]>
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> | |
29 | #include <linux/smp_lock.h> | |
30 | #include <linux/backing-dev.h> | |
31 | #include <linux/mpage.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/writeback.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/compat.h> | |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
5103e947 | 37 | #include <linux/xattr.h> |
33268eaf | 38 | #include <linux/posix_acl.h> |
d899e052 | 39 | #include <linux/falloc.h> |
4b4e25f2 | 40 | #include "compat.h" |
39279cc3 CM |
41 | #include "ctree.h" |
42 | #include "disk-io.h" | |
43 | #include "transaction.h" | |
44 | #include "btrfs_inode.h" | |
45 | #include "ioctl.h" | |
46 | #include "print-tree.h" | |
0b86a832 | 47 | #include "volumes.h" |
e6dcd2dc | 48 | #include "ordered-data.h" |
95819c05 | 49 | #include "xattr.h" |
e02119d5 | 50 | #include "tree-log.h" |
5b84e8d6 | 51 | #include "ref-cache.h" |
c8b97818 | 52 | #include "compression.h" |
39279cc3 CM |
53 | |
54 | struct btrfs_iget_args { | |
55 | u64 ino; | |
56 | struct btrfs_root *root; | |
57 | }; | |
58 | ||
59 | static struct inode_operations btrfs_dir_inode_operations; | |
60 | static struct inode_operations btrfs_symlink_inode_operations; | |
61 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 62 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
63 | static struct inode_operations btrfs_file_inode_operations; |
64 | static struct address_space_operations btrfs_aops; | |
65 | static struct address_space_operations btrfs_symlink_aops; | |
66 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 67 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
68 | |
69 | static struct kmem_cache *btrfs_inode_cachep; | |
70 | struct kmem_cache *btrfs_trans_handle_cachep; | |
71 | struct kmem_cache *btrfs_transaction_cachep; | |
72 | struct kmem_cache *btrfs_bit_radix_cachep; | |
73 | struct kmem_cache *btrfs_path_cachep; | |
74 | ||
75 | #define S_SHIFT 12 | |
76 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
77 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
78 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
79 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
80 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
81 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
82 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
83 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
84 | }; | |
85 | ||
7b128766 | 86 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 87 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
88 | static noinline int cow_file_range(struct inode *inode, |
89 | struct page *locked_page, | |
90 | u64 start, u64 end, int *page_started, | |
91 | unsigned long *nr_written, int unlock); | |
7b128766 | 92 | |
d352ac68 CM |
93 | /* |
94 | * a very lame attempt at stopping writes when the FS is 85% full. There | |
95 | * are countless ways this is incorrect, but it is better than nothing. | |
96 | */ | |
1832a6d5 CM |
97 | int btrfs_check_free_space(struct btrfs_root *root, u64 num_required, |
98 | int for_del) | |
99 | { | |
a2135011 CM |
100 | u64 total; |
101 | u64 used; | |
1832a6d5 CM |
102 | u64 thresh; |
103 | int ret = 0; | |
104 | ||
75eff68e | 105 | spin_lock(&root->fs_info->delalloc_lock); |
a2135011 CM |
106 | total = btrfs_super_total_bytes(&root->fs_info->super_copy); |
107 | used = btrfs_super_bytes_used(&root->fs_info->super_copy); | |
1832a6d5 | 108 | if (for_del) |
f9ef6604 | 109 | thresh = total * 90; |
1832a6d5 | 110 | else |
f9ef6604 CM |
111 | thresh = total * 85; |
112 | ||
113 | do_div(thresh, 100); | |
1832a6d5 | 114 | |
1832a6d5 CM |
115 | if (used + root->fs_info->delalloc_bytes + num_required > thresh) |
116 | ret = -ENOSPC; | |
75eff68e | 117 | spin_unlock(&root->fs_info->delalloc_lock); |
1832a6d5 CM |
118 | return ret; |
119 | } | |
120 | ||
c8b97818 CM |
121 | /* |
122 | * this does all the hard work for inserting an inline extent into | |
123 | * the btree. The caller should have done a btrfs_drop_extents so that | |
124 | * no overlapping inline items exist in the btree | |
125 | */ | |
d397712b | 126 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
127 | struct btrfs_root *root, struct inode *inode, |
128 | u64 start, size_t size, size_t compressed_size, | |
129 | struct page **compressed_pages) | |
130 | { | |
131 | struct btrfs_key key; | |
132 | struct btrfs_path *path; | |
133 | struct extent_buffer *leaf; | |
134 | struct page *page = NULL; | |
135 | char *kaddr; | |
136 | unsigned long ptr; | |
137 | struct btrfs_file_extent_item *ei; | |
138 | int err = 0; | |
139 | int ret; | |
140 | size_t cur_size = size; | |
141 | size_t datasize; | |
142 | unsigned long offset; | |
143 | int use_compress = 0; | |
144 | ||
145 | if (compressed_size && compressed_pages) { | |
146 | use_compress = 1; | |
147 | cur_size = compressed_size; | |
148 | } | |
149 | ||
d397712b CM |
150 | path = btrfs_alloc_path(); |
151 | if (!path) | |
c8b97818 CM |
152 | return -ENOMEM; |
153 | ||
154 | btrfs_set_trans_block_group(trans, inode); | |
155 | ||
156 | key.objectid = inode->i_ino; | |
157 | key.offset = start; | |
158 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
159 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
160 | ||
161 | inode_add_bytes(inode, size); | |
162 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
163 | datasize); | |
164 | BUG_ON(ret); | |
165 | if (ret) { | |
166 | err = ret; | |
c8b97818 CM |
167 | goto fail; |
168 | } | |
169 | leaf = path->nodes[0]; | |
170 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
171 | struct btrfs_file_extent_item); | |
172 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
173 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
174 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
175 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
176 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
177 | ptr = btrfs_file_extent_inline_start(ei); | |
178 | ||
179 | if (use_compress) { | |
180 | struct page *cpage; | |
181 | int i = 0; | |
d397712b | 182 | while (compressed_size > 0) { |
c8b97818 | 183 | cpage = compressed_pages[i]; |
5b050f04 | 184 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
185 | PAGE_CACHE_SIZE); |
186 | ||
187 | kaddr = kmap(cpage); | |
188 | write_extent_buffer(leaf, kaddr, ptr, cur_size); | |
189 | kunmap(cpage); | |
190 | ||
191 | i++; | |
192 | ptr += cur_size; | |
193 | compressed_size -= cur_size; | |
194 | } | |
195 | btrfs_set_file_extent_compression(leaf, ei, | |
196 | BTRFS_COMPRESS_ZLIB); | |
197 | } else { | |
198 | page = find_get_page(inode->i_mapping, | |
199 | start >> PAGE_CACHE_SHIFT); | |
200 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
201 | kaddr = kmap_atomic(page, KM_USER0); | |
202 | offset = start & (PAGE_CACHE_SIZE - 1); | |
203 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
204 | kunmap_atomic(kaddr, KM_USER0); | |
205 | page_cache_release(page); | |
206 | } | |
207 | btrfs_mark_buffer_dirty(leaf); | |
208 | btrfs_free_path(path); | |
209 | ||
210 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
211 | btrfs_update_inode(trans, root, inode); | |
212 | return 0; | |
213 | fail: | |
214 | btrfs_free_path(path); | |
215 | return err; | |
216 | } | |
217 | ||
218 | ||
219 | /* | |
220 | * conditionally insert an inline extent into the file. This | |
221 | * does the checks required to make sure the data is small enough | |
222 | * to fit as an inline extent. | |
223 | */ | |
224 | static int cow_file_range_inline(struct btrfs_trans_handle *trans, | |
225 | struct btrfs_root *root, | |
226 | struct inode *inode, u64 start, u64 end, | |
227 | size_t compressed_size, | |
228 | struct page **compressed_pages) | |
229 | { | |
230 | u64 isize = i_size_read(inode); | |
231 | u64 actual_end = min(end + 1, isize); | |
232 | u64 inline_len = actual_end - start; | |
233 | u64 aligned_end = (end + root->sectorsize - 1) & | |
234 | ~((u64)root->sectorsize - 1); | |
235 | u64 hint_byte; | |
236 | u64 data_len = inline_len; | |
237 | int ret; | |
238 | ||
239 | if (compressed_size) | |
240 | data_len = compressed_size; | |
241 | ||
242 | if (start > 0 || | |
70b99e69 | 243 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
244 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
245 | (!compressed_size && | |
246 | (actual_end & (root->sectorsize - 1)) == 0) || | |
247 | end + 1 < isize || | |
248 | data_len > root->fs_info->max_inline) { | |
249 | return 1; | |
250 | } | |
251 | ||
c8b97818 | 252 | ret = btrfs_drop_extents(trans, root, inode, start, |
70b99e69 | 253 | aligned_end, start, &hint_byte); |
c8b97818 CM |
254 | BUG_ON(ret); |
255 | ||
256 | if (isize > actual_end) | |
257 | inline_len = min_t(u64, isize, actual_end); | |
258 | ret = insert_inline_extent(trans, root, inode, start, | |
259 | inline_len, compressed_size, | |
260 | compressed_pages); | |
261 | BUG_ON(ret); | |
262 | btrfs_drop_extent_cache(inode, start, aligned_end, 0); | |
c8b97818 CM |
263 | return 0; |
264 | } | |
265 | ||
771ed689 CM |
266 | struct async_extent { |
267 | u64 start; | |
268 | u64 ram_size; | |
269 | u64 compressed_size; | |
270 | struct page **pages; | |
271 | unsigned long nr_pages; | |
272 | struct list_head list; | |
273 | }; | |
274 | ||
275 | struct async_cow { | |
276 | struct inode *inode; | |
277 | struct btrfs_root *root; | |
278 | struct page *locked_page; | |
279 | u64 start; | |
280 | u64 end; | |
281 | struct list_head extents; | |
282 | struct btrfs_work work; | |
283 | }; | |
284 | ||
285 | static noinline int add_async_extent(struct async_cow *cow, | |
286 | u64 start, u64 ram_size, | |
287 | u64 compressed_size, | |
288 | struct page **pages, | |
289 | unsigned long nr_pages) | |
290 | { | |
291 | struct async_extent *async_extent; | |
292 | ||
293 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
294 | async_extent->start = start; | |
295 | async_extent->ram_size = ram_size; | |
296 | async_extent->compressed_size = compressed_size; | |
297 | async_extent->pages = pages; | |
298 | async_extent->nr_pages = nr_pages; | |
299 | list_add_tail(&async_extent->list, &cow->extents); | |
300 | return 0; | |
301 | } | |
302 | ||
d352ac68 | 303 | /* |
771ed689 CM |
304 | * we create compressed extents in two phases. The first |
305 | * phase compresses a range of pages that have already been | |
306 | * locked (both pages and state bits are locked). | |
c8b97818 | 307 | * |
771ed689 CM |
308 | * This is done inside an ordered work queue, and the compression |
309 | * is spread across many cpus. The actual IO submission is step | |
310 | * two, and the ordered work queue takes care of making sure that | |
311 | * happens in the same order things were put onto the queue by | |
312 | * writepages and friends. | |
c8b97818 | 313 | * |
771ed689 CM |
314 | * If this code finds it can't get good compression, it puts an |
315 | * entry onto the work queue to write the uncompressed bytes. This | |
316 | * makes sure that both compressed inodes and uncompressed inodes | |
317 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 318 | */ |
771ed689 CM |
319 | static noinline int compress_file_range(struct inode *inode, |
320 | struct page *locked_page, | |
321 | u64 start, u64 end, | |
322 | struct async_cow *async_cow, | |
323 | int *num_added) | |
b888db2b CM |
324 | { |
325 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
326 | struct btrfs_trans_handle *trans; | |
db94535d | 327 | u64 num_bytes; |
c8b97818 CM |
328 | u64 orig_start; |
329 | u64 disk_num_bytes; | |
db94535d | 330 | u64 blocksize = root->sectorsize; |
c8b97818 | 331 | u64 actual_end; |
42dc7bab | 332 | u64 isize = i_size_read(inode); |
e6dcd2dc | 333 | int ret = 0; |
c8b97818 CM |
334 | struct page **pages = NULL; |
335 | unsigned long nr_pages; | |
336 | unsigned long nr_pages_ret = 0; | |
337 | unsigned long total_compressed = 0; | |
338 | unsigned long total_in = 0; | |
339 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 340 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
341 | int i; |
342 | int will_compress; | |
b888db2b | 343 | |
c8b97818 CM |
344 | orig_start = start; |
345 | ||
42dc7bab | 346 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
347 | again: |
348 | will_compress = 0; | |
349 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
350 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 351 | |
c8b97818 CM |
352 | total_compressed = actual_end - start; |
353 | ||
354 | /* we want to make sure that amount of ram required to uncompress | |
355 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
356 | * of a compressed extent to 128k. This is a crucial number |
357 | * because it also controls how easily we can spread reads across | |
358 | * cpus for decompression. | |
359 | * | |
360 | * We also want to make sure the amount of IO required to do | |
361 | * a random read is reasonably small, so we limit the size of | |
362 | * a compressed extent to 128k. | |
c8b97818 CM |
363 | */ |
364 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 365 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 366 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
367 | disk_num_bytes = num_bytes; |
368 | total_in = 0; | |
369 | ret = 0; | |
db94535d | 370 | |
771ed689 CM |
371 | /* |
372 | * we do compression for mount -o compress and when the | |
373 | * inode has not been flagged as nocompress. This flag can | |
374 | * change at any time if we discover bad compression ratios. | |
c8b97818 CM |
375 | */ |
376 | if (!btrfs_test_flag(inode, NOCOMPRESS) && | |
377 | btrfs_test_opt(root, COMPRESS)) { | |
378 | WARN_ON(pages); | |
cfbc246e | 379 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 380 | |
c8b97818 CM |
381 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
382 | total_compressed, pages, | |
383 | nr_pages, &nr_pages_ret, | |
384 | &total_in, | |
385 | &total_compressed, | |
386 | max_compressed); | |
387 | ||
388 | if (!ret) { | |
389 | unsigned long offset = total_compressed & | |
390 | (PAGE_CACHE_SIZE - 1); | |
391 | struct page *page = pages[nr_pages_ret - 1]; | |
392 | char *kaddr; | |
393 | ||
394 | /* zero the tail end of the last page, we might be | |
395 | * sending it down to disk | |
396 | */ | |
397 | if (offset) { | |
398 | kaddr = kmap_atomic(page, KM_USER0); | |
399 | memset(kaddr + offset, 0, | |
400 | PAGE_CACHE_SIZE - offset); | |
401 | kunmap_atomic(kaddr, KM_USER0); | |
402 | } | |
403 | will_compress = 1; | |
404 | } | |
405 | } | |
406 | if (start == 0) { | |
771ed689 CM |
407 | trans = btrfs_join_transaction(root, 1); |
408 | BUG_ON(!trans); | |
409 | btrfs_set_trans_block_group(trans, inode); | |
410 | ||
c8b97818 | 411 | /* lets try to make an inline extent */ |
771ed689 | 412 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 413 | /* we didn't compress the entire range, try |
771ed689 | 414 | * to make an uncompressed inline extent. |
c8b97818 CM |
415 | */ |
416 | ret = cow_file_range_inline(trans, root, inode, | |
417 | start, end, 0, NULL); | |
418 | } else { | |
771ed689 | 419 | /* try making a compressed inline extent */ |
c8b97818 CM |
420 | ret = cow_file_range_inline(trans, root, inode, |
421 | start, end, | |
422 | total_compressed, pages); | |
423 | } | |
771ed689 | 424 | btrfs_end_transaction(trans, root); |
c8b97818 | 425 | if (ret == 0) { |
771ed689 CM |
426 | /* |
427 | * inline extent creation worked, we don't need | |
428 | * to create any more async work items. Unlock | |
429 | * and free up our temp pages. | |
430 | */ | |
c8b97818 CM |
431 | extent_clear_unlock_delalloc(inode, |
432 | &BTRFS_I(inode)->io_tree, | |
771ed689 CM |
433 | start, end, NULL, 1, 0, |
434 | 0, 1, 1, 1); | |
c8b97818 CM |
435 | ret = 0; |
436 | goto free_pages_out; | |
437 | } | |
438 | } | |
439 | ||
440 | if (will_compress) { | |
441 | /* | |
442 | * we aren't doing an inline extent round the compressed size | |
443 | * up to a block size boundary so the allocator does sane | |
444 | * things | |
445 | */ | |
446 | total_compressed = (total_compressed + blocksize - 1) & | |
447 | ~(blocksize - 1); | |
448 | ||
449 | /* | |
450 | * one last check to make sure the compression is really a | |
451 | * win, compare the page count read with the blocks on disk | |
452 | */ | |
453 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
454 | ~(PAGE_CACHE_SIZE - 1); | |
455 | if (total_compressed >= total_in) { | |
456 | will_compress = 0; | |
457 | } else { | |
458 | disk_num_bytes = total_compressed; | |
459 | num_bytes = total_in; | |
460 | } | |
461 | } | |
462 | if (!will_compress && pages) { | |
463 | /* | |
464 | * the compression code ran but failed to make things smaller, | |
465 | * free any pages it allocated and our page pointer array | |
466 | */ | |
467 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 468 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
469 | page_cache_release(pages[i]); |
470 | } | |
471 | kfree(pages); | |
472 | pages = NULL; | |
473 | total_compressed = 0; | |
474 | nr_pages_ret = 0; | |
475 | ||
476 | /* flag the file so we don't compress in the future */ | |
477 | btrfs_set_flag(inode, NOCOMPRESS); | |
478 | } | |
771ed689 CM |
479 | if (will_compress) { |
480 | *num_added += 1; | |
c8b97818 | 481 | |
771ed689 CM |
482 | /* the async work queues will take care of doing actual |
483 | * allocation on disk for these compressed pages, | |
484 | * and will submit them to the elevator. | |
485 | */ | |
486 | add_async_extent(async_cow, start, num_bytes, | |
487 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 488 | |
42dc7bab | 489 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
490 | start += num_bytes; |
491 | pages = NULL; | |
492 | cond_resched(); | |
493 | goto again; | |
494 | } | |
495 | } else { | |
496 | /* | |
497 | * No compression, but we still need to write the pages in | |
498 | * the file we've been given so far. redirty the locked | |
499 | * page if it corresponds to our extent and set things up | |
500 | * for the async work queue to run cow_file_range to do | |
501 | * the normal delalloc dance | |
502 | */ | |
503 | if (page_offset(locked_page) >= start && | |
504 | page_offset(locked_page) <= end) { | |
505 | __set_page_dirty_nobuffers(locked_page); | |
506 | /* unlocked later on in the async handlers */ | |
507 | } | |
508 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
509 | *num_added += 1; | |
510 | } | |
3b951516 | 511 | |
771ed689 CM |
512 | out: |
513 | return 0; | |
514 | ||
515 | free_pages_out: | |
516 | for (i = 0; i < nr_pages_ret; i++) { | |
517 | WARN_ON(pages[i]->mapping); | |
518 | page_cache_release(pages[i]); | |
519 | } | |
d397712b | 520 | kfree(pages); |
771ed689 CM |
521 | |
522 | goto out; | |
523 | } | |
524 | ||
525 | /* | |
526 | * phase two of compressed writeback. This is the ordered portion | |
527 | * of the code, which only gets called in the order the work was | |
528 | * queued. We walk all the async extents created by compress_file_range | |
529 | * and send them down to the disk. | |
530 | */ | |
531 | static noinline int submit_compressed_extents(struct inode *inode, | |
532 | struct async_cow *async_cow) | |
533 | { | |
534 | struct async_extent *async_extent; | |
535 | u64 alloc_hint = 0; | |
536 | struct btrfs_trans_handle *trans; | |
537 | struct btrfs_key ins; | |
538 | struct extent_map *em; | |
539 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
540 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
541 | struct extent_io_tree *io_tree; | |
542 | int ret; | |
543 | ||
544 | if (list_empty(&async_cow->extents)) | |
545 | return 0; | |
546 | ||
547 | trans = btrfs_join_transaction(root, 1); | |
548 | ||
d397712b | 549 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
550 | async_extent = list_entry(async_cow->extents.next, |
551 | struct async_extent, list); | |
552 | list_del(&async_extent->list); | |
c8b97818 | 553 | |
771ed689 CM |
554 | io_tree = &BTRFS_I(inode)->io_tree; |
555 | ||
556 | /* did the compression code fall back to uncompressed IO? */ | |
557 | if (!async_extent->pages) { | |
558 | int page_started = 0; | |
559 | unsigned long nr_written = 0; | |
560 | ||
561 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
562 | async_extent->start + |
563 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
564 | |
565 | /* allocate blocks */ | |
566 | cow_file_range(inode, async_cow->locked_page, | |
567 | async_extent->start, | |
568 | async_extent->start + | |
569 | async_extent->ram_size - 1, | |
570 | &page_started, &nr_written, 0); | |
571 | ||
572 | /* | |
573 | * if page_started, cow_file_range inserted an | |
574 | * inline extent and took care of all the unlocking | |
575 | * and IO for us. Otherwise, we need to submit | |
576 | * all those pages down to the drive. | |
577 | */ | |
578 | if (!page_started) | |
579 | extent_write_locked_range(io_tree, | |
580 | inode, async_extent->start, | |
d397712b | 581 | async_extent->start + |
771ed689 CM |
582 | async_extent->ram_size - 1, |
583 | btrfs_get_extent, | |
584 | WB_SYNC_ALL); | |
585 | kfree(async_extent); | |
586 | cond_resched(); | |
587 | continue; | |
588 | } | |
589 | ||
590 | lock_extent(io_tree, async_extent->start, | |
591 | async_extent->start + async_extent->ram_size - 1, | |
592 | GFP_NOFS); | |
c8b97818 | 593 | /* |
771ed689 CM |
594 | * here we're doing allocation and writeback of the |
595 | * compressed pages | |
c8b97818 | 596 | */ |
771ed689 CM |
597 | btrfs_drop_extent_cache(inode, async_extent->start, |
598 | async_extent->start + | |
599 | async_extent->ram_size - 1, 0); | |
600 | ||
601 | ret = btrfs_reserve_extent(trans, root, | |
602 | async_extent->compressed_size, | |
603 | async_extent->compressed_size, | |
604 | 0, alloc_hint, | |
605 | (u64)-1, &ins, 1); | |
606 | BUG_ON(ret); | |
607 | em = alloc_extent_map(GFP_NOFS); | |
608 | em->start = async_extent->start; | |
609 | em->len = async_extent->ram_size; | |
445a6944 | 610 | em->orig_start = em->start; |
c8b97818 | 611 | |
771ed689 CM |
612 | em->block_start = ins.objectid; |
613 | em->block_len = ins.offset; | |
614 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
615 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
616 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
617 | ||
d397712b | 618 | while (1) { |
771ed689 CM |
619 | spin_lock(&em_tree->lock); |
620 | ret = add_extent_mapping(em_tree, em); | |
621 | spin_unlock(&em_tree->lock); | |
622 | if (ret != -EEXIST) { | |
623 | free_extent_map(em); | |
624 | break; | |
625 | } | |
626 | btrfs_drop_extent_cache(inode, async_extent->start, | |
627 | async_extent->start + | |
628 | async_extent->ram_size - 1, 0); | |
629 | } | |
630 | ||
631 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
632 | ins.objectid, | |
633 | async_extent->ram_size, | |
634 | ins.offset, | |
635 | BTRFS_ORDERED_COMPRESSED); | |
636 | BUG_ON(ret); | |
637 | ||
638 | btrfs_end_transaction(trans, root); | |
639 | ||
640 | /* | |
641 | * clear dirty, set writeback and unlock the pages. | |
642 | */ | |
643 | extent_clear_unlock_delalloc(inode, | |
644 | &BTRFS_I(inode)->io_tree, | |
645 | async_extent->start, | |
646 | async_extent->start + | |
647 | async_extent->ram_size - 1, | |
648 | NULL, 1, 1, 0, 1, 1, 0); | |
649 | ||
650 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
651 | async_extent->start, |
652 | async_extent->ram_size, | |
653 | ins.objectid, | |
654 | ins.offset, async_extent->pages, | |
655 | async_extent->nr_pages); | |
771ed689 CM |
656 | |
657 | BUG_ON(ret); | |
658 | trans = btrfs_join_transaction(root, 1); | |
659 | alloc_hint = ins.objectid + ins.offset; | |
660 | kfree(async_extent); | |
661 | cond_resched(); | |
662 | } | |
663 | ||
664 | btrfs_end_transaction(trans, root); | |
665 | return 0; | |
666 | } | |
667 | ||
668 | /* | |
669 | * when extent_io.c finds a delayed allocation range in the file, | |
670 | * the call backs end up in this code. The basic idea is to | |
671 | * allocate extents on disk for the range, and create ordered data structs | |
672 | * in ram to track those extents. | |
673 | * | |
674 | * locked_page is the page that writepage had locked already. We use | |
675 | * it to make sure we don't do extra locks or unlocks. | |
676 | * | |
677 | * *page_started is set to one if we unlock locked_page and do everything | |
678 | * required to start IO on it. It may be clean and already done with | |
679 | * IO when we return. | |
680 | */ | |
681 | static noinline int cow_file_range(struct inode *inode, | |
682 | struct page *locked_page, | |
683 | u64 start, u64 end, int *page_started, | |
684 | unsigned long *nr_written, | |
685 | int unlock) | |
686 | { | |
687 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
688 | struct btrfs_trans_handle *trans; | |
689 | u64 alloc_hint = 0; | |
690 | u64 num_bytes; | |
691 | unsigned long ram_size; | |
692 | u64 disk_num_bytes; | |
693 | u64 cur_alloc_size; | |
694 | u64 blocksize = root->sectorsize; | |
695 | u64 actual_end; | |
42dc7bab | 696 | u64 isize = i_size_read(inode); |
771ed689 CM |
697 | struct btrfs_key ins; |
698 | struct extent_map *em; | |
699 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
700 | int ret = 0; | |
701 | ||
702 | trans = btrfs_join_transaction(root, 1); | |
703 | BUG_ON(!trans); | |
704 | btrfs_set_trans_block_group(trans, inode); | |
705 | ||
42dc7bab | 706 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
707 | |
708 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
709 | num_bytes = max(blocksize, num_bytes); | |
710 | disk_num_bytes = num_bytes; | |
711 | ret = 0; | |
712 | ||
713 | if (start == 0) { | |
714 | /* lets try to make an inline extent */ | |
715 | ret = cow_file_range_inline(trans, root, inode, | |
716 | start, end, 0, NULL); | |
717 | if (ret == 0) { | |
718 | extent_clear_unlock_delalloc(inode, | |
719 | &BTRFS_I(inode)->io_tree, | |
720 | start, end, NULL, 1, 1, | |
721 | 1, 1, 1, 1); | |
722 | *nr_written = *nr_written + | |
723 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
724 | *page_started = 1; | |
725 | ret = 0; | |
726 | goto out; | |
727 | } | |
728 | } | |
729 | ||
730 | BUG_ON(disk_num_bytes > | |
731 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
732 | ||
733 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
734 | ||
d397712b | 735 | while (disk_num_bytes > 0) { |
c8b97818 | 736 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 737 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 738 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 739 | (u64)-1, &ins, 1); |
d397712b CM |
740 | BUG_ON(ret); |
741 | ||
e6dcd2dc CM |
742 | em = alloc_extent_map(GFP_NOFS); |
743 | em->start = start; | |
445a6944 | 744 | em->orig_start = em->start; |
c8b97818 | 745 | |
771ed689 CM |
746 | ram_size = ins.offset; |
747 | em->len = ins.offset; | |
c8b97818 | 748 | |
e6dcd2dc | 749 | em->block_start = ins.objectid; |
c8b97818 | 750 | em->block_len = ins.offset; |
e6dcd2dc | 751 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 752 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 753 | |
d397712b | 754 | while (1) { |
e6dcd2dc CM |
755 | spin_lock(&em_tree->lock); |
756 | ret = add_extent_mapping(em_tree, em); | |
757 | spin_unlock(&em_tree->lock); | |
758 | if (ret != -EEXIST) { | |
759 | free_extent_map(em); | |
760 | break; | |
761 | } | |
762 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 763 | start + ram_size - 1, 0); |
e6dcd2dc CM |
764 | } |
765 | ||
98d20f67 | 766 | cur_alloc_size = ins.offset; |
e6dcd2dc | 767 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 768 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 769 | BUG_ON(ret); |
c8b97818 | 770 | |
17d217fe YZ |
771 | if (root->root_key.objectid == |
772 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
773 | ret = btrfs_reloc_clone_csums(inode, start, | |
774 | cur_alloc_size); | |
775 | BUG_ON(ret); | |
776 | } | |
777 | ||
d397712b | 778 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 779 | break; |
d397712b | 780 | |
c8b97818 CM |
781 | /* we're not doing compressed IO, don't unlock the first |
782 | * page (which the caller expects to stay locked), don't | |
783 | * clear any dirty bits and don't set any writeback bits | |
784 | */ | |
785 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
786 | start, start + ram_size - 1, | |
771ed689 CM |
787 | locked_page, unlock, 1, |
788 | 1, 0, 0, 0); | |
c8b97818 | 789 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
790 | num_bytes -= cur_alloc_size; |
791 | alloc_hint = ins.objectid + ins.offset; | |
792 | start += cur_alloc_size; | |
b888db2b | 793 | } |
b888db2b | 794 | out: |
771ed689 | 795 | ret = 0; |
b888db2b | 796 | btrfs_end_transaction(trans, root); |
c8b97818 | 797 | |
be20aa9d | 798 | return ret; |
771ed689 | 799 | } |
c8b97818 | 800 | |
771ed689 CM |
801 | /* |
802 | * work queue call back to started compression on a file and pages | |
803 | */ | |
804 | static noinline void async_cow_start(struct btrfs_work *work) | |
805 | { | |
806 | struct async_cow *async_cow; | |
807 | int num_added = 0; | |
808 | async_cow = container_of(work, struct async_cow, work); | |
809 | ||
810 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
811 | async_cow->start, async_cow->end, async_cow, | |
812 | &num_added); | |
813 | if (num_added == 0) | |
814 | async_cow->inode = NULL; | |
815 | } | |
816 | ||
817 | /* | |
818 | * work queue call back to submit previously compressed pages | |
819 | */ | |
820 | static noinline void async_cow_submit(struct btrfs_work *work) | |
821 | { | |
822 | struct async_cow *async_cow; | |
823 | struct btrfs_root *root; | |
824 | unsigned long nr_pages; | |
825 | ||
826 | async_cow = container_of(work, struct async_cow, work); | |
827 | ||
828 | root = async_cow->root; | |
829 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
830 | PAGE_CACHE_SHIFT; | |
831 | ||
832 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
833 | ||
834 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
835 | 5 * 1042 * 1024 && | |
836 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
837 | wake_up(&root->fs_info->async_submit_wait); | |
838 | ||
d397712b | 839 | if (async_cow->inode) |
771ed689 | 840 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 841 | } |
c8b97818 | 842 | |
771ed689 CM |
843 | static noinline void async_cow_free(struct btrfs_work *work) |
844 | { | |
845 | struct async_cow *async_cow; | |
846 | async_cow = container_of(work, struct async_cow, work); | |
847 | kfree(async_cow); | |
848 | } | |
849 | ||
850 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
851 | u64 start, u64 end, int *page_started, | |
852 | unsigned long *nr_written) | |
853 | { | |
854 | struct async_cow *async_cow; | |
855 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
856 | unsigned long nr_pages; | |
857 | u64 cur_end; | |
858 | int limit = 10 * 1024 * 1042; | |
859 | ||
860 | if (!btrfs_test_opt(root, COMPRESS)) { | |
861 | return cow_file_range(inode, locked_page, start, end, | |
862 | page_started, nr_written, 1); | |
863 | } | |
864 | ||
865 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | | |
866 | EXTENT_DELALLOC, 1, 0, GFP_NOFS); | |
d397712b | 867 | while (start < end) { |
771ed689 CM |
868 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
869 | async_cow->inode = inode; | |
870 | async_cow->root = root; | |
871 | async_cow->locked_page = locked_page; | |
872 | async_cow->start = start; | |
873 | ||
874 | if (btrfs_test_flag(inode, NOCOMPRESS)) | |
875 | cur_end = end; | |
876 | else | |
877 | cur_end = min(end, start + 512 * 1024 - 1); | |
878 | ||
879 | async_cow->end = cur_end; | |
880 | INIT_LIST_HEAD(&async_cow->extents); | |
881 | ||
882 | async_cow->work.func = async_cow_start; | |
883 | async_cow->work.ordered_func = async_cow_submit; | |
884 | async_cow->work.ordered_free = async_cow_free; | |
885 | async_cow->work.flags = 0; | |
886 | ||
771ed689 CM |
887 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
888 | PAGE_CACHE_SHIFT; | |
889 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
890 | ||
891 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
892 | &async_cow->work); | |
893 | ||
894 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
895 | wait_event(root->fs_info->async_submit_wait, | |
896 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
897 | limit)); | |
898 | } | |
899 | ||
d397712b | 900 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
901 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
902 | wait_event(root->fs_info->async_submit_wait, | |
903 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
904 | 0)); | |
905 | } | |
906 | ||
907 | *nr_written += nr_pages; | |
908 | start = cur_end + 1; | |
909 | } | |
910 | *page_started = 1; | |
911 | return 0; | |
be20aa9d CM |
912 | } |
913 | ||
d397712b | 914 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
915 | u64 bytenr, u64 num_bytes) |
916 | { | |
917 | int ret; | |
918 | struct btrfs_ordered_sum *sums; | |
919 | LIST_HEAD(list); | |
920 | ||
07d400a6 YZ |
921 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
922 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
923 | if (ret == 0 && list_empty(&list)) |
924 | return 0; | |
925 | ||
926 | while (!list_empty(&list)) { | |
927 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
928 | list_del(&sums->list); | |
929 | kfree(sums); | |
930 | } | |
931 | return 1; | |
932 | } | |
933 | ||
d352ac68 CM |
934 | /* |
935 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
936 | * of the extents that exist in the file, and COWs the file as required. | |
937 | * | |
938 | * If no cow copies or snapshots exist, we write directly to the existing | |
939 | * blocks on disk | |
940 | */ | |
c8b97818 | 941 | static int run_delalloc_nocow(struct inode *inode, struct page *locked_page, |
771ed689 CM |
942 | u64 start, u64 end, int *page_started, int force, |
943 | unsigned long *nr_written) | |
be20aa9d | 944 | { |
be20aa9d | 945 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 946 | struct btrfs_trans_handle *trans; |
be20aa9d | 947 | struct extent_buffer *leaf; |
be20aa9d | 948 | struct btrfs_path *path; |
80ff3856 | 949 | struct btrfs_file_extent_item *fi; |
be20aa9d | 950 | struct btrfs_key found_key; |
80ff3856 YZ |
951 | u64 cow_start; |
952 | u64 cur_offset; | |
953 | u64 extent_end; | |
954 | u64 disk_bytenr; | |
955 | u64 num_bytes; | |
956 | int extent_type; | |
957 | int ret; | |
d899e052 | 958 | int type; |
80ff3856 YZ |
959 | int nocow; |
960 | int check_prev = 1; | |
be20aa9d CM |
961 | |
962 | path = btrfs_alloc_path(); | |
963 | BUG_ON(!path); | |
7ea394f1 YZ |
964 | trans = btrfs_join_transaction(root, 1); |
965 | BUG_ON(!trans); | |
be20aa9d | 966 | |
80ff3856 YZ |
967 | cow_start = (u64)-1; |
968 | cur_offset = start; | |
969 | while (1) { | |
970 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
971 | cur_offset, 0); | |
972 | BUG_ON(ret < 0); | |
973 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
974 | leaf = path->nodes[0]; | |
975 | btrfs_item_key_to_cpu(leaf, &found_key, | |
976 | path->slots[0] - 1); | |
977 | if (found_key.objectid == inode->i_ino && | |
978 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
979 | path->slots[0]--; | |
980 | } | |
981 | check_prev = 0; | |
982 | next_slot: | |
983 | leaf = path->nodes[0]; | |
984 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
985 | ret = btrfs_next_leaf(root, path); | |
986 | if (ret < 0) | |
987 | BUG_ON(1); | |
988 | if (ret > 0) | |
989 | break; | |
990 | leaf = path->nodes[0]; | |
991 | } | |
be20aa9d | 992 | |
80ff3856 YZ |
993 | nocow = 0; |
994 | disk_bytenr = 0; | |
17d217fe | 995 | num_bytes = 0; |
80ff3856 YZ |
996 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
997 | ||
998 | if (found_key.objectid > inode->i_ino || | |
999 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1000 | found_key.offset > end) | |
1001 | break; | |
1002 | ||
1003 | if (found_key.offset > cur_offset) { | |
1004 | extent_end = found_key.offset; | |
1005 | goto out_check; | |
1006 | } | |
1007 | ||
1008 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1009 | struct btrfs_file_extent_item); | |
1010 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1011 | ||
d899e052 YZ |
1012 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1013 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 YZ |
1014 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
1015 | extent_end = found_key.offset + | |
1016 | btrfs_file_extent_num_bytes(leaf, fi); | |
1017 | if (extent_end <= start) { | |
1018 | path->slots[0]++; | |
1019 | goto next_slot; | |
1020 | } | |
17d217fe YZ |
1021 | if (disk_bytenr == 0) |
1022 | goto out_check; | |
80ff3856 YZ |
1023 | if (btrfs_file_extent_compression(leaf, fi) || |
1024 | btrfs_file_extent_encryption(leaf, fi) || | |
1025 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1026 | goto out_check; | |
d899e052 YZ |
1027 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1028 | goto out_check; | |
d2fb3437 | 1029 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1030 | goto out_check; |
17d217fe YZ |
1031 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
1032 | disk_bytenr)) | |
1033 | goto out_check; | |
80ff3856 | 1034 | disk_bytenr += btrfs_file_extent_offset(leaf, fi); |
17d217fe YZ |
1035 | disk_bytenr += cur_offset - found_key.offset; |
1036 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1037 | /* | |
1038 | * force cow if csum exists in the range. | |
1039 | * this ensure that csum for a given extent are | |
1040 | * either valid or do not exist. | |
1041 | */ | |
1042 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1043 | goto out_check; | |
80ff3856 YZ |
1044 | nocow = 1; |
1045 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1046 | extent_end = found_key.offset + | |
1047 | btrfs_file_extent_inline_len(leaf, fi); | |
1048 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1049 | } else { | |
1050 | BUG_ON(1); | |
1051 | } | |
1052 | out_check: | |
1053 | if (extent_end <= start) { | |
1054 | path->slots[0]++; | |
1055 | goto next_slot; | |
1056 | } | |
1057 | if (!nocow) { | |
1058 | if (cow_start == (u64)-1) | |
1059 | cow_start = cur_offset; | |
1060 | cur_offset = extent_end; | |
1061 | if (cur_offset > end) | |
1062 | break; | |
1063 | path->slots[0]++; | |
1064 | goto next_slot; | |
7ea394f1 YZ |
1065 | } |
1066 | ||
1067 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1068 | if (cow_start != (u64)-1) { |
1069 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1070 | found_key.offset - 1, page_started, |
1071 | nr_written, 1); | |
80ff3856 YZ |
1072 | BUG_ON(ret); |
1073 | cow_start = (u64)-1; | |
7ea394f1 | 1074 | } |
80ff3856 | 1075 | |
d899e052 YZ |
1076 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1077 | struct extent_map *em; | |
1078 | struct extent_map_tree *em_tree; | |
1079 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1080 | em = alloc_extent_map(GFP_NOFS); | |
1081 | em->start = cur_offset; | |
445a6944 | 1082 | em->orig_start = em->start; |
d899e052 YZ |
1083 | em->len = num_bytes; |
1084 | em->block_len = num_bytes; | |
1085 | em->block_start = disk_bytenr; | |
1086 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1087 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1088 | while (1) { | |
1089 | spin_lock(&em_tree->lock); | |
1090 | ret = add_extent_mapping(em_tree, em); | |
1091 | spin_unlock(&em_tree->lock); | |
1092 | if (ret != -EEXIST) { | |
1093 | free_extent_map(em); | |
1094 | break; | |
1095 | } | |
1096 | btrfs_drop_extent_cache(inode, em->start, | |
1097 | em->start + em->len - 1, 0); | |
1098 | } | |
1099 | type = BTRFS_ORDERED_PREALLOC; | |
1100 | } else { | |
1101 | type = BTRFS_ORDERED_NOCOW; | |
1102 | } | |
80ff3856 YZ |
1103 | |
1104 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1105 | num_bytes, num_bytes, type); |
1106 | BUG_ON(ret); | |
771ed689 | 1107 | |
d899e052 YZ |
1108 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
1109 | cur_offset, cur_offset + num_bytes - 1, | |
771ed689 | 1110 | locked_page, 1, 1, 1, 0, 0, 0); |
80ff3856 YZ |
1111 | cur_offset = extent_end; |
1112 | if (cur_offset > end) | |
1113 | break; | |
be20aa9d | 1114 | } |
80ff3856 YZ |
1115 | btrfs_release_path(root, path); |
1116 | ||
1117 | if (cur_offset <= end && cow_start == (u64)-1) | |
1118 | cow_start = cur_offset; | |
1119 | if (cow_start != (u64)-1) { | |
1120 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1121 | page_started, nr_written, 1); |
80ff3856 YZ |
1122 | BUG_ON(ret); |
1123 | } | |
1124 | ||
1125 | ret = btrfs_end_transaction(trans, root); | |
1126 | BUG_ON(ret); | |
7ea394f1 | 1127 | btrfs_free_path(path); |
80ff3856 | 1128 | return 0; |
be20aa9d CM |
1129 | } |
1130 | ||
d352ac68 CM |
1131 | /* |
1132 | * extent_io.c call back to do delayed allocation processing | |
1133 | */ | |
c8b97818 | 1134 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1135 | u64 start, u64 end, int *page_started, |
1136 | unsigned long *nr_written) | |
be20aa9d | 1137 | { |
be20aa9d | 1138 | int ret; |
a2135011 | 1139 | |
17d217fe | 1140 | if (btrfs_test_flag(inode, NODATACOW)) |
c8b97818 | 1141 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1142 | page_started, 1, nr_written); |
d899e052 YZ |
1143 | else if (btrfs_test_flag(inode, PREALLOC)) |
1144 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
d397712b | 1145 | page_started, 0, nr_written); |
be20aa9d | 1146 | else |
771ed689 | 1147 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1148 | page_started, nr_written); |
1832a6d5 | 1149 | |
b888db2b CM |
1150 | return ret; |
1151 | } | |
1152 | ||
d352ac68 CM |
1153 | /* |
1154 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1155 | * bytes in this file, and to maintain the list of inodes that | |
1156 | * have pending delalloc work to be done. | |
1157 | */ | |
b2950863 | 1158 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1159 | unsigned long old, unsigned long bits) |
291d673e | 1160 | { |
75eff68e CM |
1161 | /* |
1162 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1163 | * but in this case, we are only testeing for the DELALLOC | |
1164 | * bit, which is only set or cleared with irqs on | |
1165 | */ | |
b0c68f8b | 1166 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1167 | struct btrfs_root *root = BTRFS_I(inode)->root; |
75eff68e | 1168 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1169 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1170 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1171 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1172 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1173 | &root->fs_info->delalloc_inodes); | |
1174 | } | |
75eff68e | 1175 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1176 | } |
1177 | return 0; | |
1178 | } | |
1179 | ||
d352ac68 CM |
1180 | /* |
1181 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1182 | */ | |
b2950863 | 1183 | static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1184 | unsigned long old, unsigned long bits) |
291d673e | 1185 | { |
75eff68e CM |
1186 | /* |
1187 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1188 | * but in this case, we are only testeing for the DELALLOC | |
1189 | * bit, which is only set or cleared with irqs on | |
1190 | */ | |
b0c68f8b | 1191 | if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1192 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1193 | |
75eff68e | 1194 | spin_lock(&root->fs_info->delalloc_lock); |
b0c68f8b | 1195 | if (end - start + 1 > root->fs_info->delalloc_bytes) { |
d397712b CM |
1196 | printk(KERN_INFO "btrfs warning: delalloc account " |
1197 | "%llu %llu\n", | |
1198 | (unsigned long long)end - start + 1, | |
1199 | (unsigned long long) | |
1200 | root->fs_info->delalloc_bytes); | |
b0c68f8b | 1201 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1202 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b CM |
1203 | } else { |
1204 | root->fs_info->delalloc_bytes -= end - start + 1; | |
9069218d | 1205 | BTRFS_I(inode)->delalloc_bytes -= end - start + 1; |
b0c68f8b | 1206 | } |
ea8c2819 CM |
1207 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1208 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1209 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1210 | } | |
75eff68e | 1211 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1212 | } |
1213 | return 0; | |
1214 | } | |
1215 | ||
d352ac68 CM |
1216 | /* |
1217 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1218 | * we don't create bios that span stripes or chunks | |
1219 | */ | |
239b14b3 | 1220 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1221 | size_t size, struct bio *bio, |
1222 | unsigned long bio_flags) | |
239b14b3 CM |
1223 | { |
1224 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1225 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1226 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1227 | u64 length = 0; |
1228 | u64 map_length; | |
239b14b3 CM |
1229 | int ret; |
1230 | ||
771ed689 CM |
1231 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1232 | return 0; | |
1233 | ||
f2d8d74d | 1234 | length = bio->bi_size; |
239b14b3 CM |
1235 | map_tree = &root->fs_info->mapping_tree; |
1236 | map_length = length; | |
cea9e445 | 1237 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1238 | &map_length, NULL, 0); |
cea9e445 | 1239 | |
d397712b | 1240 | if (map_length < length + size) |
239b14b3 | 1241 | return 1; |
239b14b3 CM |
1242 | return 0; |
1243 | } | |
1244 | ||
d352ac68 CM |
1245 | /* |
1246 | * in order to insert checksums into the metadata in large chunks, | |
1247 | * we wait until bio submission time. All the pages in the bio are | |
1248 | * checksummed and sums are attached onto the ordered extent record. | |
1249 | * | |
1250 | * At IO completion time the cums attached on the ordered extent record | |
1251 | * are inserted into the btree | |
1252 | */ | |
d397712b CM |
1253 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1254 | struct bio *bio, int mirror_num, | |
1255 | unsigned long bio_flags) | |
065631f6 | 1256 | { |
065631f6 | 1257 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1258 | int ret = 0; |
e015640f | 1259 | |
d20f7043 | 1260 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1261 | BUG_ON(ret); |
4a69a410 CM |
1262 | return 0; |
1263 | } | |
e015640f | 1264 | |
4a69a410 CM |
1265 | /* |
1266 | * in order to insert checksums into the metadata in large chunks, | |
1267 | * we wait until bio submission time. All the pages in the bio are | |
1268 | * checksummed and sums are attached onto the ordered extent record. | |
1269 | * | |
1270 | * At IO completion time the cums attached on the ordered extent record | |
1271 | * are inserted into the btree | |
1272 | */ | |
b2950863 | 1273 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1274 | int mirror_num, unsigned long bio_flags) |
1275 | { | |
1276 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1277 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1278 | } |
1279 | ||
d352ac68 | 1280 | /* |
cad321ad CM |
1281 | * extent_io.c submission hook. This does the right thing for csum calculation |
1282 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1283 | */ |
b2950863 | 1284 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1285 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1286 | { |
1287 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1288 | int ret = 0; | |
19b9bdb0 | 1289 | int skip_sum; |
44b8bd7e | 1290 | |
cad321ad CM |
1291 | skip_sum = btrfs_test_flag(inode, NODATASUM); |
1292 | ||
e6dcd2dc CM |
1293 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1294 | BUG_ON(ret); | |
065631f6 | 1295 | |
4d1b5fb4 | 1296 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1297 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1298 | return btrfs_submit_compressed_read(inode, bio, |
1299 | mirror_num, bio_flags); | |
d20f7043 CM |
1300 | } else if (!skip_sum) |
1301 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1302 | goto mapit; |
19b9bdb0 | 1303 | } else if (!skip_sum) { |
17d217fe YZ |
1304 | /* csum items have already been cloned */ |
1305 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1306 | goto mapit; | |
19b9bdb0 CM |
1307 | /* we're doing a write, do the async checksumming */ |
1308 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1309 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1310 | bio_flags, __btrfs_submit_bio_start, |
1311 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1312 | } |
1313 | ||
0b86a832 | 1314 | mapit: |
8b712842 | 1315 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1316 | } |
6885f308 | 1317 | |
d352ac68 CM |
1318 | /* |
1319 | * given a list of ordered sums record them in the inode. This happens | |
1320 | * at IO completion time based on sums calculated at bio submission time. | |
1321 | */ | |
ba1da2f4 | 1322 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1323 | struct inode *inode, u64 file_offset, |
1324 | struct list_head *list) | |
1325 | { | |
1326 | struct list_head *cur; | |
1327 | struct btrfs_ordered_sum *sum; | |
1328 | ||
1329 | btrfs_set_trans_block_group(trans, inode); | |
ba1da2f4 | 1330 | list_for_each(cur, list) { |
e6dcd2dc | 1331 | sum = list_entry(cur, struct btrfs_ordered_sum, list); |
d20f7043 CM |
1332 | btrfs_csum_file_blocks(trans, |
1333 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1334 | } |
1335 | return 0; | |
1336 | } | |
1337 | ||
ea8c2819 CM |
1338 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1339 | { | |
d397712b | 1340 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1341 | WARN_ON(1); |
ea8c2819 CM |
1342 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1343 | GFP_NOFS); | |
1344 | } | |
1345 | ||
d352ac68 | 1346 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1347 | struct btrfs_writepage_fixup { |
1348 | struct page *page; | |
1349 | struct btrfs_work work; | |
1350 | }; | |
1351 | ||
b2950863 | 1352 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1353 | { |
1354 | struct btrfs_writepage_fixup *fixup; | |
1355 | struct btrfs_ordered_extent *ordered; | |
1356 | struct page *page; | |
1357 | struct inode *inode; | |
1358 | u64 page_start; | |
1359 | u64 page_end; | |
1360 | ||
1361 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1362 | page = fixup->page; | |
4a096752 | 1363 | again: |
247e743c CM |
1364 | lock_page(page); |
1365 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1366 | ClearPageChecked(page); | |
1367 | goto out_page; | |
1368 | } | |
1369 | ||
1370 | inode = page->mapping->host; | |
1371 | page_start = page_offset(page); | |
1372 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1373 | ||
1374 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1375 | |
1376 | /* already ordered? We're done */ | |
1377 | if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
1378 | EXTENT_ORDERED, 0)) { | |
247e743c | 1379 | goto out; |
4a096752 CM |
1380 | } |
1381 | ||
1382 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1383 | if (ordered) { | |
1384 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1385 | page_end, GFP_NOFS); | |
1386 | unlock_page(page); | |
1387 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1388 | goto again; | |
1389 | } | |
247e743c | 1390 | |
ea8c2819 | 1391 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1392 | ClearPageChecked(page); |
1393 | out: | |
1394 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1395 | out_page: | |
1396 | unlock_page(page); | |
1397 | page_cache_release(page); | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * There are a few paths in the higher layers of the kernel that directly | |
1402 | * set the page dirty bit without asking the filesystem if it is a | |
1403 | * good idea. This causes problems because we want to make sure COW | |
1404 | * properly happens and the data=ordered rules are followed. | |
1405 | * | |
c8b97818 | 1406 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1407 | * hasn't been properly setup for IO. We kick off an async process |
1408 | * to fix it up. The async helper will wait for ordered extents, set | |
1409 | * the delalloc bit and make it safe to write the page. | |
1410 | */ | |
b2950863 | 1411 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1412 | { |
1413 | struct inode *inode = page->mapping->host; | |
1414 | struct btrfs_writepage_fixup *fixup; | |
1415 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1416 | int ret; | |
1417 | ||
1418 | ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1419 | EXTENT_ORDERED, 0); | |
1420 | if (ret) | |
1421 | return 0; | |
1422 | ||
1423 | if (PageChecked(page)) | |
1424 | return -EAGAIN; | |
1425 | ||
1426 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1427 | if (!fixup) | |
1428 | return -EAGAIN; | |
f421950f | 1429 | |
247e743c CM |
1430 | SetPageChecked(page); |
1431 | page_cache_get(page); | |
1432 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1433 | fixup->page = page; | |
1434 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1435 | return -EAGAIN; | |
1436 | } | |
1437 | ||
d899e052 YZ |
1438 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1439 | struct inode *inode, u64 file_pos, | |
1440 | u64 disk_bytenr, u64 disk_num_bytes, | |
1441 | u64 num_bytes, u64 ram_bytes, | |
1442 | u8 compression, u8 encryption, | |
1443 | u16 other_encoding, int extent_type) | |
1444 | { | |
1445 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1446 | struct btrfs_file_extent_item *fi; | |
1447 | struct btrfs_path *path; | |
1448 | struct extent_buffer *leaf; | |
1449 | struct btrfs_key ins; | |
1450 | u64 hint; | |
1451 | int ret; | |
1452 | ||
1453 | path = btrfs_alloc_path(); | |
1454 | BUG_ON(!path); | |
1455 | ||
1456 | ret = btrfs_drop_extents(trans, root, inode, file_pos, | |
1457 | file_pos + num_bytes, file_pos, &hint); | |
1458 | BUG_ON(ret); | |
1459 | ||
1460 | ins.objectid = inode->i_ino; | |
1461 | ins.offset = file_pos; | |
1462 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1463 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1464 | BUG_ON(ret); | |
1465 | leaf = path->nodes[0]; | |
1466 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1467 | struct btrfs_file_extent_item); | |
1468 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1469 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1470 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1471 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1472 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1473 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1474 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1475 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1476 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1477 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
1478 | btrfs_mark_buffer_dirty(leaf); | |
1479 | ||
1480 | inode_add_bytes(inode, num_bytes); | |
1481 | btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0); | |
1482 | ||
1483 | ins.objectid = disk_bytenr; | |
1484 | ins.offset = disk_num_bytes; | |
1485 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1486 | ret = btrfs_alloc_reserved_extent(trans, root, leaf->start, | |
1487 | root->root_key.objectid, | |
1488 | trans->transid, inode->i_ino, &ins); | |
1489 | BUG_ON(ret); | |
1490 | ||
1491 | btrfs_free_path(path); | |
1492 | return 0; | |
1493 | } | |
1494 | ||
d352ac68 CM |
1495 | /* as ordered data IO finishes, this gets called so we can finish |
1496 | * an ordered extent if the range of bytes in the file it covers are | |
1497 | * fully written. | |
1498 | */ | |
211f90e6 | 1499 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1500 | { |
e6dcd2dc CM |
1501 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1502 | struct btrfs_trans_handle *trans; | |
1503 | struct btrfs_ordered_extent *ordered_extent; | |
1504 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
d899e052 | 1505 | int compressed = 0; |
e6dcd2dc CM |
1506 | int ret; |
1507 | ||
1508 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1509 | if (!ret) |
e6dcd2dc | 1510 | return 0; |
e6dcd2dc | 1511 | |
f9295749 | 1512 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc CM |
1513 | |
1514 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
1515 | BUG_ON(!ordered_extent); | |
7ea394f1 YZ |
1516 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1517 | goto nocow; | |
e6dcd2dc CM |
1518 | |
1519 | lock_extent(io_tree, ordered_extent->file_offset, | |
1520 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1521 | GFP_NOFS); | |
1522 | ||
c8b97818 | 1523 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1524 | compressed = 1; |
1525 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1526 | BUG_ON(compressed); | |
1527 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1528 | ordered_extent->file_offset, | |
1529 | ordered_extent->file_offset + | |
1530 | ordered_extent->len); | |
1531 | BUG_ON(ret); | |
1532 | } else { | |
1533 | ret = insert_reserved_file_extent(trans, inode, | |
1534 | ordered_extent->file_offset, | |
1535 | ordered_extent->start, | |
1536 | ordered_extent->disk_len, | |
1537 | ordered_extent->len, | |
1538 | ordered_extent->len, | |
1539 | compressed, 0, 0, | |
1540 | BTRFS_FILE_EXTENT_REG); | |
1541 | BUG_ON(ret); | |
1542 | } | |
e6dcd2dc CM |
1543 | unlock_extent(io_tree, ordered_extent->file_offset, |
1544 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1545 | GFP_NOFS); | |
7ea394f1 | 1546 | nocow: |
e6dcd2dc CM |
1547 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1548 | &ordered_extent->list); | |
1549 | ||
34353029 | 1550 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1551 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1552 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1553 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1554 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1555 | |
e6dcd2dc CM |
1556 | /* once for us */ |
1557 | btrfs_put_ordered_extent(ordered_extent); | |
1558 | /* once for the tree */ | |
1559 | btrfs_put_ordered_extent(ordered_extent); | |
1560 | ||
e6dcd2dc CM |
1561 | btrfs_end_transaction(trans, root); |
1562 | return 0; | |
1563 | } | |
1564 | ||
b2950863 | 1565 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1566 | struct extent_state *state, int uptodate) |
1567 | { | |
1568 | return btrfs_finish_ordered_io(page->mapping->host, start, end); | |
1569 | } | |
1570 | ||
d352ac68 CM |
1571 | /* |
1572 | * When IO fails, either with EIO or csum verification fails, we | |
1573 | * try other mirrors that might have a good copy of the data. This | |
1574 | * io_failure_record is used to record state as we go through all the | |
1575 | * mirrors. If another mirror has good data, the page is set up to date | |
1576 | * and things continue. If a good mirror can't be found, the original | |
1577 | * bio end_io callback is called to indicate things have failed. | |
1578 | */ | |
7e38326f CM |
1579 | struct io_failure_record { |
1580 | struct page *page; | |
1581 | u64 start; | |
1582 | u64 len; | |
1583 | u64 logical; | |
d20f7043 | 1584 | unsigned long bio_flags; |
7e38326f CM |
1585 | int last_mirror; |
1586 | }; | |
1587 | ||
b2950863 | 1588 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1589 | struct page *page, u64 start, u64 end, |
1590 | struct extent_state *state) | |
7e38326f CM |
1591 | { |
1592 | struct io_failure_record *failrec = NULL; | |
1593 | u64 private; | |
1594 | struct extent_map *em; | |
1595 | struct inode *inode = page->mapping->host; | |
1596 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1597 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1598 | struct bio *bio; |
1599 | int num_copies; | |
1600 | int ret; | |
1259ab75 | 1601 | int rw; |
7e38326f CM |
1602 | u64 logical; |
1603 | ||
1604 | ret = get_state_private(failure_tree, start, &private); | |
1605 | if (ret) { | |
7e38326f CM |
1606 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1607 | if (!failrec) | |
1608 | return -ENOMEM; | |
1609 | failrec->start = start; | |
1610 | failrec->len = end - start + 1; | |
1611 | failrec->last_mirror = 0; | |
d20f7043 | 1612 | failrec->bio_flags = 0; |
7e38326f | 1613 | |
3b951516 CM |
1614 | spin_lock(&em_tree->lock); |
1615 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
1616 | if (em->start > start || em->start + em->len < start) { | |
1617 | free_extent_map(em); | |
1618 | em = NULL; | |
1619 | } | |
1620 | spin_unlock(&em_tree->lock); | |
7e38326f CM |
1621 | |
1622 | if (!em || IS_ERR(em)) { | |
1623 | kfree(failrec); | |
1624 | return -EIO; | |
1625 | } | |
1626 | logical = start - em->start; | |
1627 | logical = em->block_start + logical; | |
d20f7043 CM |
1628 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1629 | logical = em->block_start; | |
1630 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1631 | } | |
7e38326f CM |
1632 | failrec->logical = logical; |
1633 | free_extent_map(em); | |
1634 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1635 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1636 | set_state_private(failure_tree, start, |
1637 | (u64)(unsigned long)failrec); | |
7e38326f | 1638 | } else { |
587f7704 | 1639 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1640 | } |
1641 | num_copies = btrfs_num_copies( | |
1642 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1643 | failrec->logical, failrec->len); | |
1644 | failrec->last_mirror++; | |
1645 | if (!state) { | |
cad321ad | 1646 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1647 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1648 | failrec->start, | |
1649 | EXTENT_LOCKED); | |
1650 | if (state && state->start != failrec->start) | |
1651 | state = NULL; | |
cad321ad | 1652 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1653 | } |
1654 | if (!state || failrec->last_mirror > num_copies) { | |
1655 | set_state_private(failure_tree, failrec->start, 0); | |
1656 | clear_extent_bits(failure_tree, failrec->start, | |
1657 | failrec->start + failrec->len - 1, | |
1658 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1659 | kfree(failrec); | |
1660 | return -EIO; | |
1661 | } | |
1662 | bio = bio_alloc(GFP_NOFS, 1); | |
1663 | bio->bi_private = state; | |
1664 | bio->bi_end_io = failed_bio->bi_end_io; | |
1665 | bio->bi_sector = failrec->logical >> 9; | |
1666 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1667 | bio->bi_size = 0; |
d20f7043 | 1668 | |
7e38326f | 1669 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1670 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1671 | rw = WRITE; | |
1672 | else | |
1673 | rw = READ; | |
1674 | ||
1675 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1676 | failrec->last_mirror, |
d20f7043 | 1677 | failrec->bio_flags); |
1259ab75 CM |
1678 | return 0; |
1679 | } | |
1680 | ||
d352ac68 CM |
1681 | /* |
1682 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1683 | * to see if we need to process or clean up an io_failure_record | |
1684 | */ | |
b2950863 | 1685 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1686 | { |
1687 | u64 private; | |
1688 | u64 private_failure; | |
1689 | struct io_failure_record *failure; | |
1690 | int ret; | |
1691 | ||
1692 | private = 0; | |
1693 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1694 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1695 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1696 | start, &private_failure); | |
1697 | if (ret == 0) { | |
1698 | failure = (struct io_failure_record *)(unsigned long) | |
1699 | private_failure; | |
1700 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1701 | failure->start, 0); | |
1702 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1703 | failure->start, | |
1704 | failure->start + failure->len - 1, | |
1705 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1706 | GFP_NOFS); | |
1707 | kfree(failure); | |
1708 | } | |
1709 | } | |
7e38326f CM |
1710 | return 0; |
1711 | } | |
1712 | ||
d352ac68 CM |
1713 | /* |
1714 | * when reads are done, we need to check csums to verify the data is correct | |
1715 | * if there's a match, we allow the bio to finish. If not, we go through | |
1716 | * the io_failure_record routines to find good copies | |
1717 | */ | |
b2950863 | 1718 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1719 | struct extent_state *state) |
07157aac | 1720 | { |
35ebb934 | 1721 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1722 | struct inode *inode = page->mapping->host; |
d1310b2e | 1723 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1724 | char *kaddr; |
aadfeb6e | 1725 | u64 private = ~(u32)0; |
07157aac | 1726 | int ret; |
ff79f819 CM |
1727 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1728 | u32 csum = ~(u32)0; | |
d1310b2e | 1729 | |
d20f7043 CM |
1730 | if (PageChecked(page)) { |
1731 | ClearPageChecked(page); | |
1732 | goto good; | |
1733 | } | |
17d217fe YZ |
1734 | if (btrfs_test_flag(inode, NODATASUM)) |
1735 | return 0; | |
1736 | ||
1737 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
1738 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) { | |
1739 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, | |
1740 | GFP_NOFS); | |
b6cda9bc | 1741 | return 0; |
17d217fe | 1742 | } |
d20f7043 | 1743 | |
c2e639f0 | 1744 | if (state && state->start == start) { |
70dec807 CM |
1745 | private = state->private; |
1746 | ret = 0; | |
1747 | } else { | |
1748 | ret = get_state_private(io_tree, start, &private); | |
1749 | } | |
9ab86c8e | 1750 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1751 | if (ret) |
07157aac | 1752 | goto zeroit; |
d397712b | 1753 | |
ff79f819 CM |
1754 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1755 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1756 | if (csum != private) |
07157aac | 1757 | goto zeroit; |
d397712b | 1758 | |
9ab86c8e | 1759 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1760 | good: |
7e38326f CM |
1761 | /* if the io failure tree for this inode is non-empty, |
1762 | * check to see if we've recovered from a failed IO | |
1763 | */ | |
1259ab75 | 1764 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1765 | return 0; |
1766 | ||
1767 | zeroit: | |
d397712b CM |
1768 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " |
1769 | "private %llu\n", page->mapping->host->i_ino, | |
1770 | (unsigned long long)start, csum, | |
1771 | (unsigned long long)private); | |
db94535d CM |
1772 | memset(kaddr + offset, 1, end - start + 1); |
1773 | flush_dcache_page(page); | |
9ab86c8e | 1774 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1775 | if (private == 0) |
1776 | return 0; | |
7e38326f | 1777 | return -EIO; |
07157aac | 1778 | } |
b888db2b | 1779 | |
7b128766 JB |
1780 | /* |
1781 | * This creates an orphan entry for the given inode in case something goes | |
1782 | * wrong in the middle of an unlink/truncate. | |
1783 | */ | |
1784 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1785 | { | |
1786 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1787 | int ret = 0; | |
1788 | ||
bcc63abb | 1789 | spin_lock(&root->list_lock); |
7b128766 JB |
1790 | |
1791 | /* already on the orphan list, we're good */ | |
1792 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1793 | spin_unlock(&root->list_lock); |
7b128766 JB |
1794 | return 0; |
1795 | } | |
1796 | ||
1797 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
1798 | ||
bcc63abb | 1799 | spin_unlock(&root->list_lock); |
7b128766 JB |
1800 | |
1801 | /* | |
1802 | * insert an orphan item to track this unlinked/truncated file | |
1803 | */ | |
1804 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
1805 | ||
1806 | return ret; | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
1811 | * item for this particular inode. | |
1812 | */ | |
1813 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
1814 | { | |
1815 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1816 | int ret = 0; | |
1817 | ||
bcc63abb | 1818 | spin_lock(&root->list_lock); |
7b128766 JB |
1819 | |
1820 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1821 | spin_unlock(&root->list_lock); |
7b128766 JB |
1822 | return 0; |
1823 | } | |
1824 | ||
1825 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
1826 | if (!trans) { | |
bcc63abb | 1827 | spin_unlock(&root->list_lock); |
7b128766 JB |
1828 | return 0; |
1829 | } | |
1830 | ||
bcc63abb | 1831 | spin_unlock(&root->list_lock); |
7b128766 JB |
1832 | |
1833 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
1834 | ||
1835 | return ret; | |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * this cleans up any orphans that may be left on the list from the last use | |
1840 | * of this root. | |
1841 | */ | |
1842 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
1843 | { | |
1844 | struct btrfs_path *path; | |
1845 | struct extent_buffer *leaf; | |
1846 | struct btrfs_item *item; | |
1847 | struct btrfs_key key, found_key; | |
1848 | struct btrfs_trans_handle *trans; | |
1849 | struct inode *inode; | |
1850 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
1851 | ||
7b128766 JB |
1852 | path = btrfs_alloc_path(); |
1853 | if (!path) | |
1854 | return; | |
1855 | path->reada = -1; | |
1856 | ||
1857 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
1858 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
1859 | key.offset = (u64)-1; | |
1860 | ||
7b128766 JB |
1861 | |
1862 | while (1) { | |
1863 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1864 | if (ret < 0) { | |
1865 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
1866 | "\n", ret); | |
1867 | break; | |
1868 | } | |
1869 | ||
1870 | /* | |
1871 | * if ret == 0 means we found what we were searching for, which | |
1872 | * is weird, but possible, so only screw with path if we didnt | |
1873 | * find the key and see if we have stuff that matches | |
1874 | */ | |
1875 | if (ret > 0) { | |
1876 | if (path->slots[0] == 0) | |
1877 | break; | |
1878 | path->slots[0]--; | |
1879 | } | |
1880 | ||
1881 | /* pull out the item */ | |
1882 | leaf = path->nodes[0]; | |
1883 | item = btrfs_item_nr(leaf, path->slots[0]); | |
1884 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1885 | ||
1886 | /* make sure the item matches what we want */ | |
1887 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
1888 | break; | |
1889 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
1890 | break; | |
1891 | ||
1892 | /* release the path since we're done with it */ | |
1893 | btrfs_release_path(root, path); | |
1894 | ||
1895 | /* | |
1896 | * this is where we are basically btrfs_lookup, without the | |
1897 | * crossing root thing. we store the inode number in the | |
1898 | * offset of the orphan item. | |
1899 | */ | |
5b21f2ed | 1900 | inode = btrfs_iget_locked(root->fs_info->sb, |
7b128766 JB |
1901 | found_key.offset, root); |
1902 | if (!inode) | |
1903 | break; | |
1904 | ||
1905 | if (inode->i_state & I_NEW) { | |
1906 | BTRFS_I(inode)->root = root; | |
1907 | ||
1908 | /* have to set the location manually */ | |
1909 | BTRFS_I(inode)->location.objectid = inode->i_ino; | |
1910 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
1911 | BTRFS_I(inode)->location.offset = 0; | |
1912 | ||
1913 | btrfs_read_locked_inode(inode); | |
1914 | unlock_new_inode(inode); | |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * add this inode to the orphan list so btrfs_orphan_del does | |
1919 | * the proper thing when we hit it | |
1920 | */ | |
bcc63abb | 1921 | spin_lock(&root->list_lock); |
7b128766 | 1922 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 1923 | spin_unlock(&root->list_lock); |
7b128766 JB |
1924 | |
1925 | /* | |
1926 | * if this is a bad inode, means we actually succeeded in | |
1927 | * removing the inode, but not the orphan record, which means | |
1928 | * we need to manually delete the orphan since iput will just | |
1929 | * do a destroy_inode | |
1930 | */ | |
1931 | if (is_bad_inode(inode)) { | |
5b21f2ed | 1932 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 1933 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 1934 | btrfs_end_transaction(trans, root); |
7b128766 JB |
1935 | iput(inode); |
1936 | continue; | |
1937 | } | |
1938 | ||
1939 | /* if we have links, this was a truncate, lets do that */ | |
1940 | if (inode->i_nlink) { | |
1941 | nr_truncate++; | |
1942 | btrfs_truncate(inode); | |
1943 | } else { | |
1944 | nr_unlink++; | |
1945 | } | |
1946 | ||
1947 | /* this will do delete_inode and everything for us */ | |
1948 | iput(inode); | |
1949 | } | |
1950 | ||
1951 | if (nr_unlink) | |
1952 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
1953 | if (nr_truncate) | |
1954 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
1955 | ||
1956 | btrfs_free_path(path); | |
7b128766 JB |
1957 | } |
1958 | ||
d352ac68 CM |
1959 | /* |
1960 | * read an inode from the btree into the in-memory inode | |
1961 | */ | |
39279cc3 CM |
1962 | void btrfs_read_locked_inode(struct inode *inode) |
1963 | { | |
1964 | struct btrfs_path *path; | |
5f39d397 | 1965 | struct extent_buffer *leaf; |
39279cc3 | 1966 | struct btrfs_inode_item *inode_item; |
0b86a832 | 1967 | struct btrfs_timespec *tspec; |
39279cc3 CM |
1968 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1969 | struct btrfs_key location; | |
1970 | u64 alloc_group_block; | |
618e21d5 | 1971 | u32 rdev; |
39279cc3 CM |
1972 | int ret; |
1973 | ||
1974 | path = btrfs_alloc_path(); | |
1975 | BUG_ON(!path); | |
39279cc3 | 1976 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 1977 | |
39279cc3 | 1978 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 1979 | if (ret) |
39279cc3 | 1980 | goto make_bad; |
39279cc3 | 1981 | |
5f39d397 CM |
1982 | leaf = path->nodes[0]; |
1983 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
1984 | struct btrfs_inode_item); | |
1985 | ||
1986 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
1987 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
1988 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
1989 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 1990 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
1991 | |
1992 | tspec = btrfs_inode_atime(inode_item); | |
1993 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
1994 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
1995 | ||
1996 | tspec = btrfs_inode_mtime(inode_item); | |
1997 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
1998 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
1999 | ||
2000 | tspec = btrfs_inode_ctime(inode_item); | |
2001 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2002 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2003 | ||
a76a3cd4 | 2004 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2005 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2006 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2007 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2008 | inode->i_rdev = 0; |
5f39d397 CM |
2009 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2010 | ||
aec7477b | 2011 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2012 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2013 | |
5f39d397 | 2014 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
d2fb3437 YZ |
2015 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2016 | alloc_group_block, 0); | |
39279cc3 CM |
2017 | btrfs_free_path(path); |
2018 | inode_item = NULL; | |
2019 | ||
39279cc3 | 2020 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2021 | case S_IFREG: |
2022 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2023 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2024 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2025 | inode->i_fop = &btrfs_file_operations; |
2026 | inode->i_op = &btrfs_file_inode_operations; | |
2027 | break; | |
2028 | case S_IFDIR: | |
2029 | inode->i_fop = &btrfs_dir_file_operations; | |
2030 | if (root == root->fs_info->tree_root) | |
2031 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2032 | else | |
2033 | inode->i_op = &btrfs_dir_inode_operations; | |
2034 | break; | |
2035 | case S_IFLNK: | |
2036 | inode->i_op = &btrfs_symlink_inode_operations; | |
2037 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2038 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2039 | break; |
618e21d5 JB |
2040 | default: |
2041 | init_special_inode(inode, inode->i_mode, rdev); | |
2042 | break; | |
39279cc3 CM |
2043 | } |
2044 | return; | |
2045 | ||
2046 | make_bad: | |
39279cc3 | 2047 | btrfs_free_path(path); |
39279cc3 CM |
2048 | make_bad_inode(inode); |
2049 | } | |
2050 | ||
d352ac68 CM |
2051 | /* |
2052 | * given a leaf and an inode, copy the inode fields into the leaf | |
2053 | */ | |
e02119d5 CM |
2054 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2055 | struct extent_buffer *leaf, | |
5f39d397 | 2056 | struct btrfs_inode_item *item, |
39279cc3 CM |
2057 | struct inode *inode) |
2058 | { | |
5f39d397 CM |
2059 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2060 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2061 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2062 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2063 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2064 | ||
2065 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2066 | inode->i_atime.tv_sec); | |
2067 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2068 | inode->i_atime.tv_nsec); | |
2069 | ||
2070 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2071 | inode->i_mtime.tv_sec); | |
2072 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2073 | inode->i_mtime.tv_nsec); | |
2074 | ||
2075 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2076 | inode->i_ctime.tv_sec); | |
2077 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2078 | inode->i_ctime.tv_nsec); | |
2079 | ||
a76a3cd4 | 2080 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2081 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2082 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2083 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2084 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2085 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2086 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2087 | } |
2088 | ||
d352ac68 CM |
2089 | /* |
2090 | * copy everything in the in-memory inode into the btree. | |
2091 | */ | |
d397712b CM |
2092 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2093 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2094 | { |
2095 | struct btrfs_inode_item *inode_item; | |
2096 | struct btrfs_path *path; | |
5f39d397 | 2097 | struct extent_buffer *leaf; |
39279cc3 CM |
2098 | int ret; |
2099 | ||
2100 | path = btrfs_alloc_path(); | |
2101 | BUG_ON(!path); | |
39279cc3 CM |
2102 | ret = btrfs_lookup_inode(trans, root, path, |
2103 | &BTRFS_I(inode)->location, 1); | |
2104 | if (ret) { | |
2105 | if (ret > 0) | |
2106 | ret = -ENOENT; | |
2107 | goto failed; | |
2108 | } | |
2109 | ||
5f39d397 CM |
2110 | leaf = path->nodes[0]; |
2111 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2112 | struct btrfs_inode_item); |
2113 | ||
e02119d5 | 2114 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2115 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2116 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2117 | ret = 0; |
2118 | failed: | |
39279cc3 CM |
2119 | btrfs_free_path(path); |
2120 | return ret; | |
2121 | } | |
2122 | ||
2123 | ||
d352ac68 CM |
2124 | /* |
2125 | * unlink helper that gets used here in inode.c and in the tree logging | |
2126 | * recovery code. It remove a link in a directory with a given name, and | |
2127 | * also drops the back refs in the inode to the directory | |
2128 | */ | |
e02119d5 CM |
2129 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2130 | struct btrfs_root *root, | |
2131 | struct inode *dir, struct inode *inode, | |
2132 | const char *name, int name_len) | |
39279cc3 CM |
2133 | { |
2134 | struct btrfs_path *path; | |
39279cc3 | 2135 | int ret = 0; |
5f39d397 | 2136 | struct extent_buffer *leaf; |
39279cc3 | 2137 | struct btrfs_dir_item *di; |
5f39d397 | 2138 | struct btrfs_key key; |
aec7477b | 2139 | u64 index; |
39279cc3 CM |
2140 | |
2141 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2142 | if (!path) { |
2143 | ret = -ENOMEM; | |
2144 | goto err; | |
2145 | } | |
2146 | ||
39279cc3 CM |
2147 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2148 | name, name_len, -1); | |
2149 | if (IS_ERR(di)) { | |
2150 | ret = PTR_ERR(di); | |
2151 | goto err; | |
2152 | } | |
2153 | if (!di) { | |
2154 | ret = -ENOENT; | |
2155 | goto err; | |
2156 | } | |
5f39d397 CM |
2157 | leaf = path->nodes[0]; |
2158 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2159 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2160 | if (ret) |
2161 | goto err; | |
39279cc3 CM |
2162 | btrfs_release_path(root, path); |
2163 | ||
aec7477b | 2164 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2165 | inode->i_ino, |
2166 | dir->i_ino, &index); | |
aec7477b | 2167 | if (ret) { |
d397712b | 2168 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2169 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2170 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2171 | goto err; |
2172 | } | |
2173 | ||
39279cc3 | 2174 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2175 | index, name, name_len, -1); |
39279cc3 CM |
2176 | if (IS_ERR(di)) { |
2177 | ret = PTR_ERR(di); | |
2178 | goto err; | |
2179 | } | |
2180 | if (!di) { | |
2181 | ret = -ENOENT; | |
2182 | goto err; | |
2183 | } | |
2184 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2185 | btrfs_release_path(root, path); |
39279cc3 | 2186 | |
e02119d5 CM |
2187 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2188 | inode, dir->i_ino); | |
49eb7e46 CM |
2189 | BUG_ON(ret != 0 && ret != -ENOENT); |
2190 | if (ret != -ENOENT) | |
2191 | BTRFS_I(dir)->log_dirty_trans = trans->transid; | |
e02119d5 CM |
2192 | |
2193 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2194 | dir, index); | |
2195 | BUG_ON(ret); | |
39279cc3 CM |
2196 | err: |
2197 | btrfs_free_path(path); | |
e02119d5 CM |
2198 | if (ret) |
2199 | goto out; | |
2200 | ||
2201 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2202 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2203 | btrfs_update_inode(trans, root, dir); | |
2204 | btrfs_drop_nlink(inode); | |
2205 | ret = btrfs_update_inode(trans, root, inode); | |
2206 | dir->i_sb->s_dirt = 1; | |
2207 | out: | |
39279cc3 CM |
2208 | return ret; |
2209 | } | |
2210 | ||
2211 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2212 | { | |
2213 | struct btrfs_root *root; | |
2214 | struct btrfs_trans_handle *trans; | |
7b128766 | 2215 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2216 | int ret; |
1832a6d5 | 2217 | unsigned long nr = 0; |
39279cc3 CM |
2218 | |
2219 | root = BTRFS_I(dir)->root; | |
1832a6d5 CM |
2220 | |
2221 | ret = btrfs_check_free_space(root, 1, 1); | |
2222 | if (ret) | |
2223 | goto fail; | |
2224 | ||
39279cc3 | 2225 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2226 | |
39279cc3 | 2227 | btrfs_set_trans_block_group(trans, dir); |
e02119d5 CM |
2228 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2229 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2230 | |
2231 | if (inode->i_nlink == 0) | |
2232 | ret = btrfs_orphan_add(trans, inode); | |
2233 | ||
d3c2fdcf | 2234 | nr = trans->blocks_used; |
5f39d397 | 2235 | |
89ce8a63 | 2236 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2237 | fail: |
d3c2fdcf | 2238 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2239 | return ret; |
2240 | } | |
2241 | ||
2242 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2243 | { | |
2244 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2245 | int err = 0; |
39279cc3 CM |
2246 | int ret; |
2247 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2248 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2249 | unsigned long nr = 0; |
39279cc3 | 2250 | |
3394e160 CM |
2251 | /* |
2252 | * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir | |
2253 | * the root of a subvolume or snapshot | |
2254 | */ | |
2255 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || | |
2256 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
134d4512 | 2257 | return -ENOTEMPTY; |
925baedd | 2258 | } |
134d4512 | 2259 | |
1832a6d5 CM |
2260 | ret = btrfs_check_free_space(root, 1, 1); |
2261 | if (ret) | |
2262 | goto fail; | |
2263 | ||
39279cc3 CM |
2264 | trans = btrfs_start_transaction(root, 1); |
2265 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2266 | |
7b128766 JB |
2267 | err = btrfs_orphan_add(trans, inode); |
2268 | if (err) | |
2269 | goto fail_trans; | |
2270 | ||
39279cc3 | 2271 | /* now the directory is empty */ |
e02119d5 CM |
2272 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2273 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2274 | if (!err) |
dbe674a9 | 2275 | btrfs_i_size_write(inode, 0); |
3954401f | 2276 | |
7b128766 | 2277 | fail_trans: |
d3c2fdcf | 2278 | nr = trans->blocks_used; |
89ce8a63 | 2279 | ret = btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2280 | fail: |
d3c2fdcf | 2281 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2282 | |
39279cc3 CM |
2283 | if (ret && !err) |
2284 | err = ret; | |
2285 | return err; | |
2286 | } | |
2287 | ||
d20f7043 | 2288 | #if 0 |
323ac95b CM |
2289 | /* |
2290 | * when truncating bytes in a file, it is possible to avoid reading | |
2291 | * the leaves that contain only checksum items. This can be the | |
2292 | * majority of the IO required to delete a large file, but it must | |
2293 | * be done carefully. | |
2294 | * | |
2295 | * The keys in the level just above the leaves are checked to make sure | |
2296 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2297 | * after the new size. | |
2298 | * | |
2299 | * Then the key for the next leaf is checked to make sure it also has | |
2300 | * a checksum item for the same file. If it does, we know our target leaf | |
2301 | * contains only checksum items, and it can be safely freed without reading | |
2302 | * it. | |
2303 | * | |
2304 | * This is just an optimization targeted at large files. It may do | |
2305 | * nothing. It will return 0 unless things went badly. | |
2306 | */ | |
2307 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2308 | struct btrfs_root *root, | |
2309 | struct btrfs_path *path, | |
2310 | struct inode *inode, u64 new_size) | |
2311 | { | |
2312 | struct btrfs_key key; | |
2313 | int ret; | |
2314 | int nritems; | |
2315 | struct btrfs_key found_key; | |
2316 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2317 | struct btrfs_leaf_ref *ref; |
2318 | u64 leaf_gen; | |
2319 | u64 leaf_start; | |
323ac95b CM |
2320 | |
2321 | path->lowest_level = 1; | |
2322 | key.objectid = inode->i_ino; | |
2323 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2324 | key.offset = new_size; | |
2325 | again: | |
2326 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2327 | if (ret < 0) | |
2328 | goto out; | |
2329 | ||
2330 | if (path->nodes[1] == NULL) { | |
2331 | ret = 0; | |
2332 | goto out; | |
2333 | } | |
2334 | ret = 0; | |
2335 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2336 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2337 | ||
2338 | if (!nritems) | |
2339 | goto out; | |
2340 | ||
2341 | if (path->slots[1] >= nritems) | |
2342 | goto next_node; | |
2343 | ||
2344 | /* did we find a key greater than anything we want to delete? */ | |
2345 | if (found_key.objectid > inode->i_ino || | |
2346 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2347 | goto out; | |
2348 | ||
2349 | /* we check the next key in the node to make sure the leave contains | |
2350 | * only checksum items. This comparison doesn't work if our | |
2351 | * leaf is the last one in the node | |
2352 | */ | |
2353 | if (path->slots[1] + 1 >= nritems) { | |
2354 | next_node: | |
2355 | /* search forward from the last key in the node, this | |
2356 | * will bring us into the next node in the tree | |
2357 | */ | |
2358 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2359 | ||
2360 | /* unlikely, but we inc below, so check to be safe */ | |
2361 | if (found_key.offset == (u64)-1) | |
2362 | goto out; | |
2363 | ||
2364 | /* search_forward needs a path with locks held, do the | |
2365 | * search again for the original key. It is possible | |
2366 | * this will race with a balance and return a path that | |
2367 | * we could modify, but this drop is just an optimization | |
2368 | * and is allowed to miss some leaves. | |
2369 | */ | |
2370 | btrfs_release_path(root, path); | |
2371 | found_key.offset++; | |
2372 | ||
2373 | /* setup a max key for search_forward */ | |
2374 | other_key.offset = (u64)-1; | |
2375 | other_key.type = key.type; | |
2376 | other_key.objectid = key.objectid; | |
2377 | ||
2378 | path->keep_locks = 1; | |
2379 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2380 | path, 0, 0); | |
2381 | path->keep_locks = 0; | |
2382 | if (ret || found_key.objectid != key.objectid || | |
2383 | found_key.type != key.type) { | |
2384 | ret = 0; | |
2385 | goto out; | |
2386 | } | |
2387 | ||
2388 | key.offset = found_key.offset; | |
2389 | btrfs_release_path(root, path); | |
2390 | cond_resched(); | |
2391 | goto again; | |
2392 | } | |
2393 | ||
2394 | /* we know there's one more slot after us in the tree, | |
2395 | * read that key so we can verify it is also a checksum item | |
2396 | */ | |
2397 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2398 | ||
2399 | if (found_key.objectid < inode->i_ino) | |
2400 | goto next_key; | |
2401 | ||
2402 | if (found_key.type != key.type || found_key.offset < new_size) | |
2403 | goto next_key; | |
2404 | ||
2405 | /* | |
2406 | * if the key for the next leaf isn't a csum key from this objectid, | |
2407 | * we can't be sure there aren't good items inside this leaf. | |
2408 | * Bail out | |
2409 | */ | |
2410 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2411 | goto out; | |
2412 | ||
5b84e8d6 YZ |
2413 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2414 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2415 | /* |
2416 | * it is safe to delete this leaf, it contains only | |
2417 | * csum items from this inode at an offset >= new_size | |
2418 | */ | |
5b84e8d6 | 2419 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2420 | BUG_ON(ret); |
2421 | ||
5b84e8d6 YZ |
2422 | if (root->ref_cows && leaf_gen < trans->transid) { |
2423 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2424 | if (ref) { | |
2425 | ref->root_gen = root->root_key.offset; | |
2426 | ref->bytenr = leaf_start; | |
2427 | ref->owner = 0; | |
2428 | ref->generation = leaf_gen; | |
2429 | ref->nritems = 0; | |
2430 | ||
2431 | ret = btrfs_add_leaf_ref(root, ref, 0); | |
2432 | WARN_ON(ret); | |
2433 | btrfs_free_leaf_ref(root, ref); | |
2434 | } else { | |
2435 | WARN_ON(1); | |
2436 | } | |
2437 | } | |
323ac95b CM |
2438 | next_key: |
2439 | btrfs_release_path(root, path); | |
2440 | ||
2441 | if (other_key.objectid == inode->i_ino && | |
2442 | other_key.type == key.type && other_key.offset > key.offset) { | |
2443 | key.offset = other_key.offset; | |
2444 | cond_resched(); | |
2445 | goto again; | |
2446 | } | |
2447 | ret = 0; | |
2448 | out: | |
2449 | /* fixup any changes we've made to the path */ | |
2450 | path->lowest_level = 0; | |
2451 | path->keep_locks = 0; | |
2452 | btrfs_release_path(root, path); | |
2453 | return ret; | |
2454 | } | |
2455 | ||
d20f7043 CM |
2456 | #endif |
2457 | ||
39279cc3 CM |
2458 | /* |
2459 | * this can truncate away extent items, csum items and directory items. | |
2460 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2461 | * any higher than new_size |
39279cc3 CM |
2462 | * |
2463 | * csum items that cross the new i_size are truncated to the new size | |
2464 | * as well. | |
7b128766 JB |
2465 | * |
2466 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2467 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2468 | */ |
e02119d5 CM |
2469 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2470 | struct btrfs_root *root, | |
2471 | struct inode *inode, | |
2472 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2473 | { |
2474 | int ret; | |
2475 | struct btrfs_path *path; | |
2476 | struct btrfs_key key; | |
5f39d397 | 2477 | struct btrfs_key found_key; |
39279cc3 | 2478 | u32 found_type; |
5f39d397 | 2479 | struct extent_buffer *leaf; |
39279cc3 CM |
2480 | struct btrfs_file_extent_item *fi; |
2481 | u64 extent_start = 0; | |
db94535d | 2482 | u64 extent_num_bytes = 0; |
39279cc3 | 2483 | u64 item_end = 0; |
7bb86316 | 2484 | u64 root_gen = 0; |
d8d5f3e1 | 2485 | u64 root_owner = 0; |
39279cc3 CM |
2486 | int found_extent; |
2487 | int del_item; | |
85e21bac CM |
2488 | int pending_del_nr = 0; |
2489 | int pending_del_slot = 0; | |
179e29e4 | 2490 | int extent_type = -1; |
771ed689 | 2491 | int encoding; |
3b951516 | 2492 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2493 | |
e02119d5 | 2494 | if (root->ref_cows) |
5b21f2ed | 2495 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 | 2496 | path = btrfs_alloc_path(); |
3c69faec | 2497 | path->reada = -1; |
39279cc3 | 2498 | BUG_ON(!path); |
5f39d397 | 2499 | |
39279cc3 CM |
2500 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2501 | key.objectid = inode->i_ino; | |
2502 | key.offset = (u64)-1; | |
5f39d397 CM |
2503 | key.type = (u8)-1; |
2504 | ||
85e21bac | 2505 | btrfs_init_path(path); |
323ac95b | 2506 | |
85e21bac CM |
2507 | search_again: |
2508 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
d397712b | 2509 | if (ret < 0) |
85e21bac | 2510 | goto error; |
d397712b | 2511 | |
85e21bac | 2512 | if (ret > 0) { |
e02119d5 CM |
2513 | /* there are no items in the tree for us to truncate, we're |
2514 | * done | |
2515 | */ | |
2516 | if (path->slots[0] == 0) { | |
2517 | ret = 0; | |
2518 | goto error; | |
2519 | } | |
85e21bac CM |
2520 | path->slots[0]--; |
2521 | } | |
2522 | ||
d397712b | 2523 | while (1) { |
39279cc3 | 2524 | fi = NULL; |
5f39d397 CM |
2525 | leaf = path->nodes[0]; |
2526 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2527 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2528 | encoding = 0; |
39279cc3 | 2529 | |
5f39d397 | 2530 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2531 | break; |
5f39d397 | 2532 | |
85e21bac | 2533 | if (found_type < min_type) |
39279cc3 CM |
2534 | break; |
2535 | ||
5f39d397 | 2536 | item_end = found_key.offset; |
39279cc3 | 2537 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2538 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2539 | struct btrfs_file_extent_item); |
179e29e4 | 2540 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2541 | encoding = btrfs_file_extent_compression(leaf, fi); |
2542 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2543 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2544 | ||
179e29e4 | 2545 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2546 | item_end += |
db94535d | 2547 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2548 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2549 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2550 | fi); |
39279cc3 | 2551 | } |
008630c1 | 2552 | item_end--; |
39279cc3 | 2553 | } |
e02119d5 | 2554 | if (item_end < new_size) { |
d397712b | 2555 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2556 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2557 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2558 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2559 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2560 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2561 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2562 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2563 | else if (found_type) |
b888db2b | 2564 | found_type--; |
d397712b | 2565 | else |
b888db2b | 2566 | break; |
a61721d5 | 2567 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2568 | goto next; |
39279cc3 | 2569 | } |
e02119d5 | 2570 | if (found_key.offset >= new_size) |
39279cc3 CM |
2571 | del_item = 1; |
2572 | else | |
2573 | del_item = 0; | |
2574 | found_extent = 0; | |
2575 | ||
2576 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2577 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2578 | goto delete; | |
2579 | ||
2580 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2581 | u64 num_dec; |
db94535d | 2582 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2583 | if (!del_item && !encoding) { |
db94535d CM |
2584 | u64 orig_num_bytes = |
2585 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2586 | extent_num_bytes = new_size - |
5f39d397 | 2587 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2588 | extent_num_bytes = extent_num_bytes & |
2589 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2590 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2591 | extent_num_bytes); | |
2592 | num_dec = (orig_num_bytes - | |
9069218d | 2593 | extent_num_bytes); |
e02119d5 | 2594 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2595 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2596 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2597 | } else { |
db94535d CM |
2598 | extent_num_bytes = |
2599 | btrfs_file_extent_disk_num_bytes(leaf, | |
2600 | fi); | |
39279cc3 | 2601 | /* FIXME blocksize != 4096 */ |
9069218d | 2602 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2603 | if (extent_start != 0) { |
2604 | found_extent = 1; | |
e02119d5 | 2605 | if (root->ref_cows) |
a76a3cd4 | 2606 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2607 | } |
31840ae1 | 2608 | root_gen = btrfs_header_generation(leaf); |
d8d5f3e1 | 2609 | root_owner = btrfs_header_owner(leaf); |
39279cc3 | 2610 | } |
9069218d | 2611 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2612 | /* |
2613 | * we can't truncate inline items that have had | |
2614 | * special encodings | |
2615 | */ | |
2616 | if (!del_item && | |
2617 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2618 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2619 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2620 | u32 size = new_size - found_key.offset; |
2621 | ||
2622 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2623 | inode_sub_bytes(inode, item_end + 1 - |
2624 | new_size); | |
e02119d5 CM |
2625 | } |
2626 | size = | |
2627 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2628 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2629 | size, 1); |
9069218d | 2630 | BUG_ON(ret); |
e02119d5 | 2631 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2632 | inode_sub_bytes(inode, item_end + 1 - |
2633 | found_key.offset); | |
9069218d | 2634 | } |
39279cc3 | 2635 | } |
179e29e4 | 2636 | delete: |
39279cc3 | 2637 | if (del_item) { |
85e21bac CM |
2638 | if (!pending_del_nr) { |
2639 | /* no pending yet, add ourselves */ | |
2640 | pending_del_slot = path->slots[0]; | |
2641 | pending_del_nr = 1; | |
2642 | } else if (pending_del_nr && | |
2643 | path->slots[0] + 1 == pending_del_slot) { | |
2644 | /* hop on the pending chunk */ | |
2645 | pending_del_nr++; | |
2646 | pending_del_slot = path->slots[0]; | |
2647 | } else { | |
d397712b | 2648 | BUG(); |
85e21bac | 2649 | } |
39279cc3 CM |
2650 | } else { |
2651 | break; | |
2652 | } | |
39279cc3 CM |
2653 | if (found_extent) { |
2654 | ret = btrfs_free_extent(trans, root, extent_start, | |
7bb86316 | 2655 | extent_num_bytes, |
31840ae1 | 2656 | leaf->start, root_owner, |
3bb1a1bc | 2657 | root_gen, inode->i_ino, 0); |
39279cc3 CM |
2658 | BUG_ON(ret); |
2659 | } | |
85e21bac CM |
2660 | next: |
2661 | if (path->slots[0] == 0) { | |
2662 | if (pending_del_nr) | |
2663 | goto del_pending; | |
2664 | btrfs_release_path(root, path); | |
2665 | goto search_again; | |
2666 | } | |
2667 | ||
2668 | path->slots[0]--; | |
2669 | if (pending_del_nr && | |
2670 | path->slots[0] + 1 != pending_del_slot) { | |
2671 | struct btrfs_key debug; | |
2672 | del_pending: | |
2673 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2674 | pending_del_slot); | |
2675 | ret = btrfs_del_items(trans, root, path, | |
2676 | pending_del_slot, | |
2677 | pending_del_nr); | |
2678 | BUG_ON(ret); | |
2679 | pending_del_nr = 0; | |
2680 | btrfs_release_path(root, path); | |
2681 | goto search_again; | |
2682 | } | |
39279cc3 CM |
2683 | } |
2684 | ret = 0; | |
2685 | error: | |
85e21bac CM |
2686 | if (pending_del_nr) { |
2687 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
2688 | pending_del_nr); | |
2689 | } | |
39279cc3 CM |
2690 | btrfs_free_path(path); |
2691 | inode->i_sb->s_dirt = 1; | |
2692 | return ret; | |
2693 | } | |
2694 | ||
2695 | /* | |
2696 | * taken from block_truncate_page, but does cow as it zeros out | |
2697 | * any bytes left in the last page in the file. | |
2698 | */ | |
2699 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
2700 | { | |
2701 | struct inode *inode = mapping->host; | |
db94535d | 2702 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
2703 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2704 | struct btrfs_ordered_extent *ordered; | |
2705 | char *kaddr; | |
db94535d | 2706 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
2707 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
2708 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
2709 | struct page *page; | |
39279cc3 | 2710 | int ret = 0; |
a52d9a80 | 2711 | u64 page_start; |
e6dcd2dc | 2712 | u64 page_end; |
39279cc3 CM |
2713 | |
2714 | if ((offset & (blocksize - 1)) == 0) | |
2715 | goto out; | |
2716 | ||
2717 | ret = -ENOMEM; | |
211c17f5 | 2718 | again: |
39279cc3 CM |
2719 | page = grab_cache_page(mapping, index); |
2720 | if (!page) | |
2721 | goto out; | |
e6dcd2dc CM |
2722 | |
2723 | page_start = page_offset(page); | |
2724 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2725 | ||
39279cc3 | 2726 | if (!PageUptodate(page)) { |
9ebefb18 | 2727 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 2728 | lock_page(page); |
211c17f5 CM |
2729 | if (page->mapping != mapping) { |
2730 | unlock_page(page); | |
2731 | page_cache_release(page); | |
2732 | goto again; | |
2733 | } | |
39279cc3 CM |
2734 | if (!PageUptodate(page)) { |
2735 | ret = -EIO; | |
89642229 | 2736 | goto out_unlock; |
39279cc3 CM |
2737 | } |
2738 | } | |
211c17f5 | 2739 | wait_on_page_writeback(page); |
e6dcd2dc CM |
2740 | |
2741 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2742 | set_page_extent_mapped(page); | |
2743 | ||
2744 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
2745 | if (ordered) { | |
2746 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2747 | unlock_page(page); | |
2748 | page_cache_release(page); | |
eb84ae03 | 2749 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
2750 | btrfs_put_ordered_extent(ordered); |
2751 | goto again; | |
2752 | } | |
2753 | ||
ea8c2819 | 2754 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc CM |
2755 | ret = 0; |
2756 | if (offset != PAGE_CACHE_SIZE) { | |
2757 | kaddr = kmap(page); | |
2758 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
2759 | flush_dcache_page(page); | |
2760 | kunmap(page); | |
2761 | } | |
247e743c | 2762 | ClearPageChecked(page); |
e6dcd2dc CM |
2763 | set_page_dirty(page); |
2764 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 2765 | |
89642229 | 2766 | out_unlock: |
39279cc3 CM |
2767 | unlock_page(page); |
2768 | page_cache_release(page); | |
2769 | out: | |
2770 | return ret; | |
2771 | } | |
2772 | ||
9036c102 | 2773 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 2774 | { |
9036c102 YZ |
2775 | struct btrfs_trans_handle *trans; |
2776 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2777 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2778 | struct extent_map *em; | |
2779 | u64 mask = root->sectorsize - 1; | |
2780 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
2781 | u64 block_end = (size + mask) & ~mask; | |
2782 | u64 last_byte; | |
2783 | u64 cur_offset; | |
2784 | u64 hole_size; | |
39279cc3 CM |
2785 | int err; |
2786 | ||
9036c102 YZ |
2787 | if (size <= hole_start) |
2788 | return 0; | |
2789 | ||
2790 | err = btrfs_check_free_space(root, 1, 0); | |
39279cc3 CM |
2791 | if (err) |
2792 | return err; | |
2793 | ||
9036c102 | 2794 | btrfs_truncate_page(inode->i_mapping, inode->i_size); |
2bf5a725 | 2795 | |
9036c102 YZ |
2796 | while (1) { |
2797 | struct btrfs_ordered_extent *ordered; | |
2798 | btrfs_wait_ordered_range(inode, hole_start, | |
2799 | block_end - hole_start); | |
2800 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2801 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
2802 | if (!ordered) | |
2803 | break; | |
2804 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2805 | btrfs_put_ordered_extent(ordered); | |
2806 | } | |
39279cc3 | 2807 | |
9036c102 YZ |
2808 | trans = btrfs_start_transaction(root, 1); |
2809 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 2810 | |
9036c102 YZ |
2811 | cur_offset = hole_start; |
2812 | while (1) { | |
2813 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
2814 | block_end - cur_offset, 0); | |
2815 | BUG_ON(IS_ERR(em) || !em); | |
2816 | last_byte = min(extent_map_end(em), block_end); | |
2817 | last_byte = (last_byte + mask) & ~mask; | |
2818 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 2819 | u64 hint_byte = 0; |
9036c102 | 2820 | hole_size = last_byte - cur_offset; |
771ed689 CM |
2821 | err = btrfs_drop_extents(trans, root, inode, |
2822 | cur_offset, | |
2823 | cur_offset + hole_size, | |
2824 | cur_offset, &hint_byte); | |
2825 | if (err) | |
2826 | break; | |
9036c102 YZ |
2827 | err = btrfs_insert_file_extent(trans, root, |
2828 | inode->i_ino, cur_offset, 0, | |
2829 | 0, hole_size, 0, hole_size, | |
2830 | 0, 0, 0); | |
2831 | btrfs_drop_extent_cache(inode, hole_start, | |
2832 | last_byte - 1, 0); | |
2833 | } | |
2834 | free_extent_map(em); | |
2835 | cur_offset = last_byte; | |
2836 | if (err || cur_offset >= block_end) | |
2837 | break; | |
2838 | } | |
1832a6d5 | 2839 | |
9036c102 YZ |
2840 | btrfs_end_transaction(trans, root); |
2841 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2842 | return err; | |
2843 | } | |
39279cc3 | 2844 | |
9036c102 YZ |
2845 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
2846 | { | |
2847 | struct inode *inode = dentry->d_inode; | |
2848 | int err; | |
39279cc3 | 2849 | |
9036c102 YZ |
2850 | err = inode_change_ok(inode, attr); |
2851 | if (err) | |
2852 | return err; | |
2bf5a725 | 2853 | |
9036c102 YZ |
2854 | if (S_ISREG(inode->i_mode) && |
2855 | attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) { | |
2856 | err = btrfs_cont_expand(inode, attr->ia_size); | |
54aa1f4d CM |
2857 | if (err) |
2858 | return err; | |
39279cc3 | 2859 | } |
9036c102 | 2860 | |
39279cc3 | 2861 | err = inode_setattr(inode, attr); |
33268eaf JB |
2862 | |
2863 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
2864 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
2865 | return err; |
2866 | } | |
61295eb8 | 2867 | |
39279cc3 CM |
2868 | void btrfs_delete_inode(struct inode *inode) |
2869 | { | |
2870 | struct btrfs_trans_handle *trans; | |
2871 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 2872 | unsigned long nr; |
39279cc3 CM |
2873 | int ret; |
2874 | ||
2875 | truncate_inode_pages(&inode->i_data, 0); | |
2876 | if (is_bad_inode(inode)) { | |
7b128766 | 2877 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
2878 | goto no_delete; |
2879 | } | |
4a096752 | 2880 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 2881 | |
dbe674a9 | 2882 | btrfs_i_size_write(inode, 0); |
180591bc | 2883 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 2884 | |
39279cc3 | 2885 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 2886 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
2887 | if (ret) { |
2888 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 2889 | goto no_delete_lock; |
7b128766 JB |
2890 | } |
2891 | ||
2892 | btrfs_orphan_del(trans, inode); | |
85e21bac | 2893 | |
d3c2fdcf | 2894 | nr = trans->blocks_used; |
85e21bac | 2895 | clear_inode(inode); |
5f39d397 | 2896 | |
39279cc3 | 2897 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2898 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 2899 | return; |
54aa1f4d CM |
2900 | |
2901 | no_delete_lock: | |
d3c2fdcf | 2902 | nr = trans->blocks_used; |
54aa1f4d | 2903 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2904 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2905 | no_delete: |
2906 | clear_inode(inode); | |
2907 | } | |
2908 | ||
2909 | /* | |
2910 | * this returns the key found in the dir entry in the location pointer. | |
2911 | * If no dir entries were found, location->objectid is 0. | |
2912 | */ | |
2913 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
2914 | struct btrfs_key *location) | |
2915 | { | |
2916 | const char *name = dentry->d_name.name; | |
2917 | int namelen = dentry->d_name.len; | |
2918 | struct btrfs_dir_item *di; | |
2919 | struct btrfs_path *path; | |
2920 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 2921 | int ret = 0; |
39279cc3 CM |
2922 | |
2923 | path = btrfs_alloc_path(); | |
2924 | BUG_ON(!path); | |
3954401f | 2925 | |
39279cc3 CM |
2926 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
2927 | namelen, 0); | |
0d9f7f3e Y |
2928 | if (IS_ERR(di)) |
2929 | ret = PTR_ERR(di); | |
d397712b CM |
2930 | |
2931 | if (!di || IS_ERR(di)) | |
3954401f | 2932 | goto out_err; |
d397712b | 2933 | |
5f39d397 | 2934 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 2935 | out: |
39279cc3 CM |
2936 | btrfs_free_path(path); |
2937 | return ret; | |
3954401f CM |
2938 | out_err: |
2939 | location->objectid = 0; | |
2940 | goto out; | |
39279cc3 CM |
2941 | } |
2942 | ||
2943 | /* | |
2944 | * when we hit a tree root in a directory, the btrfs part of the inode | |
2945 | * needs to be changed to reflect the root directory of the tree root. This | |
2946 | * is kind of like crossing a mount point. | |
2947 | */ | |
2948 | static int fixup_tree_root_location(struct btrfs_root *root, | |
2949 | struct btrfs_key *location, | |
58176a96 JB |
2950 | struct btrfs_root **sub_root, |
2951 | struct dentry *dentry) | |
39279cc3 | 2952 | { |
39279cc3 CM |
2953 | struct btrfs_root_item *ri; |
2954 | ||
2955 | if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY) | |
2956 | return 0; | |
2957 | if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) | |
2958 | return 0; | |
2959 | ||
58176a96 JB |
2960 | *sub_root = btrfs_read_fs_root(root->fs_info, location, |
2961 | dentry->d_name.name, | |
2962 | dentry->d_name.len); | |
39279cc3 CM |
2963 | if (IS_ERR(*sub_root)) |
2964 | return PTR_ERR(*sub_root); | |
2965 | ||
2966 | ri = &(*sub_root)->root_item; | |
2967 | location->objectid = btrfs_root_dirid(ri); | |
39279cc3 CM |
2968 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); |
2969 | location->offset = 0; | |
2970 | ||
39279cc3 CM |
2971 | return 0; |
2972 | } | |
2973 | ||
e02119d5 | 2974 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 2975 | { |
e02119d5 CM |
2976 | struct btrfs_inode *bi = BTRFS_I(inode); |
2977 | ||
2978 | bi->i_acl = NULL; | |
2979 | bi->i_default_acl = NULL; | |
2980 | ||
2981 | bi->generation = 0; | |
c3027eb5 | 2982 | bi->sequence = 0; |
e02119d5 CM |
2983 | bi->last_trans = 0; |
2984 | bi->logged_trans = 0; | |
2985 | bi->delalloc_bytes = 0; | |
2986 | bi->disk_i_size = 0; | |
2987 | bi->flags = 0; | |
2988 | bi->index_cnt = (u64)-1; | |
49eb7e46 | 2989 | bi->log_dirty_trans = 0; |
d1310b2e CM |
2990 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
2991 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 2992 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
2993 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
2994 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 2995 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
ba1da2f4 | 2996 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 2997 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
2998 | mutex_init(&BTRFS_I(inode)->log_mutex); |
2999 | } | |
3000 | ||
3001 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3002 | { | |
3003 | struct btrfs_iget_args *args = p; | |
3004 | inode->i_ino = args->ino; | |
3005 | init_btrfs_i(inode); | |
3006 | BTRFS_I(inode)->root = args->root; | |
39279cc3 CM |
3007 | return 0; |
3008 | } | |
3009 | ||
3010 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3011 | { | |
3012 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3013 | return args->ino == inode->i_ino && |
3014 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3015 | } |
3016 | ||
5b21f2ed ZY |
3017 | struct inode *btrfs_ilookup(struct super_block *s, u64 objectid, |
3018 | struct btrfs_root *root, int wait) | |
3019 | { | |
3020 | struct inode *inode; | |
3021 | struct btrfs_iget_args args; | |
3022 | args.ino = objectid; | |
3023 | args.root = root; | |
3024 | ||
3025 | if (wait) { | |
3026 | inode = ilookup5(s, objectid, btrfs_find_actor, | |
3027 | (void *)&args); | |
3028 | } else { | |
3029 | inode = ilookup5_nowait(s, objectid, btrfs_find_actor, | |
3030 | (void *)&args); | |
3031 | } | |
3032 | return inode; | |
3033 | } | |
3034 | ||
39279cc3 CM |
3035 | struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid, |
3036 | struct btrfs_root *root) | |
3037 | { | |
3038 | struct inode *inode; | |
3039 | struct btrfs_iget_args args; | |
3040 | args.ino = objectid; | |
3041 | args.root = root; | |
3042 | ||
3043 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3044 | btrfs_init_locked_inode, | |
3045 | (void *)&args); | |
3046 | return inode; | |
3047 | } | |
3048 | ||
1a54ef8c BR |
3049 | /* Get an inode object given its location and corresponding root. |
3050 | * Returns in *is_new if the inode was read from disk | |
3051 | */ | |
3052 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
3053 | struct btrfs_root *root, int *is_new) | |
3054 | { | |
3055 | struct inode *inode; | |
3056 | ||
3057 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3058 | if (!inode) | |
3059 | return ERR_PTR(-EACCES); | |
3060 | ||
3061 | if (inode->i_state & I_NEW) { | |
3062 | BTRFS_I(inode)->root = root; | |
3063 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3064 | btrfs_read_locked_inode(inode); | |
3065 | unlock_new_inode(inode); | |
3066 | if (is_new) | |
3067 | *is_new = 1; | |
3068 | } else { | |
3069 | if (is_new) | |
3070 | *is_new = 0; | |
3071 | } | |
3072 | ||
3073 | return inode; | |
3074 | } | |
3075 | ||
3de4586c | 3076 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3077 | { |
d397712b | 3078 | struct inode *inode; |
39279cc3 CM |
3079 | struct btrfs_inode *bi = BTRFS_I(dir); |
3080 | struct btrfs_root *root = bi->root; | |
3081 | struct btrfs_root *sub_root = root; | |
3082 | struct btrfs_key location; | |
c146afad | 3083 | int ret, new; |
39279cc3 CM |
3084 | |
3085 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3086 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3087 | |
39279cc3 | 3088 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3089 | |
39279cc3 CM |
3090 | if (ret < 0) |
3091 | return ERR_PTR(ret); | |
5f39d397 | 3092 | |
39279cc3 CM |
3093 | inode = NULL; |
3094 | if (location.objectid) { | |
58176a96 JB |
3095 | ret = fixup_tree_root_location(root, &location, &sub_root, |
3096 | dentry); | |
39279cc3 CM |
3097 | if (ret < 0) |
3098 | return ERR_PTR(ret); | |
3099 | if (ret > 0) | |
3100 | return ERR_PTR(-ENOENT); | |
1a54ef8c BR |
3101 | inode = btrfs_iget(dir->i_sb, &location, sub_root, &new); |
3102 | if (IS_ERR(inode)) | |
3103 | return ERR_CAST(inode); | |
39279cc3 | 3104 | } |
3de4586c CM |
3105 | return inode; |
3106 | } | |
3107 | ||
3108 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, | |
3109 | struct nameidata *nd) | |
3110 | { | |
3111 | struct inode *inode; | |
3112 | ||
3113 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3114 | return ERR_PTR(-ENAMETOOLONG); | |
3115 | ||
3116 | inode = btrfs_lookup_dentry(dir, dentry); | |
3117 | if (IS_ERR(inode)) | |
3118 | return ERR_CAST(inode); | |
7b128766 | 3119 | |
39279cc3 CM |
3120 | return d_splice_alias(inode, dentry); |
3121 | } | |
3122 | ||
39279cc3 CM |
3123 | static unsigned char btrfs_filetype_table[] = { |
3124 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3125 | }; | |
3126 | ||
cbdf5a24 DW |
3127 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3128 | filldir_t filldir) | |
39279cc3 | 3129 | { |
6da6abae | 3130 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3131 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3132 | struct btrfs_item *item; | |
3133 | struct btrfs_dir_item *di; | |
3134 | struct btrfs_key key; | |
5f39d397 | 3135 | struct btrfs_key found_key; |
39279cc3 CM |
3136 | struct btrfs_path *path; |
3137 | int ret; | |
3138 | u32 nritems; | |
5f39d397 | 3139 | struct extent_buffer *leaf; |
39279cc3 CM |
3140 | int slot; |
3141 | int advance; | |
3142 | unsigned char d_type; | |
3143 | int over = 0; | |
3144 | u32 di_cur; | |
3145 | u32 di_total; | |
3146 | u32 di_len; | |
3147 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3148 | char tmp_name[32]; |
3149 | char *name_ptr; | |
3150 | int name_len; | |
39279cc3 CM |
3151 | |
3152 | /* FIXME, use a real flag for deciding about the key type */ | |
3153 | if (root->fs_info->tree_root == root) | |
3154 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3155 | |
3954401f CM |
3156 | /* special case for "." */ |
3157 | if (filp->f_pos == 0) { | |
3158 | over = filldir(dirent, ".", 1, | |
3159 | 1, inode->i_ino, | |
3160 | DT_DIR); | |
3161 | if (over) | |
3162 | return 0; | |
3163 | filp->f_pos = 1; | |
3164 | } | |
3954401f CM |
3165 | /* special case for .., just use the back ref */ |
3166 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3167 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3168 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3169 | 2, pino, DT_DIR); |
3954401f | 3170 | if (over) |
49593bfa | 3171 | return 0; |
3954401f CM |
3172 | filp->f_pos = 2; |
3173 | } | |
49593bfa DW |
3174 | path = btrfs_alloc_path(); |
3175 | path->reada = 2; | |
3176 | ||
39279cc3 CM |
3177 | btrfs_set_key_type(&key, key_type); |
3178 | key.offset = filp->f_pos; | |
49593bfa | 3179 | key.objectid = inode->i_ino; |
5f39d397 | 3180 | |
39279cc3 CM |
3181 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3182 | if (ret < 0) | |
3183 | goto err; | |
3184 | advance = 0; | |
49593bfa DW |
3185 | |
3186 | while (1) { | |
5f39d397 CM |
3187 | leaf = path->nodes[0]; |
3188 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3189 | slot = path->slots[0]; |
3190 | if (advance || slot >= nritems) { | |
49593bfa | 3191 | if (slot >= nritems - 1) { |
39279cc3 CM |
3192 | ret = btrfs_next_leaf(root, path); |
3193 | if (ret) | |
3194 | break; | |
5f39d397 CM |
3195 | leaf = path->nodes[0]; |
3196 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3197 | slot = path->slots[0]; |
3198 | } else { | |
3199 | slot++; | |
3200 | path->slots[0]++; | |
3201 | } | |
3202 | } | |
3de4586c | 3203 | |
39279cc3 | 3204 | advance = 1; |
5f39d397 CM |
3205 | item = btrfs_item_nr(leaf, slot); |
3206 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3207 | ||
3208 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3209 | break; |
5f39d397 | 3210 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3211 | break; |
5f39d397 | 3212 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3213 | continue; |
5f39d397 CM |
3214 | |
3215 | filp->f_pos = found_key.offset; | |
49593bfa | 3216 | |
39279cc3 CM |
3217 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3218 | di_cur = 0; | |
5f39d397 | 3219 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3220 | |
3221 | while (di_cur < di_total) { | |
5f39d397 CM |
3222 | struct btrfs_key location; |
3223 | ||
3224 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3225 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3226 | name_ptr = tmp_name; |
3227 | } else { | |
3228 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3229 | if (!name_ptr) { |
3230 | ret = -ENOMEM; | |
3231 | goto err; | |
3232 | } | |
5f39d397 CM |
3233 | } |
3234 | read_extent_buffer(leaf, name_ptr, | |
3235 | (unsigned long)(di + 1), name_len); | |
3236 | ||
3237 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3238 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3239 | |
3240 | /* is this a reference to our own snapshot? If so | |
3241 | * skip it | |
3242 | */ | |
3243 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3244 | location.objectid == root->root_key.objectid) { | |
3245 | over = 0; | |
3246 | goto skip; | |
3247 | } | |
5f39d397 | 3248 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3249 | found_key.offset, location.objectid, |
39279cc3 | 3250 | d_type); |
5f39d397 | 3251 | |
3de4586c | 3252 | skip: |
5f39d397 CM |
3253 | if (name_ptr != tmp_name) |
3254 | kfree(name_ptr); | |
3255 | ||
39279cc3 CM |
3256 | if (over) |
3257 | goto nopos; | |
5103e947 | 3258 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3259 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3260 | di_cur += di_len; |
3261 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3262 | } | |
3263 | } | |
49593bfa DW |
3264 | |
3265 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 YZ |
3266 | if (key_type == BTRFS_DIR_INDEX_KEY) |
3267 | filp->f_pos = INT_LIMIT(typeof(filp->f_pos)); | |
3268 | else | |
3269 | filp->f_pos++; | |
39279cc3 CM |
3270 | nopos: |
3271 | ret = 0; | |
3272 | err: | |
39279cc3 | 3273 | btrfs_free_path(path); |
39279cc3 CM |
3274 | return ret; |
3275 | } | |
3276 | ||
3277 | int btrfs_write_inode(struct inode *inode, int wait) | |
3278 | { | |
3279 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3280 | struct btrfs_trans_handle *trans; | |
3281 | int ret = 0; | |
3282 | ||
c146afad | 3283 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3284 | return 0; |
3285 | ||
39279cc3 | 3286 | if (wait) { |
f9295749 | 3287 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3288 | btrfs_set_trans_block_group(trans, inode); |
3289 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3290 | } |
3291 | return ret; | |
3292 | } | |
3293 | ||
3294 | /* | |
54aa1f4d | 3295 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3296 | * inode changes. But, it is most likely to find the inode in cache. |
3297 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3298 | * to keep or drop this code. | |
3299 | */ | |
3300 | void btrfs_dirty_inode(struct inode *inode) | |
3301 | { | |
3302 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3303 | struct btrfs_trans_handle *trans; | |
3304 | ||
f9295749 | 3305 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3306 | btrfs_set_trans_block_group(trans, inode); |
3307 | btrfs_update_inode(trans, root, inode); | |
3308 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3309 | } |
3310 | ||
d352ac68 CM |
3311 | /* |
3312 | * find the highest existing sequence number in a directory | |
3313 | * and then set the in-memory index_cnt variable to reflect | |
3314 | * free sequence numbers | |
3315 | */ | |
aec7477b JB |
3316 | static int btrfs_set_inode_index_count(struct inode *inode) |
3317 | { | |
3318 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3319 | struct btrfs_key key, found_key; | |
3320 | struct btrfs_path *path; | |
3321 | struct extent_buffer *leaf; | |
3322 | int ret; | |
3323 | ||
3324 | key.objectid = inode->i_ino; | |
3325 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3326 | key.offset = (u64)-1; | |
3327 | ||
3328 | path = btrfs_alloc_path(); | |
3329 | if (!path) | |
3330 | return -ENOMEM; | |
3331 | ||
3332 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3333 | if (ret < 0) | |
3334 | goto out; | |
3335 | /* FIXME: we should be able to handle this */ | |
3336 | if (ret == 0) | |
3337 | goto out; | |
3338 | ret = 0; | |
3339 | ||
3340 | /* | |
3341 | * MAGIC NUMBER EXPLANATION: | |
3342 | * since we search a directory based on f_pos we have to start at 2 | |
3343 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3344 | * else has to start at 2 | |
3345 | */ | |
3346 | if (path->slots[0] == 0) { | |
3347 | BTRFS_I(inode)->index_cnt = 2; | |
3348 | goto out; | |
3349 | } | |
3350 | ||
3351 | path->slots[0]--; | |
3352 | ||
3353 | leaf = path->nodes[0]; | |
3354 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3355 | ||
3356 | if (found_key.objectid != inode->i_ino || | |
3357 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3358 | BTRFS_I(inode)->index_cnt = 2; | |
3359 | goto out; | |
3360 | } | |
3361 | ||
3362 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3363 | out: | |
3364 | btrfs_free_path(path); | |
3365 | return ret; | |
3366 | } | |
3367 | ||
d352ac68 CM |
3368 | /* |
3369 | * helper to find a free sequence number in a given directory. This current | |
3370 | * code is very simple, later versions will do smarter things in the btree | |
3371 | */ | |
3de4586c | 3372 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3373 | { |
3374 | int ret = 0; | |
3375 | ||
3376 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3377 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3378 | if (ret) |
aec7477b JB |
3379 | return ret; |
3380 | } | |
3381 | ||
00e4e6b3 | 3382 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3383 | BTRFS_I(dir)->index_cnt++; |
3384 | ||
3385 | return ret; | |
3386 | } | |
3387 | ||
39279cc3 CM |
3388 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3389 | struct btrfs_root *root, | |
aec7477b | 3390 | struct inode *dir, |
9c58309d | 3391 | const char *name, int name_len, |
d2fb3437 YZ |
3392 | u64 ref_objectid, u64 objectid, |
3393 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3394 | { |
3395 | struct inode *inode; | |
5f39d397 | 3396 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3397 | struct btrfs_key *location; |
5f39d397 | 3398 | struct btrfs_path *path; |
9c58309d CM |
3399 | struct btrfs_inode_ref *ref; |
3400 | struct btrfs_key key[2]; | |
3401 | u32 sizes[2]; | |
3402 | unsigned long ptr; | |
39279cc3 CM |
3403 | int ret; |
3404 | int owner; | |
3405 | ||
5f39d397 CM |
3406 | path = btrfs_alloc_path(); |
3407 | BUG_ON(!path); | |
3408 | ||
39279cc3 CM |
3409 | inode = new_inode(root->fs_info->sb); |
3410 | if (!inode) | |
3411 | return ERR_PTR(-ENOMEM); | |
3412 | ||
aec7477b | 3413 | if (dir) { |
3de4586c | 3414 | ret = btrfs_set_inode_index(dir, index); |
aec7477b JB |
3415 | if (ret) |
3416 | return ERR_PTR(ret); | |
aec7477b JB |
3417 | } |
3418 | /* | |
3419 | * index_cnt is ignored for everything but a dir, | |
3420 | * btrfs_get_inode_index_count has an explanation for the magic | |
3421 | * number | |
3422 | */ | |
e02119d5 | 3423 | init_btrfs_i(inode); |
aec7477b | 3424 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3425 | BTRFS_I(inode)->root = root; |
e02119d5 | 3426 | BTRFS_I(inode)->generation = trans->transid; |
b888db2b | 3427 | |
39279cc3 CM |
3428 | if (mode & S_IFDIR) |
3429 | owner = 0; | |
3430 | else | |
3431 | owner = 1; | |
d2fb3437 YZ |
3432 | BTRFS_I(inode)->block_group = |
3433 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
17d217fe YZ |
3434 | if ((mode & S_IFREG)) { |
3435 | if (btrfs_test_opt(root, NODATASUM)) | |
3436 | btrfs_set_flag(inode, NODATASUM); | |
3437 | if (btrfs_test_opt(root, NODATACOW)) | |
3438 | btrfs_set_flag(inode, NODATACOW); | |
3439 | } | |
9c58309d CM |
3440 | |
3441 | key[0].objectid = objectid; | |
3442 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3443 | key[0].offset = 0; | |
3444 | ||
3445 | key[1].objectid = objectid; | |
3446 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3447 | key[1].offset = ref_objectid; | |
3448 | ||
3449 | sizes[0] = sizeof(struct btrfs_inode_item); | |
3450 | sizes[1] = name_len + sizeof(*ref); | |
3451 | ||
3452 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); | |
3453 | if (ret != 0) | |
5f39d397 CM |
3454 | goto fail; |
3455 | ||
9c58309d CM |
3456 | if (objectid > root->highest_inode) |
3457 | root->highest_inode = objectid; | |
3458 | ||
79683f2d CM |
3459 | inode->i_uid = current_fsuid(); |
3460 | inode->i_gid = current_fsgid(); | |
39279cc3 CM |
3461 | inode->i_mode = mode; |
3462 | inode->i_ino = objectid; | |
a76a3cd4 | 3463 | inode_set_bytes(inode, 0); |
39279cc3 | 3464 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
3465 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3466 | struct btrfs_inode_item); | |
e02119d5 | 3467 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
3468 | |
3469 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3470 | struct btrfs_inode_ref); | |
3471 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 3472 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
3473 | ptr = (unsigned long)(ref + 1); |
3474 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
3475 | ||
5f39d397 CM |
3476 | btrfs_mark_buffer_dirty(path->nodes[0]); |
3477 | btrfs_free_path(path); | |
3478 | ||
39279cc3 CM |
3479 | location = &BTRFS_I(inode)->location; |
3480 | location->objectid = objectid; | |
39279cc3 CM |
3481 | location->offset = 0; |
3482 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
3483 | ||
39279cc3 CM |
3484 | insert_inode_hash(inode); |
3485 | return inode; | |
5f39d397 | 3486 | fail: |
aec7477b JB |
3487 | if (dir) |
3488 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 CM |
3489 | btrfs_free_path(path); |
3490 | return ERR_PTR(ret); | |
39279cc3 CM |
3491 | } |
3492 | ||
3493 | static inline u8 btrfs_inode_type(struct inode *inode) | |
3494 | { | |
3495 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
3496 | } | |
3497 | ||
d352ac68 CM |
3498 | /* |
3499 | * utility function to add 'inode' into 'parent_inode' with | |
3500 | * a give name and a given sequence number. | |
3501 | * if 'add_backref' is true, also insert a backref from the | |
3502 | * inode to the parent directory. | |
3503 | */ | |
e02119d5 CM |
3504 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
3505 | struct inode *parent_inode, struct inode *inode, | |
3506 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 CM |
3507 | { |
3508 | int ret; | |
3509 | struct btrfs_key key; | |
e02119d5 | 3510 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 3511 | |
39279cc3 | 3512 | key.objectid = inode->i_ino; |
39279cc3 CM |
3513 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
3514 | key.offset = 0; | |
3515 | ||
e02119d5 CM |
3516 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
3517 | parent_inode->i_ino, | |
aec7477b | 3518 | &key, btrfs_inode_type(inode), |
00e4e6b3 | 3519 | index); |
39279cc3 | 3520 | if (ret == 0) { |
9c58309d CM |
3521 | if (add_backref) { |
3522 | ret = btrfs_insert_inode_ref(trans, root, | |
e02119d5 CM |
3523 | name, name_len, |
3524 | inode->i_ino, | |
3525 | parent_inode->i_ino, | |
3526 | index); | |
9c58309d | 3527 | } |
dbe674a9 | 3528 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 3529 | name_len * 2); |
79c44584 | 3530 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 3531 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
3532 | } |
3533 | return ret; | |
3534 | } | |
3535 | ||
3536 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 3537 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 3538 | int backref, u64 index) |
39279cc3 | 3539 | { |
e02119d5 CM |
3540 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3541 | inode, dentry->d_name.name, | |
3542 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
3543 | if (!err) { |
3544 | d_instantiate(dentry, inode); | |
3545 | return 0; | |
3546 | } | |
3547 | if (err > 0) | |
3548 | err = -EEXIST; | |
3549 | return err; | |
3550 | } | |
3551 | ||
618e21d5 JB |
3552 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
3553 | int mode, dev_t rdev) | |
3554 | { | |
3555 | struct btrfs_trans_handle *trans; | |
3556 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3557 | struct inode *inode = NULL; |
618e21d5 JB |
3558 | int err; |
3559 | int drop_inode = 0; | |
3560 | u64 objectid; | |
1832a6d5 | 3561 | unsigned long nr = 0; |
00e4e6b3 | 3562 | u64 index = 0; |
618e21d5 JB |
3563 | |
3564 | if (!new_valid_dev(rdev)) | |
3565 | return -EINVAL; | |
3566 | ||
1832a6d5 CM |
3567 | err = btrfs_check_free_space(root, 1, 0); |
3568 | if (err) | |
3569 | goto fail; | |
3570 | ||
618e21d5 JB |
3571 | trans = btrfs_start_transaction(root, 1); |
3572 | btrfs_set_trans_block_group(trans, dir); | |
3573 | ||
3574 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3575 | if (err) { | |
3576 | err = -ENOSPC; | |
3577 | goto out_unlock; | |
3578 | } | |
3579 | ||
aec7477b | 3580 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3581 | dentry->d_name.len, |
3582 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 3583 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
3584 | err = PTR_ERR(inode); |
3585 | if (IS_ERR(inode)) | |
3586 | goto out_unlock; | |
3587 | ||
33268eaf JB |
3588 | err = btrfs_init_acl(inode, dir); |
3589 | if (err) { | |
3590 | drop_inode = 1; | |
3591 | goto out_unlock; | |
3592 | } | |
3593 | ||
618e21d5 | 3594 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3595 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
3596 | if (err) |
3597 | drop_inode = 1; | |
3598 | else { | |
3599 | inode->i_op = &btrfs_special_inode_operations; | |
3600 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 3601 | btrfs_update_inode(trans, root, inode); |
618e21d5 JB |
3602 | } |
3603 | dir->i_sb->s_dirt = 1; | |
3604 | btrfs_update_inode_block_group(trans, inode); | |
3605 | btrfs_update_inode_block_group(trans, dir); | |
3606 | out_unlock: | |
d3c2fdcf | 3607 | nr = trans->blocks_used; |
89ce8a63 | 3608 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3609 | fail: |
618e21d5 JB |
3610 | if (drop_inode) { |
3611 | inode_dec_link_count(inode); | |
3612 | iput(inode); | |
3613 | } | |
d3c2fdcf | 3614 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
3615 | return err; |
3616 | } | |
3617 | ||
39279cc3 CM |
3618 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
3619 | int mode, struct nameidata *nd) | |
3620 | { | |
3621 | struct btrfs_trans_handle *trans; | |
3622 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3623 | struct inode *inode = NULL; |
39279cc3 CM |
3624 | int err; |
3625 | int drop_inode = 0; | |
1832a6d5 | 3626 | unsigned long nr = 0; |
39279cc3 | 3627 | u64 objectid; |
00e4e6b3 | 3628 | u64 index = 0; |
39279cc3 | 3629 | |
1832a6d5 CM |
3630 | err = btrfs_check_free_space(root, 1, 0); |
3631 | if (err) | |
3632 | goto fail; | |
39279cc3 CM |
3633 | trans = btrfs_start_transaction(root, 1); |
3634 | btrfs_set_trans_block_group(trans, dir); | |
3635 | ||
3636 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3637 | if (err) { | |
3638 | err = -ENOSPC; | |
3639 | goto out_unlock; | |
3640 | } | |
3641 | ||
aec7477b | 3642 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3643 | dentry->d_name.len, |
3644 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
3645 | objectid, BTRFS_I(dir)->block_group, mode, |
3646 | &index); | |
39279cc3 CM |
3647 | err = PTR_ERR(inode); |
3648 | if (IS_ERR(inode)) | |
3649 | goto out_unlock; | |
3650 | ||
33268eaf JB |
3651 | err = btrfs_init_acl(inode, dir); |
3652 | if (err) { | |
3653 | drop_inode = 1; | |
3654 | goto out_unlock; | |
3655 | } | |
3656 | ||
39279cc3 | 3657 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3658 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
3659 | if (err) |
3660 | drop_inode = 1; | |
3661 | else { | |
3662 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3663 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
3664 | inode->i_fop = &btrfs_file_operations; |
3665 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 3666 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3667 | } |
3668 | dir->i_sb->s_dirt = 1; | |
3669 | btrfs_update_inode_block_group(trans, inode); | |
3670 | btrfs_update_inode_block_group(trans, dir); | |
3671 | out_unlock: | |
d3c2fdcf | 3672 | nr = trans->blocks_used; |
ab78c84d | 3673 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3674 | fail: |
39279cc3 CM |
3675 | if (drop_inode) { |
3676 | inode_dec_link_count(inode); | |
3677 | iput(inode); | |
3678 | } | |
d3c2fdcf | 3679 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3680 | return err; |
3681 | } | |
3682 | ||
3683 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
3684 | struct dentry *dentry) | |
3685 | { | |
3686 | struct btrfs_trans_handle *trans; | |
3687 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3688 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 3689 | u64 index; |
1832a6d5 | 3690 | unsigned long nr = 0; |
39279cc3 CM |
3691 | int err; |
3692 | int drop_inode = 0; | |
3693 | ||
3694 | if (inode->i_nlink == 0) | |
3695 | return -ENOENT; | |
3696 | ||
e02119d5 | 3697 | btrfs_inc_nlink(inode); |
1832a6d5 CM |
3698 | err = btrfs_check_free_space(root, 1, 0); |
3699 | if (err) | |
3700 | goto fail; | |
3de4586c | 3701 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
3702 | if (err) |
3703 | goto fail; | |
3704 | ||
39279cc3 | 3705 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 3706 | |
39279cc3 CM |
3707 | btrfs_set_trans_block_group(trans, dir); |
3708 | atomic_inc(&inode->i_count); | |
aec7477b | 3709 | |
00e4e6b3 | 3710 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 3711 | |
39279cc3 CM |
3712 | if (err) |
3713 | drop_inode = 1; | |
5f39d397 | 3714 | |
39279cc3 CM |
3715 | dir->i_sb->s_dirt = 1; |
3716 | btrfs_update_inode_block_group(trans, dir); | |
54aa1f4d | 3717 | err = btrfs_update_inode(trans, root, inode); |
5f39d397 | 3718 | |
54aa1f4d CM |
3719 | if (err) |
3720 | drop_inode = 1; | |
39279cc3 | 3721 | |
d3c2fdcf | 3722 | nr = trans->blocks_used; |
ab78c84d | 3723 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3724 | fail: |
39279cc3 CM |
3725 | if (drop_inode) { |
3726 | inode_dec_link_count(inode); | |
3727 | iput(inode); | |
3728 | } | |
d3c2fdcf | 3729 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3730 | return err; |
3731 | } | |
3732 | ||
39279cc3 CM |
3733 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
3734 | { | |
b9d86667 | 3735 | struct inode *inode = NULL; |
39279cc3 CM |
3736 | struct btrfs_trans_handle *trans; |
3737 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3738 | int err = 0; | |
3739 | int drop_on_err = 0; | |
b9d86667 | 3740 | u64 objectid = 0; |
00e4e6b3 | 3741 | u64 index = 0; |
d3c2fdcf | 3742 | unsigned long nr = 1; |
39279cc3 | 3743 | |
1832a6d5 CM |
3744 | err = btrfs_check_free_space(root, 1, 0); |
3745 | if (err) | |
3746 | goto out_unlock; | |
3747 | ||
39279cc3 CM |
3748 | trans = btrfs_start_transaction(root, 1); |
3749 | btrfs_set_trans_block_group(trans, dir); | |
5f39d397 | 3750 | |
39279cc3 CM |
3751 | if (IS_ERR(trans)) { |
3752 | err = PTR_ERR(trans); | |
3753 | goto out_unlock; | |
3754 | } | |
3755 | ||
3756 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3757 | if (err) { | |
3758 | err = -ENOSPC; | |
3759 | goto out_unlock; | |
3760 | } | |
3761 | ||
aec7477b | 3762 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3763 | dentry->d_name.len, |
3764 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
3765 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
3766 | &index); | |
39279cc3 CM |
3767 | if (IS_ERR(inode)) { |
3768 | err = PTR_ERR(inode); | |
3769 | goto out_fail; | |
3770 | } | |
5f39d397 | 3771 | |
39279cc3 | 3772 | drop_on_err = 1; |
33268eaf JB |
3773 | |
3774 | err = btrfs_init_acl(inode, dir); | |
3775 | if (err) | |
3776 | goto out_fail; | |
3777 | ||
39279cc3 CM |
3778 | inode->i_op = &btrfs_dir_inode_operations; |
3779 | inode->i_fop = &btrfs_dir_file_operations; | |
3780 | btrfs_set_trans_block_group(trans, inode); | |
3781 | ||
dbe674a9 | 3782 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
3783 | err = btrfs_update_inode(trans, root, inode); |
3784 | if (err) | |
3785 | goto out_fail; | |
5f39d397 | 3786 | |
e02119d5 CM |
3787 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3788 | inode, dentry->d_name.name, | |
3789 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
3790 | if (err) |
3791 | goto out_fail; | |
5f39d397 | 3792 | |
39279cc3 CM |
3793 | d_instantiate(dentry, inode); |
3794 | drop_on_err = 0; | |
3795 | dir->i_sb->s_dirt = 1; | |
3796 | btrfs_update_inode_block_group(trans, inode); | |
3797 | btrfs_update_inode_block_group(trans, dir); | |
3798 | ||
3799 | out_fail: | |
d3c2fdcf | 3800 | nr = trans->blocks_used; |
ab78c84d | 3801 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 3802 | |
39279cc3 | 3803 | out_unlock: |
39279cc3 CM |
3804 | if (drop_on_err) |
3805 | iput(inode); | |
d3c2fdcf | 3806 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3807 | return err; |
3808 | } | |
3809 | ||
d352ac68 CM |
3810 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
3811 | * and an extent that you want to insert, deal with overlap and insert | |
3812 | * the new extent into the tree. | |
3813 | */ | |
3b951516 CM |
3814 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
3815 | struct extent_map *existing, | |
e6dcd2dc CM |
3816 | struct extent_map *em, |
3817 | u64 map_start, u64 map_len) | |
3b951516 CM |
3818 | { |
3819 | u64 start_diff; | |
3b951516 | 3820 | |
e6dcd2dc CM |
3821 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
3822 | start_diff = map_start - em->start; | |
3823 | em->start = map_start; | |
3824 | em->len = map_len; | |
c8b97818 CM |
3825 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
3826 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 3827 | em->block_start += start_diff; |
c8b97818 CM |
3828 | em->block_len -= start_diff; |
3829 | } | |
e6dcd2dc | 3830 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
3831 | } |
3832 | ||
c8b97818 CM |
3833 | static noinline int uncompress_inline(struct btrfs_path *path, |
3834 | struct inode *inode, struct page *page, | |
3835 | size_t pg_offset, u64 extent_offset, | |
3836 | struct btrfs_file_extent_item *item) | |
3837 | { | |
3838 | int ret; | |
3839 | struct extent_buffer *leaf = path->nodes[0]; | |
3840 | char *tmp; | |
3841 | size_t max_size; | |
3842 | unsigned long inline_size; | |
3843 | unsigned long ptr; | |
3844 | ||
3845 | WARN_ON(pg_offset != 0); | |
3846 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
3847 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
3848 | btrfs_item_nr(leaf, path->slots[0])); | |
3849 | tmp = kmalloc(inline_size, GFP_NOFS); | |
3850 | ptr = btrfs_file_extent_inline_start(item); | |
3851 | ||
3852 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
3853 | ||
5b050f04 | 3854 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
3855 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
3856 | inline_size, max_size); | |
3857 | if (ret) { | |
3858 | char *kaddr = kmap_atomic(page, KM_USER0); | |
3859 | unsigned long copy_size = min_t(u64, | |
3860 | PAGE_CACHE_SIZE - pg_offset, | |
3861 | max_size - extent_offset); | |
3862 | memset(kaddr + pg_offset, 0, copy_size); | |
3863 | kunmap_atomic(kaddr, KM_USER0); | |
3864 | } | |
3865 | kfree(tmp); | |
3866 | return 0; | |
3867 | } | |
3868 | ||
d352ac68 CM |
3869 | /* |
3870 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
3871 | * the ugly parts come from merging extents from the disk with the in-ram |
3872 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
3873 | * where the in-ram extents might be locked pending data=ordered completion. |
3874 | * | |
3875 | * This also copies inline extents directly into the page. | |
3876 | */ | |
d397712b | 3877 | |
a52d9a80 | 3878 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 3879 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
3880 | int create) |
3881 | { | |
3882 | int ret; | |
3883 | int err = 0; | |
db94535d | 3884 | u64 bytenr; |
a52d9a80 CM |
3885 | u64 extent_start = 0; |
3886 | u64 extent_end = 0; | |
3887 | u64 objectid = inode->i_ino; | |
3888 | u32 found_type; | |
f421950f | 3889 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
3890 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3891 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
3892 | struct extent_buffer *leaf; |
3893 | struct btrfs_key found_key; | |
a52d9a80 CM |
3894 | struct extent_map *em = NULL; |
3895 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 3896 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 3897 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 3898 | int compressed; |
a52d9a80 | 3899 | |
a52d9a80 | 3900 | again: |
d1310b2e CM |
3901 | spin_lock(&em_tree->lock); |
3902 | em = lookup_extent_mapping(em_tree, start, len); | |
a061fc8d CM |
3903 | if (em) |
3904 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
d1310b2e CM |
3905 | spin_unlock(&em_tree->lock); |
3906 | ||
a52d9a80 | 3907 | if (em) { |
e1c4b745 CM |
3908 | if (em->start > start || em->start + em->len <= start) |
3909 | free_extent_map(em); | |
3910 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
3911 | free_extent_map(em); |
3912 | else | |
3913 | goto out; | |
a52d9a80 | 3914 | } |
d1310b2e | 3915 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 3916 | if (!em) { |
d1310b2e CM |
3917 | err = -ENOMEM; |
3918 | goto out; | |
a52d9a80 | 3919 | } |
e6dcd2dc | 3920 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 3921 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 3922 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 3923 | em->len = (u64)-1; |
c8b97818 | 3924 | em->block_len = (u64)-1; |
f421950f CM |
3925 | |
3926 | if (!path) { | |
3927 | path = btrfs_alloc_path(); | |
3928 | BUG_ON(!path); | |
3929 | } | |
3930 | ||
179e29e4 CM |
3931 | ret = btrfs_lookup_file_extent(trans, root, path, |
3932 | objectid, start, trans != NULL); | |
a52d9a80 CM |
3933 | if (ret < 0) { |
3934 | err = ret; | |
3935 | goto out; | |
3936 | } | |
3937 | ||
3938 | if (ret != 0) { | |
3939 | if (path->slots[0] == 0) | |
3940 | goto not_found; | |
3941 | path->slots[0]--; | |
3942 | } | |
3943 | ||
5f39d397 CM |
3944 | leaf = path->nodes[0]; |
3945 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 3946 | struct btrfs_file_extent_item); |
a52d9a80 | 3947 | /* are we inside the extent that was found? */ |
5f39d397 CM |
3948 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3949 | found_type = btrfs_key_type(&found_key); | |
3950 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
3951 | found_type != BTRFS_EXTENT_DATA_KEY) { |
3952 | goto not_found; | |
3953 | } | |
3954 | ||
5f39d397 CM |
3955 | found_type = btrfs_file_extent_type(leaf, item); |
3956 | extent_start = found_key.offset; | |
c8b97818 | 3957 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
3958 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3959 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 3960 | extent_end = extent_start + |
db94535d | 3961 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
3962 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
3963 | size_t size; | |
3964 | size = btrfs_file_extent_inline_len(leaf, item); | |
3965 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
3966 | ~((u64)root->sectorsize - 1); | |
3967 | } | |
3968 | ||
3969 | if (start >= extent_end) { | |
3970 | path->slots[0]++; | |
3971 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3972 | ret = btrfs_next_leaf(root, path); | |
3973 | if (ret < 0) { | |
3974 | err = ret; | |
3975 | goto out; | |
a52d9a80 | 3976 | } |
9036c102 YZ |
3977 | if (ret > 0) |
3978 | goto not_found; | |
3979 | leaf = path->nodes[0]; | |
a52d9a80 | 3980 | } |
9036c102 YZ |
3981 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3982 | if (found_key.objectid != objectid || | |
3983 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
3984 | goto not_found; | |
3985 | if (start + len <= found_key.offset) | |
3986 | goto not_found; | |
3987 | em->start = start; | |
3988 | em->len = found_key.offset - start; | |
3989 | goto not_found_em; | |
3990 | } | |
3991 | ||
d899e052 YZ |
3992 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3993 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
3994 | em->start = extent_start; |
3995 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
3996 | em->orig_start = extent_start - |
3997 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
3998 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
3999 | if (bytenr == 0) { | |
5f39d397 | 4000 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4001 | goto insert; |
4002 | } | |
c8b97818 CM |
4003 | if (compressed) { |
4004 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4005 | em->block_start = bytenr; | |
4006 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4007 | item); | |
4008 | } else { | |
4009 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4010 | em->block_start = bytenr; | |
4011 | em->block_len = em->len; | |
d899e052 YZ |
4012 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4013 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4014 | } |
a52d9a80 CM |
4015 | goto insert; |
4016 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4017 | unsigned long ptr; |
a52d9a80 | 4018 | char *map; |
3326d1b0 CM |
4019 | size_t size; |
4020 | size_t extent_offset; | |
4021 | size_t copy_size; | |
a52d9a80 | 4022 | |
689f9346 | 4023 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4024 | if (!page || create) { |
689f9346 | 4025 | em->start = extent_start; |
9036c102 | 4026 | em->len = extent_end - extent_start; |
689f9346 Y |
4027 | goto out; |
4028 | } | |
5f39d397 | 4029 | |
9036c102 YZ |
4030 | size = btrfs_file_extent_inline_len(leaf, item); |
4031 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4032 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4033 | size - extent_offset); |
3326d1b0 | 4034 | em->start = extent_start + extent_offset; |
70dec807 CM |
4035 | em->len = (copy_size + root->sectorsize - 1) & |
4036 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4037 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4038 | if (compressed) |
4039 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4040 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4041 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4042 | if (btrfs_file_extent_compression(leaf, item) == |
4043 | BTRFS_COMPRESS_ZLIB) { | |
4044 | ret = uncompress_inline(path, inode, page, | |
4045 | pg_offset, | |
4046 | extent_offset, item); | |
4047 | BUG_ON(ret); | |
4048 | } else { | |
4049 | map = kmap(page); | |
4050 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4051 | copy_size); | |
4052 | kunmap(page); | |
4053 | } | |
179e29e4 CM |
4054 | flush_dcache_page(page); |
4055 | } else if (create && PageUptodate(page)) { | |
4056 | if (!trans) { | |
4057 | kunmap(page); | |
4058 | free_extent_map(em); | |
4059 | em = NULL; | |
4060 | btrfs_release_path(root, path); | |
f9295749 | 4061 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4062 | goto again; |
4063 | } | |
c8b97818 | 4064 | map = kmap(page); |
70dec807 | 4065 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4066 | copy_size); |
c8b97818 | 4067 | kunmap(page); |
179e29e4 | 4068 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4069 | } |
d1310b2e CM |
4070 | set_extent_uptodate(io_tree, em->start, |
4071 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4072 | goto insert; |
4073 | } else { | |
d397712b | 4074 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4075 | WARN_ON(1); |
4076 | } | |
4077 | not_found: | |
4078 | em->start = start; | |
d1310b2e | 4079 | em->len = len; |
a52d9a80 | 4080 | not_found_em: |
5f39d397 | 4081 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4082 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4083 | insert: |
4084 | btrfs_release_path(root, path); | |
d1310b2e | 4085 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4086 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4087 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4088 | (unsigned long long)em->len, | |
4089 | (unsigned long long)start, | |
4090 | (unsigned long long)len); | |
a52d9a80 CM |
4091 | err = -EIO; |
4092 | goto out; | |
4093 | } | |
d1310b2e CM |
4094 | |
4095 | err = 0; | |
4096 | spin_lock(&em_tree->lock); | |
a52d9a80 | 4097 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4098 | /* it is possible that someone inserted the extent into the tree |
4099 | * while we had the lock dropped. It is also possible that | |
4100 | * an overlapping map exists in the tree | |
4101 | */ | |
a52d9a80 | 4102 | if (ret == -EEXIST) { |
3b951516 | 4103 | struct extent_map *existing; |
e6dcd2dc CM |
4104 | |
4105 | ret = 0; | |
4106 | ||
3b951516 | 4107 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4108 | if (existing && (existing->start > start || |
4109 | existing->start + existing->len <= start)) { | |
4110 | free_extent_map(existing); | |
4111 | existing = NULL; | |
4112 | } | |
3b951516 CM |
4113 | if (!existing) { |
4114 | existing = lookup_extent_mapping(em_tree, em->start, | |
4115 | em->len); | |
4116 | if (existing) { | |
4117 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4118 | em, start, |
4119 | root->sectorsize); | |
3b951516 CM |
4120 | free_extent_map(existing); |
4121 | if (err) { | |
4122 | free_extent_map(em); | |
4123 | em = NULL; | |
4124 | } | |
4125 | } else { | |
4126 | err = -EIO; | |
3b951516 CM |
4127 | free_extent_map(em); |
4128 | em = NULL; | |
4129 | } | |
4130 | } else { | |
4131 | free_extent_map(em); | |
4132 | em = existing; | |
e6dcd2dc | 4133 | err = 0; |
a52d9a80 | 4134 | } |
a52d9a80 | 4135 | } |
d1310b2e | 4136 | spin_unlock(&em_tree->lock); |
a52d9a80 | 4137 | out: |
f421950f CM |
4138 | if (path) |
4139 | btrfs_free_path(path); | |
a52d9a80 CM |
4140 | if (trans) { |
4141 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4142 | if (!err) |
a52d9a80 CM |
4143 | err = ret; |
4144 | } | |
a52d9a80 CM |
4145 | if (err) { |
4146 | free_extent_map(em); | |
4147 | WARN_ON(1); | |
4148 | return ERR_PTR(err); | |
4149 | } | |
4150 | return em; | |
4151 | } | |
4152 | ||
16432985 CM |
4153 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4154 | const struct iovec *iov, loff_t offset, | |
4155 | unsigned long nr_segs) | |
4156 | { | |
e1c4b745 | 4157 | return -EINVAL; |
16432985 CM |
4158 | } |
4159 | ||
d396c6f5 | 4160 | static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock) |
39279cc3 | 4161 | { |
d396c6f5 | 4162 | return extent_bmap(mapping, iblock, btrfs_get_extent); |
39279cc3 CM |
4163 | } |
4164 | ||
a52d9a80 | 4165 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4166 | { |
d1310b2e CM |
4167 | struct extent_io_tree *tree; |
4168 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4169 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4170 | } |
1832a6d5 | 4171 | |
a52d9a80 | 4172 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4173 | { |
d1310b2e | 4174 | struct extent_io_tree *tree; |
b888db2b CM |
4175 | |
4176 | ||
4177 | if (current->flags & PF_MEMALLOC) { | |
4178 | redirty_page_for_writepage(wbc, page); | |
4179 | unlock_page(page); | |
4180 | return 0; | |
4181 | } | |
d1310b2e | 4182 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4183 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4184 | } |
4185 | ||
f421950f CM |
4186 | int btrfs_writepages(struct address_space *mapping, |
4187 | struct writeback_control *wbc) | |
b293f02e | 4188 | { |
d1310b2e | 4189 | struct extent_io_tree *tree; |
771ed689 | 4190 | |
d1310b2e | 4191 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4192 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4193 | } | |
4194 | ||
3ab2fb5a CM |
4195 | static int |
4196 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4197 | struct list_head *pages, unsigned nr_pages) | |
4198 | { | |
d1310b2e CM |
4199 | struct extent_io_tree *tree; |
4200 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4201 | return extent_readpages(tree, mapping, pages, nr_pages, |
4202 | btrfs_get_extent); | |
4203 | } | |
e6dcd2dc | 4204 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4205 | { |
d1310b2e CM |
4206 | struct extent_io_tree *tree; |
4207 | struct extent_map_tree *map; | |
a52d9a80 | 4208 | int ret; |
8c2383c3 | 4209 | |
d1310b2e CM |
4210 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4211 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4212 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4213 | if (ret == 1) { |
4214 | ClearPagePrivate(page); | |
4215 | set_page_private(page, 0); | |
4216 | page_cache_release(page); | |
39279cc3 | 4217 | } |
a52d9a80 | 4218 | return ret; |
39279cc3 CM |
4219 | } |
4220 | ||
e6dcd2dc CM |
4221 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4222 | { | |
98509cfc CM |
4223 | if (PageWriteback(page) || PageDirty(page)) |
4224 | return 0; | |
e6dcd2dc CM |
4225 | return __btrfs_releasepage(page, gfp_flags); |
4226 | } | |
4227 | ||
a52d9a80 | 4228 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4229 | { |
d1310b2e | 4230 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4231 | struct btrfs_ordered_extent *ordered; |
4232 | u64 page_start = page_offset(page); | |
4233 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4234 | |
e6dcd2dc | 4235 | wait_on_page_writeback(page); |
d1310b2e | 4236 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4237 | if (offset) { |
4238 | btrfs_releasepage(page, GFP_NOFS); | |
4239 | return; | |
4240 | } | |
4241 | ||
4242 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4243 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4244 | page_offset(page)); | |
4245 | if (ordered) { | |
eb84ae03 CM |
4246 | /* |
4247 | * IO on this page will never be started, so we need | |
4248 | * to account for any ordered extents now | |
4249 | */ | |
e6dcd2dc CM |
4250 | clear_extent_bit(tree, page_start, page_end, |
4251 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4252 | EXTENT_LOCKED, 1, 0, GFP_NOFS); | |
211f90e6 CM |
4253 | btrfs_finish_ordered_io(page->mapping->host, |
4254 | page_start, page_end); | |
e6dcd2dc CM |
4255 | btrfs_put_ordered_extent(ordered); |
4256 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4257 | } | |
4258 | clear_extent_bit(tree, page_start, page_end, | |
4259 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4260 | EXTENT_ORDERED, | |
4261 | 1, 1, GFP_NOFS); | |
4262 | __btrfs_releasepage(page, GFP_NOFS); | |
4263 | ||
4a096752 | 4264 | ClearPageChecked(page); |
9ad6b7bc | 4265 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4266 | ClearPagePrivate(page); |
4267 | set_page_private(page, 0); | |
4268 | page_cache_release(page); | |
4269 | } | |
39279cc3 CM |
4270 | } |
4271 | ||
9ebefb18 CM |
4272 | /* |
4273 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4274 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4275 | * be careful to check for EOF conditions here. We set the page up correctly | |
4276 | * for a written page which means we get ENOSPC checking when writing into | |
4277 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4278 | * support these features. | |
4279 | * | |
4280 | * We are not allowed to take the i_mutex here so we have to play games to | |
4281 | * protect against truncate races as the page could now be beyond EOF. Because | |
4282 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4283 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4284 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4285 | * unlock the page. | |
4286 | */ | |
4287 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) | |
4288 | { | |
6da6abae | 4289 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4290 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4291 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4292 | struct btrfs_ordered_extent *ordered; | |
4293 | char *kaddr; | |
4294 | unsigned long zero_start; | |
9ebefb18 | 4295 | loff_t size; |
1832a6d5 | 4296 | int ret; |
a52d9a80 | 4297 | u64 page_start; |
e6dcd2dc | 4298 | u64 page_end; |
9ebefb18 | 4299 | |
1832a6d5 | 4300 | ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0); |
1832a6d5 CM |
4301 | if (ret) |
4302 | goto out; | |
4303 | ||
4304 | ret = -EINVAL; | |
e6dcd2dc | 4305 | again: |
9ebefb18 | 4306 | lock_page(page); |
9ebefb18 | 4307 | size = i_size_read(inode); |
e6dcd2dc CM |
4308 | page_start = page_offset(page); |
4309 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4310 | |
9ebefb18 | 4311 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4312 | (page_start >= size)) { |
9ebefb18 CM |
4313 | /* page got truncated out from underneath us */ |
4314 | goto out_unlock; | |
4315 | } | |
e6dcd2dc CM |
4316 | wait_on_page_writeback(page); |
4317 | ||
4318 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4319 | set_page_extent_mapped(page); | |
4320 | ||
eb84ae03 CM |
4321 | /* |
4322 | * we can't set the delalloc bits if there are pending ordered | |
4323 | * extents. Drop our locks and wait for them to finish | |
4324 | */ | |
e6dcd2dc CM |
4325 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4326 | if (ordered) { | |
4327 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4328 | unlock_page(page); | |
eb84ae03 | 4329 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4330 | btrfs_put_ordered_extent(ordered); |
4331 | goto again; | |
4332 | } | |
4333 | ||
ea8c2819 | 4334 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc | 4335 | ret = 0; |
9ebefb18 CM |
4336 | |
4337 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4338 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4339 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4340 | else |
e6dcd2dc | 4341 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4342 | |
e6dcd2dc CM |
4343 | if (zero_start != PAGE_CACHE_SIZE) { |
4344 | kaddr = kmap(page); | |
4345 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4346 | flush_dcache_page(page); | |
4347 | kunmap(page); | |
4348 | } | |
247e743c | 4349 | ClearPageChecked(page); |
e6dcd2dc CM |
4350 | set_page_dirty(page); |
4351 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
9ebefb18 CM |
4352 | |
4353 | out_unlock: | |
4354 | unlock_page(page); | |
1832a6d5 | 4355 | out: |
9ebefb18 CM |
4356 | return ret; |
4357 | } | |
4358 | ||
39279cc3 CM |
4359 | static void btrfs_truncate(struct inode *inode) |
4360 | { | |
4361 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4362 | int ret; | |
4363 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 4364 | unsigned long nr; |
dbe674a9 | 4365 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
4366 | |
4367 | if (!S_ISREG(inode->i_mode)) | |
4368 | return; | |
4369 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4370 | return; | |
4371 | ||
4372 | btrfs_truncate_page(inode->i_mapping, inode->i_size); | |
4a096752 | 4373 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 4374 | |
39279cc3 CM |
4375 | trans = btrfs_start_transaction(root, 1); |
4376 | btrfs_set_trans_block_group(trans, inode); | |
dbe674a9 | 4377 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 4378 | |
7b128766 JB |
4379 | ret = btrfs_orphan_add(trans, inode); |
4380 | if (ret) | |
4381 | goto out; | |
39279cc3 | 4382 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 4383 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 4384 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 4385 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 4386 | |
7b128766 JB |
4387 | ret = btrfs_orphan_del(trans, inode); |
4388 | BUG_ON(ret); | |
4389 | ||
4390 | out: | |
4391 | nr = trans->blocks_used; | |
89ce8a63 | 4392 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 4393 | BUG_ON(ret); |
d3c2fdcf | 4394 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4395 | } |
4396 | ||
d352ac68 CM |
4397 | /* |
4398 | * create a new subvolume directory/inode (helper for the ioctl). | |
4399 | */ | |
d2fb3437 YZ |
4400 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
4401 | struct btrfs_root *new_root, struct dentry *dentry, | |
4402 | u64 new_dirid, u64 alloc_hint) | |
39279cc3 | 4403 | { |
39279cc3 | 4404 | struct inode *inode; |
cb8e7090 | 4405 | int error; |
00e4e6b3 | 4406 | u64 index = 0; |
39279cc3 | 4407 | |
aec7477b | 4408 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 4409 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 4410 | if (IS_ERR(inode)) |
f46b5a66 | 4411 | return PTR_ERR(inode); |
39279cc3 CM |
4412 | inode->i_op = &btrfs_dir_inode_operations; |
4413 | inode->i_fop = &btrfs_dir_file_operations; | |
4414 | ||
39279cc3 | 4415 | inode->i_nlink = 1; |
dbe674a9 | 4416 | btrfs_i_size_write(inode, 0); |
3b96362c | 4417 | |
cb8e7090 CH |
4418 | error = btrfs_update_inode(trans, new_root, inode); |
4419 | if (error) | |
4420 | return error; | |
4421 | ||
4422 | d_instantiate(dentry, inode); | |
4423 | return 0; | |
39279cc3 CM |
4424 | } |
4425 | ||
d352ac68 CM |
4426 | /* helper function for file defrag and space balancing. This |
4427 | * forces readahead on a given range of bytes in an inode | |
4428 | */ | |
edbd8d4e | 4429 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
4430 | struct file_ra_state *ra, struct file *file, |
4431 | pgoff_t offset, pgoff_t last_index) | |
4432 | { | |
8e7bf94f | 4433 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 4434 | |
86479a04 CM |
4435 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
4436 | return offset + req_size; | |
86479a04 CM |
4437 | } |
4438 | ||
39279cc3 CM |
4439 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
4440 | { | |
4441 | struct btrfs_inode *ei; | |
4442 | ||
4443 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
4444 | if (!ei) | |
4445 | return NULL; | |
15ee9bc7 | 4446 | ei->last_trans = 0; |
e02119d5 | 4447 | ei->logged_trans = 0; |
e6dcd2dc | 4448 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
33268eaf JB |
4449 | ei->i_acl = BTRFS_ACL_NOT_CACHED; |
4450 | ei->i_default_acl = BTRFS_ACL_NOT_CACHED; | |
7b128766 | 4451 | INIT_LIST_HEAD(&ei->i_orphan); |
39279cc3 CM |
4452 | return &ei->vfs_inode; |
4453 | } | |
4454 | ||
4455 | void btrfs_destroy_inode(struct inode *inode) | |
4456 | { | |
e6dcd2dc | 4457 | struct btrfs_ordered_extent *ordered; |
39279cc3 CM |
4458 | WARN_ON(!list_empty(&inode->i_dentry)); |
4459 | WARN_ON(inode->i_data.nrpages); | |
4460 | ||
33268eaf JB |
4461 | if (BTRFS_I(inode)->i_acl && |
4462 | BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED) | |
4463 | posix_acl_release(BTRFS_I(inode)->i_acl); | |
4464 | if (BTRFS_I(inode)->i_default_acl && | |
4465 | BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED) | |
4466 | posix_acl_release(BTRFS_I(inode)->i_default_acl); | |
4467 | ||
bcc63abb | 4468 | spin_lock(&BTRFS_I(inode)->root->list_lock); |
7b128766 JB |
4469 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
4470 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
4471 | " list\n", inode->i_ino); | |
4472 | dump_stack(); | |
4473 | } | |
bcc63abb | 4474 | spin_unlock(&BTRFS_I(inode)->root->list_lock); |
7b128766 | 4475 | |
d397712b | 4476 | while (1) { |
e6dcd2dc CM |
4477 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
4478 | if (!ordered) | |
4479 | break; | |
4480 | else { | |
d397712b CM |
4481 | printk(KERN_ERR "btrfs found ordered " |
4482 | "extent %llu %llu on inode cleanup\n", | |
4483 | (unsigned long long)ordered->file_offset, | |
4484 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
4485 | btrfs_remove_ordered_extent(inode, ordered); |
4486 | btrfs_put_ordered_extent(ordered); | |
4487 | btrfs_put_ordered_extent(ordered); | |
4488 | } | |
4489 | } | |
5b21f2ed | 4490 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
4491 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
4492 | } | |
4493 | ||
0ee0fda0 | 4494 | static void init_once(void *foo) |
39279cc3 CM |
4495 | { |
4496 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
4497 | ||
4498 | inode_init_once(&ei->vfs_inode); | |
4499 | } | |
4500 | ||
4501 | void btrfs_destroy_cachep(void) | |
4502 | { | |
4503 | if (btrfs_inode_cachep) | |
4504 | kmem_cache_destroy(btrfs_inode_cachep); | |
4505 | if (btrfs_trans_handle_cachep) | |
4506 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
4507 | if (btrfs_transaction_cachep) | |
4508 | kmem_cache_destroy(btrfs_transaction_cachep); | |
4509 | if (btrfs_bit_radix_cachep) | |
4510 | kmem_cache_destroy(btrfs_bit_radix_cachep); | |
4511 | if (btrfs_path_cachep) | |
4512 | kmem_cache_destroy(btrfs_path_cachep); | |
4513 | } | |
4514 | ||
86479a04 | 4515 | struct kmem_cache *btrfs_cache_create(const char *name, size_t size, |
92fee66d | 4516 | unsigned long extra_flags, |
2b1f55b0 | 4517 | void (*ctor)(void *)) |
92fee66d CM |
4518 | { |
4519 | return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT | | |
2b1f55b0 | 4520 | SLAB_MEM_SPREAD | extra_flags), ctor); |
92fee66d CM |
4521 | } |
4522 | ||
39279cc3 CM |
4523 | int btrfs_init_cachep(void) |
4524 | { | |
86479a04 | 4525 | btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache", |
92fee66d CM |
4526 | sizeof(struct btrfs_inode), |
4527 | 0, init_once); | |
39279cc3 CM |
4528 | if (!btrfs_inode_cachep) |
4529 | goto fail; | |
86479a04 CM |
4530 | btrfs_trans_handle_cachep = |
4531 | btrfs_cache_create("btrfs_trans_handle_cache", | |
4532 | sizeof(struct btrfs_trans_handle), | |
4533 | 0, NULL); | |
39279cc3 CM |
4534 | if (!btrfs_trans_handle_cachep) |
4535 | goto fail; | |
86479a04 | 4536 | btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache", |
39279cc3 | 4537 | sizeof(struct btrfs_transaction), |
92fee66d | 4538 | 0, NULL); |
39279cc3 CM |
4539 | if (!btrfs_transaction_cachep) |
4540 | goto fail; | |
86479a04 | 4541 | btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache", |
23223584 | 4542 | sizeof(struct btrfs_path), |
92fee66d | 4543 | 0, NULL); |
39279cc3 CM |
4544 | if (!btrfs_path_cachep) |
4545 | goto fail; | |
86479a04 | 4546 | btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256, |
92fee66d | 4547 | SLAB_DESTROY_BY_RCU, NULL); |
39279cc3 CM |
4548 | if (!btrfs_bit_radix_cachep) |
4549 | goto fail; | |
4550 | return 0; | |
4551 | fail: | |
4552 | btrfs_destroy_cachep(); | |
4553 | return -ENOMEM; | |
4554 | } | |
4555 | ||
4556 | static int btrfs_getattr(struct vfsmount *mnt, | |
4557 | struct dentry *dentry, struct kstat *stat) | |
4558 | { | |
4559 | struct inode *inode = dentry->d_inode; | |
4560 | generic_fillattr(inode, stat); | |
3394e160 | 4561 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 4562 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
4563 | stat->blocks = (inode_get_bytes(inode) + |
4564 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
4565 | return 0; |
4566 | } | |
4567 | ||
d397712b CM |
4568 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4569 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
4570 | { |
4571 | struct btrfs_trans_handle *trans; | |
4572 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4573 | struct inode *new_inode = new_dentry->d_inode; | |
4574 | struct inode *old_inode = old_dentry->d_inode; | |
4575 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 4576 | u64 index = 0; |
39279cc3 CM |
4577 | int ret; |
4578 | ||
3394e160 CM |
4579 | /* we're not allowed to rename between subvolumes */ |
4580 | if (BTRFS_I(old_inode)->root->root_key.objectid != | |
4581 | BTRFS_I(new_dir)->root->root_key.objectid) | |
4582 | return -EXDEV; | |
4583 | ||
39279cc3 CM |
4584 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
4585 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) { | |
4586 | return -ENOTEMPTY; | |
4587 | } | |
5f39d397 | 4588 | |
0660b5af CM |
4589 | /* to rename a snapshot or subvolume, we need to juggle the |
4590 | * backrefs. This isn't coded yet | |
4591 | */ | |
4592 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
4593 | return -EXDEV; | |
4594 | ||
1832a6d5 CM |
4595 | ret = btrfs_check_free_space(root, 1, 0); |
4596 | if (ret) | |
4597 | goto out_unlock; | |
4598 | ||
39279cc3 | 4599 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4600 | |
39279cc3 | 4601 | btrfs_set_trans_block_group(trans, new_dir); |
39279cc3 | 4602 | |
e02119d5 | 4603 | btrfs_inc_nlink(old_dentry->d_inode); |
39279cc3 CM |
4604 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
4605 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
4606 | old_inode->i_ctime = ctime; | |
5f39d397 | 4607 | |
e02119d5 CM |
4608 | ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, |
4609 | old_dentry->d_name.name, | |
4610 | old_dentry->d_name.len); | |
39279cc3 CM |
4611 | if (ret) |
4612 | goto out_fail; | |
4613 | ||
4614 | if (new_inode) { | |
4615 | new_inode->i_ctime = CURRENT_TIME; | |
e02119d5 CM |
4616 | ret = btrfs_unlink_inode(trans, root, new_dir, |
4617 | new_dentry->d_inode, | |
4618 | new_dentry->d_name.name, | |
4619 | new_dentry->d_name.len); | |
39279cc3 CM |
4620 | if (ret) |
4621 | goto out_fail; | |
7b128766 | 4622 | if (new_inode->i_nlink == 0) { |
e02119d5 | 4623 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
7b128766 JB |
4624 | if (ret) |
4625 | goto out_fail; | |
4626 | } | |
e02119d5 | 4627 | |
39279cc3 | 4628 | } |
3de4586c | 4629 | ret = btrfs_set_inode_index(new_dir, &index); |
aec7477b JB |
4630 | if (ret) |
4631 | goto out_fail; | |
4632 | ||
e02119d5 CM |
4633 | ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode, |
4634 | old_inode, new_dentry->d_name.name, | |
4635 | new_dentry->d_name.len, 1, index); | |
39279cc3 CM |
4636 | if (ret) |
4637 | goto out_fail; | |
4638 | ||
4639 | out_fail: | |
ab78c84d | 4640 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4641 | out_unlock: |
39279cc3 CM |
4642 | return ret; |
4643 | } | |
4644 | ||
d352ac68 CM |
4645 | /* |
4646 | * some fairly slow code that needs optimization. This walks the list | |
4647 | * of all the inodes with pending delalloc and forces them to disk. | |
4648 | */ | |
ea8c2819 CM |
4649 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
4650 | { | |
4651 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
4652 | struct btrfs_inode *binode; | |
5b21f2ed | 4653 | struct inode *inode; |
ea8c2819 | 4654 | |
c146afad YZ |
4655 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
4656 | return -EROFS; | |
4657 | ||
75eff68e | 4658 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 4659 | while (!list_empty(head)) { |
ea8c2819 CM |
4660 | binode = list_entry(head->next, struct btrfs_inode, |
4661 | delalloc_inodes); | |
5b21f2ed ZY |
4662 | inode = igrab(&binode->vfs_inode); |
4663 | if (!inode) | |
4664 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 4665 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 4666 | if (inode) { |
8c8bee1d | 4667 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
4668 | iput(inode); |
4669 | } | |
4670 | cond_resched(); | |
75eff68e | 4671 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 4672 | } |
75eff68e | 4673 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
4674 | |
4675 | /* the filemap_flush will queue IO into the worker threads, but | |
4676 | * we have to make sure the IO is actually started and that | |
4677 | * ordered extents get created before we return | |
4678 | */ | |
4679 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 4680 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 4681 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 4682 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
4683 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
4684 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
4685 | } |
4686 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
4687 | return 0; |
4688 | } | |
4689 | ||
39279cc3 CM |
4690 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
4691 | const char *symname) | |
4692 | { | |
4693 | struct btrfs_trans_handle *trans; | |
4694 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4695 | struct btrfs_path *path; | |
4696 | struct btrfs_key key; | |
1832a6d5 | 4697 | struct inode *inode = NULL; |
39279cc3 CM |
4698 | int err; |
4699 | int drop_inode = 0; | |
4700 | u64 objectid; | |
00e4e6b3 | 4701 | u64 index = 0 ; |
39279cc3 CM |
4702 | int name_len; |
4703 | int datasize; | |
5f39d397 | 4704 | unsigned long ptr; |
39279cc3 | 4705 | struct btrfs_file_extent_item *ei; |
5f39d397 | 4706 | struct extent_buffer *leaf; |
1832a6d5 | 4707 | unsigned long nr = 0; |
39279cc3 CM |
4708 | |
4709 | name_len = strlen(symname) + 1; | |
4710 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
4711 | return -ENAMETOOLONG; | |
1832a6d5 | 4712 | |
1832a6d5 CM |
4713 | err = btrfs_check_free_space(root, 1, 0); |
4714 | if (err) | |
4715 | goto out_fail; | |
4716 | ||
39279cc3 CM |
4717 | trans = btrfs_start_transaction(root, 1); |
4718 | btrfs_set_trans_block_group(trans, dir); | |
4719 | ||
4720 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4721 | if (err) { | |
4722 | err = -ENOSPC; | |
4723 | goto out_unlock; | |
4724 | } | |
4725 | ||
aec7477b | 4726 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4727 | dentry->d_name.len, |
4728 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4729 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
4730 | &index); | |
39279cc3 CM |
4731 | err = PTR_ERR(inode); |
4732 | if (IS_ERR(inode)) | |
4733 | goto out_unlock; | |
4734 | ||
33268eaf JB |
4735 | err = btrfs_init_acl(inode, dir); |
4736 | if (err) { | |
4737 | drop_inode = 1; | |
4738 | goto out_unlock; | |
4739 | } | |
4740 | ||
39279cc3 | 4741 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4742 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4743 | if (err) |
4744 | drop_inode = 1; | |
4745 | else { | |
4746 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4747 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4748 | inode->i_fop = &btrfs_file_operations; |
4749 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4750 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
4751 | } |
4752 | dir->i_sb->s_dirt = 1; | |
4753 | btrfs_update_inode_block_group(trans, inode); | |
4754 | btrfs_update_inode_block_group(trans, dir); | |
4755 | if (drop_inode) | |
4756 | goto out_unlock; | |
4757 | ||
4758 | path = btrfs_alloc_path(); | |
4759 | BUG_ON(!path); | |
4760 | key.objectid = inode->i_ino; | |
4761 | key.offset = 0; | |
39279cc3 CM |
4762 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
4763 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
4764 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
4765 | datasize); | |
54aa1f4d CM |
4766 | if (err) { |
4767 | drop_inode = 1; | |
4768 | goto out_unlock; | |
4769 | } | |
5f39d397 CM |
4770 | leaf = path->nodes[0]; |
4771 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
4772 | struct btrfs_file_extent_item); | |
4773 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
4774 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 4775 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
4776 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
4777 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
4778 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
4779 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
4780 | ||
39279cc3 | 4781 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
4782 | write_extent_buffer(leaf, symname, ptr, name_len); |
4783 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 4784 | btrfs_free_path(path); |
5f39d397 | 4785 | |
39279cc3 CM |
4786 | inode->i_op = &btrfs_symlink_inode_operations; |
4787 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 4788 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 4789 | inode_set_bytes(inode, name_len); |
dbe674a9 | 4790 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
4791 | err = btrfs_update_inode(trans, root, inode); |
4792 | if (err) | |
4793 | drop_inode = 1; | |
39279cc3 CM |
4794 | |
4795 | out_unlock: | |
d3c2fdcf | 4796 | nr = trans->blocks_used; |
ab78c84d | 4797 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4798 | out_fail: |
39279cc3 CM |
4799 | if (drop_inode) { |
4800 | inode_dec_link_count(inode); | |
4801 | iput(inode); | |
4802 | } | |
d3c2fdcf | 4803 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4804 | return err; |
4805 | } | |
16432985 | 4806 | |
d899e052 YZ |
4807 | static int prealloc_file_range(struct inode *inode, u64 start, u64 end, |
4808 | u64 alloc_hint, int mode) | |
4809 | { | |
4810 | struct btrfs_trans_handle *trans; | |
4811 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4812 | struct btrfs_key ins; | |
4813 | u64 alloc_size; | |
4814 | u64 cur_offset = start; | |
4815 | u64 num_bytes = end - start; | |
4816 | int ret = 0; | |
4817 | ||
4818 | trans = btrfs_join_transaction(root, 1); | |
4819 | BUG_ON(!trans); | |
4820 | btrfs_set_trans_block_group(trans, inode); | |
4821 | ||
4822 | while (num_bytes > 0) { | |
4823 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
4824 | ret = btrfs_reserve_extent(trans, root, alloc_size, | |
4825 | root->sectorsize, 0, alloc_hint, | |
4826 | (u64)-1, &ins, 1); | |
4827 | if (ret) { | |
4828 | WARN_ON(1); | |
4829 | goto out; | |
4830 | } | |
4831 | ret = insert_reserved_file_extent(trans, inode, | |
4832 | cur_offset, ins.objectid, | |
4833 | ins.offset, ins.offset, | |
4834 | ins.offset, 0, 0, 0, | |
4835 | BTRFS_FILE_EXTENT_PREALLOC); | |
4836 | BUG_ON(ret); | |
4837 | num_bytes -= ins.offset; | |
4838 | cur_offset += ins.offset; | |
4839 | alloc_hint = ins.objectid + ins.offset; | |
4840 | } | |
4841 | out: | |
4842 | if (cur_offset > start) { | |
4843 | inode->i_ctime = CURRENT_TIME; | |
4844 | btrfs_set_flag(inode, PREALLOC); | |
4845 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4846 | cur_offset > i_size_read(inode)) | |
4847 | btrfs_i_size_write(inode, cur_offset); | |
4848 | ret = btrfs_update_inode(trans, root, inode); | |
4849 | BUG_ON(ret); | |
4850 | } | |
4851 | ||
4852 | btrfs_end_transaction(trans, root); | |
4853 | return ret; | |
4854 | } | |
4855 | ||
4856 | static long btrfs_fallocate(struct inode *inode, int mode, | |
4857 | loff_t offset, loff_t len) | |
4858 | { | |
4859 | u64 cur_offset; | |
4860 | u64 last_byte; | |
4861 | u64 alloc_start; | |
4862 | u64 alloc_end; | |
4863 | u64 alloc_hint = 0; | |
4864 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
4865 | struct extent_map *em; | |
4866 | int ret; | |
4867 | ||
4868 | alloc_start = offset & ~mask; | |
4869 | alloc_end = (offset + len + mask) & ~mask; | |
4870 | ||
4871 | mutex_lock(&inode->i_mutex); | |
4872 | if (alloc_start > inode->i_size) { | |
4873 | ret = btrfs_cont_expand(inode, alloc_start); | |
4874 | if (ret) | |
4875 | goto out; | |
4876 | } | |
4877 | ||
4878 | while (1) { | |
4879 | struct btrfs_ordered_extent *ordered; | |
4880 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, | |
4881 | alloc_end - 1, GFP_NOFS); | |
4882 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
4883 | alloc_end - 1); | |
4884 | if (ordered && | |
4885 | ordered->file_offset + ordered->len > alloc_start && | |
4886 | ordered->file_offset < alloc_end) { | |
4887 | btrfs_put_ordered_extent(ordered); | |
4888 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4889 | alloc_start, alloc_end - 1, GFP_NOFS); | |
4890 | btrfs_wait_ordered_range(inode, alloc_start, | |
4891 | alloc_end - alloc_start); | |
4892 | } else { | |
4893 | if (ordered) | |
4894 | btrfs_put_ordered_extent(ordered); | |
4895 | break; | |
4896 | } | |
4897 | } | |
4898 | ||
4899 | cur_offset = alloc_start; | |
4900 | while (1) { | |
4901 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4902 | alloc_end - cur_offset, 0); | |
4903 | BUG_ON(IS_ERR(em) || !em); | |
4904 | last_byte = min(extent_map_end(em), alloc_end); | |
4905 | last_byte = (last_byte + mask) & ~mask; | |
4906 | if (em->block_start == EXTENT_MAP_HOLE) { | |
4907 | ret = prealloc_file_range(inode, cur_offset, | |
4908 | last_byte, alloc_hint, mode); | |
4909 | if (ret < 0) { | |
4910 | free_extent_map(em); | |
4911 | break; | |
4912 | } | |
4913 | } | |
4914 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
4915 | alloc_hint = em->block_start; | |
4916 | free_extent_map(em); | |
4917 | ||
4918 | cur_offset = last_byte; | |
4919 | if (cur_offset >= alloc_end) { | |
4920 | ret = 0; | |
4921 | break; | |
4922 | } | |
4923 | } | |
4924 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1, | |
4925 | GFP_NOFS); | |
4926 | out: | |
4927 | mutex_unlock(&inode->i_mutex); | |
4928 | return ret; | |
4929 | } | |
4930 | ||
e6dcd2dc CM |
4931 | static int btrfs_set_page_dirty(struct page *page) |
4932 | { | |
e6dcd2dc CM |
4933 | return __set_page_dirty_nobuffers(page); |
4934 | } | |
4935 | ||
0ee0fda0 | 4936 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd Y |
4937 | { |
4938 | if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE)) | |
4939 | return -EACCES; | |
33268eaf | 4940 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 4941 | } |
39279cc3 CM |
4942 | |
4943 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 4944 | .getattr = btrfs_getattr, |
39279cc3 CM |
4945 | .lookup = btrfs_lookup, |
4946 | .create = btrfs_create, | |
4947 | .unlink = btrfs_unlink, | |
4948 | .link = btrfs_link, | |
4949 | .mkdir = btrfs_mkdir, | |
4950 | .rmdir = btrfs_rmdir, | |
4951 | .rename = btrfs_rename, | |
4952 | .symlink = btrfs_symlink, | |
4953 | .setattr = btrfs_setattr, | |
618e21d5 | 4954 | .mknod = btrfs_mknod, |
95819c05 CH |
4955 | .setxattr = btrfs_setxattr, |
4956 | .getxattr = btrfs_getxattr, | |
5103e947 | 4957 | .listxattr = btrfs_listxattr, |
95819c05 | 4958 | .removexattr = btrfs_removexattr, |
fdebe2bd | 4959 | .permission = btrfs_permission, |
39279cc3 | 4960 | }; |
39279cc3 CM |
4961 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
4962 | .lookup = btrfs_lookup, | |
fdebe2bd | 4963 | .permission = btrfs_permission, |
39279cc3 | 4964 | }; |
39279cc3 CM |
4965 | static struct file_operations btrfs_dir_file_operations = { |
4966 | .llseek = generic_file_llseek, | |
4967 | .read = generic_read_dir, | |
cbdf5a24 | 4968 | .readdir = btrfs_real_readdir, |
34287aa3 | 4969 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 4970 | #ifdef CONFIG_COMPAT |
34287aa3 | 4971 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 4972 | #endif |
6bf13c0c | 4973 | .release = btrfs_release_file, |
e02119d5 | 4974 | .fsync = btrfs_sync_file, |
39279cc3 CM |
4975 | }; |
4976 | ||
d1310b2e | 4977 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 4978 | .fill_delalloc = run_delalloc_range, |
065631f6 | 4979 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 4980 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 4981 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 4982 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 4983 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 4984 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
4985 | .set_bit_hook = btrfs_set_bit_hook, |
4986 | .clear_bit_hook = btrfs_clear_bit_hook, | |
07157aac CM |
4987 | }; |
4988 | ||
39279cc3 CM |
4989 | static struct address_space_operations btrfs_aops = { |
4990 | .readpage = btrfs_readpage, | |
4991 | .writepage = btrfs_writepage, | |
b293f02e | 4992 | .writepages = btrfs_writepages, |
3ab2fb5a | 4993 | .readpages = btrfs_readpages, |
39279cc3 | 4994 | .sync_page = block_sync_page, |
39279cc3 | 4995 | .bmap = btrfs_bmap, |
16432985 | 4996 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
4997 | .invalidatepage = btrfs_invalidatepage, |
4998 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 4999 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5000 | }; |
5001 | ||
5002 | static struct address_space_operations btrfs_symlink_aops = { | |
5003 | .readpage = btrfs_readpage, | |
5004 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5005 | .invalidatepage = btrfs_invalidatepage, |
5006 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5007 | }; |
5008 | ||
5009 | static struct inode_operations btrfs_file_inode_operations = { | |
5010 | .truncate = btrfs_truncate, | |
5011 | .getattr = btrfs_getattr, | |
5012 | .setattr = btrfs_setattr, | |
95819c05 CH |
5013 | .setxattr = btrfs_setxattr, |
5014 | .getxattr = btrfs_getxattr, | |
5103e947 | 5015 | .listxattr = btrfs_listxattr, |
95819c05 | 5016 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5017 | .permission = btrfs_permission, |
d899e052 | 5018 | .fallocate = btrfs_fallocate, |
39279cc3 | 5019 | }; |
618e21d5 JB |
5020 | static struct inode_operations btrfs_special_inode_operations = { |
5021 | .getattr = btrfs_getattr, | |
5022 | .setattr = btrfs_setattr, | |
fdebe2bd | 5023 | .permission = btrfs_permission, |
95819c05 CH |
5024 | .setxattr = btrfs_setxattr, |
5025 | .getxattr = btrfs_getxattr, | |
33268eaf | 5026 | .listxattr = btrfs_listxattr, |
95819c05 | 5027 | .removexattr = btrfs_removexattr, |
618e21d5 | 5028 | }; |
39279cc3 CM |
5029 | static struct inode_operations btrfs_symlink_inode_operations = { |
5030 | .readlink = generic_readlink, | |
5031 | .follow_link = page_follow_link_light, | |
5032 | .put_link = page_put_link, | |
fdebe2bd | 5033 | .permission = btrfs_permission, |
39279cc3 | 5034 | }; |