]>
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 | ||
1ef30be1 JB |
1772 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1773 | if (!ret) { | |
1774 | ret = btrfs_update_inode(trans, root, inode); | |
1775 | BUG_ON(ret); | |
1776 | } | |
1777 | ret = 0; | |
c2167754 | 1778 | out: |
0cb59c99 JB |
1779 | if (nolock) { |
1780 | if (trans) | |
1781 | btrfs_end_transaction_nolock(trans, root); | |
1782 | } else { | |
1783 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); | |
1784 | if (trans) | |
1785 | btrfs_end_transaction(trans, root); | |
1786 | } | |
1787 | ||
e6dcd2dc CM |
1788 | /* once for us */ |
1789 | btrfs_put_ordered_extent(ordered_extent); | |
1790 | /* once for the tree */ | |
1791 | btrfs_put_ordered_extent(ordered_extent); | |
1792 | ||
e6dcd2dc CM |
1793 | return 0; |
1794 | } | |
1795 | ||
b2950863 | 1796 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1797 | struct extent_state *state, int uptodate) |
1798 | { | |
1abe9b8a | 1799 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1800 | ||
8b62b72b | 1801 | ClearPagePrivate2(page); |
211f90e6 CM |
1802 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1803 | } | |
1804 | ||
d352ac68 CM |
1805 | /* |
1806 | * When IO fails, either with EIO or csum verification fails, we | |
1807 | * try other mirrors that might have a good copy of the data. This | |
1808 | * io_failure_record is used to record state as we go through all the | |
1809 | * mirrors. If another mirror has good data, the page is set up to date | |
1810 | * and things continue. If a good mirror can't be found, the original | |
1811 | * bio end_io callback is called to indicate things have failed. | |
1812 | */ | |
7e38326f CM |
1813 | struct io_failure_record { |
1814 | struct page *page; | |
1815 | u64 start; | |
1816 | u64 len; | |
1817 | u64 logical; | |
d20f7043 | 1818 | unsigned long bio_flags; |
7e38326f CM |
1819 | int last_mirror; |
1820 | }; | |
1821 | ||
b2950863 | 1822 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1823 | struct page *page, u64 start, u64 end, |
1824 | struct extent_state *state) | |
7e38326f CM |
1825 | { |
1826 | struct io_failure_record *failrec = NULL; | |
1827 | u64 private; | |
1828 | struct extent_map *em; | |
1829 | struct inode *inode = page->mapping->host; | |
1830 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1831 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1832 | struct bio *bio; |
1833 | int num_copies; | |
1834 | int ret; | |
1259ab75 | 1835 | int rw; |
7e38326f CM |
1836 | u64 logical; |
1837 | ||
1838 | ret = get_state_private(failure_tree, start, &private); | |
1839 | if (ret) { | |
7e38326f CM |
1840 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1841 | if (!failrec) | |
1842 | return -ENOMEM; | |
1843 | failrec->start = start; | |
1844 | failrec->len = end - start + 1; | |
1845 | failrec->last_mirror = 0; | |
d20f7043 | 1846 | failrec->bio_flags = 0; |
7e38326f | 1847 | |
890871be | 1848 | read_lock(&em_tree->lock); |
3b951516 CM |
1849 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1850 | if (em->start > start || em->start + em->len < start) { | |
1851 | free_extent_map(em); | |
1852 | em = NULL; | |
1853 | } | |
890871be | 1854 | read_unlock(&em_tree->lock); |
7e38326f CM |
1855 | |
1856 | if (!em || IS_ERR(em)) { | |
1857 | kfree(failrec); | |
1858 | return -EIO; | |
1859 | } | |
1860 | logical = start - em->start; | |
1861 | logical = em->block_start + logical; | |
d20f7043 CM |
1862 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1863 | logical = em->block_start; | |
1864 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
261507a0 LZ |
1865 | extent_set_compress_type(&failrec->bio_flags, |
1866 | em->compress_type); | |
d20f7043 | 1867 | } |
7e38326f CM |
1868 | failrec->logical = logical; |
1869 | free_extent_map(em); | |
1870 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1871 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1872 | set_state_private(failure_tree, start, |
1873 | (u64)(unsigned long)failrec); | |
7e38326f | 1874 | } else { |
587f7704 | 1875 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1876 | } |
1877 | num_copies = btrfs_num_copies( | |
1878 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1879 | failrec->logical, failrec->len); | |
1880 | failrec->last_mirror++; | |
1881 | if (!state) { | |
cad321ad | 1882 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1883 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1884 | failrec->start, | |
1885 | EXTENT_LOCKED); | |
1886 | if (state && state->start != failrec->start) | |
1887 | state = NULL; | |
cad321ad | 1888 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1889 | } |
1890 | if (!state || failrec->last_mirror > num_copies) { | |
1891 | set_state_private(failure_tree, failrec->start, 0); | |
1892 | clear_extent_bits(failure_tree, failrec->start, | |
1893 | failrec->start + failrec->len - 1, | |
1894 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1895 | kfree(failrec); | |
1896 | return -EIO; | |
1897 | } | |
1898 | bio = bio_alloc(GFP_NOFS, 1); | |
1899 | bio->bi_private = state; | |
1900 | bio->bi_end_io = failed_bio->bi_end_io; | |
1901 | bio->bi_sector = failrec->logical >> 9; | |
1902 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1903 | bio->bi_size = 0; |
d20f7043 | 1904 | |
7e38326f | 1905 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
7b6d91da | 1906 | if (failed_bio->bi_rw & REQ_WRITE) |
1259ab75 CM |
1907 | rw = WRITE; |
1908 | else | |
1909 | rw = READ; | |
1910 | ||
c2db1073 | 1911 | ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, |
c8b97818 | 1912 | failrec->last_mirror, |
eaf25d93 | 1913 | failrec->bio_flags, 0); |
c2db1073 | 1914 | return ret; |
1259ab75 CM |
1915 | } |
1916 | ||
d352ac68 CM |
1917 | /* |
1918 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1919 | * to see if we need to process or clean up an io_failure_record | |
1920 | */ | |
b2950863 | 1921 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1922 | { |
1923 | u64 private; | |
1924 | u64 private_failure; | |
1925 | struct io_failure_record *failure; | |
1926 | int ret; | |
1927 | ||
1928 | private = 0; | |
1929 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
ec29ed5b | 1930 | (u64)-1, 1, EXTENT_DIRTY, 0)) { |
1259ab75 CM |
1931 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, |
1932 | start, &private_failure); | |
1933 | if (ret == 0) { | |
1934 | failure = (struct io_failure_record *)(unsigned long) | |
1935 | private_failure; | |
1936 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1937 | failure->start, 0); | |
1938 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1939 | failure->start, | |
1940 | failure->start + failure->len - 1, | |
1941 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1942 | GFP_NOFS); | |
1943 | kfree(failure); | |
1944 | } | |
1945 | } | |
7e38326f CM |
1946 | return 0; |
1947 | } | |
1948 | ||
d352ac68 CM |
1949 | /* |
1950 | * when reads are done, we need to check csums to verify the data is correct | |
1951 | * if there's a match, we allow the bio to finish. If not, we go through | |
1952 | * the io_failure_record routines to find good copies | |
1953 | */ | |
b2950863 | 1954 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1955 | struct extent_state *state) |
07157aac | 1956 | { |
35ebb934 | 1957 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1958 | struct inode *inode = page->mapping->host; |
d1310b2e | 1959 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1960 | char *kaddr; |
aadfeb6e | 1961 | u64 private = ~(u32)0; |
07157aac | 1962 | int ret; |
ff79f819 CM |
1963 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1964 | u32 csum = ~(u32)0; | |
d1310b2e | 1965 | |
d20f7043 CM |
1966 | if (PageChecked(page)) { |
1967 | ClearPageChecked(page); | |
1968 | goto good; | |
1969 | } | |
6cbff00f CH |
1970 | |
1971 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1972 | return 0; |
1973 | ||
1974 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1975 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1976 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1977 | GFP_NOFS); | |
b6cda9bc | 1978 | return 0; |
17d217fe | 1979 | } |
d20f7043 | 1980 | |
c2e639f0 | 1981 | if (state && state->start == start) { |
70dec807 CM |
1982 | private = state->private; |
1983 | ret = 0; | |
1984 | } else { | |
1985 | ret = get_state_private(io_tree, start, &private); | |
1986 | } | |
9ab86c8e | 1987 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1988 | if (ret) |
07157aac | 1989 | goto zeroit; |
d397712b | 1990 | |
ff79f819 CM |
1991 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1992 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1993 | if (csum != private) |
07157aac | 1994 | goto zeroit; |
d397712b | 1995 | |
9ab86c8e | 1996 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1997 | good: |
7e38326f CM |
1998 | /* if the io failure tree for this inode is non-empty, |
1999 | * check to see if we've recovered from a failed IO | |
2000 | */ | |
1259ab75 | 2001 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
2002 | return 0; |
2003 | ||
2004 | zeroit: | |
193f284d CM |
2005 | if (printk_ratelimit()) { |
2006 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
2007 | "private %llu\n", page->mapping->host->i_ino, | |
2008 | (unsigned long long)start, csum, | |
2009 | (unsigned long long)private); | |
2010 | } | |
db94535d CM |
2011 | memset(kaddr + offset, 1, end - start + 1); |
2012 | flush_dcache_page(page); | |
9ab86c8e | 2013 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2014 | if (private == 0) |
2015 | return 0; | |
7e38326f | 2016 | return -EIO; |
07157aac | 2017 | } |
b888db2b | 2018 | |
24bbcf04 YZ |
2019 | struct delayed_iput { |
2020 | struct list_head list; | |
2021 | struct inode *inode; | |
2022 | }; | |
2023 | ||
2024 | void btrfs_add_delayed_iput(struct inode *inode) | |
2025 | { | |
2026 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2027 | struct delayed_iput *delayed; | |
2028 | ||
2029 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2030 | return; | |
2031 | ||
2032 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2033 | delayed->inode = inode; | |
2034 | ||
2035 | spin_lock(&fs_info->delayed_iput_lock); | |
2036 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2037 | spin_unlock(&fs_info->delayed_iput_lock); | |
2038 | } | |
2039 | ||
2040 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2041 | { | |
2042 | LIST_HEAD(list); | |
2043 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2044 | struct delayed_iput *delayed; | |
2045 | int empty; | |
2046 | ||
2047 | spin_lock(&fs_info->delayed_iput_lock); | |
2048 | empty = list_empty(&fs_info->delayed_iputs); | |
2049 | spin_unlock(&fs_info->delayed_iput_lock); | |
2050 | if (empty) | |
2051 | return; | |
2052 | ||
2053 | down_read(&root->fs_info->cleanup_work_sem); | |
2054 | spin_lock(&fs_info->delayed_iput_lock); | |
2055 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2056 | spin_unlock(&fs_info->delayed_iput_lock); | |
2057 | ||
2058 | while (!list_empty(&list)) { | |
2059 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2060 | list_del(&delayed->list); | |
2061 | iput(delayed->inode); | |
2062 | kfree(delayed); | |
2063 | } | |
2064 | up_read(&root->fs_info->cleanup_work_sem); | |
2065 | } | |
2066 | ||
d68fc57b YZ |
2067 | /* |
2068 | * calculate extra metadata reservation when snapshotting a subvolume | |
2069 | * contains orphan files. | |
2070 | */ | |
2071 | void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans, | |
2072 | struct btrfs_pending_snapshot *pending, | |
2073 | u64 *bytes_to_reserve) | |
2074 | { | |
2075 | struct btrfs_root *root; | |
2076 | struct btrfs_block_rsv *block_rsv; | |
2077 | u64 num_bytes; | |
2078 | int index; | |
2079 | ||
2080 | root = pending->root; | |
2081 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2082 | return; | |
2083 | ||
2084 | block_rsv = root->orphan_block_rsv; | |
2085 | ||
2086 | /* orphan block reservation for the snapshot */ | |
2087 | num_bytes = block_rsv->size; | |
2088 | ||
2089 | /* | |
2090 | * after the snapshot is created, COWing tree blocks may use more | |
2091 | * space than it frees. So we should make sure there is enough | |
2092 | * reserved space. | |
2093 | */ | |
2094 | index = trans->transid & 0x1; | |
2095 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2096 | num_bytes += block_rsv->size - | |
2097 | (block_rsv->reserved + block_rsv->freed[index]); | |
2098 | } | |
2099 | ||
2100 | *bytes_to_reserve += num_bytes; | |
2101 | } | |
2102 | ||
2103 | void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans, | |
2104 | struct btrfs_pending_snapshot *pending) | |
2105 | { | |
2106 | struct btrfs_root *root = pending->root; | |
2107 | struct btrfs_root *snap = pending->snap; | |
2108 | struct btrfs_block_rsv *block_rsv; | |
2109 | u64 num_bytes; | |
2110 | int index; | |
2111 | int ret; | |
2112 | ||
2113 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2114 | return; | |
2115 | ||
2116 | /* refill source subvolume's orphan block reservation */ | |
2117 | block_rsv = root->orphan_block_rsv; | |
2118 | index = trans->transid & 0x1; | |
2119 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2120 | num_bytes = block_rsv->size - | |
2121 | (block_rsv->reserved + block_rsv->freed[index]); | |
2122 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2123 | root->orphan_block_rsv, | |
2124 | num_bytes); | |
2125 | BUG_ON(ret); | |
2126 | } | |
2127 | ||
2128 | /* setup orphan block reservation for the snapshot */ | |
2129 | block_rsv = btrfs_alloc_block_rsv(snap); | |
2130 | BUG_ON(!block_rsv); | |
2131 | ||
2132 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
2133 | snap->orphan_block_rsv = block_rsv; | |
2134 | ||
2135 | num_bytes = root->orphan_block_rsv->size; | |
2136 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2137 | block_rsv, num_bytes); | |
2138 | BUG_ON(ret); | |
2139 | ||
2140 | #if 0 | |
2141 | /* insert orphan item for the snapshot */ | |
2142 | WARN_ON(!root->orphan_item_inserted); | |
2143 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2144 | snap->root_key.objectid); | |
2145 | BUG_ON(ret); | |
2146 | snap->orphan_item_inserted = 1; | |
2147 | #endif | |
2148 | } | |
2149 | ||
2150 | enum btrfs_orphan_cleanup_state { | |
2151 | ORPHAN_CLEANUP_STARTED = 1, | |
2152 | ORPHAN_CLEANUP_DONE = 2, | |
2153 | }; | |
2154 | ||
2155 | /* | |
2156 | * This is called in transaction commmit time. If there are no orphan | |
2157 | * files in the subvolume, it removes orphan item and frees block_rsv | |
2158 | * structure. | |
2159 | */ | |
2160 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2161 | struct btrfs_root *root) | |
2162 | { | |
2163 | int ret; | |
2164 | ||
2165 | if (!list_empty(&root->orphan_list) || | |
2166 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2167 | return; | |
2168 | ||
2169 | if (root->orphan_item_inserted && | |
2170 | btrfs_root_refs(&root->root_item) > 0) { | |
2171 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2172 | root->root_key.objectid); | |
2173 | BUG_ON(ret); | |
2174 | root->orphan_item_inserted = 0; | |
2175 | } | |
2176 | ||
2177 | if (root->orphan_block_rsv) { | |
2178 | WARN_ON(root->orphan_block_rsv->size > 0); | |
2179 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
2180 | root->orphan_block_rsv = NULL; | |
2181 | } | |
2182 | } | |
2183 | ||
7b128766 JB |
2184 | /* |
2185 | * This creates an orphan entry for the given inode in case something goes | |
2186 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2187 | * |
2188 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2189 | * this function. | |
7b128766 JB |
2190 | */ |
2191 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2192 | { | |
2193 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2194 | struct btrfs_block_rsv *block_rsv = NULL; |
2195 | int reserve = 0; | |
2196 | int insert = 0; | |
2197 | int ret; | |
7b128766 | 2198 | |
d68fc57b YZ |
2199 | if (!root->orphan_block_rsv) { |
2200 | block_rsv = btrfs_alloc_block_rsv(root); | |
2201 | BUG_ON(!block_rsv); | |
2202 | } | |
7b128766 | 2203 | |
d68fc57b YZ |
2204 | spin_lock(&root->orphan_lock); |
2205 | if (!root->orphan_block_rsv) { | |
2206 | root->orphan_block_rsv = block_rsv; | |
2207 | } else if (block_rsv) { | |
2208 | btrfs_free_block_rsv(root, block_rsv); | |
2209 | block_rsv = NULL; | |
7b128766 | 2210 | } |
7b128766 | 2211 | |
d68fc57b YZ |
2212 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2213 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2214 | #if 0 | |
2215 | /* | |
2216 | * For proper ENOSPC handling, we should do orphan | |
2217 | * cleanup when mounting. But this introduces backward | |
2218 | * compatibility issue. | |
2219 | */ | |
2220 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2221 | insert = 2; | |
2222 | else | |
2223 | insert = 1; | |
2224 | #endif | |
2225 | insert = 1; | |
7b128766 JB |
2226 | } |
2227 | ||
d68fc57b YZ |
2228 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2229 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2230 | reserve = 1; | |
2231 | } | |
2232 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2233 | |
d68fc57b YZ |
2234 | if (block_rsv) |
2235 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
7b128766 | 2236 | |
d68fc57b YZ |
2237 | /* grab metadata reservation from transaction handle */ |
2238 | if (reserve) { | |
2239 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2240 | BUG_ON(ret); | |
2241 | } | |
7b128766 | 2242 | |
d68fc57b YZ |
2243 | /* insert an orphan item to track this unlinked/truncated file */ |
2244 | if (insert >= 1) { | |
2245 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2246 | BUG_ON(ret); | |
2247 | } | |
2248 | ||
2249 | /* insert an orphan item to track subvolume contains orphan files */ | |
2250 | if (insert >= 2) { | |
2251 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2252 | root->root_key.objectid); | |
2253 | BUG_ON(ret); | |
2254 | } | |
2255 | return 0; | |
7b128766 JB |
2256 | } |
2257 | ||
2258 | /* | |
2259 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2260 | * item for this particular inode. | |
2261 | */ | |
2262 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2263 | { | |
2264 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2265 | int delete_item = 0; |
2266 | int release_rsv = 0; | |
7b128766 JB |
2267 | int ret = 0; |
2268 | ||
d68fc57b YZ |
2269 | spin_lock(&root->orphan_lock); |
2270 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2271 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2272 | delete_item = 1; | |
7b128766 JB |
2273 | } |
2274 | ||
d68fc57b YZ |
2275 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2276 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2277 | release_rsv = 1; | |
7b128766 | 2278 | } |
d68fc57b | 2279 | spin_unlock(&root->orphan_lock); |
7b128766 | 2280 | |
d68fc57b YZ |
2281 | if (trans && delete_item) { |
2282 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2283 | BUG_ON(ret); | |
2284 | } | |
7b128766 | 2285 | |
d68fc57b YZ |
2286 | if (release_rsv) |
2287 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2288 | |
d68fc57b | 2289 | return 0; |
7b128766 JB |
2290 | } |
2291 | ||
2292 | /* | |
2293 | * this cleans up any orphans that may be left on the list from the last use | |
2294 | * of this root. | |
2295 | */ | |
66b4ffd1 | 2296 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2297 | { |
2298 | struct btrfs_path *path; | |
2299 | struct extent_buffer *leaf; | |
7b128766 JB |
2300 | struct btrfs_key key, found_key; |
2301 | struct btrfs_trans_handle *trans; | |
2302 | struct inode *inode; | |
2303 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2304 | ||
d68fc57b | 2305 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2306 | return 0; |
c71bf099 YZ |
2307 | |
2308 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2309 | if (!path) { |
2310 | ret = -ENOMEM; | |
2311 | goto out; | |
2312 | } | |
7b128766 JB |
2313 | path->reada = -1; |
2314 | ||
2315 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2316 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2317 | key.offset = (u64)-1; | |
2318 | ||
7b128766 JB |
2319 | while (1) { |
2320 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2321 | if (ret < 0) |
2322 | goto out; | |
7b128766 JB |
2323 | |
2324 | /* | |
2325 | * if ret == 0 means we found what we were searching for, which | |
2326 | * is weird, but possible, so only screw with path if we didnt | |
2327 | * find the key and see if we have stuff that matches | |
2328 | */ | |
2329 | if (ret > 0) { | |
66b4ffd1 | 2330 | ret = 0; |
7b128766 JB |
2331 | if (path->slots[0] == 0) |
2332 | break; | |
2333 | path->slots[0]--; | |
2334 | } | |
2335 | ||
2336 | /* pull out the item */ | |
2337 | leaf = path->nodes[0]; | |
7b128766 JB |
2338 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2339 | ||
2340 | /* make sure the item matches what we want */ | |
2341 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2342 | break; | |
2343 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2344 | break; | |
2345 | ||
2346 | /* release the path since we're done with it */ | |
2347 | btrfs_release_path(root, path); | |
2348 | ||
2349 | /* | |
2350 | * this is where we are basically btrfs_lookup, without the | |
2351 | * crossing root thing. we store the inode number in the | |
2352 | * offset of the orphan item. | |
2353 | */ | |
5d4f98a2 YZ |
2354 | found_key.objectid = found_key.offset; |
2355 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2356 | found_key.offset = 0; | |
73f73415 | 2357 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
66b4ffd1 JB |
2358 | if (IS_ERR(inode)) { |
2359 | ret = PTR_ERR(inode); | |
2360 | goto out; | |
2361 | } | |
7b128766 | 2362 | |
7b128766 JB |
2363 | /* |
2364 | * add this inode to the orphan list so btrfs_orphan_del does | |
2365 | * the proper thing when we hit it | |
2366 | */ | |
d68fc57b | 2367 | spin_lock(&root->orphan_lock); |
7b128766 | 2368 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
d68fc57b | 2369 | spin_unlock(&root->orphan_lock); |
7b128766 JB |
2370 | |
2371 | /* | |
2372 | * if this is a bad inode, means we actually succeeded in | |
2373 | * removing the inode, but not the orphan record, which means | |
2374 | * we need to manually delete the orphan since iput will just | |
2375 | * do a destroy_inode | |
2376 | */ | |
2377 | if (is_bad_inode(inode)) { | |
a22285a6 | 2378 | trans = btrfs_start_transaction(root, 0); |
66b4ffd1 JB |
2379 | if (IS_ERR(trans)) { |
2380 | ret = PTR_ERR(trans); | |
2381 | goto out; | |
2382 | } | |
7b128766 | 2383 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2384 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2385 | iput(inode); |
2386 | continue; | |
2387 | } | |
2388 | ||
2389 | /* if we have links, this was a truncate, lets do that */ | |
2390 | if (inode->i_nlink) { | |
a41ad394 JB |
2391 | if (!S_ISREG(inode->i_mode)) { |
2392 | WARN_ON(1); | |
2393 | iput(inode); | |
2394 | continue; | |
2395 | } | |
7b128766 | 2396 | nr_truncate++; |
66b4ffd1 | 2397 | ret = btrfs_truncate(inode); |
7b128766 JB |
2398 | } else { |
2399 | nr_unlink++; | |
2400 | } | |
2401 | ||
2402 | /* this will do delete_inode and everything for us */ | |
2403 | iput(inode); | |
66b4ffd1 JB |
2404 | if (ret) |
2405 | goto out; | |
7b128766 | 2406 | } |
d68fc57b YZ |
2407 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2408 | ||
2409 | if (root->orphan_block_rsv) | |
2410 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2411 | (u64)-1); | |
2412 | ||
2413 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
2414 | trans = btrfs_join_transaction(root, 1); | |
66b4ffd1 JB |
2415 | if (!IS_ERR(trans)) |
2416 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2417 | } |
7b128766 JB |
2418 | |
2419 | if (nr_unlink) | |
2420 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2421 | if (nr_truncate) | |
2422 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2423 | |
2424 | out: | |
2425 | if (ret) | |
2426 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2427 | btrfs_free_path(path); | |
2428 | return ret; | |
7b128766 JB |
2429 | } |
2430 | ||
46a53cca CM |
2431 | /* |
2432 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2433 | * don't find any xattrs, we know there can't be any acls. | |
2434 | * | |
2435 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2436 | */ | |
2437 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2438 | int slot, u64 objectid) | |
2439 | { | |
2440 | u32 nritems = btrfs_header_nritems(leaf); | |
2441 | struct btrfs_key found_key; | |
2442 | int scanned = 0; | |
2443 | ||
2444 | slot++; | |
2445 | while (slot < nritems) { | |
2446 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2447 | ||
2448 | /* we found a different objectid, there must not be acls */ | |
2449 | if (found_key.objectid != objectid) | |
2450 | return 0; | |
2451 | ||
2452 | /* we found an xattr, assume we've got an acl */ | |
2453 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2454 | return 1; | |
2455 | ||
2456 | /* | |
2457 | * we found a key greater than an xattr key, there can't | |
2458 | * be any acls later on | |
2459 | */ | |
2460 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2461 | return 0; | |
2462 | ||
2463 | slot++; | |
2464 | scanned++; | |
2465 | ||
2466 | /* | |
2467 | * it goes inode, inode backrefs, xattrs, extents, | |
2468 | * so if there are a ton of hard links to an inode there can | |
2469 | * be a lot of backrefs. Don't waste time searching too hard, | |
2470 | * this is just an optimization | |
2471 | */ | |
2472 | if (scanned >= 8) | |
2473 | break; | |
2474 | } | |
2475 | /* we hit the end of the leaf before we found an xattr or | |
2476 | * something larger than an xattr. We have to assume the inode | |
2477 | * has acls | |
2478 | */ | |
2479 | return 1; | |
2480 | } | |
2481 | ||
d352ac68 CM |
2482 | /* |
2483 | * read an inode from the btree into the in-memory inode | |
2484 | */ | |
5d4f98a2 | 2485 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2486 | { |
2487 | struct btrfs_path *path; | |
5f39d397 | 2488 | struct extent_buffer *leaf; |
39279cc3 | 2489 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2490 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2491 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2492 | struct btrfs_key location; | |
46a53cca | 2493 | int maybe_acls; |
39279cc3 | 2494 | u64 alloc_group_block; |
618e21d5 | 2495 | u32 rdev; |
39279cc3 CM |
2496 | int ret; |
2497 | ||
2498 | path = btrfs_alloc_path(); | |
2499 | BUG_ON(!path); | |
39279cc3 | 2500 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2501 | |
39279cc3 | 2502 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2503 | if (ret) |
39279cc3 | 2504 | goto make_bad; |
39279cc3 | 2505 | |
5f39d397 CM |
2506 | leaf = path->nodes[0]; |
2507 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2508 | struct btrfs_inode_item); | |
2509 | ||
2510 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2511 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2512 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2513 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2514 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2515 | |
2516 | tspec = btrfs_inode_atime(inode_item); | |
2517 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2518 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2519 | ||
2520 | tspec = btrfs_inode_mtime(inode_item); | |
2521 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2522 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2523 | ||
2524 | tspec = btrfs_inode_ctime(inode_item); | |
2525 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2526 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2527 | ||
a76a3cd4 | 2528 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2529 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2530 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2531 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2532 | inode->i_rdev = 0; |
5f39d397 CM |
2533 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2534 | ||
aec7477b | 2535 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2536 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2537 | |
5f39d397 | 2538 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2539 | |
46a53cca CM |
2540 | /* |
2541 | * try to precache a NULL acl entry for files that don't have | |
2542 | * any xattrs or acls | |
2543 | */ | |
2544 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2545 | if (!maybe_acls) |
2546 | cache_no_acl(inode); | |
46a53cca | 2547 | |
d2fb3437 YZ |
2548 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2549 | alloc_group_block, 0); | |
39279cc3 CM |
2550 | btrfs_free_path(path); |
2551 | inode_item = NULL; | |
2552 | ||
39279cc3 | 2553 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2554 | case S_IFREG: |
2555 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2556 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2557 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2558 | inode->i_fop = &btrfs_file_operations; |
2559 | inode->i_op = &btrfs_file_inode_operations; | |
2560 | break; | |
2561 | case S_IFDIR: | |
2562 | inode->i_fop = &btrfs_dir_file_operations; | |
2563 | if (root == root->fs_info->tree_root) | |
2564 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2565 | else | |
2566 | inode->i_op = &btrfs_dir_inode_operations; | |
2567 | break; | |
2568 | case S_IFLNK: | |
2569 | inode->i_op = &btrfs_symlink_inode_operations; | |
2570 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2571 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2572 | break; |
618e21d5 | 2573 | default: |
0279b4cd | 2574 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2575 | init_special_inode(inode, inode->i_mode, rdev); |
2576 | break; | |
39279cc3 | 2577 | } |
6cbff00f CH |
2578 | |
2579 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2580 | return; |
2581 | ||
2582 | make_bad: | |
39279cc3 | 2583 | btrfs_free_path(path); |
39279cc3 CM |
2584 | make_bad_inode(inode); |
2585 | } | |
2586 | ||
d352ac68 CM |
2587 | /* |
2588 | * given a leaf and an inode, copy the inode fields into the leaf | |
2589 | */ | |
e02119d5 CM |
2590 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2591 | struct extent_buffer *leaf, | |
5f39d397 | 2592 | struct btrfs_inode_item *item, |
39279cc3 CM |
2593 | struct inode *inode) |
2594 | { | |
12ddb96c JB |
2595 | if (!leaf->map_token) |
2596 | map_private_extent_buffer(leaf, (unsigned long)item, | |
2597 | sizeof(struct btrfs_inode_item), | |
2598 | &leaf->map_token, &leaf->kaddr, | |
2599 | &leaf->map_start, &leaf->map_len, | |
2600 | KM_USER1); | |
2601 | ||
5f39d397 CM |
2602 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2603 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2604 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2605 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2606 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2607 | ||
2608 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2609 | inode->i_atime.tv_sec); | |
2610 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2611 | inode->i_atime.tv_nsec); | |
2612 | ||
2613 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2614 | inode->i_mtime.tv_sec); | |
2615 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2616 | inode->i_mtime.tv_nsec); | |
2617 | ||
2618 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2619 | inode->i_ctime.tv_sec); | |
2620 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2621 | inode->i_ctime.tv_nsec); | |
2622 | ||
a76a3cd4 | 2623 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2624 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2625 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2626 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2627 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2628 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2629 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
12ddb96c JB |
2630 | |
2631 | if (leaf->map_token) { | |
2632 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2633 | leaf->map_token = NULL; | |
2634 | } | |
39279cc3 CM |
2635 | } |
2636 | ||
d352ac68 CM |
2637 | /* |
2638 | * copy everything in the in-memory inode into the btree. | |
2639 | */ | |
d397712b CM |
2640 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2641 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2642 | { |
2643 | struct btrfs_inode_item *inode_item; | |
2644 | struct btrfs_path *path; | |
5f39d397 | 2645 | struct extent_buffer *leaf; |
39279cc3 CM |
2646 | int ret; |
2647 | ||
2648 | path = btrfs_alloc_path(); | |
2649 | BUG_ON(!path); | |
b9473439 | 2650 | path->leave_spinning = 1; |
39279cc3 CM |
2651 | ret = btrfs_lookup_inode(trans, root, path, |
2652 | &BTRFS_I(inode)->location, 1); | |
2653 | if (ret) { | |
2654 | if (ret > 0) | |
2655 | ret = -ENOENT; | |
2656 | goto failed; | |
2657 | } | |
2658 | ||
b4ce94de | 2659 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2660 | leaf = path->nodes[0]; |
2661 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2662 | struct btrfs_inode_item); |
2663 | ||
e02119d5 | 2664 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2665 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2666 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2667 | ret = 0; |
2668 | failed: | |
39279cc3 CM |
2669 | btrfs_free_path(path); |
2670 | return ret; | |
2671 | } | |
2672 | ||
2673 | ||
d352ac68 CM |
2674 | /* |
2675 | * unlink helper that gets used here in inode.c and in the tree logging | |
2676 | * recovery code. It remove a link in a directory with a given name, and | |
2677 | * also drops the back refs in the inode to the directory | |
2678 | */ | |
92986796 AV |
2679 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2680 | struct btrfs_root *root, | |
2681 | struct inode *dir, struct inode *inode, | |
2682 | const char *name, int name_len) | |
39279cc3 CM |
2683 | { |
2684 | struct btrfs_path *path; | |
39279cc3 | 2685 | int ret = 0; |
5f39d397 | 2686 | struct extent_buffer *leaf; |
39279cc3 | 2687 | struct btrfs_dir_item *di; |
5f39d397 | 2688 | struct btrfs_key key; |
aec7477b | 2689 | u64 index; |
39279cc3 CM |
2690 | |
2691 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2692 | if (!path) { |
2693 | ret = -ENOMEM; | |
554233a6 | 2694 | goto out; |
54aa1f4d CM |
2695 | } |
2696 | ||
b9473439 | 2697 | path->leave_spinning = 1; |
39279cc3 CM |
2698 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2699 | name, name_len, -1); | |
2700 | if (IS_ERR(di)) { | |
2701 | ret = PTR_ERR(di); | |
2702 | goto err; | |
2703 | } | |
2704 | if (!di) { | |
2705 | ret = -ENOENT; | |
2706 | goto err; | |
2707 | } | |
5f39d397 CM |
2708 | leaf = path->nodes[0]; |
2709 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2710 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2711 | if (ret) |
2712 | goto err; | |
39279cc3 CM |
2713 | btrfs_release_path(root, path); |
2714 | ||
aec7477b | 2715 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2716 | inode->i_ino, |
2717 | dir->i_ino, &index); | |
aec7477b | 2718 | if (ret) { |
d397712b | 2719 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2720 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2721 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2722 | goto err; |
2723 | } | |
2724 | ||
39279cc3 | 2725 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2726 | index, name, name_len, -1); |
39279cc3 CM |
2727 | if (IS_ERR(di)) { |
2728 | ret = PTR_ERR(di); | |
2729 | goto err; | |
2730 | } | |
2731 | if (!di) { | |
2732 | ret = -ENOENT; | |
2733 | goto err; | |
2734 | } | |
2735 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2736 | btrfs_release_path(root, path); |
39279cc3 | 2737 | |
e02119d5 CM |
2738 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2739 | inode, dir->i_ino); | |
49eb7e46 | 2740 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2741 | |
2742 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2743 | dir, index); | |
6418c961 CM |
2744 | if (ret == -ENOENT) |
2745 | ret = 0; | |
39279cc3 CM |
2746 | err: |
2747 | btrfs_free_path(path); | |
e02119d5 CM |
2748 | if (ret) |
2749 | goto out; | |
2750 | ||
2751 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2752 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2753 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2754 | out: |
39279cc3 CM |
2755 | return ret; |
2756 | } | |
2757 | ||
92986796 AV |
2758 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2759 | struct btrfs_root *root, | |
2760 | struct inode *dir, struct inode *inode, | |
2761 | const char *name, int name_len) | |
2762 | { | |
2763 | int ret; | |
2764 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2765 | if (!ret) { | |
2766 | btrfs_drop_nlink(inode); | |
2767 | ret = btrfs_update_inode(trans, root, inode); | |
2768 | } | |
2769 | return ret; | |
2770 | } | |
2771 | ||
2772 | ||
a22285a6 YZ |
2773 | /* helper to check if there is any shared block in the path */ |
2774 | static int check_path_shared(struct btrfs_root *root, | |
2775 | struct btrfs_path *path) | |
39279cc3 | 2776 | { |
a22285a6 YZ |
2777 | struct extent_buffer *eb; |
2778 | int level; | |
0e4dcbef | 2779 | u64 refs = 1; |
5df6a9f6 | 2780 | |
a22285a6 | 2781 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2782 | int ret; |
2783 | ||
a22285a6 YZ |
2784 | if (!path->nodes[level]) |
2785 | break; | |
2786 | eb = path->nodes[level]; | |
2787 | if (!btrfs_block_can_be_shared(root, eb)) | |
2788 | continue; | |
2789 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2790 | &refs, NULL); | |
2791 | if (refs > 1) | |
2792 | return 1; | |
5df6a9f6 | 2793 | } |
dedefd72 | 2794 | return 0; |
39279cc3 CM |
2795 | } |
2796 | ||
a22285a6 YZ |
2797 | /* |
2798 | * helper to start transaction for unlink and rmdir. | |
2799 | * | |
2800 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2801 | * so in enospc case, we should make sure they will free space before | |
2802 | * allowing them to use the global metadata reservation. | |
2803 | */ | |
2804 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2805 | struct dentry *dentry) | |
4df27c4d | 2806 | { |
39279cc3 | 2807 | struct btrfs_trans_handle *trans; |
a22285a6 | 2808 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2809 | struct btrfs_path *path; |
a22285a6 | 2810 | struct btrfs_inode_ref *ref; |
4df27c4d | 2811 | struct btrfs_dir_item *di; |
7b128766 | 2812 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2813 | u64 index; |
a22285a6 YZ |
2814 | int check_link = 1; |
2815 | int err = -ENOSPC; | |
4df27c4d YZ |
2816 | int ret; |
2817 | ||
a22285a6 YZ |
2818 | trans = btrfs_start_transaction(root, 10); |
2819 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
2820 | return trans; | |
4df27c4d | 2821 | |
a22285a6 YZ |
2822 | if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
2823 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2824 | |
a22285a6 YZ |
2825 | /* check if there is someone else holds reference */ |
2826 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2827 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2828 | |
a22285a6 YZ |
2829 | if (atomic_read(&inode->i_count) > 2) |
2830 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2831 | |
a22285a6 YZ |
2832 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2833 | return ERR_PTR(-ENOSPC); | |
2834 | ||
2835 | path = btrfs_alloc_path(); | |
2836 | if (!path) { | |
2837 | root->fs_info->enospc_unlink = 0; | |
2838 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2839 | } |
2840 | ||
a22285a6 | 2841 | trans = btrfs_start_transaction(root, 0); |
5df6a9f6 | 2842 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2843 | btrfs_free_path(path); |
2844 | root->fs_info->enospc_unlink = 0; | |
2845 | return trans; | |
2846 | } | |
4df27c4d | 2847 | |
a22285a6 YZ |
2848 | path->skip_locking = 1; |
2849 | path->search_commit_root = 1; | |
4df27c4d | 2850 | |
a22285a6 YZ |
2851 | ret = btrfs_lookup_inode(trans, root, path, |
2852 | &BTRFS_I(dir)->location, 0); | |
2853 | if (ret < 0) { | |
2854 | err = ret; | |
2855 | goto out; | |
2856 | } | |
2857 | if (ret == 0) { | |
2858 | if (check_path_shared(root, path)) | |
2859 | goto out; | |
2860 | } else { | |
2861 | check_link = 0; | |
5df6a9f6 | 2862 | } |
a22285a6 YZ |
2863 | btrfs_release_path(root, path); |
2864 | ||
2865 | ret = btrfs_lookup_inode(trans, root, path, | |
2866 | &BTRFS_I(inode)->location, 0); | |
2867 | if (ret < 0) { | |
2868 | err = ret; | |
2869 | goto out; | |
2870 | } | |
2871 | if (ret == 0) { | |
2872 | if (check_path_shared(root, path)) | |
2873 | goto out; | |
2874 | } else { | |
2875 | check_link = 0; | |
2876 | } | |
2877 | btrfs_release_path(root, path); | |
2878 | ||
2879 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2880 | ret = btrfs_lookup_file_extent(trans, root, path, | |
2881 | inode->i_ino, (u64)-1, 0); | |
2882 | if (ret < 0) { | |
2883 | err = ret; | |
2884 | goto out; | |
2885 | } | |
2886 | BUG_ON(ret == 0); | |
2887 | if (check_path_shared(root, path)) | |
2888 | goto out; | |
2889 | btrfs_release_path(root, path); | |
2890 | } | |
2891 | ||
2892 | if (!check_link) { | |
2893 | err = 0; | |
2894 | goto out; | |
2895 | } | |
2896 | ||
2897 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2898 | dentry->d_name.name, dentry->d_name.len, 0); | |
2899 | if (IS_ERR(di)) { | |
2900 | err = PTR_ERR(di); | |
2901 | goto out; | |
2902 | } | |
2903 | if (di) { | |
2904 | if (check_path_shared(root, path)) | |
2905 | goto out; | |
2906 | } else { | |
2907 | err = 0; | |
2908 | goto out; | |
2909 | } | |
2910 | btrfs_release_path(root, path); | |
2911 | ||
2912 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2913 | dentry->d_name.name, dentry->d_name.len, | |
2914 | inode->i_ino, dir->i_ino, 0); | |
2915 | if (IS_ERR(ref)) { | |
2916 | err = PTR_ERR(ref); | |
2917 | goto out; | |
2918 | } | |
2919 | BUG_ON(!ref); | |
2920 | if (check_path_shared(root, path)) | |
2921 | goto out; | |
2922 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
2923 | btrfs_release_path(root, path); | |
2924 | ||
2925 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index, | |
2926 | dentry->d_name.name, dentry->d_name.len, 0); | |
2927 | if (IS_ERR(di)) { | |
2928 | err = PTR_ERR(di); | |
2929 | goto out; | |
2930 | } | |
2931 | BUG_ON(ret == -ENOENT); | |
2932 | if (check_path_shared(root, path)) | |
2933 | goto out; | |
2934 | ||
2935 | err = 0; | |
2936 | out: | |
2937 | btrfs_free_path(path); | |
2938 | if (err) { | |
2939 | btrfs_end_transaction(trans, root); | |
2940 | root->fs_info->enospc_unlink = 0; | |
2941 | return ERR_PTR(err); | |
2942 | } | |
2943 | ||
2944 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2945 | return trans; | |
2946 | } | |
2947 | ||
2948 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2949 | struct btrfs_root *root) | |
2950 | { | |
2951 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
2952 | BUG_ON(!root->fs_info->enospc_unlink); | |
2953 | root->fs_info->enospc_unlink = 0; | |
2954 | } | |
2955 | btrfs_end_transaction_throttle(trans, root); | |
2956 | } | |
2957 | ||
2958 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2959 | { | |
2960 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2961 | struct btrfs_trans_handle *trans; | |
2962 | struct inode *inode = dentry->d_inode; | |
2963 | int ret; | |
2964 | unsigned long nr = 0; | |
2965 | ||
2966 | trans = __unlink_start_trans(dir, dentry); | |
2967 | if (IS_ERR(trans)) | |
2968 | return PTR_ERR(trans); | |
5f39d397 | 2969 | |
39279cc3 | 2970 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2971 | |
2972 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2973 | ||
e02119d5 CM |
2974 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2975 | dentry->d_name.name, dentry->d_name.len); | |
a22285a6 | 2976 | BUG_ON(ret); |
7b128766 | 2977 | |
a22285a6 | 2978 | if (inode->i_nlink == 0) { |
7b128766 | 2979 | ret = btrfs_orphan_add(trans, inode); |
a22285a6 YZ |
2980 | BUG_ON(ret); |
2981 | } | |
7b128766 | 2982 | |
d3c2fdcf | 2983 | nr = trans->blocks_used; |
a22285a6 | 2984 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2985 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2986 | return ret; |
2987 | } | |
2988 | ||
4df27c4d YZ |
2989 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2990 | struct btrfs_root *root, | |
2991 | struct inode *dir, u64 objectid, | |
2992 | const char *name, int name_len) | |
2993 | { | |
2994 | struct btrfs_path *path; | |
2995 | struct extent_buffer *leaf; | |
2996 | struct btrfs_dir_item *di; | |
2997 | struct btrfs_key key; | |
2998 | u64 index; | |
2999 | int ret; | |
3000 | ||
3001 | path = btrfs_alloc_path(); | |
3002 | if (!path) | |
3003 | return -ENOMEM; | |
3004 | ||
3005 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
3006 | name, name_len, -1); | |
3007 | BUG_ON(!di || IS_ERR(di)); | |
3008 | ||
3009 | leaf = path->nodes[0]; | |
3010 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3011 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3012 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3013 | BUG_ON(ret); | |
3014 | btrfs_release_path(root, path); | |
3015 | ||
3016 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3017 | objectid, root->root_key.objectid, | |
3018 | dir->i_ino, &index, name, name_len); | |
3019 | if (ret < 0) { | |
3020 | BUG_ON(ret != -ENOENT); | |
3021 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
3022 | name, name_len); | |
3023 | BUG_ON(!di || IS_ERR(di)); | |
3024 | ||
3025 | leaf = path->nodes[0]; | |
3026 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3027 | btrfs_release_path(root, path); | |
3028 | index = key.offset; | |
3029 | } | |
3030 | ||
3031 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
3032 | index, name, name_len, -1); | |
3033 | BUG_ON(!di || IS_ERR(di)); | |
3034 | ||
3035 | leaf = path->nodes[0]; | |
3036 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3037 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3038 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3039 | BUG_ON(ret); | |
3040 | btrfs_release_path(root, path); | |
3041 | ||
3042 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
3043 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
3044 | ret = btrfs_update_inode(trans, root, dir); | |
3045 | BUG_ON(ret); | |
4df27c4d YZ |
3046 | |
3047 | btrfs_free_path(path); | |
3048 | return 0; | |
3049 | } | |
3050 | ||
39279cc3 CM |
3051 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3052 | { | |
3053 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3054 | int err = 0; |
39279cc3 | 3055 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3056 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3057 | unsigned long nr = 0; |
39279cc3 | 3058 | |
3394e160 | 3059 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 3060 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3061 | return -ENOTEMPTY; |
3062 | ||
a22285a6 YZ |
3063 | trans = __unlink_start_trans(dir, dentry); |
3064 | if (IS_ERR(trans)) | |
5df6a9f6 | 3065 | return PTR_ERR(trans); |
5df6a9f6 | 3066 | |
39279cc3 | 3067 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 3068 | |
4df27c4d YZ |
3069 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
3070 | err = btrfs_unlink_subvol(trans, root, dir, | |
3071 | BTRFS_I(inode)->location.objectid, | |
3072 | dentry->d_name.name, | |
3073 | dentry->d_name.len); | |
3074 | goto out; | |
3075 | } | |
3076 | ||
7b128766 JB |
3077 | err = btrfs_orphan_add(trans, inode); |
3078 | if (err) | |
4df27c4d | 3079 | goto out; |
7b128766 | 3080 | |
39279cc3 | 3081 | /* now the directory is empty */ |
e02119d5 CM |
3082 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3083 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3084 | if (!err) |
dbe674a9 | 3085 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3086 | out: |
d3c2fdcf | 3087 | nr = trans->blocks_used; |
a22285a6 | 3088 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3089 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3090 | |
39279cc3 CM |
3091 | return err; |
3092 | } | |
3093 | ||
d20f7043 | 3094 | #if 0 |
323ac95b CM |
3095 | /* |
3096 | * when truncating bytes in a file, it is possible to avoid reading | |
3097 | * the leaves that contain only checksum items. This can be the | |
3098 | * majority of the IO required to delete a large file, but it must | |
3099 | * be done carefully. | |
3100 | * | |
3101 | * The keys in the level just above the leaves are checked to make sure | |
3102 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
3103 | * after the new size. | |
3104 | * | |
3105 | * Then the key for the next leaf is checked to make sure it also has | |
3106 | * a checksum item for the same file. If it does, we know our target leaf | |
3107 | * contains only checksum items, and it can be safely freed without reading | |
3108 | * it. | |
3109 | * | |
3110 | * This is just an optimization targeted at large files. It may do | |
3111 | * nothing. It will return 0 unless things went badly. | |
3112 | */ | |
3113 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
3114 | struct btrfs_root *root, | |
3115 | struct btrfs_path *path, | |
3116 | struct inode *inode, u64 new_size) | |
3117 | { | |
3118 | struct btrfs_key key; | |
3119 | int ret; | |
3120 | int nritems; | |
3121 | struct btrfs_key found_key; | |
3122 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
3123 | struct btrfs_leaf_ref *ref; |
3124 | u64 leaf_gen; | |
3125 | u64 leaf_start; | |
323ac95b CM |
3126 | |
3127 | path->lowest_level = 1; | |
3128 | key.objectid = inode->i_ino; | |
3129 | key.type = BTRFS_CSUM_ITEM_KEY; | |
3130 | key.offset = new_size; | |
3131 | again: | |
3132 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3133 | if (ret < 0) | |
3134 | goto out; | |
3135 | ||
3136 | if (path->nodes[1] == NULL) { | |
3137 | ret = 0; | |
3138 | goto out; | |
3139 | } | |
3140 | ret = 0; | |
3141 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
3142 | nritems = btrfs_header_nritems(path->nodes[1]); | |
3143 | ||
3144 | if (!nritems) | |
3145 | goto out; | |
3146 | ||
3147 | if (path->slots[1] >= nritems) | |
3148 | goto next_node; | |
3149 | ||
3150 | /* did we find a key greater than anything we want to delete? */ | |
3151 | if (found_key.objectid > inode->i_ino || | |
3152 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
3153 | goto out; | |
3154 | ||
3155 | /* we check the next key in the node to make sure the leave contains | |
3156 | * only checksum items. This comparison doesn't work if our | |
3157 | * leaf is the last one in the node | |
3158 | */ | |
3159 | if (path->slots[1] + 1 >= nritems) { | |
3160 | next_node: | |
3161 | /* search forward from the last key in the node, this | |
3162 | * will bring us into the next node in the tree | |
3163 | */ | |
3164 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
3165 | ||
3166 | /* unlikely, but we inc below, so check to be safe */ | |
3167 | if (found_key.offset == (u64)-1) | |
3168 | goto out; | |
3169 | ||
3170 | /* search_forward needs a path with locks held, do the | |
3171 | * search again for the original key. It is possible | |
3172 | * this will race with a balance and return a path that | |
3173 | * we could modify, but this drop is just an optimization | |
3174 | * and is allowed to miss some leaves. | |
3175 | */ | |
3176 | btrfs_release_path(root, path); | |
3177 | found_key.offset++; | |
3178 | ||
3179 | /* setup a max key for search_forward */ | |
3180 | other_key.offset = (u64)-1; | |
3181 | other_key.type = key.type; | |
3182 | other_key.objectid = key.objectid; | |
3183 | ||
3184 | path->keep_locks = 1; | |
3185 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
3186 | path, 0, 0); | |
3187 | path->keep_locks = 0; | |
3188 | if (ret || found_key.objectid != key.objectid || | |
3189 | found_key.type != key.type) { | |
3190 | ret = 0; | |
3191 | goto out; | |
3192 | } | |
3193 | ||
3194 | key.offset = found_key.offset; | |
3195 | btrfs_release_path(root, path); | |
3196 | cond_resched(); | |
3197 | goto again; | |
3198 | } | |
3199 | ||
3200 | /* we know there's one more slot after us in the tree, | |
3201 | * read that key so we can verify it is also a checksum item | |
3202 | */ | |
3203 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
3204 | ||
3205 | if (found_key.objectid < inode->i_ino) | |
3206 | goto next_key; | |
3207 | ||
3208 | if (found_key.type != key.type || found_key.offset < new_size) | |
3209 | goto next_key; | |
3210 | ||
3211 | /* | |
3212 | * if the key for the next leaf isn't a csum key from this objectid, | |
3213 | * we can't be sure there aren't good items inside this leaf. | |
3214 | * Bail out | |
3215 | */ | |
3216 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
3217 | goto out; | |
3218 | ||
5b84e8d6 YZ |
3219 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
3220 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
3221 | /* |
3222 | * it is safe to delete this leaf, it contains only | |
3223 | * csum items from this inode at an offset >= new_size | |
3224 | */ | |
5b84e8d6 | 3225 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
3226 | BUG_ON(ret); |
3227 | ||
5b84e8d6 YZ |
3228 | if (root->ref_cows && leaf_gen < trans->transid) { |
3229 | ref = btrfs_alloc_leaf_ref(root, 0); | |
3230 | if (ref) { | |
3231 | ref->root_gen = root->root_key.offset; | |
3232 | ref->bytenr = leaf_start; | |
3233 | ref->owner = 0; | |
3234 | ref->generation = leaf_gen; | |
3235 | ref->nritems = 0; | |
3236 | ||
bd56b302 CM |
3237 | btrfs_sort_leaf_ref(ref); |
3238 | ||
5b84e8d6 YZ |
3239 | ret = btrfs_add_leaf_ref(root, ref, 0); |
3240 | WARN_ON(ret); | |
3241 | btrfs_free_leaf_ref(root, ref); | |
3242 | } else { | |
3243 | WARN_ON(1); | |
3244 | } | |
3245 | } | |
323ac95b CM |
3246 | next_key: |
3247 | btrfs_release_path(root, path); | |
3248 | ||
3249 | if (other_key.objectid == inode->i_ino && | |
3250 | other_key.type == key.type && other_key.offset > key.offset) { | |
3251 | key.offset = other_key.offset; | |
3252 | cond_resched(); | |
3253 | goto again; | |
3254 | } | |
3255 | ret = 0; | |
3256 | out: | |
3257 | /* fixup any changes we've made to the path */ | |
3258 | path->lowest_level = 0; | |
3259 | path->keep_locks = 0; | |
3260 | btrfs_release_path(root, path); | |
3261 | return ret; | |
3262 | } | |
3263 | ||
d20f7043 CM |
3264 | #endif |
3265 | ||
39279cc3 CM |
3266 | /* |
3267 | * this can truncate away extent items, csum items and directory items. | |
3268 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3269 | * any higher than new_size |
39279cc3 CM |
3270 | * |
3271 | * csum items that cross the new i_size are truncated to the new size | |
3272 | * as well. | |
7b128766 JB |
3273 | * |
3274 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3275 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3276 | */ |
8082510e YZ |
3277 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3278 | struct btrfs_root *root, | |
3279 | struct inode *inode, | |
3280 | u64 new_size, u32 min_type) | |
39279cc3 | 3281 | { |
39279cc3 | 3282 | struct btrfs_path *path; |
5f39d397 | 3283 | struct extent_buffer *leaf; |
39279cc3 | 3284 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3285 | struct btrfs_key key; |
3286 | struct btrfs_key found_key; | |
39279cc3 | 3287 | u64 extent_start = 0; |
db94535d | 3288 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3289 | u64 extent_offset = 0; |
39279cc3 | 3290 | u64 item_end = 0; |
8082510e YZ |
3291 | u64 mask = root->sectorsize - 1; |
3292 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3293 | int found_extent; |
3294 | int del_item; | |
85e21bac CM |
3295 | int pending_del_nr = 0; |
3296 | int pending_del_slot = 0; | |
179e29e4 | 3297 | int extent_type = -1; |
771ed689 | 3298 | int encoding; |
8082510e YZ |
3299 | int ret; |
3300 | int err = 0; | |
3301 | ||
3302 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3303 | |
0af3d00b | 3304 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3305 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3306 | |
39279cc3 CM |
3307 | path = btrfs_alloc_path(); |
3308 | BUG_ON(!path); | |
33c17ad5 | 3309 | path->reada = -1; |
5f39d397 | 3310 | |
39279cc3 CM |
3311 | key.objectid = inode->i_ino; |
3312 | key.offset = (u64)-1; | |
5f39d397 CM |
3313 | key.type = (u8)-1; |
3314 | ||
85e21bac | 3315 | search_again: |
b9473439 | 3316 | path->leave_spinning = 1; |
85e21bac | 3317 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3318 | if (ret < 0) { |
3319 | err = ret; | |
3320 | goto out; | |
3321 | } | |
d397712b | 3322 | |
85e21bac | 3323 | if (ret > 0) { |
e02119d5 CM |
3324 | /* there are no items in the tree for us to truncate, we're |
3325 | * done | |
3326 | */ | |
8082510e YZ |
3327 | if (path->slots[0] == 0) |
3328 | goto out; | |
85e21bac CM |
3329 | path->slots[0]--; |
3330 | } | |
3331 | ||
d397712b | 3332 | while (1) { |
39279cc3 | 3333 | fi = NULL; |
5f39d397 CM |
3334 | leaf = path->nodes[0]; |
3335 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3336 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 3337 | encoding = 0; |
39279cc3 | 3338 | |
5f39d397 | 3339 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 3340 | break; |
5f39d397 | 3341 | |
85e21bac | 3342 | if (found_type < min_type) |
39279cc3 CM |
3343 | break; |
3344 | ||
5f39d397 | 3345 | item_end = found_key.offset; |
39279cc3 | 3346 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3347 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3348 | struct btrfs_file_extent_item); |
179e29e4 | 3349 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3350 | encoding = btrfs_file_extent_compression(leaf, fi); |
3351 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3352 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3353 | ||
179e29e4 | 3354 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3355 | item_end += |
db94535d | 3356 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3357 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3358 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3359 | fi); |
39279cc3 | 3360 | } |
008630c1 | 3361 | item_end--; |
39279cc3 | 3362 | } |
8082510e YZ |
3363 | if (found_type > min_type) { |
3364 | del_item = 1; | |
3365 | } else { | |
3366 | if (item_end < new_size) | |
b888db2b | 3367 | break; |
8082510e YZ |
3368 | if (found_key.offset >= new_size) |
3369 | del_item = 1; | |
3370 | else | |
3371 | del_item = 0; | |
39279cc3 | 3372 | } |
39279cc3 | 3373 | found_extent = 0; |
39279cc3 | 3374 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3375 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3376 | goto delete; | |
3377 | ||
3378 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3379 | u64 num_dec; |
db94535d | 3380 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3381 | if (!del_item && !encoding) { |
db94535d CM |
3382 | u64 orig_num_bytes = |
3383 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3384 | extent_num_bytes = new_size - |
5f39d397 | 3385 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3386 | extent_num_bytes = extent_num_bytes & |
3387 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3388 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3389 | extent_num_bytes); | |
3390 | num_dec = (orig_num_bytes - | |
9069218d | 3391 | extent_num_bytes); |
e02119d5 | 3392 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3393 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3394 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3395 | } else { |
db94535d CM |
3396 | extent_num_bytes = |
3397 | btrfs_file_extent_disk_num_bytes(leaf, | |
3398 | fi); | |
5d4f98a2 YZ |
3399 | extent_offset = found_key.offset - |
3400 | btrfs_file_extent_offset(leaf, fi); | |
3401 | ||
39279cc3 | 3402 | /* FIXME blocksize != 4096 */ |
9069218d | 3403 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3404 | if (extent_start != 0) { |
3405 | found_extent = 1; | |
e02119d5 | 3406 | if (root->ref_cows) |
a76a3cd4 | 3407 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3408 | } |
39279cc3 | 3409 | } |
9069218d | 3410 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3411 | /* |
3412 | * we can't truncate inline items that have had | |
3413 | * special encodings | |
3414 | */ | |
3415 | if (!del_item && | |
3416 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3417 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3418 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3419 | u32 size = new_size - found_key.offset; |
3420 | ||
3421 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3422 | inode_sub_bytes(inode, item_end + 1 - |
3423 | new_size); | |
e02119d5 CM |
3424 | } |
3425 | size = | |
3426 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3427 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3428 | size, 1); |
9069218d | 3429 | BUG_ON(ret); |
e02119d5 | 3430 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3431 | inode_sub_bytes(inode, item_end + 1 - |
3432 | found_key.offset); | |
9069218d | 3433 | } |
39279cc3 | 3434 | } |
179e29e4 | 3435 | delete: |
39279cc3 | 3436 | if (del_item) { |
85e21bac CM |
3437 | if (!pending_del_nr) { |
3438 | /* no pending yet, add ourselves */ | |
3439 | pending_del_slot = path->slots[0]; | |
3440 | pending_del_nr = 1; | |
3441 | } else if (pending_del_nr && | |
3442 | path->slots[0] + 1 == pending_del_slot) { | |
3443 | /* hop on the pending chunk */ | |
3444 | pending_del_nr++; | |
3445 | pending_del_slot = path->slots[0]; | |
3446 | } else { | |
d397712b | 3447 | BUG(); |
85e21bac | 3448 | } |
39279cc3 CM |
3449 | } else { |
3450 | break; | |
3451 | } | |
0af3d00b JB |
3452 | if (found_extent && (root->ref_cows || |
3453 | root == root->fs_info->tree_root)) { | |
b9473439 | 3454 | btrfs_set_path_blocking(path); |
39279cc3 | 3455 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3456 | extent_num_bytes, 0, |
3457 | btrfs_header_owner(leaf), | |
3458 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3459 | BUG_ON(ret); |
3460 | } | |
85e21bac | 3461 | |
8082510e YZ |
3462 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3463 | break; | |
3464 | ||
3465 | if (path->slots[0] == 0 || | |
3466 | path->slots[0] != pending_del_slot) { | |
3467 | if (root->ref_cows) { | |
3468 | err = -EAGAIN; | |
3469 | goto out; | |
3470 | } | |
3471 | if (pending_del_nr) { | |
3472 | ret = btrfs_del_items(trans, root, path, | |
3473 | pending_del_slot, | |
3474 | pending_del_nr); | |
3475 | BUG_ON(ret); | |
3476 | pending_del_nr = 0; | |
3477 | } | |
85e21bac CM |
3478 | btrfs_release_path(root, path); |
3479 | goto search_again; | |
8082510e YZ |
3480 | } else { |
3481 | path->slots[0]--; | |
85e21bac | 3482 | } |
39279cc3 | 3483 | } |
8082510e | 3484 | out: |
85e21bac CM |
3485 | if (pending_del_nr) { |
3486 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3487 | pending_del_nr); | |
d68fc57b | 3488 | BUG_ON(ret); |
85e21bac | 3489 | } |
39279cc3 | 3490 | btrfs_free_path(path); |
8082510e | 3491 | return err; |
39279cc3 CM |
3492 | } |
3493 | ||
3494 | /* | |
3495 | * taken from block_truncate_page, but does cow as it zeros out | |
3496 | * any bytes left in the last page in the file. | |
3497 | */ | |
3498 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3499 | { | |
3500 | struct inode *inode = mapping->host; | |
db94535d | 3501 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3502 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3503 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3504 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3505 | char *kaddr; |
db94535d | 3506 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3507 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3508 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3509 | struct page *page; | |
39279cc3 | 3510 | int ret = 0; |
a52d9a80 | 3511 | u64 page_start; |
e6dcd2dc | 3512 | u64 page_end; |
39279cc3 CM |
3513 | |
3514 | if ((offset & (blocksize - 1)) == 0) | |
3515 | goto out; | |
0ca1f7ce | 3516 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3517 | if (ret) |
3518 | goto out; | |
39279cc3 CM |
3519 | |
3520 | ret = -ENOMEM; | |
211c17f5 | 3521 | again: |
39279cc3 | 3522 | page = grab_cache_page(mapping, index); |
5d5e103a | 3523 | if (!page) { |
0ca1f7ce | 3524 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3525 | goto out; |
5d5e103a | 3526 | } |
e6dcd2dc CM |
3527 | |
3528 | page_start = page_offset(page); | |
3529 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3530 | ||
39279cc3 | 3531 | if (!PageUptodate(page)) { |
9ebefb18 | 3532 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3533 | lock_page(page); |
211c17f5 CM |
3534 | if (page->mapping != mapping) { |
3535 | unlock_page(page); | |
3536 | page_cache_release(page); | |
3537 | goto again; | |
3538 | } | |
39279cc3 CM |
3539 | if (!PageUptodate(page)) { |
3540 | ret = -EIO; | |
89642229 | 3541 | goto out_unlock; |
39279cc3 CM |
3542 | } |
3543 | } | |
211c17f5 | 3544 | wait_on_page_writeback(page); |
e6dcd2dc | 3545 | |
2ac55d41 JB |
3546 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3547 | GFP_NOFS); | |
e6dcd2dc CM |
3548 | set_page_extent_mapped(page); |
3549 | ||
3550 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3551 | if (ordered) { | |
2ac55d41 JB |
3552 | unlock_extent_cached(io_tree, page_start, page_end, |
3553 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3554 | unlock_page(page); |
3555 | page_cache_release(page); | |
eb84ae03 | 3556 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3557 | btrfs_put_ordered_extent(ordered); |
3558 | goto again; | |
3559 | } | |
3560 | ||
2ac55d41 | 3561 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3562 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3563 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3564 | |
2ac55d41 JB |
3565 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3566 | &cached_state); | |
9ed74f2d | 3567 | if (ret) { |
2ac55d41 JB |
3568 | unlock_extent_cached(io_tree, page_start, page_end, |
3569 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3570 | goto out_unlock; |
3571 | } | |
3572 | ||
e6dcd2dc CM |
3573 | ret = 0; |
3574 | if (offset != PAGE_CACHE_SIZE) { | |
3575 | kaddr = kmap(page); | |
3576 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3577 | flush_dcache_page(page); | |
3578 | kunmap(page); | |
3579 | } | |
247e743c | 3580 | ClearPageChecked(page); |
e6dcd2dc | 3581 | set_page_dirty(page); |
2ac55d41 JB |
3582 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3583 | GFP_NOFS); | |
39279cc3 | 3584 | |
89642229 | 3585 | out_unlock: |
5d5e103a | 3586 | if (ret) |
0ca1f7ce | 3587 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3588 | unlock_page(page); |
3589 | page_cache_release(page); | |
3590 | out: | |
3591 | return ret; | |
3592 | } | |
3593 | ||
695a0d0d JB |
3594 | /* |
3595 | * This function puts in dummy file extents for the area we're creating a hole | |
3596 | * for. So if we are truncating this file to a larger size we need to insert | |
3597 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3598 | * the range between oldsize and size | |
3599 | */ | |
a41ad394 | 3600 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3601 | { |
9036c102 YZ |
3602 | struct btrfs_trans_handle *trans; |
3603 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3604 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3605 | struct extent_map *em = NULL; |
2ac55d41 | 3606 | struct extent_state *cached_state = NULL; |
9036c102 | 3607 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3608 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3609 | u64 block_end = (size + mask) & ~mask; |
3610 | u64 last_byte; | |
3611 | u64 cur_offset; | |
3612 | u64 hole_size; | |
9ed74f2d | 3613 | int err = 0; |
39279cc3 | 3614 | |
9036c102 YZ |
3615 | if (size <= hole_start) |
3616 | return 0; | |
3617 | ||
9036c102 YZ |
3618 | while (1) { |
3619 | struct btrfs_ordered_extent *ordered; | |
3620 | btrfs_wait_ordered_range(inode, hole_start, | |
3621 | block_end - hole_start); | |
2ac55d41 JB |
3622 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3623 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3624 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3625 | if (!ordered) | |
3626 | break; | |
2ac55d41 JB |
3627 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3628 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3629 | btrfs_put_ordered_extent(ordered); |
3630 | } | |
39279cc3 | 3631 | |
9036c102 YZ |
3632 | cur_offset = hole_start; |
3633 | while (1) { | |
3634 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3635 | block_end - cur_offset, 0); | |
3636 | BUG_ON(IS_ERR(em) || !em); | |
3637 | last_byte = min(extent_map_end(em), block_end); | |
3638 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3639 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3640 | u64 hint_byte = 0; |
9036c102 | 3641 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3642 | |
a22285a6 YZ |
3643 | trans = btrfs_start_transaction(root, 2); |
3644 | if (IS_ERR(trans)) { | |
3645 | err = PTR_ERR(trans); | |
9ed74f2d | 3646 | break; |
a22285a6 | 3647 | } |
8082510e YZ |
3648 | btrfs_set_trans_block_group(trans, inode); |
3649 | ||
3650 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3651 | cur_offset + hole_size, | |
3652 | &hint_byte, 1); | |
3893e33b JB |
3653 | if (err) |
3654 | break; | |
8082510e | 3655 | |
9036c102 YZ |
3656 | err = btrfs_insert_file_extent(trans, root, |
3657 | inode->i_ino, cur_offset, 0, | |
3658 | 0, hole_size, 0, hole_size, | |
3659 | 0, 0, 0); | |
3893e33b JB |
3660 | if (err) |
3661 | break; | |
8082510e | 3662 | |
9036c102 YZ |
3663 | btrfs_drop_extent_cache(inode, hole_start, |
3664 | last_byte - 1, 0); | |
8082510e YZ |
3665 | |
3666 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3667 | } |
3668 | free_extent_map(em); | |
a22285a6 | 3669 | em = NULL; |
9036c102 | 3670 | cur_offset = last_byte; |
8082510e | 3671 | if (cur_offset >= block_end) |
9036c102 YZ |
3672 | break; |
3673 | } | |
1832a6d5 | 3674 | |
a22285a6 | 3675 | free_extent_map(em); |
2ac55d41 JB |
3676 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3677 | GFP_NOFS); | |
9036c102 YZ |
3678 | return err; |
3679 | } | |
39279cc3 | 3680 | |
a41ad394 | 3681 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3682 | { |
a41ad394 | 3683 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3684 | int ret; |
3685 | ||
a41ad394 | 3686 | if (newsize == oldsize) |
8082510e YZ |
3687 | return 0; |
3688 | ||
a41ad394 JB |
3689 | if (newsize > oldsize) { |
3690 | i_size_write(inode, newsize); | |
3691 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3692 | truncate_pagecache(inode, oldsize, newsize); | |
3693 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
8082510e | 3694 | if (ret) { |
a41ad394 | 3695 | btrfs_setsize(inode, oldsize); |
8082510e YZ |
3696 | return ret; |
3697 | } | |
3698 | ||
930f028a | 3699 | mark_inode_dirty(inode); |
a41ad394 | 3700 | } else { |
8082510e | 3701 | |
a41ad394 JB |
3702 | /* |
3703 | * We're truncating a file that used to have good data down to | |
3704 | * zero. Make sure it gets into the ordered flush list so that | |
3705 | * any new writes get down to disk quickly. | |
3706 | */ | |
3707 | if (newsize == 0) | |
3708 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3709 | |
a41ad394 JB |
3710 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3711 | truncate_setsize(inode, newsize); | |
3712 | ret = btrfs_truncate(inode); | |
3713 | } | |
8082510e | 3714 | |
a41ad394 | 3715 | return ret; |
8082510e YZ |
3716 | } |
3717 | ||
9036c102 YZ |
3718 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3719 | { | |
3720 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3721 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3722 | int err; |
39279cc3 | 3723 | |
b83cc969 LZ |
3724 | if (btrfs_root_readonly(root)) |
3725 | return -EROFS; | |
3726 | ||
9036c102 YZ |
3727 | err = inode_change_ok(inode, attr); |
3728 | if (err) | |
3729 | return err; | |
2bf5a725 | 3730 | |
5a3f23d5 | 3731 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3732 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3733 | if (err) |
3734 | return err; | |
39279cc3 | 3735 | } |
9036c102 | 3736 | |
1025774c CH |
3737 | if (attr->ia_valid) { |
3738 | setattr_copy(inode, attr); | |
3739 | mark_inode_dirty(inode); | |
3740 | ||
3741 | if (attr->ia_valid & ATTR_MODE) | |
3742 | err = btrfs_acl_chmod(inode); | |
3743 | } | |
33268eaf | 3744 | |
39279cc3 CM |
3745 | return err; |
3746 | } | |
61295eb8 | 3747 | |
bd555975 | 3748 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3749 | { |
3750 | struct btrfs_trans_handle *trans; | |
3751 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3752 | unsigned long nr; |
39279cc3 CM |
3753 | int ret; |
3754 | ||
1abe9b8a | 3755 | trace_btrfs_inode_evict(inode); |
3756 | ||
39279cc3 | 3757 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b JB |
3758 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
3759 | root == root->fs_info->tree_root)) | |
bd555975 AV |
3760 | goto no_delete; |
3761 | ||
39279cc3 | 3762 | if (is_bad_inode(inode)) { |
7b128766 | 3763 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3764 | goto no_delete; |
3765 | } | |
bd555975 | 3766 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3767 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3768 | |
c71bf099 YZ |
3769 | if (root->fs_info->log_root_recovering) { |
3770 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3771 | goto no_delete; | |
3772 | } | |
3773 | ||
76dda93c YZ |
3774 | if (inode->i_nlink > 0) { |
3775 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3776 | goto no_delete; | |
3777 | } | |
3778 | ||
dbe674a9 | 3779 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3780 | |
8082510e | 3781 | while (1) { |
d68fc57b YZ |
3782 | trans = btrfs_start_transaction(root, 0); |
3783 | BUG_ON(IS_ERR(trans)); | |
8082510e | 3784 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b YZ |
3785 | trans->block_rsv = root->orphan_block_rsv; |
3786 | ||
3787 | ret = btrfs_block_rsv_check(trans, root, | |
3788 | root->orphan_block_rsv, 0, 5); | |
3789 | if (ret) { | |
3790 | BUG_ON(ret != -EAGAIN); | |
3791 | ret = btrfs_commit_transaction(trans, root); | |
3792 | BUG_ON(ret); | |
3793 | continue; | |
3794 | } | |
7b128766 | 3795 | |
d68fc57b | 3796 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3797 | if (ret != -EAGAIN) |
3798 | break; | |
85e21bac | 3799 | |
8082510e YZ |
3800 | nr = trans->blocks_used; |
3801 | btrfs_end_transaction(trans, root); | |
3802 | trans = NULL; | |
3803 | btrfs_btree_balance_dirty(root, nr); | |
d68fc57b | 3804 | |
8082510e | 3805 | } |
5f39d397 | 3806 | |
8082510e YZ |
3807 | if (ret == 0) { |
3808 | ret = btrfs_orphan_del(trans, inode); | |
3809 | BUG_ON(ret); | |
3810 | } | |
54aa1f4d | 3811 | |
d3c2fdcf | 3812 | nr = trans->blocks_used; |
54aa1f4d | 3813 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3814 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3815 | no_delete: |
bd555975 | 3816 | end_writeback(inode); |
8082510e | 3817 | return; |
39279cc3 CM |
3818 | } |
3819 | ||
3820 | /* | |
3821 | * this returns the key found in the dir entry in the location pointer. | |
3822 | * If no dir entries were found, location->objectid is 0. | |
3823 | */ | |
3824 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3825 | struct btrfs_key *location) | |
3826 | { | |
3827 | const char *name = dentry->d_name.name; | |
3828 | int namelen = dentry->d_name.len; | |
3829 | struct btrfs_dir_item *di; | |
3830 | struct btrfs_path *path; | |
3831 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3832 | int ret = 0; |
39279cc3 CM |
3833 | |
3834 | path = btrfs_alloc_path(); | |
3835 | BUG_ON(!path); | |
3954401f | 3836 | |
39279cc3 CM |
3837 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3838 | namelen, 0); | |
0d9f7f3e Y |
3839 | if (IS_ERR(di)) |
3840 | ret = PTR_ERR(di); | |
d397712b CM |
3841 | |
3842 | if (!di || IS_ERR(di)) | |
3954401f | 3843 | goto out_err; |
d397712b | 3844 | |
5f39d397 | 3845 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3846 | out: |
39279cc3 CM |
3847 | btrfs_free_path(path); |
3848 | return ret; | |
3954401f CM |
3849 | out_err: |
3850 | location->objectid = 0; | |
3851 | goto out; | |
39279cc3 CM |
3852 | } |
3853 | ||
3854 | /* | |
3855 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3856 | * needs to be changed to reflect the root directory of the tree root. This | |
3857 | * is kind of like crossing a mount point. | |
3858 | */ | |
3859 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3860 | struct inode *dir, |
3861 | struct dentry *dentry, | |
3862 | struct btrfs_key *location, | |
3863 | struct btrfs_root **sub_root) | |
39279cc3 | 3864 | { |
4df27c4d YZ |
3865 | struct btrfs_path *path; |
3866 | struct btrfs_root *new_root; | |
3867 | struct btrfs_root_ref *ref; | |
3868 | struct extent_buffer *leaf; | |
3869 | int ret; | |
3870 | int err = 0; | |
39279cc3 | 3871 | |
4df27c4d YZ |
3872 | path = btrfs_alloc_path(); |
3873 | if (!path) { | |
3874 | err = -ENOMEM; | |
3875 | goto out; | |
3876 | } | |
39279cc3 | 3877 | |
4df27c4d YZ |
3878 | err = -ENOENT; |
3879 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3880 | BTRFS_I(dir)->root->root_key.objectid, | |
3881 | location->objectid); | |
3882 | if (ret) { | |
3883 | if (ret < 0) | |
3884 | err = ret; | |
3885 | goto out; | |
3886 | } | |
39279cc3 | 3887 | |
4df27c4d YZ |
3888 | leaf = path->nodes[0]; |
3889 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3890 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3891 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3892 | goto out; | |
39279cc3 | 3893 | |
4df27c4d YZ |
3894 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3895 | (unsigned long)(ref + 1), | |
3896 | dentry->d_name.len); | |
3897 | if (ret) | |
3898 | goto out; | |
3899 | ||
3900 | btrfs_release_path(root->fs_info->tree_root, path); | |
3901 | ||
3902 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3903 | if (IS_ERR(new_root)) { | |
3904 | err = PTR_ERR(new_root); | |
3905 | goto out; | |
3906 | } | |
3907 | ||
3908 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3909 | err = -ENOENT; | |
3910 | goto out; | |
3911 | } | |
3912 | ||
3913 | *sub_root = new_root; | |
3914 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3915 | location->type = BTRFS_INODE_ITEM_KEY; | |
3916 | location->offset = 0; | |
3917 | err = 0; | |
3918 | out: | |
3919 | btrfs_free_path(path); | |
3920 | return err; | |
39279cc3 CM |
3921 | } |
3922 | ||
5d4f98a2 YZ |
3923 | static void inode_tree_add(struct inode *inode) |
3924 | { | |
3925 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3926 | struct btrfs_inode *entry; | |
03e860bd FNP |
3927 | struct rb_node **p; |
3928 | struct rb_node *parent; | |
03e860bd FNP |
3929 | again: |
3930 | p = &root->inode_tree.rb_node; | |
3931 | parent = NULL; | |
5d4f98a2 | 3932 | |
1d3382cb | 3933 | if (inode_unhashed(inode)) |
76dda93c YZ |
3934 | return; |
3935 | ||
5d4f98a2 YZ |
3936 | spin_lock(&root->inode_lock); |
3937 | while (*p) { | |
3938 | parent = *p; | |
3939 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3940 | ||
3941 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3942 | p = &parent->rb_left; |
5d4f98a2 | 3943 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3944 | p = &parent->rb_right; |
5d4f98a2 YZ |
3945 | else { |
3946 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3947 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3948 | rb_erase(parent, &root->inode_tree); |
3949 | RB_CLEAR_NODE(parent); | |
3950 | spin_unlock(&root->inode_lock); | |
3951 | goto again; | |
5d4f98a2 YZ |
3952 | } |
3953 | } | |
3954 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3955 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3956 | spin_unlock(&root->inode_lock); | |
3957 | } | |
3958 | ||
3959 | static void inode_tree_del(struct inode *inode) | |
3960 | { | |
3961 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3962 | int empty = 0; |
5d4f98a2 | 3963 | |
03e860bd | 3964 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3965 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3966 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3967 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3968 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3969 | } |
03e860bd | 3970 | spin_unlock(&root->inode_lock); |
76dda93c | 3971 | |
0af3d00b JB |
3972 | /* |
3973 | * Free space cache has inodes in the tree root, but the tree root has a | |
3974 | * root_refs of 0, so this could end up dropping the tree root as a | |
3975 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3976 | * make sure we don't drop it. | |
3977 | */ | |
3978 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3979 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3980 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3981 | spin_lock(&root->inode_lock); | |
3982 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3983 | spin_unlock(&root->inode_lock); | |
3984 | if (empty) | |
3985 | btrfs_add_dead_root(root); | |
3986 | } | |
3987 | } | |
3988 | ||
3989 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3990 | { | |
3991 | struct rb_node *node; | |
3992 | struct rb_node *prev; | |
3993 | struct btrfs_inode *entry; | |
3994 | struct inode *inode; | |
3995 | u64 objectid = 0; | |
3996 | ||
3997 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3998 | ||
3999 | spin_lock(&root->inode_lock); | |
4000 | again: | |
4001 | node = root->inode_tree.rb_node; | |
4002 | prev = NULL; | |
4003 | while (node) { | |
4004 | prev = node; | |
4005 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4006 | ||
4007 | if (objectid < entry->vfs_inode.i_ino) | |
4008 | node = node->rb_left; | |
4009 | else if (objectid > entry->vfs_inode.i_ino) | |
4010 | node = node->rb_right; | |
4011 | else | |
4012 | break; | |
4013 | } | |
4014 | if (!node) { | |
4015 | while (prev) { | |
4016 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4017 | if (objectid <= entry->vfs_inode.i_ino) { | |
4018 | node = prev; | |
4019 | break; | |
4020 | } | |
4021 | prev = rb_next(prev); | |
4022 | } | |
4023 | } | |
4024 | while (node) { | |
4025 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4026 | objectid = entry->vfs_inode.i_ino + 1; | |
4027 | inode = igrab(&entry->vfs_inode); | |
4028 | if (inode) { | |
4029 | spin_unlock(&root->inode_lock); | |
4030 | if (atomic_read(&inode->i_count) > 1) | |
4031 | d_prune_aliases(inode); | |
4032 | /* | |
45321ac5 | 4033 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4034 | * the inode cache when its usage count |
4035 | * hits zero. | |
4036 | */ | |
4037 | iput(inode); | |
4038 | cond_resched(); | |
4039 | spin_lock(&root->inode_lock); | |
4040 | goto again; | |
4041 | } | |
4042 | ||
4043 | if (cond_resched_lock(&root->inode_lock)) | |
4044 | goto again; | |
4045 | ||
4046 | node = rb_next(node); | |
4047 | } | |
4048 | spin_unlock(&root->inode_lock); | |
4049 | return 0; | |
5d4f98a2 YZ |
4050 | } |
4051 | ||
e02119d5 CM |
4052 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4053 | { | |
4054 | struct btrfs_iget_args *args = p; | |
4055 | inode->i_ino = args->ino; | |
e02119d5 | 4056 | BTRFS_I(inode)->root = args->root; |
6a63209f | 4057 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
4058 | return 0; |
4059 | } | |
4060 | ||
4061 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4062 | { | |
4063 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
4064 | return args->ino == inode->i_ino && |
4065 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
4066 | } |
4067 | ||
5d4f98a2 YZ |
4068 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4069 | u64 objectid, | |
4070 | struct btrfs_root *root) | |
39279cc3 CM |
4071 | { |
4072 | struct inode *inode; | |
4073 | struct btrfs_iget_args args; | |
4074 | args.ino = objectid; | |
4075 | args.root = root; | |
4076 | ||
4077 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4078 | btrfs_init_locked_inode, | |
4079 | (void *)&args); | |
4080 | return inode; | |
4081 | } | |
4082 | ||
1a54ef8c BR |
4083 | /* Get an inode object given its location and corresponding root. |
4084 | * Returns in *is_new if the inode was read from disk | |
4085 | */ | |
4086 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4087 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4088 | { |
4089 | struct inode *inode; | |
4090 | ||
4091 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4092 | if (!inode) | |
5d4f98a2 | 4093 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4094 | |
4095 | if (inode->i_state & I_NEW) { | |
4096 | BTRFS_I(inode)->root = root; | |
4097 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4098 | btrfs_read_locked_inode(inode); | |
5d4f98a2 | 4099 | inode_tree_add(inode); |
1a54ef8c | 4100 | unlock_new_inode(inode); |
73f73415 JB |
4101 | if (new) |
4102 | *new = 1; | |
1a54ef8c BR |
4103 | } |
4104 | ||
4105 | return inode; | |
4106 | } | |
4107 | ||
4df27c4d YZ |
4108 | static struct inode *new_simple_dir(struct super_block *s, |
4109 | struct btrfs_key *key, | |
4110 | struct btrfs_root *root) | |
4111 | { | |
4112 | struct inode *inode = new_inode(s); | |
4113 | ||
4114 | if (!inode) | |
4115 | return ERR_PTR(-ENOMEM); | |
4116 | ||
4df27c4d YZ |
4117 | BTRFS_I(inode)->root = root; |
4118 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
4119 | BTRFS_I(inode)->dummy_inode = 1; | |
4120 | ||
4121 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
4122 | inode->i_op = &simple_dir_inode_operations; | |
4123 | inode->i_fop = &simple_dir_operations; | |
4124 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4125 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4126 | ||
4127 | return inode; | |
4128 | } | |
4129 | ||
3de4586c | 4130 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4131 | { |
d397712b | 4132 | struct inode *inode; |
4df27c4d | 4133 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4134 | struct btrfs_root *sub_root = root; |
4135 | struct btrfs_key location; | |
76dda93c | 4136 | int index; |
5d4f98a2 | 4137 | int ret; |
39279cc3 CM |
4138 | |
4139 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4140 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4141 | |
39279cc3 | 4142 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 4143 | |
39279cc3 CM |
4144 | if (ret < 0) |
4145 | return ERR_PTR(ret); | |
5f39d397 | 4146 | |
4df27c4d YZ |
4147 | if (location.objectid == 0) |
4148 | return NULL; | |
4149 | ||
4150 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4151 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4152 | return inode; |
4153 | } | |
4154 | ||
4155 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4156 | ||
76dda93c | 4157 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4158 | ret = fixup_tree_root_location(root, dir, dentry, |
4159 | &location, &sub_root); | |
4160 | if (ret < 0) { | |
4161 | if (ret != -ENOENT) | |
4162 | inode = ERR_PTR(ret); | |
4163 | else | |
4164 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4165 | } else { | |
73f73415 | 4166 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4167 | } |
76dda93c YZ |
4168 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4169 | ||
34d19bad | 4170 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4171 | down_read(&root->fs_info->cleanup_work_sem); |
4172 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4173 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4174 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4175 | if (ret) |
4176 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4177 | } |
4178 | ||
3de4586c CM |
4179 | return inode; |
4180 | } | |
4181 | ||
fe15ce44 | 4182 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4183 | { |
4184 | struct btrfs_root *root; | |
4185 | ||
efefb143 YZ |
4186 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
4187 | dentry = dentry->d_parent; | |
76dda93c | 4188 | |
efefb143 YZ |
4189 | if (dentry->d_inode) { |
4190 | root = BTRFS_I(dentry->d_inode)->root; | |
4191 | if (btrfs_root_refs(&root->root_item) == 0) | |
4192 | return 1; | |
4193 | } | |
76dda93c YZ |
4194 | return 0; |
4195 | } | |
4196 | ||
3de4586c CM |
4197 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4198 | struct nameidata *nd) | |
4199 | { | |
4200 | struct inode *inode; | |
4201 | ||
3de4586c CM |
4202 | inode = btrfs_lookup_dentry(dir, dentry); |
4203 | if (IS_ERR(inode)) | |
4204 | return ERR_CAST(inode); | |
7b128766 | 4205 | |
39279cc3 CM |
4206 | return d_splice_alias(inode, dentry); |
4207 | } | |
4208 | ||
39279cc3 CM |
4209 | static unsigned char btrfs_filetype_table[] = { |
4210 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
4211 | }; | |
4212 | ||
cbdf5a24 DW |
4213 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4214 | filldir_t filldir) | |
39279cc3 | 4215 | { |
6da6abae | 4216 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4217 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4218 | struct btrfs_item *item; | |
4219 | struct btrfs_dir_item *di; | |
4220 | struct btrfs_key key; | |
5f39d397 | 4221 | struct btrfs_key found_key; |
39279cc3 CM |
4222 | struct btrfs_path *path; |
4223 | int ret; | |
4224 | u32 nritems; | |
5f39d397 | 4225 | struct extent_buffer *leaf; |
39279cc3 CM |
4226 | int slot; |
4227 | int advance; | |
4228 | unsigned char d_type; | |
4229 | int over = 0; | |
4230 | u32 di_cur; | |
4231 | u32 di_total; | |
4232 | u32 di_len; | |
4233 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4234 | char tmp_name[32]; |
4235 | char *name_ptr; | |
4236 | int name_len; | |
39279cc3 CM |
4237 | |
4238 | /* FIXME, use a real flag for deciding about the key type */ | |
4239 | if (root->fs_info->tree_root == root) | |
4240 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4241 | |
3954401f CM |
4242 | /* special case for "." */ |
4243 | if (filp->f_pos == 0) { | |
4244 | over = filldir(dirent, ".", 1, | |
4245 | 1, inode->i_ino, | |
4246 | DT_DIR); | |
4247 | if (over) | |
4248 | return 0; | |
4249 | filp->f_pos = 1; | |
4250 | } | |
3954401f CM |
4251 | /* special case for .., just use the back ref */ |
4252 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4253 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4254 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 4255 | 2, pino, DT_DIR); |
3954401f | 4256 | if (over) |
49593bfa | 4257 | return 0; |
3954401f CM |
4258 | filp->f_pos = 2; |
4259 | } | |
49593bfa DW |
4260 | path = btrfs_alloc_path(); |
4261 | path->reada = 2; | |
4262 | ||
39279cc3 CM |
4263 | btrfs_set_key_type(&key, key_type); |
4264 | key.offset = filp->f_pos; | |
49593bfa | 4265 | key.objectid = inode->i_ino; |
5f39d397 | 4266 | |
39279cc3 CM |
4267 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4268 | if (ret < 0) | |
4269 | goto err; | |
4270 | advance = 0; | |
49593bfa DW |
4271 | |
4272 | while (1) { | |
5f39d397 CM |
4273 | leaf = path->nodes[0]; |
4274 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4275 | slot = path->slots[0]; |
4276 | if (advance || slot >= nritems) { | |
49593bfa | 4277 | if (slot >= nritems - 1) { |
39279cc3 CM |
4278 | ret = btrfs_next_leaf(root, path); |
4279 | if (ret) | |
4280 | break; | |
5f39d397 CM |
4281 | leaf = path->nodes[0]; |
4282 | nritems = btrfs_header_nritems(leaf); | |
39279cc3 CM |
4283 | slot = path->slots[0]; |
4284 | } else { | |
4285 | slot++; | |
4286 | path->slots[0]++; | |
4287 | } | |
4288 | } | |
3de4586c | 4289 | |
39279cc3 | 4290 | advance = 1; |
5f39d397 CM |
4291 | item = btrfs_item_nr(leaf, slot); |
4292 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4293 | ||
4294 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4295 | break; |
5f39d397 | 4296 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4297 | break; |
5f39d397 | 4298 | if (found_key.offset < filp->f_pos) |
39279cc3 | 4299 | continue; |
5f39d397 CM |
4300 | |
4301 | filp->f_pos = found_key.offset; | |
49593bfa | 4302 | |
39279cc3 CM |
4303 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4304 | di_cur = 0; | |
5f39d397 | 4305 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4306 | |
4307 | while (di_cur < di_total) { | |
5f39d397 CM |
4308 | struct btrfs_key location; |
4309 | ||
22a94d44 JB |
4310 | if (verify_dir_item(root, leaf, di)) |
4311 | break; | |
4312 | ||
5f39d397 | 4313 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4314 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4315 | name_ptr = tmp_name; |
4316 | } else { | |
4317 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4318 | if (!name_ptr) { |
4319 | ret = -ENOMEM; | |
4320 | goto err; | |
4321 | } | |
5f39d397 CM |
4322 | } |
4323 | read_extent_buffer(leaf, name_ptr, | |
4324 | (unsigned long)(di + 1), name_len); | |
4325 | ||
4326 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4327 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
4328 | |
4329 | /* is this a reference to our own snapshot? If so | |
4330 | * skip it | |
4331 | */ | |
4332 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4333 | location.objectid == root->root_key.objectid) { | |
4334 | over = 0; | |
4335 | goto skip; | |
4336 | } | |
5f39d397 | 4337 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4338 | found_key.offset, location.objectid, |
39279cc3 | 4339 | d_type); |
5f39d397 | 4340 | |
3de4586c | 4341 | skip: |
5f39d397 CM |
4342 | if (name_ptr != tmp_name) |
4343 | kfree(name_ptr); | |
4344 | ||
39279cc3 CM |
4345 | if (over) |
4346 | goto nopos; | |
5103e947 | 4347 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4348 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4349 | di_cur += di_len; |
4350 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4351 | } | |
4352 | } | |
49593bfa DW |
4353 | |
4354 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 4355 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4356 | /* |
4357 | * 32-bit glibc will use getdents64, but then strtol - | |
4358 | * so the last number we can serve is this. | |
4359 | */ | |
4360 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4361 | else |
4362 | filp->f_pos++; | |
39279cc3 CM |
4363 | nopos: |
4364 | ret = 0; | |
4365 | err: | |
39279cc3 | 4366 | btrfs_free_path(path); |
39279cc3 CM |
4367 | return ret; |
4368 | } | |
4369 | ||
a9185b41 | 4370 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4371 | { |
4372 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4373 | struct btrfs_trans_handle *trans; | |
4374 | int ret = 0; | |
0af3d00b | 4375 | bool nolock = false; |
39279cc3 | 4376 | |
8929ecfa | 4377 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4378 | return 0; |
4379 | ||
0af3d00b JB |
4380 | smp_mb(); |
4381 | nolock = (root->fs_info->closing && root == root->fs_info->tree_root); | |
4382 | ||
a9185b41 | 4383 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b JB |
4384 | if (nolock) |
4385 | trans = btrfs_join_transaction_nolock(root, 1); | |
4386 | else | |
4387 | trans = btrfs_join_transaction(root, 1); | |
3612b495 TI |
4388 | if (IS_ERR(trans)) |
4389 | return PTR_ERR(trans); | |
39279cc3 | 4390 | btrfs_set_trans_block_group(trans, inode); |
0af3d00b JB |
4391 | if (nolock) |
4392 | ret = btrfs_end_transaction_nolock(trans, root); | |
4393 | else | |
4394 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4395 | } |
4396 | return ret; | |
4397 | } | |
4398 | ||
4399 | /* | |
54aa1f4d | 4400 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4401 | * inode changes. But, it is most likely to find the inode in cache. |
4402 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4403 | * to keep or drop this code. | |
4404 | */ | |
4405 | void btrfs_dirty_inode(struct inode *inode) | |
4406 | { | |
4407 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4408 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4409 | int ret; |
4410 | ||
4411 | if (BTRFS_I(inode)->dummy_inode) | |
4412 | return; | |
39279cc3 | 4413 | |
f9295749 | 4414 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 4415 | BUG_ON(IS_ERR(trans)); |
39279cc3 | 4416 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa YZ |
4417 | |
4418 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4419 | if (ret && ret == -ENOSPC) { |
4420 | /* whoops, lets try again with the full transaction */ | |
4421 | btrfs_end_transaction(trans, root); | |
4422 | trans = btrfs_start_transaction(root, 1); | |
9aeead73 CM |
4423 | if (IS_ERR(trans)) { |
4424 | if (printk_ratelimit()) { | |
4425 | printk(KERN_ERR "btrfs: fail to " | |
4426 | "dirty inode %lu error %ld\n", | |
4427 | inode->i_ino, PTR_ERR(trans)); | |
4428 | } | |
4429 | return; | |
4430 | } | |
94b60442 | 4431 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa | 4432 | |
94b60442 CM |
4433 | ret = btrfs_update_inode(trans, root, inode); |
4434 | if (ret) { | |
9aeead73 CM |
4435 | if (printk_ratelimit()) { |
4436 | printk(KERN_ERR "btrfs: fail to " | |
4437 | "dirty inode %lu error %d\n", | |
4438 | inode->i_ino, ret); | |
4439 | } | |
94b60442 CM |
4440 | } |
4441 | } | |
39279cc3 | 4442 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4443 | } |
4444 | ||
d352ac68 CM |
4445 | /* |
4446 | * find the highest existing sequence number in a directory | |
4447 | * and then set the in-memory index_cnt variable to reflect | |
4448 | * free sequence numbers | |
4449 | */ | |
aec7477b JB |
4450 | static int btrfs_set_inode_index_count(struct inode *inode) |
4451 | { | |
4452 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4453 | struct btrfs_key key, found_key; | |
4454 | struct btrfs_path *path; | |
4455 | struct extent_buffer *leaf; | |
4456 | int ret; | |
4457 | ||
4458 | key.objectid = inode->i_ino; | |
4459 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
4460 | key.offset = (u64)-1; | |
4461 | ||
4462 | path = btrfs_alloc_path(); | |
4463 | if (!path) | |
4464 | return -ENOMEM; | |
4465 | ||
4466 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4467 | if (ret < 0) | |
4468 | goto out; | |
4469 | /* FIXME: we should be able to handle this */ | |
4470 | if (ret == 0) | |
4471 | goto out; | |
4472 | ret = 0; | |
4473 | ||
4474 | /* | |
4475 | * MAGIC NUMBER EXPLANATION: | |
4476 | * since we search a directory based on f_pos we have to start at 2 | |
4477 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4478 | * else has to start at 2 | |
4479 | */ | |
4480 | if (path->slots[0] == 0) { | |
4481 | BTRFS_I(inode)->index_cnt = 2; | |
4482 | goto out; | |
4483 | } | |
4484 | ||
4485 | path->slots[0]--; | |
4486 | ||
4487 | leaf = path->nodes[0]; | |
4488 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4489 | ||
4490 | if (found_key.objectid != inode->i_ino || | |
4491 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
4492 | BTRFS_I(inode)->index_cnt = 2; | |
4493 | goto out; | |
4494 | } | |
4495 | ||
4496 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4497 | out: | |
4498 | btrfs_free_path(path); | |
4499 | return ret; | |
4500 | } | |
4501 | ||
d352ac68 CM |
4502 | /* |
4503 | * helper to find a free sequence number in a given directory. This current | |
4504 | * code is very simple, later versions will do smarter things in the btree | |
4505 | */ | |
3de4586c | 4506 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4507 | { |
4508 | int ret = 0; | |
4509 | ||
4510 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
4511 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 4512 | if (ret) |
aec7477b JB |
4513 | return ret; |
4514 | } | |
4515 | ||
00e4e6b3 | 4516 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4517 | BTRFS_I(dir)->index_cnt++; |
4518 | ||
4519 | return ret; | |
4520 | } | |
4521 | ||
39279cc3 CM |
4522 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4523 | struct btrfs_root *root, | |
aec7477b | 4524 | struct inode *dir, |
9c58309d | 4525 | const char *name, int name_len, |
d2fb3437 YZ |
4526 | u64 ref_objectid, u64 objectid, |
4527 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4528 | { |
4529 | struct inode *inode; | |
5f39d397 | 4530 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4531 | struct btrfs_key *location; |
5f39d397 | 4532 | struct btrfs_path *path; |
9c58309d CM |
4533 | struct btrfs_inode_ref *ref; |
4534 | struct btrfs_key key[2]; | |
4535 | u32 sizes[2]; | |
4536 | unsigned long ptr; | |
39279cc3 CM |
4537 | int ret; |
4538 | int owner; | |
4539 | ||
5f39d397 CM |
4540 | path = btrfs_alloc_path(); |
4541 | BUG_ON(!path); | |
4542 | ||
39279cc3 | 4543 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4544 | if (!inode) { |
4545 | btrfs_free_path(path); | |
39279cc3 | 4546 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4547 | } |
39279cc3 | 4548 | |
aec7477b | 4549 | if (dir) { |
1abe9b8a | 4550 | trace_btrfs_inode_request(dir); |
4551 | ||
3de4586c | 4552 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4553 | if (ret) { |
8fb27640 | 4554 | btrfs_free_path(path); |
09771430 | 4555 | iput(inode); |
aec7477b | 4556 | return ERR_PTR(ret); |
09771430 | 4557 | } |
aec7477b JB |
4558 | } |
4559 | /* | |
4560 | * index_cnt is ignored for everything but a dir, | |
4561 | * btrfs_get_inode_index_count has an explanation for the magic | |
4562 | * number | |
4563 | */ | |
4564 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4565 | BTRFS_I(inode)->root = root; |
e02119d5 | 4566 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4567 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4568 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4569 | |
39279cc3 CM |
4570 | if (mode & S_IFDIR) |
4571 | owner = 0; | |
4572 | else | |
4573 | owner = 1; | |
d2fb3437 YZ |
4574 | BTRFS_I(inode)->block_group = |
4575 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4576 | |
4577 | key[0].objectid = objectid; | |
4578 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4579 | key[0].offset = 0; | |
4580 | ||
4581 | key[1].objectid = objectid; | |
4582 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4583 | key[1].offset = ref_objectid; | |
4584 | ||
4585 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4586 | sizes[1] = name_len + sizeof(*ref); | |
4587 | ||
b9473439 | 4588 | path->leave_spinning = 1; |
9c58309d CM |
4589 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4590 | if (ret != 0) | |
5f39d397 CM |
4591 | goto fail; |
4592 | ||
ecc11fab | 4593 | inode_init_owner(inode, dir, mode); |
39279cc3 | 4594 | inode->i_ino = objectid; |
a76a3cd4 | 4595 | inode_set_bytes(inode, 0); |
39279cc3 | 4596 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4597 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4598 | struct btrfs_inode_item); | |
e02119d5 | 4599 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4600 | |
4601 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4602 | struct btrfs_inode_ref); | |
4603 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4604 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4605 | ptr = (unsigned long)(ref + 1); |
4606 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4607 | ||
5f39d397 CM |
4608 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4609 | btrfs_free_path(path); | |
4610 | ||
39279cc3 CM |
4611 | location = &BTRFS_I(inode)->location; |
4612 | location->objectid = objectid; | |
39279cc3 CM |
4613 | location->offset = 0; |
4614 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4615 | ||
6cbff00f CH |
4616 | btrfs_inherit_iflags(inode, dir); |
4617 | ||
94272164 CM |
4618 | if ((mode & S_IFREG)) { |
4619 | if (btrfs_test_opt(root, NODATASUM)) | |
4620 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4621 | if (btrfs_test_opt(root, NODATACOW) || |
4622 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4623 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4624 | } | |
4625 | ||
39279cc3 | 4626 | insert_inode_hash(inode); |
5d4f98a2 | 4627 | inode_tree_add(inode); |
1abe9b8a | 4628 | |
4629 | trace_btrfs_inode_new(inode); | |
4630 | ||
39279cc3 | 4631 | return inode; |
5f39d397 | 4632 | fail: |
aec7477b JB |
4633 | if (dir) |
4634 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4635 | btrfs_free_path(path); |
09771430 | 4636 | iput(inode); |
5f39d397 | 4637 | return ERR_PTR(ret); |
39279cc3 CM |
4638 | } |
4639 | ||
4640 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4641 | { | |
4642 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4643 | } | |
4644 | ||
d352ac68 CM |
4645 | /* |
4646 | * utility function to add 'inode' into 'parent_inode' with | |
4647 | * a give name and a given sequence number. | |
4648 | * if 'add_backref' is true, also insert a backref from the | |
4649 | * inode to the parent directory. | |
4650 | */ | |
e02119d5 CM |
4651 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4652 | struct inode *parent_inode, struct inode *inode, | |
4653 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4654 | { |
4df27c4d | 4655 | int ret = 0; |
39279cc3 | 4656 | struct btrfs_key key; |
e02119d5 | 4657 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4658 | |
4df27c4d YZ |
4659 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4660 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4661 | } else { | |
4662 | key.objectid = inode->i_ino; | |
4663 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4664 | key.offset = 0; | |
4665 | } | |
4666 | ||
4667 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4668 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4669 | key.objectid, root->root_key.objectid, | |
4670 | parent_inode->i_ino, | |
4671 | index, name, name_len); | |
4672 | } else if (add_backref) { | |
4673 | ret = btrfs_insert_inode_ref(trans, root, | |
4674 | name, name_len, inode->i_ino, | |
4675 | parent_inode->i_ino, index); | |
4676 | } | |
39279cc3 | 4677 | |
39279cc3 | 4678 | if (ret == 0) { |
4df27c4d YZ |
4679 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4680 | parent_inode->i_ino, &key, | |
4681 | btrfs_inode_type(inode), index); | |
4682 | BUG_ON(ret); | |
4683 | ||
dbe674a9 | 4684 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4685 | name_len * 2); |
79c44584 | 4686 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4687 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4688 | } |
4689 | return ret; | |
4690 | } | |
4691 | ||
4692 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4693 | struct inode *dir, struct dentry *dentry, |
4694 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4695 | { |
a1b075d2 JB |
4696 | int err = btrfs_add_link(trans, dir, inode, |
4697 | dentry->d_name.name, dentry->d_name.len, | |
4698 | backref, index); | |
39279cc3 CM |
4699 | if (!err) { |
4700 | d_instantiate(dentry, inode); | |
4701 | return 0; | |
4702 | } | |
4703 | if (err > 0) | |
4704 | err = -EEXIST; | |
4705 | return err; | |
4706 | } | |
4707 | ||
618e21d5 JB |
4708 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4709 | int mode, dev_t rdev) | |
4710 | { | |
4711 | struct btrfs_trans_handle *trans; | |
4712 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4713 | struct inode *inode = NULL; |
618e21d5 JB |
4714 | int err; |
4715 | int drop_inode = 0; | |
4716 | u64 objectid; | |
1832a6d5 | 4717 | unsigned long nr = 0; |
00e4e6b3 | 4718 | u64 index = 0; |
618e21d5 JB |
4719 | |
4720 | if (!new_valid_dev(rdev)) | |
4721 | return -EINVAL; | |
4722 | ||
a22285a6 YZ |
4723 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4724 | if (err) | |
4725 | return err; | |
4726 | ||
9ed74f2d JB |
4727 | /* |
4728 | * 2 for inode item and ref | |
4729 | * 2 for dir items | |
4730 | * 1 for xattr if selinux is on | |
4731 | */ | |
a22285a6 YZ |
4732 | trans = btrfs_start_transaction(root, 5); |
4733 | if (IS_ERR(trans)) | |
4734 | return PTR_ERR(trans); | |
1832a6d5 | 4735 | |
618e21d5 JB |
4736 | btrfs_set_trans_block_group(trans, dir); |
4737 | ||
aec7477b | 4738 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4739 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 | 4740 | BTRFS_I(dir)->block_group, mode, &index); |
618e21d5 JB |
4741 | err = PTR_ERR(inode); |
4742 | if (IS_ERR(inode)) | |
4743 | goto out_unlock; | |
4744 | ||
f34f57a3 | 4745 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4746 | if (err) { |
4747 | drop_inode = 1; | |
4748 | goto out_unlock; | |
4749 | } | |
4750 | ||
618e21d5 | 4751 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4752 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4753 | if (err) |
4754 | drop_inode = 1; | |
4755 | else { | |
4756 | inode->i_op = &btrfs_special_inode_operations; | |
4757 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4758 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4759 | } |
618e21d5 JB |
4760 | btrfs_update_inode_block_group(trans, inode); |
4761 | btrfs_update_inode_block_group(trans, dir); | |
4762 | out_unlock: | |
d3c2fdcf | 4763 | nr = trans->blocks_used; |
89ce8a63 | 4764 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4765 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4766 | if (drop_inode) { |
4767 | inode_dec_link_count(inode); | |
4768 | iput(inode); | |
4769 | } | |
618e21d5 JB |
4770 | return err; |
4771 | } | |
4772 | ||
39279cc3 CM |
4773 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4774 | int mode, struct nameidata *nd) | |
4775 | { | |
4776 | struct btrfs_trans_handle *trans; | |
4777 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4778 | struct inode *inode = NULL; |
39279cc3 | 4779 | int drop_inode = 0; |
a22285a6 | 4780 | int err; |
1832a6d5 | 4781 | unsigned long nr = 0; |
39279cc3 | 4782 | u64 objectid; |
00e4e6b3 | 4783 | u64 index = 0; |
39279cc3 | 4784 | |
a22285a6 YZ |
4785 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4786 | if (err) | |
4787 | return err; | |
9ed74f2d JB |
4788 | /* |
4789 | * 2 for inode item and ref | |
4790 | * 2 for dir items | |
4791 | * 1 for xattr if selinux is on | |
4792 | */ | |
a22285a6 YZ |
4793 | trans = btrfs_start_transaction(root, 5); |
4794 | if (IS_ERR(trans)) | |
4795 | return PTR_ERR(trans); | |
9ed74f2d | 4796 | |
39279cc3 CM |
4797 | btrfs_set_trans_block_group(trans, dir); |
4798 | ||
aec7477b | 4799 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 JB |
4800 | dentry->d_name.len, dir->i_ino, objectid, |
4801 | BTRFS_I(dir)->block_group, mode, &index); | |
39279cc3 CM |
4802 | err = PTR_ERR(inode); |
4803 | if (IS_ERR(inode)) | |
4804 | goto out_unlock; | |
4805 | ||
f34f57a3 | 4806 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4807 | if (err) { |
4808 | drop_inode = 1; | |
4809 | goto out_unlock; | |
4810 | } | |
4811 | ||
39279cc3 | 4812 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4813 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4814 | if (err) |
4815 | drop_inode = 1; | |
4816 | else { | |
4817 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4818 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4819 | inode->i_fop = &btrfs_file_operations; |
4820 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4821 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4822 | } |
39279cc3 CM |
4823 | btrfs_update_inode_block_group(trans, inode); |
4824 | btrfs_update_inode_block_group(trans, dir); | |
4825 | out_unlock: | |
d3c2fdcf | 4826 | nr = trans->blocks_used; |
ab78c84d | 4827 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4828 | if (drop_inode) { |
4829 | inode_dec_link_count(inode); | |
4830 | iput(inode); | |
4831 | } | |
d3c2fdcf | 4832 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4833 | return err; |
4834 | } | |
4835 | ||
4836 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4837 | struct dentry *dentry) | |
4838 | { | |
4839 | struct btrfs_trans_handle *trans; | |
4840 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4841 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4842 | u64 index; |
1832a6d5 | 4843 | unsigned long nr = 0; |
39279cc3 CM |
4844 | int err; |
4845 | int drop_inode = 0; | |
4846 | ||
4847 | if (inode->i_nlink == 0) | |
4848 | return -ENOENT; | |
4849 | ||
4a8be425 TH |
4850 | /* do not allow sys_link's with other subvols of the same device */ |
4851 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4852 | return -EXDEV; |
4a8be425 | 4853 | |
c055e99e AV |
4854 | if (inode->i_nlink == ~0U) |
4855 | return -EMLINK; | |
4856 | ||
9ed74f2d | 4857 | btrfs_inc_nlink(inode); |
bc1cbf1f | 4858 | inode->i_ctime = CURRENT_TIME; |
9ed74f2d | 4859 | |
3de4586c | 4860 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4861 | if (err) |
4862 | goto fail; | |
4863 | ||
a22285a6 | 4864 | /* |
7e6b6465 | 4865 | * 2 items for inode and inode ref |
a22285a6 | 4866 | * 2 items for dir items |
7e6b6465 | 4867 | * 1 item for parent inode |
a22285a6 | 4868 | */ |
7e6b6465 | 4869 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4870 | if (IS_ERR(trans)) { |
4871 | err = PTR_ERR(trans); | |
4872 | goto fail; | |
4873 | } | |
5f39d397 | 4874 | |
39279cc3 | 4875 | btrfs_set_trans_block_group(trans, dir); |
7de9c6ee | 4876 | ihold(inode); |
aec7477b | 4877 | |
a1b075d2 | 4878 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4879 | |
a5719521 | 4880 | if (err) { |
54aa1f4d | 4881 | drop_inode = 1; |
a5719521 | 4882 | } else { |
6a912213 | 4883 | struct dentry *parent = dget_parent(dentry); |
a5719521 YZ |
4884 | btrfs_update_inode_block_group(trans, dir); |
4885 | err = btrfs_update_inode(trans, root, inode); | |
4886 | BUG_ON(err); | |
6a912213 JB |
4887 | btrfs_log_new_name(trans, inode, NULL, parent); |
4888 | dput(parent); | |
a5719521 | 4889 | } |
39279cc3 | 4890 | |
d3c2fdcf | 4891 | nr = trans->blocks_used; |
ab78c84d | 4892 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4893 | fail: |
39279cc3 CM |
4894 | if (drop_inode) { |
4895 | inode_dec_link_count(inode); | |
4896 | iput(inode); | |
4897 | } | |
d3c2fdcf | 4898 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4899 | return err; |
4900 | } | |
4901 | ||
39279cc3 CM |
4902 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4903 | { | |
b9d86667 | 4904 | struct inode *inode = NULL; |
39279cc3 CM |
4905 | struct btrfs_trans_handle *trans; |
4906 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4907 | int err = 0; | |
4908 | int drop_on_err = 0; | |
b9d86667 | 4909 | u64 objectid = 0; |
00e4e6b3 | 4910 | u64 index = 0; |
d3c2fdcf | 4911 | unsigned long nr = 1; |
39279cc3 | 4912 | |
a22285a6 YZ |
4913 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4914 | if (err) | |
4915 | return err; | |
4916 | ||
9ed74f2d JB |
4917 | /* |
4918 | * 2 items for inode and ref | |
4919 | * 2 items for dir items | |
4920 | * 1 for xattr if selinux is on | |
4921 | */ | |
a22285a6 YZ |
4922 | trans = btrfs_start_transaction(root, 5); |
4923 | if (IS_ERR(trans)) | |
4924 | return PTR_ERR(trans); | |
9ed74f2d | 4925 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 4926 | |
aec7477b | 4927 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4928 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
4929 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4930 | &index); | |
39279cc3 CM |
4931 | if (IS_ERR(inode)) { |
4932 | err = PTR_ERR(inode); | |
4933 | goto out_fail; | |
4934 | } | |
5f39d397 | 4935 | |
39279cc3 | 4936 | drop_on_err = 1; |
33268eaf | 4937 | |
f34f57a3 | 4938 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4939 | if (err) |
4940 | goto out_fail; | |
4941 | ||
39279cc3 CM |
4942 | inode->i_op = &btrfs_dir_inode_operations; |
4943 | inode->i_fop = &btrfs_dir_file_operations; | |
4944 | btrfs_set_trans_block_group(trans, inode); | |
4945 | ||
dbe674a9 | 4946 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4947 | err = btrfs_update_inode(trans, root, inode); |
4948 | if (err) | |
4949 | goto out_fail; | |
5f39d397 | 4950 | |
a1b075d2 JB |
4951 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4952 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4953 | if (err) |
4954 | goto out_fail; | |
5f39d397 | 4955 | |
39279cc3 CM |
4956 | d_instantiate(dentry, inode); |
4957 | drop_on_err = 0; | |
39279cc3 CM |
4958 | btrfs_update_inode_block_group(trans, inode); |
4959 | btrfs_update_inode_block_group(trans, dir); | |
4960 | ||
4961 | out_fail: | |
d3c2fdcf | 4962 | nr = trans->blocks_used; |
ab78c84d | 4963 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4964 | if (drop_on_err) |
4965 | iput(inode); | |
d3c2fdcf | 4966 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4967 | return err; |
4968 | } | |
4969 | ||
d352ac68 CM |
4970 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4971 | * and an extent that you want to insert, deal with overlap and insert | |
4972 | * the new extent into the tree. | |
4973 | */ | |
3b951516 CM |
4974 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4975 | struct extent_map *existing, | |
e6dcd2dc CM |
4976 | struct extent_map *em, |
4977 | u64 map_start, u64 map_len) | |
3b951516 CM |
4978 | { |
4979 | u64 start_diff; | |
3b951516 | 4980 | |
e6dcd2dc CM |
4981 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4982 | start_diff = map_start - em->start; | |
4983 | em->start = map_start; | |
4984 | em->len = map_len; | |
c8b97818 CM |
4985 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4986 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4987 | em->block_start += start_diff; |
c8b97818 CM |
4988 | em->block_len -= start_diff; |
4989 | } | |
e6dcd2dc | 4990 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4991 | } |
4992 | ||
c8b97818 CM |
4993 | static noinline int uncompress_inline(struct btrfs_path *path, |
4994 | struct inode *inode, struct page *page, | |
4995 | size_t pg_offset, u64 extent_offset, | |
4996 | struct btrfs_file_extent_item *item) | |
4997 | { | |
4998 | int ret; | |
4999 | struct extent_buffer *leaf = path->nodes[0]; | |
5000 | char *tmp; | |
5001 | size_t max_size; | |
5002 | unsigned long inline_size; | |
5003 | unsigned long ptr; | |
261507a0 | 5004 | int compress_type; |
c8b97818 CM |
5005 | |
5006 | WARN_ON(pg_offset != 0); | |
261507a0 | 5007 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5008 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5009 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5010 | btrfs_item_nr(leaf, path->slots[0])); | |
5011 | tmp = kmalloc(inline_size, GFP_NOFS); | |
5012 | ptr = btrfs_file_extent_inline_start(item); | |
5013 | ||
5014 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5015 | ||
5b050f04 | 5016 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5017 | ret = btrfs_decompress(compress_type, tmp, page, |
5018 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
5019 | if (ret) { |
5020 | char *kaddr = kmap_atomic(page, KM_USER0); | |
5021 | unsigned long copy_size = min_t(u64, | |
5022 | PAGE_CACHE_SIZE - pg_offset, | |
5023 | max_size - extent_offset); | |
5024 | memset(kaddr + pg_offset, 0, copy_size); | |
5025 | kunmap_atomic(kaddr, KM_USER0); | |
5026 | } | |
5027 | kfree(tmp); | |
5028 | return 0; | |
5029 | } | |
5030 | ||
d352ac68 CM |
5031 | /* |
5032 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5033 | * the ugly parts come from merging extents from the disk with the in-ram |
5034 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5035 | * where the in-ram extents might be locked pending data=ordered completion. |
5036 | * | |
5037 | * This also copies inline extents directly into the page. | |
5038 | */ | |
d397712b | 5039 | |
a52d9a80 | 5040 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5041 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5042 | int create) |
5043 | { | |
5044 | int ret; | |
5045 | int err = 0; | |
db94535d | 5046 | u64 bytenr; |
a52d9a80 CM |
5047 | u64 extent_start = 0; |
5048 | u64 extent_end = 0; | |
5049 | u64 objectid = inode->i_ino; | |
5050 | u32 found_type; | |
f421950f | 5051 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5052 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5053 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5054 | struct extent_buffer *leaf; |
5055 | struct btrfs_key found_key; | |
a52d9a80 CM |
5056 | struct extent_map *em = NULL; |
5057 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5058 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5059 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5060 | int compress_type; |
a52d9a80 | 5061 | |
a52d9a80 | 5062 | again: |
890871be | 5063 | read_lock(&em_tree->lock); |
d1310b2e | 5064 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5065 | if (em) |
5066 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5067 | read_unlock(&em_tree->lock); |
d1310b2e | 5068 | |
a52d9a80 | 5069 | if (em) { |
e1c4b745 CM |
5070 | if (em->start > start || em->start + em->len <= start) |
5071 | free_extent_map(em); | |
5072 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5073 | free_extent_map(em); |
5074 | else | |
5075 | goto out; | |
a52d9a80 | 5076 | } |
d1310b2e | 5077 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 5078 | if (!em) { |
d1310b2e CM |
5079 | err = -ENOMEM; |
5080 | goto out; | |
a52d9a80 | 5081 | } |
e6dcd2dc | 5082 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5083 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5084 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5085 | em->len = (u64)-1; |
c8b97818 | 5086 | em->block_len = (u64)-1; |
f421950f CM |
5087 | |
5088 | if (!path) { | |
5089 | path = btrfs_alloc_path(); | |
5090 | BUG_ON(!path); | |
5091 | } | |
5092 | ||
179e29e4 CM |
5093 | ret = btrfs_lookup_file_extent(trans, root, path, |
5094 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5095 | if (ret < 0) { |
5096 | err = ret; | |
5097 | goto out; | |
5098 | } | |
5099 | ||
5100 | if (ret != 0) { | |
5101 | if (path->slots[0] == 0) | |
5102 | goto not_found; | |
5103 | path->slots[0]--; | |
5104 | } | |
5105 | ||
5f39d397 CM |
5106 | leaf = path->nodes[0]; |
5107 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5108 | struct btrfs_file_extent_item); |
a52d9a80 | 5109 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5110 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5111 | found_type = btrfs_key_type(&found_key); | |
5112 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5113 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5114 | goto not_found; | |
5115 | } | |
5116 | ||
5f39d397 CM |
5117 | found_type = btrfs_file_extent_type(leaf, item); |
5118 | extent_start = found_key.offset; | |
261507a0 | 5119 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5120 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5121 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5122 | extent_end = extent_start + |
db94535d | 5123 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5124 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5125 | size_t size; | |
5126 | size = btrfs_file_extent_inline_len(leaf, item); | |
5127 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5128 | ~((u64)root->sectorsize - 1); | |
5129 | } | |
5130 | ||
5131 | if (start >= extent_end) { | |
5132 | path->slots[0]++; | |
5133 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5134 | ret = btrfs_next_leaf(root, path); | |
5135 | if (ret < 0) { | |
5136 | err = ret; | |
5137 | goto out; | |
a52d9a80 | 5138 | } |
9036c102 YZ |
5139 | if (ret > 0) |
5140 | goto not_found; | |
5141 | leaf = path->nodes[0]; | |
a52d9a80 | 5142 | } |
9036c102 YZ |
5143 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5144 | if (found_key.objectid != objectid || | |
5145 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5146 | goto not_found; | |
5147 | if (start + len <= found_key.offset) | |
5148 | goto not_found; | |
5149 | em->start = start; | |
5150 | em->len = found_key.offset - start; | |
5151 | goto not_found_em; | |
5152 | } | |
5153 | ||
d899e052 YZ |
5154 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5155 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5156 | em->start = extent_start; |
5157 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5158 | em->orig_start = extent_start - |
5159 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5160 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5161 | if (bytenr == 0) { | |
5f39d397 | 5162 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5163 | goto insert; |
5164 | } | |
261507a0 | 5165 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5166 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5167 | em->compress_type = compress_type; |
c8b97818 CM |
5168 | em->block_start = bytenr; |
5169 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5170 | item); | |
5171 | } else { | |
5172 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5173 | em->block_start = bytenr; | |
5174 | em->block_len = em->len; | |
d899e052 YZ |
5175 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5176 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5177 | } |
a52d9a80 CM |
5178 | goto insert; |
5179 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5180 | unsigned long ptr; |
a52d9a80 | 5181 | char *map; |
3326d1b0 CM |
5182 | size_t size; |
5183 | size_t extent_offset; | |
5184 | size_t copy_size; | |
a52d9a80 | 5185 | |
689f9346 | 5186 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5187 | if (!page || create) { |
689f9346 | 5188 | em->start = extent_start; |
9036c102 | 5189 | em->len = extent_end - extent_start; |
689f9346 Y |
5190 | goto out; |
5191 | } | |
5f39d397 | 5192 | |
9036c102 YZ |
5193 | size = btrfs_file_extent_inline_len(leaf, item); |
5194 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5195 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5196 | size - extent_offset); |
3326d1b0 | 5197 | em->start = extent_start + extent_offset; |
70dec807 CM |
5198 | em->len = (copy_size + root->sectorsize - 1) & |
5199 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5200 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5201 | if (compress_type) { |
c8b97818 | 5202 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5203 | em->compress_type = compress_type; |
5204 | } | |
689f9346 | 5205 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5206 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5207 | if (btrfs_file_extent_compression(leaf, item) != |
5208 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5209 | ret = uncompress_inline(path, inode, page, |
5210 | pg_offset, | |
5211 | extent_offset, item); | |
5212 | BUG_ON(ret); | |
5213 | } else { | |
5214 | map = kmap(page); | |
5215 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5216 | copy_size); | |
93c82d57 CM |
5217 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5218 | memset(map + pg_offset + copy_size, 0, | |
5219 | PAGE_CACHE_SIZE - pg_offset - | |
5220 | copy_size); | |
5221 | } | |
c8b97818 CM |
5222 | kunmap(page); |
5223 | } | |
179e29e4 CM |
5224 | flush_dcache_page(page); |
5225 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 5226 | WARN_ON(1); |
179e29e4 CM |
5227 | if (!trans) { |
5228 | kunmap(page); | |
5229 | free_extent_map(em); | |
5230 | em = NULL; | |
5231 | btrfs_release_path(root, path); | |
f9295749 | 5232 | trans = btrfs_join_transaction(root, 1); |
3612b495 TI |
5233 | if (IS_ERR(trans)) |
5234 | return ERR_CAST(trans); | |
179e29e4 CM |
5235 | goto again; |
5236 | } | |
c8b97818 | 5237 | map = kmap(page); |
70dec807 | 5238 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5239 | copy_size); |
c8b97818 | 5240 | kunmap(page); |
179e29e4 | 5241 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5242 | } |
d1310b2e | 5243 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5244 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5245 | goto insert; |
5246 | } else { | |
d397712b | 5247 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5248 | WARN_ON(1); |
5249 | } | |
5250 | not_found: | |
5251 | em->start = start; | |
d1310b2e | 5252 | em->len = len; |
a52d9a80 | 5253 | not_found_em: |
5f39d397 | 5254 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5255 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
5256 | insert: |
5257 | btrfs_release_path(root, path); | |
d1310b2e | 5258 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5259 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5260 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5261 | (unsigned long long)em->len, | |
5262 | (unsigned long long)start, | |
5263 | (unsigned long long)len); | |
a52d9a80 CM |
5264 | err = -EIO; |
5265 | goto out; | |
5266 | } | |
d1310b2e CM |
5267 | |
5268 | err = 0; | |
890871be | 5269 | write_lock(&em_tree->lock); |
a52d9a80 | 5270 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5271 | /* it is possible that someone inserted the extent into the tree |
5272 | * while we had the lock dropped. It is also possible that | |
5273 | * an overlapping map exists in the tree | |
5274 | */ | |
a52d9a80 | 5275 | if (ret == -EEXIST) { |
3b951516 | 5276 | struct extent_map *existing; |
e6dcd2dc CM |
5277 | |
5278 | ret = 0; | |
5279 | ||
3b951516 | 5280 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5281 | if (existing && (existing->start > start || |
5282 | existing->start + existing->len <= start)) { | |
5283 | free_extent_map(existing); | |
5284 | existing = NULL; | |
5285 | } | |
3b951516 CM |
5286 | if (!existing) { |
5287 | existing = lookup_extent_mapping(em_tree, em->start, | |
5288 | em->len); | |
5289 | if (existing) { | |
5290 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5291 | em, start, |
5292 | root->sectorsize); | |
3b951516 CM |
5293 | free_extent_map(existing); |
5294 | if (err) { | |
5295 | free_extent_map(em); | |
5296 | em = NULL; | |
5297 | } | |
5298 | } else { | |
5299 | err = -EIO; | |
3b951516 CM |
5300 | free_extent_map(em); |
5301 | em = NULL; | |
5302 | } | |
5303 | } else { | |
5304 | free_extent_map(em); | |
5305 | em = existing; | |
e6dcd2dc | 5306 | err = 0; |
a52d9a80 | 5307 | } |
a52d9a80 | 5308 | } |
890871be | 5309 | write_unlock(&em_tree->lock); |
a52d9a80 | 5310 | out: |
1abe9b8a | 5311 | |
5312 | trace_btrfs_get_extent(root, em); | |
5313 | ||
f421950f CM |
5314 | if (path) |
5315 | btrfs_free_path(path); | |
a52d9a80 CM |
5316 | if (trans) { |
5317 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5318 | if (!err) |
a52d9a80 CM |
5319 | err = ret; |
5320 | } | |
a52d9a80 CM |
5321 | if (err) { |
5322 | free_extent_map(em); | |
a52d9a80 CM |
5323 | return ERR_PTR(err); |
5324 | } | |
5325 | return em; | |
5326 | } | |
5327 | ||
ec29ed5b CM |
5328 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5329 | size_t pg_offset, u64 start, u64 len, | |
5330 | int create) | |
5331 | { | |
5332 | struct extent_map *em; | |
5333 | struct extent_map *hole_em = NULL; | |
5334 | u64 range_start = start; | |
5335 | u64 end; | |
5336 | u64 found; | |
5337 | u64 found_end; | |
5338 | int err = 0; | |
5339 | ||
5340 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5341 | if (IS_ERR(em)) | |
5342 | return em; | |
5343 | if (em) { | |
5344 | /* | |
5345 | * if our em maps to a hole, there might | |
5346 | * actually be delalloc bytes behind it | |
5347 | */ | |
5348 | if (em->block_start != EXTENT_MAP_HOLE) | |
5349 | return em; | |
5350 | else | |
5351 | hole_em = em; | |
5352 | } | |
5353 | ||
5354 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5355 | end = start + len; | |
5356 | if (end < start) | |
5357 | end = (u64)-1; | |
5358 | else | |
5359 | end -= 1; | |
5360 | ||
5361 | em = NULL; | |
5362 | ||
5363 | /* ok, we didn't find anything, lets look for delalloc */ | |
5364 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5365 | end, len, EXTENT_DELALLOC, 1); | |
5366 | found_end = range_start + found; | |
5367 | if (found_end < range_start) | |
5368 | found_end = (u64)-1; | |
5369 | ||
5370 | /* | |
5371 | * we didn't find anything useful, return | |
5372 | * the original results from get_extent() | |
5373 | */ | |
5374 | if (range_start > end || found_end <= start) { | |
5375 | em = hole_em; | |
5376 | hole_em = NULL; | |
5377 | goto out; | |
5378 | } | |
5379 | ||
5380 | /* adjust the range_start to make sure it doesn't | |
5381 | * go backwards from the start they passed in | |
5382 | */ | |
5383 | range_start = max(start,range_start); | |
5384 | found = found_end - range_start; | |
5385 | ||
5386 | if (found > 0) { | |
5387 | u64 hole_start = start; | |
5388 | u64 hole_len = len; | |
5389 | ||
5390 | em = alloc_extent_map(GFP_NOFS); | |
5391 | if (!em) { | |
5392 | err = -ENOMEM; | |
5393 | goto out; | |
5394 | } | |
5395 | /* | |
5396 | * when btrfs_get_extent can't find anything it | |
5397 | * returns one huge hole | |
5398 | * | |
5399 | * make sure what it found really fits our range, and | |
5400 | * adjust to make sure it is based on the start from | |
5401 | * the caller | |
5402 | */ | |
5403 | if (hole_em) { | |
5404 | u64 calc_end = extent_map_end(hole_em); | |
5405 | ||
5406 | if (calc_end <= start || (hole_em->start > end)) { | |
5407 | free_extent_map(hole_em); | |
5408 | hole_em = NULL; | |
5409 | } else { | |
5410 | hole_start = max(hole_em->start, start); | |
5411 | hole_len = calc_end - hole_start; | |
5412 | } | |
5413 | } | |
5414 | em->bdev = NULL; | |
5415 | if (hole_em && range_start > hole_start) { | |
5416 | /* our hole starts before our delalloc, so we | |
5417 | * have to return just the parts of the hole | |
5418 | * that go until the delalloc starts | |
5419 | */ | |
5420 | em->len = min(hole_len, | |
5421 | range_start - hole_start); | |
5422 | em->start = hole_start; | |
5423 | em->orig_start = hole_start; | |
5424 | /* | |
5425 | * don't adjust block start at all, | |
5426 | * it is fixed at EXTENT_MAP_HOLE | |
5427 | */ | |
5428 | em->block_start = hole_em->block_start; | |
5429 | em->block_len = hole_len; | |
5430 | } else { | |
5431 | em->start = range_start; | |
5432 | em->len = found; | |
5433 | em->orig_start = range_start; | |
5434 | em->block_start = EXTENT_MAP_DELALLOC; | |
5435 | em->block_len = found; | |
5436 | } | |
5437 | } else if (hole_em) { | |
5438 | return hole_em; | |
5439 | } | |
5440 | out: | |
5441 | ||
5442 | free_extent_map(hole_em); | |
5443 | if (err) { | |
5444 | free_extent_map(em); | |
5445 | return ERR_PTR(err); | |
5446 | } | |
5447 | return em; | |
5448 | } | |
5449 | ||
4b46fce2 | 5450 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5451 | struct extent_map *em, |
4b46fce2 JB |
5452 | u64 start, u64 len) |
5453 | { | |
5454 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5455 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5456 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5457 | struct btrfs_key ins; | |
5458 | u64 alloc_hint; | |
5459 | int ret; | |
16d299ac | 5460 | bool insert = false; |
4b46fce2 | 5461 | |
16d299ac JB |
5462 | /* |
5463 | * Ok if the extent map we looked up is a hole and is for the exact | |
5464 | * range we want, there is no reason to allocate a new one, however if | |
5465 | * it is not right then we need to free this one and drop the cache for | |
5466 | * our range. | |
5467 | */ | |
5468 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5469 | em->len != len) { | |
5470 | free_extent_map(em); | |
5471 | em = NULL; | |
5472 | insert = true; | |
5473 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5474 | } | |
4b46fce2 JB |
5475 | |
5476 | trans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
5477 | if (IS_ERR(trans)) |
5478 | return ERR_CAST(trans); | |
4b46fce2 JB |
5479 | |
5480 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5481 | ||
5482 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5483 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5484 | alloc_hint, (u64)-1, &ins, 1); | |
5485 | if (ret) { | |
5486 | em = ERR_PTR(ret); | |
5487 | goto out; | |
5488 | } | |
5489 | ||
4b46fce2 | 5490 | if (!em) { |
16d299ac JB |
5491 | em = alloc_extent_map(GFP_NOFS); |
5492 | if (!em) { | |
5493 | em = ERR_PTR(-ENOMEM); | |
5494 | goto out; | |
5495 | } | |
4b46fce2 JB |
5496 | } |
5497 | ||
5498 | em->start = start; | |
5499 | em->orig_start = em->start; | |
5500 | em->len = ins.offset; | |
5501 | ||
5502 | em->block_start = ins.objectid; | |
5503 | em->block_len = ins.offset; | |
5504 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5505 | |
5506 | /* | |
5507 | * We need to do this because if we're using the original em we searched | |
5508 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5509 | */ | |
5510 | em->flags = 0; | |
4b46fce2 JB |
5511 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5512 | ||
16d299ac | 5513 | while (insert) { |
4b46fce2 JB |
5514 | write_lock(&em_tree->lock); |
5515 | ret = add_extent_mapping(em_tree, em); | |
5516 | write_unlock(&em_tree->lock); | |
5517 | if (ret != -EEXIST) | |
5518 | break; | |
5519 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5520 | } | |
5521 | ||
5522 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5523 | ins.offset, ins.offset, 0); | |
5524 | if (ret) { | |
5525 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5526 | em = ERR_PTR(ret); | |
5527 | } | |
5528 | out: | |
5529 | btrfs_end_transaction(trans, root); | |
5530 | return em; | |
5531 | } | |
5532 | ||
46bfbb5c CM |
5533 | /* |
5534 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5535 | * block must be cow'd | |
5536 | */ | |
5537 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5538 | struct inode *inode, u64 offset, u64 len) | |
5539 | { | |
5540 | struct btrfs_path *path; | |
5541 | int ret; | |
5542 | struct extent_buffer *leaf; | |
5543 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5544 | struct btrfs_file_extent_item *fi; | |
5545 | struct btrfs_key key; | |
5546 | u64 disk_bytenr; | |
5547 | u64 backref_offset; | |
5548 | u64 extent_end; | |
5549 | u64 num_bytes; | |
5550 | int slot; | |
5551 | int found_type; | |
5552 | ||
5553 | path = btrfs_alloc_path(); | |
5554 | if (!path) | |
5555 | return -ENOMEM; | |
5556 | ||
5557 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
5558 | offset, 0); | |
5559 | if (ret < 0) | |
5560 | goto out; | |
5561 | ||
5562 | slot = path->slots[0]; | |
5563 | if (ret == 1) { | |
5564 | if (slot == 0) { | |
5565 | /* can't find the item, must cow */ | |
5566 | ret = 0; | |
5567 | goto out; | |
5568 | } | |
5569 | slot--; | |
5570 | } | |
5571 | ret = 0; | |
5572 | leaf = path->nodes[0]; | |
5573 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5574 | if (key.objectid != inode->i_ino || | |
5575 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
5576 | /* not our file or wrong item type, must cow */ | |
5577 | goto out; | |
5578 | } | |
5579 | ||
5580 | if (key.offset > offset) { | |
5581 | /* Wrong offset, must cow */ | |
5582 | goto out; | |
5583 | } | |
5584 | ||
5585 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5586 | found_type = btrfs_file_extent_type(leaf, fi); | |
5587 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5588 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5589 | /* not a regular extent, must cow */ | |
5590 | goto out; | |
5591 | } | |
5592 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5593 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5594 | ||
5595 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5596 | if (extent_end < offset + len) { | |
5597 | /* extent doesn't include our full range, must cow */ | |
5598 | goto out; | |
5599 | } | |
5600 | ||
5601 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5602 | goto out; | |
5603 | ||
5604 | /* | |
5605 | * look for other files referencing this extent, if we | |
5606 | * find any we must cow | |
5607 | */ | |
5608 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, | |
5609 | key.offset - backref_offset, disk_bytenr)) | |
5610 | goto out; | |
5611 | ||
5612 | /* | |
5613 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5614 | * in this extent we are about to write. If there | |
5615 | * are any csums in that range we have to cow in order | |
5616 | * to keep the csums correct | |
5617 | */ | |
5618 | disk_bytenr += backref_offset; | |
5619 | disk_bytenr += offset - key.offset; | |
5620 | num_bytes = min(offset + len, extent_end) - offset; | |
5621 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5622 | goto out; | |
5623 | /* | |
5624 | * all of the above have passed, it is safe to overwrite this extent | |
5625 | * without cow | |
5626 | */ | |
5627 | ret = 1; | |
5628 | out: | |
5629 | btrfs_free_path(path); | |
5630 | return ret; | |
5631 | } | |
5632 | ||
4b46fce2 JB |
5633 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5634 | struct buffer_head *bh_result, int create) | |
5635 | { | |
5636 | struct extent_map *em; | |
5637 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5638 | u64 start = iblock << inode->i_blkbits; | |
5639 | u64 len = bh_result->b_size; | |
46bfbb5c | 5640 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5641 | |
5642 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5643 | if (IS_ERR(em)) | |
5644 | return PTR_ERR(em); | |
5645 | ||
5646 | /* | |
5647 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5648 | * io. INLINE is special, and we could probably kludge it in here, but | |
5649 | * it's still buffered so for safety lets just fall back to the generic | |
5650 | * buffered path. | |
5651 | * | |
5652 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5653 | * decompress it, so there will be buffering required no matter what we | |
5654 | * do, so go ahead and fallback to buffered. | |
5655 | * | |
5656 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5657 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5658 | * the generic code. | |
5659 | */ | |
5660 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5661 | em->block_start == EXTENT_MAP_INLINE) { | |
5662 | free_extent_map(em); | |
5663 | return -ENOTBLK; | |
5664 | } | |
5665 | ||
5666 | /* Just a good old fashioned hole, return */ | |
5667 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5668 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5669 | free_extent_map(em); | |
5670 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5671 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5672 | start + root->sectorsize - 1, GFP_NOFS); | |
5673 | return 0; | |
5674 | } | |
5675 | ||
5676 | /* | |
5677 | * We don't allocate a new extent in the following cases | |
5678 | * | |
5679 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5680 | * existing extent. | |
5681 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5682 | * just use the extent. | |
5683 | * | |
5684 | */ | |
46bfbb5c CM |
5685 | if (!create) { |
5686 | len = em->len - (start - em->start); | |
4b46fce2 | 5687 | goto map; |
46bfbb5c | 5688 | } |
4b46fce2 JB |
5689 | |
5690 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5691 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5692 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5693 | int type; |
5694 | int ret; | |
46bfbb5c | 5695 | u64 block_start; |
4b46fce2 JB |
5696 | |
5697 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5698 | type = BTRFS_ORDERED_PREALLOC; | |
5699 | else | |
5700 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5701 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5702 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5703 | |
5704 | /* | |
5705 | * we're not going to log anything, but we do need | |
5706 | * to make sure the current transaction stays open | |
5707 | * while we look for nocow cross refs | |
5708 | */ | |
5709 | trans = btrfs_join_transaction(root, 0); | |
3612b495 | 5710 | if (IS_ERR(trans)) |
46bfbb5c CM |
5711 | goto must_cow; |
5712 | ||
5713 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5714 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5715 | block_start, len, len, type); | |
5716 | btrfs_end_transaction(trans, root); | |
5717 | if (ret) { | |
5718 | free_extent_map(em); | |
5719 | return ret; | |
5720 | } | |
5721 | goto unlock; | |
4b46fce2 | 5722 | } |
46bfbb5c | 5723 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5724 | } |
46bfbb5c CM |
5725 | must_cow: |
5726 | /* | |
5727 | * this will cow the extent, reset the len in case we changed | |
5728 | * it above | |
5729 | */ | |
5730 | len = bh_result->b_size; | |
16d299ac | 5731 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5732 | if (IS_ERR(em)) |
5733 | return PTR_ERR(em); | |
5734 | len = min(len, em->len - (start - em->start)); | |
5735 | unlock: | |
4845e44f CM |
5736 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5737 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5738 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5739 | map: |
5740 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5741 | inode->i_blkbits; | |
46bfbb5c | 5742 | bh_result->b_size = len; |
4b46fce2 JB |
5743 | bh_result->b_bdev = em->bdev; |
5744 | set_buffer_mapped(bh_result); | |
5745 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5746 | set_buffer_new(bh_result); | |
5747 | ||
5748 | free_extent_map(em); | |
5749 | ||
5750 | return 0; | |
5751 | } | |
5752 | ||
5753 | struct btrfs_dio_private { | |
5754 | struct inode *inode; | |
5755 | u64 logical_offset; | |
5756 | u64 disk_bytenr; | |
5757 | u64 bytes; | |
5758 | u32 *csums; | |
5759 | void *private; | |
e65e1535 MX |
5760 | |
5761 | /* number of bios pending for this dio */ | |
5762 | atomic_t pending_bios; | |
5763 | ||
5764 | /* IO errors */ | |
5765 | int errors; | |
5766 | ||
5767 | struct bio *orig_bio; | |
4b46fce2 JB |
5768 | }; |
5769 | ||
5770 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5771 | { | |
e65e1535 | 5772 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5773 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5774 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5775 | struct inode *inode = dip->inode; |
5776 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5777 | u64 start; | |
5778 | u32 *private = dip->csums; | |
5779 | ||
5780 | start = dip->logical_offset; | |
5781 | do { | |
5782 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5783 | struct page *page = bvec->bv_page; | |
5784 | char *kaddr; | |
5785 | u32 csum = ~(u32)0; | |
5786 | unsigned long flags; | |
5787 | ||
5788 | local_irq_save(flags); | |
5789 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5790 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5791 | csum, bvec->bv_len); | |
5792 | btrfs_csum_final(csum, (char *)&csum); | |
5793 | kunmap_atomic(kaddr, KM_IRQ0); | |
5794 | local_irq_restore(flags); | |
5795 | ||
5796 | flush_dcache_page(bvec->bv_page); | |
5797 | if (csum != *private) { | |
5798 | printk(KERN_ERR "btrfs csum failed ino %lu off" | |
5799 | " %llu csum %u private %u\n", | |
5800 | inode->i_ino, (unsigned long long)start, | |
5801 | csum, *private); | |
5802 | err = -EIO; | |
5803 | } | |
5804 | } | |
5805 | ||
5806 | start += bvec->bv_len; | |
5807 | private++; | |
5808 | bvec++; | |
5809 | } while (bvec <= bvec_end); | |
5810 | ||
5811 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5812 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5813 | bio->bi_private = dip->private; | |
5814 | ||
5815 | kfree(dip->csums); | |
5816 | kfree(dip); | |
c0da7aa1 JB |
5817 | |
5818 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5819 | if (err) | |
5820 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5821 | dio_end_io(bio, err); |
5822 | } | |
5823 | ||
5824 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5825 | { | |
5826 | struct btrfs_dio_private *dip = bio->bi_private; | |
5827 | struct inode *inode = dip->inode; | |
5828 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5829 | struct btrfs_trans_handle *trans; | |
5830 | struct btrfs_ordered_extent *ordered = NULL; | |
5831 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5832 | u64 ordered_offset = dip->logical_offset; |
5833 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5834 | int ret; |
5835 | ||
5836 | if (err) | |
5837 | goto out_done; | |
163cf09c CM |
5838 | again: |
5839 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5840 | &ordered_offset, | |
5841 | ordered_bytes); | |
4b46fce2 | 5842 | if (!ret) |
163cf09c | 5843 | goto out_test; |
4b46fce2 JB |
5844 | |
5845 | BUG_ON(!ordered); | |
5846 | ||
5847 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 5848 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5849 | err = -ENOMEM; |
5850 | goto out; | |
5851 | } | |
5852 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5853 | ||
5854 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5855 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5856 | if (!ret) | |
5857 | ret = btrfs_update_inode(trans, root, inode); | |
5858 | err = ret; | |
5859 | goto out; | |
5860 | } | |
5861 | ||
5862 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5863 | ordered->file_offset + ordered->len - 1, 0, | |
5864 | &cached_state, GFP_NOFS); | |
5865 | ||
5866 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5867 | ret = btrfs_mark_extent_written(trans, inode, | |
5868 | ordered->file_offset, | |
5869 | ordered->file_offset + | |
5870 | ordered->len); | |
5871 | if (ret) { | |
5872 | err = ret; | |
5873 | goto out_unlock; | |
5874 | } | |
5875 | } else { | |
5876 | ret = insert_reserved_file_extent(trans, inode, | |
5877 | ordered->file_offset, | |
5878 | ordered->start, | |
5879 | ordered->disk_len, | |
5880 | ordered->len, | |
5881 | ordered->len, | |
5882 | 0, 0, 0, | |
5883 | BTRFS_FILE_EXTENT_REG); | |
5884 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5885 | ordered->file_offset, ordered->len); | |
5886 | if (ret) { | |
5887 | err = ret; | |
5888 | WARN_ON(1); | |
5889 | goto out_unlock; | |
5890 | } | |
5891 | } | |
5892 | ||
5893 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
1ef30be1 JB |
5894 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); |
5895 | if (!ret) | |
5896 | btrfs_update_inode(trans, root, inode); | |
5897 | ret = 0; | |
4b46fce2 JB |
5898 | out_unlock: |
5899 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5900 | ordered->file_offset + ordered->len - 1, | |
5901 | &cached_state, GFP_NOFS); | |
5902 | out: | |
5903 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5904 | btrfs_end_transaction(trans, root); | |
163cf09c | 5905 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5906 | btrfs_put_ordered_extent(ordered); |
5907 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5908 | |
5909 | out_test: | |
5910 | /* | |
5911 | * our bio might span multiple ordered extents. If we haven't | |
5912 | * completed the accounting for the whole dio, go back and try again | |
5913 | */ | |
5914 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5915 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5916 | ordered_offset; | |
5917 | goto again; | |
5918 | } | |
4b46fce2 JB |
5919 | out_done: |
5920 | bio->bi_private = dip->private; | |
5921 | ||
5922 | kfree(dip->csums); | |
5923 | kfree(dip); | |
c0da7aa1 JB |
5924 | |
5925 | /* If we had an error make sure to clear the uptodate flag */ | |
5926 | if (err) | |
5927 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5928 | dio_end_io(bio, err); |
5929 | } | |
5930 | ||
eaf25d93 CM |
5931 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5932 | struct bio *bio, int mirror_num, | |
5933 | unsigned long bio_flags, u64 offset) | |
5934 | { | |
5935 | int ret; | |
5936 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5937 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5938 | BUG_ON(ret); | |
5939 | return 0; | |
5940 | } | |
5941 | ||
e65e1535 MX |
5942 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5943 | { | |
5944 | struct btrfs_dio_private *dip = bio->bi_private; | |
5945 | ||
5946 | if (err) { | |
5947 | printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu " | |
3dd1462e JB |
5948 | "sector %#Lx len %u err no %d\n", |
5949 | dip->inode->i_ino, bio->bi_rw, | |
5950 | (unsigned long long)bio->bi_sector, bio->bi_size, err); | |
e65e1535 MX |
5951 | dip->errors = 1; |
5952 | ||
5953 | /* | |
5954 | * before atomic variable goto zero, we must make sure | |
5955 | * dip->errors is perceived to be set. | |
5956 | */ | |
5957 | smp_mb__before_atomic_dec(); | |
5958 | } | |
5959 | ||
5960 | /* if there are more bios still pending for this dio, just exit */ | |
5961 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5962 | goto out; | |
5963 | ||
5964 | if (dip->errors) | |
5965 | bio_io_error(dip->orig_bio); | |
5966 | else { | |
5967 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5968 | bio_endio(dip->orig_bio, 0); | |
5969 | } | |
5970 | out: | |
5971 | bio_put(bio); | |
5972 | } | |
5973 | ||
5974 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5975 | u64 first_sector, gfp_t gfp_flags) | |
5976 | { | |
5977 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5978 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5979 | } | |
5980 | ||
5981 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5982 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 5983 | u32 *csums, int async_submit) |
e65e1535 MX |
5984 | { |
5985 | int write = rw & REQ_WRITE; | |
5986 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5987 | int ret; | |
5988 | ||
5989 | bio_get(bio); | |
5990 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5991 | if (ret) | |
5992 | goto err; | |
5993 | ||
1ae39938 JB |
5994 | if (skip_sum) |
5995 | goto map; | |
5996 | ||
5997 | if (write && async_submit) { | |
e65e1535 MX |
5998 | ret = btrfs_wq_submit_bio(root->fs_info, |
5999 | inode, rw, bio, 0, 0, | |
6000 | file_offset, | |
6001 | __btrfs_submit_bio_start_direct_io, | |
6002 | __btrfs_submit_bio_done); | |
6003 | goto err; | |
1ae39938 JB |
6004 | } else if (write) { |
6005 | /* | |
6006 | * If we aren't doing async submit, calculate the csum of the | |
6007 | * bio now. | |
6008 | */ | |
6009 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6010 | if (ret) | |
6011 | goto err; | |
c2db1073 TI |
6012 | } else if (!skip_sum) { |
6013 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 6014 | file_offset, csums); |
c2db1073 TI |
6015 | if (ret) |
6016 | goto err; | |
6017 | } | |
e65e1535 | 6018 | |
1ae39938 JB |
6019 | map: |
6020 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6021 | err: |
6022 | bio_put(bio); | |
6023 | return ret; | |
6024 | } | |
6025 | ||
6026 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6027 | int skip_sum) | |
6028 | { | |
6029 | struct inode *inode = dip->inode; | |
6030 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6031 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6032 | struct bio *bio; | |
6033 | struct bio *orig_bio = dip->orig_bio; | |
6034 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6035 | u64 start_sector = orig_bio->bi_sector; | |
6036 | u64 file_offset = dip->logical_offset; | |
6037 | u64 submit_len = 0; | |
6038 | u64 map_length; | |
6039 | int nr_pages = 0; | |
6040 | u32 *csums = dip->csums; | |
6041 | int ret = 0; | |
1ae39938 | 6042 | int async_submit = 0; |
98bc3149 | 6043 | int write = rw & REQ_WRITE; |
e65e1535 | 6044 | |
e65e1535 MX |
6045 | map_length = orig_bio->bi_size; |
6046 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6047 | &map_length, NULL, 0); | |
6048 | if (ret) { | |
6049 | bio_put(bio); | |
6050 | return -EIO; | |
6051 | } | |
6052 | ||
02f57c7a JB |
6053 | if (map_length >= orig_bio->bi_size) { |
6054 | bio = orig_bio; | |
6055 | goto submit; | |
6056 | } | |
6057 | ||
1ae39938 | 6058 | async_submit = 1; |
02f57c7a JB |
6059 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6060 | if (!bio) | |
6061 | return -ENOMEM; | |
6062 | bio->bi_private = dip; | |
6063 | bio->bi_end_io = btrfs_end_dio_bio; | |
6064 | atomic_inc(&dip->pending_bios); | |
6065 | ||
e65e1535 MX |
6066 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6067 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6068 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6069 | bvec->bv_offset) < bvec->bv_len)) { | |
6070 | /* | |
6071 | * inc the count before we submit the bio so | |
6072 | * we know the end IO handler won't happen before | |
6073 | * we inc the count. Otherwise, the dip might get freed | |
6074 | * before we're done setting it up | |
6075 | */ | |
6076 | atomic_inc(&dip->pending_bios); | |
6077 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6078 | file_offset, skip_sum, | |
1ae39938 | 6079 | csums, async_submit); |
e65e1535 MX |
6080 | if (ret) { |
6081 | bio_put(bio); | |
6082 | atomic_dec(&dip->pending_bios); | |
6083 | goto out_err; | |
6084 | } | |
6085 | ||
98bc3149 JB |
6086 | /* Write's use the ordered csums */ |
6087 | if (!write && !skip_sum) | |
e65e1535 MX |
6088 | csums = csums + nr_pages; |
6089 | start_sector += submit_len >> 9; | |
6090 | file_offset += submit_len; | |
6091 | ||
6092 | submit_len = 0; | |
6093 | nr_pages = 0; | |
6094 | ||
6095 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6096 | start_sector, GFP_NOFS); | |
6097 | if (!bio) | |
6098 | goto out_err; | |
6099 | bio->bi_private = dip; | |
6100 | bio->bi_end_io = btrfs_end_dio_bio; | |
6101 | ||
6102 | map_length = orig_bio->bi_size; | |
6103 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6104 | &map_length, NULL, 0); | |
6105 | if (ret) { | |
6106 | bio_put(bio); | |
6107 | goto out_err; | |
6108 | } | |
6109 | } else { | |
6110 | submit_len += bvec->bv_len; | |
6111 | nr_pages ++; | |
6112 | bvec++; | |
6113 | } | |
6114 | } | |
6115 | ||
02f57c7a | 6116 | submit: |
e65e1535 | 6117 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6118 | csums, async_submit); |
e65e1535 MX |
6119 | if (!ret) |
6120 | return 0; | |
6121 | ||
6122 | bio_put(bio); | |
6123 | out_err: | |
6124 | dip->errors = 1; | |
6125 | /* | |
6126 | * before atomic variable goto zero, we must | |
6127 | * make sure dip->errors is perceived to be set. | |
6128 | */ | |
6129 | smp_mb__before_atomic_dec(); | |
6130 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6131 | bio_io_error(dip->orig_bio); | |
6132 | ||
6133 | /* bio_end_io() will handle error, so we needn't return it */ | |
6134 | return 0; | |
6135 | } | |
6136 | ||
4b46fce2 JB |
6137 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6138 | loff_t file_offset) | |
6139 | { | |
6140 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6141 | struct btrfs_dio_private *dip; | |
6142 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6143 | int skip_sum; |
7b6d91da | 6144 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6145 | int ret = 0; |
6146 | ||
6147 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6148 | ||
6149 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6150 | if (!dip) { | |
6151 | ret = -ENOMEM; | |
6152 | goto free_ordered; | |
6153 | } | |
6154 | dip->csums = NULL; | |
6155 | ||
98bc3149 JB |
6156 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6157 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6158 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6159 | if (!dip->csums) { | |
b4966b77 | 6160 | kfree(dip); |
4b46fce2 JB |
6161 | ret = -ENOMEM; |
6162 | goto free_ordered; | |
6163 | } | |
6164 | } | |
6165 | ||
6166 | dip->private = bio->bi_private; | |
6167 | dip->inode = inode; | |
6168 | dip->logical_offset = file_offset; | |
6169 | ||
4b46fce2 JB |
6170 | dip->bytes = 0; |
6171 | do { | |
6172 | dip->bytes += bvec->bv_len; | |
6173 | bvec++; | |
6174 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6175 | ||
46bfbb5c | 6176 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6177 | bio->bi_private = dip; |
e65e1535 MX |
6178 | dip->errors = 0; |
6179 | dip->orig_bio = bio; | |
6180 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6181 | |
6182 | if (write) | |
6183 | bio->bi_end_io = btrfs_endio_direct_write; | |
6184 | else | |
6185 | bio->bi_end_io = btrfs_endio_direct_read; | |
6186 | ||
e65e1535 MX |
6187 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6188 | if (!ret) | |
eaf25d93 | 6189 | return; |
4b46fce2 JB |
6190 | free_ordered: |
6191 | /* | |
6192 | * If this is a write, we need to clean up the reserved space and kill | |
6193 | * the ordered extent. | |
6194 | */ | |
6195 | if (write) { | |
6196 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6197 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6198 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6199 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6200 | btrfs_free_reserved_extent(root, ordered->start, | |
6201 | ordered->disk_len); | |
6202 | btrfs_put_ordered_extent(ordered); | |
6203 | btrfs_put_ordered_extent(ordered); | |
6204 | } | |
6205 | bio_endio(bio, ret); | |
6206 | } | |
6207 | ||
5a5f79b5 CM |
6208 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6209 | const struct iovec *iov, loff_t offset, | |
6210 | unsigned long nr_segs) | |
6211 | { | |
6212 | int seg; | |
a1b75f7d | 6213 | int i; |
5a5f79b5 CM |
6214 | size_t size; |
6215 | unsigned long addr; | |
6216 | unsigned blocksize_mask = root->sectorsize - 1; | |
6217 | ssize_t retval = -EINVAL; | |
6218 | loff_t end = offset; | |
6219 | ||
6220 | if (offset & blocksize_mask) | |
6221 | goto out; | |
6222 | ||
6223 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6224 | for (seg = 0; seg < nr_segs; seg++) { | |
6225 | addr = (unsigned long)iov[seg].iov_base; | |
6226 | size = iov[seg].iov_len; | |
6227 | end += size; | |
a1b75f7d | 6228 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6229 | goto out; |
a1b75f7d JB |
6230 | |
6231 | /* If this is a write we don't need to check anymore */ | |
6232 | if (rw & WRITE) | |
6233 | continue; | |
6234 | ||
6235 | /* | |
6236 | * Check to make sure we don't have duplicate iov_base's in this | |
6237 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6238 | * when reading back. | |
6239 | */ | |
6240 | for (i = seg + 1; i < nr_segs; i++) { | |
6241 | if (iov[seg].iov_base == iov[i].iov_base) | |
6242 | goto out; | |
6243 | } | |
5a5f79b5 CM |
6244 | } |
6245 | retval = 0; | |
6246 | out: | |
6247 | return retval; | |
6248 | } | |
16432985 CM |
6249 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6250 | const struct iovec *iov, loff_t offset, | |
6251 | unsigned long nr_segs) | |
6252 | { | |
4b46fce2 JB |
6253 | struct file *file = iocb->ki_filp; |
6254 | struct inode *inode = file->f_mapping->host; | |
6255 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6256 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6257 | u64 lockstart, lockend; |
6258 | ssize_t ret; | |
4845e44f CM |
6259 | int writing = rw & WRITE; |
6260 | int write_bits = 0; | |
3f7c579c | 6261 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6262 | |
5a5f79b5 CM |
6263 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6264 | offset, nr_segs)) { | |
6265 | return 0; | |
6266 | } | |
6267 | ||
4b46fce2 | 6268 | lockstart = offset; |
3f7c579c CM |
6269 | lockend = offset + count - 1; |
6270 | ||
6271 | if (writing) { | |
6272 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6273 | if (ret) | |
6274 | goto out; | |
6275 | } | |
4845e44f | 6276 | |
4b46fce2 | 6277 | while (1) { |
4845e44f CM |
6278 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6279 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6280 | /* |
6281 | * We're concerned with the entire range that we're going to be | |
6282 | * doing DIO to, so we need to make sure theres no ordered | |
6283 | * extents in this range. | |
6284 | */ | |
6285 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6286 | lockend - lockstart + 1); | |
6287 | if (!ordered) | |
6288 | break; | |
4845e44f CM |
6289 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6290 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6291 | btrfs_start_ordered_extent(inode, ordered, 1); |
6292 | btrfs_put_ordered_extent(ordered); | |
6293 | cond_resched(); | |
6294 | } | |
6295 | ||
4845e44f CM |
6296 | /* |
6297 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6298 | * the dirty or uptodate bits | |
6299 | */ | |
6300 | if (writing) { | |
6301 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6302 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6303 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6304 | GFP_NOFS); | |
6305 | if (ret) { | |
6306 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6307 | lockend, EXTENT_LOCKED | write_bits, | |
6308 | 1, 0, &cached_state, GFP_NOFS); | |
6309 | goto out; | |
6310 | } | |
6311 | } | |
6312 | ||
6313 | free_extent_state(cached_state); | |
6314 | cached_state = NULL; | |
6315 | ||
5a5f79b5 CM |
6316 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6317 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6318 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6319 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6320 | |
6321 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6322 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6323 | offset + iov_length(iov, nr_segs) - 1, | |
6324 | EXTENT_LOCKED | write_bits, 1, 0, | |
6325 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6326 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6327 | /* | |
6328 | * We're falling back to buffered, unlock the section we didn't | |
6329 | * do IO on. | |
6330 | */ | |
4845e44f CM |
6331 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6332 | offset + iov_length(iov, nr_segs) - 1, | |
6333 | EXTENT_LOCKED | write_bits, 1, 0, | |
6334 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6335 | } |
4845e44f CM |
6336 | out: |
6337 | free_extent_state(cached_state); | |
4b46fce2 | 6338 | return ret; |
16432985 CM |
6339 | } |
6340 | ||
1506fcc8 YS |
6341 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6342 | __u64 start, __u64 len) | |
6343 | { | |
ec29ed5b | 6344 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6345 | } |
6346 | ||
a52d9a80 | 6347 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6348 | { |
d1310b2e CM |
6349 | struct extent_io_tree *tree; |
6350 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 6351 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 6352 | } |
1832a6d5 | 6353 | |
a52d9a80 | 6354 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6355 | { |
d1310b2e | 6356 | struct extent_io_tree *tree; |
b888db2b CM |
6357 | |
6358 | ||
6359 | if (current->flags & PF_MEMALLOC) { | |
6360 | redirty_page_for_writepage(wbc, page); | |
6361 | unlock_page(page); | |
6362 | return 0; | |
6363 | } | |
d1310b2e | 6364 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6365 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6366 | } |
6367 | ||
f421950f CM |
6368 | int btrfs_writepages(struct address_space *mapping, |
6369 | struct writeback_control *wbc) | |
b293f02e | 6370 | { |
d1310b2e | 6371 | struct extent_io_tree *tree; |
771ed689 | 6372 | |
d1310b2e | 6373 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6374 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6375 | } | |
6376 | ||
3ab2fb5a CM |
6377 | static int |
6378 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6379 | struct list_head *pages, unsigned nr_pages) | |
6380 | { | |
d1310b2e CM |
6381 | struct extent_io_tree *tree; |
6382 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6383 | return extent_readpages(tree, mapping, pages, nr_pages, |
6384 | btrfs_get_extent); | |
6385 | } | |
e6dcd2dc | 6386 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6387 | { |
d1310b2e CM |
6388 | struct extent_io_tree *tree; |
6389 | struct extent_map_tree *map; | |
a52d9a80 | 6390 | int ret; |
8c2383c3 | 6391 | |
d1310b2e CM |
6392 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6393 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6394 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6395 | if (ret == 1) { |
6396 | ClearPagePrivate(page); | |
6397 | set_page_private(page, 0); | |
6398 | page_cache_release(page); | |
39279cc3 | 6399 | } |
a52d9a80 | 6400 | return ret; |
39279cc3 CM |
6401 | } |
6402 | ||
e6dcd2dc CM |
6403 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6404 | { | |
98509cfc CM |
6405 | if (PageWriteback(page) || PageDirty(page)) |
6406 | return 0; | |
b335b003 | 6407 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6408 | } |
6409 | ||
a52d9a80 | 6410 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6411 | { |
d1310b2e | 6412 | struct extent_io_tree *tree; |
e6dcd2dc | 6413 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6414 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6415 | u64 page_start = page_offset(page); |
6416 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6417 | |
8b62b72b CM |
6418 | |
6419 | /* | |
6420 | * we have the page locked, so new writeback can't start, | |
6421 | * and the dirty bit won't be cleared while we are here. | |
6422 | * | |
6423 | * Wait for IO on this page so that we can safely clear | |
6424 | * the PagePrivate2 bit and do ordered accounting | |
6425 | */ | |
e6dcd2dc | 6426 | wait_on_page_writeback(page); |
8b62b72b | 6427 | |
d1310b2e | 6428 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6429 | if (offset) { |
6430 | btrfs_releasepage(page, GFP_NOFS); | |
6431 | return; | |
6432 | } | |
2ac55d41 JB |
6433 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6434 | GFP_NOFS); | |
e6dcd2dc CM |
6435 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6436 | page_offset(page)); | |
6437 | if (ordered) { | |
eb84ae03 CM |
6438 | /* |
6439 | * IO on this page will never be started, so we need | |
6440 | * to account for any ordered extents now | |
6441 | */ | |
e6dcd2dc CM |
6442 | clear_extent_bit(tree, page_start, page_end, |
6443 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6444 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6445 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6446 | /* |
6447 | * whoever cleared the private bit is responsible | |
6448 | * for the finish_ordered_io | |
6449 | */ | |
6450 | if (TestClearPagePrivate2(page)) { | |
6451 | btrfs_finish_ordered_io(page->mapping->host, | |
6452 | page_start, page_end); | |
6453 | } | |
e6dcd2dc | 6454 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6455 | cached_state = NULL; |
6456 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6457 | GFP_NOFS); | |
e6dcd2dc CM |
6458 | } |
6459 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6460 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6461 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6462 | __btrfs_releasepage(page, GFP_NOFS); |
6463 | ||
4a096752 | 6464 | ClearPageChecked(page); |
9ad6b7bc | 6465 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6466 | ClearPagePrivate(page); |
6467 | set_page_private(page, 0); | |
6468 | page_cache_release(page); | |
6469 | } | |
39279cc3 CM |
6470 | } |
6471 | ||
9ebefb18 CM |
6472 | /* |
6473 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6474 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6475 | * be careful to check for EOF conditions here. We set the page up correctly | |
6476 | * for a written page which means we get ENOSPC checking when writing into | |
6477 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6478 | * support these features. | |
6479 | * | |
6480 | * We are not allowed to take the i_mutex here so we have to play games to | |
6481 | * protect against truncate races as the page could now be beyond EOF. Because | |
6482 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6483 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6484 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6485 | * unlock the page. | |
6486 | */ | |
c2ec175c | 6487 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6488 | { |
c2ec175c | 6489 | struct page *page = vmf->page; |
6da6abae | 6490 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6491 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6492 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6493 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6494 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6495 | char *kaddr; |
6496 | unsigned long zero_start; | |
9ebefb18 | 6497 | loff_t size; |
1832a6d5 | 6498 | int ret; |
a52d9a80 | 6499 | u64 page_start; |
e6dcd2dc | 6500 | u64 page_end; |
9ebefb18 | 6501 | |
0ca1f7ce | 6502 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
6503 | if (ret) { |
6504 | if (ret == -ENOMEM) | |
6505 | ret = VM_FAULT_OOM; | |
6506 | else /* -ENOSPC, -EIO, etc */ | |
6507 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 6508 | goto out; |
56a76f82 | 6509 | } |
1832a6d5 | 6510 | |
56a76f82 | 6511 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6512 | again: |
9ebefb18 | 6513 | lock_page(page); |
9ebefb18 | 6514 | size = i_size_read(inode); |
e6dcd2dc CM |
6515 | page_start = page_offset(page); |
6516 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6517 | |
9ebefb18 | 6518 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6519 | (page_start >= size)) { |
9ebefb18 CM |
6520 | /* page got truncated out from underneath us */ |
6521 | goto out_unlock; | |
6522 | } | |
e6dcd2dc CM |
6523 | wait_on_page_writeback(page); |
6524 | ||
2ac55d41 JB |
6525 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6526 | GFP_NOFS); | |
e6dcd2dc CM |
6527 | set_page_extent_mapped(page); |
6528 | ||
eb84ae03 CM |
6529 | /* |
6530 | * we can't set the delalloc bits if there are pending ordered | |
6531 | * extents. Drop our locks and wait for them to finish | |
6532 | */ | |
e6dcd2dc CM |
6533 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6534 | if (ordered) { | |
2ac55d41 JB |
6535 | unlock_extent_cached(io_tree, page_start, page_end, |
6536 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6537 | unlock_page(page); |
eb84ae03 | 6538 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6539 | btrfs_put_ordered_extent(ordered); |
6540 | goto again; | |
6541 | } | |
6542 | ||
fbf19087 JB |
6543 | /* |
6544 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6545 | * if it was already dirty, so for space accounting reasons we need to | |
6546 | * clear any delalloc bits for the range we are fixing to save. There | |
6547 | * is probably a better way to do this, but for now keep consistent with | |
6548 | * prepare_pages in the normal write path. | |
6549 | */ | |
2ac55d41 | 6550 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6551 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6552 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6553 | |
2ac55d41 JB |
6554 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6555 | &cached_state); | |
9ed74f2d | 6556 | if (ret) { |
2ac55d41 JB |
6557 | unlock_extent_cached(io_tree, page_start, page_end, |
6558 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6559 | ret = VM_FAULT_SIGBUS; |
6560 | goto out_unlock; | |
6561 | } | |
e6dcd2dc | 6562 | ret = 0; |
9ebefb18 CM |
6563 | |
6564 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6565 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6566 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6567 | else |
e6dcd2dc | 6568 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6569 | |
e6dcd2dc CM |
6570 | if (zero_start != PAGE_CACHE_SIZE) { |
6571 | kaddr = kmap(page); | |
6572 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6573 | flush_dcache_page(page); | |
6574 | kunmap(page); | |
6575 | } | |
247e743c | 6576 | ClearPageChecked(page); |
e6dcd2dc | 6577 | set_page_dirty(page); |
50a9b214 | 6578 | SetPageUptodate(page); |
5a3f23d5 | 6579 | |
257c62e1 CM |
6580 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6581 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6582 | ||
2ac55d41 | 6583 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6584 | |
6585 | out_unlock: | |
50a9b214 CM |
6586 | if (!ret) |
6587 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6588 | unlock_page(page); |
0ca1f7ce | 6589 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 6590 | out: |
9ebefb18 CM |
6591 | return ret; |
6592 | } | |
6593 | ||
a41ad394 | 6594 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6595 | { |
6596 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6597 | int ret; | |
3893e33b | 6598 | int err = 0; |
39279cc3 | 6599 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6600 | unsigned long nr; |
dbe674a9 | 6601 | u64 mask = root->sectorsize - 1; |
39279cc3 | 6602 | |
5d5e103a JB |
6603 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6604 | if (ret) | |
a41ad394 | 6605 | return ret; |
8082510e | 6606 | |
4a096752 | 6607 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6608 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6609 | |
f0cd846e JB |
6610 | trans = btrfs_start_transaction(root, 5); |
6611 | if (IS_ERR(trans)) | |
6612 | return PTR_ERR(trans); | |
6613 | ||
6614 | btrfs_set_trans_block_group(trans, inode); | |
6615 | ||
6616 | ret = btrfs_orphan_add(trans, inode); | |
6617 | if (ret) { | |
6618 | btrfs_end_transaction(trans, root); | |
6619 | return ret; | |
6620 | } | |
6621 | ||
6622 | nr = trans->blocks_used; | |
6623 | btrfs_end_transaction(trans, root); | |
6624 | btrfs_btree_balance_dirty(root, nr); | |
6625 | ||
6626 | /* Now start a transaction for the truncate */ | |
d68fc57b | 6627 | trans = btrfs_start_transaction(root, 0); |
3893e33b JB |
6628 | if (IS_ERR(trans)) |
6629 | return PTR_ERR(trans); | |
8082510e | 6630 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b | 6631 | trans->block_rsv = root->orphan_block_rsv; |
5a3f23d5 CM |
6632 | |
6633 | /* | |
6634 | * setattr is responsible for setting the ordered_data_close flag, | |
6635 | * but that is only tested during the last file release. That | |
6636 | * could happen well after the next commit, leaving a great big | |
6637 | * window where new writes may get lost if someone chooses to write | |
6638 | * to this file after truncating to zero | |
6639 | * | |
6640 | * The inode doesn't have any dirty data here, and so if we commit | |
6641 | * this is a noop. If someone immediately starts writing to the inode | |
6642 | * it is very likely we'll catch some of their writes in this | |
6643 | * transaction, and the commit will find this file on the ordered | |
6644 | * data list with good things to send down. | |
6645 | * | |
6646 | * This is a best effort solution, there is still a window where | |
6647 | * using truncate to replace the contents of the file will | |
6648 | * end up with a zero length file after a crash. | |
6649 | */ | |
6650 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6651 | btrfs_add_ordered_operation(trans, root, inode); | |
6652 | ||
8082510e | 6653 | while (1) { |
d68fc57b YZ |
6654 | if (!trans) { |
6655 | trans = btrfs_start_transaction(root, 0); | |
3893e33b JB |
6656 | if (IS_ERR(trans)) |
6657 | return PTR_ERR(trans); | |
d68fc57b YZ |
6658 | btrfs_set_trans_block_group(trans, inode); |
6659 | trans->block_rsv = root->orphan_block_rsv; | |
6660 | } | |
6661 | ||
6662 | ret = btrfs_block_rsv_check(trans, root, | |
6663 | root->orphan_block_rsv, 0, 5); | |
3893e33b | 6664 | if (ret == -EAGAIN) { |
d68fc57b | 6665 | ret = btrfs_commit_transaction(trans, root); |
3893e33b JB |
6666 | if (ret) |
6667 | return ret; | |
d68fc57b YZ |
6668 | trans = NULL; |
6669 | continue; | |
3893e33b JB |
6670 | } else if (ret) { |
6671 | err = ret; | |
6672 | break; | |
d68fc57b YZ |
6673 | } |
6674 | ||
8082510e YZ |
6675 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6676 | inode->i_size, | |
6677 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6678 | if (ret != -EAGAIN) { |
6679 | err = ret; | |
8082510e | 6680 | break; |
3893e33b | 6681 | } |
39279cc3 | 6682 | |
8082510e | 6683 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6684 | if (ret) { |
6685 | err = ret; | |
6686 | break; | |
6687 | } | |
5f39d397 | 6688 | |
8082510e YZ |
6689 | nr = trans->blocks_used; |
6690 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6691 | trans = NULL; |
8082510e | 6692 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6693 | } |
6694 | ||
6695 | if (ret == 0 && inode->i_nlink > 0) { | |
6696 | ret = btrfs_orphan_del(trans, inode); | |
3893e33b JB |
6697 | if (ret) |
6698 | err = ret; | |
ded5db9d JB |
6699 | } else if (ret && inode->i_nlink > 0) { |
6700 | /* | |
6701 | * Failed to do the truncate, remove us from the in memory | |
6702 | * orphan list. | |
6703 | */ | |
6704 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6705 | } |
6706 | ||
6707 | ret = btrfs_update_inode(trans, root, inode); | |
3893e33b JB |
6708 | if (ret && !err) |
6709 | err = ret; | |
7b128766 | 6710 | |
7b128766 | 6711 | nr = trans->blocks_used; |
89ce8a63 | 6712 | ret = btrfs_end_transaction_throttle(trans, root); |
3893e33b JB |
6713 | if (ret && !err) |
6714 | err = ret; | |
d3c2fdcf | 6715 | btrfs_btree_balance_dirty(root, nr); |
a41ad394 | 6716 | |
3893e33b | 6717 | return err; |
39279cc3 CM |
6718 | } |
6719 | ||
d352ac68 CM |
6720 | /* |
6721 | * create a new subvolume directory/inode (helper for the ioctl). | |
6722 | */ | |
d2fb3437 | 6723 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 6724 | struct btrfs_root *new_root, |
d2fb3437 | 6725 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 6726 | { |
39279cc3 | 6727 | struct inode *inode; |
76dda93c | 6728 | int err; |
00e4e6b3 | 6729 | u64 index = 0; |
39279cc3 | 6730 | |
aec7477b | 6731 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 6732 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 6733 | if (IS_ERR(inode)) |
f46b5a66 | 6734 | return PTR_ERR(inode); |
39279cc3 CM |
6735 | inode->i_op = &btrfs_dir_inode_operations; |
6736 | inode->i_fop = &btrfs_dir_file_operations; | |
6737 | ||
39279cc3 | 6738 | inode->i_nlink = 1; |
dbe674a9 | 6739 | btrfs_i_size_write(inode, 0); |
3b96362c | 6740 | |
76dda93c YZ |
6741 | err = btrfs_update_inode(trans, new_root, inode); |
6742 | BUG_ON(err); | |
cb8e7090 | 6743 | |
76dda93c | 6744 | iput(inode); |
cb8e7090 | 6745 | return 0; |
39279cc3 CM |
6746 | } |
6747 | ||
d352ac68 CM |
6748 | /* helper function for file defrag and space balancing. This |
6749 | * forces readahead on a given range of bytes in an inode | |
6750 | */ | |
edbd8d4e | 6751 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
6752 | struct file_ra_state *ra, struct file *file, |
6753 | pgoff_t offset, pgoff_t last_index) | |
6754 | { | |
8e7bf94f | 6755 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 6756 | |
86479a04 CM |
6757 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
6758 | return offset + req_size; | |
86479a04 CM |
6759 | } |
6760 | ||
39279cc3 CM |
6761 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6762 | { | |
6763 | struct btrfs_inode *ei; | |
2ead6ae7 | 6764 | struct inode *inode; |
39279cc3 CM |
6765 | |
6766 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6767 | if (!ei) | |
6768 | return NULL; | |
2ead6ae7 YZ |
6769 | |
6770 | ei->root = NULL; | |
6771 | ei->space_info = NULL; | |
6772 | ei->generation = 0; | |
6773 | ei->sequence = 0; | |
15ee9bc7 | 6774 | ei->last_trans = 0; |
257c62e1 | 6775 | ei->last_sub_trans = 0; |
e02119d5 | 6776 | ei->logged_trans = 0; |
2ead6ae7 YZ |
6777 | ei->delalloc_bytes = 0; |
6778 | ei->reserved_bytes = 0; | |
6779 | ei->disk_i_size = 0; | |
6780 | ei->flags = 0; | |
6781 | ei->index_cnt = (u64)-1; | |
6782 | ei->last_unlink_trans = 0; | |
6783 | ||
0ca1f7ce | 6784 | atomic_set(&ei->outstanding_extents, 0); |
57a45ced | 6785 | atomic_set(&ei->reserved_extents, 0); |
2ead6ae7 YZ |
6786 | |
6787 | ei->ordered_data_close = 0; | |
d68fc57b | 6788 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6789 | ei->dummy_inode = 0; |
261507a0 | 6790 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 YZ |
6791 | |
6792 | inode = &ei->vfs_inode; | |
6793 | extent_map_tree_init(&ei->extent_tree, GFP_NOFS); | |
6794 | extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS); | |
6795 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS); | |
6796 | mutex_init(&ei->log_mutex); | |
e6dcd2dc | 6797 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6798 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6799 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6800 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6801 | RB_CLEAR_NODE(&ei->rb_node); |
6802 | ||
6803 | return inode; | |
39279cc3 CM |
6804 | } |
6805 | ||
fa0d7e3d NP |
6806 | static void btrfs_i_callback(struct rcu_head *head) |
6807 | { | |
6808 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6809 | INIT_LIST_HEAD(&inode->i_dentry); | |
6810 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6811 | } | |
6812 | ||
39279cc3 CM |
6813 | void btrfs_destroy_inode(struct inode *inode) |
6814 | { | |
e6dcd2dc | 6815 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6816 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6817 | ||
39279cc3 CM |
6818 | WARN_ON(!list_empty(&inode->i_dentry)); |
6819 | WARN_ON(inode->i_data.nrpages); | |
0ca1f7ce | 6820 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents)); |
57a45ced | 6821 | WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents)); |
39279cc3 | 6822 | |
a6dbd429 JB |
6823 | /* |
6824 | * This can happen where we create an inode, but somebody else also | |
6825 | * created the same inode and we need to destroy the one we already | |
6826 | * created. | |
6827 | */ | |
6828 | if (!root) | |
6829 | goto free; | |
6830 | ||
5a3f23d5 CM |
6831 | /* |
6832 | * Make sure we're properly removed from the ordered operation | |
6833 | * lists. | |
6834 | */ | |
6835 | smp_mb(); | |
6836 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6837 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6838 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6839 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6840 | } | |
6841 | ||
0af3d00b JB |
6842 | if (root == root->fs_info->tree_root) { |
6843 | struct btrfs_block_group_cache *block_group; | |
6844 | ||
6845 | block_group = btrfs_lookup_block_group(root->fs_info, | |
6846 | BTRFS_I(inode)->block_group); | |
6847 | if (block_group && block_group->inode == inode) { | |
6848 | spin_lock(&block_group->lock); | |
6849 | block_group->inode = NULL; | |
6850 | spin_unlock(&block_group->lock); | |
6851 | btrfs_put_block_group(block_group); | |
6852 | } else if (block_group) { | |
6853 | btrfs_put_block_group(block_group); | |
6854 | } | |
6855 | } | |
6856 | ||
d68fc57b | 6857 | spin_lock(&root->orphan_lock); |
7b128766 | 6858 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
8082510e YZ |
6859 | printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", |
6860 | inode->i_ino); | |
6861 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
7b128766 | 6862 | } |
d68fc57b | 6863 | spin_unlock(&root->orphan_lock); |
7b128766 | 6864 | |
d397712b | 6865 | while (1) { |
e6dcd2dc CM |
6866 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6867 | if (!ordered) | |
6868 | break; | |
6869 | else { | |
d397712b CM |
6870 | printk(KERN_ERR "btrfs found ordered " |
6871 | "extent %llu %llu on inode cleanup\n", | |
6872 | (unsigned long long)ordered->file_offset, | |
6873 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6874 | btrfs_remove_ordered_extent(inode, ordered); |
6875 | btrfs_put_ordered_extent(ordered); | |
6876 | btrfs_put_ordered_extent(ordered); | |
6877 | } | |
6878 | } | |
5d4f98a2 | 6879 | inode_tree_del(inode); |
5b21f2ed | 6880 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6881 | free: |
fa0d7e3d | 6882 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6883 | } |
6884 | ||
45321ac5 | 6885 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6886 | { |
6887 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6888 | |
0af3d00b JB |
6889 | if (btrfs_root_refs(&root->root_item) == 0 && |
6890 | root != root->fs_info->tree_root) | |
45321ac5 | 6891 | return 1; |
76dda93c | 6892 | else |
45321ac5 | 6893 | return generic_drop_inode(inode); |
76dda93c YZ |
6894 | } |
6895 | ||
0ee0fda0 | 6896 | static void init_once(void *foo) |
39279cc3 CM |
6897 | { |
6898 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6899 | ||
6900 | inode_init_once(&ei->vfs_inode); | |
6901 | } | |
6902 | ||
6903 | void btrfs_destroy_cachep(void) | |
6904 | { | |
6905 | if (btrfs_inode_cachep) | |
6906 | kmem_cache_destroy(btrfs_inode_cachep); | |
6907 | if (btrfs_trans_handle_cachep) | |
6908 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6909 | if (btrfs_transaction_cachep) | |
6910 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6911 | if (btrfs_path_cachep) |
6912 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6913 | if (btrfs_free_space_cachep) |
6914 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6915 | } |
6916 | ||
6917 | int btrfs_init_cachep(void) | |
6918 | { | |
9601e3f6 CH |
6919 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6920 | sizeof(struct btrfs_inode), 0, | |
6921 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6922 | if (!btrfs_inode_cachep) |
6923 | goto fail; | |
9601e3f6 CH |
6924 | |
6925 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6926 | sizeof(struct btrfs_trans_handle), 0, | |
6927 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6928 | if (!btrfs_trans_handle_cachep) |
6929 | goto fail; | |
9601e3f6 CH |
6930 | |
6931 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6932 | sizeof(struct btrfs_transaction), 0, | |
6933 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6934 | if (!btrfs_transaction_cachep) |
6935 | goto fail; | |
9601e3f6 CH |
6936 | |
6937 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6938 | sizeof(struct btrfs_path), 0, | |
6939 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6940 | if (!btrfs_path_cachep) |
6941 | goto fail; | |
9601e3f6 | 6942 | |
dc89e982 JB |
6943 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6944 | sizeof(struct btrfs_free_space), 0, | |
6945 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6946 | if (!btrfs_free_space_cachep) | |
6947 | goto fail; | |
6948 | ||
39279cc3 CM |
6949 | return 0; |
6950 | fail: | |
6951 | btrfs_destroy_cachep(); | |
6952 | return -ENOMEM; | |
6953 | } | |
6954 | ||
6955 | static int btrfs_getattr(struct vfsmount *mnt, | |
6956 | struct dentry *dentry, struct kstat *stat) | |
6957 | { | |
6958 | struct inode *inode = dentry->d_inode; | |
6959 | generic_fillattr(inode, stat); | |
3394e160 | 6960 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 6961 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
6962 | stat->blocks = (inode_get_bytes(inode) + |
6963 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
6964 | return 0; |
6965 | } | |
6966 | ||
75e7cb7f LB |
6967 | /* |
6968 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6969 | * directory. | |
6970 | */ | |
6971 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6972 | { | |
6973 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6974 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6975 | ||
6976 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6977 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6978 | else | |
6979 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6980 | ||
6981 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6982 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6983 | else | |
6984 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6985 | } | |
6986 | ||
d397712b CM |
6987 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6988 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6989 | { |
6990 | struct btrfs_trans_handle *trans; | |
6991 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6992 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6993 | struct inode *new_inode = new_dentry->d_inode; |
6994 | struct inode *old_inode = old_dentry->d_inode; | |
6995 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6996 | u64 index = 0; |
4df27c4d | 6997 | u64 root_objectid; |
39279cc3 CM |
6998 | int ret; |
6999 | ||
f679a840 YZ |
7000 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
7001 | return -EPERM; | |
7002 | ||
4df27c4d YZ |
7003 | /* we only allow rename subvolume link between subvolumes */ |
7004 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
7005 | return -EXDEV; |
7006 | ||
4df27c4d YZ |
7007 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7008 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7009 | return -ENOTEMPTY; |
5f39d397 | 7010 | |
4df27c4d YZ |
7011 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7012 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7013 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7014 | /* |
7015 | * we're using rename to replace one file with another. | |
7016 | * and the replacement file is large. Start IO on it now so | |
7017 | * we don't add too much work to the end of the transaction | |
7018 | */ | |
4baf8c92 | 7019 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7020 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7021 | filemap_flush(old_inode->i_mapping); | |
7022 | ||
76dda93c YZ |
7023 | /* close the racy window with snapshot create/destroy ioctl */ |
7024 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
7025 | down_read(&root->fs_info->subvol_sem); | |
a22285a6 YZ |
7026 | /* |
7027 | * We want to reserve the absolute worst case amount of items. So if | |
7028 | * both inodes are subvols and we need to unlink them then that would | |
7029 | * require 4 item modifications, but if they are both normal inodes it | |
7030 | * would require 5 item modifications, so we'll assume their normal | |
7031 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7032 | * should cover the worst case number of items we'll modify. | |
7033 | */ | |
7034 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7035 | if (IS_ERR(trans)) { |
7036 | ret = PTR_ERR(trans); | |
7037 | goto out_notrans; | |
7038 | } | |
76dda93c | 7039 | |
a5719521 | 7040 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 7041 | |
4df27c4d YZ |
7042 | if (dest != root) |
7043 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7044 | |
a5719521 YZ |
7045 | ret = btrfs_set_inode_index(new_dir, &index); |
7046 | if (ret) | |
7047 | goto out_fail; | |
5a3f23d5 | 7048 | |
a5719521 | 7049 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7050 | /* force full log commit if subvolume involved. */ |
7051 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7052 | } else { | |
a5719521 YZ |
7053 | ret = btrfs_insert_inode_ref(trans, dest, |
7054 | new_dentry->d_name.name, | |
7055 | new_dentry->d_name.len, | |
7056 | old_inode->i_ino, | |
7057 | new_dir->i_ino, index); | |
7058 | if (ret) | |
7059 | goto out_fail; | |
4df27c4d YZ |
7060 | /* |
7061 | * this is an ugly little race, but the rename is required | |
7062 | * to make sure that if we crash, the inode is either at the | |
7063 | * old name or the new one. pinning the log transaction lets | |
7064 | * us make sure we don't allow a log commit to come in after | |
7065 | * we unlink the name but before we add the new name back in. | |
7066 | */ | |
7067 | btrfs_pin_log_trans(root); | |
7068 | } | |
5a3f23d5 CM |
7069 | /* |
7070 | * make sure the inode gets flushed if it is replacing | |
7071 | * something. | |
7072 | */ | |
7073 | if (new_inode && new_inode->i_size && | |
7074 | old_inode && S_ISREG(old_inode->i_mode)) { | |
7075 | btrfs_add_ordered_operation(trans, root, old_inode); | |
7076 | } | |
7077 | ||
39279cc3 CM |
7078 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7079 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7080 | old_inode->i_ctime = ctime; | |
5f39d397 | 7081 | |
12fcfd22 CM |
7082 | if (old_dentry->d_parent != new_dentry->d_parent) |
7083 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7084 | ||
4df27c4d YZ |
7085 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
7086 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
7087 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7088 | old_dentry->d_name.name, | |
7089 | old_dentry->d_name.len); | |
7090 | } else { | |
92986796 AV |
7091 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7092 | old_dentry->d_inode, | |
7093 | old_dentry->d_name.name, | |
7094 | old_dentry->d_name.len); | |
7095 | if (!ret) | |
7096 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
7097 | } |
7098 | BUG_ON(ret); | |
39279cc3 CM |
7099 | |
7100 | if (new_inode) { | |
7101 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
7102 | if (unlikely(new_inode->i_ino == |
7103 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
7104 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7105 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7106 | root_objectid, | |
7107 | new_dentry->d_name.name, | |
7108 | new_dentry->d_name.len); | |
7109 | BUG_ON(new_inode->i_nlink == 0); | |
7110 | } else { | |
7111 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7112 | new_dentry->d_inode, | |
7113 | new_dentry->d_name.name, | |
7114 | new_dentry->d_name.len); | |
7115 | } | |
7116 | BUG_ON(ret); | |
7b128766 | 7117 | if (new_inode->i_nlink == 0) { |
e02119d5 | 7118 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7119 | BUG_ON(ret); |
7b128766 | 7120 | } |
39279cc3 | 7121 | } |
aec7477b | 7122 | |
75e7cb7f LB |
7123 | fixup_inode_flags(new_dir, old_inode); |
7124 | ||
4df27c4d YZ |
7125 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7126 | new_dentry->d_name.name, | |
a5719521 | 7127 | new_dentry->d_name.len, 0, index); |
4df27c4d | 7128 | BUG_ON(ret); |
39279cc3 | 7129 | |
4df27c4d | 7130 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
6a912213 JB |
7131 | struct dentry *parent = dget_parent(new_dentry); |
7132 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
7133 | dput(parent); | |
4df27c4d YZ |
7134 | btrfs_end_log_trans(root); |
7135 | } | |
39279cc3 | 7136 | out_fail: |
ab78c84d | 7137 | btrfs_end_transaction_throttle(trans, root); |
b44c59a8 | 7138 | out_notrans: |
76dda93c YZ |
7139 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
7140 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 7141 | |
39279cc3 CM |
7142 | return ret; |
7143 | } | |
7144 | ||
d352ac68 CM |
7145 | /* |
7146 | * some fairly slow code that needs optimization. This walks the list | |
7147 | * of all the inodes with pending delalloc and forces them to disk. | |
7148 | */ | |
24bbcf04 | 7149 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7150 | { |
7151 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7152 | struct btrfs_inode *binode; | |
5b21f2ed | 7153 | struct inode *inode; |
ea8c2819 | 7154 | |
c146afad YZ |
7155 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7156 | return -EROFS; | |
7157 | ||
75eff68e | 7158 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7159 | while (!list_empty(head)) { |
ea8c2819 CM |
7160 | binode = list_entry(head->next, struct btrfs_inode, |
7161 | delalloc_inodes); | |
5b21f2ed ZY |
7162 | inode = igrab(&binode->vfs_inode); |
7163 | if (!inode) | |
7164 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7165 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7166 | if (inode) { |
8c8bee1d | 7167 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7168 | if (delay_iput) |
7169 | btrfs_add_delayed_iput(inode); | |
7170 | else | |
7171 | iput(inode); | |
5b21f2ed ZY |
7172 | } |
7173 | cond_resched(); | |
75eff68e | 7174 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7175 | } |
75eff68e | 7176 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7177 | |
7178 | /* the filemap_flush will queue IO into the worker threads, but | |
7179 | * we have to make sure the IO is actually started and that | |
7180 | * ordered extents get created before we return | |
7181 | */ | |
7182 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7183 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7184 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7185 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7186 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7187 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7188 | } |
7189 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7190 | return 0; |
7191 | } | |
7192 | ||
0019f10d JB |
7193 | int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput, |
7194 | int sync) | |
5da9d01b YZ |
7195 | { |
7196 | struct btrfs_inode *binode; | |
7197 | struct inode *inode = NULL; | |
7198 | ||
7199 | spin_lock(&root->fs_info->delalloc_lock); | |
7200 | while (!list_empty(&root->fs_info->delalloc_inodes)) { | |
7201 | binode = list_entry(root->fs_info->delalloc_inodes.next, | |
7202 | struct btrfs_inode, delalloc_inodes); | |
7203 | inode = igrab(&binode->vfs_inode); | |
7204 | if (inode) { | |
7205 | list_move_tail(&binode->delalloc_inodes, | |
7206 | &root->fs_info->delalloc_inodes); | |
7207 | break; | |
7208 | } | |
7209 | ||
7210 | list_del_init(&binode->delalloc_inodes); | |
7211 | cond_resched_lock(&root->fs_info->delalloc_lock); | |
7212 | } | |
7213 | spin_unlock(&root->fs_info->delalloc_lock); | |
7214 | ||
7215 | if (inode) { | |
0019f10d JB |
7216 | if (sync) { |
7217 | filemap_write_and_wait(inode->i_mapping); | |
7218 | /* | |
7219 | * We have to do this because compression doesn't | |
7220 | * actually set PG_writeback until it submits the pages | |
7221 | * for IO, which happens in an async thread, so we could | |
7222 | * race and not actually wait for any writeback pages | |
7223 | * because they've not been submitted yet. Technically | |
7224 | * this could still be the case for the ordered stuff | |
7225 | * since the async thread may not have started to do its | |
7226 | * work yet. If this becomes the case then we need to | |
7227 | * figure out a way to make sure that in writepage we | |
7228 | * wait for any async pages to be submitted before | |
7229 | * returning so that fdatawait does what its supposed to | |
7230 | * do. | |
7231 | */ | |
7232 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
7233 | } else { | |
7234 | filemap_flush(inode->i_mapping); | |
7235 | } | |
5da9d01b YZ |
7236 | if (delay_iput) |
7237 | btrfs_add_delayed_iput(inode); | |
7238 | else | |
7239 | iput(inode); | |
7240 | return 1; | |
7241 | } | |
7242 | return 0; | |
7243 | } | |
7244 | ||
39279cc3 CM |
7245 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7246 | const char *symname) | |
7247 | { | |
7248 | struct btrfs_trans_handle *trans; | |
7249 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7250 | struct btrfs_path *path; | |
7251 | struct btrfs_key key; | |
1832a6d5 | 7252 | struct inode *inode = NULL; |
39279cc3 CM |
7253 | int err; |
7254 | int drop_inode = 0; | |
7255 | u64 objectid; | |
00e4e6b3 | 7256 | u64 index = 0 ; |
39279cc3 CM |
7257 | int name_len; |
7258 | int datasize; | |
5f39d397 | 7259 | unsigned long ptr; |
39279cc3 | 7260 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7261 | struct extent_buffer *leaf; |
1832a6d5 | 7262 | unsigned long nr = 0; |
39279cc3 CM |
7263 | |
7264 | name_len = strlen(symname) + 1; | |
7265 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7266 | return -ENAMETOOLONG; | |
1832a6d5 | 7267 | |
a22285a6 YZ |
7268 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
7269 | if (err) | |
7270 | return err; | |
9ed74f2d JB |
7271 | /* |
7272 | * 2 items for inode item and ref | |
7273 | * 2 items for dir items | |
7274 | * 1 item for xattr if selinux is on | |
7275 | */ | |
a22285a6 YZ |
7276 | trans = btrfs_start_transaction(root, 5); |
7277 | if (IS_ERR(trans)) | |
7278 | return PTR_ERR(trans); | |
1832a6d5 | 7279 | |
39279cc3 CM |
7280 | btrfs_set_trans_block_group(trans, dir); |
7281 | ||
aec7477b | 7282 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 7283 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
7284 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
7285 | &index); | |
39279cc3 CM |
7286 | err = PTR_ERR(inode); |
7287 | if (IS_ERR(inode)) | |
7288 | goto out_unlock; | |
7289 | ||
f34f57a3 | 7290 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
7291 | if (err) { |
7292 | drop_inode = 1; | |
7293 | goto out_unlock; | |
7294 | } | |
7295 | ||
39279cc3 | 7296 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 7297 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7298 | if (err) |
7299 | drop_inode = 1; | |
7300 | else { | |
7301 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7302 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
7303 | inode->i_fop = &btrfs_file_operations; |
7304 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 7305 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7306 | } |
39279cc3 CM |
7307 | btrfs_update_inode_block_group(trans, inode); |
7308 | btrfs_update_inode_block_group(trans, dir); | |
7309 | if (drop_inode) | |
7310 | goto out_unlock; | |
7311 | ||
7312 | path = btrfs_alloc_path(); | |
7313 | BUG_ON(!path); | |
7314 | key.objectid = inode->i_ino; | |
7315 | key.offset = 0; | |
39279cc3 CM |
7316 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7317 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7318 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7319 | datasize); | |
54aa1f4d CM |
7320 | if (err) { |
7321 | drop_inode = 1; | |
7322 | goto out_unlock; | |
7323 | } | |
5f39d397 CM |
7324 | leaf = path->nodes[0]; |
7325 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7326 | struct btrfs_file_extent_item); | |
7327 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7328 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7329 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7330 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7331 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7332 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7333 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7334 | ||
39279cc3 | 7335 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7336 | write_extent_buffer(leaf, symname, ptr, name_len); |
7337 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7338 | btrfs_free_path(path); |
5f39d397 | 7339 | |
39279cc3 CM |
7340 | inode->i_op = &btrfs_symlink_inode_operations; |
7341 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7342 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7343 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7344 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7345 | err = btrfs_update_inode(trans, root, inode); |
7346 | if (err) | |
7347 | drop_inode = 1; | |
39279cc3 CM |
7348 | |
7349 | out_unlock: | |
d3c2fdcf | 7350 | nr = trans->blocks_used; |
ab78c84d | 7351 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
7352 | if (drop_inode) { |
7353 | inode_dec_link_count(inode); | |
7354 | iput(inode); | |
7355 | } | |
d3c2fdcf | 7356 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7357 | return err; |
7358 | } | |
16432985 | 7359 | |
0af3d00b JB |
7360 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7361 | u64 start, u64 num_bytes, u64 min_size, | |
7362 | loff_t actual_len, u64 *alloc_hint, | |
7363 | struct btrfs_trans_handle *trans) | |
d899e052 | 7364 | { |
d899e052 YZ |
7365 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7366 | struct btrfs_key ins; | |
d899e052 | 7367 | u64 cur_offset = start; |
55a61d1d | 7368 | u64 i_size; |
d899e052 | 7369 | int ret = 0; |
0af3d00b | 7370 | bool own_trans = true; |
d899e052 | 7371 | |
0af3d00b JB |
7372 | if (trans) |
7373 | own_trans = false; | |
d899e052 | 7374 | while (num_bytes > 0) { |
0af3d00b JB |
7375 | if (own_trans) { |
7376 | trans = btrfs_start_transaction(root, 3); | |
7377 | if (IS_ERR(trans)) { | |
7378 | ret = PTR_ERR(trans); | |
7379 | break; | |
7380 | } | |
5a303d5d YZ |
7381 | } |
7382 | ||
efa56464 YZ |
7383 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7384 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7385 | if (ret) { |
0af3d00b JB |
7386 | if (own_trans) |
7387 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7388 | break; |
d899e052 | 7389 | } |
5a303d5d | 7390 | |
d899e052 YZ |
7391 | ret = insert_reserved_file_extent(trans, inode, |
7392 | cur_offset, ins.objectid, | |
7393 | ins.offset, ins.offset, | |
920bbbfb | 7394 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7395 | BTRFS_FILE_EXTENT_PREALLOC); |
7396 | BUG_ON(ret); | |
a1ed835e CM |
7397 | btrfs_drop_extent_cache(inode, cur_offset, |
7398 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7399 | |
d899e052 YZ |
7400 | num_bytes -= ins.offset; |
7401 | cur_offset += ins.offset; | |
efa56464 | 7402 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7403 | |
d899e052 | 7404 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7405 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7406 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7407 | (actual_len > inode->i_size) && |
7408 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7409 | if (cur_offset > actual_len) |
55a61d1d | 7410 | i_size = actual_len; |
d1ea6a61 | 7411 | else |
55a61d1d JB |
7412 | i_size = cur_offset; |
7413 | i_size_write(inode, i_size); | |
7414 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7415 | } |
7416 | ||
d899e052 YZ |
7417 | ret = btrfs_update_inode(trans, root, inode); |
7418 | BUG_ON(ret); | |
d899e052 | 7419 | |
0af3d00b JB |
7420 | if (own_trans) |
7421 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7422 | } |
d899e052 YZ |
7423 | return ret; |
7424 | } | |
7425 | ||
0af3d00b JB |
7426 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7427 | u64 start, u64 num_bytes, u64 min_size, | |
7428 | loff_t actual_len, u64 *alloc_hint) | |
7429 | { | |
7430 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7431 | min_size, actual_len, alloc_hint, | |
7432 | NULL); | |
7433 | } | |
7434 | ||
7435 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7436 | struct btrfs_trans_handle *trans, int mode, | |
7437 | u64 start, u64 num_bytes, u64 min_size, | |
7438 | loff_t actual_len, u64 *alloc_hint) | |
7439 | { | |
7440 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7441 | min_size, actual_len, alloc_hint, trans); | |
7442 | } | |
7443 | ||
e6dcd2dc CM |
7444 | static int btrfs_set_page_dirty(struct page *page) |
7445 | { | |
e6dcd2dc CM |
7446 | return __set_page_dirty_nobuffers(page); |
7447 | } | |
7448 | ||
b74c79e9 | 7449 | static int btrfs_permission(struct inode *inode, int mask, unsigned int flags) |
fdebe2bd | 7450 | { |
b83cc969 LZ |
7451 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7452 | ||
7453 | if (btrfs_root_readonly(root) && (mask & MAY_WRITE)) | |
7454 | return -EROFS; | |
6cbff00f | 7455 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 7456 | return -EACCES; |
b74c79e9 | 7457 | return generic_permission(inode, mask, flags, btrfs_check_acl); |
fdebe2bd | 7458 | } |
39279cc3 | 7459 | |
6e1d5dcc | 7460 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7461 | .getattr = btrfs_getattr, |
39279cc3 CM |
7462 | .lookup = btrfs_lookup, |
7463 | .create = btrfs_create, | |
7464 | .unlink = btrfs_unlink, | |
7465 | .link = btrfs_link, | |
7466 | .mkdir = btrfs_mkdir, | |
7467 | .rmdir = btrfs_rmdir, | |
7468 | .rename = btrfs_rename, | |
7469 | .symlink = btrfs_symlink, | |
7470 | .setattr = btrfs_setattr, | |
618e21d5 | 7471 | .mknod = btrfs_mknod, |
95819c05 CH |
7472 | .setxattr = btrfs_setxattr, |
7473 | .getxattr = btrfs_getxattr, | |
5103e947 | 7474 | .listxattr = btrfs_listxattr, |
95819c05 | 7475 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7476 | .permission = btrfs_permission, |
39279cc3 | 7477 | }; |
6e1d5dcc | 7478 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7479 | .lookup = btrfs_lookup, |
fdebe2bd | 7480 | .permission = btrfs_permission, |
39279cc3 | 7481 | }; |
76dda93c | 7482 | |
828c0950 | 7483 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7484 | .llseek = generic_file_llseek, |
7485 | .read = generic_read_dir, | |
cbdf5a24 | 7486 | .readdir = btrfs_real_readdir, |
34287aa3 | 7487 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7488 | #ifdef CONFIG_COMPAT |
34287aa3 | 7489 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7490 | #endif |
6bf13c0c | 7491 | .release = btrfs_release_file, |
e02119d5 | 7492 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7493 | }; |
7494 | ||
d1310b2e | 7495 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7496 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7497 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7498 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7499 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7500 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7501 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 7502 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
7503 | .set_bit_hook = btrfs_set_bit_hook, |
7504 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7505 | .merge_extent_hook = btrfs_merge_extent_hook, |
7506 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7507 | }; |
7508 | ||
35054394 CM |
7509 | /* |
7510 | * btrfs doesn't support the bmap operation because swapfiles | |
7511 | * use bmap to make a mapping of extents in the file. They assume | |
7512 | * these extents won't change over the life of the file and they | |
7513 | * use the bmap result to do IO directly to the drive. | |
7514 | * | |
7515 | * the btrfs bmap call would return logical addresses that aren't | |
7516 | * suitable for IO and they also will change frequently as COW | |
7517 | * operations happen. So, swapfile + btrfs == corruption. | |
7518 | * | |
7519 | * For now we're avoiding this by dropping bmap. | |
7520 | */ | |
7f09410b | 7521 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7522 | .readpage = btrfs_readpage, |
7523 | .writepage = btrfs_writepage, | |
b293f02e | 7524 | .writepages = btrfs_writepages, |
3ab2fb5a | 7525 | .readpages = btrfs_readpages, |
39279cc3 | 7526 | .sync_page = block_sync_page, |
16432985 | 7527 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7528 | .invalidatepage = btrfs_invalidatepage, |
7529 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7530 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7531 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7532 | }; |
7533 | ||
7f09410b | 7534 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7535 | .readpage = btrfs_readpage, |
7536 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7537 | .invalidatepage = btrfs_invalidatepage, |
7538 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7539 | }; |
7540 | ||
6e1d5dcc | 7541 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7542 | .getattr = btrfs_getattr, |
7543 | .setattr = btrfs_setattr, | |
95819c05 CH |
7544 | .setxattr = btrfs_setxattr, |
7545 | .getxattr = btrfs_getxattr, | |
5103e947 | 7546 | .listxattr = btrfs_listxattr, |
95819c05 | 7547 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7548 | .permission = btrfs_permission, |
1506fcc8 | 7549 | .fiemap = btrfs_fiemap, |
39279cc3 | 7550 | }; |
6e1d5dcc | 7551 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7552 | .getattr = btrfs_getattr, |
7553 | .setattr = btrfs_setattr, | |
fdebe2bd | 7554 | .permission = btrfs_permission, |
95819c05 CH |
7555 | .setxattr = btrfs_setxattr, |
7556 | .getxattr = btrfs_getxattr, | |
33268eaf | 7557 | .listxattr = btrfs_listxattr, |
95819c05 | 7558 | .removexattr = btrfs_removexattr, |
618e21d5 | 7559 | }; |
6e1d5dcc | 7560 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7561 | .readlink = generic_readlink, |
7562 | .follow_link = page_follow_link_light, | |
7563 | .put_link = page_put_link, | |
f209561a | 7564 | .getattr = btrfs_getattr, |
fdebe2bd | 7565 | .permission = btrfs_permission, |
0279b4cd JO |
7566 | .setxattr = btrfs_setxattr, |
7567 | .getxattr = btrfs_getxattr, | |
7568 | .listxattr = btrfs_listxattr, | |
7569 | .removexattr = btrfs_removexattr, | |
39279cc3 | 7570 | }; |
76dda93c | 7571 | |
82d339d9 | 7572 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
7573 | .d_delete = btrfs_dentry_delete, |
7574 | }; |