Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
29 | #include <linux/smp_lock.h> | |
30 | #include <linux/backing-dev.h> | |
31 | #include <linux/mpage.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/writeback.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/compat.h> | |
9ebefb18 | 36 | #include <linux/bit_spinlock.h> |
92fee66d | 37 | #include <linux/version.h> |
5103e947 | 38 | #include <linux/xattr.h> |
33268eaf | 39 | #include <linux/posix_acl.h> |
d899e052 | 40 | #include <linux/falloc.h> |
39279cc3 CM |
41 | #include "ctree.h" |
42 | #include "disk-io.h" | |
43 | #include "transaction.h" | |
44 | #include "btrfs_inode.h" | |
45 | #include "ioctl.h" | |
46 | #include "print-tree.h" | |
0b86a832 | 47 | #include "volumes.h" |
e6dcd2dc | 48 | #include "ordered-data.h" |
95819c05 | 49 | #include "xattr.h" |
e02119d5 CM |
50 | #include "compat.h" |
51 | #include "tree-log.h" | |
5b84e8d6 | 52 | #include "ref-cache.h" |
c8b97818 | 53 | #include "compression.h" |
39279cc3 CM |
54 | |
55 | struct btrfs_iget_args { | |
56 | u64 ino; | |
57 | struct btrfs_root *root; | |
58 | }; | |
59 | ||
60 | static struct inode_operations btrfs_dir_inode_operations; | |
61 | static struct inode_operations btrfs_symlink_inode_operations; | |
62 | static struct inode_operations btrfs_dir_ro_inode_operations; | |
618e21d5 | 63 | static struct inode_operations btrfs_special_inode_operations; |
39279cc3 CM |
64 | static struct inode_operations btrfs_file_inode_operations; |
65 | static struct address_space_operations btrfs_aops; | |
66 | static struct address_space_operations btrfs_symlink_aops; | |
67 | static struct file_operations btrfs_dir_file_operations; | |
d1310b2e | 68 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
69 | |
70 | static struct kmem_cache *btrfs_inode_cachep; | |
71 | struct kmem_cache *btrfs_trans_handle_cachep; | |
72 | struct kmem_cache *btrfs_transaction_cachep; | |
73 | struct kmem_cache *btrfs_bit_radix_cachep; | |
74 | struct kmem_cache *btrfs_path_cachep; | |
75 | ||
76 | #define S_SHIFT 12 | |
77 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
78 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
79 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
80 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
81 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
82 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
83 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
84 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
85 | }; | |
86 | ||
7b128766 | 87 | static void btrfs_truncate(struct inode *inode); |
c8b97818 | 88 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
89 | static noinline int cow_file_range(struct inode *inode, |
90 | struct page *locked_page, | |
91 | u64 start, u64 end, int *page_started, | |
92 | unsigned long *nr_written, int unlock); | |
7b128766 | 93 | |
d352ac68 CM |
94 | /* |
95 | * a very lame attempt at stopping writes when the FS is 85% full. There | |
96 | * are countless ways this is incorrect, but it is better than nothing. | |
97 | */ | |
1832a6d5 CM |
98 | int btrfs_check_free_space(struct btrfs_root *root, u64 num_required, |
99 | int for_del) | |
100 | { | |
a2135011 CM |
101 | u64 total; |
102 | u64 used; | |
1832a6d5 | 103 | u64 thresh; |
bcbfce8a | 104 | unsigned long flags; |
1832a6d5 CM |
105 | int ret = 0; |
106 | ||
a2135011 CM |
107 | spin_lock_irqsave(&root->fs_info->delalloc_lock, flags); |
108 | total = btrfs_super_total_bytes(&root->fs_info->super_copy); | |
109 | used = btrfs_super_bytes_used(&root->fs_info->super_copy); | |
1832a6d5 | 110 | if (for_del) |
f9ef6604 | 111 | thresh = total * 90; |
1832a6d5 | 112 | else |
f9ef6604 CM |
113 | thresh = total * 85; |
114 | ||
115 | do_div(thresh, 100); | |
1832a6d5 | 116 | |
1832a6d5 CM |
117 | if (used + root->fs_info->delalloc_bytes + num_required > thresh) |
118 | ret = -ENOSPC; | |
bcbfce8a | 119 | spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags); |
1832a6d5 CM |
120 | return ret; |
121 | } | |
122 | ||
c8b97818 CM |
123 | /* |
124 | * this does all the hard work for inserting an inline extent into | |
125 | * the btree. The caller should have done a btrfs_drop_extents so that | |
126 | * no overlapping inline items exist in the btree | |
127 | */ | |
128 | static int noinline insert_inline_extent(struct btrfs_trans_handle *trans, | |
129 | struct btrfs_root *root, struct inode *inode, | |
130 | u64 start, size_t size, size_t compressed_size, | |
131 | struct page **compressed_pages) | |
132 | { | |
133 | struct btrfs_key key; | |
134 | struct btrfs_path *path; | |
135 | struct extent_buffer *leaf; | |
136 | struct page *page = NULL; | |
137 | char *kaddr; | |
138 | unsigned long ptr; | |
139 | struct btrfs_file_extent_item *ei; | |
140 | int err = 0; | |
141 | int ret; | |
142 | size_t cur_size = size; | |
143 | size_t datasize; | |
144 | unsigned long offset; | |
145 | int use_compress = 0; | |
146 | ||
147 | if (compressed_size && compressed_pages) { | |
148 | use_compress = 1; | |
149 | cur_size = compressed_size; | |
150 | } | |
151 | ||
152 | path = btrfs_alloc_path(); if (!path) | |
153 | return -ENOMEM; | |
154 | ||
155 | btrfs_set_trans_block_group(trans, inode); | |
156 | ||
157 | key.objectid = inode->i_ino; | |
158 | key.offset = start; | |
159 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
160 | inode_add_bytes(inode, size); | |
161 | datasize = btrfs_file_extent_calc_inline_size(cur_size); | |
162 | ||
163 | inode_add_bytes(inode, size); | |
164 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
165 | datasize); | |
166 | BUG_ON(ret); | |
167 | if (ret) { | |
168 | err = ret; | |
169 | printk("got bad ret %d\n", ret); | |
170 | goto fail; | |
171 | } | |
172 | leaf = path->nodes[0]; | |
173 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
174 | struct btrfs_file_extent_item); | |
175 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
176 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
177 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
178 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
179 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
180 | ptr = btrfs_file_extent_inline_start(ei); | |
181 | ||
182 | if (use_compress) { | |
183 | struct page *cpage; | |
184 | int i = 0; | |
185 | while(compressed_size > 0) { | |
186 | cpage = compressed_pages[i]; | |
5b050f04 | 187 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
188 | PAGE_CACHE_SIZE); |
189 | ||
190 | kaddr = kmap(cpage); | |
191 | write_extent_buffer(leaf, kaddr, ptr, cur_size); | |
192 | kunmap(cpage); | |
193 | ||
194 | i++; | |
195 | ptr += cur_size; | |
196 | compressed_size -= cur_size; | |
197 | } | |
198 | btrfs_set_file_extent_compression(leaf, ei, | |
199 | BTRFS_COMPRESS_ZLIB); | |
200 | } else { | |
201 | page = find_get_page(inode->i_mapping, | |
202 | start >> PAGE_CACHE_SHIFT); | |
203 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
204 | kaddr = kmap_atomic(page, KM_USER0); | |
205 | offset = start & (PAGE_CACHE_SIZE - 1); | |
206 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
207 | kunmap_atomic(kaddr, KM_USER0); | |
208 | page_cache_release(page); | |
209 | } | |
210 | btrfs_mark_buffer_dirty(leaf); | |
211 | btrfs_free_path(path); | |
212 | ||
213 | BTRFS_I(inode)->disk_i_size = inode->i_size; | |
214 | btrfs_update_inode(trans, root, inode); | |
215 | return 0; | |
216 | fail: | |
217 | btrfs_free_path(path); | |
218 | return err; | |
219 | } | |
220 | ||
221 | ||
222 | /* | |
223 | * conditionally insert an inline extent into the file. This | |
224 | * does the checks required to make sure the data is small enough | |
225 | * to fit as an inline extent. | |
226 | */ | |
227 | static int cow_file_range_inline(struct btrfs_trans_handle *trans, | |
228 | struct btrfs_root *root, | |
229 | struct inode *inode, u64 start, u64 end, | |
230 | size_t compressed_size, | |
231 | struct page **compressed_pages) | |
232 | { | |
233 | u64 isize = i_size_read(inode); | |
234 | u64 actual_end = min(end + 1, isize); | |
235 | u64 inline_len = actual_end - start; | |
236 | u64 aligned_end = (end + root->sectorsize - 1) & | |
237 | ~((u64)root->sectorsize - 1); | |
238 | u64 hint_byte; | |
239 | u64 data_len = inline_len; | |
240 | int ret; | |
241 | ||
242 | if (compressed_size) | |
243 | data_len = compressed_size; | |
244 | ||
245 | if (start > 0 || | |
70b99e69 | 246 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
247 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
248 | (!compressed_size && | |
249 | (actual_end & (root->sectorsize - 1)) == 0) || | |
250 | end + 1 < isize || | |
251 | data_len > root->fs_info->max_inline) { | |
252 | return 1; | |
253 | } | |
254 | ||
c8b97818 | 255 | ret = btrfs_drop_extents(trans, root, inode, start, |
70b99e69 | 256 | aligned_end, start, &hint_byte); |
c8b97818 CM |
257 | BUG_ON(ret); |
258 | ||
259 | if (isize > actual_end) | |
260 | inline_len = min_t(u64, isize, actual_end); | |
261 | ret = insert_inline_extent(trans, root, inode, start, | |
262 | inline_len, compressed_size, | |
263 | compressed_pages); | |
264 | BUG_ON(ret); | |
265 | btrfs_drop_extent_cache(inode, start, aligned_end, 0); | |
c8b97818 CM |
266 | return 0; |
267 | } | |
268 | ||
771ed689 CM |
269 | struct async_extent { |
270 | u64 start; | |
271 | u64 ram_size; | |
272 | u64 compressed_size; | |
273 | struct page **pages; | |
274 | unsigned long nr_pages; | |
275 | struct list_head list; | |
276 | }; | |
277 | ||
278 | struct async_cow { | |
279 | struct inode *inode; | |
280 | struct btrfs_root *root; | |
281 | struct page *locked_page; | |
282 | u64 start; | |
283 | u64 end; | |
284 | struct list_head extents; | |
285 | struct btrfs_work work; | |
286 | }; | |
287 | ||
288 | static noinline int add_async_extent(struct async_cow *cow, | |
289 | u64 start, u64 ram_size, | |
290 | u64 compressed_size, | |
291 | struct page **pages, | |
292 | unsigned long nr_pages) | |
293 | { | |
294 | struct async_extent *async_extent; | |
295 | ||
296 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
297 | async_extent->start = start; | |
298 | async_extent->ram_size = ram_size; | |
299 | async_extent->compressed_size = compressed_size; | |
300 | async_extent->pages = pages; | |
301 | async_extent->nr_pages = nr_pages; | |
302 | list_add_tail(&async_extent->list, &cow->extents); | |
303 | return 0; | |
304 | } | |
305 | ||
d352ac68 | 306 | /* |
771ed689 CM |
307 | * we create compressed extents in two phases. The first |
308 | * phase compresses a range of pages that have already been | |
309 | * locked (both pages and state bits are locked). | |
c8b97818 | 310 | * |
771ed689 CM |
311 | * This is done inside an ordered work queue, and the compression |
312 | * is spread across many cpus. The actual IO submission is step | |
313 | * two, and the ordered work queue takes care of making sure that | |
314 | * happens in the same order things were put onto the queue by | |
315 | * writepages and friends. | |
c8b97818 | 316 | * |
771ed689 CM |
317 | * If this code finds it can't get good compression, it puts an |
318 | * entry onto the work queue to write the uncompressed bytes. This | |
319 | * makes sure that both compressed inodes and uncompressed inodes | |
320 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 321 | */ |
771ed689 CM |
322 | static noinline int compress_file_range(struct inode *inode, |
323 | struct page *locked_page, | |
324 | u64 start, u64 end, | |
325 | struct async_cow *async_cow, | |
326 | int *num_added) | |
b888db2b CM |
327 | { |
328 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
329 | struct btrfs_trans_handle *trans; | |
db94535d | 330 | u64 num_bytes; |
c8b97818 CM |
331 | u64 orig_start; |
332 | u64 disk_num_bytes; | |
db94535d | 333 | u64 blocksize = root->sectorsize; |
c8b97818 | 334 | u64 actual_end; |
e6dcd2dc | 335 | int ret = 0; |
c8b97818 CM |
336 | struct page **pages = NULL; |
337 | unsigned long nr_pages; | |
338 | unsigned long nr_pages_ret = 0; | |
339 | unsigned long total_compressed = 0; | |
340 | unsigned long total_in = 0; | |
341 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 342 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
343 | int i; |
344 | int will_compress; | |
b888db2b | 345 | |
c8b97818 CM |
346 | orig_start = start; |
347 | ||
c8b97818 CM |
348 | again: |
349 | will_compress = 0; | |
350 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
351 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 352 | |
c8b97818 CM |
353 | actual_end = min_t(u64, i_size_read(inode), end + 1); |
354 | total_compressed = actual_end - start; | |
355 | ||
356 | /* we want to make sure that amount of ram required to uncompress | |
357 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
358 | * of a compressed extent to 128k. This is a crucial number |
359 | * because it also controls how easily we can spread reads across | |
360 | * cpus for decompression. | |
361 | * | |
362 | * We also want to make sure the amount of IO required to do | |
363 | * a random read is reasonably small, so we limit the size of | |
364 | * a compressed extent to 128k. | |
c8b97818 CM |
365 | */ |
366 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 367 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 368 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
369 | disk_num_bytes = num_bytes; |
370 | total_in = 0; | |
371 | ret = 0; | |
db94535d | 372 | |
771ed689 CM |
373 | /* |
374 | * we do compression for mount -o compress and when the | |
375 | * inode has not been flagged as nocompress. This flag can | |
376 | * change at any time if we discover bad compression ratios. | |
c8b97818 CM |
377 | */ |
378 | if (!btrfs_test_flag(inode, NOCOMPRESS) && | |
379 | btrfs_test_opt(root, COMPRESS)) { | |
380 | WARN_ON(pages); | |
cfbc246e | 381 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
c8b97818 | 382 | |
c8b97818 CM |
383 | ret = btrfs_zlib_compress_pages(inode->i_mapping, start, |
384 | total_compressed, pages, | |
385 | nr_pages, &nr_pages_ret, | |
386 | &total_in, | |
387 | &total_compressed, | |
388 | max_compressed); | |
389 | ||
390 | if (!ret) { | |
391 | unsigned long offset = total_compressed & | |
392 | (PAGE_CACHE_SIZE - 1); | |
393 | struct page *page = pages[nr_pages_ret - 1]; | |
394 | char *kaddr; | |
395 | ||
396 | /* zero the tail end of the last page, we might be | |
397 | * sending it down to disk | |
398 | */ | |
399 | if (offset) { | |
400 | kaddr = kmap_atomic(page, KM_USER0); | |
401 | memset(kaddr + offset, 0, | |
402 | PAGE_CACHE_SIZE - offset); | |
403 | kunmap_atomic(kaddr, KM_USER0); | |
404 | } | |
405 | will_compress = 1; | |
406 | } | |
407 | } | |
408 | if (start == 0) { | |
771ed689 CM |
409 | trans = btrfs_join_transaction(root, 1); |
410 | BUG_ON(!trans); | |
411 | btrfs_set_trans_block_group(trans, inode); | |
412 | ||
c8b97818 | 413 | /* lets try to make an inline extent */ |
771ed689 | 414 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 415 | /* we didn't compress the entire range, try |
771ed689 | 416 | * to make an uncompressed inline extent. |
c8b97818 CM |
417 | */ |
418 | ret = cow_file_range_inline(trans, root, inode, | |
419 | start, end, 0, NULL); | |
420 | } else { | |
771ed689 | 421 | /* try making a compressed inline extent */ |
c8b97818 CM |
422 | ret = cow_file_range_inline(trans, root, inode, |
423 | start, end, | |
424 | total_compressed, pages); | |
425 | } | |
771ed689 | 426 | btrfs_end_transaction(trans, root); |
c8b97818 | 427 | if (ret == 0) { |
771ed689 CM |
428 | /* |
429 | * inline extent creation worked, we don't need | |
430 | * to create any more async work items. Unlock | |
431 | * and free up our temp pages. | |
432 | */ | |
c8b97818 CM |
433 | extent_clear_unlock_delalloc(inode, |
434 | &BTRFS_I(inode)->io_tree, | |
771ed689 CM |
435 | start, end, NULL, 1, 0, |
436 | 0, 1, 1, 1); | |
c8b97818 CM |
437 | ret = 0; |
438 | goto free_pages_out; | |
439 | } | |
440 | } | |
441 | ||
442 | if (will_compress) { | |
443 | /* | |
444 | * we aren't doing an inline extent round the compressed size | |
445 | * up to a block size boundary so the allocator does sane | |
446 | * things | |
447 | */ | |
448 | total_compressed = (total_compressed + blocksize - 1) & | |
449 | ~(blocksize - 1); | |
450 | ||
451 | /* | |
452 | * one last check to make sure the compression is really a | |
453 | * win, compare the page count read with the blocks on disk | |
454 | */ | |
455 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
456 | ~(PAGE_CACHE_SIZE - 1); | |
457 | if (total_compressed >= total_in) { | |
458 | will_compress = 0; | |
459 | } else { | |
460 | disk_num_bytes = total_compressed; | |
461 | num_bytes = total_in; | |
462 | } | |
463 | } | |
464 | if (!will_compress && pages) { | |
465 | /* | |
466 | * the compression code ran but failed to make things smaller, | |
467 | * free any pages it allocated and our page pointer array | |
468 | */ | |
469 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 470 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
471 | page_cache_release(pages[i]); |
472 | } | |
473 | kfree(pages); | |
474 | pages = NULL; | |
475 | total_compressed = 0; | |
476 | nr_pages_ret = 0; | |
477 | ||
478 | /* flag the file so we don't compress in the future */ | |
479 | btrfs_set_flag(inode, NOCOMPRESS); | |
480 | } | |
771ed689 CM |
481 | if (will_compress) { |
482 | *num_added += 1; | |
c8b97818 | 483 | |
771ed689 CM |
484 | /* the async work queues will take care of doing actual |
485 | * allocation on disk for these compressed pages, | |
486 | * and will submit them to the elevator. | |
487 | */ | |
488 | add_async_extent(async_cow, start, num_bytes, | |
489 | total_compressed, pages, nr_pages_ret); | |
179e29e4 | 490 | |
771ed689 CM |
491 | if (start + num_bytes < end) { |
492 | start += num_bytes; | |
493 | pages = NULL; | |
494 | cond_resched(); | |
495 | goto again; | |
496 | } | |
497 | } else { | |
498 | /* | |
499 | * No compression, but we still need to write the pages in | |
500 | * the file we've been given so far. redirty the locked | |
501 | * page if it corresponds to our extent and set things up | |
502 | * for the async work queue to run cow_file_range to do | |
503 | * the normal delalloc dance | |
504 | */ | |
505 | if (page_offset(locked_page) >= start && | |
506 | page_offset(locked_page) <= end) { | |
507 | __set_page_dirty_nobuffers(locked_page); | |
508 | /* unlocked later on in the async handlers */ | |
509 | } | |
510 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0); | |
511 | *num_added += 1; | |
512 | } | |
3b951516 | 513 | |
771ed689 CM |
514 | out: |
515 | return 0; | |
516 | ||
517 | free_pages_out: | |
518 | for (i = 0; i < nr_pages_ret; i++) { | |
519 | WARN_ON(pages[i]->mapping); | |
520 | page_cache_release(pages[i]); | |
521 | } | |
522 | if (pages) | |
523 | kfree(pages); | |
524 | ||
525 | goto out; | |
526 | } | |
527 | ||
528 | /* | |
529 | * phase two of compressed writeback. This is the ordered portion | |
530 | * of the code, which only gets called in the order the work was | |
531 | * queued. We walk all the async extents created by compress_file_range | |
532 | * and send them down to the disk. | |
533 | */ | |
534 | static noinline int submit_compressed_extents(struct inode *inode, | |
535 | struct async_cow *async_cow) | |
536 | { | |
537 | struct async_extent *async_extent; | |
538 | u64 alloc_hint = 0; | |
539 | struct btrfs_trans_handle *trans; | |
540 | struct btrfs_key ins; | |
541 | struct extent_map *em; | |
542 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
543 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
544 | struct extent_io_tree *io_tree; | |
545 | int ret; | |
546 | ||
547 | if (list_empty(&async_cow->extents)) | |
548 | return 0; | |
549 | ||
550 | trans = btrfs_join_transaction(root, 1); | |
551 | ||
552 | while(!list_empty(&async_cow->extents)) { | |
553 | async_extent = list_entry(async_cow->extents.next, | |
554 | struct async_extent, list); | |
555 | list_del(&async_extent->list); | |
c8b97818 | 556 | |
771ed689 CM |
557 | io_tree = &BTRFS_I(inode)->io_tree; |
558 | ||
559 | /* did the compression code fall back to uncompressed IO? */ | |
560 | if (!async_extent->pages) { | |
561 | int page_started = 0; | |
562 | unsigned long nr_written = 0; | |
563 | ||
564 | lock_extent(io_tree, async_extent->start, | |
565 | async_extent->start + async_extent->ram_size - 1, | |
566 | GFP_NOFS); | |
567 | ||
568 | /* allocate blocks */ | |
569 | cow_file_range(inode, async_cow->locked_page, | |
570 | async_extent->start, | |
571 | async_extent->start + | |
572 | async_extent->ram_size - 1, | |
573 | &page_started, &nr_written, 0); | |
574 | ||
575 | /* | |
576 | * if page_started, cow_file_range inserted an | |
577 | * inline extent and took care of all the unlocking | |
578 | * and IO for us. Otherwise, we need to submit | |
579 | * all those pages down to the drive. | |
580 | */ | |
581 | if (!page_started) | |
582 | extent_write_locked_range(io_tree, | |
583 | inode, async_extent->start, | |
584 | async_extent->start + | |
585 | async_extent->ram_size - 1, | |
586 | btrfs_get_extent, | |
587 | WB_SYNC_ALL); | |
588 | kfree(async_extent); | |
589 | cond_resched(); | |
590 | continue; | |
591 | } | |
592 | ||
593 | lock_extent(io_tree, async_extent->start, | |
594 | async_extent->start + async_extent->ram_size - 1, | |
595 | GFP_NOFS); | |
c8b97818 | 596 | /* |
771ed689 CM |
597 | * here we're doing allocation and writeback of the |
598 | * compressed pages | |
c8b97818 | 599 | */ |
771ed689 CM |
600 | btrfs_drop_extent_cache(inode, async_extent->start, |
601 | async_extent->start + | |
602 | async_extent->ram_size - 1, 0); | |
603 | ||
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); | |
609 | BUG_ON(ret); | |
610 | em = alloc_extent_map(GFP_NOFS); | |
611 | em->start = async_extent->start; | |
612 | em->len = async_extent->ram_size; | |
445a6944 | 613 | em->orig_start = em->start; |
c8b97818 | 614 | |
771ed689 CM |
615 | em->block_start = ins.objectid; |
616 | em->block_len = ins.offset; | |
617 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
618 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
619 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
620 | ||
621 | while(1) { | |
622 | spin_lock(&em_tree->lock); | |
623 | ret = add_extent_mapping(em_tree, em); | |
624 | spin_unlock(&em_tree->lock); | |
625 | if (ret != -EEXIST) { | |
626 | free_extent_map(em); | |
627 | break; | |
628 | } | |
629 | btrfs_drop_extent_cache(inode, async_extent->start, | |
630 | async_extent->start + | |
631 | async_extent->ram_size - 1, 0); | |
632 | } | |
633 | ||
634 | ret = btrfs_add_ordered_extent(inode, async_extent->start, | |
635 | ins.objectid, | |
636 | async_extent->ram_size, | |
637 | ins.offset, | |
638 | BTRFS_ORDERED_COMPRESSED); | |
639 | BUG_ON(ret); | |
640 | ||
641 | btrfs_end_transaction(trans, root); | |
642 | ||
643 | /* | |
644 | * clear dirty, set writeback and unlock the pages. | |
645 | */ | |
646 | extent_clear_unlock_delalloc(inode, | |
647 | &BTRFS_I(inode)->io_tree, | |
648 | async_extent->start, | |
649 | async_extent->start + | |
650 | async_extent->ram_size - 1, | |
651 | NULL, 1, 1, 0, 1, 1, 0); | |
652 | ||
653 | ret = btrfs_submit_compressed_write(inode, | |
654 | async_extent->start, | |
655 | async_extent->ram_size, | |
656 | ins.objectid, | |
657 | ins.offset, async_extent->pages, | |
658 | async_extent->nr_pages); | |
659 | ||
660 | BUG_ON(ret); | |
661 | trans = btrfs_join_transaction(root, 1); | |
662 | alloc_hint = ins.objectid + ins.offset; | |
663 | kfree(async_extent); | |
664 | cond_resched(); | |
665 | } | |
666 | ||
667 | btrfs_end_transaction(trans, root); | |
668 | return 0; | |
669 | } | |
670 | ||
671 | /* | |
672 | * when extent_io.c finds a delayed allocation range in the file, | |
673 | * the call backs end up in this code. The basic idea is to | |
674 | * allocate extents on disk for the range, and create ordered data structs | |
675 | * in ram to track those extents. | |
676 | * | |
677 | * locked_page is the page that writepage had locked already. We use | |
678 | * it to make sure we don't do extra locks or unlocks. | |
679 | * | |
680 | * *page_started is set to one if we unlock locked_page and do everything | |
681 | * required to start IO on it. It may be clean and already done with | |
682 | * IO when we return. | |
683 | */ | |
684 | static noinline int cow_file_range(struct inode *inode, | |
685 | struct page *locked_page, | |
686 | u64 start, u64 end, int *page_started, | |
687 | unsigned long *nr_written, | |
688 | int unlock) | |
689 | { | |
690 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
691 | struct btrfs_trans_handle *trans; | |
692 | u64 alloc_hint = 0; | |
693 | u64 num_bytes; | |
694 | unsigned long ram_size; | |
695 | u64 disk_num_bytes; | |
696 | u64 cur_alloc_size; | |
697 | u64 blocksize = root->sectorsize; | |
698 | u64 actual_end; | |
699 | struct btrfs_key ins; | |
700 | struct extent_map *em; | |
701 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
702 | int ret = 0; | |
703 | ||
704 | trans = btrfs_join_transaction(root, 1); | |
705 | BUG_ON(!trans); | |
706 | btrfs_set_trans_block_group(trans, inode); | |
707 | ||
708 | actual_end = min_t(u64, i_size_read(inode), end + 1); | |
709 | ||
710 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); | |
711 | num_bytes = max(blocksize, num_bytes); | |
712 | disk_num_bytes = num_bytes; | |
713 | ret = 0; | |
714 | ||
715 | if (start == 0) { | |
716 | /* lets try to make an inline extent */ | |
717 | ret = cow_file_range_inline(trans, root, inode, | |
718 | start, end, 0, NULL); | |
719 | if (ret == 0) { | |
720 | extent_clear_unlock_delalloc(inode, | |
721 | &BTRFS_I(inode)->io_tree, | |
722 | start, end, NULL, 1, 1, | |
723 | 1, 1, 1, 1); | |
724 | *nr_written = *nr_written + | |
725 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
726 | *page_started = 1; | |
727 | ret = 0; | |
728 | goto out; | |
729 | } | |
730 | } | |
731 | ||
732 | BUG_ON(disk_num_bytes > | |
733 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
734 | ||
735 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); | |
736 | ||
737 | while(disk_num_bytes > 0) { | |
c8b97818 | 738 | cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent); |
e6dcd2dc | 739 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 740 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 741 | (u64)-1, &ins, 1); |
c59f8951 | 742 | if (ret) { |
771ed689 | 743 | BUG(); |
c59f8951 | 744 | } |
e6dcd2dc CM |
745 | em = alloc_extent_map(GFP_NOFS); |
746 | em->start = start; | |
445a6944 | 747 | em->orig_start = em->start; |
c8b97818 | 748 | |
771ed689 CM |
749 | ram_size = ins.offset; |
750 | em->len = ins.offset; | |
c8b97818 | 751 | |
e6dcd2dc | 752 | em->block_start = ins.objectid; |
c8b97818 | 753 | em->block_len = ins.offset; |
e6dcd2dc | 754 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 755 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 756 | |
e6dcd2dc CM |
757 | while(1) { |
758 | spin_lock(&em_tree->lock); | |
759 | ret = add_extent_mapping(em_tree, em); | |
760 | spin_unlock(&em_tree->lock); | |
761 | if (ret != -EEXIST) { | |
762 | free_extent_map(em); | |
763 | break; | |
764 | } | |
765 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 766 | start + ram_size - 1, 0); |
e6dcd2dc CM |
767 | } |
768 | ||
98d20f67 | 769 | cur_alloc_size = ins.offset; |
e6dcd2dc | 770 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 771 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 772 | BUG_ON(ret); |
c8b97818 CM |
773 | |
774 | if (disk_num_bytes < cur_alloc_size) { | |
775 | printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes, | |
3b951516 CM |
776 | cur_alloc_size); |
777 | break; | |
778 | } | |
c8b97818 CM |
779 | /* we're not doing compressed IO, don't unlock the first |
780 | * page (which the caller expects to stay locked), don't | |
781 | * clear any dirty bits and don't set any writeback bits | |
782 | */ | |
783 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
784 | start, start + ram_size - 1, | |
771ed689 CM |
785 | locked_page, unlock, 1, |
786 | 1, 0, 0, 0); | |
c8b97818 | 787 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
788 | num_bytes -= cur_alloc_size; |
789 | alloc_hint = ins.objectid + ins.offset; | |
790 | start += cur_alloc_size; | |
b888db2b | 791 | } |
b888db2b | 792 | out: |
771ed689 | 793 | ret = 0; |
b888db2b | 794 | btrfs_end_transaction(trans, root); |
c8b97818 | 795 | |
be20aa9d | 796 | return ret; |
771ed689 | 797 | } |
c8b97818 | 798 | |
771ed689 CM |
799 | /* |
800 | * work queue call back to started compression on a file and pages | |
801 | */ | |
802 | static noinline void async_cow_start(struct btrfs_work *work) | |
803 | { | |
804 | struct async_cow *async_cow; | |
805 | int num_added = 0; | |
806 | async_cow = container_of(work, struct async_cow, work); | |
807 | ||
808 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
809 | async_cow->start, async_cow->end, async_cow, | |
810 | &num_added); | |
811 | if (num_added == 0) | |
812 | async_cow->inode = NULL; | |
813 | } | |
814 | ||
815 | /* | |
816 | * work queue call back to submit previously compressed pages | |
817 | */ | |
818 | static noinline void async_cow_submit(struct btrfs_work *work) | |
819 | { | |
820 | struct async_cow *async_cow; | |
821 | struct btrfs_root *root; | |
822 | unsigned long nr_pages; | |
823 | ||
824 | async_cow = container_of(work, struct async_cow, work); | |
825 | ||
826 | root = async_cow->root; | |
827 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
828 | PAGE_CACHE_SHIFT; | |
829 | ||
830 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
831 | ||
832 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
833 | 5 * 1042 * 1024 && | |
834 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
835 | wake_up(&root->fs_info->async_submit_wait); | |
836 | ||
837 | if (async_cow->inode) { | |
838 | submit_compressed_extents(async_cow->inode, async_cow); | |
70b99e69 | 839 | } |
771ed689 | 840 | } |
c8b97818 | 841 | |
771ed689 CM |
842 | static noinline void async_cow_free(struct btrfs_work *work) |
843 | { | |
844 | struct async_cow *async_cow; | |
845 | async_cow = container_of(work, struct async_cow, work); | |
846 | kfree(async_cow); | |
847 | } | |
848 | ||
849 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
850 | u64 start, u64 end, int *page_started, | |
851 | unsigned long *nr_written) | |
852 | { | |
853 | struct async_cow *async_cow; | |
854 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
855 | unsigned long nr_pages; | |
856 | u64 cur_end; | |
857 | int limit = 10 * 1024 * 1042; | |
858 | ||
859 | if (!btrfs_test_opt(root, COMPRESS)) { | |
860 | return cow_file_range(inode, locked_page, start, end, | |
861 | page_started, nr_written, 1); | |
862 | } | |
863 | ||
864 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | | |
865 | EXTENT_DELALLOC, 1, 0, GFP_NOFS); | |
866 | while(start < end) { | |
867 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); | |
868 | async_cow->inode = inode; | |
869 | async_cow->root = root; | |
870 | async_cow->locked_page = locked_page; | |
871 | async_cow->start = start; | |
872 | ||
873 | if (btrfs_test_flag(inode, NOCOMPRESS)) | |
874 | cur_end = end; | |
875 | else | |
876 | cur_end = min(end, start + 512 * 1024 - 1); | |
877 | ||
878 | async_cow->end = cur_end; | |
879 | INIT_LIST_HEAD(&async_cow->extents); | |
880 | ||
881 | async_cow->work.func = async_cow_start; | |
882 | async_cow->work.ordered_func = async_cow_submit; | |
883 | async_cow->work.ordered_free = async_cow_free; | |
884 | async_cow->work.flags = 0; | |
885 | ||
886 | while(atomic_read(&root->fs_info->async_submit_draining) && | |
887 | atomic_read(&root->fs_info->async_delalloc_pages)) { | |
888 | wait_event(root->fs_info->async_submit_wait, | |
889 | (atomic_read(&root->fs_info->async_delalloc_pages) | |
890 | == 0)); | |
891 | } | |
892 | ||
893 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> | |
894 | PAGE_CACHE_SHIFT; | |
895 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
896 | ||
897 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
898 | &async_cow->work); | |
899 | ||
900 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
901 | wait_event(root->fs_info->async_submit_wait, | |
902 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
903 | limit)); | |
904 | } | |
905 | ||
906 | while(atomic_read(&root->fs_info->async_submit_draining) && | |
907 | atomic_read(&root->fs_info->async_delalloc_pages)) { | |
908 | wait_event(root->fs_info->async_submit_wait, | |
909 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
910 | 0)); | |
911 | } | |
912 | ||
913 | *nr_written += nr_pages; | |
914 | start = cur_end + 1; | |
915 | } | |
916 | *page_started = 1; | |
917 | return 0; | |
be20aa9d CM |
918 | } |
919 | ||
d352ac68 CM |
920 | /* |
921 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
922 | * of the extents that exist in the file, and COWs the file as required. | |
923 | * | |
924 | * If no cow copies or snapshots exist, we write directly to the existing | |
925 | * blocks on disk | |
926 | */ | |
c8b97818 | 927 | static int run_delalloc_nocow(struct inode *inode, struct page *locked_page, |
771ed689 CM |
928 | u64 start, u64 end, int *page_started, int force, |
929 | unsigned long *nr_written) | |
be20aa9d | 930 | { |
be20aa9d | 931 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 932 | struct btrfs_trans_handle *trans; |
be20aa9d | 933 | struct extent_buffer *leaf; |
be20aa9d | 934 | struct btrfs_path *path; |
80ff3856 | 935 | struct btrfs_file_extent_item *fi; |
be20aa9d | 936 | struct btrfs_key found_key; |
80ff3856 YZ |
937 | u64 cow_start; |
938 | u64 cur_offset; | |
939 | u64 extent_end; | |
940 | u64 disk_bytenr; | |
941 | u64 num_bytes; | |
942 | int extent_type; | |
943 | int ret; | |
d899e052 | 944 | int type; |
80ff3856 YZ |
945 | int nocow; |
946 | int check_prev = 1; | |
be20aa9d CM |
947 | |
948 | path = btrfs_alloc_path(); | |
949 | BUG_ON(!path); | |
7ea394f1 YZ |
950 | trans = btrfs_join_transaction(root, 1); |
951 | BUG_ON(!trans); | |
be20aa9d | 952 | |
80ff3856 YZ |
953 | cow_start = (u64)-1; |
954 | cur_offset = start; | |
955 | while (1) { | |
956 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
957 | cur_offset, 0); | |
958 | BUG_ON(ret < 0); | |
959 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
960 | leaf = path->nodes[0]; | |
961 | btrfs_item_key_to_cpu(leaf, &found_key, | |
962 | path->slots[0] - 1); | |
963 | if (found_key.objectid == inode->i_ino && | |
964 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
965 | path->slots[0]--; | |
966 | } | |
967 | check_prev = 0; | |
968 | next_slot: | |
969 | leaf = path->nodes[0]; | |
970 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
971 | ret = btrfs_next_leaf(root, path); | |
972 | if (ret < 0) | |
973 | BUG_ON(1); | |
974 | if (ret > 0) | |
975 | break; | |
976 | leaf = path->nodes[0]; | |
977 | } | |
be20aa9d | 978 | |
80ff3856 YZ |
979 | nocow = 0; |
980 | disk_bytenr = 0; | |
981 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
982 | ||
983 | if (found_key.objectid > inode->i_ino || | |
984 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
985 | found_key.offset > end) | |
986 | break; | |
987 | ||
988 | if (found_key.offset > cur_offset) { | |
989 | extent_end = found_key.offset; | |
990 | goto out_check; | |
991 | } | |
992 | ||
993 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
994 | struct btrfs_file_extent_item); | |
995 | extent_type = btrfs_file_extent_type(leaf, fi); | |
996 | ||
d899e052 YZ |
997 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
998 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 YZ |
999 | struct btrfs_block_group_cache *block_group; |
1000 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
1001 | extent_end = found_key.offset + | |
1002 | btrfs_file_extent_num_bytes(leaf, fi); | |
1003 | if (extent_end <= start) { | |
1004 | path->slots[0]++; | |
1005 | goto next_slot; | |
1006 | } | |
1007 | if (btrfs_file_extent_compression(leaf, fi) || | |
1008 | btrfs_file_extent_encryption(leaf, fi) || | |
1009 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1010 | goto out_check; | |
1011 | if (disk_bytenr == 0) | |
1012 | goto out_check; | |
d899e052 YZ |
1013 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1014 | goto out_check; | |
80ff3856 YZ |
1015 | if (btrfs_cross_ref_exist(trans, root, disk_bytenr)) |
1016 | goto out_check; | |
1017 | block_group = btrfs_lookup_block_group(root->fs_info, | |
1018 | disk_bytenr); | |
1019 | if (!block_group || block_group->ro) | |
1020 | goto out_check; | |
1021 | disk_bytenr += btrfs_file_extent_offset(leaf, fi); | |
1022 | nocow = 1; | |
1023 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1024 | extent_end = found_key.offset + | |
1025 | btrfs_file_extent_inline_len(leaf, fi); | |
1026 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1027 | } else { | |
1028 | BUG_ON(1); | |
1029 | } | |
1030 | out_check: | |
1031 | if (extent_end <= start) { | |
1032 | path->slots[0]++; | |
1033 | goto next_slot; | |
1034 | } | |
1035 | if (!nocow) { | |
1036 | if (cow_start == (u64)-1) | |
1037 | cow_start = cur_offset; | |
1038 | cur_offset = extent_end; | |
1039 | if (cur_offset > end) | |
1040 | break; | |
1041 | path->slots[0]++; | |
1042 | goto next_slot; | |
7ea394f1 YZ |
1043 | } |
1044 | ||
1045 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1046 | if (cow_start != (u64)-1) { |
1047 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1048 | found_key.offset - 1, page_started, |
1049 | nr_written, 1); | |
80ff3856 YZ |
1050 | BUG_ON(ret); |
1051 | cow_start = (u64)-1; | |
7ea394f1 | 1052 | } |
80ff3856 YZ |
1053 | |
1054 | disk_bytenr += cur_offset - found_key.offset; | |
1055 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
d899e052 YZ |
1056 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1057 | struct extent_map *em; | |
1058 | struct extent_map_tree *em_tree; | |
1059 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1060 | em = alloc_extent_map(GFP_NOFS); | |
1061 | em->start = cur_offset; | |
445a6944 | 1062 | em->orig_start = em->start; |
d899e052 YZ |
1063 | em->len = num_bytes; |
1064 | em->block_len = num_bytes; | |
1065 | em->block_start = disk_bytenr; | |
1066 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1067 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1068 | while (1) { | |
1069 | spin_lock(&em_tree->lock); | |
1070 | ret = add_extent_mapping(em_tree, em); | |
1071 | spin_unlock(&em_tree->lock); | |
1072 | if (ret != -EEXIST) { | |
1073 | free_extent_map(em); | |
1074 | break; | |
1075 | } | |
1076 | btrfs_drop_extent_cache(inode, em->start, | |
1077 | em->start + em->len - 1, 0); | |
1078 | } | |
1079 | type = BTRFS_ORDERED_PREALLOC; | |
1080 | } else { | |
1081 | type = BTRFS_ORDERED_NOCOW; | |
1082 | } | |
80ff3856 YZ |
1083 | |
1084 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1085 | num_bytes, num_bytes, type); |
1086 | BUG_ON(ret); | |
771ed689 | 1087 | |
d899e052 YZ |
1088 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
1089 | cur_offset, cur_offset + num_bytes - 1, | |
771ed689 | 1090 | locked_page, 1, 1, 1, 0, 0, 0); |
80ff3856 YZ |
1091 | cur_offset = extent_end; |
1092 | if (cur_offset > end) | |
1093 | break; | |
be20aa9d | 1094 | } |
80ff3856 YZ |
1095 | btrfs_release_path(root, path); |
1096 | ||
1097 | if (cur_offset <= end && cow_start == (u64)-1) | |
1098 | cow_start = cur_offset; | |
1099 | if (cow_start != (u64)-1) { | |
1100 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1101 | page_started, nr_written, 1); |
80ff3856 YZ |
1102 | BUG_ON(ret); |
1103 | } | |
1104 | ||
1105 | ret = btrfs_end_transaction(trans, root); | |
1106 | BUG_ON(ret); | |
7ea394f1 | 1107 | btrfs_free_path(path); |
80ff3856 | 1108 | return 0; |
be20aa9d CM |
1109 | } |
1110 | ||
d352ac68 CM |
1111 | /* |
1112 | * extent_io.c call back to do delayed allocation processing | |
1113 | */ | |
c8b97818 | 1114 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1115 | u64 start, u64 end, int *page_started, |
1116 | unsigned long *nr_written) | |
be20aa9d CM |
1117 | { |
1118 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1119 | int ret; | |
a2135011 | 1120 | |
b98b6767 Y |
1121 | if (btrfs_test_opt(root, NODATACOW) || |
1122 | btrfs_test_flag(inode, NODATACOW)) | |
c8b97818 | 1123 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
771ed689 | 1124 | page_started, 0, nr_written); |
d899e052 YZ |
1125 | else if (btrfs_test_flag(inode, PREALLOC)) |
1126 | ret = run_delalloc_nocow(inode, locked_page, start, end, | |
771ed689 | 1127 | page_started, 1, nr_written); |
be20aa9d | 1128 | else |
771ed689 CM |
1129 | ret = cow_file_range_async(inode, locked_page, start, end, |
1130 | page_started, nr_written); | |
1832a6d5 | 1131 | |
b888db2b CM |
1132 | return ret; |
1133 | } | |
1134 | ||
d352ac68 CM |
1135 | /* |
1136 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1137 | * bytes in this file, and to maintain the list of inodes that | |
1138 | * have pending delalloc work to be done. | |
1139 | */ | |
291d673e | 1140 | int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1141 | unsigned long old, unsigned long bits) |
291d673e | 1142 | { |
bcbfce8a | 1143 | unsigned long flags; |
b0c68f8b | 1144 | if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1145 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a | 1146 | spin_lock_irqsave(&root->fs_info->delalloc_lock, flags); |
9069218d | 1147 | BTRFS_I(inode)->delalloc_bytes += end - start + 1; |
291d673e | 1148 | root->fs_info->delalloc_bytes += end - start + 1; |
ea8c2819 CM |
1149 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1150 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1151 | &root->fs_info->delalloc_inodes); | |
1152 | } | |
bcbfce8a | 1153 | spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags); |
291d673e CM |
1154 | } |
1155 | return 0; | |
1156 | } | |
1157 | ||
d352ac68 CM |
1158 | /* |
1159 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1160 | */ | |
291d673e | 1161 | int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end, |
b0c68f8b | 1162 | unsigned long old, unsigned long bits) |
291d673e | 1163 | { |
b0c68f8b | 1164 | if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) { |
291d673e | 1165 | struct btrfs_root *root = BTRFS_I(inode)->root; |
bcbfce8a CM |
1166 | unsigned long flags; |
1167 | ||
1168 | spin_lock_irqsave(&root->fs_info->delalloc_lock, flags); | |
b0c68f8b CM |
1169 | if (end - start + 1 > root->fs_info->delalloc_bytes) { |
1170 | printk("warning: delalloc account %Lu %Lu\n", | |
1171 | end - start + 1, root->fs_info->delalloc_bytes); | |
1172 | root->fs_info->delalloc_bytes = 0; | |
9069218d | 1173 | BTRFS_I(inode)->delalloc_bytes = 0; |
b0c68f8b CM |
1174 | } else { |
1175 | root->fs_info->delalloc_bytes -= end - start + 1; | |
9069218d | 1176 | BTRFS_I(inode)->delalloc_bytes -= end - start + 1; |
b0c68f8b | 1177 | } |
ea8c2819 CM |
1178 | if (BTRFS_I(inode)->delalloc_bytes == 0 && |
1179 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1180 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1181 | } | |
bcbfce8a | 1182 | spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags); |
291d673e CM |
1183 | } |
1184 | return 0; | |
1185 | } | |
1186 | ||
d352ac68 CM |
1187 | /* |
1188 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1189 | * we don't create bios that span stripes or chunks | |
1190 | */ | |
239b14b3 | 1191 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1192 | size_t size, struct bio *bio, |
1193 | unsigned long bio_flags) | |
239b14b3 CM |
1194 | { |
1195 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1196 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1197 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1198 | u64 length = 0; |
1199 | u64 map_length; | |
239b14b3 CM |
1200 | int ret; |
1201 | ||
771ed689 CM |
1202 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1203 | return 0; | |
1204 | ||
f2d8d74d | 1205 | length = bio->bi_size; |
239b14b3 CM |
1206 | map_tree = &root->fs_info->mapping_tree; |
1207 | map_length = length; | |
cea9e445 | 1208 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1209 | &map_length, NULL, 0); |
cea9e445 | 1210 | |
239b14b3 | 1211 | if (map_length < length + size) { |
239b14b3 CM |
1212 | return 1; |
1213 | } | |
1214 | return 0; | |
1215 | } | |
1216 | ||
d352ac68 CM |
1217 | /* |
1218 | * in order to insert checksums into the metadata in large chunks, | |
1219 | * we wait until bio submission time. All the pages in the bio are | |
1220 | * checksummed and sums are attached onto the ordered extent record. | |
1221 | * | |
1222 | * At IO completion time the cums attached on the ordered extent record | |
1223 | * are inserted into the btree | |
1224 | */ | |
4a69a410 | 1225 | int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1226 | int mirror_num, unsigned long bio_flags) |
065631f6 | 1227 | { |
065631f6 | 1228 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1229 | int ret = 0; |
e015640f | 1230 | |
3edf7d33 | 1231 | ret = btrfs_csum_one_bio(root, inode, bio); |
44b8bd7e | 1232 | BUG_ON(ret); |
4a69a410 CM |
1233 | return 0; |
1234 | } | |
e015640f | 1235 | |
4a69a410 CM |
1236 | /* |
1237 | * in order to insert checksums into the metadata in large chunks, | |
1238 | * we wait until bio submission time. All the pages in the bio are | |
1239 | * checksummed and sums are attached onto the ordered extent record. | |
1240 | * | |
1241 | * At IO completion time the cums attached on the ordered extent record | |
1242 | * are inserted into the btree | |
1243 | */ | |
1244 | int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, | |
1245 | int mirror_num, unsigned long bio_flags) | |
1246 | { | |
1247 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1248 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1249 | } |
1250 | ||
d352ac68 CM |
1251 | /* |
1252 | * extent_io.c submission hook. This does the right thing for csum calculation on write, | |
1253 | * or reading the csums from the tree before a read | |
1254 | */ | |
44b8bd7e | 1255 | int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
c8b97818 | 1256 | int mirror_num, unsigned long bio_flags) |
44b8bd7e CM |
1257 | { |
1258 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1259 | int ret = 0; | |
19b9bdb0 | 1260 | int skip_sum; |
44b8bd7e | 1261 | |
e6dcd2dc CM |
1262 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
1263 | BUG_ON(ret); | |
065631f6 | 1264 | |
19b9bdb0 CM |
1265 | skip_sum = btrfs_test_opt(root, NODATASUM) || |
1266 | btrfs_test_flag(inode, NODATASUM); | |
7ea394f1 | 1267 | |
4d1b5fb4 | 1268 | if (!(rw & (1 << BIO_RW))) { |
c8b97818 | 1269 | |
19b9bdb0 | 1270 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
c8b97818 CM |
1271 | return btrfs_submit_compressed_read(inode, bio, |
1272 | mirror_num, bio_flags); | |
771ed689 CM |
1273 | else if (!skip_sum) |
1274 | btrfs_lookup_bio_sums(root, inode, bio); | |
4d1b5fb4 | 1275 | goto mapit; |
19b9bdb0 CM |
1276 | } else if (!skip_sum) { |
1277 | /* we're doing a write, do the async checksumming */ | |
1278 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1279 | inode, rw, bio, mirror_num, |
4a69a410 CM |
1280 | bio_flags, __btrfs_submit_bio_start, |
1281 | __btrfs_submit_bio_done); | |
19b9bdb0 CM |
1282 | } |
1283 | ||
0b86a832 | 1284 | mapit: |
8b712842 | 1285 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1286 | } |
6885f308 | 1287 | |
d352ac68 CM |
1288 | /* |
1289 | * given a list of ordered sums record them in the inode. This happens | |
1290 | * at IO completion time based on sums calculated at bio submission time. | |
1291 | */ | |
ba1da2f4 | 1292 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1293 | struct inode *inode, u64 file_offset, |
1294 | struct list_head *list) | |
1295 | { | |
1296 | struct list_head *cur; | |
1297 | struct btrfs_ordered_sum *sum; | |
1298 | ||
1299 | btrfs_set_trans_block_group(trans, inode); | |
ba1da2f4 | 1300 | list_for_each(cur, list) { |
e6dcd2dc | 1301 | sum = list_entry(cur, struct btrfs_ordered_sum, list); |
e6dcd2dc CM |
1302 | btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root, |
1303 | inode, sum); | |
e6dcd2dc CM |
1304 | } |
1305 | return 0; | |
1306 | } | |
1307 | ||
ea8c2819 CM |
1308 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end) |
1309 | { | |
771ed689 CM |
1310 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) { |
1311 | WARN_ON(1); | |
1312 | } | |
ea8c2819 CM |
1313 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
1314 | GFP_NOFS); | |
1315 | } | |
1316 | ||
d352ac68 | 1317 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1318 | struct btrfs_writepage_fixup { |
1319 | struct page *page; | |
1320 | struct btrfs_work work; | |
1321 | }; | |
1322 | ||
247e743c CM |
1323 | void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
1324 | { | |
1325 | struct btrfs_writepage_fixup *fixup; | |
1326 | struct btrfs_ordered_extent *ordered; | |
1327 | struct page *page; | |
1328 | struct inode *inode; | |
1329 | u64 page_start; | |
1330 | u64 page_end; | |
1331 | ||
1332 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1333 | page = fixup->page; | |
4a096752 | 1334 | again: |
247e743c CM |
1335 | lock_page(page); |
1336 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1337 | ClearPageChecked(page); | |
1338 | goto out_page; | |
1339 | } | |
1340 | ||
1341 | inode = page->mapping->host; | |
1342 | page_start = page_offset(page); | |
1343 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1344 | ||
1345 | lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
4a096752 CM |
1346 | |
1347 | /* already ordered? We're done */ | |
1348 | if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, | |
1349 | EXTENT_ORDERED, 0)) { | |
247e743c | 1350 | goto out; |
4a096752 CM |
1351 | } |
1352 | ||
1353 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1354 | if (ordered) { | |
1355 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, | |
1356 | page_end, GFP_NOFS); | |
1357 | unlock_page(page); | |
1358 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1359 | goto again; | |
1360 | } | |
247e743c | 1361 | |
ea8c2819 | 1362 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
247e743c CM |
1363 | ClearPageChecked(page); |
1364 | out: | |
1365 | unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS); | |
1366 | out_page: | |
1367 | unlock_page(page); | |
1368 | page_cache_release(page); | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * There are a few paths in the higher layers of the kernel that directly | |
1373 | * set the page dirty bit without asking the filesystem if it is a | |
1374 | * good idea. This causes problems because we want to make sure COW | |
1375 | * properly happens and the data=ordered rules are followed. | |
1376 | * | |
c8b97818 | 1377 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1378 | * hasn't been properly setup for IO. We kick off an async process |
1379 | * to fix it up. The async helper will wait for ordered extents, set | |
1380 | * the delalloc bit and make it safe to write the page. | |
1381 | */ | |
1382 | int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) | |
1383 | { | |
1384 | struct inode *inode = page->mapping->host; | |
1385 | struct btrfs_writepage_fixup *fixup; | |
1386 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1387 | int ret; | |
1388 | ||
1389 | ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1390 | EXTENT_ORDERED, 0); | |
1391 | if (ret) | |
1392 | return 0; | |
1393 | ||
1394 | if (PageChecked(page)) | |
1395 | return -EAGAIN; | |
1396 | ||
1397 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1398 | if (!fixup) | |
1399 | return -EAGAIN; | |
f421950f | 1400 | |
247e743c CM |
1401 | SetPageChecked(page); |
1402 | page_cache_get(page); | |
1403 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1404 | fixup->page = page; | |
1405 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1406 | return -EAGAIN; | |
1407 | } | |
1408 | ||
d899e052 YZ |
1409 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1410 | struct inode *inode, u64 file_pos, | |
1411 | u64 disk_bytenr, u64 disk_num_bytes, | |
1412 | u64 num_bytes, u64 ram_bytes, | |
1413 | u8 compression, u8 encryption, | |
1414 | u16 other_encoding, int extent_type) | |
1415 | { | |
1416 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1417 | struct btrfs_file_extent_item *fi; | |
1418 | struct btrfs_path *path; | |
1419 | struct extent_buffer *leaf; | |
1420 | struct btrfs_key ins; | |
1421 | u64 hint; | |
1422 | int ret; | |
1423 | ||
1424 | path = btrfs_alloc_path(); | |
1425 | BUG_ON(!path); | |
1426 | ||
1427 | ret = btrfs_drop_extents(trans, root, inode, file_pos, | |
1428 | file_pos + num_bytes, file_pos, &hint); | |
1429 | BUG_ON(ret); | |
1430 | ||
1431 | ins.objectid = inode->i_ino; | |
1432 | ins.offset = file_pos; | |
1433 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1434 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1435 | BUG_ON(ret); | |
1436 | leaf = path->nodes[0]; | |
1437 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1438 | struct btrfs_file_extent_item); | |
1439 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1440 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1441 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1442 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1443 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1444 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1445 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1446 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1447 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1448 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
1449 | btrfs_mark_buffer_dirty(leaf); | |
1450 | ||
1451 | inode_add_bytes(inode, num_bytes); | |
1452 | btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0); | |
1453 | ||
1454 | ins.objectid = disk_bytenr; | |
1455 | ins.offset = disk_num_bytes; | |
1456 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1457 | ret = btrfs_alloc_reserved_extent(trans, root, leaf->start, | |
1458 | root->root_key.objectid, | |
1459 | trans->transid, inode->i_ino, &ins); | |
1460 | BUG_ON(ret); | |
1461 | ||
1462 | btrfs_free_path(path); | |
1463 | return 0; | |
1464 | } | |
1465 | ||
d352ac68 CM |
1466 | /* as ordered data IO finishes, this gets called so we can finish |
1467 | * an ordered extent if the range of bytes in the file it covers are | |
1468 | * fully written. | |
1469 | */ | |
211f90e6 | 1470 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1471 | { |
e6dcd2dc CM |
1472 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1473 | struct btrfs_trans_handle *trans; | |
1474 | struct btrfs_ordered_extent *ordered_extent; | |
1475 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
d899e052 | 1476 | int compressed = 0; |
e6dcd2dc CM |
1477 | int ret; |
1478 | ||
1479 | ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1); | |
ba1da2f4 | 1480 | if (!ret) |
e6dcd2dc | 1481 | return 0; |
e6dcd2dc | 1482 | |
f9295749 | 1483 | trans = btrfs_join_transaction(root, 1); |
e6dcd2dc CM |
1484 | |
1485 | ordered_extent = btrfs_lookup_ordered_extent(inode, start); | |
1486 | BUG_ON(!ordered_extent); | |
7ea394f1 YZ |
1487 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) |
1488 | goto nocow; | |
e6dcd2dc CM |
1489 | |
1490 | lock_extent(io_tree, ordered_extent->file_offset, | |
1491 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1492 | GFP_NOFS); | |
1493 | ||
c8b97818 | 1494 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
d899e052 YZ |
1495 | compressed = 1; |
1496 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { | |
1497 | BUG_ON(compressed); | |
1498 | ret = btrfs_mark_extent_written(trans, root, inode, | |
1499 | ordered_extent->file_offset, | |
1500 | ordered_extent->file_offset + | |
1501 | ordered_extent->len); | |
1502 | BUG_ON(ret); | |
1503 | } else { | |
1504 | ret = insert_reserved_file_extent(trans, inode, | |
1505 | ordered_extent->file_offset, | |
1506 | ordered_extent->start, | |
1507 | ordered_extent->disk_len, | |
1508 | ordered_extent->len, | |
1509 | ordered_extent->len, | |
1510 | compressed, 0, 0, | |
1511 | BTRFS_FILE_EXTENT_REG); | |
1512 | BUG_ON(ret); | |
1513 | } | |
e6dcd2dc CM |
1514 | unlock_extent(io_tree, ordered_extent->file_offset, |
1515 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1516 | GFP_NOFS); | |
7ea394f1 | 1517 | nocow: |
e6dcd2dc CM |
1518 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1519 | &ordered_extent->list); | |
1520 | ||
34353029 | 1521 | mutex_lock(&BTRFS_I(inode)->extent_mutex); |
dbe674a9 | 1522 | btrfs_ordered_update_i_size(inode, ordered_extent); |
e02119d5 | 1523 | btrfs_update_inode(trans, root, inode); |
e6dcd2dc | 1524 | btrfs_remove_ordered_extent(inode, ordered_extent); |
34353029 | 1525 | mutex_unlock(&BTRFS_I(inode)->extent_mutex); |
7f3c74fb | 1526 | |
e6dcd2dc CM |
1527 | /* once for us */ |
1528 | btrfs_put_ordered_extent(ordered_extent); | |
1529 | /* once for the tree */ | |
1530 | btrfs_put_ordered_extent(ordered_extent); | |
1531 | ||
e6dcd2dc CM |
1532 | btrfs_end_transaction(trans, root); |
1533 | return 0; | |
1534 | } | |
1535 | ||
211f90e6 CM |
1536 | int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
1537 | struct extent_state *state, int uptodate) | |
1538 | { | |
1539 | return btrfs_finish_ordered_io(page->mapping->host, start, end); | |
1540 | } | |
1541 | ||
d352ac68 CM |
1542 | /* |
1543 | * When IO fails, either with EIO or csum verification fails, we | |
1544 | * try other mirrors that might have a good copy of the data. This | |
1545 | * io_failure_record is used to record state as we go through all the | |
1546 | * mirrors. If another mirror has good data, the page is set up to date | |
1547 | * and things continue. If a good mirror can't be found, the original | |
1548 | * bio end_io callback is called to indicate things have failed. | |
1549 | */ | |
7e38326f CM |
1550 | struct io_failure_record { |
1551 | struct page *page; | |
1552 | u64 start; | |
1553 | u64 len; | |
1554 | u64 logical; | |
1555 | int last_mirror; | |
1556 | }; | |
1557 | ||
1259ab75 CM |
1558 | int btrfs_io_failed_hook(struct bio *failed_bio, |
1559 | struct page *page, u64 start, u64 end, | |
1560 | struct extent_state *state) | |
7e38326f CM |
1561 | { |
1562 | struct io_failure_record *failrec = NULL; | |
1563 | u64 private; | |
1564 | struct extent_map *em; | |
1565 | struct inode *inode = page->mapping->host; | |
1566 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1567 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1568 | struct bio *bio; |
1569 | int num_copies; | |
1570 | int ret; | |
1259ab75 | 1571 | int rw; |
7e38326f | 1572 | u64 logical; |
c8b97818 | 1573 | unsigned long bio_flags = 0; |
7e38326f CM |
1574 | |
1575 | ret = get_state_private(failure_tree, start, &private); | |
1576 | if (ret) { | |
7e38326f CM |
1577 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1578 | if (!failrec) | |
1579 | return -ENOMEM; | |
1580 | failrec->start = start; | |
1581 | failrec->len = end - start + 1; | |
1582 | failrec->last_mirror = 0; | |
1583 | ||
3b951516 CM |
1584 | spin_lock(&em_tree->lock); |
1585 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
1586 | if (em->start > start || em->start + em->len < start) { | |
1587 | free_extent_map(em); | |
1588 | em = NULL; | |
1589 | } | |
1590 | spin_unlock(&em_tree->lock); | |
7e38326f CM |
1591 | |
1592 | if (!em || IS_ERR(em)) { | |
1593 | kfree(failrec); | |
1594 | return -EIO; | |
1595 | } | |
1596 | logical = start - em->start; | |
1597 | logical = em->block_start + logical; | |
c8b97818 CM |
1598 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
1599 | bio_flags = EXTENT_BIO_COMPRESSED; | |
7e38326f CM |
1600 | failrec->logical = logical; |
1601 | free_extent_map(em); | |
1602 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1603 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1604 | set_state_private(failure_tree, start, |
1605 | (u64)(unsigned long)failrec); | |
7e38326f | 1606 | } else { |
587f7704 | 1607 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1608 | } |
1609 | num_copies = btrfs_num_copies( | |
1610 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1611 | failrec->logical, failrec->len); | |
1612 | failrec->last_mirror++; | |
1613 | if (!state) { | |
1614 | spin_lock_irq(&BTRFS_I(inode)->io_tree.lock); | |
1615 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, | |
1616 | failrec->start, | |
1617 | EXTENT_LOCKED); | |
1618 | if (state && state->start != failrec->start) | |
1619 | state = NULL; | |
1620 | spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock); | |
1621 | } | |
1622 | if (!state || failrec->last_mirror > num_copies) { | |
1623 | set_state_private(failure_tree, failrec->start, 0); | |
1624 | clear_extent_bits(failure_tree, failrec->start, | |
1625 | failrec->start + failrec->len - 1, | |
1626 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1627 | kfree(failrec); | |
1628 | return -EIO; | |
1629 | } | |
1630 | bio = bio_alloc(GFP_NOFS, 1); | |
1631 | bio->bi_private = state; | |
1632 | bio->bi_end_io = failed_bio->bi_end_io; | |
1633 | bio->bi_sector = failrec->logical >> 9; | |
1634 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1635 | bio->bi_size = 0; |
7e38326f | 1636 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
1259ab75 CM |
1637 | if (failed_bio->bi_rw & (1 << BIO_RW)) |
1638 | rw = WRITE; | |
1639 | else | |
1640 | rw = READ; | |
1641 | ||
1642 | BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, | |
c8b97818 CM |
1643 | failrec->last_mirror, |
1644 | bio_flags); | |
1259ab75 CM |
1645 | return 0; |
1646 | } | |
1647 | ||
d352ac68 CM |
1648 | /* |
1649 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1650 | * to see if we need to process or clean up an io_failure_record | |
1651 | */ | |
1259ab75 CM |
1652 | int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1653 | { | |
1654 | u64 private; | |
1655 | u64 private_failure; | |
1656 | struct io_failure_record *failure; | |
1657 | int ret; | |
1658 | ||
1659 | private = 0; | |
1660 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
1661 | (u64)-1, 1, EXTENT_DIRTY)) { | |
1662 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1663 | start, &private_failure); | |
1664 | if (ret == 0) { | |
1665 | failure = (struct io_failure_record *)(unsigned long) | |
1666 | private_failure; | |
1667 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1668 | failure->start, 0); | |
1669 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1670 | failure->start, | |
1671 | failure->start + failure->len - 1, | |
1672 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1673 | GFP_NOFS); | |
1674 | kfree(failure); | |
1675 | } | |
1676 | } | |
7e38326f CM |
1677 | return 0; |
1678 | } | |
1679 | ||
d352ac68 CM |
1680 | /* |
1681 | * when reads are done, we need to check csums to verify the data is correct | |
1682 | * if there's a match, we allow the bio to finish. If not, we go through | |
1683 | * the io_failure_record routines to find good copies | |
1684 | */ | |
70dec807 CM |
1685 | int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
1686 | struct extent_state *state) | |
07157aac | 1687 | { |
35ebb934 | 1688 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1689 | struct inode *inode = page->mapping->host; |
d1310b2e | 1690 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1691 | char *kaddr; |
aadfeb6e | 1692 | u64 private = ~(u32)0; |
07157aac | 1693 | int ret; |
ff79f819 CM |
1694 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1695 | u32 csum = ~(u32)0; | |
bbf0d006 | 1696 | unsigned long flags; |
d1310b2e | 1697 | |
b98b6767 Y |
1698 | if (btrfs_test_opt(root, NODATASUM) || |
1699 | btrfs_test_flag(inode, NODATASUM)) | |
b6cda9bc | 1700 | return 0; |
c2e639f0 | 1701 | if (state && state->start == start) { |
70dec807 CM |
1702 | private = state->private; |
1703 | ret = 0; | |
1704 | } else { | |
1705 | ret = get_state_private(io_tree, start, &private); | |
1706 | } | |
bbf0d006 | 1707 | local_irq_save(flags); |
07157aac CM |
1708 | kaddr = kmap_atomic(page, KM_IRQ0); |
1709 | if (ret) { | |
1710 | goto zeroit; | |
1711 | } | |
ff79f819 CM |
1712 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1713 | btrfs_csum_final(csum, (char *)&csum); | |
1714 | if (csum != private) { | |
07157aac CM |
1715 | goto zeroit; |
1716 | } | |
1717 | kunmap_atomic(kaddr, KM_IRQ0); | |
bbf0d006 | 1718 | local_irq_restore(flags); |
7e38326f CM |
1719 | |
1720 | /* if the io failure tree for this inode is non-empty, | |
1721 | * check to see if we've recovered from a failed IO | |
1722 | */ | |
1259ab75 | 1723 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1724 | return 0; |
1725 | ||
1726 | zeroit: | |
aadfeb6e CM |
1727 | printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n", |
1728 | page->mapping->host->i_ino, (unsigned long long)start, csum, | |
1729 | private); | |
db94535d CM |
1730 | memset(kaddr + offset, 1, end - start + 1); |
1731 | flush_dcache_page(page); | |
07157aac | 1732 | kunmap_atomic(kaddr, KM_IRQ0); |
bbf0d006 | 1733 | local_irq_restore(flags); |
3b951516 CM |
1734 | if (private == 0) |
1735 | return 0; | |
7e38326f | 1736 | return -EIO; |
07157aac | 1737 | } |
b888db2b | 1738 | |
7b128766 JB |
1739 | /* |
1740 | * This creates an orphan entry for the given inode in case something goes | |
1741 | * wrong in the middle of an unlink/truncate. | |
1742 | */ | |
1743 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
1744 | { | |
1745 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1746 | int ret = 0; | |
1747 | ||
bcc63abb | 1748 | spin_lock(&root->list_lock); |
7b128766 JB |
1749 | |
1750 | /* already on the orphan list, we're good */ | |
1751 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1752 | spin_unlock(&root->list_lock); |
7b128766 JB |
1753 | return 0; |
1754 | } | |
1755 | ||
1756 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
1757 | ||
bcc63abb | 1758 | spin_unlock(&root->list_lock); |
7b128766 JB |
1759 | |
1760 | /* | |
1761 | * insert an orphan item to track this unlinked/truncated file | |
1762 | */ | |
1763 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
1764 | ||
1765 | return ret; | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
1770 | * item for this particular inode. | |
1771 | */ | |
1772 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
1773 | { | |
1774 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1775 | int ret = 0; | |
1776 | ||
bcc63abb | 1777 | spin_lock(&root->list_lock); |
7b128766 JB |
1778 | |
1779 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { | |
bcc63abb | 1780 | spin_unlock(&root->list_lock); |
7b128766 JB |
1781 | return 0; |
1782 | } | |
1783 | ||
1784 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
1785 | if (!trans) { | |
bcc63abb | 1786 | spin_unlock(&root->list_lock); |
7b128766 JB |
1787 | return 0; |
1788 | } | |
1789 | ||
bcc63abb | 1790 | spin_unlock(&root->list_lock); |
7b128766 JB |
1791 | |
1792 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
1793 | ||
1794 | return ret; | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * this cleans up any orphans that may be left on the list from the last use | |
1799 | * of this root. | |
1800 | */ | |
1801 | void btrfs_orphan_cleanup(struct btrfs_root *root) | |
1802 | { | |
1803 | struct btrfs_path *path; | |
1804 | struct extent_buffer *leaf; | |
1805 | struct btrfs_item *item; | |
1806 | struct btrfs_key key, found_key; | |
1807 | struct btrfs_trans_handle *trans; | |
1808 | struct inode *inode; | |
1809 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
1810 | ||
7b128766 JB |
1811 | path = btrfs_alloc_path(); |
1812 | if (!path) | |
1813 | return; | |
1814 | path->reada = -1; | |
1815 | ||
1816 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
1817 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
1818 | key.offset = (u64)-1; | |
1819 | ||
7b128766 JB |
1820 | |
1821 | while (1) { | |
1822 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1823 | if (ret < 0) { | |
1824 | printk(KERN_ERR "Error searching slot for orphan: %d" | |
1825 | "\n", ret); | |
1826 | break; | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * if ret == 0 means we found what we were searching for, which | |
1831 | * is weird, but possible, so only screw with path if we didnt | |
1832 | * find the key and see if we have stuff that matches | |
1833 | */ | |
1834 | if (ret > 0) { | |
1835 | if (path->slots[0] == 0) | |
1836 | break; | |
1837 | path->slots[0]--; | |
1838 | } | |
1839 | ||
1840 | /* pull out the item */ | |
1841 | leaf = path->nodes[0]; | |
1842 | item = btrfs_item_nr(leaf, path->slots[0]); | |
1843 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1844 | ||
1845 | /* make sure the item matches what we want */ | |
1846 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
1847 | break; | |
1848 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
1849 | break; | |
1850 | ||
1851 | /* release the path since we're done with it */ | |
1852 | btrfs_release_path(root, path); | |
1853 | ||
1854 | /* | |
1855 | * this is where we are basically btrfs_lookup, without the | |
1856 | * crossing root thing. we store the inode number in the | |
1857 | * offset of the orphan item. | |
1858 | */ | |
5b21f2ed | 1859 | inode = btrfs_iget_locked(root->fs_info->sb, |
7b128766 JB |
1860 | found_key.offset, root); |
1861 | if (!inode) | |
1862 | break; | |
1863 | ||
1864 | if (inode->i_state & I_NEW) { | |
1865 | BTRFS_I(inode)->root = root; | |
1866 | ||
1867 | /* have to set the location manually */ | |
1868 | BTRFS_I(inode)->location.objectid = inode->i_ino; | |
1869 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
1870 | BTRFS_I(inode)->location.offset = 0; | |
1871 | ||
1872 | btrfs_read_locked_inode(inode); | |
1873 | unlock_new_inode(inode); | |
1874 | } | |
1875 | ||
1876 | /* | |
1877 | * add this inode to the orphan list so btrfs_orphan_del does | |
1878 | * the proper thing when we hit it | |
1879 | */ | |
bcc63abb | 1880 | spin_lock(&root->list_lock); |
7b128766 | 1881 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
bcc63abb | 1882 | spin_unlock(&root->list_lock); |
7b128766 JB |
1883 | |
1884 | /* | |
1885 | * if this is a bad inode, means we actually succeeded in | |
1886 | * removing the inode, but not the orphan record, which means | |
1887 | * we need to manually delete the orphan since iput will just | |
1888 | * do a destroy_inode | |
1889 | */ | |
1890 | if (is_bad_inode(inode)) { | |
5b21f2ed | 1891 | trans = btrfs_start_transaction(root, 1); |
7b128766 | 1892 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 1893 | btrfs_end_transaction(trans, root); |
7b128766 JB |
1894 | iput(inode); |
1895 | continue; | |
1896 | } | |
1897 | ||
1898 | /* if we have links, this was a truncate, lets do that */ | |
1899 | if (inode->i_nlink) { | |
1900 | nr_truncate++; | |
1901 | btrfs_truncate(inode); | |
1902 | } else { | |
1903 | nr_unlink++; | |
1904 | } | |
1905 | ||
1906 | /* this will do delete_inode and everything for us */ | |
1907 | iput(inode); | |
1908 | } | |
1909 | ||
1910 | if (nr_unlink) | |
1911 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
1912 | if (nr_truncate) | |
1913 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
1914 | ||
1915 | btrfs_free_path(path); | |
7b128766 JB |
1916 | } |
1917 | ||
d352ac68 CM |
1918 | /* |
1919 | * read an inode from the btree into the in-memory inode | |
1920 | */ | |
39279cc3 CM |
1921 | void btrfs_read_locked_inode(struct inode *inode) |
1922 | { | |
1923 | struct btrfs_path *path; | |
5f39d397 | 1924 | struct extent_buffer *leaf; |
39279cc3 | 1925 | struct btrfs_inode_item *inode_item; |
0b86a832 | 1926 | struct btrfs_timespec *tspec; |
39279cc3 CM |
1927 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1928 | struct btrfs_key location; | |
1929 | u64 alloc_group_block; | |
618e21d5 | 1930 | u32 rdev; |
39279cc3 CM |
1931 | int ret; |
1932 | ||
1933 | path = btrfs_alloc_path(); | |
1934 | BUG_ON(!path); | |
39279cc3 | 1935 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 1936 | |
39279cc3 | 1937 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 1938 | if (ret) |
39279cc3 | 1939 | goto make_bad; |
39279cc3 | 1940 | |
5f39d397 CM |
1941 | leaf = path->nodes[0]; |
1942 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
1943 | struct btrfs_inode_item); | |
1944 | ||
1945 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
1946 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
1947 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
1948 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 1949 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
1950 | |
1951 | tspec = btrfs_inode_atime(inode_item); | |
1952 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
1953 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
1954 | ||
1955 | tspec = btrfs_inode_mtime(inode_item); | |
1956 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
1957 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
1958 | ||
1959 | tspec = btrfs_inode_ctime(inode_item); | |
1960 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
1961 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
1962 | ||
a76a3cd4 | 1963 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 CM |
1964 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
1965 | inode->i_generation = BTRFS_I(inode)->generation; | |
618e21d5 | 1966 | inode->i_rdev = 0; |
5f39d397 CM |
1967 | rdev = btrfs_inode_rdev(leaf, inode_item); |
1968 | ||
aec7477b JB |
1969 | BTRFS_I(inode)->index_cnt = (u64)-1; |
1970 | ||
5f39d397 | 1971 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
39279cc3 CM |
1972 | BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info, |
1973 | alloc_group_block); | |
b98b6767 | 1974 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
e52ec0eb CM |
1975 | if (!BTRFS_I(inode)->block_group) { |
1976 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, | |
0b86a832 CM |
1977 | NULL, 0, |
1978 | BTRFS_BLOCK_GROUP_METADATA, 0); | |
e52ec0eb | 1979 | } |
39279cc3 CM |
1980 | btrfs_free_path(path); |
1981 | inode_item = NULL; | |
1982 | ||
39279cc3 | 1983 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
1984 | case S_IFREG: |
1985 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 1986 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 1987 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
1988 | inode->i_fop = &btrfs_file_operations; |
1989 | inode->i_op = &btrfs_file_inode_operations; | |
1990 | break; | |
1991 | case S_IFDIR: | |
1992 | inode->i_fop = &btrfs_dir_file_operations; | |
1993 | if (root == root->fs_info->tree_root) | |
1994 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
1995 | else | |
1996 | inode->i_op = &btrfs_dir_inode_operations; | |
1997 | break; | |
1998 | case S_IFLNK: | |
1999 | inode->i_op = &btrfs_symlink_inode_operations; | |
2000 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2001 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2002 | break; |
618e21d5 JB |
2003 | default: |
2004 | init_special_inode(inode, inode->i_mode, rdev); | |
2005 | break; | |
39279cc3 CM |
2006 | } |
2007 | return; | |
2008 | ||
2009 | make_bad: | |
39279cc3 | 2010 | btrfs_free_path(path); |
39279cc3 CM |
2011 | make_bad_inode(inode); |
2012 | } | |
2013 | ||
d352ac68 CM |
2014 | /* |
2015 | * given a leaf and an inode, copy the inode fields into the leaf | |
2016 | */ | |
e02119d5 CM |
2017 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2018 | struct extent_buffer *leaf, | |
5f39d397 | 2019 | struct btrfs_inode_item *item, |
39279cc3 CM |
2020 | struct inode *inode) |
2021 | { | |
5f39d397 CM |
2022 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2023 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2024 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2025 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2026 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2027 | ||
2028 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2029 | inode->i_atime.tv_sec); | |
2030 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2031 | inode->i_atime.tv_nsec); | |
2032 | ||
2033 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2034 | inode->i_mtime.tv_sec); | |
2035 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2036 | inode->i_mtime.tv_nsec); | |
2037 | ||
2038 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2039 | inode->i_ctime.tv_sec); | |
2040 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2041 | inode->i_ctime.tv_nsec); | |
2042 | ||
a76a3cd4 | 2043 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 CM |
2044 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
2045 | btrfs_set_inode_transid(leaf, item, trans->transid); | |
5f39d397 | 2046 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2047 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
5f39d397 | 2048 | btrfs_set_inode_block_group(leaf, item, |
39279cc3 CM |
2049 | BTRFS_I(inode)->block_group->key.objectid); |
2050 | } | |
2051 | ||
d352ac68 CM |
2052 | /* |
2053 | * copy everything in the in-memory inode into the btree. | |
2054 | */ | |
ba1da2f4 | 2055 | int noinline btrfs_update_inode(struct btrfs_trans_handle *trans, |
39279cc3 CM |
2056 | struct btrfs_root *root, |
2057 | struct inode *inode) | |
2058 | { | |
2059 | struct btrfs_inode_item *inode_item; | |
2060 | struct btrfs_path *path; | |
5f39d397 | 2061 | struct extent_buffer *leaf; |
39279cc3 CM |
2062 | int ret; |
2063 | ||
2064 | path = btrfs_alloc_path(); | |
2065 | BUG_ON(!path); | |
39279cc3 CM |
2066 | ret = btrfs_lookup_inode(trans, root, path, |
2067 | &BTRFS_I(inode)->location, 1); | |
2068 | if (ret) { | |
2069 | if (ret > 0) | |
2070 | ret = -ENOENT; | |
2071 | goto failed; | |
2072 | } | |
2073 | ||
5f39d397 CM |
2074 | leaf = path->nodes[0]; |
2075 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2076 | struct btrfs_inode_item); |
2077 | ||
e02119d5 | 2078 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2079 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2080 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2081 | ret = 0; |
2082 | failed: | |
39279cc3 CM |
2083 | btrfs_free_path(path); |
2084 | return ret; | |
2085 | } | |
2086 | ||
2087 | ||
d352ac68 CM |
2088 | /* |
2089 | * unlink helper that gets used here in inode.c and in the tree logging | |
2090 | * recovery code. It remove a link in a directory with a given name, and | |
2091 | * also drops the back refs in the inode to the directory | |
2092 | */ | |
e02119d5 CM |
2093 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2094 | struct btrfs_root *root, | |
2095 | struct inode *dir, struct inode *inode, | |
2096 | const char *name, int name_len) | |
39279cc3 CM |
2097 | { |
2098 | struct btrfs_path *path; | |
39279cc3 | 2099 | int ret = 0; |
5f39d397 | 2100 | struct extent_buffer *leaf; |
39279cc3 | 2101 | struct btrfs_dir_item *di; |
5f39d397 | 2102 | struct btrfs_key key; |
aec7477b | 2103 | u64 index; |
39279cc3 CM |
2104 | |
2105 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2106 | if (!path) { |
2107 | ret = -ENOMEM; | |
2108 | goto err; | |
2109 | } | |
2110 | ||
39279cc3 CM |
2111 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2112 | name, name_len, -1); | |
2113 | if (IS_ERR(di)) { | |
2114 | ret = PTR_ERR(di); | |
2115 | goto err; | |
2116 | } | |
2117 | if (!di) { | |
2118 | ret = -ENOENT; | |
2119 | goto err; | |
2120 | } | |
5f39d397 CM |
2121 | leaf = path->nodes[0]; |
2122 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2123 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2124 | if (ret) |
2125 | goto err; | |
39279cc3 CM |
2126 | btrfs_release_path(root, path); |
2127 | ||
aec7477b | 2128 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2129 | inode->i_ino, |
2130 | dir->i_ino, &index); | |
aec7477b JB |
2131 | if (ret) { |
2132 | printk("failed to delete reference to %.*s, " | |
2133 | "inode %lu parent %lu\n", name_len, name, | |
e02119d5 | 2134 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2135 | goto err; |
2136 | } | |
2137 | ||
39279cc3 | 2138 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2139 | index, name, name_len, -1); |
39279cc3 CM |
2140 | if (IS_ERR(di)) { |
2141 | ret = PTR_ERR(di); | |
2142 | goto err; | |
2143 | } | |
2144 | if (!di) { | |
2145 | ret = -ENOENT; | |
2146 | goto err; | |
2147 | } | |
2148 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2149 | btrfs_release_path(root, path); |
39279cc3 | 2150 | |
e02119d5 CM |
2151 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2152 | inode, dir->i_ino); | |
49eb7e46 CM |
2153 | BUG_ON(ret != 0 && ret != -ENOENT); |
2154 | if (ret != -ENOENT) | |
2155 | BTRFS_I(dir)->log_dirty_trans = trans->transid; | |
e02119d5 CM |
2156 | |
2157 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2158 | dir, index); | |
2159 | BUG_ON(ret); | |
39279cc3 CM |
2160 | err: |
2161 | btrfs_free_path(path); | |
e02119d5 CM |
2162 | if (ret) |
2163 | goto out; | |
2164 | ||
2165 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2166 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2167 | btrfs_update_inode(trans, root, dir); | |
2168 | btrfs_drop_nlink(inode); | |
2169 | ret = btrfs_update_inode(trans, root, inode); | |
2170 | dir->i_sb->s_dirt = 1; | |
2171 | out: | |
39279cc3 CM |
2172 | return ret; |
2173 | } | |
2174 | ||
2175 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2176 | { | |
2177 | struct btrfs_root *root; | |
2178 | struct btrfs_trans_handle *trans; | |
7b128766 | 2179 | struct inode *inode = dentry->d_inode; |
39279cc3 | 2180 | int ret; |
1832a6d5 | 2181 | unsigned long nr = 0; |
39279cc3 CM |
2182 | |
2183 | root = BTRFS_I(dir)->root; | |
1832a6d5 CM |
2184 | |
2185 | ret = btrfs_check_free_space(root, 1, 1); | |
2186 | if (ret) | |
2187 | goto fail; | |
2188 | ||
39279cc3 | 2189 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2190 | |
39279cc3 | 2191 | btrfs_set_trans_block_group(trans, dir); |
e02119d5 CM |
2192 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2193 | dentry->d_name.name, dentry->d_name.len); | |
7b128766 JB |
2194 | |
2195 | if (inode->i_nlink == 0) | |
2196 | ret = btrfs_orphan_add(trans, inode); | |
2197 | ||
d3c2fdcf | 2198 | nr = trans->blocks_used; |
5f39d397 | 2199 | |
89ce8a63 | 2200 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2201 | fail: |
d3c2fdcf | 2202 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2203 | return ret; |
2204 | } | |
2205 | ||
2206 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) | |
2207 | { | |
2208 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2209 | int err = 0; |
39279cc3 CM |
2210 | int ret; |
2211 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
39279cc3 | 2212 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2213 | unsigned long nr = 0; |
39279cc3 | 2214 | |
3394e160 CM |
2215 | /* |
2216 | * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir | |
2217 | * the root of a subvolume or snapshot | |
2218 | */ | |
2219 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || | |
2220 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
134d4512 | 2221 | return -ENOTEMPTY; |
925baedd | 2222 | } |
134d4512 | 2223 | |
1832a6d5 CM |
2224 | ret = btrfs_check_free_space(root, 1, 1); |
2225 | if (ret) | |
2226 | goto fail; | |
2227 | ||
39279cc3 CM |
2228 | trans = btrfs_start_transaction(root, 1); |
2229 | btrfs_set_trans_block_group(trans, dir); | |
39279cc3 | 2230 | |
7b128766 JB |
2231 | err = btrfs_orphan_add(trans, inode); |
2232 | if (err) | |
2233 | goto fail_trans; | |
2234 | ||
39279cc3 | 2235 | /* now the directory is empty */ |
e02119d5 CM |
2236 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2237 | dentry->d_name.name, dentry->d_name.len); | |
39279cc3 | 2238 | if (!err) { |
dbe674a9 | 2239 | btrfs_i_size_write(inode, 0); |
39279cc3 | 2240 | } |
3954401f | 2241 | |
7b128766 | 2242 | fail_trans: |
d3c2fdcf | 2243 | nr = trans->blocks_used; |
89ce8a63 | 2244 | ret = btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 2245 | fail: |
d3c2fdcf | 2246 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2247 | |
39279cc3 CM |
2248 | if (ret && !err) |
2249 | err = ret; | |
2250 | return err; | |
2251 | } | |
2252 | ||
323ac95b CM |
2253 | /* |
2254 | * when truncating bytes in a file, it is possible to avoid reading | |
2255 | * the leaves that contain only checksum items. This can be the | |
2256 | * majority of the IO required to delete a large file, but it must | |
2257 | * be done carefully. | |
2258 | * | |
2259 | * The keys in the level just above the leaves are checked to make sure | |
2260 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
2261 | * after the new size. | |
2262 | * | |
2263 | * Then the key for the next leaf is checked to make sure it also has | |
2264 | * a checksum item for the same file. If it does, we know our target leaf | |
2265 | * contains only checksum items, and it can be safely freed without reading | |
2266 | * it. | |
2267 | * | |
2268 | * This is just an optimization targeted at large files. It may do | |
2269 | * nothing. It will return 0 unless things went badly. | |
2270 | */ | |
2271 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
2272 | struct btrfs_root *root, | |
2273 | struct btrfs_path *path, | |
2274 | struct inode *inode, u64 new_size) | |
2275 | { | |
2276 | struct btrfs_key key; | |
2277 | int ret; | |
2278 | int nritems; | |
2279 | struct btrfs_key found_key; | |
2280 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
2281 | struct btrfs_leaf_ref *ref; |
2282 | u64 leaf_gen; | |
2283 | u64 leaf_start; | |
323ac95b CM |
2284 | |
2285 | path->lowest_level = 1; | |
2286 | key.objectid = inode->i_ino; | |
2287 | key.type = BTRFS_CSUM_ITEM_KEY; | |
2288 | key.offset = new_size; | |
2289 | again: | |
2290 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2291 | if (ret < 0) | |
2292 | goto out; | |
2293 | ||
2294 | if (path->nodes[1] == NULL) { | |
2295 | ret = 0; | |
2296 | goto out; | |
2297 | } | |
2298 | ret = 0; | |
2299 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
2300 | nritems = btrfs_header_nritems(path->nodes[1]); | |
2301 | ||
2302 | if (!nritems) | |
2303 | goto out; | |
2304 | ||
2305 | if (path->slots[1] >= nritems) | |
2306 | goto next_node; | |
2307 | ||
2308 | /* did we find a key greater than anything we want to delete? */ | |
2309 | if (found_key.objectid > inode->i_ino || | |
2310 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
2311 | goto out; | |
2312 | ||
2313 | /* we check the next key in the node to make sure the leave contains | |
2314 | * only checksum items. This comparison doesn't work if our | |
2315 | * leaf is the last one in the node | |
2316 | */ | |
2317 | if (path->slots[1] + 1 >= nritems) { | |
2318 | next_node: | |
2319 | /* search forward from the last key in the node, this | |
2320 | * will bring us into the next node in the tree | |
2321 | */ | |
2322 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
2323 | ||
2324 | /* unlikely, but we inc below, so check to be safe */ | |
2325 | if (found_key.offset == (u64)-1) | |
2326 | goto out; | |
2327 | ||
2328 | /* search_forward needs a path with locks held, do the | |
2329 | * search again for the original key. It is possible | |
2330 | * this will race with a balance and return a path that | |
2331 | * we could modify, but this drop is just an optimization | |
2332 | * and is allowed to miss some leaves. | |
2333 | */ | |
2334 | btrfs_release_path(root, path); | |
2335 | found_key.offset++; | |
2336 | ||
2337 | /* setup a max key for search_forward */ | |
2338 | other_key.offset = (u64)-1; | |
2339 | other_key.type = key.type; | |
2340 | other_key.objectid = key.objectid; | |
2341 | ||
2342 | path->keep_locks = 1; | |
2343 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
2344 | path, 0, 0); | |
2345 | path->keep_locks = 0; | |
2346 | if (ret || found_key.objectid != key.objectid || | |
2347 | found_key.type != key.type) { | |
2348 | ret = 0; | |
2349 | goto out; | |
2350 | } | |
2351 | ||
2352 | key.offset = found_key.offset; | |
2353 | btrfs_release_path(root, path); | |
2354 | cond_resched(); | |
2355 | goto again; | |
2356 | } | |
2357 | ||
2358 | /* we know there's one more slot after us in the tree, | |
2359 | * read that key so we can verify it is also a checksum item | |
2360 | */ | |
2361 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
2362 | ||
2363 | if (found_key.objectid < inode->i_ino) | |
2364 | goto next_key; | |
2365 | ||
2366 | if (found_key.type != key.type || found_key.offset < new_size) | |
2367 | goto next_key; | |
2368 | ||
2369 | /* | |
2370 | * if the key for the next leaf isn't a csum key from this objectid, | |
2371 | * we can't be sure there aren't good items inside this leaf. | |
2372 | * Bail out | |
2373 | */ | |
2374 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
2375 | goto out; | |
2376 | ||
5b84e8d6 YZ |
2377 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
2378 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
2379 | /* |
2380 | * it is safe to delete this leaf, it contains only | |
2381 | * csum items from this inode at an offset >= new_size | |
2382 | */ | |
5b84e8d6 | 2383 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
2384 | BUG_ON(ret); |
2385 | ||
5b84e8d6 YZ |
2386 | if (root->ref_cows && leaf_gen < trans->transid) { |
2387 | ref = btrfs_alloc_leaf_ref(root, 0); | |
2388 | if (ref) { | |
2389 | ref->root_gen = root->root_key.offset; | |
2390 | ref->bytenr = leaf_start; | |
2391 | ref->owner = 0; | |
2392 | ref->generation = leaf_gen; | |
2393 | ref->nritems = 0; | |
2394 | ||
2395 | ret = btrfs_add_leaf_ref(root, ref, 0); | |
2396 | WARN_ON(ret); | |
2397 | btrfs_free_leaf_ref(root, ref); | |
2398 | } else { | |
2399 | WARN_ON(1); | |
2400 | } | |
2401 | } | |
323ac95b CM |
2402 | next_key: |
2403 | btrfs_release_path(root, path); | |
2404 | ||
2405 | if (other_key.objectid == inode->i_ino && | |
2406 | other_key.type == key.type && other_key.offset > key.offset) { | |
2407 | key.offset = other_key.offset; | |
2408 | cond_resched(); | |
2409 | goto again; | |
2410 | } | |
2411 | ret = 0; | |
2412 | out: | |
2413 | /* fixup any changes we've made to the path */ | |
2414 | path->lowest_level = 0; | |
2415 | path->keep_locks = 0; | |
2416 | btrfs_release_path(root, path); | |
2417 | return ret; | |
2418 | } | |
2419 | ||
39279cc3 CM |
2420 | /* |
2421 | * this can truncate away extent items, csum items and directory items. | |
2422 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 2423 | * any higher than new_size |
39279cc3 CM |
2424 | * |
2425 | * csum items that cross the new i_size are truncated to the new size | |
2426 | * as well. | |
7b128766 JB |
2427 | * |
2428 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
2429 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 2430 | */ |
e02119d5 CM |
2431 | noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
2432 | struct btrfs_root *root, | |
2433 | struct inode *inode, | |
2434 | u64 new_size, u32 min_type) | |
39279cc3 CM |
2435 | { |
2436 | int ret; | |
2437 | struct btrfs_path *path; | |
2438 | struct btrfs_key key; | |
5f39d397 | 2439 | struct btrfs_key found_key; |
39279cc3 | 2440 | u32 found_type; |
5f39d397 | 2441 | struct extent_buffer *leaf; |
39279cc3 CM |
2442 | struct btrfs_file_extent_item *fi; |
2443 | u64 extent_start = 0; | |
db94535d | 2444 | u64 extent_num_bytes = 0; |
39279cc3 | 2445 | u64 item_end = 0; |
7bb86316 | 2446 | u64 root_gen = 0; |
d8d5f3e1 | 2447 | u64 root_owner = 0; |
39279cc3 CM |
2448 | int found_extent; |
2449 | int del_item; | |
85e21bac CM |
2450 | int pending_del_nr = 0; |
2451 | int pending_del_slot = 0; | |
179e29e4 | 2452 | int extent_type = -1; |
771ed689 | 2453 | int encoding; |
3b951516 | 2454 | u64 mask = root->sectorsize - 1; |
39279cc3 | 2455 | |
e02119d5 | 2456 | if (root->ref_cows) |
5b21f2ed | 2457 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
39279cc3 | 2458 | path = btrfs_alloc_path(); |
3c69faec | 2459 | path->reada = -1; |
39279cc3 | 2460 | BUG_ON(!path); |
5f39d397 | 2461 | |
39279cc3 CM |
2462 | /* FIXME, add redo link to tree so we don't leak on crash */ |
2463 | key.objectid = inode->i_ino; | |
2464 | key.offset = (u64)-1; | |
5f39d397 CM |
2465 | key.type = (u8)-1; |
2466 | ||
85e21bac | 2467 | btrfs_init_path(path); |
323ac95b CM |
2468 | |
2469 | ret = drop_csum_leaves(trans, root, path, inode, new_size); | |
2470 | BUG_ON(ret); | |
2471 | ||
85e21bac CM |
2472 | search_again: |
2473 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2474 | if (ret < 0) { | |
2475 | goto error; | |
2476 | } | |
2477 | if (ret > 0) { | |
e02119d5 CM |
2478 | /* there are no items in the tree for us to truncate, we're |
2479 | * done | |
2480 | */ | |
2481 | if (path->slots[0] == 0) { | |
2482 | ret = 0; | |
2483 | goto error; | |
2484 | } | |
85e21bac CM |
2485 | path->slots[0]--; |
2486 | } | |
2487 | ||
39279cc3 | 2488 | while(1) { |
39279cc3 | 2489 | fi = NULL; |
5f39d397 CM |
2490 | leaf = path->nodes[0]; |
2491 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2492 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 2493 | encoding = 0; |
39279cc3 | 2494 | |
5f39d397 | 2495 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 2496 | break; |
5f39d397 | 2497 | |
85e21bac | 2498 | if (found_type < min_type) |
39279cc3 CM |
2499 | break; |
2500 | ||
5f39d397 | 2501 | item_end = found_key.offset; |
39279cc3 | 2502 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 2503 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 2504 | struct btrfs_file_extent_item); |
179e29e4 | 2505 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
2506 | encoding = btrfs_file_extent_compression(leaf, fi); |
2507 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
2508 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
2509 | ||
179e29e4 | 2510 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 2511 | item_end += |
db94535d | 2512 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 2513 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 2514 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 2515 | fi); |
39279cc3 | 2516 | } |
008630c1 | 2517 | item_end--; |
39279cc3 CM |
2518 | } |
2519 | if (found_type == BTRFS_CSUM_ITEM_KEY) { | |
2520 | ret = btrfs_csum_truncate(trans, root, path, | |
e02119d5 | 2521 | new_size); |
39279cc3 CM |
2522 | BUG_ON(ret); |
2523 | } | |
e02119d5 | 2524 | if (item_end < new_size) { |
b888db2b CM |
2525 | if (found_type == BTRFS_DIR_ITEM_KEY) { |
2526 | found_type = BTRFS_INODE_ITEM_KEY; | |
2527 | } else if (found_type == BTRFS_EXTENT_ITEM_KEY) { | |
2528 | found_type = BTRFS_CSUM_ITEM_KEY; | |
85e21bac CM |
2529 | } else if (found_type == BTRFS_EXTENT_DATA_KEY) { |
2530 | found_type = BTRFS_XATTR_ITEM_KEY; | |
2531 | } else if (found_type == BTRFS_XATTR_ITEM_KEY) { | |
2532 | found_type = BTRFS_INODE_REF_KEY; | |
b888db2b CM |
2533 | } else if (found_type) { |
2534 | found_type--; | |
2535 | } else { | |
2536 | break; | |
39279cc3 | 2537 | } |
a61721d5 | 2538 | btrfs_set_key_type(&key, found_type); |
85e21bac | 2539 | goto next; |
39279cc3 | 2540 | } |
e02119d5 | 2541 | if (found_key.offset >= new_size) |
39279cc3 CM |
2542 | del_item = 1; |
2543 | else | |
2544 | del_item = 0; | |
2545 | found_extent = 0; | |
2546 | ||
2547 | /* FIXME, shrink the extent if the ref count is only 1 */ | |
179e29e4 CM |
2548 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
2549 | goto delete; | |
2550 | ||
2551 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 2552 | u64 num_dec; |
db94535d | 2553 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 2554 | if (!del_item && !encoding) { |
db94535d CM |
2555 | u64 orig_num_bytes = |
2556 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 2557 | extent_num_bytes = new_size - |
5f39d397 | 2558 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
2559 | extent_num_bytes = extent_num_bytes & |
2560 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
2561 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2562 | extent_num_bytes); | |
2563 | num_dec = (orig_num_bytes - | |
9069218d | 2564 | extent_num_bytes); |
e02119d5 | 2565 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 2566 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 2567 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 2568 | } else { |
db94535d CM |
2569 | extent_num_bytes = |
2570 | btrfs_file_extent_disk_num_bytes(leaf, | |
2571 | fi); | |
39279cc3 | 2572 | /* FIXME blocksize != 4096 */ |
9069218d | 2573 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
2574 | if (extent_start != 0) { |
2575 | found_extent = 1; | |
e02119d5 | 2576 | if (root->ref_cows) |
a76a3cd4 | 2577 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 2578 | } |
31840ae1 | 2579 | root_gen = btrfs_header_generation(leaf); |
d8d5f3e1 | 2580 | root_owner = btrfs_header_owner(leaf); |
39279cc3 | 2581 | } |
9069218d | 2582 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
2583 | /* |
2584 | * we can't truncate inline items that have had | |
2585 | * special encodings | |
2586 | */ | |
2587 | if (!del_item && | |
2588 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
2589 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
2590 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
2591 | u32 size = new_size - found_key.offset; |
2592 | ||
2593 | if (root->ref_cows) { | |
a76a3cd4 YZ |
2594 | inode_sub_bytes(inode, item_end + 1 - |
2595 | new_size); | |
e02119d5 CM |
2596 | } |
2597 | size = | |
2598 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 2599 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 2600 | size, 1); |
9069218d | 2601 | BUG_ON(ret); |
e02119d5 | 2602 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
2603 | inode_sub_bytes(inode, item_end + 1 - |
2604 | found_key.offset); | |
9069218d | 2605 | } |
39279cc3 | 2606 | } |
179e29e4 | 2607 | delete: |
39279cc3 | 2608 | if (del_item) { |
85e21bac CM |
2609 | if (!pending_del_nr) { |
2610 | /* no pending yet, add ourselves */ | |
2611 | pending_del_slot = path->slots[0]; | |
2612 | pending_del_nr = 1; | |
2613 | } else if (pending_del_nr && | |
2614 | path->slots[0] + 1 == pending_del_slot) { | |
2615 | /* hop on the pending chunk */ | |
2616 | pending_del_nr++; | |
2617 | pending_del_slot = path->slots[0]; | |
2618 | } else { | |
2619 | printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot); | |
2620 | } | |
39279cc3 CM |
2621 | } else { |
2622 | break; | |
2623 | } | |
39279cc3 CM |
2624 | if (found_extent) { |
2625 | ret = btrfs_free_extent(trans, root, extent_start, | |
7bb86316 | 2626 | extent_num_bytes, |
31840ae1 | 2627 | leaf->start, root_owner, |
3bb1a1bc | 2628 | root_gen, inode->i_ino, 0); |
39279cc3 CM |
2629 | BUG_ON(ret); |
2630 | } | |
85e21bac CM |
2631 | next: |
2632 | if (path->slots[0] == 0) { | |
2633 | if (pending_del_nr) | |
2634 | goto del_pending; | |
2635 | btrfs_release_path(root, path); | |
2636 | goto search_again; | |
2637 | } | |
2638 | ||
2639 | path->slots[0]--; | |
2640 | if (pending_del_nr && | |
2641 | path->slots[0] + 1 != pending_del_slot) { | |
2642 | struct btrfs_key debug; | |
2643 | del_pending: | |
2644 | btrfs_item_key_to_cpu(path->nodes[0], &debug, | |
2645 | pending_del_slot); | |
2646 | ret = btrfs_del_items(trans, root, path, | |
2647 | pending_del_slot, | |
2648 | pending_del_nr); | |
2649 | BUG_ON(ret); | |
2650 | pending_del_nr = 0; | |
2651 | btrfs_release_path(root, path); | |
2652 | goto search_again; | |
2653 | } | |
39279cc3 CM |
2654 | } |
2655 | ret = 0; | |
2656 | error: | |
85e21bac CM |
2657 | if (pending_del_nr) { |
2658 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
2659 | pending_del_nr); | |
2660 | } | |
39279cc3 CM |
2661 | btrfs_free_path(path); |
2662 | inode->i_sb->s_dirt = 1; | |
2663 | return ret; | |
2664 | } | |
2665 | ||
2666 | /* | |
2667 | * taken from block_truncate_page, but does cow as it zeros out | |
2668 | * any bytes left in the last page in the file. | |
2669 | */ | |
2670 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
2671 | { | |
2672 | struct inode *inode = mapping->host; | |
db94535d | 2673 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
2674 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2675 | struct btrfs_ordered_extent *ordered; | |
2676 | char *kaddr; | |
db94535d | 2677 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
2678 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
2679 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
2680 | struct page *page; | |
39279cc3 | 2681 | int ret = 0; |
a52d9a80 | 2682 | u64 page_start; |
e6dcd2dc | 2683 | u64 page_end; |
39279cc3 CM |
2684 | |
2685 | if ((offset & (blocksize - 1)) == 0) | |
2686 | goto out; | |
2687 | ||
2688 | ret = -ENOMEM; | |
211c17f5 | 2689 | again: |
39279cc3 CM |
2690 | page = grab_cache_page(mapping, index); |
2691 | if (!page) | |
2692 | goto out; | |
e6dcd2dc CM |
2693 | |
2694 | page_start = page_offset(page); | |
2695 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2696 | ||
39279cc3 | 2697 | if (!PageUptodate(page)) { |
9ebefb18 | 2698 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 2699 | lock_page(page); |
211c17f5 CM |
2700 | if (page->mapping != mapping) { |
2701 | unlock_page(page); | |
2702 | page_cache_release(page); | |
2703 | goto again; | |
2704 | } | |
39279cc3 CM |
2705 | if (!PageUptodate(page)) { |
2706 | ret = -EIO; | |
89642229 | 2707 | goto out_unlock; |
39279cc3 CM |
2708 | } |
2709 | } | |
211c17f5 | 2710 | wait_on_page_writeback(page); |
e6dcd2dc CM |
2711 | |
2712 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2713 | set_page_extent_mapped(page); | |
2714 | ||
2715 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
2716 | if (ordered) { | |
2717 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2718 | unlock_page(page); | |
2719 | page_cache_release(page); | |
eb84ae03 | 2720 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
2721 | btrfs_put_ordered_extent(ordered); |
2722 | goto again; | |
2723 | } | |
2724 | ||
ea8c2819 | 2725 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc CM |
2726 | ret = 0; |
2727 | if (offset != PAGE_CACHE_SIZE) { | |
2728 | kaddr = kmap(page); | |
2729 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
2730 | flush_dcache_page(page); | |
2731 | kunmap(page); | |
2732 | } | |
247e743c | 2733 | ClearPageChecked(page); |
e6dcd2dc CM |
2734 | set_page_dirty(page); |
2735 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
39279cc3 | 2736 | |
89642229 | 2737 | out_unlock: |
39279cc3 CM |
2738 | unlock_page(page); |
2739 | page_cache_release(page); | |
2740 | out: | |
2741 | return ret; | |
2742 | } | |
2743 | ||
9036c102 | 2744 | int btrfs_cont_expand(struct inode *inode, loff_t size) |
39279cc3 | 2745 | { |
9036c102 YZ |
2746 | struct btrfs_trans_handle *trans; |
2747 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2748 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2749 | struct extent_map *em; | |
2750 | u64 mask = root->sectorsize - 1; | |
2751 | u64 hole_start = (inode->i_size + mask) & ~mask; | |
2752 | u64 block_end = (size + mask) & ~mask; | |
2753 | u64 last_byte; | |
2754 | u64 cur_offset; | |
2755 | u64 hole_size; | |
39279cc3 CM |
2756 | int err; |
2757 | ||
9036c102 YZ |
2758 | if (size <= hole_start) |
2759 | return 0; | |
2760 | ||
2761 | err = btrfs_check_free_space(root, 1, 0); | |
39279cc3 CM |
2762 | if (err) |
2763 | return err; | |
2764 | ||
9036c102 | 2765 | btrfs_truncate_page(inode->i_mapping, inode->i_size); |
2bf5a725 | 2766 | |
9036c102 YZ |
2767 | while (1) { |
2768 | struct btrfs_ordered_extent *ordered; | |
2769 | btrfs_wait_ordered_range(inode, hole_start, | |
2770 | block_end - hole_start); | |
2771 | lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2772 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); | |
2773 | if (!ordered) | |
2774 | break; | |
2775 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2776 | btrfs_put_ordered_extent(ordered); | |
2777 | } | |
39279cc3 | 2778 | |
9036c102 YZ |
2779 | trans = btrfs_start_transaction(root, 1); |
2780 | btrfs_set_trans_block_group(trans, inode); | |
39279cc3 | 2781 | |
9036c102 YZ |
2782 | cur_offset = hole_start; |
2783 | while (1) { | |
2784 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
2785 | block_end - cur_offset, 0); | |
2786 | BUG_ON(IS_ERR(em) || !em); | |
2787 | last_byte = min(extent_map_end(em), block_end); | |
2788 | last_byte = (last_byte + mask) & ~mask; | |
2789 | if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) { | |
771ed689 | 2790 | u64 hint_byte = 0; |
9036c102 | 2791 | hole_size = last_byte - cur_offset; |
771ed689 CM |
2792 | err = btrfs_drop_extents(trans, root, inode, |
2793 | cur_offset, | |
2794 | cur_offset + hole_size, | |
2795 | cur_offset, &hint_byte); | |
2796 | if (err) | |
2797 | break; | |
9036c102 YZ |
2798 | err = btrfs_insert_file_extent(trans, root, |
2799 | inode->i_ino, cur_offset, 0, | |
2800 | 0, hole_size, 0, hole_size, | |
2801 | 0, 0, 0); | |
2802 | btrfs_drop_extent_cache(inode, hole_start, | |
2803 | last_byte - 1, 0); | |
2804 | } | |
2805 | free_extent_map(em); | |
2806 | cur_offset = last_byte; | |
2807 | if (err || cur_offset >= block_end) | |
2808 | break; | |
2809 | } | |
1832a6d5 | 2810 | |
9036c102 YZ |
2811 | btrfs_end_transaction(trans, root); |
2812 | unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS); | |
2813 | return err; | |
2814 | } | |
39279cc3 | 2815 | |
9036c102 YZ |
2816 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
2817 | { | |
2818 | struct inode *inode = dentry->d_inode; | |
2819 | int err; | |
39279cc3 | 2820 | |
9036c102 YZ |
2821 | err = inode_change_ok(inode, attr); |
2822 | if (err) | |
2823 | return err; | |
2bf5a725 | 2824 | |
9036c102 YZ |
2825 | if (S_ISREG(inode->i_mode) && |
2826 | attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) { | |
2827 | err = btrfs_cont_expand(inode, attr->ia_size); | |
54aa1f4d CM |
2828 | if (err) |
2829 | return err; | |
39279cc3 | 2830 | } |
9036c102 | 2831 | |
39279cc3 | 2832 | err = inode_setattr(inode, attr); |
33268eaf JB |
2833 | |
2834 | if (!err && ((attr->ia_valid & ATTR_MODE))) | |
2835 | err = btrfs_acl_chmod(inode); | |
39279cc3 CM |
2836 | return err; |
2837 | } | |
61295eb8 | 2838 | |
39279cc3 CM |
2839 | void btrfs_delete_inode(struct inode *inode) |
2840 | { | |
2841 | struct btrfs_trans_handle *trans; | |
2842 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 2843 | unsigned long nr; |
39279cc3 CM |
2844 | int ret; |
2845 | ||
2846 | truncate_inode_pages(&inode->i_data, 0); | |
2847 | if (is_bad_inode(inode)) { | |
7b128766 | 2848 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
2849 | goto no_delete; |
2850 | } | |
4a096752 | 2851 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 2852 | |
dbe674a9 | 2853 | btrfs_i_size_write(inode, 0); |
39279cc3 | 2854 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 2855 | |
39279cc3 | 2856 | btrfs_set_trans_block_group(trans, inode); |
e02119d5 | 2857 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0); |
7b128766 JB |
2858 | if (ret) { |
2859 | btrfs_orphan_del(NULL, inode); | |
54aa1f4d | 2860 | goto no_delete_lock; |
7b128766 JB |
2861 | } |
2862 | ||
2863 | btrfs_orphan_del(trans, inode); | |
85e21bac | 2864 | |
d3c2fdcf | 2865 | nr = trans->blocks_used; |
85e21bac | 2866 | clear_inode(inode); |
5f39d397 | 2867 | |
39279cc3 | 2868 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2869 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 2870 | return; |
54aa1f4d CM |
2871 | |
2872 | no_delete_lock: | |
d3c2fdcf | 2873 | nr = trans->blocks_used; |
54aa1f4d | 2874 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 2875 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2876 | no_delete: |
2877 | clear_inode(inode); | |
2878 | } | |
2879 | ||
2880 | /* | |
2881 | * this returns the key found in the dir entry in the location pointer. | |
2882 | * If no dir entries were found, location->objectid is 0. | |
2883 | */ | |
2884 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
2885 | struct btrfs_key *location) | |
2886 | { | |
2887 | const char *name = dentry->d_name.name; | |
2888 | int namelen = dentry->d_name.len; | |
2889 | struct btrfs_dir_item *di; | |
2890 | struct btrfs_path *path; | |
2891 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 2892 | int ret = 0; |
39279cc3 CM |
2893 | |
2894 | path = btrfs_alloc_path(); | |
2895 | BUG_ON(!path); | |
3954401f | 2896 | |
39279cc3 CM |
2897 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
2898 | namelen, 0); | |
0d9f7f3e Y |
2899 | if (IS_ERR(di)) |
2900 | ret = PTR_ERR(di); | |
39279cc3 | 2901 | if (!di || IS_ERR(di)) { |
3954401f | 2902 | goto out_err; |
39279cc3 | 2903 | } |
5f39d397 | 2904 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 2905 | out: |
39279cc3 CM |
2906 | btrfs_free_path(path); |
2907 | return ret; | |
3954401f CM |
2908 | out_err: |
2909 | location->objectid = 0; | |
2910 | goto out; | |
39279cc3 CM |
2911 | } |
2912 | ||
2913 | /* | |
2914 | * when we hit a tree root in a directory, the btrfs part of the inode | |
2915 | * needs to be changed to reflect the root directory of the tree root. This | |
2916 | * is kind of like crossing a mount point. | |
2917 | */ | |
2918 | static int fixup_tree_root_location(struct btrfs_root *root, | |
2919 | struct btrfs_key *location, | |
58176a96 JB |
2920 | struct btrfs_root **sub_root, |
2921 | struct dentry *dentry) | |
39279cc3 | 2922 | { |
39279cc3 CM |
2923 | struct btrfs_root_item *ri; |
2924 | ||
2925 | if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY) | |
2926 | return 0; | |
2927 | if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) | |
2928 | return 0; | |
2929 | ||
58176a96 JB |
2930 | *sub_root = btrfs_read_fs_root(root->fs_info, location, |
2931 | dentry->d_name.name, | |
2932 | dentry->d_name.len); | |
39279cc3 CM |
2933 | if (IS_ERR(*sub_root)) |
2934 | return PTR_ERR(*sub_root); | |
2935 | ||
2936 | ri = &(*sub_root)->root_item; | |
2937 | location->objectid = btrfs_root_dirid(ri); | |
39279cc3 CM |
2938 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); |
2939 | location->offset = 0; | |
2940 | ||
39279cc3 CM |
2941 | return 0; |
2942 | } | |
2943 | ||
e02119d5 | 2944 | static noinline void init_btrfs_i(struct inode *inode) |
39279cc3 | 2945 | { |
e02119d5 CM |
2946 | struct btrfs_inode *bi = BTRFS_I(inode); |
2947 | ||
2948 | bi->i_acl = NULL; | |
2949 | bi->i_default_acl = NULL; | |
2950 | ||
2951 | bi->generation = 0; | |
2952 | bi->last_trans = 0; | |
2953 | bi->logged_trans = 0; | |
2954 | bi->delalloc_bytes = 0; | |
2955 | bi->disk_i_size = 0; | |
2956 | bi->flags = 0; | |
2957 | bi->index_cnt = (u64)-1; | |
49eb7e46 | 2958 | bi->log_dirty_trans = 0; |
d1310b2e CM |
2959 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); |
2960 | extent_io_tree_init(&BTRFS_I(inode)->io_tree, | |
b888db2b | 2961 | inode->i_mapping, GFP_NOFS); |
7e38326f CM |
2962 | extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, |
2963 | inode->i_mapping, GFP_NOFS); | |
ea8c2819 | 2964 | INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); |
ba1da2f4 | 2965 | btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); |
1b1e2135 | 2966 | mutex_init(&BTRFS_I(inode)->csum_mutex); |
ee6e6504 | 2967 | mutex_init(&BTRFS_I(inode)->extent_mutex); |
e02119d5 CM |
2968 | mutex_init(&BTRFS_I(inode)->log_mutex); |
2969 | } | |
2970 | ||
2971 | static int btrfs_init_locked_inode(struct inode *inode, void *p) | |
2972 | { | |
2973 | struct btrfs_iget_args *args = p; | |
2974 | inode->i_ino = args->ino; | |
2975 | init_btrfs_i(inode); | |
2976 | BTRFS_I(inode)->root = args->root; | |
39279cc3 CM |
2977 | return 0; |
2978 | } | |
2979 | ||
2980 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
2981 | { | |
2982 | struct btrfs_iget_args *args = opaque; | |
2983 | return (args->ino == inode->i_ino && | |
2984 | args->root == BTRFS_I(inode)->root); | |
2985 | } | |
2986 | ||
5b21f2ed ZY |
2987 | struct inode *btrfs_ilookup(struct super_block *s, u64 objectid, |
2988 | struct btrfs_root *root, int wait) | |
2989 | { | |
2990 | struct inode *inode; | |
2991 | struct btrfs_iget_args args; | |
2992 | args.ino = objectid; | |
2993 | args.root = root; | |
2994 | ||
2995 | if (wait) { | |
2996 | inode = ilookup5(s, objectid, btrfs_find_actor, | |
2997 | (void *)&args); | |
2998 | } else { | |
2999 | inode = ilookup5_nowait(s, objectid, btrfs_find_actor, | |
3000 | (void *)&args); | |
3001 | } | |
3002 | return inode; | |
3003 | } | |
3004 | ||
39279cc3 CM |
3005 | struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid, |
3006 | struct btrfs_root *root) | |
3007 | { | |
3008 | struct inode *inode; | |
3009 | struct btrfs_iget_args args; | |
3010 | args.ino = objectid; | |
3011 | args.root = root; | |
3012 | ||
3013 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3014 | btrfs_init_locked_inode, | |
3015 | (void *)&args); | |
3016 | return inode; | |
3017 | } | |
3018 | ||
1a54ef8c BR |
3019 | /* Get an inode object given its location and corresponding root. |
3020 | * Returns in *is_new if the inode was read from disk | |
3021 | */ | |
3022 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
3023 | struct btrfs_root *root, int *is_new) | |
3024 | { | |
3025 | struct inode *inode; | |
3026 | ||
3027 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3028 | if (!inode) | |
3029 | return ERR_PTR(-EACCES); | |
3030 | ||
3031 | if (inode->i_state & I_NEW) { | |
3032 | BTRFS_I(inode)->root = root; | |
3033 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3034 | btrfs_read_locked_inode(inode); | |
3035 | unlock_new_inode(inode); | |
3036 | if (is_new) | |
3037 | *is_new = 1; | |
3038 | } else { | |
3039 | if (is_new) | |
3040 | *is_new = 0; | |
3041 | } | |
3042 | ||
3043 | return inode; | |
3044 | } | |
3045 | ||
3de4586c | 3046 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 CM |
3047 | { |
3048 | struct inode * inode; | |
3049 | struct btrfs_inode *bi = BTRFS_I(dir); | |
3050 | struct btrfs_root *root = bi->root; | |
3051 | struct btrfs_root *sub_root = root; | |
3052 | struct btrfs_key location; | |
c146afad | 3053 | int ret, new; |
39279cc3 CM |
3054 | |
3055 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3056 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3057 | |
39279cc3 | 3058 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 3059 | |
39279cc3 CM |
3060 | if (ret < 0) |
3061 | return ERR_PTR(ret); | |
5f39d397 | 3062 | |
39279cc3 CM |
3063 | inode = NULL; |
3064 | if (location.objectid) { | |
58176a96 JB |
3065 | ret = fixup_tree_root_location(root, &location, &sub_root, |
3066 | dentry); | |
39279cc3 CM |
3067 | if (ret < 0) |
3068 | return ERR_PTR(ret); | |
3069 | if (ret > 0) | |
3070 | return ERR_PTR(-ENOENT); | |
1a54ef8c BR |
3071 | inode = btrfs_iget(dir->i_sb, &location, sub_root, &new); |
3072 | if (IS_ERR(inode)) | |
3073 | return ERR_CAST(inode); | |
39279cc3 | 3074 | } |
3de4586c CM |
3075 | return inode; |
3076 | } | |
3077 | ||
3078 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, | |
3079 | struct nameidata *nd) | |
3080 | { | |
3081 | struct inode *inode; | |
3082 | ||
3083 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3084 | return ERR_PTR(-ENAMETOOLONG); | |
3085 | ||
3086 | inode = btrfs_lookup_dentry(dir, dentry); | |
3087 | if (IS_ERR(inode)) | |
3088 | return ERR_CAST(inode); | |
7b128766 | 3089 | |
39279cc3 CM |
3090 | return d_splice_alias(inode, dentry); |
3091 | } | |
3092 | ||
39279cc3 CM |
3093 | static unsigned char btrfs_filetype_table[] = { |
3094 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
3095 | }; | |
3096 | ||
cbdf5a24 DW |
3097 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
3098 | filldir_t filldir) | |
39279cc3 | 3099 | { |
6da6abae | 3100 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
3101 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3102 | struct btrfs_item *item; | |
3103 | struct btrfs_dir_item *di; | |
3104 | struct btrfs_key key; | |
5f39d397 | 3105 | struct btrfs_key found_key; |
39279cc3 CM |
3106 | struct btrfs_path *path; |
3107 | int ret; | |
3108 | u32 nritems; | |
5f39d397 | 3109 | struct extent_buffer *leaf; |
39279cc3 CM |
3110 | int slot; |
3111 | int advance; | |
3112 | unsigned char d_type; | |
3113 | int over = 0; | |
3114 | u32 di_cur; | |
3115 | u32 di_total; | |
3116 | u32 di_len; | |
3117 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
3118 | char tmp_name[32]; |
3119 | char *name_ptr; | |
3120 | int name_len; | |
39279cc3 CM |
3121 | |
3122 | /* FIXME, use a real flag for deciding about the key type */ | |
3123 | if (root->fs_info->tree_root == root) | |
3124 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 3125 | |
3954401f CM |
3126 | /* special case for "." */ |
3127 | if (filp->f_pos == 0) { | |
3128 | over = filldir(dirent, ".", 1, | |
3129 | 1, inode->i_ino, | |
3130 | DT_DIR); | |
3131 | if (over) | |
3132 | return 0; | |
3133 | filp->f_pos = 1; | |
3134 | } | |
3954401f CM |
3135 | /* special case for .., just use the back ref */ |
3136 | if (filp->f_pos == 1) { | |
5ecc7e5d | 3137 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 3138 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 3139 | 2, pino, DT_DIR); |
3954401f | 3140 | if (over) |
49593bfa | 3141 | return 0; |
3954401f CM |
3142 | filp->f_pos = 2; |
3143 | } | |
49593bfa DW |
3144 | path = btrfs_alloc_path(); |
3145 | path->reada = 2; | |
3146 | ||
39279cc3 CM |
3147 | btrfs_set_key_type(&key, key_type); |
3148 | key.offset = filp->f_pos; | |
49593bfa | 3149 | key.objectid = inode->i_ino; |
5f39d397 | 3150 | |
39279cc3 CM |
3151 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3152 | if (ret < 0) | |
3153 | goto err; | |
3154 | advance = 0; | |
49593bfa DW |
3155 | |
3156 | while (1) { | |
5f39d397 CM |
3157 | leaf = path->nodes[0]; |
3158 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3159 | slot = path->slots[0]; |
3160 | if (advance || slot >= nritems) { | |
49593bfa | 3161 | if (slot >= nritems - 1) { |
39279cc3 CM |
3162 | ret = btrfs_next_leaf(root, path); |
3163 | if (ret) | |
3164 | break; | |
5f39d397 CM |
3165 | leaf = path->nodes[0]; |
3166 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
3167 | slot = path->slots[0]; |
3168 | } else { | |
3169 | slot++; | |
3170 | path->slots[0]++; | |
3171 | } | |
3172 | } | |
3de4586c | 3173 | |
39279cc3 | 3174 | advance = 1; |
5f39d397 CM |
3175 | item = btrfs_item_nr(leaf, slot); |
3176 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3177 | ||
3178 | if (found_key.objectid != key.objectid) | |
39279cc3 | 3179 | break; |
5f39d397 | 3180 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 3181 | break; |
5f39d397 | 3182 | if (found_key.offset < filp->f_pos) |
39279cc3 | 3183 | continue; |
5f39d397 CM |
3184 | |
3185 | filp->f_pos = found_key.offset; | |
49593bfa | 3186 | |
39279cc3 CM |
3187 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
3188 | di_cur = 0; | |
5f39d397 | 3189 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
3190 | |
3191 | while (di_cur < di_total) { | |
5f39d397 CM |
3192 | struct btrfs_key location; |
3193 | ||
3194 | name_len = btrfs_dir_name_len(leaf, di); | |
49593bfa | 3195 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
3196 | name_ptr = tmp_name; |
3197 | } else { | |
3198 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
3199 | if (!name_ptr) { |
3200 | ret = -ENOMEM; | |
3201 | goto err; | |
3202 | } | |
5f39d397 CM |
3203 | } |
3204 | read_extent_buffer(leaf, name_ptr, | |
3205 | (unsigned long)(di + 1), name_len); | |
3206 | ||
3207 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
3208 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
3209 | |
3210 | /* is this a reference to our own snapshot? If so | |
3211 | * skip it | |
3212 | */ | |
3213 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
3214 | location.objectid == root->root_key.objectid) { | |
3215 | over = 0; | |
3216 | goto skip; | |
3217 | } | |
5f39d397 | 3218 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 3219 | found_key.offset, location.objectid, |
39279cc3 | 3220 | d_type); |
5f39d397 | 3221 | |
3de4586c | 3222 | skip: |
5f39d397 CM |
3223 | if (name_ptr != tmp_name) |
3224 | kfree(name_ptr); | |
3225 | ||
39279cc3 CM |
3226 | if (over) |
3227 | goto nopos; | |
5103e947 | 3228 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 3229 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
3230 | di_cur += di_len; |
3231 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
3232 | } | |
3233 | } | |
49593bfa DW |
3234 | |
3235 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 YZ |
3236 | if (key_type == BTRFS_DIR_INDEX_KEY) |
3237 | filp->f_pos = INT_LIMIT(typeof(filp->f_pos)); | |
3238 | else | |
3239 | filp->f_pos++; | |
39279cc3 CM |
3240 | nopos: |
3241 | ret = 0; | |
3242 | err: | |
39279cc3 | 3243 | btrfs_free_path(path); |
39279cc3 CM |
3244 | return ret; |
3245 | } | |
3246 | ||
3247 | int btrfs_write_inode(struct inode *inode, int wait) | |
3248 | { | |
3249 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3250 | struct btrfs_trans_handle *trans; | |
3251 | int ret = 0; | |
3252 | ||
c146afad | 3253 | if (root->fs_info->btree_inode == inode) |
4ca8b41e CM |
3254 | return 0; |
3255 | ||
39279cc3 | 3256 | if (wait) { |
f9295749 | 3257 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3258 | btrfs_set_trans_block_group(trans, inode); |
3259 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
3260 | } |
3261 | return ret; | |
3262 | } | |
3263 | ||
3264 | /* | |
54aa1f4d | 3265 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
3266 | * inode changes. But, it is most likely to find the inode in cache. |
3267 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
3268 | * to keep or drop this code. | |
3269 | */ | |
3270 | void btrfs_dirty_inode(struct inode *inode) | |
3271 | { | |
3272 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3273 | struct btrfs_trans_handle *trans; | |
3274 | ||
f9295749 | 3275 | trans = btrfs_join_transaction(root, 1); |
39279cc3 CM |
3276 | btrfs_set_trans_block_group(trans, inode); |
3277 | btrfs_update_inode(trans, root, inode); | |
3278 | btrfs_end_transaction(trans, root); | |
39279cc3 CM |
3279 | } |
3280 | ||
d352ac68 CM |
3281 | /* |
3282 | * find the highest existing sequence number in a directory | |
3283 | * and then set the in-memory index_cnt variable to reflect | |
3284 | * free sequence numbers | |
3285 | */ | |
aec7477b JB |
3286 | static int btrfs_set_inode_index_count(struct inode *inode) |
3287 | { | |
3288 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3289 | struct btrfs_key key, found_key; | |
3290 | struct btrfs_path *path; | |
3291 | struct extent_buffer *leaf; | |
3292 | int ret; | |
3293 | ||
3294 | key.objectid = inode->i_ino; | |
3295 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
3296 | key.offset = (u64)-1; | |
3297 | ||
3298 | path = btrfs_alloc_path(); | |
3299 | if (!path) | |
3300 | return -ENOMEM; | |
3301 | ||
3302 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3303 | if (ret < 0) | |
3304 | goto out; | |
3305 | /* FIXME: we should be able to handle this */ | |
3306 | if (ret == 0) | |
3307 | goto out; | |
3308 | ret = 0; | |
3309 | ||
3310 | /* | |
3311 | * MAGIC NUMBER EXPLANATION: | |
3312 | * since we search a directory based on f_pos we have to start at 2 | |
3313 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
3314 | * else has to start at 2 | |
3315 | */ | |
3316 | if (path->slots[0] == 0) { | |
3317 | BTRFS_I(inode)->index_cnt = 2; | |
3318 | goto out; | |
3319 | } | |
3320 | ||
3321 | path->slots[0]--; | |
3322 | ||
3323 | leaf = path->nodes[0]; | |
3324 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3325 | ||
3326 | if (found_key.objectid != inode->i_ino || | |
3327 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
3328 | BTRFS_I(inode)->index_cnt = 2; | |
3329 | goto out; | |
3330 | } | |
3331 | ||
3332 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
3333 | out: | |
3334 | btrfs_free_path(path); | |
3335 | return ret; | |
3336 | } | |
3337 | ||
d352ac68 CM |
3338 | /* |
3339 | * helper to find a free sequence number in a given directory. This current | |
3340 | * code is very simple, later versions will do smarter things in the btree | |
3341 | */ | |
3de4586c | 3342 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
3343 | { |
3344 | int ret = 0; | |
3345 | ||
3346 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
3347 | ret = btrfs_set_inode_index_count(dir); | |
8d5bf1cb | 3348 | if (ret) { |
aec7477b | 3349 | return ret; |
8d5bf1cb | 3350 | } |
aec7477b JB |
3351 | } |
3352 | ||
00e4e6b3 | 3353 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
3354 | BTRFS_I(dir)->index_cnt++; |
3355 | ||
3356 | return ret; | |
3357 | } | |
3358 | ||
39279cc3 CM |
3359 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
3360 | struct btrfs_root *root, | |
aec7477b | 3361 | struct inode *dir, |
9c58309d CM |
3362 | const char *name, int name_len, |
3363 | u64 ref_objectid, | |
39279cc3 CM |
3364 | u64 objectid, |
3365 | struct btrfs_block_group_cache *group, | |
00e4e6b3 | 3366 | int mode, u64 *index) |
39279cc3 CM |
3367 | { |
3368 | struct inode *inode; | |
5f39d397 | 3369 | struct btrfs_inode_item *inode_item; |
6324fbf3 | 3370 | struct btrfs_block_group_cache *new_inode_group; |
39279cc3 | 3371 | struct btrfs_key *location; |
5f39d397 | 3372 | struct btrfs_path *path; |
9c58309d CM |
3373 | struct btrfs_inode_ref *ref; |
3374 | struct btrfs_key key[2]; | |
3375 | u32 sizes[2]; | |
3376 | unsigned long ptr; | |
39279cc3 CM |
3377 | int ret; |
3378 | int owner; | |
3379 | ||
5f39d397 CM |
3380 | path = btrfs_alloc_path(); |
3381 | BUG_ON(!path); | |
3382 | ||
39279cc3 CM |
3383 | inode = new_inode(root->fs_info->sb); |
3384 | if (!inode) | |
3385 | return ERR_PTR(-ENOMEM); | |
3386 | ||
aec7477b | 3387 | if (dir) { |
3de4586c | 3388 | ret = btrfs_set_inode_index(dir, index); |
aec7477b JB |
3389 | if (ret) |
3390 | return ERR_PTR(ret); | |
aec7477b JB |
3391 | } |
3392 | /* | |
3393 | * index_cnt is ignored for everything but a dir, | |
3394 | * btrfs_get_inode_index_count has an explanation for the magic | |
3395 | * number | |
3396 | */ | |
e02119d5 | 3397 | init_btrfs_i(inode); |
aec7477b | 3398 | BTRFS_I(inode)->index_cnt = 2; |
39279cc3 | 3399 | BTRFS_I(inode)->root = root; |
e02119d5 | 3400 | BTRFS_I(inode)->generation = trans->transid; |
b888db2b | 3401 | |
39279cc3 CM |
3402 | if (mode & S_IFDIR) |
3403 | owner = 0; | |
3404 | else | |
3405 | owner = 1; | |
6324fbf3 | 3406 | new_inode_group = btrfs_find_block_group(root, group, 0, |
0b86a832 | 3407 | BTRFS_BLOCK_GROUP_METADATA, owner); |
6324fbf3 CM |
3408 | if (!new_inode_group) { |
3409 | printk("find_block group failed\n"); | |
3410 | new_inode_group = group; | |
3411 | } | |
3412 | BTRFS_I(inode)->block_group = new_inode_group; | |
9c58309d CM |
3413 | |
3414 | key[0].objectid = objectid; | |
3415 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
3416 | key[0].offset = 0; | |
3417 | ||
3418 | key[1].objectid = objectid; | |
3419 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
3420 | key[1].offset = ref_objectid; | |
3421 | ||
3422 | sizes[0] = sizeof(struct btrfs_inode_item); | |
3423 | sizes[1] = name_len + sizeof(*ref); | |
3424 | ||
3425 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); | |
3426 | if (ret != 0) | |
5f39d397 CM |
3427 | goto fail; |
3428 | ||
9c58309d CM |
3429 | if (objectid > root->highest_inode) |
3430 | root->highest_inode = objectid; | |
3431 | ||
39279cc3 CM |
3432 | inode->i_uid = current->fsuid; |
3433 | inode->i_gid = current->fsgid; | |
3434 | inode->i_mode = mode; | |
3435 | inode->i_ino = objectid; | |
a76a3cd4 | 3436 | inode_set_bytes(inode, 0); |
39279cc3 | 3437 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
3438 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3439 | struct btrfs_inode_item); | |
e02119d5 | 3440 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
3441 | |
3442 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3443 | struct btrfs_inode_ref); | |
3444 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 3445 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
3446 | ptr = (unsigned long)(ref + 1); |
3447 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
3448 | ||
5f39d397 CM |
3449 | btrfs_mark_buffer_dirty(path->nodes[0]); |
3450 | btrfs_free_path(path); | |
3451 | ||
39279cc3 CM |
3452 | location = &BTRFS_I(inode)->location; |
3453 | location->objectid = objectid; | |
39279cc3 CM |
3454 | location->offset = 0; |
3455 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
3456 | ||
39279cc3 CM |
3457 | insert_inode_hash(inode); |
3458 | return inode; | |
5f39d397 | 3459 | fail: |
aec7477b JB |
3460 | if (dir) |
3461 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 CM |
3462 | btrfs_free_path(path); |
3463 | return ERR_PTR(ret); | |
39279cc3 CM |
3464 | } |
3465 | ||
3466 | static inline u8 btrfs_inode_type(struct inode *inode) | |
3467 | { | |
3468 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
3469 | } | |
3470 | ||
d352ac68 CM |
3471 | /* |
3472 | * utility function to add 'inode' into 'parent_inode' with | |
3473 | * a give name and a given sequence number. | |
3474 | * if 'add_backref' is true, also insert a backref from the | |
3475 | * inode to the parent directory. | |
3476 | */ | |
e02119d5 CM |
3477 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
3478 | struct inode *parent_inode, struct inode *inode, | |
3479 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 CM |
3480 | { |
3481 | int ret; | |
3482 | struct btrfs_key key; | |
e02119d5 | 3483 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 3484 | |
39279cc3 | 3485 | key.objectid = inode->i_ino; |
39279cc3 CM |
3486 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
3487 | key.offset = 0; | |
3488 | ||
e02119d5 CM |
3489 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
3490 | parent_inode->i_ino, | |
aec7477b | 3491 | &key, btrfs_inode_type(inode), |
00e4e6b3 | 3492 | index); |
39279cc3 | 3493 | if (ret == 0) { |
9c58309d CM |
3494 | if (add_backref) { |
3495 | ret = btrfs_insert_inode_ref(trans, root, | |
e02119d5 CM |
3496 | name, name_len, |
3497 | inode->i_ino, | |
3498 | parent_inode->i_ino, | |
3499 | index); | |
9c58309d | 3500 | } |
dbe674a9 | 3501 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 3502 | name_len * 2); |
79c44584 | 3503 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 3504 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
3505 | } |
3506 | return ret; | |
3507 | } | |
3508 | ||
3509 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
9c58309d | 3510 | struct dentry *dentry, struct inode *inode, |
00e4e6b3 | 3511 | int backref, u64 index) |
39279cc3 | 3512 | { |
e02119d5 CM |
3513 | int err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3514 | inode, dentry->d_name.name, | |
3515 | dentry->d_name.len, backref, index); | |
39279cc3 CM |
3516 | if (!err) { |
3517 | d_instantiate(dentry, inode); | |
3518 | return 0; | |
3519 | } | |
3520 | if (err > 0) | |
3521 | err = -EEXIST; | |
3522 | return err; | |
3523 | } | |
3524 | ||
618e21d5 JB |
3525 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
3526 | int mode, dev_t rdev) | |
3527 | { | |
3528 | struct btrfs_trans_handle *trans; | |
3529 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3530 | struct inode *inode = NULL; |
618e21d5 JB |
3531 | int err; |
3532 | int drop_inode = 0; | |
3533 | u64 objectid; | |
1832a6d5 | 3534 | unsigned long nr = 0; |
00e4e6b3 | 3535 | u64 index = 0; |
618e21d5 JB |
3536 | |
3537 | if (!new_valid_dev(rdev)) | |
3538 | return -EINVAL; | |
3539 | ||
1832a6d5 CM |
3540 | err = btrfs_check_free_space(root, 1, 0); |
3541 | if (err) | |
3542 | goto fail; | |
3543 | ||
618e21d5 JB |
3544 | trans = btrfs_start_transaction(root, 1); |
3545 | btrfs_set_trans_block_group(trans, dir); | |
3546 | ||
3547 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3548 | if (err) { | |
3549 | err = -ENOSPC; | |
3550 | goto out_unlock; | |
3551 | } | |
3552 | ||
aec7477b | 3553 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3554 | dentry->d_name.len, |
3555 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 | 3556 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
3557 | err = PTR_ERR(inode); |
3558 | if (IS_ERR(inode)) | |
3559 | goto out_unlock; | |
3560 | ||
33268eaf JB |
3561 | err = btrfs_init_acl(inode, dir); |
3562 | if (err) { | |
3563 | drop_inode = 1; | |
3564 | goto out_unlock; | |
3565 | } | |
3566 | ||
618e21d5 | 3567 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3568 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
618e21d5 JB |
3569 | if (err) |
3570 | drop_inode = 1; | |
3571 | else { | |
3572 | inode->i_op = &btrfs_special_inode_operations; | |
3573 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 3574 | btrfs_update_inode(trans, root, inode); |
618e21d5 JB |
3575 | } |
3576 | dir->i_sb->s_dirt = 1; | |
3577 | btrfs_update_inode_block_group(trans, inode); | |
3578 | btrfs_update_inode_block_group(trans, dir); | |
3579 | out_unlock: | |
d3c2fdcf | 3580 | nr = trans->blocks_used; |
89ce8a63 | 3581 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3582 | fail: |
618e21d5 JB |
3583 | if (drop_inode) { |
3584 | inode_dec_link_count(inode); | |
3585 | iput(inode); | |
3586 | } | |
d3c2fdcf | 3587 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
3588 | return err; |
3589 | } | |
3590 | ||
39279cc3 CM |
3591 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
3592 | int mode, struct nameidata *nd) | |
3593 | { | |
3594 | struct btrfs_trans_handle *trans; | |
3595 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 3596 | struct inode *inode = NULL; |
39279cc3 CM |
3597 | int err; |
3598 | int drop_inode = 0; | |
1832a6d5 | 3599 | unsigned long nr = 0; |
39279cc3 | 3600 | u64 objectid; |
00e4e6b3 | 3601 | u64 index = 0; |
39279cc3 | 3602 | |
1832a6d5 CM |
3603 | err = btrfs_check_free_space(root, 1, 0); |
3604 | if (err) | |
3605 | goto fail; | |
39279cc3 CM |
3606 | trans = btrfs_start_transaction(root, 1); |
3607 | btrfs_set_trans_block_group(trans, dir); | |
3608 | ||
3609 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3610 | if (err) { | |
3611 | err = -ENOSPC; | |
3612 | goto out_unlock; | |
3613 | } | |
3614 | ||
aec7477b | 3615 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3616 | dentry->d_name.len, |
3617 | dentry->d_parent->d_inode->i_ino, | |
00e4e6b3 CM |
3618 | objectid, BTRFS_I(dir)->block_group, mode, |
3619 | &index); | |
39279cc3 CM |
3620 | err = PTR_ERR(inode); |
3621 | if (IS_ERR(inode)) | |
3622 | goto out_unlock; | |
3623 | ||
33268eaf JB |
3624 | err = btrfs_init_acl(inode, dir); |
3625 | if (err) { | |
3626 | drop_inode = 1; | |
3627 | goto out_unlock; | |
3628 | } | |
3629 | ||
39279cc3 | 3630 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 3631 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
3632 | if (err) |
3633 | drop_inode = 1; | |
3634 | else { | |
3635 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 3636 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
3637 | inode->i_fop = &btrfs_file_operations; |
3638 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 3639 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3640 | } |
3641 | dir->i_sb->s_dirt = 1; | |
3642 | btrfs_update_inode_block_group(trans, inode); | |
3643 | btrfs_update_inode_block_group(trans, dir); | |
3644 | out_unlock: | |
d3c2fdcf | 3645 | nr = trans->blocks_used; |
ab78c84d | 3646 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3647 | fail: |
39279cc3 CM |
3648 | if (drop_inode) { |
3649 | inode_dec_link_count(inode); | |
3650 | iput(inode); | |
3651 | } | |
d3c2fdcf | 3652 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3653 | return err; |
3654 | } | |
3655 | ||
3656 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
3657 | struct dentry *dentry) | |
3658 | { | |
3659 | struct btrfs_trans_handle *trans; | |
3660 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3661 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 3662 | u64 index; |
1832a6d5 | 3663 | unsigned long nr = 0; |
39279cc3 CM |
3664 | int err; |
3665 | int drop_inode = 0; | |
3666 | ||
3667 | if (inode->i_nlink == 0) | |
3668 | return -ENOENT; | |
3669 | ||
e02119d5 | 3670 | btrfs_inc_nlink(inode); |
1832a6d5 CM |
3671 | err = btrfs_check_free_space(root, 1, 0); |
3672 | if (err) | |
3673 | goto fail; | |
3de4586c | 3674 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
3675 | if (err) |
3676 | goto fail; | |
3677 | ||
39279cc3 | 3678 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 3679 | |
39279cc3 CM |
3680 | btrfs_set_trans_block_group(trans, dir); |
3681 | atomic_inc(&inode->i_count); | |
aec7477b | 3682 | |
00e4e6b3 | 3683 | err = btrfs_add_nondir(trans, dentry, inode, 1, index); |
5f39d397 | 3684 | |
39279cc3 CM |
3685 | if (err) |
3686 | drop_inode = 1; | |
5f39d397 | 3687 | |
39279cc3 CM |
3688 | dir->i_sb->s_dirt = 1; |
3689 | btrfs_update_inode_block_group(trans, dir); | |
54aa1f4d | 3690 | err = btrfs_update_inode(trans, root, inode); |
5f39d397 | 3691 | |
54aa1f4d CM |
3692 | if (err) |
3693 | drop_inode = 1; | |
39279cc3 | 3694 | |
d3c2fdcf | 3695 | nr = trans->blocks_used; |
ab78c84d | 3696 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 3697 | fail: |
39279cc3 CM |
3698 | if (drop_inode) { |
3699 | inode_dec_link_count(inode); | |
3700 | iput(inode); | |
3701 | } | |
d3c2fdcf | 3702 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3703 | return err; |
3704 | } | |
3705 | ||
39279cc3 CM |
3706 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
3707 | { | |
b9d86667 | 3708 | struct inode *inode = NULL; |
39279cc3 CM |
3709 | struct btrfs_trans_handle *trans; |
3710 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3711 | int err = 0; | |
3712 | int drop_on_err = 0; | |
b9d86667 | 3713 | u64 objectid = 0; |
00e4e6b3 | 3714 | u64 index = 0; |
d3c2fdcf | 3715 | unsigned long nr = 1; |
39279cc3 | 3716 | |
1832a6d5 CM |
3717 | err = btrfs_check_free_space(root, 1, 0); |
3718 | if (err) | |
3719 | goto out_unlock; | |
3720 | ||
39279cc3 CM |
3721 | trans = btrfs_start_transaction(root, 1); |
3722 | btrfs_set_trans_block_group(trans, dir); | |
5f39d397 | 3723 | |
39279cc3 CM |
3724 | if (IS_ERR(trans)) { |
3725 | err = PTR_ERR(trans); | |
3726 | goto out_unlock; | |
3727 | } | |
3728 | ||
3729 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
3730 | if (err) { | |
3731 | err = -ENOSPC; | |
3732 | goto out_unlock; | |
3733 | } | |
3734 | ||
aec7477b | 3735 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
3736 | dentry->d_name.len, |
3737 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
3738 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
3739 | &index); | |
39279cc3 CM |
3740 | if (IS_ERR(inode)) { |
3741 | err = PTR_ERR(inode); | |
3742 | goto out_fail; | |
3743 | } | |
5f39d397 | 3744 | |
39279cc3 | 3745 | drop_on_err = 1; |
33268eaf JB |
3746 | |
3747 | err = btrfs_init_acl(inode, dir); | |
3748 | if (err) | |
3749 | goto out_fail; | |
3750 | ||
39279cc3 CM |
3751 | inode->i_op = &btrfs_dir_inode_operations; |
3752 | inode->i_fop = &btrfs_dir_file_operations; | |
3753 | btrfs_set_trans_block_group(trans, inode); | |
3754 | ||
dbe674a9 | 3755 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
3756 | err = btrfs_update_inode(trans, root, inode); |
3757 | if (err) | |
3758 | goto out_fail; | |
5f39d397 | 3759 | |
e02119d5 CM |
3760 | err = btrfs_add_link(trans, dentry->d_parent->d_inode, |
3761 | inode, dentry->d_name.name, | |
3762 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
3763 | if (err) |
3764 | goto out_fail; | |
5f39d397 | 3765 | |
39279cc3 CM |
3766 | d_instantiate(dentry, inode); |
3767 | drop_on_err = 0; | |
3768 | dir->i_sb->s_dirt = 1; | |
3769 | btrfs_update_inode_block_group(trans, inode); | |
3770 | btrfs_update_inode_block_group(trans, dir); | |
3771 | ||
3772 | out_fail: | |
d3c2fdcf | 3773 | nr = trans->blocks_used; |
ab78c84d | 3774 | btrfs_end_transaction_throttle(trans, root); |
5f39d397 | 3775 | |
39279cc3 | 3776 | out_unlock: |
39279cc3 CM |
3777 | if (drop_on_err) |
3778 | iput(inode); | |
d3c2fdcf | 3779 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3780 | return err; |
3781 | } | |
3782 | ||
d352ac68 CM |
3783 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
3784 | * and an extent that you want to insert, deal with overlap and insert | |
3785 | * the new extent into the tree. | |
3786 | */ | |
3b951516 CM |
3787 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
3788 | struct extent_map *existing, | |
e6dcd2dc CM |
3789 | struct extent_map *em, |
3790 | u64 map_start, u64 map_len) | |
3b951516 CM |
3791 | { |
3792 | u64 start_diff; | |
3b951516 | 3793 | |
e6dcd2dc CM |
3794 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
3795 | start_diff = map_start - em->start; | |
3796 | em->start = map_start; | |
3797 | em->len = map_len; | |
c8b97818 CM |
3798 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
3799 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 3800 | em->block_start += start_diff; |
c8b97818 CM |
3801 | em->block_len -= start_diff; |
3802 | } | |
e6dcd2dc | 3803 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
3804 | } |
3805 | ||
c8b97818 CM |
3806 | static noinline int uncompress_inline(struct btrfs_path *path, |
3807 | struct inode *inode, struct page *page, | |
3808 | size_t pg_offset, u64 extent_offset, | |
3809 | struct btrfs_file_extent_item *item) | |
3810 | { | |
3811 | int ret; | |
3812 | struct extent_buffer *leaf = path->nodes[0]; | |
3813 | char *tmp; | |
3814 | size_t max_size; | |
3815 | unsigned long inline_size; | |
3816 | unsigned long ptr; | |
3817 | ||
3818 | WARN_ON(pg_offset != 0); | |
3819 | max_size = btrfs_file_extent_ram_bytes(leaf, item); | |
3820 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
3821 | btrfs_item_nr(leaf, path->slots[0])); | |
3822 | tmp = kmalloc(inline_size, GFP_NOFS); | |
3823 | ptr = btrfs_file_extent_inline_start(item); | |
3824 | ||
3825 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
3826 | ||
5b050f04 | 3827 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
c8b97818 CM |
3828 | ret = btrfs_zlib_decompress(tmp, page, extent_offset, |
3829 | inline_size, max_size); | |
3830 | if (ret) { | |
3831 | char *kaddr = kmap_atomic(page, KM_USER0); | |
3832 | unsigned long copy_size = min_t(u64, | |
3833 | PAGE_CACHE_SIZE - pg_offset, | |
3834 | max_size - extent_offset); | |
3835 | memset(kaddr + pg_offset, 0, copy_size); | |
3836 | kunmap_atomic(kaddr, KM_USER0); | |
3837 | } | |
3838 | kfree(tmp); | |
3839 | return 0; | |
3840 | } | |
3841 | ||
d352ac68 CM |
3842 | /* |
3843 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
3844 | * the ugly parts come from merging extents from the disk with the | |
3845 | * in-ram representation. This gets more complex because of the data=ordered code, | |
3846 | * where the in-ram extents might be locked pending data=ordered completion. | |
3847 | * | |
3848 | * This also copies inline extents directly into the page. | |
3849 | */ | |
a52d9a80 | 3850 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 3851 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
3852 | int create) |
3853 | { | |
3854 | int ret; | |
3855 | int err = 0; | |
db94535d | 3856 | u64 bytenr; |
a52d9a80 CM |
3857 | u64 extent_start = 0; |
3858 | u64 extent_end = 0; | |
3859 | u64 objectid = inode->i_ino; | |
3860 | u32 found_type; | |
f421950f | 3861 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
3862 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3863 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
3864 | struct extent_buffer *leaf; |
3865 | struct btrfs_key found_key; | |
a52d9a80 CM |
3866 | struct extent_map *em = NULL; |
3867 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 3868 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 3869 | struct btrfs_trans_handle *trans = NULL; |
c8b97818 | 3870 | int compressed; |
a52d9a80 | 3871 | |
a52d9a80 | 3872 | again: |
d1310b2e CM |
3873 | spin_lock(&em_tree->lock); |
3874 | em = lookup_extent_mapping(em_tree, start, len); | |
a061fc8d CM |
3875 | if (em) |
3876 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
d1310b2e CM |
3877 | spin_unlock(&em_tree->lock); |
3878 | ||
a52d9a80 | 3879 | if (em) { |
e1c4b745 CM |
3880 | if (em->start > start || em->start + em->len <= start) |
3881 | free_extent_map(em); | |
3882 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
3883 | free_extent_map(em); |
3884 | else | |
3885 | goto out; | |
a52d9a80 | 3886 | } |
d1310b2e | 3887 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 3888 | if (!em) { |
d1310b2e CM |
3889 | err = -ENOMEM; |
3890 | goto out; | |
a52d9a80 | 3891 | } |
e6dcd2dc | 3892 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 3893 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 3894 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 3895 | em->len = (u64)-1; |
c8b97818 | 3896 | em->block_len = (u64)-1; |
f421950f CM |
3897 | |
3898 | if (!path) { | |
3899 | path = btrfs_alloc_path(); | |
3900 | BUG_ON(!path); | |
3901 | } | |
3902 | ||
179e29e4 CM |
3903 | ret = btrfs_lookup_file_extent(trans, root, path, |
3904 | objectid, start, trans != NULL); | |
a52d9a80 CM |
3905 | if (ret < 0) { |
3906 | err = ret; | |
3907 | goto out; | |
3908 | } | |
3909 | ||
3910 | if (ret != 0) { | |
3911 | if (path->slots[0] == 0) | |
3912 | goto not_found; | |
3913 | path->slots[0]--; | |
3914 | } | |
3915 | ||
5f39d397 CM |
3916 | leaf = path->nodes[0]; |
3917 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 3918 | struct btrfs_file_extent_item); |
a52d9a80 | 3919 | /* are we inside the extent that was found? */ |
5f39d397 CM |
3920 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3921 | found_type = btrfs_key_type(&found_key); | |
3922 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
3923 | found_type != BTRFS_EXTENT_DATA_KEY) { |
3924 | goto not_found; | |
3925 | } | |
3926 | ||
5f39d397 CM |
3927 | found_type = btrfs_file_extent_type(leaf, item); |
3928 | extent_start = found_key.offset; | |
c8b97818 | 3929 | compressed = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
3930 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3931 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 3932 | extent_end = extent_start + |
db94535d | 3933 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
3934 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
3935 | size_t size; | |
3936 | size = btrfs_file_extent_inline_len(leaf, item); | |
3937 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
3938 | ~((u64)root->sectorsize - 1); | |
3939 | } | |
3940 | ||
3941 | if (start >= extent_end) { | |
3942 | path->slots[0]++; | |
3943 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
3944 | ret = btrfs_next_leaf(root, path); | |
3945 | if (ret < 0) { | |
3946 | err = ret; | |
3947 | goto out; | |
a52d9a80 | 3948 | } |
9036c102 YZ |
3949 | if (ret > 0) |
3950 | goto not_found; | |
3951 | leaf = path->nodes[0]; | |
a52d9a80 | 3952 | } |
9036c102 YZ |
3953 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3954 | if (found_key.objectid != objectid || | |
3955 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
3956 | goto not_found; | |
3957 | if (start + len <= found_key.offset) | |
3958 | goto not_found; | |
3959 | em->start = start; | |
3960 | em->len = found_key.offset - start; | |
3961 | goto not_found_em; | |
3962 | } | |
3963 | ||
d899e052 YZ |
3964 | if (found_type == BTRFS_FILE_EXTENT_REG || |
3965 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
3966 | em->start = extent_start; |
3967 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
3968 | em->orig_start = extent_start - |
3969 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
3970 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
3971 | if (bytenr == 0) { | |
5f39d397 | 3972 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
3973 | goto insert; |
3974 | } | |
c8b97818 CM |
3975 | if (compressed) { |
3976 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
3977 | em->block_start = bytenr; | |
3978 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
3979 | item); | |
3980 | } else { | |
3981 | bytenr += btrfs_file_extent_offset(leaf, item); | |
3982 | em->block_start = bytenr; | |
3983 | em->block_len = em->len; | |
d899e052 YZ |
3984 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
3985 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 3986 | } |
a52d9a80 CM |
3987 | goto insert; |
3988 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3989 | unsigned long ptr; |
a52d9a80 | 3990 | char *map; |
3326d1b0 CM |
3991 | size_t size; |
3992 | size_t extent_offset; | |
3993 | size_t copy_size; | |
a52d9a80 | 3994 | |
689f9346 | 3995 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 3996 | if (!page || create) { |
689f9346 | 3997 | em->start = extent_start; |
9036c102 | 3998 | em->len = extent_end - extent_start; |
689f9346 Y |
3999 | goto out; |
4000 | } | |
5f39d397 | 4001 | |
9036c102 YZ |
4002 | size = btrfs_file_extent_inline_len(leaf, item); |
4003 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 4004 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 4005 | size - extent_offset); |
3326d1b0 | 4006 | em->start = extent_start + extent_offset; |
70dec807 CM |
4007 | em->len = (copy_size + root->sectorsize - 1) & |
4008 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 4009 | em->orig_start = EXTENT_MAP_INLINE; |
c8b97818 CM |
4010 | if (compressed) |
4011 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
689f9346 | 4012 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 4013 | if (create == 0 && !PageUptodate(page)) { |
c8b97818 CM |
4014 | if (btrfs_file_extent_compression(leaf, item) == |
4015 | BTRFS_COMPRESS_ZLIB) { | |
4016 | ret = uncompress_inline(path, inode, page, | |
4017 | pg_offset, | |
4018 | extent_offset, item); | |
4019 | BUG_ON(ret); | |
4020 | } else { | |
4021 | map = kmap(page); | |
4022 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
4023 | copy_size); | |
4024 | kunmap(page); | |
4025 | } | |
179e29e4 CM |
4026 | flush_dcache_page(page); |
4027 | } else if (create && PageUptodate(page)) { | |
4028 | if (!trans) { | |
4029 | kunmap(page); | |
4030 | free_extent_map(em); | |
4031 | em = NULL; | |
4032 | btrfs_release_path(root, path); | |
f9295749 | 4033 | trans = btrfs_join_transaction(root, 1); |
179e29e4 CM |
4034 | goto again; |
4035 | } | |
c8b97818 | 4036 | map = kmap(page); |
70dec807 | 4037 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 4038 | copy_size); |
c8b97818 | 4039 | kunmap(page); |
179e29e4 | 4040 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 4041 | } |
d1310b2e CM |
4042 | set_extent_uptodate(io_tree, em->start, |
4043 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
4044 | goto insert; |
4045 | } else { | |
4046 | printk("unkknown found_type %d\n", found_type); | |
4047 | WARN_ON(1); | |
4048 | } | |
4049 | not_found: | |
4050 | em->start = start; | |
d1310b2e | 4051 | em->len = len; |
a52d9a80 | 4052 | not_found_em: |
5f39d397 | 4053 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 4054 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
4055 | insert: |
4056 | btrfs_release_path(root, path); | |
d1310b2e CM |
4057 | if (em->start > start || extent_map_end(em) <= start) { |
4058 | printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len); | |
a52d9a80 CM |
4059 | err = -EIO; |
4060 | goto out; | |
4061 | } | |
d1310b2e CM |
4062 | |
4063 | err = 0; | |
4064 | spin_lock(&em_tree->lock); | |
a52d9a80 | 4065 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
4066 | /* it is possible that someone inserted the extent into the tree |
4067 | * while we had the lock dropped. It is also possible that | |
4068 | * an overlapping map exists in the tree | |
4069 | */ | |
a52d9a80 | 4070 | if (ret == -EEXIST) { |
3b951516 | 4071 | struct extent_map *existing; |
e6dcd2dc CM |
4072 | |
4073 | ret = 0; | |
4074 | ||
3b951516 | 4075 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
4076 | if (existing && (existing->start > start || |
4077 | existing->start + existing->len <= start)) { | |
4078 | free_extent_map(existing); | |
4079 | existing = NULL; | |
4080 | } | |
3b951516 CM |
4081 | if (!existing) { |
4082 | existing = lookup_extent_mapping(em_tree, em->start, | |
4083 | em->len); | |
4084 | if (existing) { | |
4085 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
4086 | em, start, |
4087 | root->sectorsize); | |
3b951516 CM |
4088 | free_extent_map(existing); |
4089 | if (err) { | |
4090 | free_extent_map(em); | |
4091 | em = NULL; | |
4092 | } | |
4093 | } else { | |
4094 | err = -EIO; | |
4095 | printk("failing to insert %Lu %Lu\n", | |
4096 | start, len); | |
4097 | free_extent_map(em); | |
4098 | em = NULL; | |
4099 | } | |
4100 | } else { | |
4101 | free_extent_map(em); | |
4102 | em = existing; | |
e6dcd2dc | 4103 | err = 0; |
a52d9a80 | 4104 | } |
a52d9a80 | 4105 | } |
d1310b2e | 4106 | spin_unlock(&em_tree->lock); |
a52d9a80 | 4107 | out: |
f421950f CM |
4108 | if (path) |
4109 | btrfs_free_path(path); | |
a52d9a80 CM |
4110 | if (trans) { |
4111 | ret = btrfs_end_transaction(trans, root); | |
e6dcd2dc | 4112 | if (!err) { |
a52d9a80 | 4113 | err = ret; |
e6dcd2dc | 4114 | } |
a52d9a80 | 4115 | } |
a52d9a80 CM |
4116 | if (err) { |
4117 | free_extent_map(em); | |
4118 | WARN_ON(1); | |
4119 | return ERR_PTR(err); | |
4120 | } | |
4121 | return em; | |
4122 | } | |
4123 | ||
16432985 CM |
4124 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
4125 | const struct iovec *iov, loff_t offset, | |
4126 | unsigned long nr_segs) | |
4127 | { | |
e1c4b745 | 4128 | return -EINVAL; |
16432985 CM |
4129 | } |
4130 | ||
d396c6f5 | 4131 | static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock) |
39279cc3 | 4132 | { |
d396c6f5 | 4133 | return extent_bmap(mapping, iblock, btrfs_get_extent); |
39279cc3 CM |
4134 | } |
4135 | ||
a52d9a80 | 4136 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 4137 | { |
d1310b2e CM |
4138 | struct extent_io_tree *tree; |
4139 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 4140 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 4141 | } |
1832a6d5 | 4142 | |
a52d9a80 | 4143 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 4144 | { |
d1310b2e | 4145 | struct extent_io_tree *tree; |
b888db2b CM |
4146 | |
4147 | ||
4148 | if (current->flags & PF_MEMALLOC) { | |
4149 | redirty_page_for_writepage(wbc, page); | |
4150 | unlock_page(page); | |
4151 | return 0; | |
4152 | } | |
d1310b2e | 4153 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 4154 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
4155 | } |
4156 | ||
f421950f CM |
4157 | int btrfs_writepages(struct address_space *mapping, |
4158 | struct writeback_control *wbc) | |
b293f02e | 4159 | { |
d1310b2e | 4160 | struct extent_io_tree *tree; |
771ed689 | 4161 | |
d1310b2e | 4162 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
4163 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
4164 | } | |
4165 | ||
3ab2fb5a CM |
4166 | static int |
4167 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
4168 | struct list_head *pages, unsigned nr_pages) | |
4169 | { | |
d1310b2e CM |
4170 | struct extent_io_tree *tree; |
4171 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
4172 | return extent_readpages(tree, mapping, pages, nr_pages, |
4173 | btrfs_get_extent); | |
4174 | } | |
e6dcd2dc | 4175 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 4176 | { |
d1310b2e CM |
4177 | struct extent_io_tree *tree; |
4178 | struct extent_map_tree *map; | |
a52d9a80 | 4179 | int ret; |
8c2383c3 | 4180 | |
d1310b2e CM |
4181 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
4182 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 4183 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
4184 | if (ret == 1) { |
4185 | ClearPagePrivate(page); | |
4186 | set_page_private(page, 0); | |
4187 | page_cache_release(page); | |
39279cc3 | 4188 | } |
a52d9a80 | 4189 | return ret; |
39279cc3 CM |
4190 | } |
4191 | ||
e6dcd2dc CM |
4192 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
4193 | { | |
98509cfc CM |
4194 | if (PageWriteback(page) || PageDirty(page)) |
4195 | return 0; | |
e6dcd2dc CM |
4196 | return __btrfs_releasepage(page, gfp_flags); |
4197 | } | |
4198 | ||
a52d9a80 | 4199 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 4200 | { |
d1310b2e | 4201 | struct extent_io_tree *tree; |
e6dcd2dc CM |
4202 | struct btrfs_ordered_extent *ordered; |
4203 | u64 page_start = page_offset(page); | |
4204 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 4205 | |
e6dcd2dc | 4206 | wait_on_page_writeback(page); |
d1310b2e | 4207 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
4208 | if (offset) { |
4209 | btrfs_releasepage(page, GFP_NOFS); | |
4210 | return; | |
4211 | } | |
4212 | ||
4213 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4214 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, | |
4215 | page_offset(page)); | |
4216 | if (ordered) { | |
eb84ae03 CM |
4217 | /* |
4218 | * IO on this page will never be started, so we need | |
4219 | * to account for any ordered extents now | |
4220 | */ | |
e6dcd2dc CM |
4221 | clear_extent_bit(tree, page_start, page_end, |
4222 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4223 | EXTENT_LOCKED, 1, 0, GFP_NOFS); | |
211f90e6 CM |
4224 | btrfs_finish_ordered_io(page->mapping->host, |
4225 | page_start, page_end); | |
e6dcd2dc CM |
4226 | btrfs_put_ordered_extent(ordered); |
4227 | lock_extent(tree, page_start, page_end, GFP_NOFS); | |
4228 | } | |
4229 | clear_extent_bit(tree, page_start, page_end, | |
4230 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | | |
4231 | EXTENT_ORDERED, | |
4232 | 1, 1, GFP_NOFS); | |
4233 | __btrfs_releasepage(page, GFP_NOFS); | |
4234 | ||
4a096752 | 4235 | ClearPageChecked(page); |
9ad6b7bc | 4236 | if (PagePrivate(page)) { |
9ad6b7bc CM |
4237 | ClearPagePrivate(page); |
4238 | set_page_private(page, 0); | |
4239 | page_cache_release(page); | |
4240 | } | |
39279cc3 CM |
4241 | } |
4242 | ||
9ebefb18 CM |
4243 | /* |
4244 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
4245 | * called from a page fault handler when a page is first dirtied. Hence we must | |
4246 | * be careful to check for EOF conditions here. We set the page up correctly | |
4247 | * for a written page which means we get ENOSPC checking when writing into | |
4248 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
4249 | * support these features. | |
4250 | * | |
4251 | * We are not allowed to take the i_mutex here so we have to play games to | |
4252 | * protect against truncate races as the page could now be beyond EOF. Because | |
4253 | * vmtruncate() writes the inode size before removing pages, once we have the | |
4254 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
4255 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
4256 | * unlock the page. | |
4257 | */ | |
4258 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) | |
4259 | { | |
6da6abae | 4260 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 4261 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4262 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4263 | struct btrfs_ordered_extent *ordered; | |
4264 | char *kaddr; | |
4265 | unsigned long zero_start; | |
9ebefb18 | 4266 | loff_t size; |
1832a6d5 | 4267 | int ret; |
a52d9a80 | 4268 | u64 page_start; |
e6dcd2dc | 4269 | u64 page_end; |
9ebefb18 | 4270 | |
1832a6d5 | 4271 | ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0); |
1832a6d5 CM |
4272 | if (ret) |
4273 | goto out; | |
4274 | ||
4275 | ret = -EINVAL; | |
e6dcd2dc | 4276 | again: |
9ebefb18 | 4277 | lock_page(page); |
9ebefb18 | 4278 | size = i_size_read(inode); |
e6dcd2dc CM |
4279 | page_start = page_offset(page); |
4280 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 4281 | |
9ebefb18 | 4282 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 4283 | (page_start >= size)) { |
9ebefb18 CM |
4284 | /* page got truncated out from underneath us */ |
4285 | goto out_unlock; | |
4286 | } | |
e6dcd2dc CM |
4287 | wait_on_page_writeback(page); |
4288 | ||
4289 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4290 | set_page_extent_mapped(page); | |
4291 | ||
eb84ae03 CM |
4292 | /* |
4293 | * we can't set the delalloc bits if there are pending ordered | |
4294 | * extents. Drop our locks and wait for them to finish | |
4295 | */ | |
e6dcd2dc CM |
4296 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
4297 | if (ordered) { | |
4298 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
4299 | unlock_page(page); | |
eb84ae03 | 4300 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4301 | btrfs_put_ordered_extent(ordered); |
4302 | goto again; | |
4303 | } | |
4304 | ||
ea8c2819 | 4305 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
e6dcd2dc | 4306 | ret = 0; |
9ebefb18 CM |
4307 | |
4308 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 4309 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 4310 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 4311 | else |
e6dcd2dc | 4312 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 4313 | |
e6dcd2dc CM |
4314 | if (zero_start != PAGE_CACHE_SIZE) { |
4315 | kaddr = kmap(page); | |
4316 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
4317 | flush_dcache_page(page); | |
4318 | kunmap(page); | |
4319 | } | |
247e743c | 4320 | ClearPageChecked(page); |
e6dcd2dc CM |
4321 | set_page_dirty(page); |
4322 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
9ebefb18 CM |
4323 | |
4324 | out_unlock: | |
4325 | unlock_page(page); | |
1832a6d5 | 4326 | out: |
9ebefb18 CM |
4327 | return ret; |
4328 | } | |
4329 | ||
39279cc3 CM |
4330 | static void btrfs_truncate(struct inode *inode) |
4331 | { | |
4332 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4333 | int ret; | |
4334 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 4335 | unsigned long nr; |
dbe674a9 | 4336 | u64 mask = root->sectorsize - 1; |
39279cc3 CM |
4337 | |
4338 | if (!S_ISREG(inode->i_mode)) | |
4339 | return; | |
4340 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4341 | return; | |
4342 | ||
4343 | btrfs_truncate_page(inode->i_mapping, inode->i_size); | |
4a096752 | 4344 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
39279cc3 | 4345 | |
39279cc3 CM |
4346 | trans = btrfs_start_transaction(root, 1); |
4347 | btrfs_set_trans_block_group(trans, inode); | |
dbe674a9 | 4348 | btrfs_i_size_write(inode, inode->i_size); |
39279cc3 | 4349 | |
7b128766 JB |
4350 | ret = btrfs_orphan_add(trans, inode); |
4351 | if (ret) | |
4352 | goto out; | |
39279cc3 | 4353 | /* FIXME, add redo link to tree so we don't leak on crash */ |
e02119d5 | 4354 | ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, |
85e21bac | 4355 | BTRFS_EXTENT_DATA_KEY); |
39279cc3 | 4356 | btrfs_update_inode(trans, root, inode); |
5f39d397 | 4357 | |
7b128766 JB |
4358 | ret = btrfs_orphan_del(trans, inode); |
4359 | BUG_ON(ret); | |
4360 | ||
4361 | out: | |
4362 | nr = trans->blocks_used; | |
89ce8a63 | 4363 | ret = btrfs_end_transaction_throttle(trans, root); |
39279cc3 | 4364 | BUG_ON(ret); |
d3c2fdcf | 4365 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4366 | } |
4367 | ||
3b96362c SW |
4368 | /* |
4369 | * Invalidate a single dcache entry at the root of the filesystem. | |
4370 | * Needed after creation of snapshot or subvolume. | |
4371 | */ | |
3de4586c | 4372 | void btrfs_invalidate_dcache_root(struct inode *dir, char *name, |
3b96362c SW |
4373 | int namelen) |
4374 | { | |
4375 | struct dentry *alias, *entry; | |
4376 | struct qstr qstr; | |
4377 | ||
3de4586c | 4378 | alias = d_find_alias(dir); |
3b96362c SW |
4379 | if (alias) { |
4380 | qstr.name = name; | |
4381 | qstr.len = namelen; | |
4382 | /* change me if btrfs ever gets a d_hash operation */ | |
4383 | qstr.hash = full_name_hash(qstr.name, qstr.len); | |
4384 | entry = d_lookup(alias, &qstr); | |
4385 | dput(alias); | |
4386 | if (entry) { | |
4387 | d_invalidate(entry); | |
4388 | dput(entry); | |
4389 | } | |
4390 | } | |
4391 | } | |
4392 | ||
d352ac68 CM |
4393 | /* |
4394 | * create a new subvolume directory/inode (helper for the ioctl). | |
4395 | */ | |
cb8e7090 | 4396 | int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry, |
f46b5a66 CH |
4397 | struct btrfs_trans_handle *trans, u64 new_dirid, |
4398 | struct btrfs_block_group_cache *block_group) | |
39279cc3 | 4399 | { |
39279cc3 | 4400 | struct inode *inode; |
cb8e7090 | 4401 | int error; |
00e4e6b3 | 4402 | u64 index = 0; |
39279cc3 | 4403 | |
aec7477b | 4404 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
00e4e6b3 | 4405 | new_dirid, block_group, S_IFDIR | 0700, &index); |
54aa1f4d | 4406 | if (IS_ERR(inode)) |
f46b5a66 | 4407 | return PTR_ERR(inode); |
39279cc3 CM |
4408 | inode->i_op = &btrfs_dir_inode_operations; |
4409 | inode->i_fop = &btrfs_dir_file_operations; | |
4410 | ||
39279cc3 | 4411 | inode->i_nlink = 1; |
dbe674a9 | 4412 | btrfs_i_size_write(inode, 0); |
3b96362c | 4413 | |
cb8e7090 CH |
4414 | error = btrfs_update_inode(trans, new_root, inode); |
4415 | if (error) | |
4416 | return error; | |
4417 | ||
4418 | d_instantiate(dentry, inode); | |
4419 | return 0; | |
39279cc3 CM |
4420 | } |
4421 | ||
d352ac68 CM |
4422 | /* helper function for file defrag and space balancing. This |
4423 | * forces readahead on a given range of bytes in an inode | |
4424 | */ | |
edbd8d4e | 4425 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
4426 | struct file_ra_state *ra, struct file *file, |
4427 | pgoff_t offset, pgoff_t last_index) | |
4428 | { | |
8e7bf94f | 4429 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 4430 | |
86479a04 CM |
4431 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
4432 | return offset + req_size; | |
86479a04 CM |
4433 | } |
4434 | ||
39279cc3 CM |
4435 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
4436 | { | |
4437 | struct btrfs_inode *ei; | |
4438 | ||
4439 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
4440 | if (!ei) | |
4441 | return NULL; | |
15ee9bc7 | 4442 | ei->last_trans = 0; |
e02119d5 | 4443 | ei->logged_trans = 0; |
e6dcd2dc | 4444 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
33268eaf JB |
4445 | ei->i_acl = BTRFS_ACL_NOT_CACHED; |
4446 | ei->i_default_acl = BTRFS_ACL_NOT_CACHED; | |
7b128766 | 4447 | INIT_LIST_HEAD(&ei->i_orphan); |
39279cc3 CM |
4448 | return &ei->vfs_inode; |
4449 | } | |
4450 | ||
4451 | void btrfs_destroy_inode(struct inode *inode) | |
4452 | { | |
e6dcd2dc | 4453 | struct btrfs_ordered_extent *ordered; |
39279cc3 CM |
4454 | WARN_ON(!list_empty(&inode->i_dentry)); |
4455 | WARN_ON(inode->i_data.nrpages); | |
4456 | ||
33268eaf JB |
4457 | if (BTRFS_I(inode)->i_acl && |
4458 | BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED) | |
4459 | posix_acl_release(BTRFS_I(inode)->i_acl); | |
4460 | if (BTRFS_I(inode)->i_default_acl && | |
4461 | BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED) | |
4462 | posix_acl_release(BTRFS_I(inode)->i_default_acl); | |
4463 | ||
bcc63abb | 4464 | spin_lock(&BTRFS_I(inode)->root->list_lock); |
7b128766 JB |
4465 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
4466 | printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" | |
4467 | " list\n", inode->i_ino); | |
4468 | dump_stack(); | |
4469 | } | |
bcc63abb | 4470 | spin_unlock(&BTRFS_I(inode)->root->list_lock); |
7b128766 | 4471 | |
e6dcd2dc CM |
4472 | while(1) { |
4473 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); | |
4474 | if (!ordered) | |
4475 | break; | |
4476 | else { | |
4477 | printk("found ordered extent %Lu %Lu\n", | |
4478 | ordered->file_offset, ordered->len); | |
4479 | btrfs_remove_ordered_extent(inode, ordered); | |
4480 | btrfs_put_ordered_extent(ordered); | |
4481 | btrfs_put_ordered_extent(ordered); | |
4482 | } | |
4483 | } | |
5b21f2ed | 4484 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
39279cc3 CM |
4485 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
4486 | } | |
4487 | ||
0ee0fda0 | 4488 | static void init_once(void *foo) |
39279cc3 CM |
4489 | { |
4490 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
4491 | ||
4492 | inode_init_once(&ei->vfs_inode); | |
4493 | } | |
4494 | ||
4495 | void btrfs_destroy_cachep(void) | |
4496 | { | |
4497 | if (btrfs_inode_cachep) | |
4498 | kmem_cache_destroy(btrfs_inode_cachep); | |
4499 | if (btrfs_trans_handle_cachep) | |
4500 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
4501 | if (btrfs_transaction_cachep) | |
4502 | kmem_cache_destroy(btrfs_transaction_cachep); | |
4503 | if (btrfs_bit_radix_cachep) | |
4504 | kmem_cache_destroy(btrfs_bit_radix_cachep); | |
4505 | if (btrfs_path_cachep) | |
4506 | kmem_cache_destroy(btrfs_path_cachep); | |
4507 | } | |
4508 | ||
86479a04 | 4509 | struct kmem_cache *btrfs_cache_create(const char *name, size_t size, |
92fee66d | 4510 | unsigned long extra_flags, |
2b1f55b0 | 4511 | void (*ctor)(void *)) |
92fee66d CM |
4512 | { |
4513 | return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT | | |
2b1f55b0 | 4514 | SLAB_MEM_SPREAD | extra_flags), ctor); |
92fee66d CM |
4515 | } |
4516 | ||
39279cc3 CM |
4517 | int btrfs_init_cachep(void) |
4518 | { | |
86479a04 | 4519 | btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache", |
92fee66d CM |
4520 | sizeof(struct btrfs_inode), |
4521 | 0, init_once); | |
39279cc3 CM |
4522 | if (!btrfs_inode_cachep) |
4523 | goto fail; | |
86479a04 CM |
4524 | btrfs_trans_handle_cachep = |
4525 | btrfs_cache_create("btrfs_trans_handle_cache", | |
4526 | sizeof(struct btrfs_trans_handle), | |
4527 | 0, NULL); | |
39279cc3 CM |
4528 | if (!btrfs_trans_handle_cachep) |
4529 | goto fail; | |
86479a04 | 4530 | btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache", |
39279cc3 | 4531 | sizeof(struct btrfs_transaction), |
92fee66d | 4532 | 0, NULL); |
39279cc3 CM |
4533 | if (!btrfs_transaction_cachep) |
4534 | goto fail; | |
86479a04 | 4535 | btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache", |
23223584 | 4536 | sizeof(struct btrfs_path), |
92fee66d | 4537 | 0, NULL); |
39279cc3 CM |
4538 | if (!btrfs_path_cachep) |
4539 | goto fail; | |
86479a04 | 4540 | btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256, |
92fee66d | 4541 | SLAB_DESTROY_BY_RCU, NULL); |
39279cc3 CM |
4542 | if (!btrfs_bit_radix_cachep) |
4543 | goto fail; | |
4544 | return 0; | |
4545 | fail: | |
4546 | btrfs_destroy_cachep(); | |
4547 | return -ENOMEM; | |
4548 | } | |
4549 | ||
4550 | static int btrfs_getattr(struct vfsmount *mnt, | |
4551 | struct dentry *dentry, struct kstat *stat) | |
4552 | { | |
4553 | struct inode *inode = dentry->d_inode; | |
4554 | generic_fillattr(inode, stat); | |
3394e160 | 4555 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 4556 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
4557 | stat->blocks = (inode_get_bytes(inode) + |
4558 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
4559 | return 0; |
4560 | } | |
4561 | ||
4562 | static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry, | |
4563 | struct inode * new_dir,struct dentry *new_dentry) | |
4564 | { | |
4565 | struct btrfs_trans_handle *trans; | |
4566 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4567 | struct inode *new_inode = new_dentry->d_inode; | |
4568 | struct inode *old_inode = old_dentry->d_inode; | |
4569 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 4570 | u64 index = 0; |
39279cc3 CM |
4571 | int ret; |
4572 | ||
3394e160 CM |
4573 | /* we're not allowed to rename between subvolumes */ |
4574 | if (BTRFS_I(old_inode)->root->root_key.objectid != | |
4575 | BTRFS_I(new_dir)->root->root_key.objectid) | |
4576 | return -EXDEV; | |
4577 | ||
39279cc3 CM |
4578 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
4579 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) { | |
4580 | return -ENOTEMPTY; | |
4581 | } | |
5f39d397 | 4582 | |
0660b5af CM |
4583 | /* to rename a snapshot or subvolume, we need to juggle the |
4584 | * backrefs. This isn't coded yet | |
4585 | */ | |
4586 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
4587 | return -EXDEV; | |
4588 | ||
1832a6d5 CM |
4589 | ret = btrfs_check_free_space(root, 1, 0); |
4590 | if (ret) | |
4591 | goto out_unlock; | |
4592 | ||
39279cc3 | 4593 | trans = btrfs_start_transaction(root, 1); |
5f39d397 | 4594 | |
39279cc3 | 4595 | btrfs_set_trans_block_group(trans, new_dir); |
39279cc3 | 4596 | |
e02119d5 | 4597 | btrfs_inc_nlink(old_dentry->d_inode); |
39279cc3 CM |
4598 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
4599 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
4600 | old_inode->i_ctime = ctime; | |
5f39d397 | 4601 | |
e02119d5 CM |
4602 | ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, |
4603 | old_dentry->d_name.name, | |
4604 | old_dentry->d_name.len); | |
39279cc3 CM |
4605 | if (ret) |
4606 | goto out_fail; | |
4607 | ||
4608 | if (new_inode) { | |
4609 | new_inode->i_ctime = CURRENT_TIME; | |
e02119d5 CM |
4610 | ret = btrfs_unlink_inode(trans, root, new_dir, |
4611 | new_dentry->d_inode, | |
4612 | new_dentry->d_name.name, | |
4613 | new_dentry->d_name.len); | |
39279cc3 CM |
4614 | if (ret) |
4615 | goto out_fail; | |
7b128766 | 4616 | if (new_inode->i_nlink == 0) { |
e02119d5 | 4617 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
7b128766 JB |
4618 | if (ret) |
4619 | goto out_fail; | |
4620 | } | |
e02119d5 | 4621 | |
39279cc3 | 4622 | } |
3de4586c | 4623 | ret = btrfs_set_inode_index(new_dir, &index); |
aec7477b JB |
4624 | if (ret) |
4625 | goto out_fail; | |
4626 | ||
e02119d5 CM |
4627 | ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode, |
4628 | old_inode, new_dentry->d_name.name, | |
4629 | new_dentry->d_name.len, 1, index); | |
39279cc3 CM |
4630 | if (ret) |
4631 | goto out_fail; | |
4632 | ||
4633 | out_fail: | |
ab78c84d | 4634 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4635 | out_unlock: |
39279cc3 CM |
4636 | return ret; |
4637 | } | |
4638 | ||
d352ac68 CM |
4639 | /* |
4640 | * some fairly slow code that needs optimization. This walks the list | |
4641 | * of all the inodes with pending delalloc and forces them to disk. | |
4642 | */ | |
ea8c2819 CM |
4643 | int btrfs_start_delalloc_inodes(struct btrfs_root *root) |
4644 | { | |
4645 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
4646 | struct btrfs_inode *binode; | |
5b21f2ed | 4647 | struct inode *inode; |
ea8c2819 CM |
4648 | unsigned long flags; |
4649 | ||
c146afad YZ |
4650 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
4651 | return -EROFS; | |
4652 | ||
ea8c2819 CM |
4653 | spin_lock_irqsave(&root->fs_info->delalloc_lock, flags); |
4654 | while(!list_empty(head)) { | |
4655 | binode = list_entry(head->next, struct btrfs_inode, | |
4656 | delalloc_inodes); | |
5b21f2ed ZY |
4657 | inode = igrab(&binode->vfs_inode); |
4658 | if (!inode) | |
4659 | list_del_init(&binode->delalloc_inodes); | |
ea8c2819 | 4660 | spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags); |
5b21f2ed | 4661 | if (inode) { |
8c8bee1d | 4662 | filemap_flush(inode->i_mapping); |
5b21f2ed ZY |
4663 | iput(inode); |
4664 | } | |
4665 | cond_resched(); | |
ea8c2819 CM |
4666 | spin_lock_irqsave(&root->fs_info->delalloc_lock, flags); |
4667 | } | |
4668 | spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags); | |
8c8bee1d CM |
4669 | |
4670 | /* the filemap_flush will queue IO into the worker threads, but | |
4671 | * we have to make sure the IO is actually started and that | |
4672 | * ordered extents get created before we return | |
4673 | */ | |
4674 | atomic_inc(&root->fs_info->async_submit_draining); | |
771ed689 CM |
4675 | while(atomic_read(&root->fs_info->nr_async_submits) || |
4676 | atomic_read(&root->fs_info->async_delalloc_pages)) { | |
8c8bee1d | 4677 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
4678 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
4679 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
4680 | } |
4681 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
4682 | return 0; |
4683 | } | |
4684 | ||
39279cc3 CM |
4685 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
4686 | const char *symname) | |
4687 | { | |
4688 | struct btrfs_trans_handle *trans; | |
4689 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4690 | struct btrfs_path *path; | |
4691 | struct btrfs_key key; | |
1832a6d5 | 4692 | struct inode *inode = NULL; |
39279cc3 CM |
4693 | int err; |
4694 | int drop_inode = 0; | |
4695 | u64 objectid; | |
00e4e6b3 | 4696 | u64 index = 0 ; |
39279cc3 CM |
4697 | int name_len; |
4698 | int datasize; | |
5f39d397 | 4699 | unsigned long ptr; |
39279cc3 | 4700 | struct btrfs_file_extent_item *ei; |
5f39d397 | 4701 | struct extent_buffer *leaf; |
1832a6d5 | 4702 | unsigned long nr = 0; |
39279cc3 CM |
4703 | |
4704 | name_len = strlen(symname) + 1; | |
4705 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
4706 | return -ENAMETOOLONG; | |
1832a6d5 | 4707 | |
1832a6d5 CM |
4708 | err = btrfs_check_free_space(root, 1, 0); |
4709 | if (err) | |
4710 | goto out_fail; | |
4711 | ||
39279cc3 CM |
4712 | trans = btrfs_start_transaction(root, 1); |
4713 | btrfs_set_trans_block_group(trans, dir); | |
4714 | ||
4715 | err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid); | |
4716 | if (err) { | |
4717 | err = -ENOSPC; | |
4718 | goto out_unlock; | |
4719 | } | |
4720 | ||
aec7477b | 4721 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
9c58309d CM |
4722 | dentry->d_name.len, |
4723 | dentry->d_parent->d_inode->i_ino, objectid, | |
00e4e6b3 CM |
4724 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
4725 | &index); | |
39279cc3 CM |
4726 | err = PTR_ERR(inode); |
4727 | if (IS_ERR(inode)) | |
4728 | goto out_unlock; | |
4729 | ||
33268eaf JB |
4730 | err = btrfs_init_acl(inode, dir); |
4731 | if (err) { | |
4732 | drop_inode = 1; | |
4733 | goto out_unlock; | |
4734 | } | |
4735 | ||
39279cc3 | 4736 | btrfs_set_trans_block_group(trans, inode); |
00e4e6b3 | 4737 | err = btrfs_add_nondir(trans, dentry, inode, 0, index); |
39279cc3 CM |
4738 | if (err) |
4739 | drop_inode = 1; | |
4740 | else { | |
4741 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4742 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4743 | inode->i_fop = &btrfs_file_operations; |
4744 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4745 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
4746 | } |
4747 | dir->i_sb->s_dirt = 1; | |
4748 | btrfs_update_inode_block_group(trans, inode); | |
4749 | btrfs_update_inode_block_group(trans, dir); | |
4750 | if (drop_inode) | |
4751 | goto out_unlock; | |
4752 | ||
4753 | path = btrfs_alloc_path(); | |
4754 | BUG_ON(!path); | |
4755 | key.objectid = inode->i_ino; | |
4756 | key.offset = 0; | |
39279cc3 CM |
4757 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
4758 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
4759 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
4760 | datasize); | |
54aa1f4d CM |
4761 | if (err) { |
4762 | drop_inode = 1; | |
4763 | goto out_unlock; | |
4764 | } | |
5f39d397 CM |
4765 | leaf = path->nodes[0]; |
4766 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
4767 | struct btrfs_file_extent_item); | |
4768 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
4769 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 4770 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
4771 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
4772 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
4773 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
4774 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
4775 | ||
39279cc3 | 4776 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
4777 | write_extent_buffer(leaf, symname, ptr, name_len); |
4778 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 4779 | btrfs_free_path(path); |
5f39d397 | 4780 | |
39279cc3 CM |
4781 | inode->i_op = &btrfs_symlink_inode_operations; |
4782 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 4783 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 4784 | inode_set_bytes(inode, name_len); |
dbe674a9 | 4785 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
4786 | err = btrfs_update_inode(trans, root, inode); |
4787 | if (err) | |
4788 | drop_inode = 1; | |
39279cc3 CM |
4789 | |
4790 | out_unlock: | |
d3c2fdcf | 4791 | nr = trans->blocks_used; |
ab78c84d | 4792 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4793 | out_fail: |
39279cc3 CM |
4794 | if (drop_inode) { |
4795 | inode_dec_link_count(inode); | |
4796 | iput(inode); | |
4797 | } | |
d3c2fdcf | 4798 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4799 | return err; |
4800 | } | |
16432985 | 4801 | |
d899e052 YZ |
4802 | static int prealloc_file_range(struct inode *inode, u64 start, u64 end, |
4803 | u64 alloc_hint, int mode) | |
4804 | { | |
4805 | struct btrfs_trans_handle *trans; | |
4806 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4807 | struct btrfs_key ins; | |
4808 | u64 alloc_size; | |
4809 | u64 cur_offset = start; | |
4810 | u64 num_bytes = end - start; | |
4811 | int ret = 0; | |
4812 | ||
4813 | trans = btrfs_join_transaction(root, 1); | |
4814 | BUG_ON(!trans); | |
4815 | btrfs_set_trans_block_group(trans, inode); | |
4816 | ||
4817 | while (num_bytes > 0) { | |
4818 | alloc_size = min(num_bytes, root->fs_info->max_extent); | |
4819 | ret = btrfs_reserve_extent(trans, root, alloc_size, | |
4820 | root->sectorsize, 0, alloc_hint, | |
4821 | (u64)-1, &ins, 1); | |
4822 | if (ret) { | |
4823 | WARN_ON(1); | |
4824 | goto out; | |
4825 | } | |
4826 | ret = insert_reserved_file_extent(trans, inode, | |
4827 | cur_offset, ins.objectid, | |
4828 | ins.offset, ins.offset, | |
4829 | ins.offset, 0, 0, 0, | |
4830 | BTRFS_FILE_EXTENT_PREALLOC); | |
4831 | BUG_ON(ret); | |
4832 | num_bytes -= ins.offset; | |
4833 | cur_offset += ins.offset; | |
4834 | alloc_hint = ins.objectid + ins.offset; | |
4835 | } | |
4836 | out: | |
4837 | if (cur_offset > start) { | |
4838 | inode->i_ctime = CURRENT_TIME; | |
4839 | btrfs_set_flag(inode, PREALLOC); | |
4840 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4841 | cur_offset > i_size_read(inode)) | |
4842 | btrfs_i_size_write(inode, cur_offset); | |
4843 | ret = btrfs_update_inode(trans, root, inode); | |
4844 | BUG_ON(ret); | |
4845 | } | |
4846 | ||
4847 | btrfs_end_transaction(trans, root); | |
4848 | return ret; | |
4849 | } | |
4850 | ||
4851 | static long btrfs_fallocate(struct inode *inode, int mode, | |
4852 | loff_t offset, loff_t len) | |
4853 | { | |
4854 | u64 cur_offset; | |
4855 | u64 last_byte; | |
4856 | u64 alloc_start; | |
4857 | u64 alloc_end; | |
4858 | u64 alloc_hint = 0; | |
4859 | u64 mask = BTRFS_I(inode)->root->sectorsize - 1; | |
4860 | struct extent_map *em; | |
4861 | int ret; | |
4862 | ||
4863 | alloc_start = offset & ~mask; | |
4864 | alloc_end = (offset + len + mask) & ~mask; | |
4865 | ||
4866 | mutex_lock(&inode->i_mutex); | |
4867 | if (alloc_start > inode->i_size) { | |
4868 | ret = btrfs_cont_expand(inode, alloc_start); | |
4869 | if (ret) | |
4870 | goto out; | |
4871 | } | |
4872 | ||
4873 | while (1) { | |
4874 | struct btrfs_ordered_extent *ordered; | |
4875 | lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, | |
4876 | alloc_end - 1, GFP_NOFS); | |
4877 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
4878 | alloc_end - 1); | |
4879 | if (ordered && | |
4880 | ordered->file_offset + ordered->len > alloc_start && | |
4881 | ordered->file_offset < alloc_end) { | |
4882 | btrfs_put_ordered_extent(ordered); | |
4883 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4884 | alloc_start, alloc_end - 1, GFP_NOFS); | |
4885 | btrfs_wait_ordered_range(inode, alloc_start, | |
4886 | alloc_end - alloc_start); | |
4887 | } else { | |
4888 | if (ordered) | |
4889 | btrfs_put_ordered_extent(ordered); | |
4890 | break; | |
4891 | } | |
4892 | } | |
4893 | ||
4894 | cur_offset = alloc_start; | |
4895 | while (1) { | |
4896 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4897 | alloc_end - cur_offset, 0); | |
4898 | BUG_ON(IS_ERR(em) || !em); | |
4899 | last_byte = min(extent_map_end(em), alloc_end); | |
4900 | last_byte = (last_byte + mask) & ~mask; | |
4901 | if (em->block_start == EXTENT_MAP_HOLE) { | |
4902 | ret = prealloc_file_range(inode, cur_offset, | |
4903 | last_byte, alloc_hint, mode); | |
4904 | if (ret < 0) { | |
4905 | free_extent_map(em); | |
4906 | break; | |
4907 | } | |
4908 | } | |
4909 | if (em->block_start <= EXTENT_MAP_LAST_BYTE) | |
4910 | alloc_hint = em->block_start; | |
4911 | free_extent_map(em); | |
4912 | ||
4913 | cur_offset = last_byte; | |
4914 | if (cur_offset >= alloc_end) { | |
4915 | ret = 0; | |
4916 | break; | |
4917 | } | |
4918 | } | |
4919 | unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1, | |
4920 | GFP_NOFS); | |
4921 | out: | |
4922 | mutex_unlock(&inode->i_mutex); | |
4923 | return ret; | |
4924 | } | |
4925 | ||
e6dcd2dc CM |
4926 | static int btrfs_set_page_dirty(struct page *page) |
4927 | { | |
e6dcd2dc CM |
4928 | return __set_page_dirty_nobuffers(page); |
4929 | } | |
4930 | ||
0ee0fda0 | 4931 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd Y |
4932 | { |
4933 | if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE)) | |
4934 | return -EACCES; | |
33268eaf | 4935 | return generic_permission(inode, mask, btrfs_check_acl); |
fdebe2bd | 4936 | } |
39279cc3 CM |
4937 | |
4938 | static struct inode_operations btrfs_dir_inode_operations = { | |
3394e160 | 4939 | .getattr = btrfs_getattr, |
39279cc3 CM |
4940 | .lookup = btrfs_lookup, |
4941 | .create = btrfs_create, | |
4942 | .unlink = btrfs_unlink, | |
4943 | .link = btrfs_link, | |
4944 | .mkdir = btrfs_mkdir, | |
4945 | .rmdir = btrfs_rmdir, | |
4946 | .rename = btrfs_rename, | |
4947 | .symlink = btrfs_symlink, | |
4948 | .setattr = btrfs_setattr, | |
618e21d5 | 4949 | .mknod = btrfs_mknod, |
95819c05 CH |
4950 | .setxattr = btrfs_setxattr, |
4951 | .getxattr = btrfs_getxattr, | |
5103e947 | 4952 | .listxattr = btrfs_listxattr, |
95819c05 | 4953 | .removexattr = btrfs_removexattr, |
fdebe2bd | 4954 | .permission = btrfs_permission, |
39279cc3 | 4955 | }; |
39279cc3 CM |
4956 | static struct inode_operations btrfs_dir_ro_inode_operations = { |
4957 | .lookup = btrfs_lookup, | |
fdebe2bd | 4958 | .permission = btrfs_permission, |
39279cc3 | 4959 | }; |
39279cc3 CM |
4960 | static struct file_operations btrfs_dir_file_operations = { |
4961 | .llseek = generic_file_llseek, | |
4962 | .read = generic_read_dir, | |
cbdf5a24 | 4963 | .readdir = btrfs_real_readdir, |
34287aa3 | 4964 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 4965 | #ifdef CONFIG_COMPAT |
34287aa3 | 4966 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 4967 | #endif |
6bf13c0c | 4968 | .release = btrfs_release_file, |
e02119d5 | 4969 | .fsync = btrfs_sync_file, |
39279cc3 CM |
4970 | }; |
4971 | ||
d1310b2e | 4972 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 4973 | .fill_delalloc = run_delalloc_range, |
065631f6 | 4974 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 4975 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 4976 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 4977 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 4978 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 4979 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
4980 | .set_bit_hook = btrfs_set_bit_hook, |
4981 | .clear_bit_hook = btrfs_clear_bit_hook, | |
07157aac CM |
4982 | }; |
4983 | ||
39279cc3 CM |
4984 | static struct address_space_operations btrfs_aops = { |
4985 | .readpage = btrfs_readpage, | |
4986 | .writepage = btrfs_writepage, | |
b293f02e | 4987 | .writepages = btrfs_writepages, |
3ab2fb5a | 4988 | .readpages = btrfs_readpages, |
39279cc3 | 4989 | .sync_page = block_sync_page, |
39279cc3 | 4990 | .bmap = btrfs_bmap, |
16432985 | 4991 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
4992 | .invalidatepage = btrfs_invalidatepage, |
4993 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 4994 | .set_page_dirty = btrfs_set_page_dirty, |
39279cc3 CM |
4995 | }; |
4996 | ||
4997 | static struct address_space_operations btrfs_symlink_aops = { | |
4998 | .readpage = btrfs_readpage, | |
4999 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
5000 | .invalidatepage = btrfs_invalidatepage, |
5001 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
5002 | }; |
5003 | ||
5004 | static struct inode_operations btrfs_file_inode_operations = { | |
5005 | .truncate = btrfs_truncate, | |
5006 | .getattr = btrfs_getattr, | |
5007 | .setattr = btrfs_setattr, | |
95819c05 CH |
5008 | .setxattr = btrfs_setxattr, |
5009 | .getxattr = btrfs_getxattr, | |
5103e947 | 5010 | .listxattr = btrfs_listxattr, |
95819c05 | 5011 | .removexattr = btrfs_removexattr, |
fdebe2bd | 5012 | .permission = btrfs_permission, |
d899e052 | 5013 | .fallocate = btrfs_fallocate, |
39279cc3 | 5014 | }; |
618e21d5 JB |
5015 | static struct inode_operations btrfs_special_inode_operations = { |
5016 | .getattr = btrfs_getattr, | |
5017 | .setattr = btrfs_setattr, | |
fdebe2bd | 5018 | .permission = btrfs_permission, |
95819c05 CH |
5019 | .setxattr = btrfs_setxattr, |
5020 | .getxattr = btrfs_getxattr, | |
33268eaf | 5021 | .listxattr = btrfs_listxattr, |
95819c05 | 5022 | .removexattr = btrfs_removexattr, |
618e21d5 | 5023 | }; |
39279cc3 CM |
5024 | static struct inode_operations btrfs_symlink_inode_operations = { |
5025 | .readlink = generic_readlink, | |
5026 | .follow_link = page_follow_link_light, | |
5027 | .put_link = page_put_link, | |
fdebe2bd | 5028 | .permission = btrfs_permission, |
39279cc3 | 5029 | }; |