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 | 955 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
8d413713 | 956 | BUG_ON(!async_cow); |
771ed689 CM |
957 | async_cow->inode = inode; |
958 | async_cow->root = root; | |
959 | async_cow->locked_page = locked_page; | |
960 | async_cow->start = start; | |
961 | ||
6cbff00f | 962 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
963 | cur_end = end; |
964 | else | |
965 | cur_end = min(end, start + 512 * 1024 - 1); | |
966 | ||
967 | async_cow->end = cur_end; | |
968 | INIT_LIST_HEAD(&async_cow->extents); | |
969 | ||
970 | async_cow->work.func = async_cow_start; | |
971 | async_cow->work.ordered_func = async_cow_submit; | |
972 | async_cow->work.ordered_free = async_cow_free; | |
973 | async_cow->work.flags = 0; | |
974 | ||
771ed689 CM |
975 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
976 | PAGE_CACHE_SHIFT; | |
977 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
978 | ||
979 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
980 | &async_cow->work); | |
981 | ||
982 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
983 | wait_event(root->fs_info->async_submit_wait, | |
984 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
985 | limit)); | |
986 | } | |
987 | ||
d397712b | 988 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
989 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
990 | wait_event(root->fs_info->async_submit_wait, | |
991 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
992 | 0)); | |
993 | } | |
994 | ||
995 | *nr_written += nr_pages; | |
996 | start = cur_end + 1; | |
997 | } | |
998 | *page_started = 1; | |
999 | return 0; | |
be20aa9d CM |
1000 | } |
1001 | ||
d397712b | 1002 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1003 | u64 bytenr, u64 num_bytes) |
1004 | { | |
1005 | int ret; | |
1006 | struct btrfs_ordered_sum *sums; | |
1007 | LIST_HEAD(list); | |
1008 | ||
07d400a6 YZ |
1009 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
1010 | bytenr + num_bytes - 1, &list); | |
17d217fe YZ |
1011 | if (ret == 0 && list_empty(&list)) |
1012 | return 0; | |
1013 | ||
1014 | while (!list_empty(&list)) { | |
1015 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1016 | list_del(&sums->list); | |
1017 | kfree(sums); | |
1018 | } | |
1019 | return 1; | |
1020 | } | |
1021 | ||
d352ac68 CM |
1022 | /* |
1023 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1024 | * of the extents that exist in the file, and COWs the file as required. | |
1025 | * | |
1026 | * If no cow copies or snapshots exist, we write directly to the existing | |
1027 | * blocks on disk | |
1028 | */ | |
7f366cfe CM |
1029 | static noinline int run_delalloc_nocow(struct inode *inode, |
1030 | struct page *locked_page, | |
771ed689 CM |
1031 | u64 start, u64 end, int *page_started, int force, |
1032 | unsigned long *nr_written) | |
be20aa9d | 1033 | { |
be20aa9d | 1034 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1035 | struct btrfs_trans_handle *trans; |
be20aa9d | 1036 | struct extent_buffer *leaf; |
be20aa9d | 1037 | struct btrfs_path *path; |
80ff3856 | 1038 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1039 | struct btrfs_key found_key; |
80ff3856 YZ |
1040 | u64 cow_start; |
1041 | u64 cur_offset; | |
1042 | u64 extent_end; | |
5d4f98a2 | 1043 | u64 extent_offset; |
80ff3856 YZ |
1044 | u64 disk_bytenr; |
1045 | u64 num_bytes; | |
1046 | int extent_type; | |
1047 | int ret; | |
d899e052 | 1048 | int type; |
80ff3856 YZ |
1049 | int nocow; |
1050 | int check_prev = 1; | |
0cb59c99 | 1051 | bool nolock = false; |
be20aa9d CM |
1052 | |
1053 | path = btrfs_alloc_path(); | |
1054 | BUG_ON(!path); | |
0cb59c99 JB |
1055 | if (root == root->fs_info->tree_root) { |
1056 | nolock = true; | |
1057 | trans = btrfs_join_transaction_nolock(root, 1); | |
1058 | } else { | |
1059 | trans = btrfs_join_transaction(root, 1); | |
1060 | } | |
3612b495 | 1061 | BUG_ON(IS_ERR(trans)); |
be20aa9d | 1062 | |
80ff3856 YZ |
1063 | cow_start = (u64)-1; |
1064 | cur_offset = start; | |
1065 | while (1) { | |
1066 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
1067 | cur_offset, 0); | |
1068 | BUG_ON(ret < 0); | |
1069 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1070 | leaf = path->nodes[0]; | |
1071 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1072 | path->slots[0] - 1); | |
1073 | if (found_key.objectid == inode->i_ino && | |
1074 | found_key.type == BTRFS_EXTENT_DATA_KEY) | |
1075 | path->slots[0]--; | |
1076 | } | |
1077 | check_prev = 0; | |
1078 | next_slot: | |
1079 | leaf = path->nodes[0]; | |
1080 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1081 | ret = btrfs_next_leaf(root, path); | |
1082 | if (ret < 0) | |
1083 | BUG_ON(1); | |
1084 | if (ret > 0) | |
1085 | break; | |
1086 | leaf = path->nodes[0]; | |
1087 | } | |
be20aa9d | 1088 | |
80ff3856 YZ |
1089 | nocow = 0; |
1090 | disk_bytenr = 0; | |
17d217fe | 1091 | num_bytes = 0; |
80ff3856 YZ |
1092 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1093 | ||
1094 | if (found_key.objectid > inode->i_ino || | |
1095 | found_key.type > BTRFS_EXTENT_DATA_KEY || | |
1096 | found_key.offset > end) | |
1097 | break; | |
1098 | ||
1099 | if (found_key.offset > cur_offset) { | |
1100 | extent_end = found_key.offset; | |
e9061e21 | 1101 | extent_type = 0; |
80ff3856 YZ |
1102 | goto out_check; |
1103 | } | |
1104 | ||
1105 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1106 | struct btrfs_file_extent_item); | |
1107 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1108 | ||
d899e052 YZ |
1109 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1110 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1111 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1112 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1113 | extent_end = found_key.offset + |
1114 | btrfs_file_extent_num_bytes(leaf, fi); | |
1115 | if (extent_end <= start) { | |
1116 | path->slots[0]++; | |
1117 | goto next_slot; | |
1118 | } | |
17d217fe YZ |
1119 | if (disk_bytenr == 0) |
1120 | goto out_check; | |
80ff3856 YZ |
1121 | if (btrfs_file_extent_compression(leaf, fi) || |
1122 | btrfs_file_extent_encryption(leaf, fi) || | |
1123 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1124 | goto out_check; | |
d899e052 YZ |
1125 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1126 | goto out_check; | |
d2fb3437 | 1127 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1128 | goto out_check; |
17d217fe | 1129 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, |
5d4f98a2 YZ |
1130 | found_key.offset - |
1131 | extent_offset, disk_bytenr)) | |
17d217fe | 1132 | goto out_check; |
5d4f98a2 | 1133 | disk_bytenr += extent_offset; |
17d217fe YZ |
1134 | disk_bytenr += cur_offset - found_key.offset; |
1135 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1136 | /* | |
1137 | * force cow if csum exists in the range. | |
1138 | * this ensure that csum for a given extent are | |
1139 | * either valid or do not exist. | |
1140 | */ | |
1141 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1142 | goto out_check; | |
80ff3856 YZ |
1143 | nocow = 1; |
1144 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1145 | extent_end = found_key.offset + | |
1146 | btrfs_file_extent_inline_len(leaf, fi); | |
1147 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1148 | } else { | |
1149 | BUG_ON(1); | |
1150 | } | |
1151 | out_check: | |
1152 | if (extent_end <= start) { | |
1153 | path->slots[0]++; | |
1154 | goto next_slot; | |
1155 | } | |
1156 | if (!nocow) { | |
1157 | if (cow_start == (u64)-1) | |
1158 | cow_start = cur_offset; | |
1159 | cur_offset = extent_end; | |
1160 | if (cur_offset > end) | |
1161 | break; | |
1162 | path->slots[0]++; | |
1163 | goto next_slot; | |
7ea394f1 YZ |
1164 | } |
1165 | ||
1166 | btrfs_release_path(root, path); | |
80ff3856 YZ |
1167 | if (cow_start != (u64)-1) { |
1168 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1169 | found_key.offset - 1, page_started, |
1170 | nr_written, 1); | |
80ff3856 YZ |
1171 | BUG_ON(ret); |
1172 | cow_start = (u64)-1; | |
7ea394f1 | 1173 | } |
80ff3856 | 1174 | |
d899e052 YZ |
1175 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1176 | struct extent_map *em; | |
1177 | struct extent_map_tree *em_tree; | |
1178 | em_tree = &BTRFS_I(inode)->extent_tree; | |
1179 | em = alloc_extent_map(GFP_NOFS); | |
c26a9203 | 1180 | BUG_ON(!em); |
d899e052 | 1181 | em->start = cur_offset; |
445a6944 | 1182 | em->orig_start = em->start; |
d899e052 YZ |
1183 | em->len = num_bytes; |
1184 | em->block_len = num_bytes; | |
1185 | em->block_start = disk_bytenr; | |
1186 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1187 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1188 | while (1) { | |
890871be | 1189 | write_lock(&em_tree->lock); |
d899e052 | 1190 | ret = add_extent_mapping(em_tree, em); |
890871be | 1191 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1192 | if (ret != -EEXIST) { |
1193 | free_extent_map(em); | |
1194 | break; | |
1195 | } | |
1196 | btrfs_drop_extent_cache(inode, em->start, | |
1197 | em->start + em->len - 1, 0); | |
1198 | } | |
1199 | type = BTRFS_ORDERED_PREALLOC; | |
1200 | } else { | |
1201 | type = BTRFS_ORDERED_NOCOW; | |
1202 | } | |
80ff3856 YZ |
1203 | |
1204 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1205 | num_bytes, num_bytes, type); |
1206 | BUG_ON(ret); | |
771ed689 | 1207 | |
efa56464 YZ |
1208 | if (root->root_key.objectid == |
1209 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1210 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1211 | num_bytes); | |
1212 | BUG_ON(ret); | |
1213 | } | |
1214 | ||
d899e052 | 1215 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1216 | cur_offset, cur_offset + num_bytes - 1, |
1217 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1218 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1219 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1220 | cur_offset = extent_end; |
1221 | if (cur_offset > end) | |
1222 | break; | |
be20aa9d | 1223 | } |
80ff3856 YZ |
1224 | btrfs_release_path(root, path); |
1225 | ||
1226 | if (cur_offset <= end && cow_start == (u64)-1) | |
1227 | cow_start = cur_offset; | |
1228 | if (cow_start != (u64)-1) { | |
1229 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1230 | page_started, nr_written, 1); |
80ff3856 YZ |
1231 | BUG_ON(ret); |
1232 | } | |
1233 | ||
0cb59c99 JB |
1234 | if (nolock) { |
1235 | ret = btrfs_end_transaction_nolock(trans, root); | |
1236 | BUG_ON(ret); | |
1237 | } else { | |
1238 | ret = btrfs_end_transaction(trans, root); | |
1239 | BUG_ON(ret); | |
1240 | } | |
7ea394f1 | 1241 | btrfs_free_path(path); |
80ff3856 | 1242 | return 0; |
be20aa9d CM |
1243 | } |
1244 | ||
d352ac68 CM |
1245 | /* |
1246 | * extent_io.c call back to do delayed allocation processing | |
1247 | */ | |
c8b97818 | 1248 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1249 | u64 start, u64 end, int *page_started, |
1250 | unsigned long *nr_written) | |
be20aa9d | 1251 | { |
be20aa9d | 1252 | int ret; |
7f366cfe | 1253 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1254 | |
6cbff00f | 1255 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1256 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1257 | page_started, 1, nr_written); |
6cbff00f | 1258 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1259 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1260 | page_started, 0, nr_written); |
1e701a32 | 1261 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1262 | !(BTRFS_I(inode)->force_compress) && |
1263 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1264 | ret = cow_file_range(inode, locked_page, start, end, |
1265 | page_started, nr_written, 1); | |
be20aa9d | 1266 | else |
771ed689 | 1267 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1268 | page_started, nr_written); |
b888db2b CM |
1269 | return ret; |
1270 | } | |
1271 | ||
9ed74f2d | 1272 | static int btrfs_split_extent_hook(struct inode *inode, |
0ca1f7ce | 1273 | struct extent_state *orig, u64 split) |
9ed74f2d | 1274 | { |
0ca1f7ce | 1275 | /* not delalloc, ignore it */ |
9ed74f2d JB |
1276 | if (!(orig->state & EXTENT_DELALLOC)) |
1277 | return 0; | |
1278 | ||
0ca1f7ce | 1279 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1280 | return 0; |
1281 | } | |
1282 | ||
1283 | /* | |
1284 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1285 | * extents so we can keep track of new extents that are just merged onto old | |
1286 | * extents, such as when we are doing sequential writes, so we can properly | |
1287 | * account for the metadata space we'll need. | |
1288 | */ | |
1289 | static int btrfs_merge_extent_hook(struct inode *inode, | |
1290 | struct extent_state *new, | |
1291 | struct extent_state *other) | |
1292 | { | |
9ed74f2d JB |
1293 | /* not delalloc, ignore it */ |
1294 | if (!(other->state & EXTENT_DELALLOC)) | |
1295 | return 0; | |
1296 | ||
0ca1f7ce | 1297 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); |
9ed74f2d JB |
1298 | return 0; |
1299 | } | |
1300 | ||
d352ac68 CM |
1301 | /* |
1302 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1303 | * bytes in this file, and to maintain the list of inodes that | |
1304 | * have pending delalloc work to be done. | |
1305 | */ | |
0ca1f7ce YZ |
1306 | static int btrfs_set_bit_hook(struct inode *inode, |
1307 | struct extent_state *state, int *bits) | |
291d673e | 1308 | { |
9ed74f2d | 1309 | |
75eff68e CM |
1310 | /* |
1311 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1312 | * but in this case, we are only testeing for the DELALLOC | |
1313 | * bit, which is only set or cleared with irqs on | |
1314 | */ | |
0ca1f7ce | 1315 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1316 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1317 | u64 len = state->end + 1 - state->start; |
0cb59c99 JB |
1318 | int do_list = (root->root_key.objectid != |
1319 | BTRFS_ROOT_TREE_OBJECTID); | |
9ed74f2d | 1320 | |
0ca1f7ce YZ |
1321 | if (*bits & EXTENT_FIRST_DELALLOC) |
1322 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1323 | else | |
1324 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); | |
287a0ab9 | 1325 | |
75eff68e | 1326 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1327 | BTRFS_I(inode)->delalloc_bytes += len; |
1328 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1329 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1330 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1331 | &root->fs_info->delalloc_inodes); | |
1332 | } | |
75eff68e | 1333 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1334 | } |
1335 | return 0; | |
1336 | } | |
1337 | ||
d352ac68 CM |
1338 | /* |
1339 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1340 | */ | |
9ed74f2d | 1341 | static int btrfs_clear_bit_hook(struct inode *inode, |
0ca1f7ce | 1342 | struct extent_state *state, int *bits) |
291d673e | 1343 | { |
75eff68e CM |
1344 | /* |
1345 | * set_bit and clear bit hooks normally require _irqsave/restore | |
1346 | * but in this case, we are only testeing for the DELALLOC | |
1347 | * bit, which is only set or cleared with irqs on | |
1348 | */ | |
0ca1f7ce | 1349 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1350 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1351 | u64 len = state->end + 1 - state->start; |
0cb59c99 JB |
1352 | int do_list = (root->root_key.objectid != |
1353 | BTRFS_ROOT_TREE_OBJECTID); | |
bcbfce8a | 1354 | |
0ca1f7ce YZ |
1355 | if (*bits & EXTENT_FIRST_DELALLOC) |
1356 | *bits &= ~EXTENT_FIRST_DELALLOC; | |
1357 | else if (!(*bits & EXTENT_DO_ACCOUNTING)) | |
1358 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
1359 | ||
1360 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1361 | btrfs_delalloc_release_metadata(inode, len); | |
1362 | ||
0cb59c99 JB |
1363 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1364 | && do_list) | |
0ca1f7ce | 1365 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1366 | |
75eff68e | 1367 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1368 | root->fs_info->delalloc_bytes -= len; |
1369 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1370 | ||
0cb59c99 | 1371 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1372 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1373 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1374 | } | |
75eff68e | 1375 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e CM |
1376 | } |
1377 | return 0; | |
1378 | } | |
1379 | ||
d352ac68 CM |
1380 | /* |
1381 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1382 | * we don't create bios that span stripes or chunks | |
1383 | */ | |
239b14b3 | 1384 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1385 | size_t size, struct bio *bio, |
1386 | unsigned long bio_flags) | |
239b14b3 CM |
1387 | { |
1388 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1389 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1390 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1391 | u64 length = 0; |
1392 | u64 map_length; | |
239b14b3 CM |
1393 | int ret; |
1394 | ||
771ed689 CM |
1395 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1396 | return 0; | |
1397 | ||
f2d8d74d | 1398 | length = bio->bi_size; |
239b14b3 CM |
1399 | map_tree = &root->fs_info->mapping_tree; |
1400 | map_length = length; | |
cea9e445 | 1401 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1402 | &map_length, NULL, 0); |
cea9e445 | 1403 | |
d397712b | 1404 | if (map_length < length + size) |
239b14b3 | 1405 | return 1; |
411fc6bc | 1406 | return ret; |
239b14b3 CM |
1407 | } |
1408 | ||
d352ac68 CM |
1409 | /* |
1410 | * in order to insert checksums into the metadata in large chunks, | |
1411 | * we wait until bio submission time. All the pages in the bio are | |
1412 | * checksummed and sums are attached onto the ordered extent record. | |
1413 | * | |
1414 | * At IO completion time the cums attached on the ordered extent record | |
1415 | * are inserted into the btree | |
1416 | */ | |
d397712b CM |
1417 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1418 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1419 | unsigned long bio_flags, |
1420 | u64 bio_offset) | |
065631f6 | 1421 | { |
065631f6 | 1422 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1423 | int ret = 0; |
e015640f | 1424 | |
d20f7043 | 1425 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1426 | BUG_ON(ret); |
4a69a410 CM |
1427 | return 0; |
1428 | } | |
e015640f | 1429 | |
4a69a410 CM |
1430 | /* |
1431 | * in order to insert checksums into the metadata in large chunks, | |
1432 | * we wait until bio submission time. All the pages in the bio are | |
1433 | * checksummed and sums are attached onto the ordered extent record. | |
1434 | * | |
1435 | * At IO completion time the cums attached on the ordered extent record | |
1436 | * are inserted into the btree | |
1437 | */ | |
b2950863 | 1438 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1439 | int mirror_num, unsigned long bio_flags, |
1440 | u64 bio_offset) | |
4a69a410 CM |
1441 | { |
1442 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1443 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1444 | } |
1445 | ||
d352ac68 | 1446 | /* |
cad321ad CM |
1447 | * extent_io.c submission hook. This does the right thing for csum calculation |
1448 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1449 | */ |
b2950863 | 1450 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1451 | int mirror_num, unsigned long bio_flags, |
1452 | u64 bio_offset) | |
44b8bd7e CM |
1453 | { |
1454 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1455 | int ret = 0; | |
19b9bdb0 | 1456 | int skip_sum; |
44b8bd7e | 1457 | |
6cbff00f | 1458 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1459 | |
0cb59c99 JB |
1460 | if (root == root->fs_info->tree_root) |
1461 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2); | |
1462 | else | |
1463 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
e6dcd2dc | 1464 | BUG_ON(ret); |
065631f6 | 1465 | |
7b6d91da | 1466 | if (!(rw & REQ_WRITE)) { |
d20f7043 | 1467 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1468 | return btrfs_submit_compressed_read(inode, bio, |
1469 | mirror_num, bio_flags); | |
c2db1073 TI |
1470 | } else if (!skip_sum) { |
1471 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1472 | if (ret) | |
1473 | return ret; | |
1474 | } | |
4d1b5fb4 | 1475 | goto mapit; |
19b9bdb0 | 1476 | } else if (!skip_sum) { |
17d217fe YZ |
1477 | /* csum items have already been cloned */ |
1478 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1479 | goto mapit; | |
19b9bdb0 CM |
1480 | /* we're doing a write, do the async checksumming */ |
1481 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1482 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1483 | bio_flags, bio_offset, |
1484 | __btrfs_submit_bio_start, | |
4a69a410 | 1485 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1486 | } |
1487 | ||
0b86a832 | 1488 | mapit: |
8b712842 | 1489 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1490 | } |
6885f308 | 1491 | |
d352ac68 CM |
1492 | /* |
1493 | * given a list of ordered sums record them in the inode. This happens | |
1494 | * at IO completion time based on sums calculated at bio submission time. | |
1495 | */ | |
ba1da2f4 | 1496 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1497 | struct inode *inode, u64 file_offset, |
1498 | struct list_head *list) | |
1499 | { | |
e6dcd2dc CM |
1500 | struct btrfs_ordered_sum *sum; |
1501 | ||
1502 | btrfs_set_trans_block_group(trans, inode); | |
c6e30871 QF |
1503 | |
1504 | list_for_each_entry(sum, list, list) { | |
d20f7043 CM |
1505 | btrfs_csum_file_blocks(trans, |
1506 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1507 | } |
1508 | return 0; | |
1509 | } | |
1510 | ||
2ac55d41 JB |
1511 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1512 | struct extent_state **cached_state) | |
ea8c2819 | 1513 | { |
d397712b | 1514 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1515 | WARN_ON(1); |
ea8c2819 | 1516 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1517 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1518 | } |
1519 | ||
d352ac68 | 1520 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1521 | struct btrfs_writepage_fixup { |
1522 | struct page *page; | |
1523 | struct btrfs_work work; | |
1524 | }; | |
1525 | ||
b2950863 | 1526 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1527 | { |
1528 | struct btrfs_writepage_fixup *fixup; | |
1529 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1530 | struct extent_state *cached_state = NULL; |
247e743c CM |
1531 | struct page *page; |
1532 | struct inode *inode; | |
1533 | u64 page_start; | |
1534 | u64 page_end; | |
1535 | ||
1536 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1537 | page = fixup->page; | |
4a096752 | 1538 | again: |
247e743c CM |
1539 | lock_page(page); |
1540 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1541 | ClearPageChecked(page); | |
1542 | goto out_page; | |
1543 | } | |
1544 | ||
1545 | inode = page->mapping->host; | |
1546 | page_start = page_offset(page); | |
1547 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1548 | ||
2ac55d41 JB |
1549 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1550 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1551 | |
1552 | /* already ordered? We're done */ | |
8b62b72b | 1553 | if (PagePrivate2(page)) |
247e743c | 1554 | goto out; |
4a096752 CM |
1555 | |
1556 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1557 | if (ordered) { | |
2ac55d41 JB |
1558 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1559 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1560 | unlock_page(page); |
1561 | btrfs_start_ordered_extent(inode, ordered, 1); | |
1562 | goto again; | |
1563 | } | |
247e743c | 1564 | |
0ca1f7ce | 1565 | BUG(); |
2ac55d41 | 1566 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c CM |
1567 | ClearPageChecked(page); |
1568 | out: | |
2ac55d41 JB |
1569 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1570 | &cached_state, GFP_NOFS); | |
247e743c CM |
1571 | out_page: |
1572 | unlock_page(page); | |
1573 | page_cache_release(page); | |
b897abec | 1574 | kfree(fixup); |
247e743c CM |
1575 | } |
1576 | ||
1577 | /* | |
1578 | * There are a few paths in the higher layers of the kernel that directly | |
1579 | * set the page dirty bit without asking the filesystem if it is a | |
1580 | * good idea. This causes problems because we want to make sure COW | |
1581 | * properly happens and the data=ordered rules are followed. | |
1582 | * | |
c8b97818 | 1583 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1584 | * hasn't been properly setup for IO. We kick off an async process |
1585 | * to fix it up. The async helper will wait for ordered extents, set | |
1586 | * the delalloc bit and make it safe to write the page. | |
1587 | */ | |
b2950863 | 1588 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1589 | { |
1590 | struct inode *inode = page->mapping->host; | |
1591 | struct btrfs_writepage_fixup *fixup; | |
1592 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1593 | |
8b62b72b CM |
1594 | /* this page is properly in the ordered list */ |
1595 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1596 | return 0; |
1597 | ||
1598 | if (PageChecked(page)) | |
1599 | return -EAGAIN; | |
1600 | ||
1601 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1602 | if (!fixup) | |
1603 | return -EAGAIN; | |
f421950f | 1604 | |
247e743c CM |
1605 | SetPageChecked(page); |
1606 | page_cache_get(page); | |
1607 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1608 | fixup->page = page; | |
1609 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
1610 | return -EAGAIN; | |
1611 | } | |
1612 | ||
d899e052 YZ |
1613 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1614 | struct inode *inode, u64 file_pos, | |
1615 | u64 disk_bytenr, u64 disk_num_bytes, | |
1616 | u64 num_bytes, u64 ram_bytes, | |
1617 | u8 compression, u8 encryption, | |
1618 | u16 other_encoding, int extent_type) | |
1619 | { | |
1620 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1621 | struct btrfs_file_extent_item *fi; | |
1622 | struct btrfs_path *path; | |
1623 | struct extent_buffer *leaf; | |
1624 | struct btrfs_key ins; | |
1625 | u64 hint; | |
1626 | int ret; | |
1627 | ||
1628 | path = btrfs_alloc_path(); | |
1629 | BUG_ON(!path); | |
1630 | ||
b9473439 | 1631 | path->leave_spinning = 1; |
a1ed835e CM |
1632 | |
1633 | /* | |
1634 | * we may be replacing one extent in the tree with another. | |
1635 | * The new extent is pinned in the extent map, and we don't want | |
1636 | * to drop it from the cache until it is completely in the btree. | |
1637 | * | |
1638 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1639 | * the caller is expected to unpin it and allow it to be merged | |
1640 | * with the others. | |
1641 | */ | |
920bbbfb YZ |
1642 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1643 | &hint, 0); | |
d899e052 YZ |
1644 | BUG_ON(ret); |
1645 | ||
1646 | ins.objectid = inode->i_ino; | |
1647 | ins.offset = file_pos; | |
1648 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1649 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1650 | BUG_ON(ret); | |
1651 | leaf = path->nodes[0]; | |
1652 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1653 | struct btrfs_file_extent_item); | |
1654 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1655 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1656 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1657 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1658 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1659 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1660 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1661 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1662 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1663 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1664 | |
1665 | btrfs_unlock_up_safe(path, 1); | |
1666 | btrfs_set_lock_blocking(leaf); | |
1667 | ||
d899e052 YZ |
1668 | btrfs_mark_buffer_dirty(leaf); |
1669 | ||
1670 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1671 | |
1672 | ins.objectid = disk_bytenr; | |
1673 | ins.offset = disk_num_bytes; | |
1674 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1675 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1676 | root->root_key.objectid, | |
1677 | inode->i_ino, file_pos, &ins); | |
d899e052 | 1678 | BUG_ON(ret); |
d899e052 | 1679 | btrfs_free_path(path); |
b9473439 | 1680 | |
d899e052 YZ |
1681 | return 0; |
1682 | } | |
1683 | ||
5d13a98f CM |
1684 | /* |
1685 | * helper function for btrfs_finish_ordered_io, this | |
1686 | * just reads in some of the csum leaves to prime them into ram | |
1687 | * before we start the transaction. It limits the amount of btree | |
1688 | * reads required while inside the transaction. | |
1689 | */ | |
d352ac68 CM |
1690 | /* as ordered data IO finishes, this gets called so we can finish |
1691 | * an ordered extent if the range of bytes in the file it covers are | |
1692 | * fully written. | |
1693 | */ | |
211f90e6 | 1694 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1695 | { |
e6dcd2dc | 1696 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1697 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1698 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1699 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1700 | struct extent_state *cached_state = NULL; |
261507a0 | 1701 | int compress_type = 0; |
e6dcd2dc | 1702 | int ret; |
0cb59c99 | 1703 | bool nolock = false; |
e6dcd2dc | 1704 | |
5a1a3df1 JB |
1705 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1706 | end - start + 1); | |
ba1da2f4 | 1707 | if (!ret) |
e6dcd2dc | 1708 | return 0; |
e6dcd2dc | 1709 | BUG_ON(!ordered_extent); |
efd049fb | 1710 | |
0cb59c99 JB |
1711 | nolock = (root == root->fs_info->tree_root); |
1712 | ||
c2167754 YZ |
1713 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1714 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1715 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1716 | if (!ret) { | |
0cb59c99 JB |
1717 | if (nolock) |
1718 | trans = btrfs_join_transaction_nolock(root, 1); | |
1719 | else | |
1720 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 1721 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce YZ |
1722 | btrfs_set_trans_block_group(trans, inode); |
1723 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 YZ |
1724 | ret = btrfs_update_inode(trans, root, inode); |
1725 | BUG_ON(ret); | |
c2167754 YZ |
1726 | } |
1727 | goto out; | |
1728 | } | |
e6dcd2dc | 1729 | |
2ac55d41 JB |
1730 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1731 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1732 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1733 | |
0cb59c99 JB |
1734 | if (nolock) |
1735 | trans = btrfs_join_transaction_nolock(root, 1); | |
1736 | else | |
1737 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 1738 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce YZ |
1739 | btrfs_set_trans_block_group(trans, inode); |
1740 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
c2167754 | 1741 | |
c8b97818 | 1742 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1743 | compress_type = ordered_extent->compress_type; |
d899e052 | 1744 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1745 | BUG_ON(compress_type); |
920bbbfb | 1746 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1747 | ordered_extent->file_offset, |
1748 | ordered_extent->file_offset + | |
1749 | ordered_extent->len); | |
1750 | BUG_ON(ret); | |
1751 | } else { | |
0af3d00b | 1752 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1753 | ret = insert_reserved_file_extent(trans, inode, |
1754 | ordered_extent->file_offset, | |
1755 | ordered_extent->start, | |
1756 | ordered_extent->disk_len, | |
1757 | ordered_extent->len, | |
1758 | ordered_extent->len, | |
261507a0 | 1759 | compress_type, 0, 0, |
d899e052 | 1760 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1761 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1762 | ordered_extent->file_offset, | |
1763 | ordered_extent->len); | |
d899e052 YZ |
1764 | BUG_ON(ret); |
1765 | } | |
2ac55d41 JB |
1766 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1767 | ordered_extent->file_offset + | |
1768 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1769 | ||
e6dcd2dc CM |
1770 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1771 | &ordered_extent->list); | |
1772 | ||
1ef30be1 JB |
1773 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1774 | if (!ret) { | |
1775 | ret = btrfs_update_inode(trans, root, inode); | |
1776 | BUG_ON(ret); | |
1777 | } | |
1778 | ret = 0; | |
c2167754 | 1779 | out: |
0cb59c99 JB |
1780 | if (nolock) { |
1781 | if (trans) | |
1782 | btrfs_end_transaction_nolock(trans, root); | |
1783 | } else { | |
1784 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); | |
1785 | if (trans) | |
1786 | btrfs_end_transaction(trans, root); | |
1787 | } | |
1788 | ||
e6dcd2dc CM |
1789 | /* once for us */ |
1790 | btrfs_put_ordered_extent(ordered_extent); | |
1791 | /* once for the tree */ | |
1792 | btrfs_put_ordered_extent(ordered_extent); | |
1793 | ||
e6dcd2dc CM |
1794 | return 0; |
1795 | } | |
1796 | ||
b2950863 | 1797 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1798 | struct extent_state *state, int uptodate) |
1799 | { | |
1abe9b8a | 1800 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1801 | ||
8b62b72b | 1802 | ClearPagePrivate2(page); |
211f90e6 CM |
1803 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1804 | } | |
1805 | ||
d352ac68 CM |
1806 | /* |
1807 | * When IO fails, either with EIO or csum verification fails, we | |
1808 | * try other mirrors that might have a good copy of the data. This | |
1809 | * io_failure_record is used to record state as we go through all the | |
1810 | * mirrors. If another mirror has good data, the page is set up to date | |
1811 | * and things continue. If a good mirror can't be found, the original | |
1812 | * bio end_io callback is called to indicate things have failed. | |
1813 | */ | |
7e38326f CM |
1814 | struct io_failure_record { |
1815 | struct page *page; | |
1816 | u64 start; | |
1817 | u64 len; | |
1818 | u64 logical; | |
d20f7043 | 1819 | unsigned long bio_flags; |
7e38326f CM |
1820 | int last_mirror; |
1821 | }; | |
1822 | ||
b2950863 | 1823 | static int btrfs_io_failed_hook(struct bio *failed_bio, |
1259ab75 CM |
1824 | struct page *page, u64 start, u64 end, |
1825 | struct extent_state *state) | |
7e38326f CM |
1826 | { |
1827 | struct io_failure_record *failrec = NULL; | |
1828 | u64 private; | |
1829 | struct extent_map *em; | |
1830 | struct inode *inode = page->mapping->host; | |
1831 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
3b951516 | 1832 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7e38326f CM |
1833 | struct bio *bio; |
1834 | int num_copies; | |
1835 | int ret; | |
1259ab75 | 1836 | int rw; |
7e38326f CM |
1837 | u64 logical; |
1838 | ||
1839 | ret = get_state_private(failure_tree, start, &private); | |
1840 | if (ret) { | |
7e38326f CM |
1841 | failrec = kmalloc(sizeof(*failrec), GFP_NOFS); |
1842 | if (!failrec) | |
1843 | return -ENOMEM; | |
1844 | failrec->start = start; | |
1845 | failrec->len = end - start + 1; | |
1846 | failrec->last_mirror = 0; | |
d20f7043 | 1847 | failrec->bio_flags = 0; |
7e38326f | 1848 | |
890871be | 1849 | read_lock(&em_tree->lock); |
3b951516 CM |
1850 | em = lookup_extent_mapping(em_tree, start, failrec->len); |
1851 | if (em->start > start || em->start + em->len < start) { | |
1852 | free_extent_map(em); | |
1853 | em = NULL; | |
1854 | } | |
890871be | 1855 | read_unlock(&em_tree->lock); |
7e38326f CM |
1856 | |
1857 | if (!em || IS_ERR(em)) { | |
1858 | kfree(failrec); | |
1859 | return -EIO; | |
1860 | } | |
1861 | logical = start - em->start; | |
1862 | logical = em->block_start + logical; | |
d20f7043 CM |
1863 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
1864 | logical = em->block_start; | |
1865 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
261507a0 LZ |
1866 | extent_set_compress_type(&failrec->bio_flags, |
1867 | em->compress_type); | |
d20f7043 | 1868 | } |
7e38326f CM |
1869 | failrec->logical = logical; |
1870 | free_extent_map(em); | |
1871 | set_extent_bits(failure_tree, start, end, EXTENT_LOCKED | | |
1872 | EXTENT_DIRTY, GFP_NOFS); | |
587f7704 CM |
1873 | set_state_private(failure_tree, start, |
1874 | (u64)(unsigned long)failrec); | |
7e38326f | 1875 | } else { |
587f7704 | 1876 | failrec = (struct io_failure_record *)(unsigned long)private; |
7e38326f CM |
1877 | } |
1878 | num_copies = btrfs_num_copies( | |
1879 | &BTRFS_I(inode)->root->fs_info->mapping_tree, | |
1880 | failrec->logical, failrec->len); | |
1881 | failrec->last_mirror++; | |
1882 | if (!state) { | |
cad321ad | 1883 | spin_lock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1884 | state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree, |
1885 | failrec->start, | |
1886 | EXTENT_LOCKED); | |
1887 | if (state && state->start != failrec->start) | |
1888 | state = NULL; | |
cad321ad | 1889 | spin_unlock(&BTRFS_I(inode)->io_tree.lock); |
7e38326f CM |
1890 | } |
1891 | if (!state || failrec->last_mirror > num_copies) { | |
1892 | set_state_private(failure_tree, failrec->start, 0); | |
1893 | clear_extent_bits(failure_tree, failrec->start, | |
1894 | failrec->start + failrec->len - 1, | |
1895 | EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS); | |
1896 | kfree(failrec); | |
1897 | return -EIO; | |
1898 | } | |
1899 | bio = bio_alloc(GFP_NOFS, 1); | |
1900 | bio->bi_private = state; | |
1901 | bio->bi_end_io = failed_bio->bi_end_io; | |
1902 | bio->bi_sector = failrec->logical >> 9; | |
1903 | bio->bi_bdev = failed_bio->bi_bdev; | |
e1c4b745 | 1904 | bio->bi_size = 0; |
d20f7043 | 1905 | |
7e38326f | 1906 | bio_add_page(bio, page, failrec->len, start - page_offset(page)); |
7b6d91da | 1907 | if (failed_bio->bi_rw & REQ_WRITE) |
1259ab75 CM |
1908 | rw = WRITE; |
1909 | else | |
1910 | rw = READ; | |
1911 | ||
c2db1073 | 1912 | ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio, |
c8b97818 | 1913 | failrec->last_mirror, |
eaf25d93 | 1914 | failrec->bio_flags, 0); |
c2db1073 | 1915 | return ret; |
1259ab75 CM |
1916 | } |
1917 | ||
d352ac68 CM |
1918 | /* |
1919 | * each time an IO finishes, we do a fast check in the IO failure tree | |
1920 | * to see if we need to process or clean up an io_failure_record | |
1921 | */ | |
b2950863 | 1922 | static int btrfs_clean_io_failures(struct inode *inode, u64 start) |
1259ab75 CM |
1923 | { |
1924 | u64 private; | |
1925 | u64 private_failure; | |
1926 | struct io_failure_record *failure; | |
1927 | int ret; | |
1928 | ||
1929 | private = 0; | |
1930 | if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private, | |
ec29ed5b | 1931 | (u64)-1, 1, EXTENT_DIRTY, 0)) { |
1259ab75 CM |
1932 | ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, |
1933 | start, &private_failure); | |
1934 | if (ret == 0) { | |
1935 | failure = (struct io_failure_record *)(unsigned long) | |
1936 | private_failure; | |
1937 | set_state_private(&BTRFS_I(inode)->io_failure_tree, | |
1938 | failure->start, 0); | |
1939 | clear_extent_bits(&BTRFS_I(inode)->io_failure_tree, | |
1940 | failure->start, | |
1941 | failure->start + failure->len - 1, | |
1942 | EXTENT_DIRTY | EXTENT_LOCKED, | |
1943 | GFP_NOFS); | |
1944 | kfree(failure); | |
1945 | } | |
1946 | } | |
7e38326f CM |
1947 | return 0; |
1948 | } | |
1949 | ||
d352ac68 CM |
1950 | /* |
1951 | * when reads are done, we need to check csums to verify the data is correct | |
1952 | * if there's a match, we allow the bio to finish. If not, we go through | |
1953 | * the io_failure_record routines to find good copies | |
1954 | */ | |
b2950863 | 1955 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1956 | struct extent_state *state) |
07157aac | 1957 | { |
35ebb934 | 1958 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1959 | struct inode *inode = page->mapping->host; |
d1310b2e | 1960 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1961 | char *kaddr; |
aadfeb6e | 1962 | u64 private = ~(u32)0; |
07157aac | 1963 | int ret; |
ff79f819 CM |
1964 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1965 | u32 csum = ~(u32)0; | |
d1310b2e | 1966 | |
d20f7043 CM |
1967 | if (PageChecked(page)) { |
1968 | ClearPageChecked(page); | |
1969 | goto good; | |
1970 | } | |
6cbff00f CH |
1971 | |
1972 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
17d217fe YZ |
1973 | return 0; |
1974 | ||
1975 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1976 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1977 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1978 | GFP_NOFS); | |
b6cda9bc | 1979 | return 0; |
17d217fe | 1980 | } |
d20f7043 | 1981 | |
c2e639f0 | 1982 | if (state && state->start == start) { |
70dec807 CM |
1983 | private = state->private; |
1984 | ret = 0; | |
1985 | } else { | |
1986 | ret = get_state_private(io_tree, start, &private); | |
1987 | } | |
9ab86c8e | 1988 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1989 | if (ret) |
07157aac | 1990 | goto zeroit; |
d397712b | 1991 | |
ff79f819 CM |
1992 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1993 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1994 | if (csum != private) |
07157aac | 1995 | goto zeroit; |
d397712b | 1996 | |
9ab86c8e | 1997 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1998 | good: |
7e38326f CM |
1999 | /* if the io failure tree for this inode is non-empty, |
2000 | * check to see if we've recovered from a failed IO | |
2001 | */ | |
1259ab75 | 2002 | btrfs_clean_io_failures(inode, start); |
07157aac CM |
2003 | return 0; |
2004 | ||
2005 | zeroit: | |
193f284d CM |
2006 | if (printk_ratelimit()) { |
2007 | printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u " | |
2008 | "private %llu\n", page->mapping->host->i_ino, | |
2009 | (unsigned long long)start, csum, | |
2010 | (unsigned long long)private); | |
2011 | } | |
db94535d CM |
2012 | memset(kaddr + offset, 1, end - start + 1); |
2013 | flush_dcache_page(page); | |
9ab86c8e | 2014 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
2015 | if (private == 0) |
2016 | return 0; | |
7e38326f | 2017 | return -EIO; |
07157aac | 2018 | } |
b888db2b | 2019 | |
24bbcf04 YZ |
2020 | struct delayed_iput { |
2021 | struct list_head list; | |
2022 | struct inode *inode; | |
2023 | }; | |
2024 | ||
2025 | void btrfs_add_delayed_iput(struct inode *inode) | |
2026 | { | |
2027 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2028 | struct delayed_iput *delayed; | |
2029 | ||
2030 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2031 | return; | |
2032 | ||
2033 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2034 | delayed->inode = inode; | |
2035 | ||
2036 | spin_lock(&fs_info->delayed_iput_lock); | |
2037 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2038 | spin_unlock(&fs_info->delayed_iput_lock); | |
2039 | } | |
2040 | ||
2041 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2042 | { | |
2043 | LIST_HEAD(list); | |
2044 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2045 | struct delayed_iput *delayed; | |
2046 | int empty; | |
2047 | ||
2048 | spin_lock(&fs_info->delayed_iput_lock); | |
2049 | empty = list_empty(&fs_info->delayed_iputs); | |
2050 | spin_unlock(&fs_info->delayed_iput_lock); | |
2051 | if (empty) | |
2052 | return; | |
2053 | ||
2054 | down_read(&root->fs_info->cleanup_work_sem); | |
2055 | spin_lock(&fs_info->delayed_iput_lock); | |
2056 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2057 | spin_unlock(&fs_info->delayed_iput_lock); | |
2058 | ||
2059 | while (!list_empty(&list)) { | |
2060 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2061 | list_del(&delayed->list); | |
2062 | iput(delayed->inode); | |
2063 | kfree(delayed); | |
2064 | } | |
2065 | up_read(&root->fs_info->cleanup_work_sem); | |
2066 | } | |
2067 | ||
d68fc57b YZ |
2068 | /* |
2069 | * calculate extra metadata reservation when snapshotting a subvolume | |
2070 | * contains orphan files. | |
2071 | */ | |
2072 | void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans, | |
2073 | struct btrfs_pending_snapshot *pending, | |
2074 | u64 *bytes_to_reserve) | |
2075 | { | |
2076 | struct btrfs_root *root; | |
2077 | struct btrfs_block_rsv *block_rsv; | |
2078 | u64 num_bytes; | |
2079 | int index; | |
2080 | ||
2081 | root = pending->root; | |
2082 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2083 | return; | |
2084 | ||
2085 | block_rsv = root->orphan_block_rsv; | |
2086 | ||
2087 | /* orphan block reservation for the snapshot */ | |
2088 | num_bytes = block_rsv->size; | |
2089 | ||
2090 | /* | |
2091 | * after the snapshot is created, COWing tree blocks may use more | |
2092 | * space than it frees. So we should make sure there is enough | |
2093 | * reserved space. | |
2094 | */ | |
2095 | index = trans->transid & 0x1; | |
2096 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2097 | num_bytes += block_rsv->size - | |
2098 | (block_rsv->reserved + block_rsv->freed[index]); | |
2099 | } | |
2100 | ||
2101 | *bytes_to_reserve += num_bytes; | |
2102 | } | |
2103 | ||
2104 | void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans, | |
2105 | struct btrfs_pending_snapshot *pending) | |
2106 | { | |
2107 | struct btrfs_root *root = pending->root; | |
2108 | struct btrfs_root *snap = pending->snap; | |
2109 | struct btrfs_block_rsv *block_rsv; | |
2110 | u64 num_bytes; | |
2111 | int index; | |
2112 | int ret; | |
2113 | ||
2114 | if (!root->orphan_block_rsv || list_empty(&root->orphan_list)) | |
2115 | return; | |
2116 | ||
2117 | /* refill source subvolume's orphan block reservation */ | |
2118 | block_rsv = root->orphan_block_rsv; | |
2119 | index = trans->transid & 0x1; | |
2120 | if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) { | |
2121 | num_bytes = block_rsv->size - | |
2122 | (block_rsv->reserved + block_rsv->freed[index]); | |
2123 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2124 | root->orphan_block_rsv, | |
2125 | num_bytes); | |
2126 | BUG_ON(ret); | |
2127 | } | |
2128 | ||
2129 | /* setup orphan block reservation for the snapshot */ | |
2130 | block_rsv = btrfs_alloc_block_rsv(snap); | |
2131 | BUG_ON(!block_rsv); | |
2132 | ||
2133 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
2134 | snap->orphan_block_rsv = block_rsv; | |
2135 | ||
2136 | num_bytes = root->orphan_block_rsv->size; | |
2137 | ret = btrfs_block_rsv_migrate(&pending->block_rsv, | |
2138 | block_rsv, num_bytes); | |
2139 | BUG_ON(ret); | |
2140 | ||
2141 | #if 0 | |
2142 | /* insert orphan item for the snapshot */ | |
2143 | WARN_ON(!root->orphan_item_inserted); | |
2144 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2145 | snap->root_key.objectid); | |
2146 | BUG_ON(ret); | |
2147 | snap->orphan_item_inserted = 1; | |
2148 | #endif | |
2149 | } | |
2150 | ||
2151 | enum btrfs_orphan_cleanup_state { | |
2152 | ORPHAN_CLEANUP_STARTED = 1, | |
2153 | ORPHAN_CLEANUP_DONE = 2, | |
2154 | }; | |
2155 | ||
2156 | /* | |
2157 | * This is called in transaction commmit time. If there are no orphan | |
2158 | * files in the subvolume, it removes orphan item and frees block_rsv | |
2159 | * structure. | |
2160 | */ | |
2161 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2162 | struct btrfs_root *root) | |
2163 | { | |
2164 | int ret; | |
2165 | ||
2166 | if (!list_empty(&root->orphan_list) || | |
2167 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
2168 | return; | |
2169 | ||
2170 | if (root->orphan_item_inserted && | |
2171 | btrfs_root_refs(&root->root_item) > 0) { | |
2172 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2173 | root->root_key.objectid); | |
2174 | BUG_ON(ret); | |
2175 | root->orphan_item_inserted = 0; | |
2176 | } | |
2177 | ||
2178 | if (root->orphan_block_rsv) { | |
2179 | WARN_ON(root->orphan_block_rsv->size > 0); | |
2180 | btrfs_free_block_rsv(root, root->orphan_block_rsv); | |
2181 | root->orphan_block_rsv = NULL; | |
2182 | } | |
2183 | } | |
2184 | ||
7b128766 JB |
2185 | /* |
2186 | * This creates an orphan entry for the given inode in case something goes | |
2187 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2188 | * |
2189 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2190 | * this function. | |
7b128766 JB |
2191 | */ |
2192 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2193 | { | |
2194 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2195 | struct btrfs_block_rsv *block_rsv = NULL; |
2196 | int reserve = 0; | |
2197 | int insert = 0; | |
2198 | int ret; | |
7b128766 | 2199 | |
d68fc57b YZ |
2200 | if (!root->orphan_block_rsv) { |
2201 | block_rsv = btrfs_alloc_block_rsv(root); | |
2202 | BUG_ON(!block_rsv); | |
2203 | } | |
7b128766 | 2204 | |
d68fc57b YZ |
2205 | spin_lock(&root->orphan_lock); |
2206 | if (!root->orphan_block_rsv) { | |
2207 | root->orphan_block_rsv = block_rsv; | |
2208 | } else if (block_rsv) { | |
2209 | btrfs_free_block_rsv(root, block_rsv); | |
2210 | block_rsv = NULL; | |
7b128766 | 2211 | } |
7b128766 | 2212 | |
d68fc57b YZ |
2213 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2214 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2215 | #if 0 | |
2216 | /* | |
2217 | * For proper ENOSPC handling, we should do orphan | |
2218 | * cleanup when mounting. But this introduces backward | |
2219 | * compatibility issue. | |
2220 | */ | |
2221 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2222 | insert = 2; | |
2223 | else | |
2224 | insert = 1; | |
2225 | #endif | |
2226 | insert = 1; | |
7b128766 JB |
2227 | } |
2228 | ||
d68fc57b YZ |
2229 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2230 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2231 | reserve = 1; | |
2232 | } | |
2233 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2234 | |
d68fc57b YZ |
2235 | if (block_rsv) |
2236 | btrfs_add_durable_block_rsv(root->fs_info, block_rsv); | |
7b128766 | 2237 | |
d68fc57b YZ |
2238 | /* grab metadata reservation from transaction handle */ |
2239 | if (reserve) { | |
2240 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2241 | BUG_ON(ret); | |
2242 | } | |
7b128766 | 2243 | |
d68fc57b YZ |
2244 | /* insert an orphan item to track this unlinked/truncated file */ |
2245 | if (insert >= 1) { | |
2246 | ret = btrfs_insert_orphan_item(trans, root, inode->i_ino); | |
2247 | BUG_ON(ret); | |
2248 | } | |
2249 | ||
2250 | /* insert an orphan item to track subvolume contains orphan files */ | |
2251 | if (insert >= 2) { | |
2252 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2253 | root->root_key.objectid); | |
2254 | BUG_ON(ret); | |
2255 | } | |
2256 | return 0; | |
7b128766 JB |
2257 | } |
2258 | ||
2259 | /* | |
2260 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2261 | * item for this particular inode. | |
2262 | */ | |
2263 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2264 | { | |
2265 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2266 | int delete_item = 0; |
2267 | int release_rsv = 0; | |
7b128766 JB |
2268 | int ret = 0; |
2269 | ||
d68fc57b YZ |
2270 | spin_lock(&root->orphan_lock); |
2271 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2272 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2273 | delete_item = 1; | |
7b128766 JB |
2274 | } |
2275 | ||
d68fc57b YZ |
2276 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2277 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2278 | release_rsv = 1; | |
7b128766 | 2279 | } |
d68fc57b | 2280 | spin_unlock(&root->orphan_lock); |
7b128766 | 2281 | |
d68fc57b YZ |
2282 | if (trans && delete_item) { |
2283 | ret = btrfs_del_orphan_item(trans, root, inode->i_ino); | |
2284 | BUG_ON(ret); | |
2285 | } | |
7b128766 | 2286 | |
d68fc57b YZ |
2287 | if (release_rsv) |
2288 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2289 | |
d68fc57b | 2290 | return 0; |
7b128766 JB |
2291 | } |
2292 | ||
2293 | /* | |
2294 | * this cleans up any orphans that may be left on the list from the last use | |
2295 | * of this root. | |
2296 | */ | |
66b4ffd1 | 2297 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2298 | { |
2299 | struct btrfs_path *path; | |
2300 | struct extent_buffer *leaf; | |
7b128766 JB |
2301 | struct btrfs_key key, found_key; |
2302 | struct btrfs_trans_handle *trans; | |
2303 | struct inode *inode; | |
2304 | int ret = 0, nr_unlink = 0, nr_truncate = 0; | |
2305 | ||
d68fc57b | 2306 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2307 | return 0; |
c71bf099 YZ |
2308 | |
2309 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2310 | if (!path) { |
2311 | ret = -ENOMEM; | |
2312 | goto out; | |
2313 | } | |
7b128766 JB |
2314 | path->reada = -1; |
2315 | ||
2316 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2317 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2318 | key.offset = (u64)-1; | |
2319 | ||
7b128766 JB |
2320 | while (1) { |
2321 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2322 | if (ret < 0) |
2323 | goto out; | |
7b128766 JB |
2324 | |
2325 | /* | |
2326 | * if ret == 0 means we found what we were searching for, which | |
2327 | * is weird, but possible, so only screw with path if we didnt | |
2328 | * find the key and see if we have stuff that matches | |
2329 | */ | |
2330 | if (ret > 0) { | |
66b4ffd1 | 2331 | ret = 0; |
7b128766 JB |
2332 | if (path->slots[0] == 0) |
2333 | break; | |
2334 | path->slots[0]--; | |
2335 | } | |
2336 | ||
2337 | /* pull out the item */ | |
2338 | leaf = path->nodes[0]; | |
7b128766 JB |
2339 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2340 | ||
2341 | /* make sure the item matches what we want */ | |
2342 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2343 | break; | |
2344 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2345 | break; | |
2346 | ||
2347 | /* release the path since we're done with it */ | |
2348 | btrfs_release_path(root, path); | |
2349 | ||
2350 | /* | |
2351 | * this is where we are basically btrfs_lookup, without the | |
2352 | * crossing root thing. we store the inode number in the | |
2353 | * offset of the orphan item. | |
2354 | */ | |
5d4f98a2 YZ |
2355 | found_key.objectid = found_key.offset; |
2356 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2357 | found_key.offset = 0; | |
73f73415 | 2358 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
66b4ffd1 JB |
2359 | if (IS_ERR(inode)) { |
2360 | ret = PTR_ERR(inode); | |
2361 | goto out; | |
2362 | } | |
7b128766 | 2363 | |
7b128766 JB |
2364 | /* |
2365 | * add this inode to the orphan list so btrfs_orphan_del does | |
2366 | * the proper thing when we hit it | |
2367 | */ | |
d68fc57b | 2368 | spin_lock(&root->orphan_lock); |
7b128766 | 2369 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); |
d68fc57b | 2370 | spin_unlock(&root->orphan_lock); |
7b128766 JB |
2371 | |
2372 | /* | |
2373 | * if this is a bad inode, means we actually succeeded in | |
2374 | * removing the inode, but not the orphan record, which means | |
2375 | * we need to manually delete the orphan since iput will just | |
2376 | * do a destroy_inode | |
2377 | */ | |
2378 | if (is_bad_inode(inode)) { | |
a22285a6 | 2379 | trans = btrfs_start_transaction(root, 0); |
66b4ffd1 JB |
2380 | if (IS_ERR(trans)) { |
2381 | ret = PTR_ERR(trans); | |
2382 | goto out; | |
2383 | } | |
7b128766 | 2384 | btrfs_orphan_del(trans, inode); |
5b21f2ed | 2385 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2386 | iput(inode); |
2387 | continue; | |
2388 | } | |
2389 | ||
2390 | /* if we have links, this was a truncate, lets do that */ | |
2391 | if (inode->i_nlink) { | |
a41ad394 JB |
2392 | if (!S_ISREG(inode->i_mode)) { |
2393 | WARN_ON(1); | |
2394 | iput(inode); | |
2395 | continue; | |
2396 | } | |
7b128766 | 2397 | nr_truncate++; |
66b4ffd1 | 2398 | ret = btrfs_truncate(inode); |
7b128766 JB |
2399 | } else { |
2400 | nr_unlink++; | |
2401 | } | |
2402 | ||
2403 | /* this will do delete_inode and everything for us */ | |
2404 | iput(inode); | |
66b4ffd1 JB |
2405 | if (ret) |
2406 | goto out; | |
7b128766 | 2407 | } |
d68fc57b YZ |
2408 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2409 | ||
2410 | if (root->orphan_block_rsv) | |
2411 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2412 | (u64)-1); | |
2413 | ||
2414 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
2415 | trans = btrfs_join_transaction(root, 1); | |
66b4ffd1 JB |
2416 | if (!IS_ERR(trans)) |
2417 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2418 | } |
7b128766 JB |
2419 | |
2420 | if (nr_unlink) | |
2421 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2422 | if (nr_truncate) | |
2423 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2424 | |
2425 | out: | |
2426 | if (ret) | |
2427 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2428 | btrfs_free_path(path); | |
2429 | return ret; | |
7b128766 JB |
2430 | } |
2431 | ||
46a53cca CM |
2432 | /* |
2433 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2434 | * don't find any xattrs, we know there can't be any acls. | |
2435 | * | |
2436 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2437 | */ | |
2438 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2439 | int slot, u64 objectid) | |
2440 | { | |
2441 | u32 nritems = btrfs_header_nritems(leaf); | |
2442 | struct btrfs_key found_key; | |
2443 | int scanned = 0; | |
2444 | ||
2445 | slot++; | |
2446 | while (slot < nritems) { | |
2447 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2448 | ||
2449 | /* we found a different objectid, there must not be acls */ | |
2450 | if (found_key.objectid != objectid) | |
2451 | return 0; | |
2452 | ||
2453 | /* we found an xattr, assume we've got an acl */ | |
2454 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2455 | return 1; | |
2456 | ||
2457 | /* | |
2458 | * we found a key greater than an xattr key, there can't | |
2459 | * be any acls later on | |
2460 | */ | |
2461 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2462 | return 0; | |
2463 | ||
2464 | slot++; | |
2465 | scanned++; | |
2466 | ||
2467 | /* | |
2468 | * it goes inode, inode backrefs, xattrs, extents, | |
2469 | * so if there are a ton of hard links to an inode there can | |
2470 | * be a lot of backrefs. Don't waste time searching too hard, | |
2471 | * this is just an optimization | |
2472 | */ | |
2473 | if (scanned >= 8) | |
2474 | break; | |
2475 | } | |
2476 | /* we hit the end of the leaf before we found an xattr or | |
2477 | * something larger than an xattr. We have to assume the inode | |
2478 | * has acls | |
2479 | */ | |
2480 | return 1; | |
2481 | } | |
2482 | ||
d352ac68 CM |
2483 | /* |
2484 | * read an inode from the btree into the in-memory inode | |
2485 | */ | |
5d4f98a2 | 2486 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2487 | { |
2488 | struct btrfs_path *path; | |
5f39d397 | 2489 | struct extent_buffer *leaf; |
39279cc3 | 2490 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2491 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2492 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2493 | struct btrfs_key location; | |
46a53cca | 2494 | int maybe_acls; |
39279cc3 | 2495 | u64 alloc_group_block; |
618e21d5 | 2496 | u32 rdev; |
39279cc3 CM |
2497 | int ret; |
2498 | ||
2499 | path = btrfs_alloc_path(); | |
2500 | BUG_ON(!path); | |
39279cc3 | 2501 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2502 | |
39279cc3 | 2503 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2504 | if (ret) |
39279cc3 | 2505 | goto make_bad; |
39279cc3 | 2506 | |
5f39d397 CM |
2507 | leaf = path->nodes[0]; |
2508 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
2509 | struct btrfs_inode_item); | |
2510 | ||
2511 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); | |
2512 | inode->i_nlink = btrfs_inode_nlink(leaf, inode_item); | |
2513 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); | |
2514 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2515 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2516 | |
2517 | tspec = btrfs_inode_atime(inode_item); | |
2518 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2519 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2520 | ||
2521 | tspec = btrfs_inode_mtime(inode_item); | |
2522 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2523 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2524 | ||
2525 | tspec = btrfs_inode_ctime(inode_item); | |
2526 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2527 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2528 | ||
a76a3cd4 | 2529 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2530 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2531 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2532 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2533 | inode->i_rdev = 0; |
5f39d397 CM |
2534 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2535 | ||
aec7477b | 2536 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2537 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
aec7477b | 2538 | |
5f39d397 | 2539 | alloc_group_block = btrfs_inode_block_group(leaf, inode_item); |
b4ce94de | 2540 | |
46a53cca CM |
2541 | /* |
2542 | * try to precache a NULL acl entry for files that don't have | |
2543 | * any xattrs or acls | |
2544 | */ | |
2545 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino); | |
72c04902 AV |
2546 | if (!maybe_acls) |
2547 | cache_no_acl(inode); | |
46a53cca | 2548 | |
d2fb3437 YZ |
2549 | BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0, |
2550 | alloc_group_block, 0); | |
39279cc3 CM |
2551 | btrfs_free_path(path); |
2552 | inode_item = NULL; | |
2553 | ||
39279cc3 | 2554 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2555 | case S_IFREG: |
2556 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2557 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2558 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2559 | inode->i_fop = &btrfs_file_operations; |
2560 | inode->i_op = &btrfs_file_inode_operations; | |
2561 | break; | |
2562 | case S_IFDIR: | |
2563 | inode->i_fop = &btrfs_dir_file_operations; | |
2564 | if (root == root->fs_info->tree_root) | |
2565 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2566 | else | |
2567 | inode->i_op = &btrfs_dir_inode_operations; | |
2568 | break; | |
2569 | case S_IFLNK: | |
2570 | inode->i_op = &btrfs_symlink_inode_operations; | |
2571 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2572 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2573 | break; |
618e21d5 | 2574 | default: |
0279b4cd | 2575 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2576 | init_special_inode(inode, inode->i_mode, rdev); |
2577 | break; | |
39279cc3 | 2578 | } |
6cbff00f CH |
2579 | |
2580 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2581 | return; |
2582 | ||
2583 | make_bad: | |
39279cc3 | 2584 | btrfs_free_path(path); |
39279cc3 CM |
2585 | make_bad_inode(inode); |
2586 | } | |
2587 | ||
d352ac68 CM |
2588 | /* |
2589 | * given a leaf and an inode, copy the inode fields into the leaf | |
2590 | */ | |
e02119d5 CM |
2591 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2592 | struct extent_buffer *leaf, | |
5f39d397 | 2593 | struct btrfs_inode_item *item, |
39279cc3 CM |
2594 | struct inode *inode) |
2595 | { | |
12ddb96c JB |
2596 | if (!leaf->map_token) |
2597 | map_private_extent_buffer(leaf, (unsigned long)item, | |
2598 | sizeof(struct btrfs_inode_item), | |
2599 | &leaf->map_token, &leaf->kaddr, | |
2600 | &leaf->map_start, &leaf->map_len, | |
2601 | KM_USER1); | |
2602 | ||
5f39d397 CM |
2603 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2604 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2605 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2606 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2607 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2608 | ||
2609 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2610 | inode->i_atime.tv_sec); | |
2611 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2612 | inode->i_atime.tv_nsec); | |
2613 | ||
2614 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2615 | inode->i_mtime.tv_sec); | |
2616 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2617 | inode->i_mtime.tv_nsec); | |
2618 | ||
2619 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2620 | inode->i_ctime.tv_sec); | |
2621 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2622 | inode->i_ctime.tv_nsec); | |
2623 | ||
a76a3cd4 | 2624 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2625 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2626 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2627 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2628 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2629 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d2fb3437 | 2630 | btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group); |
12ddb96c JB |
2631 | |
2632 | if (leaf->map_token) { | |
2633 | unmap_extent_buffer(leaf, leaf->map_token, KM_USER1); | |
2634 | leaf->map_token = NULL; | |
2635 | } | |
39279cc3 CM |
2636 | } |
2637 | ||
d352ac68 CM |
2638 | /* |
2639 | * copy everything in the in-memory inode into the btree. | |
2640 | */ | |
d397712b CM |
2641 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, |
2642 | struct btrfs_root *root, struct inode *inode) | |
39279cc3 CM |
2643 | { |
2644 | struct btrfs_inode_item *inode_item; | |
2645 | struct btrfs_path *path; | |
5f39d397 | 2646 | struct extent_buffer *leaf; |
39279cc3 CM |
2647 | int ret; |
2648 | ||
2649 | path = btrfs_alloc_path(); | |
2650 | BUG_ON(!path); | |
b9473439 | 2651 | path->leave_spinning = 1; |
39279cc3 CM |
2652 | ret = btrfs_lookup_inode(trans, root, path, |
2653 | &BTRFS_I(inode)->location, 1); | |
2654 | if (ret) { | |
2655 | if (ret > 0) | |
2656 | ret = -ENOENT; | |
2657 | goto failed; | |
2658 | } | |
2659 | ||
b4ce94de | 2660 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2661 | leaf = path->nodes[0]; |
2662 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
39279cc3 CM |
2663 | struct btrfs_inode_item); |
2664 | ||
e02119d5 | 2665 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2666 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2667 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2668 | ret = 0; |
2669 | failed: | |
39279cc3 CM |
2670 | btrfs_free_path(path); |
2671 | return ret; | |
2672 | } | |
2673 | ||
2674 | ||
d352ac68 CM |
2675 | /* |
2676 | * unlink helper that gets used here in inode.c and in the tree logging | |
2677 | * recovery code. It remove a link in a directory with a given name, and | |
2678 | * also drops the back refs in the inode to the directory | |
2679 | */ | |
92986796 AV |
2680 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2681 | struct btrfs_root *root, | |
2682 | struct inode *dir, struct inode *inode, | |
2683 | const char *name, int name_len) | |
39279cc3 CM |
2684 | { |
2685 | struct btrfs_path *path; | |
39279cc3 | 2686 | int ret = 0; |
5f39d397 | 2687 | struct extent_buffer *leaf; |
39279cc3 | 2688 | struct btrfs_dir_item *di; |
5f39d397 | 2689 | struct btrfs_key key; |
aec7477b | 2690 | u64 index; |
39279cc3 CM |
2691 | |
2692 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2693 | if (!path) { |
2694 | ret = -ENOMEM; | |
554233a6 | 2695 | goto out; |
54aa1f4d CM |
2696 | } |
2697 | ||
b9473439 | 2698 | path->leave_spinning = 1; |
39279cc3 CM |
2699 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, |
2700 | name, name_len, -1); | |
2701 | if (IS_ERR(di)) { | |
2702 | ret = PTR_ERR(di); | |
2703 | goto err; | |
2704 | } | |
2705 | if (!di) { | |
2706 | ret = -ENOENT; | |
2707 | goto err; | |
2708 | } | |
5f39d397 CM |
2709 | leaf = path->nodes[0]; |
2710 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2711 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2712 | if (ret) |
2713 | goto err; | |
39279cc3 CM |
2714 | btrfs_release_path(root, path); |
2715 | ||
aec7477b | 2716 | ret = btrfs_del_inode_ref(trans, root, name, name_len, |
e02119d5 CM |
2717 | inode->i_ino, |
2718 | dir->i_ino, &index); | |
aec7477b | 2719 | if (ret) { |
d397712b | 2720 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
aec7477b | 2721 | "inode %lu parent %lu\n", name_len, name, |
e02119d5 | 2722 | inode->i_ino, dir->i_ino); |
aec7477b JB |
2723 | goto err; |
2724 | } | |
2725 | ||
39279cc3 | 2726 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, |
aec7477b | 2727 | index, name, name_len, -1); |
39279cc3 CM |
2728 | if (IS_ERR(di)) { |
2729 | ret = PTR_ERR(di); | |
2730 | goto err; | |
2731 | } | |
2732 | if (!di) { | |
2733 | ret = -ENOENT; | |
2734 | goto err; | |
2735 | } | |
2736 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
925baedd | 2737 | btrfs_release_path(root, path); |
39279cc3 | 2738 | |
e02119d5 CM |
2739 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
2740 | inode, dir->i_ino); | |
49eb7e46 | 2741 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2742 | |
2743 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2744 | dir, index); | |
6418c961 CM |
2745 | if (ret == -ENOENT) |
2746 | ret = 0; | |
39279cc3 CM |
2747 | err: |
2748 | btrfs_free_path(path); | |
e02119d5 CM |
2749 | if (ret) |
2750 | goto out; | |
2751 | ||
2752 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2753 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2754 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2755 | out: |
39279cc3 CM |
2756 | return ret; |
2757 | } | |
2758 | ||
92986796 AV |
2759 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2760 | struct btrfs_root *root, | |
2761 | struct inode *dir, struct inode *inode, | |
2762 | const char *name, int name_len) | |
2763 | { | |
2764 | int ret; | |
2765 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2766 | if (!ret) { | |
2767 | btrfs_drop_nlink(inode); | |
2768 | ret = btrfs_update_inode(trans, root, inode); | |
2769 | } | |
2770 | return ret; | |
2771 | } | |
2772 | ||
2773 | ||
a22285a6 YZ |
2774 | /* helper to check if there is any shared block in the path */ |
2775 | static int check_path_shared(struct btrfs_root *root, | |
2776 | struct btrfs_path *path) | |
39279cc3 | 2777 | { |
a22285a6 YZ |
2778 | struct extent_buffer *eb; |
2779 | int level; | |
0e4dcbef | 2780 | u64 refs = 1; |
5df6a9f6 | 2781 | |
a22285a6 | 2782 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2783 | int ret; |
2784 | ||
a22285a6 YZ |
2785 | if (!path->nodes[level]) |
2786 | break; | |
2787 | eb = path->nodes[level]; | |
2788 | if (!btrfs_block_can_be_shared(root, eb)) | |
2789 | continue; | |
2790 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2791 | &refs, NULL); | |
2792 | if (refs > 1) | |
2793 | return 1; | |
5df6a9f6 | 2794 | } |
dedefd72 | 2795 | return 0; |
39279cc3 CM |
2796 | } |
2797 | ||
a22285a6 YZ |
2798 | /* |
2799 | * helper to start transaction for unlink and rmdir. | |
2800 | * | |
2801 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2802 | * so in enospc case, we should make sure they will free space before | |
2803 | * allowing them to use the global metadata reservation. | |
2804 | */ | |
2805 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2806 | struct dentry *dentry) | |
4df27c4d | 2807 | { |
39279cc3 | 2808 | struct btrfs_trans_handle *trans; |
a22285a6 | 2809 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2810 | struct btrfs_path *path; |
a22285a6 | 2811 | struct btrfs_inode_ref *ref; |
4df27c4d | 2812 | struct btrfs_dir_item *di; |
7b128766 | 2813 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2814 | u64 index; |
a22285a6 YZ |
2815 | int check_link = 1; |
2816 | int err = -ENOSPC; | |
4df27c4d YZ |
2817 | int ret; |
2818 | ||
a22285a6 YZ |
2819 | trans = btrfs_start_transaction(root, 10); |
2820 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) | |
2821 | return trans; | |
4df27c4d | 2822 | |
a22285a6 YZ |
2823 | if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
2824 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2825 | |
a22285a6 YZ |
2826 | /* check if there is someone else holds reference */ |
2827 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2828 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2829 | |
a22285a6 YZ |
2830 | if (atomic_read(&inode->i_count) > 2) |
2831 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2832 | |
a22285a6 YZ |
2833 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2834 | return ERR_PTR(-ENOSPC); | |
2835 | ||
2836 | path = btrfs_alloc_path(); | |
2837 | if (!path) { | |
2838 | root->fs_info->enospc_unlink = 0; | |
2839 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2840 | } |
2841 | ||
a22285a6 | 2842 | trans = btrfs_start_transaction(root, 0); |
5df6a9f6 | 2843 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2844 | btrfs_free_path(path); |
2845 | root->fs_info->enospc_unlink = 0; | |
2846 | return trans; | |
2847 | } | |
4df27c4d | 2848 | |
a22285a6 YZ |
2849 | path->skip_locking = 1; |
2850 | path->search_commit_root = 1; | |
4df27c4d | 2851 | |
a22285a6 YZ |
2852 | ret = btrfs_lookup_inode(trans, root, path, |
2853 | &BTRFS_I(dir)->location, 0); | |
2854 | if (ret < 0) { | |
2855 | err = ret; | |
2856 | goto out; | |
2857 | } | |
2858 | if (ret == 0) { | |
2859 | if (check_path_shared(root, path)) | |
2860 | goto out; | |
2861 | } else { | |
2862 | check_link = 0; | |
5df6a9f6 | 2863 | } |
a22285a6 YZ |
2864 | btrfs_release_path(root, path); |
2865 | ||
2866 | ret = btrfs_lookup_inode(trans, root, path, | |
2867 | &BTRFS_I(inode)->location, 0); | |
2868 | if (ret < 0) { | |
2869 | err = ret; | |
2870 | goto out; | |
2871 | } | |
2872 | if (ret == 0) { | |
2873 | if (check_path_shared(root, path)) | |
2874 | goto out; | |
2875 | } else { | |
2876 | check_link = 0; | |
2877 | } | |
2878 | btrfs_release_path(root, path); | |
2879 | ||
2880 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2881 | ret = btrfs_lookup_file_extent(trans, root, path, | |
2882 | inode->i_ino, (u64)-1, 0); | |
2883 | if (ret < 0) { | |
2884 | err = ret; | |
2885 | goto out; | |
2886 | } | |
2887 | BUG_ON(ret == 0); | |
2888 | if (check_path_shared(root, path)) | |
2889 | goto out; | |
2890 | btrfs_release_path(root, path); | |
2891 | } | |
2892 | ||
2893 | if (!check_link) { | |
2894 | err = 0; | |
2895 | goto out; | |
2896 | } | |
2897 | ||
2898 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
2899 | dentry->d_name.name, dentry->d_name.len, 0); | |
2900 | if (IS_ERR(di)) { | |
2901 | err = PTR_ERR(di); | |
2902 | goto out; | |
2903 | } | |
2904 | if (di) { | |
2905 | if (check_path_shared(root, path)) | |
2906 | goto out; | |
2907 | } else { | |
2908 | err = 0; | |
2909 | goto out; | |
2910 | } | |
2911 | btrfs_release_path(root, path); | |
2912 | ||
2913 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2914 | dentry->d_name.name, dentry->d_name.len, | |
2915 | inode->i_ino, dir->i_ino, 0); | |
2916 | if (IS_ERR(ref)) { | |
2917 | err = PTR_ERR(ref); | |
2918 | goto out; | |
2919 | } | |
2920 | BUG_ON(!ref); | |
2921 | if (check_path_shared(root, path)) | |
2922 | goto out; | |
2923 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
2924 | btrfs_release_path(root, path); | |
2925 | ||
2926 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index, | |
2927 | dentry->d_name.name, dentry->d_name.len, 0); | |
2928 | if (IS_ERR(di)) { | |
2929 | err = PTR_ERR(di); | |
2930 | goto out; | |
2931 | } | |
2932 | BUG_ON(ret == -ENOENT); | |
2933 | if (check_path_shared(root, path)) | |
2934 | goto out; | |
2935 | ||
2936 | err = 0; | |
2937 | out: | |
2938 | btrfs_free_path(path); | |
2939 | if (err) { | |
2940 | btrfs_end_transaction(trans, root); | |
2941 | root->fs_info->enospc_unlink = 0; | |
2942 | return ERR_PTR(err); | |
2943 | } | |
2944 | ||
2945 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2946 | return trans; | |
2947 | } | |
2948 | ||
2949 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2950 | struct btrfs_root *root) | |
2951 | { | |
2952 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
2953 | BUG_ON(!root->fs_info->enospc_unlink); | |
2954 | root->fs_info->enospc_unlink = 0; | |
2955 | } | |
2956 | btrfs_end_transaction_throttle(trans, root); | |
2957 | } | |
2958 | ||
2959 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2960 | { | |
2961 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2962 | struct btrfs_trans_handle *trans; | |
2963 | struct inode *inode = dentry->d_inode; | |
2964 | int ret; | |
2965 | unsigned long nr = 0; | |
2966 | ||
2967 | trans = __unlink_start_trans(dir, dentry); | |
2968 | if (IS_ERR(trans)) | |
2969 | return PTR_ERR(trans); | |
5f39d397 | 2970 | |
39279cc3 | 2971 | btrfs_set_trans_block_group(trans, dir); |
12fcfd22 CM |
2972 | |
2973 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); | |
2974 | ||
e02119d5 CM |
2975 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2976 | dentry->d_name.name, dentry->d_name.len); | |
a22285a6 | 2977 | BUG_ON(ret); |
7b128766 | 2978 | |
a22285a6 | 2979 | if (inode->i_nlink == 0) { |
7b128766 | 2980 | ret = btrfs_orphan_add(trans, inode); |
a22285a6 YZ |
2981 | BUG_ON(ret); |
2982 | } | |
7b128766 | 2983 | |
d3c2fdcf | 2984 | nr = trans->blocks_used; |
a22285a6 | 2985 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2986 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2987 | return ret; |
2988 | } | |
2989 | ||
4df27c4d YZ |
2990 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2991 | struct btrfs_root *root, | |
2992 | struct inode *dir, u64 objectid, | |
2993 | const char *name, int name_len) | |
2994 | { | |
2995 | struct btrfs_path *path; | |
2996 | struct extent_buffer *leaf; | |
2997 | struct btrfs_dir_item *di; | |
2998 | struct btrfs_key key; | |
2999 | u64 index; | |
3000 | int ret; | |
3001 | ||
3002 | path = btrfs_alloc_path(); | |
3003 | if (!path) | |
3004 | return -ENOMEM; | |
3005 | ||
3006 | di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, | |
3007 | name, name_len, -1); | |
3008 | BUG_ON(!di || IS_ERR(di)); | |
3009 | ||
3010 | leaf = path->nodes[0]; | |
3011 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3012 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3013 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3014 | BUG_ON(ret); | |
3015 | btrfs_release_path(root, path); | |
3016 | ||
3017 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3018 | objectid, root->root_key.objectid, | |
3019 | dir->i_ino, &index, name, name_len); | |
3020 | if (ret < 0) { | |
3021 | BUG_ON(ret != -ENOENT); | |
3022 | di = btrfs_search_dir_index_item(root, path, dir->i_ino, | |
3023 | name, name_len); | |
3024 | BUG_ON(!di || IS_ERR(di)); | |
3025 | ||
3026 | leaf = path->nodes[0]; | |
3027 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3028 | btrfs_release_path(root, path); | |
3029 | index = key.offset; | |
3030 | } | |
3031 | ||
3032 | di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, | |
3033 | index, name, name_len, -1); | |
3034 | BUG_ON(!di || IS_ERR(di)); | |
3035 | ||
3036 | leaf = path->nodes[0]; | |
3037 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3038 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3039 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
3040 | BUG_ON(ret); | |
3041 | btrfs_release_path(root, path); | |
3042 | ||
3043 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
3044 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
3045 | ret = btrfs_update_inode(trans, root, dir); | |
3046 | BUG_ON(ret); | |
4df27c4d YZ |
3047 | |
3048 | btrfs_free_path(path); | |
3049 | return 0; | |
3050 | } | |
3051 | ||
39279cc3 CM |
3052 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3053 | { | |
3054 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3055 | int err = 0; |
39279cc3 | 3056 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3057 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3058 | unsigned long nr = 0; |
39279cc3 | 3059 | |
3394e160 | 3060 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
4df27c4d | 3061 | inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
3062 | return -ENOTEMPTY; |
3063 | ||
a22285a6 YZ |
3064 | trans = __unlink_start_trans(dir, dentry); |
3065 | if (IS_ERR(trans)) | |
5df6a9f6 | 3066 | return PTR_ERR(trans); |
5df6a9f6 | 3067 | |
39279cc3 | 3068 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 3069 | |
4df27c4d YZ |
3070 | if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
3071 | err = btrfs_unlink_subvol(trans, root, dir, | |
3072 | BTRFS_I(inode)->location.objectid, | |
3073 | dentry->d_name.name, | |
3074 | dentry->d_name.len); | |
3075 | goto out; | |
3076 | } | |
3077 | ||
7b128766 JB |
3078 | err = btrfs_orphan_add(trans, inode); |
3079 | if (err) | |
4df27c4d | 3080 | goto out; |
7b128766 | 3081 | |
39279cc3 | 3082 | /* now the directory is empty */ |
e02119d5 CM |
3083 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3084 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3085 | if (!err) |
dbe674a9 | 3086 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3087 | out: |
d3c2fdcf | 3088 | nr = trans->blocks_used; |
a22285a6 | 3089 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3090 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3091 | |
39279cc3 CM |
3092 | return err; |
3093 | } | |
3094 | ||
d20f7043 | 3095 | #if 0 |
323ac95b CM |
3096 | /* |
3097 | * when truncating bytes in a file, it is possible to avoid reading | |
3098 | * the leaves that contain only checksum items. This can be the | |
3099 | * majority of the IO required to delete a large file, but it must | |
3100 | * be done carefully. | |
3101 | * | |
3102 | * The keys in the level just above the leaves are checked to make sure | |
3103 | * the lowest key in a given leaf is a csum key, and starts at an offset | |
3104 | * after the new size. | |
3105 | * | |
3106 | * Then the key for the next leaf is checked to make sure it also has | |
3107 | * a checksum item for the same file. If it does, we know our target leaf | |
3108 | * contains only checksum items, and it can be safely freed without reading | |
3109 | * it. | |
3110 | * | |
3111 | * This is just an optimization targeted at large files. It may do | |
3112 | * nothing. It will return 0 unless things went badly. | |
3113 | */ | |
3114 | static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans, | |
3115 | struct btrfs_root *root, | |
3116 | struct btrfs_path *path, | |
3117 | struct inode *inode, u64 new_size) | |
3118 | { | |
3119 | struct btrfs_key key; | |
3120 | int ret; | |
3121 | int nritems; | |
3122 | struct btrfs_key found_key; | |
3123 | struct btrfs_key other_key; | |
5b84e8d6 YZ |
3124 | struct btrfs_leaf_ref *ref; |
3125 | u64 leaf_gen; | |
3126 | u64 leaf_start; | |
323ac95b CM |
3127 | |
3128 | path->lowest_level = 1; | |
3129 | key.objectid = inode->i_ino; | |
3130 | key.type = BTRFS_CSUM_ITEM_KEY; | |
3131 | key.offset = new_size; | |
3132 | again: | |
3133 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3134 | if (ret < 0) | |
3135 | goto out; | |
3136 | ||
3137 | if (path->nodes[1] == NULL) { | |
3138 | ret = 0; | |
3139 | goto out; | |
3140 | } | |
3141 | ret = 0; | |
3142 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]); | |
3143 | nritems = btrfs_header_nritems(path->nodes[1]); | |
3144 | ||
3145 | if (!nritems) | |
3146 | goto out; | |
3147 | ||
3148 | if (path->slots[1] >= nritems) | |
3149 | goto next_node; | |
3150 | ||
3151 | /* did we find a key greater than anything we want to delete? */ | |
3152 | if (found_key.objectid > inode->i_ino || | |
3153 | (found_key.objectid == inode->i_ino && found_key.type > key.type)) | |
3154 | goto out; | |
3155 | ||
3156 | /* we check the next key in the node to make sure the leave contains | |
3157 | * only checksum items. This comparison doesn't work if our | |
3158 | * leaf is the last one in the node | |
3159 | */ | |
3160 | if (path->slots[1] + 1 >= nritems) { | |
3161 | next_node: | |
3162 | /* search forward from the last key in the node, this | |
3163 | * will bring us into the next node in the tree | |
3164 | */ | |
3165 | btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1); | |
3166 | ||
3167 | /* unlikely, but we inc below, so check to be safe */ | |
3168 | if (found_key.offset == (u64)-1) | |
3169 | goto out; | |
3170 | ||
3171 | /* search_forward needs a path with locks held, do the | |
3172 | * search again for the original key. It is possible | |
3173 | * this will race with a balance and return a path that | |
3174 | * we could modify, but this drop is just an optimization | |
3175 | * and is allowed to miss some leaves. | |
3176 | */ | |
3177 | btrfs_release_path(root, path); | |
3178 | found_key.offset++; | |
3179 | ||
3180 | /* setup a max key for search_forward */ | |
3181 | other_key.offset = (u64)-1; | |
3182 | other_key.type = key.type; | |
3183 | other_key.objectid = key.objectid; | |
3184 | ||
3185 | path->keep_locks = 1; | |
3186 | ret = btrfs_search_forward(root, &found_key, &other_key, | |
3187 | path, 0, 0); | |
3188 | path->keep_locks = 0; | |
3189 | if (ret || found_key.objectid != key.objectid || | |
3190 | found_key.type != key.type) { | |
3191 | ret = 0; | |
3192 | goto out; | |
3193 | } | |
3194 | ||
3195 | key.offset = found_key.offset; | |
3196 | btrfs_release_path(root, path); | |
3197 | cond_resched(); | |
3198 | goto again; | |
3199 | } | |
3200 | ||
3201 | /* we know there's one more slot after us in the tree, | |
3202 | * read that key so we can verify it is also a checksum item | |
3203 | */ | |
3204 | btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1); | |
3205 | ||
3206 | if (found_key.objectid < inode->i_ino) | |
3207 | goto next_key; | |
3208 | ||
3209 | if (found_key.type != key.type || found_key.offset < new_size) | |
3210 | goto next_key; | |
3211 | ||
3212 | /* | |
3213 | * if the key for the next leaf isn't a csum key from this objectid, | |
3214 | * we can't be sure there aren't good items inside this leaf. | |
3215 | * Bail out | |
3216 | */ | |
3217 | if (other_key.objectid != inode->i_ino || other_key.type != key.type) | |
3218 | goto out; | |
3219 | ||
5b84e8d6 YZ |
3220 | leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]); |
3221 | leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]); | |
323ac95b CM |
3222 | /* |
3223 | * it is safe to delete this leaf, it contains only | |
3224 | * csum items from this inode at an offset >= new_size | |
3225 | */ | |
5b84e8d6 | 3226 | ret = btrfs_del_leaf(trans, root, path, leaf_start); |
323ac95b CM |
3227 | BUG_ON(ret); |
3228 | ||
5b84e8d6 YZ |
3229 | if (root->ref_cows && leaf_gen < trans->transid) { |
3230 | ref = btrfs_alloc_leaf_ref(root, 0); | |
3231 | if (ref) { | |
3232 | ref->root_gen = root->root_key.offset; | |
3233 | ref->bytenr = leaf_start; | |
3234 | ref->owner = 0; | |
3235 | ref->generation = leaf_gen; | |
3236 | ref->nritems = 0; | |
3237 | ||
bd56b302 CM |
3238 | btrfs_sort_leaf_ref(ref); |
3239 | ||
5b84e8d6 YZ |
3240 | ret = btrfs_add_leaf_ref(root, ref, 0); |
3241 | WARN_ON(ret); | |
3242 | btrfs_free_leaf_ref(root, ref); | |
3243 | } else { | |
3244 | WARN_ON(1); | |
3245 | } | |
3246 | } | |
323ac95b CM |
3247 | next_key: |
3248 | btrfs_release_path(root, path); | |
3249 | ||
3250 | if (other_key.objectid == inode->i_ino && | |
3251 | other_key.type == key.type && other_key.offset > key.offset) { | |
3252 | key.offset = other_key.offset; | |
3253 | cond_resched(); | |
3254 | goto again; | |
3255 | } | |
3256 | ret = 0; | |
3257 | out: | |
3258 | /* fixup any changes we've made to the path */ | |
3259 | path->lowest_level = 0; | |
3260 | path->keep_locks = 0; | |
3261 | btrfs_release_path(root, path); | |
3262 | return ret; | |
3263 | } | |
3264 | ||
d20f7043 CM |
3265 | #endif |
3266 | ||
39279cc3 CM |
3267 | /* |
3268 | * this can truncate away extent items, csum items and directory items. | |
3269 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3270 | * any higher than new_size |
39279cc3 CM |
3271 | * |
3272 | * csum items that cross the new i_size are truncated to the new size | |
3273 | * as well. | |
7b128766 JB |
3274 | * |
3275 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3276 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3277 | */ |
8082510e YZ |
3278 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3279 | struct btrfs_root *root, | |
3280 | struct inode *inode, | |
3281 | u64 new_size, u32 min_type) | |
39279cc3 | 3282 | { |
39279cc3 | 3283 | struct btrfs_path *path; |
5f39d397 | 3284 | struct extent_buffer *leaf; |
39279cc3 | 3285 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3286 | struct btrfs_key key; |
3287 | struct btrfs_key found_key; | |
39279cc3 | 3288 | u64 extent_start = 0; |
db94535d | 3289 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3290 | u64 extent_offset = 0; |
39279cc3 | 3291 | u64 item_end = 0; |
8082510e YZ |
3292 | u64 mask = root->sectorsize - 1; |
3293 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3294 | int found_extent; |
3295 | int del_item; | |
85e21bac CM |
3296 | int pending_del_nr = 0; |
3297 | int pending_del_slot = 0; | |
179e29e4 | 3298 | int extent_type = -1; |
771ed689 | 3299 | int encoding; |
8082510e YZ |
3300 | int ret; |
3301 | int err = 0; | |
3302 | ||
3303 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3304 | |
0af3d00b | 3305 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3306 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3307 | |
39279cc3 CM |
3308 | path = btrfs_alloc_path(); |
3309 | BUG_ON(!path); | |
33c17ad5 | 3310 | path->reada = -1; |
5f39d397 | 3311 | |
39279cc3 CM |
3312 | key.objectid = inode->i_ino; |
3313 | key.offset = (u64)-1; | |
5f39d397 CM |
3314 | key.type = (u8)-1; |
3315 | ||
85e21bac | 3316 | search_again: |
b9473439 | 3317 | path->leave_spinning = 1; |
85e21bac | 3318 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3319 | if (ret < 0) { |
3320 | err = ret; | |
3321 | goto out; | |
3322 | } | |
d397712b | 3323 | |
85e21bac | 3324 | if (ret > 0) { |
e02119d5 CM |
3325 | /* there are no items in the tree for us to truncate, we're |
3326 | * done | |
3327 | */ | |
8082510e YZ |
3328 | if (path->slots[0] == 0) |
3329 | goto out; | |
85e21bac CM |
3330 | path->slots[0]--; |
3331 | } | |
3332 | ||
d397712b | 3333 | while (1) { |
39279cc3 | 3334 | fi = NULL; |
5f39d397 CM |
3335 | leaf = path->nodes[0]; |
3336 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3337 | found_type = btrfs_key_type(&found_key); | |
771ed689 | 3338 | encoding = 0; |
39279cc3 | 3339 | |
5f39d397 | 3340 | if (found_key.objectid != inode->i_ino) |
39279cc3 | 3341 | break; |
5f39d397 | 3342 | |
85e21bac | 3343 | if (found_type < min_type) |
39279cc3 CM |
3344 | break; |
3345 | ||
5f39d397 | 3346 | item_end = found_key.offset; |
39279cc3 | 3347 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3348 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3349 | struct btrfs_file_extent_item); |
179e29e4 | 3350 | extent_type = btrfs_file_extent_type(leaf, fi); |
771ed689 CM |
3351 | encoding = btrfs_file_extent_compression(leaf, fi); |
3352 | encoding |= btrfs_file_extent_encryption(leaf, fi); | |
3353 | encoding |= btrfs_file_extent_other_encoding(leaf, fi); | |
3354 | ||
179e29e4 | 3355 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
5f39d397 | 3356 | item_end += |
db94535d | 3357 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3358 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3359 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3360 | fi); |
39279cc3 | 3361 | } |
008630c1 | 3362 | item_end--; |
39279cc3 | 3363 | } |
8082510e YZ |
3364 | if (found_type > min_type) { |
3365 | del_item = 1; | |
3366 | } else { | |
3367 | if (item_end < new_size) | |
b888db2b | 3368 | break; |
8082510e YZ |
3369 | if (found_key.offset >= new_size) |
3370 | del_item = 1; | |
3371 | else | |
3372 | del_item = 0; | |
39279cc3 | 3373 | } |
39279cc3 | 3374 | found_extent = 0; |
39279cc3 | 3375 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3376 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3377 | goto delete; | |
3378 | ||
3379 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3380 | u64 num_dec; |
db94535d | 3381 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
771ed689 | 3382 | if (!del_item && !encoding) { |
db94535d CM |
3383 | u64 orig_num_bytes = |
3384 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3385 | extent_num_bytes = new_size - |
5f39d397 | 3386 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3387 | extent_num_bytes = extent_num_bytes & |
3388 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3389 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3390 | extent_num_bytes); | |
3391 | num_dec = (orig_num_bytes - | |
9069218d | 3392 | extent_num_bytes); |
e02119d5 | 3393 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3394 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3395 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3396 | } else { |
db94535d CM |
3397 | extent_num_bytes = |
3398 | btrfs_file_extent_disk_num_bytes(leaf, | |
3399 | fi); | |
5d4f98a2 YZ |
3400 | extent_offset = found_key.offset - |
3401 | btrfs_file_extent_offset(leaf, fi); | |
3402 | ||
39279cc3 | 3403 | /* FIXME blocksize != 4096 */ |
9069218d | 3404 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3405 | if (extent_start != 0) { |
3406 | found_extent = 1; | |
e02119d5 | 3407 | if (root->ref_cows) |
a76a3cd4 | 3408 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3409 | } |
39279cc3 | 3410 | } |
9069218d | 3411 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3412 | /* |
3413 | * we can't truncate inline items that have had | |
3414 | * special encodings | |
3415 | */ | |
3416 | if (!del_item && | |
3417 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3418 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3419 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3420 | u32 size = new_size - found_key.offset; |
3421 | ||
3422 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3423 | inode_sub_bytes(inode, item_end + 1 - |
3424 | new_size); | |
e02119d5 CM |
3425 | } |
3426 | size = | |
3427 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3428 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3429 | size, 1); |
9069218d | 3430 | BUG_ON(ret); |
e02119d5 | 3431 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3432 | inode_sub_bytes(inode, item_end + 1 - |
3433 | found_key.offset); | |
9069218d | 3434 | } |
39279cc3 | 3435 | } |
179e29e4 | 3436 | delete: |
39279cc3 | 3437 | if (del_item) { |
85e21bac CM |
3438 | if (!pending_del_nr) { |
3439 | /* no pending yet, add ourselves */ | |
3440 | pending_del_slot = path->slots[0]; | |
3441 | pending_del_nr = 1; | |
3442 | } else if (pending_del_nr && | |
3443 | path->slots[0] + 1 == pending_del_slot) { | |
3444 | /* hop on the pending chunk */ | |
3445 | pending_del_nr++; | |
3446 | pending_del_slot = path->slots[0]; | |
3447 | } else { | |
d397712b | 3448 | BUG(); |
85e21bac | 3449 | } |
39279cc3 CM |
3450 | } else { |
3451 | break; | |
3452 | } | |
0af3d00b JB |
3453 | if (found_extent && (root->ref_cows || |
3454 | root == root->fs_info->tree_root)) { | |
b9473439 | 3455 | btrfs_set_path_blocking(path); |
39279cc3 | 3456 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3457 | extent_num_bytes, 0, |
3458 | btrfs_header_owner(leaf), | |
3459 | inode->i_ino, extent_offset); | |
39279cc3 CM |
3460 | BUG_ON(ret); |
3461 | } | |
85e21bac | 3462 | |
8082510e YZ |
3463 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3464 | break; | |
3465 | ||
3466 | if (path->slots[0] == 0 || | |
3467 | path->slots[0] != pending_del_slot) { | |
3468 | if (root->ref_cows) { | |
3469 | err = -EAGAIN; | |
3470 | goto out; | |
3471 | } | |
3472 | if (pending_del_nr) { | |
3473 | ret = btrfs_del_items(trans, root, path, | |
3474 | pending_del_slot, | |
3475 | pending_del_nr); | |
3476 | BUG_ON(ret); | |
3477 | pending_del_nr = 0; | |
3478 | } | |
85e21bac CM |
3479 | btrfs_release_path(root, path); |
3480 | goto search_again; | |
8082510e YZ |
3481 | } else { |
3482 | path->slots[0]--; | |
85e21bac | 3483 | } |
39279cc3 | 3484 | } |
8082510e | 3485 | out: |
85e21bac CM |
3486 | if (pending_del_nr) { |
3487 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3488 | pending_del_nr); | |
d68fc57b | 3489 | BUG_ON(ret); |
85e21bac | 3490 | } |
39279cc3 | 3491 | btrfs_free_path(path); |
8082510e | 3492 | return err; |
39279cc3 CM |
3493 | } |
3494 | ||
3495 | /* | |
3496 | * taken from block_truncate_page, but does cow as it zeros out | |
3497 | * any bytes left in the last page in the file. | |
3498 | */ | |
3499 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3500 | { | |
3501 | struct inode *inode = mapping->host; | |
db94535d | 3502 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3503 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3504 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3505 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3506 | char *kaddr; |
db94535d | 3507 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3508 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3509 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3510 | struct page *page; | |
39279cc3 | 3511 | int ret = 0; |
a52d9a80 | 3512 | u64 page_start; |
e6dcd2dc | 3513 | u64 page_end; |
39279cc3 CM |
3514 | |
3515 | if ((offset & (blocksize - 1)) == 0) | |
3516 | goto out; | |
0ca1f7ce | 3517 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3518 | if (ret) |
3519 | goto out; | |
39279cc3 CM |
3520 | |
3521 | ret = -ENOMEM; | |
211c17f5 | 3522 | again: |
39279cc3 | 3523 | page = grab_cache_page(mapping, index); |
5d5e103a | 3524 | if (!page) { |
0ca1f7ce | 3525 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3526 | goto out; |
5d5e103a | 3527 | } |
e6dcd2dc CM |
3528 | |
3529 | page_start = page_offset(page); | |
3530 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3531 | ||
39279cc3 | 3532 | if (!PageUptodate(page)) { |
9ebefb18 | 3533 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3534 | lock_page(page); |
211c17f5 CM |
3535 | if (page->mapping != mapping) { |
3536 | unlock_page(page); | |
3537 | page_cache_release(page); | |
3538 | goto again; | |
3539 | } | |
39279cc3 CM |
3540 | if (!PageUptodate(page)) { |
3541 | ret = -EIO; | |
89642229 | 3542 | goto out_unlock; |
39279cc3 CM |
3543 | } |
3544 | } | |
211c17f5 | 3545 | wait_on_page_writeback(page); |
e6dcd2dc | 3546 | |
2ac55d41 JB |
3547 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3548 | GFP_NOFS); | |
e6dcd2dc CM |
3549 | set_page_extent_mapped(page); |
3550 | ||
3551 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3552 | if (ordered) { | |
2ac55d41 JB |
3553 | unlock_extent_cached(io_tree, page_start, page_end, |
3554 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3555 | unlock_page(page); |
3556 | page_cache_release(page); | |
eb84ae03 | 3557 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3558 | btrfs_put_ordered_extent(ordered); |
3559 | goto again; | |
3560 | } | |
3561 | ||
2ac55d41 | 3562 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3563 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3564 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3565 | |
2ac55d41 JB |
3566 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3567 | &cached_state); | |
9ed74f2d | 3568 | if (ret) { |
2ac55d41 JB |
3569 | unlock_extent_cached(io_tree, page_start, page_end, |
3570 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3571 | goto out_unlock; |
3572 | } | |
3573 | ||
e6dcd2dc CM |
3574 | ret = 0; |
3575 | if (offset != PAGE_CACHE_SIZE) { | |
3576 | kaddr = kmap(page); | |
3577 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3578 | flush_dcache_page(page); | |
3579 | kunmap(page); | |
3580 | } | |
247e743c | 3581 | ClearPageChecked(page); |
e6dcd2dc | 3582 | set_page_dirty(page); |
2ac55d41 JB |
3583 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3584 | GFP_NOFS); | |
39279cc3 | 3585 | |
89642229 | 3586 | out_unlock: |
5d5e103a | 3587 | if (ret) |
0ca1f7ce | 3588 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3589 | unlock_page(page); |
3590 | page_cache_release(page); | |
3591 | out: | |
3592 | return ret; | |
3593 | } | |
3594 | ||
695a0d0d JB |
3595 | /* |
3596 | * This function puts in dummy file extents for the area we're creating a hole | |
3597 | * for. So if we are truncating this file to a larger size we need to insert | |
3598 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3599 | * the range between oldsize and size | |
3600 | */ | |
a41ad394 | 3601 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3602 | { |
9036c102 YZ |
3603 | struct btrfs_trans_handle *trans; |
3604 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3605 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3606 | struct extent_map *em = NULL; |
2ac55d41 | 3607 | struct extent_state *cached_state = NULL; |
9036c102 | 3608 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3609 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3610 | u64 block_end = (size + mask) & ~mask; |
3611 | u64 last_byte; | |
3612 | u64 cur_offset; | |
3613 | u64 hole_size; | |
9ed74f2d | 3614 | int err = 0; |
39279cc3 | 3615 | |
9036c102 YZ |
3616 | if (size <= hole_start) |
3617 | return 0; | |
3618 | ||
9036c102 YZ |
3619 | while (1) { |
3620 | struct btrfs_ordered_extent *ordered; | |
3621 | btrfs_wait_ordered_range(inode, hole_start, | |
3622 | block_end - hole_start); | |
2ac55d41 JB |
3623 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3624 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3625 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3626 | if (!ordered) | |
3627 | break; | |
2ac55d41 JB |
3628 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3629 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3630 | btrfs_put_ordered_extent(ordered); |
3631 | } | |
39279cc3 | 3632 | |
9036c102 YZ |
3633 | cur_offset = hole_start; |
3634 | while (1) { | |
3635 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3636 | block_end - cur_offset, 0); | |
3637 | BUG_ON(IS_ERR(em) || !em); | |
3638 | last_byte = min(extent_map_end(em), block_end); | |
3639 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3640 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3641 | u64 hint_byte = 0; |
9036c102 | 3642 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3643 | |
a22285a6 YZ |
3644 | trans = btrfs_start_transaction(root, 2); |
3645 | if (IS_ERR(trans)) { | |
3646 | err = PTR_ERR(trans); | |
9ed74f2d | 3647 | break; |
a22285a6 | 3648 | } |
8082510e YZ |
3649 | btrfs_set_trans_block_group(trans, inode); |
3650 | ||
3651 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3652 | cur_offset + hole_size, | |
3653 | &hint_byte, 1); | |
3893e33b JB |
3654 | if (err) |
3655 | break; | |
8082510e | 3656 | |
9036c102 YZ |
3657 | err = btrfs_insert_file_extent(trans, root, |
3658 | inode->i_ino, cur_offset, 0, | |
3659 | 0, hole_size, 0, hole_size, | |
3660 | 0, 0, 0); | |
3893e33b JB |
3661 | if (err) |
3662 | break; | |
8082510e | 3663 | |
9036c102 YZ |
3664 | btrfs_drop_extent_cache(inode, hole_start, |
3665 | last_byte - 1, 0); | |
8082510e YZ |
3666 | |
3667 | btrfs_end_transaction(trans, root); | |
9036c102 YZ |
3668 | } |
3669 | free_extent_map(em); | |
a22285a6 | 3670 | em = NULL; |
9036c102 | 3671 | cur_offset = last_byte; |
8082510e | 3672 | if (cur_offset >= block_end) |
9036c102 YZ |
3673 | break; |
3674 | } | |
1832a6d5 | 3675 | |
a22285a6 | 3676 | free_extent_map(em); |
2ac55d41 JB |
3677 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3678 | GFP_NOFS); | |
9036c102 YZ |
3679 | return err; |
3680 | } | |
39279cc3 | 3681 | |
a41ad394 | 3682 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3683 | { |
a41ad394 | 3684 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3685 | int ret; |
3686 | ||
a41ad394 | 3687 | if (newsize == oldsize) |
8082510e YZ |
3688 | return 0; |
3689 | ||
a41ad394 JB |
3690 | if (newsize > oldsize) { |
3691 | i_size_write(inode, newsize); | |
3692 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3693 | truncate_pagecache(inode, oldsize, newsize); | |
3694 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
8082510e | 3695 | if (ret) { |
a41ad394 | 3696 | btrfs_setsize(inode, oldsize); |
8082510e YZ |
3697 | return ret; |
3698 | } | |
3699 | ||
930f028a | 3700 | mark_inode_dirty(inode); |
a41ad394 | 3701 | } else { |
8082510e | 3702 | |
a41ad394 JB |
3703 | /* |
3704 | * We're truncating a file that used to have good data down to | |
3705 | * zero. Make sure it gets into the ordered flush list so that | |
3706 | * any new writes get down to disk quickly. | |
3707 | */ | |
3708 | if (newsize == 0) | |
3709 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3710 | |
a41ad394 JB |
3711 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3712 | truncate_setsize(inode, newsize); | |
3713 | ret = btrfs_truncate(inode); | |
3714 | } | |
8082510e | 3715 | |
a41ad394 | 3716 | return ret; |
8082510e YZ |
3717 | } |
3718 | ||
9036c102 YZ |
3719 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3720 | { | |
3721 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3722 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3723 | int err; |
39279cc3 | 3724 | |
b83cc969 LZ |
3725 | if (btrfs_root_readonly(root)) |
3726 | return -EROFS; | |
3727 | ||
9036c102 YZ |
3728 | err = inode_change_ok(inode, attr); |
3729 | if (err) | |
3730 | return err; | |
2bf5a725 | 3731 | |
5a3f23d5 | 3732 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3733 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3734 | if (err) |
3735 | return err; | |
39279cc3 | 3736 | } |
9036c102 | 3737 | |
1025774c CH |
3738 | if (attr->ia_valid) { |
3739 | setattr_copy(inode, attr); | |
3740 | mark_inode_dirty(inode); | |
3741 | ||
3742 | if (attr->ia_valid & ATTR_MODE) | |
3743 | err = btrfs_acl_chmod(inode); | |
3744 | } | |
33268eaf | 3745 | |
39279cc3 CM |
3746 | return err; |
3747 | } | |
61295eb8 | 3748 | |
bd555975 | 3749 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3750 | { |
3751 | struct btrfs_trans_handle *trans; | |
3752 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d3c2fdcf | 3753 | unsigned long nr; |
39279cc3 CM |
3754 | int ret; |
3755 | ||
1abe9b8a | 3756 | trace_btrfs_inode_evict(inode); |
3757 | ||
39279cc3 | 3758 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b JB |
3759 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
3760 | root == root->fs_info->tree_root)) | |
bd555975 AV |
3761 | goto no_delete; |
3762 | ||
39279cc3 | 3763 | if (is_bad_inode(inode)) { |
7b128766 | 3764 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3765 | goto no_delete; |
3766 | } | |
bd555975 | 3767 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3768 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3769 | |
c71bf099 YZ |
3770 | if (root->fs_info->log_root_recovering) { |
3771 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3772 | goto no_delete; | |
3773 | } | |
3774 | ||
76dda93c YZ |
3775 | if (inode->i_nlink > 0) { |
3776 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3777 | goto no_delete; | |
3778 | } | |
3779 | ||
dbe674a9 | 3780 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3781 | |
8082510e | 3782 | while (1) { |
d68fc57b YZ |
3783 | trans = btrfs_start_transaction(root, 0); |
3784 | BUG_ON(IS_ERR(trans)); | |
8082510e | 3785 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b YZ |
3786 | trans->block_rsv = root->orphan_block_rsv; |
3787 | ||
3788 | ret = btrfs_block_rsv_check(trans, root, | |
3789 | root->orphan_block_rsv, 0, 5); | |
3790 | if (ret) { | |
3791 | BUG_ON(ret != -EAGAIN); | |
3792 | ret = btrfs_commit_transaction(trans, root); | |
3793 | BUG_ON(ret); | |
3794 | continue; | |
3795 | } | |
7b128766 | 3796 | |
d68fc57b | 3797 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3798 | if (ret != -EAGAIN) |
3799 | break; | |
85e21bac | 3800 | |
8082510e YZ |
3801 | nr = trans->blocks_used; |
3802 | btrfs_end_transaction(trans, root); | |
3803 | trans = NULL; | |
3804 | btrfs_btree_balance_dirty(root, nr); | |
d68fc57b | 3805 | |
8082510e | 3806 | } |
5f39d397 | 3807 | |
8082510e YZ |
3808 | if (ret == 0) { |
3809 | ret = btrfs_orphan_del(trans, inode); | |
3810 | BUG_ON(ret); | |
3811 | } | |
54aa1f4d | 3812 | |
d3c2fdcf | 3813 | nr = trans->blocks_used; |
54aa1f4d | 3814 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3815 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3816 | no_delete: |
bd555975 | 3817 | end_writeback(inode); |
8082510e | 3818 | return; |
39279cc3 CM |
3819 | } |
3820 | ||
3821 | /* | |
3822 | * this returns the key found in the dir entry in the location pointer. | |
3823 | * If no dir entries were found, location->objectid is 0. | |
3824 | */ | |
3825 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3826 | struct btrfs_key *location) | |
3827 | { | |
3828 | const char *name = dentry->d_name.name; | |
3829 | int namelen = dentry->d_name.len; | |
3830 | struct btrfs_dir_item *di; | |
3831 | struct btrfs_path *path; | |
3832 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3833 | int ret = 0; |
39279cc3 CM |
3834 | |
3835 | path = btrfs_alloc_path(); | |
3836 | BUG_ON(!path); | |
3954401f | 3837 | |
39279cc3 CM |
3838 | di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name, |
3839 | namelen, 0); | |
0d9f7f3e Y |
3840 | if (IS_ERR(di)) |
3841 | ret = PTR_ERR(di); | |
d397712b CM |
3842 | |
3843 | if (!di || IS_ERR(di)) | |
3954401f | 3844 | goto out_err; |
d397712b | 3845 | |
5f39d397 | 3846 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3847 | out: |
39279cc3 CM |
3848 | btrfs_free_path(path); |
3849 | return ret; | |
3954401f CM |
3850 | out_err: |
3851 | location->objectid = 0; | |
3852 | goto out; | |
39279cc3 CM |
3853 | } |
3854 | ||
3855 | /* | |
3856 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3857 | * needs to be changed to reflect the root directory of the tree root. This | |
3858 | * is kind of like crossing a mount point. | |
3859 | */ | |
3860 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3861 | struct inode *dir, |
3862 | struct dentry *dentry, | |
3863 | struct btrfs_key *location, | |
3864 | struct btrfs_root **sub_root) | |
39279cc3 | 3865 | { |
4df27c4d YZ |
3866 | struct btrfs_path *path; |
3867 | struct btrfs_root *new_root; | |
3868 | struct btrfs_root_ref *ref; | |
3869 | struct extent_buffer *leaf; | |
3870 | int ret; | |
3871 | int err = 0; | |
39279cc3 | 3872 | |
4df27c4d YZ |
3873 | path = btrfs_alloc_path(); |
3874 | if (!path) { | |
3875 | err = -ENOMEM; | |
3876 | goto out; | |
3877 | } | |
39279cc3 | 3878 | |
4df27c4d YZ |
3879 | err = -ENOENT; |
3880 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3881 | BTRFS_I(dir)->root->root_key.objectid, | |
3882 | location->objectid); | |
3883 | if (ret) { | |
3884 | if (ret < 0) | |
3885 | err = ret; | |
3886 | goto out; | |
3887 | } | |
39279cc3 | 3888 | |
4df27c4d YZ |
3889 | leaf = path->nodes[0]; |
3890 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
3891 | if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino || | |
3892 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) | |
3893 | goto out; | |
39279cc3 | 3894 | |
4df27c4d YZ |
3895 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3896 | (unsigned long)(ref + 1), | |
3897 | dentry->d_name.len); | |
3898 | if (ret) | |
3899 | goto out; | |
3900 | ||
3901 | btrfs_release_path(root->fs_info->tree_root, path); | |
3902 | ||
3903 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3904 | if (IS_ERR(new_root)) { | |
3905 | err = PTR_ERR(new_root); | |
3906 | goto out; | |
3907 | } | |
3908 | ||
3909 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3910 | err = -ENOENT; | |
3911 | goto out; | |
3912 | } | |
3913 | ||
3914 | *sub_root = new_root; | |
3915 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3916 | location->type = BTRFS_INODE_ITEM_KEY; | |
3917 | location->offset = 0; | |
3918 | err = 0; | |
3919 | out: | |
3920 | btrfs_free_path(path); | |
3921 | return err; | |
39279cc3 CM |
3922 | } |
3923 | ||
5d4f98a2 YZ |
3924 | static void inode_tree_add(struct inode *inode) |
3925 | { | |
3926 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3927 | struct btrfs_inode *entry; | |
03e860bd FNP |
3928 | struct rb_node **p; |
3929 | struct rb_node *parent; | |
03e860bd FNP |
3930 | again: |
3931 | p = &root->inode_tree.rb_node; | |
3932 | parent = NULL; | |
5d4f98a2 | 3933 | |
1d3382cb | 3934 | if (inode_unhashed(inode)) |
76dda93c YZ |
3935 | return; |
3936 | ||
5d4f98a2 YZ |
3937 | spin_lock(&root->inode_lock); |
3938 | while (*p) { | |
3939 | parent = *p; | |
3940 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3941 | ||
3942 | if (inode->i_ino < entry->vfs_inode.i_ino) | |
03e860bd | 3943 | p = &parent->rb_left; |
5d4f98a2 | 3944 | else if (inode->i_ino > entry->vfs_inode.i_ino) |
03e860bd | 3945 | p = &parent->rb_right; |
5d4f98a2 YZ |
3946 | else { |
3947 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3948 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3949 | rb_erase(parent, &root->inode_tree); |
3950 | RB_CLEAR_NODE(parent); | |
3951 | spin_unlock(&root->inode_lock); | |
3952 | goto again; | |
5d4f98a2 YZ |
3953 | } |
3954 | } | |
3955 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3956 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3957 | spin_unlock(&root->inode_lock); | |
3958 | } | |
3959 | ||
3960 | static void inode_tree_del(struct inode *inode) | |
3961 | { | |
3962 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3963 | int empty = 0; |
5d4f98a2 | 3964 | |
03e860bd | 3965 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3966 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3967 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3968 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3969 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3970 | } |
03e860bd | 3971 | spin_unlock(&root->inode_lock); |
76dda93c | 3972 | |
0af3d00b JB |
3973 | /* |
3974 | * Free space cache has inodes in the tree root, but the tree root has a | |
3975 | * root_refs of 0, so this could end up dropping the tree root as a | |
3976 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3977 | * make sure we don't drop it. | |
3978 | */ | |
3979 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3980 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3981 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3982 | spin_lock(&root->inode_lock); | |
3983 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3984 | spin_unlock(&root->inode_lock); | |
3985 | if (empty) | |
3986 | btrfs_add_dead_root(root); | |
3987 | } | |
3988 | } | |
3989 | ||
3990 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3991 | { | |
3992 | struct rb_node *node; | |
3993 | struct rb_node *prev; | |
3994 | struct btrfs_inode *entry; | |
3995 | struct inode *inode; | |
3996 | u64 objectid = 0; | |
3997 | ||
3998 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3999 | ||
4000 | spin_lock(&root->inode_lock); | |
4001 | again: | |
4002 | node = root->inode_tree.rb_node; | |
4003 | prev = NULL; | |
4004 | while (node) { | |
4005 | prev = node; | |
4006 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4007 | ||
4008 | if (objectid < entry->vfs_inode.i_ino) | |
4009 | node = node->rb_left; | |
4010 | else if (objectid > entry->vfs_inode.i_ino) | |
4011 | node = node->rb_right; | |
4012 | else | |
4013 | break; | |
4014 | } | |
4015 | if (!node) { | |
4016 | while (prev) { | |
4017 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4018 | if (objectid <= entry->vfs_inode.i_ino) { | |
4019 | node = prev; | |
4020 | break; | |
4021 | } | |
4022 | prev = rb_next(prev); | |
4023 | } | |
4024 | } | |
4025 | while (node) { | |
4026 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4027 | objectid = entry->vfs_inode.i_ino + 1; | |
4028 | inode = igrab(&entry->vfs_inode); | |
4029 | if (inode) { | |
4030 | spin_unlock(&root->inode_lock); | |
4031 | if (atomic_read(&inode->i_count) > 1) | |
4032 | d_prune_aliases(inode); | |
4033 | /* | |
45321ac5 | 4034 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4035 | * the inode cache when its usage count |
4036 | * hits zero. | |
4037 | */ | |
4038 | iput(inode); | |
4039 | cond_resched(); | |
4040 | spin_lock(&root->inode_lock); | |
4041 | goto again; | |
4042 | } | |
4043 | ||
4044 | if (cond_resched_lock(&root->inode_lock)) | |
4045 | goto again; | |
4046 | ||
4047 | node = rb_next(node); | |
4048 | } | |
4049 | spin_unlock(&root->inode_lock); | |
4050 | return 0; | |
5d4f98a2 YZ |
4051 | } |
4052 | ||
e02119d5 CM |
4053 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4054 | { | |
4055 | struct btrfs_iget_args *args = p; | |
4056 | inode->i_ino = args->ino; | |
e02119d5 | 4057 | BTRFS_I(inode)->root = args->root; |
6a63209f | 4058 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
4059 | return 0; |
4060 | } | |
4061 | ||
4062 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4063 | { | |
4064 | struct btrfs_iget_args *args = opaque; | |
d397712b CM |
4065 | return args->ino == inode->i_ino && |
4066 | args->root == BTRFS_I(inode)->root; | |
39279cc3 CM |
4067 | } |
4068 | ||
5d4f98a2 YZ |
4069 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4070 | u64 objectid, | |
4071 | struct btrfs_root *root) | |
39279cc3 CM |
4072 | { |
4073 | struct inode *inode; | |
4074 | struct btrfs_iget_args args; | |
4075 | args.ino = objectid; | |
4076 | args.root = root; | |
4077 | ||
4078 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4079 | btrfs_init_locked_inode, | |
4080 | (void *)&args); | |
4081 | return inode; | |
4082 | } | |
4083 | ||
1a54ef8c BR |
4084 | /* Get an inode object given its location and corresponding root. |
4085 | * Returns in *is_new if the inode was read from disk | |
4086 | */ | |
4087 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4088 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4089 | { |
4090 | struct inode *inode; | |
4091 | ||
4092 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4093 | if (!inode) | |
5d4f98a2 | 4094 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4095 | |
4096 | if (inode->i_state & I_NEW) { | |
4097 | BTRFS_I(inode)->root = root; | |
4098 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4099 | btrfs_read_locked_inode(inode); | |
5d4f98a2 | 4100 | inode_tree_add(inode); |
1a54ef8c | 4101 | unlock_new_inode(inode); |
73f73415 JB |
4102 | if (new) |
4103 | *new = 1; | |
1a54ef8c BR |
4104 | } |
4105 | ||
4106 | return inode; | |
4107 | } | |
4108 | ||
4df27c4d YZ |
4109 | static struct inode *new_simple_dir(struct super_block *s, |
4110 | struct btrfs_key *key, | |
4111 | struct btrfs_root *root) | |
4112 | { | |
4113 | struct inode *inode = new_inode(s); | |
4114 | ||
4115 | if (!inode) | |
4116 | return ERR_PTR(-ENOMEM); | |
4117 | ||
4df27c4d YZ |
4118 | BTRFS_I(inode)->root = root; |
4119 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
4120 | BTRFS_I(inode)->dummy_inode = 1; | |
4121 | ||
4122 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
4123 | inode->i_op = &simple_dir_inode_operations; | |
4124 | inode->i_fop = &simple_dir_operations; | |
4125 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4126 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4127 | ||
4128 | return inode; | |
4129 | } | |
4130 | ||
3de4586c | 4131 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4132 | { |
d397712b | 4133 | struct inode *inode; |
4df27c4d | 4134 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4135 | struct btrfs_root *sub_root = root; |
4136 | struct btrfs_key location; | |
76dda93c | 4137 | int index; |
5d4f98a2 | 4138 | int ret; |
39279cc3 CM |
4139 | |
4140 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4141 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4142 | |
39279cc3 | 4143 | ret = btrfs_inode_by_name(dir, dentry, &location); |
5f39d397 | 4144 | |
39279cc3 CM |
4145 | if (ret < 0) |
4146 | return ERR_PTR(ret); | |
5f39d397 | 4147 | |
4df27c4d YZ |
4148 | if (location.objectid == 0) |
4149 | return NULL; | |
4150 | ||
4151 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4152 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4153 | return inode; |
4154 | } | |
4155 | ||
4156 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4157 | ||
76dda93c | 4158 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4159 | ret = fixup_tree_root_location(root, dir, dentry, |
4160 | &location, &sub_root); | |
4161 | if (ret < 0) { | |
4162 | if (ret != -ENOENT) | |
4163 | inode = ERR_PTR(ret); | |
4164 | else | |
4165 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4166 | } else { | |
73f73415 | 4167 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4168 | } |
76dda93c YZ |
4169 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4170 | ||
34d19bad | 4171 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4172 | down_read(&root->fs_info->cleanup_work_sem); |
4173 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4174 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4175 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4176 | if (ret) |
4177 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4178 | } |
4179 | ||
3de4586c CM |
4180 | return inode; |
4181 | } | |
4182 | ||
fe15ce44 | 4183 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4184 | { |
4185 | struct btrfs_root *root; | |
4186 | ||
efefb143 YZ |
4187 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
4188 | dentry = dentry->d_parent; | |
76dda93c | 4189 | |
efefb143 YZ |
4190 | if (dentry->d_inode) { |
4191 | root = BTRFS_I(dentry->d_inode)->root; | |
4192 | if (btrfs_root_refs(&root->root_item) == 0) | |
4193 | return 1; | |
4194 | } | |
76dda93c YZ |
4195 | return 0; |
4196 | } | |
4197 | ||
3de4586c CM |
4198 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4199 | struct nameidata *nd) | |
4200 | { | |
4201 | struct inode *inode; | |
4202 | ||
3de4586c CM |
4203 | inode = btrfs_lookup_dentry(dir, dentry); |
4204 | if (IS_ERR(inode)) | |
4205 | return ERR_CAST(inode); | |
7b128766 | 4206 | |
39279cc3 CM |
4207 | return d_splice_alias(inode, dentry); |
4208 | } | |
4209 | ||
39279cc3 CM |
4210 | static unsigned char btrfs_filetype_table[] = { |
4211 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK | |
4212 | }; | |
4213 | ||
cbdf5a24 DW |
4214 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4215 | filldir_t filldir) | |
39279cc3 | 4216 | { |
6da6abae | 4217 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4218 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4219 | struct btrfs_item *item; | |
4220 | struct btrfs_dir_item *di; | |
4221 | struct btrfs_key key; | |
5f39d397 | 4222 | struct btrfs_key found_key; |
39279cc3 CM |
4223 | struct btrfs_path *path; |
4224 | int ret; | |
5f39d397 | 4225 | struct extent_buffer *leaf; |
39279cc3 | 4226 | int slot; |
39279cc3 CM |
4227 | unsigned char d_type; |
4228 | int over = 0; | |
4229 | u32 di_cur; | |
4230 | u32 di_total; | |
4231 | u32 di_len; | |
4232 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4233 | char tmp_name[32]; |
4234 | char *name_ptr; | |
4235 | int name_len; | |
39279cc3 CM |
4236 | |
4237 | /* FIXME, use a real flag for deciding about the key type */ | |
4238 | if (root->fs_info->tree_root == root) | |
4239 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4240 | |
3954401f CM |
4241 | /* special case for "." */ |
4242 | if (filp->f_pos == 0) { | |
4243 | over = filldir(dirent, ".", 1, | |
4244 | 1, inode->i_ino, | |
4245 | DT_DIR); | |
4246 | if (over) | |
4247 | return 0; | |
4248 | filp->f_pos = 1; | |
4249 | } | |
3954401f CM |
4250 | /* special case for .., just use the back ref */ |
4251 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4252 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4253 | over = filldir(dirent, "..", 2, |
5ecc7e5d | 4254 | 2, pino, DT_DIR); |
3954401f | 4255 | if (over) |
49593bfa | 4256 | return 0; |
3954401f CM |
4257 | filp->f_pos = 2; |
4258 | } | |
49593bfa DW |
4259 | path = btrfs_alloc_path(); |
4260 | path->reada = 2; | |
4261 | ||
39279cc3 CM |
4262 | btrfs_set_key_type(&key, key_type); |
4263 | key.offset = filp->f_pos; | |
49593bfa | 4264 | key.objectid = inode->i_ino; |
5f39d397 | 4265 | |
39279cc3 CM |
4266 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4267 | if (ret < 0) | |
4268 | goto err; | |
49593bfa DW |
4269 | |
4270 | while (1) { | |
5f39d397 | 4271 | leaf = path->nodes[0]; |
39279cc3 | 4272 | slot = path->slots[0]; |
b9e03af0 LZ |
4273 | if (slot >= btrfs_header_nritems(leaf)) { |
4274 | ret = btrfs_next_leaf(root, path); | |
4275 | if (ret < 0) | |
4276 | goto err; | |
4277 | else if (ret > 0) | |
4278 | break; | |
4279 | continue; | |
39279cc3 | 4280 | } |
3de4586c | 4281 | |
5f39d397 CM |
4282 | item = btrfs_item_nr(leaf, slot); |
4283 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4284 | ||
4285 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4286 | break; |
5f39d397 | 4287 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4288 | break; |
5f39d397 | 4289 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4290 | goto next; |
5f39d397 CM |
4291 | |
4292 | filp->f_pos = found_key.offset; | |
49593bfa | 4293 | |
39279cc3 CM |
4294 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4295 | di_cur = 0; | |
5f39d397 | 4296 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4297 | |
4298 | while (di_cur < di_total) { | |
5f39d397 CM |
4299 | struct btrfs_key location; |
4300 | ||
22a94d44 JB |
4301 | if (verify_dir_item(root, leaf, di)) |
4302 | break; | |
4303 | ||
5f39d397 | 4304 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4305 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4306 | name_ptr = tmp_name; |
4307 | } else { | |
4308 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4309 | if (!name_ptr) { |
4310 | ret = -ENOMEM; | |
4311 | goto err; | |
4312 | } | |
5f39d397 CM |
4313 | } |
4314 | read_extent_buffer(leaf, name_ptr, | |
4315 | (unsigned long)(di + 1), name_len); | |
4316 | ||
4317 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4318 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c CM |
4319 | |
4320 | /* is this a reference to our own snapshot? If so | |
4321 | * skip it | |
4322 | */ | |
4323 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4324 | location.objectid == root->root_key.objectid) { | |
4325 | over = 0; | |
4326 | goto skip; | |
4327 | } | |
5f39d397 | 4328 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4329 | found_key.offset, location.objectid, |
39279cc3 | 4330 | d_type); |
5f39d397 | 4331 | |
3de4586c | 4332 | skip: |
5f39d397 CM |
4333 | if (name_ptr != tmp_name) |
4334 | kfree(name_ptr); | |
4335 | ||
39279cc3 CM |
4336 | if (over) |
4337 | goto nopos; | |
5103e947 | 4338 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4339 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4340 | di_cur += di_len; |
4341 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4342 | } | |
b9e03af0 LZ |
4343 | next: |
4344 | path->slots[0]++; | |
39279cc3 | 4345 | } |
49593bfa DW |
4346 | |
4347 | /* Reached end of directory/root. Bump pos past the last item. */ | |
5e591a07 | 4348 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4349 | /* |
4350 | * 32-bit glibc will use getdents64, but then strtol - | |
4351 | * so the last number we can serve is this. | |
4352 | */ | |
4353 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4354 | else |
4355 | filp->f_pos++; | |
39279cc3 CM |
4356 | nopos: |
4357 | ret = 0; | |
4358 | err: | |
39279cc3 | 4359 | btrfs_free_path(path); |
39279cc3 CM |
4360 | return ret; |
4361 | } | |
4362 | ||
a9185b41 | 4363 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4364 | { |
4365 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4366 | struct btrfs_trans_handle *trans; | |
4367 | int ret = 0; | |
0af3d00b | 4368 | bool nolock = false; |
39279cc3 | 4369 | |
8929ecfa | 4370 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4371 | return 0; |
4372 | ||
0af3d00b JB |
4373 | smp_mb(); |
4374 | nolock = (root->fs_info->closing && root == root->fs_info->tree_root); | |
4375 | ||
a9185b41 | 4376 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b JB |
4377 | if (nolock) |
4378 | trans = btrfs_join_transaction_nolock(root, 1); | |
4379 | else | |
4380 | trans = btrfs_join_transaction(root, 1); | |
3612b495 TI |
4381 | if (IS_ERR(trans)) |
4382 | return PTR_ERR(trans); | |
39279cc3 | 4383 | btrfs_set_trans_block_group(trans, inode); |
0af3d00b JB |
4384 | if (nolock) |
4385 | ret = btrfs_end_transaction_nolock(trans, root); | |
4386 | else | |
4387 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4388 | } |
4389 | return ret; | |
4390 | } | |
4391 | ||
4392 | /* | |
54aa1f4d | 4393 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4394 | * inode changes. But, it is most likely to find the inode in cache. |
4395 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4396 | * to keep or drop this code. | |
4397 | */ | |
4398 | void btrfs_dirty_inode(struct inode *inode) | |
4399 | { | |
4400 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4401 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4402 | int ret; |
4403 | ||
4404 | if (BTRFS_I(inode)->dummy_inode) | |
4405 | return; | |
39279cc3 | 4406 | |
f9295749 | 4407 | trans = btrfs_join_transaction(root, 1); |
3612b495 | 4408 | BUG_ON(IS_ERR(trans)); |
39279cc3 | 4409 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa YZ |
4410 | |
4411 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4412 | if (ret && ret == -ENOSPC) { |
4413 | /* whoops, lets try again with the full transaction */ | |
4414 | btrfs_end_transaction(trans, root); | |
4415 | trans = btrfs_start_transaction(root, 1); | |
9aeead73 CM |
4416 | if (IS_ERR(trans)) { |
4417 | if (printk_ratelimit()) { | |
4418 | printk(KERN_ERR "btrfs: fail to " | |
4419 | "dirty inode %lu error %ld\n", | |
4420 | inode->i_ino, PTR_ERR(trans)); | |
4421 | } | |
4422 | return; | |
4423 | } | |
94b60442 | 4424 | btrfs_set_trans_block_group(trans, inode); |
8929ecfa | 4425 | |
94b60442 CM |
4426 | ret = btrfs_update_inode(trans, root, inode); |
4427 | if (ret) { | |
9aeead73 CM |
4428 | if (printk_ratelimit()) { |
4429 | printk(KERN_ERR "btrfs: fail to " | |
4430 | "dirty inode %lu error %d\n", | |
4431 | inode->i_ino, ret); | |
4432 | } | |
94b60442 CM |
4433 | } |
4434 | } | |
39279cc3 | 4435 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4436 | } |
4437 | ||
d352ac68 CM |
4438 | /* |
4439 | * find the highest existing sequence number in a directory | |
4440 | * and then set the in-memory index_cnt variable to reflect | |
4441 | * free sequence numbers | |
4442 | */ | |
aec7477b JB |
4443 | static int btrfs_set_inode_index_count(struct inode *inode) |
4444 | { | |
4445 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4446 | struct btrfs_key key, found_key; | |
4447 | struct btrfs_path *path; | |
4448 | struct extent_buffer *leaf; | |
4449 | int ret; | |
4450 | ||
4451 | key.objectid = inode->i_ino; | |
4452 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); | |
4453 | key.offset = (u64)-1; | |
4454 | ||
4455 | path = btrfs_alloc_path(); | |
4456 | if (!path) | |
4457 | return -ENOMEM; | |
4458 | ||
4459 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4460 | if (ret < 0) | |
4461 | goto out; | |
4462 | /* FIXME: we should be able to handle this */ | |
4463 | if (ret == 0) | |
4464 | goto out; | |
4465 | ret = 0; | |
4466 | ||
4467 | /* | |
4468 | * MAGIC NUMBER EXPLANATION: | |
4469 | * since we search a directory based on f_pos we have to start at 2 | |
4470 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4471 | * else has to start at 2 | |
4472 | */ | |
4473 | if (path->slots[0] == 0) { | |
4474 | BTRFS_I(inode)->index_cnt = 2; | |
4475 | goto out; | |
4476 | } | |
4477 | ||
4478 | path->slots[0]--; | |
4479 | ||
4480 | leaf = path->nodes[0]; | |
4481 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4482 | ||
4483 | if (found_key.objectid != inode->i_ino || | |
4484 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { | |
4485 | BTRFS_I(inode)->index_cnt = 2; | |
4486 | goto out; | |
4487 | } | |
4488 | ||
4489 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4490 | out: | |
4491 | btrfs_free_path(path); | |
4492 | return ret; | |
4493 | } | |
4494 | ||
d352ac68 CM |
4495 | /* |
4496 | * helper to find a free sequence number in a given directory. This current | |
4497 | * code is very simple, later versions will do smarter things in the btree | |
4498 | */ | |
3de4586c | 4499 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4500 | { |
4501 | int ret = 0; | |
4502 | ||
4503 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
4504 | ret = btrfs_set_inode_index_count(dir); | |
d397712b | 4505 | if (ret) |
aec7477b JB |
4506 | return ret; |
4507 | } | |
4508 | ||
00e4e6b3 | 4509 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4510 | BTRFS_I(dir)->index_cnt++; |
4511 | ||
4512 | return ret; | |
4513 | } | |
4514 | ||
39279cc3 CM |
4515 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4516 | struct btrfs_root *root, | |
aec7477b | 4517 | struct inode *dir, |
9c58309d | 4518 | const char *name, int name_len, |
d2fb3437 YZ |
4519 | u64 ref_objectid, u64 objectid, |
4520 | u64 alloc_hint, int mode, u64 *index) | |
39279cc3 CM |
4521 | { |
4522 | struct inode *inode; | |
5f39d397 | 4523 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4524 | struct btrfs_key *location; |
5f39d397 | 4525 | struct btrfs_path *path; |
9c58309d CM |
4526 | struct btrfs_inode_ref *ref; |
4527 | struct btrfs_key key[2]; | |
4528 | u32 sizes[2]; | |
4529 | unsigned long ptr; | |
39279cc3 CM |
4530 | int ret; |
4531 | int owner; | |
4532 | ||
5f39d397 CM |
4533 | path = btrfs_alloc_path(); |
4534 | BUG_ON(!path); | |
4535 | ||
39279cc3 | 4536 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4537 | if (!inode) { |
4538 | btrfs_free_path(path); | |
39279cc3 | 4539 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4540 | } |
39279cc3 | 4541 | |
aec7477b | 4542 | if (dir) { |
1abe9b8a | 4543 | trace_btrfs_inode_request(dir); |
4544 | ||
3de4586c | 4545 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4546 | if (ret) { |
8fb27640 | 4547 | btrfs_free_path(path); |
09771430 | 4548 | iput(inode); |
aec7477b | 4549 | return ERR_PTR(ret); |
09771430 | 4550 | } |
aec7477b JB |
4551 | } |
4552 | /* | |
4553 | * index_cnt is ignored for everything but a dir, | |
4554 | * btrfs_get_inode_index_count has an explanation for the magic | |
4555 | * number | |
4556 | */ | |
4557 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4558 | BTRFS_I(inode)->root = root; |
e02119d5 | 4559 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4560 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4561 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4562 | |
39279cc3 CM |
4563 | if (mode & S_IFDIR) |
4564 | owner = 0; | |
4565 | else | |
4566 | owner = 1; | |
d2fb3437 YZ |
4567 | BTRFS_I(inode)->block_group = |
4568 | btrfs_find_block_group(root, 0, alloc_hint, owner); | |
9c58309d CM |
4569 | |
4570 | key[0].objectid = objectid; | |
4571 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4572 | key[0].offset = 0; | |
4573 | ||
4574 | key[1].objectid = objectid; | |
4575 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4576 | key[1].offset = ref_objectid; | |
4577 | ||
4578 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4579 | sizes[1] = name_len + sizeof(*ref); | |
4580 | ||
b9473439 | 4581 | path->leave_spinning = 1; |
9c58309d CM |
4582 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4583 | if (ret != 0) | |
5f39d397 CM |
4584 | goto fail; |
4585 | ||
ecc11fab | 4586 | inode_init_owner(inode, dir, mode); |
39279cc3 | 4587 | inode->i_ino = objectid; |
a76a3cd4 | 4588 | inode_set_bytes(inode, 0); |
39279cc3 | 4589 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4590 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4591 | struct btrfs_inode_item); | |
e02119d5 | 4592 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4593 | |
4594 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4595 | struct btrfs_inode_ref); | |
4596 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4597 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4598 | ptr = (unsigned long)(ref + 1); |
4599 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4600 | ||
5f39d397 CM |
4601 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4602 | btrfs_free_path(path); | |
4603 | ||
39279cc3 CM |
4604 | location = &BTRFS_I(inode)->location; |
4605 | location->objectid = objectid; | |
39279cc3 CM |
4606 | location->offset = 0; |
4607 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4608 | ||
6cbff00f CH |
4609 | btrfs_inherit_iflags(inode, dir); |
4610 | ||
94272164 CM |
4611 | if ((mode & S_IFREG)) { |
4612 | if (btrfs_test_opt(root, NODATASUM)) | |
4613 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4614 | if (btrfs_test_opt(root, NODATACOW) || |
4615 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4616 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4617 | } | |
4618 | ||
39279cc3 | 4619 | insert_inode_hash(inode); |
5d4f98a2 | 4620 | inode_tree_add(inode); |
1abe9b8a | 4621 | |
4622 | trace_btrfs_inode_new(inode); | |
4623 | ||
39279cc3 | 4624 | return inode; |
5f39d397 | 4625 | fail: |
aec7477b JB |
4626 | if (dir) |
4627 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4628 | btrfs_free_path(path); |
09771430 | 4629 | iput(inode); |
5f39d397 | 4630 | return ERR_PTR(ret); |
39279cc3 CM |
4631 | } |
4632 | ||
4633 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4634 | { | |
4635 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4636 | } | |
4637 | ||
d352ac68 CM |
4638 | /* |
4639 | * utility function to add 'inode' into 'parent_inode' with | |
4640 | * a give name and a given sequence number. | |
4641 | * if 'add_backref' is true, also insert a backref from the | |
4642 | * inode to the parent directory. | |
4643 | */ | |
e02119d5 CM |
4644 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4645 | struct inode *parent_inode, struct inode *inode, | |
4646 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4647 | { |
4df27c4d | 4648 | int ret = 0; |
39279cc3 | 4649 | struct btrfs_key key; |
e02119d5 | 4650 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
5f39d397 | 4651 | |
4df27c4d YZ |
4652 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4653 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); | |
4654 | } else { | |
4655 | key.objectid = inode->i_ino; | |
4656 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); | |
4657 | key.offset = 0; | |
4658 | } | |
4659 | ||
4660 | if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4661 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, | |
4662 | key.objectid, root->root_key.objectid, | |
4663 | parent_inode->i_ino, | |
4664 | index, name, name_len); | |
4665 | } else if (add_backref) { | |
4666 | ret = btrfs_insert_inode_ref(trans, root, | |
4667 | name, name_len, inode->i_ino, | |
4668 | parent_inode->i_ino, index); | |
4669 | } | |
39279cc3 | 4670 | |
39279cc3 | 4671 | if (ret == 0) { |
4df27c4d YZ |
4672 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4673 | parent_inode->i_ino, &key, | |
4674 | btrfs_inode_type(inode), index); | |
4675 | BUG_ON(ret); | |
4676 | ||
dbe674a9 | 4677 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4678 | name_len * 2); |
79c44584 | 4679 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4680 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4681 | } |
4682 | return ret; | |
4683 | } | |
4684 | ||
4685 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4686 | struct inode *dir, struct dentry *dentry, |
4687 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4688 | { |
a1b075d2 JB |
4689 | int err = btrfs_add_link(trans, dir, inode, |
4690 | dentry->d_name.name, dentry->d_name.len, | |
4691 | backref, index); | |
39279cc3 CM |
4692 | if (!err) { |
4693 | d_instantiate(dentry, inode); | |
4694 | return 0; | |
4695 | } | |
4696 | if (err > 0) | |
4697 | err = -EEXIST; | |
4698 | return err; | |
4699 | } | |
4700 | ||
618e21d5 JB |
4701 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4702 | int mode, dev_t rdev) | |
4703 | { | |
4704 | struct btrfs_trans_handle *trans; | |
4705 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4706 | struct inode *inode = NULL; |
618e21d5 JB |
4707 | int err; |
4708 | int drop_inode = 0; | |
4709 | u64 objectid; | |
1832a6d5 | 4710 | unsigned long nr = 0; |
00e4e6b3 | 4711 | u64 index = 0; |
618e21d5 JB |
4712 | |
4713 | if (!new_valid_dev(rdev)) | |
4714 | return -EINVAL; | |
4715 | ||
a22285a6 YZ |
4716 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4717 | if (err) | |
4718 | return err; | |
4719 | ||
9ed74f2d JB |
4720 | /* |
4721 | * 2 for inode item and ref | |
4722 | * 2 for dir items | |
4723 | * 1 for xattr if selinux is on | |
4724 | */ | |
a22285a6 YZ |
4725 | trans = btrfs_start_transaction(root, 5); |
4726 | if (IS_ERR(trans)) | |
4727 | return PTR_ERR(trans); | |
1832a6d5 | 4728 | |
618e21d5 JB |
4729 | btrfs_set_trans_block_group(trans, dir); |
4730 | ||
aec7477b | 4731 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4732 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 | 4733 | BTRFS_I(dir)->block_group, mode, &index); |
7cf96da3 TI |
4734 | if (IS_ERR(inode)) { |
4735 | err = PTR_ERR(inode); | |
618e21d5 | 4736 | goto out_unlock; |
7cf96da3 | 4737 | } |
618e21d5 | 4738 | |
f34f57a3 | 4739 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4740 | if (err) { |
4741 | drop_inode = 1; | |
4742 | goto out_unlock; | |
4743 | } | |
4744 | ||
618e21d5 | 4745 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4746 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4747 | if (err) |
4748 | drop_inode = 1; | |
4749 | else { | |
4750 | inode->i_op = &btrfs_special_inode_operations; | |
4751 | init_special_inode(inode, inode->i_mode, rdev); | |
1b4ab1bb | 4752 | btrfs_update_inode(trans, root, inode); |
618e21d5 | 4753 | } |
618e21d5 JB |
4754 | btrfs_update_inode_block_group(trans, inode); |
4755 | btrfs_update_inode_block_group(trans, dir); | |
4756 | out_unlock: | |
d3c2fdcf | 4757 | nr = trans->blocks_used; |
89ce8a63 | 4758 | btrfs_end_transaction_throttle(trans, root); |
a22285a6 | 4759 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4760 | if (drop_inode) { |
4761 | inode_dec_link_count(inode); | |
4762 | iput(inode); | |
4763 | } | |
618e21d5 JB |
4764 | return err; |
4765 | } | |
4766 | ||
39279cc3 CM |
4767 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4768 | int mode, struct nameidata *nd) | |
4769 | { | |
4770 | struct btrfs_trans_handle *trans; | |
4771 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4772 | struct inode *inode = NULL; |
39279cc3 | 4773 | int drop_inode = 0; |
a22285a6 | 4774 | int err; |
1832a6d5 | 4775 | unsigned long nr = 0; |
39279cc3 | 4776 | u64 objectid; |
00e4e6b3 | 4777 | u64 index = 0; |
39279cc3 | 4778 | |
a22285a6 YZ |
4779 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4780 | if (err) | |
4781 | return err; | |
9ed74f2d JB |
4782 | /* |
4783 | * 2 for inode item and ref | |
4784 | * 2 for dir items | |
4785 | * 1 for xattr if selinux is on | |
4786 | */ | |
a22285a6 YZ |
4787 | trans = btrfs_start_transaction(root, 5); |
4788 | if (IS_ERR(trans)) | |
4789 | return PTR_ERR(trans); | |
9ed74f2d | 4790 | |
39279cc3 CM |
4791 | btrfs_set_trans_block_group(trans, dir); |
4792 | ||
aec7477b | 4793 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 JB |
4794 | dentry->d_name.len, dir->i_ino, objectid, |
4795 | BTRFS_I(dir)->block_group, mode, &index); | |
7cf96da3 TI |
4796 | if (IS_ERR(inode)) { |
4797 | err = PTR_ERR(inode); | |
39279cc3 | 4798 | goto out_unlock; |
7cf96da3 | 4799 | } |
39279cc3 | 4800 | |
f34f57a3 | 4801 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4802 | if (err) { |
4803 | drop_inode = 1; | |
4804 | goto out_unlock; | |
4805 | } | |
4806 | ||
39279cc3 | 4807 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 4808 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4809 | if (err) |
4810 | drop_inode = 1; | |
4811 | else { | |
4812 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4813 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
4814 | inode->i_fop = &btrfs_file_operations; |
4815 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 4816 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 4817 | } |
39279cc3 CM |
4818 | btrfs_update_inode_block_group(trans, inode); |
4819 | btrfs_update_inode_block_group(trans, dir); | |
4820 | out_unlock: | |
d3c2fdcf | 4821 | nr = trans->blocks_used; |
ab78c84d | 4822 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4823 | if (drop_inode) { |
4824 | inode_dec_link_count(inode); | |
4825 | iput(inode); | |
4826 | } | |
d3c2fdcf | 4827 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4828 | return err; |
4829 | } | |
4830 | ||
4831 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4832 | struct dentry *dentry) | |
4833 | { | |
4834 | struct btrfs_trans_handle *trans; | |
4835 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4836 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4837 | u64 index; |
1832a6d5 | 4838 | unsigned long nr = 0; |
39279cc3 CM |
4839 | int err; |
4840 | int drop_inode = 0; | |
4841 | ||
4842 | if (inode->i_nlink == 0) | |
4843 | return -ENOENT; | |
4844 | ||
4a8be425 TH |
4845 | /* do not allow sys_link's with other subvols of the same device */ |
4846 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4847 | return -EXDEV; |
4a8be425 | 4848 | |
c055e99e AV |
4849 | if (inode->i_nlink == ~0U) |
4850 | return -EMLINK; | |
4851 | ||
3de4586c | 4852 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4853 | if (err) |
4854 | goto fail; | |
4855 | ||
a22285a6 | 4856 | /* |
7e6b6465 | 4857 | * 2 items for inode and inode ref |
a22285a6 | 4858 | * 2 items for dir items |
7e6b6465 | 4859 | * 1 item for parent inode |
a22285a6 | 4860 | */ |
7e6b6465 | 4861 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4862 | if (IS_ERR(trans)) { |
4863 | err = PTR_ERR(trans); | |
4864 | goto fail; | |
4865 | } | |
5f39d397 | 4866 | |
3153495d MX |
4867 | btrfs_inc_nlink(inode); |
4868 | inode->i_ctime = CURRENT_TIME; | |
4869 | ||
39279cc3 | 4870 | btrfs_set_trans_block_group(trans, dir); |
7de9c6ee | 4871 | ihold(inode); |
aec7477b | 4872 | |
a1b075d2 | 4873 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4874 | |
a5719521 | 4875 | if (err) { |
54aa1f4d | 4876 | drop_inode = 1; |
a5719521 | 4877 | } else { |
6a912213 | 4878 | struct dentry *parent = dget_parent(dentry); |
a5719521 YZ |
4879 | btrfs_update_inode_block_group(trans, dir); |
4880 | err = btrfs_update_inode(trans, root, inode); | |
4881 | BUG_ON(err); | |
6a912213 JB |
4882 | btrfs_log_new_name(trans, inode, NULL, parent); |
4883 | dput(parent); | |
a5719521 | 4884 | } |
39279cc3 | 4885 | |
d3c2fdcf | 4886 | nr = trans->blocks_used; |
ab78c84d | 4887 | btrfs_end_transaction_throttle(trans, root); |
1832a6d5 | 4888 | fail: |
39279cc3 CM |
4889 | if (drop_inode) { |
4890 | inode_dec_link_count(inode); | |
4891 | iput(inode); | |
4892 | } | |
d3c2fdcf | 4893 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4894 | return err; |
4895 | } | |
4896 | ||
39279cc3 CM |
4897 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4898 | { | |
b9d86667 | 4899 | struct inode *inode = NULL; |
39279cc3 CM |
4900 | struct btrfs_trans_handle *trans; |
4901 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4902 | int err = 0; | |
4903 | int drop_on_err = 0; | |
b9d86667 | 4904 | u64 objectid = 0; |
00e4e6b3 | 4905 | u64 index = 0; |
d3c2fdcf | 4906 | unsigned long nr = 1; |
39279cc3 | 4907 | |
a22285a6 YZ |
4908 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
4909 | if (err) | |
4910 | return err; | |
4911 | ||
9ed74f2d JB |
4912 | /* |
4913 | * 2 items for inode and ref | |
4914 | * 2 items for dir items | |
4915 | * 1 for xattr if selinux is on | |
4916 | */ | |
a22285a6 YZ |
4917 | trans = btrfs_start_transaction(root, 5); |
4918 | if (IS_ERR(trans)) | |
4919 | return PTR_ERR(trans); | |
9ed74f2d | 4920 | btrfs_set_trans_block_group(trans, dir); |
39279cc3 | 4921 | |
aec7477b | 4922 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 4923 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
4924 | BTRFS_I(dir)->block_group, S_IFDIR | mode, |
4925 | &index); | |
39279cc3 CM |
4926 | if (IS_ERR(inode)) { |
4927 | err = PTR_ERR(inode); | |
4928 | goto out_fail; | |
4929 | } | |
5f39d397 | 4930 | |
39279cc3 | 4931 | drop_on_err = 1; |
33268eaf | 4932 | |
f34f57a3 | 4933 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
4934 | if (err) |
4935 | goto out_fail; | |
4936 | ||
39279cc3 CM |
4937 | inode->i_op = &btrfs_dir_inode_operations; |
4938 | inode->i_fop = &btrfs_dir_file_operations; | |
4939 | btrfs_set_trans_block_group(trans, inode); | |
4940 | ||
dbe674a9 | 4941 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4942 | err = btrfs_update_inode(trans, root, inode); |
4943 | if (err) | |
4944 | goto out_fail; | |
5f39d397 | 4945 | |
a1b075d2 JB |
4946 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4947 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4948 | if (err) |
4949 | goto out_fail; | |
5f39d397 | 4950 | |
39279cc3 CM |
4951 | d_instantiate(dentry, inode); |
4952 | drop_on_err = 0; | |
39279cc3 CM |
4953 | btrfs_update_inode_block_group(trans, inode); |
4954 | btrfs_update_inode_block_group(trans, dir); | |
4955 | ||
4956 | out_fail: | |
d3c2fdcf | 4957 | nr = trans->blocks_used; |
ab78c84d | 4958 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
4959 | if (drop_on_err) |
4960 | iput(inode); | |
d3c2fdcf | 4961 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4962 | return err; |
4963 | } | |
4964 | ||
d352ac68 CM |
4965 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4966 | * and an extent that you want to insert, deal with overlap and insert | |
4967 | * the new extent into the tree. | |
4968 | */ | |
3b951516 CM |
4969 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4970 | struct extent_map *existing, | |
e6dcd2dc CM |
4971 | struct extent_map *em, |
4972 | u64 map_start, u64 map_len) | |
3b951516 CM |
4973 | { |
4974 | u64 start_diff; | |
3b951516 | 4975 | |
e6dcd2dc CM |
4976 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4977 | start_diff = map_start - em->start; | |
4978 | em->start = map_start; | |
4979 | em->len = map_len; | |
c8b97818 CM |
4980 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4981 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4982 | em->block_start += start_diff; |
c8b97818 CM |
4983 | em->block_len -= start_diff; |
4984 | } | |
e6dcd2dc | 4985 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4986 | } |
4987 | ||
c8b97818 CM |
4988 | static noinline int uncompress_inline(struct btrfs_path *path, |
4989 | struct inode *inode, struct page *page, | |
4990 | size_t pg_offset, u64 extent_offset, | |
4991 | struct btrfs_file_extent_item *item) | |
4992 | { | |
4993 | int ret; | |
4994 | struct extent_buffer *leaf = path->nodes[0]; | |
4995 | char *tmp; | |
4996 | size_t max_size; | |
4997 | unsigned long inline_size; | |
4998 | unsigned long ptr; | |
261507a0 | 4999 | int compress_type; |
c8b97818 CM |
5000 | |
5001 | WARN_ON(pg_offset != 0); | |
261507a0 | 5002 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5003 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5004 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5005 | btrfs_item_nr(leaf, path->slots[0])); | |
5006 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5007 | if (!tmp) |
5008 | return -ENOMEM; | |
c8b97818 CM |
5009 | ptr = btrfs_file_extent_inline_start(item); |
5010 | ||
5011 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5012 | ||
5b050f04 | 5013 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5014 | ret = btrfs_decompress(compress_type, tmp, page, |
5015 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
5016 | if (ret) { |
5017 | char *kaddr = kmap_atomic(page, KM_USER0); | |
5018 | unsigned long copy_size = min_t(u64, | |
5019 | PAGE_CACHE_SIZE - pg_offset, | |
5020 | max_size - extent_offset); | |
5021 | memset(kaddr + pg_offset, 0, copy_size); | |
5022 | kunmap_atomic(kaddr, KM_USER0); | |
5023 | } | |
5024 | kfree(tmp); | |
5025 | return 0; | |
5026 | } | |
5027 | ||
d352ac68 CM |
5028 | /* |
5029 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5030 | * the ugly parts come from merging extents from the disk with the in-ram |
5031 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5032 | * where the in-ram extents might be locked pending data=ordered completion. |
5033 | * | |
5034 | * This also copies inline extents directly into the page. | |
5035 | */ | |
d397712b | 5036 | |
a52d9a80 | 5037 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5038 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5039 | int create) |
5040 | { | |
5041 | int ret; | |
5042 | int err = 0; | |
db94535d | 5043 | u64 bytenr; |
a52d9a80 CM |
5044 | u64 extent_start = 0; |
5045 | u64 extent_end = 0; | |
5046 | u64 objectid = inode->i_ino; | |
5047 | u32 found_type; | |
f421950f | 5048 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5049 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5050 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5051 | struct extent_buffer *leaf; |
5052 | struct btrfs_key found_key; | |
a52d9a80 CM |
5053 | struct extent_map *em = NULL; |
5054 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5055 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5056 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5057 | int compress_type; |
a52d9a80 | 5058 | |
a52d9a80 | 5059 | again: |
890871be | 5060 | read_lock(&em_tree->lock); |
d1310b2e | 5061 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5062 | if (em) |
5063 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5064 | read_unlock(&em_tree->lock); |
d1310b2e | 5065 | |
a52d9a80 | 5066 | if (em) { |
e1c4b745 CM |
5067 | if (em->start > start || em->start + em->len <= start) |
5068 | free_extent_map(em); | |
5069 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5070 | free_extent_map(em); |
5071 | else | |
5072 | goto out; | |
a52d9a80 | 5073 | } |
d1310b2e | 5074 | em = alloc_extent_map(GFP_NOFS); |
a52d9a80 | 5075 | if (!em) { |
d1310b2e CM |
5076 | err = -ENOMEM; |
5077 | goto out; | |
a52d9a80 | 5078 | } |
e6dcd2dc | 5079 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5080 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5081 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5082 | em->len = (u64)-1; |
c8b97818 | 5083 | em->block_len = (u64)-1; |
f421950f CM |
5084 | |
5085 | if (!path) { | |
5086 | path = btrfs_alloc_path(); | |
5087 | BUG_ON(!path); | |
5088 | } | |
5089 | ||
179e29e4 CM |
5090 | ret = btrfs_lookup_file_extent(trans, root, path, |
5091 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5092 | if (ret < 0) { |
5093 | err = ret; | |
5094 | goto out; | |
5095 | } | |
5096 | ||
5097 | if (ret != 0) { | |
5098 | if (path->slots[0] == 0) | |
5099 | goto not_found; | |
5100 | path->slots[0]--; | |
5101 | } | |
5102 | ||
5f39d397 CM |
5103 | leaf = path->nodes[0]; |
5104 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5105 | struct btrfs_file_extent_item); |
a52d9a80 | 5106 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5107 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5108 | found_type = btrfs_key_type(&found_key); | |
5109 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5110 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5111 | goto not_found; | |
5112 | } | |
5113 | ||
5f39d397 CM |
5114 | found_type = btrfs_file_extent_type(leaf, item); |
5115 | extent_start = found_key.offset; | |
261507a0 | 5116 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5117 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5118 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5119 | extent_end = extent_start + |
db94535d | 5120 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5121 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5122 | size_t size; | |
5123 | size = btrfs_file_extent_inline_len(leaf, item); | |
5124 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5125 | ~((u64)root->sectorsize - 1); | |
5126 | } | |
5127 | ||
5128 | if (start >= extent_end) { | |
5129 | path->slots[0]++; | |
5130 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5131 | ret = btrfs_next_leaf(root, path); | |
5132 | if (ret < 0) { | |
5133 | err = ret; | |
5134 | goto out; | |
a52d9a80 | 5135 | } |
9036c102 YZ |
5136 | if (ret > 0) |
5137 | goto not_found; | |
5138 | leaf = path->nodes[0]; | |
a52d9a80 | 5139 | } |
9036c102 YZ |
5140 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5141 | if (found_key.objectid != objectid || | |
5142 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5143 | goto not_found; | |
5144 | if (start + len <= found_key.offset) | |
5145 | goto not_found; | |
5146 | em->start = start; | |
5147 | em->len = found_key.offset - start; | |
5148 | goto not_found_em; | |
5149 | } | |
5150 | ||
d899e052 YZ |
5151 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5152 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5153 | em->start = extent_start; |
5154 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5155 | em->orig_start = extent_start - |
5156 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5157 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5158 | if (bytenr == 0) { | |
5f39d397 | 5159 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5160 | goto insert; |
5161 | } | |
261507a0 | 5162 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5163 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5164 | em->compress_type = compress_type; |
c8b97818 CM |
5165 | em->block_start = bytenr; |
5166 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5167 | item); | |
5168 | } else { | |
5169 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5170 | em->block_start = bytenr; | |
5171 | em->block_len = em->len; | |
d899e052 YZ |
5172 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5173 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5174 | } |
a52d9a80 CM |
5175 | goto insert; |
5176 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5177 | unsigned long ptr; |
a52d9a80 | 5178 | char *map; |
3326d1b0 CM |
5179 | size_t size; |
5180 | size_t extent_offset; | |
5181 | size_t copy_size; | |
a52d9a80 | 5182 | |
689f9346 | 5183 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5184 | if (!page || create) { |
689f9346 | 5185 | em->start = extent_start; |
9036c102 | 5186 | em->len = extent_end - extent_start; |
689f9346 Y |
5187 | goto out; |
5188 | } | |
5f39d397 | 5189 | |
9036c102 YZ |
5190 | size = btrfs_file_extent_inline_len(leaf, item); |
5191 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5192 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5193 | size - extent_offset); |
3326d1b0 | 5194 | em->start = extent_start + extent_offset; |
70dec807 CM |
5195 | em->len = (copy_size + root->sectorsize - 1) & |
5196 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5197 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5198 | if (compress_type) { |
c8b97818 | 5199 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5200 | em->compress_type = compress_type; |
5201 | } | |
689f9346 | 5202 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5203 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5204 | if (btrfs_file_extent_compression(leaf, item) != |
5205 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5206 | ret = uncompress_inline(path, inode, page, |
5207 | pg_offset, | |
5208 | extent_offset, item); | |
5209 | BUG_ON(ret); | |
5210 | } else { | |
5211 | map = kmap(page); | |
5212 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5213 | copy_size); | |
93c82d57 CM |
5214 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5215 | memset(map + pg_offset + copy_size, 0, | |
5216 | PAGE_CACHE_SIZE - pg_offset - | |
5217 | copy_size); | |
5218 | } | |
c8b97818 CM |
5219 | kunmap(page); |
5220 | } | |
179e29e4 CM |
5221 | flush_dcache_page(page); |
5222 | } else if (create && PageUptodate(page)) { | |
0ca1f7ce | 5223 | WARN_ON(1); |
179e29e4 CM |
5224 | if (!trans) { |
5225 | kunmap(page); | |
5226 | free_extent_map(em); | |
5227 | em = NULL; | |
5228 | btrfs_release_path(root, path); | |
f9295749 | 5229 | trans = btrfs_join_transaction(root, 1); |
3612b495 TI |
5230 | if (IS_ERR(trans)) |
5231 | return ERR_CAST(trans); | |
179e29e4 CM |
5232 | goto again; |
5233 | } | |
c8b97818 | 5234 | map = kmap(page); |
70dec807 | 5235 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5236 | copy_size); |
c8b97818 | 5237 | kunmap(page); |
179e29e4 | 5238 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5239 | } |
d1310b2e | 5240 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5241 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5242 | goto insert; |
5243 | } else { | |
d397712b | 5244 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5245 | WARN_ON(1); |
5246 | } | |
5247 | not_found: | |
5248 | em->start = start; | |
d1310b2e | 5249 | em->len = len; |
a52d9a80 | 5250 | not_found_em: |
5f39d397 | 5251 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5252 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 CM |
5253 | insert: |
5254 | btrfs_release_path(root, path); | |
d1310b2e | 5255 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5256 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5257 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5258 | (unsigned long long)em->len, | |
5259 | (unsigned long long)start, | |
5260 | (unsigned long long)len); | |
a52d9a80 CM |
5261 | err = -EIO; |
5262 | goto out; | |
5263 | } | |
d1310b2e CM |
5264 | |
5265 | err = 0; | |
890871be | 5266 | write_lock(&em_tree->lock); |
a52d9a80 | 5267 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5268 | /* it is possible that someone inserted the extent into the tree |
5269 | * while we had the lock dropped. It is also possible that | |
5270 | * an overlapping map exists in the tree | |
5271 | */ | |
a52d9a80 | 5272 | if (ret == -EEXIST) { |
3b951516 | 5273 | struct extent_map *existing; |
e6dcd2dc CM |
5274 | |
5275 | ret = 0; | |
5276 | ||
3b951516 | 5277 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5278 | if (existing && (existing->start > start || |
5279 | existing->start + existing->len <= start)) { | |
5280 | free_extent_map(existing); | |
5281 | existing = NULL; | |
5282 | } | |
3b951516 CM |
5283 | if (!existing) { |
5284 | existing = lookup_extent_mapping(em_tree, em->start, | |
5285 | em->len); | |
5286 | if (existing) { | |
5287 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5288 | em, start, |
5289 | root->sectorsize); | |
3b951516 CM |
5290 | free_extent_map(existing); |
5291 | if (err) { | |
5292 | free_extent_map(em); | |
5293 | em = NULL; | |
5294 | } | |
5295 | } else { | |
5296 | err = -EIO; | |
3b951516 CM |
5297 | free_extent_map(em); |
5298 | em = NULL; | |
5299 | } | |
5300 | } else { | |
5301 | free_extent_map(em); | |
5302 | em = existing; | |
e6dcd2dc | 5303 | err = 0; |
a52d9a80 | 5304 | } |
a52d9a80 | 5305 | } |
890871be | 5306 | write_unlock(&em_tree->lock); |
a52d9a80 | 5307 | out: |
1abe9b8a | 5308 | |
5309 | trace_btrfs_get_extent(root, em); | |
5310 | ||
f421950f CM |
5311 | if (path) |
5312 | btrfs_free_path(path); | |
a52d9a80 CM |
5313 | if (trans) { |
5314 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5315 | if (!err) |
a52d9a80 CM |
5316 | err = ret; |
5317 | } | |
a52d9a80 CM |
5318 | if (err) { |
5319 | free_extent_map(em); | |
a52d9a80 CM |
5320 | return ERR_PTR(err); |
5321 | } | |
5322 | return em; | |
5323 | } | |
5324 | ||
ec29ed5b CM |
5325 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5326 | size_t pg_offset, u64 start, u64 len, | |
5327 | int create) | |
5328 | { | |
5329 | struct extent_map *em; | |
5330 | struct extent_map *hole_em = NULL; | |
5331 | u64 range_start = start; | |
5332 | u64 end; | |
5333 | u64 found; | |
5334 | u64 found_end; | |
5335 | int err = 0; | |
5336 | ||
5337 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5338 | if (IS_ERR(em)) | |
5339 | return em; | |
5340 | if (em) { | |
5341 | /* | |
5342 | * if our em maps to a hole, there might | |
5343 | * actually be delalloc bytes behind it | |
5344 | */ | |
5345 | if (em->block_start != EXTENT_MAP_HOLE) | |
5346 | return em; | |
5347 | else | |
5348 | hole_em = em; | |
5349 | } | |
5350 | ||
5351 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5352 | end = start + len; | |
5353 | if (end < start) | |
5354 | end = (u64)-1; | |
5355 | else | |
5356 | end -= 1; | |
5357 | ||
5358 | em = NULL; | |
5359 | ||
5360 | /* ok, we didn't find anything, lets look for delalloc */ | |
5361 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5362 | end, len, EXTENT_DELALLOC, 1); | |
5363 | found_end = range_start + found; | |
5364 | if (found_end < range_start) | |
5365 | found_end = (u64)-1; | |
5366 | ||
5367 | /* | |
5368 | * we didn't find anything useful, return | |
5369 | * the original results from get_extent() | |
5370 | */ | |
5371 | if (range_start > end || found_end <= start) { | |
5372 | em = hole_em; | |
5373 | hole_em = NULL; | |
5374 | goto out; | |
5375 | } | |
5376 | ||
5377 | /* adjust the range_start to make sure it doesn't | |
5378 | * go backwards from the start they passed in | |
5379 | */ | |
5380 | range_start = max(start,range_start); | |
5381 | found = found_end - range_start; | |
5382 | ||
5383 | if (found > 0) { | |
5384 | u64 hole_start = start; | |
5385 | u64 hole_len = len; | |
5386 | ||
5387 | em = alloc_extent_map(GFP_NOFS); | |
5388 | if (!em) { | |
5389 | err = -ENOMEM; | |
5390 | goto out; | |
5391 | } | |
5392 | /* | |
5393 | * when btrfs_get_extent can't find anything it | |
5394 | * returns one huge hole | |
5395 | * | |
5396 | * make sure what it found really fits our range, and | |
5397 | * adjust to make sure it is based on the start from | |
5398 | * the caller | |
5399 | */ | |
5400 | if (hole_em) { | |
5401 | u64 calc_end = extent_map_end(hole_em); | |
5402 | ||
5403 | if (calc_end <= start || (hole_em->start > end)) { | |
5404 | free_extent_map(hole_em); | |
5405 | hole_em = NULL; | |
5406 | } else { | |
5407 | hole_start = max(hole_em->start, start); | |
5408 | hole_len = calc_end - hole_start; | |
5409 | } | |
5410 | } | |
5411 | em->bdev = NULL; | |
5412 | if (hole_em && range_start > hole_start) { | |
5413 | /* our hole starts before our delalloc, so we | |
5414 | * have to return just the parts of the hole | |
5415 | * that go until the delalloc starts | |
5416 | */ | |
5417 | em->len = min(hole_len, | |
5418 | range_start - hole_start); | |
5419 | em->start = hole_start; | |
5420 | em->orig_start = hole_start; | |
5421 | /* | |
5422 | * don't adjust block start at all, | |
5423 | * it is fixed at EXTENT_MAP_HOLE | |
5424 | */ | |
5425 | em->block_start = hole_em->block_start; | |
5426 | em->block_len = hole_len; | |
5427 | } else { | |
5428 | em->start = range_start; | |
5429 | em->len = found; | |
5430 | em->orig_start = range_start; | |
5431 | em->block_start = EXTENT_MAP_DELALLOC; | |
5432 | em->block_len = found; | |
5433 | } | |
5434 | } else if (hole_em) { | |
5435 | return hole_em; | |
5436 | } | |
5437 | out: | |
5438 | ||
5439 | free_extent_map(hole_em); | |
5440 | if (err) { | |
5441 | free_extent_map(em); | |
5442 | return ERR_PTR(err); | |
5443 | } | |
5444 | return em; | |
5445 | } | |
5446 | ||
4b46fce2 | 5447 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5448 | struct extent_map *em, |
4b46fce2 JB |
5449 | u64 start, u64 len) |
5450 | { | |
5451 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5452 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5453 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5454 | struct btrfs_key ins; | |
5455 | u64 alloc_hint; | |
5456 | int ret; | |
16d299ac | 5457 | bool insert = false; |
4b46fce2 | 5458 | |
16d299ac JB |
5459 | /* |
5460 | * Ok if the extent map we looked up is a hole and is for the exact | |
5461 | * range we want, there is no reason to allocate a new one, however if | |
5462 | * it is not right then we need to free this one and drop the cache for | |
5463 | * our range. | |
5464 | */ | |
5465 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5466 | em->len != len) { | |
5467 | free_extent_map(em); | |
5468 | em = NULL; | |
5469 | insert = true; | |
5470 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5471 | } | |
4b46fce2 JB |
5472 | |
5473 | trans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
5474 | if (IS_ERR(trans)) |
5475 | return ERR_CAST(trans); | |
4b46fce2 JB |
5476 | |
5477 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5478 | ||
5479 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5480 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5481 | alloc_hint, (u64)-1, &ins, 1); | |
5482 | if (ret) { | |
5483 | em = ERR_PTR(ret); | |
5484 | goto out; | |
5485 | } | |
5486 | ||
4b46fce2 | 5487 | if (!em) { |
16d299ac JB |
5488 | em = alloc_extent_map(GFP_NOFS); |
5489 | if (!em) { | |
5490 | em = ERR_PTR(-ENOMEM); | |
5491 | goto out; | |
5492 | } | |
4b46fce2 JB |
5493 | } |
5494 | ||
5495 | em->start = start; | |
5496 | em->orig_start = em->start; | |
5497 | em->len = ins.offset; | |
5498 | ||
5499 | em->block_start = ins.objectid; | |
5500 | em->block_len = ins.offset; | |
5501 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5502 | |
5503 | /* | |
5504 | * We need to do this because if we're using the original em we searched | |
5505 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5506 | */ | |
5507 | em->flags = 0; | |
4b46fce2 JB |
5508 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5509 | ||
16d299ac | 5510 | while (insert) { |
4b46fce2 JB |
5511 | write_lock(&em_tree->lock); |
5512 | ret = add_extent_mapping(em_tree, em); | |
5513 | write_unlock(&em_tree->lock); | |
5514 | if (ret != -EEXIST) | |
5515 | break; | |
5516 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5517 | } | |
5518 | ||
5519 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5520 | ins.offset, ins.offset, 0); | |
5521 | if (ret) { | |
5522 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5523 | em = ERR_PTR(ret); | |
5524 | } | |
5525 | out: | |
5526 | btrfs_end_transaction(trans, root); | |
5527 | return em; | |
5528 | } | |
5529 | ||
46bfbb5c CM |
5530 | /* |
5531 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5532 | * block must be cow'd | |
5533 | */ | |
5534 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5535 | struct inode *inode, u64 offset, u64 len) | |
5536 | { | |
5537 | struct btrfs_path *path; | |
5538 | int ret; | |
5539 | struct extent_buffer *leaf; | |
5540 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5541 | struct btrfs_file_extent_item *fi; | |
5542 | struct btrfs_key key; | |
5543 | u64 disk_bytenr; | |
5544 | u64 backref_offset; | |
5545 | u64 extent_end; | |
5546 | u64 num_bytes; | |
5547 | int slot; | |
5548 | int found_type; | |
5549 | ||
5550 | path = btrfs_alloc_path(); | |
5551 | if (!path) | |
5552 | return -ENOMEM; | |
5553 | ||
5554 | ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino, | |
5555 | offset, 0); | |
5556 | if (ret < 0) | |
5557 | goto out; | |
5558 | ||
5559 | slot = path->slots[0]; | |
5560 | if (ret == 1) { | |
5561 | if (slot == 0) { | |
5562 | /* can't find the item, must cow */ | |
5563 | ret = 0; | |
5564 | goto out; | |
5565 | } | |
5566 | slot--; | |
5567 | } | |
5568 | ret = 0; | |
5569 | leaf = path->nodes[0]; | |
5570 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5571 | if (key.objectid != inode->i_ino || | |
5572 | key.type != BTRFS_EXTENT_DATA_KEY) { | |
5573 | /* not our file or wrong item type, must cow */ | |
5574 | goto out; | |
5575 | } | |
5576 | ||
5577 | if (key.offset > offset) { | |
5578 | /* Wrong offset, must cow */ | |
5579 | goto out; | |
5580 | } | |
5581 | ||
5582 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5583 | found_type = btrfs_file_extent_type(leaf, fi); | |
5584 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5585 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5586 | /* not a regular extent, must cow */ | |
5587 | goto out; | |
5588 | } | |
5589 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5590 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5591 | ||
5592 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5593 | if (extent_end < offset + len) { | |
5594 | /* extent doesn't include our full range, must cow */ | |
5595 | goto out; | |
5596 | } | |
5597 | ||
5598 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5599 | goto out; | |
5600 | ||
5601 | /* | |
5602 | * look for other files referencing this extent, if we | |
5603 | * find any we must cow | |
5604 | */ | |
5605 | if (btrfs_cross_ref_exist(trans, root, inode->i_ino, | |
5606 | key.offset - backref_offset, disk_bytenr)) | |
5607 | goto out; | |
5608 | ||
5609 | /* | |
5610 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5611 | * in this extent we are about to write. If there | |
5612 | * are any csums in that range we have to cow in order | |
5613 | * to keep the csums correct | |
5614 | */ | |
5615 | disk_bytenr += backref_offset; | |
5616 | disk_bytenr += offset - key.offset; | |
5617 | num_bytes = min(offset + len, extent_end) - offset; | |
5618 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5619 | goto out; | |
5620 | /* | |
5621 | * all of the above have passed, it is safe to overwrite this extent | |
5622 | * without cow | |
5623 | */ | |
5624 | ret = 1; | |
5625 | out: | |
5626 | btrfs_free_path(path); | |
5627 | return ret; | |
5628 | } | |
5629 | ||
4b46fce2 JB |
5630 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5631 | struct buffer_head *bh_result, int create) | |
5632 | { | |
5633 | struct extent_map *em; | |
5634 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5635 | u64 start = iblock << inode->i_blkbits; | |
5636 | u64 len = bh_result->b_size; | |
46bfbb5c | 5637 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5638 | |
5639 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5640 | if (IS_ERR(em)) | |
5641 | return PTR_ERR(em); | |
5642 | ||
5643 | /* | |
5644 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5645 | * io. INLINE is special, and we could probably kludge it in here, but | |
5646 | * it's still buffered so for safety lets just fall back to the generic | |
5647 | * buffered path. | |
5648 | * | |
5649 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5650 | * decompress it, so there will be buffering required no matter what we | |
5651 | * do, so go ahead and fallback to buffered. | |
5652 | * | |
5653 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5654 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5655 | * the generic code. | |
5656 | */ | |
5657 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5658 | em->block_start == EXTENT_MAP_INLINE) { | |
5659 | free_extent_map(em); | |
5660 | return -ENOTBLK; | |
5661 | } | |
5662 | ||
5663 | /* Just a good old fashioned hole, return */ | |
5664 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5665 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5666 | free_extent_map(em); | |
5667 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5668 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5669 | start + root->sectorsize - 1, GFP_NOFS); | |
5670 | return 0; | |
5671 | } | |
5672 | ||
5673 | /* | |
5674 | * We don't allocate a new extent in the following cases | |
5675 | * | |
5676 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5677 | * existing extent. | |
5678 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5679 | * just use the extent. | |
5680 | * | |
5681 | */ | |
46bfbb5c CM |
5682 | if (!create) { |
5683 | len = em->len - (start - em->start); | |
4b46fce2 | 5684 | goto map; |
46bfbb5c | 5685 | } |
4b46fce2 JB |
5686 | |
5687 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5688 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5689 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5690 | int type; |
5691 | int ret; | |
46bfbb5c | 5692 | u64 block_start; |
4b46fce2 JB |
5693 | |
5694 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5695 | type = BTRFS_ORDERED_PREALLOC; | |
5696 | else | |
5697 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5698 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5699 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5700 | |
5701 | /* | |
5702 | * we're not going to log anything, but we do need | |
5703 | * to make sure the current transaction stays open | |
5704 | * while we look for nocow cross refs | |
5705 | */ | |
5706 | trans = btrfs_join_transaction(root, 0); | |
3612b495 | 5707 | if (IS_ERR(trans)) |
46bfbb5c CM |
5708 | goto must_cow; |
5709 | ||
5710 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5711 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5712 | block_start, len, len, type); | |
5713 | btrfs_end_transaction(trans, root); | |
5714 | if (ret) { | |
5715 | free_extent_map(em); | |
5716 | return ret; | |
5717 | } | |
5718 | goto unlock; | |
4b46fce2 | 5719 | } |
46bfbb5c | 5720 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5721 | } |
46bfbb5c CM |
5722 | must_cow: |
5723 | /* | |
5724 | * this will cow the extent, reset the len in case we changed | |
5725 | * it above | |
5726 | */ | |
5727 | len = bh_result->b_size; | |
16d299ac | 5728 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5729 | if (IS_ERR(em)) |
5730 | return PTR_ERR(em); | |
5731 | len = min(len, em->len - (start - em->start)); | |
5732 | unlock: | |
4845e44f CM |
5733 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5734 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5735 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5736 | map: |
5737 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5738 | inode->i_blkbits; | |
46bfbb5c | 5739 | bh_result->b_size = len; |
4b46fce2 JB |
5740 | bh_result->b_bdev = em->bdev; |
5741 | set_buffer_mapped(bh_result); | |
5742 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5743 | set_buffer_new(bh_result); | |
5744 | ||
5745 | free_extent_map(em); | |
5746 | ||
5747 | return 0; | |
5748 | } | |
5749 | ||
5750 | struct btrfs_dio_private { | |
5751 | struct inode *inode; | |
5752 | u64 logical_offset; | |
5753 | u64 disk_bytenr; | |
5754 | u64 bytes; | |
5755 | u32 *csums; | |
5756 | void *private; | |
e65e1535 MX |
5757 | |
5758 | /* number of bios pending for this dio */ | |
5759 | atomic_t pending_bios; | |
5760 | ||
5761 | /* IO errors */ | |
5762 | int errors; | |
5763 | ||
5764 | struct bio *orig_bio; | |
4b46fce2 JB |
5765 | }; |
5766 | ||
5767 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5768 | { | |
e65e1535 | 5769 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5770 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5771 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5772 | struct inode *inode = dip->inode; |
5773 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5774 | u64 start; | |
5775 | u32 *private = dip->csums; | |
5776 | ||
5777 | start = dip->logical_offset; | |
5778 | do { | |
5779 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5780 | struct page *page = bvec->bv_page; | |
5781 | char *kaddr; | |
5782 | u32 csum = ~(u32)0; | |
5783 | unsigned long flags; | |
5784 | ||
5785 | local_irq_save(flags); | |
5786 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5787 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5788 | csum, bvec->bv_len); | |
5789 | btrfs_csum_final(csum, (char *)&csum); | |
5790 | kunmap_atomic(kaddr, KM_IRQ0); | |
5791 | local_irq_restore(flags); | |
5792 | ||
5793 | flush_dcache_page(bvec->bv_page); | |
5794 | if (csum != *private) { | |
5795 | printk(KERN_ERR "btrfs csum failed ino %lu off" | |
5796 | " %llu csum %u private %u\n", | |
5797 | inode->i_ino, (unsigned long long)start, | |
5798 | csum, *private); | |
5799 | err = -EIO; | |
5800 | } | |
5801 | } | |
5802 | ||
5803 | start += bvec->bv_len; | |
5804 | private++; | |
5805 | bvec++; | |
5806 | } while (bvec <= bvec_end); | |
5807 | ||
5808 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5809 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5810 | bio->bi_private = dip->private; | |
5811 | ||
5812 | kfree(dip->csums); | |
5813 | kfree(dip); | |
c0da7aa1 JB |
5814 | |
5815 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5816 | if (err) | |
5817 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5818 | dio_end_io(bio, err); |
5819 | } | |
5820 | ||
5821 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5822 | { | |
5823 | struct btrfs_dio_private *dip = bio->bi_private; | |
5824 | struct inode *inode = dip->inode; | |
5825 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5826 | struct btrfs_trans_handle *trans; | |
5827 | struct btrfs_ordered_extent *ordered = NULL; | |
5828 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5829 | u64 ordered_offset = dip->logical_offset; |
5830 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5831 | int ret; |
5832 | ||
5833 | if (err) | |
5834 | goto out_done; | |
163cf09c CM |
5835 | again: |
5836 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5837 | &ordered_offset, | |
5838 | ordered_bytes); | |
4b46fce2 | 5839 | if (!ret) |
163cf09c | 5840 | goto out_test; |
4b46fce2 JB |
5841 | |
5842 | BUG_ON(!ordered); | |
5843 | ||
5844 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 5845 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5846 | err = -ENOMEM; |
5847 | goto out; | |
5848 | } | |
5849 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5850 | ||
5851 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5852 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5853 | if (!ret) | |
5854 | ret = btrfs_update_inode(trans, root, inode); | |
5855 | err = ret; | |
5856 | goto out; | |
5857 | } | |
5858 | ||
5859 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5860 | ordered->file_offset + ordered->len - 1, 0, | |
5861 | &cached_state, GFP_NOFS); | |
5862 | ||
5863 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5864 | ret = btrfs_mark_extent_written(trans, inode, | |
5865 | ordered->file_offset, | |
5866 | ordered->file_offset + | |
5867 | ordered->len); | |
5868 | if (ret) { | |
5869 | err = ret; | |
5870 | goto out_unlock; | |
5871 | } | |
5872 | } else { | |
5873 | ret = insert_reserved_file_extent(trans, inode, | |
5874 | ordered->file_offset, | |
5875 | ordered->start, | |
5876 | ordered->disk_len, | |
5877 | ordered->len, | |
5878 | ordered->len, | |
5879 | 0, 0, 0, | |
5880 | BTRFS_FILE_EXTENT_REG); | |
5881 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5882 | ordered->file_offset, ordered->len); | |
5883 | if (ret) { | |
5884 | err = ret; | |
5885 | WARN_ON(1); | |
5886 | goto out_unlock; | |
5887 | } | |
5888 | } | |
5889 | ||
5890 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
1ef30be1 JB |
5891 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); |
5892 | if (!ret) | |
5893 | btrfs_update_inode(trans, root, inode); | |
5894 | ret = 0; | |
4b46fce2 JB |
5895 | out_unlock: |
5896 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5897 | ordered->file_offset + ordered->len - 1, | |
5898 | &cached_state, GFP_NOFS); | |
5899 | out: | |
5900 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5901 | btrfs_end_transaction(trans, root); | |
163cf09c | 5902 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5903 | btrfs_put_ordered_extent(ordered); |
5904 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5905 | |
5906 | out_test: | |
5907 | /* | |
5908 | * our bio might span multiple ordered extents. If we haven't | |
5909 | * completed the accounting for the whole dio, go back and try again | |
5910 | */ | |
5911 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5912 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5913 | ordered_offset; | |
5914 | goto again; | |
5915 | } | |
4b46fce2 JB |
5916 | out_done: |
5917 | bio->bi_private = dip->private; | |
5918 | ||
5919 | kfree(dip->csums); | |
5920 | kfree(dip); | |
c0da7aa1 JB |
5921 | |
5922 | /* If we had an error make sure to clear the uptodate flag */ | |
5923 | if (err) | |
5924 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5925 | dio_end_io(bio, err); |
5926 | } | |
5927 | ||
eaf25d93 CM |
5928 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5929 | struct bio *bio, int mirror_num, | |
5930 | unsigned long bio_flags, u64 offset) | |
5931 | { | |
5932 | int ret; | |
5933 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5934 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5935 | BUG_ON(ret); | |
5936 | return 0; | |
5937 | } | |
5938 | ||
e65e1535 MX |
5939 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5940 | { | |
5941 | struct btrfs_dio_private *dip = bio->bi_private; | |
5942 | ||
5943 | if (err) { | |
5944 | printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu " | |
3dd1462e JB |
5945 | "sector %#Lx len %u err no %d\n", |
5946 | dip->inode->i_ino, bio->bi_rw, | |
5947 | (unsigned long long)bio->bi_sector, bio->bi_size, err); | |
e65e1535 MX |
5948 | dip->errors = 1; |
5949 | ||
5950 | /* | |
5951 | * before atomic variable goto zero, we must make sure | |
5952 | * dip->errors is perceived to be set. | |
5953 | */ | |
5954 | smp_mb__before_atomic_dec(); | |
5955 | } | |
5956 | ||
5957 | /* if there are more bios still pending for this dio, just exit */ | |
5958 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5959 | goto out; | |
5960 | ||
5961 | if (dip->errors) | |
5962 | bio_io_error(dip->orig_bio); | |
5963 | else { | |
5964 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5965 | bio_endio(dip->orig_bio, 0); | |
5966 | } | |
5967 | out: | |
5968 | bio_put(bio); | |
5969 | } | |
5970 | ||
5971 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5972 | u64 first_sector, gfp_t gfp_flags) | |
5973 | { | |
5974 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5975 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5976 | } | |
5977 | ||
5978 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5979 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 5980 | u32 *csums, int async_submit) |
e65e1535 MX |
5981 | { |
5982 | int write = rw & REQ_WRITE; | |
5983 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5984 | int ret; | |
5985 | ||
5986 | bio_get(bio); | |
5987 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5988 | if (ret) | |
5989 | goto err; | |
5990 | ||
1ae39938 JB |
5991 | if (skip_sum) |
5992 | goto map; | |
5993 | ||
5994 | if (write && async_submit) { | |
e65e1535 MX |
5995 | ret = btrfs_wq_submit_bio(root->fs_info, |
5996 | inode, rw, bio, 0, 0, | |
5997 | file_offset, | |
5998 | __btrfs_submit_bio_start_direct_io, | |
5999 | __btrfs_submit_bio_done); | |
6000 | goto err; | |
1ae39938 JB |
6001 | } else if (write) { |
6002 | /* | |
6003 | * If we aren't doing async submit, calculate the csum of the | |
6004 | * bio now. | |
6005 | */ | |
6006 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6007 | if (ret) | |
6008 | goto err; | |
c2db1073 TI |
6009 | } else if (!skip_sum) { |
6010 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 6011 | file_offset, csums); |
c2db1073 TI |
6012 | if (ret) |
6013 | goto err; | |
6014 | } | |
e65e1535 | 6015 | |
1ae39938 JB |
6016 | map: |
6017 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6018 | err: |
6019 | bio_put(bio); | |
6020 | return ret; | |
6021 | } | |
6022 | ||
6023 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6024 | int skip_sum) | |
6025 | { | |
6026 | struct inode *inode = dip->inode; | |
6027 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6028 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6029 | struct bio *bio; | |
6030 | struct bio *orig_bio = dip->orig_bio; | |
6031 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6032 | u64 start_sector = orig_bio->bi_sector; | |
6033 | u64 file_offset = dip->logical_offset; | |
6034 | u64 submit_len = 0; | |
6035 | u64 map_length; | |
6036 | int nr_pages = 0; | |
6037 | u32 *csums = dip->csums; | |
6038 | int ret = 0; | |
1ae39938 | 6039 | int async_submit = 0; |
98bc3149 | 6040 | int write = rw & REQ_WRITE; |
e65e1535 | 6041 | |
e65e1535 MX |
6042 | map_length = orig_bio->bi_size; |
6043 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6044 | &map_length, NULL, 0); | |
6045 | if (ret) { | |
64728bbb | 6046 | bio_put(orig_bio); |
e65e1535 MX |
6047 | return -EIO; |
6048 | } | |
6049 | ||
02f57c7a JB |
6050 | if (map_length >= orig_bio->bi_size) { |
6051 | bio = orig_bio; | |
6052 | goto submit; | |
6053 | } | |
6054 | ||
1ae39938 | 6055 | async_submit = 1; |
02f57c7a JB |
6056 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6057 | if (!bio) | |
6058 | return -ENOMEM; | |
6059 | bio->bi_private = dip; | |
6060 | bio->bi_end_io = btrfs_end_dio_bio; | |
6061 | atomic_inc(&dip->pending_bios); | |
6062 | ||
e65e1535 MX |
6063 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6064 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6065 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6066 | bvec->bv_offset) < bvec->bv_len)) { | |
6067 | /* | |
6068 | * inc the count before we submit the bio so | |
6069 | * we know the end IO handler won't happen before | |
6070 | * we inc the count. Otherwise, the dip might get freed | |
6071 | * before we're done setting it up | |
6072 | */ | |
6073 | atomic_inc(&dip->pending_bios); | |
6074 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6075 | file_offset, skip_sum, | |
1ae39938 | 6076 | csums, async_submit); |
e65e1535 MX |
6077 | if (ret) { |
6078 | bio_put(bio); | |
6079 | atomic_dec(&dip->pending_bios); | |
6080 | goto out_err; | |
6081 | } | |
6082 | ||
98bc3149 JB |
6083 | /* Write's use the ordered csums */ |
6084 | if (!write && !skip_sum) | |
e65e1535 MX |
6085 | csums = csums + nr_pages; |
6086 | start_sector += submit_len >> 9; | |
6087 | file_offset += submit_len; | |
6088 | ||
6089 | submit_len = 0; | |
6090 | nr_pages = 0; | |
6091 | ||
6092 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6093 | start_sector, GFP_NOFS); | |
6094 | if (!bio) | |
6095 | goto out_err; | |
6096 | bio->bi_private = dip; | |
6097 | bio->bi_end_io = btrfs_end_dio_bio; | |
6098 | ||
6099 | map_length = orig_bio->bi_size; | |
6100 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6101 | &map_length, NULL, 0); | |
6102 | if (ret) { | |
6103 | bio_put(bio); | |
6104 | goto out_err; | |
6105 | } | |
6106 | } else { | |
6107 | submit_len += bvec->bv_len; | |
6108 | nr_pages ++; | |
6109 | bvec++; | |
6110 | } | |
6111 | } | |
6112 | ||
02f57c7a | 6113 | submit: |
e65e1535 | 6114 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6115 | csums, async_submit); |
e65e1535 MX |
6116 | if (!ret) |
6117 | return 0; | |
6118 | ||
6119 | bio_put(bio); | |
6120 | out_err: | |
6121 | dip->errors = 1; | |
6122 | /* | |
6123 | * before atomic variable goto zero, we must | |
6124 | * make sure dip->errors is perceived to be set. | |
6125 | */ | |
6126 | smp_mb__before_atomic_dec(); | |
6127 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6128 | bio_io_error(dip->orig_bio); | |
6129 | ||
6130 | /* bio_end_io() will handle error, so we needn't return it */ | |
6131 | return 0; | |
6132 | } | |
6133 | ||
4b46fce2 JB |
6134 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6135 | loff_t file_offset) | |
6136 | { | |
6137 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6138 | struct btrfs_dio_private *dip; | |
6139 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6140 | int skip_sum; |
7b6d91da | 6141 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6142 | int ret = 0; |
6143 | ||
6144 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6145 | ||
6146 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6147 | if (!dip) { | |
6148 | ret = -ENOMEM; | |
6149 | goto free_ordered; | |
6150 | } | |
6151 | dip->csums = NULL; | |
6152 | ||
98bc3149 JB |
6153 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6154 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6155 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6156 | if (!dip->csums) { | |
b4966b77 | 6157 | kfree(dip); |
4b46fce2 JB |
6158 | ret = -ENOMEM; |
6159 | goto free_ordered; | |
6160 | } | |
6161 | } | |
6162 | ||
6163 | dip->private = bio->bi_private; | |
6164 | dip->inode = inode; | |
6165 | dip->logical_offset = file_offset; | |
6166 | ||
4b46fce2 JB |
6167 | dip->bytes = 0; |
6168 | do { | |
6169 | dip->bytes += bvec->bv_len; | |
6170 | bvec++; | |
6171 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6172 | ||
46bfbb5c | 6173 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6174 | bio->bi_private = dip; |
e65e1535 MX |
6175 | dip->errors = 0; |
6176 | dip->orig_bio = bio; | |
6177 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6178 | |
6179 | if (write) | |
6180 | bio->bi_end_io = btrfs_endio_direct_write; | |
6181 | else | |
6182 | bio->bi_end_io = btrfs_endio_direct_read; | |
6183 | ||
e65e1535 MX |
6184 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6185 | if (!ret) | |
eaf25d93 | 6186 | return; |
4b46fce2 JB |
6187 | free_ordered: |
6188 | /* | |
6189 | * If this is a write, we need to clean up the reserved space and kill | |
6190 | * the ordered extent. | |
6191 | */ | |
6192 | if (write) { | |
6193 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6194 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6195 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6196 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6197 | btrfs_free_reserved_extent(root, ordered->start, | |
6198 | ordered->disk_len); | |
6199 | btrfs_put_ordered_extent(ordered); | |
6200 | btrfs_put_ordered_extent(ordered); | |
6201 | } | |
6202 | bio_endio(bio, ret); | |
6203 | } | |
6204 | ||
5a5f79b5 CM |
6205 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6206 | const struct iovec *iov, loff_t offset, | |
6207 | unsigned long nr_segs) | |
6208 | { | |
6209 | int seg; | |
a1b75f7d | 6210 | int i; |
5a5f79b5 CM |
6211 | size_t size; |
6212 | unsigned long addr; | |
6213 | unsigned blocksize_mask = root->sectorsize - 1; | |
6214 | ssize_t retval = -EINVAL; | |
6215 | loff_t end = offset; | |
6216 | ||
6217 | if (offset & blocksize_mask) | |
6218 | goto out; | |
6219 | ||
6220 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6221 | for (seg = 0; seg < nr_segs; seg++) { | |
6222 | addr = (unsigned long)iov[seg].iov_base; | |
6223 | size = iov[seg].iov_len; | |
6224 | end += size; | |
a1b75f7d | 6225 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6226 | goto out; |
a1b75f7d JB |
6227 | |
6228 | /* If this is a write we don't need to check anymore */ | |
6229 | if (rw & WRITE) | |
6230 | continue; | |
6231 | ||
6232 | /* | |
6233 | * Check to make sure we don't have duplicate iov_base's in this | |
6234 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6235 | * when reading back. | |
6236 | */ | |
6237 | for (i = seg + 1; i < nr_segs; i++) { | |
6238 | if (iov[seg].iov_base == iov[i].iov_base) | |
6239 | goto out; | |
6240 | } | |
5a5f79b5 CM |
6241 | } |
6242 | retval = 0; | |
6243 | out: | |
6244 | return retval; | |
6245 | } | |
16432985 CM |
6246 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6247 | const struct iovec *iov, loff_t offset, | |
6248 | unsigned long nr_segs) | |
6249 | { | |
4b46fce2 JB |
6250 | struct file *file = iocb->ki_filp; |
6251 | struct inode *inode = file->f_mapping->host; | |
6252 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6253 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6254 | u64 lockstart, lockend; |
6255 | ssize_t ret; | |
4845e44f CM |
6256 | int writing = rw & WRITE; |
6257 | int write_bits = 0; | |
3f7c579c | 6258 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6259 | |
5a5f79b5 CM |
6260 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6261 | offset, nr_segs)) { | |
6262 | return 0; | |
6263 | } | |
6264 | ||
4b46fce2 | 6265 | lockstart = offset; |
3f7c579c CM |
6266 | lockend = offset + count - 1; |
6267 | ||
6268 | if (writing) { | |
6269 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6270 | if (ret) | |
6271 | goto out; | |
6272 | } | |
4845e44f | 6273 | |
4b46fce2 | 6274 | while (1) { |
4845e44f CM |
6275 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6276 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6277 | /* |
6278 | * We're concerned with the entire range that we're going to be | |
6279 | * doing DIO to, so we need to make sure theres no ordered | |
6280 | * extents in this range. | |
6281 | */ | |
6282 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6283 | lockend - lockstart + 1); | |
6284 | if (!ordered) | |
6285 | break; | |
4845e44f CM |
6286 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6287 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6288 | btrfs_start_ordered_extent(inode, ordered, 1); |
6289 | btrfs_put_ordered_extent(ordered); | |
6290 | cond_resched(); | |
6291 | } | |
6292 | ||
4845e44f CM |
6293 | /* |
6294 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6295 | * the dirty or uptodate bits | |
6296 | */ | |
6297 | if (writing) { | |
6298 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6299 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6300 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6301 | GFP_NOFS); | |
6302 | if (ret) { | |
6303 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6304 | lockend, EXTENT_LOCKED | write_bits, | |
6305 | 1, 0, &cached_state, GFP_NOFS); | |
6306 | goto out; | |
6307 | } | |
6308 | } | |
6309 | ||
6310 | free_extent_state(cached_state); | |
6311 | cached_state = NULL; | |
6312 | ||
5a5f79b5 CM |
6313 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6314 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6315 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6316 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6317 | |
6318 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6319 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6320 | offset + iov_length(iov, nr_segs) - 1, | |
6321 | EXTENT_LOCKED | write_bits, 1, 0, | |
6322 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6323 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6324 | /* | |
6325 | * We're falling back to buffered, unlock the section we didn't | |
6326 | * do IO on. | |
6327 | */ | |
4845e44f CM |
6328 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6329 | offset + iov_length(iov, nr_segs) - 1, | |
6330 | EXTENT_LOCKED | write_bits, 1, 0, | |
6331 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6332 | } |
4845e44f CM |
6333 | out: |
6334 | free_extent_state(cached_state); | |
4b46fce2 | 6335 | return ret; |
16432985 CM |
6336 | } |
6337 | ||
1506fcc8 YS |
6338 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6339 | __u64 start, __u64 len) | |
6340 | { | |
ec29ed5b | 6341 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6342 | } |
6343 | ||
a52d9a80 | 6344 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6345 | { |
d1310b2e CM |
6346 | struct extent_io_tree *tree; |
6347 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
a52d9a80 | 6348 | return extent_read_full_page(tree, page, btrfs_get_extent); |
9ebefb18 | 6349 | } |
1832a6d5 | 6350 | |
a52d9a80 | 6351 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6352 | { |
d1310b2e | 6353 | struct extent_io_tree *tree; |
b888db2b CM |
6354 | |
6355 | ||
6356 | if (current->flags & PF_MEMALLOC) { | |
6357 | redirty_page_for_writepage(wbc, page); | |
6358 | unlock_page(page); | |
6359 | return 0; | |
6360 | } | |
d1310b2e | 6361 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6362 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6363 | } |
6364 | ||
f421950f CM |
6365 | int btrfs_writepages(struct address_space *mapping, |
6366 | struct writeback_control *wbc) | |
b293f02e | 6367 | { |
d1310b2e | 6368 | struct extent_io_tree *tree; |
771ed689 | 6369 | |
d1310b2e | 6370 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6371 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6372 | } | |
6373 | ||
3ab2fb5a CM |
6374 | static int |
6375 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6376 | struct list_head *pages, unsigned nr_pages) | |
6377 | { | |
d1310b2e CM |
6378 | struct extent_io_tree *tree; |
6379 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6380 | return extent_readpages(tree, mapping, pages, nr_pages, |
6381 | btrfs_get_extent); | |
6382 | } | |
e6dcd2dc | 6383 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6384 | { |
d1310b2e CM |
6385 | struct extent_io_tree *tree; |
6386 | struct extent_map_tree *map; | |
a52d9a80 | 6387 | int ret; |
8c2383c3 | 6388 | |
d1310b2e CM |
6389 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6390 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6391 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6392 | if (ret == 1) { |
6393 | ClearPagePrivate(page); | |
6394 | set_page_private(page, 0); | |
6395 | page_cache_release(page); | |
39279cc3 | 6396 | } |
a52d9a80 | 6397 | return ret; |
39279cc3 CM |
6398 | } |
6399 | ||
e6dcd2dc CM |
6400 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6401 | { | |
98509cfc CM |
6402 | if (PageWriteback(page) || PageDirty(page)) |
6403 | return 0; | |
b335b003 | 6404 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6405 | } |
6406 | ||
a52d9a80 | 6407 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6408 | { |
d1310b2e | 6409 | struct extent_io_tree *tree; |
e6dcd2dc | 6410 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6411 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6412 | u64 page_start = page_offset(page); |
6413 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6414 | |
8b62b72b CM |
6415 | |
6416 | /* | |
6417 | * we have the page locked, so new writeback can't start, | |
6418 | * and the dirty bit won't be cleared while we are here. | |
6419 | * | |
6420 | * Wait for IO on this page so that we can safely clear | |
6421 | * the PagePrivate2 bit and do ordered accounting | |
6422 | */ | |
e6dcd2dc | 6423 | wait_on_page_writeback(page); |
8b62b72b | 6424 | |
d1310b2e | 6425 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6426 | if (offset) { |
6427 | btrfs_releasepage(page, GFP_NOFS); | |
6428 | return; | |
6429 | } | |
2ac55d41 JB |
6430 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6431 | GFP_NOFS); | |
e6dcd2dc CM |
6432 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6433 | page_offset(page)); | |
6434 | if (ordered) { | |
eb84ae03 CM |
6435 | /* |
6436 | * IO on this page will never be started, so we need | |
6437 | * to account for any ordered extents now | |
6438 | */ | |
e6dcd2dc CM |
6439 | clear_extent_bit(tree, page_start, page_end, |
6440 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6441 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6442 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6443 | /* |
6444 | * whoever cleared the private bit is responsible | |
6445 | * for the finish_ordered_io | |
6446 | */ | |
6447 | if (TestClearPagePrivate2(page)) { | |
6448 | btrfs_finish_ordered_io(page->mapping->host, | |
6449 | page_start, page_end); | |
6450 | } | |
e6dcd2dc | 6451 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6452 | cached_state = NULL; |
6453 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6454 | GFP_NOFS); | |
e6dcd2dc CM |
6455 | } |
6456 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6457 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6458 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6459 | __btrfs_releasepage(page, GFP_NOFS); |
6460 | ||
4a096752 | 6461 | ClearPageChecked(page); |
9ad6b7bc | 6462 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6463 | ClearPagePrivate(page); |
6464 | set_page_private(page, 0); | |
6465 | page_cache_release(page); | |
6466 | } | |
39279cc3 CM |
6467 | } |
6468 | ||
9ebefb18 CM |
6469 | /* |
6470 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6471 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6472 | * be careful to check for EOF conditions here. We set the page up correctly | |
6473 | * for a written page which means we get ENOSPC checking when writing into | |
6474 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6475 | * support these features. | |
6476 | * | |
6477 | * We are not allowed to take the i_mutex here so we have to play games to | |
6478 | * protect against truncate races as the page could now be beyond EOF. Because | |
6479 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6480 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6481 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6482 | * unlock the page. | |
6483 | */ | |
c2ec175c | 6484 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6485 | { |
c2ec175c | 6486 | struct page *page = vmf->page; |
6da6abae | 6487 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6488 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6489 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6490 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6491 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6492 | char *kaddr; |
6493 | unsigned long zero_start; | |
9ebefb18 | 6494 | loff_t size; |
1832a6d5 | 6495 | int ret; |
a52d9a80 | 6496 | u64 page_start; |
e6dcd2dc | 6497 | u64 page_end; |
9ebefb18 | 6498 | |
0ca1f7ce | 6499 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
56a76f82 NP |
6500 | if (ret) { |
6501 | if (ret == -ENOMEM) | |
6502 | ret = VM_FAULT_OOM; | |
6503 | else /* -ENOSPC, -EIO, etc */ | |
6504 | ret = VM_FAULT_SIGBUS; | |
1832a6d5 | 6505 | goto out; |
56a76f82 | 6506 | } |
1832a6d5 | 6507 | |
56a76f82 | 6508 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6509 | again: |
9ebefb18 | 6510 | lock_page(page); |
9ebefb18 | 6511 | size = i_size_read(inode); |
e6dcd2dc CM |
6512 | page_start = page_offset(page); |
6513 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6514 | |
9ebefb18 | 6515 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6516 | (page_start >= size)) { |
9ebefb18 CM |
6517 | /* page got truncated out from underneath us */ |
6518 | goto out_unlock; | |
6519 | } | |
e6dcd2dc CM |
6520 | wait_on_page_writeback(page); |
6521 | ||
2ac55d41 JB |
6522 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6523 | GFP_NOFS); | |
e6dcd2dc CM |
6524 | set_page_extent_mapped(page); |
6525 | ||
eb84ae03 CM |
6526 | /* |
6527 | * we can't set the delalloc bits if there are pending ordered | |
6528 | * extents. Drop our locks and wait for them to finish | |
6529 | */ | |
e6dcd2dc CM |
6530 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6531 | if (ordered) { | |
2ac55d41 JB |
6532 | unlock_extent_cached(io_tree, page_start, page_end, |
6533 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6534 | unlock_page(page); |
eb84ae03 | 6535 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6536 | btrfs_put_ordered_extent(ordered); |
6537 | goto again; | |
6538 | } | |
6539 | ||
fbf19087 JB |
6540 | /* |
6541 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6542 | * if it was already dirty, so for space accounting reasons we need to | |
6543 | * clear any delalloc bits for the range we are fixing to save. There | |
6544 | * is probably a better way to do this, but for now keep consistent with | |
6545 | * prepare_pages in the normal write path. | |
6546 | */ | |
2ac55d41 | 6547 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6548 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6549 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6550 | |
2ac55d41 JB |
6551 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6552 | &cached_state); | |
9ed74f2d | 6553 | if (ret) { |
2ac55d41 JB |
6554 | unlock_extent_cached(io_tree, page_start, page_end, |
6555 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6556 | ret = VM_FAULT_SIGBUS; |
6557 | goto out_unlock; | |
6558 | } | |
e6dcd2dc | 6559 | ret = 0; |
9ebefb18 CM |
6560 | |
6561 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6562 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6563 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6564 | else |
e6dcd2dc | 6565 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6566 | |
e6dcd2dc CM |
6567 | if (zero_start != PAGE_CACHE_SIZE) { |
6568 | kaddr = kmap(page); | |
6569 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6570 | flush_dcache_page(page); | |
6571 | kunmap(page); | |
6572 | } | |
247e743c | 6573 | ClearPageChecked(page); |
e6dcd2dc | 6574 | set_page_dirty(page); |
50a9b214 | 6575 | SetPageUptodate(page); |
5a3f23d5 | 6576 | |
257c62e1 CM |
6577 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6578 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6579 | ||
2ac55d41 | 6580 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6581 | |
6582 | out_unlock: | |
50a9b214 CM |
6583 | if (!ret) |
6584 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6585 | unlock_page(page); |
0ca1f7ce | 6586 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
1832a6d5 | 6587 | out: |
9ebefb18 CM |
6588 | return ret; |
6589 | } | |
6590 | ||
a41ad394 | 6591 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6592 | { |
6593 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6594 | int ret; | |
3893e33b | 6595 | int err = 0; |
39279cc3 | 6596 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6597 | unsigned long nr; |
dbe674a9 | 6598 | u64 mask = root->sectorsize - 1; |
39279cc3 | 6599 | |
5d5e103a JB |
6600 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6601 | if (ret) | |
a41ad394 | 6602 | return ret; |
8082510e | 6603 | |
4a096752 | 6604 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6605 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6606 | |
f0cd846e JB |
6607 | trans = btrfs_start_transaction(root, 5); |
6608 | if (IS_ERR(trans)) | |
6609 | return PTR_ERR(trans); | |
6610 | ||
6611 | btrfs_set_trans_block_group(trans, inode); | |
6612 | ||
6613 | ret = btrfs_orphan_add(trans, inode); | |
6614 | if (ret) { | |
6615 | btrfs_end_transaction(trans, root); | |
6616 | return ret; | |
6617 | } | |
6618 | ||
6619 | nr = trans->blocks_used; | |
6620 | btrfs_end_transaction(trans, root); | |
6621 | btrfs_btree_balance_dirty(root, nr); | |
6622 | ||
6623 | /* Now start a transaction for the truncate */ | |
d68fc57b | 6624 | trans = btrfs_start_transaction(root, 0); |
3893e33b JB |
6625 | if (IS_ERR(trans)) |
6626 | return PTR_ERR(trans); | |
8082510e | 6627 | btrfs_set_trans_block_group(trans, inode); |
d68fc57b | 6628 | trans->block_rsv = root->orphan_block_rsv; |
5a3f23d5 CM |
6629 | |
6630 | /* | |
6631 | * setattr is responsible for setting the ordered_data_close flag, | |
6632 | * but that is only tested during the last file release. That | |
6633 | * could happen well after the next commit, leaving a great big | |
6634 | * window where new writes may get lost if someone chooses to write | |
6635 | * to this file after truncating to zero | |
6636 | * | |
6637 | * The inode doesn't have any dirty data here, and so if we commit | |
6638 | * this is a noop. If someone immediately starts writing to the inode | |
6639 | * it is very likely we'll catch some of their writes in this | |
6640 | * transaction, and the commit will find this file on the ordered | |
6641 | * data list with good things to send down. | |
6642 | * | |
6643 | * This is a best effort solution, there is still a window where | |
6644 | * using truncate to replace the contents of the file will | |
6645 | * end up with a zero length file after a crash. | |
6646 | */ | |
6647 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6648 | btrfs_add_ordered_operation(trans, root, inode); | |
6649 | ||
8082510e | 6650 | while (1) { |
d68fc57b YZ |
6651 | if (!trans) { |
6652 | trans = btrfs_start_transaction(root, 0); | |
3893e33b JB |
6653 | if (IS_ERR(trans)) |
6654 | return PTR_ERR(trans); | |
d68fc57b YZ |
6655 | btrfs_set_trans_block_group(trans, inode); |
6656 | trans->block_rsv = root->orphan_block_rsv; | |
6657 | } | |
6658 | ||
6659 | ret = btrfs_block_rsv_check(trans, root, | |
6660 | root->orphan_block_rsv, 0, 5); | |
3893e33b | 6661 | if (ret == -EAGAIN) { |
d68fc57b | 6662 | ret = btrfs_commit_transaction(trans, root); |
3893e33b JB |
6663 | if (ret) |
6664 | return ret; | |
d68fc57b YZ |
6665 | trans = NULL; |
6666 | continue; | |
3893e33b JB |
6667 | } else if (ret) { |
6668 | err = ret; | |
6669 | break; | |
d68fc57b YZ |
6670 | } |
6671 | ||
8082510e YZ |
6672 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6673 | inode->i_size, | |
6674 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6675 | if (ret != -EAGAIN) { |
6676 | err = ret; | |
8082510e | 6677 | break; |
3893e33b | 6678 | } |
39279cc3 | 6679 | |
8082510e | 6680 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6681 | if (ret) { |
6682 | err = ret; | |
6683 | break; | |
6684 | } | |
5f39d397 | 6685 | |
8082510e YZ |
6686 | nr = trans->blocks_used; |
6687 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6688 | trans = NULL; |
8082510e | 6689 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6690 | } |
6691 | ||
6692 | if (ret == 0 && inode->i_nlink > 0) { | |
6693 | ret = btrfs_orphan_del(trans, inode); | |
3893e33b JB |
6694 | if (ret) |
6695 | err = ret; | |
ded5db9d JB |
6696 | } else if (ret && inode->i_nlink > 0) { |
6697 | /* | |
6698 | * Failed to do the truncate, remove us from the in memory | |
6699 | * orphan list. | |
6700 | */ | |
6701 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6702 | } |
6703 | ||
6704 | ret = btrfs_update_inode(trans, root, inode); | |
3893e33b JB |
6705 | if (ret && !err) |
6706 | err = ret; | |
7b128766 | 6707 | |
7b128766 | 6708 | nr = trans->blocks_used; |
89ce8a63 | 6709 | ret = btrfs_end_transaction_throttle(trans, root); |
3893e33b JB |
6710 | if (ret && !err) |
6711 | err = ret; | |
d3c2fdcf | 6712 | btrfs_btree_balance_dirty(root, nr); |
a41ad394 | 6713 | |
3893e33b | 6714 | return err; |
39279cc3 CM |
6715 | } |
6716 | ||
d352ac68 CM |
6717 | /* |
6718 | * create a new subvolume directory/inode (helper for the ioctl). | |
6719 | */ | |
d2fb3437 | 6720 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
76dda93c | 6721 | struct btrfs_root *new_root, |
d2fb3437 | 6722 | u64 new_dirid, u64 alloc_hint) |
39279cc3 | 6723 | { |
39279cc3 | 6724 | struct inode *inode; |
76dda93c | 6725 | int err; |
00e4e6b3 | 6726 | u64 index = 0; |
39279cc3 | 6727 | |
aec7477b | 6728 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid, |
d2fb3437 | 6729 | new_dirid, alloc_hint, S_IFDIR | 0700, &index); |
54aa1f4d | 6730 | if (IS_ERR(inode)) |
f46b5a66 | 6731 | return PTR_ERR(inode); |
39279cc3 CM |
6732 | inode->i_op = &btrfs_dir_inode_operations; |
6733 | inode->i_fop = &btrfs_dir_file_operations; | |
6734 | ||
39279cc3 | 6735 | inode->i_nlink = 1; |
dbe674a9 | 6736 | btrfs_i_size_write(inode, 0); |
3b96362c | 6737 | |
76dda93c YZ |
6738 | err = btrfs_update_inode(trans, new_root, inode); |
6739 | BUG_ON(err); | |
cb8e7090 | 6740 | |
76dda93c | 6741 | iput(inode); |
cb8e7090 | 6742 | return 0; |
39279cc3 CM |
6743 | } |
6744 | ||
d352ac68 CM |
6745 | /* helper function for file defrag and space balancing. This |
6746 | * forces readahead on a given range of bytes in an inode | |
6747 | */ | |
edbd8d4e | 6748 | unsigned long btrfs_force_ra(struct address_space *mapping, |
86479a04 CM |
6749 | struct file_ra_state *ra, struct file *file, |
6750 | pgoff_t offset, pgoff_t last_index) | |
6751 | { | |
8e7bf94f | 6752 | pgoff_t req_size = last_index - offset + 1; |
86479a04 | 6753 | |
86479a04 CM |
6754 | page_cache_sync_readahead(mapping, ra, file, offset, req_size); |
6755 | return offset + req_size; | |
86479a04 CM |
6756 | } |
6757 | ||
39279cc3 CM |
6758 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6759 | { | |
6760 | struct btrfs_inode *ei; | |
2ead6ae7 | 6761 | struct inode *inode; |
39279cc3 CM |
6762 | |
6763 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6764 | if (!ei) | |
6765 | return NULL; | |
2ead6ae7 YZ |
6766 | |
6767 | ei->root = NULL; | |
6768 | ei->space_info = NULL; | |
6769 | ei->generation = 0; | |
6770 | ei->sequence = 0; | |
15ee9bc7 | 6771 | ei->last_trans = 0; |
257c62e1 | 6772 | ei->last_sub_trans = 0; |
e02119d5 | 6773 | ei->logged_trans = 0; |
2ead6ae7 YZ |
6774 | ei->delalloc_bytes = 0; |
6775 | ei->reserved_bytes = 0; | |
6776 | ei->disk_i_size = 0; | |
6777 | ei->flags = 0; | |
6778 | ei->index_cnt = (u64)-1; | |
6779 | ei->last_unlink_trans = 0; | |
6780 | ||
0ca1f7ce | 6781 | atomic_set(&ei->outstanding_extents, 0); |
57a45ced | 6782 | atomic_set(&ei->reserved_extents, 0); |
2ead6ae7 YZ |
6783 | |
6784 | ei->ordered_data_close = 0; | |
d68fc57b | 6785 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6786 | ei->dummy_inode = 0; |
261507a0 | 6787 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 YZ |
6788 | |
6789 | inode = &ei->vfs_inode; | |
6790 | extent_map_tree_init(&ei->extent_tree, GFP_NOFS); | |
6791 | extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS); | |
6792 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS); | |
6793 | mutex_init(&ei->log_mutex); | |
e6dcd2dc | 6794 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6795 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6796 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6797 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6798 | RB_CLEAR_NODE(&ei->rb_node); |
6799 | ||
6800 | return inode; | |
39279cc3 CM |
6801 | } |
6802 | ||
fa0d7e3d NP |
6803 | static void btrfs_i_callback(struct rcu_head *head) |
6804 | { | |
6805 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6806 | INIT_LIST_HEAD(&inode->i_dentry); | |
6807 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6808 | } | |
6809 | ||
39279cc3 CM |
6810 | void btrfs_destroy_inode(struct inode *inode) |
6811 | { | |
e6dcd2dc | 6812 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6813 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6814 | ||
39279cc3 CM |
6815 | WARN_ON(!list_empty(&inode->i_dentry)); |
6816 | WARN_ON(inode->i_data.nrpages); | |
0ca1f7ce | 6817 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents)); |
57a45ced | 6818 | WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents)); |
39279cc3 | 6819 | |
a6dbd429 JB |
6820 | /* |
6821 | * This can happen where we create an inode, but somebody else also | |
6822 | * created the same inode and we need to destroy the one we already | |
6823 | * created. | |
6824 | */ | |
6825 | if (!root) | |
6826 | goto free; | |
6827 | ||
5a3f23d5 CM |
6828 | /* |
6829 | * Make sure we're properly removed from the ordered operation | |
6830 | * lists. | |
6831 | */ | |
6832 | smp_mb(); | |
6833 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6834 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6835 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6836 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6837 | } | |
6838 | ||
0af3d00b JB |
6839 | if (root == root->fs_info->tree_root) { |
6840 | struct btrfs_block_group_cache *block_group; | |
6841 | ||
6842 | block_group = btrfs_lookup_block_group(root->fs_info, | |
6843 | BTRFS_I(inode)->block_group); | |
6844 | if (block_group && block_group->inode == inode) { | |
6845 | spin_lock(&block_group->lock); | |
6846 | block_group->inode = NULL; | |
6847 | spin_unlock(&block_group->lock); | |
6848 | btrfs_put_block_group(block_group); | |
6849 | } else if (block_group) { | |
6850 | btrfs_put_block_group(block_group); | |
6851 | } | |
6852 | } | |
6853 | ||
d68fc57b | 6854 | spin_lock(&root->orphan_lock); |
7b128766 | 6855 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
8082510e YZ |
6856 | printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n", |
6857 | inode->i_ino); | |
6858 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
7b128766 | 6859 | } |
d68fc57b | 6860 | spin_unlock(&root->orphan_lock); |
7b128766 | 6861 | |
d397712b | 6862 | while (1) { |
e6dcd2dc CM |
6863 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6864 | if (!ordered) | |
6865 | break; | |
6866 | else { | |
d397712b CM |
6867 | printk(KERN_ERR "btrfs found ordered " |
6868 | "extent %llu %llu on inode cleanup\n", | |
6869 | (unsigned long long)ordered->file_offset, | |
6870 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6871 | btrfs_remove_ordered_extent(inode, ordered); |
6872 | btrfs_put_ordered_extent(ordered); | |
6873 | btrfs_put_ordered_extent(ordered); | |
6874 | } | |
6875 | } | |
5d4f98a2 | 6876 | inode_tree_del(inode); |
5b21f2ed | 6877 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6878 | free: |
fa0d7e3d | 6879 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6880 | } |
6881 | ||
45321ac5 | 6882 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6883 | { |
6884 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6885 | |
0af3d00b JB |
6886 | if (btrfs_root_refs(&root->root_item) == 0 && |
6887 | root != root->fs_info->tree_root) | |
45321ac5 | 6888 | return 1; |
76dda93c | 6889 | else |
45321ac5 | 6890 | return generic_drop_inode(inode); |
76dda93c YZ |
6891 | } |
6892 | ||
0ee0fda0 | 6893 | static void init_once(void *foo) |
39279cc3 CM |
6894 | { |
6895 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6896 | ||
6897 | inode_init_once(&ei->vfs_inode); | |
6898 | } | |
6899 | ||
6900 | void btrfs_destroy_cachep(void) | |
6901 | { | |
6902 | if (btrfs_inode_cachep) | |
6903 | kmem_cache_destroy(btrfs_inode_cachep); | |
6904 | if (btrfs_trans_handle_cachep) | |
6905 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6906 | if (btrfs_transaction_cachep) | |
6907 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6908 | if (btrfs_path_cachep) |
6909 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6910 | if (btrfs_free_space_cachep) |
6911 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6912 | } |
6913 | ||
6914 | int btrfs_init_cachep(void) | |
6915 | { | |
9601e3f6 CH |
6916 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6917 | sizeof(struct btrfs_inode), 0, | |
6918 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6919 | if (!btrfs_inode_cachep) |
6920 | goto fail; | |
9601e3f6 CH |
6921 | |
6922 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6923 | sizeof(struct btrfs_trans_handle), 0, | |
6924 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6925 | if (!btrfs_trans_handle_cachep) |
6926 | goto fail; | |
9601e3f6 CH |
6927 | |
6928 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6929 | sizeof(struct btrfs_transaction), 0, | |
6930 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6931 | if (!btrfs_transaction_cachep) |
6932 | goto fail; | |
9601e3f6 CH |
6933 | |
6934 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6935 | sizeof(struct btrfs_path), 0, | |
6936 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6937 | if (!btrfs_path_cachep) |
6938 | goto fail; | |
9601e3f6 | 6939 | |
dc89e982 JB |
6940 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6941 | sizeof(struct btrfs_free_space), 0, | |
6942 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6943 | if (!btrfs_free_space_cachep) | |
6944 | goto fail; | |
6945 | ||
39279cc3 CM |
6946 | return 0; |
6947 | fail: | |
6948 | btrfs_destroy_cachep(); | |
6949 | return -ENOMEM; | |
6950 | } | |
6951 | ||
6952 | static int btrfs_getattr(struct vfsmount *mnt, | |
6953 | struct dentry *dentry, struct kstat *stat) | |
6954 | { | |
6955 | struct inode *inode = dentry->d_inode; | |
6956 | generic_fillattr(inode, stat); | |
3394e160 | 6957 | stat->dev = BTRFS_I(inode)->root->anon_super.s_dev; |
d6667462 | 6958 | stat->blksize = PAGE_CACHE_SIZE; |
a76a3cd4 YZ |
6959 | stat->blocks = (inode_get_bytes(inode) + |
6960 | BTRFS_I(inode)->delalloc_bytes) >> 9; | |
39279cc3 CM |
6961 | return 0; |
6962 | } | |
6963 | ||
75e7cb7f LB |
6964 | /* |
6965 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6966 | * directory. | |
6967 | */ | |
6968 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6969 | { | |
6970 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6971 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6972 | ||
6973 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6974 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6975 | else | |
6976 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6977 | ||
6978 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6979 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6980 | else | |
6981 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6982 | } | |
6983 | ||
d397712b CM |
6984 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6985 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6986 | { |
6987 | struct btrfs_trans_handle *trans; | |
6988 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6989 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6990 | struct inode *new_inode = new_dentry->d_inode; |
6991 | struct inode *old_inode = old_dentry->d_inode; | |
6992 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6993 | u64 index = 0; |
4df27c4d | 6994 | u64 root_objectid; |
39279cc3 CM |
6995 | int ret; |
6996 | ||
f679a840 YZ |
6997 | if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
6998 | return -EPERM; | |
6999 | ||
4df27c4d YZ |
7000 | /* we only allow rename subvolume link between subvolumes */ |
7001 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
3394e160 CM |
7002 | return -EXDEV; |
7003 | ||
4df27c4d YZ |
7004 | if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7005 | (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7006 | return -ENOTEMPTY; |
5f39d397 | 7007 | |
4df27c4d YZ |
7008 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7009 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7010 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7011 | /* |
7012 | * we're using rename to replace one file with another. | |
7013 | * and the replacement file is large. Start IO on it now so | |
7014 | * we don't add too much work to the end of the transaction | |
7015 | */ | |
4baf8c92 | 7016 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7017 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7018 | filemap_flush(old_inode->i_mapping); | |
7019 | ||
76dda93c YZ |
7020 | /* close the racy window with snapshot create/destroy ioctl */ |
7021 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) | |
7022 | down_read(&root->fs_info->subvol_sem); | |
a22285a6 YZ |
7023 | /* |
7024 | * We want to reserve the absolute worst case amount of items. So if | |
7025 | * both inodes are subvols and we need to unlink them then that would | |
7026 | * require 4 item modifications, but if they are both normal inodes it | |
7027 | * would require 5 item modifications, so we'll assume their normal | |
7028 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7029 | * should cover the worst case number of items we'll modify. | |
7030 | */ | |
7031 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7032 | if (IS_ERR(trans)) { |
7033 | ret = PTR_ERR(trans); | |
7034 | goto out_notrans; | |
7035 | } | |
76dda93c | 7036 | |
a5719521 | 7037 | btrfs_set_trans_block_group(trans, new_dir); |
5f39d397 | 7038 | |
4df27c4d YZ |
7039 | if (dest != root) |
7040 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7041 | |
a5719521 YZ |
7042 | ret = btrfs_set_inode_index(new_dir, &index); |
7043 | if (ret) | |
7044 | goto out_fail; | |
5a3f23d5 | 7045 | |
a5719521 | 7046 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7047 | /* force full log commit if subvolume involved. */ |
7048 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7049 | } else { | |
a5719521 YZ |
7050 | ret = btrfs_insert_inode_ref(trans, dest, |
7051 | new_dentry->d_name.name, | |
7052 | new_dentry->d_name.len, | |
7053 | old_inode->i_ino, | |
7054 | new_dir->i_ino, index); | |
7055 | if (ret) | |
7056 | goto out_fail; | |
4df27c4d YZ |
7057 | /* |
7058 | * this is an ugly little race, but the rename is required | |
7059 | * to make sure that if we crash, the inode is either at the | |
7060 | * old name or the new one. pinning the log transaction lets | |
7061 | * us make sure we don't allow a log commit to come in after | |
7062 | * we unlink the name but before we add the new name back in. | |
7063 | */ | |
7064 | btrfs_pin_log_trans(root); | |
7065 | } | |
5a3f23d5 CM |
7066 | /* |
7067 | * make sure the inode gets flushed if it is replacing | |
7068 | * something. | |
7069 | */ | |
7070 | if (new_inode && new_inode->i_size && | |
7071 | old_inode && S_ISREG(old_inode->i_mode)) { | |
7072 | btrfs_add_ordered_operation(trans, root, old_inode); | |
7073 | } | |
7074 | ||
39279cc3 CM |
7075 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7076 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7077 | old_inode->i_ctime = ctime; | |
5f39d397 | 7078 | |
12fcfd22 CM |
7079 | if (old_dentry->d_parent != new_dentry->d_parent) |
7080 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7081 | ||
4df27c4d YZ |
7082 | if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
7083 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
7084 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7085 | old_dentry->d_name.name, | |
7086 | old_dentry->d_name.len); | |
7087 | } else { | |
92986796 AV |
7088 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7089 | old_dentry->d_inode, | |
7090 | old_dentry->d_name.name, | |
7091 | old_dentry->d_name.len); | |
7092 | if (!ret) | |
7093 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
7094 | } |
7095 | BUG_ON(ret); | |
39279cc3 CM |
7096 | |
7097 | if (new_inode) { | |
7098 | new_inode->i_ctime = CURRENT_TIME; | |
4df27c4d YZ |
7099 | if (unlikely(new_inode->i_ino == |
7100 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { | |
7101 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7102 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7103 | root_objectid, | |
7104 | new_dentry->d_name.name, | |
7105 | new_dentry->d_name.len); | |
7106 | BUG_ON(new_inode->i_nlink == 0); | |
7107 | } else { | |
7108 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7109 | new_dentry->d_inode, | |
7110 | new_dentry->d_name.name, | |
7111 | new_dentry->d_name.len); | |
7112 | } | |
7113 | BUG_ON(ret); | |
7b128766 | 7114 | if (new_inode->i_nlink == 0) { |
e02119d5 | 7115 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7116 | BUG_ON(ret); |
7b128766 | 7117 | } |
39279cc3 | 7118 | } |
aec7477b | 7119 | |
75e7cb7f LB |
7120 | fixup_inode_flags(new_dir, old_inode); |
7121 | ||
4df27c4d YZ |
7122 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7123 | new_dentry->d_name.name, | |
a5719521 | 7124 | new_dentry->d_name.len, 0, index); |
4df27c4d | 7125 | BUG_ON(ret); |
39279cc3 | 7126 | |
4df27c4d | 7127 | if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { |
6a912213 JB |
7128 | struct dentry *parent = dget_parent(new_dentry); |
7129 | btrfs_log_new_name(trans, old_inode, old_dir, parent); | |
7130 | dput(parent); | |
4df27c4d YZ |
7131 | btrfs_end_log_trans(root); |
7132 | } | |
39279cc3 | 7133 | out_fail: |
ab78c84d | 7134 | btrfs_end_transaction_throttle(trans, root); |
b44c59a8 | 7135 | out_notrans: |
76dda93c YZ |
7136 | if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) |
7137 | up_read(&root->fs_info->subvol_sem); | |
9ed74f2d | 7138 | |
39279cc3 CM |
7139 | return ret; |
7140 | } | |
7141 | ||
d352ac68 CM |
7142 | /* |
7143 | * some fairly slow code that needs optimization. This walks the list | |
7144 | * of all the inodes with pending delalloc and forces them to disk. | |
7145 | */ | |
24bbcf04 | 7146 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7147 | { |
7148 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7149 | struct btrfs_inode *binode; | |
5b21f2ed | 7150 | struct inode *inode; |
ea8c2819 | 7151 | |
c146afad YZ |
7152 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7153 | return -EROFS; | |
7154 | ||
75eff68e | 7155 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7156 | while (!list_empty(head)) { |
ea8c2819 CM |
7157 | binode = list_entry(head->next, struct btrfs_inode, |
7158 | delalloc_inodes); | |
5b21f2ed ZY |
7159 | inode = igrab(&binode->vfs_inode); |
7160 | if (!inode) | |
7161 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7162 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7163 | if (inode) { |
8c8bee1d | 7164 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7165 | if (delay_iput) |
7166 | btrfs_add_delayed_iput(inode); | |
7167 | else | |
7168 | iput(inode); | |
5b21f2ed ZY |
7169 | } |
7170 | cond_resched(); | |
75eff68e | 7171 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7172 | } |
75eff68e | 7173 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7174 | |
7175 | /* the filemap_flush will queue IO into the worker threads, but | |
7176 | * we have to make sure the IO is actually started and that | |
7177 | * ordered extents get created before we return | |
7178 | */ | |
7179 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7180 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7181 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7182 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7183 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7184 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7185 | } |
7186 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7187 | return 0; |
7188 | } | |
7189 | ||
0019f10d JB |
7190 | int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput, |
7191 | int sync) | |
5da9d01b YZ |
7192 | { |
7193 | struct btrfs_inode *binode; | |
7194 | struct inode *inode = NULL; | |
7195 | ||
7196 | spin_lock(&root->fs_info->delalloc_lock); | |
7197 | while (!list_empty(&root->fs_info->delalloc_inodes)) { | |
7198 | binode = list_entry(root->fs_info->delalloc_inodes.next, | |
7199 | struct btrfs_inode, delalloc_inodes); | |
7200 | inode = igrab(&binode->vfs_inode); | |
7201 | if (inode) { | |
7202 | list_move_tail(&binode->delalloc_inodes, | |
7203 | &root->fs_info->delalloc_inodes); | |
7204 | break; | |
7205 | } | |
7206 | ||
7207 | list_del_init(&binode->delalloc_inodes); | |
7208 | cond_resched_lock(&root->fs_info->delalloc_lock); | |
7209 | } | |
7210 | spin_unlock(&root->fs_info->delalloc_lock); | |
7211 | ||
7212 | if (inode) { | |
0019f10d JB |
7213 | if (sync) { |
7214 | filemap_write_and_wait(inode->i_mapping); | |
7215 | /* | |
7216 | * We have to do this because compression doesn't | |
7217 | * actually set PG_writeback until it submits the pages | |
7218 | * for IO, which happens in an async thread, so we could | |
7219 | * race and not actually wait for any writeback pages | |
7220 | * because they've not been submitted yet. Technically | |
7221 | * this could still be the case for the ordered stuff | |
7222 | * since the async thread may not have started to do its | |
7223 | * work yet. If this becomes the case then we need to | |
7224 | * figure out a way to make sure that in writepage we | |
7225 | * wait for any async pages to be submitted before | |
7226 | * returning so that fdatawait does what its supposed to | |
7227 | * do. | |
7228 | */ | |
7229 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
7230 | } else { | |
7231 | filemap_flush(inode->i_mapping); | |
7232 | } | |
5da9d01b YZ |
7233 | if (delay_iput) |
7234 | btrfs_add_delayed_iput(inode); | |
7235 | else | |
7236 | iput(inode); | |
7237 | return 1; | |
7238 | } | |
7239 | return 0; | |
7240 | } | |
7241 | ||
39279cc3 CM |
7242 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7243 | const char *symname) | |
7244 | { | |
7245 | struct btrfs_trans_handle *trans; | |
7246 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7247 | struct btrfs_path *path; | |
7248 | struct btrfs_key key; | |
1832a6d5 | 7249 | struct inode *inode = NULL; |
39279cc3 CM |
7250 | int err; |
7251 | int drop_inode = 0; | |
7252 | u64 objectid; | |
00e4e6b3 | 7253 | u64 index = 0 ; |
39279cc3 CM |
7254 | int name_len; |
7255 | int datasize; | |
5f39d397 | 7256 | unsigned long ptr; |
39279cc3 | 7257 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7258 | struct extent_buffer *leaf; |
1832a6d5 | 7259 | unsigned long nr = 0; |
39279cc3 CM |
7260 | |
7261 | name_len = strlen(symname) + 1; | |
7262 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7263 | return -ENAMETOOLONG; | |
1832a6d5 | 7264 | |
a22285a6 YZ |
7265 | err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid); |
7266 | if (err) | |
7267 | return err; | |
9ed74f2d JB |
7268 | /* |
7269 | * 2 items for inode item and ref | |
7270 | * 2 items for dir items | |
7271 | * 1 item for xattr if selinux is on | |
7272 | */ | |
a22285a6 YZ |
7273 | trans = btrfs_start_transaction(root, 5); |
7274 | if (IS_ERR(trans)) | |
7275 | return PTR_ERR(trans); | |
1832a6d5 | 7276 | |
39279cc3 CM |
7277 | btrfs_set_trans_block_group(trans, dir); |
7278 | ||
aec7477b | 7279 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
a1b075d2 | 7280 | dentry->d_name.len, dir->i_ino, objectid, |
00e4e6b3 CM |
7281 | BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO, |
7282 | &index); | |
7cf96da3 TI |
7283 | if (IS_ERR(inode)) { |
7284 | err = PTR_ERR(inode); | |
39279cc3 | 7285 | goto out_unlock; |
7cf96da3 | 7286 | } |
39279cc3 | 7287 | |
f34f57a3 | 7288 | err = btrfs_init_inode_security(trans, inode, dir); |
33268eaf JB |
7289 | if (err) { |
7290 | drop_inode = 1; | |
7291 | goto out_unlock; | |
7292 | } | |
7293 | ||
39279cc3 | 7294 | btrfs_set_trans_block_group(trans, inode); |
a1b075d2 | 7295 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7296 | if (err) |
7297 | drop_inode = 1; | |
7298 | else { | |
7299 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7300 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 CM |
7301 | inode->i_fop = &btrfs_file_operations; |
7302 | inode->i_op = &btrfs_file_inode_operations; | |
d1310b2e | 7303 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7304 | } |
39279cc3 CM |
7305 | btrfs_update_inode_block_group(trans, inode); |
7306 | btrfs_update_inode_block_group(trans, dir); | |
7307 | if (drop_inode) | |
7308 | goto out_unlock; | |
7309 | ||
7310 | path = btrfs_alloc_path(); | |
7311 | BUG_ON(!path); | |
7312 | key.objectid = inode->i_ino; | |
7313 | key.offset = 0; | |
39279cc3 CM |
7314 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7315 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7316 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7317 | datasize); | |
54aa1f4d CM |
7318 | if (err) { |
7319 | drop_inode = 1; | |
7320 | goto out_unlock; | |
7321 | } | |
5f39d397 CM |
7322 | leaf = path->nodes[0]; |
7323 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7324 | struct btrfs_file_extent_item); | |
7325 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7326 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7327 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7328 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7329 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7330 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7331 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7332 | ||
39279cc3 | 7333 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7334 | write_extent_buffer(leaf, symname, ptr, name_len); |
7335 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7336 | btrfs_free_path(path); |
5f39d397 | 7337 | |
39279cc3 CM |
7338 | inode->i_op = &btrfs_symlink_inode_operations; |
7339 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7340 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7341 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7342 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7343 | err = btrfs_update_inode(trans, root, inode); |
7344 | if (err) | |
7345 | drop_inode = 1; | |
39279cc3 CM |
7346 | |
7347 | out_unlock: | |
d3c2fdcf | 7348 | nr = trans->blocks_used; |
ab78c84d | 7349 | btrfs_end_transaction_throttle(trans, root); |
39279cc3 CM |
7350 | if (drop_inode) { |
7351 | inode_dec_link_count(inode); | |
7352 | iput(inode); | |
7353 | } | |
d3c2fdcf | 7354 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7355 | return err; |
7356 | } | |
16432985 | 7357 | |
0af3d00b JB |
7358 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7359 | u64 start, u64 num_bytes, u64 min_size, | |
7360 | loff_t actual_len, u64 *alloc_hint, | |
7361 | struct btrfs_trans_handle *trans) | |
d899e052 | 7362 | { |
d899e052 YZ |
7363 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7364 | struct btrfs_key ins; | |
d899e052 | 7365 | u64 cur_offset = start; |
55a61d1d | 7366 | u64 i_size; |
d899e052 | 7367 | int ret = 0; |
0af3d00b | 7368 | bool own_trans = true; |
d899e052 | 7369 | |
0af3d00b JB |
7370 | if (trans) |
7371 | own_trans = false; | |
d899e052 | 7372 | while (num_bytes > 0) { |
0af3d00b JB |
7373 | if (own_trans) { |
7374 | trans = btrfs_start_transaction(root, 3); | |
7375 | if (IS_ERR(trans)) { | |
7376 | ret = PTR_ERR(trans); | |
7377 | break; | |
7378 | } | |
5a303d5d YZ |
7379 | } |
7380 | ||
efa56464 YZ |
7381 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7382 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7383 | if (ret) { |
0af3d00b JB |
7384 | if (own_trans) |
7385 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7386 | break; |
d899e052 | 7387 | } |
5a303d5d | 7388 | |
d899e052 YZ |
7389 | ret = insert_reserved_file_extent(trans, inode, |
7390 | cur_offset, ins.objectid, | |
7391 | ins.offset, ins.offset, | |
920bbbfb | 7392 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7393 | BTRFS_FILE_EXTENT_PREALLOC); |
7394 | BUG_ON(ret); | |
a1ed835e CM |
7395 | btrfs_drop_extent_cache(inode, cur_offset, |
7396 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7397 | |
d899e052 YZ |
7398 | num_bytes -= ins.offset; |
7399 | cur_offset += ins.offset; | |
efa56464 | 7400 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7401 | |
d899e052 | 7402 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7403 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7404 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7405 | (actual_len > inode->i_size) && |
7406 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7407 | if (cur_offset > actual_len) |
55a61d1d | 7408 | i_size = actual_len; |
d1ea6a61 | 7409 | else |
55a61d1d JB |
7410 | i_size = cur_offset; |
7411 | i_size_write(inode, i_size); | |
7412 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7413 | } |
7414 | ||
d899e052 YZ |
7415 | ret = btrfs_update_inode(trans, root, inode); |
7416 | BUG_ON(ret); | |
d899e052 | 7417 | |
0af3d00b JB |
7418 | if (own_trans) |
7419 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7420 | } |
d899e052 YZ |
7421 | return ret; |
7422 | } | |
7423 | ||
0af3d00b JB |
7424 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7425 | u64 start, u64 num_bytes, u64 min_size, | |
7426 | loff_t actual_len, u64 *alloc_hint) | |
7427 | { | |
7428 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7429 | min_size, actual_len, alloc_hint, | |
7430 | NULL); | |
7431 | } | |
7432 | ||
7433 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7434 | struct btrfs_trans_handle *trans, int mode, | |
7435 | u64 start, u64 num_bytes, u64 min_size, | |
7436 | loff_t actual_len, u64 *alloc_hint) | |
7437 | { | |
7438 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7439 | min_size, actual_len, alloc_hint, trans); | |
7440 | } | |
7441 | ||
e6dcd2dc CM |
7442 | static int btrfs_set_page_dirty(struct page *page) |
7443 | { | |
e6dcd2dc CM |
7444 | return __set_page_dirty_nobuffers(page); |
7445 | } | |
7446 | ||
b74c79e9 | 7447 | static int btrfs_permission(struct inode *inode, int mask, unsigned int flags) |
fdebe2bd | 7448 | { |
b83cc969 LZ |
7449 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7450 | ||
7451 | if (btrfs_root_readonly(root) && (mask & MAY_WRITE)) | |
7452 | return -EROFS; | |
6cbff00f | 7453 | if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE)) |
fdebe2bd | 7454 | return -EACCES; |
b74c79e9 | 7455 | return generic_permission(inode, mask, flags, btrfs_check_acl); |
fdebe2bd | 7456 | } |
39279cc3 | 7457 | |
6e1d5dcc | 7458 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7459 | .getattr = btrfs_getattr, |
39279cc3 CM |
7460 | .lookup = btrfs_lookup, |
7461 | .create = btrfs_create, | |
7462 | .unlink = btrfs_unlink, | |
7463 | .link = btrfs_link, | |
7464 | .mkdir = btrfs_mkdir, | |
7465 | .rmdir = btrfs_rmdir, | |
7466 | .rename = btrfs_rename, | |
7467 | .symlink = btrfs_symlink, | |
7468 | .setattr = btrfs_setattr, | |
618e21d5 | 7469 | .mknod = btrfs_mknod, |
95819c05 CH |
7470 | .setxattr = btrfs_setxattr, |
7471 | .getxattr = btrfs_getxattr, | |
5103e947 | 7472 | .listxattr = btrfs_listxattr, |
95819c05 | 7473 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7474 | .permission = btrfs_permission, |
39279cc3 | 7475 | }; |
6e1d5dcc | 7476 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7477 | .lookup = btrfs_lookup, |
fdebe2bd | 7478 | .permission = btrfs_permission, |
39279cc3 | 7479 | }; |
76dda93c | 7480 | |
828c0950 | 7481 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7482 | .llseek = generic_file_llseek, |
7483 | .read = generic_read_dir, | |
cbdf5a24 | 7484 | .readdir = btrfs_real_readdir, |
34287aa3 | 7485 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7486 | #ifdef CONFIG_COMPAT |
34287aa3 | 7487 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7488 | #endif |
6bf13c0c | 7489 | .release = btrfs_release_file, |
e02119d5 | 7490 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7491 | }; |
7492 | ||
d1310b2e | 7493 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7494 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7495 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7496 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7497 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7498 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7499 | .writepage_start_hook = btrfs_writepage_start_hook, |
1259ab75 | 7500 | .readpage_io_failed_hook = btrfs_io_failed_hook, |
b0c68f8b CM |
7501 | .set_bit_hook = btrfs_set_bit_hook, |
7502 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7503 | .merge_extent_hook = btrfs_merge_extent_hook, |
7504 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7505 | }; |
7506 | ||
35054394 CM |
7507 | /* |
7508 | * btrfs doesn't support the bmap operation because swapfiles | |
7509 | * use bmap to make a mapping of extents in the file. They assume | |
7510 | * these extents won't change over the life of the file and they | |
7511 | * use the bmap result to do IO directly to the drive. | |
7512 | * | |
7513 | * the btrfs bmap call would return logical addresses that aren't | |
7514 | * suitable for IO and they also will change frequently as COW | |
7515 | * operations happen. So, swapfile + btrfs == corruption. | |
7516 | * | |
7517 | * For now we're avoiding this by dropping bmap. | |
7518 | */ | |
7f09410b | 7519 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7520 | .readpage = btrfs_readpage, |
7521 | .writepage = btrfs_writepage, | |
b293f02e | 7522 | .writepages = btrfs_writepages, |
3ab2fb5a | 7523 | .readpages = btrfs_readpages, |
39279cc3 | 7524 | .sync_page = block_sync_page, |
16432985 | 7525 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7526 | .invalidatepage = btrfs_invalidatepage, |
7527 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7528 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7529 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7530 | }; |
7531 | ||
7f09410b | 7532 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7533 | .readpage = btrfs_readpage, |
7534 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7535 | .invalidatepage = btrfs_invalidatepage, |
7536 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7537 | }; |
7538 | ||
6e1d5dcc | 7539 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7540 | .getattr = btrfs_getattr, |
7541 | .setattr = btrfs_setattr, | |
95819c05 CH |
7542 | .setxattr = btrfs_setxattr, |
7543 | .getxattr = btrfs_getxattr, | |
5103e947 | 7544 | .listxattr = btrfs_listxattr, |
95819c05 | 7545 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7546 | .permission = btrfs_permission, |
1506fcc8 | 7547 | .fiemap = btrfs_fiemap, |
39279cc3 | 7548 | }; |
6e1d5dcc | 7549 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7550 | .getattr = btrfs_getattr, |
7551 | .setattr = btrfs_setattr, | |
fdebe2bd | 7552 | .permission = btrfs_permission, |
95819c05 CH |
7553 | .setxattr = btrfs_setxattr, |
7554 | .getxattr = btrfs_getxattr, | |
33268eaf | 7555 | .listxattr = btrfs_listxattr, |
95819c05 | 7556 | .removexattr = btrfs_removexattr, |
618e21d5 | 7557 | }; |
6e1d5dcc | 7558 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7559 | .readlink = generic_readlink, |
7560 | .follow_link = page_follow_link_light, | |
7561 | .put_link = page_put_link, | |
f209561a | 7562 | .getattr = btrfs_getattr, |
fdebe2bd | 7563 | .permission = btrfs_permission, |
0279b4cd JO |
7564 | .setxattr = btrfs_setxattr, |
7565 | .getxattr = btrfs_getxattr, | |
7566 | .listxattr = btrfs_listxattr, | |
7567 | .removexattr = btrfs_removexattr, | |
39279cc3 | 7568 | }; |
76dda93c | 7569 | |
82d339d9 | 7570 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c YZ |
7571 | .d_delete = btrfs_dentry_delete, |
7572 | }; |