Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
4b4e25f2 | 39 | #include "compat.h" |
39279cc3 CM |
40 | #include "ctree.h" |
41 | #include "disk-io.h" | |
42 | #include "transaction.h" | |
43 | #include "btrfs_inode.h" | |
44 | #include "ioctl.h" | |
45 | #include "print-tree.h" | |
0b86a832 | 46 | #include "volumes.h" |
e6dcd2dc | 47 | #include "ordered-data.h" |
95819c05 | 48 | #include "xattr.h" |
e02119d5 | 49 | #include "tree-log.h" |
c8b97818 | 50 | #include "compression.h" |
b4ce94de | 51 | #include "locking.h" |
39279cc3 CM |
52 | |
53 | struct btrfs_iget_args { | |
54 | u64 ino; | |
55 | struct btrfs_root *root; | |
56 | }; | |
57 | ||
58 | static struct inode_operations btrfs_dir_inode_operations; | |
59 | static struct inode_operations btrfs_symlink_inode_operations; | |
60 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 61 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
62 | static struct inode_operations btrfs_file_inode_operations; |
63 | static struct address_space_operations btrfs_aops; | |
64 | static struct address_space_operations btrfs_symlink_aops; | |
65 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 66 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
67 | |
68 | static struct kmem_cache *btrfs_inode_cachep; | |
69 | struct kmem_cache *btrfs_trans_handle_cachep; | |
70 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 CM |
71 | struct kmem_cache *btrfs_path_cachep; |
72 | ||
73 | #define S_SHIFT 12 | |
74 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
75 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
76 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
77 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
78 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
79 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
80 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
81 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
82 | }; | |
83 | ||
7b128766 | 84 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 85 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
86 | static noinline int cow_file_range(struct inode *inode, |
87 | struct page *locked_page, | |
88 | u64 start, u64 end, int *page_started, | |
89 | unsigned long *nr_written, int unlock); | |
7b128766 | 90 | |
0279b4cd JO |
91 | static int btrfs_init_inode_security(struct inode *inode, struct inode *dir) |
92 | { | |
93 | int err; | |
94 | ||
95 | err = btrfs_init_acl(inode, dir); | |
96 | if (!err) | |
97 | err = btrfs_xattr_security_init(inode, dir); | |
98 | return err; | |
99 | } | |
100 | ||
c8b97818 CM |
101 | /* |
102 | * this does all the hard work for inserting an inline extent into | |
103 | * the btree. The caller should have done a btrfs_drop_extents so that | |
104 | * no overlapping inline items exist in the btree | |
105 | */ | |
d397712b | 106 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
107 | struct btrfs_root *root, struct inode *inode, |
108 | u64 start, size_t size, size_t compressed_size, | |
109 | struct page **compressed_pages) | |
110 | { | |
111 | struct btrfs_key key; | |
112 | struct btrfs_path *path; | |
113 | struct extent_buffer *leaf; | |
114 | struct page *page = NULL; | |
115 | char *kaddr; | |
116 | unsigned long ptr; | |
117 | struct btrfs_file_extent_item *ei; | |
118 | int err = 0; | |
119 | int ret; | |
120 | size_t cur_size = size; | |
121 | size_t datasize; | |
122 | unsigned long offset; | |
123 | int use_compress = 0; | |
124 | ||
125 | if (compressed_size && compressed_pages) { | |
126 | use_compress = 1; | |
127 | cur_size = compressed_size; | |
128 | } | |
129 | ||
d397712b CM |
130 | path = btrfs_alloc_path(); |
131 | if (!path) | |
c8b97818 CM |
132 | return -ENOMEM; |
133 | ||
b9473439 | 134 | path->leave_spinning = 1; |
c8b97818 CM |
135 | btrfs_set_trans_block_group(trans, inode); |
136 | ||
137 | key.objectid = inode->i_ino; | |
138 | key.offset = start; | |
139 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
140 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
141 | ||
142 | inode_add_bytes(inode, size); | |
143 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
144 | datasize); | |
145 | BUG_ON(ret); | |
146 | if (ret) { | |
147 | err = ret; | |
c8b97818 CM |
148 | goto fail; |
149 | } | |
150 | leaf = path->nodes[0]; | |
151 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
152 | struct btrfs_file_extent_item); | |
153 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
154 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
155 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
156 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
157 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
158 | ptr = btrfs_file_extent_inline_start(ei); | |
159 | ||
160 | if (use_compress) { | |
161 | struct page *cpage; | |
162 | int i = 0; | |
d397712b | 163 | while (compressed_size > 0) { |
c8b97818 | 164 | cpage = compressed_pages[i]; |
5b050f04 | 165 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
166 | PAGE_CACHE_SIZE); |
167 | ||
b9473439 | 168 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 169 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 170 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
171 | |
172 | i++; | |
173 | ptr += cur_size; | |
174 | compressed_size -= cur_size; | |
175 | } | |
176 | btrfs_set_file_extent_compression(leaf, ei, | |
177 | BTRFS_COMPRESS_ZLIB); | |
178 | } else { | |
179 | page = find_get_page(inode->i_mapping, | |
180 | start >> PAGE_CACHE_SHIFT); | |
181 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
182 | kaddr = kmap_atomic(page, KM_USER0); | |
183 | offset = start & (PAGE_CACHE_SIZE - 1); | |
184 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
185 | kunmap_atomic(kaddr, KM_USER0); | |
186 | page_cache_release(page); | |
187 | } | |
188 | btrfs_mark_buffer_dirty(leaf); | |
189 | btrfs_free_path(path); | |
190 | ||
191 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
192 | btrfs_update_inode(trans, root, inode); | |
193 | return 0; | |
194 | fail: | |
195 | btrfs_free_path(path); | |
196 | return err; | |
197 | } | |
198 | ||
199 | ||
200 | /* | |
201 | * conditionally insert an inline extent into the file. This | |
202 | * does the checks required to make sure the data is small enough | |
203 | * to fit as an inline extent. | |
204 | */ | |
7f366cfe | 205 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
206 | struct btrfs_root *root, |
207 | struct inode *inode, u64 start, u64 end, | |
208 | size_t compressed_size, | |
209 | struct page **compressed_pages) | |
210 | { | |
211 | u64 isize = i_size_read(inode); | |
212 | u64 actual_end = min(end + 1, isize); | |
213 | u64 inline_len = actual_end - start; | |
214 | u64 aligned_end = (end + root->sectorsize - 1) & | |
215 | ~((u64)root->sectorsize - 1); | |
216 | u64 hint_byte; | |
217 | u64 data_len = inline_len; | |
218 | int ret; | |
219 | ||
220 | if (compressed_size) | |
221 | data_len = compressed_size; | |
222 | ||
223 | if (start > 0 || | |
70b99e69 | 224 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
225 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
226 | (!compressed_size && | |
227 | (actual_end & (root->sectorsize - 1)) == 0) || | |
228 | end + 1 < isize || | |
229 | data_len > root->fs_info->max_inline) { | |
230 | return 1; | |
231 | } | |
232 | ||
c8b97818 | 233 | ret = btrfs_drop_extents(trans, root, inode, start, |
a1ed835e CM |
234 | aligned_end, aligned_end, start, |
235 | &hint_byte, 1); | |
c8b97818 CM |
236 | BUG_ON(ret); |
237 | ||
238 | if (isize > actual_end) | |
239 | inline_len = min_t(u64, isize, actual_end); | |
240 | ret = insert_inline_extent(trans, root, inode, start, | |
241 | inline_len, compressed_size, | |
242 | compressed_pages); | |
243 | BUG_ON(ret); | |
a1ed835e | 244 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
245 | return 0; |
246 | } | |
247 | ||
771ed689 CM |
248 | struct async_extent { |
249 | u64 start; | |
250 | u64 ram_size; | |
251 | u64 compressed_size; | |
252 | struct page **pages; | |
253 | unsigned long nr_pages; | |
254 | struct list_head list; | |
255 | }; | |
256 | ||
257 | struct async_cow { | |
258 | struct inode *inode; | |
259 | struct btrfs_root *root; | |
260 | struct page *locked_page; | |
261 | u64 start; | |
262 | u64 end; | |
263 | struct list_head extents; | |
264 | struct btrfs_work work; | |
265 | }; | |
266 | ||
267 | static noinline int add_async_extent(struct async_cow *cow, | |
268 | u64 start, u64 ram_size, | |
269 | u64 compressed_size, | |
270 | struct page **pages, | |
271 | unsigned long nr_pages) | |
272 | { | |
273 | struct async_extent *async_extent; | |
274 | ||
275 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
276 | async_extent->start = start; | |
277 | async_extent->ram_size = ram_size; | |
278 | async_extent->compressed_size = compressed_size; | |
279 | async_extent->pages = pages; | |
280 | async_extent->nr_pages = nr_pages; | |
281 | list_add_tail(&async_extent->list, &cow->extents); | |
282 | return 0; | |
283 | } | |
284 | ||
d352ac68 | 285 | /* |
771ed689 CM |
286 | * we create compressed extents in two phases. The first |
287 | * phase compresses a range of pages that have already been | |
288 | * locked (both pages and state bits are locked). | |
c8b97818 | 289 | * |
771ed689 CM |
290 | * This is done inside an ordered work queue, and the compression |
291 | * is spread across many cpus. The actual IO submission is step | |
292 | * two, and the ordered work queue takes care of making sure that | |
293 | * happens in the same order things were put onto the queue by | |
294 | * writepages and friends. | |
c8b97818 | 295 | * |
771ed689 CM |
296 | * If this code finds it can't get good compression, it puts an |
297 | * entry onto the work queue to write the uncompressed bytes. This | |
298 | * makes sure that both compressed inodes and uncompressed inodes | |
299 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 300 | */ |
771ed689 CM |
301 | static noinline int compress_file_range(struct inode *inode, |
302 | struct page *locked_page, | |
303 | u64 start, u64 end, | |
304 | struct async_cow *async_cow, | |
305 | int *num_added) | |
b888db2b CM |
306 | { |
307 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
308 | struct btrfs_trans_handle *trans; | |
db94535d | 309 | u64 num_bytes; |
c8b97818 CM |
310 | u64 orig_start; |
311 | u64 disk_num_bytes; | |
db94535d | 312 | u64 blocksize = root->sectorsize; |
c8b97818 | 313 | u64 actual_end; |
42dc7bab | 314 | u64 isize = i_size_read(inode); |
e6dcd2dc | 315 | int ret = 0; |
c8b97818 CM |
316 | struct page **pages = NULL; |
317 | unsigned long nr_pages; | |
318 | unsigned long nr_pages_ret = 0; | |
319 | unsigned long total_compressed = 0; | |
320 | unsigned long total_in = 0; | |
321 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 322 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
323 | int i; |
324 | int will_compress; | |
b888db2b | 325 | |
c8b97818 CM |
326 | orig_start = start; |
327 | ||
42dc7bab | 328 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
329 | again: |
330 | will_compress = 0; | |
331 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
332 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 333 | |
f03d9301 CM |
334 | /* |
335 | * we don't want to send crud past the end of i_size through | |
336 | * compression, that's just a waste of CPU time. So, if the | |
337 | * end of the file is before the start of our current | |
338 | * requested range of bytes, we bail out to the uncompressed | |
339 | * cleanup code that can deal with all of this. | |
340 | * | |
341 | * It isn't really the fastest way to fix things, but this is a | |
342 | * very uncommon corner. | |
343 | */ | |
344 | if (actual_end <= start) | |
345 | goto cleanup_and_bail_uncompressed; | |
346 | ||
c8b97818 CM |
347 | total_compressed = actual_end - start; |
348 | ||
349 | /* we want to make sure that amount of ram required to uncompress | |
350 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
351 | * of a compressed extent to 128k. This is a crucial number |
352 | * because it also controls how easily we can spread reads across | |
353 | * cpus for decompression. | |
354 | * | |
355 | * We also want to make sure the amount of IO required to do | |
356 | * a random read is reasonably small, so we limit the size of | |
357 | * a compressed extent to 128k. | |
c8b97818 CM |
358 | */ |
359 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 360 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 361 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
362 | disk_num_bytes = num_bytes; |
363 | total_in = 0; | |
364 | ret = 0; | |
db94535d | 365 | |
771ed689 CM |
366 | /* |
367 | * we do compression for mount -o compress and when the | |
368 | * inode has not been flagged as nocompress. This flag can | |
369 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 370 | */ |
6cbff00f | 371 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
372 | btrfs_test_opt(root, COMPRESS)) { |
373 | WARN_ON(pages); | |
cfbc246e | 374 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 375 | |
c8b97818 CM |
376 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
377 | total_compressed, pages, | |
378 | nr_pages, &nr_pages_ret, | |
379 | &total_in, | |
380 | &total_compressed, | |
381 | max_compressed); | |
382 | ||
383 | if (!ret) { | |
384 | unsigned long offset = total_compressed & | |
385 | (PAGE_CACHE_SIZE - 1); | |
386 | struct page *page = pages[nr_pages_ret - 1]; | |
387 | char *kaddr; | |
388 | ||
389 | /* zero the tail end of the last page, we might be | |
390 | * sending it down to disk | |
391 | */ | |
392 | if (offset) { | |
393 | kaddr = kmap_atomic(page, KM_USER0); | |
394 | memset(kaddr + offset, 0, | |
395 | PAGE_CACHE_SIZE - offset); | |
396 | kunmap_atomic(kaddr, KM_USER0); | |
397 | } | |
398 | will_compress = 1; | |
399 | } | |
400 | } | |
401 | if (start == 0) { | |
771ed689 CM |
402 | trans = btrfs_join_transaction(root, 1); |
403 | BUG_ON(!trans); | |
404 | btrfs_set_trans_block_group(trans, inode); | |
405 | ||
c8b97818 | 406 | /* lets try to make an inline extent */ |
771ed689 | 407 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 408 | /* we didn't compress the entire range, try |
771ed689 | 409 | * to make an uncompressed inline extent. |
c8b97818 CM |
410 | */ |
411 | ret = cow_file_range_inline(trans, root, inode, | |
412 | start, end, 0, NULL); | |
413 | } else { | |
771ed689 | 414 | /* try making a compressed inline extent */ |
c8b97818 CM |
415 | ret = cow_file_range_inline(trans, root, inode, |
416 | start, end, | |
417 | total_compressed, pages); | |
418 | } | |
771ed689 | 419 | btrfs_end_transaction(trans, root); |
c8b97818 | 420 | if (ret == 0) { |
771ed689 CM |
421 | /* |
422 | * inline extent creation worked, we don't need | |
423 | * to create any more async work items. Unlock | |
424 | * and free up our temp pages. | |
425 | */ | |
c8b97818 CM |
426 | extent_clear_unlock_delalloc(inode, |
427 | &BTRFS_I(inode)->io_tree, | |
771ed689 | 428 | start, end, NULL, 1, 0, |
8b62b72b | 429 | 0, 1, 1, 1, 0); |
c8b97818 CM |
430 | ret = 0; |
431 | goto free_pages_out; | |
432 | } | |
433 | } | |
434 | ||
435 | if (will_compress) { | |
436 | /* | |
437 | * we aren't doing an inline extent round the compressed size | |
438 | * up to a block size boundary so the allocator does sane | |
439 | * things | |
440 | */ | |
441 | total_compressed = (total_compressed + blocksize - 1) & | |
442 | ~(blocksize - 1); | |
443 | ||
444 | /* | |
445 | * one last check to make sure the compression is really a | |
446 | * win, compare the page count read with the blocks on disk | |
447 | */ | |
448 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
449 | ~(PAGE_CACHE_SIZE - 1); | |
450 | if (total_compressed >= total_in) { | |
451 | will_compress = 0; | |
452 | } else { | |
453 | disk_num_bytes = total_compressed; | |
454 | num_bytes = total_in; | |
455 | } | |
456 | } | |
457 | if (!will_compress && pages) { | |
458 | /* | |
459 | * the compression code ran but failed to make things smaller, | |
460 | * free any pages it allocated and our page pointer array | |
461 | */ | |
462 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 463 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
464 | page_cache_release(pages[i]); |
465 | } | |
466 | kfree(pages); | |
467 | pages = NULL; | |
468 | total_compressed = 0; | |
469 | nr_pages_ret = 0; | |
470 | ||
471 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 472 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 473 | } |
771ed689 CM |
474 | if (will_compress) { |
475 | *num_added += 1; | |
c8b97818 | 476 | |
771ed689 CM |
477 | /* the async work queues will take care of doing actual |
478 | * allocation on disk for these compressed pages, | |
479 | * and will submit them to the elevator. | |
480 | */ | |
481 | add_async_extent(async_cow, start, num_bytes, | |
482 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 483 | |
42dc7bab | 484 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
485 | start += num_bytes; |
486 | pages = NULL; | |
487 | cond_resched(); | |
488 | goto again; | |
489 | } | |
490 | } else { | |
f03d9301 | 491 | cleanup_and_bail_uncompressed: |
771ed689 CM |
492 | /* |
493 | * No compression, but we still need to write the pages in | |
494 | * the file we've been given so far. redirty the locked | |
495 | * page if it corresponds to our extent and set things up | |
496 | * for the async work queue to run cow_file_range to do | |
497 | * the normal delalloc dance | |
498 | */ | |
499 | if (page_offset(locked_page) >= start && | |
500 | page_offset(locked_page) <= end) { | |
501 | __set_page_dirty_nobuffers(locked_page); | |
502 | /* unlocked later on in the async handlers */ | |
503 | } | |
504 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
505 | *num_added += 1; | |
506 | } | |
3b951516 | 507 | |
771ed689 CM |
508 | out: |
509 | return 0; | |
510 | ||
511 | free_pages_out: | |
512 | for (i = 0; i < nr_pages_ret; i++) { | |
513 | WARN_ON(pages[i]->mapping); | |
514 | page_cache_release(pages[i]); | |
515 | } | |
d397712b | 516 | kfree(pages); |
771ed689 CM |
517 | |
518 | goto out; | |
519 | } | |
520 | ||
521 | /* | |
522 | * phase two of compressed writeback. This is the ordered portion | |
523 | * of the code, which only gets called in the order the work was | |
524 | * queued. We walk all the async extents created by compress_file_range | |
525 | * and send them down to the disk. | |
526 | */ | |
527 | static noinline int submit_compressed_extents(struct inode *inode, | |
528 | struct async_cow *async_cow) | |
529 | { | |
530 | struct async_extent *async_extent; | |
531 | u64 alloc_hint = 0; | |
532 | struct btrfs_trans_handle *trans; | |
533 | struct btrfs_key ins; | |
534 | struct extent_map *em; | |
535 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
536 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
537 | struct extent_io_tree *io_tree; | |
538 | int ret; | |
539 | ||
540 | if (list_empty(&async_cow->extents)) | |
541 | return 0; | |
542 | ||
543 | trans = btrfs_join_transaction(root, 1); | |
544 | ||
d397712b | 545 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
546 | async_extent = list_entry(async_cow->extents.next, |
547 | struct async_extent, list); | |
548 | list_del(&async_extent->list); | |
c8b97818 | 549 | |
771ed689 CM |
550 | io_tree = &BTRFS_I(inode)->io_tree; |
551 | ||
552 | /* did the compression code fall back to uncompressed IO? */ | |
553 | if (!async_extent->pages) { | |
554 | int page_started = 0; | |
555 | unsigned long nr_written = 0; | |
556 | ||
557 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
558 | async_extent->start + |
559 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
560 | |
561 | /* allocate blocks */ | |
562 | cow_file_range(inode, async_cow->locked_page, | |
563 | async_extent->start, | |
564 | async_extent->start + | |
565 | async_extent->ram_size - 1, | |
566 | &page_started, &nr_written, 0); | |
567 | ||
568 | /* | |
569 | * if page_started, cow_file_range inserted an | |
570 | * inline extent and took care of all the unlocking | |
571 | * and IO for us. Otherwise, we need to submit | |
572 | * all those pages down to the drive. | |
573 | */ | |
574 | if (!page_started) | |
575 | extent_write_locked_range(io_tree, | |
576 | inode, async_extent->start, | |
d397712b | 577 | async_extent->start + |
771ed689 CM |
578 | async_extent->ram_size - 1, |
579 | btrfs_get_extent, | |
580 | WB_SYNC_ALL); | |
581 | kfree(async_extent); | |
582 | cond_resched(); | |
583 | continue; | |
584 | } | |
585 | ||
586 | lock_extent(io_tree, async_extent->start, | |
587 | async_extent->start + async_extent->ram_size - 1, | |
588 | GFP_NOFS); | |
c8b97818 | 589 | /* |
771ed689 CM |
590 | * here we're doing allocation and writeback of the |
591 | * compressed pages | |
c8b97818 | 592 | */ |
771ed689 CM |
593 | btrfs_drop_extent_cache(inode, async_extent->start, |
594 | async_extent->start + | |
595 | async_extent->ram_size - 1, 0); | |
596 | ||
597 | ret = btrfs_reserve_extent(trans, root, | |
598 | async_extent->compressed_size, | |
599 | async_extent->compressed_size, | |
600 | 0, alloc_hint, | |
601 | (u64)-1, &ins, 1); | |
602 | BUG_ON(ret); | |
603 | em = alloc_extent_map(GFP_NOFS); | |
604 | em->start = async_extent->start; | |
605 | em->len = async_extent->ram_size; | |
445a6944 | 606 | em->orig_start = em->start; |
c8b97818 | 607 | |
771ed689 CM |
608 | em->block_start = ins.objectid; |
609 | em->block_len = ins.offset; | |
610 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
611 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
612 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
613 | ||
d397712b | 614 | while (1) { |
890871be | 615 | write_lock(&em_tree->lock); |
771ed689 | 616 | ret = add_extent_mapping(em_tree, em); |
890871be | 617 | write_unlock(&em_tree->lock); |
771ed689 CM |
618 | if (ret != -EEXIST) { |
619 | free_extent_map(em); | |
620 | break; | |
621 | } | |
622 | btrfs_drop_extent_cache(inode, async_extent->start, | |
623 | async_extent->start + | |
624 | async_extent->ram_size - 1, 0); | |
625 | } | |
626 | ||
627 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
628 | ins.objectid, | |
629 | async_extent->ram_size, | |
630 | ins.offset, | |
631 | BTRFS_ORDERED_COMPRESSED); | |
632 | BUG_ON(ret); | |
633 | ||
634 | btrfs_end_transaction(trans, root); | |
635 | ||
636 | /* | |
637 | * clear dirty, set writeback and unlock the pages. | |
638 | */ | |
639 | extent_clear_unlock_delalloc(inode, | |
640 | &BTRFS_I(inode)->io_tree, | |
641 | async_extent->start, | |
642 | async_extent->start + | |
643 | async_extent->ram_size - 1, | |
8b62b72b | 644 | NULL, 1, 1, 0, 1, 1, 0, 0); |
771ed689 CM |
645 | |
646 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
647 | async_extent->start, |
648 | async_extent->ram_size, | |
649 | ins.objectid, | |
650 | ins.offset, async_extent->pages, | |
651 | async_extent->nr_pages); | |
771ed689 CM |
652 | |
653 | BUG_ON(ret); | |
654 | trans = btrfs_join_transaction(root, 1); | |
655 | alloc_hint = ins.objectid + ins.offset; | |
656 | kfree(async_extent); | |
657 | cond_resched(); | |
658 | } | |
659 | ||
660 | btrfs_end_transaction(trans, root); | |
661 | return 0; | |
662 | } | |
663 | ||
664 | /* | |
665 | * when extent_io.c finds a delayed allocation range in the file, | |
666 | * the call backs end up in this code. The basic idea is to | |
667 | * allocate extents on disk for the range, and create ordered data structs | |
668 | * in ram to track those extents. | |
669 | * | |
670 | * locked_page is the page that writepage had locked already. We use | |
671 | * it to make sure we don't do extra locks or unlocks. | |
672 | * | |
673 | * *page_started is set to one if we unlock locked_page and do everything | |
674 | * required to start IO on it. It may be clean and already done with | |
675 | * IO when we return. | |
676 | */ | |
677 | static noinline int cow_file_range(struct inode *inode, | |
678 | struct page *locked_page, | |
679 | u64 start, u64 end, int *page_started, | |
680 | unsigned long *nr_written, | |
681 | int unlock) | |
682 | { | |
683 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
684 | struct btrfs_trans_handle *trans; | |
685 | u64 alloc_hint = 0; | |
686 | u64 num_bytes; | |
687 | unsigned long ram_size; | |
688 | u64 disk_num_bytes; | |
689 | u64 cur_alloc_size; | |
690 | u64 blocksize = root->sectorsize; | |
691 | u64 actual_end; | |
42dc7bab | 692 | u64 isize = i_size_read(inode); |
771ed689 CM |
693 | struct btrfs_key ins; |
694 | struct extent_map *em; | |
695 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
696 | int ret = 0; | |
697 | ||
698 | trans = btrfs_join_transaction(root, 1); | |
699 | BUG_ON(!trans); | |
700 | btrfs_set_trans_block_group(trans, inode); | |
701 | ||
42dc7bab | 702 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
703 | |
704 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
705 | num_bytes = max(blocksize, num_bytes); | |
706 | disk_num_bytes = num_bytes; | |
707 | ret = 0; | |
708 | ||
709 | if (start == 0) { | |
710 | /* lets try to make an inline extent */ | |
711 | ret = cow_file_range_inline(trans, root, inode, | |
712 | start, end, 0, NULL); | |
713 | if (ret == 0) { | |
714 | extent_clear_unlock_delalloc(inode, | |
715 | &BTRFS_I(inode)->io_tree, | |
716 | start, end, NULL, 1, 1, | |
8b62b72b | 717 | 1, 1, 1, 1, 0); |
771ed689 CM |
718 | *nr_written = *nr_written + |
719 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
720 | *page_started = 1; | |
721 | ret = 0; | |
722 | goto out; | |
723 | } | |
724 | } | |
725 | ||
726 | BUG_ON(disk_num_bytes > | |
727 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
728 | ||
b917b7c3 CM |
729 | |
730 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
731 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
732 | start, num_bytes); | |
733 | if (em) { | |
734 | alloc_hint = em->block_start; | |
735 | free_extent_map(em); | |
736 | } | |
737 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
738 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
739 | ||
d397712b | 740 | while (disk_num_bytes > 0) { |
c8b97818 | 741 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 742 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 743 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 744 | (u64)-1, &ins, 1); |
d397712b CM |
745 | BUG_ON(ret); |
746 | ||
e6dcd2dc CM |
747 | em = alloc_extent_map(GFP_NOFS); |
748 | em->start = start; | |
445a6944 | 749 | em->orig_start = em->start; |
771ed689 CM |
750 | ram_size = ins.offset; |
751 | em->len = ins.offset; | |
c8b97818 | 752 | |
e6dcd2dc | 753 | em->block_start = ins.objectid; |
c8b97818 | 754 | em->block_len = ins.offset; |
e6dcd2dc | 755 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 756 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 757 | |
d397712b | 758 | while (1) { |
890871be | 759 | write_lock(&em_tree->lock); |
e6dcd2dc | 760 | ret = add_extent_mapping(em_tree, em); |
890871be | 761 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
762 | if (ret != -EEXIST) { |
763 | free_extent_map(em); | |
764 | break; | |
765 | } | |
766 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 767 | start + ram_size - 1, 0); |
e6dcd2dc CM |
768 | } |
769 | ||
98d20f67 | 770 | cur_alloc_size = ins.offset; |
e6dcd2dc | 771 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 772 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 773 | BUG_ON(ret); |
c8b97818 | 774 | |
17d217fe YZ |
775 | if (root->root_key.objectid == |
776 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
777 | ret = btrfs_reloc_clone_csums(inode, start, | |
778 | cur_alloc_size); | |
779 | BUG_ON(ret); | |
780 | } | |
781 | ||
d397712b | 782 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 783 | break; |
d397712b | 784 | |
c8b97818 CM |
785 | /* we're not doing compressed IO, don't unlock the first |
786 | * page (which the caller expects to stay locked), don't | |
787 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
788 | * |
789 | * Do set the Private2 bit so we know this page was properly | |
790 | * setup for writepage | |
c8b97818 CM |
791 | */ |
792 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
793 | start, start + ram_size - 1, | |
771ed689 | 794 | locked_page, unlock, 1, |
8b62b72b | 795 | 1, 0, 0, 0, 1); |
c8b97818 | 796 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
797 | num_bytes -= cur_alloc_size; |
798 | alloc_hint = ins.objectid + ins.offset; | |
799 | start += cur_alloc_size; | |
b888db2b | 800 | } |
b888db2b | 801 | out: |
771ed689 | 802 | ret = 0; |
b888db2b | 803 | btrfs_end_transaction(trans, root); |
c8b97818 | 804 | |
be20aa9d | 805 | return ret; |
771ed689 | 806 | } |
c8b97818 | 807 | |
771ed689 CM |
808 | /* |
809 | * work queue call back to started compression on a file and pages | |
810 | */ | |
811 | static noinline void async_cow_start(struct btrfs_work *work) | |
812 | { | |
813 | struct async_cow *async_cow; | |
814 | int num_added = 0; | |
815 | async_cow = container_of(work, struct async_cow, work); | |
816 | ||
817 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
818 | async_cow->start, async_cow->end, async_cow, | |
819 | &num_added); | |
820 | if (num_added == 0) | |
821 | async_cow->inode = NULL; | |
822 | } | |
823 | ||
824 | /* | |
825 | * work queue call back to submit previously compressed pages | |
826 | */ | |
827 | static noinline void async_cow_submit(struct btrfs_work *work) | |
828 | { | |
829 | struct async_cow *async_cow; | |
830 | struct btrfs_root *root; | |
831 | unsigned long nr_pages; | |
832 | ||
833 | async_cow = container_of(work, struct async_cow, work); | |
834 | ||
835 | root = async_cow->root; | |
836 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
837 | PAGE_CACHE_SHIFT; | |
838 | ||
839 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
840 | ||
841 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
842 | 5 * 1042 * 1024 && | |
843 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
844 | wake_up(&root->fs_info->async_submit_wait); | |
845 | ||
d397712b | 846 | if (async_cow->inode) |
771ed689 | 847 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 848 | } |
c8b97818 | 849 | |
771ed689 CM |
850 | static noinline void async_cow_free(struct btrfs_work *work) |
851 | { | |
852 | struct async_cow *async_cow; | |
853 | async_cow = container_of(work, struct async_cow, work); | |
854 | kfree(async_cow); | |
855 | } | |
856 | ||
857 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
858 | u64 start, u64 end, int *page_started, | |
859 | unsigned long *nr_written) | |
860 | { | |
861 | struct async_cow *async_cow; | |
862 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
863 | unsigned long nr_pages; | |
864 | u64 cur_end; | |
865 | int limit = 10 * 1024 * 1042; | |
866 | ||
771ed689 | 867 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | |
2c64c53d | 868 | EXTENT_DELALLOC, 1, 0, NULL, GFP_NOFS); |
d397712b | 869 | while (start < end) { |
771ed689 CM |
870 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
871 | async_cow->inode = inode; | |
872 | async_cow->root = root; | |
873 | async_cow->locked_page = locked_page; | |
874 | async_cow->start = start; | |
875 | ||
6cbff00f | 876 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
877 | cur_end = end; |
878 | else | |
879 | cur_end = min(end, start + 512 * 1024 - 1); | |
880 | ||
881 | async_cow->end = cur_end; | |
882 | INIT_LIST_HEAD(&async_cow->extents); | |
883 | ||
884 | async_cow->work.func = async_cow_start; | |
885 | async_cow->work.ordered_func = async_cow_submit; | |
886 | async_cow->work.ordered_free = async_cow_free; | |
887 | async_cow->work.flags = 0; | |
888 | ||
771ed689 CM |
889 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
890 | PAGE_CACHE_SHIFT; | |
891 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
892 | ||
893 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
894 | &async_cow->work); | |
895 | ||
896 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
897 | wait_event(root->fs_info->async_submit_wait, | |
898 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
899 | limit)); | |
900 | } | |
901 | ||
d397712b | 902 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
903 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
904 | wait_event(root->fs_info->async_submit_wait, | |
905 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
906 | 0)); | |
907 | } | |
908 | ||
909 | *nr_written += nr_pages; | |
910 | start = cur_end + 1; | |
911 | } | |
912 | *page_started = 1; | |
913 | return 0; | |
be20aa9d CM |
914 | } |
915 | ||
d397712b | 916 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
917 | u64 bytenr, u64 num_bytes) |
918 | { | |
919 | int ret; | |
920 | struct btrfs_ordered_sum *sums; | |
921 | LIST_HEAD(list); | |
922 | ||
07d400a6 YZ |
923 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
924 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
925 | if (ret == 0 && list_empty(&list)) |
926 | return 0; | |
927 | ||
928 | while (!list_empty(&list)) { | |
929 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
930 | list_del(&sums->list); | |
931 | kfree(sums); | |
932 | } | |
933 | return 1; | |
934 | } | |
935 | ||
d352ac68 CM |
936 | /* |
937 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
938 | * of the extents that exist in the file, and COWs the file as required. | |
939 | * | |
940 | * If no cow copies or snapshots exist, we write directly to the existing | |
941 | * blocks on disk | |
942 | */ | |
7f366cfe CM |
943 | static noinline int run_delalloc_nocow(struct inode *inode, |
944 | struct page *locked_page, | |
771ed689 CM |
945 | u64 start, u64 end, int *page_started, int force, |
946 | unsigned long *nr_written) | |
be20aa9d | 947 | { |
be20aa9d | 948 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 949 | struct btrfs_trans_handle *trans; |
be20aa9d | 950 | struct extent_buffer *leaf; |
be20aa9d | 951 | struct btrfs_path *path; |
80ff3856 | 952 | struct btrfs_file_extent_item *fi; |
be20aa9d | 953 | struct btrfs_key found_key; |
80ff3856 YZ |
954 | u64 cow_start; |
955 | u64 cur_offset; | |
956 | u64 extent_end; | |
5d4f98a2 | 957 | u64 extent_offset; |
80ff3856 YZ |
958 | u64 disk_bytenr; |
959 | u64 num_bytes; | |
960 | int extent_type; | |
961 | int ret; | |
d899e052 | 962 | int type; |
80ff3856 YZ |
963 | int nocow; |
964 | int check_prev = 1; | |
be20aa9d CM |
965 | |
966 | path = btrfs_alloc_path(); | |
967 | BUG_ON(!path); | |
7ea394f1 YZ |
968 | trans = btrfs_join_transaction(root, 1); |
969 | BUG_ON(!trans); | |
be20aa9d | 970 | |
80ff3856 YZ |
971 | cow_start = (u64)-1; |
972 | cur_offset = start; | |
973 | while (1) { | |
974 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
975 | cur_offset, 0); | |
976 | BUG_ON(ret < 0); | |
977 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
978 | leaf = path->nodes[0]; | |
979 | btrfs_item_key_to_cpu(leaf, &found_key, | |
980 | path->slots[0] - 1); | |
981 | if (found_key.objectid == inode->i_ino && | |
982 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
983 | path->slots[0]--; | |
984 | } | |
985 | check_prev = 0; | |
986 | next_slot: | |
987 | leaf = path->nodes[0]; | |
988 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
989 | ret = btrfs_next_leaf(root, path); | |
990 | if (ret < 0) | |
991 | BUG_ON(1); | |
992 | if (ret > 0) | |
993 | break; | |
994 | leaf = path->nodes[0]; | |
995 | } | |
be20aa9d | 996 | |
80ff3856 YZ |
997 | nocow = 0; |
998 | disk_bytenr = 0; | |
17d217fe | 999 | num_bytes = 0; |
80ff3856 YZ |
1000 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1001 | ||
1002 | if (found_key.objectid > inode->i_ino || | |
1003 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1004 | found_key.offset > end) | |
1005 | break; | |
1006 | ||
1007 | if (found_key.offset > cur_offset) { | |
1008 | extent_end = found_key.offset; | |
1009 | goto out_check; | |
1010 | } | |
1011 | ||
1012 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1013 | struct btrfs_file_extent_item); | |
1014 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1015 | ||
d899e052 YZ |
1016 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1017 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1018 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1019 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1020 | extent_end = found_key.offset + |
1021 | btrfs_file_extent_num_bytes(leaf, fi); | |
1022 | if (extent_end <= start) { | |
1023 | path->slots[0]++; | |
1024 | goto next_slot; | |
1025 | } | |
17d217fe YZ |
1026 | if (disk_bytenr == 0) |
1027 | goto out_check; | |
80ff3856 YZ |
1028 | if (btrfs_file_extent_compression(leaf, fi) || |
1029 | btrfs_file_extent_encryption(leaf, fi) || | |
1030 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1031 | goto out_check; | |
d899e052 YZ |
1032 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1033 | goto out_check; | |
d2fb3437 | 1034 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1035 | goto out_check; |
17d217fe | 1036 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1037 | found_key.offset - |
1038 | extent_offset, disk_bytenr)) | |
17d217fe | 1039 | goto out_check; |
5d4f98a2 | 1040 | disk_bytenr += extent_offset; |
17d217fe YZ |
1041 | disk_bytenr += cur_offset - found_key.offset; |
1042 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1043 | /* | |
1044 | * force cow if csum exists in the range. | |
1045 | * this ensure that csum for a given extent are | |
1046 | * either valid or do not exist. | |
1047 | */ | |
1048 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1049 | goto out_check; | |
80ff3856 YZ |
1050 | nocow = 1; |
1051 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1052 | extent_end = found_key.offset + | |
1053 | btrfs_file_extent_inline_len(leaf, fi); | |
1054 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1055 | } else { | |
1056 | BUG_ON(1); | |
1057 | } | |
1058 | out_check: | |
1059 | if (extent_end <= start) { | |
1060 | path->slots[0]++; | |
1061 | goto next_slot; | |
1062 | } | |
1063 | if (!nocow) { | |
1064 | if (cow_start == (u64)-1) | |
1065 | cow_start = cur_offset; | |
1066 | cur_offset = extent_end; | |
1067 | if (cur_offset > end) | |
1068 | break; | |
1069 | path->slots[0]++; | |
1070 | goto next_slot; | |
7ea394f1 YZ |
1071 | } |
1072 | ||
1073 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1074 | if (cow_start != (u64)-1) { |
1075 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1076 | found_key.offset - 1, page_started, |
1077 | nr_written, 1); | |
80ff3856 YZ |
1078 | BUG_ON(ret); |
1079 | cow_start = (u64)-1; | |
7ea394f1 | 1080 | } |
80ff3856 | 1081 | |
d899e052 YZ |
1082 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1083 | struct extent_map *em; | |
1084 | struct extent_map_tree *em_tree; | |
1085 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1086 | em = alloc_extent_map(GFP_NOFS); | |
1087 | em->start = cur_offset; | |
445a6944 | 1088 | em->orig_start = em->start; |
d899e052 YZ |
1089 | em->len = num_bytes; |
1090 | em->block_len = num_bytes; | |
1091 | em->block_start = disk_bytenr; | |
1092 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1093 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1094 | while (1) { | |
890871be | 1095 | write_lock(&em_tree->lock); |
d899e052 | 1096 | ret = add_extent_mapping(em_tree, em); |
890871be | 1097 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1098 | if (ret != -EEXIST) { |
1099 | free_extent_map(em); | |
1100 | break; | |
1101 | } | |
1102 | btrfs_drop_extent_cache(inode, em->start, | |
1103 | em->start + em->len - 1, 0); | |
1104 | } | |
1105 | type = BTRFS_ORDERED_PREALLOC; | |
1106 | } else { | |
1107 | type = BTRFS_ORDERED_NOCOW; | |
1108 | } | |
80ff3856 YZ |
1109 | |
1110 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1111 | num_bytes, num_bytes, type); |
1112 | BUG_ON(ret); | |
771ed689 | 1113 | |
d899e052 YZ |
1114 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
1115 | cur_offset, cur_offset + num_bytes - 1, | |
8b62b72b | 1116 | locked_page, 1, 1, 1, 0, 0, 0, 1); |
80ff3856 YZ |
1117 | cur_offset = extent_end; |
1118 | if (cur_offset > end) | |
1119 | break; | |
be20aa9d | 1120 | } |
80ff3856 YZ |
1121 | btrfs_release_path(root, path); |
1122 | ||
1123 | if (cur_offset <= end && cow_start == (u64)-1) | |
1124 | cow_start = cur_offset; | |
1125 | if (cow_start != (u64)-1) { | |
1126 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1127 | page_started, nr_written, 1); |
80ff3856 YZ |
1128 | BUG_ON(ret); |
1129 | } | |
1130 | ||
1131 | ret = btrfs_end_transaction(trans, root); | |
1132 | BUG_ON(ret); | |
7ea394f1 | 1133 | btrfs_free_path(path); |
80ff3856 | 1134 | return 0; |
be20aa9d CM |
1135 | } |
1136 | ||
d352ac68 CM |
1137 | /* |
1138 | * extent_io.c call back to do delayed allocation processing | |
1139 | */ | |
c8b97818 | 1140 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1141 | u64 start, u64 end, int *page_started, |
1142 | unsigned long *nr_written) | |
be20aa9d | 1143 | { |
be20aa9d | 1144 | int ret; |
7f366cfe | 1145 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1146 | |
6cbff00f | 1147 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1148 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1149 | page_started, 1, nr_written); |
6cbff00f | 1150 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1151 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1152 | page_started, 0, nr_written); |
7f366cfe CM |
1153 | else if (!btrfs_test_opt(root, COMPRESS)) |
1154 | ret = cow_file_range(inode, locked_page, start, end, | |
1155 | page_started, nr_written, 1); | |
be20aa9d | 1156 | else |
771ed689 | 1157 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1158 | page_started, nr_written); |
b888db2b CM |
1159 | return ret; |
1160 | } | |
1161 | ||
d352ac68 CM |
1162 | /* |
1163 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1164 | * bytes in this file, and to maintain the list of inodes that | |
1165 | * have pending delalloc work to be done. | |
1166 | */ | |
b2950863 | 1167 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1168 | unsigned long old, unsigned long bits) |
291d673e | 1169 | { |
75eff68e CM |
1170 | /* |
1171 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1172 | * but in this case, we are only testeing for the DELALLOC | |
1173 | * bit, which is only set or cleared with irqs on | |
1174 | */ | |
b0c68f8b | 1175 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1176 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 1177 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1178 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1179 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1180 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1181 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1182 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1183 | &root->fs_info->delalloc_inodes); | |
1184 | } | |
75eff68e | 1185 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1186 | } |
1187 | return 0; | |
1188 | } | |
1189 | ||
d352ac68 CM |
1190 | /* |
1191 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1192 | */ | |
b2950863 | 1193 | static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1194 | unsigned long old, unsigned long bits) |
291d673e | 1195 | { |
75eff68e CM |
1196 | /* |
1197 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1198 | * but in this case, we are only testeing for the DELALLOC | |
1199 | * bit, which is only set or cleared with irqs on | |
1200 | */ | |
b0c68f8b | 1201 | if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1202 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1203 | |
75eff68e | 1204 | spin_lock(&root->fs_info->delalloc_lock); |
b0c68f8b | 1205 | if (end - start + 1 > root->fs_info->delalloc_bytes) { |
d397712b CM |
1206 | printk(KERN_INFO "btrfs warning: delalloc account " |
1207 | "%llu %llu\n", | |
1208 | (unsigned long long)end - start + 1, | |
1209 | (unsigned long long) | |
1210 | root->fs_info->delalloc_bytes); | |
6a63209f | 1211 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1212 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1213 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1214 | } else { |
6a63209f JB |
1215 | btrfs_delalloc_free_space(root, inode, |
1216 | end - start + 1); | |
b0c68f8b | 1217 | root->fs_info->delalloc_bytes -= end - start + 1; |
9069218d | 1218 | BTRFS_I(inode)->delalloc_bytes -= end - start + 1; |
b0c68f8b | 1219 | } |
ea8c2819 CM |
1220 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1221 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1222 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1223 | } | |
75eff68e | 1224 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1225 | } |
1226 | return 0; | |
1227 | } | |
1228 | ||
d352ac68 CM |
1229 | /* |
1230 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1231 | * we don't create bios that span stripes or chunks | |
1232 | */ | |
239b14b3 | 1233 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1234 | size_t size, struct bio *bio, |
1235 | unsigned long bio_flags) | |
239b14b3 CM |
1236 | { |
1237 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1238 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1239 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1240 | u64 length = 0; |
1241 | u64 map_length; | |
239b14b3 CM |
1242 | int ret; |
1243 | ||
771ed689 CM |
1244 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1245 | return 0; | |
1246 | ||
f2d8d74d | 1247 | length = bio->bi_size; |
239b14b3 CM |
1248 | map_tree = &root->fs_info->mapping_tree; |
1249 | map_length = length; | |
cea9e445 | 1250 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1251 | &map_length, NULL, 0); |
cea9e445 | 1252 | |
d397712b | 1253 | if (map_length < length + size) |
239b14b3 | 1254 | return 1; |
239b14b3 CM |
1255 | return 0; |
1256 | } | |
1257 | ||
d352ac68 CM |
1258 | /* |
1259 | * in order to insert checksums into the metadata in large chunks, | |
1260 | * we wait until bio submission time. All the pages in the bio are | |
1261 | * checksummed and sums are attached onto the ordered extent record. | |
1262 | * | |
1263 | * At IO completion time the cums attached on the ordered extent record | |
1264 | * are inserted into the btree | |
1265 | */ | |
d397712b CM |
1266 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1267 | struct bio *bio, int mirror_num, | |
1268 | unsigned long bio_flags) | |
065631f6 | 1269 | { |
065631f6 | 1270 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1271 | int ret = 0; |
e015640f | 1272 | |
d20f7043 | 1273 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1274 | BUG_ON(ret); |
4a69a410 CM |
1275 | return 0; |
1276 | } | |
e015640f | 1277 | |
4a69a410 CM |
1278 | /* |
1279 | * in order to insert checksums into the metadata in large chunks, | |
1280 | * we wait until bio submission time. All the pages in the bio are | |
1281 | * checksummed and sums are attached onto the ordered extent record. | |
1282 | * | |
1283 | * At IO completion time the cums attached on the ordered extent record | |
1284 | * are inserted into the btree | |
1285 | */ | |
b2950863 | 1286 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1287 | int mirror_num, unsigned long bio_flags) |
1288 | { | |
1289 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1290 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1291 | } |
1292 | ||
d352ac68 | 1293 | /* |
cad321ad CM |
1294 | * extent_io.c submission hook. This does the right thing for csum calculation |
1295 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1296 | */ |
b2950863 | 1297 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1298 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1299 | { |
1300 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1301 | int ret = 0; | |
19b9bdb0 | 1302 | int skip_sum; |
44b8bd7e | 1303 | |
6cbff00f | 1304 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1305 | |
e6dcd2dc CM |
1306 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1307 | BUG_ON(ret); | |
065631f6 | 1308 | |
4d1b5fb4 | 1309 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1310 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1311 | return btrfs_submit_compressed_read(inode, bio, |
1312 | mirror_num, bio_flags); | |
d20f7043 CM |
1313 | } else if (!skip_sum) |
1314 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1315 | goto mapit; |
19b9bdb0 | 1316 | } else if (!skip_sum) { |
17d217fe YZ |
1317 | /* csum items have already been cloned */ |
1318 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1319 | goto mapit; | |
19b9bdb0 CM |
1320 | /* we're doing a write, do the async checksumming */ |
1321 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1322 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1323 | bio_flags, __btrfs_submit_bio_start, |
1324 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1325 | } |
1326 | ||
0b86a832 | 1327 | mapit: |
8b712842 | 1328 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1329 | } |
6885f308 | 1330 | |
d352ac68 CM |
1331 | /* |
1332 | * given a list of ordered sums record them in the inode. This happens | |
1333 | * at IO completion time based on sums calculated at bio submission time. | |
1334 | */ | |
ba1da2f4 | 1335 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1336 | struct inode *inode, u64 file_offset, |
1337 | struct list_head *list) | |
1338 | { | |
e6dcd2dc CM |
1339 | struct btrfs_ordered_sum *sum; |
1340 | ||
1341 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1342 | |
1343 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1344 | btrfs_csum_file_blocks(trans, |
1345 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1346 | } |
1347 | return 0; | |
1348 | } | |
1349 | ||
ea8c2819 CM |
1350 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1351 | { | |
d397712b | 1352 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1353 | WARN_ON(1); |
ea8c2819 CM |
1354 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1355 | GFP_NOFS); | |
1356 | } | |
1357 | ||
d352ac68 | 1358 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1359 | struct btrfs_writepage_fixup { |
1360 | struct page *page; | |
1361 | struct btrfs_work work; | |
1362 | }; | |
1363 | ||
b2950863 | 1364 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1365 | { |
1366 | struct btrfs_writepage_fixup *fixup; | |
1367 | struct btrfs_ordered_extent *ordered; | |
1368 | struct page *page; | |
1369 | struct inode *inode; | |
1370 | u64 page_start; | |
1371 | u64 page_end; | |
1372 | ||
1373 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1374 | page = fixup->page; | |
4a096752 | 1375 | again: |
247e743c CM |
1376 | lock_page(page); |
1377 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1378 | ClearPageChecked(page); | |
1379 | goto out_page; | |
1380 | } | |
1381 | ||
1382 | inode = page->mapping->host; | |
1383 | page_start = page_offset(page); | |
1384 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1385 | ||
1386 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1387 | |
1388 | /* already ordered? We're done */ | |
8b62b72b | 1389 | if (PagePrivate2(page)) |
247e743c | 1390 | goto out; |
4a096752 CM |
1391 | |
1392 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1393 | if (ordered) { | |
1394 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1395 | page_end, GFP_NOFS); | |
1396 | unlock_page(page); | |
1397 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1398 | goto again; | |
1399 | } | |
247e743c | 1400 | |
ea8c2819 | 1401 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1402 | ClearPageChecked(page); |
1403 | out: | |
1404 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1405 | out_page: | |
1406 | unlock_page(page); | |
1407 | page_cache_release(page); | |
1408 | } | |
1409 | ||
1410 | /* | |
1411 | * There are a few paths in the higher layers of the kernel that directly | |
1412 | * set the page dirty bit without asking the filesystem if it is a | |
1413 | * good idea. This causes problems because we want to make sure COW | |
1414 | * properly happens and the data=ordered rules are followed. | |
1415 | * | |
c8b97818 | 1416 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1417 | * hasn't been properly setup for IO. We kick off an async process |
1418 | * to fix it up. The async helper will wait for ordered extents, set | |
1419 | * the delalloc bit and make it safe to write the page. | |
1420 | */ | |
b2950863 | 1421 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1422 | { |
1423 | struct inode *inode = page->mapping->host; | |
1424 | struct btrfs_writepage_fixup *fixup; | |
1425 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1426 | |
8b62b72b CM |
1427 | /* this page is properly in the ordered list */ |
1428 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1429 | return 0; |
1430 | ||
1431 | if (PageChecked(page)) | |
1432 | return -EAGAIN; | |
1433 | ||
1434 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1435 | if (!fixup) | |
1436 | return -EAGAIN; | |
f421950f | 1437 | |
247e743c CM |
1438 | SetPageChecked(page); |
1439 | page_cache_get(page); | |
1440 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1441 | fixup->page = page; | |
1442 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1443 | return -EAGAIN; | |
1444 | } | |
1445 | ||
d899e052 YZ |
1446 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1447 | struct inode *inode, u64 file_pos, | |
1448 | u64 disk_bytenr, u64 disk_num_bytes, | |
1449 | u64 num_bytes, u64 ram_bytes, | |
e980b50c | 1450 | u64 locked_end, |
d899e052 YZ |
1451 | u8 compression, u8 encryption, |
1452 | u16 other_encoding, int extent_type) | |
1453 | { | |
1454 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1455 | struct btrfs_file_extent_item *fi; | |
1456 | struct btrfs_path *path; | |
1457 | struct extent_buffer *leaf; | |
1458 | struct btrfs_key ins; | |
1459 | u64 hint; | |
1460 | int ret; | |
1461 | ||
1462 | path = btrfs_alloc_path(); | |
1463 | BUG_ON(!path); | |
1464 | ||
b9473439 | 1465 | path->leave_spinning = 1; |
a1ed835e CM |
1466 | |
1467 | /* | |
1468 | * we may be replacing one extent in the tree with another. | |
1469 | * The new extent is pinned in the extent map, and we don't want | |
1470 | * to drop it from the cache until it is completely in the btree. | |
1471 | * | |
1472 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1473 | * the caller is expected to unpin it and allow it to be merged | |
1474 | * with the others. | |
1475 | */ | |
d899e052 | 1476 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
e980b50c | 1477 | file_pos + num_bytes, locked_end, |
a1ed835e | 1478 | file_pos, &hint, 0); |
d899e052 YZ |
1479 | BUG_ON(ret); |
1480 | ||
1481 | ins.objectid = inode->i_ino; | |
1482 | ins.offset = file_pos; | |
1483 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1484 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1485 | BUG_ON(ret); | |
1486 | leaf = path->nodes[0]; | |
1487 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1488 | struct btrfs_file_extent_item); | |
1489 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1490 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1491 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1492 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1493 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1494 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1495 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1496 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1497 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1498 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1499 | |
1500 | btrfs_unlock_up_safe(path, 1); | |
1501 | btrfs_set_lock_blocking(leaf); | |
1502 | ||
d899e052 YZ |
1503 | btrfs_mark_buffer_dirty(leaf); |
1504 | ||
1505 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1506 | |
1507 | ins.objectid = disk_bytenr; | |
1508 | ins.offset = disk_num_bytes; | |
1509 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1510 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1511 | root->root_key.objectid, | |
1512 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1513 | BUG_ON(ret); |
d899e052 | 1514 | btrfs_free_path(path); |
b9473439 | 1515 | |
d899e052 YZ |
1516 | return 0; |
1517 | } | |
1518 | ||
5d13a98f CM |
1519 | /* |
1520 | * helper function for btrfs_finish_ordered_io, this | |
1521 | * just reads in some of the csum leaves to prime them into ram | |
1522 | * before we start the transaction. It limits the amount of btree | |
1523 | * reads required while inside the transaction. | |
1524 | */ | |
1525 | static noinline void reada_csum(struct btrfs_root *root, | |
1526 | struct btrfs_path *path, | |
1527 | struct btrfs_ordered_extent *ordered_extent) | |
1528 | { | |
1529 | struct btrfs_ordered_sum *sum; | |
1530 | u64 bytenr; | |
1531 | ||
1532 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1533 | list); | |
1534 | bytenr = sum->sums[0].bytenr; | |
1535 | ||
1536 | /* | |
1537 | * we don't care about the results, the point of this search is | |
1538 | * just to get the btree leaves into ram | |
1539 | */ | |
1540 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1541 | } | |
1542 | ||
d352ac68 CM |
1543 | /* as ordered data IO finishes, this gets called so we can finish |
1544 | * an ordered extent if the range of bytes in the file it covers are | |
1545 | * fully written. | |
1546 | */ | |
211f90e6 | 1547 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1548 | { |
e6dcd2dc CM |
1549 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1550 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1551 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1552 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1553 | struct btrfs_path *path; |
d899e052 | 1554 | int compressed = 0; |
e6dcd2dc CM |
1555 | int ret; |
1556 | ||
1557 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1558 | if (!ret) |
e6dcd2dc | 1559 | return 0; |
e6dcd2dc | 1560 | |
b7ec40d7 CM |
1561 | /* |
1562 | * before we join the transaction, try to do some of our IO. | |
1563 | * This will limit the amount of IO that we have to do with | |
1564 | * the transaction running. We're unlikely to need to do any | |
1565 | * IO if the file extents are new, the disk_i_size checks | |
1566 | * covers the most common case. | |
1567 | */ | |
1568 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1569 | path = btrfs_alloc_path(); | |
1570 | if (path) { | |
1571 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1572 | inode->i_ino, | |
1573 | start, 0); | |
5d13a98f CM |
1574 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1575 | start); | |
1576 | if (!list_empty(&ordered_extent->list)) { | |
1577 | btrfs_release_path(root, path); | |
1578 | reada_csum(root, path, ordered_extent); | |
1579 | } | |
b7ec40d7 CM |
1580 | btrfs_free_path(path); |
1581 | } | |
1582 | } | |
1583 | ||
f9295749 | 1584 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc | 1585 | |
5d13a98f CM |
1586 | if (!ordered_extent) |
1587 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1588 | BUG_ON(!ordered_extent); |
7ea394f1 YZ |
1589 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1590 | goto nocow; | |
e6dcd2dc CM |
1591 | |
1592 | lock_extent(io_tree, ordered_extent->file_offset, | |
1593 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1594 | GFP_NOFS); | |
1595 | ||
c8b97818 | 1596 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1597 | compressed = 1; |
1598 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1599 | BUG_ON(compressed); | |
1600 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1601 | ordered_extent->file_offset, | |
1602 | ordered_extent->file_offset + | |
1603 | ordered_extent->len); | |
1604 | BUG_ON(ret); | |
1605 | } else { | |
1606 | ret = insert_reserved_file_extent(trans, inode, | |
1607 | ordered_extent->file_offset, | |
1608 | ordered_extent->start, | |
1609 | ordered_extent->disk_len, | |
1610 | ordered_extent->len, | |
1611 | ordered_extent->len, | |
e980b50c CM |
1612 | ordered_extent->file_offset + |
1613 | ordered_extent->len, | |
d899e052 YZ |
1614 | compressed, 0, 0, |
1615 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1616 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1617 | ordered_extent->file_offset, | |
1618 | ordered_extent->len); | |
d899e052 YZ |
1619 | BUG_ON(ret); |
1620 | } | |
e6dcd2dc CM |
1621 | unlock_extent(io_tree, ordered_extent->file_offset, |
1622 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1623 | GFP_NOFS); | |
7ea394f1 | 1624 | nocow: |
e6dcd2dc CM |
1625 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1626 | &ordered_extent->list); | |
1627 | ||
34353029 | 1628 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1629 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1630 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1631 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1632 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1633 | |
e6dcd2dc CM |
1634 | /* once for us */ |
1635 | btrfs_put_ordered_extent(ordered_extent); | |
1636 | /* once for the tree */ | |
1637 | btrfs_put_ordered_extent(ordered_extent); | |
1638 | ||
e6dcd2dc CM |
1639 | btrfs_end_transaction(trans, root); |
1640 | return 0; | |
1641 | } | |
1642 | ||
b2950863 | 1643 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1644 | struct extent_state *state, int uptodate) |
1645 | { | |
8b62b72b | 1646 | ClearPagePrivate2(page); |
211f90e6 CM |
1647 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1648 | } | |
1649 | ||
d352ac68 CM |
1650 | /* |
1651 | * When IO fails, either with EIO or csum verification fails, we | |
1652 | * try other mirrors that might have a good copy of the data. This | |
1653 | * io_failure_record is used to record state as we go through all the | |
1654 | * mirrors. If another mirror has good data, the page is set up to date | |
1655 | * and things continue. If a good mirror can't be found, the original | |
1656 | * bio end_io callback is called to indicate things have failed. | |
1657 | */ | |
7e38326f CM |
1658 | struct io_failure_record { |
1659 | struct page *page; | |
1660 | u64 start; | |
1661 | u64 len; | |
1662 | u64 logical; | |
d20f7043 | 1663 | unsigned long bio_flags; |
7e38326f CM |
1664 | int last_mirror; |
1665 | }; | |
1666 | ||
b2950863 | 1667 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1668 | struct page *page, u64 start, u64 end, |
1669 | struct extent_state *state) | |
7e38326f CM |
1670 | { |
1671 | struct io_failure_record *failrec = NULL; | |
1672 | u64 private; | |
1673 | struct extent_map *em; | |
1674 | struct inode *inode = page->mapping->host; | |
1675 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1676 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1677 | struct bio *bio; |
1678 | int num_copies; | |
1679 | int ret; | |
1259ab75 | 1680 | int rw; |
7e38326f CM |
1681 | u64 logical; |
1682 | ||
1683 | ret = get_state_private(failure_tree, start, &private); | |
1684 | if (ret) { | |
7e38326f CM |
1685 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1686 | if (!failrec) | |
1687 | return -ENOMEM; | |
1688 | failrec->start = start; | |
1689 | failrec->len = end - start + 1; | |
1690 | failrec->last_mirror = 0; | |
d20f7043 | 1691 | failrec->bio_flags = 0; |
7e38326f | 1692 | |
890871be | 1693 | read_lock(&em_tree->lock); |
3b951516 CM |
1694 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1695 | if (em->start > start || em->start + em->len < start) { | |
1696 | free_extent_map(em); | |
1697 | em = NULL; | |
1698 | } | |
890871be | 1699 | read_unlock(&em_tree->lock); |
7e38326f CM |
1700 | |
1701 | if (!em || IS_ERR(em)) { | |
1702 | kfree(failrec); | |
1703 | return -EIO; | |
1704 | } | |
1705 | logical = start - em->start; | |
1706 | logical = em->block_start + logical; | |
d20f7043 CM |
1707 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1708 | logical = em->block_start; | |
1709 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1710 | } | |
7e38326f CM |
1711 | failrec->logical = logical; |
1712 | free_extent_map(em); | |
1713 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1714 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1715 | set_state_private(failure_tree, start, |
1716 | (u64)(unsigned long)failrec); | |
7e38326f | 1717 | } else { |
587f7704 | 1718 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1719 | } |
1720 | num_copies = btrfs_num_copies( | |
1721 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1722 | failrec->logical, failrec->len); | |
1723 | failrec->last_mirror++; | |
1724 | if (!state) { | |
cad321ad | 1725 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1726 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1727 | failrec->start, | |
1728 | EXTENT_LOCKED); | |
1729 | if (state && state->start != failrec->start) | |
1730 | state = NULL; | |
cad321ad | 1731 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1732 | } |
1733 | if (!state || failrec->last_mirror > num_copies) { | |
1734 | set_state_private(failure_tree, failrec->start, 0); | |
1735 | clear_extent_bits(failure_tree, failrec->start, | |
1736 | failrec->start + failrec->len - 1, | |
1737 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1738 | kfree(failrec); | |
1739 | return -EIO; | |
1740 | } | |
1741 | bio = bio_alloc(GFP_NOFS, 1); | |
1742 | bio->bi_private = state; | |
1743 | bio->bi_end_io = failed_bio->bi_end_io; | |
1744 | bio->bi_sector = failrec->logical >> 9; | |
1745 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1746 | bio->bi_size = 0; |
d20f7043 | 1747 | |
7e38326f | 1748 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1749 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1750 | rw = WRITE; | |
1751 | else | |
1752 | rw = READ; | |
1753 | ||
1754 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1755 | failrec->last_mirror, |
d20f7043 | 1756 | failrec->bio_flags); |
1259ab75 CM |
1757 | return 0; |
1758 | } | |
1759 | ||
d352ac68 CM |
1760 | /* |
1761 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1762 | * to see if we need to process or clean up an io_failure_record | |
1763 | */ | |
b2950863 | 1764 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1765 | { |
1766 | u64 private; | |
1767 | u64 private_failure; | |
1768 | struct io_failure_record *failure; | |
1769 | int ret; | |
1770 | ||
1771 | private = 0; | |
1772 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1773 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1774 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1775 | start, &private_failure); | |
1776 | if (ret == 0) { | |
1777 | failure = (struct io_failure_record *)(unsigned long) | |
1778 | private_failure; | |
1779 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1780 | failure->start, 0); | |
1781 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1782 | failure->start, | |
1783 | failure->start + failure->len - 1, | |
1784 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1785 | GFP_NOFS); | |
1786 | kfree(failure); | |
1787 | } | |
1788 | } | |
7e38326f CM |
1789 | return 0; |
1790 | } | |
1791 | ||
d352ac68 CM |
1792 | /* |
1793 | * when reads are done, we need to check csums to verify the data is correct | |
1794 | * if there's a match, we allow the bio to finish. If not, we go through | |
1795 | * the io_failure_record routines to find good copies | |
1796 | */ | |
b2950863 | 1797 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1798 | struct extent_state *state) |
07157aac | 1799 | { |
35ebb934 | 1800 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1801 | struct inode *inode = page->mapping->host; |
d1310b2e | 1802 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1803 | char *kaddr; |
aadfeb6e | 1804 | u64 private = ~(u32)0; |
07157aac | 1805 | int ret; |
ff79f819 CM |
1806 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1807 | u32 csum = ~(u32)0; | |
d1310b2e | 1808 | |
d20f7043 CM |
1809 | if (PageChecked(page)) { |
1810 | ClearPageChecked(page); | |
1811 | goto good; | |
1812 | } | |
6cbff00f CH |
1813 | |
1814 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1815 | return 0; |
1816 | ||
1817 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1818 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1819 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1820 | GFP_NOFS); | |
b6cda9bc | 1821 | return 0; |
17d217fe | 1822 | } |
d20f7043 | 1823 | |
c2e639f0 | 1824 | if (state && state->start == start) { |
70dec807 CM |
1825 | private = state->private; |
1826 | ret = 0; | |
1827 | } else { | |
1828 | ret = get_state_private(io_tree, start, &private); | |
1829 | } | |
9ab86c8e | 1830 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1831 | if (ret) |
07157aac | 1832 | goto zeroit; |
d397712b | 1833 | |
ff79f819 CM |
1834 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1835 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1836 | if (csum != private) |
07157aac | 1837 | goto zeroit; |
d397712b | 1838 | |
9ab86c8e | 1839 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1840 | good: |
7e38326f CM |
1841 | /* if the io failure tree for this inode is non-empty, |
1842 | * check to see if we've recovered from a failed IO | |
1843 | */ | |
1259ab75 | 1844 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1845 | return 0; |
1846 | ||
1847 | zeroit: | |
193f284d CM |
1848 | if (printk_ratelimit()) { |
1849 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
1850 | "private %llu\n", page->mapping->host->i_ino, | |
1851 | (unsigned long long)start, csum, | |
1852 | (unsigned long long)private); | |
1853 | } | |
db94535d CM |
1854 | memset(kaddr + offset, 1, end - start + 1); |
1855 | flush_dcache_page(page); | |
9ab86c8e | 1856 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1857 | if (private == 0) |
1858 | return 0; | |
7e38326f | 1859 | return -EIO; |
07157aac | 1860 | } |
b888db2b | 1861 | |
7b128766 JB |
1862 | /* |
1863 | * This creates an orphan entry for the given inode in case something goes | |
1864 | * wrong in the middle of an unlink/truncate. | |
1865 | */ | |
1866 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1867 | { | |
1868 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1869 | int ret = 0; | |
1870 | ||
bcc63abb | 1871 | spin_lock(&root->list_lock); |
7b128766 JB |
1872 | |
1873 | /* already on the orphan list, we're good */ | |
1874 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1875 | spin_unlock(&root->list_lock); |
7b128766 JB |
1876 | return 0; |
1877 | } | |
1878 | ||
1879 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
1880 | ||
bcc63abb | 1881 | spin_unlock(&root->list_lock); |
7b128766 JB |
1882 | |
1883 | /* | |
1884 | * insert an orphan item to track this unlinked/truncated file | |
1885 | */ | |
1886 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
1887 | ||
1888 | return ret; | |
1889 | } | |
1890 | ||
1891 | /* | |
1892 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
1893 | * item for this particular inode. | |
1894 | */ | |
1895 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
1896 | { | |
1897 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1898 | int ret = 0; | |
1899 | ||
bcc63abb | 1900 | spin_lock(&root->list_lock); |
7b128766 JB |
1901 | |
1902 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1903 | spin_unlock(&root->list_lock); |
7b128766 JB |
1904 | return 0; |
1905 | } | |
1906 | ||
1907 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
1908 | if (!trans) { | |
bcc63abb | 1909 | spin_unlock(&root->list_lock); |
7b128766 JB |
1910 | return 0; |
1911 | } | |
1912 | ||
bcc63abb | 1913 | spin_unlock(&root->list_lock); |
7b128766 JB |
1914 | |
1915 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
1916 | ||
1917 | return ret; | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * this cleans up any orphans that may be left on the list from the last use | |
1922 | * of this root. | |
1923 | */ | |
1924 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
1925 | { | |
1926 | struct btrfs_path *path; | |
1927 | struct extent_buffer *leaf; | |
1928 | struct btrfs_item *item; | |
1929 | struct btrfs_key key, found_key; | |
1930 | struct btrfs_trans_handle *trans; | |
1931 | struct inode *inode; | |
1932 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
1933 | ||
7b128766 JB |
1934 | path = btrfs_alloc_path(); |
1935 | if (!path) | |
1936 | return; | |
1937 | path->reada = -1; | |
1938 | ||
1939 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
1940 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
1941 | key.offset = (u64)-1; | |
1942 | ||
7b128766 JB |
1943 | |
1944 | while (1) { | |
1945 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1946 | if (ret < 0) { | |
1947 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
1948 | "\n", ret); | |
1949 | break; | |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * if ret == 0 means we found what we were searching for, which | |
1954 | * is weird, but possible, so only screw with path if we didnt | |
1955 | * find the key and see if we have stuff that matches | |
1956 | */ | |
1957 | if (ret > 0) { | |
1958 | if (path->slots[0] == 0) | |
1959 | break; | |
1960 | path->slots[0]--; | |
1961 | } | |
1962 | ||
1963 | /* pull out the item */ | |
1964 | leaf = path->nodes[0]; | |
1965 | item = btrfs_item_nr(leaf, path->slots[0]); | |
1966 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1967 | ||
1968 | /* make sure the item matches what we want */ | |
1969 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
1970 | break; | |
1971 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
1972 | break; | |
1973 | ||
1974 | /* release the path since we're done with it */ | |
1975 | btrfs_release_path(root, path); | |
1976 | ||
1977 | /* | |
1978 | * this is where we are basically btrfs_lookup, without the | |
1979 | * crossing root thing. we store the inode number in the | |
1980 | * offset of the orphan item. | |
1981 | */ | |
5d4f98a2 YZ |
1982 | found_key.objectid = found_key.offset; |
1983 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
1984 | found_key.offset = 0; | |
1985 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
1986 | if (IS_ERR(inode)) | |
7b128766 JB |
1987 | break; |
1988 | ||
7b128766 JB |
1989 | /* |
1990 | * add this inode to the orphan list so btrfs_orphan_del does | |
1991 | * the proper thing when we hit it | |
1992 | */ | |
bcc63abb | 1993 | spin_lock(&root->list_lock); |
7b128766 | 1994 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 1995 | spin_unlock(&root->list_lock); |
7b128766 JB |
1996 | |
1997 | /* | |
1998 | * if this is a bad inode, means we actually succeeded in | |
1999 | * removing the inode, but not the orphan record, which means | |
2000 | * we need to manually delete the orphan since iput will just | |
2001 | * do a destroy_inode | |
2002 | */ | |
2003 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2004 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2005 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2006 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2007 | iput(inode); |
2008 | continue; | |
2009 | } | |
2010 | ||
2011 | /* if we have links, this was a truncate, lets do that */ | |
2012 | if (inode->i_nlink) { | |
2013 | nr_truncate++; | |
2014 | btrfs_truncate(inode); | |
2015 | } else { | |
2016 | nr_unlink++; | |
2017 | } | |
2018 | ||
2019 | /* this will do delete_inode and everything for us */ | |
2020 | iput(inode); | |
2021 | } | |
2022 | ||
2023 | if (nr_unlink) | |
2024 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2025 | if (nr_truncate) | |
2026 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2027 | ||
2028 | btrfs_free_path(path); | |
7b128766 JB |
2029 | } |
2030 | ||
46a53cca CM |
2031 | /* |
2032 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2033 | * don't find any xattrs, we know there can't be any acls. | |
2034 | * | |
2035 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2036 | */ | |
2037 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2038 | int slot, u64 objectid) | |
2039 | { | |
2040 | u32 nritems = btrfs_header_nritems(leaf); | |
2041 | struct btrfs_key found_key; | |
2042 | int scanned = 0; | |
2043 | ||
2044 | slot++; | |
2045 | while (slot < nritems) { | |
2046 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2047 | ||
2048 | /* we found a different objectid, there must not be acls */ | |
2049 | if (found_key.objectid != objectid) | |
2050 | return 0; | |
2051 | ||
2052 | /* we found an xattr, assume we've got an acl */ | |
2053 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2054 | return 1; | |
2055 | ||
2056 | /* | |
2057 | * we found a key greater than an xattr key, there can't | |
2058 | * be any acls later on | |
2059 | */ | |
2060 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2061 | return 0; | |
2062 | ||
2063 | slot++; | |
2064 | scanned++; | |
2065 | ||
2066 | /* | |
2067 | * it goes inode, inode backrefs, xattrs, extents, | |
2068 | * so if there are a ton of hard links to an inode there can | |
2069 | * be a lot of backrefs. Don't waste time searching too hard, | |
2070 | * this is just an optimization | |
2071 | */ | |
2072 | if (scanned >= 8) | |
2073 | break; | |
2074 | } | |
2075 | /* we hit the end of the leaf before we found an xattr or | |
2076 | * something larger than an xattr. We have to assume the inode | |
2077 | * has acls | |
2078 | */ | |
2079 | return 1; | |
2080 | } | |
2081 | ||
d352ac68 CM |
2082 | /* |
2083 | * read an inode from the btree into the in-memory inode | |
2084 | */ | |
5d4f98a2 | 2085 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2086 | { |
2087 | struct btrfs_path *path; | |
5f39d397 | 2088 | struct extent_buffer *leaf; |
39279cc3 | 2089 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2090 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2091 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2092 | struct btrfs_key location; | |
46a53cca | 2093 | int maybe_acls; |
39279cc3 | 2094 | u64 alloc_group_block; |
618e21d5 | 2095 | u32 rdev; |
39279cc3 CM |
2096 | int ret; |
2097 | ||
2098 | path = btrfs_alloc_path(); | |
2099 | BUG_ON(!path); | |
39279cc3 | 2100 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2101 | |
39279cc3 | 2102 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2103 | if (ret) |
39279cc3 | 2104 | goto make_bad; |
39279cc3 | 2105 | |
5f39d397 CM |
2106 | leaf = path->nodes[0]; |
2107 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2108 | struct btrfs_inode_item); | |
2109 | ||
2110 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2111 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2112 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2113 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2114 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2115 | |
2116 | tspec = btrfs_inode_atime(inode_item); | |
2117 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2118 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2119 | ||
2120 | tspec = btrfs_inode_mtime(inode_item); | |
2121 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2122 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2123 | ||
2124 | tspec = btrfs_inode_ctime(inode_item); | |
2125 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2126 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2127 | ||
a76a3cd4 | 2128 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2129 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2130 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2131 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2132 | inode->i_rdev = 0; |
5f39d397 CM |
2133 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2134 | ||
aec7477b | 2135 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2136 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2137 | |
5f39d397 | 2138 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2139 | |
46a53cca CM |
2140 | /* |
2141 | * try to precache a NULL acl entry for files that don't have | |
2142 | * any xattrs or acls | |
2143 | */ | |
2144 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2145 | if (!maybe_acls) |
2146 | cache_no_acl(inode); | |
46a53cca | 2147 | |
d2fb3437 YZ |
2148 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2149 | alloc_group_block, 0); | |
39279cc3 CM |
2150 | btrfs_free_path(path); |
2151 | inode_item = NULL; | |
2152 | ||
39279cc3 | 2153 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2154 | case S_IFREG: |
2155 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2156 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2157 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2158 | inode->i_fop = &btrfs_file_operations; |
2159 | inode->i_op = &btrfs_file_inode_operations; | |
2160 | break; | |
2161 | case S_IFDIR: | |
2162 | inode->i_fop = &btrfs_dir_file_operations; | |
2163 | if (root == root->fs_info->tree_root) | |
2164 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2165 | else | |
2166 | inode->i_op = &btrfs_dir_inode_operations; | |
2167 | break; | |
2168 | case S_IFLNK: | |
2169 | inode->i_op = &btrfs_symlink_inode_operations; | |
2170 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2171 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2172 | break; |
618e21d5 | 2173 | default: |
0279b4cd | 2174 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2175 | init_special_inode(inode, inode->i_mode, rdev); |
2176 | break; | |
39279cc3 | 2177 | } |
6cbff00f CH |
2178 | |
2179 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2180 | return; |
2181 | ||
2182 | make_bad: | |
39279cc3 | 2183 | btrfs_free_path(path); |
39279cc3 CM |
2184 | make_bad_inode(inode); |
2185 | } | |
2186 | ||
d352ac68 CM |
2187 | /* |
2188 | * given a leaf and an inode, copy the inode fields into the leaf | |
2189 | */ | |
e02119d5 CM |
2190 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2191 | struct extent_buffer *leaf, | |
5f39d397 | 2192 | struct btrfs_inode_item *item, |
39279cc3 CM |
2193 | struct inode *inode) |
2194 | { | |
5f39d397 CM |
2195 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2196 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2197 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2198 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2199 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2200 | ||
2201 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2202 | inode->i_atime.tv_sec); | |
2203 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2204 | inode->i_atime.tv_nsec); | |
2205 | ||
2206 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2207 | inode->i_mtime.tv_sec); | |
2208 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2209 | inode->i_mtime.tv_nsec); | |
2210 | ||
2211 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2212 | inode->i_ctime.tv_sec); | |
2213 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2214 | inode->i_ctime.tv_nsec); | |
2215 | ||
a76a3cd4 | 2216 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2217 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2218 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2219 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2220 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2221 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2222 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2223 | } |
2224 | ||
d352ac68 CM |
2225 | /* |
2226 | * copy everything in the in-memory inode into the btree. | |
2227 | */ | |
d397712b CM |
2228 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2229 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2230 | { |
2231 | struct btrfs_inode_item *inode_item; | |
2232 | struct btrfs_path *path; | |
5f39d397 | 2233 | struct extent_buffer *leaf; |
39279cc3 CM |
2234 | int ret; |
2235 | ||
2236 | path = btrfs_alloc_path(); | |
2237 | BUG_ON(!path); | |
b9473439 | 2238 | path->leave_spinning = 1; |
39279cc3 CM |
2239 | ret = btrfs_lookup_inode(trans, root, path, |
2240 | &BTRFS_I(inode)->location, 1); | |
2241 | if (ret) { | |
2242 | if (ret > 0) | |
2243 | ret = -ENOENT; | |
2244 | goto failed; | |
2245 | } | |
2246 | ||
b4ce94de | 2247 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2248 | leaf = path->nodes[0]; |
2249 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2250 | struct btrfs_inode_item); |
2251 | ||
e02119d5 | 2252 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2253 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2254 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2255 | ret = 0; |
2256 | failed: | |
39279cc3 CM |
2257 | btrfs_free_path(path); |
2258 | return ret; | |
2259 | } | |
2260 | ||
2261 | ||
d352ac68 CM |
2262 | /* |
2263 | * unlink helper that gets used here in inode.c and in the tree logging | |
2264 | * recovery code. It remove a link in a directory with a given name, and | |
2265 | * also drops the back refs in the inode to the directory | |
2266 | */ | |
e02119d5 CM |
2267 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2268 | struct btrfs_root *root, | |
2269 | struct inode *dir, struct inode *inode, | |
2270 | const char *name, int name_len) | |
39279cc3 CM |
2271 | { |
2272 | struct btrfs_path *path; | |
39279cc3 | 2273 | int ret = 0; |
5f39d397 | 2274 | struct extent_buffer *leaf; |
39279cc3 | 2275 | struct btrfs_dir_item *di; |
5f39d397 | 2276 | struct btrfs_key key; |
aec7477b | 2277 | u64 index; |
39279cc3 CM |
2278 | |
2279 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2280 | if (!path) { |
2281 | ret = -ENOMEM; | |
2282 | goto err; | |
2283 | } | |
2284 | ||
b9473439 | 2285 | path->leave_spinning = 1; |
39279cc3 CM |
2286 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2287 | name, name_len, -1); | |
2288 | if (IS_ERR(di)) { | |
2289 | ret = PTR_ERR(di); | |
2290 | goto err; | |
2291 | } | |
2292 | if (!di) { | |
2293 | ret = -ENOENT; | |
2294 | goto err; | |
2295 | } | |
5f39d397 CM |
2296 | leaf = path->nodes[0]; |
2297 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2298 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2299 | if (ret) |
2300 | goto err; | |
39279cc3 CM |
2301 | btrfs_release_path(root, path); |
2302 | ||
aec7477b | 2303 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2304 | inode->i_ino, |
2305 | dir->i_ino, &index); | |
aec7477b | 2306 | if (ret) { |
d397712b | 2307 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2308 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2309 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2310 | goto err; |
2311 | } | |
2312 | ||
39279cc3 | 2313 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2314 | index, name, name_len, -1); |
39279cc3 CM |
2315 | if (IS_ERR(di)) { |
2316 | ret = PTR_ERR(di); | |
2317 | goto err; | |
2318 | } | |
2319 | if (!di) { | |
2320 | ret = -ENOENT; | |
2321 | goto err; | |
2322 | } | |
2323 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2324 | btrfs_release_path(root, path); |
39279cc3 | 2325 | |
e02119d5 CM |
2326 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2327 | inode, dir->i_ino); | |
49eb7e46 | 2328 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2329 | |
2330 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2331 | dir, index); | |
2332 | BUG_ON(ret); | |
39279cc3 CM |
2333 | err: |
2334 | btrfs_free_path(path); | |
e02119d5 CM |
2335 | if (ret) |
2336 | goto out; | |
2337 | ||
2338 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2339 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2340 | btrfs_update_inode(trans, root, dir); | |
2341 | btrfs_drop_nlink(inode); | |
2342 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2343 | out: |
39279cc3 CM |
2344 | return ret; |
2345 | } | |
2346 | ||
2347 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2348 | { | |
2349 | struct btrfs_root *root; | |
2350 | struct btrfs_trans_handle *trans; | |
7b128766 | 2351 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2352 | int ret; |
1832a6d5 | 2353 | unsigned long nr = 0; |
39279cc3 CM |
2354 | |
2355 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2356 | |
39279cc3 | 2357 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2358 | |
39279cc3 | 2359 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2360 | |
2361 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2362 | ||
e02119d5 CM |
2363 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2364 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2365 | |
2366 | if (inode->i_nlink == 0) | |
2367 | ret = btrfs_orphan_add(trans, inode); | |
2368 | ||
d3c2fdcf | 2369 | nr = trans->blocks_used; |
5f39d397 | 2370 | |
89ce8a63 | 2371 | btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2372 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2373 | return ret; |
2374 | } | |
2375 | ||
4df27c4d YZ |
2376 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2377 | struct btrfs_root *root, | |
2378 | struct inode *dir, u64 objectid, | |
2379 | const char *name, int name_len) | |
2380 | { | |
2381 | struct btrfs_path *path; | |
2382 | struct extent_buffer *leaf; | |
2383 | struct btrfs_dir_item *di; | |
2384 | struct btrfs_key key; | |
2385 | u64 index; | |
2386 | int ret; | |
2387 | ||
2388 | path = btrfs_alloc_path(); | |
2389 | if (!path) | |
2390 | return -ENOMEM; | |
2391 | ||
2392 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2393 | name, name_len, -1); | |
2394 | BUG_ON(!di || IS_ERR(di)); | |
2395 | ||
2396 | leaf = path->nodes[0]; | |
2397 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2398 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2399 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2400 | BUG_ON(ret); | |
2401 | btrfs_release_path(root, path); | |
2402 | ||
2403 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2404 | objectid, root->root_key.objectid, | |
2405 | dir->i_ino, &index, name, name_len); | |
2406 | if (ret < 0) { | |
2407 | BUG_ON(ret != -ENOENT); | |
2408 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2409 | name, name_len); | |
2410 | BUG_ON(!di || IS_ERR(di)); | |
2411 | ||
2412 | leaf = path->nodes[0]; | |
2413 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2414 | btrfs_release_path(root, path); | |
2415 | index = key.offset; | |
2416 | } | |
2417 | ||
2418 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2419 | index, name, name_len, -1); | |
2420 | BUG_ON(!di || IS_ERR(di)); | |
2421 | ||
2422 | leaf = path->nodes[0]; | |
2423 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2424 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2425 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2426 | BUG_ON(ret); | |
2427 | btrfs_release_path(root, path); | |
2428 | ||
2429 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2430 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2431 | ret = btrfs_update_inode(trans, root, dir); | |
2432 | BUG_ON(ret); | |
2433 | dir->i_sb->s_dirt = 1; | |
2434 | ||
2435 | btrfs_free_path(path); | |
2436 | return 0; | |
2437 | } | |
2438 | ||
39279cc3 CM |
2439 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2440 | { | |
2441 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2442 | int err = 0; |
39279cc3 CM |
2443 | int ret; |
2444 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2445 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2446 | unsigned long nr = 0; |
39279cc3 | 2447 | |
3394e160 | 2448 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2449 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2450 | return -ENOTEMPTY; |
2451 | ||
39279cc3 CM |
2452 | trans = btrfs_start_transaction(root, 1); |
2453 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2454 | |
4df27c4d YZ |
2455 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2456 | err = btrfs_unlink_subvol(trans, root, dir, | |
2457 | BTRFS_I(inode)->location.objectid, | |
2458 | dentry->d_name.name, | |
2459 | dentry->d_name.len); | |
2460 | goto out; | |
2461 | } | |
2462 | ||
7b128766 JB |
2463 | err = btrfs_orphan_add(trans, inode); |
2464 | if (err) | |
4df27c4d | 2465 | goto out; |
7b128766 | 2466 | |
39279cc3 | 2467 | /* now the directory is empty */ |
e02119d5 CM |
2468 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2469 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2470 | if (!err) |
dbe674a9 | 2471 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2472 | out: |
d3c2fdcf | 2473 | nr = trans->blocks_used; |
89ce8a63 | 2474 | ret = btrfs_end_transaction_throttle(trans, root); |
d3c2fdcf | 2475 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2476 | |
39279cc3 CM |
2477 | if (ret && !err) |
2478 | err = ret; | |
2479 | return err; | |
2480 | } | |
2481 | ||
d20f7043 | 2482 | #if 0 |
323ac95b CM |
2483 | /* |
2484 | * when truncating bytes in a file, it is possible to avoid reading | |
2485 | * the leaves that contain only checksum items. This can be the | |
2486 | * majority of the IO required to delete a large file, but it must | |
2487 | * be done carefully. | |
2488 | * | |
2489 | * The keys in the level just above the leaves are checked to make sure | |
2490 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2491 | * after the new size. | |
2492 | * | |
2493 | * Then the key for the next leaf is checked to make sure it also has | |
2494 | * a checksum item for the same file. If it does, we know our target leaf | |
2495 | * contains only checksum items, and it can be safely freed without reading | |
2496 | * it. | |
2497 | * | |
2498 | * This is just an optimization targeted at large files. It may do | |
2499 | * nothing. It will return 0 unless things went badly. | |
2500 | */ | |
2501 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2502 | struct btrfs_root *root, | |
2503 | struct btrfs_path *path, | |
2504 | struct inode *inode, u64 new_size) | |
2505 | { | |
2506 | struct btrfs_key key; | |
2507 | int ret; | |
2508 | int nritems; | |
2509 | struct btrfs_key found_key; | |
2510 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2511 | struct btrfs_leaf_ref *ref; |
2512 | u64 leaf_gen; | |
2513 | u64 leaf_start; | |
323ac95b CM |
2514 | |
2515 | path->lowest_level = 1; | |
2516 | key.objectid = inode->i_ino; | |
2517 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2518 | key.offset = new_size; | |
2519 | again: | |
2520 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2521 | if (ret < 0) | |
2522 | goto out; | |
2523 | ||
2524 | if (path->nodes[1] == NULL) { | |
2525 | ret = 0; | |
2526 | goto out; | |
2527 | } | |
2528 | ret = 0; | |
2529 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2530 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2531 | ||
2532 | if (!nritems) | |
2533 | goto out; | |
2534 | ||
2535 | if (path->slots[1] >= nritems) | |
2536 | goto next_node; | |
2537 | ||
2538 | /* did we find a key greater than anything we want to delete? */ | |
2539 | if (found_key.objectid > inode->i_ino || | |
2540 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2541 | goto out; | |
2542 | ||
2543 | /* we check the next key in the node to make sure the leave contains | |
2544 | * only checksum items. This comparison doesn't work if our | |
2545 | * leaf is the last one in the node | |
2546 | */ | |
2547 | if (path->slots[1] + 1 >= nritems) { | |
2548 | next_node: | |
2549 | /* search forward from the last key in the node, this | |
2550 | * will bring us into the next node in the tree | |
2551 | */ | |
2552 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2553 | ||
2554 | /* unlikely, but we inc below, so check to be safe */ | |
2555 | if (found_key.offset == (u64)-1) | |
2556 | goto out; | |
2557 | ||
2558 | /* search_forward needs a path with locks held, do the | |
2559 | * search again for the original key. It is possible | |
2560 | * this will race with a balance and return a path that | |
2561 | * we could modify, but this drop is just an optimization | |
2562 | * and is allowed to miss some leaves. | |
2563 | */ | |
2564 | btrfs_release_path(root, path); | |
2565 | found_key.offset++; | |
2566 | ||
2567 | /* setup a max key for search_forward */ | |
2568 | other_key.offset = (u64)-1; | |
2569 | other_key.type = key.type; | |
2570 | other_key.objectid = key.objectid; | |
2571 | ||
2572 | path->keep_locks = 1; | |
2573 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2574 | path, 0, 0); | |
2575 | path->keep_locks = 0; | |
2576 | if (ret || found_key.objectid != key.objectid || | |
2577 | found_key.type != key.type) { | |
2578 | ret = 0; | |
2579 | goto out; | |
2580 | } | |
2581 | ||
2582 | key.offset = found_key.offset; | |
2583 | btrfs_release_path(root, path); | |
2584 | cond_resched(); | |
2585 | goto again; | |
2586 | } | |
2587 | ||
2588 | /* we know there's one more slot after us in the tree, | |
2589 | * read that key so we can verify it is also a checksum item | |
2590 | */ | |
2591 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2592 | ||
2593 | if (found_key.objectid < inode->i_ino) | |
2594 | goto next_key; | |
2595 | ||
2596 | if (found_key.type != key.type || found_key.offset < new_size) | |
2597 | goto next_key; | |
2598 | ||
2599 | /* | |
2600 | * if the key for the next leaf isn't a csum key from this objectid, | |
2601 | * we can't be sure there aren't good items inside this leaf. | |
2602 | * Bail out | |
2603 | */ | |
2604 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2605 | goto out; | |
2606 | ||
5b84e8d6 YZ |
2607 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2608 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2609 | /* |
2610 | * it is safe to delete this leaf, it contains only | |
2611 | * csum items from this inode at an offset >= new_size | |
2612 | */ | |
5b84e8d6 | 2613 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2614 | BUG_ON(ret); |
2615 | ||
5b84e8d6 YZ |
2616 | if (root->ref_cows && leaf_gen < trans->transid) { |
2617 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2618 | if (ref) { | |
2619 | ref->root_gen = root->root_key.offset; | |
2620 | ref->bytenr = leaf_start; | |
2621 | ref->owner = 0; | |
2622 | ref->generation = leaf_gen; | |
2623 | ref->nritems = 0; | |
2624 | ||
bd56b302 CM |
2625 | btrfs_sort_leaf_ref(ref); |
2626 | ||
5b84e8d6 YZ |
2627 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2628 | WARN_ON(ret); | |
2629 | btrfs_free_leaf_ref(root, ref); | |
2630 | } else { | |
2631 | WARN_ON(1); | |
2632 | } | |
2633 | } | |
323ac95b CM |
2634 | next_key: |
2635 | btrfs_release_path(root, path); | |
2636 | ||
2637 | if (other_key.objectid == inode->i_ino && | |
2638 | other_key.type == key.type && other_key.offset > key.offset) { | |
2639 | key.offset = other_key.offset; | |
2640 | cond_resched(); | |
2641 | goto again; | |
2642 | } | |
2643 | ret = 0; | |
2644 | out: | |
2645 | /* fixup any changes we've made to the path */ | |
2646 | path->lowest_level = 0; | |
2647 | path->keep_locks = 0; | |
2648 | btrfs_release_path(root, path); | |
2649 | return ret; | |
2650 | } | |
2651 | ||
d20f7043 CM |
2652 | #endif |
2653 | ||
39279cc3 CM |
2654 | /* |
2655 | * this can truncate away extent items, csum items and directory items. | |
2656 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2657 | * any higher than new_size |
39279cc3 CM |
2658 | * |
2659 | * csum items that cross the new i_size are truncated to the new size | |
2660 | * as well. | |
7b128766 JB |
2661 | * |
2662 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2663 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2664 | */ |
e02119d5 CM |
2665 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2666 | struct btrfs_root *root, | |
2667 | struct inode *inode, | |
2668 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2669 | { |
2670 | int ret; | |
2671 | struct btrfs_path *path; | |
2672 | struct btrfs_key key; | |
5f39d397 | 2673 | struct btrfs_key found_key; |
06d9a8d7 | 2674 | u32 found_type = (u8)-1; |
5f39d397 | 2675 | struct extent_buffer *leaf; |
39279cc3 CM |
2676 | struct btrfs_file_extent_item *fi; |
2677 | u64 extent_start = 0; | |
db94535d | 2678 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2679 | u64 extent_offset = 0; |
39279cc3 CM |
2680 | u64 item_end = 0; |
2681 | int found_extent; | |
2682 | int del_item; | |
85e21bac CM |
2683 | int pending_del_nr = 0; |
2684 | int pending_del_slot = 0; | |
179e29e4 | 2685 | int extent_type = -1; |
771ed689 | 2686 | int encoding; |
3b951516 | 2687 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2688 | |
e02119d5 | 2689 | if (root->ref_cows) |
5b21f2ed | 2690 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2691 | path = btrfs_alloc_path(); |
2692 | BUG_ON(!path); | |
33c17ad5 | 2693 | path->reada = -1; |
5f39d397 | 2694 | |
39279cc3 CM |
2695 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2696 | key.objectid = inode->i_ino; | |
2697 | key.offset = (u64)-1; | |
5f39d397 CM |
2698 | key.type = (u8)-1; |
2699 | ||
85e21bac | 2700 | search_again: |
b9473439 | 2701 | path->leave_spinning = 1; |
85e21bac | 2702 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2703 | if (ret < 0) |
85e21bac | 2704 | goto error; |
d397712b | 2705 | |
85e21bac | 2706 | if (ret > 0) { |
e02119d5 CM |
2707 | /* there are no items in the tree for us to truncate, we're |
2708 | * done | |
2709 | */ | |
2710 | if (path->slots[0] == 0) { | |
2711 | ret = 0; | |
2712 | goto error; | |
2713 | } | |
85e21bac CM |
2714 | path->slots[0]--; |
2715 | } | |
2716 | ||
d397712b | 2717 | while (1) { |
39279cc3 | 2718 | fi = NULL; |
5f39d397 CM |
2719 | leaf = path->nodes[0]; |
2720 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2721 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2722 | encoding = 0; |
39279cc3 | 2723 | |
5f39d397 | 2724 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2725 | break; |
5f39d397 | 2726 | |
85e21bac | 2727 | if (found_type < min_type) |
39279cc3 CM |
2728 | break; |
2729 | ||
5f39d397 | 2730 | item_end = found_key.offset; |
39279cc3 | 2731 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2732 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2733 | struct btrfs_file_extent_item); |
179e29e4 | 2734 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2735 | encoding = btrfs_file_extent_compression(leaf, fi); |
2736 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2737 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2738 | ||
179e29e4 | 2739 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2740 | item_end += |
db94535d | 2741 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2742 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2743 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2744 | fi); |
39279cc3 | 2745 | } |
008630c1 | 2746 | item_end--; |
39279cc3 | 2747 | } |
e02119d5 | 2748 | if (item_end < new_size) { |
d397712b | 2749 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2750 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2751 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2752 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2753 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2754 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2755 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2756 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2757 | else if (found_type) |
b888db2b | 2758 | found_type--; |
d397712b | 2759 | else |
b888db2b | 2760 | break; |
a61721d5 | 2761 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2762 | goto next; |
39279cc3 | 2763 | } |
e02119d5 | 2764 | if (found_key.offset >= new_size) |
39279cc3 CM |
2765 | del_item = 1; |
2766 | else | |
2767 | del_item = 0; | |
2768 | found_extent = 0; | |
2769 | ||
2770 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2771 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2772 | goto delete; | |
2773 | ||
2774 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2775 | u64 num_dec; |
db94535d | 2776 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2777 | if (!del_item && !encoding) { |
db94535d CM |
2778 | u64 orig_num_bytes = |
2779 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2780 | extent_num_bytes = new_size - |
5f39d397 | 2781 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2782 | extent_num_bytes = extent_num_bytes & |
2783 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2784 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2785 | extent_num_bytes); | |
2786 | num_dec = (orig_num_bytes - | |
9069218d | 2787 | extent_num_bytes); |
e02119d5 | 2788 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2789 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2790 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2791 | } else { |
db94535d CM |
2792 | extent_num_bytes = |
2793 | btrfs_file_extent_disk_num_bytes(leaf, | |
2794 | fi); | |
5d4f98a2 YZ |
2795 | extent_offset = found_key.offset - |
2796 | btrfs_file_extent_offset(leaf, fi); | |
2797 | ||
39279cc3 | 2798 | /* FIXME blocksize != 4096 */ |
9069218d | 2799 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2800 | if (extent_start != 0) { |
2801 | found_extent = 1; | |
e02119d5 | 2802 | if (root->ref_cows) |
a76a3cd4 | 2803 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2804 | } |
39279cc3 | 2805 | } |
9069218d | 2806 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2807 | /* |
2808 | * we can't truncate inline items that have had | |
2809 | * special encodings | |
2810 | */ | |
2811 | if (!del_item && | |
2812 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2813 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2814 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2815 | u32 size = new_size - found_key.offset; |
2816 | ||
2817 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2818 | inode_sub_bytes(inode, item_end + 1 - |
2819 | new_size); | |
e02119d5 CM |
2820 | } |
2821 | size = | |
2822 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2823 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2824 | size, 1); |
9069218d | 2825 | BUG_ON(ret); |
e02119d5 | 2826 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2827 | inode_sub_bytes(inode, item_end + 1 - |
2828 | found_key.offset); | |
9069218d | 2829 | } |
39279cc3 | 2830 | } |
179e29e4 | 2831 | delete: |
39279cc3 | 2832 | if (del_item) { |
85e21bac CM |
2833 | if (!pending_del_nr) { |
2834 | /* no pending yet, add ourselves */ | |
2835 | pending_del_slot = path->slots[0]; | |
2836 | pending_del_nr = 1; | |
2837 | } else if (pending_del_nr && | |
2838 | path->slots[0] + 1 == pending_del_slot) { | |
2839 | /* hop on the pending chunk */ | |
2840 | pending_del_nr++; | |
2841 | pending_del_slot = path->slots[0]; | |
2842 | } else { | |
d397712b | 2843 | BUG(); |
85e21bac | 2844 | } |
39279cc3 CM |
2845 | } else { |
2846 | break; | |
2847 | } | |
5d4f98a2 | 2848 | if (found_extent && root->ref_cows) { |
b9473439 | 2849 | btrfs_set_path_blocking(path); |
39279cc3 | 2850 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
2851 | extent_num_bytes, 0, |
2852 | btrfs_header_owner(leaf), | |
2853 | inode->i_ino, extent_offset); | |
39279cc3 CM |
2854 | BUG_ON(ret); |
2855 | } | |
85e21bac CM |
2856 | next: |
2857 | if (path->slots[0] == 0) { | |
2858 | if (pending_del_nr) | |
2859 | goto del_pending; | |
2860 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2861 | if (found_type == BTRFS_INODE_ITEM_KEY) |
2862 | break; | |
85e21bac CM |
2863 | goto search_again; |
2864 | } | |
2865 | ||
2866 | path->slots[0]--; | |
2867 | if (pending_del_nr && | |
2868 | path->slots[0] + 1 != pending_del_slot) { | |
2869 | struct btrfs_key debug; | |
2870 | del_pending: | |
2871 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2872 | pending_del_slot); | |
2873 | ret = btrfs_del_items(trans, root, path, | |
2874 | pending_del_slot, | |
2875 | pending_del_nr); | |
2876 | BUG_ON(ret); | |
2877 | pending_del_nr = 0; | |
2878 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
2879 | if (found_type == BTRFS_INODE_ITEM_KEY) |
2880 | break; | |
85e21bac CM |
2881 | goto search_again; |
2882 | } | |
39279cc3 CM |
2883 | } |
2884 | ret = 0; | |
2885 | error: | |
85e21bac CM |
2886 | if (pending_del_nr) { |
2887 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
2888 | pending_del_nr); | |
2889 | } | |
39279cc3 | 2890 | btrfs_free_path(path); |
39279cc3 CM |
2891 | return ret; |
2892 | } | |
2893 | ||
2894 | /* | |
2895 | * taken from block_truncate_page, but does cow as it zeros out | |
2896 | * any bytes left in the last page in the file. | |
2897 | */ | |
2898 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
2899 | { | |
2900 | struct inode *inode = mapping->host; | |
db94535d | 2901 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
2902 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2903 | struct btrfs_ordered_extent *ordered; | |
2904 | char *kaddr; | |
db94535d | 2905 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
2906 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
2907 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
2908 | struct page *page; | |
39279cc3 | 2909 | int ret = 0; |
a52d9a80 | 2910 | u64 page_start; |
e6dcd2dc | 2911 | u64 page_end; |
39279cc3 CM |
2912 | |
2913 | if ((offset & (blocksize - 1)) == 0) | |
2914 | goto out; | |
2915 | ||
2916 | ret = -ENOMEM; | |
211c17f5 | 2917 | again: |
39279cc3 CM |
2918 | page = grab_cache_page(mapping, index); |
2919 | if (!page) | |
2920 | goto out; | |
e6dcd2dc CM |
2921 | |
2922 | page_start = page_offset(page); | |
2923 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2924 | ||
39279cc3 | 2925 | if (!PageUptodate(page)) { |
9ebefb18 | 2926 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 2927 | lock_page(page); |
211c17f5 CM |
2928 | if (page->mapping != mapping) { |
2929 | unlock_page(page); | |
2930 | page_cache_release(page); | |
2931 | goto again; | |
2932 | } | |
39279cc3 CM |
2933 | if (!PageUptodate(page)) { |
2934 | ret = -EIO; | |
89642229 | 2935 | goto out_unlock; |
39279cc3 CM |
2936 | } |
2937 | } | |
211c17f5 | 2938 | wait_on_page_writeback(page); |
e6dcd2dc CM |
2939 | |
2940 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2941 | set_page_extent_mapped(page); | |
2942 | ||
2943 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
2944 | if (ordered) { | |
2945 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2946 | unlock_page(page); | |
2947 | page_cache_release(page); | |
eb84ae03 | 2948 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
2949 | btrfs_put_ordered_extent(ordered); |
2950 | goto again; | |
2951 | } | |
2952 | ||
ea8c2819 | 2953 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc CM |
2954 | ret = 0; |
2955 | if (offset != PAGE_CACHE_SIZE) { | |
2956 | kaddr = kmap(page); | |
2957 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
2958 | flush_dcache_page(page); | |
2959 | kunmap(page); | |
2960 | } | |
247e743c | 2961 | ClearPageChecked(page); |
e6dcd2dc CM |
2962 | set_page_dirty(page); |
2963 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 2964 | |
89642229 | 2965 | out_unlock: |
39279cc3 CM |
2966 | unlock_page(page); |
2967 | page_cache_release(page); | |
2968 | out: | |
2969 | return ret; | |
2970 | } | |
2971 | ||
9036c102 | 2972 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 2973 | { |
9036c102 YZ |
2974 | struct btrfs_trans_handle *trans; |
2975 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2976 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2977 | struct extent_map *em; | |
2978 | u64 mask = root->sectorsize - 1; | |
2979 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
2980 | u64 block_end = (size + mask) & ~mask; | |
2981 | u64 last_byte; | |
2982 | u64 cur_offset; | |
2983 | u64 hole_size; | |
39279cc3 CM |
2984 | int err; |
2985 | ||
9036c102 YZ |
2986 | if (size <= hole_start) |
2987 | return 0; | |
2988 | ||
6a63209f | 2989 | err = btrfs_check_metadata_free_space(root); |
39279cc3 CM |
2990 | if (err) |
2991 | return err; | |
2992 | ||
9036c102 | 2993 | btrfs_truncate_page(inode->i_mapping, inode->i_size); |
2bf5a725 | 2994 | |
9036c102 YZ |
2995 | while (1) { |
2996 | struct btrfs_ordered_extent *ordered; | |
2997 | btrfs_wait_ordered_range(inode, hole_start, | |
2998 | block_end - hole_start); | |
2999 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3000 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3001 | if (!ordered) | |
3002 | break; | |
3003 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3004 | btrfs_put_ordered_extent(ordered); | |
3005 | } | |
39279cc3 | 3006 | |
9036c102 YZ |
3007 | trans = btrfs_start_transaction(root, 1); |
3008 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3009 | |
9036c102 YZ |
3010 | cur_offset = hole_start; |
3011 | while (1) { | |
3012 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3013 | block_end - cur_offset, 0); | |
3014 | BUG_ON(IS_ERR(em) || !em); | |
3015 | last_byte = min(extent_map_end(em), block_end); | |
3016 | last_byte = (last_byte + mask) & ~mask; | |
3017 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3018 | u64 hint_byte = 0; |
9036c102 | 3019 | hole_size = last_byte - cur_offset; |
771ed689 CM |
3020 | err = btrfs_drop_extents(trans, root, inode, |
3021 | cur_offset, | |
3022 | cur_offset + hole_size, | |
e980b50c | 3023 | block_end, |
a1ed835e | 3024 | cur_offset, &hint_byte, 1); |
771ed689 CM |
3025 | if (err) |
3026 | break; | |
9036c102 YZ |
3027 | err = btrfs_insert_file_extent(trans, root, |
3028 | inode->i_ino, cur_offset, 0, | |
3029 | 0, hole_size, 0, hole_size, | |
3030 | 0, 0, 0); | |
3031 | btrfs_drop_extent_cache(inode, hole_start, | |
3032 | last_byte - 1, 0); | |
3033 | } | |
3034 | free_extent_map(em); | |
3035 | cur_offset = last_byte; | |
3036 | if (err || cur_offset >= block_end) | |
3037 | break; | |
3038 | } | |
1832a6d5 | 3039 | |
9036c102 YZ |
3040 | btrfs_end_transaction(trans, root); |
3041 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3042 | return err; | |
3043 | } | |
39279cc3 | 3044 | |
9036c102 YZ |
3045 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3046 | { | |
3047 | struct inode *inode = dentry->d_inode; | |
3048 | int err; | |
39279cc3 | 3049 | |
9036c102 YZ |
3050 | err = inode_change_ok(inode, attr); |
3051 | if (err) | |
3052 | return err; | |
2bf5a725 | 3053 | |
5a3f23d5 CM |
3054 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3055 | if (attr->ia_size > inode->i_size) { | |
3056 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3057 | if (err) | |
3058 | return err; | |
3059 | } else if (inode->i_size > 0 && | |
3060 | attr->ia_size == 0) { | |
3061 | ||
3062 | /* we're truncating a file that used to have good | |
3063 | * data down to zero. Make sure it gets into | |
3064 | * the ordered flush list so that any new writes | |
3065 | * get down to disk quickly. | |
3066 | */ | |
3067 | BTRFS_I(inode)->ordered_data_close = 1; | |
3068 | } | |
39279cc3 | 3069 | } |
9036c102 | 3070 | |
39279cc3 | 3071 | err = inode_setattr(inode, attr); |
33268eaf JB |
3072 | |
3073 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3074 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3075 | return err; |
3076 | } | |
61295eb8 | 3077 | |
39279cc3 CM |
3078 | void btrfs_delete_inode(struct inode *inode) |
3079 | { | |
3080 | struct btrfs_trans_handle *trans; | |
3081 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3082 | unsigned long nr; |
39279cc3 CM |
3083 | int ret; |
3084 | ||
3085 | truncate_inode_pages(&inode->i_data, 0); | |
3086 | if (is_bad_inode(inode)) { | |
7b128766 | 3087 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3088 | goto no_delete; |
3089 | } | |
4a096752 | 3090 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3091 | |
dbe674a9 | 3092 | btrfs_i_size_write(inode, 0); |
180591bc | 3093 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3094 | |
39279cc3 | 3095 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3096 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3097 | if (ret) { |
3098 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3099 | goto no_delete_lock; |
7b128766 JB |
3100 | } |
3101 | ||
3102 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3103 | |
d3c2fdcf | 3104 | nr = trans->blocks_used; |
85e21bac | 3105 | clear_inode(inode); |
5f39d397 | 3106 | |
39279cc3 | 3107 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3108 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3109 | return; |
54aa1f4d CM |
3110 | |
3111 | no_delete_lock: | |
d3c2fdcf | 3112 | nr = trans->blocks_used; |
54aa1f4d | 3113 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3114 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3115 | no_delete: |
3116 | clear_inode(inode); | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * this returns the key found in the dir entry in the location pointer. | |
3121 | * If no dir entries were found, location->objectid is 0. | |
3122 | */ | |
3123 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3124 | struct btrfs_key *location) | |
3125 | { | |
3126 | const char *name = dentry->d_name.name; | |
3127 | int namelen = dentry->d_name.len; | |
3128 | struct btrfs_dir_item *di; | |
3129 | struct btrfs_path *path; | |
3130 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3131 | int ret = 0; |
39279cc3 CM |
3132 | |
3133 | path = btrfs_alloc_path(); | |
3134 | BUG_ON(!path); | |
3954401f | 3135 | |
39279cc3 CM |
3136 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3137 | namelen, 0); | |
0d9f7f3e Y |
3138 | if (IS_ERR(di)) |
3139 | ret = PTR_ERR(di); | |
d397712b CM |
3140 | |
3141 | if (!di || IS_ERR(di)) | |
3954401f | 3142 | goto out_err; |
d397712b | 3143 | |
5f39d397 | 3144 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3145 | out: |
39279cc3 CM |
3146 | btrfs_free_path(path); |
3147 | return ret; | |
3954401f CM |
3148 | out_err: |
3149 | location->objectid = 0; | |
3150 | goto out; | |
39279cc3 CM |
3151 | } |
3152 | ||
3153 | /* | |
3154 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3155 | * needs to be changed to reflect the root directory of the tree root. This | |
3156 | * is kind of like crossing a mount point. | |
3157 | */ | |
3158 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3159 | struct inode *dir, |
3160 | struct dentry *dentry, | |
3161 | struct btrfs_key *location, | |
3162 | struct btrfs_root **sub_root) | |
39279cc3 | 3163 | { |
4df27c4d YZ |
3164 | struct btrfs_path *path; |
3165 | struct btrfs_root *new_root; | |
3166 | struct btrfs_root_ref *ref; | |
3167 | struct extent_buffer *leaf; | |
3168 | int ret; | |
3169 | int err = 0; | |
39279cc3 | 3170 | |
4df27c4d YZ |
3171 | path = btrfs_alloc_path(); |
3172 | if (!path) { | |
3173 | err = -ENOMEM; | |
3174 | goto out; | |
3175 | } | |
39279cc3 | 3176 | |
4df27c4d YZ |
3177 | err = -ENOENT; |
3178 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3179 | BTRFS_I(dir)->root->root_key.objectid, | |
3180 | location->objectid); | |
3181 | if (ret) { | |
3182 | if (ret < 0) | |
3183 | err = ret; | |
3184 | goto out; | |
3185 | } | |
39279cc3 | 3186 | |
4df27c4d YZ |
3187 | leaf = path->nodes[0]; |
3188 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3189 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3190 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3191 | goto out; | |
39279cc3 | 3192 | |
4df27c4d YZ |
3193 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3194 | (unsigned long)(ref + 1), | |
3195 | dentry->d_name.len); | |
3196 | if (ret) | |
3197 | goto out; | |
3198 | ||
3199 | btrfs_release_path(root->fs_info->tree_root, path); | |
3200 | ||
3201 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3202 | if (IS_ERR(new_root)) { | |
3203 | err = PTR_ERR(new_root); | |
3204 | goto out; | |
3205 | } | |
3206 | ||
3207 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3208 | err = -ENOENT; | |
3209 | goto out; | |
3210 | } | |
3211 | ||
3212 | *sub_root = new_root; | |
3213 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3214 | location->type = BTRFS_INODE_ITEM_KEY; | |
3215 | location->offset = 0; | |
3216 | err = 0; | |
3217 | out: | |
3218 | btrfs_free_path(path); | |
3219 | return err; | |
39279cc3 CM |
3220 | } |
3221 | ||
5d4f98a2 YZ |
3222 | static void inode_tree_add(struct inode *inode) |
3223 | { | |
3224 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3225 | struct btrfs_inode *entry; | |
03e860bd FNP |
3226 | struct rb_node **p; |
3227 | struct rb_node *parent; | |
3228 | ||
3229 | again: | |
3230 | p = &root->inode_tree.rb_node; | |
3231 | parent = NULL; | |
5d4f98a2 YZ |
3232 | |
3233 | spin_lock(&root->inode_lock); | |
3234 | while (*p) { | |
3235 | parent = *p; | |
3236 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3237 | ||
3238 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3239 | p = &parent->rb_left; |
5d4f98a2 | 3240 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3241 | p = &parent->rb_right; |
5d4f98a2 YZ |
3242 | else { |
3243 | WARN_ON(!(entry->vfs_inode.i_state & | |
3244 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd FNP |
3245 | rb_erase(parent, &root->inode_tree); |
3246 | RB_CLEAR_NODE(parent); | |
3247 | spin_unlock(&root->inode_lock); | |
3248 | goto again; | |
5d4f98a2 YZ |
3249 | } |
3250 | } | |
3251 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3252 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3253 | spin_unlock(&root->inode_lock); | |
3254 | } | |
3255 | ||
3256 | static void inode_tree_del(struct inode *inode) | |
3257 | { | |
3258 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3259 | ||
03e860bd | 3260 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3261 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3262 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 YZ |
3263 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
3264 | } | |
03e860bd | 3265 | spin_unlock(&root->inode_lock); |
5d4f98a2 YZ |
3266 | } |
3267 | ||
e02119d5 | 3268 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3269 | { |
e02119d5 CM |
3270 | struct btrfs_inode *bi = BTRFS_I(inode); |
3271 | ||
e02119d5 | 3272 | bi->generation = 0; |
c3027eb5 | 3273 | bi->sequence = 0; |
e02119d5 CM |
3274 | bi->last_trans = 0; |
3275 | bi->logged_trans = 0; | |
3276 | bi->delalloc_bytes = 0; | |
6a63209f | 3277 | bi->reserved_bytes = 0; |
e02119d5 CM |
3278 | bi->disk_i_size = 0; |
3279 | bi->flags = 0; | |
3280 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3281 | bi->last_unlink_trans = 0; |
2757495c | 3282 | bi->ordered_data_close = 0; |
d1310b2e CM |
3283 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3284 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3285 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3286 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3287 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3288 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3289 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3290 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3291 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
ee6e6504 | 3292 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
3293 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3294 | } | |
3295 | ||
3296 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3297 | { | |
3298 | struct btrfs_iget_args *args = p; | |
3299 | inode->i_ino = args->ino; | |
3300 | init_btrfs_i(inode); | |
3301 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3302 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3303 | return 0; |
3304 | } | |
3305 | ||
3306 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3307 | { | |
3308 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3309 | return args->ino == inode->i_ino && |
3310 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3311 | } |
3312 | ||
5d4f98a2 YZ |
3313 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3314 | u64 objectid, | |
3315 | struct btrfs_root *root) | |
39279cc3 CM |
3316 | { |
3317 | struct inode *inode; | |
3318 | struct btrfs_iget_args args; | |
3319 | args.ino = objectid; | |
3320 | args.root = root; | |
3321 | ||
3322 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3323 | btrfs_init_locked_inode, | |
3324 | (void *)&args); | |
3325 | return inode; | |
3326 | } | |
3327 | ||
1a54ef8c BR |
3328 | /* Get an inode object given its location and corresponding root. |
3329 | * Returns in *is_new if the inode was read from disk | |
3330 | */ | |
3331 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3332 | struct btrfs_root *root) |
1a54ef8c BR |
3333 | { |
3334 | struct inode *inode; | |
3335 | ||
3336 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3337 | if (!inode) | |
5d4f98a2 | 3338 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3339 | |
3340 | if (inode->i_state & I_NEW) { | |
3341 | BTRFS_I(inode)->root = root; | |
3342 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3343 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3344 | |
3345 | inode_tree_add(inode); | |
1a54ef8c | 3346 | unlock_new_inode(inode); |
1a54ef8c BR |
3347 | } |
3348 | ||
3349 | return inode; | |
3350 | } | |
3351 | ||
4df27c4d YZ |
3352 | static struct inode *new_simple_dir(struct super_block *s, |
3353 | struct btrfs_key *key, | |
3354 | struct btrfs_root *root) | |
3355 | { | |
3356 | struct inode *inode = new_inode(s); | |
3357 | ||
3358 | if (!inode) | |
3359 | return ERR_PTR(-ENOMEM); | |
3360 | ||
3361 | init_btrfs_i(inode); | |
3362 | ||
3363 | BTRFS_I(inode)->root = root; | |
3364 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3365 | BTRFS_I(inode)->dummy_inode = 1; | |
3366 | ||
3367 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3368 | inode->i_op = &simple_dir_inode_operations; | |
3369 | inode->i_fop = &simple_dir_operations; | |
3370 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3371 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3372 | ||
3373 | return inode; | |
3374 | } | |
3375 | ||
3de4586c | 3376 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3377 | { |
d397712b | 3378 | struct inode *inode; |
4df27c4d | 3379 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3380 | struct btrfs_root *sub_root = root; |
3381 | struct btrfs_key location; | |
5d4f98a2 | 3382 | int ret; |
39279cc3 CM |
3383 | |
3384 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3385 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3386 | |
39279cc3 | 3387 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3388 | |
39279cc3 CM |
3389 | if (ret < 0) |
3390 | return ERR_PTR(ret); | |
5f39d397 | 3391 | |
4df27c4d YZ |
3392 | if (location.objectid == 0) |
3393 | return NULL; | |
3394 | ||
3395 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3396 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3397 | return inode; | |
3398 | } | |
3399 | ||
3400 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3401 | ||
3402 | ret = fixup_tree_root_location(root, dir, dentry, | |
3403 | &location, &sub_root); | |
3404 | if (ret < 0) { | |
3405 | if (ret != -ENOENT) | |
3406 | inode = ERR_PTR(ret); | |
3407 | else | |
3408 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3409 | } else { | |
5d4f98a2 | 3410 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3411 | } |
3de4586c CM |
3412 | return inode; |
3413 | } | |
3414 | ||
3415 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, | |
3416 | struct nameidata *nd) | |
3417 | { | |
3418 | struct inode *inode; | |
3419 | ||
3de4586c CM |
3420 | inode = btrfs_lookup_dentry(dir, dentry); |
3421 | if (IS_ERR(inode)) | |
3422 | return ERR_CAST(inode); | |
7b128766 | 3423 | |
39279cc3 CM |
3424 | return d_splice_alias(inode, dentry); |
3425 | } | |
3426 | ||
39279cc3 CM |
3427 | static unsigned char btrfs_filetype_table[] = { |
3428 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3429 | }; | |
3430 | ||
cbdf5a24 DW |
3431 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3432 | filldir_t filldir) | |
39279cc3 | 3433 | { |
6da6abae | 3434 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3435 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3436 | struct btrfs_item *item; | |
3437 | struct btrfs_dir_item *di; | |
3438 | struct btrfs_key key; | |
5f39d397 | 3439 | struct btrfs_key found_key; |
39279cc3 CM |
3440 | struct btrfs_path *path; |
3441 | int ret; | |
3442 | u32 nritems; | |
5f39d397 | 3443 | struct extent_buffer *leaf; |
39279cc3 CM |
3444 | int slot; |
3445 | int advance; | |
3446 | unsigned char d_type; | |
3447 | int over = 0; | |
3448 | u32 di_cur; | |
3449 | u32 di_total; | |
3450 | u32 di_len; | |
3451 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3452 | char tmp_name[32]; |
3453 | char *name_ptr; | |
3454 | int name_len; | |
39279cc3 CM |
3455 | |
3456 | /* FIXME, use a real flag for deciding about the key type */ | |
3457 | if (root->fs_info->tree_root == root) | |
3458 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3459 | |
3954401f CM |
3460 | /* special case for "." */ |
3461 | if (filp->f_pos == 0) { | |
3462 | over = filldir(dirent, ".", 1, | |
3463 | 1, inode->i_ino, | |
3464 | DT_DIR); | |
3465 | if (over) | |
3466 | return 0; | |
3467 | filp->f_pos = 1; | |
3468 | } | |
3954401f CM |
3469 | /* special case for .., just use the back ref */ |
3470 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3471 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3472 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3473 | 2, pino, DT_DIR); |
3954401f | 3474 | if (over) |
49593bfa | 3475 | return 0; |
3954401f CM |
3476 | filp->f_pos = 2; |
3477 | } | |
49593bfa DW |
3478 | path = btrfs_alloc_path(); |
3479 | path->reada = 2; | |
3480 | ||
39279cc3 CM |
3481 | btrfs_set_key_type(&key, key_type); |
3482 | key.offset = filp->f_pos; | |
49593bfa | 3483 | key.objectid = inode->i_ino; |
5f39d397 | 3484 | |
39279cc3 CM |
3485 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3486 | if (ret < 0) | |
3487 | goto err; | |
3488 | advance = 0; | |
49593bfa DW |
3489 | |
3490 | while (1) { | |
5f39d397 CM |
3491 | leaf = path->nodes[0]; |
3492 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3493 | slot = path->slots[0]; |
3494 | if (advance || slot >= nritems) { | |
49593bfa | 3495 | if (slot >= nritems - 1) { |
39279cc3 CM |
3496 | ret = btrfs_next_leaf(root, path); |
3497 | if (ret) | |
3498 | break; | |
5f39d397 CM |
3499 | leaf = path->nodes[0]; |
3500 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3501 | slot = path->slots[0]; |
3502 | } else { | |
3503 | slot++; | |
3504 | path->slots[0]++; | |
3505 | } | |
3506 | } | |
3de4586c | 3507 | |
39279cc3 | 3508 | advance = 1; |
5f39d397 CM |
3509 | item = btrfs_item_nr(leaf, slot); |
3510 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3511 | ||
3512 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3513 | break; |
5f39d397 | 3514 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3515 | break; |
5f39d397 | 3516 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3517 | continue; |
5f39d397 CM |
3518 | |
3519 | filp->f_pos = found_key.offset; | |
49593bfa | 3520 | |
39279cc3 CM |
3521 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3522 | di_cur = 0; | |
5f39d397 | 3523 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3524 | |
3525 | while (di_cur < di_total) { | |
5f39d397 CM |
3526 | struct btrfs_key location; |
3527 | ||
3528 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3529 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3530 | name_ptr = tmp_name; |
3531 | } else { | |
3532 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3533 | if (!name_ptr) { |
3534 | ret = -ENOMEM; | |
3535 | goto err; | |
3536 | } | |
5f39d397 CM |
3537 | } |
3538 | read_extent_buffer(leaf, name_ptr, | |
3539 | (unsigned long)(di + 1), name_len); | |
3540 | ||
3541 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3542 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3543 | |
3544 | /* is this a reference to our own snapshot? If so | |
3545 | * skip it | |
3546 | */ | |
3547 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3548 | location.objectid == root->root_key.objectid) { | |
3549 | over = 0; | |
3550 | goto skip; | |
3551 | } | |
5f39d397 | 3552 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3553 | found_key.offset, location.objectid, |
39279cc3 | 3554 | d_type); |
5f39d397 | 3555 | |
3de4586c | 3556 | skip: |
5f39d397 CM |
3557 | if (name_ptr != tmp_name) |
3558 | kfree(name_ptr); | |
3559 | ||
39279cc3 CM |
3560 | if (over) |
3561 | goto nopos; | |
5103e947 | 3562 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3563 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3564 | di_cur += di_len; |
3565 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3566 | } | |
3567 | } | |
49593bfa DW |
3568 | |
3569 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3570 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3571 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3572 | else |
3573 | filp->f_pos++; | |
39279cc3 CM |
3574 | nopos: |
3575 | ret = 0; | |
3576 | err: | |
39279cc3 | 3577 | btrfs_free_path(path); |
39279cc3 CM |
3578 | return ret; |
3579 | } | |
3580 | ||
3581 | int btrfs_write_inode(struct inode *inode, int wait) | |
3582 | { | |
3583 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3584 | struct btrfs_trans_handle *trans; | |
3585 | int ret = 0; | |
3586 | ||
c146afad | 3587 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3588 | return 0; |
3589 | ||
39279cc3 | 3590 | if (wait) { |
f9295749 | 3591 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3592 | btrfs_set_trans_block_group(trans, inode); |
3593 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3594 | } |
3595 | return ret; | |
3596 | } | |
3597 | ||
3598 | /* | |
54aa1f4d | 3599 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3600 | * inode changes. But, it is most likely to find the inode in cache. |
3601 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3602 | * to keep or drop this code. | |
3603 | */ | |
3604 | void btrfs_dirty_inode(struct inode *inode) | |
3605 | { | |
3606 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3607 | struct btrfs_trans_handle *trans; | |
3608 | ||
f9295749 | 3609 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3610 | btrfs_set_trans_block_group(trans, inode); |
3611 | btrfs_update_inode(trans, root, inode); | |
3612 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3613 | } |
3614 | ||
d352ac68 CM |
3615 | /* |
3616 | * find the highest existing sequence number in a directory | |
3617 | * and then set the in-memory index_cnt variable to reflect | |
3618 | * free sequence numbers | |
3619 | */ | |
aec7477b JB |
3620 | static int btrfs_set_inode_index_count(struct inode *inode) |
3621 | { | |
3622 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3623 | struct btrfs_key key, found_key; | |
3624 | struct btrfs_path *path; | |
3625 | struct extent_buffer *leaf; | |
3626 | int ret; | |
3627 | ||
3628 | key.objectid = inode->i_ino; | |
3629 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3630 | key.offset = (u64)-1; | |
3631 | ||
3632 | path = btrfs_alloc_path(); | |
3633 | if (!path) | |
3634 | return -ENOMEM; | |
3635 | ||
3636 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3637 | if (ret < 0) | |
3638 | goto out; | |
3639 | /* FIXME: we should be able to handle this */ | |
3640 | if (ret == 0) | |
3641 | goto out; | |
3642 | ret = 0; | |
3643 | ||
3644 | /* | |
3645 | * MAGIC NUMBER EXPLANATION: | |
3646 | * since we search a directory based on f_pos we have to start at 2 | |
3647 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3648 | * else has to start at 2 | |
3649 | */ | |
3650 | if (path->slots[0] == 0) { | |
3651 | BTRFS_I(inode)->index_cnt = 2; | |
3652 | goto out; | |
3653 | } | |
3654 | ||
3655 | path->slots[0]--; | |
3656 | ||
3657 | leaf = path->nodes[0]; | |
3658 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3659 | ||
3660 | if (found_key.objectid != inode->i_ino || | |
3661 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3662 | BTRFS_I(inode)->index_cnt = 2; | |
3663 | goto out; | |
3664 | } | |
3665 | ||
3666 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3667 | out: | |
3668 | btrfs_free_path(path); | |
3669 | return ret; | |
3670 | } | |
3671 | ||
d352ac68 CM |
3672 | /* |
3673 | * helper to find a free sequence number in a given directory. This current | |
3674 | * code is very simple, later versions will do smarter things in the btree | |
3675 | */ | |
3de4586c | 3676 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3677 | { |
3678 | int ret = 0; | |
3679 | ||
3680 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3681 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3682 | if (ret) |
aec7477b JB |
3683 | return ret; |
3684 | } | |
3685 | ||
00e4e6b3 | 3686 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3687 | BTRFS_I(dir)->index_cnt++; |
3688 | ||
3689 | return ret; | |
3690 | } | |
3691 | ||
39279cc3 CM |
3692 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3693 | struct btrfs_root *root, | |
aec7477b | 3694 | struct inode *dir, |
9c58309d | 3695 | const char *name, int name_len, |
d2fb3437 YZ |
3696 | u64 ref_objectid, u64 objectid, |
3697 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
3698 | { |
3699 | struct inode *inode; | |
5f39d397 | 3700 | struct btrfs_inode_item *inode_item; |
39279cc3 | 3701 | struct btrfs_key *location; |
5f39d397 | 3702 | struct btrfs_path *path; |
9c58309d CM |
3703 | struct btrfs_inode_ref *ref; |
3704 | struct btrfs_key key[2]; | |
3705 | u32 sizes[2]; | |
3706 | unsigned long ptr; | |
39279cc3 CM |
3707 | int ret; |
3708 | int owner; | |
3709 | ||
5f39d397 CM |
3710 | path = btrfs_alloc_path(); |
3711 | BUG_ON(!path); | |
3712 | ||
39279cc3 CM |
3713 | inode = new_inode(root->fs_info->sb); |
3714 | if (!inode) | |
3715 | return ERR_PTR(-ENOMEM); | |
3716 | ||
aec7477b | 3717 | if (dir) { |
3de4586c | 3718 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
3719 | if (ret) { |
3720 | iput(inode); | |
aec7477b | 3721 | return ERR_PTR(ret); |
09771430 | 3722 | } |
aec7477b JB |
3723 | } |
3724 | /* | |
3725 | * index_cnt is ignored for everything but a dir, | |
3726 | * btrfs_get_inode_index_count has an explanation for the magic | |
3727 | * number | |
3728 | */ | |
e02119d5 | 3729 | init_btrfs_i(inode); |
aec7477b | 3730 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3731 | BTRFS_I(inode)->root = root; |
e02119d5 | 3732 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 3733 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 3734 | |
39279cc3 CM |
3735 | if (mode & S_IFDIR) |
3736 | owner = 0; | |
3737 | else | |
3738 | owner = 1; | |
d2fb3437 YZ |
3739 | BTRFS_I(inode)->block_group = |
3740 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
3741 | |
3742 | key[0].objectid = objectid; | |
3743 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3744 | key[0].offset = 0; | |
3745 | ||
3746 | key[1].objectid = objectid; | |
3747 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3748 | key[1].offset = ref_objectid; | |
3749 | ||
3750 | sizes[0] = sizeof(struct btrfs_inode_item); | |
3751 | sizes[1] = name_len + sizeof(*ref); | |
3752 | ||
b9473439 | 3753 | path->leave_spinning = 1; |
9c58309d CM |
3754 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
3755 | if (ret != 0) | |
5f39d397 CM |
3756 | goto fail; |
3757 | ||
79683f2d | 3758 | inode->i_uid = current_fsuid(); |
8c087b51 | 3759 | |
42f15d77 | 3760 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
3761 | inode->i_gid = dir->i_gid; |
3762 | if (S_ISDIR(mode)) | |
3763 | mode |= S_ISGID; | |
3764 | } else | |
3765 | inode->i_gid = current_fsgid(); | |
3766 | ||
39279cc3 CM |
3767 | inode->i_mode = mode; |
3768 | inode->i_ino = objectid; | |
a76a3cd4 | 3769 | inode_set_bytes(inode, 0); |
39279cc3 | 3770 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
3771 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3772 | struct btrfs_inode_item); | |
e02119d5 | 3773 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
3774 | |
3775 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3776 | struct btrfs_inode_ref); | |
3777 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 3778 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
3779 | ptr = (unsigned long)(ref + 1); |
3780 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
3781 | ||
5f39d397 CM |
3782 | btrfs_mark_buffer_dirty(path->nodes[0]); |
3783 | btrfs_free_path(path); | |
3784 | ||
39279cc3 CM |
3785 | location = &BTRFS_I(inode)->location; |
3786 | location->objectid = objectid; | |
39279cc3 CM |
3787 | location->offset = 0; |
3788 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
3789 | ||
6cbff00f CH |
3790 | btrfs_inherit_iflags(inode, dir); |
3791 | ||
94272164 CM |
3792 | if ((mode & S_IFREG)) { |
3793 | if (btrfs_test_opt(root, NODATASUM)) | |
3794 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
3795 | if (btrfs_test_opt(root, NODATACOW)) | |
3796 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
3797 | } | |
3798 | ||
39279cc3 | 3799 | insert_inode_hash(inode); |
5d4f98a2 | 3800 | inode_tree_add(inode); |
39279cc3 | 3801 | return inode; |
5f39d397 | 3802 | fail: |
aec7477b JB |
3803 | if (dir) |
3804 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 3805 | btrfs_free_path(path); |
09771430 | 3806 | iput(inode); |
5f39d397 | 3807 | return ERR_PTR(ret); |
39279cc3 CM |
3808 | } |
3809 | ||
3810 | static inline u8 btrfs_inode_type(struct inode *inode) | |
3811 | { | |
3812 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
3813 | } | |
3814 | ||
d352ac68 CM |
3815 | /* |
3816 | * utility function to add 'inode' into 'parent_inode' with | |
3817 | * a give name and a given sequence number. | |
3818 | * if 'add_backref' is true, also insert a backref from the | |
3819 | * inode to the parent directory. | |
3820 | */ | |
e02119d5 CM |
3821 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
3822 | struct inode *parent_inode, struct inode *inode, | |
3823 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 3824 | { |
4df27c4d | 3825 | int ret = 0; |
39279cc3 | 3826 | struct btrfs_key key; |
e02119d5 | 3827 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 3828 | |
4df27c4d YZ |
3829 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
3830 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
3831 | } else { | |
3832 | key.objectid = inode->i_ino; | |
3833 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
3834 | key.offset = 0; | |
3835 | } | |
3836 | ||
3837 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
3838 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
3839 | key.objectid, root->root_key.objectid, | |
3840 | parent_inode->i_ino, | |
3841 | index, name, name_len); | |
3842 | } else if (add_backref) { | |
3843 | ret = btrfs_insert_inode_ref(trans, root, | |
3844 | name, name_len, inode->i_ino, | |
3845 | parent_inode->i_ino, index); | |
3846 | } | |
39279cc3 | 3847 | |
39279cc3 | 3848 | if (ret == 0) { |
4df27c4d YZ |
3849 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
3850 | parent_inode->i_ino, &key, | |
3851 | btrfs_inode_type(inode), index); | |
3852 | BUG_ON(ret); | |
3853 | ||
dbe674a9 | 3854 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 3855 | name_len * 2); |
79c44584 | 3856 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 3857 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
3858 | } |
3859 | return ret; | |
3860 | } | |
3861 | ||
3862 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 3863 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 3864 | int backref, u64 index) |
39279cc3 | 3865 | { |
e02119d5 CM |
3866 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3867 | inode, dentry->d_name.name, | |
3868 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
3869 | if (!err) { |
3870 | d_instantiate(dentry, inode); | |
3871 | return 0; | |
3872 | } | |
3873 | if (err > 0) | |
3874 | err = -EEXIST; | |
3875 | return err; | |
3876 | } | |
3877 | ||
618e21d5 JB |
3878 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
3879 | int mode, dev_t rdev) | |
3880 | { | |
3881 | struct btrfs_trans_handle *trans; | |
3882 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3883 | struct inode *inode = NULL; |
618e21d5 JB |
3884 | int err; |
3885 | int drop_inode = 0; | |
3886 | u64 objectid; | |
1832a6d5 | 3887 | unsigned long nr = 0; |
00e4e6b3 | 3888 | u64 index = 0; |
618e21d5 JB |
3889 | |
3890 | if (!new_valid_dev(rdev)) | |
3891 | return -EINVAL; | |
3892 | ||
6a63209f | 3893 | err = btrfs_check_metadata_free_space(root); |
1832a6d5 CM |
3894 | if (err) |
3895 | goto fail; | |
3896 | ||
618e21d5 JB |
3897 | trans = btrfs_start_transaction(root, 1); |
3898 | btrfs_set_trans_block_group(trans, dir); | |
3899 | ||
3900 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3901 | if (err) { | |
3902 | err = -ENOSPC; | |
3903 | goto out_unlock; | |
3904 | } | |
3905 | ||
aec7477b | 3906 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3907 | dentry->d_name.len, |
3908 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 3909 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
3910 | err = PTR_ERR(inode); |
3911 | if (IS_ERR(inode)) | |
3912 | goto out_unlock; | |
3913 | ||
0279b4cd | 3914 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
3915 | if (err) { |
3916 | drop_inode = 1; | |
3917 | goto out_unlock; | |
3918 | } | |
3919 | ||
618e21d5 | 3920 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3921 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
3922 | if (err) |
3923 | drop_inode = 1; | |
3924 | else { | |
3925 | inode->i_op = &btrfs_special_inode_operations; | |
3926 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 3927 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 3928 | } |
618e21d5 JB |
3929 | btrfs_update_inode_block_group(trans, inode); |
3930 | btrfs_update_inode_block_group(trans, dir); | |
3931 | out_unlock: | |
d3c2fdcf | 3932 | nr = trans->blocks_used; |
89ce8a63 | 3933 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3934 | fail: |
618e21d5 JB |
3935 | if (drop_inode) { |
3936 | inode_dec_link_count(inode); | |
3937 | iput(inode); | |
3938 | } | |
d3c2fdcf | 3939 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
3940 | return err; |
3941 | } | |
3942 | ||
39279cc3 CM |
3943 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
3944 | int mode, struct nameidata *nd) | |
3945 | { | |
3946 | struct btrfs_trans_handle *trans; | |
3947 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3948 | struct inode *inode = NULL; |
39279cc3 CM |
3949 | int err; |
3950 | int drop_inode = 0; | |
1832a6d5 | 3951 | unsigned long nr = 0; |
39279cc3 | 3952 | u64 objectid; |
00e4e6b3 | 3953 | u64 index = 0; |
39279cc3 | 3954 | |
6a63209f | 3955 | err = btrfs_check_metadata_free_space(root); |
1832a6d5 CM |
3956 | if (err) |
3957 | goto fail; | |
39279cc3 CM |
3958 | trans = btrfs_start_transaction(root, 1); |
3959 | btrfs_set_trans_block_group(trans, dir); | |
3960 | ||
3961 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3962 | if (err) { | |
3963 | err = -ENOSPC; | |
3964 | goto out_unlock; | |
3965 | } | |
3966 | ||
aec7477b | 3967 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3968 | dentry->d_name.len, |
3969 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
3970 | objectid, BTRFS_I(dir)->block_group, mode, |
3971 | &index); | |
39279cc3 CM |
3972 | err = PTR_ERR(inode); |
3973 | if (IS_ERR(inode)) | |
3974 | goto out_unlock; | |
3975 | ||
0279b4cd | 3976 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
3977 | if (err) { |
3978 | drop_inode = 1; | |
3979 | goto out_unlock; | |
3980 | } | |
3981 | ||
39279cc3 | 3982 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3983 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
3984 | if (err) |
3985 | drop_inode = 1; | |
3986 | else { | |
3987 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3988 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
3989 | inode->i_fop = &btrfs_file_operations; |
3990 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 3991 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 3992 | } |
39279cc3 CM |
3993 | btrfs_update_inode_block_group(trans, inode); |
3994 | btrfs_update_inode_block_group(trans, dir); | |
3995 | out_unlock: | |
d3c2fdcf | 3996 | nr = trans->blocks_used; |
ab78c84d | 3997 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3998 | fail: |
39279cc3 CM |
3999 | if (drop_inode) { |
4000 | inode_dec_link_count(inode); | |
4001 | iput(inode); | |
4002 | } | |
d3c2fdcf | 4003 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4004 | return err; |
4005 | } | |
4006 | ||
4007 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4008 | struct dentry *dentry) | |
4009 | { | |
4010 | struct btrfs_trans_handle *trans; | |
4011 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4012 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4013 | u64 index; |
1832a6d5 | 4014 | unsigned long nr = 0; |
39279cc3 CM |
4015 | int err; |
4016 | int drop_inode = 0; | |
4017 | ||
4018 | if (inode->i_nlink == 0) | |
4019 | return -ENOENT; | |
4020 | ||
e02119d5 | 4021 | btrfs_inc_nlink(inode); |
6a63209f | 4022 | err = btrfs_check_metadata_free_space(root); |
1832a6d5 CM |
4023 | if (err) |
4024 | goto fail; | |
3de4586c | 4025 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4026 | if (err) |
4027 | goto fail; | |
4028 | ||
39279cc3 | 4029 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4030 | |
39279cc3 CM |
4031 | btrfs_set_trans_block_group(trans, dir); |
4032 | atomic_inc(&inode->i_count); | |
aec7477b | 4033 | |
00e4e6b3 | 4034 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4035 | |
39279cc3 CM |
4036 | if (err) |
4037 | drop_inode = 1; | |
5f39d397 | 4038 | |
39279cc3 | 4039 | btrfs_update_inode_block_group(trans, dir); |
54aa1f4d | 4040 | err = btrfs_update_inode(trans, root, inode); |
5f39d397 | 4041 | |
54aa1f4d CM |
4042 | if (err) |
4043 | drop_inode = 1; | |
39279cc3 | 4044 | |
d3c2fdcf | 4045 | nr = trans->blocks_used; |
12fcfd22 CM |
4046 | |
4047 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
ab78c84d | 4048 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4049 | fail: |
39279cc3 CM |
4050 | if (drop_inode) { |
4051 | inode_dec_link_count(inode); | |
4052 | iput(inode); | |
4053 | } | |
d3c2fdcf | 4054 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4055 | return err; |
4056 | } | |
4057 | ||
39279cc3 CM |
4058 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4059 | { | |
b9d86667 | 4060 | struct inode *inode = NULL; |
39279cc3 CM |
4061 | struct btrfs_trans_handle *trans; |
4062 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4063 | int err = 0; | |
4064 | int drop_on_err = 0; | |
b9d86667 | 4065 | u64 objectid = 0; |
00e4e6b3 | 4066 | u64 index = 0; |
d3c2fdcf | 4067 | unsigned long nr = 1; |
39279cc3 | 4068 | |
6a63209f | 4069 | err = btrfs_check_metadata_free_space(root); |
1832a6d5 CM |
4070 | if (err) |
4071 | goto out_unlock; | |
4072 | ||
39279cc3 CM |
4073 | trans = btrfs_start_transaction(root, 1); |
4074 | btrfs_set_trans_block_group(trans, dir); | |
5f39d397 | 4075 | |
39279cc3 CM |
4076 | if (IS_ERR(trans)) { |
4077 | err = PTR_ERR(trans); | |
4078 | goto out_unlock; | |
4079 | } | |
4080 | ||
4081 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4082 | if (err) { | |
4083 | err = -ENOSPC; | |
4084 | goto out_unlock; | |
4085 | } | |
4086 | ||
aec7477b | 4087 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4088 | dentry->d_name.len, |
4089 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4090 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4091 | &index); | |
39279cc3 CM |
4092 | if (IS_ERR(inode)) { |
4093 | err = PTR_ERR(inode); | |
4094 | goto out_fail; | |
4095 | } | |
5f39d397 | 4096 | |
39279cc3 | 4097 | drop_on_err = 1; |
33268eaf | 4098 | |
0279b4cd | 4099 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4100 | if (err) |
4101 | goto out_fail; | |
4102 | ||
39279cc3 CM |
4103 | inode->i_op = &btrfs_dir_inode_operations; |
4104 | inode->i_fop = &btrfs_dir_file_operations; | |
4105 | btrfs_set_trans_block_group(trans, inode); | |
4106 | ||
dbe674a9 | 4107 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4108 | err = btrfs_update_inode(trans, root, inode); |
4109 | if (err) | |
4110 | goto out_fail; | |
5f39d397 | 4111 | |
e02119d5 CM |
4112 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4113 | inode, dentry->d_name.name, | |
4114 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4115 | if (err) |
4116 | goto out_fail; | |
5f39d397 | 4117 | |
39279cc3 CM |
4118 | d_instantiate(dentry, inode); |
4119 | drop_on_err = 0; | |
39279cc3 CM |
4120 | btrfs_update_inode_block_group(trans, inode); |
4121 | btrfs_update_inode_block_group(trans, dir); | |
4122 | ||
4123 | out_fail: | |
d3c2fdcf | 4124 | nr = trans->blocks_used; |
ab78c84d | 4125 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4126 | |
39279cc3 | 4127 | out_unlock: |
39279cc3 CM |
4128 | if (drop_on_err) |
4129 | iput(inode); | |
d3c2fdcf | 4130 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4131 | return err; |
4132 | } | |
4133 | ||
d352ac68 CM |
4134 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4135 | * and an extent that you want to insert, deal with overlap and insert | |
4136 | * the new extent into the tree. | |
4137 | */ | |
3b951516 CM |
4138 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4139 | struct extent_map *existing, | |
e6dcd2dc CM |
4140 | struct extent_map *em, |
4141 | u64 map_start, u64 map_len) | |
3b951516 CM |
4142 | { |
4143 | u64 start_diff; | |
3b951516 | 4144 | |
e6dcd2dc CM |
4145 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4146 | start_diff = map_start - em->start; | |
4147 | em->start = map_start; | |
4148 | em->len = map_len; | |
c8b97818 CM |
4149 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4150 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4151 | em->block_start += start_diff; |
c8b97818 CM |
4152 | em->block_len -= start_diff; |
4153 | } | |
e6dcd2dc | 4154 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4155 | } |
4156 | ||
c8b97818 CM |
4157 | static noinline int uncompress_inline(struct btrfs_path *path, |
4158 | struct inode *inode, struct page *page, | |
4159 | size_t pg_offset, u64 extent_offset, | |
4160 | struct btrfs_file_extent_item *item) | |
4161 | { | |
4162 | int ret; | |
4163 | struct extent_buffer *leaf = path->nodes[0]; | |
4164 | char *tmp; | |
4165 | size_t max_size; | |
4166 | unsigned long inline_size; | |
4167 | unsigned long ptr; | |
4168 | ||
4169 | WARN_ON(pg_offset != 0); | |
4170 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4171 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4172 | btrfs_item_nr(leaf, path->slots[0])); | |
4173 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4174 | ptr = btrfs_file_extent_inline_start(item); | |
4175 | ||
4176 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4177 | ||
5b050f04 | 4178 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4179 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4180 | inline_size, max_size); | |
4181 | if (ret) { | |
4182 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4183 | unsigned long copy_size = min_t(u64, | |
4184 | PAGE_CACHE_SIZE - pg_offset, | |
4185 | max_size - extent_offset); | |
4186 | memset(kaddr + pg_offset, 0, copy_size); | |
4187 | kunmap_atomic(kaddr, KM_USER0); | |
4188 | } | |
4189 | kfree(tmp); | |
4190 | return 0; | |
4191 | } | |
4192 | ||
d352ac68 CM |
4193 | /* |
4194 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4195 | * the ugly parts come from merging extents from the disk with the in-ram |
4196 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4197 | * where the in-ram extents might be locked pending data=ordered completion. |
4198 | * | |
4199 | * This also copies inline extents directly into the page. | |
4200 | */ | |
d397712b | 4201 | |
a52d9a80 | 4202 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4203 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4204 | int create) |
4205 | { | |
4206 | int ret; | |
4207 | int err = 0; | |
db94535d | 4208 | u64 bytenr; |
a52d9a80 CM |
4209 | u64 extent_start = 0; |
4210 | u64 extent_end = 0; | |
4211 | u64 objectid = inode->i_ino; | |
4212 | u32 found_type; | |
f421950f | 4213 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4214 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4215 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4216 | struct extent_buffer *leaf; |
4217 | struct btrfs_key found_key; | |
a52d9a80 CM |
4218 | struct extent_map *em = NULL; |
4219 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4220 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4221 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4222 | int compressed; |
a52d9a80 | 4223 | |
a52d9a80 | 4224 | again: |
890871be | 4225 | read_lock(&em_tree->lock); |
d1310b2e | 4226 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4227 | if (em) |
4228 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4229 | read_unlock(&em_tree->lock); |
d1310b2e | 4230 | |
a52d9a80 | 4231 | if (em) { |
e1c4b745 CM |
4232 | if (em->start > start || em->start + em->len <= start) |
4233 | free_extent_map(em); | |
4234 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4235 | free_extent_map(em); |
4236 | else | |
4237 | goto out; | |
a52d9a80 | 4238 | } |
d1310b2e | 4239 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4240 | if (!em) { |
d1310b2e CM |
4241 | err = -ENOMEM; |
4242 | goto out; | |
a52d9a80 | 4243 | } |
e6dcd2dc | 4244 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4245 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4246 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4247 | em->len = (u64)-1; |
c8b97818 | 4248 | em->block_len = (u64)-1; |
f421950f CM |
4249 | |
4250 | if (!path) { | |
4251 | path = btrfs_alloc_path(); | |
4252 | BUG_ON(!path); | |
4253 | } | |
4254 | ||
179e29e4 CM |
4255 | ret = btrfs_lookup_file_extent(trans, root, path, |
4256 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4257 | if (ret < 0) { |
4258 | err = ret; | |
4259 | goto out; | |
4260 | } | |
4261 | ||
4262 | if (ret != 0) { | |
4263 | if (path->slots[0] == 0) | |
4264 | goto not_found; | |
4265 | path->slots[0]--; | |
4266 | } | |
4267 | ||
5f39d397 CM |
4268 | leaf = path->nodes[0]; |
4269 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4270 | struct btrfs_file_extent_item); |
a52d9a80 | 4271 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4272 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4273 | found_type = btrfs_key_type(&found_key); | |
4274 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4275 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4276 | goto not_found; | |
4277 | } | |
4278 | ||
5f39d397 CM |
4279 | found_type = btrfs_file_extent_type(leaf, item); |
4280 | extent_start = found_key.offset; | |
c8b97818 | 4281 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4282 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4283 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4284 | extent_end = extent_start + |
db94535d | 4285 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4286 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4287 | size_t size; | |
4288 | size = btrfs_file_extent_inline_len(leaf, item); | |
4289 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4290 | ~((u64)root->sectorsize - 1); | |
4291 | } | |
4292 | ||
4293 | if (start >= extent_end) { | |
4294 | path->slots[0]++; | |
4295 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4296 | ret = btrfs_next_leaf(root, path); | |
4297 | if (ret < 0) { | |
4298 | err = ret; | |
4299 | goto out; | |
a52d9a80 | 4300 | } |
9036c102 YZ |
4301 | if (ret > 0) |
4302 | goto not_found; | |
4303 | leaf = path->nodes[0]; | |
a52d9a80 | 4304 | } |
9036c102 YZ |
4305 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4306 | if (found_key.objectid != objectid || | |
4307 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4308 | goto not_found; | |
4309 | if (start + len <= found_key.offset) | |
4310 | goto not_found; | |
4311 | em->start = start; | |
4312 | em->len = found_key.offset - start; | |
4313 | goto not_found_em; | |
4314 | } | |
4315 | ||
d899e052 YZ |
4316 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4317 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4318 | em->start = extent_start; |
4319 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4320 | em->orig_start = extent_start - |
4321 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4322 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4323 | if (bytenr == 0) { | |
5f39d397 | 4324 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4325 | goto insert; |
4326 | } | |
c8b97818 CM |
4327 | if (compressed) { |
4328 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4329 | em->block_start = bytenr; | |
4330 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4331 | item); | |
4332 | } else { | |
4333 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4334 | em->block_start = bytenr; | |
4335 | em->block_len = em->len; | |
d899e052 YZ |
4336 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4337 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4338 | } |
a52d9a80 CM |
4339 | goto insert; |
4340 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4341 | unsigned long ptr; |
a52d9a80 | 4342 | char *map; |
3326d1b0 CM |
4343 | size_t size; |
4344 | size_t extent_offset; | |
4345 | size_t copy_size; | |
a52d9a80 | 4346 | |
689f9346 | 4347 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4348 | if (!page || create) { |
689f9346 | 4349 | em->start = extent_start; |
9036c102 | 4350 | em->len = extent_end - extent_start; |
689f9346 Y |
4351 | goto out; |
4352 | } | |
5f39d397 | 4353 | |
9036c102 YZ |
4354 | size = btrfs_file_extent_inline_len(leaf, item); |
4355 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4356 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4357 | size - extent_offset); |
3326d1b0 | 4358 | em->start = extent_start + extent_offset; |
70dec807 CM |
4359 | em->len = (copy_size + root->sectorsize - 1) & |
4360 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4361 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4362 | if (compressed) |
4363 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4364 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4365 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4366 | if (btrfs_file_extent_compression(leaf, item) == |
4367 | BTRFS_COMPRESS_ZLIB) { | |
4368 | ret = uncompress_inline(path, inode, page, | |
4369 | pg_offset, | |
4370 | extent_offset, item); | |
4371 | BUG_ON(ret); | |
4372 | } else { | |
4373 | map = kmap(page); | |
4374 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4375 | copy_size); | |
93c82d57 CM |
4376 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4377 | memset(map + pg_offset + copy_size, 0, | |
4378 | PAGE_CACHE_SIZE - pg_offset - | |
4379 | copy_size); | |
4380 | } | |
c8b97818 CM |
4381 | kunmap(page); |
4382 | } | |
179e29e4 CM |
4383 | flush_dcache_page(page); |
4384 | } else if (create && PageUptodate(page)) { | |
4385 | if (!trans) { | |
4386 | kunmap(page); | |
4387 | free_extent_map(em); | |
4388 | em = NULL; | |
4389 | btrfs_release_path(root, path); | |
f9295749 | 4390 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4391 | goto again; |
4392 | } | |
c8b97818 | 4393 | map = kmap(page); |
70dec807 | 4394 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4395 | copy_size); |
c8b97818 | 4396 | kunmap(page); |
179e29e4 | 4397 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4398 | } |
d1310b2e CM |
4399 | set_extent_uptodate(io_tree, em->start, |
4400 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4401 | goto insert; |
4402 | } else { | |
d397712b | 4403 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4404 | WARN_ON(1); |
4405 | } | |
4406 | not_found: | |
4407 | em->start = start; | |
d1310b2e | 4408 | em->len = len; |
a52d9a80 | 4409 | not_found_em: |
5f39d397 | 4410 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4411 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4412 | insert: |
4413 | btrfs_release_path(root, path); | |
d1310b2e | 4414 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4415 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4416 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4417 | (unsigned long long)em->len, | |
4418 | (unsigned long long)start, | |
4419 | (unsigned long long)len); | |
a52d9a80 CM |
4420 | err = -EIO; |
4421 | goto out; | |
4422 | } | |
d1310b2e CM |
4423 | |
4424 | err = 0; | |
890871be | 4425 | write_lock(&em_tree->lock); |
a52d9a80 | 4426 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4427 | /* it is possible that someone inserted the extent into the tree |
4428 | * while we had the lock dropped. It is also possible that | |
4429 | * an overlapping map exists in the tree | |
4430 | */ | |
a52d9a80 | 4431 | if (ret == -EEXIST) { |
3b951516 | 4432 | struct extent_map *existing; |
e6dcd2dc CM |
4433 | |
4434 | ret = 0; | |
4435 | ||
3b951516 | 4436 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4437 | if (existing && (existing->start > start || |
4438 | existing->start + existing->len <= start)) { | |
4439 | free_extent_map(existing); | |
4440 | existing = NULL; | |
4441 | } | |
3b951516 CM |
4442 | if (!existing) { |
4443 | existing = lookup_extent_mapping(em_tree, em->start, | |
4444 | em->len); | |
4445 | if (existing) { | |
4446 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4447 | em, start, |
4448 | root->sectorsize); | |
3b951516 CM |
4449 | free_extent_map(existing); |
4450 | if (err) { | |
4451 | free_extent_map(em); | |
4452 | em = NULL; | |
4453 | } | |
4454 | } else { | |
4455 | err = -EIO; | |
3b951516 CM |
4456 | free_extent_map(em); |
4457 | em = NULL; | |
4458 | } | |
4459 | } else { | |
4460 | free_extent_map(em); | |
4461 | em = existing; | |
e6dcd2dc | 4462 | err = 0; |
a52d9a80 | 4463 | } |
a52d9a80 | 4464 | } |
890871be | 4465 | write_unlock(&em_tree->lock); |
a52d9a80 | 4466 | out: |
f421950f CM |
4467 | if (path) |
4468 | btrfs_free_path(path); | |
a52d9a80 CM |
4469 | if (trans) { |
4470 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4471 | if (!err) |
a52d9a80 CM |
4472 | err = ret; |
4473 | } | |
a52d9a80 CM |
4474 | if (err) { |
4475 | free_extent_map(em); | |
a52d9a80 CM |
4476 | return ERR_PTR(err); |
4477 | } | |
4478 | return em; | |
4479 | } | |
4480 | ||
16432985 CM |
4481 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4482 | const struct iovec *iov, loff_t offset, | |
4483 | unsigned long nr_segs) | |
4484 | { | |
e1c4b745 | 4485 | return -EINVAL; |
16432985 CM |
4486 | } |
4487 | ||
1506fcc8 YS |
4488 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4489 | __u64 start, __u64 len) | |
4490 | { | |
4491 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4492 | } | |
4493 | ||
a52d9a80 | 4494 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4495 | { |
d1310b2e CM |
4496 | struct extent_io_tree *tree; |
4497 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4498 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4499 | } |
1832a6d5 | 4500 | |
a52d9a80 | 4501 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4502 | { |
d1310b2e | 4503 | struct extent_io_tree *tree; |
b888db2b CM |
4504 | |
4505 | ||
4506 | if (current->flags & PF_MEMALLOC) { | |
4507 | redirty_page_for_writepage(wbc, page); | |
4508 | unlock_page(page); | |
4509 | return 0; | |
4510 | } | |
d1310b2e | 4511 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4512 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4513 | } |
4514 | ||
f421950f CM |
4515 | int btrfs_writepages(struct address_space *mapping, |
4516 | struct writeback_control *wbc) | |
b293f02e | 4517 | { |
d1310b2e | 4518 | struct extent_io_tree *tree; |
771ed689 | 4519 | |
d1310b2e | 4520 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4521 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4522 | } | |
4523 | ||
3ab2fb5a CM |
4524 | static int |
4525 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4526 | struct list_head *pages, unsigned nr_pages) | |
4527 | { | |
d1310b2e CM |
4528 | struct extent_io_tree *tree; |
4529 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4530 | return extent_readpages(tree, mapping, pages, nr_pages, |
4531 | btrfs_get_extent); | |
4532 | } | |
e6dcd2dc | 4533 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4534 | { |
d1310b2e CM |
4535 | struct extent_io_tree *tree; |
4536 | struct extent_map_tree *map; | |
a52d9a80 | 4537 | int ret; |
8c2383c3 | 4538 | |
d1310b2e CM |
4539 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4540 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4541 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4542 | if (ret == 1) { |
4543 | ClearPagePrivate(page); | |
4544 | set_page_private(page, 0); | |
4545 | page_cache_release(page); | |
39279cc3 | 4546 | } |
a52d9a80 | 4547 | return ret; |
39279cc3 CM |
4548 | } |
4549 | ||
e6dcd2dc CM |
4550 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4551 | { | |
98509cfc CM |
4552 | if (PageWriteback(page) || PageDirty(page)) |
4553 | return 0; | |
b335b003 | 4554 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4555 | } |
4556 | ||
a52d9a80 | 4557 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4558 | { |
d1310b2e | 4559 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4560 | struct btrfs_ordered_extent *ordered; |
4561 | u64 page_start = page_offset(page); | |
4562 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4563 | |
8b62b72b CM |
4564 | |
4565 | /* | |
4566 | * we have the page locked, so new writeback can't start, | |
4567 | * and the dirty bit won't be cleared while we are here. | |
4568 | * | |
4569 | * Wait for IO on this page so that we can safely clear | |
4570 | * the PagePrivate2 bit and do ordered accounting | |
4571 | */ | |
e6dcd2dc | 4572 | wait_on_page_writeback(page); |
8b62b72b | 4573 | |
d1310b2e | 4574 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4575 | if (offset) { |
4576 | btrfs_releasepage(page, GFP_NOFS); | |
4577 | return; | |
4578 | } | |
e6dcd2dc CM |
4579 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4580 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4581 | page_offset(page)); | |
4582 | if (ordered) { | |
eb84ae03 CM |
4583 | /* |
4584 | * IO on this page will never be started, so we need | |
4585 | * to account for any ordered extents now | |
4586 | */ | |
e6dcd2dc CM |
4587 | clear_extent_bit(tree, page_start, page_end, |
4588 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
2c64c53d | 4589 | EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS); |
8b62b72b CM |
4590 | /* |
4591 | * whoever cleared the private bit is responsible | |
4592 | * for the finish_ordered_io | |
4593 | */ | |
4594 | if (TestClearPagePrivate2(page)) { | |
4595 | btrfs_finish_ordered_io(page->mapping->host, | |
4596 | page_start, page_end); | |
4597 | } | |
e6dcd2dc CM |
4598 | btrfs_put_ordered_extent(ordered); |
4599 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4600 | } | |
4601 | clear_extent_bit(tree, page_start, page_end, | |
8b62b72b | 4602 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC, |
2c64c53d | 4603 | 1, 1, NULL, GFP_NOFS); |
e6dcd2dc CM |
4604 | __btrfs_releasepage(page, GFP_NOFS); |
4605 | ||
4a096752 | 4606 | ClearPageChecked(page); |
9ad6b7bc | 4607 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4608 | ClearPagePrivate(page); |
4609 | set_page_private(page, 0); | |
4610 | page_cache_release(page); | |
4611 | } | |
39279cc3 CM |
4612 | } |
4613 | ||
9ebefb18 CM |
4614 | /* |
4615 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4616 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4617 | * be careful to check for EOF conditions here. We set the page up correctly | |
4618 | * for a written page which means we get ENOSPC checking when writing into | |
4619 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4620 | * support these features. | |
4621 | * | |
4622 | * We are not allowed to take the i_mutex here so we have to play games to | |
4623 | * protect against truncate races as the page could now be beyond EOF. Because | |
4624 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4625 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4626 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4627 | * unlock the page. | |
4628 | */ | |
c2ec175c | 4629 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4630 | { |
c2ec175c | 4631 | struct page *page = vmf->page; |
6da6abae | 4632 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4633 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4634 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4635 | struct btrfs_ordered_extent *ordered; | |
4636 | char *kaddr; | |
4637 | unsigned long zero_start; | |
9ebefb18 | 4638 | loff_t size; |
1832a6d5 | 4639 | int ret; |
a52d9a80 | 4640 | u64 page_start; |
e6dcd2dc | 4641 | u64 page_end; |
9ebefb18 | 4642 | |
6a63209f | 4643 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4644 | if (ret) { |
4645 | if (ret == -ENOMEM) | |
4646 | ret = VM_FAULT_OOM; | |
4647 | else /* -ENOSPC, -EIO, etc */ | |
4648 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4649 | goto out; |
56a76f82 | 4650 | } |
1832a6d5 | 4651 | |
56a76f82 | 4652 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4653 | again: |
9ebefb18 | 4654 | lock_page(page); |
9ebefb18 | 4655 | size = i_size_read(inode); |
e6dcd2dc CM |
4656 | page_start = page_offset(page); |
4657 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4658 | |
9ebefb18 | 4659 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4660 | (page_start >= size)) { |
6a63209f | 4661 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
4662 | /* page got truncated out from underneath us */ |
4663 | goto out_unlock; | |
4664 | } | |
e6dcd2dc CM |
4665 | wait_on_page_writeback(page); |
4666 | ||
4667 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4668 | set_page_extent_mapped(page); | |
4669 | ||
eb84ae03 CM |
4670 | /* |
4671 | * we can't set the delalloc bits if there are pending ordered | |
4672 | * extents. Drop our locks and wait for them to finish | |
4673 | */ | |
e6dcd2dc CM |
4674 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4675 | if (ordered) { | |
4676 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4677 | unlock_page(page); | |
eb84ae03 | 4678 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4679 | btrfs_put_ordered_extent(ordered); |
4680 | goto again; | |
4681 | } | |
4682 | ||
ea8c2819 | 4683 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc | 4684 | ret = 0; |
9ebefb18 CM |
4685 | |
4686 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4687 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4688 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4689 | else |
e6dcd2dc | 4690 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4691 | |
e6dcd2dc CM |
4692 | if (zero_start != PAGE_CACHE_SIZE) { |
4693 | kaddr = kmap(page); | |
4694 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4695 | flush_dcache_page(page); | |
4696 | kunmap(page); | |
4697 | } | |
247e743c | 4698 | ClearPageChecked(page); |
e6dcd2dc | 4699 | set_page_dirty(page); |
50a9b214 | 4700 | SetPageUptodate(page); |
5a3f23d5 CM |
4701 | |
4702 | BTRFS_I(inode)->last_trans = root->fs_info->generation + 1; | |
e6dcd2dc | 4703 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
4704 | |
4705 | out_unlock: | |
50a9b214 CM |
4706 | if (!ret) |
4707 | return VM_FAULT_LOCKED; | |
9ebefb18 | 4708 | unlock_page(page); |
1832a6d5 | 4709 | out: |
9ebefb18 CM |
4710 | return ret; |
4711 | } | |
4712 | ||
39279cc3 CM |
4713 | static void btrfs_truncate(struct inode *inode) |
4714 | { | |
4715 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4716 | int ret; | |
4717 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 4718 | unsigned long nr; |
dbe674a9 | 4719 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
4720 | |
4721 | if (!S_ISREG(inode->i_mode)) | |
4722 | return; | |
4723 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4724 | return; | |
4725 | ||
4726 | btrfs_truncate_page(inode->i_mapping, inode->i_size); | |
4a096752 | 4727 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 4728 | |
39279cc3 | 4729 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
4730 | |
4731 | /* | |
4732 | * setattr is responsible for setting the ordered_data_close flag, | |
4733 | * but that is only tested during the last file release. That | |
4734 | * could happen well after the next commit, leaving a great big | |
4735 | * window where new writes may get lost if someone chooses to write | |
4736 | * to this file after truncating to zero | |
4737 | * | |
4738 | * The inode doesn't have any dirty data here, and so if we commit | |
4739 | * this is a noop. If someone immediately starts writing to the inode | |
4740 | * it is very likely we'll catch some of their writes in this | |
4741 | * transaction, and the commit will find this file on the ordered | |
4742 | * data list with good things to send down. | |
4743 | * | |
4744 | * This is a best effort solution, there is still a window where | |
4745 | * using truncate to replace the contents of the file will | |
4746 | * end up with a zero length file after a crash. | |
4747 | */ | |
4748 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
4749 | btrfs_add_ordered_operation(trans, root, inode); | |
4750 | ||
39279cc3 | 4751 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 4752 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 4753 | |
7b128766 JB |
4754 | ret = btrfs_orphan_add(trans, inode); |
4755 | if (ret) | |
4756 | goto out; | |
39279cc3 | 4757 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 4758 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 4759 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 4760 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 4761 | |
7b128766 JB |
4762 | ret = btrfs_orphan_del(trans, inode); |
4763 | BUG_ON(ret); | |
4764 | ||
4765 | out: | |
4766 | nr = trans->blocks_used; | |
89ce8a63 | 4767 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 4768 | BUG_ON(ret); |
d3c2fdcf | 4769 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4770 | } |
4771 | ||
d352ac68 CM |
4772 | /* |
4773 | * create a new subvolume directory/inode (helper for the ioctl). | |
4774 | */ | |
d2fb3437 YZ |
4775 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
4776 | struct btrfs_root *new_root, struct dentry *dentry, | |
4777 | u64 new_dirid, u64 alloc_hint) | |
39279cc3 | 4778 | { |
39279cc3 | 4779 | struct inode *inode; |
cb8e7090 | 4780 | int error; |
00e4e6b3 | 4781 | u64 index = 0; |
39279cc3 | 4782 | |
aec7477b | 4783 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 4784 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 4785 | if (IS_ERR(inode)) |
f46b5a66 | 4786 | return PTR_ERR(inode); |
39279cc3 CM |
4787 | inode->i_op = &btrfs_dir_inode_operations; |
4788 | inode->i_fop = &btrfs_dir_file_operations; | |
4789 | ||
39279cc3 | 4790 | inode->i_nlink = 1; |
dbe674a9 | 4791 | btrfs_i_size_write(inode, 0); |
3b96362c | 4792 | |
cb8e7090 CH |
4793 | error = btrfs_update_inode(trans, new_root, inode); |
4794 | if (error) | |
4795 | return error; | |
4796 | ||
4797 | d_instantiate(dentry, inode); | |
4798 | return 0; | |
39279cc3 CM |
4799 | } |
4800 | ||
d352ac68 CM |
4801 | /* helper function for file defrag and space balancing. This |
4802 | * forces readahead on a given range of bytes in an inode | |
4803 | */ | |
edbd8d4e | 4804 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
4805 | struct file_ra_state *ra, struct file *file, |
4806 | pgoff_t offset, pgoff_t last_index) | |
4807 | { | |
8e7bf94f | 4808 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 4809 | |
86479a04 CM |
4810 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
4811 | return offset + req_size; | |
86479a04 CM |
4812 | } |
4813 | ||
39279cc3 CM |
4814 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
4815 | { | |
4816 | struct btrfs_inode *ei; | |
4817 | ||
4818 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
4819 | if (!ei) | |
4820 | return NULL; | |
15ee9bc7 | 4821 | ei->last_trans = 0; |
e02119d5 | 4822 | ei->logged_trans = 0; |
e6dcd2dc | 4823 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 4824 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 4825 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
4826 | return &ei->vfs_inode; |
4827 | } | |
4828 | ||
4829 | void btrfs_destroy_inode(struct inode *inode) | |
4830 | { | |
e6dcd2dc | 4831 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
4832 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4833 | ||
39279cc3 CM |
4834 | WARN_ON(!list_empty(&inode->i_dentry)); |
4835 | WARN_ON(inode->i_data.nrpages); | |
4836 | ||
5a3f23d5 CM |
4837 | /* |
4838 | * Make sure we're properly removed from the ordered operation | |
4839 | * lists. | |
4840 | */ | |
4841 | smp_mb(); | |
4842 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
4843 | spin_lock(&root->fs_info->ordered_extent_lock); | |
4844 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
4845 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
4846 | } | |
4847 | ||
4848 | spin_lock(&root->list_lock); | |
7b128766 JB |
4849 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
4850 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
4851 | " list\n", inode->i_ino); | |
4852 | dump_stack(); | |
4853 | } | |
5a3f23d5 | 4854 | spin_unlock(&root->list_lock); |
7b128766 | 4855 | |
d397712b | 4856 | while (1) { |
e6dcd2dc CM |
4857 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
4858 | if (!ordered) | |
4859 | break; | |
4860 | else { | |
d397712b CM |
4861 | printk(KERN_ERR "btrfs found ordered " |
4862 | "extent %llu %llu on inode cleanup\n", | |
4863 | (unsigned long long)ordered->file_offset, | |
4864 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
4865 | btrfs_remove_ordered_extent(inode, ordered); |
4866 | btrfs_put_ordered_extent(ordered); | |
4867 | btrfs_put_ordered_extent(ordered); | |
4868 | } | |
4869 | } | |
5d4f98a2 | 4870 | inode_tree_del(inode); |
5b21f2ed | 4871 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
4872 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
4873 | } | |
4874 | ||
0ee0fda0 | 4875 | static void init_once(void *foo) |
39279cc3 CM |
4876 | { |
4877 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
4878 | ||
4879 | inode_init_once(&ei->vfs_inode); | |
4880 | } | |
4881 | ||
4882 | void btrfs_destroy_cachep(void) | |
4883 | { | |
4884 | if (btrfs_inode_cachep) | |
4885 | kmem_cache_destroy(btrfs_inode_cachep); | |
4886 | if (btrfs_trans_handle_cachep) | |
4887 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
4888 | if (btrfs_transaction_cachep) | |
4889 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
4890 | if (btrfs_path_cachep) |
4891 | kmem_cache_destroy(btrfs_path_cachep); | |
4892 | } | |
4893 | ||
4894 | int btrfs_init_cachep(void) | |
4895 | { | |
9601e3f6 CH |
4896 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
4897 | sizeof(struct btrfs_inode), 0, | |
4898 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
4899 | if (!btrfs_inode_cachep) |
4900 | goto fail; | |
9601e3f6 CH |
4901 | |
4902 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
4903 | sizeof(struct btrfs_trans_handle), 0, | |
4904 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
4905 | if (!btrfs_trans_handle_cachep) |
4906 | goto fail; | |
9601e3f6 CH |
4907 | |
4908 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
4909 | sizeof(struct btrfs_transaction), 0, | |
4910 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
4911 | if (!btrfs_transaction_cachep) |
4912 | goto fail; | |
9601e3f6 CH |
4913 | |
4914 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
4915 | sizeof(struct btrfs_path), 0, | |
4916 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
4917 | if (!btrfs_path_cachep) |
4918 | goto fail; | |
9601e3f6 | 4919 | |
39279cc3 CM |
4920 | return 0; |
4921 | fail: | |
4922 | btrfs_destroy_cachep(); | |
4923 | return -ENOMEM; | |
4924 | } | |
4925 | ||
4926 | static int btrfs_getattr(struct vfsmount *mnt, | |
4927 | struct dentry *dentry, struct kstat *stat) | |
4928 | { | |
4929 | struct inode *inode = dentry->d_inode; | |
4930 | generic_fillattr(inode, stat); | |
3394e160 | 4931 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 4932 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
4933 | stat->blocks = (inode_get_bytes(inode) + |
4934 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
4935 | return 0; |
4936 | } | |
4937 | ||
d397712b CM |
4938 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
4939 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
4940 | { |
4941 | struct btrfs_trans_handle *trans; | |
4942 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 4943 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
4944 | struct inode *new_inode = new_dentry->d_inode; |
4945 | struct inode *old_inode = old_dentry->d_inode; | |
4946 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 4947 | u64 index = 0; |
4df27c4d | 4948 | u64 root_objectid; |
39279cc3 CM |
4949 | int ret; |
4950 | ||
4df27c4d YZ |
4951 | /* we only allow rename subvolume link between subvolumes */ |
4952 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
4953 | return -EXDEV; |
4954 | ||
4df27c4d YZ |
4955 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4956 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 4957 | return -ENOTEMPTY; |
5f39d397 | 4958 | |
4df27c4d YZ |
4959 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
4960 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
4961 | return -ENOTEMPTY; | |
0660b5af | 4962 | |
6a63209f | 4963 | ret = btrfs_check_metadata_free_space(root); |
1832a6d5 | 4964 | if (ret) |
4df27c4d | 4965 | return ret; |
1832a6d5 | 4966 | |
5a3f23d5 CM |
4967 | /* |
4968 | * we're using rename to replace one file with another. | |
4969 | * and the replacement file is large. Start IO on it now so | |
4970 | * we don't add too much work to the end of the transaction | |
4971 | */ | |
4baf8c92 | 4972 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
4973 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
4974 | filemap_flush(old_inode->i_mapping); | |
4975 | ||
39279cc3 | 4976 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4977 | |
4df27c4d YZ |
4978 | if (dest != root) |
4979 | btrfs_record_root_in_trans(trans, dest); | |
4980 | ||
5a3f23d5 CM |
4981 | /* |
4982 | * make sure the inode gets flushed if it is replacing | |
4983 | * something. | |
4984 | */ | |
4985 | if (new_inode && new_inode->i_size && | |
4986 | old_inode && S_ISREG(old_inode->i_mode)) { | |
4987 | btrfs_add_ordered_operation(trans, root, old_inode); | |
4988 | } | |
4989 | ||
4df27c4d YZ |
4990 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { |
4991 | /* force full log commit if subvolume involved. */ | |
4992 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
4993 | } else { | |
4994 | /* | |
4995 | * this is an ugly little race, but the rename is required | |
4996 | * to make sure that if we crash, the inode is either at the | |
4997 | * old name or the new one. pinning the log transaction lets | |
4998 | * us make sure we don't allow a log commit to come in after | |
4999 | * we unlink the name but before we add the new name back in. | |
5000 | */ | |
5001 | btrfs_pin_log_trans(root); | |
5002 | } | |
12fcfd22 | 5003 | |
39279cc3 | 5004 | btrfs_set_trans_block_group(trans, new_dir); |
39279cc3 | 5005 | |
39279cc3 CM |
5006 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5007 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5008 | old_inode->i_ctime = ctime; | |
5f39d397 | 5009 | |
12fcfd22 CM |
5010 | if (old_dentry->d_parent != new_dentry->d_parent) |
5011 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5012 | ||
4df27c4d YZ |
5013 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5014 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5015 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5016 | old_dentry->d_name.name, | |
5017 | old_dentry->d_name.len); | |
5018 | } else { | |
5019 | btrfs_inc_nlink(old_dentry->d_inode); | |
5020 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5021 | old_dentry->d_inode, | |
5022 | old_dentry->d_name.name, | |
5023 | old_dentry->d_name.len); | |
5024 | } | |
5025 | BUG_ON(ret); | |
39279cc3 CM |
5026 | |
5027 | if (new_inode) { | |
5028 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5029 | if (unlikely(new_inode->i_ino == |
5030 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5031 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5032 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5033 | root_objectid, | |
5034 | new_dentry->d_name.name, | |
5035 | new_dentry->d_name.len); | |
5036 | BUG_ON(new_inode->i_nlink == 0); | |
5037 | } else { | |
5038 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5039 | new_dentry->d_inode, | |
5040 | new_dentry->d_name.name, | |
5041 | new_dentry->d_name.len); | |
5042 | } | |
5043 | BUG_ON(ret); | |
7b128766 | 5044 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5045 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5046 | BUG_ON(ret); |
7b128766 | 5047 | } |
39279cc3 | 5048 | } |
3de4586c | 5049 | ret = btrfs_set_inode_index(new_dir, &index); |
4df27c4d | 5050 | BUG_ON(ret); |
aec7477b | 5051 | |
4df27c4d YZ |
5052 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5053 | new_dentry->d_name.name, | |
e02119d5 | 5054 | new_dentry->d_name.len, 1, index); |
4df27c4d | 5055 | BUG_ON(ret); |
39279cc3 | 5056 | |
4df27c4d YZ |
5057 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5058 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5059 | new_dentry->d_parent); | |
5060 | btrfs_end_log_trans(root); | |
5061 | } | |
12fcfd22 | 5062 | |
ab78c84d | 5063 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5064 | |
39279cc3 CM |
5065 | return ret; |
5066 | } | |
5067 | ||
d352ac68 CM |
5068 | /* |
5069 | * some fairly slow code that needs optimization. This walks the list | |
5070 | * of all the inodes with pending delalloc and forces them to disk. | |
5071 | */ | |
ea8c2819 CM |
5072 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5073 | { | |
5074 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5075 | struct btrfs_inode *binode; | |
5b21f2ed | 5076 | struct inode *inode; |
ea8c2819 | 5077 | |
c146afad YZ |
5078 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5079 | return -EROFS; | |
5080 | ||
75eff68e | 5081 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5082 | while (!list_empty(head)) { |
ea8c2819 CM |
5083 | binode = list_entry(head->next, struct btrfs_inode, |
5084 | delalloc_inodes); | |
5b21f2ed ZY |
5085 | inode = igrab(&binode->vfs_inode); |
5086 | if (!inode) | |
5087 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5088 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5089 | if (inode) { |
8c8bee1d | 5090 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5091 | iput(inode); |
5092 | } | |
5093 | cond_resched(); | |
75eff68e | 5094 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5095 | } |
75eff68e | 5096 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5097 | |
5098 | /* the filemap_flush will queue IO into the worker threads, but | |
5099 | * we have to make sure the IO is actually started and that | |
5100 | * ordered extents get created before we return | |
5101 | */ | |
5102 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5103 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5104 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5105 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5106 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5107 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5108 | } |
5109 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5110 | return 0; |
5111 | } | |
5112 | ||
39279cc3 CM |
5113 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5114 | const char *symname) | |
5115 | { | |
5116 | struct btrfs_trans_handle *trans; | |
5117 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5118 | struct btrfs_path *path; | |
5119 | struct btrfs_key key; | |
1832a6d5 | 5120 | struct inode *inode = NULL; |
39279cc3 CM |
5121 | int err; |
5122 | int drop_inode = 0; | |
5123 | u64 objectid; | |
00e4e6b3 | 5124 | u64 index = 0 ; |
39279cc3 CM |
5125 | int name_len; |
5126 | int datasize; | |
5f39d397 | 5127 | unsigned long ptr; |
39279cc3 | 5128 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5129 | struct extent_buffer *leaf; |
1832a6d5 | 5130 | unsigned long nr = 0; |
39279cc3 CM |
5131 | |
5132 | name_len = strlen(symname) + 1; | |
5133 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5134 | return -ENAMETOOLONG; | |
1832a6d5 | 5135 | |
6a63209f | 5136 | err = btrfs_check_metadata_free_space(root); |
1832a6d5 CM |
5137 | if (err) |
5138 | goto out_fail; | |
5139 | ||
39279cc3 CM |
5140 | trans = btrfs_start_transaction(root, 1); |
5141 | btrfs_set_trans_block_group(trans, dir); | |
5142 | ||
5143 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5144 | if (err) { | |
5145 | err = -ENOSPC; | |
5146 | goto out_unlock; | |
5147 | } | |
5148 | ||
aec7477b | 5149 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5150 | dentry->d_name.len, |
5151 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5152 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5153 | &index); | |
39279cc3 CM |
5154 | err = PTR_ERR(inode); |
5155 | if (IS_ERR(inode)) | |
5156 | goto out_unlock; | |
5157 | ||
0279b4cd | 5158 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5159 | if (err) { |
5160 | drop_inode = 1; | |
5161 | goto out_unlock; | |
5162 | } | |
5163 | ||
39279cc3 | 5164 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5165 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5166 | if (err) |
5167 | drop_inode = 1; | |
5168 | else { | |
5169 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5170 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5171 | inode->i_fop = &btrfs_file_operations; |
5172 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5173 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5174 | } |
39279cc3 CM |
5175 | btrfs_update_inode_block_group(trans, inode); |
5176 | btrfs_update_inode_block_group(trans, dir); | |
5177 | if (drop_inode) | |
5178 | goto out_unlock; | |
5179 | ||
5180 | path = btrfs_alloc_path(); | |
5181 | BUG_ON(!path); | |
5182 | key.objectid = inode->i_ino; | |
5183 | key.offset = 0; | |
39279cc3 CM |
5184 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5185 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5186 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5187 | datasize); | |
54aa1f4d CM |
5188 | if (err) { |
5189 | drop_inode = 1; | |
5190 | goto out_unlock; | |
5191 | } | |
5f39d397 CM |
5192 | leaf = path->nodes[0]; |
5193 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5194 | struct btrfs_file_extent_item); | |
5195 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5196 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5197 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5198 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5199 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5200 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5201 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5202 | ||
39279cc3 | 5203 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5204 | write_extent_buffer(leaf, symname, ptr, name_len); |
5205 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5206 | btrfs_free_path(path); |
5f39d397 | 5207 | |
39279cc3 CM |
5208 | inode->i_op = &btrfs_symlink_inode_operations; |
5209 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5210 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5211 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5212 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5213 | err = btrfs_update_inode(trans, root, inode); |
5214 | if (err) | |
5215 | drop_inode = 1; | |
39279cc3 CM |
5216 | |
5217 | out_unlock: | |
d3c2fdcf | 5218 | nr = trans->blocks_used; |
ab78c84d | 5219 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5220 | out_fail: |
39279cc3 CM |
5221 | if (drop_inode) { |
5222 | inode_dec_link_count(inode); | |
5223 | iput(inode); | |
5224 | } | |
d3c2fdcf | 5225 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5226 | return err; |
5227 | } | |
16432985 | 5228 | |
546888da CM |
5229 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5230 | struct inode *inode, u64 start, u64 end, | |
e980b50c | 5231 | u64 locked_end, u64 alloc_hint, int mode) |
d899e052 | 5232 | { |
d899e052 YZ |
5233 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5234 | struct btrfs_key ins; | |
5235 | u64 alloc_size; | |
5236 | u64 cur_offset = start; | |
5237 | u64 num_bytes = end - start; | |
5238 | int ret = 0; | |
5239 | ||
d899e052 YZ |
5240 | while (num_bytes > 0) { |
5241 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
5242 | ret = btrfs_reserve_extent(trans, root, alloc_size, | |
5243 | root->sectorsize, 0, alloc_hint, | |
5244 | (u64)-1, &ins, 1); | |
5245 | if (ret) { | |
5246 | WARN_ON(1); | |
5247 | goto out; | |
5248 | } | |
5249 | ret = insert_reserved_file_extent(trans, inode, | |
5250 | cur_offset, ins.objectid, | |
5251 | ins.offset, ins.offset, | |
e980b50c CM |
5252 | ins.offset, locked_end, |
5253 | 0, 0, 0, | |
d899e052 YZ |
5254 | BTRFS_FILE_EXTENT_PREALLOC); |
5255 | BUG_ON(ret); | |
a1ed835e CM |
5256 | btrfs_drop_extent_cache(inode, cur_offset, |
5257 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5258 | num_bytes -= ins.offset; |
5259 | cur_offset += ins.offset; | |
5260 | alloc_hint = ins.objectid + ins.offset; | |
5261 | } | |
5262 | out: | |
5263 | if (cur_offset > start) { | |
5264 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5265 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5266 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5267 | cur_offset > i_size_read(inode)) | |
5268 | btrfs_i_size_write(inode, cur_offset); | |
5269 | ret = btrfs_update_inode(trans, root, inode); | |
5270 | BUG_ON(ret); | |
5271 | } | |
5272 | ||
d899e052 YZ |
5273 | return ret; |
5274 | } | |
5275 | ||
5276 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5277 | loff_t offset, loff_t len) | |
5278 | { | |
5279 | u64 cur_offset; | |
5280 | u64 last_byte; | |
5281 | u64 alloc_start; | |
5282 | u64 alloc_end; | |
5283 | u64 alloc_hint = 0; | |
e980b50c | 5284 | u64 locked_end; |
d899e052 YZ |
5285 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5286 | struct extent_map *em; | |
546888da | 5287 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5288 | struct btrfs_root *root; |
d899e052 YZ |
5289 | int ret; |
5290 | ||
5291 | alloc_start = offset & ~mask; | |
5292 | alloc_end = (offset + len + mask) & ~mask; | |
5293 | ||
546888da CM |
5294 | /* |
5295 | * wait for ordered IO before we have any locks. We'll loop again | |
5296 | * below with the locks held. | |
5297 | */ | |
5298 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5299 | ||
d899e052 YZ |
5300 | mutex_lock(&inode->i_mutex); |
5301 | if (alloc_start > inode->i_size) { | |
5302 | ret = btrfs_cont_expand(inode, alloc_start); | |
5303 | if (ret) | |
5304 | goto out; | |
5305 | } | |
5306 | ||
a970b0a1 JB |
5307 | root = BTRFS_I(inode)->root; |
5308 | ||
5309 | ret = btrfs_check_data_free_space(root, inode, | |
5310 | alloc_end - alloc_start); | |
5311 | if (ret) | |
5312 | goto out; | |
5313 | ||
e980b50c | 5314 | locked_end = alloc_end - 1; |
d899e052 YZ |
5315 | while (1) { |
5316 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5317 | |
5318 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5319 | if (!trans) { | |
5320 | ret = -EIO; | |
a970b0a1 | 5321 | goto out_free; |
546888da CM |
5322 | } |
5323 | ||
5324 | /* the extent lock is ordered inside the running | |
5325 | * transaction | |
5326 | */ | |
e980b50c CM |
5327 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5328 | GFP_NOFS); | |
d899e052 YZ |
5329 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5330 | alloc_end - 1); | |
5331 | if (ordered && | |
5332 | ordered->file_offset + ordered->len > alloc_start && | |
5333 | ordered->file_offset < alloc_end) { | |
5334 | btrfs_put_ordered_extent(ordered); | |
5335 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5336 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5337 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5338 | ||
5339 | /* | |
5340 | * we can't wait on the range with the transaction | |
5341 | * running or with the extent lock held | |
5342 | */ | |
d899e052 YZ |
5343 | btrfs_wait_ordered_range(inode, alloc_start, |
5344 | alloc_end - alloc_start); | |
5345 | } else { | |
5346 | if (ordered) | |
5347 | btrfs_put_ordered_extent(ordered); | |
5348 | break; | |
5349 | } | |
5350 | } | |
5351 | ||
5352 | cur_offset = alloc_start; | |
5353 | while (1) { | |
5354 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5355 | alloc_end - cur_offset, 0); | |
5356 | BUG_ON(IS_ERR(em) || !em); | |
5357 | last_byte = min(extent_map_end(em), alloc_end); | |
5358 | last_byte = (last_byte + mask) & ~mask; | |
5359 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5360 | ret = prealloc_file_range(trans, inode, cur_offset, |
e980b50c CM |
5361 | last_byte, locked_end + 1, |
5362 | alloc_hint, mode); | |
d899e052 YZ |
5363 | if (ret < 0) { |
5364 | free_extent_map(em); | |
5365 | break; | |
5366 | } | |
5367 | } | |
5368 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5369 | alloc_hint = em->block_start; | |
5370 | free_extent_map(em); | |
5371 | ||
5372 | cur_offset = last_byte; | |
5373 | if (cur_offset >= alloc_end) { | |
5374 | ret = 0; | |
5375 | break; | |
5376 | } | |
5377 | } | |
e980b50c | 5378 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5379 | GFP_NOFS); |
546888da CM |
5380 | |
5381 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5382 | out_free: |
5383 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5384 | out: |
5385 | mutex_unlock(&inode->i_mutex); | |
5386 | return ret; | |
5387 | } | |
5388 | ||
e6dcd2dc CM |
5389 | static int btrfs_set_page_dirty(struct page *page) |
5390 | { | |
e6dcd2dc CM |
5391 | return __set_page_dirty_nobuffers(page); |
5392 | } | |
5393 | ||
0ee0fda0 | 5394 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5395 | { |
6cbff00f | 5396 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5397 | return -EACCES; |
33268eaf | 5398 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5399 | } |
39279cc3 CM |
5400 | |
5401 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 5402 | .getattr = btrfs_getattr, |
39279cc3 CM |
5403 | .lookup = btrfs_lookup, |
5404 | .create = btrfs_create, | |
5405 | .unlink = btrfs_unlink, | |
5406 | .link = btrfs_link, | |
5407 | .mkdir = btrfs_mkdir, | |
5408 | .rmdir = btrfs_rmdir, | |
5409 | .rename = btrfs_rename, | |
5410 | .symlink = btrfs_symlink, | |
5411 | .setattr = btrfs_setattr, | |
618e21d5 | 5412 | .mknod = btrfs_mknod, |
95819c05 CH |
5413 | .setxattr = btrfs_setxattr, |
5414 | .getxattr = btrfs_getxattr, | |
5103e947 | 5415 | .listxattr = btrfs_listxattr, |
95819c05 | 5416 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5417 | .permission = btrfs_permission, |
39279cc3 | 5418 | }; |
39279cc3 CM |
5419 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
5420 | .lookup = btrfs_lookup, | |
fdebe2bd | 5421 | .permission = btrfs_permission, |
39279cc3 | 5422 | }; |
39279cc3 CM |
5423 | static struct file_operations btrfs_dir_file_operations = { |
5424 | .llseek = generic_file_llseek, | |
5425 | .read = generic_read_dir, | |
cbdf5a24 | 5426 | .readdir = btrfs_real_readdir, |
34287aa3 | 5427 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5428 | #ifdef CONFIG_COMPAT |
34287aa3 | 5429 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5430 | #endif |
6bf13c0c | 5431 | .release = btrfs_release_file, |
e02119d5 | 5432 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5433 | }; |
5434 | ||
d1310b2e | 5435 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5436 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5437 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5438 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5439 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5440 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5441 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5442 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5443 | .set_bit_hook = btrfs_set_bit_hook, |
5444 | .clear_bit_hook = btrfs_clear_bit_hook, | |
07157aac CM |
5445 | }; |
5446 | ||
35054394 CM |
5447 | /* |
5448 | * btrfs doesn't support the bmap operation because swapfiles | |
5449 | * use bmap to make a mapping of extents in the file. They assume | |
5450 | * these extents won't change over the life of the file and they | |
5451 | * use the bmap result to do IO directly to the drive. | |
5452 | * | |
5453 | * the btrfs bmap call would return logical addresses that aren't | |
5454 | * suitable for IO and they also will change frequently as COW | |
5455 | * operations happen. So, swapfile + btrfs == corruption. | |
5456 | * | |
5457 | * For now we're avoiding this by dropping bmap. | |
5458 | */ | |
39279cc3 CM |
5459 | static struct address_space_operations btrfs_aops = { |
5460 | .readpage = btrfs_readpage, | |
5461 | .writepage = btrfs_writepage, | |
b293f02e | 5462 | .writepages = btrfs_writepages, |
3ab2fb5a | 5463 | .readpages = btrfs_readpages, |
39279cc3 | 5464 | .sync_page = block_sync_page, |
16432985 | 5465 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5466 | .invalidatepage = btrfs_invalidatepage, |
5467 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5468 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
5469 | }; |
5470 | ||
5471 | static struct address_space_operations btrfs_symlink_aops = { | |
5472 | .readpage = btrfs_readpage, | |
5473 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5474 | .invalidatepage = btrfs_invalidatepage, |
5475 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5476 | }; |
5477 | ||
5478 | static struct inode_operations btrfs_file_inode_operations = { | |
5479 | .truncate = btrfs_truncate, | |
5480 | .getattr = btrfs_getattr, | |
5481 | .setattr = btrfs_setattr, | |
95819c05 CH |
5482 | .setxattr = btrfs_setxattr, |
5483 | .getxattr = btrfs_getxattr, | |
5103e947 | 5484 | .listxattr = btrfs_listxattr, |
95819c05 | 5485 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5486 | .permission = btrfs_permission, |
d899e052 | 5487 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5488 | .fiemap = btrfs_fiemap, |
39279cc3 | 5489 | }; |
618e21d5 JB |
5490 | static struct inode_operations btrfs_special_inode_operations = { |
5491 | .getattr = btrfs_getattr, | |
5492 | .setattr = btrfs_setattr, | |
fdebe2bd | 5493 | .permission = btrfs_permission, |
95819c05 CH |
5494 | .setxattr = btrfs_setxattr, |
5495 | .getxattr = btrfs_getxattr, | |
33268eaf | 5496 | .listxattr = btrfs_listxattr, |
95819c05 | 5497 | .removexattr = btrfs_removexattr, |
618e21d5 | 5498 | }; |
39279cc3 CM |
5499 | static struct inode_operations btrfs_symlink_inode_operations = { |
5500 | .readlink = generic_readlink, | |
5501 | .follow_link = page_follow_link_light, | |
5502 | .put_link = page_put_link, | |
fdebe2bd | 5503 | .permission = btrfs_permission, |
0279b4cd JO |
5504 | .setxattr = btrfs_setxattr, |
5505 | .getxattr = btrfs_getxattr, | |
5506 | .listxattr = btrfs_listxattr, | |
5507 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5508 | }; |