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 | ||
6e1d5dcc AD |
58 | static const struct inode_operations btrfs_dir_inode_operations; |
59 | static const struct inode_operations btrfs_symlink_inode_operations; | |
60 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
61 | static const struct inode_operations btrfs_special_inode_operations; | |
62 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
63 | static const struct address_space_operations btrfs_aops; |
64 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 65 | static const 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 | ||
c2167754 YZ |
191 | /* |
192 | * we're an inline extent, so nobody can | |
193 | * extend the file past i_size without locking | |
194 | * a page we already have locked. | |
195 | * | |
196 | * We must do any isize and inode updates | |
197 | * before we unlock the pages. Otherwise we | |
198 | * could end up racing with unlink. | |
199 | */ | |
c8b97818 CM |
200 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
201 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 202 | |
c8b97818 CM |
203 | return 0; |
204 | fail: | |
205 | btrfs_free_path(path); | |
206 | return err; | |
207 | } | |
208 | ||
209 | ||
210 | /* | |
211 | * conditionally insert an inline extent into the file. This | |
212 | * does the checks required to make sure the data is small enough | |
213 | * to fit as an inline extent. | |
214 | */ | |
7f366cfe | 215 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
216 | struct btrfs_root *root, |
217 | struct inode *inode, u64 start, u64 end, | |
218 | size_t compressed_size, | |
219 | struct page **compressed_pages) | |
220 | { | |
221 | u64 isize = i_size_read(inode); | |
222 | u64 actual_end = min(end + 1, isize); | |
223 | u64 inline_len = actual_end - start; | |
224 | u64 aligned_end = (end + root->sectorsize - 1) & | |
225 | ~((u64)root->sectorsize - 1); | |
226 | u64 hint_byte; | |
227 | u64 data_len = inline_len; | |
228 | int ret; | |
229 | ||
230 | if (compressed_size) | |
231 | data_len = compressed_size; | |
232 | ||
233 | if (start > 0 || | |
70b99e69 | 234 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
235 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
236 | (!compressed_size && | |
237 | (actual_end & (root->sectorsize - 1)) == 0) || | |
238 | end + 1 < isize || | |
239 | data_len > root->fs_info->max_inline) { | |
240 | return 1; | |
241 | } | |
242 | ||
920bbbfb | 243 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 244 | &hint_byte, 1); |
c8b97818 CM |
245 | BUG_ON(ret); |
246 | ||
247 | if (isize > actual_end) | |
248 | inline_len = min_t(u64, isize, actual_end); | |
249 | ret = insert_inline_extent(trans, root, inode, start, | |
250 | inline_len, compressed_size, | |
251 | compressed_pages); | |
252 | BUG_ON(ret); | |
a1ed835e | 253 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
254 | return 0; |
255 | } | |
256 | ||
771ed689 CM |
257 | struct async_extent { |
258 | u64 start; | |
259 | u64 ram_size; | |
260 | u64 compressed_size; | |
261 | struct page **pages; | |
262 | unsigned long nr_pages; | |
263 | struct list_head list; | |
264 | }; | |
265 | ||
266 | struct async_cow { | |
267 | struct inode *inode; | |
268 | struct btrfs_root *root; | |
269 | struct page *locked_page; | |
270 | u64 start; | |
271 | u64 end; | |
272 | struct list_head extents; | |
273 | struct btrfs_work work; | |
274 | }; | |
275 | ||
276 | static noinline int add_async_extent(struct async_cow *cow, | |
277 | u64 start, u64 ram_size, | |
278 | u64 compressed_size, | |
279 | struct page **pages, | |
280 | unsigned long nr_pages) | |
281 | { | |
282 | struct async_extent *async_extent; | |
283 | ||
284 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
285 | async_extent->start = start; | |
286 | async_extent->ram_size = ram_size; | |
287 | async_extent->compressed_size = compressed_size; | |
288 | async_extent->pages = pages; | |
289 | async_extent->nr_pages = nr_pages; | |
290 | list_add_tail(&async_extent->list, &cow->extents); | |
291 | return 0; | |
292 | } | |
293 | ||
d352ac68 | 294 | /* |
771ed689 CM |
295 | * we create compressed extents in two phases. The first |
296 | * phase compresses a range of pages that have already been | |
297 | * locked (both pages and state bits are locked). | |
c8b97818 | 298 | * |
771ed689 CM |
299 | * This is done inside an ordered work queue, and the compression |
300 | * is spread across many cpus. The actual IO submission is step | |
301 | * two, and the ordered work queue takes care of making sure that | |
302 | * happens in the same order things were put onto the queue by | |
303 | * writepages and friends. | |
c8b97818 | 304 | * |
771ed689 CM |
305 | * If this code finds it can't get good compression, it puts an |
306 | * entry onto the work queue to write the uncompressed bytes. This | |
307 | * makes sure that both compressed inodes and uncompressed inodes | |
308 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 309 | */ |
771ed689 CM |
310 | static noinline int compress_file_range(struct inode *inode, |
311 | struct page *locked_page, | |
312 | u64 start, u64 end, | |
313 | struct async_cow *async_cow, | |
314 | int *num_added) | |
b888db2b CM |
315 | { |
316 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
317 | struct btrfs_trans_handle *trans; | |
db94535d | 318 | u64 num_bytes; |
c8b97818 CM |
319 | u64 orig_start; |
320 | u64 disk_num_bytes; | |
db94535d | 321 | u64 blocksize = root->sectorsize; |
c8b97818 | 322 | u64 actual_end; |
42dc7bab | 323 | u64 isize = i_size_read(inode); |
e6dcd2dc | 324 | int ret = 0; |
c8b97818 CM |
325 | struct page **pages = NULL; |
326 | unsigned long nr_pages; | |
327 | unsigned long nr_pages_ret = 0; | |
328 | unsigned long total_compressed = 0; | |
329 | unsigned long total_in = 0; | |
330 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 331 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
332 | int i; |
333 | int will_compress; | |
b888db2b | 334 | |
c8b97818 CM |
335 | orig_start = start; |
336 | ||
42dc7bab | 337 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
338 | again: |
339 | will_compress = 0; | |
340 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
341 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 342 | |
f03d9301 CM |
343 | /* |
344 | * we don't want to send crud past the end of i_size through | |
345 | * compression, that's just a waste of CPU time. So, if the | |
346 | * end of the file is before the start of our current | |
347 | * requested range of bytes, we bail out to the uncompressed | |
348 | * cleanup code that can deal with all of this. | |
349 | * | |
350 | * It isn't really the fastest way to fix things, but this is a | |
351 | * very uncommon corner. | |
352 | */ | |
353 | if (actual_end <= start) | |
354 | goto cleanup_and_bail_uncompressed; | |
355 | ||
c8b97818 CM |
356 | total_compressed = actual_end - start; |
357 | ||
358 | /* we want to make sure that amount of ram required to uncompress | |
359 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
360 | * of a compressed extent to 128k. This is a crucial number |
361 | * because it also controls how easily we can spread reads across | |
362 | * cpus for decompression. | |
363 | * | |
364 | * We also want to make sure the amount of IO required to do | |
365 | * a random read is reasonably small, so we limit the size of | |
366 | * a compressed extent to 128k. | |
c8b97818 CM |
367 | */ |
368 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 369 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 370 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
371 | disk_num_bytes = num_bytes; |
372 | total_in = 0; | |
373 | ret = 0; | |
db94535d | 374 | |
771ed689 CM |
375 | /* |
376 | * we do compression for mount -o compress and when the | |
377 | * inode has not been flagged as nocompress. This flag can | |
378 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 379 | */ |
6cbff00f | 380 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
c8b97818 CM |
381 | btrfs_test_opt(root, COMPRESS)) { |
382 | WARN_ON(pages); | |
cfbc246e | 383 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 384 | |
c8b97818 CM |
385 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
386 | total_compressed, pages, | |
387 | nr_pages, &nr_pages_ret, | |
388 | &total_in, | |
389 | &total_compressed, | |
390 | max_compressed); | |
391 | ||
392 | if (!ret) { | |
393 | unsigned long offset = total_compressed & | |
394 | (PAGE_CACHE_SIZE - 1); | |
395 | struct page *page = pages[nr_pages_ret - 1]; | |
396 | char *kaddr; | |
397 | ||
398 | /* zero the tail end of the last page, we might be | |
399 | * sending it down to disk | |
400 | */ | |
401 | if (offset) { | |
402 | kaddr = kmap_atomic(page, KM_USER0); | |
403 | memset(kaddr + offset, 0, | |
404 | PAGE_CACHE_SIZE - offset); | |
405 | kunmap_atomic(kaddr, KM_USER0); | |
406 | } | |
407 | will_compress = 1; | |
408 | } | |
409 | } | |
410 | if (start == 0) { | |
771ed689 CM |
411 | trans = btrfs_join_transaction(root, 1); |
412 | BUG_ON(!trans); | |
413 | btrfs_set_trans_block_group(trans, inode); | |
414 | ||
c8b97818 | 415 | /* lets try to make an inline extent */ |
771ed689 | 416 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 417 | /* we didn't compress the entire range, try |
771ed689 | 418 | * to make an uncompressed inline extent. |
c8b97818 CM |
419 | */ |
420 | ret = cow_file_range_inline(trans, root, inode, | |
421 | start, end, 0, NULL); | |
422 | } else { | |
771ed689 | 423 | /* try making a compressed inline extent */ |
c8b97818 CM |
424 | ret = cow_file_range_inline(trans, root, inode, |
425 | start, end, | |
426 | total_compressed, pages); | |
427 | } | |
428 | if (ret == 0) { | |
771ed689 CM |
429 | /* |
430 | * inline extent creation worked, we don't need | |
431 | * to create any more async work items. Unlock | |
432 | * and free up our temp pages. | |
433 | */ | |
c8b97818 | 434 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
435 | &BTRFS_I(inode)->io_tree, |
436 | start, end, NULL, | |
437 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 438 | EXTENT_CLEAR_DELALLOC | |
32c00aff | 439 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e | 440 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
441 | |
442 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
443 | goto free_pages_out; |
444 | } | |
c2167754 | 445 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
446 | } |
447 | ||
448 | if (will_compress) { | |
449 | /* | |
450 | * we aren't doing an inline extent round the compressed size | |
451 | * up to a block size boundary so the allocator does sane | |
452 | * things | |
453 | */ | |
454 | total_compressed = (total_compressed + blocksize - 1) & | |
455 | ~(blocksize - 1); | |
456 | ||
457 | /* | |
458 | * one last check to make sure the compression is really a | |
459 | * win, compare the page count read with the blocks on disk | |
460 | */ | |
461 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
462 | ~(PAGE_CACHE_SIZE - 1); | |
463 | if (total_compressed >= total_in) { | |
464 | will_compress = 0; | |
465 | } else { | |
466 | disk_num_bytes = total_compressed; | |
467 | num_bytes = total_in; | |
468 | } | |
469 | } | |
470 | if (!will_compress && pages) { | |
471 | /* | |
472 | * the compression code ran but failed to make things smaller, | |
473 | * free any pages it allocated and our page pointer array | |
474 | */ | |
475 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 476 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
477 | page_cache_release(pages[i]); |
478 | } | |
479 | kfree(pages); | |
480 | pages = NULL; | |
481 | total_compressed = 0; | |
482 | nr_pages_ret = 0; | |
483 | ||
484 | /* flag the file so we don't compress in the future */ | |
6cbff00f | 485 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
c8b97818 | 486 | } |
771ed689 CM |
487 | if (will_compress) { |
488 | *num_added += 1; | |
c8b97818 | 489 | |
771ed689 CM |
490 | /* the async work queues will take care of doing actual |
491 | * allocation on disk for these compressed pages, | |
492 | * and will submit them to the elevator. | |
493 | */ | |
494 | add_async_extent(async_cow, start, num_bytes, | |
495 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 496 | |
42dc7bab | 497 | if (start + num_bytes < end && start + num_bytes < actual_end) { |
771ed689 CM |
498 | start += num_bytes; |
499 | pages = NULL; | |
500 | cond_resched(); | |
501 | goto again; | |
502 | } | |
503 | } else { | |
f03d9301 | 504 | cleanup_and_bail_uncompressed: |
771ed689 CM |
505 | /* |
506 | * No compression, but we still need to write the pages in | |
507 | * the file we've been given so far. redirty the locked | |
508 | * page if it corresponds to our extent and set things up | |
509 | * for the async work queue to run cow_file_range to do | |
510 | * the normal delalloc dance | |
511 | */ | |
512 | if (page_offset(locked_page) >= start && | |
513 | page_offset(locked_page) <= end) { | |
514 | __set_page_dirty_nobuffers(locked_page); | |
515 | /* unlocked later on in the async handlers */ | |
516 | } | |
517 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
518 | *num_added += 1; | |
519 | } | |
3b951516 | 520 | |
771ed689 CM |
521 | out: |
522 | return 0; | |
523 | ||
524 | free_pages_out: | |
525 | for (i = 0; i < nr_pages_ret; i++) { | |
526 | WARN_ON(pages[i]->mapping); | |
527 | page_cache_release(pages[i]); | |
528 | } | |
d397712b | 529 | kfree(pages); |
771ed689 CM |
530 | |
531 | goto out; | |
532 | } | |
533 | ||
534 | /* | |
535 | * phase two of compressed writeback. This is the ordered portion | |
536 | * of the code, which only gets called in the order the work was | |
537 | * queued. We walk all the async extents created by compress_file_range | |
538 | * and send them down to the disk. | |
539 | */ | |
540 | static noinline int submit_compressed_extents(struct inode *inode, | |
541 | struct async_cow *async_cow) | |
542 | { | |
543 | struct async_extent *async_extent; | |
544 | u64 alloc_hint = 0; | |
545 | struct btrfs_trans_handle *trans; | |
546 | struct btrfs_key ins; | |
547 | struct extent_map *em; | |
548 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
549 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
550 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 551 | int ret = 0; |
771ed689 CM |
552 | |
553 | if (list_empty(&async_cow->extents)) | |
554 | return 0; | |
555 | ||
771ed689 | 556 | |
d397712b | 557 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
558 | async_extent = list_entry(async_cow->extents.next, |
559 | struct async_extent, list); | |
560 | list_del(&async_extent->list); | |
c8b97818 | 561 | |
771ed689 CM |
562 | io_tree = &BTRFS_I(inode)->io_tree; |
563 | ||
f5a84ee3 | 564 | retry: |
771ed689 CM |
565 | /* did the compression code fall back to uncompressed IO? */ |
566 | if (!async_extent->pages) { | |
567 | int page_started = 0; | |
568 | unsigned long nr_written = 0; | |
569 | ||
570 | lock_extent(io_tree, async_extent->start, | |
d397712b CM |
571 | async_extent->start + |
572 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
573 | |
574 | /* allocate blocks */ | |
f5a84ee3 JB |
575 | ret = cow_file_range(inode, async_cow->locked_page, |
576 | async_extent->start, | |
577 | async_extent->start + | |
578 | async_extent->ram_size - 1, | |
579 | &page_started, &nr_written, 0); | |
771ed689 CM |
580 | |
581 | /* | |
582 | * if page_started, cow_file_range inserted an | |
583 | * inline extent and took care of all the unlocking | |
584 | * and IO for us. Otherwise, we need to submit | |
585 | * all those pages down to the drive. | |
586 | */ | |
f5a84ee3 | 587 | if (!page_started && !ret) |
771ed689 CM |
588 | extent_write_locked_range(io_tree, |
589 | inode, async_extent->start, | |
d397712b | 590 | async_extent->start + |
771ed689 CM |
591 | async_extent->ram_size - 1, |
592 | btrfs_get_extent, | |
593 | WB_SYNC_ALL); | |
594 | kfree(async_extent); | |
595 | cond_resched(); | |
596 | continue; | |
597 | } | |
598 | ||
599 | lock_extent(io_tree, async_extent->start, | |
600 | async_extent->start + async_extent->ram_size - 1, | |
601 | GFP_NOFS); | |
771ed689 | 602 | |
c2167754 | 603 | trans = btrfs_join_transaction(root, 1); |
771ed689 CM |
604 | ret = btrfs_reserve_extent(trans, root, |
605 | async_extent->compressed_size, | |
606 | async_extent->compressed_size, | |
607 | 0, alloc_hint, | |
608 | (u64)-1, &ins, 1); | |
c2167754 YZ |
609 | btrfs_end_transaction(trans, root); |
610 | ||
f5a84ee3 JB |
611 | if (ret) { |
612 | int i; | |
613 | for (i = 0; i < async_extent->nr_pages; i++) { | |
614 | WARN_ON(async_extent->pages[i]->mapping); | |
615 | page_cache_release(async_extent->pages[i]); | |
616 | } | |
617 | kfree(async_extent->pages); | |
618 | async_extent->nr_pages = 0; | |
619 | async_extent->pages = NULL; | |
620 | unlock_extent(io_tree, async_extent->start, | |
621 | async_extent->start + | |
622 | async_extent->ram_size - 1, GFP_NOFS); | |
623 | goto retry; | |
624 | } | |
625 | ||
c2167754 YZ |
626 | /* |
627 | * here we're doing allocation and writeback of the | |
628 | * compressed pages | |
629 | */ | |
630 | btrfs_drop_extent_cache(inode, async_extent->start, | |
631 | async_extent->start + | |
632 | async_extent->ram_size - 1, 0); | |
633 | ||
771ed689 CM |
634 | em = alloc_extent_map(GFP_NOFS); |
635 | em->start = async_extent->start; | |
636 | em->len = async_extent->ram_size; | |
445a6944 | 637 | em->orig_start = em->start; |
c8b97818 | 638 | |
771ed689 CM |
639 | em->block_start = ins.objectid; |
640 | em->block_len = ins.offset; | |
641 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
642 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
643 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
644 | ||
d397712b | 645 | while (1) { |
890871be | 646 | write_lock(&em_tree->lock); |
771ed689 | 647 | ret = add_extent_mapping(em_tree, em); |
890871be | 648 | write_unlock(&em_tree->lock); |
771ed689 CM |
649 | if (ret != -EEXIST) { |
650 | free_extent_map(em); | |
651 | break; | |
652 | } | |
653 | btrfs_drop_extent_cache(inode, async_extent->start, | |
654 | async_extent->start + | |
655 | async_extent->ram_size - 1, 0); | |
656 | } | |
657 | ||
658 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
659 | ins.objectid, | |
660 | async_extent->ram_size, | |
661 | ins.offset, | |
662 | BTRFS_ORDERED_COMPRESSED); | |
663 | BUG_ON(ret); | |
664 | ||
771ed689 CM |
665 | /* |
666 | * clear dirty, set writeback and unlock the pages. | |
667 | */ | |
668 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
669 | &BTRFS_I(inode)->io_tree, |
670 | async_extent->start, | |
671 | async_extent->start + | |
672 | async_extent->ram_size - 1, | |
673 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
674 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 675 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 676 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
677 | |
678 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
679 | async_extent->start, |
680 | async_extent->ram_size, | |
681 | ins.objectid, | |
682 | ins.offset, async_extent->pages, | |
683 | async_extent->nr_pages); | |
771ed689 CM |
684 | |
685 | BUG_ON(ret); | |
771ed689 CM |
686 | alloc_hint = ins.objectid + ins.offset; |
687 | kfree(async_extent); | |
688 | cond_resched(); | |
689 | } | |
690 | ||
771ed689 CM |
691 | return 0; |
692 | } | |
693 | ||
694 | /* | |
695 | * when extent_io.c finds a delayed allocation range in the file, | |
696 | * the call backs end up in this code. The basic idea is to | |
697 | * allocate extents on disk for the range, and create ordered data structs | |
698 | * in ram to track those extents. | |
699 | * | |
700 | * locked_page is the page that writepage had locked already. We use | |
701 | * it to make sure we don't do extra locks or unlocks. | |
702 | * | |
703 | * *page_started is set to one if we unlock locked_page and do everything | |
704 | * required to start IO on it. It may be clean and already done with | |
705 | * IO when we return. | |
706 | */ | |
707 | static noinline int cow_file_range(struct inode *inode, | |
708 | struct page *locked_page, | |
709 | u64 start, u64 end, int *page_started, | |
710 | unsigned long *nr_written, | |
711 | int unlock) | |
712 | { | |
713 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
714 | struct btrfs_trans_handle *trans; | |
715 | u64 alloc_hint = 0; | |
716 | u64 num_bytes; | |
717 | unsigned long ram_size; | |
718 | u64 disk_num_bytes; | |
719 | u64 cur_alloc_size; | |
720 | u64 blocksize = root->sectorsize; | |
721 | u64 actual_end; | |
42dc7bab | 722 | u64 isize = i_size_read(inode); |
771ed689 CM |
723 | struct btrfs_key ins; |
724 | struct extent_map *em; | |
725 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
726 | int ret = 0; | |
727 | ||
728 | trans = btrfs_join_transaction(root, 1); | |
729 | BUG_ON(!trans); | |
730 | btrfs_set_trans_block_group(trans, inode); | |
731 | ||
42dc7bab | 732 | actual_end = min_t(u64, isize, end + 1); |
771ed689 CM |
733 | |
734 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
735 | num_bytes = max(blocksize, num_bytes); | |
736 | disk_num_bytes = num_bytes; | |
737 | ret = 0; | |
738 | ||
739 | if (start == 0) { | |
740 | /* lets try to make an inline extent */ | |
741 | ret = cow_file_range_inline(trans, root, inode, | |
742 | start, end, 0, NULL); | |
743 | if (ret == 0) { | |
744 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
745 | &BTRFS_I(inode)->io_tree, |
746 | start, end, NULL, | |
747 | EXTENT_CLEAR_UNLOCK_PAGE | | |
748 | EXTENT_CLEAR_UNLOCK | | |
749 | EXTENT_CLEAR_DELALLOC | | |
32c00aff | 750 | EXTENT_CLEAR_ACCOUNTING | |
a791e35e CM |
751 | EXTENT_CLEAR_DIRTY | |
752 | EXTENT_SET_WRITEBACK | | |
753 | EXTENT_END_WRITEBACK); | |
c2167754 | 754 | |
771ed689 CM |
755 | *nr_written = *nr_written + |
756 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
757 | *page_started = 1; | |
758 | ret = 0; | |
759 | goto out; | |
760 | } | |
761 | } | |
762 | ||
763 | BUG_ON(disk_num_bytes > | |
764 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
765 | ||
b917b7c3 CM |
766 | |
767 | read_lock(&BTRFS_I(inode)->extent_tree.lock); | |
768 | em = search_extent_mapping(&BTRFS_I(inode)->extent_tree, | |
769 | start, num_bytes); | |
770 | if (em) { | |
6346c939 JB |
771 | /* |
772 | * if block start isn't an actual block number then find the | |
773 | * first block in this inode and use that as a hint. If that | |
774 | * block is also bogus then just don't worry about it. | |
775 | */ | |
776 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
777 | free_extent_map(em); | |
778 | em = search_extent_mapping(em_tree, 0, 0); | |
779 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
780 | alloc_hint = em->block_start; | |
781 | if (em) | |
782 | free_extent_map(em); | |
783 | } else { | |
784 | alloc_hint = em->block_start; | |
785 | free_extent_map(em); | |
786 | } | |
b917b7c3 CM |
787 | } |
788 | read_unlock(&BTRFS_I(inode)->extent_tree.lock); | |
771ed689 CM |
789 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
790 | ||
d397712b | 791 | while (disk_num_bytes > 0) { |
a791e35e CM |
792 | unsigned long op; |
793 | ||
c8b97818 | 794 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 795 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 796 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 797 | (u64)-1, &ins, 1); |
d397712b CM |
798 | BUG_ON(ret); |
799 | ||
e6dcd2dc CM |
800 | em = alloc_extent_map(GFP_NOFS); |
801 | em->start = start; | |
445a6944 | 802 | em->orig_start = em->start; |
771ed689 CM |
803 | ram_size = ins.offset; |
804 | em->len = ins.offset; | |
c8b97818 | 805 | |
e6dcd2dc | 806 | em->block_start = ins.objectid; |
c8b97818 | 807 | em->block_len = ins.offset; |
e6dcd2dc | 808 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 809 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 810 | |
d397712b | 811 | while (1) { |
890871be | 812 | write_lock(&em_tree->lock); |
e6dcd2dc | 813 | ret = add_extent_mapping(em_tree, em); |
890871be | 814 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
815 | if (ret != -EEXIST) { |
816 | free_extent_map(em); | |
817 | break; | |
818 | } | |
819 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 820 | start + ram_size - 1, 0); |
e6dcd2dc CM |
821 | } |
822 | ||
98d20f67 | 823 | cur_alloc_size = ins.offset; |
e6dcd2dc | 824 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 825 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 826 | BUG_ON(ret); |
c8b97818 | 827 | |
17d217fe YZ |
828 | if (root->root_key.objectid == |
829 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
830 | ret = btrfs_reloc_clone_csums(inode, start, | |
831 | cur_alloc_size); | |
832 | BUG_ON(ret); | |
833 | } | |
834 | ||
d397712b | 835 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 836 | break; |
d397712b | 837 | |
c8b97818 CM |
838 | /* we're not doing compressed IO, don't unlock the first |
839 | * page (which the caller expects to stay locked), don't | |
840 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
841 | * |
842 | * Do set the Private2 bit so we know this page was properly | |
843 | * setup for writepage | |
c8b97818 | 844 | */ |
a791e35e CM |
845 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
846 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
847 | EXTENT_SET_PRIVATE2; | |
848 | ||
c8b97818 CM |
849 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
850 | start, start + ram_size - 1, | |
a791e35e | 851 | locked_page, op); |
c8b97818 | 852 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
853 | num_bytes -= cur_alloc_size; |
854 | alloc_hint = ins.objectid + ins.offset; | |
855 | start += cur_alloc_size; | |
b888db2b | 856 | } |
b888db2b | 857 | out: |
771ed689 | 858 | ret = 0; |
b888db2b | 859 | btrfs_end_transaction(trans, root); |
c8b97818 | 860 | |
be20aa9d | 861 | return ret; |
771ed689 | 862 | } |
c8b97818 | 863 | |
771ed689 CM |
864 | /* |
865 | * work queue call back to started compression on a file and pages | |
866 | */ | |
867 | static noinline void async_cow_start(struct btrfs_work *work) | |
868 | { | |
869 | struct async_cow *async_cow; | |
870 | int num_added = 0; | |
871 | async_cow = container_of(work, struct async_cow, work); | |
872 | ||
873 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
874 | async_cow->start, async_cow->end, async_cow, | |
875 | &num_added); | |
876 | if (num_added == 0) | |
877 | async_cow->inode = NULL; | |
878 | } | |
879 | ||
880 | /* | |
881 | * work queue call back to submit previously compressed pages | |
882 | */ | |
883 | static noinline void async_cow_submit(struct btrfs_work *work) | |
884 | { | |
885 | struct async_cow *async_cow; | |
886 | struct btrfs_root *root; | |
887 | unsigned long nr_pages; | |
888 | ||
889 | async_cow = container_of(work, struct async_cow, work); | |
890 | ||
891 | root = async_cow->root; | |
892 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
893 | PAGE_CACHE_SHIFT; | |
894 | ||
895 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
896 | ||
897 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
898 | 5 * 1042 * 1024 && | |
899 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
900 | wake_up(&root->fs_info->async_submit_wait); | |
901 | ||
d397712b | 902 | if (async_cow->inode) |
771ed689 | 903 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 904 | } |
c8b97818 | 905 | |
771ed689 CM |
906 | static noinline void async_cow_free(struct btrfs_work *work) |
907 | { | |
908 | struct async_cow *async_cow; | |
909 | async_cow = container_of(work, struct async_cow, work); | |
910 | kfree(async_cow); | |
911 | } | |
912 | ||
913 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
914 | u64 start, u64 end, int *page_started, | |
915 | unsigned long *nr_written) | |
916 | { | |
917 | struct async_cow *async_cow; | |
918 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
919 | unsigned long nr_pages; | |
920 | u64 cur_end; | |
921 | int limit = 10 * 1024 * 1042; | |
922 | ||
a3429ab7 CM |
923 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
924 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 925 | while (start < end) { |
771ed689 CM |
926 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
927 | async_cow->inode = inode; | |
928 | async_cow->root = root; | |
929 | async_cow->locked_page = locked_page; | |
930 | async_cow->start = start; | |
931 | ||
6cbff00f | 932 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
933 | cur_end = end; |
934 | else | |
935 | cur_end = min(end, start + 512 * 1024 - 1); | |
936 | ||
937 | async_cow->end = cur_end; | |
938 | INIT_LIST_HEAD(&async_cow->extents); | |
939 | ||
940 | async_cow->work.func = async_cow_start; | |
941 | async_cow->work.ordered_func = async_cow_submit; | |
942 | async_cow->work.ordered_free = async_cow_free; | |
943 | async_cow->work.flags = 0; | |
944 | ||
771ed689 CM |
945 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
946 | PAGE_CACHE_SHIFT; | |
947 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
948 | ||
949 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
950 | &async_cow->work); | |
951 | ||
952 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
953 | wait_event(root->fs_info->async_submit_wait, | |
954 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
955 | limit)); | |
956 | } | |
957 | ||
d397712b | 958 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
959 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
960 | wait_event(root->fs_info->async_submit_wait, | |
961 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
962 | 0)); | |
963 | } | |
964 | ||
965 | *nr_written += nr_pages; | |
966 | start = cur_end + 1; | |
967 | } | |
968 | *page_started = 1; | |
969 | return 0; | |
be20aa9d CM |
970 | } |
971 | ||
d397712b | 972 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
973 | u64 bytenr, u64 num_bytes) |
974 | { | |
975 | int ret; | |
976 | struct btrfs_ordered_sum *sums; | |
977 | LIST_HEAD(list); | |
978 | ||
07d400a6 YZ |
979 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
980 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
981 | if (ret == 0 && list_empty(&list)) |
982 | return 0; | |
983 | ||
984 | while (!list_empty(&list)) { | |
985 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
986 | list_del(&sums->list); | |
987 | kfree(sums); | |
988 | } | |
989 | return 1; | |
990 | } | |
991 | ||
d352ac68 CM |
992 | /* |
993 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
994 | * of the extents that exist in the file, and COWs the file as required. | |
995 | * | |
996 | * If no cow copies or snapshots exist, we write directly to the existing | |
997 | * blocks on disk | |
998 | */ | |
7f366cfe CM |
999 | static noinline int run_delalloc_nocow(struct inode *inode, |
1000 | struct page *locked_page, | |
771ed689 CM |
1001 | u64 start, u64 end, int *page_started, int force, |
1002 | unsigned long *nr_written) | |
be20aa9d | 1003 | { |
be20aa9d | 1004 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1005 | struct btrfs_trans_handle *trans; |
be20aa9d | 1006 | struct extent_buffer *leaf; |
be20aa9d | 1007 | struct btrfs_path *path; |
80ff3856 | 1008 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1009 | struct btrfs_key found_key; |
80ff3856 YZ |
1010 | u64 cow_start; |
1011 | u64 cur_offset; | |
1012 | u64 extent_end; | |
5d4f98a2 | 1013 | u64 extent_offset; |
80ff3856 YZ |
1014 | u64 disk_bytenr; |
1015 | u64 num_bytes; | |
1016 | int extent_type; | |
1017 | int ret; | |
d899e052 | 1018 | int type; |
80ff3856 YZ |
1019 | int nocow; |
1020 | int check_prev = 1; | |
be20aa9d CM |
1021 | |
1022 | path = btrfs_alloc_path(); | |
1023 | BUG_ON(!path); | |
7ea394f1 YZ |
1024 | trans = btrfs_join_transaction(root, 1); |
1025 | BUG_ON(!trans); | |
be20aa9d | 1026 | |
80ff3856 YZ |
1027 | cow_start = (u64)-1; |
1028 | cur_offset = start; | |
1029 | while (1) { | |
1030 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1031 | cur_offset, 0); | |
1032 | BUG_ON(ret < 0); | |
1033 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1034 | leaf = path->nodes[0]; | |
1035 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1036 | path->slots[0] - 1); | |
1037 | if (found_key.objectid == inode->i_ino && | |
1038 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1039 | path->slots[0]--; | |
1040 | } | |
1041 | check_prev = 0; | |
1042 | next_slot: | |
1043 | leaf = path->nodes[0]; | |
1044 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1045 | ret = btrfs_next_leaf(root, path); | |
1046 | if (ret < 0) | |
1047 | BUG_ON(1); | |
1048 | if (ret > 0) | |
1049 | break; | |
1050 | leaf = path->nodes[0]; | |
1051 | } | |
be20aa9d | 1052 | |
80ff3856 YZ |
1053 | nocow = 0; |
1054 | disk_bytenr = 0; | |
17d217fe | 1055 | num_bytes = 0; |
80ff3856 YZ |
1056 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1057 | ||
1058 | if (found_key.objectid > inode->i_ino || | |
1059 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1060 | found_key.offset > end) | |
1061 | break; | |
1062 | ||
1063 | if (found_key.offset > cur_offset) { | |
1064 | extent_end = found_key.offset; | |
e9061e21 | 1065 | extent_type = 0; |
80ff3856 YZ |
1066 | goto out_check; |
1067 | } | |
1068 | ||
1069 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1070 | struct btrfs_file_extent_item); | |
1071 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1072 | ||
d899e052 YZ |
1073 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1074 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1075 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1076 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1077 | extent_end = found_key.offset + |
1078 | btrfs_file_extent_num_bytes(leaf, fi); | |
1079 | if (extent_end <= start) { | |
1080 | path->slots[0]++; | |
1081 | goto next_slot; | |
1082 | } | |
17d217fe YZ |
1083 | if (disk_bytenr == 0) |
1084 | goto out_check; | |
80ff3856 YZ |
1085 | if (btrfs_file_extent_compression(leaf, fi) || |
1086 | btrfs_file_extent_encryption(leaf, fi) || | |
1087 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1088 | goto out_check; | |
d899e052 YZ |
1089 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1090 | goto out_check; | |
d2fb3437 | 1091 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1092 | goto out_check; |
17d217fe | 1093 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1094 | found_key.offset - |
1095 | extent_offset, disk_bytenr)) | |
17d217fe | 1096 | goto out_check; |
5d4f98a2 | 1097 | disk_bytenr += extent_offset; |
17d217fe YZ |
1098 | disk_bytenr += cur_offset - found_key.offset; |
1099 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1100 | /* | |
1101 | * force cow if csum exists in the range. | |
1102 | * this ensure that csum for a given extent are | |
1103 | * either valid or do not exist. | |
1104 | */ | |
1105 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1106 | goto out_check; | |
80ff3856 YZ |
1107 | nocow = 1; |
1108 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1109 | extent_end = found_key.offset + | |
1110 | btrfs_file_extent_inline_len(leaf, fi); | |
1111 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1112 | } else { | |
1113 | BUG_ON(1); | |
1114 | } | |
1115 | out_check: | |
1116 | if (extent_end <= start) { | |
1117 | path->slots[0]++; | |
1118 | goto next_slot; | |
1119 | } | |
1120 | if (!nocow) { | |
1121 | if (cow_start == (u64)-1) | |
1122 | cow_start = cur_offset; | |
1123 | cur_offset = extent_end; | |
1124 | if (cur_offset > end) | |
1125 | break; | |
1126 | path->slots[0]++; | |
1127 | goto next_slot; | |
7ea394f1 YZ |
1128 | } |
1129 | ||
1130 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1131 | if (cow_start != (u64)-1) { |
1132 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1133 | found_key.offset - 1, page_started, |
1134 | nr_written, 1); | |
80ff3856 YZ |
1135 | BUG_ON(ret); |
1136 | cow_start = (u64)-1; | |
7ea394f1 | 1137 | } |
80ff3856 | 1138 | |
d899e052 YZ |
1139 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1140 | struct extent_map *em; | |
1141 | struct extent_map_tree *em_tree; | |
1142 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1143 | em = alloc_extent_map(GFP_NOFS); | |
1144 | em->start = cur_offset; | |
445a6944 | 1145 | em->orig_start = em->start; |
d899e052 YZ |
1146 | em->len = num_bytes; |
1147 | em->block_len = num_bytes; | |
1148 | em->block_start = disk_bytenr; | |
1149 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1150 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1151 | while (1) { | |
890871be | 1152 | write_lock(&em_tree->lock); |
d899e052 | 1153 | ret = add_extent_mapping(em_tree, em); |
890871be | 1154 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1155 | if (ret != -EEXIST) { |
1156 | free_extent_map(em); | |
1157 | break; | |
1158 | } | |
1159 | btrfs_drop_extent_cache(inode, em->start, | |
1160 | em->start + em->len - 1, 0); | |
1161 | } | |
1162 | type = BTRFS_ORDERED_PREALLOC; | |
1163 | } else { | |
1164 | type = BTRFS_ORDERED_NOCOW; | |
1165 | } | |
80ff3856 YZ |
1166 | |
1167 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1168 | num_bytes, num_bytes, type); |
1169 | BUG_ON(ret); | |
771ed689 | 1170 | |
d899e052 | 1171 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1172 | cur_offset, cur_offset + num_bytes - 1, |
1173 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1174 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1175 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1176 | cur_offset = extent_end; |
1177 | if (cur_offset > end) | |
1178 | break; | |
be20aa9d | 1179 | } |
80ff3856 YZ |
1180 | btrfs_release_path(root, path); |
1181 | ||
1182 | if (cur_offset <= end && cow_start == (u64)-1) | |
1183 | cow_start = cur_offset; | |
1184 | if (cow_start != (u64)-1) { | |
1185 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1186 | page_started, nr_written, 1); |
80ff3856 YZ |
1187 | BUG_ON(ret); |
1188 | } | |
1189 | ||
1190 | ret = btrfs_end_transaction(trans, root); | |
1191 | BUG_ON(ret); | |
7ea394f1 | 1192 | btrfs_free_path(path); |
80ff3856 | 1193 | return 0; |
be20aa9d CM |
1194 | } |
1195 | ||
d352ac68 CM |
1196 | /* |
1197 | * extent_io.c call back to do delayed allocation processing | |
1198 | */ | |
c8b97818 | 1199 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1200 | u64 start, u64 end, int *page_started, |
1201 | unsigned long *nr_written) | |
be20aa9d | 1202 | { |
be20aa9d | 1203 | int ret; |
7f366cfe | 1204 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1205 | |
6cbff00f | 1206 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1207 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1208 | page_started, 1, nr_written); |
6cbff00f | 1209 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1210 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1211 | page_started, 0, nr_written); |
7f366cfe CM |
1212 | else if (!btrfs_test_opt(root, COMPRESS)) |
1213 | ret = cow_file_range(inode, locked_page, start, end, | |
1214 | page_started, nr_written, 1); | |
be20aa9d | 1215 | else |
771ed689 | 1216 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1217 | page_started, nr_written); |
b888db2b CM |
1218 | return ret; |
1219 | } | |
1220 | ||
9ed74f2d JB |
1221 | static int btrfs_split_extent_hook(struct inode *inode, |
1222 | struct extent_state *orig, u64 split) | |
1223 | { | |
1224 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1225 | u64 size; | |
1226 | ||
1227 | if (!(orig->state & EXTENT_DELALLOC)) | |
1228 | return 0; | |
1229 | ||
1230 | size = orig->end - orig->start + 1; | |
1231 | if (size > root->fs_info->max_extent) { | |
1232 | u64 num_extents; | |
1233 | u64 new_size; | |
1234 | ||
1235 | new_size = orig->end - split + 1; | |
1236 | num_extents = div64_u64(size + root->fs_info->max_extent - 1, | |
1237 | root->fs_info->max_extent); | |
1238 | ||
1239 | /* | |
32c00aff JB |
1240 | * if we break a large extent up then leave oustanding_extents |
1241 | * be, since we've already accounted for the large extent. | |
9ed74f2d JB |
1242 | */ |
1243 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1244 | root->fs_info->max_extent) < num_extents) | |
1245 | return 0; | |
1246 | } | |
1247 | ||
32c00aff JB |
1248 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1249 | BTRFS_I(inode)->outstanding_extents++; | |
1250 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1251 | |
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1257 | * extents so we can keep track of new extents that are just merged onto old | |
1258 | * extents, such as when we are doing sequential writes, so we can properly | |
1259 | * account for the metadata space we'll need. | |
1260 | */ | |
1261 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1262 | struct extent_state *new, | |
1263 | struct extent_state *other) | |
1264 | { | |
1265 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1266 | u64 new_size, old_size; | |
1267 | u64 num_extents; | |
1268 | ||
1269 | /* not delalloc, ignore it */ | |
1270 | if (!(other->state & EXTENT_DELALLOC)) | |
1271 | return 0; | |
1272 | ||
1273 | old_size = other->end - other->start + 1; | |
1274 | if (new->start < other->start) | |
1275 | new_size = other->end - new->start + 1; | |
1276 | else | |
1277 | new_size = new->end - other->start + 1; | |
1278 | ||
1279 | /* we're not bigger than the max, unreserve the space and go */ | |
1280 | if (new_size <= root->fs_info->max_extent) { | |
32c00aff JB |
1281 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1282 | BTRFS_I(inode)->outstanding_extents--; | |
1283 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1284 | return 0; |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * If we grew by another max_extent, just return, we want to keep that | |
1289 | * reserved amount. | |
1290 | */ | |
1291 | num_extents = div64_u64(old_size + root->fs_info->max_extent - 1, | |
1292 | root->fs_info->max_extent); | |
1293 | if (div64_u64(new_size + root->fs_info->max_extent - 1, | |
1294 | root->fs_info->max_extent) > num_extents) | |
1295 | return 0; | |
1296 | ||
32c00aff JB |
1297 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1298 | BTRFS_I(inode)->outstanding_extents--; | |
1299 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
1300 | |
1301 | return 0; | |
1302 | } | |
1303 | ||
d352ac68 CM |
1304 | /* |
1305 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1306 | * bytes in this file, and to maintain the list of inodes that | |
1307 | * have pending delalloc work to be done. | |
1308 | */ | |
b2950863 | 1309 | static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1310 | unsigned long old, unsigned long bits) |
291d673e | 1311 | { |
9ed74f2d | 1312 | |
75eff68e CM |
1313 | /* |
1314 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1315 | * but in this case, we are only testeing for the DELALLOC | |
1316 | * bit, which is only set or cleared with irqs on | |
1317 | */ | |
b0c68f8b | 1318 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1319 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9ed74f2d | 1320 | |
32c00aff JB |
1321 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
1322 | BTRFS_I(inode)->outstanding_extents++; | |
1323 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
6a63209f | 1324 | btrfs_delalloc_reserve_space(root, inode, end - start + 1); |
75eff68e | 1325 | spin_lock(&root->fs_info->delalloc_lock); |
9069218d | 1326 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1327 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1328 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1329 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1330 | &root->fs_info->delalloc_inodes); | |
1331 | } | |
75eff68e | 1332 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1333 | } |
1334 | return 0; | |
1335 | } | |
1336 | ||
d352ac68 CM |
1337 | /* |
1338 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1339 | */ | |
9ed74f2d JB |
1340 | static int btrfs_clear_bit_hook(struct inode *inode, |
1341 | struct extent_state *state, unsigned long bits) | |
291d673e | 1342 | { |
75eff68e CM |
1343 | /* |
1344 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1345 | * but in this case, we are only testeing for the DELALLOC | |
1346 | * bit, which is only set or cleared with irqs on | |
1347 | */ | |
9ed74f2d | 1348 | if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1349 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1350 | |
32c00aff JB |
1351 | if (bits & EXTENT_DO_ACCOUNTING) { |
1352 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
1353 | BTRFS_I(inode)->outstanding_extents--; | |
1354 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
1355 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
1356 | } | |
9ed74f2d | 1357 | |
75eff68e | 1358 | spin_lock(&root->fs_info->delalloc_lock); |
9ed74f2d JB |
1359 | if (state->end - state->start + 1 > |
1360 | root->fs_info->delalloc_bytes) { | |
d397712b CM |
1361 | printk(KERN_INFO "btrfs warning: delalloc account " |
1362 | "%llu %llu\n", | |
9ed74f2d JB |
1363 | (unsigned long long) |
1364 | state->end - state->start + 1, | |
d397712b CM |
1365 | (unsigned long long) |
1366 | root->fs_info->delalloc_bytes); | |
6a63209f | 1367 | btrfs_delalloc_free_space(root, inode, (u64)-1); |
b0c68f8b | 1368 | root->fs_info->delalloc_bytes = 0; |
9069218d | 1369 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b | 1370 | } else { |
6a63209f | 1371 | btrfs_delalloc_free_space(root, inode, |
9ed74f2d JB |
1372 | state->end - |
1373 | state->start + 1); | |
1374 | root->fs_info->delalloc_bytes -= state->end - | |
1375 | state->start + 1; | |
1376 | BTRFS_I(inode)->delalloc_bytes -= state->end - | |
1377 | state->start + 1; | |
b0c68f8b | 1378 | } |
ea8c2819 CM |
1379 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1380 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1381 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1382 | } | |
75eff68e | 1383 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1384 | } |
1385 | return 0; | |
1386 | } | |
1387 | ||
d352ac68 CM |
1388 | /* |
1389 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1390 | * we don't create bios that span stripes or chunks | |
1391 | */ | |
239b14b3 | 1392 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1393 | size_t size, struct bio *bio, |
1394 | unsigned long bio_flags) | |
239b14b3 CM |
1395 | { |
1396 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1397 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1398 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1399 | u64 length = 0; |
1400 | u64 map_length; | |
239b14b3 CM |
1401 | int ret; |
1402 | ||
771ed689 CM |
1403 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1404 | return 0; | |
1405 | ||
f2d8d74d | 1406 | length = bio->bi_size; |
239b14b3 CM |
1407 | map_tree = &root->fs_info->mapping_tree; |
1408 | map_length = length; | |
cea9e445 | 1409 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1410 | &map_length, NULL, 0); |
cea9e445 | 1411 | |
d397712b | 1412 | if (map_length < length + size) |
239b14b3 | 1413 | return 1; |
239b14b3 CM |
1414 | return 0; |
1415 | } | |
1416 | ||
d352ac68 CM |
1417 | /* |
1418 | * in order to insert checksums into the metadata in large chunks, | |
1419 | * we wait until bio submission time. All the pages in the bio are | |
1420 | * checksummed and sums are attached onto the ordered extent record. | |
1421 | * | |
1422 | * At IO completion time the cums attached on the ordered extent record | |
1423 | * are inserted into the btree | |
1424 | */ | |
d397712b CM |
1425 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1426 | struct bio *bio, int mirror_num, | |
1427 | unsigned long bio_flags) | |
065631f6 | 1428 | { |
065631f6 | 1429 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1430 | int ret = 0; |
e015640f | 1431 | |
d20f7043 | 1432 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1433 | BUG_ON(ret); |
4a69a410 CM |
1434 | return 0; |
1435 | } | |
e015640f | 1436 | |
4a69a410 CM |
1437 | /* |
1438 | * in order to insert checksums into the metadata in large chunks, | |
1439 | * we wait until bio submission time. All the pages in the bio are | |
1440 | * checksummed and sums are attached onto the ordered extent record. | |
1441 | * | |
1442 | * At IO completion time the cums attached on the ordered extent record | |
1443 | * are inserted into the btree | |
1444 | */ | |
b2950863 | 1445 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
4a69a410 CM |
1446 | int mirror_num, unsigned long bio_flags) |
1447 | { | |
1448 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1449 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1450 | } |
1451 | ||
d352ac68 | 1452 | /* |
cad321ad CM |
1453 | * extent_io.c submission hook. This does the right thing for csum calculation |
1454 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1455 | */ |
b2950863 | 1456 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1457 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1458 | { |
1459 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1460 | int ret = 0; | |
19b9bdb0 | 1461 | int skip_sum; |
44b8bd7e | 1462 | |
6cbff00f | 1463 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1464 | |
e6dcd2dc CM |
1465 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1466 | BUG_ON(ret); | |
065631f6 | 1467 | |
4d1b5fb4 | 1468 | if (!(rw & (1 << BIO_RW))) { |
d20f7043 | 1469 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1470 | return btrfs_submit_compressed_read(inode, bio, |
1471 | mirror_num, bio_flags); | |
d20f7043 CM |
1472 | } else if (!skip_sum) |
1473 | btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
4d1b5fb4 | 1474 | goto mapit; |
19b9bdb0 | 1475 | } else if (!skip_sum) { |
17d217fe YZ |
1476 | /* csum items have already been cloned */ |
1477 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1478 | goto mapit; | |
19b9bdb0 CM |
1479 | /* we're doing a write, do the async checksumming */ |
1480 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1481 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1482 | bio_flags, __btrfs_submit_bio_start, |
1483 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1484 | } |
1485 | ||
0b86a832 | 1486 | mapit: |
8b712842 | 1487 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1488 | } |
6885f308 | 1489 | |
d352ac68 CM |
1490 | /* |
1491 | * given a list of ordered sums record them in the inode. This happens | |
1492 | * at IO completion time based on sums calculated at bio submission time. | |
1493 | */ | |
ba1da2f4 | 1494 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1495 | struct inode *inode, u64 file_offset, |
1496 | struct list_head *list) | |
1497 | { | |
e6dcd2dc CM |
1498 | struct btrfs_ordered_sum *sum; |
1499 | ||
1500 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1501 | |
1502 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1503 | btrfs_csum_file_blocks(trans, |
1504 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1505 | } |
1506 | return 0; | |
1507 | } | |
1508 | ||
ea8c2819 CM |
1509 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1510 | { | |
d397712b | 1511 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1512 | WARN_ON(1); |
ea8c2819 CM |
1513 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1514 | GFP_NOFS); | |
1515 | } | |
1516 | ||
d352ac68 | 1517 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1518 | struct btrfs_writepage_fixup { |
1519 | struct page *page; | |
1520 | struct btrfs_work work; | |
1521 | }; | |
1522 | ||
b2950863 | 1523 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1524 | { |
1525 | struct btrfs_writepage_fixup *fixup; | |
1526 | struct btrfs_ordered_extent *ordered; | |
1527 | struct page *page; | |
1528 | struct inode *inode; | |
1529 | u64 page_start; | |
1530 | u64 page_end; | |
1531 | ||
1532 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1533 | page = fixup->page; | |
4a096752 | 1534 | again: |
247e743c CM |
1535 | lock_page(page); |
1536 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1537 | ClearPageChecked(page); | |
1538 | goto out_page; | |
1539 | } | |
1540 | ||
1541 | inode = page->mapping->host; | |
1542 | page_start = page_offset(page); | |
1543 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1544 | ||
1545 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1546 | |
1547 | /* already ordered? We're done */ | |
8b62b72b | 1548 | if (PagePrivate2(page)) |
247e743c | 1549 | goto out; |
4a096752 CM |
1550 | |
1551 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1552 | if (ordered) { | |
1553 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1554 | page_end, GFP_NOFS); | |
1555 | unlock_page(page); | |
1556 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1557 | goto again; | |
1558 | } | |
247e743c | 1559 | |
ea8c2819 | 1560 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1561 | ClearPageChecked(page); |
1562 | out: | |
1563 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1564 | out_page: | |
1565 | unlock_page(page); | |
1566 | page_cache_release(page); | |
1567 | } | |
1568 | ||
1569 | /* | |
1570 | * There are a few paths in the higher layers of the kernel that directly | |
1571 | * set the page dirty bit without asking the filesystem if it is a | |
1572 | * good idea. This causes problems because we want to make sure COW | |
1573 | * properly happens and the data=ordered rules are followed. | |
1574 | * | |
c8b97818 | 1575 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1576 | * hasn't been properly setup for IO. We kick off an async process |
1577 | * to fix it up. The async helper will wait for ordered extents, set | |
1578 | * the delalloc bit and make it safe to write the page. | |
1579 | */ | |
b2950863 | 1580 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1581 | { |
1582 | struct inode *inode = page->mapping->host; | |
1583 | struct btrfs_writepage_fixup *fixup; | |
1584 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1585 | |
8b62b72b CM |
1586 | /* this page is properly in the ordered list */ |
1587 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1588 | return 0; |
1589 | ||
1590 | if (PageChecked(page)) | |
1591 | return -EAGAIN; | |
1592 | ||
1593 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1594 | if (!fixup) | |
1595 | return -EAGAIN; | |
f421950f | 1596 | |
247e743c CM |
1597 | SetPageChecked(page); |
1598 | page_cache_get(page); | |
1599 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1600 | fixup->page = page; | |
1601 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1602 | return -EAGAIN; | |
1603 | } | |
1604 | ||
d899e052 YZ |
1605 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1606 | struct inode *inode, u64 file_pos, | |
1607 | u64 disk_bytenr, u64 disk_num_bytes, | |
1608 | u64 num_bytes, u64 ram_bytes, | |
1609 | u8 compression, u8 encryption, | |
1610 | u16 other_encoding, int extent_type) | |
1611 | { | |
1612 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1613 | struct btrfs_file_extent_item *fi; | |
1614 | struct btrfs_path *path; | |
1615 | struct extent_buffer *leaf; | |
1616 | struct btrfs_key ins; | |
1617 | u64 hint; | |
1618 | int ret; | |
1619 | ||
1620 | path = btrfs_alloc_path(); | |
1621 | BUG_ON(!path); | |
1622 | ||
b9473439 | 1623 | path->leave_spinning = 1; |
a1ed835e CM |
1624 | |
1625 | /* | |
1626 | * we may be replacing one extent in the tree with another. | |
1627 | * The new extent is pinned in the extent map, and we don't want | |
1628 | * to drop it from the cache until it is completely in the btree. | |
1629 | * | |
1630 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1631 | * the caller is expected to unpin it and allow it to be merged | |
1632 | * with the others. | |
1633 | */ | |
920bbbfb YZ |
1634 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1635 | &hint, 0); | |
d899e052 YZ |
1636 | BUG_ON(ret); |
1637 | ||
1638 | ins.objectid = inode->i_ino; | |
1639 | ins.offset = file_pos; | |
1640 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1641 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1642 | BUG_ON(ret); | |
1643 | leaf = path->nodes[0]; | |
1644 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1645 | struct btrfs_file_extent_item); | |
1646 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1647 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1648 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1649 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1650 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1651 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1652 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1653 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1654 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1655 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1656 | |
1657 | btrfs_unlock_up_safe(path, 1); | |
1658 | btrfs_set_lock_blocking(leaf); | |
1659 | ||
d899e052 YZ |
1660 | btrfs_mark_buffer_dirty(leaf); |
1661 | ||
1662 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1663 | |
1664 | ins.objectid = disk_bytenr; | |
1665 | ins.offset = disk_num_bytes; | |
1666 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1667 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1668 | root->root_key.objectid, | |
1669 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1670 | BUG_ON(ret); |
d899e052 | 1671 | btrfs_free_path(path); |
b9473439 | 1672 | |
d899e052 YZ |
1673 | return 0; |
1674 | } | |
1675 | ||
5d13a98f CM |
1676 | /* |
1677 | * helper function for btrfs_finish_ordered_io, this | |
1678 | * just reads in some of the csum leaves to prime them into ram | |
1679 | * before we start the transaction. It limits the amount of btree | |
1680 | * reads required while inside the transaction. | |
1681 | */ | |
1682 | static noinline void reada_csum(struct btrfs_root *root, | |
1683 | struct btrfs_path *path, | |
1684 | struct btrfs_ordered_extent *ordered_extent) | |
1685 | { | |
1686 | struct btrfs_ordered_sum *sum; | |
1687 | u64 bytenr; | |
1688 | ||
1689 | sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum, | |
1690 | list); | |
1691 | bytenr = sum->sums[0].bytenr; | |
1692 | ||
1693 | /* | |
1694 | * we don't care about the results, the point of this search is | |
1695 | * just to get the btree leaves into ram | |
1696 | */ | |
1697 | btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0); | |
1698 | } | |
1699 | ||
d352ac68 CM |
1700 | /* as ordered data IO finishes, this gets called so we can finish |
1701 | * an ordered extent if the range of bytes in the file it covers are | |
1702 | * fully written. | |
1703 | */ | |
211f90e6 | 1704 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1705 | { |
e6dcd2dc CM |
1706 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1707 | struct btrfs_trans_handle *trans; | |
5d13a98f | 1708 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1709 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
b7ec40d7 | 1710 | struct btrfs_path *path; |
d899e052 | 1711 | int compressed = 0; |
e6dcd2dc CM |
1712 | int ret; |
1713 | ||
1714 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1715 | if (!ret) |
e6dcd2dc | 1716 | return 0; |
e6dcd2dc | 1717 | |
b7ec40d7 CM |
1718 | /* |
1719 | * before we join the transaction, try to do some of our IO. | |
1720 | * This will limit the amount of IO that we have to do with | |
1721 | * the transaction running. We're unlikely to need to do any | |
1722 | * IO if the file extents are new, the disk_i_size checks | |
1723 | * covers the most common case. | |
1724 | */ | |
1725 | if (start < BTRFS_I(inode)->disk_i_size) { | |
1726 | path = btrfs_alloc_path(); | |
1727 | if (path) { | |
1728 | ret = btrfs_lookup_file_extent(NULL, root, path, | |
1729 | inode->i_ino, | |
1730 | start, 0); | |
5d13a98f CM |
1731 | ordered_extent = btrfs_lookup_ordered_extent(inode, |
1732 | start); | |
1733 | if (!list_empty(&ordered_extent->list)) { | |
1734 | btrfs_release_path(root, path); | |
1735 | reada_csum(root, path, ordered_extent); | |
1736 | } | |
b7ec40d7 CM |
1737 | btrfs_free_path(path); |
1738 | } | |
1739 | } | |
1740 | ||
5d13a98f CM |
1741 | if (!ordered_extent) |
1742 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
e6dcd2dc | 1743 | BUG_ON(!ordered_extent); |
c2167754 YZ |
1744 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1745 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1746 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1747 | if (!ret) { | |
1748 | trans = btrfs_join_transaction(root, 1); | |
1749 | ret = btrfs_update_inode(trans, root, inode); | |
1750 | BUG_ON(ret); | |
1751 | btrfs_end_transaction(trans, root); | |
1752 | } | |
1753 | goto out; | |
1754 | } | |
e6dcd2dc CM |
1755 | |
1756 | lock_extent(io_tree, ordered_extent->file_offset, | |
1757 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1758 | GFP_NOFS); | |
1759 | ||
c2167754 YZ |
1760 | trans = btrfs_join_transaction(root, 1); |
1761 | ||
c8b97818 | 1762 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1763 | compressed = 1; |
1764 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1765 | BUG_ON(compressed); | |
920bbbfb | 1766 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1767 | ordered_extent->file_offset, |
1768 | ordered_extent->file_offset + | |
1769 | ordered_extent->len); | |
1770 | BUG_ON(ret); | |
1771 | } else { | |
1772 | ret = insert_reserved_file_extent(trans, inode, | |
1773 | ordered_extent->file_offset, | |
1774 | ordered_extent->start, | |
1775 | ordered_extent->disk_len, | |
1776 | ordered_extent->len, | |
1777 | ordered_extent->len, | |
1778 | compressed, 0, 0, | |
1779 | BTRFS_FILE_EXTENT_REG); | |
a1ed835e CM |
1780 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1781 | ordered_extent->file_offset, | |
1782 | ordered_extent->len); | |
d899e052 YZ |
1783 | BUG_ON(ret); |
1784 | } | |
e6dcd2dc CM |
1785 | unlock_extent(io_tree, ordered_extent->file_offset, |
1786 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1787 | GFP_NOFS); | |
1788 | add_pending_csums(trans, inode, ordered_extent->file_offset, | |
1789 | &ordered_extent->list); | |
1790 | ||
c2167754 YZ |
1791 | /* this also removes the ordered extent from the tree */ |
1792 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1793 | ret = btrfs_update_inode(trans, root, inode); | |
1794 | BUG_ON(ret); | |
1795 | btrfs_end_transaction(trans, root); | |
1796 | out: | |
e6dcd2dc CM |
1797 | /* once for us */ |
1798 | btrfs_put_ordered_extent(ordered_extent); | |
1799 | /* once for the tree */ | |
1800 | btrfs_put_ordered_extent(ordered_extent); | |
1801 | ||
e6dcd2dc CM |
1802 | return 0; |
1803 | } | |
1804 | ||
b2950863 | 1805 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1806 | struct extent_state *state, int uptodate) |
1807 | { | |
8b62b72b | 1808 | ClearPagePrivate2(page); |
211f90e6 CM |
1809 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1810 | } | |
1811 | ||
d352ac68 CM |
1812 | /* |
1813 | * When IO fails, either with EIO or csum verification fails, we | |
1814 | * try other mirrors that might have a good copy of the data. This | |
1815 | * io_failure_record is used to record state as we go through all the | |
1816 | * mirrors. If another mirror has good data, the page is set up to date | |
1817 | * and things continue. If a good mirror can't be found, the original | |
1818 | * bio end_io callback is called to indicate things have failed. | |
1819 | */ | |
7e38326f CM |
1820 | struct io_failure_record { |
1821 | struct page *page; | |
1822 | u64 start; | |
1823 | u64 len; | |
1824 | u64 logical; | |
d20f7043 | 1825 | unsigned long bio_flags; |
7e38326f CM |
1826 | int last_mirror; |
1827 | }; | |
1828 | ||
b2950863 | 1829 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1830 | struct page *page, u64 start, u64 end, |
1831 | struct extent_state *state) | |
7e38326f CM |
1832 | { |
1833 | struct io_failure_record *failrec = NULL; | |
1834 | u64 private; | |
1835 | struct extent_map *em; | |
1836 | struct inode *inode = page->mapping->host; | |
1837 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1838 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1839 | struct bio *bio; |
1840 | int num_copies; | |
1841 | int ret; | |
1259ab75 | 1842 | int rw; |
7e38326f CM |
1843 | u64 logical; |
1844 | ||
1845 | ret = get_state_private(failure_tree, start, &private); | |
1846 | if (ret) { | |
7e38326f CM |
1847 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1848 | if (!failrec) | |
1849 | return -ENOMEM; | |
1850 | failrec->start = start; | |
1851 | failrec->len = end - start + 1; | |
1852 | failrec->last_mirror = 0; | |
d20f7043 | 1853 | failrec->bio_flags = 0; |
7e38326f | 1854 | |
890871be | 1855 | read_lock(&em_tree->lock); |
3b951516 CM |
1856 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1857 | if (em->start > start || em->start + em->len < start) { | |
1858 | free_extent_map(em); | |
1859 | em = NULL; | |
1860 | } | |
890871be | 1861 | read_unlock(&em_tree->lock); |
7e38326f CM |
1862 | |
1863 | if (!em || IS_ERR(em)) { | |
1864 | kfree(failrec); | |
1865 | return -EIO; | |
1866 | } | |
1867 | logical = start - em->start; | |
1868 | logical = em->block_start + logical; | |
d20f7043 CM |
1869 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1870 | logical = em->block_start; | |
1871 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
1872 | } | |
7e38326f CM |
1873 | failrec->logical = logical; |
1874 | free_extent_map(em); | |
1875 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1876 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1877 | set_state_private(failure_tree, start, |
1878 | (u64)(unsigned long)failrec); | |
7e38326f | 1879 | } else { |
587f7704 | 1880 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1881 | } |
1882 | num_copies = btrfs_num_copies( | |
1883 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1884 | failrec->logical, failrec->len); | |
1885 | failrec->last_mirror++; | |
1886 | if (!state) { | |
cad321ad | 1887 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1888 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1889 | failrec->start, | |
1890 | EXTENT_LOCKED); | |
1891 | if (state && state->start != failrec->start) | |
1892 | state = NULL; | |
cad321ad | 1893 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1894 | } |
1895 | if (!state || failrec->last_mirror > num_copies) { | |
1896 | set_state_private(failure_tree, failrec->start, 0); | |
1897 | clear_extent_bits(failure_tree, failrec->start, | |
1898 | failrec->start + failrec->len - 1, | |
1899 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1900 | kfree(failrec); | |
1901 | return -EIO; | |
1902 | } | |
1903 | bio = bio_alloc(GFP_NOFS, 1); | |
1904 | bio->bi_private = state; | |
1905 | bio->bi_end_io = failed_bio->bi_end_io; | |
1906 | bio->bi_sector = failrec->logical >> 9; | |
1907 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1908 | bio->bi_size = 0; |
d20f7043 | 1909 | |
7e38326f | 1910 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1911 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1912 | rw = WRITE; | |
1913 | else | |
1914 | rw = READ; | |
1915 | ||
1916 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 | 1917 | failrec->last_mirror, |
d20f7043 | 1918 | failrec->bio_flags); |
1259ab75 CM |
1919 | return 0; |
1920 | } | |
1921 | ||
d352ac68 CM |
1922 | /* |
1923 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1924 | * to see if we need to process or clean up an io_failure_record | |
1925 | */ | |
b2950863 | 1926 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1927 | { |
1928 | u64 private; | |
1929 | u64 private_failure; | |
1930 | struct io_failure_record *failure; | |
1931 | int ret; | |
1932 | ||
1933 | private = 0; | |
1934 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1935 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1936 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1937 | start, &private_failure); | |
1938 | if (ret == 0) { | |
1939 | failure = (struct io_failure_record *)(unsigned long) | |
1940 | private_failure; | |
1941 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1942 | failure->start, 0); | |
1943 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1944 | failure->start, | |
1945 | failure->start + failure->len - 1, | |
1946 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1947 | GFP_NOFS); | |
1948 | kfree(failure); | |
1949 | } | |
1950 | } | |
7e38326f CM |
1951 | return 0; |
1952 | } | |
1953 | ||
d352ac68 CM |
1954 | /* |
1955 | * when reads are done, we need to check csums to verify the data is correct | |
1956 | * if there's a match, we allow the bio to finish. If not, we go through | |
1957 | * the io_failure_record routines to find good copies | |
1958 | */ | |
b2950863 | 1959 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1960 | struct extent_state *state) |
07157aac | 1961 | { |
35ebb934 | 1962 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1963 | struct inode *inode = page->mapping->host; |
d1310b2e | 1964 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1965 | char *kaddr; |
aadfeb6e | 1966 | u64 private = ~(u32)0; |
07157aac | 1967 | int ret; |
ff79f819 CM |
1968 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1969 | u32 csum = ~(u32)0; | |
d1310b2e | 1970 | |
d20f7043 CM |
1971 | if (PageChecked(page)) { |
1972 | ClearPageChecked(page); | |
1973 | goto good; | |
1974 | } | |
6cbff00f CH |
1975 | |
1976 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1977 | return 0; |
1978 | ||
1979 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1980 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1981 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1982 | GFP_NOFS); | |
b6cda9bc | 1983 | return 0; |
17d217fe | 1984 | } |
d20f7043 | 1985 | |
c2e639f0 | 1986 | if (state && state->start == start) { |
70dec807 CM |
1987 | private = state->private; |
1988 | ret = 0; | |
1989 | } else { | |
1990 | ret = get_state_private(io_tree, start, &private); | |
1991 | } | |
9ab86c8e | 1992 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1993 | if (ret) |
07157aac | 1994 | goto zeroit; |
d397712b | 1995 | |
ff79f819 CM |
1996 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1997 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1998 | if (csum != private) |
07157aac | 1999 | goto zeroit; |
d397712b | 2000 | |
9ab86c8e | 2001 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 2002 | good: |
7e38326f CM |
2003 | /* if the io failure tree for this inode is non-empty, |
2004 | * check to see if we've recovered from a failed IO | |
2005 | */ | |
1259ab75 | 2006 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
2007 | return 0; |
2008 | ||
2009 | zeroit: | |
193f284d CM |
2010 | if (printk_ratelimit()) { |
2011 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
2012 | "private %llu\n", page->mapping->host->i_ino, | |
2013 | (unsigned long long)start, csum, | |
2014 | (unsigned long long)private); | |
2015 | } | |
db94535d CM |
2016 | memset(kaddr + offset, 1, end - start + 1); |
2017 | flush_dcache_page(page); | |
9ab86c8e | 2018 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2019 | if (private == 0) |
2020 | return 0; | |
7e38326f | 2021 | return -EIO; |
07157aac | 2022 | } |
b888db2b | 2023 | |
7b128766 JB |
2024 | /* |
2025 | * This creates an orphan entry for the given inode in case something goes | |
2026 | * wrong in the middle of an unlink/truncate. | |
2027 | */ | |
2028 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2029 | { | |
2030 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2031 | int ret = 0; | |
2032 | ||
bcc63abb | 2033 | spin_lock(&root->list_lock); |
7b128766 JB |
2034 | |
2035 | /* already on the orphan list, we're good */ | |
2036 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2037 | spin_unlock(&root->list_lock); |
7b128766 JB |
2038 | return 0; |
2039 | } | |
2040 | ||
2041 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2042 | ||
bcc63abb | 2043 | spin_unlock(&root->list_lock); |
7b128766 JB |
2044 | |
2045 | /* | |
2046 | * insert an orphan item to track this unlinked/truncated file | |
2047 | */ | |
2048 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2049 | ||
2050 | return ret; | |
2051 | } | |
2052 | ||
2053 | /* | |
2054 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2055 | * item for this particular inode. | |
2056 | */ | |
2057 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2058 | { | |
2059 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2060 | int ret = 0; | |
2061 | ||
bcc63abb | 2062 | spin_lock(&root->list_lock); |
7b128766 JB |
2063 | |
2064 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 2065 | spin_unlock(&root->list_lock); |
7b128766 JB |
2066 | return 0; |
2067 | } | |
2068 | ||
2069 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2070 | if (!trans) { | |
bcc63abb | 2071 | spin_unlock(&root->list_lock); |
7b128766 JB |
2072 | return 0; |
2073 | } | |
2074 | ||
bcc63abb | 2075 | spin_unlock(&root->list_lock); |
7b128766 JB |
2076 | |
2077 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2078 | ||
2079 | return ret; | |
2080 | } | |
2081 | ||
2082 | /* | |
2083 | * this cleans up any orphans that may be left on the list from the last use | |
2084 | * of this root. | |
2085 | */ | |
2086 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
2087 | { | |
2088 | struct btrfs_path *path; | |
2089 | struct extent_buffer *leaf; | |
2090 | struct btrfs_item *item; | |
2091 | struct btrfs_key key, found_key; | |
2092 | struct btrfs_trans_handle *trans; | |
2093 | struct inode *inode; | |
2094 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2095 | ||
7b128766 JB |
2096 | path = btrfs_alloc_path(); |
2097 | if (!path) | |
2098 | return; | |
2099 | path->reada = -1; | |
2100 | ||
2101 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2102 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2103 | key.offset = (u64)-1; | |
2104 | ||
7b128766 JB |
2105 | |
2106 | while (1) { | |
2107 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2108 | if (ret < 0) { | |
2109 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
2110 | "\n", ret); | |
2111 | break; | |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * if ret == 0 means we found what we were searching for, which | |
2116 | * is weird, but possible, so only screw with path if we didnt | |
2117 | * find the key and see if we have stuff that matches | |
2118 | */ | |
2119 | if (ret > 0) { | |
2120 | if (path->slots[0] == 0) | |
2121 | break; | |
2122 | path->slots[0]--; | |
2123 | } | |
2124 | ||
2125 | /* pull out the item */ | |
2126 | leaf = path->nodes[0]; | |
2127 | item = btrfs_item_nr(leaf, path->slots[0]); | |
2128 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2129 | ||
2130 | /* make sure the item matches what we want */ | |
2131 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2132 | break; | |
2133 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2134 | break; | |
2135 | ||
2136 | /* release the path since we're done with it */ | |
2137 | btrfs_release_path(root, path); | |
2138 | ||
2139 | /* | |
2140 | * this is where we are basically btrfs_lookup, without the | |
2141 | * crossing root thing. we store the inode number in the | |
2142 | * offset of the orphan item. | |
2143 | */ | |
5d4f98a2 YZ |
2144 | found_key.objectid = found_key.offset; |
2145 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2146 | found_key.offset = 0; | |
2147 | inode = btrfs_iget(root->fs_info->sb, &found_key, root); | |
2148 | if (IS_ERR(inode)) | |
7b128766 JB |
2149 | break; |
2150 | ||
7b128766 JB |
2151 | /* |
2152 | * add this inode to the orphan list so btrfs_orphan_del does | |
2153 | * the proper thing when we hit it | |
2154 | */ | |
bcc63abb | 2155 | spin_lock(&root->list_lock); |
7b128766 | 2156 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 2157 | spin_unlock(&root->list_lock); |
7b128766 JB |
2158 | |
2159 | /* | |
2160 | * if this is a bad inode, means we actually succeeded in | |
2161 | * removing the inode, but not the orphan record, which means | |
2162 | * we need to manually delete the orphan since iput will just | |
2163 | * do a destroy_inode | |
2164 | */ | |
2165 | if (is_bad_inode(inode)) { | |
5b21f2ed | 2166 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 2167 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2168 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2169 | iput(inode); |
2170 | continue; | |
2171 | } | |
2172 | ||
2173 | /* if we have links, this was a truncate, lets do that */ | |
2174 | if (inode->i_nlink) { | |
2175 | nr_truncate++; | |
2176 | btrfs_truncate(inode); | |
2177 | } else { | |
2178 | nr_unlink++; | |
2179 | } | |
2180 | ||
2181 | /* this will do delete_inode and everything for us */ | |
2182 | iput(inode); | |
2183 | } | |
2184 | ||
2185 | if (nr_unlink) | |
2186 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2187 | if (nr_truncate) | |
2188 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
2189 | ||
2190 | btrfs_free_path(path); | |
7b128766 JB |
2191 | } |
2192 | ||
46a53cca CM |
2193 | /* |
2194 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2195 | * don't find any xattrs, we know there can't be any acls. | |
2196 | * | |
2197 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2198 | */ | |
2199 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2200 | int slot, u64 objectid) | |
2201 | { | |
2202 | u32 nritems = btrfs_header_nritems(leaf); | |
2203 | struct btrfs_key found_key; | |
2204 | int scanned = 0; | |
2205 | ||
2206 | slot++; | |
2207 | while (slot < nritems) { | |
2208 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2209 | ||
2210 | /* we found a different objectid, there must not be acls */ | |
2211 | if (found_key.objectid != objectid) | |
2212 | return 0; | |
2213 | ||
2214 | /* we found an xattr, assume we've got an acl */ | |
2215 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2216 | return 1; | |
2217 | ||
2218 | /* | |
2219 | * we found a key greater than an xattr key, there can't | |
2220 | * be any acls later on | |
2221 | */ | |
2222 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2223 | return 0; | |
2224 | ||
2225 | slot++; | |
2226 | scanned++; | |
2227 | ||
2228 | /* | |
2229 | * it goes inode, inode backrefs, xattrs, extents, | |
2230 | * so if there are a ton of hard links to an inode there can | |
2231 | * be a lot of backrefs. Don't waste time searching too hard, | |
2232 | * this is just an optimization | |
2233 | */ | |
2234 | if (scanned >= 8) | |
2235 | break; | |
2236 | } | |
2237 | /* we hit the end of the leaf before we found an xattr or | |
2238 | * something larger than an xattr. We have to assume the inode | |
2239 | * has acls | |
2240 | */ | |
2241 | return 1; | |
2242 | } | |
2243 | ||
d352ac68 CM |
2244 | /* |
2245 | * read an inode from the btree into the in-memory inode | |
2246 | */ | |
5d4f98a2 | 2247 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2248 | { |
2249 | struct btrfs_path *path; | |
5f39d397 | 2250 | struct extent_buffer *leaf; |
39279cc3 | 2251 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2252 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2253 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2254 | struct btrfs_key location; | |
46a53cca | 2255 | int maybe_acls; |
39279cc3 | 2256 | u64 alloc_group_block; |
618e21d5 | 2257 | u32 rdev; |
39279cc3 CM |
2258 | int ret; |
2259 | ||
2260 | path = btrfs_alloc_path(); | |
2261 | BUG_ON(!path); | |
39279cc3 | 2262 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2263 | |
39279cc3 | 2264 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2265 | if (ret) |
39279cc3 | 2266 | goto make_bad; |
39279cc3 | 2267 | |
5f39d397 CM |
2268 | leaf = path->nodes[0]; |
2269 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2270 | struct btrfs_inode_item); | |
2271 | ||
2272 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2273 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2274 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2275 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2276 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2277 | |
2278 | tspec = btrfs_inode_atime(inode_item); | |
2279 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2280 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2281 | ||
2282 | tspec = btrfs_inode_mtime(inode_item); | |
2283 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2284 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2285 | ||
2286 | tspec = btrfs_inode_ctime(inode_item); | |
2287 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2288 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2289 | ||
a76a3cd4 | 2290 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2291 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2292 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2293 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2294 | inode->i_rdev = 0; |
5f39d397 CM |
2295 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2296 | ||
aec7477b | 2297 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2298 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2299 | |
5f39d397 | 2300 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2301 | |
46a53cca CM |
2302 | /* |
2303 | * try to precache a NULL acl entry for files that don't have | |
2304 | * any xattrs or acls | |
2305 | */ | |
2306 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2307 | if (!maybe_acls) |
2308 | cache_no_acl(inode); | |
46a53cca | 2309 | |
d2fb3437 YZ |
2310 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2311 | alloc_group_block, 0); | |
39279cc3 CM |
2312 | btrfs_free_path(path); |
2313 | inode_item = NULL; | |
2314 | ||
39279cc3 | 2315 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2316 | case S_IFREG: |
2317 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2318 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2319 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2320 | inode->i_fop = &btrfs_file_operations; |
2321 | inode->i_op = &btrfs_file_inode_operations; | |
2322 | break; | |
2323 | case S_IFDIR: | |
2324 | inode->i_fop = &btrfs_dir_file_operations; | |
2325 | if (root == root->fs_info->tree_root) | |
2326 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2327 | else | |
2328 | inode->i_op = &btrfs_dir_inode_operations; | |
2329 | break; | |
2330 | case S_IFLNK: | |
2331 | inode->i_op = &btrfs_symlink_inode_operations; | |
2332 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2333 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2334 | break; |
618e21d5 | 2335 | default: |
0279b4cd | 2336 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2337 | init_special_inode(inode, inode->i_mode, rdev); |
2338 | break; | |
39279cc3 | 2339 | } |
6cbff00f CH |
2340 | |
2341 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2342 | return; |
2343 | ||
2344 | make_bad: | |
39279cc3 | 2345 | btrfs_free_path(path); |
39279cc3 CM |
2346 | make_bad_inode(inode); |
2347 | } | |
2348 | ||
d352ac68 CM |
2349 | /* |
2350 | * given a leaf and an inode, copy the inode fields into the leaf | |
2351 | */ | |
e02119d5 CM |
2352 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2353 | struct extent_buffer *leaf, | |
5f39d397 | 2354 | struct btrfs_inode_item *item, |
39279cc3 CM |
2355 | struct inode *inode) |
2356 | { | |
5f39d397 CM |
2357 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2358 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2359 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2360 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2361 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2362 | ||
2363 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2364 | inode->i_atime.tv_sec); | |
2365 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2366 | inode->i_atime.tv_nsec); | |
2367 | ||
2368 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2369 | inode->i_mtime.tv_sec); | |
2370 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2371 | inode->i_mtime.tv_nsec); | |
2372 | ||
2373 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2374 | inode->i_ctime.tv_sec); | |
2375 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2376 | inode->i_ctime.tv_nsec); | |
2377 | ||
a76a3cd4 | 2378 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2379 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2380 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2381 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2382 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2383 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2384 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2385 | } |
2386 | ||
d352ac68 CM |
2387 | /* |
2388 | * copy everything in the in-memory inode into the btree. | |
2389 | */ | |
d397712b CM |
2390 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2391 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2392 | { |
2393 | struct btrfs_inode_item *inode_item; | |
2394 | struct btrfs_path *path; | |
5f39d397 | 2395 | struct extent_buffer *leaf; |
39279cc3 CM |
2396 | int ret; |
2397 | ||
2398 | path = btrfs_alloc_path(); | |
2399 | BUG_ON(!path); | |
b9473439 | 2400 | path->leave_spinning = 1; |
39279cc3 CM |
2401 | ret = btrfs_lookup_inode(trans, root, path, |
2402 | &BTRFS_I(inode)->location, 1); | |
2403 | if (ret) { | |
2404 | if (ret > 0) | |
2405 | ret = -ENOENT; | |
2406 | goto failed; | |
2407 | } | |
2408 | ||
b4ce94de | 2409 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2410 | leaf = path->nodes[0]; |
2411 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2412 | struct btrfs_inode_item); |
2413 | ||
e02119d5 | 2414 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2415 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2416 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2417 | ret = 0; |
2418 | failed: | |
39279cc3 CM |
2419 | btrfs_free_path(path); |
2420 | return ret; | |
2421 | } | |
2422 | ||
2423 | ||
d352ac68 CM |
2424 | /* |
2425 | * unlink helper that gets used here in inode.c and in the tree logging | |
2426 | * recovery code. It remove a link in a directory with a given name, and | |
2427 | * also drops the back refs in the inode to the directory | |
2428 | */ | |
e02119d5 CM |
2429 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2430 | struct btrfs_root *root, | |
2431 | struct inode *dir, struct inode *inode, | |
2432 | const char *name, int name_len) | |
39279cc3 CM |
2433 | { |
2434 | struct btrfs_path *path; | |
39279cc3 | 2435 | int ret = 0; |
5f39d397 | 2436 | struct extent_buffer *leaf; |
39279cc3 | 2437 | struct btrfs_dir_item *di; |
5f39d397 | 2438 | struct btrfs_key key; |
aec7477b | 2439 | u64 index; |
39279cc3 CM |
2440 | |
2441 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2442 | if (!path) { |
2443 | ret = -ENOMEM; | |
2444 | goto err; | |
2445 | } | |
2446 | ||
b9473439 | 2447 | path->leave_spinning = 1; |
39279cc3 CM |
2448 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2449 | name, name_len, -1); | |
2450 | if (IS_ERR(di)) { | |
2451 | ret = PTR_ERR(di); | |
2452 | goto err; | |
2453 | } | |
2454 | if (!di) { | |
2455 | ret = -ENOENT; | |
2456 | goto err; | |
2457 | } | |
5f39d397 CM |
2458 | leaf = path->nodes[0]; |
2459 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2460 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2461 | if (ret) |
2462 | goto err; | |
39279cc3 CM |
2463 | btrfs_release_path(root, path); |
2464 | ||
aec7477b | 2465 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2466 | inode->i_ino, |
2467 | dir->i_ino, &index); | |
aec7477b | 2468 | if (ret) { |
d397712b | 2469 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2470 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2471 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2472 | goto err; |
2473 | } | |
2474 | ||
39279cc3 | 2475 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2476 | index, name, name_len, -1); |
39279cc3 CM |
2477 | if (IS_ERR(di)) { |
2478 | ret = PTR_ERR(di); | |
2479 | goto err; | |
2480 | } | |
2481 | if (!di) { | |
2482 | ret = -ENOENT; | |
2483 | goto err; | |
2484 | } | |
2485 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2486 | btrfs_release_path(root, path); |
39279cc3 | 2487 | |
e02119d5 CM |
2488 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2489 | inode, dir->i_ino); | |
49eb7e46 | 2490 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2491 | |
2492 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2493 | dir, index); | |
2494 | BUG_ON(ret); | |
39279cc3 CM |
2495 | err: |
2496 | btrfs_free_path(path); | |
e02119d5 CM |
2497 | if (ret) |
2498 | goto out; | |
2499 | ||
2500 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2501 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2502 | btrfs_update_inode(trans, root, dir); | |
2503 | btrfs_drop_nlink(inode); | |
2504 | ret = btrfs_update_inode(trans, root, inode); | |
e02119d5 | 2505 | out: |
39279cc3 CM |
2506 | return ret; |
2507 | } | |
2508 | ||
2509 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2510 | { | |
2511 | struct btrfs_root *root; | |
2512 | struct btrfs_trans_handle *trans; | |
7b128766 | 2513 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2514 | int ret; |
1832a6d5 | 2515 | unsigned long nr = 0; |
39279cc3 CM |
2516 | |
2517 | root = BTRFS_I(dir)->root; | |
1832a6d5 | 2518 | |
5df6a9f6 JB |
2519 | /* |
2520 | * 5 items for unlink inode | |
2521 | * 1 for orphan | |
2522 | */ | |
2523 | ret = btrfs_reserve_metadata_space(root, 6); | |
2524 | if (ret) | |
2525 | return ret; | |
2526 | ||
39279cc3 | 2527 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2528 | if (IS_ERR(trans)) { |
2529 | btrfs_unreserve_metadata_space(root, 6); | |
2530 | return PTR_ERR(trans); | |
2531 | } | |
5f39d397 | 2532 | |
39279cc3 | 2533 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2534 | |
2535 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2536 | ||
e02119d5 CM |
2537 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2538 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2539 | |
2540 | if (inode->i_nlink == 0) | |
2541 | ret = btrfs_orphan_add(trans, inode); | |
2542 | ||
d3c2fdcf | 2543 | nr = trans->blocks_used; |
5f39d397 | 2544 | |
89ce8a63 | 2545 | btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2546 | btrfs_unreserve_metadata_space(root, 6); |
d3c2fdcf | 2547 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2548 | return ret; |
2549 | } | |
2550 | ||
4df27c4d YZ |
2551 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2552 | struct btrfs_root *root, | |
2553 | struct inode *dir, u64 objectid, | |
2554 | const char *name, int name_len) | |
2555 | { | |
2556 | struct btrfs_path *path; | |
2557 | struct extent_buffer *leaf; | |
2558 | struct btrfs_dir_item *di; | |
2559 | struct btrfs_key key; | |
2560 | u64 index; | |
2561 | int ret; | |
2562 | ||
2563 | path = btrfs_alloc_path(); | |
2564 | if (!path) | |
2565 | return -ENOMEM; | |
2566 | ||
2567 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2568 | name, name_len, -1); | |
2569 | BUG_ON(!di || IS_ERR(di)); | |
2570 | ||
2571 | leaf = path->nodes[0]; | |
2572 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2573 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2574 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2575 | BUG_ON(ret); | |
2576 | btrfs_release_path(root, path); | |
2577 | ||
2578 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2579 | objectid, root->root_key.objectid, | |
2580 | dir->i_ino, &index, name, name_len); | |
2581 | if (ret < 0) { | |
2582 | BUG_ON(ret != -ENOENT); | |
2583 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
2584 | name, name_len); | |
2585 | BUG_ON(!di || IS_ERR(di)); | |
2586 | ||
2587 | leaf = path->nodes[0]; | |
2588 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2589 | btrfs_release_path(root, path); | |
2590 | index = key.offset; | |
2591 | } | |
2592 | ||
2593 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
2594 | index, name, name_len, -1); | |
2595 | BUG_ON(!di || IS_ERR(di)); | |
2596 | ||
2597 | leaf = path->nodes[0]; | |
2598 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2599 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2600 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2601 | BUG_ON(ret); | |
2602 | btrfs_release_path(root, path); | |
2603 | ||
2604 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2605 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2606 | ret = btrfs_update_inode(trans, root, dir); | |
2607 | BUG_ON(ret); | |
2608 | dir->i_sb->s_dirt = 1; | |
2609 | ||
2610 | btrfs_free_path(path); | |
2611 | return 0; | |
2612 | } | |
2613 | ||
39279cc3 CM |
2614 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2615 | { | |
2616 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2617 | int err = 0; |
39279cc3 CM |
2618 | int ret; |
2619 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2620 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2621 | unsigned long nr = 0; |
39279cc3 | 2622 | |
3394e160 | 2623 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 2624 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2625 | return -ENOTEMPTY; |
2626 | ||
5df6a9f6 JB |
2627 | ret = btrfs_reserve_metadata_space(root, 5); |
2628 | if (ret) | |
2629 | return ret; | |
2630 | ||
39279cc3 | 2631 | trans = btrfs_start_transaction(root, 1); |
5df6a9f6 JB |
2632 | if (IS_ERR(trans)) { |
2633 | btrfs_unreserve_metadata_space(root, 5); | |
2634 | return PTR_ERR(trans); | |
2635 | } | |
2636 | ||
39279cc3 | 2637 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 2638 | |
4df27c4d YZ |
2639 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
2640 | err = btrfs_unlink_subvol(trans, root, dir, | |
2641 | BTRFS_I(inode)->location.objectid, | |
2642 | dentry->d_name.name, | |
2643 | dentry->d_name.len); | |
2644 | goto out; | |
2645 | } | |
2646 | ||
7b128766 JB |
2647 | err = btrfs_orphan_add(trans, inode); |
2648 | if (err) | |
4df27c4d | 2649 | goto out; |
7b128766 | 2650 | |
39279cc3 | 2651 | /* now the directory is empty */ |
e02119d5 CM |
2652 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2653 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2654 | if (!err) |
dbe674a9 | 2655 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2656 | out: |
d3c2fdcf | 2657 | nr = trans->blocks_used; |
89ce8a63 | 2658 | ret = btrfs_end_transaction_throttle(trans, root); |
5df6a9f6 | 2659 | btrfs_unreserve_metadata_space(root, 5); |
d3c2fdcf | 2660 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2661 | |
39279cc3 CM |
2662 | if (ret && !err) |
2663 | err = ret; | |
2664 | return err; | |
2665 | } | |
2666 | ||
d20f7043 | 2667 | #if 0 |
323ac95b CM |
2668 | /* |
2669 | * when truncating bytes in a file, it is possible to avoid reading | |
2670 | * the leaves that contain only checksum items. This can be the | |
2671 | * majority of the IO required to delete a large file, but it must | |
2672 | * be done carefully. | |
2673 | * | |
2674 | * The keys in the level just above the leaves are checked to make sure | |
2675 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2676 | * after the new size. | |
2677 | * | |
2678 | * Then the key for the next leaf is checked to make sure it also has | |
2679 | * a checksum item for the same file. If it does, we know our target leaf | |
2680 | * contains only checksum items, and it can be safely freed without reading | |
2681 | * it. | |
2682 | * | |
2683 | * This is just an optimization targeted at large files. It may do | |
2684 | * nothing. It will return 0 unless things went badly. | |
2685 | */ | |
2686 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2687 | struct btrfs_root *root, | |
2688 | struct btrfs_path *path, | |
2689 | struct inode *inode, u64 new_size) | |
2690 | { | |
2691 | struct btrfs_key key; | |
2692 | int ret; | |
2693 | int nritems; | |
2694 | struct btrfs_key found_key; | |
2695 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2696 | struct btrfs_leaf_ref *ref; |
2697 | u64 leaf_gen; | |
2698 | u64 leaf_start; | |
323ac95b CM |
2699 | |
2700 | path->lowest_level = 1; | |
2701 | key.objectid = inode->i_ino; | |
2702 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2703 | key.offset = new_size; | |
2704 | again: | |
2705 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2706 | if (ret < 0) | |
2707 | goto out; | |
2708 | ||
2709 | if (path->nodes[1] == NULL) { | |
2710 | ret = 0; | |
2711 | goto out; | |
2712 | } | |
2713 | ret = 0; | |
2714 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2715 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2716 | ||
2717 | if (!nritems) | |
2718 | goto out; | |
2719 | ||
2720 | if (path->slots[1] >= nritems) | |
2721 | goto next_node; | |
2722 | ||
2723 | /* did we find a key greater than anything we want to delete? */ | |
2724 | if (found_key.objectid > inode->i_ino || | |
2725 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2726 | goto out; | |
2727 | ||
2728 | /* we check the next key in the node to make sure the leave contains | |
2729 | * only checksum items. This comparison doesn't work if our | |
2730 | * leaf is the last one in the node | |
2731 | */ | |
2732 | if (path->slots[1] + 1 >= nritems) { | |
2733 | next_node: | |
2734 | /* search forward from the last key in the node, this | |
2735 | * will bring us into the next node in the tree | |
2736 | */ | |
2737 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2738 | ||
2739 | /* unlikely, but we inc below, so check to be safe */ | |
2740 | if (found_key.offset == (u64)-1) | |
2741 | goto out; | |
2742 | ||
2743 | /* search_forward needs a path with locks held, do the | |
2744 | * search again for the original key. It is possible | |
2745 | * this will race with a balance and return a path that | |
2746 | * we could modify, but this drop is just an optimization | |
2747 | * and is allowed to miss some leaves. | |
2748 | */ | |
2749 | btrfs_release_path(root, path); | |
2750 | found_key.offset++; | |
2751 | ||
2752 | /* setup a max key for search_forward */ | |
2753 | other_key.offset = (u64)-1; | |
2754 | other_key.type = key.type; | |
2755 | other_key.objectid = key.objectid; | |
2756 | ||
2757 | path->keep_locks = 1; | |
2758 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2759 | path, 0, 0); | |
2760 | path->keep_locks = 0; | |
2761 | if (ret || found_key.objectid != key.objectid || | |
2762 | found_key.type != key.type) { | |
2763 | ret = 0; | |
2764 | goto out; | |
2765 | } | |
2766 | ||
2767 | key.offset = found_key.offset; | |
2768 | btrfs_release_path(root, path); | |
2769 | cond_resched(); | |
2770 | goto again; | |
2771 | } | |
2772 | ||
2773 | /* we know there's one more slot after us in the tree, | |
2774 | * read that key so we can verify it is also a checksum item | |
2775 | */ | |
2776 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2777 | ||
2778 | if (found_key.objectid < inode->i_ino) | |
2779 | goto next_key; | |
2780 | ||
2781 | if (found_key.type != key.type || found_key.offset < new_size) | |
2782 | goto next_key; | |
2783 | ||
2784 | /* | |
2785 | * if the key for the next leaf isn't a csum key from this objectid, | |
2786 | * we can't be sure there aren't good items inside this leaf. | |
2787 | * Bail out | |
2788 | */ | |
2789 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2790 | goto out; | |
2791 | ||
5b84e8d6 YZ |
2792 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2793 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2794 | /* |
2795 | * it is safe to delete this leaf, it contains only | |
2796 | * csum items from this inode at an offset >= new_size | |
2797 | */ | |
5b84e8d6 | 2798 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2799 | BUG_ON(ret); |
2800 | ||
5b84e8d6 YZ |
2801 | if (root->ref_cows && leaf_gen < trans->transid) { |
2802 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2803 | if (ref) { | |
2804 | ref->root_gen = root->root_key.offset; | |
2805 | ref->bytenr = leaf_start; | |
2806 | ref->owner = 0; | |
2807 | ref->generation = leaf_gen; | |
2808 | ref->nritems = 0; | |
2809 | ||
bd56b302 CM |
2810 | btrfs_sort_leaf_ref(ref); |
2811 | ||
5b84e8d6 YZ |
2812 | ret = btrfs_add_leaf_ref(root, ref, 0); |
2813 | WARN_ON(ret); | |
2814 | btrfs_free_leaf_ref(root, ref); | |
2815 | } else { | |
2816 | WARN_ON(1); | |
2817 | } | |
2818 | } | |
323ac95b CM |
2819 | next_key: |
2820 | btrfs_release_path(root, path); | |
2821 | ||
2822 | if (other_key.objectid == inode->i_ino && | |
2823 | other_key.type == key.type && other_key.offset > key.offset) { | |
2824 | key.offset = other_key.offset; | |
2825 | cond_resched(); | |
2826 | goto again; | |
2827 | } | |
2828 | ret = 0; | |
2829 | out: | |
2830 | /* fixup any changes we've made to the path */ | |
2831 | path->lowest_level = 0; | |
2832 | path->keep_locks = 0; | |
2833 | btrfs_release_path(root, path); | |
2834 | return ret; | |
2835 | } | |
2836 | ||
d20f7043 CM |
2837 | #endif |
2838 | ||
39279cc3 CM |
2839 | /* |
2840 | * this can truncate away extent items, csum items and directory items. | |
2841 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2842 | * any higher than new_size |
39279cc3 CM |
2843 | * |
2844 | * csum items that cross the new i_size are truncated to the new size | |
2845 | * as well. | |
7b128766 JB |
2846 | * |
2847 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2848 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2849 | */ |
e02119d5 CM |
2850 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2851 | struct btrfs_root *root, | |
2852 | struct inode *inode, | |
2853 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2854 | { |
2855 | int ret; | |
2856 | struct btrfs_path *path; | |
2857 | struct btrfs_key key; | |
5f39d397 | 2858 | struct btrfs_key found_key; |
06d9a8d7 | 2859 | u32 found_type = (u8)-1; |
5f39d397 | 2860 | struct extent_buffer *leaf; |
39279cc3 CM |
2861 | struct btrfs_file_extent_item *fi; |
2862 | u64 extent_start = 0; | |
db94535d | 2863 | u64 extent_num_bytes = 0; |
5d4f98a2 | 2864 | u64 extent_offset = 0; |
39279cc3 CM |
2865 | u64 item_end = 0; |
2866 | int found_extent; | |
2867 | int del_item; | |
85e21bac CM |
2868 | int pending_del_nr = 0; |
2869 | int pending_del_slot = 0; | |
179e29e4 | 2870 | int extent_type = -1; |
771ed689 | 2871 | int encoding; |
3b951516 | 2872 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2873 | |
e02119d5 | 2874 | if (root->ref_cows) |
5b21f2ed | 2875 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 CM |
2876 | path = btrfs_alloc_path(); |
2877 | BUG_ON(!path); | |
33c17ad5 | 2878 | path->reada = -1; |
5f39d397 | 2879 | |
39279cc3 CM |
2880 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2881 | key.objectid = inode->i_ino; | |
2882 | key.offset = (u64)-1; | |
5f39d397 CM |
2883 | key.type = (u8)-1; |
2884 | ||
85e21bac | 2885 | search_again: |
b9473439 | 2886 | path->leave_spinning = 1; |
85e21bac | 2887 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
d397712b | 2888 | if (ret < 0) |
85e21bac | 2889 | goto error; |
d397712b | 2890 | |
85e21bac | 2891 | if (ret > 0) { |
e02119d5 CM |
2892 | /* there are no items in the tree for us to truncate, we're |
2893 | * done | |
2894 | */ | |
2895 | if (path->slots[0] == 0) { | |
2896 | ret = 0; | |
2897 | goto error; | |
2898 | } | |
85e21bac CM |
2899 | path->slots[0]--; |
2900 | } | |
2901 | ||
d397712b | 2902 | while (1) { |
39279cc3 | 2903 | fi = NULL; |
5f39d397 CM |
2904 | leaf = path->nodes[0]; |
2905 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2906 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2907 | encoding = 0; |
39279cc3 | 2908 | |
5f39d397 | 2909 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2910 | break; |
5f39d397 | 2911 | |
85e21bac | 2912 | if (found_type < min_type) |
39279cc3 CM |
2913 | break; |
2914 | ||
5f39d397 | 2915 | item_end = found_key.offset; |
39279cc3 | 2916 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2917 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2918 | struct btrfs_file_extent_item); |
179e29e4 | 2919 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2920 | encoding = btrfs_file_extent_compression(leaf, fi); |
2921 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2922 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2923 | ||
179e29e4 | 2924 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2925 | item_end += |
db94535d | 2926 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2927 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2928 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2929 | fi); |
39279cc3 | 2930 | } |
008630c1 | 2931 | item_end--; |
39279cc3 | 2932 | } |
e02119d5 | 2933 | if (item_end < new_size) { |
d397712b | 2934 | if (found_type == BTRFS_DIR_ITEM_KEY) |
b888db2b | 2935 | found_type = BTRFS_INODE_ITEM_KEY; |
d397712b | 2936 | else if (found_type == BTRFS_EXTENT_ITEM_KEY) |
d20f7043 | 2937 | found_type = BTRFS_EXTENT_DATA_KEY; |
d397712b | 2938 | else if (found_type == BTRFS_EXTENT_DATA_KEY) |
85e21bac | 2939 | found_type = BTRFS_XATTR_ITEM_KEY; |
d397712b | 2940 | else if (found_type == BTRFS_XATTR_ITEM_KEY) |
85e21bac | 2941 | found_type = BTRFS_INODE_REF_KEY; |
d397712b | 2942 | else if (found_type) |
b888db2b | 2943 | found_type--; |
d397712b | 2944 | else |
b888db2b | 2945 | break; |
a61721d5 | 2946 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2947 | goto next; |
39279cc3 | 2948 | } |
e02119d5 | 2949 | if (found_key.offset >= new_size) |
39279cc3 CM |
2950 | del_item = 1; |
2951 | else | |
2952 | del_item = 0; | |
2953 | found_extent = 0; | |
2954 | ||
2955 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2956 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2957 | goto delete; | |
2958 | ||
2959 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2960 | u64 num_dec; |
db94535d | 2961 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2962 | if (!del_item && !encoding) { |
db94535d CM |
2963 | u64 orig_num_bytes = |
2964 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2965 | extent_num_bytes = new_size - |
5f39d397 | 2966 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2967 | extent_num_bytes = extent_num_bytes & |
2968 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2969 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2970 | extent_num_bytes); | |
2971 | num_dec = (orig_num_bytes - | |
9069218d | 2972 | extent_num_bytes); |
e02119d5 | 2973 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2974 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2975 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2976 | } else { |
db94535d CM |
2977 | extent_num_bytes = |
2978 | btrfs_file_extent_disk_num_bytes(leaf, | |
2979 | fi); | |
5d4f98a2 YZ |
2980 | extent_offset = found_key.offset - |
2981 | btrfs_file_extent_offset(leaf, fi); | |
2982 | ||
39279cc3 | 2983 | /* FIXME blocksize != 4096 */ |
9069218d | 2984 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2985 | if (extent_start != 0) { |
2986 | found_extent = 1; | |
e02119d5 | 2987 | if (root->ref_cows) |
a76a3cd4 | 2988 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2989 | } |
39279cc3 | 2990 | } |
9069218d | 2991 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2992 | /* |
2993 | * we can't truncate inline items that have had | |
2994 | * special encodings | |
2995 | */ | |
2996 | if (!del_item && | |
2997 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2998 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2999 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3000 | u32 size = new_size - found_key.offset; |
3001 | ||
3002 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3003 | inode_sub_bytes(inode, item_end + 1 - |
3004 | new_size); | |
e02119d5 CM |
3005 | } |
3006 | size = | |
3007 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3008 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3009 | size, 1); |
9069218d | 3010 | BUG_ON(ret); |
e02119d5 | 3011 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3012 | inode_sub_bytes(inode, item_end + 1 - |
3013 | found_key.offset); | |
9069218d | 3014 | } |
39279cc3 | 3015 | } |
179e29e4 | 3016 | delete: |
39279cc3 | 3017 | if (del_item) { |
85e21bac CM |
3018 | if (!pending_del_nr) { |
3019 | /* no pending yet, add ourselves */ | |
3020 | pending_del_slot = path->slots[0]; | |
3021 | pending_del_nr = 1; | |
3022 | } else if (pending_del_nr && | |
3023 | path->slots[0] + 1 == pending_del_slot) { | |
3024 | /* hop on the pending chunk */ | |
3025 | pending_del_nr++; | |
3026 | pending_del_slot = path->slots[0]; | |
3027 | } else { | |
d397712b | 3028 | BUG(); |
85e21bac | 3029 | } |
39279cc3 CM |
3030 | } else { |
3031 | break; | |
3032 | } | |
5d4f98a2 | 3033 | if (found_extent && root->ref_cows) { |
b9473439 | 3034 | btrfs_set_path_blocking(path); |
39279cc3 | 3035 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3036 | extent_num_bytes, 0, |
3037 | btrfs_header_owner(leaf), | |
3038 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3039 | BUG_ON(ret); |
3040 | } | |
85e21bac CM |
3041 | next: |
3042 | if (path->slots[0] == 0) { | |
3043 | if (pending_del_nr) | |
3044 | goto del_pending; | |
3045 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3046 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3047 | break; | |
85e21bac CM |
3048 | goto search_again; |
3049 | } | |
3050 | ||
3051 | path->slots[0]--; | |
3052 | if (pending_del_nr && | |
3053 | path->slots[0] + 1 != pending_del_slot) { | |
3054 | struct btrfs_key debug; | |
3055 | del_pending: | |
3056 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
3057 | pending_del_slot); | |
3058 | ret = btrfs_del_items(trans, root, path, | |
3059 | pending_del_slot, | |
3060 | pending_del_nr); | |
3061 | BUG_ON(ret); | |
3062 | pending_del_nr = 0; | |
3063 | btrfs_release_path(root, path); | |
06d9a8d7 CM |
3064 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3065 | break; | |
85e21bac CM |
3066 | goto search_again; |
3067 | } | |
39279cc3 CM |
3068 | } |
3069 | ret = 0; | |
3070 | error: | |
85e21bac CM |
3071 | if (pending_del_nr) { |
3072 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3073 | pending_del_nr); | |
3074 | } | |
39279cc3 | 3075 | btrfs_free_path(path); |
39279cc3 CM |
3076 | return ret; |
3077 | } | |
3078 | ||
3079 | /* | |
3080 | * taken from block_truncate_page, but does cow as it zeros out | |
3081 | * any bytes left in the last page in the file. | |
3082 | */ | |
3083 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3084 | { | |
3085 | struct inode *inode = mapping->host; | |
db94535d | 3086 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3087 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3088 | struct btrfs_ordered_extent *ordered; | |
3089 | char *kaddr; | |
db94535d | 3090 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3091 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3092 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3093 | struct page *page; | |
39279cc3 | 3094 | int ret = 0; |
a52d9a80 | 3095 | u64 page_start; |
e6dcd2dc | 3096 | u64 page_end; |
39279cc3 CM |
3097 | |
3098 | if ((offset & (blocksize - 1)) == 0) | |
3099 | goto out; | |
5d5e103a JB |
3100 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
3101 | if (ret) | |
3102 | goto out; | |
3103 | ||
3104 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); | |
3105 | if (ret) | |
3106 | goto out; | |
39279cc3 CM |
3107 | |
3108 | ret = -ENOMEM; | |
211c17f5 | 3109 | again: |
39279cc3 | 3110 | page = grab_cache_page(mapping, index); |
5d5e103a JB |
3111 | if (!page) { |
3112 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3113 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 | 3114 | goto out; |
5d5e103a | 3115 | } |
e6dcd2dc CM |
3116 | |
3117 | page_start = page_offset(page); | |
3118 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3119 | ||
39279cc3 | 3120 | if (!PageUptodate(page)) { |
9ebefb18 | 3121 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3122 | lock_page(page); |
211c17f5 CM |
3123 | if (page->mapping != mapping) { |
3124 | unlock_page(page); | |
3125 | page_cache_release(page); | |
3126 | goto again; | |
3127 | } | |
39279cc3 CM |
3128 | if (!PageUptodate(page)) { |
3129 | ret = -EIO; | |
89642229 | 3130 | goto out_unlock; |
39279cc3 CM |
3131 | } |
3132 | } | |
211c17f5 | 3133 | wait_on_page_writeback(page); |
e6dcd2dc CM |
3134 | |
3135 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3136 | set_page_extent_mapped(page); | |
3137 | ||
3138 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3139 | if (ordered) { | |
3140 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3141 | unlock_page(page); | |
3142 | page_cache_release(page); | |
eb84ae03 | 3143 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3144 | btrfs_put_ordered_extent(ordered); |
3145 | goto again; | |
3146 | } | |
3147 | ||
5d5e103a JB |
3148 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
3149 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, | |
3150 | GFP_NOFS); | |
3151 | ||
9ed74f2d JB |
3152 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
3153 | if (ret) { | |
3154 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3155 | goto out_unlock; | |
3156 | } | |
3157 | ||
e6dcd2dc CM |
3158 | ret = 0; |
3159 | if (offset != PAGE_CACHE_SIZE) { | |
3160 | kaddr = kmap(page); | |
3161 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3162 | flush_dcache_page(page); | |
3163 | kunmap(page); | |
3164 | } | |
247e743c | 3165 | ClearPageChecked(page); |
e6dcd2dc CM |
3166 | set_page_dirty(page); |
3167 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 3168 | |
89642229 | 3169 | out_unlock: |
5d5e103a JB |
3170 | if (ret) |
3171 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
3172 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); | |
39279cc3 CM |
3173 | unlock_page(page); |
3174 | page_cache_release(page); | |
3175 | out: | |
3176 | return ret; | |
3177 | } | |
3178 | ||
9036c102 | 3179 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 3180 | { |
9036c102 YZ |
3181 | struct btrfs_trans_handle *trans; |
3182 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3183 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
3184 | struct extent_map *em; | |
3185 | u64 mask = root->sectorsize - 1; | |
3186 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
3187 | u64 block_end = (size + mask) & ~mask; | |
3188 | u64 last_byte; | |
3189 | u64 cur_offset; | |
3190 | u64 hole_size; | |
9ed74f2d | 3191 | int err = 0; |
39279cc3 | 3192 | |
9036c102 YZ |
3193 | if (size <= hole_start) |
3194 | return 0; | |
3195 | ||
5d5e103a JB |
3196 | err = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
3197 | if (err) | |
3198 | return err; | |
2bf5a725 | 3199 | |
9036c102 YZ |
3200 | while (1) { |
3201 | struct btrfs_ordered_extent *ordered; | |
3202 | btrfs_wait_ordered_range(inode, hole_start, | |
3203 | block_end - hole_start); | |
3204 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3205 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
3206 | if (!ordered) | |
3207 | break; | |
3208 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3209 | btrfs_put_ordered_extent(ordered); | |
3210 | } | |
39279cc3 | 3211 | |
9036c102 YZ |
3212 | trans = btrfs_start_transaction(root, 1); |
3213 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 3214 | |
9036c102 YZ |
3215 | cur_offset = hole_start; |
3216 | while (1) { | |
3217 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3218 | block_end - cur_offset, 0); | |
3219 | BUG_ON(IS_ERR(em) || !em); | |
3220 | last_byte = min(extent_map_end(em), block_end); | |
3221 | last_byte = (last_byte + mask) & ~mask; | |
3222 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 3223 | u64 hint_byte = 0; |
9036c102 | 3224 | hole_size = last_byte - cur_offset; |
920bbbfb | 3225 | err = btrfs_drop_extents(trans, inode, cur_offset, |
771ed689 | 3226 | cur_offset + hole_size, |
920bbbfb | 3227 | &hint_byte, 1); |
771ed689 CM |
3228 | if (err) |
3229 | break; | |
9ed74f2d JB |
3230 | |
3231 | err = btrfs_reserve_metadata_space(root, 1); | |
3232 | if (err) | |
3233 | break; | |
3234 | ||
9036c102 YZ |
3235 | err = btrfs_insert_file_extent(trans, root, |
3236 | inode->i_ino, cur_offset, 0, | |
3237 | 0, hole_size, 0, hole_size, | |
3238 | 0, 0, 0); | |
3239 | btrfs_drop_extent_cache(inode, hole_start, | |
3240 | last_byte - 1, 0); | |
9ed74f2d | 3241 | btrfs_unreserve_metadata_space(root, 1); |
9036c102 YZ |
3242 | } |
3243 | free_extent_map(em); | |
3244 | cur_offset = last_byte; | |
3245 | if (err || cur_offset >= block_end) | |
3246 | break; | |
3247 | } | |
1832a6d5 | 3248 | |
9036c102 YZ |
3249 | btrfs_end_transaction(trans, root); |
3250 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
3251 | return err; | |
3252 | } | |
39279cc3 | 3253 | |
9036c102 YZ |
3254 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3255 | { | |
3256 | struct inode *inode = dentry->d_inode; | |
3257 | int err; | |
39279cc3 | 3258 | |
9036c102 YZ |
3259 | err = inode_change_ok(inode, attr); |
3260 | if (err) | |
3261 | return err; | |
2bf5a725 | 3262 | |
5a3f23d5 CM |
3263 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3264 | if (attr->ia_size > inode->i_size) { | |
3265 | err = btrfs_cont_expand(inode, attr->ia_size); | |
3266 | if (err) | |
3267 | return err; | |
3268 | } else if (inode->i_size > 0 && | |
3269 | attr->ia_size == 0) { | |
3270 | ||
3271 | /* we're truncating a file that used to have good | |
3272 | * data down to zero. Make sure it gets into | |
3273 | * the ordered flush list so that any new writes | |
3274 | * get down to disk quickly. | |
3275 | */ | |
3276 | BTRFS_I(inode)->ordered_data_close = 1; | |
3277 | } | |
39279cc3 | 3278 | } |
9036c102 | 3279 | |
39279cc3 | 3280 | err = inode_setattr(inode, attr); |
33268eaf JB |
3281 | |
3282 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
3283 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
3284 | return err; |
3285 | } | |
61295eb8 | 3286 | |
39279cc3 CM |
3287 | void btrfs_delete_inode(struct inode *inode) |
3288 | { | |
3289 | struct btrfs_trans_handle *trans; | |
3290 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3291 | unsigned long nr; |
39279cc3 CM |
3292 | int ret; |
3293 | ||
3294 | truncate_inode_pages(&inode->i_data, 0); | |
3295 | if (is_bad_inode(inode)) { | |
7b128766 | 3296 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3297 | goto no_delete; |
3298 | } | |
4a096752 | 3299 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3300 | |
76dda93c YZ |
3301 | if (inode->i_nlink > 0) { |
3302 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3303 | goto no_delete; | |
3304 | } | |
3305 | ||
dbe674a9 | 3306 | btrfs_i_size_write(inode, 0); |
180591bc | 3307 | trans = btrfs_join_transaction(root, 1); |
5f39d397 | 3308 | |
39279cc3 | 3309 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 3310 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
3311 | if (ret) { |
3312 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 3313 | goto no_delete_lock; |
7b128766 JB |
3314 | } |
3315 | ||
3316 | btrfs_orphan_del(trans, inode); | |
85e21bac | 3317 | |
d3c2fdcf | 3318 | nr = trans->blocks_used; |
85e21bac | 3319 | clear_inode(inode); |
5f39d397 | 3320 | |
39279cc3 | 3321 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3322 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3323 | return; |
54aa1f4d CM |
3324 | |
3325 | no_delete_lock: | |
d3c2fdcf | 3326 | nr = trans->blocks_used; |
54aa1f4d | 3327 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3328 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3329 | no_delete: |
3330 | clear_inode(inode); | |
3331 | } | |
3332 | ||
3333 | /* | |
3334 | * this returns the key found in the dir entry in the location pointer. | |
3335 | * If no dir entries were found, location->objectid is 0. | |
3336 | */ | |
3337 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3338 | struct btrfs_key *location) | |
3339 | { | |
3340 | const char *name = dentry->d_name.name; | |
3341 | int namelen = dentry->d_name.len; | |
3342 | struct btrfs_dir_item *di; | |
3343 | struct btrfs_path *path; | |
3344 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3345 | int ret = 0; |
39279cc3 CM |
3346 | |
3347 | path = btrfs_alloc_path(); | |
3348 | BUG_ON(!path); | |
3954401f | 3349 | |
39279cc3 CM |
3350 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3351 | namelen, 0); | |
0d9f7f3e Y |
3352 | if (IS_ERR(di)) |
3353 | ret = PTR_ERR(di); | |
d397712b CM |
3354 | |
3355 | if (!di || IS_ERR(di)) | |
3954401f | 3356 | goto out_err; |
d397712b | 3357 | |
5f39d397 | 3358 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3359 | out: |
39279cc3 CM |
3360 | btrfs_free_path(path); |
3361 | return ret; | |
3954401f CM |
3362 | out_err: |
3363 | location->objectid = 0; | |
3364 | goto out; | |
39279cc3 CM |
3365 | } |
3366 | ||
3367 | /* | |
3368 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3369 | * needs to be changed to reflect the root directory of the tree root. This | |
3370 | * is kind of like crossing a mount point. | |
3371 | */ | |
3372 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3373 | struct inode *dir, |
3374 | struct dentry *dentry, | |
3375 | struct btrfs_key *location, | |
3376 | struct btrfs_root **sub_root) | |
39279cc3 | 3377 | { |
4df27c4d YZ |
3378 | struct btrfs_path *path; |
3379 | struct btrfs_root *new_root; | |
3380 | struct btrfs_root_ref *ref; | |
3381 | struct extent_buffer *leaf; | |
3382 | int ret; | |
3383 | int err = 0; | |
39279cc3 | 3384 | |
4df27c4d YZ |
3385 | path = btrfs_alloc_path(); |
3386 | if (!path) { | |
3387 | err = -ENOMEM; | |
3388 | goto out; | |
3389 | } | |
39279cc3 | 3390 | |
4df27c4d YZ |
3391 | err = -ENOENT; |
3392 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3393 | BTRFS_I(dir)->root->root_key.objectid, | |
3394 | location->objectid); | |
3395 | if (ret) { | |
3396 | if (ret < 0) | |
3397 | err = ret; | |
3398 | goto out; | |
3399 | } | |
39279cc3 | 3400 | |
4df27c4d YZ |
3401 | leaf = path->nodes[0]; |
3402 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3403 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3404 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3405 | goto out; | |
39279cc3 | 3406 | |
4df27c4d YZ |
3407 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3408 | (unsigned long)(ref + 1), | |
3409 | dentry->d_name.len); | |
3410 | if (ret) | |
3411 | goto out; | |
3412 | ||
3413 | btrfs_release_path(root->fs_info->tree_root, path); | |
3414 | ||
3415 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3416 | if (IS_ERR(new_root)) { | |
3417 | err = PTR_ERR(new_root); | |
3418 | goto out; | |
3419 | } | |
3420 | ||
3421 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3422 | err = -ENOENT; | |
3423 | goto out; | |
3424 | } | |
3425 | ||
3426 | *sub_root = new_root; | |
3427 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3428 | location->type = BTRFS_INODE_ITEM_KEY; | |
3429 | location->offset = 0; | |
3430 | err = 0; | |
3431 | out: | |
3432 | btrfs_free_path(path); | |
3433 | return err; | |
39279cc3 CM |
3434 | } |
3435 | ||
5d4f98a2 YZ |
3436 | static void inode_tree_add(struct inode *inode) |
3437 | { | |
3438 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3439 | struct btrfs_inode *entry; | |
03e860bd FNP |
3440 | struct rb_node **p; |
3441 | struct rb_node *parent; | |
03e860bd FNP |
3442 | again: |
3443 | p = &root->inode_tree.rb_node; | |
3444 | parent = NULL; | |
5d4f98a2 | 3445 | |
76dda93c YZ |
3446 | if (hlist_unhashed(&inode->i_hash)) |
3447 | return; | |
3448 | ||
5d4f98a2 YZ |
3449 | spin_lock(&root->inode_lock); |
3450 | while (*p) { | |
3451 | parent = *p; | |
3452 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3453 | ||
3454 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3455 | p = &parent->rb_left; |
5d4f98a2 | 3456 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3457 | p = &parent->rb_right; |
5d4f98a2 YZ |
3458 | else { |
3459 | WARN_ON(!(entry->vfs_inode.i_state & | |
3460 | (I_WILL_FREE | I_FREEING | I_CLEAR))); | |
03e860bd FNP |
3461 | rb_erase(parent, &root->inode_tree); |
3462 | RB_CLEAR_NODE(parent); | |
3463 | spin_unlock(&root->inode_lock); | |
3464 | goto again; | |
5d4f98a2 YZ |
3465 | } |
3466 | } | |
3467 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3468 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3469 | spin_unlock(&root->inode_lock); | |
3470 | } | |
3471 | ||
3472 | static void inode_tree_del(struct inode *inode) | |
3473 | { | |
3474 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3475 | int empty = 0; |
5d4f98a2 | 3476 | |
03e860bd | 3477 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3478 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3479 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3480 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3481 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3482 | } |
03e860bd | 3483 | spin_unlock(&root->inode_lock); |
76dda93c YZ |
3484 | |
3485 | if (empty && btrfs_root_refs(&root->root_item) == 0) { | |
3486 | synchronize_srcu(&root->fs_info->subvol_srcu); | |
3487 | spin_lock(&root->inode_lock); | |
3488 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3489 | spin_unlock(&root->inode_lock); | |
3490 | if (empty) | |
3491 | btrfs_add_dead_root(root); | |
3492 | } | |
3493 | } | |
3494 | ||
3495 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3496 | { | |
3497 | struct rb_node *node; | |
3498 | struct rb_node *prev; | |
3499 | struct btrfs_inode *entry; | |
3500 | struct inode *inode; | |
3501 | u64 objectid = 0; | |
3502 | ||
3503 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3504 | ||
3505 | spin_lock(&root->inode_lock); | |
3506 | again: | |
3507 | node = root->inode_tree.rb_node; | |
3508 | prev = NULL; | |
3509 | while (node) { | |
3510 | prev = node; | |
3511 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3512 | ||
3513 | if (objectid < entry->vfs_inode.i_ino) | |
3514 | node = node->rb_left; | |
3515 | else if (objectid > entry->vfs_inode.i_ino) | |
3516 | node = node->rb_right; | |
3517 | else | |
3518 | break; | |
3519 | } | |
3520 | if (!node) { | |
3521 | while (prev) { | |
3522 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
3523 | if (objectid <= entry->vfs_inode.i_ino) { | |
3524 | node = prev; | |
3525 | break; | |
3526 | } | |
3527 | prev = rb_next(prev); | |
3528 | } | |
3529 | } | |
3530 | while (node) { | |
3531 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3532 | objectid = entry->vfs_inode.i_ino + 1; | |
3533 | inode = igrab(&entry->vfs_inode); | |
3534 | if (inode) { | |
3535 | spin_unlock(&root->inode_lock); | |
3536 | if (atomic_read(&inode->i_count) > 1) | |
3537 | d_prune_aliases(inode); | |
3538 | /* | |
3539 | * btrfs_drop_inode will remove it from | |
3540 | * the inode cache when its usage count | |
3541 | * hits zero. | |
3542 | */ | |
3543 | iput(inode); | |
3544 | cond_resched(); | |
3545 | spin_lock(&root->inode_lock); | |
3546 | goto again; | |
3547 | } | |
3548 | ||
3549 | if (cond_resched_lock(&root->inode_lock)) | |
3550 | goto again; | |
3551 | ||
3552 | node = rb_next(node); | |
3553 | } | |
3554 | spin_unlock(&root->inode_lock); | |
3555 | return 0; | |
5d4f98a2 YZ |
3556 | } |
3557 | ||
e02119d5 | 3558 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 3559 | { |
e02119d5 CM |
3560 | struct btrfs_inode *bi = BTRFS_I(inode); |
3561 | ||
e02119d5 | 3562 | bi->generation = 0; |
c3027eb5 | 3563 | bi->sequence = 0; |
e02119d5 | 3564 | bi->last_trans = 0; |
257c62e1 | 3565 | bi->last_sub_trans = 0; |
e02119d5 CM |
3566 | bi->logged_trans = 0; |
3567 | bi->delalloc_bytes = 0; | |
6a63209f | 3568 | bi->reserved_bytes = 0; |
e02119d5 CM |
3569 | bi->disk_i_size = 0; |
3570 | bi->flags = 0; | |
3571 | bi->index_cnt = (u64)-1; | |
12fcfd22 | 3572 | bi->last_unlink_trans = 0; |
2757495c | 3573 | bi->ordered_data_close = 0; |
d1310b2e CM |
3574 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
3575 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 3576 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
3577 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
3578 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 3579 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
5a3f23d5 | 3580 | INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations); |
5d4f98a2 | 3581 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
ba1da2f4 | 3582 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
e02119d5 CM |
3583 | mutex_init(&BTRFS_I(inode)->log_mutex); |
3584 | } | |
3585 | ||
3586 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
3587 | { | |
3588 | struct btrfs_iget_args *args = p; | |
3589 | inode->i_ino = args->ino; | |
3590 | init_btrfs_i(inode); | |
3591 | BTRFS_I(inode)->root = args->root; | |
6a63209f | 3592 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3593 | return 0; |
3594 | } | |
3595 | ||
3596 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3597 | { | |
3598 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
3599 | return args->ino == inode->i_ino && |
3600 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
3601 | } |
3602 | ||
5d4f98a2 YZ |
3603 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3604 | u64 objectid, | |
3605 | struct btrfs_root *root) | |
39279cc3 CM |
3606 | { |
3607 | struct inode *inode; | |
3608 | struct btrfs_iget_args args; | |
3609 | args.ino = objectid; | |
3610 | args.root = root; | |
3611 | ||
3612 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3613 | btrfs_init_locked_inode, | |
3614 | (void *)&args); | |
3615 | return inode; | |
3616 | } | |
3617 | ||
1a54ef8c BR |
3618 | /* Get an inode object given its location and corresponding root. |
3619 | * Returns in *is_new if the inode was read from disk | |
3620 | */ | |
3621 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
5d4f98a2 | 3622 | struct btrfs_root *root) |
1a54ef8c BR |
3623 | { |
3624 | struct inode *inode; | |
3625 | ||
3626 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3627 | if (!inode) | |
5d4f98a2 | 3628 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3629 | |
3630 | if (inode->i_state & I_NEW) { | |
3631 | BTRFS_I(inode)->root = root; | |
3632 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3633 | btrfs_read_locked_inode(inode); | |
5d4f98a2 YZ |
3634 | |
3635 | inode_tree_add(inode); | |
1a54ef8c | 3636 | unlock_new_inode(inode); |
1a54ef8c BR |
3637 | } |
3638 | ||
3639 | return inode; | |
3640 | } | |
3641 | ||
4df27c4d YZ |
3642 | static struct inode *new_simple_dir(struct super_block *s, |
3643 | struct btrfs_key *key, | |
3644 | struct btrfs_root *root) | |
3645 | { | |
3646 | struct inode *inode = new_inode(s); | |
3647 | ||
3648 | if (!inode) | |
3649 | return ERR_PTR(-ENOMEM); | |
3650 | ||
3651 | init_btrfs_i(inode); | |
3652 | ||
3653 | BTRFS_I(inode)->root = root; | |
3654 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3655 | BTRFS_I(inode)->dummy_inode = 1; | |
3656 | ||
3657 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3658 | inode->i_op = &simple_dir_inode_operations; | |
3659 | inode->i_fop = &simple_dir_operations; | |
3660 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3661 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3662 | ||
3663 | return inode; | |
3664 | } | |
3665 | ||
3de4586c | 3666 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3667 | { |
d397712b | 3668 | struct inode *inode; |
4df27c4d | 3669 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3670 | struct btrfs_root *sub_root = root; |
3671 | struct btrfs_key location; | |
76dda93c | 3672 | int index; |
5d4f98a2 | 3673 | int ret; |
39279cc3 | 3674 | |
76dda93c YZ |
3675 | dentry->d_op = &btrfs_dentry_operations; |
3676 | ||
39279cc3 CM |
3677 | if (dentry->d_name.len > BTRFS_NAME_LEN) |
3678 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3679 | |
39279cc3 | 3680 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3681 | |
39279cc3 CM |
3682 | if (ret < 0) |
3683 | return ERR_PTR(ret); | |
5f39d397 | 3684 | |
4df27c4d YZ |
3685 | if (location.objectid == 0) |
3686 | return NULL; | |
3687 | ||
3688 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
3689 | inode = btrfs_iget(dir->i_sb, &location, root); | |
3690 | return inode; | |
3691 | } | |
3692 | ||
3693 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3694 | ||
76dda93c | 3695 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3696 | ret = fixup_tree_root_location(root, dir, dentry, |
3697 | &location, &sub_root); | |
3698 | if (ret < 0) { | |
3699 | if (ret != -ENOENT) | |
3700 | inode = ERR_PTR(ret); | |
3701 | else | |
3702 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3703 | } else { | |
5d4f98a2 | 3704 | inode = btrfs_iget(dir->i_sb, &location, sub_root); |
39279cc3 | 3705 | } |
76dda93c YZ |
3706 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3707 | ||
3de4586c CM |
3708 | return inode; |
3709 | } | |
3710 | ||
76dda93c YZ |
3711 | static int btrfs_dentry_delete(struct dentry *dentry) |
3712 | { | |
3713 | struct btrfs_root *root; | |
3714 | ||
efefb143 YZ |
3715 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3716 | dentry = dentry->d_parent; | |
76dda93c | 3717 | |
efefb143 YZ |
3718 | if (dentry->d_inode) { |
3719 | root = BTRFS_I(dentry->d_inode)->root; | |
3720 | if (btrfs_root_refs(&root->root_item) == 0) | |
3721 | return 1; | |
3722 | } | |
76dda93c YZ |
3723 | return 0; |
3724 | } | |
3725 | ||
3de4586c CM |
3726 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
3727 | struct nameidata *nd) | |
3728 | { | |
3729 | struct inode *inode; | |
3730 | ||
3de4586c CM |
3731 | inode = btrfs_lookup_dentry(dir, dentry); |
3732 | if (IS_ERR(inode)) | |
3733 | return ERR_CAST(inode); | |
7b128766 | 3734 | |
39279cc3 CM |
3735 | return d_splice_alias(inode, dentry); |
3736 | } | |
3737 | ||
39279cc3 CM |
3738 | static unsigned char btrfs_filetype_table[] = { |
3739 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3740 | }; | |
3741 | ||
cbdf5a24 DW |
3742 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3743 | filldir_t filldir) | |
39279cc3 | 3744 | { |
6da6abae | 3745 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3746 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3747 | struct btrfs_item *item; | |
3748 | struct btrfs_dir_item *di; | |
3749 | struct btrfs_key key; | |
5f39d397 | 3750 | struct btrfs_key found_key; |
39279cc3 CM |
3751 | struct btrfs_path *path; |
3752 | int ret; | |
3753 | u32 nritems; | |
5f39d397 | 3754 | struct extent_buffer *leaf; |
39279cc3 CM |
3755 | int slot; |
3756 | int advance; | |
3757 | unsigned char d_type; | |
3758 | int over = 0; | |
3759 | u32 di_cur; | |
3760 | u32 di_total; | |
3761 | u32 di_len; | |
3762 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3763 | char tmp_name[32]; |
3764 | char *name_ptr; | |
3765 | int name_len; | |
39279cc3 CM |
3766 | |
3767 | /* FIXME, use a real flag for deciding about the key type */ | |
3768 | if (root->fs_info->tree_root == root) | |
3769 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3770 | |
3954401f CM |
3771 | /* special case for "." */ |
3772 | if (filp->f_pos == 0) { | |
3773 | over = filldir(dirent, ".", 1, | |
3774 | 1, inode->i_ino, | |
3775 | DT_DIR); | |
3776 | if (over) | |
3777 | return 0; | |
3778 | filp->f_pos = 1; | |
3779 | } | |
3954401f CM |
3780 | /* special case for .., just use the back ref */ |
3781 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3782 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3783 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3784 | 2, pino, DT_DIR); |
3954401f | 3785 | if (over) |
49593bfa | 3786 | return 0; |
3954401f CM |
3787 | filp->f_pos = 2; |
3788 | } | |
49593bfa DW |
3789 | path = btrfs_alloc_path(); |
3790 | path->reada = 2; | |
3791 | ||
39279cc3 CM |
3792 | btrfs_set_key_type(&key, key_type); |
3793 | key.offset = filp->f_pos; | |
49593bfa | 3794 | key.objectid = inode->i_ino; |
5f39d397 | 3795 | |
39279cc3 CM |
3796 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3797 | if (ret < 0) | |
3798 | goto err; | |
3799 | advance = 0; | |
49593bfa DW |
3800 | |
3801 | while (1) { | |
5f39d397 CM |
3802 | leaf = path->nodes[0]; |
3803 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3804 | slot = path->slots[0]; |
3805 | if (advance || slot >= nritems) { | |
49593bfa | 3806 | if (slot >= nritems - 1) { |
39279cc3 CM |
3807 | ret = btrfs_next_leaf(root, path); |
3808 | if (ret) | |
3809 | break; | |
5f39d397 CM |
3810 | leaf = path->nodes[0]; |
3811 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3812 | slot = path->slots[0]; |
3813 | } else { | |
3814 | slot++; | |
3815 | path->slots[0]++; | |
3816 | } | |
3817 | } | |
3de4586c | 3818 | |
39279cc3 | 3819 | advance = 1; |
5f39d397 CM |
3820 | item = btrfs_item_nr(leaf, slot); |
3821 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3822 | ||
3823 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3824 | break; |
5f39d397 | 3825 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3826 | break; |
5f39d397 | 3827 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3828 | continue; |
5f39d397 CM |
3829 | |
3830 | filp->f_pos = found_key.offset; | |
49593bfa | 3831 | |
39279cc3 CM |
3832 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3833 | di_cur = 0; | |
5f39d397 | 3834 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3835 | |
3836 | while (di_cur < di_total) { | |
5f39d397 CM |
3837 | struct btrfs_key location; |
3838 | ||
3839 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3840 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3841 | name_ptr = tmp_name; |
3842 | } else { | |
3843 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3844 | if (!name_ptr) { |
3845 | ret = -ENOMEM; | |
3846 | goto err; | |
3847 | } | |
5f39d397 CM |
3848 | } |
3849 | read_extent_buffer(leaf, name_ptr, | |
3850 | (unsigned long)(di + 1), name_len); | |
3851 | ||
3852 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3853 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3854 | |
3855 | /* is this a reference to our own snapshot? If so | |
3856 | * skip it | |
3857 | */ | |
3858 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3859 | location.objectid == root->root_key.objectid) { | |
3860 | over = 0; | |
3861 | goto skip; | |
3862 | } | |
5f39d397 | 3863 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3864 | found_key.offset, location.objectid, |
39279cc3 | 3865 | d_type); |
5f39d397 | 3866 | |
3de4586c | 3867 | skip: |
5f39d397 CM |
3868 | if (name_ptr != tmp_name) |
3869 | kfree(name_ptr); | |
3870 | ||
39279cc3 CM |
3871 | if (over) |
3872 | goto nopos; | |
5103e947 | 3873 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3874 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3875 | di_cur += di_len; |
3876 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3877 | } | |
3878 | } | |
49593bfa DW |
3879 | |
3880 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 3881 | if (key_type == BTRFS_DIR_INDEX_KEY) |
89f135d8 | 3882 | filp->f_pos = INT_LIMIT(off_t); |
5e591a07 YZ |
3883 | else |
3884 | filp->f_pos++; | |
39279cc3 CM |
3885 | nopos: |
3886 | ret = 0; | |
3887 | err: | |
39279cc3 | 3888 | btrfs_free_path(path); |
39279cc3 CM |
3889 | return ret; |
3890 | } | |
3891 | ||
3892 | int btrfs_write_inode(struct inode *inode, int wait) | |
3893 | { | |
3894 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3895 | struct btrfs_trans_handle *trans; | |
3896 | int ret = 0; | |
3897 | ||
c146afad | 3898 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3899 | return 0; |
3900 | ||
39279cc3 | 3901 | if (wait) { |
f9295749 | 3902 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3903 | btrfs_set_trans_block_group(trans, inode); |
3904 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3905 | } |
3906 | return ret; | |
3907 | } | |
3908 | ||
3909 | /* | |
54aa1f4d | 3910 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3911 | * inode changes. But, it is most likely to find the inode in cache. |
3912 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3913 | * to keep or drop this code. | |
3914 | */ | |
3915 | void btrfs_dirty_inode(struct inode *inode) | |
3916 | { | |
3917 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3918 | struct btrfs_trans_handle *trans; | |
3919 | ||
f9295749 | 3920 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3921 | btrfs_set_trans_block_group(trans, inode); |
3922 | btrfs_update_inode(trans, root, inode); | |
3923 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3924 | } |
3925 | ||
d352ac68 CM |
3926 | /* |
3927 | * find the highest existing sequence number in a directory | |
3928 | * and then set the in-memory index_cnt variable to reflect | |
3929 | * free sequence numbers | |
3930 | */ | |
aec7477b JB |
3931 | static int btrfs_set_inode_index_count(struct inode *inode) |
3932 | { | |
3933 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3934 | struct btrfs_key key, found_key; | |
3935 | struct btrfs_path *path; | |
3936 | struct extent_buffer *leaf; | |
3937 | int ret; | |
3938 | ||
3939 | key.objectid = inode->i_ino; | |
3940 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3941 | key.offset = (u64)-1; | |
3942 | ||
3943 | path = btrfs_alloc_path(); | |
3944 | if (!path) | |
3945 | return -ENOMEM; | |
3946 | ||
3947 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3948 | if (ret < 0) | |
3949 | goto out; | |
3950 | /* FIXME: we should be able to handle this */ | |
3951 | if (ret == 0) | |
3952 | goto out; | |
3953 | ret = 0; | |
3954 | ||
3955 | /* | |
3956 | * MAGIC NUMBER EXPLANATION: | |
3957 | * since we search a directory based on f_pos we have to start at 2 | |
3958 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3959 | * else has to start at 2 | |
3960 | */ | |
3961 | if (path->slots[0] == 0) { | |
3962 | BTRFS_I(inode)->index_cnt = 2; | |
3963 | goto out; | |
3964 | } | |
3965 | ||
3966 | path->slots[0]--; | |
3967 | ||
3968 | leaf = path->nodes[0]; | |
3969 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3970 | ||
3971 | if (found_key.objectid != inode->i_ino || | |
3972 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3973 | BTRFS_I(inode)->index_cnt = 2; | |
3974 | goto out; | |
3975 | } | |
3976 | ||
3977 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3978 | out: | |
3979 | btrfs_free_path(path); | |
3980 | return ret; | |
3981 | } | |
3982 | ||
d352ac68 CM |
3983 | /* |
3984 | * helper to find a free sequence number in a given directory. This current | |
3985 | * code is very simple, later versions will do smarter things in the btree | |
3986 | */ | |
3de4586c | 3987 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3988 | { |
3989 | int ret = 0; | |
3990 | ||
3991 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3992 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 3993 | if (ret) |
aec7477b JB |
3994 | return ret; |
3995 | } | |
3996 | ||
00e4e6b3 | 3997 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3998 | BTRFS_I(dir)->index_cnt++; |
3999 | ||
4000 | return ret; | |
4001 | } | |
4002 | ||
39279cc3 CM |
4003 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4004 | struct btrfs_root *root, | |
aec7477b | 4005 | struct inode *dir, |
9c58309d | 4006 | const char *name, int name_len, |
d2fb3437 YZ |
4007 | u64 ref_objectid, u64 objectid, |
4008 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4009 | { |
4010 | struct inode *inode; | |
5f39d397 | 4011 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4012 | struct btrfs_key *location; |
5f39d397 | 4013 | struct btrfs_path *path; |
9c58309d CM |
4014 | struct btrfs_inode_ref *ref; |
4015 | struct btrfs_key key[2]; | |
4016 | u32 sizes[2]; | |
4017 | unsigned long ptr; | |
39279cc3 CM |
4018 | int ret; |
4019 | int owner; | |
4020 | ||
5f39d397 CM |
4021 | path = btrfs_alloc_path(); |
4022 | BUG_ON(!path); | |
4023 | ||
39279cc3 CM |
4024 | inode = new_inode(root->fs_info->sb); |
4025 | if (!inode) | |
4026 | return ERR_PTR(-ENOMEM); | |
4027 | ||
aec7477b | 4028 | if (dir) { |
3de4586c | 4029 | ret = btrfs_set_inode_index(dir, index); |
09771430 SF |
4030 | if (ret) { |
4031 | iput(inode); | |
aec7477b | 4032 | return ERR_PTR(ret); |
09771430 | 4033 | } |
aec7477b JB |
4034 | } |
4035 | /* | |
4036 | * index_cnt is ignored for everything but a dir, | |
4037 | * btrfs_get_inode_index_count has an explanation for the magic | |
4038 | * number | |
4039 | */ | |
e02119d5 | 4040 | init_btrfs_i(inode); |
aec7477b | 4041 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 4042 | BTRFS_I(inode)->root = root; |
e02119d5 | 4043 | BTRFS_I(inode)->generation = trans->transid; |
6a63209f | 4044 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4045 | |
39279cc3 CM |
4046 | if (mode & S_IFDIR) |
4047 | owner = 0; | |
4048 | else | |
4049 | owner = 1; | |
d2fb3437 YZ |
4050 | BTRFS_I(inode)->block_group = |
4051 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4052 | |
4053 | key[0].objectid = objectid; | |
4054 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4055 | key[0].offset = 0; | |
4056 | ||
4057 | key[1].objectid = objectid; | |
4058 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4059 | key[1].offset = ref_objectid; | |
4060 | ||
4061 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4062 | sizes[1] = name_len + sizeof(*ref); | |
4063 | ||
b9473439 | 4064 | path->leave_spinning = 1; |
9c58309d CM |
4065 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4066 | if (ret != 0) | |
5f39d397 CM |
4067 | goto fail; |
4068 | ||
79683f2d | 4069 | inode->i_uid = current_fsuid(); |
8c087b51 | 4070 | |
42f15d77 | 4071 | if (dir && (dir->i_mode & S_ISGID)) { |
8c087b51 CB |
4072 | inode->i_gid = dir->i_gid; |
4073 | if (S_ISDIR(mode)) | |
4074 | mode |= S_ISGID; | |
4075 | } else | |
4076 | inode->i_gid = current_fsgid(); | |
4077 | ||
39279cc3 CM |
4078 | inode->i_mode = mode; |
4079 | inode->i_ino = objectid; | |
a76a3cd4 | 4080 | inode_set_bytes(inode, 0); |
39279cc3 | 4081 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4082 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4083 | struct btrfs_inode_item); | |
e02119d5 | 4084 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4085 | |
4086 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4087 | struct btrfs_inode_ref); | |
4088 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4089 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4090 | ptr = (unsigned long)(ref + 1); |
4091 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4092 | ||
5f39d397 CM |
4093 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4094 | btrfs_free_path(path); | |
4095 | ||
39279cc3 CM |
4096 | location = &BTRFS_I(inode)->location; |
4097 | location->objectid = objectid; | |
39279cc3 CM |
4098 | location->offset = 0; |
4099 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4100 | ||
6cbff00f CH |
4101 | btrfs_inherit_iflags(inode, dir); |
4102 | ||
94272164 CM |
4103 | if ((mode & S_IFREG)) { |
4104 | if (btrfs_test_opt(root, NODATASUM)) | |
4105 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
4106 | if (btrfs_test_opt(root, NODATACOW)) | |
4107 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
4108 | } | |
4109 | ||
39279cc3 | 4110 | insert_inode_hash(inode); |
5d4f98a2 | 4111 | inode_tree_add(inode); |
39279cc3 | 4112 | return inode; |
5f39d397 | 4113 | fail: |
aec7477b JB |
4114 | if (dir) |
4115 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4116 | btrfs_free_path(path); |
09771430 | 4117 | iput(inode); |
5f39d397 | 4118 | return ERR_PTR(ret); |
39279cc3 CM |
4119 | } |
4120 | ||
4121 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4122 | { | |
4123 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4124 | } | |
4125 | ||
d352ac68 CM |
4126 | /* |
4127 | * utility function to add 'inode' into 'parent_inode' with | |
4128 | * a give name and a given sequence number. | |
4129 | * if 'add_backref' is true, also insert a backref from the | |
4130 | * inode to the parent directory. | |
4131 | */ | |
e02119d5 CM |
4132 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4133 | struct inode *parent_inode, struct inode *inode, | |
4134 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4135 | { |
4df27c4d | 4136 | int ret = 0; |
39279cc3 | 4137 | struct btrfs_key key; |
e02119d5 | 4138 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4139 | |
4df27c4d YZ |
4140 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4141 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4142 | } else { | |
4143 | key.objectid = inode->i_ino; | |
4144 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4145 | key.offset = 0; | |
4146 | } | |
4147 | ||
4148 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4149 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4150 | key.objectid, root->root_key.objectid, | |
4151 | parent_inode->i_ino, | |
4152 | index, name, name_len); | |
4153 | } else if (add_backref) { | |
4154 | ret = btrfs_insert_inode_ref(trans, root, | |
4155 | name, name_len, inode->i_ino, | |
4156 | parent_inode->i_ino, index); | |
4157 | } | |
39279cc3 | 4158 | |
39279cc3 | 4159 | if (ret == 0) { |
4df27c4d YZ |
4160 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4161 | parent_inode->i_ino, &key, | |
4162 | btrfs_inode_type(inode), index); | |
4163 | BUG_ON(ret); | |
4164 | ||
dbe674a9 | 4165 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4166 | name_len * 2); |
79c44584 | 4167 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4168 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4169 | } |
4170 | return ret; | |
4171 | } | |
4172 | ||
4173 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 4174 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 4175 | int backref, u64 index) |
39279cc3 | 4176 | { |
e02119d5 CM |
4177 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4178 | inode, dentry->d_name.name, | |
4179 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
4180 | if (!err) { |
4181 | d_instantiate(dentry, inode); | |
4182 | return 0; | |
4183 | } | |
4184 | if (err > 0) | |
4185 | err = -EEXIST; | |
4186 | return err; | |
4187 | } | |
4188 | ||
618e21d5 JB |
4189 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4190 | int mode, dev_t rdev) | |
4191 | { | |
4192 | struct btrfs_trans_handle *trans; | |
4193 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4194 | struct inode *inode = NULL; |
618e21d5 JB |
4195 | int err; |
4196 | int drop_inode = 0; | |
4197 | u64 objectid; | |
1832a6d5 | 4198 | unsigned long nr = 0; |
00e4e6b3 | 4199 | u64 index = 0; |
618e21d5 JB |
4200 | |
4201 | if (!new_valid_dev(rdev)) | |
4202 | return -EINVAL; | |
4203 | ||
9ed74f2d JB |
4204 | /* |
4205 | * 2 for inode item and ref | |
4206 | * 2 for dir items | |
4207 | * 1 for xattr if selinux is on | |
4208 | */ | |
4209 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4210 | if (err) |
9ed74f2d | 4211 | return err; |
1832a6d5 | 4212 | |
618e21d5 | 4213 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4214 | if (!trans) |
4215 | goto fail; | |
618e21d5 JB |
4216 | btrfs_set_trans_block_group(trans, dir); |
4217 | ||
4218 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4219 | if (err) { | |
4220 | err = -ENOSPC; | |
4221 | goto out_unlock; | |
4222 | } | |
4223 | ||
aec7477b | 4224 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4225 | dentry->d_name.len, |
4226 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 4227 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4228 | err = PTR_ERR(inode); |
4229 | if (IS_ERR(inode)) | |
4230 | goto out_unlock; | |
4231 | ||
0279b4cd | 4232 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4233 | if (err) { |
4234 | drop_inode = 1; | |
4235 | goto out_unlock; | |
4236 | } | |
4237 | ||
618e21d5 | 4238 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4239 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
4240 | if (err) |
4241 | drop_inode = 1; | |
4242 | else { | |
4243 | inode->i_op = &btrfs_special_inode_operations; | |
4244 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4245 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4246 | } |
618e21d5 JB |
4247 | btrfs_update_inode_block_group(trans, inode); |
4248 | btrfs_update_inode_block_group(trans, dir); | |
4249 | out_unlock: | |
d3c2fdcf | 4250 | nr = trans->blocks_used; |
89ce8a63 | 4251 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4252 | fail: |
9ed74f2d | 4253 | btrfs_unreserve_metadata_space(root, 5); |
618e21d5 JB |
4254 | if (drop_inode) { |
4255 | inode_dec_link_count(inode); | |
4256 | iput(inode); | |
4257 | } | |
d3c2fdcf | 4258 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4259 | return err; |
4260 | } | |
4261 | ||
39279cc3 CM |
4262 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4263 | int mode, struct nameidata *nd) | |
4264 | { | |
4265 | struct btrfs_trans_handle *trans; | |
4266 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4267 | struct inode *inode = NULL; |
39279cc3 CM |
4268 | int err; |
4269 | int drop_inode = 0; | |
1832a6d5 | 4270 | unsigned long nr = 0; |
39279cc3 | 4271 | u64 objectid; |
00e4e6b3 | 4272 | u64 index = 0; |
39279cc3 | 4273 | |
9ed74f2d JB |
4274 | /* |
4275 | * 2 for inode item and ref | |
4276 | * 2 for dir items | |
4277 | * 1 for xattr if selinux is on | |
4278 | */ | |
4279 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4280 | if (err) |
9ed74f2d JB |
4281 | return err; |
4282 | ||
39279cc3 | 4283 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4284 | if (!trans) |
4285 | goto fail; | |
39279cc3 CM |
4286 | btrfs_set_trans_block_group(trans, dir); |
4287 | ||
4288 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4289 | if (err) { | |
4290 | err = -ENOSPC; | |
4291 | goto out_unlock; | |
4292 | } | |
4293 | ||
aec7477b | 4294 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4295 | dentry->d_name.len, |
4296 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
4297 | objectid, BTRFS_I(dir)->block_group, mode, |
4298 | &index); | |
39279cc3 CM |
4299 | err = PTR_ERR(inode); |
4300 | if (IS_ERR(inode)) | |
4301 | goto out_unlock; | |
4302 | ||
0279b4cd | 4303 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4304 | if (err) { |
4305 | drop_inode = 1; | |
4306 | goto out_unlock; | |
4307 | } | |
4308 | ||
39279cc3 | 4309 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4310 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4311 | if (err) |
4312 | drop_inode = 1; | |
4313 | else { | |
4314 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4315 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4316 | inode->i_fop = &btrfs_file_operations; |
4317 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4318 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4319 | } |
39279cc3 CM |
4320 | btrfs_update_inode_block_group(trans, inode); |
4321 | btrfs_update_inode_block_group(trans, dir); | |
4322 | out_unlock: | |
d3c2fdcf | 4323 | nr = trans->blocks_used; |
ab78c84d | 4324 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4325 | fail: |
9ed74f2d | 4326 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4327 | if (drop_inode) { |
4328 | inode_dec_link_count(inode); | |
4329 | iput(inode); | |
4330 | } | |
d3c2fdcf | 4331 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4332 | return err; |
4333 | } | |
4334 | ||
4335 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4336 | struct dentry *dentry) | |
4337 | { | |
4338 | struct btrfs_trans_handle *trans; | |
4339 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4340 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4341 | u64 index; |
1832a6d5 | 4342 | unsigned long nr = 0; |
39279cc3 CM |
4343 | int err; |
4344 | int drop_inode = 0; | |
4345 | ||
4346 | if (inode->i_nlink == 0) | |
4347 | return -ENOENT; | |
4348 | ||
9ed74f2d JB |
4349 | /* |
4350 | * 1 item for inode ref | |
4351 | * 2 items for dir items | |
4352 | */ | |
4353 | err = btrfs_reserve_metadata_space(root, 3); | |
1832a6d5 | 4354 | if (err) |
9ed74f2d JB |
4355 | return err; |
4356 | ||
4357 | btrfs_inc_nlink(inode); | |
4358 | ||
3de4586c | 4359 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4360 | if (err) |
4361 | goto fail; | |
4362 | ||
39279cc3 | 4363 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4364 | |
39279cc3 CM |
4365 | btrfs_set_trans_block_group(trans, dir); |
4366 | atomic_inc(&inode->i_count); | |
aec7477b | 4367 | |
00e4e6b3 | 4368 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 4369 | |
a5719521 | 4370 | if (err) { |
54aa1f4d | 4371 | drop_inode = 1; |
a5719521 YZ |
4372 | } else { |
4373 | btrfs_update_inode_block_group(trans, dir); | |
4374 | err = btrfs_update_inode(trans, root, inode); | |
4375 | BUG_ON(err); | |
4376 | btrfs_log_new_name(trans, inode, NULL, dentry->d_parent); | |
4377 | } | |
39279cc3 | 4378 | |
d3c2fdcf | 4379 | nr = trans->blocks_used; |
ab78c84d | 4380 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4381 | fail: |
9ed74f2d | 4382 | btrfs_unreserve_metadata_space(root, 3); |
39279cc3 CM |
4383 | if (drop_inode) { |
4384 | inode_dec_link_count(inode); | |
4385 | iput(inode); | |
4386 | } | |
d3c2fdcf | 4387 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4388 | return err; |
4389 | } | |
4390 | ||
39279cc3 CM |
4391 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4392 | { | |
b9d86667 | 4393 | struct inode *inode = NULL; |
39279cc3 CM |
4394 | struct btrfs_trans_handle *trans; |
4395 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4396 | int err = 0; | |
4397 | int drop_on_err = 0; | |
b9d86667 | 4398 | u64 objectid = 0; |
00e4e6b3 | 4399 | u64 index = 0; |
d3c2fdcf | 4400 | unsigned long nr = 1; |
39279cc3 | 4401 | |
9ed74f2d JB |
4402 | /* |
4403 | * 2 items for inode and ref | |
4404 | * 2 items for dir items | |
4405 | * 1 for xattr if selinux is on | |
4406 | */ | |
4407 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 4408 | if (err) |
9ed74f2d | 4409 | return err; |
1832a6d5 | 4410 | |
39279cc3 | 4411 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
4412 | if (!trans) { |
4413 | err = -ENOMEM; | |
39279cc3 CM |
4414 | goto out_unlock; |
4415 | } | |
9ed74f2d | 4416 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 CM |
4417 | |
4418 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4419 | if (err) { | |
4420 | err = -ENOSPC; | |
4421 | goto out_unlock; | |
4422 | } | |
4423 | ||
aec7477b | 4424 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4425 | dentry->d_name.len, |
4426 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4427 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4428 | &index); | |
39279cc3 CM |
4429 | if (IS_ERR(inode)) { |
4430 | err = PTR_ERR(inode); | |
4431 | goto out_fail; | |
4432 | } | |
5f39d397 | 4433 | |
39279cc3 | 4434 | drop_on_err = 1; |
33268eaf | 4435 | |
0279b4cd | 4436 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
4437 | if (err) |
4438 | goto out_fail; | |
4439 | ||
39279cc3 CM |
4440 | inode->i_op = &btrfs_dir_inode_operations; |
4441 | inode->i_fop = &btrfs_dir_file_operations; | |
4442 | btrfs_set_trans_block_group(trans, inode); | |
4443 | ||
dbe674a9 | 4444 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4445 | err = btrfs_update_inode(trans, root, inode); |
4446 | if (err) | |
4447 | goto out_fail; | |
5f39d397 | 4448 | |
e02119d5 CM |
4449 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
4450 | inode, dentry->d_name.name, | |
4451 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4452 | if (err) |
4453 | goto out_fail; | |
5f39d397 | 4454 | |
39279cc3 CM |
4455 | d_instantiate(dentry, inode); |
4456 | drop_on_err = 0; | |
39279cc3 CM |
4457 | btrfs_update_inode_block_group(trans, inode); |
4458 | btrfs_update_inode_block_group(trans, dir); | |
4459 | ||
4460 | out_fail: | |
d3c2fdcf | 4461 | nr = trans->blocks_used; |
ab78c84d | 4462 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 4463 | |
39279cc3 | 4464 | out_unlock: |
9ed74f2d | 4465 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
4466 | if (drop_on_err) |
4467 | iput(inode); | |
d3c2fdcf | 4468 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4469 | return err; |
4470 | } | |
4471 | ||
d352ac68 CM |
4472 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4473 | * and an extent that you want to insert, deal with overlap and insert | |
4474 | * the new extent into the tree. | |
4475 | */ | |
3b951516 CM |
4476 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4477 | struct extent_map *existing, | |
e6dcd2dc CM |
4478 | struct extent_map *em, |
4479 | u64 map_start, u64 map_len) | |
3b951516 CM |
4480 | { |
4481 | u64 start_diff; | |
3b951516 | 4482 | |
e6dcd2dc CM |
4483 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4484 | start_diff = map_start - em->start; | |
4485 | em->start = map_start; | |
4486 | em->len = map_len; | |
c8b97818 CM |
4487 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4488 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4489 | em->block_start += start_diff; |
c8b97818 CM |
4490 | em->block_len -= start_diff; |
4491 | } | |
e6dcd2dc | 4492 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4493 | } |
4494 | ||
c8b97818 CM |
4495 | static noinline int uncompress_inline(struct btrfs_path *path, |
4496 | struct inode *inode, struct page *page, | |
4497 | size_t pg_offset, u64 extent_offset, | |
4498 | struct btrfs_file_extent_item *item) | |
4499 | { | |
4500 | int ret; | |
4501 | struct extent_buffer *leaf = path->nodes[0]; | |
4502 | char *tmp; | |
4503 | size_t max_size; | |
4504 | unsigned long inline_size; | |
4505 | unsigned long ptr; | |
4506 | ||
4507 | WARN_ON(pg_offset != 0); | |
4508 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
4509 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4510 | btrfs_item_nr(leaf, path->slots[0])); | |
4511 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4512 | ptr = btrfs_file_extent_inline_start(item); | |
4513 | ||
4514 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4515 | ||
5b050f04 | 4516 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
4517 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
4518 | inline_size, max_size); | |
4519 | if (ret) { | |
4520 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4521 | unsigned long copy_size = min_t(u64, | |
4522 | PAGE_CACHE_SIZE - pg_offset, | |
4523 | max_size - extent_offset); | |
4524 | memset(kaddr + pg_offset, 0, copy_size); | |
4525 | kunmap_atomic(kaddr, KM_USER0); | |
4526 | } | |
4527 | kfree(tmp); | |
4528 | return 0; | |
4529 | } | |
4530 | ||
d352ac68 CM |
4531 | /* |
4532 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4533 | * the ugly parts come from merging extents from the disk with the in-ram |
4534 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4535 | * where the in-ram extents might be locked pending data=ordered completion. |
4536 | * | |
4537 | * This also copies inline extents directly into the page. | |
4538 | */ | |
d397712b | 4539 | |
a52d9a80 | 4540 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4541 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4542 | int create) |
4543 | { | |
4544 | int ret; | |
4545 | int err = 0; | |
db94535d | 4546 | u64 bytenr; |
a52d9a80 CM |
4547 | u64 extent_start = 0; |
4548 | u64 extent_end = 0; | |
4549 | u64 objectid = inode->i_ino; | |
4550 | u32 found_type; | |
f421950f | 4551 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4552 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4553 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4554 | struct extent_buffer *leaf; |
4555 | struct btrfs_key found_key; | |
a52d9a80 CM |
4556 | struct extent_map *em = NULL; |
4557 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4558 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4559 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 4560 | int compressed; |
a52d9a80 | 4561 | |
a52d9a80 | 4562 | again: |
890871be | 4563 | read_lock(&em_tree->lock); |
d1310b2e | 4564 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4565 | if (em) |
4566 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4567 | read_unlock(&em_tree->lock); |
d1310b2e | 4568 | |
a52d9a80 | 4569 | if (em) { |
e1c4b745 CM |
4570 | if (em->start > start || em->start + em->len <= start) |
4571 | free_extent_map(em); | |
4572 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4573 | free_extent_map(em); |
4574 | else | |
4575 | goto out; | |
a52d9a80 | 4576 | } |
d1310b2e | 4577 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 4578 | if (!em) { |
d1310b2e CM |
4579 | err = -ENOMEM; |
4580 | goto out; | |
a52d9a80 | 4581 | } |
e6dcd2dc | 4582 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4583 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4584 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4585 | em->len = (u64)-1; |
c8b97818 | 4586 | em->block_len = (u64)-1; |
f421950f CM |
4587 | |
4588 | if (!path) { | |
4589 | path = btrfs_alloc_path(); | |
4590 | BUG_ON(!path); | |
4591 | } | |
4592 | ||
179e29e4 CM |
4593 | ret = btrfs_lookup_file_extent(trans, root, path, |
4594 | objectid, start, trans != NULL); | |
a52d9a80 CM |
4595 | if (ret < 0) { |
4596 | err = ret; | |
4597 | goto out; | |
4598 | } | |
4599 | ||
4600 | if (ret != 0) { | |
4601 | if (path->slots[0] == 0) | |
4602 | goto not_found; | |
4603 | path->slots[0]--; | |
4604 | } | |
4605 | ||
5f39d397 CM |
4606 | leaf = path->nodes[0]; |
4607 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 4608 | struct btrfs_file_extent_item); |
a52d9a80 | 4609 | /* are we inside the extent that was found? */ |
5f39d397 CM |
4610 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4611 | found_type = btrfs_key_type(&found_key); | |
4612 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
4613 | found_type != BTRFS_EXTENT_DATA_KEY) { |
4614 | goto not_found; | |
4615 | } | |
4616 | ||
5f39d397 CM |
4617 | found_type = btrfs_file_extent_type(leaf, item); |
4618 | extent_start = found_key.offset; | |
c8b97818 | 4619 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
4620 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4621 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 4622 | extent_end = extent_start + |
db94535d | 4623 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
4624 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
4625 | size_t size; | |
4626 | size = btrfs_file_extent_inline_len(leaf, item); | |
4627 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
4628 | ~((u64)root->sectorsize - 1); | |
4629 | } | |
4630 | ||
4631 | if (start >= extent_end) { | |
4632 | path->slots[0]++; | |
4633 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
4634 | ret = btrfs_next_leaf(root, path); | |
4635 | if (ret < 0) { | |
4636 | err = ret; | |
4637 | goto out; | |
a52d9a80 | 4638 | } |
9036c102 YZ |
4639 | if (ret > 0) |
4640 | goto not_found; | |
4641 | leaf = path->nodes[0]; | |
a52d9a80 | 4642 | } |
9036c102 YZ |
4643 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4644 | if (found_key.objectid != objectid || | |
4645 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
4646 | goto not_found; | |
4647 | if (start + len <= found_key.offset) | |
4648 | goto not_found; | |
4649 | em->start = start; | |
4650 | em->len = found_key.offset - start; | |
4651 | goto not_found_em; | |
4652 | } | |
4653 | ||
d899e052 YZ |
4654 | if (found_type == BTRFS_FILE_EXTENT_REG || |
4655 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
4656 | em->start = extent_start; |
4657 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
4658 | em->orig_start = extent_start - |
4659 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
4660 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
4661 | if (bytenr == 0) { | |
5f39d397 | 4662 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
4663 | goto insert; |
4664 | } | |
c8b97818 CM |
4665 | if (compressed) { |
4666 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
4667 | em->block_start = bytenr; | |
4668 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
4669 | item); | |
4670 | } else { | |
4671 | bytenr += btrfs_file_extent_offset(leaf, item); | |
4672 | em->block_start = bytenr; | |
4673 | em->block_len = em->len; | |
d899e052 YZ |
4674 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
4675 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 4676 | } |
a52d9a80 CM |
4677 | goto insert; |
4678 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4679 | unsigned long ptr; |
a52d9a80 | 4680 | char *map; |
3326d1b0 CM |
4681 | size_t size; |
4682 | size_t extent_offset; | |
4683 | size_t copy_size; | |
a52d9a80 | 4684 | |
689f9346 | 4685 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 4686 | if (!page || create) { |
689f9346 | 4687 | em->start = extent_start; |
9036c102 | 4688 | em->len = extent_end - extent_start; |
689f9346 Y |
4689 | goto out; |
4690 | } | |
5f39d397 | 4691 | |
9036c102 YZ |
4692 | size = btrfs_file_extent_inline_len(leaf, item); |
4693 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4694 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4695 | size - extent_offset); |
3326d1b0 | 4696 | em->start = extent_start + extent_offset; |
70dec807 CM |
4697 | em->len = (copy_size + root->sectorsize - 1) & |
4698 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4699 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4700 | if (compressed) |
4701 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4702 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4703 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4704 | if (btrfs_file_extent_compression(leaf, item) == |
4705 | BTRFS_COMPRESS_ZLIB) { | |
4706 | ret = uncompress_inline(path, inode, page, | |
4707 | pg_offset, | |
4708 | extent_offset, item); | |
4709 | BUG_ON(ret); | |
4710 | } else { | |
4711 | map = kmap(page); | |
4712 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4713 | copy_size); | |
93c82d57 CM |
4714 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
4715 | memset(map + pg_offset + copy_size, 0, | |
4716 | PAGE_CACHE_SIZE - pg_offset - | |
4717 | copy_size); | |
4718 | } | |
c8b97818 CM |
4719 | kunmap(page); |
4720 | } | |
179e29e4 CM |
4721 | flush_dcache_page(page); |
4722 | } else if (create && PageUptodate(page)) { | |
4723 | if (!trans) { | |
4724 | kunmap(page); | |
4725 | free_extent_map(em); | |
4726 | em = NULL; | |
4727 | btrfs_release_path(root, path); | |
f9295749 | 4728 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4729 | goto again; |
4730 | } | |
c8b97818 | 4731 | map = kmap(page); |
70dec807 | 4732 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4733 | copy_size); |
c8b97818 | 4734 | kunmap(page); |
179e29e4 | 4735 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4736 | } |
d1310b2e CM |
4737 | set_extent_uptodate(io_tree, em->start, |
4738 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4739 | goto insert; |
4740 | } else { | |
d397712b | 4741 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
4742 | WARN_ON(1); |
4743 | } | |
4744 | not_found: | |
4745 | em->start = start; | |
d1310b2e | 4746 | em->len = len; |
a52d9a80 | 4747 | not_found_em: |
5f39d397 | 4748 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4749 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4750 | insert: |
4751 | btrfs_release_path(root, path); | |
d1310b2e | 4752 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
4753 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
4754 | "[%llu %llu]\n", (unsigned long long)em->start, | |
4755 | (unsigned long long)em->len, | |
4756 | (unsigned long long)start, | |
4757 | (unsigned long long)len); | |
a52d9a80 CM |
4758 | err = -EIO; |
4759 | goto out; | |
4760 | } | |
d1310b2e CM |
4761 | |
4762 | err = 0; | |
890871be | 4763 | write_lock(&em_tree->lock); |
a52d9a80 | 4764 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4765 | /* it is possible that someone inserted the extent into the tree |
4766 | * while we had the lock dropped. It is also possible that | |
4767 | * an overlapping map exists in the tree | |
4768 | */ | |
a52d9a80 | 4769 | if (ret == -EEXIST) { |
3b951516 | 4770 | struct extent_map *existing; |
e6dcd2dc CM |
4771 | |
4772 | ret = 0; | |
4773 | ||
3b951516 | 4774 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4775 | if (existing && (existing->start > start || |
4776 | existing->start + existing->len <= start)) { | |
4777 | free_extent_map(existing); | |
4778 | existing = NULL; | |
4779 | } | |
3b951516 CM |
4780 | if (!existing) { |
4781 | existing = lookup_extent_mapping(em_tree, em->start, | |
4782 | em->len); | |
4783 | if (existing) { | |
4784 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4785 | em, start, |
4786 | root->sectorsize); | |
3b951516 CM |
4787 | free_extent_map(existing); |
4788 | if (err) { | |
4789 | free_extent_map(em); | |
4790 | em = NULL; | |
4791 | } | |
4792 | } else { | |
4793 | err = -EIO; | |
3b951516 CM |
4794 | free_extent_map(em); |
4795 | em = NULL; | |
4796 | } | |
4797 | } else { | |
4798 | free_extent_map(em); | |
4799 | em = existing; | |
e6dcd2dc | 4800 | err = 0; |
a52d9a80 | 4801 | } |
a52d9a80 | 4802 | } |
890871be | 4803 | write_unlock(&em_tree->lock); |
a52d9a80 | 4804 | out: |
f421950f CM |
4805 | if (path) |
4806 | btrfs_free_path(path); | |
a52d9a80 CM |
4807 | if (trans) { |
4808 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 4809 | if (!err) |
a52d9a80 CM |
4810 | err = ret; |
4811 | } | |
a52d9a80 CM |
4812 | if (err) { |
4813 | free_extent_map(em); | |
a52d9a80 CM |
4814 | return ERR_PTR(err); |
4815 | } | |
4816 | return em; | |
4817 | } | |
4818 | ||
16432985 CM |
4819 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4820 | const struct iovec *iov, loff_t offset, | |
4821 | unsigned long nr_segs) | |
4822 | { | |
e1c4b745 | 4823 | return -EINVAL; |
16432985 CM |
4824 | } |
4825 | ||
1506fcc8 YS |
4826 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
4827 | __u64 start, __u64 len) | |
4828 | { | |
4829 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent); | |
4830 | } | |
4831 | ||
a52d9a80 | 4832 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4833 | { |
d1310b2e CM |
4834 | struct extent_io_tree *tree; |
4835 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4836 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4837 | } |
1832a6d5 | 4838 | |
a52d9a80 | 4839 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4840 | { |
d1310b2e | 4841 | struct extent_io_tree *tree; |
b888db2b CM |
4842 | |
4843 | ||
4844 | if (current->flags & PF_MEMALLOC) { | |
4845 | redirty_page_for_writepage(wbc, page); | |
4846 | unlock_page(page); | |
4847 | return 0; | |
4848 | } | |
d1310b2e | 4849 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4850 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4851 | } |
4852 | ||
f421950f CM |
4853 | int btrfs_writepages(struct address_space *mapping, |
4854 | struct writeback_control *wbc) | |
b293f02e | 4855 | { |
d1310b2e | 4856 | struct extent_io_tree *tree; |
771ed689 | 4857 | |
d1310b2e | 4858 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4859 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4860 | } | |
4861 | ||
3ab2fb5a CM |
4862 | static int |
4863 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4864 | struct list_head *pages, unsigned nr_pages) | |
4865 | { | |
d1310b2e CM |
4866 | struct extent_io_tree *tree; |
4867 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4868 | return extent_readpages(tree, mapping, pages, nr_pages, |
4869 | btrfs_get_extent); | |
4870 | } | |
e6dcd2dc | 4871 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4872 | { |
d1310b2e CM |
4873 | struct extent_io_tree *tree; |
4874 | struct extent_map_tree *map; | |
a52d9a80 | 4875 | int ret; |
8c2383c3 | 4876 | |
d1310b2e CM |
4877 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4878 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4879 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4880 | if (ret == 1) { |
4881 | ClearPagePrivate(page); | |
4882 | set_page_private(page, 0); | |
4883 | page_cache_release(page); | |
39279cc3 | 4884 | } |
a52d9a80 | 4885 | return ret; |
39279cc3 CM |
4886 | } |
4887 | ||
e6dcd2dc CM |
4888 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4889 | { | |
98509cfc CM |
4890 | if (PageWriteback(page) || PageDirty(page)) |
4891 | return 0; | |
b335b003 | 4892 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
4893 | } |
4894 | ||
a52d9a80 | 4895 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4896 | { |
d1310b2e | 4897 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4898 | struct btrfs_ordered_extent *ordered; |
4899 | u64 page_start = page_offset(page); | |
4900 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4901 | |
8b62b72b CM |
4902 | |
4903 | /* | |
4904 | * we have the page locked, so new writeback can't start, | |
4905 | * and the dirty bit won't be cleared while we are here. | |
4906 | * | |
4907 | * Wait for IO on this page so that we can safely clear | |
4908 | * the PagePrivate2 bit and do ordered accounting | |
4909 | */ | |
e6dcd2dc | 4910 | wait_on_page_writeback(page); |
8b62b72b | 4911 | |
d1310b2e | 4912 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4913 | if (offset) { |
4914 | btrfs_releasepage(page, GFP_NOFS); | |
4915 | return; | |
4916 | } | |
e6dcd2dc CM |
4917 | lock_extent(tree, page_start, page_end, GFP_NOFS); |
4918 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4919 | page_offset(page)); | |
4920 | if (ordered) { | |
eb84ae03 CM |
4921 | /* |
4922 | * IO on this page will never be started, so we need | |
4923 | * to account for any ordered extents now | |
4924 | */ | |
e6dcd2dc CM |
4925 | clear_extent_bit(tree, page_start, page_end, |
4926 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff JB |
4927 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
4928 | NULL, GFP_NOFS); | |
8b62b72b CM |
4929 | /* |
4930 | * whoever cleared the private bit is responsible | |
4931 | * for the finish_ordered_io | |
4932 | */ | |
4933 | if (TestClearPagePrivate2(page)) { | |
4934 | btrfs_finish_ordered_io(page->mapping->host, | |
4935 | page_start, page_end); | |
4936 | } | |
e6dcd2dc CM |
4937 | btrfs_put_ordered_extent(ordered); |
4938 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4939 | } | |
4940 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff JB |
4941 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
4942 | EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS); | |
e6dcd2dc CM |
4943 | __btrfs_releasepage(page, GFP_NOFS); |
4944 | ||
4a096752 | 4945 | ClearPageChecked(page); |
9ad6b7bc | 4946 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4947 | ClearPagePrivate(page); |
4948 | set_page_private(page, 0); | |
4949 | page_cache_release(page); | |
4950 | } | |
39279cc3 CM |
4951 | } |
4952 | ||
9ebefb18 CM |
4953 | /* |
4954 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4955 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4956 | * be careful to check for EOF conditions here. We set the page up correctly | |
4957 | * for a written page which means we get ENOSPC checking when writing into | |
4958 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4959 | * support these features. | |
4960 | * | |
4961 | * We are not allowed to take the i_mutex here so we have to play games to | |
4962 | * protect against truncate races as the page could now be beyond EOF. Because | |
4963 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4964 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4965 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4966 | * unlock the page. | |
4967 | */ | |
c2ec175c | 4968 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 4969 | { |
c2ec175c | 4970 | struct page *page = vmf->page; |
6da6abae | 4971 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4972 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4973 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4974 | struct btrfs_ordered_extent *ordered; | |
4975 | char *kaddr; | |
4976 | unsigned long zero_start; | |
9ebefb18 | 4977 | loff_t size; |
1832a6d5 | 4978 | int ret; |
a52d9a80 | 4979 | u64 page_start; |
e6dcd2dc | 4980 | u64 page_end; |
9ebefb18 | 4981 | |
6a63209f | 4982 | ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
4983 | if (ret) { |
4984 | if (ret == -ENOMEM) | |
4985 | ret = VM_FAULT_OOM; | |
4986 | else /* -ENOSPC, -EIO, etc */ | |
4987 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 4988 | goto out; |
56a76f82 | 4989 | } |
1832a6d5 | 4990 | |
9ed74f2d JB |
4991 | ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); |
4992 | if (ret) { | |
4993 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); | |
4994 | ret = VM_FAULT_SIGBUS; | |
4995 | goto out; | |
4996 | } | |
4997 | ||
56a76f82 | 4998 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 4999 | again: |
9ebefb18 | 5000 | lock_page(page); |
9ebefb18 | 5001 | size = i_size_read(inode); |
e6dcd2dc CM |
5002 | page_start = page_offset(page); |
5003 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 5004 | |
9ebefb18 | 5005 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 5006 | (page_start >= size)) { |
6a63209f | 5007 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ebefb18 CM |
5008 | /* page got truncated out from underneath us */ |
5009 | goto out_unlock; | |
5010 | } | |
e6dcd2dc CM |
5011 | wait_on_page_writeback(page); |
5012 | ||
5013 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5014 | set_page_extent_mapped(page); | |
5015 | ||
eb84ae03 CM |
5016 | /* |
5017 | * we can't set the delalloc bits if there are pending ordered | |
5018 | * extents. Drop our locks and wait for them to finish | |
5019 | */ | |
e6dcd2dc CM |
5020 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5021 | if (ordered) { | |
5022 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5023 | unlock_page(page); | |
eb84ae03 | 5024 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
5025 | btrfs_put_ordered_extent(ordered); |
5026 | goto again; | |
5027 | } | |
5028 | ||
fbf19087 JB |
5029 | /* |
5030 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
5031 | * if it was already dirty, so for space accounting reasons we need to | |
5032 | * clear any delalloc bits for the range we are fixing to save. There | |
5033 | * is probably a better way to do this, but for now keep consistent with | |
5034 | * prepare_pages in the normal write path. | |
5035 | */ | |
5036 | clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
32c00aff JB |
5037 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
5038 | GFP_NOFS); | |
fbf19087 | 5039 | |
9ed74f2d JB |
5040 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end); |
5041 | if (ret) { | |
5042 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5043 | ret = VM_FAULT_SIGBUS; | |
fbf19087 | 5044 | btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE); |
9ed74f2d JB |
5045 | goto out_unlock; |
5046 | } | |
e6dcd2dc | 5047 | ret = 0; |
9ebefb18 CM |
5048 | |
5049 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 5050 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 5051 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 5052 | else |
e6dcd2dc | 5053 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 5054 | |
e6dcd2dc CM |
5055 | if (zero_start != PAGE_CACHE_SIZE) { |
5056 | kaddr = kmap(page); | |
5057 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
5058 | flush_dcache_page(page); | |
5059 | kunmap(page); | |
5060 | } | |
247e743c | 5061 | ClearPageChecked(page); |
e6dcd2dc | 5062 | set_page_dirty(page); |
50a9b214 | 5063 | SetPageUptodate(page); |
5a3f23d5 | 5064 | |
257c62e1 CM |
5065 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
5066 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
5067 | ||
e6dcd2dc | 5068 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
9ebefb18 CM |
5069 | |
5070 | out_unlock: | |
9ed74f2d | 5071 | btrfs_unreserve_metadata_for_delalloc(root, inode, 1); |
50a9b214 CM |
5072 | if (!ret) |
5073 | return VM_FAULT_LOCKED; | |
9ebefb18 | 5074 | unlock_page(page); |
1832a6d5 | 5075 | out: |
9ebefb18 CM |
5076 | return ret; |
5077 | } | |
5078 | ||
39279cc3 CM |
5079 | static void btrfs_truncate(struct inode *inode) |
5080 | { | |
5081 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5082 | int ret; | |
5083 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 5084 | unsigned long nr; |
dbe674a9 | 5085 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
5086 | |
5087 | if (!S_ISREG(inode->i_mode)) | |
5088 | return; | |
5089 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
5090 | return; | |
5091 | ||
5d5e103a JB |
5092 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
5093 | if (ret) | |
5094 | return; | |
4a096752 | 5095 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 5096 | |
39279cc3 | 5097 | trans = btrfs_start_transaction(root, 1); |
5a3f23d5 CM |
5098 | |
5099 | /* | |
5100 | * setattr is responsible for setting the ordered_data_close flag, | |
5101 | * but that is only tested during the last file release. That | |
5102 | * could happen well after the next commit, leaving a great big | |
5103 | * window where new writes may get lost if someone chooses to write | |
5104 | * to this file after truncating to zero | |
5105 | * | |
5106 | * The inode doesn't have any dirty data here, and so if we commit | |
5107 | * this is a noop. If someone immediately starts writing to the inode | |
5108 | * it is very likely we'll catch some of their writes in this | |
5109 | * transaction, and the commit will find this file on the ordered | |
5110 | * data list with good things to send down. | |
5111 | * | |
5112 | * This is a best effort solution, there is still a window where | |
5113 | * using truncate to replace the contents of the file will | |
5114 | * end up with a zero length file after a crash. | |
5115 | */ | |
5116 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
5117 | btrfs_add_ordered_operation(trans, root, inode); | |
5118 | ||
39279cc3 | 5119 | btrfs_set_trans_block_group(trans, inode); |
dbe674a9 | 5120 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 5121 | |
7b128766 JB |
5122 | ret = btrfs_orphan_add(trans, inode); |
5123 | if (ret) | |
5124 | goto out; | |
39279cc3 | 5125 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 5126 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 5127 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 5128 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 5129 | |
7b128766 JB |
5130 | ret = btrfs_orphan_del(trans, inode); |
5131 | BUG_ON(ret); | |
5132 | ||
5133 | out: | |
5134 | nr = trans->blocks_used; | |
89ce8a63 | 5135 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 5136 | BUG_ON(ret); |
d3c2fdcf | 5137 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5138 | } |
5139 | ||
d352ac68 CM |
5140 | /* |
5141 | * create a new subvolume directory/inode (helper for the ioctl). | |
5142 | */ | |
d2fb3437 | 5143 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 5144 | struct btrfs_root *new_root, |
d2fb3437 | 5145 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 5146 | { |
39279cc3 | 5147 | struct inode *inode; |
76dda93c | 5148 | int err; |
00e4e6b3 | 5149 | u64 index = 0; |
39279cc3 | 5150 | |
aec7477b | 5151 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 5152 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 5153 | if (IS_ERR(inode)) |
f46b5a66 | 5154 | return PTR_ERR(inode); |
39279cc3 CM |
5155 | inode->i_op = &btrfs_dir_inode_operations; |
5156 | inode->i_fop = &btrfs_dir_file_operations; | |
5157 | ||
39279cc3 | 5158 | inode->i_nlink = 1; |
dbe674a9 | 5159 | btrfs_i_size_write(inode, 0); |
3b96362c | 5160 | |
76dda93c YZ |
5161 | err = btrfs_update_inode(trans, new_root, inode); |
5162 | BUG_ON(err); | |
cb8e7090 | 5163 | |
76dda93c | 5164 | iput(inode); |
cb8e7090 | 5165 | return 0; |
39279cc3 CM |
5166 | } |
5167 | ||
d352ac68 CM |
5168 | /* helper function for file defrag and space balancing. This |
5169 | * forces readahead on a given range of bytes in an inode | |
5170 | */ | |
edbd8d4e | 5171 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
5172 | struct file_ra_state *ra, struct file *file, |
5173 | pgoff_t offset, pgoff_t last_index) | |
5174 | { | |
8e7bf94f | 5175 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 5176 | |
86479a04 CM |
5177 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
5178 | return offset + req_size; | |
86479a04 CM |
5179 | } |
5180 | ||
39279cc3 CM |
5181 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
5182 | { | |
5183 | struct btrfs_inode *ei; | |
5184 | ||
5185 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
5186 | if (!ei) | |
5187 | return NULL; | |
15ee9bc7 | 5188 | ei->last_trans = 0; |
257c62e1 | 5189 | ei->last_sub_trans = 0; |
e02119d5 | 5190 | ei->logged_trans = 0; |
32c00aff JB |
5191 | ei->outstanding_extents = 0; |
5192 | ei->reserved_extents = 0; | |
a6dbd429 | 5193 | ei->root = NULL; |
32c00aff | 5194 | spin_lock_init(&ei->accounting_lock); |
e6dcd2dc | 5195 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 5196 | INIT_LIST_HEAD(&ei->i_orphan); |
5a3f23d5 | 5197 | INIT_LIST_HEAD(&ei->ordered_operations); |
39279cc3 CM |
5198 | return &ei->vfs_inode; |
5199 | } | |
5200 | ||
5201 | void btrfs_destroy_inode(struct inode *inode) | |
5202 | { | |
e6dcd2dc | 5203 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
5204 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5205 | ||
39279cc3 CM |
5206 | WARN_ON(!list_empty(&inode->i_dentry)); |
5207 | WARN_ON(inode->i_data.nrpages); | |
5208 | ||
a6dbd429 JB |
5209 | /* |
5210 | * This can happen where we create an inode, but somebody else also | |
5211 | * created the same inode and we need to destroy the one we already | |
5212 | * created. | |
5213 | */ | |
5214 | if (!root) | |
5215 | goto free; | |
5216 | ||
5a3f23d5 CM |
5217 | /* |
5218 | * Make sure we're properly removed from the ordered operation | |
5219 | * lists. | |
5220 | */ | |
5221 | smp_mb(); | |
5222 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
5223 | spin_lock(&root->fs_info->ordered_extent_lock); | |
5224 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
5225 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
5226 | } | |
5227 | ||
5228 | spin_lock(&root->list_lock); | |
7b128766 JB |
5229 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
5230 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
5231 | " list\n", inode->i_ino); | |
5232 | dump_stack(); | |
5233 | } | |
5a3f23d5 | 5234 | spin_unlock(&root->list_lock); |
7b128766 | 5235 | |
d397712b | 5236 | while (1) { |
e6dcd2dc CM |
5237 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
5238 | if (!ordered) | |
5239 | break; | |
5240 | else { | |
d397712b CM |
5241 | printk(KERN_ERR "btrfs found ordered " |
5242 | "extent %llu %llu on inode cleanup\n", | |
5243 | (unsigned long long)ordered->file_offset, | |
5244 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
5245 | btrfs_remove_ordered_extent(inode, ordered); |
5246 | btrfs_put_ordered_extent(ordered); | |
5247 | btrfs_put_ordered_extent(ordered); | |
5248 | } | |
5249 | } | |
5d4f98a2 | 5250 | inode_tree_del(inode); |
5b21f2ed | 5251 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 5252 | free: |
39279cc3 CM |
5253 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
5254 | } | |
5255 | ||
76dda93c YZ |
5256 | void btrfs_drop_inode(struct inode *inode) |
5257 | { | |
5258 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5259 | ||
5260 | if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0) | |
5261 | generic_delete_inode(inode); | |
5262 | else | |
5263 | generic_drop_inode(inode); | |
5264 | } | |
5265 | ||
0ee0fda0 | 5266 | static void init_once(void *foo) |
39279cc3 CM |
5267 | { |
5268 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
5269 | ||
5270 | inode_init_once(&ei->vfs_inode); | |
5271 | } | |
5272 | ||
5273 | void btrfs_destroy_cachep(void) | |
5274 | { | |
5275 | if (btrfs_inode_cachep) | |
5276 | kmem_cache_destroy(btrfs_inode_cachep); | |
5277 | if (btrfs_trans_handle_cachep) | |
5278 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5279 | if (btrfs_transaction_cachep) | |
5280 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
5281 | if (btrfs_path_cachep) |
5282 | kmem_cache_destroy(btrfs_path_cachep); | |
5283 | } | |
5284 | ||
5285 | int btrfs_init_cachep(void) | |
5286 | { | |
9601e3f6 CH |
5287 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
5288 | sizeof(struct btrfs_inode), 0, | |
5289 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
5290 | if (!btrfs_inode_cachep) |
5291 | goto fail; | |
9601e3f6 CH |
5292 | |
5293 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
5294 | sizeof(struct btrfs_trans_handle), 0, | |
5295 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5296 | if (!btrfs_trans_handle_cachep) |
5297 | goto fail; | |
9601e3f6 CH |
5298 | |
5299 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
5300 | sizeof(struct btrfs_transaction), 0, | |
5301 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5302 | if (!btrfs_transaction_cachep) |
5303 | goto fail; | |
9601e3f6 CH |
5304 | |
5305 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
5306 | sizeof(struct btrfs_path), 0, | |
5307 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
5308 | if (!btrfs_path_cachep) |
5309 | goto fail; | |
9601e3f6 | 5310 | |
39279cc3 CM |
5311 | return 0; |
5312 | fail: | |
5313 | btrfs_destroy_cachep(); | |
5314 | return -ENOMEM; | |
5315 | } | |
5316 | ||
5317 | static int btrfs_getattr(struct vfsmount *mnt, | |
5318 | struct dentry *dentry, struct kstat *stat) | |
5319 | { | |
5320 | struct inode *inode = dentry->d_inode; | |
5321 | generic_fillattr(inode, stat); | |
3394e160 | 5322 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 5323 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
5324 | stat->blocks = (inode_get_bytes(inode) + |
5325 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
5326 | return 0; |
5327 | } | |
5328 | ||
d397712b CM |
5329 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
5330 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
5331 | { |
5332 | struct btrfs_trans_handle *trans; | |
5333 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 5334 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
5335 | struct inode *new_inode = new_dentry->d_inode; |
5336 | struct inode *old_inode = old_dentry->d_inode; | |
5337 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 5338 | u64 index = 0; |
4df27c4d | 5339 | u64 root_objectid; |
39279cc3 CM |
5340 | int ret; |
5341 | ||
f679a840 YZ |
5342 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
5343 | return -EPERM; | |
5344 | ||
4df27c4d YZ |
5345 | /* we only allow rename subvolume link between subvolumes */ |
5346 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
5347 | return -EXDEV; |
5348 | ||
4df27c4d YZ |
5349 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
5350 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 5351 | return -ENOTEMPTY; |
5f39d397 | 5352 | |
4df27c4d YZ |
5353 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
5354 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
5355 | return -ENOTEMPTY; | |
0660b5af | 5356 | |
9ed74f2d | 5357 | /* |
5df6a9f6 JB |
5358 | * We want to reserve the absolute worst case amount of items. So if |
5359 | * both inodes are subvols and we need to unlink them then that would | |
5360 | * require 4 item modifications, but if they are both normal inodes it | |
5361 | * would require 5 item modifications, so we'll assume their normal | |
5362 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
5363 | * should cover the worst case number of items we'll modify. | |
9ed74f2d | 5364 | */ |
5df6a9f6 | 5365 | ret = btrfs_reserve_metadata_space(root, 11); |
1832a6d5 | 5366 | if (ret) |
4df27c4d | 5367 | return ret; |
1832a6d5 | 5368 | |
5a3f23d5 CM |
5369 | /* |
5370 | * we're using rename to replace one file with another. | |
5371 | * and the replacement file is large. Start IO on it now so | |
5372 | * we don't add too much work to the end of the transaction | |
5373 | */ | |
4baf8c92 | 5374 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
5375 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
5376 | filemap_flush(old_inode->i_mapping); | |
5377 | ||
76dda93c YZ |
5378 | /* close the racy window with snapshot create/destroy ioctl */ |
5379 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
5380 | down_read(&root->fs_info->subvol_sem); | |
5381 | ||
39279cc3 | 5382 | trans = btrfs_start_transaction(root, 1); |
a5719521 | 5383 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 5384 | |
4df27c4d YZ |
5385 | if (dest != root) |
5386 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 5387 | |
a5719521 YZ |
5388 | ret = btrfs_set_inode_index(new_dir, &index); |
5389 | if (ret) | |
5390 | goto out_fail; | |
5a3f23d5 | 5391 | |
a5719521 | 5392 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
5393 | /* force full log commit if subvolume involved. */ |
5394 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
5395 | } else { | |
a5719521 YZ |
5396 | ret = btrfs_insert_inode_ref(trans, dest, |
5397 | new_dentry->d_name.name, | |
5398 | new_dentry->d_name.len, | |
5399 | old_inode->i_ino, | |
5400 | new_dir->i_ino, index); | |
5401 | if (ret) | |
5402 | goto out_fail; | |
4df27c4d YZ |
5403 | /* |
5404 | * this is an ugly little race, but the rename is required | |
5405 | * to make sure that if we crash, the inode is either at the | |
5406 | * old name or the new one. pinning the log transaction lets | |
5407 | * us make sure we don't allow a log commit to come in after | |
5408 | * we unlink the name but before we add the new name back in. | |
5409 | */ | |
5410 | btrfs_pin_log_trans(root); | |
5411 | } | |
5a3f23d5 CM |
5412 | /* |
5413 | * make sure the inode gets flushed if it is replacing | |
5414 | * something. | |
5415 | */ | |
5416 | if (new_inode && new_inode->i_size && | |
5417 | old_inode && S_ISREG(old_inode->i_mode)) { | |
5418 | btrfs_add_ordered_operation(trans, root, old_inode); | |
5419 | } | |
5420 | ||
39279cc3 CM |
5421 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
5422 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
5423 | old_inode->i_ctime = ctime; | |
5f39d397 | 5424 | |
12fcfd22 CM |
5425 | if (old_dentry->d_parent != new_dentry->d_parent) |
5426 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
5427 | ||
4df27c4d YZ |
5428 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
5429 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
5430 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
5431 | old_dentry->d_name.name, | |
5432 | old_dentry->d_name.len); | |
5433 | } else { | |
5434 | btrfs_inc_nlink(old_dentry->d_inode); | |
5435 | ret = btrfs_unlink_inode(trans, root, old_dir, | |
5436 | old_dentry->d_inode, | |
5437 | old_dentry->d_name.name, | |
5438 | old_dentry->d_name.len); | |
5439 | } | |
5440 | BUG_ON(ret); | |
39279cc3 CM |
5441 | |
5442 | if (new_inode) { | |
5443 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
5444 | if (unlikely(new_inode->i_ino == |
5445 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
5446 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
5447 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
5448 | root_objectid, | |
5449 | new_dentry->d_name.name, | |
5450 | new_dentry->d_name.len); | |
5451 | BUG_ON(new_inode->i_nlink == 0); | |
5452 | } else { | |
5453 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
5454 | new_dentry->d_inode, | |
5455 | new_dentry->d_name.name, | |
5456 | new_dentry->d_name.len); | |
5457 | } | |
5458 | BUG_ON(ret); | |
7b128766 | 5459 | if (new_inode->i_nlink == 0) { |
e02119d5 | 5460 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 5461 | BUG_ON(ret); |
7b128766 | 5462 | } |
39279cc3 | 5463 | } |
aec7477b | 5464 | |
4df27c4d YZ |
5465 | ret = btrfs_add_link(trans, new_dir, old_inode, |
5466 | new_dentry->d_name.name, | |
a5719521 | 5467 | new_dentry->d_name.len, 0, index); |
4df27c4d | 5468 | BUG_ON(ret); |
39279cc3 | 5469 | |
4df27c4d YZ |
5470 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
5471 | btrfs_log_new_name(trans, old_inode, old_dir, | |
5472 | new_dentry->d_parent); | |
5473 | btrfs_end_log_trans(root); | |
5474 | } | |
39279cc3 | 5475 | out_fail: |
ab78c84d | 5476 | btrfs_end_transaction_throttle(trans, root); |
4df27c4d | 5477 | |
76dda93c YZ |
5478 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
5479 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 5480 | |
5df6a9f6 | 5481 | btrfs_unreserve_metadata_space(root, 11); |
39279cc3 CM |
5482 | return ret; |
5483 | } | |
5484 | ||
d352ac68 CM |
5485 | /* |
5486 | * some fairly slow code that needs optimization. This walks the list | |
5487 | * of all the inodes with pending delalloc and forces them to disk. | |
5488 | */ | |
ea8c2819 CM |
5489 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
5490 | { | |
5491 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
5492 | struct btrfs_inode *binode; | |
5b21f2ed | 5493 | struct inode *inode; |
ea8c2819 | 5494 | |
c146afad YZ |
5495 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
5496 | return -EROFS; | |
5497 | ||
75eff68e | 5498 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 5499 | while (!list_empty(head)) { |
ea8c2819 CM |
5500 | binode = list_entry(head->next, struct btrfs_inode, |
5501 | delalloc_inodes); | |
5b21f2ed ZY |
5502 | inode = igrab(&binode->vfs_inode); |
5503 | if (!inode) | |
5504 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 5505 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 5506 | if (inode) { |
8c8bee1d | 5507 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
5508 | iput(inode); |
5509 | } | |
5510 | cond_resched(); | |
75eff68e | 5511 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 5512 | } |
75eff68e | 5513 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
5514 | |
5515 | /* the filemap_flush will queue IO into the worker threads, but | |
5516 | * we have to make sure the IO is actually started and that | |
5517 | * ordered extents get created before we return | |
5518 | */ | |
5519 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 5520 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 5521 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 5522 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
5523 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
5524 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
5525 | } |
5526 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
5527 | return 0; |
5528 | } | |
5529 | ||
39279cc3 CM |
5530 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
5531 | const char *symname) | |
5532 | { | |
5533 | struct btrfs_trans_handle *trans; | |
5534 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5535 | struct btrfs_path *path; | |
5536 | struct btrfs_key key; | |
1832a6d5 | 5537 | struct inode *inode = NULL; |
39279cc3 CM |
5538 | int err; |
5539 | int drop_inode = 0; | |
5540 | u64 objectid; | |
00e4e6b3 | 5541 | u64 index = 0 ; |
39279cc3 CM |
5542 | int name_len; |
5543 | int datasize; | |
5f39d397 | 5544 | unsigned long ptr; |
39279cc3 | 5545 | struct btrfs_file_extent_item *ei; |
5f39d397 | 5546 | struct extent_buffer *leaf; |
1832a6d5 | 5547 | unsigned long nr = 0; |
39279cc3 CM |
5548 | |
5549 | name_len = strlen(symname) + 1; | |
5550 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
5551 | return -ENAMETOOLONG; | |
1832a6d5 | 5552 | |
9ed74f2d JB |
5553 | /* |
5554 | * 2 items for inode item and ref | |
5555 | * 2 items for dir items | |
5556 | * 1 item for xattr if selinux is on | |
5557 | */ | |
5558 | err = btrfs_reserve_metadata_space(root, 5); | |
1832a6d5 | 5559 | if (err) |
9ed74f2d | 5560 | return err; |
1832a6d5 | 5561 | |
39279cc3 | 5562 | trans = btrfs_start_transaction(root, 1); |
9ed74f2d JB |
5563 | if (!trans) |
5564 | goto out_fail; | |
39279cc3 CM |
5565 | btrfs_set_trans_block_group(trans, dir); |
5566 | ||
5567 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
5568 | if (err) { | |
5569 | err = -ENOSPC; | |
5570 | goto out_unlock; | |
5571 | } | |
5572 | ||
aec7477b | 5573 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
5574 | dentry->d_name.len, |
5575 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
5576 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
5577 | &index); | |
39279cc3 CM |
5578 | err = PTR_ERR(inode); |
5579 | if (IS_ERR(inode)) | |
5580 | goto out_unlock; | |
5581 | ||
0279b4cd | 5582 | err = btrfs_init_inode_security(inode, dir); |
33268eaf JB |
5583 | if (err) { |
5584 | drop_inode = 1; | |
5585 | goto out_unlock; | |
5586 | } | |
5587 | ||
39279cc3 | 5588 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 5589 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
5590 | if (err) |
5591 | drop_inode = 1; | |
5592 | else { | |
5593 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5594 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
5595 | inode->i_fop = &btrfs_file_operations; |
5596 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 5597 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 5598 | } |
39279cc3 CM |
5599 | btrfs_update_inode_block_group(trans, inode); |
5600 | btrfs_update_inode_block_group(trans, dir); | |
5601 | if (drop_inode) | |
5602 | goto out_unlock; | |
5603 | ||
5604 | path = btrfs_alloc_path(); | |
5605 | BUG_ON(!path); | |
5606 | key.objectid = inode->i_ino; | |
5607 | key.offset = 0; | |
39279cc3 CM |
5608 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
5609 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
5610 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
5611 | datasize); | |
54aa1f4d CM |
5612 | if (err) { |
5613 | drop_inode = 1; | |
5614 | goto out_unlock; | |
5615 | } | |
5f39d397 CM |
5616 | leaf = path->nodes[0]; |
5617 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
5618 | struct btrfs_file_extent_item); | |
5619 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
5620 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 5621 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
5622 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
5623 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
5624 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
5625 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
5626 | ||
39279cc3 | 5627 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
5628 | write_extent_buffer(leaf, symname, ptr, name_len); |
5629 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 5630 | btrfs_free_path(path); |
5f39d397 | 5631 | |
39279cc3 CM |
5632 | inode->i_op = &btrfs_symlink_inode_operations; |
5633 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 5634 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 5635 | inode_set_bytes(inode, name_len); |
dbe674a9 | 5636 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
5637 | err = btrfs_update_inode(trans, root, inode); |
5638 | if (err) | |
5639 | drop_inode = 1; | |
39279cc3 CM |
5640 | |
5641 | out_unlock: | |
d3c2fdcf | 5642 | nr = trans->blocks_used; |
ab78c84d | 5643 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 5644 | out_fail: |
9ed74f2d | 5645 | btrfs_unreserve_metadata_space(root, 5); |
39279cc3 CM |
5646 | if (drop_inode) { |
5647 | inode_dec_link_count(inode); | |
5648 | iput(inode); | |
5649 | } | |
d3c2fdcf | 5650 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5651 | return err; |
5652 | } | |
16432985 | 5653 | |
546888da CM |
5654 | static int prealloc_file_range(struct btrfs_trans_handle *trans, |
5655 | struct inode *inode, u64 start, u64 end, | |
920bbbfb | 5656 | u64 alloc_hint, int mode) |
d899e052 | 5657 | { |
d899e052 YZ |
5658 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5659 | struct btrfs_key ins; | |
5660 | u64 alloc_size; | |
5661 | u64 cur_offset = start; | |
5662 | u64 num_bytes = end - start; | |
5663 | int ret = 0; | |
5664 | ||
d899e052 YZ |
5665 | while (num_bytes > 0) { |
5666 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
9ed74f2d JB |
5667 | |
5668 | ret = btrfs_reserve_metadata_space(root, 1); | |
5669 | if (ret) | |
5670 | goto out; | |
5671 | ||
d899e052 YZ |
5672 | ret = btrfs_reserve_extent(trans, root, alloc_size, |
5673 | root->sectorsize, 0, alloc_hint, | |
5674 | (u64)-1, &ins, 1); | |
5675 | if (ret) { | |
5676 | WARN_ON(1); | |
5677 | goto out; | |
5678 | } | |
5679 | ret = insert_reserved_file_extent(trans, inode, | |
5680 | cur_offset, ins.objectid, | |
5681 | ins.offset, ins.offset, | |
920bbbfb | 5682 | ins.offset, 0, 0, 0, |
d899e052 YZ |
5683 | BTRFS_FILE_EXTENT_PREALLOC); |
5684 | BUG_ON(ret); | |
a1ed835e CM |
5685 | btrfs_drop_extent_cache(inode, cur_offset, |
5686 | cur_offset + ins.offset -1, 0); | |
d899e052 YZ |
5687 | num_bytes -= ins.offset; |
5688 | cur_offset += ins.offset; | |
5689 | alloc_hint = ins.objectid + ins.offset; | |
9ed74f2d | 5690 | btrfs_unreserve_metadata_space(root, 1); |
d899e052 YZ |
5691 | } |
5692 | out: | |
5693 | if (cur_offset > start) { | |
5694 | inode->i_ctime = CURRENT_TIME; | |
6cbff00f | 5695 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 YZ |
5696 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
5697 | cur_offset > i_size_read(inode)) | |
5698 | btrfs_i_size_write(inode, cur_offset); | |
5699 | ret = btrfs_update_inode(trans, root, inode); | |
5700 | BUG_ON(ret); | |
5701 | } | |
5702 | ||
d899e052 YZ |
5703 | return ret; |
5704 | } | |
5705 | ||
5706 | static long btrfs_fallocate(struct inode *inode, int mode, | |
5707 | loff_t offset, loff_t len) | |
5708 | { | |
5709 | u64 cur_offset; | |
5710 | u64 last_byte; | |
5711 | u64 alloc_start; | |
5712 | u64 alloc_end; | |
5713 | u64 alloc_hint = 0; | |
e980b50c | 5714 | u64 locked_end; |
d899e052 YZ |
5715 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; |
5716 | struct extent_map *em; | |
546888da | 5717 | struct btrfs_trans_handle *trans; |
a970b0a1 | 5718 | struct btrfs_root *root; |
d899e052 YZ |
5719 | int ret; |
5720 | ||
5721 | alloc_start = offset & ~mask; | |
5722 | alloc_end = (offset + len + mask) & ~mask; | |
5723 | ||
546888da CM |
5724 | /* |
5725 | * wait for ordered IO before we have any locks. We'll loop again | |
5726 | * below with the locks held. | |
5727 | */ | |
5728 | btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start); | |
5729 | ||
d899e052 YZ |
5730 | mutex_lock(&inode->i_mutex); |
5731 | if (alloc_start > inode->i_size) { | |
5732 | ret = btrfs_cont_expand(inode, alloc_start); | |
5733 | if (ret) | |
5734 | goto out; | |
5735 | } | |
5736 | ||
a970b0a1 JB |
5737 | root = BTRFS_I(inode)->root; |
5738 | ||
5739 | ret = btrfs_check_data_free_space(root, inode, | |
5740 | alloc_end - alloc_start); | |
5741 | if (ret) | |
5742 | goto out; | |
5743 | ||
e980b50c | 5744 | locked_end = alloc_end - 1; |
d899e052 YZ |
5745 | while (1) { |
5746 | struct btrfs_ordered_extent *ordered; | |
546888da CM |
5747 | |
5748 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
5749 | if (!trans) { | |
5750 | ret = -EIO; | |
a970b0a1 | 5751 | goto out_free; |
546888da CM |
5752 | } |
5753 | ||
5754 | /* the extent lock is ordered inside the running | |
5755 | * transaction | |
5756 | */ | |
e980b50c CM |
5757 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
5758 | GFP_NOFS); | |
d899e052 YZ |
5759 | ordered = btrfs_lookup_first_ordered_extent(inode, |
5760 | alloc_end - 1); | |
5761 | if (ordered && | |
5762 | ordered->file_offset + ordered->len > alloc_start && | |
5763 | ordered->file_offset < alloc_end) { | |
5764 | btrfs_put_ordered_extent(ordered); | |
5765 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
e980b50c | 5766 | alloc_start, locked_end, GFP_NOFS); |
546888da CM |
5767 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
5768 | ||
5769 | /* | |
5770 | * we can't wait on the range with the transaction | |
5771 | * running or with the extent lock held | |
5772 | */ | |
d899e052 YZ |
5773 | btrfs_wait_ordered_range(inode, alloc_start, |
5774 | alloc_end - alloc_start); | |
5775 | } else { | |
5776 | if (ordered) | |
5777 | btrfs_put_ordered_extent(ordered); | |
5778 | break; | |
5779 | } | |
5780 | } | |
5781 | ||
5782 | cur_offset = alloc_start; | |
5783 | while (1) { | |
5784 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
5785 | alloc_end - cur_offset, 0); | |
5786 | BUG_ON(IS_ERR(em) || !em); | |
5787 | last_byte = min(extent_map_end(em), alloc_end); | |
5788 | last_byte = (last_byte + mask) & ~mask; | |
5789 | if (em->block_start == EXTENT_MAP_HOLE) { | |
546888da | 5790 | ret = prealloc_file_range(trans, inode, cur_offset, |
920bbbfb | 5791 | last_byte, alloc_hint, mode); |
d899e052 YZ |
5792 | if (ret < 0) { |
5793 | free_extent_map(em); | |
5794 | break; | |
5795 | } | |
5796 | } | |
5797 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
5798 | alloc_hint = em->block_start; | |
5799 | free_extent_map(em); | |
5800 | ||
5801 | cur_offset = last_byte; | |
5802 | if (cur_offset >= alloc_end) { | |
5803 | ret = 0; | |
5804 | break; | |
5805 | } | |
5806 | } | |
e980b50c | 5807 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
d899e052 | 5808 | GFP_NOFS); |
546888da CM |
5809 | |
5810 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
a970b0a1 JB |
5811 | out_free: |
5812 | btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start); | |
d899e052 YZ |
5813 | out: |
5814 | mutex_unlock(&inode->i_mutex); | |
5815 | return ret; | |
5816 | } | |
5817 | ||
e6dcd2dc CM |
5818 | static int btrfs_set_page_dirty(struct page *page) |
5819 | { | |
e6dcd2dc CM |
5820 | return __set_page_dirty_nobuffers(page); |
5821 | } | |
5822 | ||
0ee0fda0 | 5823 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 5824 | { |
6cbff00f | 5825 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 5826 | return -EACCES; |
33268eaf | 5827 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 5828 | } |
39279cc3 | 5829 | |
6e1d5dcc | 5830 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 5831 | .getattr = btrfs_getattr, |
39279cc3 CM |
5832 | .lookup = btrfs_lookup, |
5833 | .create = btrfs_create, | |
5834 | .unlink = btrfs_unlink, | |
5835 | .link = btrfs_link, | |
5836 | .mkdir = btrfs_mkdir, | |
5837 | .rmdir = btrfs_rmdir, | |
5838 | .rename = btrfs_rename, | |
5839 | .symlink = btrfs_symlink, | |
5840 | .setattr = btrfs_setattr, | |
618e21d5 | 5841 | .mknod = btrfs_mknod, |
95819c05 CH |
5842 | .setxattr = btrfs_setxattr, |
5843 | .getxattr = btrfs_getxattr, | |
5103e947 | 5844 | .listxattr = btrfs_listxattr, |
95819c05 | 5845 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5846 | .permission = btrfs_permission, |
39279cc3 | 5847 | }; |
6e1d5dcc | 5848 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 5849 | .lookup = btrfs_lookup, |
fdebe2bd | 5850 | .permission = btrfs_permission, |
39279cc3 | 5851 | }; |
76dda93c | 5852 | |
828c0950 | 5853 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
5854 | .llseek = generic_file_llseek, |
5855 | .read = generic_read_dir, | |
cbdf5a24 | 5856 | .readdir = btrfs_real_readdir, |
34287aa3 | 5857 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 5858 | #ifdef CONFIG_COMPAT |
34287aa3 | 5859 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 5860 | #endif |
6bf13c0c | 5861 | .release = btrfs_release_file, |
e02119d5 | 5862 | .fsync = btrfs_sync_file, |
39279cc3 CM |
5863 | }; |
5864 | ||
d1310b2e | 5865 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 5866 | .fill_delalloc = run_delalloc_range, |
065631f6 | 5867 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 5868 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 5869 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 5870 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 5871 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 5872 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
5873 | .set_bit_hook = btrfs_set_bit_hook, |
5874 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
5875 | .merge_extent_hook = btrfs_merge_extent_hook, |
5876 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
5877 | }; |
5878 | ||
35054394 CM |
5879 | /* |
5880 | * btrfs doesn't support the bmap operation because swapfiles | |
5881 | * use bmap to make a mapping of extents in the file. They assume | |
5882 | * these extents won't change over the life of the file and they | |
5883 | * use the bmap result to do IO directly to the drive. | |
5884 | * | |
5885 | * the btrfs bmap call would return logical addresses that aren't | |
5886 | * suitable for IO and they also will change frequently as COW | |
5887 | * operations happen. So, swapfile + btrfs == corruption. | |
5888 | * | |
5889 | * For now we're avoiding this by dropping bmap. | |
5890 | */ | |
7f09410b | 5891 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
5892 | .readpage = btrfs_readpage, |
5893 | .writepage = btrfs_writepage, | |
b293f02e | 5894 | .writepages = btrfs_writepages, |
3ab2fb5a | 5895 | .readpages = btrfs_readpages, |
39279cc3 | 5896 | .sync_page = block_sync_page, |
16432985 | 5897 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
5898 | .invalidatepage = btrfs_invalidatepage, |
5899 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 5900 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 5901 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
5902 | }; |
5903 | ||
7f09410b | 5904 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
5905 | .readpage = btrfs_readpage, |
5906 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5907 | .invalidatepage = btrfs_invalidatepage, |
5908 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5909 | }; |
5910 | ||
6e1d5dcc | 5911 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
5912 | .truncate = btrfs_truncate, |
5913 | .getattr = btrfs_getattr, | |
5914 | .setattr = btrfs_setattr, | |
95819c05 CH |
5915 | .setxattr = btrfs_setxattr, |
5916 | .getxattr = btrfs_getxattr, | |
5103e947 | 5917 | .listxattr = btrfs_listxattr, |
95819c05 | 5918 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5919 | .permission = btrfs_permission, |
d899e052 | 5920 | .fallocate = btrfs_fallocate, |
1506fcc8 | 5921 | .fiemap = btrfs_fiemap, |
39279cc3 | 5922 | }; |
6e1d5dcc | 5923 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
5924 | .getattr = btrfs_getattr, |
5925 | .setattr = btrfs_setattr, | |
fdebe2bd | 5926 | .permission = btrfs_permission, |
95819c05 CH |
5927 | .setxattr = btrfs_setxattr, |
5928 | .getxattr = btrfs_getxattr, | |
33268eaf | 5929 | .listxattr = btrfs_listxattr, |
95819c05 | 5930 | .removexattr = btrfs_removexattr, |
618e21d5 | 5931 | }; |
6e1d5dcc | 5932 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
5933 | .readlink = generic_readlink, |
5934 | .follow_link = page_follow_link_light, | |
5935 | .put_link = page_put_link, | |
fdebe2bd | 5936 | .permission = btrfs_permission, |
0279b4cd JO |
5937 | .setxattr = btrfs_setxattr, |
5938 | .getxattr = btrfs_getxattr, | |
5939 | .listxattr = btrfs_listxattr, | |
5940 | .removexattr = btrfs_removexattr, | |
39279cc3 | 5941 | }; |
76dda93c | 5942 | |
82d339d9 | 5943 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
5944 | .d_delete = btrfs_dentry_delete, |
5945 | }; |