]>
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> |
4b4e25f2 | 39 | #include "compat.h" |
39279cc3 CM |
40 | #include "ctree.h" |
41 | #include "disk-io.h" | |
42 | #include "transaction.h" | |
43 | #include "btrfs_inode.h" | |
44 | #include "ioctl.h" | |
45 | #include "print-tree.h" | |
0b86a832 | 46 | #include "volumes.h" |
e6dcd2dc | 47 | #include "ordered-data.h" |
95819c05 | 48 | #include "xattr.h" |
e02119d5 | 49 | #include "tree-log.h" |
c8b97818 | 50 | #include "compression.h" |
b4ce94de | 51 | #include "locking.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
54 | u64 ino; | |
55 | struct btrfs_root *root; | |
56 | }; | |
57 | ||
58 | static struct inode_operations btrfs_dir_inode_operations; | |
59 | static struct inode_operations btrfs_symlink_inode_operations; | |
60 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 61 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
62 | static struct inode_operations btrfs_file_inode_operations; |
63 | static struct address_space_operations btrfs_aops; | |
64 | static struct address_space_operations btrfs_symlink_aops; | |
65 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 66 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
67 | |
68 | static struct kmem_cache *btrfs_inode_cachep; | |
69 | struct kmem_cache *btrfs_trans_handle_cachep; | |
70 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
71 | struct kmem_cache *btrfs_path_cachep; |
72 | ||
73 | #define S_SHIFT 12 | |
74 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
75 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
76 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
77 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
78 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
79 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
80 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
81 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
82 | }; | |
83 | ||
7b128766 | 84 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 85 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
86 | static noinline int cow_file_range(struct inode *inode, |
87 | struct page *locked_page, | |
88 | u64 start, u64 end, int *page_started, | |
89 | unsigned long *nr_written, int unlock); | |
7b128766 | 90 | |
0279b4cd JO |
91 | static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) |
92 | { | |
93 | int err; | |
94 | ||
95 | err = btrfs_init_acl(inode, dir); | |
96 | if (!err) | |
97 | err = btrfs_xattr_security_init(inode, dir); | |
98 | return err; | |
99 | } | |
100 | ||
c8b97818 CM |
101 | /* |
102 | * this does all the hard work for inserting an inline extent into | |
103 | * the btree. The caller should have done a btrfs_drop_extents so that | |
104 | * no overlapping inline items exist in the btree | |
105 | */ | |
d397712b | 106 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
107 | struct btrfs_root *root, struct inode *inode, |
108 | u64 start, size_t size, size_t compressed_size, | |
109 | struct page **compressed_pages) | |
110 | { | |
111 | struct btrfs_key key; | |
112 | struct btrfs_path *path; | |
113 | struct extent_buffer *leaf; | |
114 | struct page *page = NULL; | |
115 | char *kaddr; | |
116 | unsigned long ptr; | |
117 | struct btrfs_file_extent_item *ei; | |
118 | int err = 0; | |
119 | int ret; | |
120 | size_t cur_size = size; | |
121 | size_t datasize; | |
122 | unsigned long offset; | |
123 | int use_compress = 0; | |
124 | ||
125 | if (compressed_size && compressed_pages) { | |
126 | use_compress = 1; | |
127 | cur_size = compressed_size; | |
128 | } | |
129 | ||
d397712b CM |
130 | path = btrfs_alloc_path(); |
131 | if (!path) | |
c8b97818 CM |
132 | return -ENOMEM; |
133 | ||
b9473439 | 134 | path->leave_spinning = 1; |
c8b97818 CM |
135 | btrfs_set_trans_block_group(trans, inode); |
136 | ||
137 | key.objectid = inode->i_ino; | |
138 | key.offset = start; | |
139 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
140 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
141 | ||
142 | inode_add_bytes(inode, size); | |
143 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
144 | datasize); | |
145 | BUG_ON(ret); | |
146 | if (ret) { | |
147 | err = ret; | |
c8b97818 CM |
148 | goto fail; |
149 | } | |
150 | leaf = path->nodes[0]; | |
151 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
152 | struct btrfs_file_extent_item); | |
153 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
154 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
155 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
156 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
157 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
158 | ptr = btrfs_file_extent_inline_start(ei); | |
159 | ||
160 | if (use_compress) { | |
161 | struct page *cpage; | |
162 | int i = 0; | |
d397712b | 163 | while (compressed_size > 0) { |
c8b97818 | 164 | cpage = compressed_pages[i]; |
5b050f04 | 165 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
166 | PAGE_CACHE_SIZE); |
167 | ||
b9473439 | 168 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 169 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 170 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
171 | |
172 | i++; | |
173 | ptr += cur_size; | |
174 | compressed_size -= cur_size; | |
175 | } | |
176 | btrfs_set_file_extent_compression(leaf, ei, | |
177 | BTRFS_COMPRESS_ZLIB); | |
178 | } else { | |
179 | page = find_get_page(inode->i_mapping, | |
180 | start >> PAGE_CACHE_SHIFT); | |
181 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
182 | kaddr = kmap_atomic(page, KM_USER0); | |
183 | offset = start & (PAGE_CACHE_SIZE - 1); | |
184 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
185 | kunmap_atomic(kaddr, KM_USER0); | |
186 | page_cache_release(page); | |
187 | } | |
188 | btrfs_mark_buffer_dirty(leaf); | |
189 | btrfs_free_path(path); | |
190 | ||
191 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
192 | btrfs_update_inode(trans, root, inode); | |
193 | return 0; | |
194 | fail: | |
195 | btrfs_free_path(path); | |
196 | return err; | |
197 | } | |
198 | ||
199 | ||
200 | /* | |
201 | * conditionally insert an inline extent into the file. This | |
202 | * does the checks required to make sure the data is small enough | |
203 | * to fit as an inline extent. | |
204 | */ | |
7f366cfe | 205 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
206 | struct btrfs_root *root, |
207 | struct inode *inode, u64 start, u64 end, | |
208 | size_t compressed_size, | |
209 | struct page **compressed_pages) | |
210 | { | |
211 | u64 isize = i_size_read(inode); | |
212 | u64 actual_end = min(end + 1, isize); | |
213 | u64 inline_len = actual_end - start; | |
214 | u64 aligned_end = (end + root->sectorsize - 1) & | |
215 | ~((u64)root->sectorsize - 1); | |
216 | u64 hint_byte; | |
217 | u64 data_len = inline_len; | |
218 | int ret; | |
219 | ||
220 | if (compressed_size) | |
221 | data_len = compressed_size; | |
222 | ||
223 | if (start > 0 || | |
70b99e69 | 224 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
225 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
226 | (!compressed_size && | |
227 | (actual_end & (root->sectorsize - 1)) == 0) || | |
228 | end + 1 < isize || | |
229 | data_len > root->fs_info->max_inline) { | |
230 | return 1; | |
231 | } | |
232 | ||
c8b97818 | 233 | ret = btrfs_drop_extents(trans, root, inode, start, |
a1ed835e CM |
234 | aligned_end, aligned_end, start, |
235 | &hint_byte, 1); | |
c8b97818 CM |
236 | BUG_ON(ret); |
237 | ||
238 | if (isize > actual_end) | |
239 | inline_len = min_t(u64, isize, actual_end); | |
240 | ret = insert_inline_extent(trans, root, inode, start, | |
241 | inline_len, compressed_size, | |
242 | compressed_pages); | |
243 | BUG_ON(ret); | |
a1ed835e | 244 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
245 | return 0; |
246 | } | |
247 | ||
771ed689 CM |
248 | struct async_extent { |
249 | u64 start; | |
250 | u64 ram_size; | |
251 | u64 compressed_size; | |
252 | struct page **pages; | |
253 | unsigned long nr_pages; | |
254 | struct list_head list; | |
255 | }; | |
256 | ||
257 | struct async_cow { | |
258 | struct inode *inode; | |
259 | struct btrfs_root *root; | |
260 | struct page *locked_page; | |
261 | u64 start; | |
262 | u64 end; | |
263 | struct list_head extents; | |
264 | struct btrfs_work work; | |
265 | }; | |
266 | ||
267 | static noinline int add_async_extent(struct async_cow *cow, | |
268 | u64 start, u64 ram_size, | |
269 | u64 compressed_size, | |
270 | struct page **pages, | |
271 | unsigned long nr_pages) | |
272 | { | |
273 | struct async_extent *async_extent; | |
274 | ||
275 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
276 | async_extent->start = start; | |
277 | async_extent->ram_size = ram_size; | |
278 | async_extent->compressed_size = compressed_size; | |
279 | async_extent->pages = pages; | |
280 | async_extent->nr_pages = nr_pages; | |
281 | list_add_tail(&async_extent->list, &cow->extents); | |
282 | return 0; | |
283 | } | |
284 | ||
d352ac68 | 285 | /* |
771ed689 CM |
286 | * we create compressed extents in two phases. The first |
287 | * phase compresses a range of pages that have already been | |
288 | * locked (both pages and state bits are locked). | |
c8b97818 | 289 | * |
771ed689 CM |
290 | * This is done inside an ordered work queue, and the compression |
291 | * is spread across many cpus. The actual IO submission is step | |
292 | * two, and the ordered work queue takes care of making sure that | |
293 | * happens in the same order things were put onto the queue by | |
294 | * writepages and friends. | |
c8b97818 | 295 | * |
771ed689 CM |
296 | * If this code finds it can't get good compression, it puts an |
297 | * entry onto the work queue to write the uncompressed bytes. This | |
298 | * makes sure that both compressed inodes and uncompressed inodes | |
299 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 300 | */ |
771ed689 CM |
301 | static noinline int compress_file_range(struct inode *inode, |
302 | struct page *locked_page, | |
303 | u64 start, u64 end, | |
304 | struct async_cow *async_cow, | |
305 | int *num_added) | |
b888db2b CM |
306 | { |
307 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
308 | struct btrfs_trans_handle *trans; | |
db94535d | 309 | u64 num_bytes; |
c8b97818 CM |
310 | u64 orig_start; |
311 | u64 disk_num_bytes; | |
db94535d | 312 | u64 blocksize = root->sectorsize; |
c8b97818 | 313 | u64 actual_end; |
42dc7bab | 314 | u64 isize = i_size_read(inode); |
e6dcd2dc | 315 | int ret = 0; |
c8b97818 CM |
316 | struct page **pages = NULL; |
317 | unsigned long nr_pages; | |
318 | unsigned long nr_pages_ret = 0; | |
319 | unsigned long total_compressed = 0; | |
320 | unsigned long total_in = 0; | |
321 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 322 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
323 | int i; |
324 | int will_compress; | |
b888db2b | 325 | |
c8b97818 CM |
326 | orig_start = start; |
327 | ||
42dc7bab | 328 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
329 | again: |
330 | will_compress = 0; | |
331 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
332 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 333 | |
f03d9301 CM |
334 | /* |
335 | * we don't want to send crud past the end of i_size through | |
336 | * compression, that's just a waste of CPU time. So, if the | |
337 | * end of the file is before the start of our current | |
338 | * requested range of bytes, we bail out to the uncompressed | |
339 | * cleanup code that can deal with all of this. | |
340 | * | |
341 | * It isn't really the fastest way to fix things, but this is a | |
342 | * very uncommon corner. | |
343 | */ | |
344 | if (actual_end <= start) | |
345 | goto cleanup_and_bail_uncompressed; | |
346 | ||
c8b97818 CM |
347 | total_compressed = actual_end - start; |
348 | ||
349 | /* we want to make sure that amount of ram required to uncompress | |
350 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
351 | * of a compressed extent to 128k. This is a crucial number |
352 | * because it also controls how easily we can spread reads across | |
353 | * cpus for decompression. | |
354 | * | |
355 | * We also want to make sure the amount of IO required to do | |
356 | * a random read is reasonably small, so we limit the size of | |
357 | * a compressed extent to 128k. | |
c8b97818 CM |
358 | */ |
359 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 360 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 361 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
362 | disk_num_bytes = num_bytes; |
363 | total_in = 0; | |
364 | ret = 0; | |
db94535d | 365 | |
771ed689 CM |
366 | /* |
367 | * we do compression for mount -o compress and when the | |
368 | * inode has not been flagged as nocompress. This flag can | |
369 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 370 | */ |
6cbff00f | 371 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
372 | btrfs_test_opt(root, COMPRESS)) { |
373 | WARN_ON(pages); | |
cfbc246e | 374 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 375 | |
c8b97818 CM |
376 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
377 | total_compressed, pages, | |
378 | nr_pages, &nr_pages_ret, | |
379 | &total_in, | |
380 | &total_compressed, | |
381 | max_compressed); | |
382 | ||
383 | if (!ret) { | |
384 | unsigned long offset = total_compressed & | |
385 | (PAGE_CACHE_SIZE - 1); | |
386 | struct page *page = pages[nr_pages_ret - 1]; | |
387 | char *kaddr; | |
388 | ||
389 | /* zero the tail end of the last page, we might be | |
390 | * sending it down to disk | |
391 | */ | |
392 | if (offset) { | |
393 | kaddr = kmap_atomic(page, KM_USER0); | |
394 | memset(kaddr + offset, 0, | |
395 | PAGE_CACHE_SIZE - offset); | |
396 | kunmap_atomic(kaddr, KM_USER0); | |
397 | } | |
398 | will_compress = 1; | |
399 | } | |
400 | } | |
401 | if (start == 0) { | |
771ed689 CM |
402 | trans = btrfs_join_transaction(root, 1); |
403 | BUG_ON(!trans); | |
404 | btrfs_set_trans_block_group(trans, inode); | |
405 | ||
c8b97818 | 406 | /* lets try to make an inline extent */ |
771ed689 | 407 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 408 | /* we didn't compress the entire range, try |
771ed689 | 409 | * to make an uncompressed inline extent. |
c8b97818 CM |
410 | */ |
411 | ret = cow_file_range_inline(trans, root, inode, | |
412 | start, end, 0, NULL); | |
413 | } else { | |
771ed689 | 414 | /* try making a compressed inline extent */ |
c8b97818 CM |
415 | ret = cow_file_range_inline(trans, root, inode, |
416 | start, end, | |
417 | total_compressed, pages); | |
418 | } | |
771ed689 | 419 | btrfs_end_transaction(trans, root); |
c8b97818 | 420 | if (ret == 0) { |
771ed689 CM |
421 | /* |
422 | * inline extent creation worked, we don't need | |
423 | * to create any more async work items. Unlock | |
424 | * and free up our temp pages. | |
425 | */ | |
c8b97818 | 426 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
427 | &BTRFS_I(inode)->io_tree, |
428 | start, end, NULL, | |
429 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 430 | EXTENT_CLEAR_DELALLOC | |
32c00aff | 431 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e | 432 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c8b97818 CM |
433 | ret = 0; |
434 | goto free_pages_out; | |
435 | } | |
436 | } | |
437 | ||
438 | if (will_compress) { | |
439 | /* | |
440 | * we aren't doing an inline extent round the compressed size | |
441 | * up to a block size boundary so the allocator does sane | |
442 | * things | |
443 | */ | |
444 | total_compressed = (total_compressed + blocksize - 1) & | |
445 | ~(blocksize - 1); | |
446 | ||
447 | /* | |
448 | * one last check to make sure the compression is really a | |
449 | * win, compare the page count read with the blocks on disk | |
450 | */ | |
451 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
452 | ~(PAGE_CACHE_SIZE - 1); | |
453 | if (total_compressed >= total_in) { | |
454 | will_compress = 0; | |
455 | } else { | |
456 | disk_num_bytes = total_compressed; | |
457 | num_bytes = total_in; | |
458 | } | |
459 | } | |
460 | if (!will_compress && pages) { | |
461 | /* | |
462 | * the compression code ran but failed to make things smaller, | |
463 | * free any pages it allocated and our page pointer array | |
464 | */ | |
465 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 466 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
467 | page_cache_release(pages[i]); |
468 | } | |
469 | kfree(pages); | |
470 | pages = NULL; | |
471 | total_compressed = 0; | |
472 | nr_pages_ret = 0; | |
473 | ||
474 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 475 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 476 | } |
771ed689 CM |
477 | if (will_compress) { |
478 | *num_added += 1; | |
c8b97818 | 479 | |
771ed689 CM |
480 | /* the async work queues will take care of doing actual |
481 | * allocation on disk for these compressed pages, | |
482 | * and will submit them to the elevator. | |
483 | */ | |
484 | add_async_extent(async_cow, start, num_bytes, | |
485 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 486 | |
42dc7bab | 487 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
488 | start += num_bytes; |
489 | pages = NULL; | |
490 | cond_resched(); | |
491 | goto again; | |
492 | } | |
493 | } else { | |
f03d9301 | 494 | cleanup_and_bail_uncompressed: |
771ed689 CM |
495 | /* |
496 | * No compression, but we still need to write the pages in | |
497 | * the file we've been given so far. redirty the locked | |
498 | * page if it corresponds to our extent and set things up | |
499 | * for the async work queue to run cow_file_range to do | |
500 | * the normal delalloc dance | |
501 | */ | |
502 | if (page_offset(locked_page) >= start && | |
503 | page_offset(locked_page) <= end) { | |
504 | __set_page_dirty_nobuffers(locked_page); | |
505 | /* unlocked later on in the async handlers */ | |
506 | } | |
507 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
508 | *num_added += 1; | |
509 | } | |
3b951516 | 510 | |
771ed689 CM |
511 | out: |
512 | return 0; | |
513 | ||
514 | free_pages_out: | |
515 | for (i = 0; i < nr_pages_ret; i++) { | |
516 | WARN_ON(pages[i]->mapping); | |
517 | page_cache_release(pages[i]); | |
518 | } | |
d397712b | 519 | kfree(pages); |
771ed689 CM |
520 | |
521 | goto out; | |
522 | } | |
523 | ||
524 | /* | |
525 | * phase two of compressed writeback. This is the ordered portion | |
526 | * of the code, which only gets called in the order the work was | |
527 | * queued. We walk all the async extents created by compress_file_range | |
528 | * and send them down to the disk. | |
529 | */ | |
530 | static noinline int submit_compressed_extents(struct inode *inode, | |
531 | struct async_cow *async_cow) | |
532 | { | |
533 | struct async_extent *async_extent; | |
534 | u64 alloc_hint = 0; | |
535 | struct btrfs_trans_handle *trans; | |
536 | struct btrfs_key ins; | |
537 | struct extent_map *em; | |
538 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
539 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
540 | struct extent_io_tree *io_tree; | |
541 | int ret; | |
542 | ||
543 | if (list_empty(&async_cow->extents)) | |
544 | return 0; | |
545 | ||
546 | trans = btrfs_join_transaction(root, 1); | |
547 | ||
d397712b | 548 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
549 | async_extent = list_entry(async_cow->extents.next, |
550 | struct async_extent, list); | |
551 | list_del(&async_extent->list); | |
c8b97818 | 552 | |
771ed689 CM |
553 | io_tree = &BTRFS_I(inode)->io_tree; |
554 | ||
555 | /* did the compression code fall back to uncompressed IO? */ | |
556 | if (!async_extent->pages) { | |
557 | int page_started = 0; | |
558 | unsigned long nr_written = 0; | |
559 | ||
560 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
561 | async_extent->start + |
562 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
563 | |
564 | /* allocate blocks */ | |
565 | cow_file_range(inode, async_cow->locked_page, | |
566 | async_extent->start, | |
567 | async_extent->start + | |
568 | async_extent->ram_size - 1, | |
569 | &page_started, &nr_written, 0); | |
570 | ||
571 | /* | |
572 | * if page_started, cow_file_range inserted an | |
573 | * inline extent and took care of all the unlocking | |
574 | * and IO for us. Otherwise, we need to submit | |
575 | * all those pages down to the drive. | |
576 | */ | |
577 | if (!page_started) | |
578 | extent_write_locked_range(io_tree, | |
579 | inode, async_extent->start, | |
d397712b | 580 | async_extent->start + |
771ed689 CM |
581 | async_extent->ram_size - 1, |
582 | btrfs_get_extent, | |
583 | WB_SYNC_ALL); | |
584 | kfree(async_extent); | |
585 | cond_resched(); | |
586 | continue; | |
587 | } | |
588 | ||
589 | lock_extent(io_tree, async_extent->start, | |
590 | async_extent->start + async_extent->ram_size - 1, | |
591 | GFP_NOFS); | |
c8b97818 | 592 | /* |
771ed689 CM |
593 | * here we're doing allocation and writeback of the |
594 | * compressed pages | |
c8b97818 | 595 | */ |
771ed689 CM |
596 | btrfs_drop_extent_cache(inode, async_extent->start, |
597 | async_extent->start + | |
598 | async_extent->ram_size - 1, 0); | |
599 | ||
600 | ret = btrfs_reserve_extent(trans, root, | |
601 | async_extent->compressed_size, | |
602 | async_extent->compressed_size, | |
603 | 0, alloc_hint, | |
604 | (u64)-1, &ins, 1); | |
605 | BUG_ON(ret); | |
606 | em = alloc_extent_map(GFP_NOFS); | |
607 | em->start = async_extent->start; | |
608 | em->len = async_extent->ram_size; | |
445a6944 | 609 | em->orig_start = em->start; |
c8b97818 | 610 | |
771ed689 CM |
611 | em->block_start = ins.objectid; |
612 | em->block_len = ins.offset; | |
613 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
614 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
615 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
616 | ||
d397712b | 617 | while (1) { |
890871be | 618 | write_lock(&em_tree->lock); |
771ed689 | 619 | ret = add_extent_mapping(em_tree, em); |
890871be | 620 | write_unlock(&em_tree->lock); |
771ed689 CM |
621 | if (ret != -EEXIST) { |
622 | free_extent_map(em); | |
623 | break; | |
624 | } | |
625 | btrfs_drop_extent_cache(inode, async_extent->start, | |
626 | async_extent->start + | |
627 | async_extent->ram_size - 1, 0); | |
628 | } | |
629 | ||
630 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
631 | ins.objectid, | |
632 | async_extent->ram_size, | |
633 | ins.offset, | |
634 | BTRFS_ORDERED_COMPRESSED); | |
635 | BUG_ON(ret); | |
636 | ||
637 | btrfs_end_transaction(trans, root); | |
638 | ||
639 | /* | |
640 | * clear dirty, set writeback and unlock the pages. | |
641 | */ | |
642 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
643 | &BTRFS_I(inode)->io_tree, |
644 | async_extent->start, | |
645 | async_extent->start + | |
646 | async_extent->ram_size - 1, | |
647 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
648 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 649 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 650 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
651 | |
652 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
653 | async_extent->start, |
654 | async_extent->ram_size, | |
655 | ins.objectid, | |
656 | ins.offset, async_extent->pages, | |
657 | async_extent->nr_pages); | |
771ed689 CM |
658 | |
659 | BUG_ON(ret); | |
660 | trans = btrfs_join_transaction(root, 1); | |
661 | alloc_hint = ins.objectid + ins.offset; | |
662 | kfree(async_extent); | |
663 | cond_resched(); | |
664 | } | |
665 | ||
666 | btrfs_end_transaction(trans, root); | |
667 | return 0; | |
668 | } | |
669 | ||
670 | /* | |
671 | * when extent_io.c finds a delayed allocation range in the file, | |
672 | * the call backs end up in this code. The basic idea is to | |
673 | * allocate extents on disk for the range, and create ordered data structs | |
674 | * in ram to track those extents. | |
675 | * | |
676 | * locked_page is the page that writepage had locked already. We use | |
677 | * it to make sure we don't do extra locks or unlocks. | |
678 | * | |
679 | * *page_started is set to one if we unlock locked_page and do everything | |
680 | * required to start IO on it. It may be clean and already done with | |
681 | * IO when we return. | |
682 | */ | |
683 | static noinline int cow_file_range(struct inode *inode, | |
684 | struct page *locked_page, | |
685 | u64 start, u64 end, int *page_started, | |
686 | unsigned long *nr_written, | |
687 | int unlock) | |
688 | { | |
689 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
690 | struct btrfs_trans_handle *trans; | |
691 | u64 alloc_hint = 0; | |
692 | u64 num_bytes; | |
693 | unsigned long ram_size; | |
694 | u64 disk_num_bytes; | |
695 | u64 cur_alloc_size; | |
696 | u64 blocksize = root->sectorsize; | |
697 | u64 actual_end; | |
42dc7bab | 698 | u64 isize = i_size_read(inode); |
771ed689 CM |
699 | struct btrfs_key ins; |
700 | struct extent_map *em; | |
701 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
702 | int ret = 0; | |
703 | ||
704 | trans = btrfs_join_transaction(root, 1); | |
705 | BUG_ON(!trans); | |
706 | btrfs_set_trans_block_group(trans, inode); | |
707 | ||
42dc7bab | 708 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
709 | |
710 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
711 | num_bytes = max(blocksize, num_bytes); | |
712 | disk_num_bytes = num_bytes; | |
713 | ret = 0; | |
714 | ||
715 | if (start == 0) { | |
716 | /* lets try to make an inline extent */ | |
717 | ret = cow_file_range_inline(trans, root, inode, | |
718 | start, end, 0, NULL); | |
719 | if (ret == 0) { | |
720 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
721 | &BTRFS_I(inode)->io_tree, |
722 | start, end, NULL, | |
723 | EXTENT_CLEAR_UNLOCK_PAGE | | |
724 | EXTENT_CLEAR_UNLOCK | | |
725 | EXTENT_CLEAR_DELALLOC | | |
32c00aff | 726 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e CM |
727 | EXTENT_CLEAR_DIRTY | |
728 | EXTENT_SET_WRITEBACK | | |
729 | EXTENT_END_WRITEBACK); | |
771ed689 CM |
730 | *nr_written = *nr_written + |
731 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
732 | *page_started = 1; | |
733 | ret = 0; | |
734 | goto out; | |
735 | } | |
736 | } | |
737 | ||
738 | BUG_ON(disk_num_bytes > | |
739 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
740 | ||
b917b7c3 CM |
741 | |
742 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
743 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
744 | start, num_bytes); | |
745 | if (em) { | |
6346c939 JB |
746 | /* |
747 | * if block start isn't an actual block number then find the | |
748 | * first block in this inode and use that as a hint. If that | |
749 | * block is also bogus then just don't worry about it. | |
750 | */ | |
751 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
752 | free_extent_map(em); | |
753 | em = search_extent_mapping(em_tree, 0, 0); | |
754 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
755 | alloc_hint = em->block_start; | |
756 | if (em) | |
757 | free_extent_map(em); | |
758 | } else { | |
759 | alloc_hint = em->block_start; | |
760 | free_extent_map(em); | |
761 | } | |
b917b7c3 CM |
762 | } |
763 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
764 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
765 | ||
d397712b | 766 | while (disk_num_bytes > 0) { |
a791e35e CM |
767 | unsigned long op; |
768 | ||
c8b97818 | 769 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 770 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 771 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 772 | (u64)-1, &ins, 1); |
d397712b CM |
773 | BUG_ON(ret); |
774 | ||
e6dcd2dc CM |
775 | em = alloc_extent_map(GFP_NOFS); |
776 | em->start = start; | |
445a6944 | 777 | em->orig_start = em->start; |
771ed689 CM |
778 | ram_size = ins.offset; |
779 | em->len = ins.offset; | |
c8b97818 | 780 | |
e6dcd2dc | 781 | em->block_start = ins.objectid; |
c8b97818 | 782 | em->block_len = ins.offset; |
e6dcd2dc | 783 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 784 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 785 | |
d397712b | 786 | while (1) { |
890871be | 787 | write_lock(&em_tree->lock); |
e6dcd2dc | 788 | ret = add_extent_mapping(em_tree, em); |
890871be | 789 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
790 | if (ret != -EEXIST) { |
791 | free_extent_map(em); | |
792 | break; | |
793 | } | |
794 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 795 | start + ram_size - 1, 0); |
e6dcd2dc CM |
796 | } |
797 | ||
98d20f67 | 798 | cur_alloc_size = ins.offset; |
e6dcd2dc | 799 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 800 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 801 | BUG_ON(ret); |
c8b97818 | 802 | |
17d217fe YZ |
803 | if (root->root_key.objectid == |
804 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
805 | ret = btrfs_reloc_clone_csums(inode, start, | |
806 | cur_alloc_size); | |
807 | BUG_ON(ret); | |
808 | } | |
809 | ||
d397712b | 810 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 811 | break; |
d397712b | 812 | |
c8b97818 CM |
813 | /* we're not doing compressed IO, don't unlock the first |
814 | * page (which the caller expects to stay locked), don't | |
815 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
816 | * |
817 | * Do set the Private2 bit so we know this page was properly | |
818 | * setup for writepage | |
c8b97818 | 819 | */ |
a791e35e CM |
820 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
821 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
822 | EXTENT_SET_PRIVATE2; | |
823 | ||
c8b97818 CM |
824 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
825 | start, start + ram_size - 1, | |
a791e35e | 826 | locked_page, op); |
c8b97818 | 827 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
828 | num_bytes -= cur_alloc_size; |
829 | alloc_hint = ins.objectid + ins.offset; | |
830 | start += cur_alloc_size; | |
b888db2b | 831 | } |
b888db2b | 832 | out: |
771ed689 | 833 | ret = 0; |
b888db2b | 834 | btrfs_end_transaction(trans, root); |
c8b97818 | 835 | |
be20aa9d | 836 | return ret; |
771ed689 | 837 | } |
c8b97818 | 838 | |
771ed689 CM |
839 | /* |
840 | * work queue call back to started compression on a file and pages | |
841 | */ | |
842 | static noinline void async_cow_start(struct btrfs_work *work) | |
843 | { | |
844 | struct async_cow *async_cow; | |
845 | int num_added = 0; | |
846 | async_cow = container_of(work, struct async_cow, work); | |
847 | ||
848 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
849 | async_cow->start, async_cow->end, async_cow, | |
850 | &num_added); | |
851 | if (num_added == 0) | |
852 | async_cow->inode = NULL; | |
853 | } | |
854 | ||
855 | /* | |
856 | * work queue call back to submit previously compressed pages | |
857 | */ | |
858 | static noinline void async_cow_submit(struct btrfs_work *work) | |
859 | { | |
860 | struct async_cow *async_cow; | |
861 | struct btrfs_root *root; | |
862 | unsigned long nr_pages; | |
863 | ||
864 | async_cow = container_of(work, struct async_cow, work); | |
865 | ||
866 | root = async_cow->root; | |
867 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
868 | PAGE_CACHE_SHIFT; | |
869 | ||
870 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
871 | ||
872 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
873 | 5 * 1042 * 1024 && | |
874 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
875 | wake_up(&root->fs_info->async_submit_wait); | |
876 | ||
d397712b | 877 | if (async_cow->inode) |
771ed689 | 878 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 879 | } |
c8b97818 | 880 | |
771ed689 CM |
881 | static noinline void async_cow_free(struct btrfs_work *work) |
882 | { | |
883 | struct async_cow *async_cow; | |
884 | async_cow = container_of(work, struct async_cow, work); | |
885 | kfree(async_cow); | |
886 | } | |
887 | ||
888 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
889 | u64 start, u64 end, int *page_started, | |
890 | unsigned long *nr_written) | |
891 | { | |
892 | struct async_cow *async_cow; | |
893 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
894 | unsigned long nr_pages; | |
895 | u64 cur_end; | |
896 | int limit = 10 * 1024 * 1042; | |
897 | ||
a3429ab7 CM |
898 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
899 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 900 | while (start < end) { |
771ed689 CM |
901 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
902 | async_cow->inode = inode; | |
903 | async_cow->root = root; | |
904 | async_cow->locked_page = locked_page; | |
905 | async_cow->start = start; | |
906 | ||
6cbff00f | 907 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
908 | cur_end = end; |
909 | else | |
910 | cur_end = min(end, start + 512 * 1024 - 1); | |
911 | ||
912 | async_cow->end = cur_end; | |
913 | INIT_LIST_HEAD(&async_cow->extents); | |
914 | ||
915 | async_cow->work.func = async_cow_start; | |
916 | async_cow->work.ordered_func = async_cow_submit; | |
917 | async_cow->work.ordered_free = async_cow_free; | |
918 | async_cow->work.flags = 0; | |
919 | ||
771ed689 CM |
920 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
921 | PAGE_CACHE_SHIFT; | |
922 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
923 | ||
924 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
925 | &async_cow->work); | |
926 | ||
927 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
928 | wait_event(root->fs_info->async_submit_wait, | |
929 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
930 | limit)); | |
931 | } | |
932 | ||
d397712b | 933 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
934 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
935 | wait_event(root->fs_info->async_submit_wait, | |
936 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
937 | 0)); | |
938 | } | |
939 | ||
940 | *nr_written += nr_pages; | |
941 | start = cur_end + 1; | |
942 | } | |
943 | *page_started = 1; | |
944 | return 0; | |
be20aa9d CM |
945 | } |
946 | ||
d397712b | 947 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
948 | u64 bytenr, u64 num_bytes) |
949 | { | |
950 | int ret; | |
951 | struct btrfs_ordered_sum *sums; | |
952 | LIST_HEAD(list); | |
953 | ||
07d400a6 YZ |
954 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
955 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
956 | if (ret == 0 && list_empty(&list)) |
957 | return 0; | |
958 | ||
959 | while (!list_empty(&list)) { | |
960 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
961 | list_del(&sums->list); | |
962 | kfree(sums); | |
963 | } | |
964 | return 1; | |
965 | } | |
966 | ||
d352ac68 CM |
967 | /* |
968 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
969 | * of the extents that exist in the file, and COWs the file as required. | |
970 | * | |
971 | * If no cow copies or snapshots exist, we write directly to the existing | |
972 | * blocks on disk | |
973 | */ | |
7f366cfe CM |
974 | static noinline int run_delalloc_nocow(struct inode *inode, |
975 | struct page *locked_page, | |
771ed689 CM |
976 | u64 start, u64 end, int *page_started, int force, |
977 | unsigned long *nr_written) | |
be20aa9d | 978 | { |
be20aa9d | 979 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 980 | struct btrfs_trans_handle *trans; |
be20aa9d | 981 | struct extent_buffer *leaf; |
be20aa9d | 982 | struct btrfs_path *path; |
80ff3856 | 983 | struct btrfs_file_extent_item *fi; |
be20aa9d | 984 | struct btrfs_key found_key; |
80ff3856 YZ |
985 | u64 cow_start; |
986 | u64 cur_offset; | |
987 | u64 extent_end; | |
5d4f98a2 | 988 | u64 extent_offset; |
80ff3856 YZ |
989 | u64 disk_bytenr; |
990 | u64 num_bytes; | |
991 | int extent_type; | |
992 | int ret; | |
d899e052 | 993 | int type; |
80ff3856 YZ |
994 | int nocow; |
995 | int check_prev = 1; | |
be20aa9d CM |
996 | |
997 | path = btrfs_alloc_path(); | |
998 | BUG_ON(!path); | |
7ea394f1 YZ |
999 | trans = btrfs_join_transaction(root, 1); |
1000 | BUG_ON(!trans); | |
be20aa9d | 1001 | |
80ff3856 YZ |
1002 | cow_start = (u64)-1; |
1003 | cur_offset = start; | |
1004 | while (1) { | |
1005 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1006 | cur_offset, 0); | |
1007 | BUG_ON(ret < 0); | |
1008 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1009 | leaf = path->nodes[0]; | |
1010 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1011 | path->slots[0] - 1); | |
1012 | if (found_key.objectid == inode->i_ino && | |
1013 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1014 | path->slots[0]--; | |
1015 | } | |
1016 | check_prev = 0; | |
1017 | next_slot: | |
1018 | leaf = path->nodes[0]; | |
1019 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1020 | ret = btrfs_next_leaf(root, path); | |
1021 | if (ret < 0) | |
1022 | BUG_ON(1); | |
1023 | if (ret > 0) | |
1024 | break; | |
1025 | leaf = path->nodes[0]; | |
1026 | } | |
be20aa9d | 1027 | |
80ff3856 YZ |
1028 | nocow = 0; |
1029 | disk_bytenr = 0; | |
17d217fe | 1030 | num_bytes = 0; |
80ff3856 YZ |
1031 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1032 | ||
1033 | if (found_key.objectid > inode->i_ino || | |
1034 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1035 | found_key.offset > end) | |
1036 | break; | |
1037 | ||
1038 | if (found_key.offset > cur_offset) { | |
1039 | extent_end = found_key.offset; | |
e9061e21 | 1040 | extent_type = 0; |
80ff3856 YZ |
1041 | goto out_check; |
1042 | } | |
1043 | ||
1044 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1045 | struct btrfs_file_extent_item); | |
1046 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1047 | ||
d899e052 YZ |
1048 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1049 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1050 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1051 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1052 | extent_end = found_key.offset + |
1053 | btrfs_file_extent_num_bytes(leaf, fi); | |
1054 | if (extent_end <= start) { | |
1055 | path->slots[0]++; | |
1056 | goto next_slot; | |
1057 | } | |
17d217fe YZ |
1058 | if (disk_bytenr == 0) |
1059 | goto out_check; | |
80ff3856 YZ |
1060 | if (btrfs_file_extent_compression(leaf, fi) || |
1061 | btrfs_file_extent_encryption(leaf, fi) || | |
1062 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1063 | goto out_check; | |
d899e052 YZ |
1064 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1065 | goto out_check; | |
d2fb3437 | 1066 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1067 | goto out_check; |
17d217fe | 1068 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1069 | found_key.offset - |
1070 | extent_offset, disk_bytenr)) | |
17d217fe | 1071 | goto out_check; |
5d4f98a2 | 1072 | disk_bytenr += extent_offset; |
17d217fe YZ |
1073 | disk_bytenr += cur_offset - found_key.offset; |
1074 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1075 | /* | |
1076 | * force cow if csum exists in the range. | |
1077 | * this ensure that csum for a given extent are | |
1078 | * either valid or do not exist. | |
1079 | */ | |
1080 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1081 | goto out_check; | |
80ff3856 YZ |
1082 | nocow = 1; |
1083 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1084 | extent_end = found_key.offset + | |
1085 | btrfs_file_extent_inline_len(leaf, fi); | |
1086 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1087 | } else { | |
1088 | BUG_ON(1); | |
1089 | } | |
1090 | out_check: | |
1091 | if (extent_end <= start) { | |
1092 | path->slots[0]++; | |
1093 | goto next_slot; | |
1094 | } | |
1095 | if (!nocow) { | |
1096 | if (cow_start == (u64)-1) | |
1097 | cow_start = cur_offset; | |
1098 | cur_offset = extent_end; | |
1099 | if (cur_offset > end) | |
1100 | break; | |
1101 | path->slots[0]++; | |
1102 | goto next_slot; | |
7ea394f1 YZ |
1103 | } |
1104 | ||
1105 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1106 | if (cow_start != (u64)-1) { |
1107 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1108 | found_key.offset - 1, page_started, |
1109 | nr_written, 1); | |
80ff3856 YZ |
1110 | BUG_ON(ret); |
1111 | cow_start = (u64)-1; | |
7ea394f1 | 1112 | } |
80ff3856 | 1113 | |
d899e052 YZ |
1114 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1115 | struct extent_map *em; | |
1116 | struct extent_map_tree *em_tree; | |
1117 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1118 | em = alloc_extent_map(GFP_NOFS); | |
1119 | em->start = cur_offset; | |
445a6944 | 1120 | em->orig_start = em->start; |
d899e052 YZ |
1121 | em->len = num_bytes; |
1122 | em->block_len = num_bytes; | |
1123 | em->block_start = disk_bytenr; | |
1124 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1125 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1126 | while (1) { | |
890871be | 1127 | write_lock(&em_tree->lock); |
d899e052 | 1128 | ret = add_extent_mapping(em_tree, em); |
890871be | 1129 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1130 | if (ret != -EEXIST) { |
1131 | free_extent_map(em); | |
1132 | break; | |
1133 | } | |
1134 | btrfs_drop_extent_cache(inode, em->start, | |
1135 | em->start + em->len - 1, 0); | |
1136 | } | |
1137 | type = BTRFS_ORDERED_PREALLOC; | |
1138 | } else { | |
1139 | type = BTRFS_ORDERED_NOCOW; | |
1140 | } | |
80ff3856 YZ |
1141 | |
1142 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1143 | num_bytes, num_bytes, type); |
1144 | BUG_ON(ret); | |
771ed689 | 1145 | |
d899e052 | 1146 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1147 | cur_offset, cur_offset + num_bytes - 1, |
1148 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1149 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1150 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1151 | cur_offset = extent_end; |
1152 | if (cur_offset > end) | |
1153 | break; | |
be20aa9d | 1154 | } |
80ff3856 YZ |
1155 | btrfs_release_path(root, path); |
1156 | ||
1157 | if (cur_offset <= end && cow_start == (u64)-1) | |
1158 | cow_start = cur_offset; | |
1159 | if (cow_start != (u64)-1) { | |
1160 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1161 | page_started, nr_written, 1); |
80ff3856 YZ |
1162 | BUG_ON(ret); |
1163 | } | |
1164 | ||
1165 | ret = btrfs_end_transaction(trans, root); | |
1166 | BUG_ON(ret); | |
7ea394f1 | 1167 | btrfs_free_path(path); |
80ff3856 | 1168 | return 0; |
be20aa9d CM |
1169 | } |
1170 | ||
d352ac68 CM |
1171 | /* |
1172 | * extent_io.c call back to do delayed allocation processing | |
1173 | */ | |
c8b97818 | 1174 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1175 | u64 start, u64 end, int *page_started, |
1176 | unsigned long *nr_written) | |
be20aa9d | 1177 | { |
be20aa9d | 1178 | int ret; |
7f366cfe | 1179 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1180 | |
6cbff00f | 1181 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1182 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1183 | page_started, 1, nr_written); |
6cbff00f | 1184 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1185 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1186 | page_started, 0, nr_written); |
7f366cfe CM |
1187 | else if (!btrfs_test_opt(root, COMPRESS)) |
1188 | ret = cow_file_range(inode, locked_page, start, end, | |
1189 | page_started, nr_written, 1); | |
be20aa9d | 1190 | else |
771ed689 | 1191 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1192 | page_started, nr_written); |
b888db2b CM |
1193 | return ret; |
1194 | } | |
1195 | ||
9ed74f2d JB |
1196 | static int btrfs_split_extent_hook(struct inode *inode, |
1197 | struct extent_state *orig, u64 split) | |
1198 | { | |
1199 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1200 | u64 size; | |
1201 | ||
1202 | if (!(orig->state & EXTENT_DELALLOC)) | |
1203 | return 0; | |
1204 | ||
1205 | size = orig->end - orig->start + 1; | |
1206 | if (size > root->fs_info->max_extent) { | |
1207 | u64 num_extents; | |
1208 | u64 new_size; | |
1209 | ||
1210 | new_size = orig->end - split + 1; | |
1211 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1212 | root->fs_info->max_extent); | |
1213 | ||
1214 | /* | |
32c00aff JB |
1215 | * if we break a large extent up then leave oustanding_extents |
1216 | * be, since we've already accounted for the large extent. | |
9ed74f2d JB |
1217 | */ |
1218 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1219 | root->fs_info->max_extent) < num_extents) | |
1220 | return 0; | |
1221 | } | |
1222 | ||
32c00aff JB |
1223 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1224 | BTRFS_I(inode)->outstanding_extents++; | |
1225 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1226 | |
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1232 | * extents so we can keep track of new extents that are just merged onto old | |
1233 | * extents, such as when we are doing sequential writes, so we can properly | |
1234 | * account for the metadata space we'll need. | |
1235 | */ | |
1236 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1237 | struct extent_state *new, | |
1238 | struct extent_state *other) | |
1239 | { | |
1240 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1241 | u64 new_size, old_size; | |
1242 | u64 num_extents; | |
1243 | ||
1244 | /* not delalloc, ignore it */ | |
1245 | if (!(other->state & EXTENT_DELALLOC)) | |
1246 | return 0; | |
1247 | ||
1248 | old_size = other->end - other->start + 1; | |
1249 | if (new->start < other->start) | |
1250 | new_size = other->end - new->start + 1; | |
1251 | else | |
1252 | new_size = new->end - other->start + 1; | |
1253 | ||
1254 | /* we're not bigger than the max, unreserve the space and go */ | |
1255 | if (new_size <= root->fs_info->max_extent) { | |
32c00aff JB |
1256 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1257 | BTRFS_I(inode)->outstanding_extents--; | |
1258 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1259 | return 0; |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * If we grew by another max_extent, just return, we want to keep that | |
1264 | * reserved amount. | |
1265 | */ | |
1266 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1267 | root->fs_info->max_extent); | |
1268 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1269 | root->fs_info->max_extent) > num_extents) | |
1270 | return 0; | |
1271 | ||
32c00aff JB |
1272 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1273 | BTRFS_I(inode)->outstanding_extents--; | |
1274 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1275 | |
1276 | return 0; | |
1277 | } | |
1278 | ||
d352ac68 CM |
1279 | /* |
1280 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1281 | * bytes in this file, and to maintain the list of inodes that | |
1282 | * have pending delalloc work to be done. | |
1283 | */ | |
b2950863 | 1284 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1285 | unsigned long old, unsigned long bits) |
291d673e | 1286 | { |
9ed74f2d | 1287 | |
75eff68e CM |
1288 | /* |
1289 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1290 | * but in this case, we are only testeing for the DELALLOC | |
1291 | * bit, which is only set or cleared with irqs on | |
1292 | */ | |
b0c68f8b | 1293 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1294 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d | 1295 | |
32c00aff JB |
1296 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1297 | BTRFS_I(inode)->outstanding_extents++; | |
1298 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
6a63209f | 1299 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1300 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1301 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1302 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1303 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1304 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1305 | &root->fs_info->delalloc_inodes); | |
1306 | } | |
75eff68e | 1307 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1308 | } |
1309 | return 0; | |
1310 | } | |
1311 | ||
d352ac68 CM |
1312 | /* |
1313 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1314 | */ | |
9ed74f2d JB |
1315 | static int btrfs_clear_bit_hook(struct inode *inode, |
1316 | struct extent_state *state, unsigned long bits) | |
291d673e | 1317 | { |
75eff68e CM |
1318 | /* |
1319 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1320 | * but in this case, we are only testeing for the DELALLOC | |
1321 | * bit, which is only set or cleared with irqs on | |
1322 | */ | |
9ed74f2d | 1323 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1324 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1325 | |
32c00aff JB |
1326 | if (bits & EXTENT_DO_ACCOUNTING) { |
1327 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
1328 | BTRFS_I(inode)->outstanding_extents--; | |
1329 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
1330 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1331 | } | |
9ed74f2d | 1332 | |
75eff68e | 1333 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1334 | if (state->end - state->start + 1 > |
1335 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1336 | printk(KERN_INFO "btrfs warning: delalloc account " |
1337 | "%llu %llu\n", | |
9ed74f2d JB |
1338 | (unsigned long long) |
1339 | state->end - state->start + 1, | |
d397712b CM |
1340 | (unsigned long long) |
1341 | root->fs_info->delalloc_bytes); | |
6a63209f | 1342 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1343 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1344 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1345 | } else { |
6a63209f | 1346 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1347 | state->end - |
1348 | state->start + 1); | |
1349 | root->fs_info->delalloc_bytes -= state->end - | |
1350 | state->start + 1; | |
1351 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1352 | state->start + 1; | |
b0c68f8b | 1353 | } |
ea8c2819 CM |
1354 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1355 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1356 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1357 | } | |
75eff68e | 1358 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1359 | } |
1360 | return 0; | |
1361 | } | |
1362 | ||
d352ac68 CM |
1363 | /* |
1364 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1365 | * we don't create bios that span stripes or chunks | |
1366 | */ | |
239b14b3 | 1367 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1368 | size_t size, struct bio *bio, |
1369 | unsigned long bio_flags) | |
239b14b3 CM |
1370 | { |
1371 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1372 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1373 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1374 | u64 length = 0; |
1375 | u64 map_length; | |
239b14b3 CM |
1376 | int ret; |
1377 | ||
771ed689 CM |
1378 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1379 | return 0; | |
1380 | ||
f2d8d74d | 1381 | length = bio->bi_size; |
239b14b3 CM |
1382 | map_tree = &root->fs_info->mapping_tree; |
1383 | map_length = length; | |
cea9e445 | 1384 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1385 | &map_length, NULL, 0); |
cea9e445 | 1386 | |
d397712b | 1387 | if (map_length < length + size) |
239b14b3 | 1388 | return 1; |
239b14b3 CM |
1389 | return 0; |
1390 | } | |
1391 | ||
d352ac68 CM |
1392 | /* |
1393 | * in order to insert checksums into the metadata in large chunks, | |
1394 | * we wait until bio submission time. All the pages in the bio are | |
1395 | * checksummed and sums are attached onto the ordered extent record. | |
1396 | * | |
1397 | * At IO completion time the cums attached on the ordered extent record | |
1398 | * are inserted into the btree | |
1399 | */ | |
d397712b CM |
1400 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1401 | struct bio *bio, int mirror_num, | |
1402 | unsigned long bio_flags) | |
065631f6 | 1403 | { |
065631f6 | 1404 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1405 | int ret = 0; |
e015640f | 1406 | |
d20f7043 | 1407 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1408 | BUG_ON(ret); |
4a69a410 CM |
1409 | return 0; |
1410 | } | |
e015640f | 1411 | |
4a69a410 CM |
1412 | /* |
1413 | * in order to insert checksums into the metadata in large chunks, | |
1414 | * we wait until bio submission time. All the pages in the bio are | |
1415 | * checksummed and sums are attached onto the ordered extent record. | |
1416 | * | |
1417 | * At IO completion time the cums attached on the ordered extent record | |
1418 | * are inserted into the btree | |
1419 | */ | |
b2950863 | 1420 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1421 | int mirror_num, unsigned long bio_flags) |
1422 | { | |
1423 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1424 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1425 | } |
1426 | ||
d352ac68 | 1427 | /* |
cad321ad CM |
1428 | * extent_io.c submission hook. This does the right thing for csum calculation |
1429 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1430 | */ |
b2950863 | 1431 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1432 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1433 | { |
1434 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1435 | int ret = 0; | |
19b9bdb0 | 1436 | int skip_sum; |
44b8bd7e | 1437 | |
6cbff00f | 1438 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1439 | |
e6dcd2dc CM |
1440 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1441 | BUG_ON(ret); | |
065631f6 | 1442 | |
4d1b5fb4 | 1443 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1444 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1445 | return btrfs_submit_compressed_read(inode, bio, |
1446 | mirror_num, bio_flags); | |
d20f7043 CM |
1447 | } else if (!skip_sum) |
1448 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1449 | goto mapit; |
19b9bdb0 | 1450 | } else if (!skip_sum) { |
17d217fe YZ |
1451 | /* csum items have already been cloned */ |
1452 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1453 | goto mapit; | |
19b9bdb0 CM |
1454 | /* we're doing a write, do the async checksumming */ |
1455 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1456 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1457 | bio_flags, __btrfs_submit_bio_start, |
1458 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1459 | } |
1460 | ||
0b86a832 | 1461 | mapit: |
8b712842 | 1462 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1463 | } |
6885f308 | 1464 | |
d352ac68 CM |
1465 | /* |
1466 | * given a list of ordered sums record them in the inode. This happens | |
1467 | * at IO completion time based on sums calculated at bio submission time. | |
1468 | */ | |
ba1da2f4 | 1469 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1470 | struct inode *inode, u64 file_offset, |
1471 | struct list_head *list) | |
1472 | { | |
e6dcd2dc CM |
1473 | struct btrfs_ordered_sum *sum; |
1474 | ||
1475 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1476 | |
1477 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1478 | btrfs_csum_file_blocks(trans, |
1479 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1480 | } |
1481 | return 0; | |
1482 | } | |
1483 | ||
ea8c2819 CM |
1484 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1485 | { | |
d397712b | 1486 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1487 | WARN_ON(1); |
ea8c2819 CM |
1488 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1489 | GFP_NOFS); | |
1490 | } | |
1491 | ||
d352ac68 | 1492 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1493 | struct btrfs_writepage_fixup { |
1494 | struct page *page; | |
1495 | struct btrfs_work work; | |
1496 | }; | |
1497 | ||
b2950863 | 1498 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1499 | { |
1500 | struct btrfs_writepage_fixup *fixup; | |
1501 | struct btrfs_ordered_extent *ordered; | |
1502 | struct page *page; | |
1503 | struct inode *inode; | |
1504 | u64 page_start; | |
1505 | u64 page_end; | |
1506 | ||
1507 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1508 | page = fixup->page; | |
4a096752 | 1509 | again: |
247e743c CM |
1510 | lock_page(page); |
1511 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1512 | ClearPageChecked(page); | |
1513 | goto out_page; | |
1514 | } | |
1515 | ||
1516 | inode = page->mapping->host; | |
1517 | page_start = page_offset(page); | |
1518 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1519 | ||
1520 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1521 | |
1522 | /* already ordered? We're done */ | |
8b62b72b | 1523 | if (PagePrivate2(page)) |
247e743c | 1524 | goto out; |
4a096752 CM |
1525 | |
1526 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1527 | if (ordered) { | |
1528 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1529 | page_end, GFP_NOFS); | |
1530 | unlock_page(page); | |
1531 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1532 | goto again; | |
1533 | } | |
247e743c | 1534 | |
ea8c2819 | 1535 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1536 | ClearPageChecked(page); |
1537 | out: | |
1538 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1539 | out_page: | |
1540 | unlock_page(page); | |
1541 | page_cache_release(page); | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * There are a few paths in the higher layers of the kernel that directly | |
1546 | * set the page dirty bit without asking the filesystem if it is a | |
1547 | * good idea. This causes problems because we want to make sure COW | |
1548 | * properly happens and the data=ordered rules are followed. | |
1549 | * | |
c8b97818 | 1550 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1551 | * hasn't been properly setup for IO. We kick off an async process |
1552 | * to fix it up. The async helper will wait for ordered extents, set | |
1553 | * the delalloc bit and make it safe to write the page. | |
1554 | */ | |
b2950863 | 1555 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1556 | { |
1557 | struct inode *inode = page->mapping->host; | |
1558 | struct btrfs_writepage_fixup *fixup; | |
1559 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1560 | |
8b62b72b CM |
1561 | /* this page is properly in the ordered list */ |
1562 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1563 | return 0; |
1564 | ||
1565 | if (PageChecked(page)) | |
1566 | return -EAGAIN; | |
1567 | ||
1568 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1569 | if (!fixup) | |
1570 | return -EAGAIN; | |
f421950f | 1571 | |
247e743c CM |
1572 | SetPageChecked(page); |
1573 | page_cache_get(page); | |
1574 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1575 | fixup->page = page; | |
1576 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1577 | return -EAGAIN; | |
1578 | } | |
1579 | ||
d899e052 YZ |
1580 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1581 | struct inode *inode, u64 file_pos, | |
1582 | u64 disk_bytenr, u64 disk_num_bytes, | |
1583 | u64 num_bytes, u64 ram_bytes, | |
e980b50c | 1584 | u64 locked_end, |
d899e052 YZ |
1585 | u8 compression, u8 encryption, |
1586 | u16 other_encoding, int extent_type) | |
1587 | { | |
1588 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1589 | struct btrfs_file_extent_item *fi; | |
1590 | struct btrfs_path *path; | |
1591 | struct extent_buffer *leaf; | |
1592 | struct btrfs_key ins; | |
1593 | u64 hint; | |
1594 | int ret; | |
1595 | ||
1596 | path = btrfs_alloc_path(); | |
1597 | BUG_ON(!path); | |
1598 | ||
b9473439 | 1599 | path->leave_spinning = 1; |
a1ed835e CM |
1600 | |
1601 | /* | |
1602 | * we may be replacing one extent in the tree with another. | |
1603 | * The new extent is pinned in the extent map, and we don't want | |
1604 | * to drop it from the cache until it is completely in the btree. | |
1605 | * | |
1606 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1607 | * the caller is expected to unpin it and allow it to be merged | |
1608 | * with the others. | |
1609 | */ | |
d899e052 | 1610 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
e980b50c | 1611 | file_pos + num_bytes, locked_end, |
a1ed835e | 1612 | file_pos, &hint, 0); |
d899e052 YZ |
1613 | BUG_ON(ret); |
1614 | ||
1615 | ins.objectid = inode->i_ino; | |
1616 | ins.offset = file_pos; | |
1617 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1618 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1619 | BUG_ON(ret); | |
1620 | leaf = path->nodes[0]; | |
1621 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1622 | struct btrfs_file_extent_item); | |
1623 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1624 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1625 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1626 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1627 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1628 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1629 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1630 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1631 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1632 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1633 | |
1634 | btrfs_unlock_up_safe(path, 1); | |
1635 | btrfs_set_lock_blocking(leaf); | |
1636 | ||
d899e052 YZ |
1637 | btrfs_mark_buffer_dirty(leaf); |
1638 | ||
1639 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1640 | |
1641 | ins.objectid = disk_bytenr; | |
1642 | ins.offset = disk_num_bytes; | |
1643 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1644 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1645 | root->root_key.objectid, | |
1646 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1647 | BUG_ON(ret); |
d899e052 | 1648 | btrfs_free_path(path); |
b9473439 | 1649 | |
d899e052 YZ |
1650 | return 0; |
1651 | } | |
1652 | ||
5d13a98f CM |
1653 | /* |
1654 | * helper function for btrfs_finish_ordered_io, this | |
1655 | * just reads in some of the csum leaves to prime them into ram | |
1656 | * before we start the transaction. It limits the amount of btree | |
1657 | * reads required while inside the transaction. | |
1658 | */ | |
1659 | static noinline void reada_csum(struct btrfs_root *root, | |
1660 | struct btrfs_path *path, | |
1661 | struct btrfs_ordered_extent *ordered_extent) | |
1662 | { | |
1663 | struct btrfs_ordered_sum *sum; | |
1664 | u64 bytenr; | |
1665 | ||
1666 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1667 | list); | |
1668 | bytenr = sum->sums[0].bytenr; | |
1669 | ||
1670 | /* | |
1671 | * we don't care about the results, the point of this search is | |
1672 | * just to get the btree leaves into ram | |
1673 | */ | |
1674 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1675 | } | |
1676 | ||
d352ac68 CM |
1677 | /* as ordered data IO finishes, this gets called so we can finish |
1678 | * an ordered extent if the range of bytes in the file it covers are | |
1679 | * fully written. | |
1680 | */ | |
211f90e6 | 1681 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1682 | { |
e6dcd2dc CM |
1683 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1684 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1685 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1686 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1687 | struct btrfs_path *path; |
d899e052 | 1688 | int compressed = 0; |
e6dcd2dc CM |
1689 | int ret; |
1690 | ||
1691 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1692 | if (!ret) |
e6dcd2dc | 1693 | return 0; |
e6dcd2dc | 1694 | |
b7ec40d7 CM |
1695 | /* |
1696 | * before we join the transaction, try to do some of our IO. | |
1697 | * This will limit the amount of IO that we have to do with | |
1698 | * the transaction running. We're unlikely to need to do any | |
1699 | * IO if the file extents are new, the disk_i_size checks | |
1700 | * covers the most common case. | |
1701 | */ | |
1702 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1703 | path = btrfs_alloc_path(); | |
1704 | if (path) { | |
1705 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1706 | inode->i_ino, | |
1707 | start, 0); | |
5d13a98f CM |
1708 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1709 | start); | |
1710 | if (!list_empty(&ordered_extent->list)) { | |
1711 | btrfs_release_path(root, path); | |
1712 | reada_csum(root, path, ordered_extent); | |
1713 | } | |
b7ec40d7 CM |
1714 | btrfs_free_path(path); |
1715 | } | |
1716 | } | |
1717 | ||
f9295749 | 1718 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc | 1719 | |
5d13a98f CM |
1720 | if (!ordered_extent) |
1721 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1722 | BUG_ON(!ordered_extent); |
7ea394f1 YZ |
1723 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1724 | goto nocow; | |
e6dcd2dc CM |
1725 | |
1726 | lock_extent(io_tree, ordered_extent->file_offset, | |
1727 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1728 | GFP_NOFS); | |
1729 | ||
c8b97818 | 1730 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1731 | compressed = 1; |
1732 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1733 | BUG_ON(compressed); | |
1734 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1735 | ordered_extent->file_offset, | |
1736 | ordered_extent->file_offset + | |
1737 | ordered_extent->len); | |
1738 | BUG_ON(ret); | |
1739 | } else { | |
1740 | ret = insert_reserved_file_extent(trans, inode, | |
1741 | ordered_extent->file_offset, | |
1742 | ordered_extent->start, | |
1743 | ordered_extent->disk_len, | |
1744 | ordered_extent->len, | |
1745 | ordered_extent->len, | |
e980b50c CM |
1746 | ordered_extent->file_offset + |
1747 | ordered_extent->len, | |
d899e052 YZ |
1748 | compressed, 0, 0, |
1749 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1750 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1751 | ordered_extent->file_offset, | |
1752 | ordered_extent->len); | |
d899e052 YZ |
1753 | BUG_ON(ret); |
1754 | } | |
e6dcd2dc CM |
1755 | unlock_extent(io_tree, ordered_extent->file_offset, |
1756 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1757 | GFP_NOFS); | |
7ea394f1 | 1758 | nocow: |
e6dcd2dc CM |
1759 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1760 | &ordered_extent->list); | |
1761 | ||
34353029 | 1762 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1763 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1764 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1765 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1766 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1767 | |
e6dcd2dc CM |
1768 | /* once for us */ |
1769 | btrfs_put_ordered_extent(ordered_extent); | |
1770 | /* once for the tree */ | |
1771 | btrfs_put_ordered_extent(ordered_extent); | |
1772 | ||
e6dcd2dc CM |
1773 | btrfs_end_transaction(trans, root); |
1774 | return 0; | |
1775 | } | |
1776 | ||
b2950863 | 1777 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1778 | struct extent_state *state, int uptodate) |
1779 | { | |
8b62b72b | 1780 | ClearPagePrivate2(page); |
211f90e6 CM |
1781 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1782 | } | |
1783 | ||
d352ac68 CM |
1784 | /* |
1785 | * When IO fails, either with EIO or csum verification fails, we | |
1786 | * try other mirrors that might have a good copy of the data. This | |
1787 | * io_failure_record is used to record state as we go through all the | |
1788 | * mirrors. If another mirror has good data, the page is set up to date | |
1789 | * and things continue. If a good mirror can't be found, the original | |
1790 | * bio end_io callback is called to indicate things have failed. | |
1791 | */ | |
7e38326f CM |
1792 | struct io_failure_record { |
1793 | struct page *page; | |
1794 | u64 start; | |
1795 | u64 len; | |
1796 | u64 logical; | |
d20f7043 | 1797 | unsigned long bio_flags; |
7e38326f CM |
1798 | int last_mirror; |
1799 | }; | |
1800 | ||
b2950863 | 1801 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1802 | struct page *page, u64 start, u64 end, |
1803 | struct extent_state *state) | |
7e38326f CM |
1804 | { |
1805 | struct io_failure_record *failrec = NULL; | |
1806 | u64 private; | |
1807 | struct extent_map *em; | |
1808 | struct inode *inode = page->mapping->host; | |
1809 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1810 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1811 | struct bio *bio; |
1812 | int num_copies; | |
1813 | int ret; | |
1259ab75 | 1814 | int rw; |
7e38326f CM |
1815 | u64 logical; |
1816 | ||
1817 | ret = get_state_private(failure_tree, start, &private); | |
1818 | if (ret) { | |
7e38326f CM |
1819 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1820 | if (!failrec) | |
1821 | return -ENOMEM; | |
1822 | failrec->start = start; | |
1823 | failrec->len = end - start + 1; | |
1824 | failrec->last_mirror = 0; | |
d20f7043 | 1825 | failrec->bio_flags = 0; |
7e38326f | 1826 | |
890871be | 1827 | read_lock(&em_tree->lock); |
3b951516 CM |
1828 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1829 | if (em->start > start || em->start + em->len < start) { | |
1830 | free_extent_map(em); | |
1831 | em = NULL; | |
1832 | } | |
890871be | 1833 | read_unlock(&em_tree->lock); |
7e38326f CM |
1834 | |
1835 | if (!em || IS_ERR(em)) { | |
1836 | kfree(failrec); | |
1837 | return -EIO; | |
1838 | } | |
1839 | logical = start - em->start; | |
1840 | logical = em->block_start + logical; | |
d20f7043 CM |
1841 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1842 | logical = em->block_start; | |
1843 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1844 | } | |
7e38326f CM |
1845 | failrec->logical = logical; |
1846 | free_extent_map(em); | |
1847 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1848 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1849 | set_state_private(failure_tree, start, |
1850 | (u64)(unsigned long)failrec); | |
7e38326f | 1851 | } else { |
587f7704 | 1852 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1853 | } |
1854 | num_copies = btrfs_num_copies( | |
1855 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1856 | failrec->logical, failrec->len); | |
1857 | failrec->last_mirror++; | |
1858 | if (!state) { | |
cad321ad | 1859 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1860 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1861 | failrec->start, | |
1862 | EXTENT_LOCKED); | |
1863 | if (state && state->start != failrec->start) | |
1864 | state = NULL; | |
cad321ad | 1865 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1866 | } |
1867 | if (!state || failrec->last_mirror > num_copies) { | |
1868 | set_state_private(failure_tree, failrec->start, 0); | |
1869 | clear_extent_bits(failure_tree, failrec->start, | |
1870 | failrec->start + failrec->len - 1, | |
1871 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1872 | kfree(failrec); | |
1873 | return -EIO; | |
1874 | } | |
1875 | bio = bio_alloc(GFP_NOFS, 1); | |
1876 | bio->bi_private = state; | |
1877 | bio->bi_end_io = failed_bio->bi_end_io; | |
1878 | bio->bi_sector = failrec->logical >> 9; | |
1879 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1880 | bio->bi_size = 0; |
d20f7043 | 1881 | |
7e38326f | 1882 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1883 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1884 | rw = WRITE; | |
1885 | else | |
1886 | rw = READ; | |
1887 | ||
1888 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1889 | failrec->last_mirror, |
d20f7043 | 1890 | failrec->bio_flags); |
1259ab75 CM |
1891 | return 0; |
1892 | } | |
1893 | ||
d352ac68 CM |
1894 | /* |
1895 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1896 | * to see if we need to process or clean up an io_failure_record | |
1897 | */ | |
b2950863 | 1898 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1899 | { |
1900 | u64 private; | |
1901 | u64 private_failure; | |
1902 | struct io_failure_record *failure; | |
1903 | int ret; | |
1904 | ||
1905 | private = 0; | |
1906 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1907 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1908 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1909 | start, &private_failure); | |
1910 | if (ret == 0) { | |
1911 | failure = (struct io_failure_record *)(unsigned long) | |
1912 | private_failure; | |
1913 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1914 | failure->start, 0); | |
1915 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1916 | failure->start, | |
1917 | failure->start + failure->len - 1, | |
1918 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1919 | GFP_NOFS); | |
1920 | kfree(failure); | |
1921 | } | |
1922 | } | |
7e38326f CM |
1923 | return 0; |
1924 | } | |
1925 | ||
d352ac68 CM |
1926 | /* |
1927 | * when reads are done, we need to check csums to verify the data is correct | |
1928 | * if there's a match, we allow the bio to finish. If not, we go through | |
1929 | * the io_failure_record routines to find good copies | |
1930 | */ | |
b2950863 | 1931 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1932 | struct extent_state *state) |
07157aac | 1933 | { |
35ebb934 | 1934 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1935 | struct inode *inode = page->mapping->host; |
d1310b2e | 1936 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1937 | char *kaddr; |
aadfeb6e | 1938 | u64 private = ~(u32)0; |
07157aac | 1939 | int ret; |
ff79f819 CM |
1940 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1941 | u32 csum = ~(u32)0; | |
d1310b2e | 1942 | |
d20f7043 CM |
1943 | if (PageChecked(page)) { |
1944 | ClearPageChecked(page); | |
1945 | goto good; | |
1946 | } | |
6cbff00f CH |
1947 | |
1948 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1949 | return 0; |
1950 | ||
1951 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1952 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1953 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1954 | GFP_NOFS); | |
b6cda9bc | 1955 | return 0; |
17d217fe | 1956 | } |
d20f7043 | 1957 | |
c2e639f0 | 1958 | if (state && state->start == start) { |
70dec807 CM |
1959 | private = state->private; |
1960 | ret = 0; | |
1961 | } else { | |
1962 | ret = get_state_private(io_tree, start, &private); | |
1963 | } | |
9ab86c8e | 1964 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1965 | if (ret) |
07157aac | 1966 | goto zeroit; |
d397712b | 1967 | |
ff79f819 CM |
1968 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1969 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1970 | if (csum != private) |
07157aac | 1971 | goto zeroit; |
d397712b | 1972 | |
9ab86c8e | 1973 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1974 | good: |
7e38326f CM |
1975 | /* if the io failure tree for this inode is non-empty, |
1976 | * check to see if we've recovered from a failed IO | |
1977 | */ | |
1259ab75 | 1978 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1979 | return 0; |
1980 | ||
1981 | zeroit: | |
193f284d CM |
1982 | if (printk_ratelimit()) { |
1983 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1984 | "private %llu\n", page->mapping->host->i_ino, | |
1985 | (unsigned long long)start, csum, | |
1986 | (unsigned long long)private); | |
1987 | } | |
db94535d CM |
1988 | memset(kaddr + offset, 1, end - start + 1); |
1989 | flush_dcache_page(page); | |
9ab86c8e | 1990 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1991 | if (private == 0) |
1992 | return 0; | |
7e38326f | 1993 | return -EIO; |
07157aac | 1994 | } |
b888db2b | 1995 | |
7b128766 JB |
1996 | /* |
1997 | * This creates an orphan entry for the given inode in case something goes | |
1998 | * wrong in the middle of an unlink/truncate. | |
1999 | */ | |
2000 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2001 | { | |
2002 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2003 | int ret = 0; | |
2004 | ||
bcc63abb | 2005 | spin_lock(&root->list_lock); |
7b128766 JB |
2006 | |
2007 | /* already on the orphan list, we're good */ | |
2008 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2009 | spin_unlock(&root->list_lock); |
7b128766 JB |
2010 | return 0; |
2011 | } | |
2012 | ||
2013 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2014 | ||
bcc63abb | 2015 | spin_unlock(&root->list_lock); |
7b128766 JB |
2016 | |
2017 | /* | |
2018 | * insert an orphan item to track this unlinked/truncated file | |
2019 | */ | |
2020 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2021 | ||
2022 | return ret; | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2027 | * item for this particular inode. | |
2028 | */ | |
2029 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2030 | { | |
2031 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2032 | int ret = 0; | |
2033 | ||
bcc63abb | 2034 | spin_lock(&root->list_lock); |
7b128766 JB |
2035 | |
2036 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2037 | spin_unlock(&root->list_lock); |
7b128766 JB |
2038 | return 0; |
2039 | } | |
2040 | ||
2041 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2042 | if (!trans) { | |
bcc63abb | 2043 | spin_unlock(&root->list_lock); |
7b128766 JB |
2044 | return 0; |
2045 | } | |
2046 | ||
bcc63abb | 2047 | spin_unlock(&root->list_lock); |
7b128766 JB |
2048 | |
2049 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2050 | ||
2051 | return ret; | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * this cleans up any orphans that may be left on the list from the last use | |
2056 | * of this root. | |
2057 | */ | |
2058 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2059 | { | |
2060 | struct btrfs_path *path; | |
2061 | struct extent_buffer *leaf; | |
2062 | struct btrfs_item *item; | |
2063 | struct btrfs_key key, found_key; | |
2064 | struct btrfs_trans_handle *trans; | |
2065 | struct inode *inode; | |
2066 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2067 | ||
7b128766 JB |
2068 | path = btrfs_alloc_path(); |
2069 | if (!path) | |
2070 | return; | |
2071 | path->reada = -1; | |
2072 | ||
2073 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2074 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2075 | key.offset = (u64)-1; | |
2076 | ||
7b128766 JB |
2077 | |
2078 | while (1) { | |
2079 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2080 | if (ret < 0) { | |
2081 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2082 | "\n", ret); | |
2083 | break; | |
2084 | } | |
2085 | ||
2086 | /* | |
2087 | * if ret == 0 means we found what we were searching for, which | |
2088 | * is weird, but possible, so only screw with path if we didnt | |
2089 | * find the key and see if we have stuff that matches | |
2090 | */ | |
2091 | if (ret > 0) { | |
2092 | if (path->slots[0] == 0) | |
2093 | break; | |
2094 | path->slots[0]--; | |
2095 | } | |
2096 | ||
2097 | /* pull out the item */ | |
2098 | leaf = path->nodes[0]; | |
2099 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2100 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2101 | ||
2102 | /* make sure the item matches what we want */ | |
2103 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2104 | break; | |
2105 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2106 | break; | |
2107 | ||
2108 | /* release the path since we're done with it */ | |
2109 | btrfs_release_path(root, path); | |
2110 | ||
2111 | /* | |
2112 | * this is where we are basically btrfs_lookup, without the | |
2113 | * crossing root thing. we store the inode number in the | |
2114 | * offset of the orphan item. | |
2115 | */ | |
5d4f98a2 YZ |
2116 | found_key.objectid = found_key.offset; |
2117 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2118 | found_key.offset = 0; | |
2119 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2120 | if (IS_ERR(inode)) | |
7b128766 JB |
2121 | break; |
2122 | ||
7b128766 JB |
2123 | /* |
2124 | * add this inode to the orphan list so btrfs_orphan_del does | |
2125 | * the proper thing when we hit it | |
2126 | */ | |
bcc63abb | 2127 | spin_lock(&root->list_lock); |
7b128766 | 2128 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2129 | spin_unlock(&root->list_lock); |
7b128766 JB |
2130 | |
2131 | /* | |
2132 | * if this is a bad inode, means we actually succeeded in | |
2133 | * removing the inode, but not the orphan record, which means | |
2134 | * we need to manually delete the orphan since iput will just | |
2135 | * do a destroy_inode | |
2136 | */ | |
2137 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2138 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2139 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2140 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2141 | iput(inode); |
2142 | continue; | |
2143 | } | |
2144 | ||
2145 | /* if we have links, this was a truncate, lets do that */ | |
2146 | if (inode->i_nlink) { | |
2147 | nr_truncate++; | |
2148 | btrfs_truncate(inode); | |
2149 | } else { | |
2150 | nr_unlink++; | |
2151 | } | |
2152 | ||
2153 | /* this will do delete_inode and everything for us */ | |
2154 | iput(inode); | |
2155 | } | |
2156 | ||
2157 | if (nr_unlink) | |
2158 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2159 | if (nr_truncate) | |
2160 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2161 | ||
2162 | btrfs_free_path(path); | |
7b128766 JB |
2163 | } |
2164 | ||
46a53cca CM |
2165 | /* |
2166 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2167 | * don't find any xattrs, we know there can't be any acls. | |
2168 | * | |
2169 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2170 | */ | |
2171 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2172 | int slot, u64 objectid) | |
2173 | { | |
2174 | u32 nritems = btrfs_header_nritems(leaf); | |
2175 | struct btrfs_key found_key; | |
2176 | int scanned = 0; | |
2177 | ||
2178 | slot++; | |
2179 | while (slot < nritems) { | |
2180 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2181 | ||
2182 | /* we found a different objectid, there must not be acls */ | |
2183 | if (found_key.objectid != objectid) | |
2184 | return 0; | |
2185 | ||
2186 | /* we found an xattr, assume we've got an acl */ | |
2187 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2188 | return 1; | |
2189 | ||
2190 | /* | |
2191 | * we found a key greater than an xattr key, there can't | |
2192 | * be any acls later on | |
2193 | */ | |
2194 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2195 | return 0; | |
2196 | ||
2197 | slot++; | |
2198 | scanned++; | |
2199 | ||
2200 | /* | |
2201 | * it goes inode, inode backrefs, xattrs, extents, | |
2202 | * so if there are a ton of hard links to an inode there can | |
2203 | * be a lot of backrefs. Don't waste time searching too hard, | |
2204 | * this is just an optimization | |
2205 | */ | |
2206 | if (scanned >= 8) | |
2207 | break; | |
2208 | } | |
2209 | /* we hit the end of the leaf before we found an xattr or | |
2210 | * something larger than an xattr. We have to assume the inode | |
2211 | * has acls | |
2212 | */ | |
2213 | return 1; | |
2214 | } | |
2215 | ||
d352ac68 CM |
2216 | /* |
2217 | * read an inode from the btree into the in-memory inode | |
2218 | */ | |
5d4f98a2 | 2219 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2220 | { |
2221 | struct btrfs_path *path; | |
5f39d397 | 2222 | struct extent_buffer *leaf; |
39279cc3 | 2223 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2224 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2225 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2226 | struct btrfs_key location; | |
46a53cca | 2227 | int maybe_acls; |
39279cc3 | 2228 | u64 alloc_group_block; |
618e21d5 | 2229 | u32 rdev; |
39279cc3 CM |
2230 | int ret; |
2231 | ||
2232 | path = btrfs_alloc_path(); | |
2233 | BUG_ON(!path); | |
39279cc3 | 2234 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2235 | |
39279cc3 | 2236 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2237 | if (ret) |
39279cc3 | 2238 | goto make_bad; |
39279cc3 | 2239 | |
5f39d397 CM |
2240 | leaf = path->nodes[0]; |
2241 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2242 | struct btrfs_inode_item); | |
2243 | ||
2244 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2245 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2246 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2247 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2248 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2249 | |
2250 | tspec = btrfs_inode_atime(inode_item); | |
2251 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2252 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2253 | ||
2254 | tspec = btrfs_inode_mtime(inode_item); | |
2255 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2256 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2257 | ||
2258 | tspec = btrfs_inode_ctime(inode_item); | |
2259 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2260 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2261 | ||
a76a3cd4 | 2262 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2263 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2264 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2265 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2266 | inode->i_rdev = 0; |
5f39d397 CM |
2267 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2268 | ||
aec7477b | 2269 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2270 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2271 | |
5f39d397 | 2272 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2273 | |
46a53cca CM |
2274 | /* |
2275 | * try to precache a NULL acl entry for files that don't have | |
2276 | * any xattrs or acls | |
2277 | */ | |
2278 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2279 | if (!maybe_acls) |
2280 | cache_no_acl(inode); | |
46a53cca | 2281 | |
d2fb3437 YZ |
2282 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2283 | alloc_group_block, 0); | |
39279cc3 CM |
2284 | btrfs_free_path(path); |
2285 | inode_item = NULL; | |
2286 | ||
39279cc3 | 2287 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2288 | case S_IFREG: |
2289 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2290 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2291 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2292 | inode->i_fop = &btrfs_file_operations; |
2293 | inode->i_op = &btrfs_file_inode_operations; | |
2294 | break; | |
2295 | case S_IFDIR: | |
2296 | inode->i_fop = &btrfs_dir_file_operations; | |
2297 | if (root == root->fs_info->tree_root) | |
2298 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2299 | else | |
2300 | inode->i_op = &btrfs_dir_inode_operations; | |
2301 | break; | |
2302 | case S_IFLNK: | |
2303 | inode->i_op = &btrfs_symlink_inode_operations; | |
2304 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2305 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2306 | break; |
618e21d5 | 2307 | default: |
0279b4cd | 2308 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2309 | init_special_inode(inode, inode->i_mode, rdev); |
2310 | break; | |
39279cc3 | 2311 | } |
6cbff00f CH |
2312 | |
2313 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2314 | return; |
2315 | ||
2316 | make_bad: | |
39279cc3 | 2317 | btrfs_free_path(path); |
39279cc3 CM |
2318 | make_bad_inode(inode); |
2319 | } | |
2320 | ||
d352ac68 CM |
2321 | /* |
2322 | * given a leaf and an inode, copy the inode fields into the leaf | |
2323 | */ | |
e02119d5 CM |
2324 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2325 | struct extent_buffer *leaf, | |
5f39d397 | 2326 | struct btrfs_inode_item *item, |
39279cc3 CM |
2327 | struct inode *inode) |
2328 | { | |
5f39d397 CM |
2329 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2330 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2331 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2332 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2333 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2334 | ||
2335 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2336 | inode->i_atime.tv_sec); | |
2337 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2338 | inode->i_atime.tv_nsec); | |
2339 | ||
2340 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2341 | inode->i_mtime.tv_sec); | |
2342 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2343 | inode->i_mtime.tv_nsec); | |
2344 | ||
2345 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2346 | inode->i_ctime.tv_sec); | |
2347 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2348 | inode->i_ctime.tv_nsec); | |
2349 | ||
a76a3cd4 | 2350 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2351 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2352 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2353 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2354 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2355 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2356 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2357 | } |
2358 | ||
d352ac68 CM |
2359 | /* |
2360 | * copy everything in the in-memory inode into the btree. | |
2361 | */ | |
d397712b CM |
2362 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2363 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2364 | { |
2365 | struct btrfs_inode_item *inode_item; | |
2366 | struct btrfs_path *path; | |
5f39d397 | 2367 | struct extent_buffer *leaf; |
39279cc3 CM |
2368 | int ret; |
2369 | ||
2370 | path = btrfs_alloc_path(); | |
2371 | BUG_ON(!path); | |
b9473439 | 2372 | path->leave_spinning = 1; |
39279cc3 CM |
2373 | ret = btrfs_lookup_inode(trans, root, path, |
2374 | &BTRFS_I(inode)->location, 1); | |
2375 | if (ret) { | |
2376 | if (ret > 0) | |
2377 | ret = -ENOENT; | |
2378 | goto failed; | |
2379 | } | |
2380 | ||
b4ce94de | 2381 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2382 | leaf = path->nodes[0]; |
2383 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2384 | struct btrfs_inode_item); |
2385 | ||
e02119d5 | 2386 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2387 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2388 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2389 | ret = 0; |
2390 | failed: | |
39279cc3 CM |
2391 | btrfs_free_path(path); |
2392 | return ret; | |
2393 | } | |
2394 | ||
2395 | ||
d352ac68 CM |
2396 | /* |
2397 | * unlink helper that gets used here in inode.c and in the tree logging | |
2398 | * recovery code. It remove a link in a directory with a given name, and | |
2399 | * also drops the back refs in the inode to the directory | |
2400 | */ | |
e02119d5 CM |
2401 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2402 | struct btrfs_root *root, | |
2403 | struct inode *dir, struct inode *inode, | |
2404 | const char *name, int name_len) | |
39279cc3 CM |
2405 | { |
2406 | struct btrfs_path *path; | |
39279cc3 | 2407 | int ret = 0; |
5f39d397 | 2408 | struct extent_buffer *leaf; |
39279cc3 | 2409 | struct btrfs_dir_item *di; |
5f39d397 | 2410 | struct btrfs_key key; |
aec7477b | 2411 | u64 index; |
39279cc3 CM |
2412 | |
2413 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2414 | if (!path) { |
2415 | ret = -ENOMEM; | |
2416 | goto err; | |
2417 | } | |
2418 | ||
b9473439 | 2419 | path->leave_spinning = 1; |
39279cc3 CM |
2420 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2421 | name, name_len, -1); | |
2422 | if (IS_ERR(di)) { | |
2423 | ret = PTR_ERR(di); | |
2424 | goto err; | |
2425 | } | |
2426 | if (!di) { | |
2427 | ret = -ENOENT; | |
2428 | goto err; | |
2429 | } | |
5f39d397 CM |
2430 | leaf = path->nodes[0]; |
2431 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2432 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2433 | if (ret) |
2434 | goto err; | |
39279cc3 CM |
2435 | btrfs_release_path(root, path); |
2436 | ||
aec7477b | 2437 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2438 | inode->i_ino, |
2439 | dir->i_ino, &index); | |
aec7477b | 2440 | if (ret) { |
d397712b | 2441 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2442 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2443 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2444 | goto err; |
2445 | } | |
2446 | ||
39279cc3 | 2447 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2448 | index, name, name_len, -1); |
39279cc3 CM |
2449 | if (IS_ERR(di)) { |
2450 | ret = PTR_ERR(di); | |
2451 | goto err; | |
2452 | } | |
2453 | if (!di) { | |
2454 | ret = -ENOENT; | |
2455 | goto err; | |
2456 | } | |
2457 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2458 | btrfs_release_path(root, path); |
39279cc3 | 2459 | |
e02119d5 CM |
2460 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2461 | inode, dir->i_ino); | |
49eb7e46 | 2462 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2463 | |
2464 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2465 | dir, index); | |
2466 | BUG_ON(ret); | |
39279cc3 CM |
2467 | err: |
2468 | btrfs_free_path(path); | |
e02119d5 CM |
2469 | if (ret) |
2470 | goto out; | |
2471 | ||
2472 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2473 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2474 | btrfs_update_inode(trans, root, dir); | |
2475 | btrfs_drop_nlink(inode); | |
2476 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2477 | out: |
39279cc3 CM |
2478 | return ret; |
2479 | } | |
2480 | ||
2481 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2482 | { | |
2483 | struct btrfs_root *root; | |
2484 | struct btrfs_trans_handle *trans; | |
7b128766 | 2485 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2486 | int ret; |
1832a6d5 | 2487 | unsigned long nr = 0; |
39279cc3 CM |
2488 | |
2489 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2490 | |
5df6a9f6 JB |
2491 | /* |
2492 | * 5 items for unlink inode | |
2493 | * 1 for orphan | |
2494 | */ | |
2495 | ret = btrfs_reserve_metadata_space(root, 6); | |
2496 | if (ret) | |
2497 | return ret; | |
2498 | ||
39279cc3 | 2499 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2500 | if (IS_ERR(trans)) { |
2501 | btrfs_unreserve_metadata_space(root, 6); | |
2502 | return PTR_ERR(trans); | |
2503 | } | |
5f39d397 | 2504 | |
39279cc3 | 2505 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2506 | |
2507 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2508 | ||
e02119d5 CM |
2509 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2510 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2511 | |
2512 | if (inode->i_nlink == 0) | |
2513 | ret = btrfs_orphan_add(trans, inode); | |
2514 | ||
d3c2fdcf | 2515 | nr = trans->blocks_used; |
5f39d397 | 2516 | |
89ce8a63 | 2517 | btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2518 | btrfs_unreserve_metadata_space(root, 6); |
d3c2fdcf | 2519 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2520 | return ret; |
2521 | } | |
2522 | ||
4df27c4d YZ |
2523 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2524 | struct btrfs_root *root, | |
2525 | struct inode *dir, u64 objectid, | |
2526 | const char *name, int name_len) | |
2527 | { | |
2528 | struct btrfs_path *path; | |
2529 | struct extent_buffer *leaf; | |
2530 | struct btrfs_dir_item *di; | |
2531 | struct btrfs_key key; | |
2532 | u64 index; | |
2533 | int ret; | |
2534 | ||
2535 | path = btrfs_alloc_path(); | |
2536 | if (!path) | |
2537 | return -ENOMEM; | |
2538 | ||
2539 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2540 | name, name_len, -1); | |
2541 | BUG_ON(!di || IS_ERR(di)); | |
2542 | ||
2543 | leaf = path->nodes[0]; | |
2544 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2545 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2546 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2547 | BUG_ON(ret); | |
2548 | btrfs_release_path(root, path); | |
2549 | ||
2550 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2551 | objectid, root->root_key.objectid, | |
2552 | dir->i_ino, &index, name, name_len); | |
2553 | if (ret < 0) { | |
2554 | BUG_ON(ret != -ENOENT); | |
2555 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2556 | name, name_len); | |
2557 | BUG_ON(!di || IS_ERR(di)); | |
2558 | ||
2559 | leaf = path->nodes[0]; | |
2560 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2561 | btrfs_release_path(root, path); | |
2562 | index = key.offset; | |
2563 | } | |
2564 | ||
2565 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2566 | index, name, name_len, -1); | |
2567 | BUG_ON(!di || IS_ERR(di)); | |
2568 | ||
2569 | leaf = path->nodes[0]; | |
2570 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2571 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2572 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2573 | BUG_ON(ret); | |
2574 | btrfs_release_path(root, path); | |
2575 | ||
2576 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2577 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2578 | ret = btrfs_update_inode(trans, root, dir); | |
2579 | BUG_ON(ret); | |
2580 | dir->i_sb->s_dirt = 1; | |
2581 | ||
2582 | btrfs_free_path(path); | |
2583 | return 0; | |
2584 | } | |
2585 | ||
39279cc3 CM |
2586 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2587 | { | |
2588 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2589 | int err = 0; |
39279cc3 CM |
2590 | int ret; |
2591 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2592 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2593 | unsigned long nr = 0; |
39279cc3 | 2594 | |
3394e160 | 2595 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2596 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2597 | return -ENOTEMPTY; |
2598 | ||
5df6a9f6 JB |
2599 | ret = btrfs_reserve_metadata_space(root, 5); |
2600 | if (ret) | |
2601 | return ret; | |
2602 | ||
39279cc3 | 2603 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2604 | if (IS_ERR(trans)) { |
2605 | btrfs_unreserve_metadata_space(root, 5); | |
2606 | return PTR_ERR(trans); | |
2607 | } | |
2608 | ||
39279cc3 | 2609 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 2610 | |
4df27c4d YZ |
2611 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2612 | err = btrfs_unlink_subvol(trans, root, dir, | |
2613 | BTRFS_I(inode)->location.objectid, | |
2614 | dentry->d_name.name, | |
2615 | dentry->d_name.len); | |
2616 | goto out; | |
2617 | } | |
2618 | ||
7b128766 JB |
2619 | err = btrfs_orphan_add(trans, inode); |
2620 | if (err) | |
4df27c4d | 2621 | goto out; |
7b128766 | 2622 | |
39279cc3 | 2623 | /* now the directory is empty */ |
e02119d5 CM |
2624 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2625 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2626 | if (!err) |
dbe674a9 | 2627 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2628 | out: |
d3c2fdcf | 2629 | nr = trans->blocks_used; |
89ce8a63 | 2630 | ret = btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2631 | btrfs_unreserve_metadata_space(root, 5); |
d3c2fdcf | 2632 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2633 | |
39279cc3 CM |
2634 | if (ret && !err) |
2635 | err = ret; | |
2636 | return err; | |
2637 | } | |
2638 | ||
d20f7043 | 2639 | #if 0 |
323ac95b CM |
2640 | /* |
2641 | * when truncating bytes in a file, it is possible to avoid reading | |
2642 | * the leaves that contain only checksum items. This can be the | |
2643 | * majority of the IO required to delete a large file, but it must | |
2644 | * be done carefully. | |
2645 | * | |
2646 | * The keys in the level just above the leaves are checked to make sure | |
2647 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2648 | * after the new size. | |
2649 | * | |
2650 | * Then the key for the next leaf is checked to make sure it also has | |
2651 | * a checksum item for the same file. If it does, we know our target leaf | |
2652 | * contains only checksum items, and it can be safely freed without reading | |
2653 | * it. | |
2654 | * | |
2655 | * This is just an optimization targeted at large files. It may do | |
2656 | * nothing. It will return 0 unless things went badly. | |
2657 | */ | |
2658 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2659 | struct btrfs_root *root, | |
2660 | struct btrfs_path *path, | |
2661 | struct inode *inode, u64 new_size) | |
2662 | { | |
2663 | struct btrfs_key key; | |
2664 | int ret; | |
2665 | int nritems; | |
2666 | struct btrfs_key found_key; | |
2667 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2668 | struct btrfs_leaf_ref *ref; |
2669 | u64 leaf_gen; | |
2670 | u64 leaf_start; | |
323ac95b CM |
2671 | |
2672 | path->lowest_level = 1; | |
2673 | key.objectid = inode->i_ino; | |
2674 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2675 | key.offset = new_size; | |
2676 | again: | |
2677 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2678 | if (ret < 0) | |
2679 | goto out; | |
2680 | ||
2681 | if (path->nodes[1] == NULL) { | |
2682 | ret = 0; | |
2683 | goto out; | |
2684 | } | |
2685 | ret = 0; | |
2686 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2687 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2688 | ||
2689 | if (!nritems) | |
2690 | goto out; | |
2691 | ||
2692 | if (path->slots[1] >= nritems) | |
2693 | goto next_node; | |
2694 | ||
2695 | /* did we find a key greater than anything we want to delete? */ | |
2696 | if (found_key.objectid > inode->i_ino || | |
2697 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2698 | goto out; | |
2699 | ||
2700 | /* we check the next key in the node to make sure the leave contains | |
2701 | * only checksum items. This comparison doesn't work if our | |
2702 | * leaf is the last one in the node | |
2703 | */ | |
2704 | if (path->slots[1] + 1 >= nritems) { | |
2705 | next_node: | |
2706 | /* search forward from the last key in the node, this | |
2707 | * will bring us into the next node in the tree | |
2708 | */ | |
2709 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2710 | ||
2711 | /* unlikely, but we inc below, so check to be safe */ | |
2712 | if (found_key.offset == (u64)-1) | |
2713 | goto out; | |
2714 | ||
2715 | /* search_forward needs a path with locks held, do the | |
2716 | * search again for the original key. It is possible | |
2717 | * this will race with a balance and return a path that | |
2718 | * we could modify, but this drop is just an optimization | |
2719 | * and is allowed to miss some leaves. | |
2720 | */ | |
2721 | btrfs_release_path(root, path); | |
2722 | found_key.offset++; | |
2723 | ||
2724 | /* setup a max key for search_forward */ | |
2725 | other_key.offset = (u64)-1; | |
2726 | other_key.type = key.type; | |
2727 | other_key.objectid = key.objectid; | |
2728 | ||
2729 | path->keep_locks = 1; | |
2730 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2731 | path, 0, 0); | |
2732 | path->keep_locks = 0; | |
2733 | if (ret || found_key.objectid != key.objectid || | |
2734 | found_key.type != key.type) { | |
2735 | ret = 0; | |
2736 | goto out; | |
2737 | } | |
2738 | ||
2739 | key.offset = found_key.offset; | |
2740 | btrfs_release_path(root, path); | |
2741 | cond_resched(); | |
2742 | goto again; | |
2743 | } | |
2744 | ||
2745 | /* we know there's one more slot after us in the tree, | |
2746 | * read that key so we can verify it is also a checksum item | |
2747 | */ | |
2748 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2749 | ||
2750 | if (found_key.objectid < inode->i_ino) | |
2751 | goto next_key; | |
2752 | ||
2753 | if (found_key.type != key.type || found_key.offset < new_size) | |
2754 | goto next_key; | |
2755 | ||
2756 | /* | |
2757 | * if the key for the next leaf isn't a csum key from this objectid, | |
2758 | * we can't be sure there aren't good items inside this leaf. | |
2759 | * Bail out | |
2760 | */ | |
2761 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2762 | goto out; | |
2763 | ||
5b84e8d6 YZ |
2764 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2765 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2766 | /* |
2767 | * it is safe to delete this leaf, it contains only | |
2768 | * csum items from this inode at an offset >= new_size | |
2769 | */ | |
5b84e8d6 | 2770 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2771 | BUG_ON(ret); |
2772 | ||
5b84e8d6 YZ |
2773 | if (root->ref_cows && leaf_gen < trans->transid) { |
2774 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2775 | if (ref) { | |
2776 | ref->root_gen = root->root_key.offset; | |
2777 | ref->bytenr = leaf_start; | |
2778 | ref->owner = 0; | |
2779 | ref->generation = leaf_gen; | |
2780 | ref->nritems = 0; | |
2781 | ||
bd56b302 CM |
2782 | btrfs_sort_leaf_ref(ref); |
2783 | ||
5b84e8d6 YZ |
2784 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2785 | WARN_ON(ret); | |
2786 | btrfs_free_leaf_ref(root, ref); | |
2787 | } else { | |
2788 | WARN_ON(1); | |
2789 | } | |
2790 | } | |
323ac95b CM |
2791 | next_key: |
2792 | btrfs_release_path(root, path); | |
2793 | ||
2794 | if (other_key.objectid == inode->i_ino && | |
2795 | other_key.type == key.type && other_key.offset > key.offset) { | |
2796 | key.offset = other_key.offset; | |
2797 | cond_resched(); | |
2798 | goto again; | |
2799 | } | |
2800 | ret = 0; | |
2801 | out: | |
2802 | /* fixup any changes we've made to the path */ | |
2803 | path->lowest_level = 0; | |
2804 | path->keep_locks = 0; | |
2805 | btrfs_release_path(root, path); | |
2806 | return ret; | |
2807 | } | |
2808 | ||
d20f7043 CM |
2809 | #endif |
2810 | ||
39279cc3 CM |
2811 | /* |
2812 | * this can truncate away extent items, csum items and directory items. | |
2813 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2814 | * any higher than new_size |
39279cc3 CM |
2815 | * |
2816 | * csum items that cross the new i_size are truncated to the new size | |
2817 | * as well. | |
7b128766 JB |
2818 | * |
2819 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2820 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2821 | */ |
e02119d5 CM |
2822 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2823 | struct btrfs_root *root, | |
2824 | struct inode *inode, | |
2825 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2826 | { |
2827 | int ret; | |
2828 | struct btrfs_path *path; | |
2829 | struct btrfs_key key; | |
5f39d397 | 2830 | struct btrfs_key found_key; |
06d9a8d7 | 2831 | u32 found_type = (u8)-1; |
5f39d397 | 2832 | struct extent_buffer *leaf; |
39279cc3 CM |
2833 | struct btrfs_file_extent_item *fi; |
2834 | u64 extent_start = 0; | |
db94535d | 2835 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2836 | u64 extent_offset = 0; |
39279cc3 CM |
2837 | u64 item_end = 0; |
2838 | int found_extent; | |
2839 | int del_item; | |
85e21bac CM |
2840 | int pending_del_nr = 0; |
2841 | int pending_del_slot = 0; | |
179e29e4 | 2842 | int extent_type = -1; |
771ed689 | 2843 | int encoding; |
3b951516 | 2844 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2845 | |
e02119d5 | 2846 | if (root->ref_cows) |
5b21f2ed | 2847 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2848 | path = btrfs_alloc_path(); |
2849 | BUG_ON(!path); | |
33c17ad5 | 2850 | path->reada = -1; |
5f39d397 | 2851 | |
39279cc3 CM |
2852 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2853 | key.objectid = inode->i_ino; | |
2854 | key.offset = (u64)-1; | |
5f39d397 CM |
2855 | key.type = (u8)-1; |
2856 | ||
85e21bac | 2857 | search_again: |
b9473439 | 2858 | path->leave_spinning = 1; |
85e21bac | 2859 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2860 | if (ret < 0) |
85e21bac | 2861 | goto error; |
d397712b | 2862 | |
85e21bac | 2863 | if (ret > 0) { |
e02119d5 CM |
2864 | /* there are no items in the tree for us to truncate, we're |
2865 | * done | |
2866 | */ | |
2867 | if (path->slots[0] == 0) { | |
2868 | ret = 0; | |
2869 | goto error; | |
2870 | } | |
85e21bac CM |
2871 | path->slots[0]--; |
2872 | } | |
2873 | ||
d397712b | 2874 | while (1) { |
39279cc3 | 2875 | fi = NULL; |
5f39d397 CM |
2876 | leaf = path->nodes[0]; |
2877 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2878 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2879 | encoding = 0; |
39279cc3 | 2880 | |
5f39d397 | 2881 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2882 | break; |
5f39d397 | 2883 | |
85e21bac | 2884 | if (found_type < min_type) |
39279cc3 CM |
2885 | break; |
2886 | ||
5f39d397 | 2887 | item_end = found_key.offset; |
39279cc3 | 2888 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2889 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2890 | struct btrfs_file_extent_item); |
179e29e4 | 2891 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2892 | encoding = btrfs_file_extent_compression(leaf, fi); |
2893 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2894 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2895 | ||
179e29e4 | 2896 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2897 | item_end += |
db94535d | 2898 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2899 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2900 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2901 | fi); |
39279cc3 | 2902 | } |
008630c1 | 2903 | item_end--; |
39279cc3 | 2904 | } |
e02119d5 | 2905 | if (item_end < new_size) { |
d397712b | 2906 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2907 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2908 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2909 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2910 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2911 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2912 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2913 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2914 | else if (found_type) |
b888db2b | 2915 | found_type--; |
d397712b | 2916 | else |
b888db2b | 2917 | break; |
a61721d5 | 2918 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2919 | goto next; |
39279cc3 | 2920 | } |
e02119d5 | 2921 | if (found_key.offset >= new_size) |
39279cc3 CM |
2922 | del_item = 1; |
2923 | else | |
2924 | del_item = 0; | |
2925 | found_extent = 0; | |
2926 | ||
2927 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2928 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2929 | goto delete; | |
2930 | ||
2931 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2932 | u64 num_dec; |
db94535d | 2933 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2934 | if (!del_item && !encoding) { |
db94535d CM |
2935 | u64 orig_num_bytes = |
2936 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2937 | extent_num_bytes = new_size - |
5f39d397 | 2938 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2939 | extent_num_bytes = extent_num_bytes & |
2940 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2941 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2942 | extent_num_bytes); | |
2943 | num_dec = (orig_num_bytes - | |
9069218d | 2944 | extent_num_bytes); |
e02119d5 | 2945 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2946 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2947 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2948 | } else { |
db94535d CM |
2949 | extent_num_bytes = |
2950 | btrfs_file_extent_disk_num_bytes(leaf, | |
2951 | fi); | |
5d4f98a2 YZ |
2952 | extent_offset = found_key.offset - |
2953 | btrfs_file_extent_offset(leaf, fi); | |
2954 | ||
39279cc3 | 2955 | /* FIXME blocksize != 4096 */ |
9069218d | 2956 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2957 | if (extent_start != 0) { |
2958 | found_extent = 1; | |
e02119d5 | 2959 | if (root->ref_cows) |
a76a3cd4 | 2960 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2961 | } |
39279cc3 | 2962 | } |
9069218d | 2963 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2964 | /* |
2965 | * we can't truncate inline items that have had | |
2966 | * special encodings | |
2967 | */ | |
2968 | if (!del_item && | |
2969 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2970 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2971 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2972 | u32 size = new_size - found_key.offset; |
2973 | ||
2974 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2975 | inode_sub_bytes(inode, item_end + 1 - |
2976 | new_size); | |
e02119d5 CM |
2977 | } |
2978 | size = | |
2979 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2980 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2981 | size, 1); |
9069218d | 2982 | BUG_ON(ret); |
e02119d5 | 2983 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2984 | inode_sub_bytes(inode, item_end + 1 - |
2985 | found_key.offset); | |
9069218d | 2986 | } |
39279cc3 | 2987 | } |
179e29e4 | 2988 | delete: |
39279cc3 | 2989 | if (del_item) { |
85e21bac CM |
2990 | if (!pending_del_nr) { |
2991 | /* no pending yet, add ourselves */ | |
2992 | pending_del_slot = path->slots[0]; | |
2993 | pending_del_nr = 1; | |
2994 | } else if (pending_del_nr && | |
2995 | path->slots[0] + 1 == pending_del_slot) { | |
2996 | /* hop on the pending chunk */ | |
2997 | pending_del_nr++; | |
2998 | pending_del_slot = path->slots[0]; | |
2999 | } else { | |
d397712b | 3000 | BUG(); |
85e21bac | 3001 | } |
39279cc3 CM |
3002 | } else { |
3003 | break; | |
3004 | } | |
5d4f98a2 | 3005 | if (found_extent && root->ref_cows) { |
b9473439 | 3006 | btrfs_set_path_blocking(path); |
39279cc3 | 3007 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3008 | extent_num_bytes, 0, |
3009 | btrfs_header_owner(leaf), | |
3010 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3011 | BUG_ON(ret); |
3012 | } | |
85e21bac CM |
3013 | next: |
3014 | if (path->slots[0] == 0) { | |
3015 | if (pending_del_nr) | |
3016 | goto del_pending; | |
3017 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3018 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3019 | break; | |
85e21bac CM |
3020 | goto search_again; |
3021 | } | |
3022 | ||
3023 | path->slots[0]--; | |
3024 | if (pending_del_nr && | |
3025 | path->slots[0] + 1 != pending_del_slot) { | |
3026 | struct btrfs_key debug; | |
3027 | del_pending: | |
3028 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
3029 | pending_del_slot); | |
3030 | ret = btrfs_del_items(trans, root, path, | |
3031 | pending_del_slot, | |
3032 | pending_del_nr); | |
3033 | BUG_ON(ret); | |
3034 | pending_del_nr = 0; | |
3035 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3036 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3037 | break; | |
85e21bac CM |
3038 | goto search_again; |
3039 | } | |
39279cc3 CM |
3040 | } |
3041 | ret = 0; | |
3042 | error: | |
85e21bac CM |
3043 | if (pending_del_nr) { |
3044 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3045 | pending_del_nr); | |
3046 | } | |
39279cc3 | 3047 | btrfs_free_path(path); |
39279cc3 CM |
3048 | return ret; |
3049 | } | |
3050 | ||
3051 | /* | |
3052 | * taken from block_truncate_page, but does cow as it zeros out | |
3053 | * any bytes left in the last page in the file. | |
3054 | */ | |
3055 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3056 | { | |
3057 | struct inode *inode = mapping->host; | |
db94535d | 3058 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3059 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3060 | struct btrfs_ordered_extent *ordered; | |
3061 | char *kaddr; | |
db94535d | 3062 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3063 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3064 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3065 | struct page *page; | |
39279cc3 | 3066 | int ret = 0; |
a52d9a80 | 3067 | u64 page_start; |
e6dcd2dc | 3068 | u64 page_end; |
39279cc3 CM |
3069 | |
3070 | if ((offset & (blocksize - 1)) == 0) | |
3071 | goto out; | |
5d5e103a JB |
3072 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
3073 | if (ret) | |
3074 | goto out; | |
3075 | ||
3076 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); | |
3077 | if (ret) | |
3078 | goto out; | |
39279cc3 CM |
3079 | |
3080 | ret = -ENOMEM; | |
211c17f5 | 3081 | again: |
39279cc3 | 3082 | page = grab_cache_page(mapping, index); |
5d5e103a JB |
3083 | if (!page) { |
3084 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3085 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 | 3086 | goto out; |
5d5e103a | 3087 | } |
e6dcd2dc CM |
3088 | |
3089 | page_start = page_offset(page); | |
3090 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3091 | ||
39279cc3 | 3092 | if (!PageUptodate(page)) { |
9ebefb18 | 3093 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3094 | lock_page(page); |
211c17f5 CM |
3095 | if (page->mapping != mapping) { |
3096 | unlock_page(page); | |
3097 | page_cache_release(page); | |
3098 | goto again; | |
3099 | } | |
39279cc3 CM |
3100 | if (!PageUptodate(page)) { |
3101 | ret = -EIO; | |
89642229 | 3102 | goto out_unlock; |
39279cc3 CM |
3103 | } |
3104 | } | |
211c17f5 | 3105 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3106 | |
3107 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3108 | set_page_extent_mapped(page); | |
3109 | ||
3110 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3111 | if (ordered) { | |
3112 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3113 | unlock_page(page); | |
3114 | page_cache_release(page); | |
eb84ae03 | 3115 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3116 | btrfs_put_ordered_extent(ordered); |
3117 | goto again; | |
3118 | } | |
3119 | ||
5d5e103a JB |
3120 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
3121 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, | |
3122 | GFP_NOFS); | |
3123 | ||
9ed74f2d JB |
3124 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3125 | if (ret) { | |
3126 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3127 | goto out_unlock; | |
3128 | } | |
3129 | ||
e6dcd2dc CM |
3130 | ret = 0; |
3131 | if (offset != PAGE_CACHE_SIZE) { | |
3132 | kaddr = kmap(page); | |
3133 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3134 | flush_dcache_page(page); | |
3135 | kunmap(page); | |
3136 | } | |
247e743c | 3137 | ClearPageChecked(page); |
e6dcd2dc CM |
3138 | set_page_dirty(page); |
3139 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3140 | |
89642229 | 3141 | out_unlock: |
5d5e103a JB |
3142 | if (ret) |
3143 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3144 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 CM |
3145 | unlock_page(page); |
3146 | page_cache_release(page); | |
3147 | out: | |
3148 | return ret; | |
3149 | } | |
3150 | ||
9036c102 | 3151 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3152 | { |
9036c102 YZ |
3153 | struct btrfs_trans_handle *trans; |
3154 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3155 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3156 | struct extent_map *em; | |
3157 | u64 mask = root->sectorsize - 1; | |
3158 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3159 | u64 block_end = (size + mask) & ~mask; | |
3160 | u64 last_byte; | |
3161 | u64 cur_offset; | |
3162 | u64 hole_size; | |
9ed74f2d | 3163 | int err = 0; |
39279cc3 | 3164 | |
9036c102 YZ |
3165 | if (size <= hole_start) |
3166 | return 0; | |
3167 | ||
5d5e103a JB |
3168 | err = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
3169 | if (err) | |
3170 | return err; | |
2bf5a725 | 3171 | |
9036c102 YZ |
3172 | while (1) { |
3173 | struct btrfs_ordered_extent *ordered; | |
3174 | btrfs_wait_ordered_range(inode, hole_start, | |
3175 | block_end - hole_start); | |
3176 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3177 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3178 | if (!ordered) | |
3179 | break; | |
3180 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3181 | btrfs_put_ordered_extent(ordered); | |
3182 | } | |
39279cc3 | 3183 | |
9036c102 YZ |
3184 | trans = btrfs_start_transaction(root, 1); |
3185 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3186 | |
9036c102 YZ |
3187 | cur_offset = hole_start; |
3188 | while (1) { | |
3189 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3190 | block_end - cur_offset, 0); | |
3191 | BUG_ON(IS_ERR(em) || !em); | |
3192 | last_byte = min(extent_map_end(em), block_end); | |
3193 | last_byte = (last_byte + mask) & ~mask; | |
3194 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3195 | u64 hint_byte = 0; |
9036c102 | 3196 | hole_size = last_byte - cur_offset; |
771ed689 CM |
3197 | err = btrfs_drop_extents(trans, root, inode, |
3198 | cur_offset, | |
3199 | cur_offset + hole_size, | |
e980b50c | 3200 | block_end, |
a1ed835e | 3201 | cur_offset, &hint_byte, 1); |
771ed689 CM |
3202 | if (err) |
3203 | break; | |
9ed74f2d JB |
3204 | |
3205 | err = btrfs_reserve_metadata_space(root, 1); | |
3206 | if (err) | |
3207 | break; | |
3208 | ||
9036c102 YZ |
3209 | err = btrfs_insert_file_extent(trans, root, |
3210 | inode->i_ino, cur_offset, 0, | |
3211 | 0, hole_size, 0, hole_size, | |
3212 | 0, 0, 0); | |
3213 | btrfs_drop_extent_cache(inode, hole_start, | |
3214 | last_byte - 1, 0); | |
9ed74f2d | 3215 | btrfs_unreserve_metadata_space(root, 1); |
9036c102 YZ |
3216 | } |
3217 | free_extent_map(em); | |
3218 | cur_offset = last_byte; | |
3219 | if (err || cur_offset >= block_end) | |
3220 | break; | |
3221 | } | |
1832a6d5 | 3222 | |
9036c102 YZ |
3223 | btrfs_end_transaction(trans, root); |
3224 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3225 | return err; | |
3226 | } | |
39279cc3 | 3227 | |
9036c102 YZ |
3228 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3229 | { | |
3230 | struct inode *inode = dentry->d_inode; | |
3231 | int err; | |
39279cc3 | 3232 | |
9036c102 YZ |
3233 | err = inode_change_ok(inode, attr); |
3234 | if (err) | |
3235 | return err; | |
2bf5a725 | 3236 | |
5a3f23d5 CM |
3237 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3238 | if (attr->ia_size > inode->i_size) { | |
3239 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3240 | if (err) | |
3241 | return err; | |
3242 | } else if (inode->i_size > 0 && | |
3243 | attr->ia_size == 0) { | |
3244 | ||
3245 | /* we're truncating a file that used to have good | |
3246 | * data down to zero. Make sure it gets into | |
3247 | * the ordered flush list so that any new writes | |
3248 | * get down to disk quickly. | |
3249 | */ | |
3250 | BTRFS_I(inode)->ordered_data_close = 1; | |
3251 | } | |
39279cc3 | 3252 | } |
9036c102 | 3253 | |
39279cc3 | 3254 | err = inode_setattr(inode, attr); |
33268eaf JB |
3255 | |
3256 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3257 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3258 | return err; |
3259 | } | |
61295eb8 | 3260 | |
39279cc3 CM |
3261 | void btrfs_delete_inode(struct inode *inode) |
3262 | { | |
3263 | struct btrfs_trans_handle *trans; | |
3264 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3265 | unsigned long nr; |
39279cc3 CM |
3266 | int ret; |
3267 | ||
3268 | truncate_inode_pages(&inode->i_data, 0); | |
3269 | if (is_bad_inode(inode)) { | |
7b128766 | 3270 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3271 | goto no_delete; |
3272 | } | |
4a096752 | 3273 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3274 | |
76dda93c YZ |
3275 | if (inode->i_nlink > 0) { |
3276 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3277 | goto no_delete; | |
3278 | } | |
3279 | ||
dbe674a9 | 3280 | btrfs_i_size_write(inode, 0); |
180591bc | 3281 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3282 | |
39279cc3 | 3283 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3284 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3285 | if (ret) { |
3286 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3287 | goto no_delete_lock; |
7b128766 JB |
3288 | } |
3289 | ||
3290 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3291 | |
d3c2fdcf | 3292 | nr = trans->blocks_used; |
85e21bac | 3293 | clear_inode(inode); |
5f39d397 | 3294 | |
39279cc3 | 3295 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3296 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3297 | return; |
54aa1f4d CM |
3298 | |
3299 | no_delete_lock: | |
d3c2fdcf | 3300 | nr = trans->blocks_used; |
54aa1f4d | 3301 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3302 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3303 | no_delete: |
3304 | clear_inode(inode); | |
3305 | } | |
3306 | ||
3307 | /* | |
3308 | * this returns the key found in the dir entry in the location pointer. | |
3309 | * If no dir entries were found, location->objectid is 0. | |
3310 | */ | |
3311 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3312 | struct btrfs_key *location) | |
3313 | { | |
3314 | const char *name = dentry->d_name.name; | |
3315 | int namelen = dentry->d_name.len; | |
3316 | struct btrfs_dir_item *di; | |
3317 | struct btrfs_path *path; | |
3318 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3319 | int ret = 0; |
39279cc3 CM |
3320 | |
3321 | path = btrfs_alloc_path(); | |
3322 | BUG_ON(!path); | |
3954401f | 3323 | |
39279cc3 CM |
3324 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3325 | namelen, 0); | |
0d9f7f3e Y |
3326 | if (IS_ERR(di)) |
3327 | ret = PTR_ERR(di); | |
d397712b CM |
3328 | |
3329 | if (!di || IS_ERR(di)) | |
3954401f | 3330 | goto out_err; |
d397712b | 3331 | |
5f39d397 | 3332 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3333 | out: |
39279cc3 CM |
3334 | btrfs_free_path(path); |
3335 | return ret; | |
3954401f CM |
3336 | out_err: |
3337 | location->objectid = 0; | |
3338 | goto out; | |
39279cc3 CM |
3339 | } |
3340 | ||
3341 | /* | |
3342 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3343 | * needs to be changed to reflect the root directory of the tree root. This | |
3344 | * is kind of like crossing a mount point. | |
3345 | */ | |
3346 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3347 | struct inode *dir, |
3348 | struct dentry *dentry, | |
3349 | struct btrfs_key *location, | |
3350 | struct btrfs_root **sub_root) | |
39279cc3 | 3351 | { |
4df27c4d YZ |
3352 | struct btrfs_path *path; |
3353 | struct btrfs_root *new_root; | |
3354 | struct btrfs_root_ref *ref; | |
3355 | struct extent_buffer *leaf; | |
3356 | int ret; | |
3357 | int err = 0; | |
39279cc3 | 3358 | |
4df27c4d YZ |
3359 | path = btrfs_alloc_path(); |
3360 | if (!path) { | |
3361 | err = -ENOMEM; | |
3362 | goto out; | |
3363 | } | |
39279cc3 | 3364 | |
4df27c4d YZ |
3365 | err = -ENOENT; |
3366 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3367 | BTRFS_I(dir)->root->root_key.objectid, | |
3368 | location->objectid); | |
3369 | if (ret) { | |
3370 | if (ret < 0) | |
3371 | err = ret; | |
3372 | goto out; | |
3373 | } | |
39279cc3 | 3374 | |
4df27c4d YZ |
3375 | leaf = path->nodes[0]; |
3376 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3377 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3378 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3379 | goto out; | |
39279cc3 | 3380 | |
4df27c4d YZ |
3381 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3382 | (unsigned long)(ref + 1), | |
3383 | dentry->d_name.len); | |
3384 | if (ret) | |
3385 | goto out; | |
3386 | ||
3387 | btrfs_release_path(root->fs_info->tree_root, path); | |
3388 | ||
3389 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3390 | if (IS_ERR(new_root)) { | |
3391 | err = PTR_ERR(new_root); | |
3392 | goto out; | |
3393 | } | |
3394 | ||
3395 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3396 | err = -ENOENT; | |
3397 | goto out; | |
3398 | } | |
3399 | ||
3400 | *sub_root = new_root; | |
3401 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3402 | location->type = BTRFS_INODE_ITEM_KEY; | |
3403 | location->offset = 0; | |
3404 | err = 0; | |
3405 | out: | |
3406 | btrfs_free_path(path); | |
3407 | return err; | |
39279cc3 CM |
3408 | } |
3409 | ||
5d4f98a2 YZ |
3410 | static void inode_tree_add(struct inode *inode) |
3411 | { | |
3412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3413 | struct btrfs_inode *entry; | |
03e860bd NP |
3414 | struct rb_node **p; |
3415 | struct rb_node *parent; | |
03e860bd NP |
3416 | again: |
3417 | p = &root->inode_tree.rb_node; | |
3418 | parent = NULL; | |
5d4f98a2 | 3419 | |
76dda93c YZ |
3420 | if (hlist_unhashed(&inode->i_hash)) |
3421 | return; | |
3422 | ||
5d4f98a2 YZ |
3423 | spin_lock(&root->inode_lock); |
3424 | while (*p) { | |
3425 | parent = *p; | |
3426 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3427 | ||
3428 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3429 | p = &parent->rb_left; |
5d4f98a2 | 3430 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3431 | p = &parent->rb_right; |
5d4f98a2 YZ |
3432 | else { |
3433 | WARN_ON(!(entry->vfs_inode.i_state & | |
3434 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd NP |
3435 | rb_erase(parent, &root->inode_tree); |
3436 | RB_CLEAR_NODE(parent); | |
3437 | spin_unlock(&root->inode_lock); | |
3438 | goto again; | |
5d4f98a2 YZ |
3439 | } |
3440 | } | |
3441 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3442 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3443 | spin_unlock(&root->inode_lock); | |
3444 | } | |
3445 | ||
3446 | static void inode_tree_del(struct inode *inode) | |
3447 | { | |
3448 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3449 | int empty = 0; |
5d4f98a2 | 3450 | |
03e860bd | 3451 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3452 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3453 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3454 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3455 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3456 | } |
03e860bd | 3457 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3458 | |
3459 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3460 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3461 | spin_lock(&root->inode_lock); | |
3462 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3463 | spin_unlock(&root->inode_lock); | |
3464 | if (empty) | |
3465 | btrfs_add_dead_root(root); | |
3466 | } | |
3467 | } | |
3468 | ||
3469 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3470 | { | |
3471 | struct rb_node *node; | |
3472 | struct rb_node *prev; | |
3473 | struct btrfs_inode *entry; | |
3474 | struct inode *inode; | |
3475 | u64 objectid = 0; | |
3476 | ||
3477 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3478 | ||
3479 | spin_lock(&root->inode_lock); | |
3480 | again: | |
3481 | node = root->inode_tree.rb_node; | |
3482 | prev = NULL; | |
3483 | while (node) { | |
3484 | prev = node; | |
3485 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3486 | ||
3487 | if (objectid < entry->vfs_inode.i_ino) | |
3488 | node = node->rb_left; | |
3489 | else if (objectid > entry->vfs_inode.i_ino) | |
3490 | node = node->rb_right; | |
3491 | else | |
3492 | break; | |
3493 | } | |
3494 | if (!node) { | |
3495 | while (prev) { | |
3496 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3497 | if (objectid <= entry->vfs_inode.i_ino) { | |
3498 | node = prev; | |
3499 | break; | |
3500 | } | |
3501 | prev = rb_next(prev); | |
3502 | } | |
3503 | } | |
3504 | while (node) { | |
3505 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3506 | objectid = entry->vfs_inode.i_ino + 1; | |
3507 | inode = igrab(&entry->vfs_inode); | |
3508 | if (inode) { | |
3509 | spin_unlock(&root->inode_lock); | |
3510 | if (atomic_read(&inode->i_count) > 1) | |
3511 | d_prune_aliases(inode); | |
3512 | /* | |
3513 | * btrfs_drop_inode will remove it from | |
3514 | * the inode cache when its usage count | |
3515 | * hits zero. | |
3516 | */ | |
3517 | iput(inode); | |
3518 | cond_resched(); | |
3519 | spin_lock(&root->inode_lock); | |
3520 | goto again; | |
3521 | } | |
3522 | ||
3523 | if (cond_resched_lock(&root->inode_lock)) | |
3524 | goto again; | |
3525 | ||
3526 | node = rb_next(node); | |
3527 | } | |
3528 | spin_unlock(&root->inode_lock); | |
3529 | return 0; | |
5d4f98a2 YZ |
3530 | } |
3531 | ||
e02119d5 | 3532 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3533 | { |
e02119d5 CM |
3534 | struct btrfs_inode *bi = BTRFS_I(inode); |
3535 | ||
e02119d5 | 3536 | bi->generation = 0; |
c3027eb5 | 3537 | bi->sequence = 0; |
e02119d5 | 3538 | bi->last_trans = 0; |
257c62e1 | 3539 | bi->last_sub_trans = 0; |
e02119d5 CM |
3540 | bi->logged_trans = 0; |
3541 | bi->delalloc_bytes = 0; | |
6a63209f | 3542 | bi->reserved_bytes = 0; |
e02119d5 CM |
3543 | bi->disk_i_size = 0; |
3544 | bi->flags = 0; | |
3545 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3546 | bi->last_unlink_trans = 0; |
2757495c | 3547 | bi->ordered_data_close = 0; |
d1310b2e CM |
3548 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3549 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3550 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3551 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3552 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3553 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3554 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3555 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3556 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3557 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3558 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3559 | } | |
3560 | ||
3561 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3562 | { | |
3563 | struct btrfs_iget_args *args = p; | |
3564 | inode->i_ino = args->ino; | |
3565 | init_btrfs_i(inode); | |
3566 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3567 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3568 | return 0; |
3569 | } | |
3570 | ||
3571 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3572 | { | |
3573 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3574 | return args->ino == inode->i_ino && |
3575 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3576 | } |
3577 | ||
5d4f98a2 YZ |
3578 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3579 | u64 objectid, | |
3580 | struct btrfs_root *root) | |
39279cc3 CM |
3581 | { |
3582 | struct inode *inode; | |
3583 | struct btrfs_iget_args args; | |
3584 | args.ino = objectid; | |
3585 | args.root = root; | |
3586 | ||
3587 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3588 | btrfs_init_locked_inode, | |
3589 | (void *)&args); | |
3590 | return inode; | |
3591 | } | |
3592 | ||
1a54ef8c BR |
3593 | /* Get an inode object given its location and corresponding root. |
3594 | * Returns in *is_new if the inode was read from disk | |
3595 | */ | |
3596 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3597 | struct btrfs_root *root) |
1a54ef8c BR |
3598 | { |
3599 | struct inode *inode; | |
3600 | ||
3601 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3602 | if (!inode) | |
5d4f98a2 | 3603 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3604 | |
3605 | if (inode->i_state & I_NEW) { | |
3606 | BTRFS_I(inode)->root = root; | |
3607 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3608 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3609 | |
3610 | inode_tree_add(inode); | |
1a54ef8c | 3611 | unlock_new_inode(inode); |
1a54ef8c BR |
3612 | } |
3613 | ||
3614 | return inode; | |
3615 | } | |
3616 | ||
4df27c4d YZ |
3617 | static struct inode *new_simple_dir(struct super_block *s, |
3618 | struct btrfs_key *key, | |
3619 | struct btrfs_root *root) | |
3620 | { | |
3621 | struct inode *inode = new_inode(s); | |
3622 | ||
3623 | if (!inode) | |
3624 | return ERR_PTR(-ENOMEM); | |
3625 | ||
3626 | init_btrfs_i(inode); | |
3627 | ||
3628 | BTRFS_I(inode)->root = root; | |
3629 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3630 | BTRFS_I(inode)->dummy_inode = 1; | |
3631 | ||
3632 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3633 | inode->i_op = &simple_dir_inode_operations; | |
3634 | inode->i_fop = &simple_dir_operations; | |
3635 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3636 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3637 | ||
3638 | return inode; | |
3639 | } | |
3640 | ||
3de4586c | 3641 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3642 | { |
d397712b | 3643 | struct inode *inode; |
4df27c4d | 3644 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3645 | struct btrfs_root *sub_root = root; |
3646 | struct btrfs_key location; | |
76dda93c | 3647 | int index; |
5d4f98a2 | 3648 | int ret; |
39279cc3 | 3649 | |
76dda93c YZ |
3650 | dentry->d_op = &btrfs_dentry_operations; |
3651 | ||
39279cc3 CM |
3652 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3653 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3654 | |
39279cc3 | 3655 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3656 | |
39279cc3 CM |
3657 | if (ret < 0) |
3658 | return ERR_PTR(ret); | |
5f39d397 | 3659 | |
4df27c4d YZ |
3660 | if (location.objectid == 0) |
3661 | return NULL; | |
3662 | ||
3663 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3664 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3665 | return inode; | |
3666 | } | |
3667 | ||
3668 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3669 | ||
76dda93c | 3670 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3671 | ret = fixup_tree_root_location(root, dir, dentry, |
3672 | &location, &sub_root); | |
3673 | if (ret < 0) { | |
3674 | if (ret != -ENOENT) | |
3675 | inode = ERR_PTR(ret); | |
3676 | else | |
3677 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3678 | } else { | |
5d4f98a2 | 3679 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3680 | } |
76dda93c YZ |
3681 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3682 | ||
3de4586c CM |
3683 | return inode; |
3684 | } | |
3685 | ||
76dda93c YZ |
3686 | static int btrfs_dentry_delete(struct dentry *dentry) |
3687 | { | |
3688 | struct btrfs_root *root; | |
3689 | ||
efefb143 YZ |
3690 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3691 | dentry = dentry->d_parent; | |
76dda93c | 3692 | |
efefb143 YZ |
3693 | if (dentry->d_inode) { |
3694 | root = BTRFS_I(dentry->d_inode)->root; | |
3695 | if (btrfs_root_refs(&root->root_item) == 0) | |
3696 | return 1; | |
3697 | } | |
76dda93c YZ |
3698 | return 0; |
3699 | } | |
3700 | ||
3de4586c CM |
3701 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3702 | struct nameidata *nd) | |
3703 | { | |
3704 | struct inode *inode; | |
3705 | ||
3de4586c CM |
3706 | inode = btrfs_lookup_dentry(dir, dentry); |
3707 | if (IS_ERR(inode)) | |
3708 | return ERR_CAST(inode); | |
7b128766 | 3709 | |
39279cc3 CM |
3710 | return d_splice_alias(inode, dentry); |
3711 | } | |
3712 | ||
39279cc3 CM |
3713 | static unsigned char btrfs_filetype_table[] = { |
3714 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3715 | }; | |
3716 | ||
cbdf5a24 DW |
3717 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3718 | filldir_t filldir) | |
39279cc3 | 3719 | { |
6da6abae | 3720 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3721 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3722 | struct btrfs_item *item; | |
3723 | struct btrfs_dir_item *di; | |
3724 | struct btrfs_key key; | |
5f39d397 | 3725 | struct btrfs_key found_key; |
39279cc3 CM |
3726 | struct btrfs_path *path; |
3727 | int ret; | |
3728 | u32 nritems; | |
5f39d397 | 3729 | struct extent_buffer *leaf; |
39279cc3 CM |
3730 | int slot; |
3731 | int advance; | |
3732 | unsigned char d_type; | |
3733 | int over = 0; | |
3734 | u32 di_cur; | |
3735 | u32 di_total; | |
3736 | u32 di_len; | |
3737 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3738 | char tmp_name[32]; |
3739 | char *name_ptr; | |
3740 | int name_len; | |
39279cc3 CM |
3741 | |
3742 | /* FIXME, use a real flag for deciding about the key type */ | |
3743 | if (root->fs_info->tree_root == root) | |
3744 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3745 | |
3954401f CM |
3746 | /* special case for "." */ |
3747 | if (filp->f_pos == 0) { | |
3748 | over = filldir(dirent, ".", 1, | |
3749 | 1, inode->i_ino, | |
3750 | DT_DIR); | |
3751 | if (over) | |
3752 | return 0; | |
3753 | filp->f_pos = 1; | |
3754 | } | |
3954401f CM |
3755 | /* special case for .., just use the back ref */ |
3756 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3757 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3758 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3759 | 2, pino, DT_DIR); |
3954401f | 3760 | if (over) |
49593bfa | 3761 | return 0; |
3954401f CM |
3762 | filp->f_pos = 2; |
3763 | } | |
49593bfa DW |
3764 | path = btrfs_alloc_path(); |
3765 | path->reada = 2; | |
3766 | ||
39279cc3 CM |
3767 | btrfs_set_key_type(&key, key_type); |
3768 | key.offset = filp->f_pos; | |
49593bfa | 3769 | key.objectid = inode->i_ino; |
5f39d397 | 3770 | |
39279cc3 CM |
3771 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3772 | if (ret < 0) | |
3773 | goto err; | |
3774 | advance = 0; | |
49593bfa DW |
3775 | |
3776 | while (1) { | |
5f39d397 CM |
3777 | leaf = path->nodes[0]; |
3778 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3779 | slot = path->slots[0]; |
3780 | if (advance || slot >= nritems) { | |
49593bfa | 3781 | if (slot >= nritems - 1) { |
39279cc3 CM |
3782 | ret = btrfs_next_leaf(root, path); |
3783 | if (ret) | |
3784 | break; | |
5f39d397 CM |
3785 | leaf = path->nodes[0]; |
3786 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3787 | slot = path->slots[0]; |
3788 | } else { | |
3789 | slot++; | |
3790 | path->slots[0]++; | |
3791 | } | |
3792 | } | |
3de4586c | 3793 | |
39279cc3 | 3794 | advance = 1; |
5f39d397 CM |
3795 | item = btrfs_item_nr(leaf, slot); |
3796 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3797 | ||
3798 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3799 | break; |
5f39d397 | 3800 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3801 | break; |
5f39d397 | 3802 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3803 | continue; |
5f39d397 CM |
3804 | |
3805 | filp->f_pos = found_key.offset; | |
49593bfa | 3806 | |
39279cc3 CM |
3807 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3808 | di_cur = 0; | |
5f39d397 | 3809 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3810 | |
3811 | while (di_cur < di_total) { | |
5f39d397 CM |
3812 | struct btrfs_key location; |
3813 | ||
3814 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3815 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3816 | name_ptr = tmp_name; |
3817 | } else { | |
3818 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3819 | if (!name_ptr) { |
3820 | ret = -ENOMEM; | |
3821 | goto err; | |
3822 | } | |
5f39d397 CM |
3823 | } |
3824 | read_extent_buffer(leaf, name_ptr, | |
3825 | (unsigned long)(di + 1), name_len); | |
3826 | ||
3827 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3828 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3829 | |
3830 | /* is this a reference to our own snapshot? If so | |
3831 | * skip it | |
3832 | */ | |
3833 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3834 | location.objectid == root->root_key.objectid) { | |
3835 | over = 0; | |
3836 | goto skip; | |
3837 | } | |
5f39d397 | 3838 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3839 | found_key.offset, location.objectid, |
39279cc3 | 3840 | d_type); |
5f39d397 | 3841 | |
3de4586c | 3842 | skip: |
5f39d397 CM |
3843 | if (name_ptr != tmp_name) |
3844 | kfree(name_ptr); | |
3845 | ||
39279cc3 CM |
3846 | if (over) |
3847 | goto nopos; | |
5103e947 | 3848 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3849 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3850 | di_cur += di_len; |
3851 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3852 | } | |
3853 | } | |
49593bfa DW |
3854 | |
3855 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3856 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3857 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3858 | else |
3859 | filp->f_pos++; | |
39279cc3 CM |
3860 | nopos: |
3861 | ret = 0; | |
3862 | err: | |
39279cc3 | 3863 | btrfs_free_path(path); |
39279cc3 CM |
3864 | return ret; |
3865 | } | |
3866 | ||
3867 | int btrfs_write_inode(struct inode *inode, int wait) | |
3868 | { | |
3869 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3870 | struct btrfs_trans_handle *trans; | |
3871 | int ret = 0; | |
3872 | ||
c146afad | 3873 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3874 | return 0; |
3875 | ||
39279cc3 | 3876 | if (wait) { |
f9295749 | 3877 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3878 | btrfs_set_trans_block_group(trans, inode); |
3879 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3880 | } |
3881 | return ret; | |
3882 | } | |
3883 | ||
3884 | /* | |
54aa1f4d | 3885 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3886 | * inode changes. But, it is most likely to find the inode in cache. |
3887 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3888 | * to keep or drop this code. | |
3889 | */ | |
3890 | void btrfs_dirty_inode(struct inode *inode) | |
3891 | { | |
3892 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3893 | struct btrfs_trans_handle *trans; | |
3894 | ||
f9295749 | 3895 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3896 | btrfs_set_trans_block_group(trans, inode); |
3897 | btrfs_update_inode(trans, root, inode); | |
3898 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3899 | } |
3900 | ||
d352ac68 CM |
3901 | /* |
3902 | * find the highest existing sequence number in a directory | |
3903 | * and then set the in-memory index_cnt variable to reflect | |
3904 | * free sequence numbers | |
3905 | */ | |
aec7477b JB |
3906 | static int btrfs_set_inode_index_count(struct inode *inode) |
3907 | { | |
3908 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3909 | struct btrfs_key key, found_key; | |
3910 | struct btrfs_path *path; | |
3911 | struct extent_buffer *leaf; | |
3912 | int ret; | |
3913 | ||
3914 | key.objectid = inode->i_ino; | |
3915 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3916 | key.offset = (u64)-1; | |
3917 | ||
3918 | path = btrfs_alloc_path(); | |
3919 | if (!path) | |
3920 | return -ENOMEM; | |
3921 | ||
3922 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3923 | if (ret < 0) | |
3924 | goto out; | |
3925 | /* FIXME: we should be able to handle this */ | |
3926 | if (ret == 0) | |
3927 | goto out; | |
3928 | ret = 0; | |
3929 | ||
3930 | /* | |
3931 | * MAGIC NUMBER EXPLANATION: | |
3932 | * since we search a directory based on f_pos we have to start at 2 | |
3933 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3934 | * else has to start at 2 | |
3935 | */ | |
3936 | if (path->slots[0] == 0) { | |
3937 | BTRFS_I(inode)->index_cnt = 2; | |
3938 | goto out; | |
3939 | } | |
3940 | ||
3941 | path->slots[0]--; | |
3942 | ||
3943 | leaf = path->nodes[0]; | |
3944 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3945 | ||
3946 | if (found_key.objectid != inode->i_ino || | |
3947 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3948 | BTRFS_I(inode)->index_cnt = 2; | |
3949 | goto out; | |
3950 | } | |
3951 | ||
3952 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3953 | out: | |
3954 | btrfs_free_path(path); | |
3955 | return ret; | |
3956 | } | |
3957 | ||
d352ac68 CM |
3958 | /* |
3959 | * helper to find a free sequence number in a given directory. This current | |
3960 | * code is very simple, later versions will do smarter things in the btree | |
3961 | */ | |
3de4586c | 3962 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3963 | { |
3964 | int ret = 0; | |
3965 | ||
3966 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3967 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3968 | if (ret) |
aec7477b JB |
3969 | return ret; |
3970 | } | |
3971 | ||
00e4e6b3 | 3972 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3973 | BTRFS_I(dir)->index_cnt++; |
3974 | ||
3975 | return ret; | |
3976 | } | |
3977 | ||
39279cc3 CM |
3978 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3979 | struct btrfs_root *root, | |
aec7477b | 3980 | struct inode *dir, |
9c58309d | 3981 | const char *name, int name_len, |
d2fb3437 YZ |
3982 | u64 ref_objectid, u64 objectid, |
3983 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3984 | { |
3985 | struct inode *inode; | |
5f39d397 | 3986 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3987 | struct btrfs_key *location; |
5f39d397 | 3988 | struct btrfs_path *path; |
9c58309d CM |
3989 | struct btrfs_inode_ref *ref; |
3990 | struct btrfs_key key[2]; | |
3991 | u32 sizes[2]; | |
3992 | unsigned long ptr; | |
39279cc3 CM |
3993 | int ret; |
3994 | int owner; | |
3995 | ||
5f39d397 CM |
3996 | path = btrfs_alloc_path(); |
3997 | BUG_ON(!path); | |
3998 | ||
39279cc3 CM |
3999 | inode = new_inode(root->fs_info->sb); |
4000 | if (!inode) | |
4001 | return ERR_PTR(-ENOMEM); | |
4002 | ||
aec7477b | 4003 | if (dir) { |
3de4586c | 4004 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
4005 | if (ret) { |
4006 | iput(inode); | |
aec7477b | 4007 | return ERR_PTR(ret); |
09771430 | 4008 | } |
aec7477b JB |
4009 | } |
4010 | /* | |
4011 | * index_cnt is ignored for everything but a dir, | |
4012 | * btrfs_get_inode_index_count has an explanation for the magic | |
4013 | * number | |
4014 | */ | |
e02119d5 | 4015 | init_btrfs_i(inode); |
aec7477b | 4016 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 4017 | BTRFS_I(inode)->root = root; |
e02119d5 | 4018 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 4019 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4020 | |
39279cc3 CM |
4021 | if (mode & S_IFDIR) |
4022 | owner = 0; | |
4023 | else | |
4024 | owner = 1; | |
d2fb3437 YZ |
4025 | BTRFS_I(inode)->block_group = |
4026 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4027 | |
4028 | key[0].objectid = objectid; | |
4029 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4030 | key[0].offset = 0; | |
4031 | ||
4032 | key[1].objectid = objectid; | |
4033 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4034 | key[1].offset = ref_objectid; | |
4035 | ||
4036 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4037 | sizes[1] = name_len + sizeof(*ref); | |
4038 | ||
b9473439 | 4039 | path->leave_spinning = 1; |
9c58309d CM |
4040 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4041 | if (ret != 0) | |
5f39d397 CM |
4042 | goto fail; |
4043 | ||
79683f2d | 4044 | inode->i_uid = current_fsuid(); |
8c087b51 | 4045 | |
42f15d77 | 4046 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4047 | inode->i_gid = dir->i_gid; |
4048 | if (S_ISDIR(mode)) | |
4049 | mode |= S_ISGID; | |
4050 | } else | |
4051 | inode->i_gid = current_fsgid(); | |
4052 | ||
39279cc3 CM |
4053 | inode->i_mode = mode; |
4054 | inode->i_ino = objectid; | |
a76a3cd4 | 4055 | inode_set_bytes(inode, 0); |
39279cc3 | 4056 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4057 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4058 | struct btrfs_inode_item); | |
e02119d5 | 4059 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4060 | |
4061 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4062 | struct btrfs_inode_ref); | |
4063 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4064 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4065 | ptr = (unsigned long)(ref + 1); |
4066 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4067 | ||
5f39d397 CM |
4068 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4069 | btrfs_free_path(path); | |
4070 | ||
39279cc3 CM |
4071 | location = &BTRFS_I(inode)->location; |
4072 | location->objectid = objectid; | |
39279cc3 CM |
4073 | location->offset = 0; |
4074 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4075 | ||
6cbff00f CH |
4076 | btrfs_inherit_iflags(inode, dir); |
4077 | ||
94272164 CM |
4078 | if ((mode & S_IFREG)) { |
4079 | if (btrfs_test_opt(root, NODATASUM)) | |
4080 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4081 | if (btrfs_test_opt(root, NODATACOW)) | |
4082 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4083 | } | |
4084 | ||
39279cc3 | 4085 | insert_inode_hash(inode); |
5d4f98a2 | 4086 | inode_tree_add(inode); |
39279cc3 | 4087 | return inode; |
5f39d397 | 4088 | fail: |
aec7477b JB |
4089 | if (dir) |
4090 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4091 | btrfs_free_path(path); |
09771430 | 4092 | iput(inode); |
5f39d397 | 4093 | return ERR_PTR(ret); |
39279cc3 CM |
4094 | } |
4095 | ||
4096 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4097 | { | |
4098 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4099 | } | |
4100 | ||
d352ac68 CM |
4101 | /* |
4102 | * utility function to add 'inode' into 'parent_inode' with | |
4103 | * a give name and a given sequence number. | |
4104 | * if 'add_backref' is true, also insert a backref from the | |
4105 | * inode to the parent directory. | |
4106 | */ | |
e02119d5 CM |
4107 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4108 | struct inode *parent_inode, struct inode *inode, | |
4109 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4110 | { |
4df27c4d | 4111 | int ret = 0; |
39279cc3 | 4112 | struct btrfs_key key; |
e02119d5 | 4113 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4114 | |
4df27c4d YZ |
4115 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4116 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4117 | } else { | |
4118 | key.objectid = inode->i_ino; | |
4119 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4120 | key.offset = 0; | |
4121 | } | |
4122 | ||
4123 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4124 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4125 | key.objectid, root->root_key.objectid, | |
4126 | parent_inode->i_ino, | |
4127 | index, name, name_len); | |
4128 | } else if (add_backref) { | |
4129 | ret = btrfs_insert_inode_ref(trans, root, | |
4130 | name, name_len, inode->i_ino, | |
4131 | parent_inode->i_ino, index); | |
4132 | } | |
39279cc3 | 4133 | |
39279cc3 | 4134 | if (ret == 0) { |
4df27c4d YZ |
4135 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4136 | parent_inode->i_ino, &key, | |
4137 | btrfs_inode_type(inode), index); | |
4138 | BUG_ON(ret); | |
4139 | ||
dbe674a9 | 4140 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4141 | name_len * 2); |
79c44584 | 4142 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4143 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4144 | } |
4145 | return ret; | |
4146 | } | |
4147 | ||
4148 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4149 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4150 | int backref, u64 index) |
39279cc3 | 4151 | { |
e02119d5 CM |
4152 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4153 | inode, dentry->d_name.name, | |
4154 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4155 | if (!err) { |
4156 | d_instantiate(dentry, inode); | |
4157 | return 0; | |
4158 | } | |
4159 | if (err > 0) | |
4160 | err = -EEXIST; | |
4161 | return err; | |
4162 | } | |
4163 | ||
618e21d5 JB |
4164 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4165 | int mode, dev_t rdev) | |
4166 | { | |
4167 | struct btrfs_trans_handle *trans; | |
4168 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4169 | struct inode *inode = NULL; |
618e21d5 JB |
4170 | int err; |
4171 | int drop_inode = 0; | |
4172 | u64 objectid; | |
1832a6d5 | 4173 | unsigned long nr = 0; |
00e4e6b3 | 4174 | u64 index = 0; |
618e21d5 JB |
4175 | |
4176 | if (!new_valid_dev(rdev)) | |
4177 | return -EINVAL; | |
4178 | ||
9ed74f2d JB |
4179 | /* |
4180 | * 2 for inode item and ref | |
4181 | * 2 for dir items | |
4182 | * 1 for xattr if selinux is on | |
4183 | */ | |
4184 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4185 | if (err) |
9ed74f2d | 4186 | return err; |
1832a6d5 | 4187 | |
618e21d5 | 4188 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4189 | if (!trans) |
4190 | goto fail; | |
618e21d5 JB |
4191 | btrfs_set_trans_block_group(trans, dir); |
4192 | ||
4193 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4194 | if (err) { | |
4195 | err = -ENOSPC; | |
4196 | goto out_unlock; | |
4197 | } | |
4198 | ||
aec7477b | 4199 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4200 | dentry->d_name.len, |
4201 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4202 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4203 | err = PTR_ERR(inode); |
4204 | if (IS_ERR(inode)) | |
4205 | goto out_unlock; | |
4206 | ||
0279b4cd | 4207 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4208 | if (err) { |
4209 | drop_inode = 1; | |
4210 | goto out_unlock; | |
4211 | } | |
4212 | ||
618e21d5 | 4213 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4214 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4215 | if (err) |
4216 | drop_inode = 1; | |
4217 | else { | |
4218 | inode->i_op = &btrfs_special_inode_operations; | |
4219 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4220 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4221 | } |
618e21d5 JB |
4222 | btrfs_update_inode_block_group(trans, inode); |
4223 | btrfs_update_inode_block_group(trans, dir); | |
4224 | out_unlock: | |
d3c2fdcf | 4225 | nr = trans->blocks_used; |
89ce8a63 | 4226 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4227 | fail: |
9ed74f2d | 4228 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4229 | if (drop_inode) { |
4230 | inode_dec_link_count(inode); | |
4231 | iput(inode); | |
4232 | } | |
d3c2fdcf | 4233 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4234 | return err; |
4235 | } | |
4236 | ||
39279cc3 CM |
4237 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4238 | int mode, struct nameidata *nd) | |
4239 | { | |
4240 | struct btrfs_trans_handle *trans; | |
4241 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4242 | struct inode *inode = NULL; |
39279cc3 CM |
4243 | int err; |
4244 | int drop_inode = 0; | |
1832a6d5 | 4245 | unsigned long nr = 0; |
39279cc3 | 4246 | u64 objectid; |
00e4e6b3 | 4247 | u64 index = 0; |
39279cc3 | 4248 | |
9ed74f2d JB |
4249 | /* |
4250 | * 2 for inode item and ref | |
4251 | * 2 for dir items | |
4252 | * 1 for xattr if selinux is on | |
4253 | */ | |
4254 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4255 | if (err) |
9ed74f2d JB |
4256 | return err; |
4257 | ||
39279cc3 | 4258 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4259 | if (!trans) |
4260 | goto fail; | |
39279cc3 CM |
4261 | btrfs_set_trans_block_group(trans, dir); |
4262 | ||
4263 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4264 | if (err) { | |
4265 | err = -ENOSPC; | |
4266 | goto out_unlock; | |
4267 | } | |
4268 | ||
aec7477b | 4269 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4270 | dentry->d_name.len, |
4271 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4272 | objectid, BTRFS_I(dir)->block_group, mode, |
4273 | &index); | |
39279cc3 CM |
4274 | err = PTR_ERR(inode); |
4275 | if (IS_ERR(inode)) | |
4276 | goto out_unlock; | |
4277 | ||
0279b4cd | 4278 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4279 | if (err) { |
4280 | drop_inode = 1; | |
4281 | goto out_unlock; | |
4282 | } | |
4283 | ||
39279cc3 | 4284 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4285 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4286 | if (err) |
4287 | drop_inode = 1; | |
4288 | else { | |
4289 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4290 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4291 | inode->i_fop = &btrfs_file_operations; |
4292 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4293 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4294 | } |
39279cc3 CM |
4295 | btrfs_update_inode_block_group(trans, inode); |
4296 | btrfs_update_inode_block_group(trans, dir); | |
4297 | out_unlock: | |
d3c2fdcf | 4298 | nr = trans->blocks_used; |
ab78c84d | 4299 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4300 | fail: |
9ed74f2d | 4301 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4302 | if (drop_inode) { |
4303 | inode_dec_link_count(inode); | |
4304 | iput(inode); | |
4305 | } | |
d3c2fdcf | 4306 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4307 | return err; |
4308 | } | |
4309 | ||
4310 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4311 | struct dentry *dentry) | |
4312 | { | |
4313 | struct btrfs_trans_handle *trans; | |
4314 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4315 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4316 | u64 index; |
1832a6d5 | 4317 | unsigned long nr = 0; |
39279cc3 CM |
4318 | int err; |
4319 | int drop_inode = 0; | |
4320 | ||
4321 | if (inode->i_nlink == 0) | |
4322 | return -ENOENT; | |
4323 | ||
9ed74f2d JB |
4324 | /* |
4325 | * 1 item for inode ref | |
4326 | * 2 items for dir items | |
4327 | */ | |
4328 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4329 | if (err) |
9ed74f2d JB |
4330 | return err; |
4331 | ||
4332 | btrfs_inc_nlink(inode); | |
4333 | ||
3de4586c | 4334 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4335 | if (err) |
4336 | goto fail; | |
4337 | ||
39279cc3 | 4338 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4339 | |
39279cc3 CM |
4340 | btrfs_set_trans_block_group(trans, dir); |
4341 | atomic_inc(&inode->i_count); | |
aec7477b | 4342 | |
00e4e6b3 | 4343 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4344 | |
a5719521 | 4345 | if (err) { |
54aa1f4d | 4346 | drop_inode = 1; |
a5719521 YZ |
4347 | } else { |
4348 | btrfs_update_inode_block_group(trans, dir); | |
4349 | err = btrfs_update_inode(trans, root, inode); | |
4350 | BUG_ON(err); | |
4351 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4352 | } | |
39279cc3 | 4353 | |
d3c2fdcf | 4354 | nr = trans->blocks_used; |
ab78c84d | 4355 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4356 | fail: |
9ed74f2d | 4357 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4358 | if (drop_inode) { |
4359 | inode_dec_link_count(inode); | |
4360 | iput(inode); | |
4361 | } | |
d3c2fdcf | 4362 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4363 | return err; |
4364 | } | |
4365 | ||
39279cc3 CM |
4366 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4367 | { | |
b9d86667 | 4368 | struct inode *inode = NULL; |
39279cc3 CM |
4369 | struct btrfs_trans_handle *trans; |
4370 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4371 | int err = 0; | |
4372 | int drop_on_err = 0; | |
b9d86667 | 4373 | u64 objectid = 0; |
00e4e6b3 | 4374 | u64 index = 0; |
d3c2fdcf | 4375 | unsigned long nr = 1; |
39279cc3 | 4376 | |
9ed74f2d JB |
4377 | /* |
4378 | * 2 items for inode and ref | |
4379 | * 2 items for dir items | |
4380 | * 1 for xattr if selinux is on | |
4381 | */ | |
4382 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4383 | if (err) |
9ed74f2d | 4384 | return err; |
1832a6d5 | 4385 | |
39279cc3 | 4386 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4387 | if (!trans) { |
4388 | err = -ENOMEM; | |
39279cc3 CM |
4389 | goto out_unlock; |
4390 | } | |
9ed74f2d | 4391 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4392 | |
4393 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4394 | if (err) { | |
4395 | err = -ENOSPC; | |
4396 | goto out_unlock; | |
4397 | } | |
4398 | ||
aec7477b | 4399 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4400 | dentry->d_name.len, |
4401 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4402 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4403 | &index); | |
39279cc3 CM |
4404 | if (IS_ERR(inode)) { |
4405 | err = PTR_ERR(inode); | |
4406 | goto out_fail; | |
4407 | } | |
5f39d397 | 4408 | |
39279cc3 | 4409 | drop_on_err = 1; |
33268eaf | 4410 | |
0279b4cd | 4411 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4412 | if (err) |
4413 | goto out_fail; | |
4414 | ||
39279cc3 CM |
4415 | inode->i_op = &btrfs_dir_inode_operations; |
4416 | inode->i_fop = &btrfs_dir_file_operations; | |
4417 | btrfs_set_trans_block_group(trans, inode); | |
4418 | ||
dbe674a9 | 4419 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4420 | err = btrfs_update_inode(trans, root, inode); |
4421 | if (err) | |
4422 | goto out_fail; | |
5f39d397 | 4423 | |
e02119d5 CM |
4424 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4425 | inode, dentry->d_name.name, | |
4426 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4427 | if (err) |
4428 | goto out_fail; | |
5f39d397 | 4429 | |
39279cc3 CM |
4430 | d_instantiate(dentry, inode); |
4431 | drop_on_err = 0; | |
39279cc3 CM |
4432 | btrfs_update_inode_block_group(trans, inode); |
4433 | btrfs_update_inode_block_group(trans, dir); | |
4434 | ||
4435 | out_fail: | |
d3c2fdcf | 4436 | nr = trans->blocks_used; |
ab78c84d | 4437 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4438 | |
39279cc3 | 4439 | out_unlock: |
9ed74f2d | 4440 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4441 | if (drop_on_err) |
4442 | iput(inode); | |
d3c2fdcf | 4443 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4444 | return err; |
4445 | } | |
4446 | ||
d352ac68 CM |
4447 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4448 | * and an extent that you want to insert, deal with overlap and insert | |
4449 | * the new extent into the tree. | |
4450 | */ | |
3b951516 CM |
4451 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4452 | struct extent_map *existing, | |
e6dcd2dc CM |
4453 | struct extent_map *em, |
4454 | u64 map_start, u64 map_len) | |
3b951516 CM |
4455 | { |
4456 | u64 start_diff; | |
3b951516 | 4457 | |
e6dcd2dc CM |
4458 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4459 | start_diff = map_start - em->start; | |
4460 | em->start = map_start; | |
4461 | em->len = map_len; | |
c8b97818 CM |
4462 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4463 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4464 | em->block_start += start_diff; |
c8b97818 CM |
4465 | em->block_len -= start_diff; |
4466 | } | |
e6dcd2dc | 4467 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4468 | } |
4469 | ||
c8b97818 CM |
4470 | static noinline int uncompress_inline(struct btrfs_path *path, |
4471 | struct inode *inode, struct page *page, | |
4472 | size_t pg_offset, u64 extent_offset, | |
4473 | struct btrfs_file_extent_item *item) | |
4474 | { | |
4475 | int ret; | |
4476 | struct extent_buffer *leaf = path->nodes[0]; | |
4477 | char *tmp; | |
4478 | size_t max_size; | |
4479 | unsigned long inline_size; | |
4480 | unsigned long ptr; | |
4481 | ||
4482 | WARN_ON(pg_offset != 0); | |
4483 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4484 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4485 | btrfs_item_nr(leaf, path->slots[0])); | |
4486 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4487 | ptr = btrfs_file_extent_inline_start(item); | |
4488 | ||
4489 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4490 | ||
5b050f04 | 4491 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4492 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4493 | inline_size, max_size); | |
4494 | if (ret) { | |
4495 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4496 | unsigned long copy_size = min_t(u64, | |
4497 | PAGE_CACHE_SIZE - pg_offset, | |
4498 | max_size - extent_offset); | |
4499 | memset(kaddr + pg_offset, 0, copy_size); | |
4500 | kunmap_atomic(kaddr, KM_USER0); | |
4501 | } | |
4502 | kfree(tmp); | |
4503 | return 0; | |
4504 | } | |
4505 | ||
d352ac68 CM |
4506 | /* |
4507 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4508 | * the ugly parts come from merging extents from the disk with the in-ram |
4509 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4510 | * where the in-ram extents might be locked pending data=ordered completion. |
4511 | * | |
4512 | * This also copies inline extents directly into the page. | |
4513 | */ | |
d397712b | 4514 | |
a52d9a80 | 4515 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4516 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4517 | int create) |
4518 | { | |
4519 | int ret; | |
4520 | int err = 0; | |
db94535d | 4521 | u64 bytenr; |
a52d9a80 CM |
4522 | u64 extent_start = 0; |
4523 | u64 extent_end = 0; | |
4524 | u64 objectid = inode->i_ino; | |
4525 | u32 found_type; | |
f421950f | 4526 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4527 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4528 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4529 | struct extent_buffer *leaf; |
4530 | struct btrfs_key found_key; | |
a52d9a80 CM |
4531 | struct extent_map *em = NULL; |
4532 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4533 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4534 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4535 | int compressed; |
a52d9a80 | 4536 | |
a52d9a80 | 4537 | again: |
890871be | 4538 | read_lock(&em_tree->lock); |
d1310b2e | 4539 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4540 | if (em) |
4541 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4542 | read_unlock(&em_tree->lock); |
d1310b2e | 4543 | |
a52d9a80 | 4544 | if (em) { |
e1c4b745 CM |
4545 | if (em->start > start || em->start + em->len <= start) |
4546 | free_extent_map(em); | |
4547 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4548 | free_extent_map(em); |
4549 | else | |
4550 | goto out; | |
a52d9a80 | 4551 | } |
d1310b2e | 4552 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4553 | if (!em) { |
d1310b2e CM |
4554 | err = -ENOMEM; |
4555 | goto out; | |
a52d9a80 | 4556 | } |
e6dcd2dc | 4557 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4558 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4559 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4560 | em->len = (u64)-1; |
c8b97818 | 4561 | em->block_len = (u64)-1; |
f421950f CM |
4562 | |
4563 | if (!path) { | |
4564 | path = btrfs_alloc_path(); | |
4565 | BUG_ON(!path); | |
4566 | } | |
4567 | ||
179e29e4 CM |
4568 | ret = btrfs_lookup_file_extent(trans, root, path, |
4569 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4570 | if (ret < 0) { |
4571 | err = ret; | |
4572 | goto out; | |
4573 | } | |
4574 | ||
4575 | if (ret != 0) { | |
4576 | if (path->slots[0] == 0) | |
4577 | goto not_found; | |
4578 | path->slots[0]--; | |
4579 | } | |
4580 | ||
5f39d397 CM |
4581 | leaf = path->nodes[0]; |
4582 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4583 | struct btrfs_file_extent_item); |
a52d9a80 | 4584 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4585 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4586 | found_type = btrfs_key_type(&found_key); | |
4587 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4588 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4589 | goto not_found; | |
4590 | } | |
4591 | ||
5f39d397 CM |
4592 | found_type = btrfs_file_extent_type(leaf, item); |
4593 | extent_start = found_key.offset; | |
c8b97818 | 4594 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4595 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4596 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4597 | extent_end = extent_start + |
db94535d | 4598 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4599 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4600 | size_t size; | |
4601 | size = btrfs_file_extent_inline_len(leaf, item); | |
4602 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4603 | ~((u64)root->sectorsize - 1); | |
4604 | } | |
4605 | ||
4606 | if (start >= extent_end) { | |
4607 | path->slots[0]++; | |
4608 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4609 | ret = btrfs_next_leaf(root, path); | |
4610 | if (ret < 0) { | |
4611 | err = ret; | |
4612 | goto out; | |
a52d9a80 | 4613 | } |
9036c102 YZ |
4614 | if (ret > 0) |
4615 | goto not_found; | |
4616 | leaf = path->nodes[0]; | |
a52d9a80 | 4617 | } |
9036c102 YZ |
4618 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4619 | if (found_key.objectid != objectid || | |
4620 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4621 | goto not_found; | |
4622 | if (start + len <= found_key.offset) | |
4623 | goto not_found; | |
4624 | em->start = start; | |
4625 | em->len = found_key.offset - start; | |
4626 | goto not_found_em; | |
4627 | } | |
4628 | ||
d899e052 YZ |
4629 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4630 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4631 | em->start = extent_start; |
4632 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4633 | em->orig_start = extent_start - |
4634 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4635 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4636 | if (bytenr == 0) { | |
5f39d397 | 4637 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4638 | goto insert; |
4639 | } | |
c8b97818 CM |
4640 | if (compressed) { |
4641 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4642 | em->block_start = bytenr; | |
4643 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4644 | item); | |
4645 | } else { | |
4646 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4647 | em->block_start = bytenr; | |
4648 | em->block_len = em->len; | |
d899e052 YZ |
4649 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4650 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4651 | } |
a52d9a80 CM |
4652 | goto insert; |
4653 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4654 | unsigned long ptr; |
a52d9a80 | 4655 | char *map; |
3326d1b0 CM |
4656 | size_t size; |
4657 | size_t extent_offset; | |
4658 | size_t copy_size; | |
a52d9a80 | 4659 | |
689f9346 | 4660 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4661 | if (!page || create) { |
689f9346 | 4662 | em->start = extent_start; |
9036c102 | 4663 | em->len = extent_end - extent_start; |
689f9346 Y |
4664 | goto out; |
4665 | } | |
5f39d397 | 4666 | |
9036c102 YZ |
4667 | size = btrfs_file_extent_inline_len(leaf, item); |
4668 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4669 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4670 | size - extent_offset); |
3326d1b0 | 4671 | em->start = extent_start + extent_offset; |
70dec807 CM |
4672 | em->len = (copy_size + root->sectorsize - 1) & |
4673 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4674 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4675 | if (compressed) |
4676 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4677 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4678 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4679 | if (btrfs_file_extent_compression(leaf, item) == |
4680 | BTRFS_COMPRESS_ZLIB) { | |
4681 | ret = uncompress_inline(path, inode, page, | |
4682 | pg_offset, | |
4683 | extent_offset, item); | |
4684 | BUG_ON(ret); | |
4685 | } else { | |
4686 | map = kmap(page); | |
4687 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4688 | copy_size); | |
93c82d57 CM |
4689 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4690 | memset(map + pg_offset + copy_size, 0, | |
4691 | PAGE_CACHE_SIZE - pg_offset - | |
4692 | copy_size); | |
4693 | } | |
c8b97818 CM |
4694 | kunmap(page); |
4695 | } | |
179e29e4 CM |
4696 | flush_dcache_page(page); |
4697 | } else if (create && PageUptodate(page)) { | |
4698 | if (!trans) { | |
4699 | kunmap(page); | |
4700 | free_extent_map(em); | |
4701 | em = NULL; | |
4702 | btrfs_release_path(root, path); | |
f9295749 | 4703 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4704 | goto again; |
4705 | } | |
c8b97818 | 4706 | map = kmap(page); |
70dec807 | 4707 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4708 | copy_size); |
c8b97818 | 4709 | kunmap(page); |
179e29e4 | 4710 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4711 | } |
d1310b2e CM |
4712 | set_extent_uptodate(io_tree, em->start, |
4713 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4714 | goto insert; |
4715 | } else { | |
d397712b | 4716 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4717 | WARN_ON(1); |
4718 | } | |
4719 | not_found: | |
4720 | em->start = start; | |
d1310b2e | 4721 | em->len = len; |
a52d9a80 | 4722 | not_found_em: |
5f39d397 | 4723 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4724 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4725 | insert: |
4726 | btrfs_release_path(root, path); | |
d1310b2e | 4727 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4728 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4729 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4730 | (unsigned long long)em->len, | |
4731 | (unsigned long long)start, | |
4732 | (unsigned long long)len); | |
a52d9a80 CM |
4733 | err = -EIO; |
4734 | goto out; | |
4735 | } | |
d1310b2e CM |
4736 | |
4737 | err = 0; | |
890871be | 4738 | write_lock(&em_tree->lock); |
a52d9a80 | 4739 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4740 | /* it is possible that someone inserted the extent into the tree |
4741 | * while we had the lock dropped. It is also possible that | |
4742 | * an overlapping map exists in the tree | |
4743 | */ | |
a52d9a80 | 4744 | if (ret == -EEXIST) { |
3b951516 | 4745 | struct extent_map *existing; |
e6dcd2dc CM |
4746 | |
4747 | ret = 0; | |
4748 | ||
3b951516 | 4749 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4750 | if (existing && (existing->start > start || |
4751 | existing->start + existing->len <= start)) { | |
4752 | free_extent_map(existing); | |
4753 | existing = NULL; | |
4754 | } | |
3b951516 CM |
4755 | if (!existing) { |
4756 | existing = lookup_extent_mapping(em_tree, em->start, | |
4757 | em->len); | |
4758 | if (existing) { | |
4759 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4760 | em, start, |
4761 | root->sectorsize); | |
3b951516 CM |
4762 | free_extent_map(existing); |
4763 | if (err) { | |
4764 | free_extent_map(em); | |
4765 | em = NULL; | |
4766 | } | |
4767 | } else { | |
4768 | err = -EIO; | |
3b951516 CM |
4769 | free_extent_map(em); |
4770 | em = NULL; | |
4771 | } | |
4772 | } else { | |
4773 | free_extent_map(em); | |
4774 | em = existing; | |
e6dcd2dc | 4775 | err = 0; |
a52d9a80 | 4776 | } |
a52d9a80 | 4777 | } |
890871be | 4778 | write_unlock(&em_tree->lock); |
a52d9a80 | 4779 | out: |
f421950f CM |
4780 | if (path) |
4781 | btrfs_free_path(path); | |
a52d9a80 CM |
4782 | if (trans) { |
4783 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4784 | if (!err) |
a52d9a80 CM |
4785 | err = ret; |
4786 | } | |
a52d9a80 CM |
4787 | if (err) { |
4788 | free_extent_map(em); | |
a52d9a80 CM |
4789 | return ERR_PTR(err); |
4790 | } | |
4791 | return em; | |
4792 | } | |
4793 | ||
16432985 CM |
4794 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4795 | const struct iovec *iov, loff_t offset, | |
4796 | unsigned long nr_segs) | |
4797 | { | |
e1c4b745 | 4798 | return -EINVAL; |
16432985 CM |
4799 | } |
4800 | ||
1506fcc8 YS |
4801 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4802 | __u64 start, __u64 len) | |
4803 | { | |
4804 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4805 | } | |
4806 | ||
a52d9a80 | 4807 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4808 | { |
d1310b2e CM |
4809 | struct extent_io_tree *tree; |
4810 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4811 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4812 | } |
1832a6d5 | 4813 | |
a52d9a80 | 4814 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4815 | { |
d1310b2e | 4816 | struct extent_io_tree *tree; |
b888db2b CM |
4817 | |
4818 | ||
4819 | if (current->flags & PF_MEMALLOC) { | |
4820 | redirty_page_for_writepage(wbc, page); | |
4821 | unlock_page(page); | |
4822 | return 0; | |
4823 | } | |
d1310b2e | 4824 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4825 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4826 | } |
4827 | ||
f421950f CM |
4828 | int btrfs_writepages(struct address_space *mapping, |
4829 | struct writeback_control *wbc) | |
b293f02e | 4830 | { |
d1310b2e | 4831 | struct extent_io_tree *tree; |
771ed689 | 4832 | |
d1310b2e | 4833 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4834 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4835 | } | |
4836 | ||
3ab2fb5a CM |
4837 | static int |
4838 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4839 | struct list_head *pages, unsigned nr_pages) | |
4840 | { | |
d1310b2e CM |
4841 | struct extent_io_tree *tree; |
4842 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4843 | return extent_readpages(tree, mapping, pages, nr_pages, |
4844 | btrfs_get_extent); | |
4845 | } | |
e6dcd2dc | 4846 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4847 | { |
d1310b2e CM |
4848 | struct extent_io_tree *tree; |
4849 | struct extent_map_tree *map; | |
a52d9a80 | 4850 | int ret; |
8c2383c3 | 4851 | |
d1310b2e CM |
4852 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4853 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4854 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4855 | if (ret == 1) { |
4856 | ClearPagePrivate(page); | |
4857 | set_page_private(page, 0); | |
4858 | page_cache_release(page); | |
39279cc3 | 4859 | } |
a52d9a80 | 4860 | return ret; |
39279cc3 CM |
4861 | } |
4862 | ||
e6dcd2dc CM |
4863 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4864 | { | |
98509cfc CM |
4865 | if (PageWriteback(page) || PageDirty(page)) |
4866 | return 0; | |
b335b003 | 4867 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4868 | } |
4869 | ||
a52d9a80 | 4870 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4871 | { |
d1310b2e | 4872 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4873 | struct btrfs_ordered_extent *ordered; |
4874 | u64 page_start = page_offset(page); | |
4875 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4876 | |
8b62b72b CM |
4877 | |
4878 | /* | |
4879 | * we have the page locked, so new writeback can't start, | |
4880 | * and the dirty bit won't be cleared while we are here. | |
4881 | * | |
4882 | * Wait for IO on this page so that we can safely clear | |
4883 | * the PagePrivate2 bit and do ordered accounting | |
4884 | */ | |
e6dcd2dc | 4885 | wait_on_page_writeback(page); |
8b62b72b | 4886 | |
d1310b2e | 4887 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4888 | if (offset) { |
4889 | btrfs_releasepage(page, GFP_NOFS); | |
4890 | return; | |
4891 | } | |
e6dcd2dc CM |
4892 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4893 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4894 | page_offset(page)); | |
4895 | if (ordered) { | |
eb84ae03 CM |
4896 | /* |
4897 | * IO on this page will never be started, so we need | |
4898 | * to account for any ordered extents now | |
4899 | */ | |
e6dcd2dc CM |
4900 | clear_extent_bit(tree, page_start, page_end, |
4901 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff JB |
4902 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
4903 | NULL, GFP_NOFS); | |
8b62b72b CM |
4904 | /* |
4905 | * whoever cleared the private bit is responsible | |
4906 | * for the finish_ordered_io | |
4907 | */ | |
4908 | if (TestClearPagePrivate2(page)) { | |
4909 | btrfs_finish_ordered_io(page->mapping->host, | |
4910 | page_start, page_end); | |
4911 | } | |
e6dcd2dc CM |
4912 | btrfs_put_ordered_extent(ordered); |
4913 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4914 | } | |
4915 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff JB |
4916 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4917 | EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); | |
e6dcd2dc CM |
4918 | __btrfs_releasepage(page, GFP_NOFS); |
4919 | ||
4a096752 | 4920 | ClearPageChecked(page); |
9ad6b7bc | 4921 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4922 | ClearPagePrivate(page); |
4923 | set_page_private(page, 0); | |
4924 | page_cache_release(page); | |
4925 | } | |
39279cc3 CM |
4926 | } |
4927 | ||
9ebefb18 CM |
4928 | /* |
4929 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4930 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4931 | * be careful to check for EOF conditions here. We set the page up correctly | |
4932 | * for a written page which means we get ENOSPC checking when writing into | |
4933 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4934 | * support these features. | |
4935 | * | |
4936 | * We are not allowed to take the i_mutex here so we have to play games to | |
4937 | * protect against truncate races as the page could now be beyond EOF. Because | |
4938 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4939 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4940 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4941 | * unlock the page. | |
4942 | */ | |
c2ec175c | 4943 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4944 | { |
c2ec175c | 4945 | struct page *page = vmf->page; |
6da6abae | 4946 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4947 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4948 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4949 | struct btrfs_ordered_extent *ordered; | |
4950 | char *kaddr; | |
4951 | unsigned long zero_start; | |
9ebefb18 | 4952 | loff_t size; |
1832a6d5 | 4953 | int ret; |
a52d9a80 | 4954 | u64 page_start; |
e6dcd2dc | 4955 | u64 page_end; |
9ebefb18 | 4956 | |
6a63209f | 4957 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4958 | if (ret) { |
4959 | if (ret == -ENOMEM) | |
4960 | ret = VM_FAULT_OOM; | |
4961 | else /* -ENOSPC, -EIO, etc */ | |
4962 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4963 | goto out; |
56a76f82 | 4964 | } |
1832a6d5 | 4965 | |
9ed74f2d JB |
4966 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
4967 | if (ret) { | |
4968 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
4969 | ret = VM_FAULT_SIGBUS; | |
4970 | goto out; | |
4971 | } | |
4972 | ||
56a76f82 | 4973 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4974 | again: |
9ebefb18 | 4975 | lock_page(page); |
9ebefb18 | 4976 | size = i_size_read(inode); |
e6dcd2dc CM |
4977 | page_start = page_offset(page); |
4978 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4979 | |
9ebefb18 | 4980 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4981 | (page_start >= size)) { |
6a63209f | 4982 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
4983 | /* page got truncated out from underneath us */ |
4984 | goto out_unlock; | |
4985 | } | |
e6dcd2dc CM |
4986 | wait_on_page_writeback(page); |
4987 | ||
4988 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4989 | set_page_extent_mapped(page); | |
4990 | ||
eb84ae03 CM |
4991 | /* |
4992 | * we can't set the delalloc bits if there are pending ordered | |
4993 | * extents. Drop our locks and wait for them to finish | |
4994 | */ | |
e6dcd2dc CM |
4995 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4996 | if (ordered) { | |
4997 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4998 | unlock_page(page); | |
eb84ae03 | 4999 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5000 | btrfs_put_ordered_extent(ordered); |
5001 | goto again; | |
5002 | } | |
5003 | ||
fbf19087 JB |
5004 | /* |
5005 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
5006 | * if it was already dirty, so for space accounting reasons we need to | |
5007 | * clear any delalloc bits for the range we are fixing to save. There | |
5008 | * is probably a better way to do this, but for now keep consistent with | |
5009 | * prepare_pages in the normal write path. | |
5010 | */ | |
5011 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
32c00aff JB |
5012 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
5013 | GFP_NOFS); | |
fbf19087 | 5014 | |
9ed74f2d JB |
5015 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
5016 | if (ret) { | |
5017 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5018 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 5019 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
5020 | goto out_unlock; |
5021 | } | |
e6dcd2dc | 5022 | ret = 0; |
9ebefb18 CM |
5023 | |
5024 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 5025 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 5026 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 5027 | else |
e6dcd2dc | 5028 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 5029 | |
e6dcd2dc CM |
5030 | if (zero_start != PAGE_CACHE_SIZE) { |
5031 | kaddr = kmap(page); | |
5032 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
5033 | flush_dcache_page(page); | |
5034 | kunmap(page); | |
5035 | } | |
247e743c | 5036 | ClearPageChecked(page); |
e6dcd2dc | 5037 | set_page_dirty(page); |
50a9b214 | 5038 | SetPageUptodate(page); |
5a3f23d5 | 5039 | |
257c62e1 CM |
5040 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5041 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5042 | ||
e6dcd2dc | 5043 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
5044 | |
5045 | out_unlock: | |
9ed74f2d | 5046 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
5047 | if (!ret) |
5048 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5049 | unlock_page(page); |
1832a6d5 | 5050 | out: |
9ebefb18 CM |
5051 | return ret; |
5052 | } | |
5053 | ||
39279cc3 CM |
5054 | static void btrfs_truncate(struct inode *inode) |
5055 | { | |
5056 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5057 | int ret; | |
5058 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5059 | unsigned long nr; |
dbe674a9 | 5060 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
5061 | |
5062 | if (!S_ISREG(inode->i_mode)) | |
5063 | return; | |
5064 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
5065 | return; | |
5066 | ||
5d5e103a JB |
5067 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5068 | if (ret) | |
5069 | return; | |
4a096752 | 5070 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 5071 | |
39279cc3 | 5072 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
5073 | |
5074 | /* | |
5075 | * setattr is responsible for setting the ordered_data_close flag, | |
5076 | * but that is only tested during the last file release. That | |
5077 | * could happen well after the next commit, leaving a great big | |
5078 | * window where new writes may get lost if someone chooses to write | |
5079 | * to this file after truncating to zero | |
5080 | * | |
5081 | * The inode doesn't have any dirty data here, and so if we commit | |
5082 | * this is a noop. If someone immediately starts writing to the inode | |
5083 | * it is very likely we'll catch some of their writes in this | |
5084 | * transaction, and the commit will find this file on the ordered | |
5085 | * data list with good things to send down. | |
5086 | * | |
5087 | * This is a best effort solution, there is still a window where | |
5088 | * using truncate to replace the contents of the file will | |
5089 | * end up with a zero length file after a crash. | |
5090 | */ | |
5091 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5092 | btrfs_add_ordered_operation(trans, root, inode); | |
5093 | ||
39279cc3 | 5094 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 5095 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 5096 | |
7b128766 JB |
5097 | ret = btrfs_orphan_add(trans, inode); |
5098 | if (ret) | |
5099 | goto out; | |
39279cc3 | 5100 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 5101 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 5102 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 5103 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 5104 | |
7b128766 JB |
5105 | ret = btrfs_orphan_del(trans, inode); |
5106 | BUG_ON(ret); | |
5107 | ||
5108 | out: | |
5109 | nr = trans->blocks_used; | |
89ce8a63 | 5110 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5111 | BUG_ON(ret); |
d3c2fdcf | 5112 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5113 | } |
5114 | ||
d352ac68 CM |
5115 | /* |
5116 | * create a new subvolume directory/inode (helper for the ioctl). | |
5117 | */ | |
d2fb3437 | 5118 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5119 | struct btrfs_root *new_root, |
d2fb3437 | 5120 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5121 | { |
39279cc3 | 5122 | struct inode *inode; |
76dda93c | 5123 | int err; |
00e4e6b3 | 5124 | u64 index = 0; |
39279cc3 | 5125 | |
aec7477b | 5126 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5127 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5128 | if (IS_ERR(inode)) |
f46b5a66 | 5129 | return PTR_ERR(inode); |
39279cc3 CM |
5130 | inode->i_op = &btrfs_dir_inode_operations; |
5131 | inode->i_fop = &btrfs_dir_file_operations; | |
5132 | ||
39279cc3 | 5133 | inode->i_nlink = 1; |
dbe674a9 | 5134 | btrfs_i_size_write(inode, 0); |
3b96362c | 5135 | |
76dda93c YZ |
5136 | err = btrfs_update_inode(trans, new_root, inode); |
5137 | BUG_ON(err); | |
cb8e7090 | 5138 | |
76dda93c | 5139 | iput(inode); |
cb8e7090 | 5140 | return 0; |
39279cc3 CM |
5141 | } |
5142 | ||
d352ac68 CM |
5143 | /* helper function for file defrag and space balancing. This |
5144 | * forces readahead on a given range of bytes in an inode | |
5145 | */ | |
edbd8d4e | 5146 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5147 | struct file_ra_state *ra, struct file *file, |
5148 | pgoff_t offset, pgoff_t last_index) | |
5149 | { | |
8e7bf94f | 5150 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5151 | |
86479a04 CM |
5152 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5153 | return offset + req_size; | |
86479a04 CM |
5154 | } |
5155 | ||
39279cc3 CM |
5156 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5157 | { | |
5158 | struct btrfs_inode *ei; | |
5159 | ||
5160 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5161 | if (!ei) | |
5162 | return NULL; | |
15ee9bc7 | 5163 | ei->last_trans = 0; |
257c62e1 | 5164 | ei->last_sub_trans = 0; |
e02119d5 | 5165 | ei->logged_trans = 0; |
32c00aff JB |
5166 | ei->outstanding_extents = 0; |
5167 | ei->reserved_extents = 0; | |
5168 | spin_lock_init(&ei->accounting_lock); | |
e6dcd2dc | 5169 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5170 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5171 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5172 | return &ei->vfs_inode; |
5173 | } | |
5174 | ||
5175 | void btrfs_destroy_inode(struct inode *inode) | |
5176 | { | |
e6dcd2dc | 5177 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5178 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5179 | ||
39279cc3 CM |
5180 | WARN_ON(!list_empty(&inode->i_dentry)); |
5181 | WARN_ON(inode->i_data.nrpages); | |
5182 | ||
5a3f23d5 CM |
5183 | /* |
5184 | * Make sure we're properly removed from the ordered operation | |
5185 | * lists. | |
5186 | */ | |
5187 | smp_mb(); | |
5188 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5189 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5190 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5191 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5192 | } | |
5193 | ||
5194 | spin_lock(&root->list_lock); | |
7b128766 JB |
5195 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
5196 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
5197 | " list\n", inode->i_ino); | |
5198 | dump_stack(); | |
5199 | } | |
5a3f23d5 | 5200 | spin_unlock(&root->list_lock); |
7b128766 | 5201 | |
d397712b | 5202 | while (1) { |
e6dcd2dc CM |
5203 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5204 | if (!ordered) | |
5205 | break; | |
5206 | else { | |
d397712b CM |
5207 | printk(KERN_ERR "btrfs found ordered " |
5208 | "extent %llu %llu on inode cleanup\n", | |
5209 | (unsigned long long)ordered->file_offset, | |
5210 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5211 | btrfs_remove_ordered_extent(inode, ordered); |
5212 | btrfs_put_ordered_extent(ordered); | |
5213 | btrfs_put_ordered_extent(ordered); | |
5214 | } | |
5215 | } | |
5d4f98a2 | 5216 | inode_tree_del(inode); |
5b21f2ed | 5217 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
5218 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5219 | } | |
5220 | ||
76dda93c YZ |
5221 | void btrfs_drop_inode(struct inode *inode) |
5222 | { | |
5223 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5224 | ||
5225 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5226 | generic_delete_inode(inode); | |
5227 | else | |
5228 | generic_drop_inode(inode); | |
5229 | } | |
5230 | ||
0ee0fda0 | 5231 | static void init_once(void *foo) |
39279cc3 CM |
5232 | { |
5233 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5234 | ||
5235 | inode_init_once(&ei->vfs_inode); | |
5236 | } | |
5237 | ||
5238 | void btrfs_destroy_cachep(void) | |
5239 | { | |
5240 | if (btrfs_inode_cachep) | |
5241 | kmem_cache_destroy(btrfs_inode_cachep); | |
5242 | if (btrfs_trans_handle_cachep) | |
5243 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5244 | if (btrfs_transaction_cachep) | |
5245 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5246 | if (btrfs_path_cachep) |
5247 | kmem_cache_destroy(btrfs_path_cachep); | |
5248 | } | |
5249 | ||
5250 | int btrfs_init_cachep(void) | |
5251 | { | |
9601e3f6 CH |
5252 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5253 | sizeof(struct btrfs_inode), 0, | |
5254 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5255 | if (!btrfs_inode_cachep) |
5256 | goto fail; | |
9601e3f6 CH |
5257 | |
5258 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5259 | sizeof(struct btrfs_trans_handle), 0, | |
5260 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5261 | if (!btrfs_trans_handle_cachep) |
5262 | goto fail; | |
9601e3f6 CH |
5263 | |
5264 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5265 | sizeof(struct btrfs_transaction), 0, | |
5266 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5267 | if (!btrfs_transaction_cachep) |
5268 | goto fail; | |
9601e3f6 CH |
5269 | |
5270 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5271 | sizeof(struct btrfs_path), 0, | |
5272 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5273 | if (!btrfs_path_cachep) |
5274 | goto fail; | |
9601e3f6 | 5275 | |
39279cc3 CM |
5276 | return 0; |
5277 | fail: | |
5278 | btrfs_destroy_cachep(); | |
5279 | return -ENOMEM; | |
5280 | } | |
5281 | ||
5282 | static int btrfs_getattr(struct vfsmount *mnt, | |
5283 | struct dentry *dentry, struct kstat *stat) | |
5284 | { | |
5285 | struct inode *inode = dentry->d_inode; | |
5286 | generic_fillattr(inode, stat); | |
3394e160 | 5287 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5288 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5289 | stat->blocks = (inode_get_bytes(inode) + |
5290 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5291 | return 0; |
5292 | } | |
5293 | ||
d397712b CM |
5294 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5295 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5296 | { |
5297 | struct btrfs_trans_handle *trans; | |
5298 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5299 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5300 | struct inode *new_inode = new_dentry->d_inode; |
5301 | struct inode *old_inode = old_dentry->d_inode; | |
5302 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5303 | u64 index = 0; |
4df27c4d | 5304 | u64 root_objectid; |
39279cc3 CM |
5305 | int ret; |
5306 | ||
f679a840 YZ |
5307 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5308 | return -EPERM; | |
5309 | ||
4df27c4d YZ |
5310 | /* we only allow rename subvolume link between subvolumes */ |
5311 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5312 | return -EXDEV; |
5313 | ||
4df27c4d YZ |
5314 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5315 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5316 | return -ENOTEMPTY; |
5f39d397 | 5317 | |
4df27c4d YZ |
5318 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5319 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5320 | return -ENOTEMPTY; | |
0660b5af | 5321 | |
9ed74f2d | 5322 | /* |
5df6a9f6 JB |
5323 | * We want to reserve the absolute worst case amount of items. So if |
5324 | * both inodes are subvols and we need to unlink them then that would | |
5325 | * require 4 item modifications, but if they are both normal inodes it | |
5326 | * would require 5 item modifications, so we'll assume their normal | |
5327 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
5328 | * should cover the worst case number of items we'll modify. | |
9ed74f2d | 5329 | */ |
5df6a9f6 | 5330 | ret = btrfs_reserve_metadata_space(root, 11); |
1832a6d5 | 5331 | if (ret) |
4df27c4d | 5332 | return ret; |
1832a6d5 | 5333 | |
5a3f23d5 CM |
5334 | /* |
5335 | * we're using rename to replace one file with another. | |
5336 | * and the replacement file is large. Start IO on it now so | |
5337 | * we don't add too much work to the end of the transaction | |
5338 | */ | |
4baf8c92 | 5339 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5340 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5341 | filemap_flush(old_inode->i_mapping); | |
5342 | ||
76dda93c YZ |
5343 | /* close the racy window with snapshot create/destroy ioctl */ |
5344 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5345 | down_read(&root->fs_info->subvol_sem); | |
5346 | ||
39279cc3 | 5347 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5348 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5349 | |
4df27c4d YZ |
5350 | if (dest != root) |
5351 | btrfs_record_root_in_trans(trans, dest); | |
5352 | ||
a5719521 YZ |
5353 | ret = btrfs_set_inode_index(new_dir, &index); |
5354 | if (ret) | |
5355 | goto out_fail; | |
5a3f23d5 | 5356 | |
a5719521 | 5357 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5358 | /* force full log commit if subvolume involved. */ |
5359 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5360 | } else { | |
a5719521 YZ |
5361 | ret = btrfs_insert_inode_ref(trans, dest, |
5362 | new_dentry->d_name.name, | |
5363 | new_dentry->d_name.len, | |
5364 | old_inode->i_ino, | |
5365 | new_dir->i_ino, index); | |
5366 | if (ret) | |
5367 | goto out_fail; | |
4df27c4d YZ |
5368 | /* |
5369 | * this is an ugly little race, but the rename is required | |
5370 | * to make sure that if we crash, the inode is either at the | |
5371 | * old name or the new one. pinning the log transaction lets | |
5372 | * us make sure we don't allow a log commit to come in after | |
5373 | * we unlink the name but before we add the new name back in. | |
5374 | */ | |
5375 | btrfs_pin_log_trans(root); | |
5376 | } | |
a5719521 YZ |
5377 | /* |
5378 | * make sure the inode gets flushed if it is replacing | |
5379 | * something. | |
5380 | */ | |
5381 | if (new_inode && new_inode->i_size && | |
5382 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5383 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5384 | } | |
39279cc3 | 5385 | |
39279cc3 CM |
5386 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5387 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5388 | old_inode->i_ctime = ctime; | |
5f39d397 | 5389 | |
12fcfd22 CM |
5390 | if (old_dentry->d_parent != new_dentry->d_parent) |
5391 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5392 | ||
4df27c4d YZ |
5393 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5394 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5395 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5396 | old_dentry->d_name.name, | |
5397 | old_dentry->d_name.len); | |
5398 | } else { | |
5399 | btrfs_inc_nlink(old_dentry->d_inode); | |
5400 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5401 | old_dentry->d_inode, | |
5402 | old_dentry->d_name.name, | |
5403 | old_dentry->d_name.len); | |
5404 | } | |
5405 | BUG_ON(ret); | |
39279cc3 CM |
5406 | |
5407 | if (new_inode) { | |
5408 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5409 | if (unlikely(new_inode->i_ino == |
5410 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5411 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5412 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5413 | root_objectid, | |
5414 | new_dentry->d_name.name, | |
5415 | new_dentry->d_name.len); | |
5416 | BUG_ON(new_inode->i_nlink == 0); | |
5417 | } else { | |
5418 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5419 | new_dentry->d_inode, | |
5420 | new_dentry->d_name.name, | |
5421 | new_dentry->d_name.len); | |
5422 | } | |
5423 | BUG_ON(ret); | |
7b128766 | 5424 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5425 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5426 | BUG_ON(ret); |
7b128766 | 5427 | } |
39279cc3 | 5428 | } |
aec7477b | 5429 | |
4df27c4d YZ |
5430 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5431 | new_dentry->d_name.name, | |
a5719521 | 5432 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5433 | BUG_ON(ret); |
39279cc3 | 5434 | |
4df27c4d YZ |
5435 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5436 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5437 | new_dentry->d_parent); | |
5438 | btrfs_end_log_trans(root); | |
5439 | } | |
a5719521 | 5440 | out_fail: |
ab78c84d | 5441 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5442 | |
76dda93c YZ |
5443 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5444 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 5445 | |
5df6a9f6 | 5446 | btrfs_unreserve_metadata_space(root, 11); |
39279cc3 CM |
5447 | return ret; |
5448 | } | |
5449 | ||
d352ac68 CM |
5450 | /* |
5451 | * some fairly slow code that needs optimization. This walks the list | |
5452 | * of all the inodes with pending delalloc and forces them to disk. | |
5453 | */ | |
ea8c2819 CM |
5454 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5455 | { | |
5456 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5457 | struct btrfs_inode *binode; | |
5b21f2ed | 5458 | struct inode *inode; |
ea8c2819 | 5459 | |
c146afad YZ |
5460 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5461 | return -EROFS; | |
5462 | ||
75eff68e | 5463 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5464 | while (!list_empty(head)) { |
ea8c2819 CM |
5465 | binode = list_entry(head->next, struct btrfs_inode, |
5466 | delalloc_inodes); | |
5b21f2ed ZY |
5467 | inode = igrab(&binode->vfs_inode); |
5468 | if (!inode) | |
5469 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5470 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5471 | if (inode) { |
8c8bee1d | 5472 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5473 | iput(inode); |
5474 | } | |
5475 | cond_resched(); | |
75eff68e | 5476 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5477 | } |
75eff68e | 5478 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5479 | |
5480 | /* the filemap_flush will queue IO into the worker threads, but | |
5481 | * we have to make sure the IO is actually started and that | |
5482 | * ordered extents get created before we return | |
5483 | */ | |
5484 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5485 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5486 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5487 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5488 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5489 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5490 | } |
5491 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5492 | return 0; |
5493 | } | |
5494 | ||
39279cc3 CM |
5495 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5496 | const char *symname) | |
5497 | { | |
5498 | struct btrfs_trans_handle *trans; | |
5499 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5500 | struct btrfs_path *path; | |
5501 | struct btrfs_key key; | |
1832a6d5 | 5502 | struct inode *inode = NULL; |
39279cc3 CM |
5503 | int err; |
5504 | int drop_inode = 0; | |
5505 | u64 objectid; | |
00e4e6b3 | 5506 | u64 index = 0 ; |
39279cc3 CM |
5507 | int name_len; |
5508 | int datasize; | |
5f39d397 | 5509 | unsigned long ptr; |
39279cc3 | 5510 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5511 | struct extent_buffer *leaf; |
1832a6d5 | 5512 | unsigned long nr = 0; |
39279cc3 CM |
5513 | |
5514 | name_len = strlen(symname) + 1; | |
5515 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5516 | return -ENAMETOOLONG; | |
1832a6d5 | 5517 | |
9ed74f2d JB |
5518 | /* |
5519 | * 2 items for inode item and ref | |
5520 | * 2 items for dir items | |
5521 | * 1 item for xattr if selinux is on | |
5522 | */ | |
5523 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5524 | if (err) |
9ed74f2d | 5525 | return err; |
1832a6d5 | 5526 | |
39279cc3 | 5527 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5528 | if (!trans) |
5529 | goto out_fail; | |
39279cc3 CM |
5530 | btrfs_set_trans_block_group(trans, dir); |
5531 | ||
5532 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5533 | if (err) { | |
5534 | err = -ENOSPC; | |
5535 | goto out_unlock; | |
5536 | } | |
5537 | ||
aec7477b | 5538 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5539 | dentry->d_name.len, |
5540 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5541 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5542 | &index); | |
39279cc3 CM |
5543 | err = PTR_ERR(inode); |
5544 | if (IS_ERR(inode)) | |
5545 | goto out_unlock; | |
5546 | ||
0279b4cd | 5547 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5548 | if (err) { |
5549 | drop_inode = 1; | |
5550 | goto out_unlock; | |
5551 | } | |
5552 | ||
39279cc3 | 5553 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5554 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5555 | if (err) |
5556 | drop_inode = 1; | |
5557 | else { | |
5558 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5559 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5560 | inode->i_fop = &btrfs_file_operations; |
5561 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5562 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5563 | } |
39279cc3 CM |
5564 | btrfs_update_inode_block_group(trans, inode); |
5565 | btrfs_update_inode_block_group(trans, dir); | |
5566 | if (drop_inode) | |
5567 | goto out_unlock; | |
5568 | ||
5569 | path = btrfs_alloc_path(); | |
5570 | BUG_ON(!path); | |
5571 | key.objectid = inode->i_ino; | |
5572 | key.offset = 0; | |
39279cc3 CM |
5573 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5574 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5575 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5576 | datasize); | |
54aa1f4d CM |
5577 | if (err) { |
5578 | drop_inode = 1; | |
5579 | goto out_unlock; | |
5580 | } | |
5f39d397 CM |
5581 | leaf = path->nodes[0]; |
5582 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5583 | struct btrfs_file_extent_item); | |
5584 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5585 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5586 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5587 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5588 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5589 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5590 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5591 | ||
39279cc3 | 5592 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5593 | write_extent_buffer(leaf, symname, ptr, name_len); |
5594 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5595 | btrfs_free_path(path); |
5f39d397 | 5596 | |
39279cc3 CM |
5597 | inode->i_op = &btrfs_symlink_inode_operations; |
5598 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5599 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5600 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5601 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5602 | err = btrfs_update_inode(trans, root, inode); |
5603 | if (err) | |
5604 | drop_inode = 1; | |
39279cc3 CM |
5605 | |
5606 | out_unlock: | |
d3c2fdcf | 5607 | nr = trans->blocks_used; |
ab78c84d | 5608 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5609 | out_fail: |
9ed74f2d | 5610 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5611 | if (drop_inode) { |
5612 | inode_dec_link_count(inode); | |
5613 | iput(inode); | |
5614 | } | |
d3c2fdcf | 5615 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5616 | return err; |
5617 | } | |
16432985 | 5618 | |
546888da CM |
5619 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5620 | struct inode *inode, u64 start, u64 end, | |
e980b50c | 5621 | u64 locked_end, u64 alloc_hint, int mode) |
d899e052 | 5622 | { |
d899e052 YZ |
5623 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5624 | struct btrfs_key ins; | |
5625 | u64 alloc_size; | |
5626 | u64 cur_offset = start; | |
5627 | u64 num_bytes = end - start; | |
5628 | int ret = 0; | |
5629 | ||
d899e052 YZ |
5630 | while (num_bytes > 0) { |
5631 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d JB |
5632 | |
5633 | ret = btrfs_reserve_metadata_space(root, 1); | |
5634 | if (ret) | |
5635 | goto out; | |
5636 | ||
d899e052 YZ |
5637 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5638 | root->sectorsize, 0, alloc_hint, | |
5639 | (u64)-1, &ins, 1); | |
5640 | if (ret) { | |
5641 | WARN_ON(1); | |
5642 | goto out; | |
5643 | } | |
5644 | ret = insert_reserved_file_extent(trans, inode, | |
5645 | cur_offset, ins.objectid, | |
5646 | ins.offset, ins.offset, | |
e980b50c CM |
5647 | ins.offset, locked_end, |
5648 | 0, 0, 0, | |
d899e052 YZ |
5649 | BTRFS_FILE_EXTENT_PREALLOC); |
5650 | BUG_ON(ret); | |
a1ed835e CM |
5651 | btrfs_drop_extent_cache(inode, cur_offset, |
5652 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5653 | num_bytes -= ins.offset; |
5654 | cur_offset += ins.offset; | |
5655 | alloc_hint = ins.objectid + ins.offset; | |
9ed74f2d | 5656 | btrfs_unreserve_metadata_space(root, 1); |
d899e052 YZ |
5657 | } |
5658 | out: | |
5659 | if (cur_offset > start) { | |
5660 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5661 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5662 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5663 | cur_offset > i_size_read(inode)) | |
5664 | btrfs_i_size_write(inode, cur_offset); | |
5665 | ret = btrfs_update_inode(trans, root, inode); | |
5666 | BUG_ON(ret); | |
5667 | } | |
5668 | ||
d899e052 YZ |
5669 | return ret; |
5670 | } | |
5671 | ||
5672 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5673 | loff_t offset, loff_t len) | |
5674 | { | |
5675 | u64 cur_offset; | |
5676 | u64 last_byte; | |
5677 | u64 alloc_start; | |
5678 | u64 alloc_end; | |
5679 | u64 alloc_hint = 0; | |
e980b50c | 5680 | u64 locked_end; |
d899e052 YZ |
5681 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5682 | struct extent_map *em; | |
546888da | 5683 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5684 | struct btrfs_root *root; |
d899e052 YZ |
5685 | int ret; |
5686 | ||
5687 | alloc_start = offset & ~mask; | |
5688 | alloc_end = (offset + len + mask) & ~mask; | |
5689 | ||
546888da CM |
5690 | /* |
5691 | * wait for ordered IO before we have any locks. We'll loop again | |
5692 | * below with the locks held. | |
5693 | */ | |
5694 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5695 | ||
d899e052 YZ |
5696 | mutex_lock(&inode->i_mutex); |
5697 | if (alloc_start > inode->i_size) { | |
5698 | ret = btrfs_cont_expand(inode, alloc_start); | |
5699 | if (ret) | |
5700 | goto out; | |
5701 | } | |
5702 | ||
a970b0a1 JB |
5703 | root = BTRFS_I(inode)->root; |
5704 | ||
5705 | ret = btrfs_check_data_free_space(root, inode, | |
5706 | alloc_end - alloc_start); | |
5707 | if (ret) | |
5708 | goto out; | |
5709 | ||
e980b50c | 5710 | locked_end = alloc_end - 1; |
d899e052 YZ |
5711 | while (1) { |
5712 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5713 | |
5714 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5715 | if (!trans) { | |
5716 | ret = -EIO; | |
a970b0a1 | 5717 | goto out_free; |
546888da CM |
5718 | } |
5719 | ||
5720 | /* the extent lock is ordered inside the running | |
5721 | * transaction | |
5722 | */ | |
e980b50c CM |
5723 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5724 | GFP_NOFS); | |
d899e052 YZ |
5725 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5726 | alloc_end - 1); | |
5727 | if (ordered && | |
5728 | ordered->file_offset + ordered->len > alloc_start && | |
5729 | ordered->file_offset < alloc_end) { | |
5730 | btrfs_put_ordered_extent(ordered); | |
5731 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5732 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5733 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5734 | ||
5735 | /* | |
5736 | * we can't wait on the range with the transaction | |
5737 | * running or with the extent lock held | |
5738 | */ | |
d899e052 YZ |
5739 | btrfs_wait_ordered_range(inode, alloc_start, |
5740 | alloc_end - alloc_start); | |
5741 | } else { | |
5742 | if (ordered) | |
5743 | btrfs_put_ordered_extent(ordered); | |
5744 | break; | |
5745 | } | |
5746 | } | |
5747 | ||
5748 | cur_offset = alloc_start; | |
5749 | while (1) { | |
5750 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5751 | alloc_end - cur_offset, 0); | |
5752 | BUG_ON(IS_ERR(em) || !em); | |
5753 | last_byte = min(extent_map_end(em), alloc_end); | |
5754 | last_byte = (last_byte + mask) & ~mask; | |
5755 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5756 | ret = prealloc_file_range(trans, inode, cur_offset, |
e980b50c CM |
5757 | last_byte, locked_end + 1, |
5758 | alloc_hint, mode); | |
d899e052 YZ |
5759 | if (ret < 0) { |
5760 | free_extent_map(em); | |
5761 | break; | |
5762 | } | |
5763 | } | |
5764 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5765 | alloc_hint = em->block_start; | |
5766 | free_extent_map(em); | |
5767 | ||
5768 | cur_offset = last_byte; | |
5769 | if (cur_offset >= alloc_end) { | |
5770 | ret = 0; | |
5771 | break; | |
5772 | } | |
5773 | } | |
e980b50c | 5774 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5775 | GFP_NOFS); |
546888da CM |
5776 | |
5777 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5778 | out_free: |
5779 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5780 | out: |
5781 | mutex_unlock(&inode->i_mutex); | |
5782 | return ret; | |
5783 | } | |
5784 | ||
e6dcd2dc CM |
5785 | static int btrfs_set_page_dirty(struct page *page) |
5786 | { | |
e6dcd2dc CM |
5787 | return __set_page_dirty_nobuffers(page); |
5788 | } | |
5789 | ||
0ee0fda0 | 5790 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5791 | { |
6cbff00f | 5792 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5793 | return -EACCES; |
33268eaf | 5794 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5795 | } |
39279cc3 CM |
5796 | |
5797 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 5798 | .getattr = btrfs_getattr, |
39279cc3 CM |
5799 | .lookup = btrfs_lookup, |
5800 | .create = btrfs_create, | |
5801 | .unlink = btrfs_unlink, | |
5802 | .link = btrfs_link, | |
5803 | .mkdir = btrfs_mkdir, | |
5804 | .rmdir = btrfs_rmdir, | |
5805 | .rename = btrfs_rename, | |
5806 | .symlink = btrfs_symlink, | |
5807 | .setattr = btrfs_setattr, | |
618e21d5 | 5808 | .mknod = btrfs_mknod, |
95819c05 CH |
5809 | .setxattr = btrfs_setxattr, |
5810 | .getxattr = btrfs_getxattr, | |
5103e947 | 5811 | .listxattr = btrfs_listxattr, |
95819c05 | 5812 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5813 | .permission = btrfs_permission, |
39279cc3 | 5814 | }; |
39279cc3 CM |
5815 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
5816 | .lookup = btrfs_lookup, | |
fdebe2bd | 5817 | .permission = btrfs_permission, |
39279cc3 | 5818 | }; |
76dda93c | 5819 | |
39279cc3 CM |
5820 | static struct file_operations btrfs_dir_file_operations = { |
5821 | .llseek = generic_file_llseek, | |
5822 | .read = generic_read_dir, | |
cbdf5a24 | 5823 | .readdir = btrfs_real_readdir, |
34287aa3 | 5824 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5825 | #ifdef CONFIG_COMPAT |
34287aa3 | 5826 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5827 | #endif |
6bf13c0c | 5828 | .release = btrfs_release_file, |
e02119d5 | 5829 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5830 | }; |
5831 | ||
d1310b2e | 5832 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5833 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5834 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5835 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5836 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5837 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5838 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5839 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5840 | .set_bit_hook = btrfs_set_bit_hook, |
5841 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
5842 | .merge_extent_hook = btrfs_merge_extent_hook, |
5843 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
5844 | }; |
5845 | ||
35054394 CM |
5846 | /* |
5847 | * btrfs doesn't support the bmap operation because swapfiles | |
5848 | * use bmap to make a mapping of extents in the file. They assume | |
5849 | * these extents won't change over the life of the file and they | |
5850 | * use the bmap result to do IO directly to the drive. | |
5851 | * | |
5852 | * the btrfs bmap call would return logical addresses that aren't | |
5853 | * suitable for IO and they also will change frequently as COW | |
5854 | * operations happen. So, swapfile + btrfs == corruption. | |
5855 | * | |
5856 | * For now we're avoiding this by dropping bmap. | |
5857 | */ | |
39279cc3 CM |
5858 | static struct address_space_operations btrfs_aops = { |
5859 | .readpage = btrfs_readpage, | |
5860 | .writepage = btrfs_writepage, | |
b293f02e | 5861 | .writepages = btrfs_writepages, |
3ab2fb5a | 5862 | .readpages = btrfs_readpages, |
39279cc3 | 5863 | .sync_page = block_sync_page, |
16432985 | 5864 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5865 | .invalidatepage = btrfs_invalidatepage, |
5866 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5867 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5868 | }; |
5869 | ||
5870 | static struct address_space_operations btrfs_symlink_aops = { | |
5871 | .readpage = btrfs_readpage, | |
5872 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5873 | .invalidatepage = btrfs_invalidatepage, |
5874 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5875 | }; |
5876 | ||
5877 | static struct inode_operations btrfs_file_inode_operations = { | |
5878 | .truncate = btrfs_truncate, | |
5879 | .getattr = btrfs_getattr, | |
5880 | .setattr = btrfs_setattr, | |
95819c05 CH |
5881 | .setxattr = btrfs_setxattr, |
5882 | .getxattr = btrfs_getxattr, | |
5103e947 | 5883 | .listxattr = btrfs_listxattr, |
95819c05 | 5884 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5885 | .permission = btrfs_permission, |
d899e052 | 5886 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5887 | .fiemap = btrfs_fiemap, |
39279cc3 | 5888 | }; |
618e21d5 JB |
5889 | static struct inode_operations btrfs_special_inode_operations = { |
5890 | .getattr = btrfs_getattr, | |
5891 | .setattr = btrfs_setattr, | |
fdebe2bd | 5892 | .permission = btrfs_permission, |
95819c05 CH |
5893 | .setxattr = btrfs_setxattr, |
5894 | .getxattr = btrfs_getxattr, | |
33268eaf | 5895 | .listxattr = btrfs_listxattr, |
95819c05 | 5896 | .removexattr = btrfs_removexattr, |
618e21d5 | 5897 | }; |
39279cc3 CM |
5898 | static struct inode_operations btrfs_symlink_inode_operations = { |
5899 | .readlink = generic_readlink, | |
5900 | .follow_link = page_follow_link_light, | |
5901 | .put_link = page_put_link, | |
fdebe2bd | 5902 | .permission = btrfs_permission, |
0279b4cd JO |
5903 | .setxattr = btrfs_setxattr, |
5904 | .getxattr = btrfs_getxattr, | |
5905 | .listxattr = btrfs_listxattr, | |
5906 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5907 | }; |
76dda93c | 5908 | |
82d339d9 | 5909 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
5910 | .d_delete = btrfs_dentry_delete, |
5911 | }; |