Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
4b4e25f2 | 40 | #include "compat.h" |
39279cc3 CM |
41 | #include "ctree.h" |
42 | #include "disk-io.h" | |
43 | #include "transaction.h" | |
44 | #include "btrfs_inode.h" | |
45 | #include "ioctl.h" | |
46 | #include "print-tree.h" | |
0b86a832 | 47 | #include "volumes.h" |
e6dcd2dc | 48 | #include "ordered-data.h" |
95819c05 | 49 | #include "xattr.h" |
e02119d5 | 50 | #include "tree-log.h" |
c8b97818 | 51 | #include "compression.h" |
b4ce94de | 52 | #include "locking.h" |
dc89e982 | 53 | #include "free-space-cache.h" |
39279cc3 CM |
54 | |
55 | struct btrfs_iget_args { | |
56 | u64 ino; | |
57 | struct btrfs_root *root; | |
58 | }; | |
59 | ||
6e1d5dcc AD |
60 | static const struct inode_operations btrfs_dir_inode_operations; |
61 | static const struct inode_operations btrfs_symlink_inode_operations; | |
62 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
63 | static const struct inode_operations btrfs_special_inode_operations; | |
64 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
65 | static const struct address_space_operations btrfs_aops; |
66 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 67 | static const 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; | |
39279cc3 | 73 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 74 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
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 | ||
a41ad394 JB |
87 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
88 | static int btrfs_truncate(struct inode *inode); | |
c8b97818 | 89 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
90 | static noinline int cow_file_range(struct inode *inode, |
91 | struct page *locked_page, | |
92 | u64 start, u64 end, int *page_started, | |
93 | unsigned long *nr_written, int unlock); | |
7b128766 | 94 | |
f34f57a3 YZ |
95 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
96 | struct inode *inode, struct inode *dir) | |
0279b4cd JO |
97 | { |
98 | int err; | |
99 | ||
f34f57a3 | 100 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 101 | if (!err) |
f34f57a3 | 102 | err = btrfs_xattr_security_init(trans, inode, dir); |
0279b4cd JO |
103 | return err; |
104 | } | |
105 | ||
c8b97818 CM |
106 | /* |
107 | * this does all the hard work for inserting an inline extent into | |
108 | * the btree. The caller should have done a btrfs_drop_extents so that | |
109 | * no overlapping inline items exist in the btree | |
110 | */ | |
d397712b | 111 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
112 | struct btrfs_root *root, struct inode *inode, |
113 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 114 | int compress_type, |
c8b97818 CM |
115 | struct page **compressed_pages) |
116 | { | |
117 | struct btrfs_key key; | |
118 | struct btrfs_path *path; | |
119 | struct extent_buffer *leaf; | |
120 | struct page *page = NULL; | |
121 | char *kaddr; | |
122 | unsigned long ptr; | |
123 | struct btrfs_file_extent_item *ei; | |
124 | int err = 0; | |
125 | int ret; | |
126 | size_t cur_size = size; | |
127 | size_t datasize; | |
128 | unsigned long offset; | |
c8b97818 | 129 | |
fe3f566c | 130 | if (compressed_size && compressed_pages) |
c8b97818 | 131 | cur_size = compressed_size; |
c8b97818 | 132 | |
d397712b CM |
133 | path = btrfs_alloc_path(); |
134 | if (!path) | |
c8b97818 CM |
135 | return -ENOMEM; |
136 | ||
b9473439 | 137 | path->leave_spinning = 1; |
c8b97818 CM |
138 | btrfs_set_trans_block_group(trans, inode); |
139 | ||
140 | key.objectid = inode->i_ino; | |
141 | key.offset = start; | |
142 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
143 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
144 | ||
145 | inode_add_bytes(inode, size); | |
146 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
147 | datasize); | |
148 | BUG_ON(ret); | |
149 | if (ret) { | |
150 | err = ret; | |
c8b97818 CM |
151 | goto fail; |
152 | } | |
153 | leaf = path->nodes[0]; | |
154 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
155 | struct btrfs_file_extent_item); | |
156 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
157 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
158 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
159 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
160 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
161 | ptr = btrfs_file_extent_inline_start(ei); | |
162 | ||
261507a0 | 163 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
164 | struct page *cpage; |
165 | int i = 0; | |
d397712b | 166 | while (compressed_size > 0) { |
c8b97818 | 167 | cpage = compressed_pages[i]; |
5b050f04 | 168 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
169 | PAGE_CACHE_SIZE); |
170 | ||
b9473439 | 171 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 172 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 173 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
174 | |
175 | i++; | |
176 | ptr += cur_size; | |
177 | compressed_size -= cur_size; | |
178 | } | |
179 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 180 | compress_type); |
c8b97818 CM |
181 | } else { |
182 | page = find_get_page(inode->i_mapping, | |
183 | start >> PAGE_CACHE_SHIFT); | |
184 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
185 | kaddr = kmap_atomic(page, KM_USER0); | |
186 | offset = start & (PAGE_CACHE_SIZE - 1); | |
187 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
188 | kunmap_atomic(kaddr, KM_USER0); | |
189 | page_cache_release(page); | |
190 | } | |
191 | btrfs_mark_buffer_dirty(leaf); | |
192 | btrfs_free_path(path); | |
193 | ||
c2167754 YZ |
194 | /* |
195 | * we're an inline extent, so nobody can | |
196 | * extend the file past i_size without locking | |
197 | * a page we already have locked. | |
198 | * | |
199 | * We must do any isize and inode updates | |
200 | * before we unlock the pages. Otherwise we | |
201 | * could end up racing with unlink. | |
202 | */ | |
c8b97818 CM |
203 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
204 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 205 | |
c8b97818 CM |
206 | return 0; |
207 | fail: | |
208 | btrfs_free_path(path); | |
209 | return err; | |
210 | } | |
211 | ||
212 | ||
213 | /* | |
214 | * conditionally insert an inline extent into the file. This | |
215 | * does the checks required to make sure the data is small enough | |
216 | * to fit as an inline extent. | |
217 | */ | |
7f366cfe | 218 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
219 | struct btrfs_root *root, |
220 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 221 | size_t compressed_size, int compress_type, |
c8b97818 CM |
222 | struct page **compressed_pages) |
223 | { | |
224 | u64 isize = i_size_read(inode); | |
225 | u64 actual_end = min(end + 1, isize); | |
226 | u64 inline_len = actual_end - start; | |
227 | u64 aligned_end = (end + root->sectorsize - 1) & | |
228 | ~((u64)root->sectorsize - 1); | |
229 | u64 hint_byte; | |
230 | u64 data_len = inline_len; | |
231 | int ret; | |
232 | ||
233 | if (compressed_size) | |
234 | data_len = compressed_size; | |
235 | ||
236 | if (start > 0 || | |
70b99e69 | 237 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
238 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
239 | (!compressed_size && | |
240 | (actual_end & (root->sectorsize - 1)) == 0) || | |
241 | end + 1 < isize || | |
242 | data_len > root->fs_info->max_inline) { | |
243 | return 1; | |
244 | } | |
245 | ||
920bbbfb | 246 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 247 | &hint_byte, 1); |
c8b97818 CM |
248 | BUG_ON(ret); |
249 | ||
250 | if (isize > actual_end) | |
251 | inline_len = min_t(u64, isize, actual_end); | |
252 | ret = insert_inline_extent(trans, root, inode, start, | |
253 | inline_len, compressed_size, | |
fe3f566c | 254 | compress_type, compressed_pages); |
c8b97818 | 255 | BUG_ON(ret); |
0ca1f7ce | 256 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 257 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
258 | return 0; |
259 | } | |
260 | ||
771ed689 CM |
261 | struct async_extent { |
262 | u64 start; | |
263 | u64 ram_size; | |
264 | u64 compressed_size; | |
265 | struct page **pages; | |
266 | unsigned long nr_pages; | |
261507a0 | 267 | int compress_type; |
771ed689 CM |
268 | struct list_head list; |
269 | }; | |
270 | ||
271 | struct async_cow { | |
272 | struct inode *inode; | |
273 | struct btrfs_root *root; | |
274 | struct page *locked_page; | |
275 | u64 start; | |
276 | u64 end; | |
277 | struct list_head extents; | |
278 | struct btrfs_work work; | |
279 | }; | |
280 | ||
281 | static noinline int add_async_extent(struct async_cow *cow, | |
282 | u64 start, u64 ram_size, | |
283 | u64 compressed_size, | |
284 | struct page **pages, | |
261507a0 LZ |
285 | unsigned long nr_pages, |
286 | int compress_type) | |
771ed689 CM |
287 | { |
288 | struct async_extent *async_extent; | |
289 | ||
290 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
dac97e51 | 291 | BUG_ON(!async_extent); |
771ed689 CM |
292 | async_extent->start = start; |
293 | async_extent->ram_size = ram_size; | |
294 | async_extent->compressed_size = compressed_size; | |
295 | async_extent->pages = pages; | |
296 | async_extent->nr_pages = nr_pages; | |
261507a0 | 297 | async_extent->compress_type = compress_type; |
771ed689 CM |
298 | list_add_tail(&async_extent->list, &cow->extents); |
299 | return 0; | |
300 | } | |
301 | ||
d352ac68 | 302 | /* |
771ed689 CM |
303 | * we create compressed extents in two phases. The first |
304 | * phase compresses a range of pages that have already been | |
305 | * locked (both pages and state bits are locked). | |
c8b97818 | 306 | * |
771ed689 CM |
307 | * This is done inside an ordered work queue, and the compression |
308 | * is spread across many cpus. The actual IO submission is step | |
309 | * two, and the ordered work queue takes care of making sure that | |
310 | * happens in the same order things were put onto the queue by | |
311 | * writepages and friends. | |
c8b97818 | 312 | * |
771ed689 CM |
313 | * If this code finds it can't get good compression, it puts an |
314 | * entry onto the work queue to write the uncompressed bytes. This | |
315 | * makes sure that both compressed inodes and uncompressed inodes | |
316 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 317 | */ |
771ed689 CM |
318 | static noinline int compress_file_range(struct inode *inode, |
319 | struct page *locked_page, | |
320 | u64 start, u64 end, | |
321 | struct async_cow *async_cow, | |
322 | int *num_added) | |
b888db2b CM |
323 | { |
324 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
325 | struct btrfs_trans_handle *trans; | |
db94535d | 326 | u64 num_bytes; |
db94535d | 327 | u64 blocksize = root->sectorsize; |
c8b97818 | 328 | u64 actual_end; |
42dc7bab | 329 | u64 isize = i_size_read(inode); |
e6dcd2dc | 330 | int ret = 0; |
c8b97818 CM |
331 | struct page **pages = NULL; |
332 | unsigned long nr_pages; | |
333 | unsigned long nr_pages_ret = 0; | |
334 | unsigned long total_compressed = 0; | |
335 | unsigned long total_in = 0; | |
336 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 337 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
338 | int i; |
339 | int will_compress; | |
261507a0 | 340 | int compress_type = root->fs_info->compress_type; |
b888db2b | 341 | |
42dc7bab | 342 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
343 | again: |
344 | will_compress = 0; | |
345 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
346 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 347 | |
f03d9301 CM |
348 | /* |
349 | * we don't want to send crud past the end of i_size through | |
350 | * compression, that's just a waste of CPU time. So, if the | |
351 | * end of the file is before the start of our current | |
352 | * requested range of bytes, we bail out to the uncompressed | |
353 | * cleanup code that can deal with all of this. | |
354 | * | |
355 | * It isn't really the fastest way to fix things, but this is a | |
356 | * very uncommon corner. | |
357 | */ | |
358 | if (actual_end <= start) | |
359 | goto cleanup_and_bail_uncompressed; | |
360 | ||
c8b97818 CM |
361 | total_compressed = actual_end - start; |
362 | ||
363 | /* we want to make sure that amount of ram required to uncompress | |
364 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
365 | * of a compressed extent to 128k. This is a crucial number |
366 | * because it also controls how easily we can spread reads across | |
367 | * cpus for decompression. | |
368 | * | |
369 | * We also want to make sure the amount of IO required to do | |
370 | * a random read is reasonably small, so we limit the size of | |
371 | * a compressed extent to 128k. | |
c8b97818 CM |
372 | */ |
373 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 374 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 375 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
376 | total_in = 0; |
377 | ret = 0; | |
db94535d | 378 | |
771ed689 CM |
379 | /* |
380 | * we do compression for mount -o compress and when the | |
381 | * inode has not been flagged as nocompress. This flag can | |
382 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 383 | */ |
6cbff00f | 384 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 385 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
386 | (BTRFS_I(inode)->force_compress) || |
387 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 388 | WARN_ON(pages); |
cfbc246e | 389 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
dac97e51 | 390 | BUG_ON(!pages); |
c8b97818 | 391 | |
261507a0 LZ |
392 | if (BTRFS_I(inode)->force_compress) |
393 | compress_type = BTRFS_I(inode)->force_compress; | |
394 | ||
395 | ret = btrfs_compress_pages(compress_type, | |
396 | inode->i_mapping, start, | |
397 | total_compressed, pages, | |
398 | nr_pages, &nr_pages_ret, | |
399 | &total_in, | |
400 | &total_compressed, | |
401 | max_compressed); | |
c8b97818 CM |
402 | |
403 | if (!ret) { | |
404 | unsigned long offset = total_compressed & | |
405 | (PAGE_CACHE_SIZE - 1); | |
406 | struct page *page = pages[nr_pages_ret - 1]; | |
407 | char *kaddr; | |
408 | ||
409 | /* zero the tail end of the last page, we might be | |
410 | * sending it down to disk | |
411 | */ | |
412 | if (offset) { | |
413 | kaddr = kmap_atomic(page, KM_USER0); | |
414 | memset(kaddr + offset, 0, | |
415 | PAGE_CACHE_SIZE - offset); | |
416 | kunmap_atomic(kaddr, KM_USER0); | |
417 | } | |
418 | will_compress = 1; | |
419 | } | |
420 | } | |
421 | if (start == 0) { | |
771ed689 | 422 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 423 | BUG_ON(IS_ERR(trans)); |
771ed689 | 424 | btrfs_set_trans_block_group(trans, inode); |
0ca1f7ce | 425 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 426 | |
c8b97818 | 427 | /* lets try to make an inline extent */ |
771ed689 | 428 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 429 | /* we didn't compress the entire range, try |
771ed689 | 430 | * to make an uncompressed inline extent. |
c8b97818 CM |
431 | */ |
432 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 433 | start, end, 0, 0, NULL); |
c8b97818 | 434 | } else { |
771ed689 | 435 | /* try making a compressed inline extent */ |
c8b97818 CM |
436 | ret = cow_file_range_inline(trans, root, inode, |
437 | start, end, | |
fe3f566c LZ |
438 | total_compressed, |
439 | compress_type, pages); | |
c8b97818 CM |
440 | } |
441 | if (ret == 0) { | |
771ed689 CM |
442 | /* |
443 | * inline extent creation worked, we don't need | |
444 | * to create any more async work items. Unlock | |
445 | * and free up our temp pages. | |
446 | */ | |
c8b97818 | 447 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
448 | &BTRFS_I(inode)->io_tree, |
449 | start, end, NULL, | |
450 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 451 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 452 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
453 | |
454 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
455 | goto free_pages_out; |
456 | } | |
c2167754 | 457 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
458 | } |
459 | ||
460 | if (will_compress) { | |
461 | /* | |
462 | * we aren't doing an inline extent round the compressed size | |
463 | * up to a block size boundary so the allocator does sane | |
464 | * things | |
465 | */ | |
466 | total_compressed = (total_compressed + blocksize - 1) & | |
467 | ~(blocksize - 1); | |
468 | ||
469 | /* | |
470 | * one last check to make sure the compression is really a | |
471 | * win, compare the page count read with the blocks on disk | |
472 | */ | |
473 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
474 | ~(PAGE_CACHE_SIZE - 1); | |
475 | if (total_compressed >= total_in) { | |
476 | will_compress = 0; | |
477 | } else { | |
c8b97818 CM |
478 | num_bytes = total_in; |
479 | } | |
480 | } | |
481 | if (!will_compress && pages) { | |
482 | /* | |
483 | * the compression code ran but failed to make things smaller, | |
484 | * free any pages it allocated and our page pointer array | |
485 | */ | |
486 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 487 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
488 | page_cache_release(pages[i]); |
489 | } | |
490 | kfree(pages); | |
491 | pages = NULL; | |
492 | total_compressed = 0; | |
493 | nr_pages_ret = 0; | |
494 | ||
495 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
496 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
497 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 498 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 499 | } |
c8b97818 | 500 | } |
771ed689 CM |
501 | if (will_compress) { |
502 | *num_added += 1; | |
c8b97818 | 503 | |
771ed689 CM |
504 | /* the async work queues will take care of doing actual |
505 | * allocation on disk for these compressed pages, | |
506 | * and will submit them to the elevator. | |
507 | */ | |
508 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
509 | total_compressed, pages, nr_pages_ret, |
510 | compress_type); | |
179e29e4 | 511 | |
24ae6365 | 512 | if (start + num_bytes < end) { |
771ed689 CM |
513 | start += num_bytes; |
514 | pages = NULL; | |
515 | cond_resched(); | |
516 | goto again; | |
517 | } | |
518 | } else { | |
f03d9301 | 519 | cleanup_and_bail_uncompressed: |
771ed689 CM |
520 | /* |
521 | * No compression, but we still need to write the pages in | |
522 | * the file we've been given so far. redirty the locked | |
523 | * page if it corresponds to our extent and set things up | |
524 | * for the async work queue to run cow_file_range to do | |
525 | * the normal delalloc dance | |
526 | */ | |
527 | if (page_offset(locked_page) >= start && | |
528 | page_offset(locked_page) <= end) { | |
529 | __set_page_dirty_nobuffers(locked_page); | |
530 | /* unlocked later on in the async handlers */ | |
531 | } | |
261507a0 LZ |
532 | add_async_extent(async_cow, start, end - start + 1, |
533 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
534 | *num_added += 1; |
535 | } | |
3b951516 | 536 | |
771ed689 CM |
537 | out: |
538 | return 0; | |
539 | ||
540 | free_pages_out: | |
541 | for (i = 0; i < nr_pages_ret; i++) { | |
542 | WARN_ON(pages[i]->mapping); | |
543 | page_cache_release(pages[i]); | |
544 | } | |
d397712b | 545 | kfree(pages); |
771ed689 CM |
546 | |
547 | goto out; | |
548 | } | |
549 | ||
550 | /* | |
551 | * phase two of compressed writeback. This is the ordered portion | |
552 | * of the code, which only gets called in the order the work was | |
553 | * queued. We walk all the async extents created by compress_file_range | |
554 | * and send them down to the disk. | |
555 | */ | |
556 | static noinline int submit_compressed_extents(struct inode *inode, | |
557 | struct async_cow *async_cow) | |
558 | { | |
559 | struct async_extent *async_extent; | |
560 | u64 alloc_hint = 0; | |
561 | struct btrfs_trans_handle *trans; | |
562 | struct btrfs_key ins; | |
563 | struct extent_map *em; | |
564 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
565 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
566 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 567 | int ret = 0; |
771ed689 CM |
568 | |
569 | if (list_empty(&async_cow->extents)) | |
570 | return 0; | |
571 | ||
771ed689 | 572 | |
d397712b | 573 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
574 | async_extent = list_entry(async_cow->extents.next, |
575 | struct async_extent, list); | |
576 | list_del(&async_extent->list); | |
c8b97818 | 577 | |
771ed689 CM |
578 | io_tree = &BTRFS_I(inode)->io_tree; |
579 | ||
f5a84ee3 | 580 | retry: |
771ed689 CM |
581 | /* did the compression code fall back to uncompressed IO? */ |
582 | if (!async_extent->pages) { | |
583 | int page_started = 0; | |
584 | unsigned long nr_written = 0; | |
585 | ||
586 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 JB |
587 | async_extent->start + |
588 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
589 | |
590 | /* allocate blocks */ | |
f5a84ee3 JB |
591 | ret = cow_file_range(inode, async_cow->locked_page, |
592 | async_extent->start, | |
593 | async_extent->start + | |
594 | async_extent->ram_size - 1, | |
595 | &page_started, &nr_written, 0); | |
771ed689 CM |
596 | |
597 | /* | |
598 | * if page_started, cow_file_range inserted an | |
599 | * inline extent and took care of all the unlocking | |
600 | * and IO for us. Otherwise, we need to submit | |
601 | * all those pages down to the drive. | |
602 | */ | |
f5a84ee3 | 603 | if (!page_started && !ret) |
771ed689 CM |
604 | extent_write_locked_range(io_tree, |
605 | inode, async_extent->start, | |
d397712b | 606 | async_extent->start + |
771ed689 CM |
607 | async_extent->ram_size - 1, |
608 | btrfs_get_extent, | |
609 | WB_SYNC_ALL); | |
610 | kfree(async_extent); | |
611 | cond_resched(); | |
612 | continue; | |
613 | } | |
614 | ||
615 | lock_extent(io_tree, async_extent->start, | |
616 | async_extent->start + async_extent->ram_size - 1, | |
617 | GFP_NOFS); | |
771ed689 | 618 | |
c2167754 | 619 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 620 | BUG_ON(IS_ERR(trans)); |
771ed689 CM |
621 | ret = btrfs_reserve_extent(trans, root, |
622 | async_extent->compressed_size, | |
623 | async_extent->compressed_size, | |
624 | 0, alloc_hint, | |
625 | (u64)-1, &ins, 1); | |
c2167754 YZ |
626 | btrfs_end_transaction(trans, root); |
627 | ||
f5a84ee3 JB |
628 | if (ret) { |
629 | int i; | |
630 | for (i = 0; i < async_extent->nr_pages; i++) { | |
631 | WARN_ON(async_extent->pages[i]->mapping); | |
632 | page_cache_release(async_extent->pages[i]); | |
633 | } | |
634 | kfree(async_extent->pages); | |
635 | async_extent->nr_pages = 0; | |
636 | async_extent->pages = NULL; | |
637 | unlock_extent(io_tree, async_extent->start, | |
638 | async_extent->start + | |
639 | async_extent->ram_size - 1, GFP_NOFS); | |
640 | goto retry; | |
641 | } | |
642 | ||
c2167754 YZ |
643 | /* |
644 | * here we're doing allocation and writeback of the | |
645 | * compressed pages | |
646 | */ | |
647 | btrfs_drop_extent_cache(inode, async_extent->start, | |
648 | async_extent->start + | |
649 | async_extent->ram_size - 1, 0); | |
650 | ||
771ed689 | 651 | em = alloc_extent_map(GFP_NOFS); |
c26a9203 | 652 | BUG_ON(!em); |
771ed689 CM |
653 | em->start = async_extent->start; |
654 | em->len = async_extent->ram_size; | |
445a6944 | 655 | em->orig_start = em->start; |
c8b97818 | 656 | |
771ed689 CM |
657 | em->block_start = ins.objectid; |
658 | em->block_len = ins.offset; | |
659 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 660 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
661 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
662 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
663 | ||
d397712b | 664 | while (1) { |
890871be | 665 | write_lock(&em_tree->lock); |
771ed689 | 666 | ret = add_extent_mapping(em_tree, em); |
890871be | 667 | write_unlock(&em_tree->lock); |
771ed689 CM |
668 | if (ret != -EEXIST) { |
669 | free_extent_map(em); | |
670 | break; | |
671 | } | |
672 | btrfs_drop_extent_cache(inode, async_extent->start, | |
673 | async_extent->start + | |
674 | async_extent->ram_size - 1, 0); | |
675 | } | |
676 | ||
261507a0 LZ |
677 | ret = btrfs_add_ordered_extent_compress(inode, |
678 | async_extent->start, | |
679 | ins.objectid, | |
680 | async_extent->ram_size, | |
681 | ins.offset, | |
682 | BTRFS_ORDERED_COMPRESSED, | |
683 | async_extent->compress_type); | |
771ed689 CM |
684 | BUG_ON(ret); |
685 | ||
771ed689 CM |
686 | /* |
687 | * clear dirty, set writeback and unlock the pages. | |
688 | */ | |
689 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
690 | &BTRFS_I(inode)->io_tree, |
691 | async_extent->start, | |
692 | async_extent->start + | |
693 | async_extent->ram_size - 1, | |
694 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
695 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 696 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 697 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
698 | |
699 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
700 | async_extent->start, |
701 | async_extent->ram_size, | |
702 | ins.objectid, | |
703 | ins.offset, async_extent->pages, | |
704 | async_extent->nr_pages); | |
771ed689 CM |
705 | |
706 | BUG_ON(ret); | |
771ed689 CM |
707 | alloc_hint = ins.objectid + ins.offset; |
708 | kfree(async_extent); | |
709 | cond_resched(); | |
710 | } | |
711 | ||
771ed689 CM |
712 | return 0; |
713 | } | |
714 | ||
4b46fce2 JB |
715 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
716 | u64 num_bytes) | |
717 | { | |
718 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
719 | struct extent_map *em; | |
720 | u64 alloc_hint = 0; | |
721 | ||
722 | read_lock(&em_tree->lock); | |
723 | em = search_extent_mapping(em_tree, start, num_bytes); | |
724 | if (em) { | |
725 | /* | |
726 | * if block start isn't an actual block number then find the | |
727 | * first block in this inode and use that as a hint. If that | |
728 | * block is also bogus then just don't worry about it. | |
729 | */ | |
730 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
731 | free_extent_map(em); | |
732 | em = search_extent_mapping(em_tree, 0, 0); | |
733 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
734 | alloc_hint = em->block_start; | |
735 | if (em) | |
736 | free_extent_map(em); | |
737 | } else { | |
738 | alloc_hint = em->block_start; | |
739 | free_extent_map(em); | |
740 | } | |
741 | } | |
742 | read_unlock(&em_tree->lock); | |
743 | ||
744 | return alloc_hint; | |
745 | } | |
746 | ||
771ed689 CM |
747 | /* |
748 | * when extent_io.c finds a delayed allocation range in the file, | |
749 | * the call backs end up in this code. The basic idea is to | |
750 | * allocate extents on disk for the range, and create ordered data structs | |
751 | * in ram to track those extents. | |
752 | * | |
753 | * locked_page is the page that writepage had locked already. We use | |
754 | * it to make sure we don't do extra locks or unlocks. | |
755 | * | |
756 | * *page_started is set to one if we unlock locked_page and do everything | |
757 | * required to start IO on it. It may be clean and already done with | |
758 | * IO when we return. | |
759 | */ | |
760 | static noinline int cow_file_range(struct inode *inode, | |
761 | struct page *locked_page, | |
762 | u64 start, u64 end, int *page_started, | |
763 | unsigned long *nr_written, | |
764 | int unlock) | |
765 | { | |
766 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
767 | struct btrfs_trans_handle *trans; | |
768 | u64 alloc_hint = 0; | |
769 | u64 num_bytes; | |
770 | unsigned long ram_size; | |
771 | u64 disk_num_bytes; | |
772 | u64 cur_alloc_size; | |
773 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
774 | struct btrfs_key ins; |
775 | struct extent_map *em; | |
776 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
777 | int ret = 0; | |
778 | ||
0cb59c99 | 779 | BUG_ON(root == root->fs_info->tree_root); |
771ed689 | 780 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 781 | BUG_ON(IS_ERR(trans)); |
771ed689 | 782 | btrfs_set_trans_block_group(trans, inode); |
0ca1f7ce | 783 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 784 | |
771ed689 CM |
785 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
786 | num_bytes = max(blocksize, num_bytes); | |
787 | disk_num_bytes = num_bytes; | |
788 | ret = 0; | |
789 | ||
790 | if (start == 0) { | |
791 | /* lets try to make an inline extent */ | |
792 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 793 | start, end, 0, 0, NULL); |
771ed689 CM |
794 | if (ret == 0) { |
795 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
796 | &BTRFS_I(inode)->io_tree, |
797 | start, end, NULL, | |
798 | EXTENT_CLEAR_UNLOCK_PAGE | | |
799 | EXTENT_CLEAR_UNLOCK | | |
800 | EXTENT_CLEAR_DELALLOC | | |
801 | EXTENT_CLEAR_DIRTY | | |
802 | EXTENT_SET_WRITEBACK | | |
803 | EXTENT_END_WRITEBACK); | |
c2167754 | 804 | |
771ed689 CM |
805 | *nr_written = *nr_written + |
806 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
807 | *page_started = 1; | |
808 | ret = 0; | |
809 | goto out; | |
810 | } | |
811 | } | |
812 | ||
813 | BUG_ON(disk_num_bytes > | |
814 | btrfs_super_total_bytes(&root->fs_info->super_copy)); | |
815 | ||
4b46fce2 | 816 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
817 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
818 | ||
d397712b | 819 | while (disk_num_bytes > 0) { |
a791e35e CM |
820 | unsigned long op; |
821 | ||
287a0ab9 | 822 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 823 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 824 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 825 | (u64)-1, &ins, 1); |
d397712b CM |
826 | BUG_ON(ret); |
827 | ||
e6dcd2dc | 828 | em = alloc_extent_map(GFP_NOFS); |
c26a9203 | 829 | BUG_ON(!em); |
e6dcd2dc | 830 | em->start = start; |
445a6944 | 831 | em->orig_start = em->start; |
771ed689 CM |
832 | ram_size = ins.offset; |
833 | em->len = ins.offset; | |
c8b97818 | 834 | |
e6dcd2dc | 835 | em->block_start = ins.objectid; |
c8b97818 | 836 | em->block_len = ins.offset; |
e6dcd2dc | 837 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 838 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 839 | |
d397712b | 840 | while (1) { |
890871be | 841 | write_lock(&em_tree->lock); |
e6dcd2dc | 842 | ret = add_extent_mapping(em_tree, em); |
890871be | 843 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
844 | if (ret != -EEXIST) { |
845 | free_extent_map(em); | |
846 | break; | |
847 | } | |
848 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 849 | start + ram_size - 1, 0); |
e6dcd2dc CM |
850 | } |
851 | ||
98d20f67 | 852 | cur_alloc_size = ins.offset; |
e6dcd2dc | 853 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 854 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 855 | BUG_ON(ret); |
c8b97818 | 856 | |
17d217fe YZ |
857 | if (root->root_key.objectid == |
858 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
859 | ret = btrfs_reloc_clone_csums(inode, start, | |
860 | cur_alloc_size); | |
861 | BUG_ON(ret); | |
862 | } | |
863 | ||
d397712b | 864 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 865 | break; |
d397712b | 866 | |
c8b97818 CM |
867 | /* we're not doing compressed IO, don't unlock the first |
868 | * page (which the caller expects to stay locked), don't | |
869 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
870 | * |
871 | * Do set the Private2 bit so we know this page was properly | |
872 | * setup for writepage | |
c8b97818 | 873 | */ |
a791e35e CM |
874 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
875 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
876 | EXTENT_SET_PRIVATE2; | |
877 | ||
c8b97818 CM |
878 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
879 | start, start + ram_size - 1, | |
a791e35e | 880 | locked_page, op); |
c8b97818 | 881 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
882 | num_bytes -= cur_alloc_size; |
883 | alloc_hint = ins.objectid + ins.offset; | |
884 | start += cur_alloc_size; | |
b888db2b | 885 | } |
b888db2b | 886 | out: |
771ed689 | 887 | ret = 0; |
b888db2b | 888 | btrfs_end_transaction(trans, root); |
c8b97818 | 889 | |
be20aa9d | 890 | return ret; |
771ed689 | 891 | } |
c8b97818 | 892 | |
771ed689 CM |
893 | /* |
894 | * work queue call back to started compression on a file and pages | |
895 | */ | |
896 | static noinline void async_cow_start(struct btrfs_work *work) | |
897 | { | |
898 | struct async_cow *async_cow; | |
899 | int num_added = 0; | |
900 | async_cow = container_of(work, struct async_cow, work); | |
901 | ||
902 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
903 | async_cow->start, async_cow->end, async_cow, | |
904 | &num_added); | |
905 | if (num_added == 0) | |
906 | async_cow->inode = NULL; | |
907 | } | |
908 | ||
909 | /* | |
910 | * work queue call back to submit previously compressed pages | |
911 | */ | |
912 | static noinline void async_cow_submit(struct btrfs_work *work) | |
913 | { | |
914 | struct async_cow *async_cow; | |
915 | struct btrfs_root *root; | |
916 | unsigned long nr_pages; | |
917 | ||
918 | async_cow = container_of(work, struct async_cow, work); | |
919 | ||
920 | root = async_cow->root; | |
921 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
922 | PAGE_CACHE_SHIFT; | |
923 | ||
924 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
925 | ||
926 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
927 | 5 * 1042 * 1024 && | |
928 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
929 | wake_up(&root->fs_info->async_submit_wait); | |
930 | ||
d397712b | 931 | if (async_cow->inode) |
771ed689 | 932 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 933 | } |
c8b97818 | 934 | |
771ed689 CM |
935 | static noinline void async_cow_free(struct btrfs_work *work) |
936 | { | |
937 | struct async_cow *async_cow; | |
938 | async_cow = container_of(work, struct async_cow, work); | |
939 | kfree(async_cow); | |
940 | } | |
941 | ||
942 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
943 | u64 start, u64 end, int *page_started, | |
944 | unsigned long *nr_written) | |
945 | { | |
946 | struct async_cow *async_cow; | |
947 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
948 | unsigned long nr_pages; | |
949 | u64 cur_end; | |
950 | int limit = 10 * 1024 * 1042; | |
951 | ||
a3429ab7 CM |
952 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
953 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 954 | while (start < end) { |
771ed689 CM |
955 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
956 | async_cow->inode = inode; | |
957 | async_cow->root = root; | |
958 | async_cow->locked_page = locked_page; | |
959 | async_cow->start = start; | |
960 | ||
6cbff00f | 961 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
962 | cur_end = end; |
963 | else | |
964 | cur_end = min(end, start + 512 * 1024 - 1); | |
965 | ||
966 | async_cow->end = cur_end; | |
967 | INIT_LIST_HEAD(&async_cow->extents); | |
968 | ||
969 | async_cow->work.func = async_cow_start; | |
970 | async_cow->work.ordered_func = async_cow_submit; | |
971 | async_cow->work.ordered_free = async_cow_free; | |
972 | async_cow->work.flags = 0; | |
973 | ||
771ed689 CM |
974 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
975 | PAGE_CACHE_SHIFT; | |
976 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
977 | ||
978 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
979 | &async_cow->work); | |
980 | ||
981 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
982 | wait_event(root->fs_info->async_submit_wait, | |
983 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
984 | limit)); | |
985 | } | |
986 | ||
d397712b | 987 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
988 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
989 | wait_event(root->fs_info->async_submit_wait, | |
990 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
991 | 0)); | |
992 | } | |
993 | ||
994 | *nr_written += nr_pages; | |
995 | start = cur_end + 1; | |
996 | } | |
997 | *page_started = 1; | |
998 | return 0; | |
be20aa9d CM |
999 | } |
1000 | ||
d397712b | 1001 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1002 | u64 bytenr, u64 num_bytes) |
1003 | { | |
1004 | int ret; | |
1005 | struct btrfs_ordered_sum *sums; | |
1006 | LIST_HEAD(list); | |
1007 | ||
07d400a6 YZ |
1008 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
1009 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
1010 | if (ret == 0 && list_empty(&list)) |
1011 | return 0; | |
1012 | ||
1013 | while (!list_empty(&list)) { | |
1014 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1015 | list_del(&sums->list); | |
1016 | kfree(sums); | |
1017 | } | |
1018 | return 1; | |
1019 | } | |
1020 | ||
d352ac68 CM |
1021 | /* |
1022 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1023 | * of the extents that exist in the file, and COWs the file as required. | |
1024 | * | |
1025 | * If no cow copies or snapshots exist, we write directly to the existing | |
1026 | * blocks on disk | |
1027 | */ | |
7f366cfe CM |
1028 | static noinline int run_delalloc_nocow(struct inode *inode, |
1029 | struct page *locked_page, | |
771ed689 CM |
1030 | u64 start, u64 end, int *page_started, int force, |
1031 | unsigned long *nr_written) | |
be20aa9d | 1032 | { |
be20aa9d | 1033 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1034 | struct btrfs_trans_handle *trans; |
be20aa9d | 1035 | struct extent_buffer *leaf; |
be20aa9d | 1036 | struct btrfs_path *path; |
80ff3856 | 1037 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1038 | struct btrfs_key found_key; |
80ff3856 YZ |
1039 | u64 cow_start; |
1040 | u64 cur_offset; | |
1041 | u64 extent_end; | |
5d4f98a2 | 1042 | u64 extent_offset; |
80ff3856 YZ |
1043 | u64 disk_bytenr; |
1044 | u64 num_bytes; | |
1045 | int extent_type; | |
1046 | int ret; | |
d899e052 | 1047 | int type; |
80ff3856 YZ |
1048 | int nocow; |
1049 | int check_prev = 1; | |
0cb59c99 | 1050 | bool nolock = false; |
be20aa9d CM |
1051 | |
1052 | path = btrfs_alloc_path(); | |
1053 | BUG_ON(!path); | |
0cb59c99 JB |
1054 | if (root == root->fs_info->tree_root) { |
1055 | nolock = true; | |
1056 | trans = btrfs_join_transaction_nolock(root, 1); | |
1057 | } else { | |
1058 | trans = btrfs_join_transaction(root, 1); | |
1059 | } | |
3612b495 | 1060 | BUG_ON(IS_ERR(trans)); |
be20aa9d | 1061 | |
80ff3856 YZ |
1062 | cow_start = (u64)-1; |
1063 | cur_offset = start; | |
1064 | while (1) { | |
1065 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1066 | cur_offset, 0); | |
1067 | BUG_ON(ret < 0); | |
1068 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1069 | leaf = path->nodes[0]; | |
1070 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1071 | path->slots[0] - 1); | |
1072 | if (found_key.objectid == inode->i_ino && | |
1073 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1074 | path->slots[0]--; | |
1075 | } | |
1076 | check_prev = 0; | |
1077 | next_slot: | |
1078 | leaf = path->nodes[0]; | |
1079 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1080 | ret = btrfs_next_leaf(root, path); | |
1081 | if (ret < 0) | |
1082 | BUG_ON(1); | |
1083 | if (ret > 0) | |
1084 | break; | |
1085 | leaf = path->nodes[0]; | |
1086 | } | |
be20aa9d | 1087 | |
80ff3856 YZ |
1088 | nocow = 0; |
1089 | disk_bytenr = 0; | |
17d217fe | 1090 | num_bytes = 0; |
80ff3856 YZ |
1091 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1092 | ||
1093 | if (found_key.objectid > inode->i_ino || | |
1094 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1095 | found_key.offset > end) | |
1096 | break; | |
1097 | ||
1098 | if (found_key.offset > cur_offset) { | |
1099 | extent_end = found_key.offset; | |
e9061e21 | 1100 | extent_type = 0; |
80ff3856 YZ |
1101 | goto out_check; |
1102 | } | |
1103 | ||
1104 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1105 | struct btrfs_file_extent_item); | |
1106 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1107 | ||
d899e052 YZ |
1108 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1109 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1110 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1111 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1112 | extent_end = found_key.offset + |
1113 | btrfs_file_extent_num_bytes(leaf, fi); | |
1114 | if (extent_end <= start) { | |
1115 | path->slots[0]++; | |
1116 | goto next_slot; | |
1117 | } | |
17d217fe YZ |
1118 | if (disk_bytenr == 0) |
1119 | goto out_check; | |
80ff3856 YZ |
1120 | if (btrfs_file_extent_compression(leaf, fi) || |
1121 | btrfs_file_extent_encryption(leaf, fi) || | |
1122 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1123 | goto out_check; | |
d899e052 YZ |
1124 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1125 | goto out_check; | |
d2fb3437 | 1126 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1127 | goto out_check; |
17d217fe | 1128 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1129 | found_key.offset - |
1130 | extent_offset, disk_bytenr)) | |
17d217fe | 1131 | goto out_check; |
5d4f98a2 | 1132 | disk_bytenr += extent_offset; |
17d217fe YZ |
1133 | disk_bytenr += cur_offset - found_key.offset; |
1134 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1135 | /* | |
1136 | * force cow if csum exists in the range. | |
1137 | * this ensure that csum for a given extent are | |
1138 | * either valid or do not exist. | |
1139 | */ | |
1140 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1141 | goto out_check; | |
80ff3856 YZ |
1142 | nocow = 1; |
1143 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1144 | extent_end = found_key.offset + | |
1145 | btrfs_file_extent_inline_len(leaf, fi); | |
1146 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1147 | } else { | |
1148 | BUG_ON(1); | |
1149 | } | |
1150 | out_check: | |
1151 | if (extent_end <= start) { | |
1152 | path->slots[0]++; | |
1153 | goto next_slot; | |
1154 | } | |
1155 | if (!nocow) { | |
1156 | if (cow_start == (u64)-1) | |
1157 | cow_start = cur_offset; | |
1158 | cur_offset = extent_end; | |
1159 | if (cur_offset > end) | |
1160 | break; | |
1161 | path->slots[0]++; | |
1162 | goto next_slot; | |
7ea394f1 YZ |
1163 | } |
1164 | ||
1165 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1166 | if (cow_start != (u64)-1) { |
1167 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1168 | found_key.offset - 1, page_started, |
1169 | nr_written, 1); | |
80ff3856 YZ |
1170 | BUG_ON(ret); |
1171 | cow_start = (u64)-1; | |
7ea394f1 | 1172 | } |
80ff3856 | 1173 | |
d899e052 YZ |
1174 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1175 | struct extent_map *em; | |
1176 | struct extent_map_tree *em_tree; | |
1177 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1178 | em = alloc_extent_map(GFP_NOFS); | |
c26a9203 | 1179 | BUG_ON(!em); |
d899e052 | 1180 | em->start = cur_offset; |
445a6944 | 1181 | em->orig_start = em->start; |
d899e052 YZ |
1182 | em->len = num_bytes; |
1183 | em->block_len = num_bytes; | |
1184 | em->block_start = disk_bytenr; | |
1185 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1186 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1187 | while (1) { | |
890871be | 1188 | write_lock(&em_tree->lock); |
d899e052 | 1189 | ret = add_extent_mapping(em_tree, em); |
890871be | 1190 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1191 | if (ret != -EEXIST) { |
1192 | free_extent_map(em); | |
1193 | break; | |
1194 | } | |
1195 | btrfs_drop_extent_cache(inode, em->start, | |
1196 | em->start + em->len - 1, 0); | |
1197 | } | |
1198 | type = BTRFS_ORDERED_PREALLOC; | |
1199 | } else { | |
1200 | type = BTRFS_ORDERED_NOCOW; | |
1201 | } | |
80ff3856 YZ |
1202 | |
1203 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1204 | num_bytes, num_bytes, type); |
1205 | BUG_ON(ret); | |
771ed689 | 1206 | |
efa56464 YZ |
1207 | if (root->root_key.objectid == |
1208 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1209 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1210 | num_bytes); | |
1211 | BUG_ON(ret); | |
1212 | } | |
1213 | ||
d899e052 | 1214 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1215 | cur_offset, cur_offset + num_bytes - 1, |
1216 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1217 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1218 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1219 | cur_offset = extent_end; |
1220 | if (cur_offset > end) | |
1221 | break; | |
be20aa9d | 1222 | } |
80ff3856 YZ |
1223 | btrfs_release_path(root, path); |
1224 | ||
1225 | if (cur_offset <= end && cow_start == (u64)-1) | |
1226 | cow_start = cur_offset; | |
1227 | if (cow_start != (u64)-1) { | |
1228 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1229 | page_started, nr_written, 1); |
80ff3856 YZ |
1230 | BUG_ON(ret); |
1231 | } | |
1232 | ||
0cb59c99 JB |
1233 | if (nolock) { |
1234 | ret = btrfs_end_transaction_nolock(trans, root); | |
1235 | BUG_ON(ret); | |
1236 | } else { | |
1237 | ret = btrfs_end_transaction(trans, root); | |
1238 | BUG_ON(ret); | |
1239 | } | |
7ea394f1 | 1240 | btrfs_free_path(path); |
80ff3856 | 1241 | return 0; |
be20aa9d CM |
1242 | } |
1243 | ||
d352ac68 CM |
1244 | /* |
1245 | * extent_io.c call back to do delayed allocation processing | |
1246 | */ | |
c8b97818 | 1247 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1248 | u64 start, u64 end, int *page_started, |
1249 | unsigned long *nr_written) | |
be20aa9d | 1250 | { |
be20aa9d | 1251 | int ret; |
7f366cfe | 1252 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1253 | |
6cbff00f | 1254 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1255 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1256 | page_started, 1, nr_written); |
6cbff00f | 1257 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1258 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1259 | page_started, 0, nr_written); |
1e701a32 | 1260 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1261 | !(BTRFS_I(inode)->force_compress) && |
1262 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1263 | ret = cow_file_range(inode, locked_page, start, end, |
1264 | page_started, nr_written, 1); | |
be20aa9d | 1265 | else |
771ed689 | 1266 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1267 | page_started, nr_written); |
b888db2b CM |
1268 | return ret; |
1269 | } | |
1270 | ||
9ed74f2d | 1271 | static int btrfs_split_extent_hook(struct inode *inode, |
0ca1f7ce | 1272 | struct extent_state *orig, u64 split) |
9ed74f2d | 1273 | { |
0ca1f7ce | 1274 | /* not delalloc, ignore it */ |
9ed74f2d JB |
1275 | if (!(orig->state & EXTENT_DELALLOC)) |
1276 | return 0; | |
1277 | ||
0ca1f7ce | 1278 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1279 | return 0; |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1284 | * extents so we can keep track of new extents that are just merged onto old | |
1285 | * extents, such as when we are doing sequential writes, so we can properly | |
1286 | * account for the metadata space we'll need. | |
1287 | */ | |
1288 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1289 | struct extent_state *new, | |
1290 | struct extent_state *other) | |
1291 | { | |
9ed74f2d JB |
1292 | /* not delalloc, ignore it */ |
1293 | if (!(other->state & EXTENT_DELALLOC)) | |
1294 | return 0; | |
1295 | ||
0ca1f7ce | 1296 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1297 | return 0; |
1298 | } | |
1299 | ||
d352ac68 CM |
1300 | /* |
1301 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1302 | * bytes in this file, and to maintain the list of inodes that | |
1303 | * have pending delalloc work to be done. | |
1304 | */ | |
0ca1f7ce YZ |
1305 | static int btrfs_set_bit_hook(struct inode *inode, |
1306 | struct extent_state *state, int *bits) | |
291d673e | 1307 | { |
9ed74f2d | 1308 | |
75eff68e CM |
1309 | /* |
1310 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1311 | * but in this case, we are only testeing for the DELALLOC | |
1312 | * bit, which is only set or cleared with irqs on | |
1313 | */ | |
0ca1f7ce | 1314 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1315 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1316 | u64 len = state->end + 1 - state->start; |
0cb59c99 JB |
1317 | int do_list = (root->root_key.objectid != |
1318 | BTRFS_ROOT_TREE_OBJECTID); | |
9ed74f2d | 1319 | |
0ca1f7ce YZ |
1320 | if (*bits & EXTENT_FIRST_DELALLOC) |
1321 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1322 | else | |
1323 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); | |
287a0ab9 | 1324 | |
75eff68e | 1325 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1326 | BTRFS_I(inode)->delalloc_bytes += len; |
1327 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1328 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1329 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1330 | &root->fs_info->delalloc_inodes); | |
1331 | } | |
75eff68e | 1332 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1333 | } |
1334 | return 0; | |
1335 | } | |
1336 | ||
d352ac68 CM |
1337 | /* |
1338 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1339 | */ | |
9ed74f2d | 1340 | static int btrfs_clear_bit_hook(struct inode *inode, |
0ca1f7ce | 1341 | struct extent_state *state, int *bits) |
291d673e | 1342 | { |
75eff68e CM |
1343 | /* |
1344 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1345 | * but in this case, we are only testeing for the DELALLOC | |
1346 | * bit, which is only set or cleared with irqs on | |
1347 | */ | |
0ca1f7ce | 1348 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1349 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1350 | u64 len = state->end + 1 - state->start; |
0cb59c99 JB |
1351 | int do_list = (root->root_key.objectid != |
1352 | BTRFS_ROOT_TREE_OBJECTID); | |
bcbfce8a | 1353 | |
0ca1f7ce YZ |
1354 | if (*bits & EXTENT_FIRST_DELALLOC) |
1355 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1356 | else if (!(*bits & EXTENT_DO_ACCOUNTING)) | |
1357 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
1358 | ||
1359 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1360 | btrfs_delalloc_release_metadata(inode, len); | |
1361 | ||
0cb59c99 JB |
1362 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1363 | && do_list) | |
0ca1f7ce | 1364 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1365 | |
75eff68e | 1366 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1367 | root->fs_info->delalloc_bytes -= len; |
1368 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1369 | ||
0cb59c99 | 1370 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1371 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1372 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1373 | } | |
75eff68e | 1374 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1375 | } |
1376 | return 0; | |
1377 | } | |
1378 | ||
d352ac68 CM |
1379 | /* |
1380 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1381 | * we don't create bios that span stripes or chunks | |
1382 | */ | |
239b14b3 | 1383 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1384 | size_t size, struct bio *bio, |
1385 | unsigned long bio_flags) | |
239b14b3 CM |
1386 | { |
1387 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1388 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1389 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1390 | u64 length = 0; |
1391 | u64 map_length; | |
239b14b3 CM |
1392 | int ret; |
1393 | ||
771ed689 CM |
1394 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1395 | return 0; | |
1396 | ||
f2d8d74d | 1397 | length = bio->bi_size; |
239b14b3 CM |
1398 | map_tree = &root->fs_info->mapping_tree; |
1399 | map_length = length; | |
cea9e445 | 1400 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1401 | &map_length, NULL, 0); |
cea9e445 | 1402 | |
d397712b | 1403 | if (map_length < length + size) |
239b14b3 | 1404 | return 1; |
411fc6bc | 1405 | return ret; |
239b14b3 CM |
1406 | } |
1407 | ||
d352ac68 CM |
1408 | /* |
1409 | * in order to insert checksums into the metadata in large chunks, | |
1410 | * we wait until bio submission time. All the pages in the bio are | |
1411 | * checksummed and sums are attached onto the ordered extent record. | |
1412 | * | |
1413 | * At IO completion time the cums attached on the ordered extent record | |
1414 | * are inserted into the btree | |
1415 | */ | |
d397712b CM |
1416 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1417 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1418 | unsigned long bio_flags, |
1419 | u64 bio_offset) | |
065631f6 | 1420 | { |
065631f6 | 1421 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1422 | int ret = 0; |
e015640f | 1423 | |
d20f7043 | 1424 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1425 | BUG_ON(ret); |
4a69a410 CM |
1426 | return 0; |
1427 | } | |
e015640f | 1428 | |
4a69a410 CM |
1429 | /* |
1430 | * in order to insert checksums into the metadata in large chunks, | |
1431 | * we wait until bio submission time. All the pages in the bio are | |
1432 | * checksummed and sums are attached onto the ordered extent record. | |
1433 | * | |
1434 | * At IO completion time the cums attached on the ordered extent record | |
1435 | * are inserted into the btree | |
1436 | */ | |
b2950863 | 1437 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1438 | int mirror_num, unsigned long bio_flags, |
1439 | u64 bio_offset) | |
4a69a410 CM |
1440 | { |
1441 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1442 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1443 | } |
1444 | ||
d352ac68 | 1445 | /* |
cad321ad CM |
1446 | * extent_io.c submission hook. This does the right thing for csum calculation |
1447 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1448 | */ |
b2950863 | 1449 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1450 | int mirror_num, unsigned long bio_flags, |
1451 | u64 bio_offset) | |
44b8bd7e CM |
1452 | { |
1453 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1454 | int ret = 0; | |
19b9bdb0 | 1455 | int skip_sum; |
44b8bd7e | 1456 | |
6cbff00f | 1457 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1458 | |
0cb59c99 JB |
1459 | if (root == root->fs_info->tree_root) |
1460 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2); | |
1461 | else | |
1462 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
e6dcd2dc | 1463 | BUG_ON(ret); |
065631f6 | 1464 | |
7b6d91da | 1465 | if (!(rw & REQ_WRITE)) { |
d20f7043 | 1466 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1467 | return btrfs_submit_compressed_read(inode, bio, |
1468 | mirror_num, bio_flags); | |
c2db1073 TI |
1469 | } else if (!skip_sum) { |
1470 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1471 | if (ret) | |
1472 | return ret; | |
1473 | } | |
4d1b5fb4 | 1474 | goto mapit; |
19b9bdb0 | 1475 | } else if (!skip_sum) { |
17d217fe YZ |
1476 | /* csum items have already been cloned */ |
1477 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1478 | goto mapit; | |
19b9bdb0 CM |
1479 | /* we're doing a write, do the async checksumming */ |
1480 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1481 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1482 | bio_flags, bio_offset, |
1483 | __btrfs_submit_bio_start, | |
4a69a410 | 1484 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1485 | } |
1486 | ||
0b86a832 | 1487 | mapit: |
8b712842 | 1488 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1489 | } |
6885f308 | 1490 | |
d352ac68 CM |
1491 | /* |
1492 | * given a list of ordered sums record them in the inode. This happens | |
1493 | * at IO completion time based on sums calculated at bio submission time. | |
1494 | */ | |
ba1da2f4 | 1495 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1496 | struct inode *inode, u64 file_offset, |
1497 | struct list_head *list) | |
1498 | { | |
e6dcd2dc CM |
1499 | struct btrfs_ordered_sum *sum; |
1500 | ||
1501 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1502 | |
1503 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1504 | btrfs_csum_file_blocks(trans, |
1505 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1506 | } |
1507 | return 0; | |
1508 | } | |
1509 | ||
2ac55d41 JB |
1510 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1511 | struct extent_state **cached_state) | |
ea8c2819 | 1512 | { |
d397712b | 1513 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1514 | WARN_ON(1); |
ea8c2819 | 1515 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1516 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1517 | } |
1518 | ||
d352ac68 | 1519 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1520 | struct btrfs_writepage_fixup { |
1521 | struct page *page; | |
1522 | struct btrfs_work work; | |
1523 | }; | |
1524 | ||
b2950863 | 1525 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1526 | { |
1527 | struct btrfs_writepage_fixup *fixup; | |
1528 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1529 | struct extent_state *cached_state = NULL; |
247e743c CM |
1530 | struct page *page; |
1531 | struct inode *inode; | |
1532 | u64 page_start; | |
1533 | u64 page_end; | |
1534 | ||
1535 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1536 | page = fixup->page; | |
4a096752 | 1537 | again: |
247e743c CM |
1538 | lock_page(page); |
1539 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1540 | ClearPageChecked(page); | |
1541 | goto out_page; | |
1542 | } | |
1543 | ||
1544 | inode = page->mapping->host; | |
1545 | page_start = page_offset(page); | |
1546 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1547 | ||
2ac55d41 JB |
1548 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1549 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1550 | |
1551 | /* already ordered? We're done */ | |
8b62b72b | 1552 | if (PagePrivate2(page)) |
247e743c | 1553 | goto out; |
4a096752 CM |
1554 | |
1555 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1556 | if (ordered) { | |
2ac55d41 JB |
1557 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1558 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1559 | unlock_page(page); |
1560 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1561 | goto again; | |
1562 | } | |
247e743c | 1563 | |
0ca1f7ce | 1564 | BUG(); |
2ac55d41 | 1565 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c CM |
1566 | ClearPageChecked(page); |
1567 | out: | |
2ac55d41 JB |
1568 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1569 | &cached_state, GFP_NOFS); | |
247e743c CM |
1570 | out_page: |
1571 | unlock_page(page); | |
1572 | page_cache_release(page); | |
b897abec | 1573 | kfree(fixup); |
247e743c CM |
1574 | } |
1575 | ||
1576 | /* | |
1577 | * There are a few paths in the higher layers of the kernel that directly | |
1578 | * set the page dirty bit without asking the filesystem if it is a | |
1579 | * good idea. This causes problems because we want to make sure COW | |
1580 | * properly happens and the data=ordered rules are followed. | |
1581 | * | |
c8b97818 | 1582 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1583 | * hasn't been properly setup for IO. We kick off an async process |
1584 | * to fix it up. The async helper will wait for ordered extents, set | |
1585 | * the delalloc bit and make it safe to write the page. | |
1586 | */ | |
b2950863 | 1587 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1588 | { |
1589 | struct inode *inode = page->mapping->host; | |
1590 | struct btrfs_writepage_fixup *fixup; | |
1591 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1592 | |
8b62b72b CM |
1593 | /* this page is properly in the ordered list */ |
1594 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1595 | return 0; |
1596 | ||
1597 | if (PageChecked(page)) | |
1598 | return -EAGAIN; | |
1599 | ||
1600 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1601 | if (!fixup) | |
1602 | return -EAGAIN; | |
f421950f | 1603 | |
247e743c CM |
1604 | SetPageChecked(page); |
1605 | page_cache_get(page); | |
1606 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1607 | fixup->page = page; | |
1608 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1609 | return -EAGAIN; | |
1610 | } | |
1611 | ||
d899e052 YZ |
1612 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1613 | struct inode *inode, u64 file_pos, | |
1614 | u64 disk_bytenr, u64 disk_num_bytes, | |
1615 | u64 num_bytes, u64 ram_bytes, | |
1616 | u8 compression, u8 encryption, | |
1617 | u16 other_encoding, int extent_type) | |
1618 | { | |
1619 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1620 | struct btrfs_file_extent_item *fi; | |
1621 | struct btrfs_path *path; | |
1622 | struct extent_buffer *leaf; | |
1623 | struct btrfs_key ins; | |
1624 | u64 hint; | |
1625 | int ret; | |
1626 | ||
1627 | path = btrfs_alloc_path(); | |
1628 | BUG_ON(!path); | |
1629 | ||
b9473439 | 1630 | path->leave_spinning = 1; |
a1ed835e CM |
1631 | |
1632 | /* | |
1633 | * we may be replacing one extent in the tree with another. | |
1634 | * The new extent is pinned in the extent map, and we don't want | |
1635 | * to drop it from the cache until it is completely in the btree. | |
1636 | * | |
1637 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1638 | * the caller is expected to unpin it and allow it to be merged | |
1639 | * with the others. | |
1640 | */ | |
920bbbfb YZ |
1641 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1642 | &hint, 0); | |
d899e052 YZ |
1643 | BUG_ON(ret); |
1644 | ||
1645 | ins.objectid = inode->i_ino; | |
1646 | ins.offset = file_pos; | |
1647 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1648 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1649 | BUG_ON(ret); | |
1650 | leaf = path->nodes[0]; | |
1651 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1652 | struct btrfs_file_extent_item); | |
1653 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1654 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1655 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1656 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1657 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1658 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1659 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1660 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1661 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1662 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1663 | |
1664 | btrfs_unlock_up_safe(path, 1); | |
1665 | btrfs_set_lock_blocking(leaf); | |
1666 | ||
d899e052 YZ |
1667 | btrfs_mark_buffer_dirty(leaf); |
1668 | ||
1669 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1670 | |
1671 | ins.objectid = disk_bytenr; | |
1672 | ins.offset = disk_num_bytes; | |
1673 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1674 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1675 | root->root_key.objectid, | |
1676 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1677 | BUG_ON(ret); |
d899e052 | 1678 | btrfs_free_path(path); |
b9473439 | 1679 | |
d899e052 YZ |
1680 | return 0; |
1681 | } | |
1682 | ||
5d13a98f CM |
1683 | /* |
1684 | * helper function for btrfs_finish_ordered_io, this | |
1685 | * just reads in some of the csum leaves to prime them into ram | |
1686 | * before we start the transaction. It limits the amount of btree | |
1687 | * reads required while inside the transaction. | |
1688 | */ | |
d352ac68 CM |
1689 | /* as ordered data IO finishes, this gets called so we can finish |
1690 | * an ordered extent if the range of bytes in the file it covers are | |
1691 | * fully written. | |
1692 | */ | |
211f90e6 | 1693 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1694 | { |
e6dcd2dc | 1695 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1696 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1697 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1698 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1699 | struct extent_state *cached_state = NULL; |
261507a0 | 1700 | int compress_type = 0; |
e6dcd2dc | 1701 | int ret; |
0cb59c99 | 1702 | bool nolock = false; |
e6dcd2dc | 1703 | |
5a1a3df1 JB |
1704 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1705 | end - start + 1); | |
ba1da2f4 | 1706 | if (!ret) |
e6dcd2dc | 1707 | return 0; |
e6dcd2dc | 1708 | BUG_ON(!ordered_extent); |
efd049fb | 1709 | |
0cb59c99 JB |
1710 | nolock = (root == root->fs_info->tree_root); |
1711 | ||
c2167754 YZ |
1712 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1713 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1714 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1715 | if (!ret) { | |
0cb59c99 JB |
1716 | if (nolock) |
1717 | trans = btrfs_join_transaction_nolock(root, 1); | |
1718 | else | |
1719 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 1720 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce YZ |
1721 | btrfs_set_trans_block_group(trans, inode); |
1722 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 YZ |
1723 | ret = btrfs_update_inode(trans, root, inode); |
1724 | BUG_ON(ret); | |
c2167754 YZ |
1725 | } |
1726 | goto out; | |
1727 | } | |
e6dcd2dc | 1728 | |
2ac55d41 JB |
1729 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1730 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1731 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1732 | |
0cb59c99 JB |
1733 | if (nolock) |
1734 | trans = btrfs_join_transaction_nolock(root, 1); | |
1735 | else | |
1736 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 1737 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce YZ |
1738 | btrfs_set_trans_block_group(trans, inode); |
1739 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 | 1740 | |
c8b97818 | 1741 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1742 | compress_type = ordered_extent->compress_type; |
d899e052 | 1743 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1744 | BUG_ON(compress_type); |
920bbbfb | 1745 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1746 | ordered_extent->file_offset, |
1747 | ordered_extent->file_offset + | |
1748 | ordered_extent->len); | |
1749 | BUG_ON(ret); | |
1750 | } else { | |
0af3d00b | 1751 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1752 | ret = insert_reserved_file_extent(trans, inode, |
1753 | ordered_extent->file_offset, | |
1754 | ordered_extent->start, | |
1755 | ordered_extent->disk_len, | |
1756 | ordered_extent->len, | |
1757 | ordered_extent->len, | |
261507a0 | 1758 | compress_type, 0, 0, |
d899e052 | 1759 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1760 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1761 | ordered_extent->file_offset, | |
1762 | ordered_extent->len); | |
d899e052 YZ |
1763 | BUG_ON(ret); |
1764 | } | |
2ac55d41 JB |
1765 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1766 | ordered_extent->file_offset + | |
1767 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1768 | ||
e6dcd2dc CM |
1769 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1770 | &ordered_extent->list); | |
1771 | ||
c2167754 YZ |
1772 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1773 | ret = btrfs_update_inode(trans, root, inode); | |
1774 | BUG_ON(ret); | |
c2167754 | 1775 | out: |
0cb59c99 JB |
1776 | if (nolock) { |
1777 | if (trans) | |
1778 | btrfs_end_transaction_nolock(trans, root); | |
1779 | } else { | |
1780 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); | |
1781 | if (trans) | |
1782 | btrfs_end_transaction(trans, root); | |
1783 | } | |
1784 | ||
e6dcd2dc CM |
1785 | /* once for us */ |
1786 | btrfs_put_ordered_extent(ordered_extent); | |
1787 | /* once for the tree */ | |
1788 | btrfs_put_ordered_extent(ordered_extent); | |
1789 | ||
e6dcd2dc CM |
1790 | return 0; |
1791 | } | |
1792 | ||
b2950863 | 1793 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1794 | struct extent_state *state, int uptodate) |
1795 | { | |
1abe9b8a | 1796 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1797 | ||
8b62b72b | 1798 | ClearPagePrivate2(page); |
211f90e6 CM |
1799 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1800 | } | |
1801 | ||
d352ac68 CM |
1802 | /* |
1803 | * When IO fails, either with EIO or csum verification fails, we | |
1804 | * try other mirrors that might have a good copy of the data. This | |
1805 | * io_failure_record is used to record state as we go through all the | |
1806 | * mirrors. If another mirror has good data, the page is set up to date | |
1807 | * and things continue. If a good mirror can't be found, the original | |
1808 | * bio end_io callback is called to indicate things have failed. | |
1809 | */ | |
7e38326f CM |
1810 | struct io_failure_record { |
1811 | struct page *page; | |
1812 | u64 start; | |
1813 | u64 len; | |
1814 | u64 logical; | |
d20f7043 | 1815 | unsigned long bio_flags; |
7e38326f CM |
1816 | int last_mirror; |
1817 | }; | |
1818 | ||
b2950863 | 1819 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1820 | struct page *page, u64 start, u64 end, |
1821 | struct extent_state *state) | |
7e38326f CM |
1822 | { |
1823 | struct io_failure_record *failrec = NULL; | |
1824 | u64 private; | |
1825 | struct extent_map *em; | |
1826 | struct inode *inode = page->mapping->host; | |
1827 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1828 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1829 | struct bio *bio; |
1830 | int num_copies; | |
1831 | int ret; | |
1259ab75 | 1832 | int rw; |
7e38326f CM |
1833 | u64 logical; |
1834 | ||
1835 | ret = get_state_private(failure_tree, start, &private); | |
1836 | if (ret) { | |
7e38326f CM |
1837 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1838 | if (!failrec) | |
1839 | return -ENOMEM; | |
1840 | failrec->start = start; | |
1841 | failrec->len = end - start + 1; | |
1842 | failrec->last_mirror = 0; | |
d20f7043 | 1843 | failrec->bio_flags = 0; |
7e38326f | 1844 | |
890871be | 1845 | read_lock(&em_tree->lock); |
3b951516 CM |
1846 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1847 | if (em->start > start || em->start + em->len < start) { | |
1848 | free_extent_map(em); | |
1849 | em = NULL; | |
1850 | } | |
890871be | 1851 | read_unlock(&em_tree->lock); |
7e38326f CM |
1852 | |
1853 | if (!em || IS_ERR(em)) { | |
1854 | kfree(failrec); | |
1855 | return -EIO; | |
1856 | } | |
1857 | logical = start - em->start; | |
1858 | logical = em->block_start + logical; | |
d20f7043 CM |
1859 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1860 | logical = em->block_start; | |
1861 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
261507a0 LZ |
1862 | extent_set_compress_type(&failrec->bio_flags, |
1863 | em->compress_type); | |
d20f7043 | 1864 | } |
7e38326f CM |
1865 | failrec->logical = logical; |
1866 | free_extent_map(em); | |
1867 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1868 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1869 | set_state_private(failure_tree, start, |
1870 | (u64)(unsigned long)failrec); | |
7e38326f | 1871 | } else { |
587f7704 | 1872 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1873 | } |
1874 | num_copies = btrfs_num_copies( | |
1875 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1876 | failrec->logical, failrec->len); | |
1877 | failrec->last_mirror++; | |
1878 | if (!state) { | |
cad321ad | 1879 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1880 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1881 | failrec->start, | |
1882 | EXTENT_LOCKED); | |
1883 | if (state && state->start != failrec->start) | |
1884 | state = NULL; | |
cad321ad | 1885 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1886 | } |
1887 | if (!state || failrec->last_mirror > num_copies) { | |
1888 | set_state_private(failure_tree, failrec->start, 0); | |
1889 | clear_extent_bits(failure_tree, failrec->start, | |
1890 | failrec->start + failrec->len - 1, | |
1891 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1892 | kfree(failrec); | |
1893 | return -EIO; | |
1894 | } | |
1895 | bio = bio_alloc(GFP_NOFS, 1); | |
1896 | bio->bi_private = state; | |
1897 | bio->bi_end_io = failed_bio->bi_end_io; | |
1898 | bio->bi_sector = failrec->logical >> 9; | |
1899 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1900 | bio->bi_size = 0; |
d20f7043 | 1901 | |
7e38326f | 1902 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
7b6d91da | 1903 | if (failed_bio->bi_rw & REQ_WRITE) |
1259ab75 CM |
1904 | rw = WRITE; |
1905 | else | |
1906 | rw = READ; | |
1907 | ||
c2db1073 | 1908 | ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, |
c8b97818 | 1909 | failrec->last_mirror, |
eaf25d93 | 1910 | failrec->bio_flags, 0); |
c2db1073 | 1911 | return ret; |
1259ab75 CM |
1912 | } |
1913 | ||
d352ac68 CM |
1914 | /* |
1915 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1916 | * to see if we need to process or clean up an io_failure_record | |
1917 | */ | |
b2950863 | 1918 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1919 | { |
1920 | u64 private; | |
1921 | u64 private_failure; | |
1922 | struct io_failure_record *failure; | |
1923 | int ret; | |
1924 | ||
1925 | private = 0; | |
1926 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
ec29ed5b | 1927 | (u64)-1, 1, EXTENT_DIRTY, 0)) { |
1259ab75 CM |
1928 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, |
1929 | start, &private_failure); | |
1930 | if (ret == 0) { | |
1931 | failure = (struct io_failure_record *)(unsigned long) | |
1932 | private_failure; | |
1933 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1934 | failure->start, 0); | |
1935 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1936 | failure->start, | |
1937 | failure->start + failure->len - 1, | |
1938 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1939 | GFP_NOFS); | |
1940 | kfree(failure); | |
1941 | } | |
1942 | } | |
7e38326f CM |
1943 | return 0; |
1944 | } | |
1945 | ||
d352ac68 CM |
1946 | /* |
1947 | * when reads are done, we need to check csums to verify the data is correct | |
1948 | * if there's a match, we allow the bio to finish. If not, we go through | |
1949 | * the io_failure_record routines to find good copies | |
1950 | */ | |
b2950863 | 1951 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1952 | struct extent_state *state) |
07157aac | 1953 | { |
35ebb934 | 1954 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1955 | struct inode *inode = page->mapping->host; |
d1310b2e | 1956 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1957 | char *kaddr; |
aadfeb6e | 1958 | u64 private = ~(u32)0; |
07157aac | 1959 | int ret; |
ff79f819 CM |
1960 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1961 | u32 csum = ~(u32)0; | |
d1310b2e | 1962 | |
d20f7043 CM |
1963 | if (PageChecked(page)) { |
1964 | ClearPageChecked(page); | |
1965 | goto good; | |
1966 | } | |
6cbff00f CH |
1967 | |
1968 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1969 | return 0; |
1970 | ||
1971 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1972 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1973 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1974 | GFP_NOFS); | |
b6cda9bc | 1975 | return 0; |
17d217fe | 1976 | } |
d20f7043 | 1977 | |
c2e639f0 | 1978 | if (state && state->start == start) { |
70dec807 CM |
1979 | private = state->private; |
1980 | ret = 0; | |
1981 | } else { | |
1982 | ret = get_state_private(io_tree, start, &private); | |
1983 | } | |
9ab86c8e | 1984 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1985 | if (ret) |
07157aac | 1986 | goto zeroit; |
d397712b | 1987 | |
ff79f819 CM |
1988 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1989 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1990 | if (csum != private) |
07157aac | 1991 | goto zeroit; |
d397712b | 1992 | |
9ab86c8e | 1993 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1994 | good: |
7e38326f CM |
1995 | /* if the io failure tree for this inode is non-empty, |
1996 | * check to see if we've recovered from a failed IO | |
1997 | */ | |
1259ab75 | 1998 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
1999 | return 0; |
2000 | ||
2001 | zeroit: | |
193f284d CM |
2002 | if (printk_ratelimit()) { |
2003 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
2004 | "private %llu\n", page->mapping->host->i_ino, | |
2005 | (unsigned long long)start, csum, | |
2006 | (unsigned long long)private); | |
2007 | } | |
db94535d CM |
2008 | memset(kaddr + offset, 1, end - start + 1); |
2009 | flush_dcache_page(page); | |
9ab86c8e | 2010 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2011 | if (private == 0) |
2012 | return 0; | |
7e38326f | 2013 | return -EIO; |
07157aac | 2014 | } |
b888db2b | 2015 | |
24bbcf04 YZ |
2016 | struct delayed_iput { |
2017 | struct list_head list; | |
2018 | struct inode *inode; | |
2019 | }; | |
2020 | ||
2021 | void btrfs_add_delayed_iput(struct inode *inode) | |
2022 | { | |
2023 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2024 | struct delayed_iput *delayed; | |
2025 | ||
2026 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2027 | return; | |
2028 | ||
2029 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2030 | delayed->inode = inode; | |
2031 | ||
2032 | spin_lock(&fs_info->delayed_iput_lock); | |
2033 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2034 | spin_unlock(&fs_info->delayed_iput_lock); | |
2035 | } | |
2036 | ||
2037 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2038 | { | |
2039 | LIST_HEAD(list); | |
2040 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2041 | struct delayed_iput *delayed; | |
2042 | int empty; | |
2043 | ||
2044 | spin_lock(&fs_info->delayed_iput_lock); | |
2045 | empty = list_empty(&fs_info->delayed_iputs); | |
2046 | spin_unlock(&fs_info->delayed_iput_lock); | |
2047 | if (empty) | |
2048 | return; | |
2049 | ||
2050 | down_read(&root->fs_info->cleanup_work_sem); | |
2051 | spin_lock(&fs_info->delayed_iput_lock); | |
2052 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2053 | spin_unlock(&fs_info->delayed_iput_lock); | |
2054 | ||
2055 | while (!list_empty(&list)) { | |
2056 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2057 | list_del(&delayed->list); | |
2058 | iput(delayed->inode); | |
2059 | kfree(delayed); | |
2060 | } | |
2061 | up_read(&root->fs_info->cleanup_work_sem); | |
2062 | } | |
2063 | ||
d68fc57b YZ |
2064 | /* |
2065 | * calculate extra metadata reservation when snapshotting a subvolume | |
2066 | * contains orphan files. | |
2067 | */ | |
2068 | void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans, | |
2069 | struct btrfs_pending_snapshot *pending, | |
2070 | u64 *bytes_to_reserve) | |
2071 | { | |
2072 | struct btrfs_root *root; | |
2073 | struct btrfs_block_rsv *block_rsv; | |
2074 | u64 num_bytes; | |
2075 | int index; | |
2076 | ||
2077 | root = pending->root; | |
2078 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2079 | return; | |
2080 | ||
2081 | block_rsv = root->orphan_block_rsv; | |
2082 | ||
2083 | /* orphan block reservation for the snapshot */ | |
2084 | num_bytes = block_rsv->size; | |
2085 | ||
2086 | /* | |
2087 | * after the snapshot is created, COWing tree blocks may use more | |
2088 | * space than it frees. So we should make sure there is enough | |
2089 | * reserved space. | |
2090 | */ | |
2091 | index = trans->transid & 0x1; | |
2092 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2093 | num_bytes += block_rsv->size - | |
2094 | (block_rsv->reserved + block_rsv->freed[index]); | |
2095 | } | |
2096 | ||
2097 | *bytes_to_reserve += num_bytes; | |
2098 | } | |
2099 | ||
2100 | void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans, | |
2101 | struct btrfs_pending_snapshot *pending) | |
2102 | { | |
2103 | struct btrfs_root *root = pending->root; | |
2104 | struct btrfs_root *snap = pending->snap; | |
2105 | struct btrfs_block_rsv *block_rsv; | |
2106 | u64 num_bytes; | |
2107 | int index; | |
2108 | int ret; | |
2109 | ||
2110 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2111 | return; | |
2112 | ||
2113 | /* refill source subvolume's orphan block reservation */ | |
2114 | block_rsv = root->orphan_block_rsv; | |
2115 | index = trans->transid & 0x1; | |
2116 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2117 | num_bytes = block_rsv->size - | |
2118 | (block_rsv->reserved + block_rsv->freed[index]); | |
2119 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2120 | root->orphan_block_rsv, | |
2121 | num_bytes); | |
2122 | BUG_ON(ret); | |
2123 | } | |
2124 | ||
2125 | /* setup orphan block reservation for the snapshot */ | |
2126 | block_rsv = btrfs_alloc_block_rsv(snap); | |
2127 | BUG_ON(!block_rsv); | |
2128 | ||
2129 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
2130 | snap->orphan_block_rsv = block_rsv; | |
2131 | ||
2132 | num_bytes = root->orphan_block_rsv->size; | |
2133 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2134 | block_rsv, num_bytes); | |
2135 | BUG_ON(ret); | |
2136 | ||
2137 | #if 0 | |
2138 | /* insert orphan item for the snapshot */ | |
2139 | WARN_ON(!root->orphan_item_inserted); | |
2140 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2141 | snap->root_key.objectid); | |
2142 | BUG_ON(ret); | |
2143 | snap->orphan_item_inserted = 1; | |
2144 | #endif | |
2145 | } | |
2146 | ||
2147 | enum btrfs_orphan_cleanup_state { | |
2148 | ORPHAN_CLEANUP_STARTED = 1, | |
2149 | ORPHAN_CLEANUP_DONE = 2, | |
2150 | }; | |
2151 | ||
2152 | /* | |
2153 | * This is called in transaction commmit time. If there are no orphan | |
2154 | * files in the subvolume, it removes orphan item and frees block_rsv | |
2155 | * structure. | |
2156 | */ | |
2157 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2158 | struct btrfs_root *root) | |
2159 | { | |
2160 | int ret; | |
2161 | ||
2162 | if (!list_empty(&root->orphan_list) || | |
2163 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2164 | return; | |
2165 | ||
2166 | if (root->orphan_item_inserted && | |
2167 | btrfs_root_refs(&root->root_item) > 0) { | |
2168 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2169 | root->root_key.objectid); | |
2170 | BUG_ON(ret); | |
2171 | root->orphan_item_inserted = 0; | |
2172 | } | |
2173 | ||
2174 | if (root->orphan_block_rsv) { | |
2175 | WARN_ON(root->orphan_block_rsv->size > 0); | |
2176 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
2177 | root->orphan_block_rsv = NULL; | |
2178 | } | |
2179 | } | |
2180 | ||
7b128766 JB |
2181 | /* |
2182 | * This creates an orphan entry for the given inode in case something goes | |
2183 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2184 | * |
2185 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2186 | * this function. | |
7b128766 JB |
2187 | */ |
2188 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2189 | { | |
2190 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2191 | struct btrfs_block_rsv *block_rsv = NULL; |
2192 | int reserve = 0; | |
2193 | int insert = 0; | |
2194 | int ret; | |
7b128766 | 2195 | |
d68fc57b YZ |
2196 | if (!root->orphan_block_rsv) { |
2197 | block_rsv = btrfs_alloc_block_rsv(root); | |
2198 | BUG_ON(!block_rsv); | |
2199 | } | |
7b128766 | 2200 | |
d68fc57b YZ |
2201 | spin_lock(&root->orphan_lock); |
2202 | if (!root->orphan_block_rsv) { | |
2203 | root->orphan_block_rsv = block_rsv; | |
2204 | } else if (block_rsv) { | |
2205 | btrfs_free_block_rsv(root, block_rsv); | |
2206 | block_rsv = NULL; | |
7b128766 | 2207 | } |
7b128766 | 2208 | |
d68fc57b YZ |
2209 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2210 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2211 | #if 0 | |
2212 | /* | |
2213 | * For proper ENOSPC handling, we should do orphan | |
2214 | * cleanup when mounting. But this introduces backward | |
2215 | * compatibility issue. | |
2216 | */ | |
2217 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2218 | insert = 2; | |
2219 | else | |
2220 | insert = 1; | |
2221 | #endif | |
2222 | insert = 1; | |
7b128766 JB |
2223 | } |
2224 | ||
d68fc57b YZ |
2225 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2226 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2227 | reserve = 1; | |
2228 | } | |
2229 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2230 | |
d68fc57b YZ |
2231 | if (block_rsv) |
2232 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
7b128766 | 2233 | |
d68fc57b YZ |
2234 | /* grab metadata reservation from transaction handle */ |
2235 | if (reserve) { | |
2236 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2237 | BUG_ON(ret); | |
2238 | } | |
7b128766 | 2239 | |
d68fc57b YZ |
2240 | /* insert an orphan item to track this unlinked/truncated file */ |
2241 | if (insert >= 1) { | |
2242 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2243 | BUG_ON(ret); | |
2244 | } | |
2245 | ||
2246 | /* insert an orphan item to track subvolume contains orphan files */ | |
2247 | if (insert >= 2) { | |
2248 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2249 | root->root_key.objectid); | |
2250 | BUG_ON(ret); | |
2251 | } | |
2252 | return 0; | |
7b128766 JB |
2253 | } |
2254 | ||
2255 | /* | |
2256 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2257 | * item for this particular inode. | |
2258 | */ | |
2259 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2260 | { | |
2261 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2262 | int delete_item = 0; |
2263 | int release_rsv = 0; | |
7b128766 JB |
2264 | int ret = 0; |
2265 | ||
d68fc57b YZ |
2266 | spin_lock(&root->orphan_lock); |
2267 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2268 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2269 | delete_item = 1; | |
7b128766 JB |
2270 | } |
2271 | ||
d68fc57b YZ |
2272 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2273 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2274 | release_rsv = 1; | |
7b128766 | 2275 | } |
d68fc57b | 2276 | spin_unlock(&root->orphan_lock); |
7b128766 | 2277 | |
d68fc57b YZ |
2278 | if (trans && delete_item) { |
2279 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2280 | BUG_ON(ret); | |
2281 | } | |
7b128766 | 2282 | |
d68fc57b YZ |
2283 | if (release_rsv) |
2284 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2285 | |
d68fc57b | 2286 | return 0; |
7b128766 JB |
2287 | } |
2288 | ||
2289 | /* | |
2290 | * this cleans up any orphans that may be left on the list from the last use | |
2291 | * of this root. | |
2292 | */ | |
66b4ffd1 | 2293 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2294 | { |
2295 | struct btrfs_path *path; | |
2296 | struct extent_buffer *leaf; | |
7b128766 JB |
2297 | struct btrfs_key key, found_key; |
2298 | struct btrfs_trans_handle *trans; | |
2299 | struct inode *inode; | |
2300 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2301 | ||
d68fc57b | 2302 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2303 | return 0; |
c71bf099 YZ |
2304 | |
2305 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2306 | if (!path) { |
2307 | ret = -ENOMEM; | |
2308 | goto out; | |
2309 | } | |
7b128766 JB |
2310 | path->reada = -1; |
2311 | ||
2312 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2313 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2314 | key.offset = (u64)-1; | |
2315 | ||
7b128766 JB |
2316 | while (1) { |
2317 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2318 | if (ret < 0) |
2319 | goto out; | |
7b128766 JB |
2320 | |
2321 | /* | |
2322 | * if ret == 0 means we found what we were searching for, which | |
2323 | * is weird, but possible, so only screw with path if we didnt | |
2324 | * find the key and see if we have stuff that matches | |
2325 | */ | |
2326 | if (ret > 0) { | |
66b4ffd1 | 2327 | ret = 0; |
7b128766 JB |
2328 | if (path->slots[0] == 0) |
2329 | break; | |
2330 | path->slots[0]--; | |
2331 | } | |
2332 | ||
2333 | /* pull out the item */ | |
2334 | leaf = path->nodes[0]; | |
7b128766 JB |
2335 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2336 | ||
2337 | /* make sure the item matches what we want */ | |
2338 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2339 | break; | |
2340 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2341 | break; | |
2342 | ||
2343 | /* release the path since we're done with it */ | |
2344 | btrfs_release_path(root, path); | |
2345 | ||
2346 | /* | |
2347 | * this is where we are basically btrfs_lookup, without the | |
2348 | * crossing root thing. we store the inode number in the | |
2349 | * offset of the orphan item. | |
2350 | */ | |
5d4f98a2 YZ |
2351 | found_key.objectid = found_key.offset; |
2352 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2353 | found_key.offset = 0; | |
73f73415 | 2354 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
66b4ffd1 JB |
2355 | if (IS_ERR(inode)) { |
2356 | ret = PTR_ERR(inode); | |
2357 | goto out; | |
2358 | } | |
7b128766 | 2359 | |
7b128766 JB |
2360 | /* |
2361 | * add this inode to the orphan list so btrfs_orphan_del does | |
2362 | * the proper thing when we hit it | |
2363 | */ | |
d68fc57b | 2364 | spin_lock(&root->orphan_lock); |
7b128766 | 2365 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
d68fc57b | 2366 | spin_unlock(&root->orphan_lock); |
7b128766 JB |
2367 | |
2368 | /* | |
2369 | * if this is a bad inode, means we actually succeeded in | |
2370 | * removing the inode, but not the orphan record, which means | |
2371 | * we need to manually delete the orphan since iput will just | |
2372 | * do a destroy_inode | |
2373 | */ | |
2374 | if (is_bad_inode(inode)) { | |
a22285a6 | 2375 | trans = btrfs_start_transaction(root, 0); |
66b4ffd1 JB |
2376 | if (IS_ERR(trans)) { |
2377 | ret = PTR_ERR(trans); | |
2378 | goto out; | |
2379 | } | |
7b128766 | 2380 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2381 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2382 | iput(inode); |
2383 | continue; | |
2384 | } | |
2385 | ||
2386 | /* if we have links, this was a truncate, lets do that */ | |
2387 | if (inode->i_nlink) { | |
a41ad394 JB |
2388 | if (!S_ISREG(inode->i_mode)) { |
2389 | WARN_ON(1); | |
2390 | iput(inode); | |
2391 | continue; | |
2392 | } | |
7b128766 | 2393 | nr_truncate++; |
66b4ffd1 | 2394 | ret = btrfs_truncate(inode); |
7b128766 JB |
2395 | } else { |
2396 | nr_unlink++; | |
2397 | } | |
2398 | ||
2399 | /* this will do delete_inode and everything for us */ | |
2400 | iput(inode); | |
66b4ffd1 JB |
2401 | if (ret) |
2402 | goto out; | |
7b128766 | 2403 | } |
d68fc57b YZ |
2404 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2405 | ||
2406 | if (root->orphan_block_rsv) | |
2407 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2408 | (u64)-1); | |
2409 | ||
2410 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
2411 | trans = btrfs_join_transaction(root, 1); | |
66b4ffd1 JB |
2412 | if (!IS_ERR(trans)) |
2413 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2414 | } |
7b128766 JB |
2415 | |
2416 | if (nr_unlink) | |
2417 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2418 | if (nr_truncate) | |
2419 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2420 | |
2421 | out: | |
2422 | if (ret) | |
2423 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2424 | btrfs_free_path(path); | |
2425 | return ret; | |
7b128766 JB |
2426 | } |
2427 | ||
46a53cca CM |
2428 | /* |
2429 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2430 | * don't find any xattrs, we know there can't be any acls. | |
2431 | * | |
2432 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2433 | */ | |
2434 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2435 | int slot, u64 objectid) | |
2436 | { | |
2437 | u32 nritems = btrfs_header_nritems(leaf); | |
2438 | struct btrfs_key found_key; | |
2439 | int scanned = 0; | |
2440 | ||
2441 | slot++; | |
2442 | while (slot < nritems) { | |
2443 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2444 | ||
2445 | /* we found a different objectid, there must not be acls */ | |
2446 | if (found_key.objectid != objectid) | |
2447 | return 0; | |
2448 | ||
2449 | /* we found an xattr, assume we've got an acl */ | |
2450 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2451 | return 1; | |
2452 | ||
2453 | /* | |
2454 | * we found a key greater than an xattr key, there can't | |
2455 | * be any acls later on | |
2456 | */ | |
2457 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2458 | return 0; | |
2459 | ||
2460 | slot++; | |
2461 | scanned++; | |
2462 | ||
2463 | /* | |
2464 | * it goes inode, inode backrefs, xattrs, extents, | |
2465 | * so if there are a ton of hard links to an inode there can | |
2466 | * be a lot of backrefs. Don't waste time searching too hard, | |
2467 | * this is just an optimization | |
2468 | */ | |
2469 | if (scanned >= 8) | |
2470 | break; | |
2471 | } | |
2472 | /* we hit the end of the leaf before we found an xattr or | |
2473 | * something larger than an xattr. We have to assume the inode | |
2474 | * has acls | |
2475 | */ | |
2476 | return 1; | |
2477 | } | |
2478 | ||
d352ac68 CM |
2479 | /* |
2480 | * read an inode from the btree into the in-memory inode | |
2481 | */ | |
5d4f98a2 | 2482 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2483 | { |
2484 | struct btrfs_path *path; | |
5f39d397 | 2485 | struct extent_buffer *leaf; |
39279cc3 | 2486 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2487 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2488 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2489 | struct btrfs_key location; | |
46a53cca | 2490 | int maybe_acls; |
39279cc3 | 2491 | u64 alloc_group_block; |
618e21d5 | 2492 | u32 rdev; |
39279cc3 CM |
2493 | int ret; |
2494 | ||
2495 | path = btrfs_alloc_path(); | |
2496 | BUG_ON(!path); | |
39279cc3 | 2497 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2498 | |
39279cc3 | 2499 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2500 | if (ret) |
39279cc3 | 2501 | goto make_bad; |
39279cc3 | 2502 | |
5f39d397 CM |
2503 | leaf = path->nodes[0]; |
2504 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2505 | struct btrfs_inode_item); | |
2506 | ||
2507 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2508 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2509 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2510 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2511 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2512 | |
2513 | tspec = btrfs_inode_atime(inode_item); | |
2514 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2515 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2516 | ||
2517 | tspec = btrfs_inode_mtime(inode_item); | |
2518 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2519 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2520 | ||
2521 | tspec = btrfs_inode_ctime(inode_item); | |
2522 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2523 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2524 | ||
a76a3cd4 | 2525 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2526 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2527 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2528 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2529 | inode->i_rdev = 0; |
5f39d397 CM |
2530 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2531 | ||
aec7477b | 2532 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2533 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2534 | |
5f39d397 | 2535 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2536 | |
46a53cca CM |
2537 | /* |
2538 | * try to precache a NULL acl entry for files that don't have | |
2539 | * any xattrs or acls | |
2540 | */ | |
2541 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2542 | if (!maybe_acls) |
2543 | cache_no_acl(inode); | |
46a53cca | 2544 | |
d2fb3437 YZ |
2545 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2546 | alloc_group_block, 0); | |
39279cc3 CM |
2547 | btrfs_free_path(path); |
2548 | inode_item = NULL; | |
2549 | ||
39279cc3 | 2550 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2551 | case S_IFREG: |
2552 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2553 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2554 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2555 | inode->i_fop = &btrfs_file_operations; |
2556 | inode->i_op = &btrfs_file_inode_operations; | |
2557 | break; | |
2558 | case S_IFDIR: | |
2559 | inode->i_fop = &btrfs_dir_file_operations; | |
2560 | if (root == root->fs_info->tree_root) | |
2561 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2562 | else | |
2563 | inode->i_op = &btrfs_dir_inode_operations; | |
2564 | break; | |
2565 | case S_IFLNK: | |
2566 | inode->i_op = &btrfs_symlink_inode_operations; | |
2567 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2568 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2569 | break; |
618e21d5 | 2570 | default: |
0279b4cd | 2571 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2572 | init_special_inode(inode, inode->i_mode, rdev); |
2573 | break; | |
39279cc3 | 2574 | } |
6cbff00f CH |
2575 | |
2576 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2577 | return; |
2578 | ||
2579 | make_bad: | |
39279cc3 | 2580 | btrfs_free_path(path); |
39279cc3 CM |
2581 | make_bad_inode(inode); |
2582 | } | |
2583 | ||
d352ac68 CM |
2584 | /* |
2585 | * given a leaf and an inode, copy the inode fields into the leaf | |
2586 | */ | |
e02119d5 CM |
2587 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2588 | struct extent_buffer *leaf, | |
5f39d397 | 2589 | struct btrfs_inode_item *item, |
39279cc3 CM |
2590 | struct inode *inode) |
2591 | { | |
5f39d397 CM |
2592 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2593 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2594 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2595 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2596 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2597 | ||
2598 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2599 | inode->i_atime.tv_sec); | |
2600 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2601 | inode->i_atime.tv_nsec); | |
2602 | ||
2603 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2604 | inode->i_mtime.tv_sec); | |
2605 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2606 | inode->i_mtime.tv_nsec); | |
2607 | ||
2608 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2609 | inode->i_ctime.tv_sec); | |
2610 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2611 | inode->i_ctime.tv_nsec); | |
2612 | ||
a76a3cd4 | 2613 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2614 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2615 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2616 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2617 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2618 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2619 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
39279cc3 CM |
2620 | } |
2621 | ||
d352ac68 CM |
2622 | /* |
2623 | * copy everything in the in-memory inode into the btree. | |
2624 | */ | |
d397712b CM |
2625 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2626 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2627 | { |
2628 | struct btrfs_inode_item *inode_item; | |
2629 | struct btrfs_path *path; | |
5f39d397 | 2630 | struct extent_buffer *leaf; |
39279cc3 CM |
2631 | int ret; |
2632 | ||
2633 | path = btrfs_alloc_path(); | |
2634 | BUG_ON(!path); | |
b9473439 | 2635 | path->leave_spinning = 1; |
39279cc3 CM |
2636 | ret = btrfs_lookup_inode(trans, root, path, |
2637 | &BTRFS_I(inode)->location, 1); | |
2638 | if (ret) { | |
2639 | if (ret > 0) | |
2640 | ret = -ENOENT; | |
2641 | goto failed; | |
2642 | } | |
2643 | ||
b4ce94de | 2644 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2645 | leaf = path->nodes[0]; |
2646 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2647 | struct btrfs_inode_item); |
2648 | ||
e02119d5 | 2649 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2650 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2651 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2652 | ret = 0; |
2653 | failed: | |
39279cc3 CM |
2654 | btrfs_free_path(path); |
2655 | return ret; | |
2656 | } | |
2657 | ||
2658 | ||
d352ac68 CM |
2659 | /* |
2660 | * unlink helper that gets used here in inode.c and in the tree logging | |
2661 | * recovery code. It remove a link in a directory with a given name, and | |
2662 | * also drops the back refs in the inode to the directory | |
2663 | */ | |
92986796 AV |
2664 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2665 | struct btrfs_root *root, | |
2666 | struct inode *dir, struct inode *inode, | |
2667 | const char *name, int name_len) | |
39279cc3 CM |
2668 | { |
2669 | struct btrfs_path *path; | |
39279cc3 | 2670 | int ret = 0; |
5f39d397 | 2671 | struct extent_buffer *leaf; |
39279cc3 | 2672 | struct btrfs_dir_item *di; |
5f39d397 | 2673 | struct btrfs_key key; |
aec7477b | 2674 | u64 index; |
39279cc3 CM |
2675 | |
2676 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2677 | if (!path) { |
2678 | ret = -ENOMEM; | |
554233a6 | 2679 | goto out; |
54aa1f4d CM |
2680 | } |
2681 | ||
b9473439 | 2682 | path->leave_spinning = 1; |
39279cc3 CM |
2683 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2684 | name, name_len, -1); | |
2685 | if (IS_ERR(di)) { | |
2686 | ret = PTR_ERR(di); | |
2687 | goto err; | |
2688 | } | |
2689 | if (!di) { | |
2690 | ret = -ENOENT; | |
2691 | goto err; | |
2692 | } | |
5f39d397 CM |
2693 | leaf = path->nodes[0]; |
2694 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2695 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2696 | if (ret) |
2697 | goto err; | |
39279cc3 CM |
2698 | btrfs_release_path(root, path); |
2699 | ||
aec7477b | 2700 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2701 | inode->i_ino, |
2702 | dir->i_ino, &index); | |
aec7477b | 2703 | if (ret) { |
d397712b | 2704 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2705 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2706 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2707 | goto err; |
2708 | } | |
2709 | ||
39279cc3 | 2710 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2711 | index, name, name_len, -1); |
39279cc3 CM |
2712 | if (IS_ERR(di)) { |
2713 | ret = PTR_ERR(di); | |
2714 | goto err; | |
2715 | } | |
2716 | if (!di) { | |
2717 | ret = -ENOENT; | |
2718 | goto err; | |
2719 | } | |
2720 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2721 | btrfs_release_path(root, path); |
39279cc3 | 2722 | |
e02119d5 CM |
2723 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2724 | inode, dir->i_ino); | |
49eb7e46 | 2725 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2726 | |
2727 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2728 | dir, index); | |
6418c961 CM |
2729 | if (ret == -ENOENT) |
2730 | ret = 0; | |
39279cc3 CM |
2731 | err: |
2732 | btrfs_free_path(path); | |
e02119d5 CM |
2733 | if (ret) |
2734 | goto out; | |
2735 | ||
2736 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2737 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2738 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2739 | out: |
39279cc3 CM |
2740 | return ret; |
2741 | } | |
2742 | ||
92986796 AV |
2743 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2744 | struct btrfs_root *root, | |
2745 | struct inode *dir, struct inode *inode, | |
2746 | const char *name, int name_len) | |
2747 | { | |
2748 | int ret; | |
2749 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2750 | if (!ret) { | |
2751 | btrfs_drop_nlink(inode); | |
2752 | ret = btrfs_update_inode(trans, root, inode); | |
2753 | } | |
2754 | return ret; | |
2755 | } | |
2756 | ||
2757 | ||
a22285a6 YZ |
2758 | /* helper to check if there is any shared block in the path */ |
2759 | static int check_path_shared(struct btrfs_root *root, | |
2760 | struct btrfs_path *path) | |
39279cc3 | 2761 | { |
a22285a6 YZ |
2762 | struct extent_buffer *eb; |
2763 | int level; | |
0e4dcbef | 2764 | u64 refs = 1; |
5df6a9f6 | 2765 | |
a22285a6 | 2766 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2767 | int ret; |
2768 | ||
a22285a6 YZ |
2769 | if (!path->nodes[level]) |
2770 | break; | |
2771 | eb = path->nodes[level]; | |
2772 | if (!btrfs_block_can_be_shared(root, eb)) | |
2773 | continue; | |
2774 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2775 | &refs, NULL); | |
2776 | if (refs > 1) | |
2777 | return 1; | |
5df6a9f6 | 2778 | } |
dedefd72 | 2779 | return 0; |
39279cc3 CM |
2780 | } |
2781 | ||
a22285a6 YZ |
2782 | /* |
2783 | * helper to start transaction for unlink and rmdir. | |
2784 | * | |
2785 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2786 | * so in enospc case, we should make sure they will free space before | |
2787 | * allowing them to use the global metadata reservation. | |
2788 | */ | |
2789 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2790 | struct dentry *dentry) | |
4df27c4d | 2791 | { |
39279cc3 | 2792 | struct btrfs_trans_handle *trans; |
a22285a6 | 2793 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2794 | struct btrfs_path *path; |
a22285a6 | 2795 | struct btrfs_inode_ref *ref; |
4df27c4d | 2796 | struct btrfs_dir_item *di; |
7b128766 | 2797 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2798 | u64 index; |
a22285a6 YZ |
2799 | int check_link = 1; |
2800 | int err = -ENOSPC; | |
4df27c4d YZ |
2801 | int ret; |
2802 | ||
a22285a6 YZ |
2803 | trans = btrfs_start_transaction(root, 10); |
2804 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
2805 | return trans; | |
4df27c4d | 2806 | |
a22285a6 YZ |
2807 | if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
2808 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2809 | |
a22285a6 YZ |
2810 | /* check if there is someone else holds reference */ |
2811 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2812 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2813 | |
a22285a6 YZ |
2814 | if (atomic_read(&inode->i_count) > 2) |
2815 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2816 | |
a22285a6 YZ |
2817 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2818 | return ERR_PTR(-ENOSPC); | |
2819 | ||
2820 | path = btrfs_alloc_path(); | |
2821 | if (!path) { | |
2822 | root->fs_info->enospc_unlink = 0; | |
2823 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2824 | } |
2825 | ||
a22285a6 | 2826 | trans = btrfs_start_transaction(root, 0); |
5df6a9f6 | 2827 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2828 | btrfs_free_path(path); |
2829 | root->fs_info->enospc_unlink = 0; | |
2830 | return trans; | |
2831 | } | |
4df27c4d | 2832 | |
a22285a6 YZ |
2833 | path->skip_locking = 1; |
2834 | path->search_commit_root = 1; | |
4df27c4d | 2835 | |
a22285a6 YZ |
2836 | ret = btrfs_lookup_inode(trans, root, path, |
2837 | &BTRFS_I(dir)->location, 0); | |
2838 | if (ret < 0) { | |
2839 | err = ret; | |
2840 | goto out; | |
2841 | } | |
2842 | if (ret == 0) { | |
2843 | if (check_path_shared(root, path)) | |
2844 | goto out; | |
2845 | } else { | |
2846 | check_link = 0; | |
5df6a9f6 | 2847 | } |
a22285a6 YZ |
2848 | btrfs_release_path(root, path); |
2849 | ||
2850 | ret = btrfs_lookup_inode(trans, root, path, | |
2851 | &BTRFS_I(inode)->location, 0); | |
2852 | if (ret < 0) { | |
2853 | err = ret; | |
2854 | goto out; | |
2855 | } | |
2856 | if (ret == 0) { | |
2857 | if (check_path_shared(root, path)) | |
2858 | goto out; | |
2859 | } else { | |
2860 | check_link = 0; | |
2861 | } | |
2862 | btrfs_release_path(root, path); | |
2863 | ||
2864 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2865 | ret = btrfs_lookup_file_extent(trans, root, path, | |
2866 | inode->i_ino, (u64)-1, 0); | |
2867 | if (ret < 0) { | |
2868 | err = ret; | |
2869 | goto out; | |
2870 | } | |
2871 | BUG_ON(ret == 0); | |
2872 | if (check_path_shared(root, path)) | |
2873 | goto out; | |
2874 | btrfs_release_path(root, path); | |
2875 | } | |
2876 | ||
2877 | if (!check_link) { | |
2878 | err = 0; | |
2879 | goto out; | |
2880 | } | |
2881 | ||
2882 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2883 | dentry->d_name.name, dentry->d_name.len, 0); | |
2884 | if (IS_ERR(di)) { | |
2885 | err = PTR_ERR(di); | |
2886 | goto out; | |
2887 | } | |
2888 | if (di) { | |
2889 | if (check_path_shared(root, path)) | |
2890 | goto out; | |
2891 | } else { | |
2892 | err = 0; | |
2893 | goto out; | |
2894 | } | |
2895 | btrfs_release_path(root, path); | |
2896 | ||
2897 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2898 | dentry->d_name.name, dentry->d_name.len, | |
2899 | inode->i_ino, dir->i_ino, 0); | |
2900 | if (IS_ERR(ref)) { | |
2901 | err = PTR_ERR(ref); | |
2902 | goto out; | |
2903 | } | |
2904 | BUG_ON(!ref); | |
2905 | if (check_path_shared(root, path)) | |
2906 | goto out; | |
2907 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
2908 | btrfs_release_path(root, path); | |
2909 | ||
2910 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index, | |
2911 | dentry->d_name.name, dentry->d_name.len, 0); | |
2912 | if (IS_ERR(di)) { | |
2913 | err = PTR_ERR(di); | |
2914 | goto out; | |
2915 | } | |
2916 | BUG_ON(ret == -ENOENT); | |
2917 | if (check_path_shared(root, path)) | |
2918 | goto out; | |
2919 | ||
2920 | err = 0; | |
2921 | out: | |
2922 | btrfs_free_path(path); | |
2923 | if (err) { | |
2924 | btrfs_end_transaction(trans, root); | |
2925 | root->fs_info->enospc_unlink = 0; | |
2926 | return ERR_PTR(err); | |
2927 | } | |
2928 | ||
2929 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2930 | return trans; | |
2931 | } | |
2932 | ||
2933 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2934 | struct btrfs_root *root) | |
2935 | { | |
2936 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
2937 | BUG_ON(!root->fs_info->enospc_unlink); | |
2938 | root->fs_info->enospc_unlink = 0; | |
2939 | } | |
2940 | btrfs_end_transaction_throttle(trans, root); | |
2941 | } | |
2942 | ||
2943 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2944 | { | |
2945 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2946 | struct btrfs_trans_handle *trans; | |
2947 | struct inode *inode = dentry->d_inode; | |
2948 | int ret; | |
2949 | unsigned long nr = 0; | |
2950 | ||
2951 | trans = __unlink_start_trans(dir, dentry); | |
2952 | if (IS_ERR(trans)) | |
2953 | return PTR_ERR(trans); | |
5f39d397 | 2954 | |
39279cc3 | 2955 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2956 | |
2957 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2958 | ||
e02119d5 CM |
2959 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2960 | dentry->d_name.name, dentry->d_name.len); | |
a22285a6 | 2961 | BUG_ON(ret); |
7b128766 | 2962 | |
a22285a6 | 2963 | if (inode->i_nlink == 0) { |
7b128766 | 2964 | ret = btrfs_orphan_add(trans, inode); |
a22285a6 YZ |
2965 | BUG_ON(ret); |
2966 | } | |
7b128766 | 2967 | |
d3c2fdcf | 2968 | nr = trans->blocks_used; |
a22285a6 | 2969 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2970 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2971 | return ret; |
2972 | } | |
2973 | ||
4df27c4d YZ |
2974 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2975 | struct btrfs_root *root, | |
2976 | struct inode *dir, u64 objectid, | |
2977 | const char *name, int name_len) | |
2978 | { | |
2979 | struct btrfs_path *path; | |
2980 | struct extent_buffer *leaf; | |
2981 | struct btrfs_dir_item *di; | |
2982 | struct btrfs_key key; | |
2983 | u64 index; | |
2984 | int ret; | |
2985 | ||
2986 | path = btrfs_alloc_path(); | |
2987 | if (!path) | |
2988 | return -ENOMEM; | |
2989 | ||
2990 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2991 | name, name_len, -1); | |
2992 | BUG_ON(!di || IS_ERR(di)); | |
2993 | ||
2994 | leaf = path->nodes[0]; | |
2995 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2996 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2997 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2998 | BUG_ON(ret); | |
2999 | btrfs_release_path(root, path); | |
3000 | ||
3001 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3002 | objectid, root->root_key.objectid, | |
3003 | dir->i_ino, &index, name, name_len); | |
3004 | if (ret < 0) { | |
3005 | BUG_ON(ret != -ENOENT); | |
3006 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
3007 | name, name_len); | |
3008 | BUG_ON(!di || IS_ERR(di)); | |
3009 | ||
3010 | leaf = path->nodes[0]; | |
3011 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3012 | btrfs_release_path(root, path); | |
3013 | index = key.offset; | |
3014 | } | |
3015 | ||
3016 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
3017 | index, name, name_len, -1); | |
3018 | BUG_ON(!di || IS_ERR(di)); | |
3019 | ||
3020 | leaf = path->nodes[0]; | |
3021 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3022 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3023 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3024 | BUG_ON(ret); | |
3025 | btrfs_release_path(root, path); | |
3026 | ||
3027 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
3028 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
3029 | ret = btrfs_update_inode(trans, root, dir); | |
3030 | BUG_ON(ret); | |
4df27c4d YZ |
3031 | |
3032 | btrfs_free_path(path); | |
3033 | return 0; | |
3034 | } | |
3035 | ||
39279cc3 CM |
3036 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3037 | { | |
3038 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3039 | int err = 0; |
39279cc3 | 3040 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3041 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3042 | unsigned long nr = 0; |
39279cc3 | 3043 | |
3394e160 | 3044 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 3045 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3046 | return -ENOTEMPTY; |
3047 | ||
a22285a6 YZ |
3048 | trans = __unlink_start_trans(dir, dentry); |
3049 | if (IS_ERR(trans)) | |
5df6a9f6 | 3050 | return PTR_ERR(trans); |
5df6a9f6 | 3051 | |
39279cc3 | 3052 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 3053 | |
4df27c4d YZ |
3054 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
3055 | err = btrfs_unlink_subvol(trans, root, dir, | |
3056 | BTRFS_I(inode)->location.objectid, | |
3057 | dentry->d_name.name, | |
3058 | dentry->d_name.len); | |
3059 | goto out; | |
3060 | } | |
3061 | ||
7b128766 JB |
3062 | err = btrfs_orphan_add(trans, inode); |
3063 | if (err) | |
4df27c4d | 3064 | goto out; |
7b128766 | 3065 | |
39279cc3 | 3066 | /* now the directory is empty */ |
e02119d5 CM |
3067 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3068 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3069 | if (!err) |
dbe674a9 | 3070 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3071 | out: |
d3c2fdcf | 3072 | nr = trans->blocks_used; |
a22285a6 | 3073 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3074 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3075 | |
39279cc3 CM |
3076 | return err; |
3077 | } | |
3078 | ||
d20f7043 | 3079 | #if 0 |
323ac95b CM |
3080 | /* |
3081 | * when truncating bytes in a file, it is possible to avoid reading | |
3082 | * the leaves that contain only checksum items. This can be the | |
3083 | * majority of the IO required to delete a large file, but it must | |
3084 | * be done carefully. | |
3085 | * | |
3086 | * The keys in the level just above the leaves are checked to make sure | |
3087 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
3088 | * after the new size. | |
3089 | * | |
3090 | * Then the key for the next leaf is checked to make sure it also has | |
3091 | * a checksum item for the same file. If it does, we know our target leaf | |
3092 | * contains only checksum items, and it can be safely freed without reading | |
3093 | * it. | |
3094 | * | |
3095 | * This is just an optimization targeted at large files. It may do | |
3096 | * nothing. It will return 0 unless things went badly. | |
3097 | */ | |
3098 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
3099 | struct btrfs_root *root, | |
3100 | struct btrfs_path *path, | |
3101 | struct inode *inode, u64 new_size) | |
3102 | { | |
3103 | struct btrfs_key key; | |
3104 | int ret; | |
3105 | int nritems; | |
3106 | struct btrfs_key found_key; | |
3107 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
3108 | struct btrfs_leaf_ref *ref; |
3109 | u64 leaf_gen; | |
3110 | u64 leaf_start; | |
323ac95b CM |
3111 | |
3112 | path->lowest_level = 1; | |
3113 | key.objectid = inode->i_ino; | |
3114 | key.type = BTRFS_CSUM_ITEM_KEY; | |
3115 | key.offset = new_size; | |
3116 | again: | |
3117 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3118 | if (ret < 0) | |
3119 | goto out; | |
3120 | ||
3121 | if (path->nodes[1] == NULL) { | |
3122 | ret = 0; | |
3123 | goto out; | |
3124 | } | |
3125 | ret = 0; | |
3126 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
3127 | nritems = btrfs_header_nritems(path->nodes[1]); | |
3128 | ||
3129 | if (!nritems) | |
3130 | goto out; | |
3131 | ||
3132 | if (path->slots[1] >= nritems) | |
3133 | goto next_node; | |
3134 | ||
3135 | /* did we find a key greater than anything we want to delete? */ | |
3136 | if (found_key.objectid > inode->i_ino || | |
3137 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
3138 | goto out; | |
3139 | ||
3140 | /* we check the next key in the node to make sure the leave contains | |
3141 | * only checksum items. This comparison doesn't work if our | |
3142 | * leaf is the last one in the node | |
3143 | */ | |
3144 | if (path->slots[1] + 1 >= nritems) { | |
3145 | next_node: | |
3146 | /* search forward from the last key in the node, this | |
3147 | * will bring us into the next node in the tree | |
3148 | */ | |
3149 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
3150 | ||
3151 | /* unlikely, but we inc below, so check to be safe */ | |
3152 | if (found_key.offset == (u64)-1) | |
3153 | goto out; | |
3154 | ||
3155 | /* search_forward needs a path with locks held, do the | |
3156 | * search again for the original key. It is possible | |
3157 | * this will race with a balance and return a path that | |
3158 | * we could modify, but this drop is just an optimization | |
3159 | * and is allowed to miss some leaves. | |
3160 | */ | |
3161 | btrfs_release_path(root, path); | |
3162 | found_key.offset++; | |
3163 | ||
3164 | /* setup a max key for search_forward */ | |
3165 | other_key.offset = (u64)-1; | |
3166 | other_key.type = key.type; | |
3167 | other_key.objectid = key.objectid; | |
3168 | ||
3169 | path->keep_locks = 1; | |
3170 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
3171 | path, 0, 0); | |
3172 | path->keep_locks = 0; | |
3173 | if (ret || found_key.objectid != key.objectid || | |
3174 | found_key.type != key.type) { | |
3175 | ret = 0; | |
3176 | goto out; | |
3177 | } | |
3178 | ||
3179 | key.offset = found_key.offset; | |
3180 | btrfs_release_path(root, path); | |
3181 | cond_resched(); | |
3182 | goto again; | |
3183 | } | |
3184 | ||
3185 | /* we know there's one more slot after us in the tree, | |
3186 | * read that key so we can verify it is also a checksum item | |
3187 | */ | |
3188 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
3189 | ||
3190 | if (found_key.objectid < inode->i_ino) | |
3191 | goto next_key; | |
3192 | ||
3193 | if (found_key.type != key.type || found_key.offset < new_size) | |
3194 | goto next_key; | |
3195 | ||
3196 | /* | |
3197 | * if the key for the next leaf isn't a csum key from this objectid, | |
3198 | * we can't be sure there aren't good items inside this leaf. | |
3199 | * Bail out | |
3200 | */ | |
3201 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
3202 | goto out; | |
3203 | ||
5b84e8d6 YZ |
3204 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
3205 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
3206 | /* |
3207 | * it is safe to delete this leaf, it contains only | |
3208 | * csum items from this inode at an offset >= new_size | |
3209 | */ | |
5b84e8d6 | 3210 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
3211 | BUG_ON(ret); |
3212 | ||
5b84e8d6 YZ |
3213 | if (root->ref_cows && leaf_gen < trans->transid) { |
3214 | ref = btrfs_alloc_leaf_ref(root, 0); | |
3215 | if (ref) { | |
3216 | ref->root_gen = root->root_key.offset; | |
3217 | ref->bytenr = leaf_start; | |
3218 | ref->owner = 0; | |
3219 | ref->generation = leaf_gen; | |
3220 | ref->nritems = 0; | |
3221 | ||
bd56b302 CM |
3222 | btrfs_sort_leaf_ref(ref); |
3223 | ||
5b84e8d6 YZ |
3224 | ret = btrfs_add_leaf_ref(root, ref, 0); |
3225 | WARN_ON(ret); | |
3226 | btrfs_free_leaf_ref(root, ref); | |
3227 | } else { | |
3228 | WARN_ON(1); | |
3229 | } | |
3230 | } | |
323ac95b CM |
3231 | next_key: |
3232 | btrfs_release_path(root, path); | |
3233 | ||
3234 | if (other_key.objectid == inode->i_ino && | |
3235 | other_key.type == key.type && other_key.offset > key.offset) { | |
3236 | key.offset = other_key.offset; | |
3237 | cond_resched(); | |
3238 | goto again; | |
3239 | } | |
3240 | ret = 0; | |
3241 | out: | |
3242 | /* fixup any changes we've made to the path */ | |
3243 | path->lowest_level = 0; | |
3244 | path->keep_locks = 0; | |
3245 | btrfs_release_path(root, path); | |
3246 | return ret; | |
3247 | } | |
3248 | ||
d20f7043 CM |
3249 | #endif |
3250 | ||
39279cc3 CM |
3251 | /* |
3252 | * this can truncate away extent items, csum items and directory items. | |
3253 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3254 | * any higher than new_size |
39279cc3 CM |
3255 | * |
3256 | * csum items that cross the new i_size are truncated to the new size | |
3257 | * as well. | |
7b128766 JB |
3258 | * |
3259 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3260 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3261 | */ |
8082510e YZ |
3262 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3263 | struct btrfs_root *root, | |
3264 | struct inode *inode, | |
3265 | u64 new_size, u32 min_type) | |
39279cc3 | 3266 | { |
39279cc3 | 3267 | struct btrfs_path *path; |
5f39d397 | 3268 | struct extent_buffer *leaf; |
39279cc3 | 3269 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3270 | struct btrfs_key key; |
3271 | struct btrfs_key found_key; | |
39279cc3 | 3272 | u64 extent_start = 0; |
db94535d | 3273 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3274 | u64 extent_offset = 0; |
39279cc3 | 3275 | u64 item_end = 0; |
8082510e YZ |
3276 | u64 mask = root->sectorsize - 1; |
3277 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3278 | int found_extent; |
3279 | int del_item; | |
85e21bac CM |
3280 | int pending_del_nr = 0; |
3281 | int pending_del_slot = 0; | |
179e29e4 | 3282 | int extent_type = -1; |
771ed689 | 3283 | int encoding; |
8082510e YZ |
3284 | int ret; |
3285 | int err = 0; | |
3286 | ||
3287 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3288 | |
0af3d00b | 3289 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3290 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3291 | |
39279cc3 CM |
3292 | path = btrfs_alloc_path(); |
3293 | BUG_ON(!path); | |
33c17ad5 | 3294 | path->reada = -1; |
5f39d397 | 3295 | |
39279cc3 CM |
3296 | key.objectid = inode->i_ino; |
3297 | key.offset = (u64)-1; | |
5f39d397 CM |
3298 | key.type = (u8)-1; |
3299 | ||
85e21bac | 3300 | search_again: |
b9473439 | 3301 | path->leave_spinning = 1; |
85e21bac | 3302 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3303 | if (ret < 0) { |
3304 | err = ret; | |
3305 | goto out; | |
3306 | } | |
d397712b | 3307 | |
85e21bac | 3308 | if (ret > 0) { |
e02119d5 CM |
3309 | /* there are no items in the tree for us to truncate, we're |
3310 | * done | |
3311 | */ | |
8082510e YZ |
3312 | if (path->slots[0] == 0) |
3313 | goto out; | |
85e21bac CM |
3314 | path->slots[0]--; |
3315 | } | |
3316 | ||
d397712b | 3317 | while (1) { |
39279cc3 | 3318 | fi = NULL; |
5f39d397 CM |
3319 | leaf = path->nodes[0]; |
3320 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3321 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 3322 | encoding = 0; |
39279cc3 | 3323 | |
5f39d397 | 3324 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 3325 | break; |
5f39d397 | 3326 | |
85e21bac | 3327 | if (found_type < min_type) |
39279cc3 CM |
3328 | break; |
3329 | ||
5f39d397 | 3330 | item_end = found_key.offset; |
39279cc3 | 3331 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3332 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3333 | struct btrfs_file_extent_item); |
179e29e4 | 3334 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3335 | encoding = btrfs_file_extent_compression(leaf, fi); |
3336 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3337 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3338 | ||
179e29e4 | 3339 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3340 | item_end += |
db94535d | 3341 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3342 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3343 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3344 | fi); |
39279cc3 | 3345 | } |
008630c1 | 3346 | item_end--; |
39279cc3 | 3347 | } |
8082510e YZ |
3348 | if (found_type > min_type) { |
3349 | del_item = 1; | |
3350 | } else { | |
3351 | if (item_end < new_size) | |
b888db2b | 3352 | break; |
8082510e YZ |
3353 | if (found_key.offset >= new_size) |
3354 | del_item = 1; | |
3355 | else | |
3356 | del_item = 0; | |
39279cc3 | 3357 | } |
39279cc3 | 3358 | found_extent = 0; |
39279cc3 | 3359 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3360 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3361 | goto delete; | |
3362 | ||
3363 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3364 | u64 num_dec; |
db94535d | 3365 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3366 | if (!del_item && !encoding) { |
db94535d CM |
3367 | u64 orig_num_bytes = |
3368 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3369 | extent_num_bytes = new_size - |
5f39d397 | 3370 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3371 | extent_num_bytes = extent_num_bytes & |
3372 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3373 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3374 | extent_num_bytes); | |
3375 | num_dec = (orig_num_bytes - | |
9069218d | 3376 | extent_num_bytes); |
e02119d5 | 3377 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3378 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3379 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3380 | } else { |
db94535d CM |
3381 | extent_num_bytes = |
3382 | btrfs_file_extent_disk_num_bytes(leaf, | |
3383 | fi); | |
5d4f98a2 YZ |
3384 | extent_offset = found_key.offset - |
3385 | btrfs_file_extent_offset(leaf, fi); | |
3386 | ||
39279cc3 | 3387 | /* FIXME blocksize != 4096 */ |
9069218d | 3388 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3389 | if (extent_start != 0) { |
3390 | found_extent = 1; | |
e02119d5 | 3391 | if (root->ref_cows) |
a76a3cd4 | 3392 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3393 | } |
39279cc3 | 3394 | } |
9069218d | 3395 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3396 | /* |
3397 | * we can't truncate inline items that have had | |
3398 | * special encodings | |
3399 | */ | |
3400 | if (!del_item && | |
3401 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3402 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3403 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3404 | u32 size = new_size - found_key.offset; |
3405 | ||
3406 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3407 | inode_sub_bytes(inode, item_end + 1 - |
3408 | new_size); | |
e02119d5 CM |
3409 | } |
3410 | size = | |
3411 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3412 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3413 | size, 1); |
9069218d | 3414 | BUG_ON(ret); |
e02119d5 | 3415 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3416 | inode_sub_bytes(inode, item_end + 1 - |
3417 | found_key.offset); | |
9069218d | 3418 | } |
39279cc3 | 3419 | } |
179e29e4 | 3420 | delete: |
39279cc3 | 3421 | if (del_item) { |
85e21bac CM |
3422 | if (!pending_del_nr) { |
3423 | /* no pending yet, add ourselves */ | |
3424 | pending_del_slot = path->slots[0]; | |
3425 | pending_del_nr = 1; | |
3426 | } else if (pending_del_nr && | |
3427 | path->slots[0] + 1 == pending_del_slot) { | |
3428 | /* hop on the pending chunk */ | |
3429 | pending_del_nr++; | |
3430 | pending_del_slot = path->slots[0]; | |
3431 | } else { | |
d397712b | 3432 | BUG(); |
85e21bac | 3433 | } |
39279cc3 CM |
3434 | } else { |
3435 | break; | |
3436 | } | |
0af3d00b JB |
3437 | if (found_extent && (root->ref_cows || |
3438 | root == root->fs_info->tree_root)) { | |
b9473439 | 3439 | btrfs_set_path_blocking(path); |
39279cc3 | 3440 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3441 | extent_num_bytes, 0, |
3442 | btrfs_header_owner(leaf), | |
3443 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3444 | BUG_ON(ret); |
3445 | } | |
85e21bac | 3446 | |
8082510e YZ |
3447 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3448 | break; | |
3449 | ||
3450 | if (path->slots[0] == 0 || | |
3451 | path->slots[0] != pending_del_slot) { | |
3452 | if (root->ref_cows) { | |
3453 | err = -EAGAIN; | |
3454 | goto out; | |
3455 | } | |
3456 | if (pending_del_nr) { | |
3457 | ret = btrfs_del_items(trans, root, path, | |
3458 | pending_del_slot, | |
3459 | pending_del_nr); | |
3460 | BUG_ON(ret); | |
3461 | pending_del_nr = 0; | |
3462 | } | |
85e21bac CM |
3463 | btrfs_release_path(root, path); |
3464 | goto search_again; | |
8082510e YZ |
3465 | } else { |
3466 | path->slots[0]--; | |
85e21bac | 3467 | } |
39279cc3 | 3468 | } |
8082510e | 3469 | out: |
85e21bac CM |
3470 | if (pending_del_nr) { |
3471 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3472 | pending_del_nr); | |
d68fc57b | 3473 | BUG_ON(ret); |
85e21bac | 3474 | } |
39279cc3 | 3475 | btrfs_free_path(path); |
8082510e | 3476 | return err; |
39279cc3 CM |
3477 | } |
3478 | ||
3479 | /* | |
3480 | * taken from block_truncate_page, but does cow as it zeros out | |
3481 | * any bytes left in the last page in the file. | |
3482 | */ | |
3483 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3484 | { | |
3485 | struct inode *inode = mapping->host; | |
db94535d | 3486 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3487 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3488 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3489 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3490 | char *kaddr; |
db94535d | 3491 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3492 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3493 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3494 | struct page *page; | |
39279cc3 | 3495 | int ret = 0; |
a52d9a80 | 3496 | u64 page_start; |
e6dcd2dc | 3497 | u64 page_end; |
39279cc3 CM |
3498 | |
3499 | if ((offset & (blocksize - 1)) == 0) | |
3500 | goto out; | |
0ca1f7ce | 3501 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3502 | if (ret) |
3503 | goto out; | |
39279cc3 CM |
3504 | |
3505 | ret = -ENOMEM; | |
211c17f5 | 3506 | again: |
39279cc3 | 3507 | page = grab_cache_page(mapping, index); |
5d5e103a | 3508 | if (!page) { |
0ca1f7ce | 3509 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3510 | goto out; |
5d5e103a | 3511 | } |
e6dcd2dc CM |
3512 | |
3513 | page_start = page_offset(page); | |
3514 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3515 | ||
39279cc3 | 3516 | if (!PageUptodate(page)) { |
9ebefb18 | 3517 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3518 | lock_page(page); |
211c17f5 CM |
3519 | if (page->mapping != mapping) { |
3520 | unlock_page(page); | |
3521 | page_cache_release(page); | |
3522 | goto again; | |
3523 | } | |
39279cc3 CM |
3524 | if (!PageUptodate(page)) { |
3525 | ret = -EIO; | |
89642229 | 3526 | goto out_unlock; |
39279cc3 CM |
3527 | } |
3528 | } | |
211c17f5 | 3529 | wait_on_page_writeback(page); |
e6dcd2dc | 3530 | |
2ac55d41 JB |
3531 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3532 | GFP_NOFS); | |
e6dcd2dc CM |
3533 | set_page_extent_mapped(page); |
3534 | ||
3535 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3536 | if (ordered) { | |
2ac55d41 JB |
3537 | unlock_extent_cached(io_tree, page_start, page_end, |
3538 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3539 | unlock_page(page); |
3540 | page_cache_release(page); | |
eb84ae03 | 3541 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3542 | btrfs_put_ordered_extent(ordered); |
3543 | goto again; | |
3544 | } | |
3545 | ||
2ac55d41 | 3546 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3547 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3548 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3549 | |
2ac55d41 JB |
3550 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3551 | &cached_state); | |
9ed74f2d | 3552 | if (ret) { |
2ac55d41 JB |
3553 | unlock_extent_cached(io_tree, page_start, page_end, |
3554 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3555 | goto out_unlock; |
3556 | } | |
3557 | ||
e6dcd2dc CM |
3558 | ret = 0; |
3559 | if (offset != PAGE_CACHE_SIZE) { | |
3560 | kaddr = kmap(page); | |
3561 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3562 | flush_dcache_page(page); | |
3563 | kunmap(page); | |
3564 | } | |
247e743c | 3565 | ClearPageChecked(page); |
e6dcd2dc | 3566 | set_page_dirty(page); |
2ac55d41 JB |
3567 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3568 | GFP_NOFS); | |
39279cc3 | 3569 | |
89642229 | 3570 | out_unlock: |
5d5e103a | 3571 | if (ret) |
0ca1f7ce | 3572 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3573 | unlock_page(page); |
3574 | page_cache_release(page); | |
3575 | out: | |
3576 | return ret; | |
3577 | } | |
3578 | ||
695a0d0d JB |
3579 | /* |
3580 | * This function puts in dummy file extents for the area we're creating a hole | |
3581 | * for. So if we are truncating this file to a larger size we need to insert | |
3582 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3583 | * the range between oldsize and size | |
3584 | */ | |
a41ad394 | 3585 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3586 | { |
9036c102 YZ |
3587 | struct btrfs_trans_handle *trans; |
3588 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3589 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3590 | struct extent_map *em = NULL; |
2ac55d41 | 3591 | struct extent_state *cached_state = NULL; |
9036c102 | 3592 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3593 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3594 | u64 block_end = (size + mask) & ~mask; |
3595 | u64 last_byte; | |
3596 | u64 cur_offset; | |
3597 | u64 hole_size; | |
9ed74f2d | 3598 | int err = 0; |
39279cc3 | 3599 | |
9036c102 YZ |
3600 | if (size <= hole_start) |
3601 | return 0; | |
3602 | ||
9036c102 YZ |
3603 | while (1) { |
3604 | struct btrfs_ordered_extent *ordered; | |
3605 | btrfs_wait_ordered_range(inode, hole_start, | |
3606 | block_end - hole_start); | |
2ac55d41 JB |
3607 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3608 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3609 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3610 | if (!ordered) | |
3611 | break; | |
2ac55d41 JB |
3612 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3613 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3614 | btrfs_put_ordered_extent(ordered); |
3615 | } | |
39279cc3 | 3616 | |
9036c102 YZ |
3617 | cur_offset = hole_start; |
3618 | while (1) { | |
3619 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3620 | block_end - cur_offset, 0); | |
3621 | BUG_ON(IS_ERR(em) || !em); | |
3622 | last_byte = min(extent_map_end(em), block_end); | |
3623 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3624 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3625 | u64 hint_byte = 0; |
9036c102 | 3626 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3627 | |
a22285a6 YZ |
3628 | trans = btrfs_start_transaction(root, 2); |
3629 | if (IS_ERR(trans)) { | |
3630 | err = PTR_ERR(trans); | |
9ed74f2d | 3631 | break; |
a22285a6 | 3632 | } |
8082510e YZ |
3633 | btrfs_set_trans_block_group(trans, inode); |
3634 | ||
3635 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3636 | cur_offset + hole_size, | |
3637 | &hint_byte, 1); | |
3893e33b JB |
3638 | if (err) |
3639 | break; | |
8082510e | 3640 | |
9036c102 YZ |
3641 | err = btrfs_insert_file_extent(trans, root, |
3642 | inode->i_ino, cur_offset, 0, | |
3643 | 0, hole_size, 0, hole_size, | |
3644 | 0, 0, 0); | |
3893e33b JB |
3645 | if (err) |
3646 | break; | |
8082510e | 3647 | |
9036c102 YZ |
3648 | btrfs_drop_extent_cache(inode, hole_start, |
3649 | last_byte - 1, 0); | |
8082510e YZ |
3650 | |
3651 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3652 | } |
3653 | free_extent_map(em); | |
a22285a6 | 3654 | em = NULL; |
9036c102 | 3655 | cur_offset = last_byte; |
8082510e | 3656 | if (cur_offset >= block_end) |
9036c102 YZ |
3657 | break; |
3658 | } | |
1832a6d5 | 3659 | |
a22285a6 | 3660 | free_extent_map(em); |
2ac55d41 JB |
3661 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3662 | GFP_NOFS); | |
9036c102 YZ |
3663 | return err; |
3664 | } | |
39279cc3 | 3665 | |
a41ad394 | 3666 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3667 | { |
a41ad394 | 3668 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3669 | int ret; |
3670 | ||
a41ad394 | 3671 | if (newsize == oldsize) |
8082510e YZ |
3672 | return 0; |
3673 | ||
a41ad394 JB |
3674 | if (newsize > oldsize) { |
3675 | i_size_write(inode, newsize); | |
3676 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3677 | truncate_pagecache(inode, oldsize, newsize); | |
3678 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
8082510e | 3679 | if (ret) { |
a41ad394 | 3680 | btrfs_setsize(inode, oldsize); |
8082510e YZ |
3681 | return ret; |
3682 | } | |
3683 | ||
930f028a | 3684 | mark_inode_dirty(inode); |
a41ad394 | 3685 | } else { |
8082510e | 3686 | |
a41ad394 JB |
3687 | /* |
3688 | * We're truncating a file that used to have good data down to | |
3689 | * zero. Make sure it gets into the ordered flush list so that | |
3690 | * any new writes get down to disk quickly. | |
3691 | */ | |
3692 | if (newsize == 0) | |
3693 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3694 | |
a41ad394 JB |
3695 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3696 | truncate_setsize(inode, newsize); | |
3697 | ret = btrfs_truncate(inode); | |
3698 | } | |
8082510e | 3699 | |
a41ad394 | 3700 | return ret; |
8082510e YZ |
3701 | } |
3702 | ||
9036c102 YZ |
3703 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3704 | { | |
3705 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3706 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3707 | int err; |
39279cc3 | 3708 | |
b83cc969 LZ |
3709 | if (btrfs_root_readonly(root)) |
3710 | return -EROFS; | |
3711 | ||
9036c102 YZ |
3712 | err = inode_change_ok(inode, attr); |
3713 | if (err) | |
3714 | return err; | |
2bf5a725 | 3715 | |
5a3f23d5 | 3716 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3717 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3718 | if (err) |
3719 | return err; | |
39279cc3 | 3720 | } |
9036c102 | 3721 | |
1025774c CH |
3722 | if (attr->ia_valid) { |
3723 | setattr_copy(inode, attr); | |
3724 | mark_inode_dirty(inode); | |
3725 | ||
3726 | if (attr->ia_valid & ATTR_MODE) | |
3727 | err = btrfs_acl_chmod(inode); | |
3728 | } | |
33268eaf | 3729 | |
39279cc3 CM |
3730 | return err; |
3731 | } | |
61295eb8 | 3732 | |
bd555975 | 3733 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3734 | { |
3735 | struct btrfs_trans_handle *trans; | |
3736 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3737 | unsigned long nr; |
39279cc3 CM |
3738 | int ret; |
3739 | ||
1abe9b8a | 3740 | trace_btrfs_inode_evict(inode); |
3741 | ||
39279cc3 | 3742 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b JB |
3743 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
3744 | root == root->fs_info->tree_root)) | |
bd555975 AV |
3745 | goto no_delete; |
3746 | ||
39279cc3 | 3747 | if (is_bad_inode(inode)) { |
7b128766 | 3748 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3749 | goto no_delete; |
3750 | } | |
bd555975 | 3751 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3752 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3753 | |
c71bf099 YZ |
3754 | if (root->fs_info->log_root_recovering) { |
3755 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3756 | goto no_delete; | |
3757 | } | |
3758 | ||
76dda93c YZ |
3759 | if (inode->i_nlink > 0) { |
3760 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3761 | goto no_delete; | |
3762 | } | |
3763 | ||
dbe674a9 | 3764 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3765 | |
8082510e | 3766 | while (1) { |
d68fc57b YZ |
3767 | trans = btrfs_start_transaction(root, 0); |
3768 | BUG_ON(IS_ERR(trans)); | |
8082510e | 3769 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b YZ |
3770 | trans->block_rsv = root->orphan_block_rsv; |
3771 | ||
3772 | ret = btrfs_block_rsv_check(trans, root, | |
3773 | root->orphan_block_rsv, 0, 5); | |
3774 | if (ret) { | |
3775 | BUG_ON(ret != -EAGAIN); | |
3776 | ret = btrfs_commit_transaction(trans, root); | |
3777 | BUG_ON(ret); | |
3778 | continue; | |
3779 | } | |
7b128766 | 3780 | |
d68fc57b | 3781 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3782 | if (ret != -EAGAIN) |
3783 | break; | |
85e21bac | 3784 | |
8082510e YZ |
3785 | nr = trans->blocks_used; |
3786 | btrfs_end_transaction(trans, root); | |
3787 | trans = NULL; | |
3788 | btrfs_btree_balance_dirty(root, nr); | |
d68fc57b | 3789 | |
8082510e | 3790 | } |
5f39d397 | 3791 | |
8082510e YZ |
3792 | if (ret == 0) { |
3793 | ret = btrfs_orphan_del(trans, inode); | |
3794 | BUG_ON(ret); | |
3795 | } | |
54aa1f4d | 3796 | |
d3c2fdcf | 3797 | nr = trans->blocks_used; |
54aa1f4d | 3798 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3799 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3800 | no_delete: |
bd555975 | 3801 | end_writeback(inode); |
8082510e | 3802 | return; |
39279cc3 CM |
3803 | } |
3804 | ||
3805 | /* | |
3806 | * this returns the key found in the dir entry in the location pointer. | |
3807 | * If no dir entries were found, location->objectid is 0. | |
3808 | */ | |
3809 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3810 | struct btrfs_key *location) | |
3811 | { | |
3812 | const char *name = dentry->d_name.name; | |
3813 | int namelen = dentry->d_name.len; | |
3814 | struct btrfs_dir_item *di; | |
3815 | struct btrfs_path *path; | |
3816 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3817 | int ret = 0; |
39279cc3 CM |
3818 | |
3819 | path = btrfs_alloc_path(); | |
3820 | BUG_ON(!path); | |
3954401f | 3821 | |
39279cc3 CM |
3822 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3823 | namelen, 0); | |
0d9f7f3e Y |
3824 | if (IS_ERR(di)) |
3825 | ret = PTR_ERR(di); | |
d397712b CM |
3826 | |
3827 | if (!di || IS_ERR(di)) | |
3954401f | 3828 | goto out_err; |
d397712b | 3829 | |
5f39d397 | 3830 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3831 | out: |
39279cc3 CM |
3832 | btrfs_free_path(path); |
3833 | return ret; | |
3954401f CM |
3834 | out_err: |
3835 | location->objectid = 0; | |
3836 | goto out; | |
39279cc3 CM |
3837 | } |
3838 | ||
3839 | /* | |
3840 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3841 | * needs to be changed to reflect the root directory of the tree root. This | |
3842 | * is kind of like crossing a mount point. | |
3843 | */ | |
3844 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3845 | struct inode *dir, |
3846 | struct dentry *dentry, | |
3847 | struct btrfs_key *location, | |
3848 | struct btrfs_root **sub_root) | |
39279cc3 | 3849 | { |
4df27c4d YZ |
3850 | struct btrfs_path *path; |
3851 | struct btrfs_root *new_root; | |
3852 | struct btrfs_root_ref *ref; | |
3853 | struct extent_buffer *leaf; | |
3854 | int ret; | |
3855 | int err = 0; | |
39279cc3 | 3856 | |
4df27c4d YZ |
3857 | path = btrfs_alloc_path(); |
3858 | if (!path) { | |
3859 | err = -ENOMEM; | |
3860 | goto out; | |
3861 | } | |
39279cc3 | 3862 | |
4df27c4d YZ |
3863 | err = -ENOENT; |
3864 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3865 | BTRFS_I(dir)->root->root_key.objectid, | |
3866 | location->objectid); | |
3867 | if (ret) { | |
3868 | if (ret < 0) | |
3869 | err = ret; | |
3870 | goto out; | |
3871 | } | |
39279cc3 | 3872 | |
4df27c4d YZ |
3873 | leaf = path->nodes[0]; |
3874 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3875 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3876 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3877 | goto out; | |
39279cc3 | 3878 | |
4df27c4d YZ |
3879 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3880 | (unsigned long)(ref + 1), | |
3881 | dentry->d_name.len); | |
3882 | if (ret) | |
3883 | goto out; | |
3884 | ||
3885 | btrfs_release_path(root->fs_info->tree_root, path); | |
3886 | ||
3887 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3888 | if (IS_ERR(new_root)) { | |
3889 | err = PTR_ERR(new_root); | |
3890 | goto out; | |
3891 | } | |
3892 | ||
3893 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3894 | err = -ENOENT; | |
3895 | goto out; | |
3896 | } | |
3897 | ||
3898 | *sub_root = new_root; | |
3899 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3900 | location->type = BTRFS_INODE_ITEM_KEY; | |
3901 | location->offset = 0; | |
3902 | err = 0; | |
3903 | out: | |
3904 | btrfs_free_path(path); | |
3905 | return err; | |
39279cc3 CM |
3906 | } |
3907 | ||
5d4f98a2 YZ |
3908 | static void inode_tree_add(struct inode *inode) |
3909 | { | |
3910 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3911 | struct btrfs_inode *entry; | |
03e860bd FNP |
3912 | struct rb_node **p; |
3913 | struct rb_node *parent; | |
03e860bd FNP |
3914 | again: |
3915 | p = &root->inode_tree.rb_node; | |
3916 | parent = NULL; | |
5d4f98a2 | 3917 | |
1d3382cb | 3918 | if (inode_unhashed(inode)) |
76dda93c YZ |
3919 | return; |
3920 | ||
5d4f98a2 YZ |
3921 | spin_lock(&root->inode_lock); |
3922 | while (*p) { | |
3923 | parent = *p; | |
3924 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3925 | ||
3926 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3927 | p = &parent->rb_left; |
5d4f98a2 | 3928 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3929 | p = &parent->rb_right; |
5d4f98a2 YZ |
3930 | else { |
3931 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3932 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3933 | rb_erase(parent, &root->inode_tree); |
3934 | RB_CLEAR_NODE(parent); | |
3935 | spin_unlock(&root->inode_lock); | |
3936 | goto again; | |
5d4f98a2 YZ |
3937 | } |
3938 | } | |
3939 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3940 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3941 | spin_unlock(&root->inode_lock); | |
3942 | } | |
3943 | ||
3944 | static void inode_tree_del(struct inode *inode) | |
3945 | { | |
3946 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3947 | int empty = 0; |
5d4f98a2 | 3948 | |
03e860bd | 3949 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3950 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3951 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3952 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3953 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3954 | } |
03e860bd | 3955 | spin_unlock(&root->inode_lock); |
76dda93c | 3956 | |
0af3d00b JB |
3957 | /* |
3958 | * Free space cache has inodes in the tree root, but the tree root has a | |
3959 | * root_refs of 0, so this could end up dropping the tree root as a | |
3960 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3961 | * make sure we don't drop it. | |
3962 | */ | |
3963 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3964 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3965 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3966 | spin_lock(&root->inode_lock); | |
3967 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3968 | spin_unlock(&root->inode_lock); | |
3969 | if (empty) | |
3970 | btrfs_add_dead_root(root); | |
3971 | } | |
3972 | } | |
3973 | ||
3974 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3975 | { | |
3976 | struct rb_node *node; | |
3977 | struct rb_node *prev; | |
3978 | struct btrfs_inode *entry; | |
3979 | struct inode *inode; | |
3980 | u64 objectid = 0; | |
3981 | ||
3982 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3983 | ||
3984 | spin_lock(&root->inode_lock); | |
3985 | again: | |
3986 | node = root->inode_tree.rb_node; | |
3987 | prev = NULL; | |
3988 | while (node) { | |
3989 | prev = node; | |
3990 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3991 | ||
3992 | if (objectid < entry->vfs_inode.i_ino) | |
3993 | node = node->rb_left; | |
3994 | else if (objectid > entry->vfs_inode.i_ino) | |
3995 | node = node->rb_right; | |
3996 | else | |
3997 | break; | |
3998 | } | |
3999 | if (!node) { | |
4000 | while (prev) { | |
4001 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4002 | if (objectid <= entry->vfs_inode.i_ino) { | |
4003 | node = prev; | |
4004 | break; | |
4005 | } | |
4006 | prev = rb_next(prev); | |
4007 | } | |
4008 | } | |
4009 | while (node) { | |
4010 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4011 | objectid = entry->vfs_inode.i_ino + 1; | |
4012 | inode = igrab(&entry->vfs_inode); | |
4013 | if (inode) { | |
4014 | spin_unlock(&root->inode_lock); | |
4015 | if (atomic_read(&inode->i_count) > 1) | |
4016 | d_prune_aliases(inode); | |
4017 | /* | |
45321ac5 | 4018 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4019 | * the inode cache when its usage count |
4020 | * hits zero. | |
4021 | */ | |
4022 | iput(inode); | |
4023 | cond_resched(); | |
4024 | spin_lock(&root->inode_lock); | |
4025 | goto again; | |
4026 | } | |
4027 | ||
4028 | if (cond_resched_lock(&root->inode_lock)) | |
4029 | goto again; | |
4030 | ||
4031 | node = rb_next(node); | |
4032 | } | |
4033 | spin_unlock(&root->inode_lock); | |
4034 | return 0; | |
5d4f98a2 YZ |
4035 | } |
4036 | ||
e02119d5 CM |
4037 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4038 | { | |
4039 | struct btrfs_iget_args *args = p; | |
4040 | inode->i_ino = args->ino; | |
e02119d5 | 4041 | BTRFS_I(inode)->root = args->root; |
6a63209f | 4042 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
4043 | return 0; |
4044 | } | |
4045 | ||
4046 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4047 | { | |
4048 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
4049 | return args->ino == inode->i_ino && |
4050 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
4051 | } |
4052 | ||
5d4f98a2 YZ |
4053 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4054 | u64 objectid, | |
4055 | struct btrfs_root *root) | |
39279cc3 CM |
4056 | { |
4057 | struct inode *inode; | |
4058 | struct btrfs_iget_args args; | |
4059 | args.ino = objectid; | |
4060 | args.root = root; | |
4061 | ||
4062 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4063 | btrfs_init_locked_inode, | |
4064 | (void *)&args); | |
4065 | return inode; | |
4066 | } | |
4067 | ||
1a54ef8c BR |
4068 | /* Get an inode object given its location and corresponding root. |
4069 | * Returns in *is_new if the inode was read from disk | |
4070 | */ | |
4071 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4072 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4073 | { |
4074 | struct inode *inode; | |
4075 | ||
4076 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4077 | if (!inode) | |
5d4f98a2 | 4078 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4079 | |
4080 | if (inode->i_state & I_NEW) { | |
4081 | BTRFS_I(inode)->root = root; | |
4082 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4083 | btrfs_read_locked_inode(inode); | |
5d4f98a2 | 4084 | inode_tree_add(inode); |
1a54ef8c | 4085 | unlock_new_inode(inode); |
73f73415 JB |
4086 | if (new) |
4087 | *new = 1; | |
1a54ef8c BR |
4088 | } |
4089 | ||
4090 | return inode; | |
4091 | } | |
4092 | ||
4df27c4d YZ |
4093 | static struct inode *new_simple_dir(struct super_block *s, |
4094 | struct btrfs_key *key, | |
4095 | struct btrfs_root *root) | |
4096 | { | |
4097 | struct inode *inode = new_inode(s); | |
4098 | ||
4099 | if (!inode) | |
4100 | return ERR_PTR(-ENOMEM); | |
4101 | ||
4df27c4d YZ |
4102 | BTRFS_I(inode)->root = root; |
4103 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
4104 | BTRFS_I(inode)->dummy_inode = 1; | |
4105 | ||
4106 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
4107 | inode->i_op = &simple_dir_inode_operations; | |
4108 | inode->i_fop = &simple_dir_operations; | |
4109 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4110 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4111 | ||
4112 | return inode; | |
4113 | } | |
4114 | ||
3de4586c | 4115 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4116 | { |
d397712b | 4117 | struct inode *inode; |
4df27c4d | 4118 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4119 | struct btrfs_root *sub_root = root; |
4120 | struct btrfs_key location; | |
76dda93c | 4121 | int index; |
5d4f98a2 | 4122 | int ret; |
39279cc3 CM |
4123 | |
4124 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4125 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4126 | |
39279cc3 | 4127 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 4128 | |
39279cc3 CM |
4129 | if (ret < 0) |
4130 | return ERR_PTR(ret); | |
5f39d397 | 4131 | |
4df27c4d YZ |
4132 | if (location.objectid == 0) |
4133 | return NULL; | |
4134 | ||
4135 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4136 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4137 | return inode; |
4138 | } | |
4139 | ||
4140 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4141 | ||
76dda93c | 4142 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4143 | ret = fixup_tree_root_location(root, dir, dentry, |
4144 | &location, &sub_root); | |
4145 | if (ret < 0) { | |
4146 | if (ret != -ENOENT) | |
4147 | inode = ERR_PTR(ret); | |
4148 | else | |
4149 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4150 | } else { | |
73f73415 | 4151 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4152 | } |
76dda93c YZ |
4153 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4154 | ||
34d19bad | 4155 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4156 | down_read(&root->fs_info->cleanup_work_sem); |
4157 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4158 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4159 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4160 | if (ret) |
4161 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4162 | } |
4163 | ||
3de4586c CM |
4164 | return inode; |
4165 | } | |
4166 | ||
fe15ce44 | 4167 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4168 | { |
4169 | struct btrfs_root *root; | |
4170 | ||
efefb143 YZ |
4171 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
4172 | dentry = dentry->d_parent; | |
76dda93c | 4173 | |
efefb143 YZ |
4174 | if (dentry->d_inode) { |
4175 | root = BTRFS_I(dentry->d_inode)->root; | |
4176 | if (btrfs_root_refs(&root->root_item) == 0) | |
4177 | return 1; | |
4178 | } | |
76dda93c YZ |
4179 | return 0; |
4180 | } | |
4181 | ||
3de4586c CM |
4182 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4183 | struct nameidata *nd) | |
4184 | { | |
4185 | struct inode *inode; | |
4186 | ||
3de4586c CM |
4187 | inode = btrfs_lookup_dentry(dir, dentry); |
4188 | if (IS_ERR(inode)) | |
4189 | return ERR_CAST(inode); | |
7b128766 | 4190 | |
39279cc3 CM |
4191 | return d_splice_alias(inode, dentry); |
4192 | } | |
4193 | ||
39279cc3 CM |
4194 | static unsigned char btrfs_filetype_table[] = { |
4195 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
4196 | }; | |
4197 | ||
cbdf5a24 DW |
4198 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4199 | filldir_t filldir) | |
39279cc3 | 4200 | { |
6da6abae | 4201 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4202 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4203 | struct btrfs_item *item; | |
4204 | struct btrfs_dir_item *di; | |
4205 | struct btrfs_key key; | |
5f39d397 | 4206 | struct btrfs_key found_key; |
39279cc3 CM |
4207 | struct btrfs_path *path; |
4208 | int ret; | |
4209 | u32 nritems; | |
5f39d397 | 4210 | struct extent_buffer *leaf; |
39279cc3 CM |
4211 | int slot; |
4212 | int advance; | |
4213 | unsigned char d_type; | |
4214 | int over = 0; | |
4215 | u32 di_cur; | |
4216 | u32 di_total; | |
4217 | u32 di_len; | |
4218 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4219 | char tmp_name[32]; |
4220 | char *name_ptr; | |
4221 | int name_len; | |
39279cc3 CM |
4222 | |
4223 | /* FIXME, use a real flag for deciding about the key type */ | |
4224 | if (root->fs_info->tree_root == root) | |
4225 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4226 | |
3954401f CM |
4227 | /* special case for "." */ |
4228 | if (filp->f_pos == 0) { | |
4229 | over = filldir(dirent, ".", 1, | |
4230 | 1, inode->i_ino, | |
4231 | DT_DIR); | |
4232 | if (over) | |
4233 | return 0; | |
4234 | filp->f_pos = 1; | |
4235 | } | |
3954401f CM |
4236 | /* special case for .., just use the back ref */ |
4237 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4238 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4239 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 4240 | 2, pino, DT_DIR); |
3954401f | 4241 | if (over) |
49593bfa | 4242 | return 0; |
3954401f CM |
4243 | filp->f_pos = 2; |
4244 | } | |
49593bfa DW |
4245 | path = btrfs_alloc_path(); |
4246 | path->reada = 2; | |
4247 | ||
39279cc3 CM |
4248 | btrfs_set_key_type(&key, key_type); |
4249 | key.offset = filp->f_pos; | |
49593bfa | 4250 | key.objectid = inode->i_ino; |
5f39d397 | 4251 | |
39279cc3 CM |
4252 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4253 | if (ret < 0) | |
4254 | goto err; | |
4255 | advance = 0; | |
49593bfa DW |
4256 | |
4257 | while (1) { | |
5f39d397 CM |
4258 | leaf = path->nodes[0]; |
4259 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4260 | slot = path->slots[0]; |
4261 | if (advance || slot >= nritems) { | |
49593bfa | 4262 | if (slot >= nritems - 1) { |
39279cc3 CM |
4263 | ret = btrfs_next_leaf(root, path); |
4264 | if (ret) | |
4265 | break; | |
5f39d397 CM |
4266 | leaf = path->nodes[0]; |
4267 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4268 | slot = path->slots[0]; |
4269 | } else { | |
4270 | slot++; | |
4271 | path->slots[0]++; | |
4272 | } | |
4273 | } | |
3de4586c | 4274 | |
39279cc3 | 4275 | advance = 1; |
5f39d397 CM |
4276 | item = btrfs_item_nr(leaf, slot); |
4277 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4278 | ||
4279 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4280 | break; |
5f39d397 | 4281 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4282 | break; |
5f39d397 | 4283 | if (found_key.offset < filp->f_pos) |
39279cc3 | 4284 | continue; |
5f39d397 CM |
4285 | |
4286 | filp->f_pos = found_key.offset; | |
49593bfa | 4287 | |
39279cc3 CM |
4288 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4289 | di_cur = 0; | |
5f39d397 | 4290 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4291 | |
4292 | while (di_cur < di_total) { | |
5f39d397 CM |
4293 | struct btrfs_key location; |
4294 | ||
22a94d44 JB |
4295 | if (verify_dir_item(root, leaf, di)) |
4296 | break; | |
4297 | ||
5f39d397 | 4298 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4299 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4300 | name_ptr = tmp_name; |
4301 | } else { | |
4302 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4303 | if (!name_ptr) { |
4304 | ret = -ENOMEM; | |
4305 | goto err; | |
4306 | } | |
5f39d397 CM |
4307 | } |
4308 | read_extent_buffer(leaf, name_ptr, | |
4309 | (unsigned long)(di + 1), name_len); | |
4310 | ||
4311 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4312 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
4313 | |
4314 | /* is this a reference to our own snapshot? If so | |
4315 | * skip it | |
4316 | */ | |
4317 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4318 | location.objectid == root->root_key.objectid) { | |
4319 | over = 0; | |
4320 | goto skip; | |
4321 | } | |
5f39d397 | 4322 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4323 | found_key.offset, location.objectid, |
39279cc3 | 4324 | d_type); |
5f39d397 | 4325 | |
3de4586c | 4326 | skip: |
5f39d397 CM |
4327 | if (name_ptr != tmp_name) |
4328 | kfree(name_ptr); | |
4329 | ||
39279cc3 CM |
4330 | if (over) |
4331 | goto nopos; | |
5103e947 | 4332 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4333 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4334 | di_cur += di_len; |
4335 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4336 | } | |
4337 | } | |
49593bfa DW |
4338 | |
4339 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 4340 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4341 | /* |
4342 | * 32-bit glibc will use getdents64, but then strtol - | |
4343 | * so the last number we can serve is this. | |
4344 | */ | |
4345 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4346 | else |
4347 | filp->f_pos++; | |
39279cc3 CM |
4348 | nopos: |
4349 | ret = 0; | |
4350 | err: | |
39279cc3 | 4351 | btrfs_free_path(path); |
39279cc3 CM |
4352 | return ret; |
4353 | } | |
4354 | ||
a9185b41 | 4355 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4356 | { |
4357 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4358 | struct btrfs_trans_handle *trans; | |
4359 | int ret = 0; | |
0af3d00b | 4360 | bool nolock = false; |
39279cc3 | 4361 | |
8929ecfa | 4362 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4363 | return 0; |
4364 | ||
0af3d00b JB |
4365 | smp_mb(); |
4366 | nolock = (root->fs_info->closing && root == root->fs_info->tree_root); | |
4367 | ||
a9185b41 | 4368 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b JB |
4369 | if (nolock) |
4370 | trans = btrfs_join_transaction_nolock(root, 1); | |
4371 | else | |
4372 | trans = btrfs_join_transaction(root, 1); | |
3612b495 TI |
4373 | if (IS_ERR(trans)) |
4374 | return PTR_ERR(trans); | |
39279cc3 | 4375 | btrfs_set_trans_block_group(trans, inode); |
0af3d00b JB |
4376 | if (nolock) |
4377 | ret = btrfs_end_transaction_nolock(trans, root); | |
4378 | else | |
4379 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4380 | } |
4381 | return ret; | |
4382 | } | |
4383 | ||
4384 | /* | |
54aa1f4d | 4385 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4386 | * inode changes. But, it is most likely to find the inode in cache. |
4387 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4388 | * to keep or drop this code. | |
4389 | */ | |
4390 | void btrfs_dirty_inode(struct inode *inode) | |
4391 | { | |
4392 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4393 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4394 | int ret; |
4395 | ||
4396 | if (BTRFS_I(inode)->dummy_inode) | |
4397 | return; | |
39279cc3 | 4398 | |
f9295749 | 4399 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 4400 | BUG_ON(IS_ERR(trans)); |
39279cc3 | 4401 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa YZ |
4402 | |
4403 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4404 | if (ret && ret == -ENOSPC) { |
4405 | /* whoops, lets try again with the full transaction */ | |
4406 | btrfs_end_transaction(trans, root); | |
4407 | trans = btrfs_start_transaction(root, 1); | |
9aeead73 CM |
4408 | if (IS_ERR(trans)) { |
4409 | if (printk_ratelimit()) { | |
4410 | printk(KERN_ERR "btrfs: fail to " | |
4411 | "dirty inode %lu error %ld\n", | |
4412 | inode->i_ino, PTR_ERR(trans)); | |
4413 | } | |
4414 | return; | |
4415 | } | |
94b60442 | 4416 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa | 4417 | |
94b60442 CM |
4418 | ret = btrfs_update_inode(trans, root, inode); |
4419 | if (ret) { | |
9aeead73 CM |
4420 | if (printk_ratelimit()) { |
4421 | printk(KERN_ERR "btrfs: fail to " | |
4422 | "dirty inode %lu error %d\n", | |
4423 | inode->i_ino, ret); | |
4424 | } | |
94b60442 CM |
4425 | } |
4426 | } | |
39279cc3 | 4427 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4428 | } |
4429 | ||
d352ac68 CM |
4430 | /* |
4431 | * find the highest existing sequence number in a directory | |
4432 | * and then set the in-memory index_cnt variable to reflect | |
4433 | * free sequence numbers | |
4434 | */ | |
aec7477b JB |
4435 | static int btrfs_set_inode_index_count(struct inode *inode) |
4436 | { | |
4437 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4438 | struct btrfs_key key, found_key; | |
4439 | struct btrfs_path *path; | |
4440 | struct extent_buffer *leaf; | |
4441 | int ret; | |
4442 | ||
4443 | key.objectid = inode->i_ino; | |
4444 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
4445 | key.offset = (u64)-1; | |
4446 | ||
4447 | path = btrfs_alloc_path(); | |
4448 | if (!path) | |
4449 | return -ENOMEM; | |
4450 | ||
4451 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4452 | if (ret < 0) | |
4453 | goto out; | |
4454 | /* FIXME: we should be able to handle this */ | |
4455 | if (ret == 0) | |
4456 | goto out; | |
4457 | ret = 0; | |
4458 | ||
4459 | /* | |
4460 | * MAGIC NUMBER EXPLANATION: | |
4461 | * since we search a directory based on f_pos we have to start at 2 | |
4462 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4463 | * else has to start at 2 | |
4464 | */ | |
4465 | if (path->slots[0] == 0) { | |
4466 | BTRFS_I(inode)->index_cnt = 2; | |
4467 | goto out; | |
4468 | } | |
4469 | ||
4470 | path->slots[0]--; | |
4471 | ||
4472 | leaf = path->nodes[0]; | |
4473 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4474 | ||
4475 | if (found_key.objectid != inode->i_ino || | |
4476 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
4477 | BTRFS_I(inode)->index_cnt = 2; | |
4478 | goto out; | |
4479 | } | |
4480 | ||
4481 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4482 | out: | |
4483 | btrfs_free_path(path); | |
4484 | return ret; | |
4485 | } | |
4486 | ||
d352ac68 CM |
4487 | /* |
4488 | * helper to find a free sequence number in a given directory. This current | |
4489 | * code is very simple, later versions will do smarter things in the btree | |
4490 | */ | |
3de4586c | 4491 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4492 | { |
4493 | int ret = 0; | |
4494 | ||
4495 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
4496 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 4497 | if (ret) |
aec7477b JB |
4498 | return ret; |
4499 | } | |
4500 | ||
00e4e6b3 | 4501 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4502 | BTRFS_I(dir)->index_cnt++; |
4503 | ||
4504 | return ret; | |
4505 | } | |
4506 | ||
39279cc3 CM |
4507 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4508 | struct btrfs_root *root, | |
aec7477b | 4509 | struct inode *dir, |
9c58309d | 4510 | const char *name, int name_len, |
d2fb3437 YZ |
4511 | u64 ref_objectid, u64 objectid, |
4512 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4513 | { |
4514 | struct inode *inode; | |
5f39d397 | 4515 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4516 | struct btrfs_key *location; |
5f39d397 | 4517 | struct btrfs_path *path; |
9c58309d CM |
4518 | struct btrfs_inode_ref *ref; |
4519 | struct btrfs_key key[2]; | |
4520 | u32 sizes[2]; | |
4521 | unsigned long ptr; | |
39279cc3 CM |
4522 | int ret; |
4523 | int owner; | |
4524 | ||
5f39d397 CM |
4525 | path = btrfs_alloc_path(); |
4526 | BUG_ON(!path); | |
4527 | ||
39279cc3 | 4528 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4529 | if (!inode) { |
4530 | btrfs_free_path(path); | |
39279cc3 | 4531 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4532 | } |
39279cc3 | 4533 | |
aec7477b | 4534 | if (dir) { |
1abe9b8a | 4535 | trace_btrfs_inode_request(dir); |
4536 | ||
3de4586c | 4537 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4538 | if (ret) { |
8fb27640 | 4539 | btrfs_free_path(path); |
09771430 | 4540 | iput(inode); |
aec7477b | 4541 | return ERR_PTR(ret); |
09771430 | 4542 | } |
aec7477b JB |
4543 | } |
4544 | /* | |
4545 | * index_cnt is ignored for everything but a dir, | |
4546 | * btrfs_get_inode_index_count has an explanation for the magic | |
4547 | * number | |
4548 | */ | |
4549 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4550 | BTRFS_I(inode)->root = root; |
e02119d5 | 4551 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4552 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4553 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4554 | |
39279cc3 CM |
4555 | if (mode & S_IFDIR) |
4556 | owner = 0; | |
4557 | else | |
4558 | owner = 1; | |
d2fb3437 YZ |
4559 | BTRFS_I(inode)->block_group = |
4560 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4561 | |
4562 | key[0].objectid = objectid; | |
4563 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4564 | key[0].offset = 0; | |
4565 | ||
4566 | key[1].objectid = objectid; | |
4567 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4568 | key[1].offset = ref_objectid; | |
4569 | ||
4570 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4571 | sizes[1] = name_len + sizeof(*ref); | |
4572 | ||
b9473439 | 4573 | path->leave_spinning = 1; |
9c58309d CM |
4574 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4575 | if (ret != 0) | |
5f39d397 CM |
4576 | goto fail; |
4577 | ||
ecc11fab | 4578 | inode_init_owner(inode, dir, mode); |
39279cc3 | 4579 | inode->i_ino = objectid; |
a76a3cd4 | 4580 | inode_set_bytes(inode, 0); |
39279cc3 | 4581 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4582 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4583 | struct btrfs_inode_item); | |
e02119d5 | 4584 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4585 | |
4586 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4587 | struct btrfs_inode_ref); | |
4588 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4589 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4590 | ptr = (unsigned long)(ref + 1); |
4591 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4592 | ||
5f39d397 CM |
4593 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4594 | btrfs_free_path(path); | |
4595 | ||
39279cc3 CM |
4596 | location = &BTRFS_I(inode)->location; |
4597 | location->objectid = objectid; | |
39279cc3 CM |
4598 | location->offset = 0; |
4599 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4600 | ||
6cbff00f CH |
4601 | btrfs_inherit_iflags(inode, dir); |
4602 | ||
94272164 CM |
4603 | if ((mode & S_IFREG)) { |
4604 | if (btrfs_test_opt(root, NODATASUM)) | |
4605 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4606 | if (btrfs_test_opt(root, NODATACOW) || |
4607 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4608 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4609 | } | |
4610 | ||
39279cc3 | 4611 | insert_inode_hash(inode); |
5d4f98a2 | 4612 | inode_tree_add(inode); |
1abe9b8a | 4613 | |
4614 | trace_btrfs_inode_new(inode); | |
4615 | ||
39279cc3 | 4616 | return inode; |
5f39d397 | 4617 | fail: |
aec7477b JB |
4618 | if (dir) |
4619 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4620 | btrfs_free_path(path); |
09771430 | 4621 | iput(inode); |
5f39d397 | 4622 | return ERR_PTR(ret); |
39279cc3 CM |
4623 | } |
4624 | ||
4625 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4626 | { | |
4627 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4628 | } | |
4629 | ||
d352ac68 CM |
4630 | /* |
4631 | * utility function to add 'inode' into 'parent_inode' with | |
4632 | * a give name and a given sequence number. | |
4633 | * if 'add_backref' is true, also insert a backref from the | |
4634 | * inode to the parent directory. | |
4635 | */ | |
e02119d5 CM |
4636 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4637 | struct inode *parent_inode, struct inode *inode, | |
4638 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4639 | { |
4df27c4d | 4640 | int ret = 0; |
39279cc3 | 4641 | struct btrfs_key key; |
e02119d5 | 4642 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4643 | |
4df27c4d YZ |
4644 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4645 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4646 | } else { | |
4647 | key.objectid = inode->i_ino; | |
4648 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4649 | key.offset = 0; | |
4650 | } | |
4651 | ||
4652 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4653 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4654 | key.objectid, root->root_key.objectid, | |
4655 | parent_inode->i_ino, | |
4656 | index, name, name_len); | |
4657 | } else if (add_backref) { | |
4658 | ret = btrfs_insert_inode_ref(trans, root, | |
4659 | name, name_len, inode->i_ino, | |
4660 | parent_inode->i_ino, index); | |
4661 | } | |
39279cc3 | 4662 | |
39279cc3 | 4663 | if (ret == 0) { |
4df27c4d YZ |
4664 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4665 | parent_inode->i_ino, &key, | |
4666 | btrfs_inode_type(inode), index); | |
4667 | BUG_ON(ret); | |
4668 | ||
dbe674a9 | 4669 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4670 | name_len * 2); |
79c44584 | 4671 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4672 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4673 | } |
4674 | return ret; | |
4675 | } | |
4676 | ||
4677 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4678 | struct inode *dir, struct dentry *dentry, |
4679 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4680 | { |
a1b075d2 JB |
4681 | int err = btrfs_add_link(trans, dir, inode, |
4682 | dentry->d_name.name, dentry->d_name.len, | |
4683 | backref, index); | |
39279cc3 CM |
4684 | if (!err) { |
4685 | d_instantiate(dentry, inode); | |
4686 | return 0; | |
4687 | } | |
4688 | if (err > 0) | |
4689 | err = -EEXIST; | |
4690 | return err; | |
4691 | } | |
4692 | ||
618e21d5 JB |
4693 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4694 | int mode, dev_t rdev) | |
4695 | { | |
4696 | struct btrfs_trans_handle *trans; | |
4697 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4698 | struct inode *inode = NULL; |
618e21d5 JB |
4699 | int err; |
4700 | int drop_inode = 0; | |
4701 | u64 objectid; | |
1832a6d5 | 4702 | unsigned long nr = 0; |
00e4e6b3 | 4703 | u64 index = 0; |
618e21d5 JB |
4704 | |
4705 | if (!new_valid_dev(rdev)) | |
4706 | return -EINVAL; | |
4707 | ||
a22285a6 YZ |
4708 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4709 | if (err) | |
4710 | return err; | |
4711 | ||
9ed74f2d JB |
4712 | /* |
4713 | * 2 for inode item and ref | |
4714 | * 2 for dir items | |
4715 | * 1 for xattr if selinux is on | |
4716 | */ | |
a22285a6 YZ |
4717 | trans = btrfs_start_transaction(root, 5); |
4718 | if (IS_ERR(trans)) | |
4719 | return PTR_ERR(trans); | |
1832a6d5 | 4720 | |
618e21d5 JB |
4721 | btrfs_set_trans_block_group(trans, dir); |
4722 | ||
aec7477b | 4723 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4724 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 | 4725 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4726 | err = PTR_ERR(inode); |
4727 | if (IS_ERR(inode)) | |
4728 | goto out_unlock; | |
4729 | ||
f34f57a3 | 4730 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4731 | if (err) { |
4732 | drop_inode = 1; | |
4733 | goto out_unlock; | |
4734 | } | |
4735 | ||
618e21d5 | 4736 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4737 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4738 | if (err) |
4739 | drop_inode = 1; | |
4740 | else { | |
4741 | inode->i_op = &btrfs_special_inode_operations; | |
4742 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4743 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4744 | } |
618e21d5 JB |
4745 | btrfs_update_inode_block_group(trans, inode); |
4746 | btrfs_update_inode_block_group(trans, dir); | |
4747 | out_unlock: | |
d3c2fdcf | 4748 | nr = trans->blocks_used; |
89ce8a63 | 4749 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4750 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4751 | if (drop_inode) { |
4752 | inode_dec_link_count(inode); | |
4753 | iput(inode); | |
4754 | } | |
618e21d5 JB |
4755 | return err; |
4756 | } | |
4757 | ||
39279cc3 CM |
4758 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4759 | int mode, struct nameidata *nd) | |
4760 | { | |
4761 | struct btrfs_trans_handle *trans; | |
4762 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4763 | struct inode *inode = NULL; |
39279cc3 | 4764 | int drop_inode = 0; |
a22285a6 | 4765 | int err; |
1832a6d5 | 4766 | unsigned long nr = 0; |
39279cc3 | 4767 | u64 objectid; |
00e4e6b3 | 4768 | u64 index = 0; |
39279cc3 | 4769 | |
a22285a6 YZ |
4770 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4771 | if (err) | |
4772 | return err; | |
9ed74f2d JB |
4773 | /* |
4774 | * 2 for inode item and ref | |
4775 | * 2 for dir items | |
4776 | * 1 for xattr if selinux is on | |
4777 | */ | |
a22285a6 YZ |
4778 | trans = btrfs_start_transaction(root, 5); |
4779 | if (IS_ERR(trans)) | |
4780 | return PTR_ERR(trans); | |
9ed74f2d | 4781 | |
39279cc3 CM |
4782 | btrfs_set_trans_block_group(trans, dir); |
4783 | ||
aec7477b | 4784 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 JB |
4785 | dentry->d_name.len, dir->i_ino, objectid, |
4786 | BTRFS_I(dir)->block_group, mode, &index); | |
39279cc3 CM |
4787 | err = PTR_ERR(inode); |
4788 | if (IS_ERR(inode)) | |
4789 | goto out_unlock; | |
4790 | ||
f34f57a3 | 4791 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4792 | if (err) { |
4793 | drop_inode = 1; | |
4794 | goto out_unlock; | |
4795 | } | |
4796 | ||
39279cc3 | 4797 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4798 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4799 | if (err) |
4800 | drop_inode = 1; | |
4801 | else { | |
4802 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4803 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4804 | inode->i_fop = &btrfs_file_operations; |
4805 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4806 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4807 | } |
39279cc3 CM |
4808 | btrfs_update_inode_block_group(trans, inode); |
4809 | btrfs_update_inode_block_group(trans, dir); | |
4810 | out_unlock: | |
d3c2fdcf | 4811 | nr = trans->blocks_used; |
ab78c84d | 4812 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4813 | if (drop_inode) { |
4814 | inode_dec_link_count(inode); | |
4815 | iput(inode); | |
4816 | } | |
d3c2fdcf | 4817 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4818 | return err; |
4819 | } | |
4820 | ||
4821 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4822 | struct dentry *dentry) | |
4823 | { | |
4824 | struct btrfs_trans_handle *trans; | |
4825 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4826 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4827 | u64 index; |
1832a6d5 | 4828 | unsigned long nr = 0; |
39279cc3 CM |
4829 | int err; |
4830 | int drop_inode = 0; | |
4831 | ||
4832 | if (inode->i_nlink == 0) | |
4833 | return -ENOENT; | |
4834 | ||
4a8be425 TH |
4835 | /* do not allow sys_link's with other subvols of the same device */ |
4836 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4837 | return -EXDEV; |
4a8be425 | 4838 | |
c055e99e AV |
4839 | if (inode->i_nlink == ~0U) |
4840 | return -EMLINK; | |
4841 | ||
9ed74f2d | 4842 | btrfs_inc_nlink(inode); |
bc1cbf1f | 4843 | inode->i_ctime = CURRENT_TIME; |
9ed74f2d | 4844 | |
3de4586c | 4845 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4846 | if (err) |
4847 | goto fail; | |
4848 | ||
a22285a6 | 4849 | /* |
7e6b6465 | 4850 | * 2 items for inode and inode ref |
a22285a6 | 4851 | * 2 items for dir items |
7e6b6465 | 4852 | * 1 item for parent inode |
a22285a6 | 4853 | */ |
7e6b6465 | 4854 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4855 | if (IS_ERR(trans)) { |
4856 | err = PTR_ERR(trans); | |
4857 | goto fail; | |
4858 | } | |
5f39d397 | 4859 | |
39279cc3 | 4860 | btrfs_set_trans_block_group(trans, dir); |
7de9c6ee | 4861 | ihold(inode); |
aec7477b | 4862 | |
a1b075d2 | 4863 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4864 | |
a5719521 | 4865 | if (err) { |
54aa1f4d | 4866 | drop_inode = 1; |
a5719521 | 4867 | } else { |
6a912213 | 4868 | struct dentry *parent = dget_parent(dentry); |
a5719521 YZ |
4869 | btrfs_update_inode_block_group(trans, dir); |
4870 | err = btrfs_update_inode(trans, root, inode); | |
4871 | BUG_ON(err); | |
6a912213 JB |
4872 | btrfs_log_new_name(trans, inode, NULL, parent); |
4873 | dput(parent); | |
a5719521 | 4874 | } |
39279cc3 | 4875 | |
d3c2fdcf | 4876 | nr = trans->blocks_used; |
ab78c84d | 4877 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4878 | fail: |
39279cc3 CM |
4879 | if (drop_inode) { |
4880 | inode_dec_link_count(inode); | |
4881 | iput(inode); | |
4882 | } | |
d3c2fdcf | 4883 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4884 | return err; |
4885 | } | |
4886 | ||
39279cc3 CM |
4887 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4888 | { | |
b9d86667 | 4889 | struct inode *inode = NULL; |
39279cc3 CM |
4890 | struct btrfs_trans_handle *trans; |
4891 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4892 | int err = 0; | |
4893 | int drop_on_err = 0; | |
b9d86667 | 4894 | u64 objectid = 0; |
00e4e6b3 | 4895 | u64 index = 0; |
d3c2fdcf | 4896 | unsigned long nr = 1; |
39279cc3 | 4897 | |
a22285a6 YZ |
4898 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4899 | if (err) | |
4900 | return err; | |
4901 | ||
9ed74f2d JB |
4902 | /* |
4903 | * 2 items for inode and ref | |
4904 | * 2 items for dir items | |
4905 | * 1 for xattr if selinux is on | |
4906 | */ | |
a22285a6 YZ |
4907 | trans = btrfs_start_transaction(root, 5); |
4908 | if (IS_ERR(trans)) | |
4909 | return PTR_ERR(trans); | |
9ed74f2d | 4910 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 4911 | |
aec7477b | 4912 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4913 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
4914 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4915 | &index); | |
39279cc3 CM |
4916 | if (IS_ERR(inode)) { |
4917 | err = PTR_ERR(inode); | |
4918 | goto out_fail; | |
4919 | } | |
5f39d397 | 4920 | |
39279cc3 | 4921 | drop_on_err = 1; |
33268eaf | 4922 | |
f34f57a3 | 4923 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4924 | if (err) |
4925 | goto out_fail; | |
4926 | ||
39279cc3 CM |
4927 | inode->i_op = &btrfs_dir_inode_operations; |
4928 | inode->i_fop = &btrfs_dir_file_operations; | |
4929 | btrfs_set_trans_block_group(trans, inode); | |
4930 | ||
dbe674a9 | 4931 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4932 | err = btrfs_update_inode(trans, root, inode); |
4933 | if (err) | |
4934 | goto out_fail; | |
5f39d397 | 4935 | |
a1b075d2 JB |
4936 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4937 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4938 | if (err) |
4939 | goto out_fail; | |
5f39d397 | 4940 | |
39279cc3 CM |
4941 | d_instantiate(dentry, inode); |
4942 | drop_on_err = 0; | |
39279cc3 CM |
4943 | btrfs_update_inode_block_group(trans, inode); |
4944 | btrfs_update_inode_block_group(trans, dir); | |
4945 | ||
4946 | out_fail: | |
d3c2fdcf | 4947 | nr = trans->blocks_used; |
ab78c84d | 4948 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4949 | if (drop_on_err) |
4950 | iput(inode); | |
d3c2fdcf | 4951 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4952 | return err; |
4953 | } | |
4954 | ||
d352ac68 CM |
4955 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4956 | * and an extent that you want to insert, deal with overlap and insert | |
4957 | * the new extent into the tree. | |
4958 | */ | |
3b951516 CM |
4959 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4960 | struct extent_map *existing, | |
e6dcd2dc CM |
4961 | struct extent_map *em, |
4962 | u64 map_start, u64 map_len) | |
3b951516 CM |
4963 | { |
4964 | u64 start_diff; | |
3b951516 | 4965 | |
e6dcd2dc CM |
4966 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4967 | start_diff = map_start - em->start; | |
4968 | em->start = map_start; | |
4969 | em->len = map_len; | |
c8b97818 CM |
4970 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4971 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4972 | em->block_start += start_diff; |
c8b97818 CM |
4973 | em->block_len -= start_diff; |
4974 | } | |
e6dcd2dc | 4975 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4976 | } |
4977 | ||
c8b97818 CM |
4978 | static noinline int uncompress_inline(struct btrfs_path *path, |
4979 | struct inode *inode, struct page *page, | |
4980 | size_t pg_offset, u64 extent_offset, | |
4981 | struct btrfs_file_extent_item *item) | |
4982 | { | |
4983 | int ret; | |
4984 | struct extent_buffer *leaf = path->nodes[0]; | |
4985 | char *tmp; | |
4986 | size_t max_size; | |
4987 | unsigned long inline_size; | |
4988 | unsigned long ptr; | |
261507a0 | 4989 | int compress_type; |
c8b97818 CM |
4990 | |
4991 | WARN_ON(pg_offset != 0); | |
261507a0 | 4992 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
4993 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
4994 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4995 | btrfs_item_nr(leaf, path->slots[0])); | |
4996 | tmp = kmalloc(inline_size, GFP_NOFS); | |
4997 | ptr = btrfs_file_extent_inline_start(item); | |
4998 | ||
4999 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5000 | ||
5b050f04 | 5001 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5002 | ret = btrfs_decompress(compress_type, tmp, page, |
5003 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
5004 | if (ret) { |
5005 | char *kaddr = kmap_atomic(page, KM_USER0); | |
5006 | unsigned long copy_size = min_t(u64, | |
5007 | PAGE_CACHE_SIZE - pg_offset, | |
5008 | max_size - extent_offset); | |
5009 | memset(kaddr + pg_offset, 0, copy_size); | |
5010 | kunmap_atomic(kaddr, KM_USER0); | |
5011 | } | |
5012 | kfree(tmp); | |
5013 | return 0; | |
5014 | } | |
5015 | ||
d352ac68 CM |
5016 | /* |
5017 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5018 | * the ugly parts come from merging extents from the disk with the in-ram |
5019 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5020 | * where the in-ram extents might be locked pending data=ordered completion. |
5021 | * | |
5022 | * This also copies inline extents directly into the page. | |
5023 | */ | |
d397712b | 5024 | |
a52d9a80 | 5025 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5026 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5027 | int create) |
5028 | { | |
5029 | int ret; | |
5030 | int err = 0; | |
db94535d | 5031 | u64 bytenr; |
a52d9a80 CM |
5032 | u64 extent_start = 0; |
5033 | u64 extent_end = 0; | |
5034 | u64 objectid = inode->i_ino; | |
5035 | u32 found_type; | |
f421950f | 5036 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5037 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5038 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5039 | struct extent_buffer *leaf; |
5040 | struct btrfs_key found_key; | |
a52d9a80 CM |
5041 | struct extent_map *em = NULL; |
5042 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5043 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5044 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5045 | int compress_type; |
a52d9a80 | 5046 | |
a52d9a80 | 5047 | again: |
890871be | 5048 | read_lock(&em_tree->lock); |
d1310b2e | 5049 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5050 | if (em) |
5051 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5052 | read_unlock(&em_tree->lock); |
d1310b2e | 5053 | |
a52d9a80 | 5054 | if (em) { |
e1c4b745 CM |
5055 | if (em->start > start || em->start + em->len <= start) |
5056 | free_extent_map(em); | |
5057 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5058 | free_extent_map(em); |
5059 | else | |
5060 | goto out; | |
a52d9a80 | 5061 | } |
d1310b2e | 5062 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 5063 | if (!em) { |
d1310b2e CM |
5064 | err = -ENOMEM; |
5065 | goto out; | |
a52d9a80 | 5066 | } |
e6dcd2dc | 5067 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5068 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5069 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5070 | em->len = (u64)-1; |
c8b97818 | 5071 | em->block_len = (u64)-1; |
f421950f CM |
5072 | |
5073 | if (!path) { | |
5074 | path = btrfs_alloc_path(); | |
5075 | BUG_ON(!path); | |
5076 | } | |
5077 | ||
179e29e4 CM |
5078 | ret = btrfs_lookup_file_extent(trans, root, path, |
5079 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5080 | if (ret < 0) { |
5081 | err = ret; | |
5082 | goto out; | |
5083 | } | |
5084 | ||
5085 | if (ret != 0) { | |
5086 | if (path->slots[0] == 0) | |
5087 | goto not_found; | |
5088 | path->slots[0]--; | |
5089 | } | |
5090 | ||
5f39d397 CM |
5091 | leaf = path->nodes[0]; |
5092 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5093 | struct btrfs_file_extent_item); |
a52d9a80 | 5094 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5095 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5096 | found_type = btrfs_key_type(&found_key); | |
5097 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5098 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5099 | goto not_found; | |
5100 | } | |
5101 | ||
5f39d397 CM |
5102 | found_type = btrfs_file_extent_type(leaf, item); |
5103 | extent_start = found_key.offset; | |
261507a0 | 5104 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5105 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5106 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5107 | extent_end = extent_start + |
db94535d | 5108 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5109 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5110 | size_t size; | |
5111 | size = btrfs_file_extent_inline_len(leaf, item); | |
5112 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5113 | ~((u64)root->sectorsize - 1); | |
5114 | } | |
5115 | ||
5116 | if (start >= extent_end) { | |
5117 | path->slots[0]++; | |
5118 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5119 | ret = btrfs_next_leaf(root, path); | |
5120 | if (ret < 0) { | |
5121 | err = ret; | |
5122 | goto out; | |
a52d9a80 | 5123 | } |
9036c102 YZ |
5124 | if (ret > 0) |
5125 | goto not_found; | |
5126 | leaf = path->nodes[0]; | |
a52d9a80 | 5127 | } |
9036c102 YZ |
5128 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5129 | if (found_key.objectid != objectid || | |
5130 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5131 | goto not_found; | |
5132 | if (start + len <= found_key.offset) | |
5133 | goto not_found; | |
5134 | em->start = start; | |
5135 | em->len = found_key.offset - start; | |
5136 | goto not_found_em; | |
5137 | } | |
5138 | ||
d899e052 YZ |
5139 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5140 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5141 | em->start = extent_start; |
5142 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5143 | em->orig_start = extent_start - |
5144 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5145 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5146 | if (bytenr == 0) { | |
5f39d397 | 5147 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5148 | goto insert; |
5149 | } | |
261507a0 | 5150 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5151 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5152 | em->compress_type = compress_type; |
c8b97818 CM |
5153 | em->block_start = bytenr; |
5154 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5155 | item); | |
5156 | } else { | |
5157 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5158 | em->block_start = bytenr; | |
5159 | em->block_len = em->len; | |
d899e052 YZ |
5160 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5161 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5162 | } |
a52d9a80 CM |
5163 | goto insert; |
5164 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5165 | unsigned long ptr; |
a52d9a80 | 5166 | char *map; |
3326d1b0 CM |
5167 | size_t size; |
5168 | size_t extent_offset; | |
5169 | size_t copy_size; | |
a52d9a80 | 5170 | |
689f9346 | 5171 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5172 | if (!page || create) { |
689f9346 | 5173 | em->start = extent_start; |
9036c102 | 5174 | em->len = extent_end - extent_start; |
689f9346 Y |
5175 | goto out; |
5176 | } | |
5f39d397 | 5177 | |
9036c102 YZ |
5178 | size = btrfs_file_extent_inline_len(leaf, item); |
5179 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5180 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5181 | size - extent_offset); |
3326d1b0 | 5182 | em->start = extent_start + extent_offset; |
70dec807 CM |
5183 | em->len = (copy_size + root->sectorsize - 1) & |
5184 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5185 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5186 | if (compress_type) { |
c8b97818 | 5187 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5188 | em->compress_type = compress_type; |
5189 | } | |
689f9346 | 5190 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5191 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5192 | if (btrfs_file_extent_compression(leaf, item) != |
5193 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5194 | ret = uncompress_inline(path, inode, page, |
5195 | pg_offset, | |
5196 | extent_offset, item); | |
5197 | BUG_ON(ret); | |
5198 | } else { | |
5199 | map = kmap(page); | |
5200 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5201 | copy_size); | |
93c82d57 CM |
5202 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5203 | memset(map + pg_offset + copy_size, 0, | |
5204 | PAGE_CACHE_SIZE - pg_offset - | |
5205 | copy_size); | |
5206 | } | |
c8b97818 CM |
5207 | kunmap(page); |
5208 | } | |
179e29e4 CM |
5209 | flush_dcache_page(page); |
5210 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 5211 | WARN_ON(1); |
179e29e4 CM |
5212 | if (!trans) { |
5213 | kunmap(page); | |
5214 | free_extent_map(em); | |
5215 | em = NULL; | |
5216 | btrfs_release_path(root, path); | |
f9295749 | 5217 | trans = btrfs_join_transaction(root, 1); |
3612b495 TI |
5218 | if (IS_ERR(trans)) |
5219 | return ERR_CAST(trans); | |
179e29e4 CM |
5220 | goto again; |
5221 | } | |
c8b97818 | 5222 | map = kmap(page); |
70dec807 | 5223 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5224 | copy_size); |
c8b97818 | 5225 | kunmap(page); |
179e29e4 | 5226 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5227 | } |
d1310b2e CM |
5228 | set_extent_uptodate(io_tree, em->start, |
5229 | extent_map_end(em) - 1, GFP_NOFS); | |
a52d9a80 CM |
5230 | goto insert; |
5231 | } else { | |
d397712b | 5232 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5233 | WARN_ON(1); |
5234 | } | |
5235 | not_found: | |
5236 | em->start = start; | |
d1310b2e | 5237 | em->len = len; |
a52d9a80 | 5238 | not_found_em: |
5f39d397 | 5239 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5240 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
5241 | insert: |
5242 | btrfs_release_path(root, path); | |
d1310b2e | 5243 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5244 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5245 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5246 | (unsigned long long)em->len, | |
5247 | (unsigned long long)start, | |
5248 | (unsigned long long)len); | |
a52d9a80 CM |
5249 | err = -EIO; |
5250 | goto out; | |
5251 | } | |
d1310b2e CM |
5252 | |
5253 | err = 0; | |
890871be | 5254 | write_lock(&em_tree->lock); |
a52d9a80 | 5255 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5256 | /* it is possible that someone inserted the extent into the tree |
5257 | * while we had the lock dropped. It is also possible that | |
5258 | * an overlapping map exists in the tree | |
5259 | */ | |
a52d9a80 | 5260 | if (ret == -EEXIST) { |
3b951516 | 5261 | struct extent_map *existing; |
e6dcd2dc CM |
5262 | |
5263 | ret = 0; | |
5264 | ||
3b951516 | 5265 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5266 | if (existing && (existing->start > start || |
5267 | existing->start + existing->len <= start)) { | |
5268 | free_extent_map(existing); | |
5269 | existing = NULL; | |
5270 | } | |
3b951516 CM |
5271 | if (!existing) { |
5272 | existing = lookup_extent_mapping(em_tree, em->start, | |
5273 | em->len); | |
5274 | if (existing) { | |
5275 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5276 | em, start, |
5277 | root->sectorsize); | |
3b951516 CM |
5278 | free_extent_map(existing); |
5279 | if (err) { | |
5280 | free_extent_map(em); | |
5281 | em = NULL; | |
5282 | } | |
5283 | } else { | |
5284 | err = -EIO; | |
3b951516 CM |
5285 | free_extent_map(em); |
5286 | em = NULL; | |
5287 | } | |
5288 | } else { | |
5289 | free_extent_map(em); | |
5290 | em = existing; | |
e6dcd2dc | 5291 | err = 0; |
a52d9a80 | 5292 | } |
a52d9a80 | 5293 | } |
890871be | 5294 | write_unlock(&em_tree->lock); |
a52d9a80 | 5295 | out: |
1abe9b8a | 5296 | |
5297 | trace_btrfs_get_extent(root, em); | |
5298 | ||
f421950f CM |
5299 | if (path) |
5300 | btrfs_free_path(path); | |
a52d9a80 CM |
5301 | if (trans) { |
5302 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5303 | if (!err) |
a52d9a80 CM |
5304 | err = ret; |
5305 | } | |
a52d9a80 CM |
5306 | if (err) { |
5307 | free_extent_map(em); | |
a52d9a80 CM |
5308 | return ERR_PTR(err); |
5309 | } | |
5310 | return em; | |
5311 | } | |
5312 | ||
ec29ed5b CM |
5313 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5314 | size_t pg_offset, u64 start, u64 len, | |
5315 | int create) | |
5316 | { | |
5317 | struct extent_map *em; | |
5318 | struct extent_map *hole_em = NULL; | |
5319 | u64 range_start = start; | |
5320 | u64 end; | |
5321 | u64 found; | |
5322 | u64 found_end; | |
5323 | int err = 0; | |
5324 | ||
5325 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5326 | if (IS_ERR(em)) | |
5327 | return em; | |
5328 | if (em) { | |
5329 | /* | |
5330 | * if our em maps to a hole, there might | |
5331 | * actually be delalloc bytes behind it | |
5332 | */ | |
5333 | if (em->block_start != EXTENT_MAP_HOLE) | |
5334 | return em; | |
5335 | else | |
5336 | hole_em = em; | |
5337 | } | |
5338 | ||
5339 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5340 | end = start + len; | |
5341 | if (end < start) | |
5342 | end = (u64)-1; | |
5343 | else | |
5344 | end -= 1; | |
5345 | ||
5346 | em = NULL; | |
5347 | ||
5348 | /* ok, we didn't find anything, lets look for delalloc */ | |
5349 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5350 | end, len, EXTENT_DELALLOC, 1); | |
5351 | found_end = range_start + found; | |
5352 | if (found_end < range_start) | |
5353 | found_end = (u64)-1; | |
5354 | ||
5355 | /* | |
5356 | * we didn't find anything useful, return | |
5357 | * the original results from get_extent() | |
5358 | */ | |
5359 | if (range_start > end || found_end <= start) { | |
5360 | em = hole_em; | |
5361 | hole_em = NULL; | |
5362 | goto out; | |
5363 | } | |
5364 | ||
5365 | /* adjust the range_start to make sure it doesn't | |
5366 | * go backwards from the start they passed in | |
5367 | */ | |
5368 | range_start = max(start,range_start); | |
5369 | found = found_end - range_start; | |
5370 | ||
5371 | if (found > 0) { | |
5372 | u64 hole_start = start; | |
5373 | u64 hole_len = len; | |
5374 | ||
5375 | em = alloc_extent_map(GFP_NOFS); | |
5376 | if (!em) { | |
5377 | err = -ENOMEM; | |
5378 | goto out; | |
5379 | } | |
5380 | /* | |
5381 | * when btrfs_get_extent can't find anything it | |
5382 | * returns one huge hole | |
5383 | * | |
5384 | * make sure what it found really fits our range, and | |
5385 | * adjust to make sure it is based on the start from | |
5386 | * the caller | |
5387 | */ | |
5388 | if (hole_em) { | |
5389 | u64 calc_end = extent_map_end(hole_em); | |
5390 | ||
5391 | if (calc_end <= start || (hole_em->start > end)) { | |
5392 | free_extent_map(hole_em); | |
5393 | hole_em = NULL; | |
5394 | } else { | |
5395 | hole_start = max(hole_em->start, start); | |
5396 | hole_len = calc_end - hole_start; | |
5397 | } | |
5398 | } | |
5399 | em->bdev = NULL; | |
5400 | if (hole_em && range_start > hole_start) { | |
5401 | /* our hole starts before our delalloc, so we | |
5402 | * have to return just the parts of the hole | |
5403 | * that go until the delalloc starts | |
5404 | */ | |
5405 | em->len = min(hole_len, | |
5406 | range_start - hole_start); | |
5407 | em->start = hole_start; | |
5408 | em->orig_start = hole_start; | |
5409 | /* | |
5410 | * don't adjust block start at all, | |
5411 | * it is fixed at EXTENT_MAP_HOLE | |
5412 | */ | |
5413 | em->block_start = hole_em->block_start; | |
5414 | em->block_len = hole_len; | |
5415 | } else { | |
5416 | em->start = range_start; | |
5417 | em->len = found; | |
5418 | em->orig_start = range_start; | |
5419 | em->block_start = EXTENT_MAP_DELALLOC; | |
5420 | em->block_len = found; | |
5421 | } | |
5422 | } else if (hole_em) { | |
5423 | return hole_em; | |
5424 | } | |
5425 | out: | |
5426 | ||
5427 | free_extent_map(hole_em); | |
5428 | if (err) { | |
5429 | free_extent_map(em); | |
5430 | return ERR_PTR(err); | |
5431 | } | |
5432 | return em; | |
5433 | } | |
5434 | ||
4b46fce2 JB |
5435 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
5436 | u64 start, u64 len) | |
5437 | { | |
5438 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5439 | struct btrfs_trans_handle *trans; | |
5440 | struct extent_map *em; | |
5441 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
5442 | struct btrfs_key ins; | |
5443 | u64 alloc_hint; | |
5444 | int ret; | |
5445 | ||
5446 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5447 | ||
5448 | trans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
5449 | if (IS_ERR(trans)) |
5450 | return ERR_CAST(trans); | |
4b46fce2 JB |
5451 | |
5452 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5453 | ||
5454 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5455 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5456 | alloc_hint, (u64)-1, &ins, 1); | |
5457 | if (ret) { | |
5458 | em = ERR_PTR(ret); | |
5459 | goto out; | |
5460 | } | |
5461 | ||
5462 | em = alloc_extent_map(GFP_NOFS); | |
5463 | if (!em) { | |
5464 | em = ERR_PTR(-ENOMEM); | |
5465 | goto out; | |
5466 | } | |
5467 | ||
5468 | em->start = start; | |
5469 | em->orig_start = em->start; | |
5470 | em->len = ins.offset; | |
5471 | ||
5472 | em->block_start = ins.objectid; | |
5473 | em->block_len = ins.offset; | |
5474 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5475 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5476 | ||
5477 | while (1) { | |
5478 | write_lock(&em_tree->lock); | |
5479 | ret = add_extent_mapping(em_tree, em); | |
5480 | write_unlock(&em_tree->lock); | |
5481 | if (ret != -EEXIST) | |
5482 | break; | |
5483 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5484 | } | |
5485 | ||
5486 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5487 | ins.offset, ins.offset, 0); | |
5488 | if (ret) { | |
5489 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5490 | em = ERR_PTR(ret); | |
5491 | } | |
5492 | out: | |
5493 | btrfs_end_transaction(trans, root); | |
5494 | return em; | |
5495 | } | |
5496 | ||
46bfbb5c CM |
5497 | /* |
5498 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5499 | * block must be cow'd | |
5500 | */ | |
5501 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5502 | struct inode *inode, u64 offset, u64 len) | |
5503 | { | |
5504 | struct btrfs_path *path; | |
5505 | int ret; | |
5506 | struct extent_buffer *leaf; | |
5507 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5508 | struct btrfs_file_extent_item *fi; | |
5509 | struct btrfs_key key; | |
5510 | u64 disk_bytenr; | |
5511 | u64 backref_offset; | |
5512 | u64 extent_end; | |
5513 | u64 num_bytes; | |
5514 | int slot; | |
5515 | int found_type; | |
5516 | ||
5517 | path = btrfs_alloc_path(); | |
5518 | if (!path) | |
5519 | return -ENOMEM; | |
5520 | ||
5521 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
5522 | offset, 0); | |
5523 | if (ret < 0) | |
5524 | goto out; | |
5525 | ||
5526 | slot = path->slots[0]; | |
5527 | if (ret == 1) { | |
5528 | if (slot == 0) { | |
5529 | /* can't find the item, must cow */ | |
5530 | ret = 0; | |
5531 | goto out; | |
5532 | } | |
5533 | slot--; | |
5534 | } | |
5535 | ret = 0; | |
5536 | leaf = path->nodes[0]; | |
5537 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5538 | if (key.objectid != inode->i_ino || | |
5539 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
5540 | /* not our file or wrong item type, must cow */ | |
5541 | goto out; | |
5542 | } | |
5543 | ||
5544 | if (key.offset > offset) { | |
5545 | /* Wrong offset, must cow */ | |
5546 | goto out; | |
5547 | } | |
5548 | ||
5549 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5550 | found_type = btrfs_file_extent_type(leaf, fi); | |
5551 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5552 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5553 | /* not a regular extent, must cow */ | |
5554 | goto out; | |
5555 | } | |
5556 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5557 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5558 | ||
5559 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5560 | if (extent_end < offset + len) { | |
5561 | /* extent doesn't include our full range, must cow */ | |
5562 | goto out; | |
5563 | } | |
5564 | ||
5565 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5566 | goto out; | |
5567 | ||
5568 | /* | |
5569 | * look for other files referencing this extent, if we | |
5570 | * find any we must cow | |
5571 | */ | |
5572 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, | |
5573 | key.offset - backref_offset, disk_bytenr)) | |
5574 | goto out; | |
5575 | ||
5576 | /* | |
5577 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5578 | * in this extent we are about to write. If there | |
5579 | * are any csums in that range we have to cow in order | |
5580 | * to keep the csums correct | |
5581 | */ | |
5582 | disk_bytenr += backref_offset; | |
5583 | disk_bytenr += offset - key.offset; | |
5584 | num_bytes = min(offset + len, extent_end) - offset; | |
5585 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5586 | goto out; | |
5587 | /* | |
5588 | * all of the above have passed, it is safe to overwrite this extent | |
5589 | * without cow | |
5590 | */ | |
5591 | ret = 1; | |
5592 | out: | |
5593 | btrfs_free_path(path); | |
5594 | return ret; | |
5595 | } | |
5596 | ||
4b46fce2 JB |
5597 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5598 | struct buffer_head *bh_result, int create) | |
5599 | { | |
5600 | struct extent_map *em; | |
5601 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5602 | u64 start = iblock << inode->i_blkbits; | |
5603 | u64 len = bh_result->b_size; | |
46bfbb5c | 5604 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5605 | |
5606 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5607 | if (IS_ERR(em)) | |
5608 | return PTR_ERR(em); | |
5609 | ||
5610 | /* | |
5611 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5612 | * io. INLINE is special, and we could probably kludge it in here, but | |
5613 | * it's still buffered so for safety lets just fall back to the generic | |
5614 | * buffered path. | |
5615 | * | |
5616 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5617 | * decompress it, so there will be buffering required no matter what we | |
5618 | * do, so go ahead and fallback to buffered. | |
5619 | * | |
5620 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5621 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5622 | * the generic code. | |
5623 | */ | |
5624 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5625 | em->block_start == EXTENT_MAP_INLINE) { | |
5626 | free_extent_map(em); | |
5627 | return -ENOTBLK; | |
5628 | } | |
5629 | ||
5630 | /* Just a good old fashioned hole, return */ | |
5631 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5632 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5633 | free_extent_map(em); | |
5634 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5635 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5636 | start + root->sectorsize - 1, GFP_NOFS); | |
5637 | return 0; | |
5638 | } | |
5639 | ||
5640 | /* | |
5641 | * We don't allocate a new extent in the following cases | |
5642 | * | |
5643 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5644 | * existing extent. | |
5645 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5646 | * just use the extent. | |
5647 | * | |
5648 | */ | |
46bfbb5c CM |
5649 | if (!create) { |
5650 | len = em->len - (start - em->start); | |
4b46fce2 | 5651 | goto map; |
46bfbb5c | 5652 | } |
4b46fce2 JB |
5653 | |
5654 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5655 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5656 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5657 | int type; |
5658 | int ret; | |
46bfbb5c | 5659 | u64 block_start; |
4b46fce2 JB |
5660 | |
5661 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5662 | type = BTRFS_ORDERED_PREALLOC; | |
5663 | else | |
5664 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5665 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5666 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5667 | |
5668 | /* | |
5669 | * we're not going to log anything, but we do need | |
5670 | * to make sure the current transaction stays open | |
5671 | * while we look for nocow cross refs | |
5672 | */ | |
5673 | trans = btrfs_join_transaction(root, 0); | |
3612b495 | 5674 | if (IS_ERR(trans)) |
46bfbb5c CM |
5675 | goto must_cow; |
5676 | ||
5677 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5678 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5679 | block_start, len, len, type); | |
5680 | btrfs_end_transaction(trans, root); | |
5681 | if (ret) { | |
5682 | free_extent_map(em); | |
5683 | return ret; | |
5684 | } | |
5685 | goto unlock; | |
4b46fce2 | 5686 | } |
46bfbb5c | 5687 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5688 | } |
46bfbb5c CM |
5689 | must_cow: |
5690 | /* | |
5691 | * this will cow the extent, reset the len in case we changed | |
5692 | * it above | |
5693 | */ | |
5694 | len = bh_result->b_size; | |
5695 | free_extent_map(em); | |
5696 | em = btrfs_new_extent_direct(inode, start, len); | |
5697 | if (IS_ERR(em)) | |
5698 | return PTR_ERR(em); | |
5699 | len = min(len, em->len - (start - em->start)); | |
5700 | unlock: | |
4845e44f CM |
5701 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5702 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5703 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5704 | map: |
5705 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5706 | inode->i_blkbits; | |
46bfbb5c | 5707 | bh_result->b_size = len; |
4b46fce2 JB |
5708 | bh_result->b_bdev = em->bdev; |
5709 | set_buffer_mapped(bh_result); | |
5710 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5711 | set_buffer_new(bh_result); | |
5712 | ||
5713 | free_extent_map(em); | |
5714 | ||
5715 | return 0; | |
5716 | } | |
5717 | ||
5718 | struct btrfs_dio_private { | |
5719 | struct inode *inode; | |
5720 | u64 logical_offset; | |
5721 | u64 disk_bytenr; | |
5722 | u64 bytes; | |
5723 | u32 *csums; | |
5724 | void *private; | |
e65e1535 MX |
5725 | |
5726 | /* number of bios pending for this dio */ | |
5727 | atomic_t pending_bios; | |
5728 | ||
5729 | /* IO errors */ | |
5730 | int errors; | |
5731 | ||
5732 | struct bio *orig_bio; | |
4b46fce2 JB |
5733 | }; |
5734 | ||
5735 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5736 | { | |
e65e1535 | 5737 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5738 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5739 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5740 | struct inode *inode = dip->inode; |
5741 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5742 | u64 start; | |
5743 | u32 *private = dip->csums; | |
5744 | ||
5745 | start = dip->logical_offset; | |
5746 | do { | |
5747 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5748 | struct page *page = bvec->bv_page; | |
5749 | char *kaddr; | |
5750 | u32 csum = ~(u32)0; | |
5751 | unsigned long flags; | |
5752 | ||
5753 | local_irq_save(flags); | |
5754 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5755 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5756 | csum, bvec->bv_len); | |
5757 | btrfs_csum_final(csum, (char *)&csum); | |
5758 | kunmap_atomic(kaddr, KM_IRQ0); | |
5759 | local_irq_restore(flags); | |
5760 | ||
5761 | flush_dcache_page(bvec->bv_page); | |
5762 | if (csum != *private) { | |
5763 | printk(KERN_ERR "btrfs csum failed ino %lu off" | |
5764 | " %llu csum %u private %u\n", | |
5765 | inode->i_ino, (unsigned long long)start, | |
5766 | csum, *private); | |
5767 | err = -EIO; | |
5768 | } | |
5769 | } | |
5770 | ||
5771 | start += bvec->bv_len; | |
5772 | private++; | |
5773 | bvec++; | |
5774 | } while (bvec <= bvec_end); | |
5775 | ||
5776 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5777 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5778 | bio->bi_private = dip->private; | |
5779 | ||
5780 | kfree(dip->csums); | |
5781 | kfree(dip); | |
c0da7aa1 JB |
5782 | |
5783 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5784 | if (err) | |
5785 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5786 | dio_end_io(bio, err); |
5787 | } | |
5788 | ||
5789 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5790 | { | |
5791 | struct btrfs_dio_private *dip = bio->bi_private; | |
5792 | struct inode *inode = dip->inode; | |
5793 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5794 | struct btrfs_trans_handle *trans; | |
5795 | struct btrfs_ordered_extent *ordered = NULL; | |
5796 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5797 | u64 ordered_offset = dip->logical_offset; |
5798 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5799 | int ret; |
5800 | ||
5801 | if (err) | |
5802 | goto out_done; | |
163cf09c CM |
5803 | again: |
5804 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5805 | &ordered_offset, | |
5806 | ordered_bytes); | |
4b46fce2 | 5807 | if (!ret) |
163cf09c | 5808 | goto out_test; |
4b46fce2 JB |
5809 | |
5810 | BUG_ON(!ordered); | |
5811 | ||
5812 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 5813 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5814 | err = -ENOMEM; |
5815 | goto out; | |
5816 | } | |
5817 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5818 | ||
5819 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5820 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5821 | if (!ret) | |
5822 | ret = btrfs_update_inode(trans, root, inode); | |
5823 | err = ret; | |
5824 | goto out; | |
5825 | } | |
5826 | ||
5827 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5828 | ordered->file_offset + ordered->len - 1, 0, | |
5829 | &cached_state, GFP_NOFS); | |
5830 | ||
5831 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5832 | ret = btrfs_mark_extent_written(trans, inode, | |
5833 | ordered->file_offset, | |
5834 | ordered->file_offset + | |
5835 | ordered->len); | |
5836 | if (ret) { | |
5837 | err = ret; | |
5838 | goto out_unlock; | |
5839 | } | |
5840 | } else { | |
5841 | ret = insert_reserved_file_extent(trans, inode, | |
5842 | ordered->file_offset, | |
5843 | ordered->start, | |
5844 | ordered->disk_len, | |
5845 | ordered->len, | |
5846 | ordered->len, | |
5847 | 0, 0, 0, | |
5848 | BTRFS_FILE_EXTENT_REG); | |
5849 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5850 | ordered->file_offset, ordered->len); | |
5851 | if (ret) { | |
5852 | err = ret; | |
5853 | WARN_ON(1); | |
5854 | goto out_unlock; | |
5855 | } | |
5856 | } | |
5857 | ||
5858 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
5859 | btrfs_ordered_update_i_size(inode, 0, ordered); | |
5860 | btrfs_update_inode(trans, root, inode); | |
5861 | out_unlock: | |
5862 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5863 | ordered->file_offset + ordered->len - 1, | |
5864 | &cached_state, GFP_NOFS); | |
5865 | out: | |
5866 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5867 | btrfs_end_transaction(trans, root); | |
163cf09c | 5868 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5869 | btrfs_put_ordered_extent(ordered); |
5870 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5871 | |
5872 | out_test: | |
5873 | /* | |
5874 | * our bio might span multiple ordered extents. If we haven't | |
5875 | * completed the accounting for the whole dio, go back and try again | |
5876 | */ | |
5877 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5878 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5879 | ordered_offset; | |
5880 | goto again; | |
5881 | } | |
4b46fce2 JB |
5882 | out_done: |
5883 | bio->bi_private = dip->private; | |
5884 | ||
5885 | kfree(dip->csums); | |
5886 | kfree(dip); | |
c0da7aa1 JB |
5887 | |
5888 | /* If we had an error make sure to clear the uptodate flag */ | |
5889 | if (err) | |
5890 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5891 | dio_end_io(bio, err); |
5892 | } | |
5893 | ||
eaf25d93 CM |
5894 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5895 | struct bio *bio, int mirror_num, | |
5896 | unsigned long bio_flags, u64 offset) | |
5897 | { | |
5898 | int ret; | |
5899 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5900 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5901 | BUG_ON(ret); | |
5902 | return 0; | |
5903 | } | |
5904 | ||
e65e1535 MX |
5905 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5906 | { | |
5907 | struct btrfs_dio_private *dip = bio->bi_private; | |
5908 | ||
5909 | if (err) { | |
5910 | printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu " | |
3dd1462e JB |
5911 | "sector %#Lx len %u err no %d\n", |
5912 | dip->inode->i_ino, bio->bi_rw, | |
5913 | (unsigned long long)bio->bi_sector, bio->bi_size, err); | |
e65e1535 MX |
5914 | dip->errors = 1; |
5915 | ||
5916 | /* | |
5917 | * before atomic variable goto zero, we must make sure | |
5918 | * dip->errors is perceived to be set. | |
5919 | */ | |
5920 | smp_mb__before_atomic_dec(); | |
5921 | } | |
5922 | ||
5923 | /* if there are more bios still pending for this dio, just exit */ | |
5924 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5925 | goto out; | |
5926 | ||
5927 | if (dip->errors) | |
5928 | bio_io_error(dip->orig_bio); | |
5929 | else { | |
5930 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5931 | bio_endio(dip->orig_bio, 0); | |
5932 | } | |
5933 | out: | |
5934 | bio_put(bio); | |
5935 | } | |
5936 | ||
5937 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5938 | u64 first_sector, gfp_t gfp_flags) | |
5939 | { | |
5940 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5941 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5942 | } | |
5943 | ||
5944 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5945 | int rw, u64 file_offset, int skip_sum, | |
5946 | u32 *csums) | |
5947 | { | |
5948 | int write = rw & REQ_WRITE; | |
5949 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5950 | int ret; | |
5951 | ||
5952 | bio_get(bio); | |
5953 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5954 | if (ret) | |
5955 | goto err; | |
5956 | ||
5957 | if (write && !skip_sum) { | |
5958 | ret = btrfs_wq_submit_bio(root->fs_info, | |
5959 | inode, rw, bio, 0, 0, | |
5960 | file_offset, | |
5961 | __btrfs_submit_bio_start_direct_io, | |
5962 | __btrfs_submit_bio_done); | |
5963 | goto err; | |
c2db1073 TI |
5964 | } else if (!skip_sum) { |
5965 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 5966 | file_offset, csums); |
c2db1073 TI |
5967 | if (ret) |
5968 | goto err; | |
5969 | } | |
e65e1535 MX |
5970 | |
5971 | ret = btrfs_map_bio(root, rw, bio, 0, 1); | |
5972 | err: | |
5973 | bio_put(bio); | |
5974 | return ret; | |
5975 | } | |
5976 | ||
5977 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
5978 | int skip_sum) | |
5979 | { | |
5980 | struct inode *inode = dip->inode; | |
5981 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5982 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5983 | struct bio *bio; | |
5984 | struct bio *orig_bio = dip->orig_bio; | |
5985 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
5986 | u64 start_sector = orig_bio->bi_sector; | |
5987 | u64 file_offset = dip->logical_offset; | |
5988 | u64 submit_len = 0; | |
5989 | u64 map_length; | |
5990 | int nr_pages = 0; | |
5991 | u32 *csums = dip->csums; | |
5992 | int ret = 0; | |
98bc3149 | 5993 | int write = rw & REQ_WRITE; |
e65e1535 MX |
5994 | |
5995 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); | |
5996 | if (!bio) | |
5997 | return -ENOMEM; | |
5998 | bio->bi_private = dip; | |
5999 | bio->bi_end_io = btrfs_end_dio_bio; | |
6000 | atomic_inc(&dip->pending_bios); | |
6001 | ||
6002 | map_length = orig_bio->bi_size; | |
6003 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6004 | &map_length, NULL, 0); | |
6005 | if (ret) { | |
6006 | bio_put(bio); | |
6007 | return -EIO; | |
6008 | } | |
6009 | ||
6010 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { | |
6011 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6012 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6013 | bvec->bv_offset) < bvec->bv_len)) { | |
6014 | /* | |
6015 | * inc the count before we submit the bio so | |
6016 | * we know the end IO handler won't happen before | |
6017 | * we inc the count. Otherwise, the dip might get freed | |
6018 | * before we're done setting it up | |
6019 | */ | |
6020 | atomic_inc(&dip->pending_bios); | |
6021 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6022 | file_offset, skip_sum, | |
6023 | csums); | |
6024 | if (ret) { | |
6025 | bio_put(bio); | |
6026 | atomic_dec(&dip->pending_bios); | |
6027 | goto out_err; | |
6028 | } | |
6029 | ||
98bc3149 JB |
6030 | /* Write's use the ordered csums */ |
6031 | if (!write && !skip_sum) | |
e65e1535 MX |
6032 | csums = csums + nr_pages; |
6033 | start_sector += submit_len >> 9; | |
6034 | file_offset += submit_len; | |
6035 | ||
6036 | submit_len = 0; | |
6037 | nr_pages = 0; | |
6038 | ||
6039 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6040 | start_sector, GFP_NOFS); | |
6041 | if (!bio) | |
6042 | goto out_err; | |
6043 | bio->bi_private = dip; | |
6044 | bio->bi_end_io = btrfs_end_dio_bio; | |
6045 | ||
6046 | map_length = orig_bio->bi_size; | |
6047 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6048 | &map_length, NULL, 0); | |
6049 | if (ret) { | |
6050 | bio_put(bio); | |
6051 | goto out_err; | |
6052 | } | |
6053 | } else { | |
6054 | submit_len += bvec->bv_len; | |
6055 | nr_pages ++; | |
6056 | bvec++; | |
6057 | } | |
6058 | } | |
6059 | ||
6060 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, | |
6061 | csums); | |
6062 | if (!ret) | |
6063 | return 0; | |
6064 | ||
6065 | bio_put(bio); | |
6066 | out_err: | |
6067 | dip->errors = 1; | |
6068 | /* | |
6069 | * before atomic variable goto zero, we must | |
6070 | * make sure dip->errors is perceived to be set. | |
6071 | */ | |
6072 | smp_mb__before_atomic_dec(); | |
6073 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6074 | bio_io_error(dip->orig_bio); | |
6075 | ||
6076 | /* bio_end_io() will handle error, so we needn't return it */ | |
6077 | return 0; | |
6078 | } | |
6079 | ||
4b46fce2 JB |
6080 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6081 | loff_t file_offset) | |
6082 | { | |
6083 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6084 | struct btrfs_dio_private *dip; | |
6085 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6086 | int skip_sum; |
7b6d91da | 6087 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6088 | int ret = 0; |
6089 | ||
6090 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6091 | ||
6092 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6093 | if (!dip) { | |
6094 | ret = -ENOMEM; | |
6095 | goto free_ordered; | |
6096 | } | |
6097 | dip->csums = NULL; | |
6098 | ||
98bc3149 JB |
6099 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6100 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6101 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6102 | if (!dip->csums) { | |
b4966b77 | 6103 | kfree(dip); |
4b46fce2 JB |
6104 | ret = -ENOMEM; |
6105 | goto free_ordered; | |
6106 | } | |
6107 | } | |
6108 | ||
6109 | dip->private = bio->bi_private; | |
6110 | dip->inode = inode; | |
6111 | dip->logical_offset = file_offset; | |
6112 | ||
4b46fce2 JB |
6113 | dip->bytes = 0; |
6114 | do { | |
6115 | dip->bytes += bvec->bv_len; | |
6116 | bvec++; | |
6117 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6118 | ||
46bfbb5c | 6119 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6120 | bio->bi_private = dip; |
e65e1535 MX |
6121 | dip->errors = 0; |
6122 | dip->orig_bio = bio; | |
6123 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6124 | |
6125 | if (write) | |
6126 | bio->bi_end_io = btrfs_endio_direct_write; | |
6127 | else | |
6128 | bio->bi_end_io = btrfs_endio_direct_read; | |
6129 | ||
e65e1535 MX |
6130 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6131 | if (!ret) | |
eaf25d93 | 6132 | return; |
4b46fce2 JB |
6133 | free_ordered: |
6134 | /* | |
6135 | * If this is a write, we need to clean up the reserved space and kill | |
6136 | * the ordered extent. | |
6137 | */ | |
6138 | if (write) { | |
6139 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6140 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6141 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6142 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6143 | btrfs_free_reserved_extent(root, ordered->start, | |
6144 | ordered->disk_len); | |
6145 | btrfs_put_ordered_extent(ordered); | |
6146 | btrfs_put_ordered_extent(ordered); | |
6147 | } | |
6148 | bio_endio(bio, ret); | |
6149 | } | |
6150 | ||
5a5f79b5 CM |
6151 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6152 | const struct iovec *iov, loff_t offset, | |
6153 | unsigned long nr_segs) | |
6154 | { | |
6155 | int seg; | |
a1b75f7d | 6156 | int i; |
5a5f79b5 CM |
6157 | size_t size; |
6158 | unsigned long addr; | |
6159 | unsigned blocksize_mask = root->sectorsize - 1; | |
6160 | ssize_t retval = -EINVAL; | |
6161 | loff_t end = offset; | |
6162 | ||
6163 | if (offset & blocksize_mask) | |
6164 | goto out; | |
6165 | ||
6166 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6167 | for (seg = 0; seg < nr_segs; seg++) { | |
6168 | addr = (unsigned long)iov[seg].iov_base; | |
6169 | size = iov[seg].iov_len; | |
6170 | end += size; | |
a1b75f7d | 6171 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6172 | goto out; |
a1b75f7d JB |
6173 | |
6174 | /* If this is a write we don't need to check anymore */ | |
6175 | if (rw & WRITE) | |
6176 | continue; | |
6177 | ||
6178 | /* | |
6179 | * Check to make sure we don't have duplicate iov_base's in this | |
6180 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6181 | * when reading back. | |
6182 | */ | |
6183 | for (i = seg + 1; i < nr_segs; i++) { | |
6184 | if (iov[seg].iov_base == iov[i].iov_base) | |
6185 | goto out; | |
6186 | } | |
5a5f79b5 CM |
6187 | } |
6188 | retval = 0; | |
6189 | out: | |
6190 | return retval; | |
6191 | } | |
16432985 CM |
6192 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6193 | const struct iovec *iov, loff_t offset, | |
6194 | unsigned long nr_segs) | |
6195 | { | |
4b46fce2 JB |
6196 | struct file *file = iocb->ki_filp; |
6197 | struct inode *inode = file->f_mapping->host; | |
6198 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6199 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6200 | u64 lockstart, lockend; |
6201 | ssize_t ret; | |
4845e44f CM |
6202 | int writing = rw & WRITE; |
6203 | int write_bits = 0; | |
3f7c579c | 6204 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6205 | |
5a5f79b5 CM |
6206 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6207 | offset, nr_segs)) { | |
6208 | return 0; | |
6209 | } | |
6210 | ||
4b46fce2 | 6211 | lockstart = offset; |
3f7c579c CM |
6212 | lockend = offset + count - 1; |
6213 | ||
6214 | if (writing) { | |
6215 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6216 | if (ret) | |
6217 | goto out; | |
6218 | } | |
4845e44f | 6219 | |
4b46fce2 | 6220 | while (1) { |
4845e44f CM |
6221 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6222 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6223 | /* |
6224 | * We're concerned with the entire range that we're going to be | |
6225 | * doing DIO to, so we need to make sure theres no ordered | |
6226 | * extents in this range. | |
6227 | */ | |
6228 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6229 | lockend - lockstart + 1); | |
6230 | if (!ordered) | |
6231 | break; | |
4845e44f CM |
6232 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6233 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6234 | btrfs_start_ordered_extent(inode, ordered, 1); |
6235 | btrfs_put_ordered_extent(ordered); | |
6236 | cond_resched(); | |
6237 | } | |
6238 | ||
4845e44f CM |
6239 | /* |
6240 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6241 | * the dirty or uptodate bits | |
6242 | */ | |
6243 | if (writing) { | |
6244 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6245 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6246 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6247 | GFP_NOFS); | |
6248 | if (ret) { | |
6249 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6250 | lockend, EXTENT_LOCKED | write_bits, | |
6251 | 1, 0, &cached_state, GFP_NOFS); | |
6252 | goto out; | |
6253 | } | |
6254 | } | |
6255 | ||
6256 | free_extent_state(cached_state); | |
6257 | cached_state = NULL; | |
6258 | ||
5a5f79b5 CM |
6259 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6260 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6261 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6262 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6263 | |
6264 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6265 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6266 | offset + iov_length(iov, nr_segs) - 1, | |
6267 | EXTENT_LOCKED | write_bits, 1, 0, | |
6268 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6269 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6270 | /* | |
6271 | * We're falling back to buffered, unlock the section we didn't | |
6272 | * do IO on. | |
6273 | */ | |
4845e44f CM |
6274 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6275 | offset + iov_length(iov, nr_segs) - 1, | |
6276 | EXTENT_LOCKED | write_bits, 1, 0, | |
6277 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6278 | } |
4845e44f CM |
6279 | out: |
6280 | free_extent_state(cached_state); | |
4b46fce2 | 6281 | return ret; |
16432985 CM |
6282 | } |
6283 | ||
1506fcc8 YS |
6284 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6285 | __u64 start, __u64 len) | |
6286 | { | |
ec29ed5b | 6287 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6288 | } |
6289 | ||
a52d9a80 | 6290 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6291 | { |
d1310b2e CM |
6292 | struct extent_io_tree *tree; |
6293 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 6294 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 6295 | } |
1832a6d5 | 6296 | |
a52d9a80 | 6297 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6298 | { |
d1310b2e | 6299 | struct extent_io_tree *tree; |
b888db2b CM |
6300 | |
6301 | ||
6302 | if (current->flags & PF_MEMALLOC) { | |
6303 | redirty_page_for_writepage(wbc, page); | |
6304 | unlock_page(page); | |
6305 | return 0; | |
6306 | } | |
d1310b2e | 6307 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6308 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6309 | } |
6310 | ||
f421950f CM |
6311 | int btrfs_writepages(struct address_space *mapping, |
6312 | struct writeback_control *wbc) | |
b293f02e | 6313 | { |
d1310b2e | 6314 | struct extent_io_tree *tree; |
771ed689 | 6315 | |
d1310b2e | 6316 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6317 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6318 | } | |
6319 | ||
3ab2fb5a CM |
6320 | static int |
6321 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6322 | struct list_head *pages, unsigned nr_pages) | |
6323 | { | |
d1310b2e CM |
6324 | struct extent_io_tree *tree; |
6325 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6326 | return extent_readpages(tree, mapping, pages, nr_pages, |
6327 | btrfs_get_extent); | |
6328 | } | |
e6dcd2dc | 6329 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6330 | { |
d1310b2e CM |
6331 | struct extent_io_tree *tree; |
6332 | struct extent_map_tree *map; | |
a52d9a80 | 6333 | int ret; |
8c2383c3 | 6334 | |
d1310b2e CM |
6335 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6336 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6337 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6338 | if (ret == 1) { |
6339 | ClearPagePrivate(page); | |
6340 | set_page_private(page, 0); | |
6341 | page_cache_release(page); | |
39279cc3 | 6342 | } |
a52d9a80 | 6343 | return ret; |
39279cc3 CM |
6344 | } |
6345 | ||
e6dcd2dc CM |
6346 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6347 | { | |
98509cfc CM |
6348 | if (PageWriteback(page) || PageDirty(page)) |
6349 | return 0; | |
b335b003 | 6350 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6351 | } |
6352 | ||
a52d9a80 | 6353 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6354 | { |
d1310b2e | 6355 | struct extent_io_tree *tree; |
e6dcd2dc | 6356 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6357 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6358 | u64 page_start = page_offset(page); |
6359 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6360 | |
8b62b72b CM |
6361 | |
6362 | /* | |
6363 | * we have the page locked, so new writeback can't start, | |
6364 | * and the dirty bit won't be cleared while we are here. | |
6365 | * | |
6366 | * Wait for IO on this page so that we can safely clear | |
6367 | * the PagePrivate2 bit and do ordered accounting | |
6368 | */ | |
e6dcd2dc | 6369 | wait_on_page_writeback(page); |
8b62b72b | 6370 | |
d1310b2e | 6371 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6372 | if (offset) { |
6373 | btrfs_releasepage(page, GFP_NOFS); | |
6374 | return; | |
6375 | } | |
2ac55d41 JB |
6376 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6377 | GFP_NOFS); | |
e6dcd2dc CM |
6378 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6379 | page_offset(page)); | |
6380 | if (ordered) { | |
eb84ae03 CM |
6381 | /* |
6382 | * IO on this page will never be started, so we need | |
6383 | * to account for any ordered extents now | |
6384 | */ | |
e6dcd2dc CM |
6385 | clear_extent_bit(tree, page_start, page_end, |
6386 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6387 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6388 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6389 | /* |
6390 | * whoever cleared the private bit is responsible | |
6391 | * for the finish_ordered_io | |
6392 | */ | |
6393 | if (TestClearPagePrivate2(page)) { | |
6394 | btrfs_finish_ordered_io(page->mapping->host, | |
6395 | page_start, page_end); | |
6396 | } | |
e6dcd2dc | 6397 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6398 | cached_state = NULL; |
6399 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6400 | GFP_NOFS); | |
e6dcd2dc CM |
6401 | } |
6402 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6403 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6404 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6405 | __btrfs_releasepage(page, GFP_NOFS); |
6406 | ||
4a096752 | 6407 | ClearPageChecked(page); |
9ad6b7bc | 6408 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6409 | ClearPagePrivate(page); |
6410 | set_page_private(page, 0); | |
6411 | page_cache_release(page); | |
6412 | } | |
39279cc3 CM |
6413 | } |
6414 | ||
9ebefb18 CM |
6415 | /* |
6416 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6417 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6418 | * be careful to check for EOF conditions here. We set the page up correctly | |
6419 | * for a written page which means we get ENOSPC checking when writing into | |
6420 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6421 | * support these features. | |
6422 | * | |
6423 | * We are not allowed to take the i_mutex here so we have to play games to | |
6424 | * protect against truncate races as the page could now be beyond EOF. Because | |
6425 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6426 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6427 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6428 | * unlock the page. | |
6429 | */ | |
c2ec175c | 6430 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6431 | { |
c2ec175c | 6432 | struct page *page = vmf->page; |
6da6abae | 6433 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6434 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6435 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6436 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6437 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6438 | char *kaddr; |
6439 | unsigned long zero_start; | |
9ebefb18 | 6440 | loff_t size; |
1832a6d5 | 6441 | int ret; |
a52d9a80 | 6442 | u64 page_start; |
e6dcd2dc | 6443 | u64 page_end; |
9ebefb18 | 6444 | |
0ca1f7ce | 6445 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
6446 | if (ret) { |
6447 | if (ret == -ENOMEM) | |
6448 | ret = VM_FAULT_OOM; | |
6449 | else /* -ENOSPC, -EIO, etc */ | |
6450 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 6451 | goto out; |
56a76f82 | 6452 | } |
1832a6d5 | 6453 | |
56a76f82 | 6454 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6455 | again: |
9ebefb18 | 6456 | lock_page(page); |
9ebefb18 | 6457 | size = i_size_read(inode); |
e6dcd2dc CM |
6458 | page_start = page_offset(page); |
6459 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6460 | |
9ebefb18 | 6461 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6462 | (page_start >= size)) { |
9ebefb18 CM |
6463 | /* page got truncated out from underneath us */ |
6464 | goto out_unlock; | |
6465 | } | |
e6dcd2dc CM |
6466 | wait_on_page_writeback(page); |
6467 | ||
2ac55d41 JB |
6468 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6469 | GFP_NOFS); | |
e6dcd2dc CM |
6470 | set_page_extent_mapped(page); |
6471 | ||
eb84ae03 CM |
6472 | /* |
6473 | * we can't set the delalloc bits if there are pending ordered | |
6474 | * extents. Drop our locks and wait for them to finish | |
6475 | */ | |
e6dcd2dc CM |
6476 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6477 | if (ordered) { | |
2ac55d41 JB |
6478 | unlock_extent_cached(io_tree, page_start, page_end, |
6479 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6480 | unlock_page(page); |
eb84ae03 | 6481 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6482 | btrfs_put_ordered_extent(ordered); |
6483 | goto again; | |
6484 | } | |
6485 | ||
fbf19087 JB |
6486 | /* |
6487 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6488 | * if it was already dirty, so for space accounting reasons we need to | |
6489 | * clear any delalloc bits for the range we are fixing to save. There | |
6490 | * is probably a better way to do this, but for now keep consistent with | |
6491 | * prepare_pages in the normal write path. | |
6492 | */ | |
2ac55d41 | 6493 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6494 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6495 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6496 | |
2ac55d41 JB |
6497 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6498 | &cached_state); | |
9ed74f2d | 6499 | if (ret) { |
2ac55d41 JB |
6500 | unlock_extent_cached(io_tree, page_start, page_end, |
6501 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6502 | ret = VM_FAULT_SIGBUS; |
6503 | goto out_unlock; | |
6504 | } | |
e6dcd2dc | 6505 | ret = 0; |
9ebefb18 CM |
6506 | |
6507 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6508 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6509 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6510 | else |
e6dcd2dc | 6511 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6512 | |
e6dcd2dc CM |
6513 | if (zero_start != PAGE_CACHE_SIZE) { |
6514 | kaddr = kmap(page); | |
6515 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6516 | flush_dcache_page(page); | |
6517 | kunmap(page); | |
6518 | } | |
247e743c | 6519 | ClearPageChecked(page); |
e6dcd2dc | 6520 | set_page_dirty(page); |
50a9b214 | 6521 | SetPageUptodate(page); |
5a3f23d5 | 6522 | |
257c62e1 CM |
6523 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6524 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6525 | ||
2ac55d41 | 6526 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6527 | |
6528 | out_unlock: | |
50a9b214 CM |
6529 | if (!ret) |
6530 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6531 | unlock_page(page); |
0ca1f7ce | 6532 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 6533 | out: |
9ebefb18 CM |
6534 | return ret; |
6535 | } | |
6536 | ||
a41ad394 | 6537 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6538 | { |
6539 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6540 | int ret; | |
3893e33b | 6541 | int err = 0; |
39279cc3 | 6542 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6543 | unsigned long nr; |
dbe674a9 | 6544 | u64 mask = root->sectorsize - 1; |
39279cc3 | 6545 | |
5d5e103a JB |
6546 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6547 | if (ret) | |
a41ad394 | 6548 | return ret; |
8082510e | 6549 | |
4a096752 | 6550 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6551 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6552 | |
f0cd846e JB |
6553 | trans = btrfs_start_transaction(root, 5); |
6554 | if (IS_ERR(trans)) | |
6555 | return PTR_ERR(trans); | |
6556 | ||
6557 | btrfs_set_trans_block_group(trans, inode); | |
6558 | ||
6559 | ret = btrfs_orphan_add(trans, inode); | |
6560 | if (ret) { | |
6561 | btrfs_end_transaction(trans, root); | |
6562 | return ret; | |
6563 | } | |
6564 | ||
6565 | nr = trans->blocks_used; | |
6566 | btrfs_end_transaction(trans, root); | |
6567 | btrfs_btree_balance_dirty(root, nr); | |
6568 | ||
6569 | /* Now start a transaction for the truncate */ | |
d68fc57b | 6570 | trans = btrfs_start_transaction(root, 0); |
3893e33b JB |
6571 | if (IS_ERR(trans)) |
6572 | return PTR_ERR(trans); | |
8082510e | 6573 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b | 6574 | trans->block_rsv = root->orphan_block_rsv; |
5a3f23d5 CM |
6575 | |
6576 | /* | |
6577 | * setattr is responsible for setting the ordered_data_close flag, | |
6578 | * but that is only tested during the last file release. That | |
6579 | * could happen well after the next commit, leaving a great big | |
6580 | * window where new writes may get lost if someone chooses to write | |
6581 | * to this file after truncating to zero | |
6582 | * | |
6583 | * The inode doesn't have any dirty data here, and so if we commit | |
6584 | * this is a noop. If someone immediately starts writing to the inode | |
6585 | * it is very likely we'll catch some of their writes in this | |
6586 | * transaction, and the commit will find this file on the ordered | |
6587 | * data list with good things to send down. | |
6588 | * | |
6589 | * This is a best effort solution, there is still a window where | |
6590 | * using truncate to replace the contents of the file will | |
6591 | * end up with a zero length file after a crash. | |
6592 | */ | |
6593 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6594 | btrfs_add_ordered_operation(trans, root, inode); | |
6595 | ||
8082510e | 6596 | while (1) { |
d68fc57b YZ |
6597 | if (!trans) { |
6598 | trans = btrfs_start_transaction(root, 0); | |
3893e33b JB |
6599 | if (IS_ERR(trans)) |
6600 | return PTR_ERR(trans); | |
d68fc57b YZ |
6601 | btrfs_set_trans_block_group(trans, inode); |
6602 | trans->block_rsv = root->orphan_block_rsv; | |
6603 | } | |
6604 | ||
6605 | ret = btrfs_block_rsv_check(trans, root, | |
6606 | root->orphan_block_rsv, 0, 5); | |
3893e33b | 6607 | if (ret == -EAGAIN) { |
d68fc57b | 6608 | ret = btrfs_commit_transaction(trans, root); |
3893e33b JB |
6609 | if (ret) |
6610 | return ret; | |
d68fc57b YZ |
6611 | trans = NULL; |
6612 | continue; | |
3893e33b JB |
6613 | } else if (ret) { |
6614 | err = ret; | |
6615 | break; | |
d68fc57b YZ |
6616 | } |
6617 | ||
8082510e YZ |
6618 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6619 | inode->i_size, | |
6620 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6621 | if (ret != -EAGAIN) { |
6622 | err = ret; | |
8082510e | 6623 | break; |
3893e33b | 6624 | } |
39279cc3 | 6625 | |
8082510e | 6626 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6627 | if (ret) { |
6628 | err = ret; | |
6629 | break; | |
6630 | } | |
5f39d397 | 6631 | |
8082510e YZ |
6632 | nr = trans->blocks_used; |
6633 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6634 | trans = NULL; |
8082510e | 6635 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6636 | } |
6637 | ||
6638 | if (ret == 0 && inode->i_nlink > 0) { | |
6639 | ret = btrfs_orphan_del(trans, inode); | |
3893e33b JB |
6640 | if (ret) |
6641 | err = ret; | |
ded5db9d JB |
6642 | } else if (ret && inode->i_nlink > 0) { |
6643 | /* | |
6644 | * Failed to do the truncate, remove us from the in memory | |
6645 | * orphan list. | |
6646 | */ | |
6647 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6648 | } |
6649 | ||
6650 | ret = btrfs_update_inode(trans, root, inode); | |
3893e33b JB |
6651 | if (ret && !err) |
6652 | err = ret; | |
7b128766 | 6653 | |
7b128766 | 6654 | nr = trans->blocks_used; |
89ce8a63 | 6655 | ret = btrfs_end_transaction_throttle(trans, root); |
3893e33b JB |
6656 | if (ret && !err) |
6657 | err = ret; | |
d3c2fdcf | 6658 | btrfs_btree_balance_dirty(root, nr); |
a41ad394 | 6659 | |
3893e33b | 6660 | return err; |
39279cc3 CM |
6661 | } |
6662 | ||
d352ac68 CM |
6663 | /* |
6664 | * create a new subvolume directory/inode (helper for the ioctl). | |
6665 | */ | |
d2fb3437 | 6666 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 6667 | struct btrfs_root *new_root, |
d2fb3437 | 6668 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 6669 | { |
39279cc3 | 6670 | struct inode *inode; |
76dda93c | 6671 | int err; |
00e4e6b3 | 6672 | u64 index = 0; |
39279cc3 | 6673 | |
aec7477b | 6674 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 6675 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 6676 | if (IS_ERR(inode)) |
f46b5a66 | 6677 | return PTR_ERR(inode); |
39279cc3 CM |
6678 | inode->i_op = &btrfs_dir_inode_operations; |
6679 | inode->i_fop = &btrfs_dir_file_operations; | |
6680 | ||
39279cc3 | 6681 | inode->i_nlink = 1; |
dbe674a9 | 6682 | btrfs_i_size_write(inode, 0); |
3b96362c | 6683 | |
76dda93c YZ |
6684 | err = btrfs_update_inode(trans, new_root, inode); |
6685 | BUG_ON(err); | |
cb8e7090 | 6686 | |
76dda93c | 6687 | iput(inode); |
cb8e7090 | 6688 | return 0; |
39279cc3 CM |
6689 | } |
6690 | ||
d352ac68 CM |
6691 | /* helper function for file defrag and space balancing. This |
6692 | * forces readahead on a given range of bytes in an inode | |
6693 | */ | |
edbd8d4e | 6694 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
6695 | struct file_ra_state *ra, struct file *file, |
6696 | pgoff_t offset, pgoff_t last_index) | |
6697 | { | |
8e7bf94f | 6698 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 6699 | |
86479a04 CM |
6700 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
6701 | return offset + req_size; | |
86479a04 CM |
6702 | } |
6703 | ||
39279cc3 CM |
6704 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6705 | { | |
6706 | struct btrfs_inode *ei; | |
2ead6ae7 | 6707 | struct inode *inode; |
39279cc3 CM |
6708 | |
6709 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6710 | if (!ei) | |
6711 | return NULL; | |
2ead6ae7 YZ |
6712 | |
6713 | ei->root = NULL; | |
6714 | ei->space_info = NULL; | |
6715 | ei->generation = 0; | |
6716 | ei->sequence = 0; | |
15ee9bc7 | 6717 | ei->last_trans = 0; |
257c62e1 | 6718 | ei->last_sub_trans = 0; |
e02119d5 | 6719 | ei->logged_trans = 0; |
2ead6ae7 YZ |
6720 | ei->delalloc_bytes = 0; |
6721 | ei->reserved_bytes = 0; | |
6722 | ei->disk_i_size = 0; | |
6723 | ei->flags = 0; | |
6724 | ei->index_cnt = (u64)-1; | |
6725 | ei->last_unlink_trans = 0; | |
6726 | ||
0ca1f7ce | 6727 | atomic_set(&ei->outstanding_extents, 0); |
57a45ced | 6728 | atomic_set(&ei->reserved_extents, 0); |
2ead6ae7 YZ |
6729 | |
6730 | ei->ordered_data_close = 0; | |
d68fc57b | 6731 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6732 | ei->dummy_inode = 0; |
261507a0 | 6733 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 YZ |
6734 | |
6735 | inode = &ei->vfs_inode; | |
6736 | extent_map_tree_init(&ei->extent_tree, GFP_NOFS); | |
6737 | extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS); | |
6738 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS); | |
6739 | mutex_init(&ei->log_mutex); | |
e6dcd2dc | 6740 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6741 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6742 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6743 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6744 | RB_CLEAR_NODE(&ei->rb_node); |
6745 | ||
6746 | return inode; | |
39279cc3 CM |
6747 | } |
6748 | ||
fa0d7e3d NP |
6749 | static void btrfs_i_callback(struct rcu_head *head) |
6750 | { | |
6751 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6752 | INIT_LIST_HEAD(&inode->i_dentry); | |
6753 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6754 | } | |
6755 | ||
39279cc3 CM |
6756 | void btrfs_destroy_inode(struct inode *inode) |
6757 | { | |
e6dcd2dc | 6758 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6759 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6760 | ||
39279cc3 CM |
6761 | WARN_ON(!list_empty(&inode->i_dentry)); |
6762 | WARN_ON(inode->i_data.nrpages); | |
0ca1f7ce | 6763 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents)); |
57a45ced | 6764 | WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents)); |
39279cc3 | 6765 | |
a6dbd429 JB |
6766 | /* |
6767 | * This can happen where we create an inode, but somebody else also | |
6768 | * created the same inode and we need to destroy the one we already | |
6769 | * created. | |
6770 | */ | |
6771 | if (!root) | |
6772 | goto free; | |
6773 | ||
5a3f23d5 CM |
6774 | /* |
6775 | * Make sure we're properly removed from the ordered operation | |
6776 | * lists. | |
6777 | */ | |
6778 | smp_mb(); | |
6779 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6780 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6781 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6782 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6783 | } | |
6784 | ||
0af3d00b JB |
6785 | if (root == root->fs_info->tree_root) { |
6786 | struct btrfs_block_group_cache *block_group; | |
6787 | ||
6788 | block_group = btrfs_lookup_block_group(root->fs_info, | |
6789 | BTRFS_I(inode)->block_group); | |
6790 | if (block_group && block_group->inode == inode) { | |
6791 | spin_lock(&block_group->lock); | |
6792 | block_group->inode = NULL; | |
6793 | spin_unlock(&block_group->lock); | |
6794 | btrfs_put_block_group(block_group); | |
6795 | } else if (block_group) { | |
6796 | btrfs_put_block_group(block_group); | |
6797 | } | |
6798 | } | |
6799 | ||
d68fc57b | 6800 | spin_lock(&root->orphan_lock); |
7b128766 | 6801 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
8082510e YZ |
6802 | printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", |
6803 | inode->i_ino); | |
6804 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
7b128766 | 6805 | } |
d68fc57b | 6806 | spin_unlock(&root->orphan_lock); |
7b128766 | 6807 | |
d397712b | 6808 | while (1) { |
e6dcd2dc CM |
6809 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6810 | if (!ordered) | |
6811 | break; | |
6812 | else { | |
d397712b CM |
6813 | printk(KERN_ERR "btrfs found ordered " |
6814 | "extent %llu %llu on inode cleanup\n", | |
6815 | (unsigned long long)ordered->file_offset, | |
6816 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6817 | btrfs_remove_ordered_extent(inode, ordered); |
6818 | btrfs_put_ordered_extent(ordered); | |
6819 | btrfs_put_ordered_extent(ordered); | |
6820 | } | |
6821 | } | |
5d4f98a2 | 6822 | inode_tree_del(inode); |
5b21f2ed | 6823 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6824 | free: |
fa0d7e3d | 6825 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6826 | } |
6827 | ||
45321ac5 | 6828 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6829 | { |
6830 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6831 | |
0af3d00b JB |
6832 | if (btrfs_root_refs(&root->root_item) == 0 && |
6833 | root != root->fs_info->tree_root) | |
45321ac5 | 6834 | return 1; |
76dda93c | 6835 | else |
45321ac5 | 6836 | return generic_drop_inode(inode); |
76dda93c YZ |
6837 | } |
6838 | ||
0ee0fda0 | 6839 | static void init_once(void *foo) |
39279cc3 CM |
6840 | { |
6841 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6842 | ||
6843 | inode_init_once(&ei->vfs_inode); | |
6844 | } | |
6845 | ||
6846 | void btrfs_destroy_cachep(void) | |
6847 | { | |
6848 | if (btrfs_inode_cachep) | |
6849 | kmem_cache_destroy(btrfs_inode_cachep); | |
6850 | if (btrfs_trans_handle_cachep) | |
6851 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6852 | if (btrfs_transaction_cachep) | |
6853 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6854 | if (btrfs_path_cachep) |
6855 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6856 | if (btrfs_free_space_cachep) |
6857 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6858 | } |
6859 | ||
6860 | int btrfs_init_cachep(void) | |
6861 | { | |
9601e3f6 CH |
6862 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6863 | sizeof(struct btrfs_inode), 0, | |
6864 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6865 | if (!btrfs_inode_cachep) |
6866 | goto fail; | |
9601e3f6 CH |
6867 | |
6868 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6869 | sizeof(struct btrfs_trans_handle), 0, | |
6870 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6871 | if (!btrfs_trans_handle_cachep) |
6872 | goto fail; | |
9601e3f6 CH |
6873 | |
6874 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6875 | sizeof(struct btrfs_transaction), 0, | |
6876 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6877 | if (!btrfs_transaction_cachep) |
6878 | goto fail; | |
9601e3f6 CH |
6879 | |
6880 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6881 | sizeof(struct btrfs_path), 0, | |
6882 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6883 | if (!btrfs_path_cachep) |
6884 | goto fail; | |
9601e3f6 | 6885 | |
dc89e982 JB |
6886 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6887 | sizeof(struct btrfs_free_space), 0, | |
6888 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6889 | if (!btrfs_free_space_cachep) | |
6890 | goto fail; | |
6891 | ||
39279cc3 CM |
6892 | return 0; |
6893 | fail: | |
6894 | btrfs_destroy_cachep(); | |
6895 | return -ENOMEM; | |
6896 | } | |
6897 | ||
6898 | static int btrfs_getattr(struct vfsmount *mnt, | |
6899 | struct dentry *dentry, struct kstat *stat) | |
6900 | { | |
6901 | struct inode *inode = dentry->d_inode; | |
6902 | generic_fillattr(inode, stat); | |
3394e160 | 6903 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 6904 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
6905 | stat->blocks = (inode_get_bytes(inode) + |
6906 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
6907 | return 0; |
6908 | } | |
6909 | ||
75e7cb7f LB |
6910 | /* |
6911 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6912 | * directory. | |
6913 | */ | |
6914 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6915 | { | |
6916 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6917 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6918 | ||
6919 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6920 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6921 | else | |
6922 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6923 | ||
6924 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6925 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6926 | else | |
6927 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6928 | } | |
6929 | ||
d397712b CM |
6930 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6931 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6932 | { |
6933 | struct btrfs_trans_handle *trans; | |
6934 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6935 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6936 | struct inode *new_inode = new_dentry->d_inode; |
6937 | struct inode *old_inode = old_dentry->d_inode; | |
6938 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6939 | u64 index = 0; |
4df27c4d | 6940 | u64 root_objectid; |
39279cc3 CM |
6941 | int ret; |
6942 | ||
f679a840 YZ |
6943 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
6944 | return -EPERM; | |
6945 | ||
4df27c4d YZ |
6946 | /* we only allow rename subvolume link between subvolumes */ |
6947 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
6948 | return -EXDEV; |
6949 | ||
4df27c4d YZ |
6950 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
6951 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 6952 | return -ENOTEMPTY; |
5f39d397 | 6953 | |
4df27c4d YZ |
6954 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
6955 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
6956 | return -ENOTEMPTY; | |
5a3f23d5 CM |
6957 | /* |
6958 | * we're using rename to replace one file with another. | |
6959 | * and the replacement file is large. Start IO on it now so | |
6960 | * we don't add too much work to the end of the transaction | |
6961 | */ | |
4baf8c92 | 6962 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
6963 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
6964 | filemap_flush(old_inode->i_mapping); | |
6965 | ||
76dda93c YZ |
6966 | /* close the racy window with snapshot create/destroy ioctl */ |
6967 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
6968 | down_read(&root->fs_info->subvol_sem); | |
a22285a6 YZ |
6969 | /* |
6970 | * We want to reserve the absolute worst case amount of items. So if | |
6971 | * both inodes are subvols and we need to unlink them then that would | |
6972 | * require 4 item modifications, but if they are both normal inodes it | |
6973 | * would require 5 item modifications, so we'll assume their normal | |
6974 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
6975 | * should cover the worst case number of items we'll modify. | |
6976 | */ | |
6977 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
6978 | if (IS_ERR(trans)) { |
6979 | ret = PTR_ERR(trans); | |
6980 | goto out_notrans; | |
6981 | } | |
76dda93c | 6982 | |
a5719521 | 6983 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 6984 | |
4df27c4d YZ |
6985 | if (dest != root) |
6986 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 6987 | |
a5719521 YZ |
6988 | ret = btrfs_set_inode_index(new_dir, &index); |
6989 | if (ret) | |
6990 | goto out_fail; | |
5a3f23d5 | 6991 | |
a5719521 | 6992 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6993 | /* force full log commit if subvolume involved. */ |
6994 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
6995 | } else { | |
a5719521 YZ |
6996 | ret = btrfs_insert_inode_ref(trans, dest, |
6997 | new_dentry->d_name.name, | |
6998 | new_dentry->d_name.len, | |
6999 | old_inode->i_ino, | |
7000 | new_dir->i_ino, index); | |
7001 | if (ret) | |
7002 | goto out_fail; | |
4df27c4d YZ |
7003 | /* |
7004 | * this is an ugly little race, but the rename is required | |
7005 | * to make sure that if we crash, the inode is either at the | |
7006 | * old name or the new one. pinning the log transaction lets | |
7007 | * us make sure we don't allow a log commit to come in after | |
7008 | * we unlink the name but before we add the new name back in. | |
7009 | */ | |
7010 | btrfs_pin_log_trans(root); | |
7011 | } | |
5a3f23d5 CM |
7012 | /* |
7013 | * make sure the inode gets flushed if it is replacing | |
7014 | * something. | |
7015 | */ | |
7016 | if (new_inode && new_inode->i_size && | |
7017 | old_inode && S_ISREG(old_inode->i_mode)) { | |
7018 | btrfs_add_ordered_operation(trans, root, old_inode); | |
7019 | } | |
7020 | ||
39279cc3 CM |
7021 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7022 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7023 | old_inode->i_ctime = ctime; | |
5f39d397 | 7024 | |
12fcfd22 CM |
7025 | if (old_dentry->d_parent != new_dentry->d_parent) |
7026 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7027 | ||
4df27c4d YZ |
7028 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
7029 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
7030 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7031 | old_dentry->d_name.name, | |
7032 | old_dentry->d_name.len); | |
7033 | } else { | |
92986796 AV |
7034 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7035 | old_dentry->d_inode, | |
7036 | old_dentry->d_name.name, | |
7037 | old_dentry->d_name.len); | |
7038 | if (!ret) | |
7039 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
7040 | } |
7041 | BUG_ON(ret); | |
39279cc3 CM |
7042 | |
7043 | if (new_inode) { | |
7044 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
7045 | if (unlikely(new_inode->i_ino == |
7046 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
7047 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7048 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7049 | root_objectid, | |
7050 | new_dentry->d_name.name, | |
7051 | new_dentry->d_name.len); | |
7052 | BUG_ON(new_inode->i_nlink == 0); | |
7053 | } else { | |
7054 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7055 | new_dentry->d_inode, | |
7056 | new_dentry->d_name.name, | |
7057 | new_dentry->d_name.len); | |
7058 | } | |
7059 | BUG_ON(ret); | |
7b128766 | 7060 | if (new_inode->i_nlink == 0) { |
e02119d5 | 7061 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7062 | BUG_ON(ret); |
7b128766 | 7063 | } |
39279cc3 | 7064 | } |
aec7477b | 7065 | |
75e7cb7f LB |
7066 | fixup_inode_flags(new_dir, old_inode); |
7067 | ||
4df27c4d YZ |
7068 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7069 | new_dentry->d_name.name, | |
a5719521 | 7070 | new_dentry->d_name.len, 0, index); |
4df27c4d | 7071 | BUG_ON(ret); |
39279cc3 | 7072 | |
4df27c4d | 7073 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
6a912213 JB |
7074 | struct dentry *parent = dget_parent(new_dentry); |
7075 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
7076 | dput(parent); | |
4df27c4d YZ |
7077 | btrfs_end_log_trans(root); |
7078 | } | |
39279cc3 | 7079 | out_fail: |
ab78c84d | 7080 | btrfs_end_transaction_throttle(trans, root); |
b44c59a8 | 7081 | out_notrans: |
76dda93c YZ |
7082 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
7083 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 7084 | |
39279cc3 CM |
7085 | return ret; |
7086 | } | |
7087 | ||
d352ac68 CM |
7088 | /* |
7089 | * some fairly slow code that needs optimization. This walks the list | |
7090 | * of all the inodes with pending delalloc and forces them to disk. | |
7091 | */ | |
24bbcf04 | 7092 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7093 | { |
7094 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7095 | struct btrfs_inode *binode; | |
5b21f2ed | 7096 | struct inode *inode; |
ea8c2819 | 7097 | |
c146afad YZ |
7098 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7099 | return -EROFS; | |
7100 | ||
75eff68e | 7101 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7102 | while (!list_empty(head)) { |
ea8c2819 CM |
7103 | binode = list_entry(head->next, struct btrfs_inode, |
7104 | delalloc_inodes); | |
5b21f2ed ZY |
7105 | inode = igrab(&binode->vfs_inode); |
7106 | if (!inode) | |
7107 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7108 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7109 | if (inode) { |
8c8bee1d | 7110 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7111 | if (delay_iput) |
7112 | btrfs_add_delayed_iput(inode); | |
7113 | else | |
7114 | iput(inode); | |
5b21f2ed ZY |
7115 | } |
7116 | cond_resched(); | |
75eff68e | 7117 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7118 | } |
75eff68e | 7119 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7120 | |
7121 | /* the filemap_flush will queue IO into the worker threads, but | |
7122 | * we have to make sure the IO is actually started and that | |
7123 | * ordered extents get created before we return | |
7124 | */ | |
7125 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7126 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7127 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7128 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7129 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7130 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7131 | } |
7132 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7133 | return 0; |
7134 | } | |
7135 | ||
0019f10d JB |
7136 | int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput, |
7137 | int sync) | |
5da9d01b YZ |
7138 | { |
7139 | struct btrfs_inode *binode; | |
7140 | struct inode *inode = NULL; | |
7141 | ||
7142 | spin_lock(&root->fs_info->delalloc_lock); | |
7143 | while (!list_empty(&root->fs_info->delalloc_inodes)) { | |
7144 | binode = list_entry(root->fs_info->delalloc_inodes.next, | |
7145 | struct btrfs_inode, delalloc_inodes); | |
7146 | inode = igrab(&binode->vfs_inode); | |
7147 | if (inode) { | |
7148 | list_move_tail(&binode->delalloc_inodes, | |
7149 | &root->fs_info->delalloc_inodes); | |
7150 | break; | |
7151 | } | |
7152 | ||
7153 | list_del_init(&binode->delalloc_inodes); | |
7154 | cond_resched_lock(&root->fs_info->delalloc_lock); | |
7155 | } | |
7156 | spin_unlock(&root->fs_info->delalloc_lock); | |
7157 | ||
7158 | if (inode) { | |
0019f10d JB |
7159 | if (sync) { |
7160 | filemap_write_and_wait(inode->i_mapping); | |
7161 | /* | |
7162 | * We have to do this because compression doesn't | |
7163 | * actually set PG_writeback until it submits the pages | |
7164 | * for IO, which happens in an async thread, so we could | |
7165 | * race and not actually wait for any writeback pages | |
7166 | * because they've not been submitted yet. Technically | |
7167 | * this could still be the case for the ordered stuff | |
7168 | * since the async thread may not have started to do its | |
7169 | * work yet. If this becomes the case then we need to | |
7170 | * figure out a way to make sure that in writepage we | |
7171 | * wait for any async pages to be submitted before | |
7172 | * returning so that fdatawait does what its supposed to | |
7173 | * do. | |
7174 | */ | |
7175 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
7176 | } else { | |
7177 | filemap_flush(inode->i_mapping); | |
7178 | } | |
5da9d01b YZ |
7179 | if (delay_iput) |
7180 | btrfs_add_delayed_iput(inode); | |
7181 | else | |
7182 | iput(inode); | |
7183 | return 1; | |
7184 | } | |
7185 | return 0; | |
7186 | } | |
7187 | ||
39279cc3 CM |
7188 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7189 | const char *symname) | |
7190 | { | |
7191 | struct btrfs_trans_handle *trans; | |
7192 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7193 | struct btrfs_path *path; | |
7194 | struct btrfs_key key; | |
1832a6d5 | 7195 | struct inode *inode = NULL; |
39279cc3 CM |
7196 | int err; |
7197 | int drop_inode = 0; | |
7198 | u64 objectid; | |
00e4e6b3 | 7199 | u64 index = 0 ; |
39279cc3 CM |
7200 | int name_len; |
7201 | int datasize; | |
5f39d397 | 7202 | unsigned long ptr; |
39279cc3 | 7203 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7204 | struct extent_buffer *leaf; |
1832a6d5 | 7205 | unsigned long nr = 0; |
39279cc3 CM |
7206 | |
7207 | name_len = strlen(symname) + 1; | |
7208 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7209 | return -ENAMETOOLONG; | |
1832a6d5 | 7210 | |
a22285a6 YZ |
7211 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
7212 | if (err) | |
7213 | return err; | |
9ed74f2d JB |
7214 | /* |
7215 | * 2 items for inode item and ref | |
7216 | * 2 items for dir items | |
7217 | * 1 item for xattr if selinux is on | |
7218 | */ | |
a22285a6 YZ |
7219 | trans = btrfs_start_transaction(root, 5); |
7220 | if (IS_ERR(trans)) | |
7221 | return PTR_ERR(trans); | |
1832a6d5 | 7222 | |
39279cc3 CM |
7223 | btrfs_set_trans_block_group(trans, dir); |
7224 | ||
aec7477b | 7225 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 7226 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
7227 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
7228 | &index); | |
39279cc3 CM |
7229 | err = PTR_ERR(inode); |
7230 | if (IS_ERR(inode)) | |
7231 | goto out_unlock; | |
7232 | ||
f34f57a3 | 7233 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
7234 | if (err) { |
7235 | drop_inode = 1; | |
7236 | goto out_unlock; | |
7237 | } | |
7238 | ||
39279cc3 | 7239 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 7240 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7241 | if (err) |
7242 | drop_inode = 1; | |
7243 | else { | |
7244 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7245 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
7246 | inode->i_fop = &btrfs_file_operations; |
7247 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 7248 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7249 | } |
39279cc3 CM |
7250 | btrfs_update_inode_block_group(trans, inode); |
7251 | btrfs_update_inode_block_group(trans, dir); | |
7252 | if (drop_inode) | |
7253 | goto out_unlock; | |
7254 | ||
7255 | path = btrfs_alloc_path(); | |
7256 | BUG_ON(!path); | |
7257 | key.objectid = inode->i_ino; | |
7258 | key.offset = 0; | |
39279cc3 CM |
7259 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7260 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7261 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7262 | datasize); | |
54aa1f4d CM |
7263 | if (err) { |
7264 | drop_inode = 1; | |
7265 | goto out_unlock; | |
7266 | } | |
5f39d397 CM |
7267 | leaf = path->nodes[0]; |
7268 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7269 | struct btrfs_file_extent_item); | |
7270 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7271 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7272 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7273 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7274 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7275 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7276 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7277 | ||
39279cc3 | 7278 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7279 | write_extent_buffer(leaf, symname, ptr, name_len); |
7280 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7281 | btrfs_free_path(path); |
5f39d397 | 7282 | |
39279cc3 CM |
7283 | inode->i_op = &btrfs_symlink_inode_operations; |
7284 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7285 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7286 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7287 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7288 | err = btrfs_update_inode(trans, root, inode); |
7289 | if (err) | |
7290 | drop_inode = 1; | |
39279cc3 CM |
7291 | |
7292 | out_unlock: | |
d3c2fdcf | 7293 | nr = trans->blocks_used; |
ab78c84d | 7294 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
7295 | if (drop_inode) { |
7296 | inode_dec_link_count(inode); | |
7297 | iput(inode); | |
7298 | } | |
d3c2fdcf | 7299 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7300 | return err; |
7301 | } | |
16432985 | 7302 | |
0af3d00b JB |
7303 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7304 | u64 start, u64 num_bytes, u64 min_size, | |
7305 | loff_t actual_len, u64 *alloc_hint, | |
7306 | struct btrfs_trans_handle *trans) | |
d899e052 | 7307 | { |
d899e052 YZ |
7308 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7309 | struct btrfs_key ins; | |
d899e052 | 7310 | u64 cur_offset = start; |
55a61d1d | 7311 | u64 i_size; |
d899e052 | 7312 | int ret = 0; |
0af3d00b | 7313 | bool own_trans = true; |
d899e052 | 7314 | |
0af3d00b JB |
7315 | if (trans) |
7316 | own_trans = false; | |
d899e052 | 7317 | while (num_bytes > 0) { |
0af3d00b JB |
7318 | if (own_trans) { |
7319 | trans = btrfs_start_transaction(root, 3); | |
7320 | if (IS_ERR(trans)) { | |
7321 | ret = PTR_ERR(trans); | |
7322 | break; | |
7323 | } | |
5a303d5d YZ |
7324 | } |
7325 | ||
efa56464 YZ |
7326 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7327 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7328 | if (ret) { |
0af3d00b JB |
7329 | if (own_trans) |
7330 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7331 | break; |
d899e052 | 7332 | } |
5a303d5d | 7333 | |
d899e052 YZ |
7334 | ret = insert_reserved_file_extent(trans, inode, |
7335 | cur_offset, ins.objectid, | |
7336 | ins.offset, ins.offset, | |
920bbbfb | 7337 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7338 | BTRFS_FILE_EXTENT_PREALLOC); |
7339 | BUG_ON(ret); | |
a1ed835e CM |
7340 | btrfs_drop_extent_cache(inode, cur_offset, |
7341 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7342 | |
d899e052 YZ |
7343 | num_bytes -= ins.offset; |
7344 | cur_offset += ins.offset; | |
efa56464 | 7345 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7346 | |
d899e052 | 7347 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7348 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7349 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7350 | (actual_len > inode->i_size) && |
7351 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7352 | if (cur_offset > actual_len) |
55a61d1d | 7353 | i_size = actual_len; |
d1ea6a61 | 7354 | else |
55a61d1d JB |
7355 | i_size = cur_offset; |
7356 | i_size_write(inode, i_size); | |
7357 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7358 | } |
7359 | ||
d899e052 YZ |
7360 | ret = btrfs_update_inode(trans, root, inode); |
7361 | BUG_ON(ret); | |
d899e052 | 7362 | |
0af3d00b JB |
7363 | if (own_trans) |
7364 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7365 | } |
d899e052 YZ |
7366 | return ret; |
7367 | } | |
7368 | ||
0af3d00b JB |
7369 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7370 | u64 start, u64 num_bytes, u64 min_size, | |
7371 | loff_t actual_len, u64 *alloc_hint) | |
7372 | { | |
7373 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7374 | min_size, actual_len, alloc_hint, | |
7375 | NULL); | |
7376 | } | |
7377 | ||
7378 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7379 | struct btrfs_trans_handle *trans, int mode, | |
7380 | u64 start, u64 num_bytes, u64 min_size, | |
7381 | loff_t actual_len, u64 *alloc_hint) | |
7382 | { | |
7383 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7384 | min_size, actual_len, alloc_hint, trans); | |
7385 | } | |
7386 | ||
e6dcd2dc CM |
7387 | static int btrfs_set_page_dirty(struct page *page) |
7388 | { | |
e6dcd2dc CM |
7389 | return __set_page_dirty_nobuffers(page); |
7390 | } | |
7391 | ||
b74c79e9 | 7392 | static int btrfs_permission(struct inode *inode, int mask, unsigned int flags) |
fdebe2bd | 7393 | { |
b83cc969 LZ |
7394 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7395 | ||
7396 | if (btrfs_root_readonly(root) && (mask & MAY_WRITE)) | |
7397 | return -EROFS; | |
6cbff00f | 7398 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 7399 | return -EACCES; |
b74c79e9 | 7400 | return generic_permission(inode, mask, flags, btrfs_check_acl); |
fdebe2bd | 7401 | } |
39279cc3 | 7402 | |
6e1d5dcc | 7403 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7404 | .getattr = btrfs_getattr, |
39279cc3 CM |
7405 | .lookup = btrfs_lookup, |
7406 | .create = btrfs_create, | |
7407 | .unlink = btrfs_unlink, | |
7408 | .link = btrfs_link, | |
7409 | .mkdir = btrfs_mkdir, | |
7410 | .rmdir = btrfs_rmdir, | |
7411 | .rename = btrfs_rename, | |
7412 | .symlink = btrfs_symlink, | |
7413 | .setattr = btrfs_setattr, | |
618e21d5 | 7414 | .mknod = btrfs_mknod, |
95819c05 CH |
7415 | .setxattr = btrfs_setxattr, |
7416 | .getxattr = btrfs_getxattr, | |
5103e947 | 7417 | .listxattr = btrfs_listxattr, |
95819c05 | 7418 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7419 | .permission = btrfs_permission, |
39279cc3 | 7420 | }; |
6e1d5dcc | 7421 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7422 | .lookup = btrfs_lookup, |
fdebe2bd | 7423 | .permission = btrfs_permission, |
39279cc3 | 7424 | }; |
76dda93c | 7425 | |
828c0950 | 7426 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7427 | .llseek = generic_file_llseek, |
7428 | .read = generic_read_dir, | |
cbdf5a24 | 7429 | .readdir = btrfs_real_readdir, |
34287aa3 | 7430 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7431 | #ifdef CONFIG_COMPAT |
34287aa3 | 7432 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7433 | #endif |
6bf13c0c | 7434 | .release = btrfs_release_file, |
e02119d5 | 7435 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7436 | }; |
7437 | ||
d1310b2e | 7438 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7439 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7440 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7441 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7442 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7443 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7444 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 7445 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
7446 | .set_bit_hook = btrfs_set_bit_hook, |
7447 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7448 | .merge_extent_hook = btrfs_merge_extent_hook, |
7449 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7450 | }; |
7451 | ||
35054394 CM |
7452 | /* |
7453 | * btrfs doesn't support the bmap operation because swapfiles | |
7454 | * use bmap to make a mapping of extents in the file. They assume | |
7455 | * these extents won't change over the life of the file and they | |
7456 | * use the bmap result to do IO directly to the drive. | |
7457 | * | |
7458 | * the btrfs bmap call would return logical addresses that aren't | |
7459 | * suitable for IO and they also will change frequently as COW | |
7460 | * operations happen. So, swapfile + btrfs == corruption. | |
7461 | * | |
7462 | * For now we're avoiding this by dropping bmap. | |
7463 | */ | |
7f09410b | 7464 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7465 | .readpage = btrfs_readpage, |
7466 | .writepage = btrfs_writepage, | |
b293f02e | 7467 | .writepages = btrfs_writepages, |
3ab2fb5a | 7468 | .readpages = btrfs_readpages, |
39279cc3 | 7469 | .sync_page = block_sync_page, |
16432985 | 7470 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7471 | .invalidatepage = btrfs_invalidatepage, |
7472 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7473 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7474 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7475 | }; |
7476 | ||
7f09410b | 7477 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7478 | .readpage = btrfs_readpage, |
7479 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7480 | .invalidatepage = btrfs_invalidatepage, |
7481 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7482 | }; |
7483 | ||
6e1d5dcc | 7484 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7485 | .getattr = btrfs_getattr, |
7486 | .setattr = btrfs_setattr, | |
95819c05 CH |
7487 | .setxattr = btrfs_setxattr, |
7488 | .getxattr = btrfs_getxattr, | |
5103e947 | 7489 | .listxattr = btrfs_listxattr, |
95819c05 | 7490 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7491 | .permission = btrfs_permission, |
1506fcc8 | 7492 | .fiemap = btrfs_fiemap, |
39279cc3 | 7493 | }; |
6e1d5dcc | 7494 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7495 | .getattr = btrfs_getattr, |
7496 | .setattr = btrfs_setattr, | |
fdebe2bd | 7497 | .permission = btrfs_permission, |
95819c05 CH |
7498 | .setxattr = btrfs_setxattr, |
7499 | .getxattr = btrfs_getxattr, | |
33268eaf | 7500 | .listxattr = btrfs_listxattr, |
95819c05 | 7501 | .removexattr = btrfs_removexattr, |
618e21d5 | 7502 | }; |
6e1d5dcc | 7503 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7504 | .readlink = generic_readlink, |
7505 | .follow_link = page_follow_link_light, | |
7506 | .put_link = page_put_link, | |
f209561a | 7507 | .getattr = btrfs_getattr, |
fdebe2bd | 7508 | .permission = btrfs_permission, |
0279b4cd JO |
7509 | .setxattr = btrfs_setxattr, |
7510 | .getxattr = btrfs_getxattr, | |
7511 | .listxattr = btrfs_listxattr, | |
7512 | .removexattr = btrfs_removexattr, | |
39279cc3 | 7513 | }; |
76dda93c | 7514 | |
82d339d9 | 7515 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
7516 | .d_delete = btrfs_dentry_delete, |
7517 | }; |