'll catch some of their writes in this | |
7644 | * transaction, and the commit will find this file on the ordered | |
7645 | * data list with good things to send down. | |
7646 | * | |
7647 | * This is a best effort solution, there is still a window where | |
7648 | * using truncate to replace the contents of the file will | |
7649 | * end up with a zero length file after a crash. | |
7650 | */ | |
72ac3c0d JB |
7651 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7652 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7653 | btrfs_add_ordered_operation(trans, root, inode); |
7654 | ||
5dc562c5 JB |
7655 | /* |
7656 | * So if we truncate and then write and fsync we normally would just | |
7657 | * write the extents that changed, which is a problem if we need to | |
7658 | * first truncate that entire inode. So set this flag so we write out | |
7659 | * all of the extents in the inode to the sync log so we're completely | |
7660 | * safe. | |
7661 | */ | |
7662 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7663 | trans->block_rsv = rsv; |
907cbceb | 7664 | |
8082510e YZ |
7665 | while (1) { |
7666 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7667 | inode->i_size, | |
7668 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7669 | if (ret != -ENOSPC) { |
3893e33b | 7670 | err = ret; |
8082510e | 7671 | break; |
3893e33b | 7672 | } |
39279cc3 | 7673 | |
fcb80c2a | 7674 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7675 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7676 | if (ret) { |
7677 | err = ret; | |
7678 | break; | |
7679 | } | |
ca7e70f5 | 7680 | |
8082510e | 7681 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7682 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7683 | |
7684 | trans = btrfs_start_transaction(root, 2); | |
7685 | if (IS_ERR(trans)) { | |
7686 | ret = err = PTR_ERR(trans); | |
7687 | trans = NULL; | |
7688 | break; | |
7689 | } | |
7690 | ||
7691 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7692 | rsv, min_size); | |
7693 | BUG_ON(ret); /* shouldn't happen */ | |
7694 | trans->block_rsv = rsv; | |
8082510e YZ |
7695 | } |
7696 | ||
7697 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7698 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7699 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7700 | if (ret) |
7701 | err = ret; | |
8082510e YZ |
7702 | } |
7703 | ||
917c16b2 CM |
7704 | if (trans) { |
7705 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7706 | ret = btrfs_update_inode(trans, root, inode); | |
7707 | if (ret && !err) | |
7708 | err = ret; | |
7b128766 | 7709 | |
7ad85bb7 | 7710 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7711 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7712 | } |
fcb80c2a JB |
7713 | |
7714 | out: | |
7715 | btrfs_free_block_rsv(root, rsv); | |
7716 | ||
3893e33b JB |
7717 | if (ret && !err) |
7718 | err = ret; | |
a41ad394 | 7719 | |
3893e33b | 7720 | return err; |
39279cc3 CM |
7721 | } |
7722 | ||
d352ac68 CM |
7723 | /* |
7724 | * create a new subvolume directory/inode (helper for the ioctl). | |
7725 | */ | |
d2fb3437 | 7726 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7727 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7728 | { |
39279cc3 | 7729 | struct inode *inode; |
76dda93c | 7730 | int err; |
00e4e6b3 | 7731 | u64 index = 0; |
39279cc3 | 7732 | |
12fc9d09 FA |
7733 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7734 | new_dirid, new_dirid, | |
7735 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7736 | &index); | |
54aa1f4d | 7737 | if (IS_ERR(inode)) |
f46b5a66 | 7738 | return PTR_ERR(inode); |
39279cc3 CM |
7739 | inode->i_op = &btrfs_dir_inode_operations; |
7740 | inode->i_fop = &btrfs_dir_file_operations; | |
7741 | ||
bfe86848 | 7742 | set_nlink(inode, 1); |
dbe674a9 | 7743 | btrfs_i_size_write(inode, 0); |
3b96362c | 7744 | |
76dda93c | 7745 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7746 | |
76dda93c | 7747 | iput(inode); |
ce598979 | 7748 | return err; |
39279cc3 CM |
7749 | } |
7750 | ||
39279cc3 CM |
7751 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7752 | { | |
7753 | struct btrfs_inode *ei; | |
2ead6ae7 | 7754 | struct inode *inode; |
39279cc3 CM |
7755 | |
7756 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7757 | if (!ei) | |
7758 | return NULL; | |
2ead6ae7 YZ |
7759 | |
7760 | ei->root = NULL; | |
2ead6ae7 | 7761 | ei->generation = 0; |
15ee9bc7 | 7762 | ei->last_trans = 0; |
257c62e1 | 7763 | ei->last_sub_trans = 0; |
e02119d5 | 7764 | ei->logged_trans = 0; |
2ead6ae7 | 7765 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7766 | ei->disk_i_size = 0; |
7767 | ei->flags = 0; | |
7709cde3 | 7768 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7769 | ei->index_cnt = (u64)-1; |
7770 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7771 | ei->last_log_commit = 0; |
2ead6ae7 | 7772 | |
9e0baf60 JB |
7773 | spin_lock_init(&ei->lock); |
7774 | ei->outstanding_extents = 0; | |
7775 | ei->reserved_extents = 0; | |
2ead6ae7 | 7776 | |
72ac3c0d | 7777 | ei->runtime_flags = 0; |
261507a0 | 7778 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7779 | |
16cdcec7 MX |
7780 | ei->delayed_node = NULL; |
7781 | ||
2ead6ae7 | 7782 | inode = &ei->vfs_inode; |
a8067e02 | 7783 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7784 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7785 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7786 | ei->io_tree.track_uptodate = 1; |
7787 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7788 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7789 | mutex_init(&ei->log_mutex); |
f248679e | 7790 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7791 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7792 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7793 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7794 | RB_CLEAR_NODE(&ei->rb_node); |
7795 | ||
7796 | return inode; | |
39279cc3 CM |
7797 | } |
7798 | ||
fa0d7e3d NP |
7799 | static void btrfs_i_callback(struct rcu_head *head) |
7800 | { | |
7801 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7802 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7803 | } | |
7804 | ||
39279cc3 CM |
7805 | void btrfs_destroy_inode(struct inode *inode) |
7806 | { | |
e6dcd2dc | 7807 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7808 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7809 | ||
b3d9b7a3 | 7810 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7811 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7812 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7813 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7814 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7815 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7816 | |
a6dbd429 JB |
7817 | /* |
7818 | * This can happen where we create an inode, but somebody else also | |
7819 | * created the same inode and we need to destroy the one we already | |
7820 | * created. | |
7821 | */ | |
7822 | if (!root) | |
7823 | goto free; | |
7824 | ||
5a3f23d5 CM |
7825 | /* |
7826 | * Make sure we're properly removed from the ordered operation | |
7827 | * lists. | |
7828 | */ | |
7829 | smp_mb(); | |
7830 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
199c2a9c | 7831 | spin_lock(&root->fs_info->ordered_root_lock); |
5a3f23d5 | 7832 | list_del_init(&BTRFS_I(inode)->ordered_operations); |
199c2a9c | 7833 | spin_unlock(&root->fs_info->ordered_root_lock); |
5a3f23d5 CM |
7834 | } |
7835 | ||
8a35d95f JB |
7836 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7837 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb SK |
7838 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
7839 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7840 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7841 | } |
7b128766 | 7842 | |
d397712b | 7843 | while (1) { |
e6dcd2dc CM |
7844 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7845 | if (!ordered) | |
7846 | break; | |
7847 | else { | |
c2cf52eb SK |
7848 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
7849 | (unsigned long long)ordered->file_offset, | |
7850 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7851 | btrfs_remove_ordered_extent(inode, ordered); |
7852 | btrfs_put_ordered_extent(ordered); | |
7853 | btrfs_put_ordered_extent(ordered); | |
7854 | } | |
7855 | } | |
5d4f98a2 | 7856 | inode_tree_del(inode); |
5b21f2ed | 7857 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7858 | free: |
fa0d7e3d | 7859 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7860 | } |
7861 | ||
45321ac5 | 7862 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7863 | { |
7864 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7865 | |
6379ef9f NA |
7866 | if (root == NULL) |
7867 | return 1; | |
7868 | ||
fa6ac876 | 7869 | /* the snap/subvol tree is on deleting */ |
0af3d00b | 7870 | if (btrfs_root_refs(&root->root_item) == 0 && |
fa6ac876 | 7871 | root != root->fs_info->tree_root) |
45321ac5 | 7872 | return 1; |
76dda93c | 7873 | else |
45321ac5 | 7874 | return generic_drop_inode(inode); |
76dda93c YZ |
7875 | } |
7876 | ||
0ee0fda0 | 7877 | static void init_once(void *foo) |
39279cc3 CM |
7878 | { |
7879 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7880 | ||
7881 | inode_init_once(&ei->vfs_inode); | |
7882 | } | |
7883 | ||
7884 | void btrfs_destroy_cachep(void) | |
7885 | { | |
8c0a8537 KS |
7886 | /* |
7887 | * Make sure all delayed rcu free inodes are flushed before we | |
7888 | * destroy cache. | |
7889 | */ | |
7890 | rcu_barrier(); | |
39279cc3 CM |
7891 | if (btrfs_inode_cachep) |
7892 | kmem_cache_destroy(btrfs_inode_cachep); | |
7893 | if (btrfs_trans_handle_cachep) | |
7894 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7895 | if (btrfs_transaction_cachep) | |
7896 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7897 | if (btrfs_path_cachep) |
7898 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7899 | if (btrfs_free_space_cachep) |
7900 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7901 | if (btrfs_delalloc_work_cachep) |
7902 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7903 | } |
7904 | ||
7905 | int btrfs_init_cachep(void) | |
7906 | { | |
837e1972 | 7907 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7908 | sizeof(struct btrfs_inode), 0, |
7909 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7910 | if (!btrfs_inode_cachep) |
7911 | goto fail; | |
9601e3f6 | 7912 | |
837e1972 | 7913 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7914 | sizeof(struct btrfs_trans_handle), 0, |
7915 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7916 | if (!btrfs_trans_handle_cachep) |
7917 | goto fail; | |
9601e3f6 | 7918 | |
837e1972 | 7919 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7920 | sizeof(struct btrfs_transaction), 0, |
7921 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7922 | if (!btrfs_transaction_cachep) |
7923 | goto fail; | |
9601e3f6 | 7924 | |
837e1972 | 7925 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7926 | sizeof(struct btrfs_path), 0, |
7927 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7928 | if (!btrfs_path_cachep) |
7929 | goto fail; | |
9601e3f6 | 7930 | |
837e1972 | 7931 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7932 | sizeof(struct btrfs_free_space), 0, |
7933 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7934 | if (!btrfs_free_space_cachep) | |
7935 | goto fail; | |
7936 | ||
8ccf6f19 MX |
7937 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7938 | sizeof(struct btrfs_delalloc_work), 0, | |
7939 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7940 | NULL); | |
7941 | if (!btrfs_delalloc_work_cachep) | |
7942 | goto fail; | |
7943 | ||
39279cc3 CM |
7944 | return 0; |
7945 | fail: | |
7946 | btrfs_destroy_cachep(); | |
7947 | return -ENOMEM; | |
7948 | } | |
7949 | ||
7950 | static int btrfs_getattr(struct vfsmount *mnt, | |
7951 | struct dentry *dentry, struct kstat *stat) | |
7952 | { | |
df0af1a5 | 7953 | u64 delalloc_bytes; |
39279cc3 | 7954 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
7955 | u32 blocksize = inode->i_sb->s_blocksize; |
7956 | ||
39279cc3 | 7957 | generic_fillattr(inode, stat); |
0ee5dc67 | 7958 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7959 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
7960 | |
7961 | spin_lock(&BTRFS_I(inode)->lock); | |
7962 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
7963 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 7964 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 7965 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
7966 | return 0; |
7967 | } | |
7968 | ||
d397712b CM |
7969 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7970 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7971 | { |
7972 | struct btrfs_trans_handle *trans; | |
7973 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7974 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7975 | struct inode *new_inode = new_dentry->d_inode; |
7976 | struct inode *old_inode = old_dentry->d_inode; | |
7977 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7978 | u64 index = 0; |
4df27c4d | 7979 | u64 root_objectid; |
39279cc3 | 7980 | int ret; |
33345d01 | 7981 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7982 | |
33345d01 | 7983 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7984 | return -EPERM; |
7985 | ||
4df27c4d | 7986 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7987 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7988 | return -EXDEV; |
7989 | ||
33345d01 LZ |
7990 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7991 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7992 | return -ENOTEMPTY; |
5f39d397 | 7993 | |
4df27c4d YZ |
7994 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7995 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7996 | return -ENOTEMPTY; | |
9c52057c CM |
7997 | |
7998 | ||
7999 | /* check for collisions, even if the name isn't there */ | |
8000 | ret = btrfs_check_dir_item_collision(root, new_dir->i_ino, | |
8001 | new_dentry->d_name.name, | |
8002 | new_dentry->d_name.len); | |
8003 | ||
8004 | if (ret) { | |
8005 | if (ret == -EEXIST) { | |
8006 | /* we shouldn't get | |
8007 | * eexist without a new_inode */ | |
8008 | if (!new_inode) { | |
8009 | WARN_ON(1); | |
8010 | return ret; | |
8011 | } | |
8012 | } else { | |
8013 | /* maybe -EOVERFLOW */ | |
8014 | return ret; | |
8015 | } | |
8016 | } | |
8017 | ret = 0; | |
8018 | ||
5a3f23d5 CM |
8019 | /* |
8020 | * we're using rename to replace one file with another. | |
8021 | * and the replacement file is large. Start IO on it now so | |
8022 | * we don't add too much work to the end of the transaction | |
8023 | */ | |
4baf8c92 | 8024 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
8025 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
8026 | filemap_flush(old_inode->i_mapping); | |
8027 | ||
76dda93c | 8028 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 8029 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8030 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
8031 | /* |
8032 | * We want to reserve the absolute worst case amount of items. So if | |
8033 | * both inodes are subvols and we need to unlink them then that would | |
8034 | * require 4 item modifications, but if they are both normal inodes it | |
8035 | * would require 5 item modifications, so we'll assume their normal | |
8036 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
8037 | * should cover the worst case number of items we'll modify. | |
8038 | */ | |
6e137ed3 | 8039 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
8040 | if (IS_ERR(trans)) { |
8041 | ret = PTR_ERR(trans); | |
8042 | goto out_notrans; | |
8043 | } | |
76dda93c | 8044 | |
4df27c4d YZ |
8045 | if (dest != root) |
8046 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 8047 | |
a5719521 YZ |
8048 | ret = btrfs_set_inode_index(new_dir, &index); |
8049 | if (ret) | |
8050 | goto out_fail; | |
5a3f23d5 | 8051 | |
33345d01 | 8052 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8053 | /* force full log commit if subvolume involved. */ |
8054 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
8055 | } else { | |
a5719521 YZ |
8056 | ret = btrfs_insert_inode_ref(trans, dest, |
8057 | new_dentry->d_name.name, | |
8058 | new_dentry->d_name.len, | |
33345d01 LZ |
8059 | old_ino, |
8060 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
8061 | if (ret) |
8062 | goto out_fail; | |
4df27c4d YZ |
8063 | /* |
8064 | * this is an ugly little race, but the rename is required | |
8065 | * to make sure that if we crash, the inode is either at the | |
8066 | * old name or the new one. pinning the log transaction lets | |
8067 | * us make sure we don't allow a log commit to come in after | |
8068 | * we unlink the name but before we add the new name back in. | |
8069 | */ | |
8070 | btrfs_pin_log_trans(root); | |
8071 | } | |
5a3f23d5 CM |
8072 | /* |
8073 | * make sure the inode gets flushed if it is replacing | |
8074 | * something. | |
8075 | */ | |
33345d01 | 8076 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 8077 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 8078 | |
0c4d2d95 JB |
8079 | inode_inc_iversion(old_dir); |
8080 | inode_inc_iversion(new_dir); | |
8081 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
8082 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
8083 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
8084 | old_inode->i_ctime = ctime; | |
5f39d397 | 8085 | |
12fcfd22 CM |
8086 | if (old_dentry->d_parent != new_dentry->d_parent) |
8087 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
8088 | ||
33345d01 | 8089 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
8090 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
8091 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
8092 | old_dentry->d_name.name, | |
8093 | old_dentry->d_name.len); | |
8094 | } else { | |
92986796 AV |
8095 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
8096 | old_dentry->d_inode, | |
8097 | old_dentry->d_name.name, | |
8098 | old_dentry->d_name.len); | |
8099 | if (!ret) | |
8100 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 8101 | } |
79787eaa JM |
8102 | if (ret) { |
8103 | btrfs_abort_transaction(trans, root, ret); | |
8104 | goto out_fail; | |
8105 | } | |
39279cc3 CM |
8106 | |
8107 | if (new_inode) { | |
0c4d2d95 | 8108 | inode_inc_iversion(new_inode); |
39279cc3 | 8109 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 8110 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
8111 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
8112 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
8113 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
8114 | root_objectid, | |
8115 | new_dentry->d_name.name, | |
8116 | new_dentry->d_name.len); | |
8117 | BUG_ON(new_inode->i_nlink == 0); | |
8118 | } else { | |
8119 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
8120 | new_dentry->d_inode, | |
8121 | new_dentry->d_name.name, | |
8122 | new_dentry->d_name.len); | |
8123 | } | |
79787eaa | 8124 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 8125 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 8126 | BUG_ON(ret); |
7b128766 | 8127 | } |
79787eaa JM |
8128 | if (ret) { |
8129 | btrfs_abort_transaction(trans, root, ret); | |
8130 | goto out_fail; | |
8131 | } | |
39279cc3 | 8132 | } |
aec7477b | 8133 | |
4df27c4d YZ |
8134 | ret = btrfs_add_link(trans, new_dir, old_inode, |
8135 | new_dentry->d_name.name, | |
a5719521 | 8136 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
8137 | if (ret) { |
8138 | btrfs_abort_transaction(trans, root, ret); | |
8139 | goto out_fail; | |
8140 | } | |
39279cc3 | 8141 | |
33345d01 | 8142 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 8143 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 8144 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
8145 | btrfs_end_log_trans(root); |
8146 | } | |
39279cc3 | 8147 | out_fail: |
7ad85bb7 | 8148 | btrfs_end_transaction(trans, root); |
b44c59a8 | 8149 | out_notrans: |
33345d01 | 8150 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 8151 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 8152 | |
39279cc3 CM |
8153 | return ret; |
8154 | } | |
8155 | ||
8ccf6f19 MX |
8156 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
8157 | { | |
8158 | struct btrfs_delalloc_work *delalloc_work; | |
8159 | ||
8160 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
8161 | work); | |
8162 | if (delalloc_work->wait) | |
8163 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
8164 | else | |
8165 | filemap_flush(delalloc_work->inode->i_mapping); | |
8166 | ||
8167 | if (delalloc_work->delay_iput) | |
8168 | btrfs_add_delayed_iput(delalloc_work->inode); | |
8169 | else | |
8170 | iput(delalloc_work->inode); | |
8171 | complete(&delalloc_work->completion); | |
8172 | } | |
8173 | ||
8174 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
8175 | int wait, int delay_iput) | |
8176 | { | |
8177 | struct btrfs_delalloc_work *work; | |
8178 | ||
8179 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
8180 | if (!work) | |
8181 | return NULL; | |
8182 | ||
8183 | init_completion(&work->completion); | |
8184 | INIT_LIST_HEAD(&work->list); | |
8185 | work->inode = inode; | |
8186 | work->wait = wait; | |
8187 | work->delay_iput = delay_iput; | |
8188 | work->work.func = btrfs_run_delalloc_work; | |
8189 | ||
8190 | return work; | |
8191 | } | |
8192 | ||
8193 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
8194 | { | |
8195 | wait_for_completion(&work->completion); | |
8196 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
8197 | } | |
8198 | ||
d352ac68 CM |
8199 | /* |
8200 | * some fairly slow code that needs optimization. This walks the list | |
8201 | * of all the inodes with pending delalloc and forces them to disk. | |
8202 | */ | |
eb73c1b7 | 8203 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 8204 | { |
ea8c2819 | 8205 | struct btrfs_inode *binode; |
5b21f2ed | 8206 | struct inode *inode; |
8ccf6f19 MX |
8207 | struct btrfs_delalloc_work *work, *next; |
8208 | struct list_head works; | |
1eafa6c7 | 8209 | struct list_head splice; |
8ccf6f19 | 8210 | int ret = 0; |
ea8c2819 | 8211 | |
8ccf6f19 | 8212 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 8213 | INIT_LIST_HEAD(&splice); |
63607cc8 | 8214 | |
eb73c1b7 MX |
8215 | spin_lock(&root->delalloc_lock); |
8216 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
8217 | while (!list_empty(&splice)) { |
8218 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 8219 | delalloc_inodes); |
1eafa6c7 | 8220 | |
eb73c1b7 MX |
8221 | list_move_tail(&binode->delalloc_inodes, |
8222 | &root->delalloc_inodes); | |
5b21f2ed | 8223 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 8224 | if (!inode) { |
eb73c1b7 | 8225 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 8226 | continue; |
df0af1a5 | 8227 | } |
eb73c1b7 | 8228 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
8229 | |
8230 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
8231 | if (unlikely(!work)) { | |
8232 | ret = -ENOMEM; | |
8233 | goto out; | |
5b21f2ed | 8234 | } |
1eafa6c7 MX |
8235 | list_add_tail(&work->list, &works); |
8236 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
8237 | &work->work); | |
8238 | ||
5b21f2ed | 8239 | cond_resched(); |
eb73c1b7 | 8240 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 8241 | } |
eb73c1b7 | 8242 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 8243 | |
1eafa6c7 MX |
8244 | list_for_each_entry_safe(work, next, &works, list) { |
8245 | list_del_init(&work->list); | |
8246 | btrfs_wait_and_free_delalloc_work(work); | |
8247 | } | |
eb73c1b7 MX |
8248 | return 0; |
8249 | out: | |
8250 | list_for_each_entry_safe(work, next, &works, list) { | |
8251 | list_del_init(&work->list); | |
8252 | btrfs_wait_and_free_delalloc_work(work); | |
8253 | } | |
8254 | ||
8255 | if (!list_empty_careful(&splice)) { | |
8256 | spin_lock(&root->delalloc_lock); | |
8257 | list_splice_tail(&splice, &root->delalloc_inodes); | |
8258 | spin_unlock(&root->delalloc_lock); | |
8259 | } | |
8260 | return ret; | |
8261 | } | |
1eafa6c7 | 8262 | |
eb73c1b7 MX |
8263 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
8264 | { | |
8265 | int ret; | |
8266 | ||
8267 | if (root->fs_info->sb->s_flags & MS_RDONLY) | |
8268 | return -EROFS; | |
8269 | ||
8270 | ret = __start_delalloc_inodes(root, delay_iput); | |
8271 | /* | |
8272 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
8273 | * we have to make sure the IO is actually started and that |
8274 | * ordered extents get created before we return | |
8275 | */ | |
8276 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 8277 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 8278 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 8279 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
8280 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
8281 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
8282 | } |
8283 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
8284 | return ret; |
8285 | } | |
8286 | ||
8287 | int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info, | |
8288 | int delay_iput) | |
8289 | { | |
8290 | struct btrfs_root *root; | |
8291 | struct list_head splice; | |
8292 | int ret; | |
8293 | ||
8294 | if (fs_info->sb->s_flags & MS_RDONLY) | |
8295 | return -EROFS; | |
8296 | ||
8297 | INIT_LIST_HEAD(&splice); | |
8298 | ||
8299 | spin_lock(&fs_info->delalloc_root_lock); | |
8300 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
8301 | while (!list_empty(&splice)) { | |
8302 | root = list_first_entry(&splice, struct btrfs_root, | |
8303 | delalloc_root); | |
8304 | root = btrfs_grab_fs_root(root); | |
8305 | BUG_ON(!root); | |
8306 | list_move_tail(&root->delalloc_root, | |
8307 | &fs_info->delalloc_roots); | |
8308 | spin_unlock(&fs_info->delalloc_root_lock); | |
8309 | ||
8310 | ret = __start_delalloc_inodes(root, delay_iput); | |
8311 | btrfs_put_fs_root(root); | |
8312 | if (ret) | |
8313 | goto out; | |
8314 | ||
8315 | spin_lock(&fs_info->delalloc_root_lock); | |
8ccf6f19 | 8316 | } |
eb73c1b7 | 8317 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 8318 | |
eb73c1b7 MX |
8319 | atomic_inc(&fs_info->async_submit_draining); |
8320 | while (atomic_read(&fs_info->nr_async_submits) || | |
8321 | atomic_read(&fs_info->async_delalloc_pages)) { | |
8322 | wait_event(fs_info->async_submit_wait, | |
8323 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
8324 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
8325 | } | |
8326 | atomic_dec(&fs_info->async_submit_draining); | |
8327 | return 0; | |
8328 | out: | |
1eafa6c7 | 8329 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
8330 | spin_lock(&fs_info->delalloc_root_lock); |
8331 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
8332 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 8333 | } |
8ccf6f19 | 8334 | return ret; |
ea8c2819 CM |
8335 | } |
8336 | ||
39279cc3 CM |
8337 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
8338 | const char *symname) | |
8339 | { | |
8340 | struct btrfs_trans_handle *trans; | |
8341 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
8342 | struct btrfs_path *path; | |
8343 | struct btrfs_key key; | |
1832a6d5 | 8344 | struct inode *inode = NULL; |
39279cc3 CM |
8345 | int err; |
8346 | int drop_inode = 0; | |
8347 | u64 objectid; | |
00e4e6b3 | 8348 | u64 index = 0 ; |
39279cc3 CM |
8349 | int name_len; |
8350 | int datasize; | |
5f39d397 | 8351 | unsigned long ptr; |
39279cc3 | 8352 | struct btrfs_file_extent_item *ei; |
5f39d397 | 8353 | struct extent_buffer *leaf; |
39279cc3 CM |
8354 | |
8355 | name_len = strlen(symname) + 1; | |
8356 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
8357 | return -ENAMETOOLONG; | |
1832a6d5 | 8358 | |
9ed74f2d JB |
8359 | /* |
8360 | * 2 items for inode item and ref | |
8361 | * 2 items for dir items | |
8362 | * 1 item for xattr if selinux is on | |
8363 | */ | |
a22285a6 YZ |
8364 | trans = btrfs_start_transaction(root, 5); |
8365 | if (IS_ERR(trans)) | |
8366 | return PTR_ERR(trans); | |
1832a6d5 | 8367 | |
581bb050 LZ |
8368 | err = btrfs_find_free_ino(root, &objectid); |
8369 | if (err) | |
8370 | goto out_unlock; | |
8371 | ||
aec7477b | 8372 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 8373 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 8374 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
8375 | if (IS_ERR(inode)) { |
8376 | err = PTR_ERR(inode); | |
39279cc3 | 8377 | goto out_unlock; |
7cf96da3 | 8378 | } |
39279cc3 | 8379 | |
2a7dba39 | 8380 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
8381 | if (err) { |
8382 | drop_inode = 1; | |
8383 | goto out_unlock; | |
8384 | } | |
8385 | ||
ad19db71 CS |
8386 | /* |
8387 | * If the active LSM wants to access the inode during | |
8388 | * d_instantiate it needs these. Smack checks to see | |
8389 | * if the filesystem supports xattrs by looking at the | |
8390 | * ops vector. | |
8391 | */ | |
8392 | inode->i_fop = &btrfs_file_operations; | |
8393 | inode->i_op = &btrfs_file_inode_operations; | |
8394 | ||
a1b075d2 | 8395 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
8396 | if (err) |
8397 | drop_inode = 1; | |
8398 | else { | |
8399 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 8400 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 8401 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 8402 | } |
39279cc3 CM |
8403 | if (drop_inode) |
8404 | goto out_unlock; | |
8405 | ||
8406 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
8407 | if (!path) { |
8408 | err = -ENOMEM; | |
8409 | drop_inode = 1; | |
8410 | goto out_unlock; | |
8411 | } | |
33345d01 | 8412 | key.objectid = btrfs_ino(inode); |
39279cc3 | 8413 | key.offset = 0; |
39279cc3 CM |
8414 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
8415 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
8416 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
8417 | datasize); | |
54aa1f4d CM |
8418 | if (err) { |
8419 | drop_inode = 1; | |
b0839166 | 8420 | btrfs_free_path(path); |
54aa1f4d CM |
8421 | goto out_unlock; |
8422 | } | |
5f39d397 CM |
8423 | leaf = path->nodes[0]; |
8424 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
8425 | struct btrfs_file_extent_item); | |
8426 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
8427 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 8428 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
8429 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
8430 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
8431 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
8432 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
8433 | ||
39279cc3 | 8434 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
8435 | write_extent_buffer(leaf, symname, ptr, name_len); |
8436 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 8437 | btrfs_free_path(path); |
5f39d397 | 8438 | |
39279cc3 CM |
8439 | inode->i_op = &btrfs_symlink_inode_operations; |
8440 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 8441 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 8442 | inode_set_bytes(inode, name_len); |
dbe674a9 | 8443 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
8444 | err = btrfs_update_inode(trans, root, inode); |
8445 | if (err) | |
8446 | drop_inode = 1; | |
39279cc3 CM |
8447 | |
8448 | out_unlock: | |
08c422c2 AV |
8449 | if (!err) |
8450 | d_instantiate(dentry, inode); | |
7ad85bb7 | 8451 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
8452 | if (drop_inode) { |
8453 | inode_dec_link_count(inode); | |
8454 | iput(inode); | |
8455 | } | |
b53d3f5d | 8456 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
8457 | return err; |
8458 | } | |
16432985 | 8459 | |
0af3d00b JB |
8460 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
8461 | u64 start, u64 num_bytes, u64 min_size, | |
8462 | loff_t actual_len, u64 *alloc_hint, | |
8463 | struct btrfs_trans_handle *trans) | |
d899e052 | 8464 | { |
5dc562c5 JB |
8465 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
8466 | struct extent_map *em; | |
d899e052 YZ |
8467 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8468 | struct btrfs_key ins; | |
d899e052 | 8469 | u64 cur_offset = start; |
55a61d1d | 8470 | u64 i_size; |
154ea289 | 8471 | u64 cur_bytes; |
d899e052 | 8472 | int ret = 0; |
0af3d00b | 8473 | bool own_trans = true; |
d899e052 | 8474 | |
0af3d00b JB |
8475 | if (trans) |
8476 | own_trans = false; | |
d899e052 | 8477 | while (num_bytes > 0) { |
0af3d00b JB |
8478 | if (own_trans) { |
8479 | trans = btrfs_start_transaction(root, 3); | |
8480 | if (IS_ERR(trans)) { | |
8481 | ret = PTR_ERR(trans); | |
8482 | break; | |
8483 | } | |
5a303d5d YZ |
8484 | } |
8485 | ||
154ea289 CM |
8486 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
8487 | cur_bytes = max(cur_bytes, min_size); | |
8488 | ret = btrfs_reserve_extent(trans, root, cur_bytes, | |
24542bf7 | 8489 | min_size, 0, *alloc_hint, &ins, 1); |
5a303d5d | 8490 | if (ret) { |
0af3d00b JB |
8491 | if (own_trans) |
8492 | btrfs_end_transaction(trans, root); | |
a22285a6 | 8493 | break; |
d899e052 | 8494 | } |
5a303d5d | 8495 | |
d899e052 YZ |
8496 | ret = insert_reserved_file_extent(trans, inode, |
8497 | cur_offset, ins.objectid, | |
8498 | ins.offset, ins.offset, | |
920bbbfb | 8499 | ins.offset, 0, 0, 0, |
d899e052 | 8500 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
8501 | if (ret) { |
8502 | btrfs_abort_transaction(trans, root, ret); | |
8503 | if (own_trans) | |
8504 | btrfs_end_transaction(trans, root); | |
8505 | break; | |
8506 | } | |
a1ed835e CM |
8507 | btrfs_drop_extent_cache(inode, cur_offset, |
8508 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 8509 | |
5dc562c5 JB |
8510 | em = alloc_extent_map(); |
8511 | if (!em) { | |
8512 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
8513 | &BTRFS_I(inode)->runtime_flags); | |
8514 | goto next; | |
8515 | } | |
8516 | ||
8517 | em->start = cur_offset; | |
8518 | em->orig_start = cur_offset; | |
8519 | em->len = ins.offset; | |
8520 | em->block_start = ins.objectid; | |
8521 | em->block_len = ins.offset; | |
b4939680 | 8522 | em->orig_block_len = ins.offset; |
cc95bef6 | 8523 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
8524 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
8525 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
8526 | em->generation = trans->transid; | |
8527 | ||
8528 | while (1) { | |
8529 | write_lock(&em_tree->lock); | |
09a2a8f9 | 8530 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
8531 | write_unlock(&em_tree->lock); |
8532 | if (ret != -EEXIST) | |
8533 | break; | |
8534 | btrfs_drop_extent_cache(inode, cur_offset, | |
8535 | cur_offset + ins.offset - 1, | |
8536 | 0); | |
8537 | } | |
8538 | free_extent_map(em); | |
8539 | next: | |
d899e052 YZ |
8540 | num_bytes -= ins.offset; |
8541 | cur_offset += ins.offset; | |
efa56464 | 8542 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 8543 | |
0c4d2d95 | 8544 | inode_inc_iversion(inode); |
d899e052 | 8545 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 8546 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 8547 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
8548 | (actual_len > inode->i_size) && |
8549 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 8550 | if (cur_offset > actual_len) |
55a61d1d | 8551 | i_size = actual_len; |
d1ea6a61 | 8552 | else |
55a61d1d JB |
8553 | i_size = cur_offset; |
8554 | i_size_write(inode, i_size); | |
8555 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
8556 | } |
8557 | ||
d899e052 | 8558 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
8559 | |
8560 | if (ret) { | |
8561 | btrfs_abort_transaction(trans, root, ret); | |
8562 | if (own_trans) | |
8563 | btrfs_end_transaction(trans, root); | |
8564 | break; | |
8565 | } | |
d899e052 | 8566 | |
0af3d00b JB |
8567 | if (own_trans) |
8568 | btrfs_end_transaction(trans, root); | |
5a303d5d | 8569 | } |
d899e052 YZ |
8570 | return ret; |
8571 | } | |
8572 | ||
0af3d00b JB |
8573 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
8574 | u64 start, u64 num_bytes, u64 min_size, | |
8575 | loff_t actual_len, u64 *alloc_hint) | |
8576 | { | |
8577 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8578 | min_size, actual_len, alloc_hint, | |
8579 | NULL); | |
8580 | } | |
8581 | ||
8582 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
8583 | struct btrfs_trans_handle *trans, int mode, | |
8584 | u64 start, u64 num_bytes, u64 min_size, | |
8585 | loff_t actual_len, u64 *alloc_hint) | |
8586 | { | |
8587 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
8588 | min_size, actual_len, alloc_hint, trans); | |
8589 | } | |
8590 | ||
e6dcd2dc CM |
8591 | static int btrfs_set_page_dirty(struct page *page) |
8592 | { | |
e6dcd2dc CM |
8593 | return __set_page_dirty_nobuffers(page); |
8594 | } | |
8595 | ||
10556cb2 | 8596 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 8597 | { |
b83cc969 | 8598 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 8599 | umode_t mode = inode->i_mode; |
b83cc969 | 8600 | |
cb6db4e5 JM |
8601 | if (mask & MAY_WRITE && |
8602 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
8603 | if (btrfs_root_readonly(root)) | |
8604 | return -EROFS; | |
8605 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
8606 | return -EACCES; | |
8607 | } | |
2830ba7f | 8608 | return generic_permission(inode, mask); |
fdebe2bd | 8609 | } |
39279cc3 | 8610 | |
6e1d5dcc | 8611 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 8612 | .getattr = btrfs_getattr, |
39279cc3 CM |
8613 | .lookup = btrfs_lookup, |
8614 | .create = btrfs_create, | |
8615 | .unlink = btrfs_unlink, | |
8616 | .link = btrfs_link, | |
8617 | .mkdir = btrfs_mkdir, | |
8618 | .rmdir = btrfs_rmdir, | |
8619 | .rename = btrfs_rename, | |
8620 | .symlink = btrfs_symlink, | |
8621 | .setattr = btrfs_setattr, | |
618e21d5 | 8622 | .mknod = btrfs_mknod, |
95819c05 CH |
8623 | .setxattr = btrfs_setxattr, |
8624 | .getxattr = btrfs_getxattr, | |
5103e947 | 8625 | .listxattr = btrfs_listxattr, |
95819c05 | 8626 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8627 | .permission = btrfs_permission, |
4e34e719 | 8628 | .get_acl = btrfs_get_acl, |
39279cc3 | 8629 | }; |
6e1d5dcc | 8630 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8631 | .lookup = btrfs_lookup, |
fdebe2bd | 8632 | .permission = btrfs_permission, |
4e34e719 | 8633 | .get_acl = btrfs_get_acl, |
39279cc3 | 8634 | }; |
76dda93c | 8635 | |
828c0950 | 8636 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8637 | .llseek = generic_file_llseek, |
8638 | .read = generic_read_dir, | |
cbdf5a24 | 8639 | .readdir = btrfs_real_readdir, |
34287aa3 | 8640 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8641 | #ifdef CONFIG_COMPAT |
34287aa3 | 8642 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8643 | #endif |
6bf13c0c | 8644 | .release = btrfs_release_file, |
e02119d5 | 8645 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8646 | }; |
8647 | ||
d1310b2e | 8648 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8649 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8650 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8651 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8652 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8653 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8654 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8655 | .set_bit_hook = btrfs_set_bit_hook, |
8656 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8657 | .merge_extent_hook = btrfs_merge_extent_hook, |
8658 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8659 | }; |
8660 | ||
35054394 CM |
8661 | /* |
8662 | * btrfs doesn't support the bmap operation because swapfiles | |
8663 | * use bmap to make a mapping of extents in the file. They assume | |
8664 | * these extents won't change over the life of the file and they | |
8665 | * use the bmap result to do IO directly to the drive. | |
8666 | * | |
8667 | * the btrfs bmap call would return logical addresses that aren't | |
8668 | * suitable for IO and they also will change frequently as COW | |
8669 | * operations happen. So, swapfile + btrfs == corruption. | |
8670 | * | |
8671 | * For now we're avoiding this by dropping bmap. | |
8672 | */ | |
7f09410b | 8673 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8674 | .readpage = btrfs_readpage, |
8675 | .writepage = btrfs_writepage, | |
b293f02e | 8676 | .writepages = btrfs_writepages, |
3ab2fb5a | 8677 | .readpages = btrfs_readpages, |
16432985 | 8678 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8679 | .invalidatepage = btrfs_invalidatepage, |
8680 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8681 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8682 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8683 | }; |
8684 | ||
7f09410b | 8685 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8686 | .readpage = btrfs_readpage, |
8687 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8688 | .invalidatepage = btrfs_invalidatepage, |
8689 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8690 | }; |
8691 | ||
6e1d5dcc | 8692 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8693 | .getattr = btrfs_getattr, |
8694 | .setattr = btrfs_setattr, | |
95819c05 CH |
8695 | .setxattr = btrfs_setxattr, |
8696 | .getxattr = btrfs_getxattr, | |
5103e947 | 8697 | .listxattr = btrfs_listxattr, |
95819c05 | 8698 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8699 | .permission = btrfs_permission, |
1506fcc8 | 8700 | .fiemap = btrfs_fiemap, |
4e34e719 | 8701 | .get_acl = btrfs_get_acl, |
e41f941a | 8702 | .update_time = btrfs_update_time, |
39279cc3 | 8703 | }; |
6e1d5dcc | 8704 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8705 | .getattr = btrfs_getattr, |
8706 | .setattr = btrfs_setattr, | |
fdebe2bd | 8707 | .permission = btrfs_permission, |
95819c05 CH |
8708 | .setxattr = btrfs_setxattr, |
8709 | .getxattr = btrfs_getxattr, | |
33268eaf | 8710 | .listxattr = btrfs_listxattr, |
95819c05 | 8711 | .removexattr = btrfs_removexattr, |
4e34e719 | 8712 | .get_acl = btrfs_get_acl, |
e41f941a | 8713 | .update_time = btrfs_update_time, |
618e21d5 | 8714 | }; |
6e1d5dcc | 8715 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8716 | .readlink = generic_readlink, |
8717 | .follow_link = page_follow_link_light, | |
8718 | .put_link = page_put_link, | |
f209561a | 8719 | .getattr = btrfs_getattr, |
22c44fe6 | 8720 | .setattr = btrfs_setattr, |
fdebe2bd | 8721 | .permission = btrfs_permission, |
0279b4cd JO |
8722 | .setxattr = btrfs_setxattr, |
8723 | .getxattr = btrfs_getxattr, | |
8724 | .listxattr = btrfs_listxattr, | |
8725 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8726 | .get_acl = btrfs_get_acl, |
e41f941a | 8727 | .update_time = btrfs_update_time, |
39279cc3 | 8728 | }; |
76dda93c | 8729 | |
82d339d9 | 8730 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8731 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8732 | .d_release = btrfs_dentry_release, |
76dda93c | 8733 | }; |