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> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
4b4e25f2 | 43 | #include "compat.h" |
39279cc3 CM |
44 | #include "ctree.h" |
45 | #include "disk-io.h" | |
46 | #include "transaction.h" | |
47 | #include "btrfs_inode.h" | |
39279cc3 | 48 | #include "print-tree.h" |
e6dcd2dc | 49 | #include "ordered-data.h" |
95819c05 | 50 | #include "xattr.h" |
e02119d5 | 51 | #include "tree-log.h" |
4a54c8c1 | 52 | #include "volumes.h" |
c8b97818 | 53 | #include "compression.h" |
b4ce94de | 54 | #include "locking.h" |
dc89e982 | 55 | #include "free-space-cache.h" |
581bb050 | 56 | #include "inode-map.h" |
39279cc3 CM |
57 | |
58 | struct btrfs_iget_args { | |
59 | u64 ino; | |
60 | struct btrfs_root *root; | |
61 | }; | |
62 | ||
6e1d5dcc AD |
63 | static const struct inode_operations btrfs_dir_inode_operations; |
64 | static const struct inode_operations btrfs_symlink_inode_operations; | |
65 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
66 | static const struct inode_operations btrfs_special_inode_operations; | |
67 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
68 | static const struct address_space_operations btrfs_aops; |
69 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 70 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 71 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
72 | |
73 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 74 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
75 | struct kmem_cache *btrfs_trans_handle_cachep; |
76 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 77 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 78 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
79 | |
80 | #define S_SHIFT 12 | |
81 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
82 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
83 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
84 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
85 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
86 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
87 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
88 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
89 | }; | |
90 | ||
3972f260 | 91 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 92 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 93 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
94 | static noinline int cow_file_range(struct inode *inode, |
95 | struct page *locked_page, | |
96 | u64 start, u64 end, int *page_started, | |
97 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
98 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
99 | u64 len, u64 orig_start, | |
100 | u64 block_start, u64 block_len, | |
101 | u64 orig_block_len, int type); | |
7b128766 | 102 | |
f34f57a3 | 103 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
104 | struct inode *inode, struct inode *dir, |
105 | const struct qstr *qstr) | |
0279b4cd JO |
106 | { |
107 | int err; | |
108 | ||
f34f57a3 | 109 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 110 | if (!err) |
2a7dba39 | 111 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
112 | return err; |
113 | } | |
114 | ||
c8b97818 CM |
115 | /* |
116 | * this does all the hard work for inserting an inline extent into | |
117 | * the btree. The caller should have done a btrfs_drop_extents so that | |
118 | * no overlapping inline items exist in the btree | |
119 | */ | |
d397712b | 120 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
121 | struct btrfs_root *root, struct inode *inode, |
122 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 123 | int compress_type, |
c8b97818 CM |
124 | struct page **compressed_pages) |
125 | { | |
126 | struct btrfs_key key; | |
127 | struct btrfs_path *path; | |
128 | struct extent_buffer *leaf; | |
129 | struct page *page = NULL; | |
130 | char *kaddr; | |
131 | unsigned long ptr; | |
132 | struct btrfs_file_extent_item *ei; | |
133 | int err = 0; | |
134 | int ret; | |
135 | size_t cur_size = size; | |
136 | size_t datasize; | |
137 | unsigned long offset; | |
c8b97818 | 138 | |
fe3f566c | 139 | if (compressed_size && compressed_pages) |
c8b97818 | 140 | cur_size = compressed_size; |
c8b97818 | 141 | |
d397712b CM |
142 | path = btrfs_alloc_path(); |
143 | if (!path) | |
c8b97818 CM |
144 | return -ENOMEM; |
145 | ||
b9473439 | 146 | path->leave_spinning = 1; |
c8b97818 | 147 | |
33345d01 | 148 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
149 | key.offset = start; |
150 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
151 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
152 | ||
153 | inode_add_bytes(inode, size); | |
154 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
155 | datasize); | |
c8b97818 CM |
156 | if (ret) { |
157 | err = ret; | |
c8b97818 CM |
158 | goto fail; |
159 | } | |
160 | leaf = path->nodes[0]; | |
161 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
162 | struct btrfs_file_extent_item); | |
163 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
164 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
165 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
166 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
167 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
168 | ptr = btrfs_file_extent_inline_start(ei); | |
169 | ||
261507a0 | 170 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
171 | struct page *cpage; |
172 | int i = 0; | |
d397712b | 173 | while (compressed_size > 0) { |
c8b97818 | 174 | cpage = compressed_pages[i]; |
5b050f04 | 175 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
176 | PAGE_CACHE_SIZE); |
177 | ||
7ac687d9 | 178 | kaddr = kmap_atomic(cpage); |
c8b97818 | 179 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 180 | kunmap_atomic(kaddr); |
c8b97818 CM |
181 | |
182 | i++; | |
183 | ptr += cur_size; | |
184 | compressed_size -= cur_size; | |
185 | } | |
186 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 187 | compress_type); |
c8b97818 CM |
188 | } else { |
189 | page = find_get_page(inode->i_mapping, | |
190 | start >> PAGE_CACHE_SHIFT); | |
191 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 192 | kaddr = kmap_atomic(page); |
c8b97818 CM |
193 | offset = start & (PAGE_CACHE_SIZE - 1); |
194 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 195 | kunmap_atomic(kaddr); |
c8b97818 CM |
196 | page_cache_release(page); |
197 | } | |
198 | btrfs_mark_buffer_dirty(leaf); | |
199 | btrfs_free_path(path); | |
200 | ||
c2167754 YZ |
201 | /* |
202 | * we're an inline extent, so nobody can | |
203 | * extend the file past i_size without locking | |
204 | * a page we already have locked. | |
205 | * | |
206 | * We must do any isize and inode updates | |
207 | * before we unlock the pages. Otherwise we | |
208 | * could end up racing with unlink. | |
209 | */ | |
c8b97818 | 210 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 211 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 212 | |
79787eaa | 213 | return ret; |
c8b97818 CM |
214 | fail: |
215 | btrfs_free_path(path); | |
216 | return err; | |
217 | } | |
218 | ||
219 | ||
220 | /* | |
221 | * conditionally insert an inline extent into the file. This | |
222 | * does the checks required to make sure the data is small enough | |
223 | * to fit as an inline extent. | |
224 | */ | |
7f366cfe | 225 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
226 | struct btrfs_root *root, |
227 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 228 | size_t compressed_size, int compress_type, |
c8b97818 CM |
229 | struct page **compressed_pages) |
230 | { | |
231 | u64 isize = i_size_read(inode); | |
232 | u64 actual_end = min(end + 1, isize); | |
233 | u64 inline_len = actual_end - start; | |
234 | u64 aligned_end = (end + root->sectorsize - 1) & | |
235 | ~((u64)root->sectorsize - 1); | |
c8b97818 CM |
236 | u64 data_len = inline_len; |
237 | int ret; | |
238 | ||
239 | if (compressed_size) | |
240 | data_len = compressed_size; | |
241 | ||
242 | if (start > 0 || | |
70b99e69 | 243 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
244 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
245 | (!compressed_size && | |
246 | (actual_end & (root->sectorsize - 1)) == 0) || | |
247 | end + 1 < isize || | |
248 | data_len > root->fs_info->max_inline) { | |
249 | return 1; | |
250 | } | |
251 | ||
2671485d | 252 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
253 | if (ret) |
254 | return ret; | |
c8b97818 CM |
255 | |
256 | if (isize > actual_end) | |
257 | inline_len = min_t(u64, isize, actual_end); | |
258 | ret = insert_inline_extent(trans, root, inode, start, | |
259 | inline_len, compressed_size, | |
fe3f566c | 260 | compress_type, compressed_pages); |
2adcac1a | 261 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
262 | btrfs_abort_transaction(trans, root, ret); |
263 | return ret; | |
2adcac1a JB |
264 | } else if (ret == -ENOSPC) { |
265 | return 1; | |
79787eaa | 266 | } |
2adcac1a | 267 | |
0ca1f7ce | 268 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 269 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
270 | return 0; |
271 | } | |
272 | ||
771ed689 CM |
273 | struct async_extent { |
274 | u64 start; | |
275 | u64 ram_size; | |
276 | u64 compressed_size; | |
277 | struct page **pages; | |
278 | unsigned long nr_pages; | |
261507a0 | 279 | int compress_type; |
771ed689 CM |
280 | struct list_head list; |
281 | }; | |
282 | ||
283 | struct async_cow { | |
284 | struct inode *inode; | |
285 | struct btrfs_root *root; | |
286 | struct page *locked_page; | |
287 | u64 start; | |
288 | u64 end; | |
289 | struct list_head extents; | |
290 | struct btrfs_work work; | |
291 | }; | |
292 | ||
293 | static noinline int add_async_extent(struct async_cow *cow, | |
294 | u64 start, u64 ram_size, | |
295 | u64 compressed_size, | |
296 | struct page **pages, | |
261507a0 LZ |
297 | unsigned long nr_pages, |
298 | int compress_type) | |
771ed689 CM |
299 | { |
300 | struct async_extent *async_extent; | |
301 | ||
302 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 303 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
304 | async_extent->start = start; |
305 | async_extent->ram_size = ram_size; | |
306 | async_extent->compressed_size = compressed_size; | |
307 | async_extent->pages = pages; | |
308 | async_extent->nr_pages = nr_pages; | |
261507a0 | 309 | async_extent->compress_type = compress_type; |
771ed689 CM |
310 | list_add_tail(&async_extent->list, &cow->extents); |
311 | return 0; | |
312 | } | |
313 | ||
d352ac68 | 314 | /* |
771ed689 CM |
315 | * we create compressed extents in two phases. The first |
316 | * phase compresses a range of pages that have already been | |
317 | * locked (both pages and state bits are locked). | |
c8b97818 | 318 | * |
771ed689 CM |
319 | * This is done inside an ordered work queue, and the compression |
320 | * is spread across many cpus. The actual IO submission is step | |
321 | * two, and the ordered work queue takes care of making sure that | |
322 | * happens in the same order things were put onto the queue by | |
323 | * writepages and friends. | |
c8b97818 | 324 | * |
771ed689 CM |
325 | * If this code finds it can't get good compression, it puts an |
326 | * entry onto the work queue to write the uncompressed bytes. This | |
327 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
328 | * are written in the same order that the flusher thread sent them |
329 | * down. | |
d352ac68 | 330 | */ |
771ed689 CM |
331 | static noinline int compress_file_range(struct inode *inode, |
332 | struct page *locked_page, | |
333 | u64 start, u64 end, | |
334 | struct async_cow *async_cow, | |
335 | int *num_added) | |
b888db2b CM |
336 | { |
337 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
338 | struct btrfs_trans_handle *trans; | |
db94535d | 339 | u64 num_bytes; |
db94535d | 340 | u64 blocksize = root->sectorsize; |
c8b97818 | 341 | u64 actual_end; |
42dc7bab | 342 | u64 isize = i_size_read(inode); |
e6dcd2dc | 343 | int ret = 0; |
c8b97818 CM |
344 | struct page **pages = NULL; |
345 | unsigned long nr_pages; | |
346 | unsigned long nr_pages_ret = 0; | |
347 | unsigned long total_compressed = 0; | |
348 | unsigned long total_in = 0; | |
349 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 350 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
351 | int i; |
352 | int will_compress; | |
261507a0 | 353 | int compress_type = root->fs_info->compress_type; |
b888db2b | 354 | |
4cb13e5d LB |
355 | /* if this is a small write inside eof, kick off a defrag */ |
356 | if ((end - start + 1) < 16 * 1024 && | |
357 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
358 | btrfs_add_inode_defrag(NULL, inode); |
359 | ||
42dc7bab | 360 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
361 | again: |
362 | will_compress = 0; | |
363 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
364 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 365 | |
f03d9301 CM |
366 | /* |
367 | * we don't want to send crud past the end of i_size through | |
368 | * compression, that's just a waste of CPU time. So, if the | |
369 | * end of the file is before the start of our current | |
370 | * requested range of bytes, we bail out to the uncompressed | |
371 | * cleanup code that can deal with all of this. | |
372 | * | |
373 | * It isn't really the fastest way to fix things, but this is a | |
374 | * very uncommon corner. | |
375 | */ | |
376 | if (actual_end <= start) | |
377 | goto cleanup_and_bail_uncompressed; | |
378 | ||
c8b97818 CM |
379 | total_compressed = actual_end - start; |
380 | ||
381 | /* we want to make sure that amount of ram required to uncompress | |
382 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
383 | * of a compressed extent to 128k. This is a crucial number |
384 | * because it also controls how easily we can spread reads across | |
385 | * cpus for decompression. | |
386 | * | |
387 | * We also want to make sure the amount of IO required to do | |
388 | * a random read is reasonably small, so we limit the size of | |
389 | * a compressed extent to 128k. | |
c8b97818 CM |
390 | */ |
391 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 392 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 393 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
394 | total_in = 0; |
395 | ret = 0; | |
db94535d | 396 | |
771ed689 CM |
397 | /* |
398 | * we do compression for mount -o compress and when the | |
399 | * inode has not been flagged as nocompress. This flag can | |
400 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 401 | */ |
6cbff00f | 402 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 403 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
404 | (BTRFS_I(inode)->force_compress) || |
405 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 406 | WARN_ON(pages); |
cfbc246e | 407 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
408 | if (!pages) { |
409 | /* just bail out to the uncompressed code */ | |
410 | goto cont; | |
411 | } | |
c8b97818 | 412 | |
261507a0 LZ |
413 | if (BTRFS_I(inode)->force_compress) |
414 | compress_type = BTRFS_I(inode)->force_compress; | |
415 | ||
416 | ret = btrfs_compress_pages(compress_type, | |
417 | inode->i_mapping, start, | |
418 | total_compressed, pages, | |
419 | nr_pages, &nr_pages_ret, | |
420 | &total_in, | |
421 | &total_compressed, | |
422 | max_compressed); | |
c8b97818 CM |
423 | |
424 | if (!ret) { | |
425 | unsigned long offset = total_compressed & | |
426 | (PAGE_CACHE_SIZE - 1); | |
427 | struct page *page = pages[nr_pages_ret - 1]; | |
428 | char *kaddr; | |
429 | ||
430 | /* zero the tail end of the last page, we might be | |
431 | * sending it down to disk | |
432 | */ | |
433 | if (offset) { | |
7ac687d9 | 434 | kaddr = kmap_atomic(page); |
c8b97818 CM |
435 | memset(kaddr + offset, 0, |
436 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 437 | kunmap_atomic(kaddr); |
c8b97818 CM |
438 | } |
439 | will_compress = 1; | |
440 | } | |
441 | } | |
560f7d75 | 442 | cont: |
c8b97818 | 443 | if (start == 0) { |
7a7eaa40 | 444 | trans = btrfs_join_transaction(root); |
79787eaa JM |
445 | if (IS_ERR(trans)) { |
446 | ret = PTR_ERR(trans); | |
447 | trans = NULL; | |
448 | goto cleanup_and_out; | |
449 | } | |
0ca1f7ce | 450 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 451 | |
c8b97818 | 452 | /* lets try to make an inline extent */ |
771ed689 | 453 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 454 | /* we didn't compress the entire range, try |
771ed689 | 455 | * to make an uncompressed inline extent. |
c8b97818 CM |
456 | */ |
457 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 458 | start, end, 0, 0, NULL); |
c8b97818 | 459 | } else { |
771ed689 | 460 | /* try making a compressed inline extent */ |
c8b97818 CM |
461 | ret = cow_file_range_inline(trans, root, inode, |
462 | start, end, | |
fe3f566c LZ |
463 | total_compressed, |
464 | compress_type, pages); | |
c8b97818 | 465 | } |
79787eaa | 466 | if (ret <= 0) { |
771ed689 | 467 | /* |
79787eaa JM |
468 | * inline extent creation worked or returned error, |
469 | * we don't need to create any more async work items. | |
470 | * Unlock and free up our temp pages. | |
771ed689 | 471 | */ |
c8b97818 | 472 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
473 | &BTRFS_I(inode)->io_tree, |
474 | start, end, NULL, | |
475 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 476 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 477 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
478 | |
479 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
480 | goto free_pages_out; |
481 | } | |
c2167754 | 482 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
483 | } |
484 | ||
485 | if (will_compress) { | |
486 | /* | |
487 | * we aren't doing an inline extent round the compressed size | |
488 | * up to a block size boundary so the allocator does sane | |
489 | * things | |
490 | */ | |
491 | total_compressed = (total_compressed + blocksize - 1) & | |
492 | ~(blocksize - 1); | |
493 | ||
494 | /* | |
495 | * one last check to make sure the compression is really a | |
496 | * win, compare the page count read with the blocks on disk | |
497 | */ | |
498 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
499 | ~(PAGE_CACHE_SIZE - 1); | |
500 | if (total_compressed >= total_in) { | |
501 | will_compress = 0; | |
502 | } else { | |
c8b97818 CM |
503 | num_bytes = total_in; |
504 | } | |
505 | } | |
506 | if (!will_compress && pages) { | |
507 | /* | |
508 | * the compression code ran but failed to make things smaller, | |
509 | * free any pages it allocated and our page pointer array | |
510 | */ | |
511 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 512 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
513 | page_cache_release(pages[i]); |
514 | } | |
515 | kfree(pages); | |
516 | pages = NULL; | |
517 | total_compressed = 0; | |
518 | nr_pages_ret = 0; | |
519 | ||
520 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
521 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
522 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 523 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 524 | } |
c8b97818 | 525 | } |
771ed689 CM |
526 | if (will_compress) { |
527 | *num_added += 1; | |
c8b97818 | 528 | |
771ed689 CM |
529 | /* the async work queues will take care of doing actual |
530 | * allocation on disk for these compressed pages, | |
531 | * and will submit them to the elevator. | |
532 | */ | |
533 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
534 | total_compressed, pages, nr_pages_ret, |
535 | compress_type); | |
179e29e4 | 536 | |
24ae6365 | 537 | if (start + num_bytes < end) { |
771ed689 CM |
538 | start += num_bytes; |
539 | pages = NULL; | |
540 | cond_resched(); | |
541 | goto again; | |
542 | } | |
543 | } else { | |
f03d9301 | 544 | cleanup_and_bail_uncompressed: |
771ed689 CM |
545 | /* |
546 | * No compression, but we still need to write the pages in | |
547 | * the file we've been given so far. redirty the locked | |
548 | * page if it corresponds to our extent and set things up | |
549 | * for the async work queue to run cow_file_range to do | |
550 | * the normal delalloc dance | |
551 | */ | |
552 | if (page_offset(locked_page) >= start && | |
553 | page_offset(locked_page) <= end) { | |
554 | __set_page_dirty_nobuffers(locked_page); | |
555 | /* unlocked later on in the async handlers */ | |
556 | } | |
261507a0 LZ |
557 | add_async_extent(async_cow, start, end - start + 1, |
558 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
559 | *num_added += 1; |
560 | } | |
3b951516 | 561 | |
771ed689 | 562 | out: |
79787eaa | 563 | return ret; |
771ed689 CM |
564 | |
565 | free_pages_out: | |
566 | for (i = 0; i < nr_pages_ret; i++) { | |
567 | WARN_ON(pages[i]->mapping); | |
568 | page_cache_release(pages[i]); | |
569 | } | |
d397712b | 570 | kfree(pages); |
771ed689 CM |
571 | |
572 | goto out; | |
79787eaa JM |
573 | |
574 | cleanup_and_out: | |
575 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
576 | start, end, NULL, | |
577 | EXTENT_CLEAR_UNLOCK_PAGE | | |
578 | EXTENT_CLEAR_DIRTY | | |
579 | EXTENT_CLEAR_DELALLOC | | |
580 | EXTENT_SET_WRITEBACK | | |
581 | EXTENT_END_WRITEBACK); | |
582 | if (!trans || IS_ERR(trans)) | |
583 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
584 | else | |
585 | btrfs_abort_transaction(trans, root, ret); | |
586 | goto free_pages_out; | |
771ed689 CM |
587 | } |
588 | ||
589 | /* | |
590 | * phase two of compressed writeback. This is the ordered portion | |
591 | * of the code, which only gets called in the order the work was | |
592 | * queued. We walk all the async extents created by compress_file_range | |
593 | * and send them down to the disk. | |
594 | */ | |
595 | static noinline int submit_compressed_extents(struct inode *inode, | |
596 | struct async_cow *async_cow) | |
597 | { | |
598 | struct async_extent *async_extent; | |
599 | u64 alloc_hint = 0; | |
600 | struct btrfs_trans_handle *trans; | |
601 | struct btrfs_key ins; | |
602 | struct extent_map *em; | |
603 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
604 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
605 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 606 | int ret = 0; |
771ed689 CM |
607 | |
608 | if (list_empty(&async_cow->extents)) | |
609 | return 0; | |
610 | ||
771ed689 | 611 | |
d397712b | 612 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
613 | async_extent = list_entry(async_cow->extents.next, |
614 | struct async_extent, list); | |
615 | list_del(&async_extent->list); | |
c8b97818 | 616 | |
771ed689 CM |
617 | io_tree = &BTRFS_I(inode)->io_tree; |
618 | ||
f5a84ee3 | 619 | retry: |
771ed689 CM |
620 | /* did the compression code fall back to uncompressed IO? */ |
621 | if (!async_extent->pages) { | |
622 | int page_started = 0; | |
623 | unsigned long nr_written = 0; | |
624 | ||
625 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 626 | async_extent->start + |
d0082371 | 627 | async_extent->ram_size - 1); |
771ed689 CM |
628 | |
629 | /* allocate blocks */ | |
f5a84ee3 JB |
630 | ret = cow_file_range(inode, async_cow->locked_page, |
631 | async_extent->start, | |
632 | async_extent->start + | |
633 | async_extent->ram_size - 1, | |
634 | &page_started, &nr_written, 0); | |
771ed689 | 635 | |
79787eaa JM |
636 | /* JDM XXX */ |
637 | ||
771ed689 CM |
638 | /* |
639 | * if page_started, cow_file_range inserted an | |
640 | * inline extent and took care of all the unlocking | |
641 | * and IO for us. Otherwise, we need to submit | |
642 | * all those pages down to the drive. | |
643 | */ | |
f5a84ee3 | 644 | if (!page_started && !ret) |
771ed689 CM |
645 | extent_write_locked_range(io_tree, |
646 | inode, async_extent->start, | |
d397712b | 647 | async_extent->start + |
771ed689 CM |
648 | async_extent->ram_size - 1, |
649 | btrfs_get_extent, | |
650 | WB_SYNC_ALL); | |
651 | kfree(async_extent); | |
652 | cond_resched(); | |
653 | continue; | |
654 | } | |
655 | ||
656 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 657 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 658 | |
7a7eaa40 | 659 | trans = btrfs_join_transaction(root); |
79787eaa JM |
660 | if (IS_ERR(trans)) { |
661 | ret = PTR_ERR(trans); | |
662 | } else { | |
663 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
664 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
665 | async_extent->compressed_size, |
666 | async_extent->compressed_size, | |
81c9ad23 | 667 | 0, alloc_hint, &ins, 1); |
962197ba | 668 | if (ret && ret != -ENOSPC) |
79787eaa JM |
669 | btrfs_abort_transaction(trans, root, ret); |
670 | btrfs_end_transaction(trans, root); | |
671 | } | |
c2167754 | 672 | |
f5a84ee3 JB |
673 | if (ret) { |
674 | int i; | |
675 | for (i = 0; i < async_extent->nr_pages; i++) { | |
676 | WARN_ON(async_extent->pages[i]->mapping); | |
677 | page_cache_release(async_extent->pages[i]); | |
678 | } | |
679 | kfree(async_extent->pages); | |
680 | async_extent->nr_pages = 0; | |
681 | async_extent->pages = NULL; | |
682 | unlock_extent(io_tree, async_extent->start, | |
683 | async_extent->start + | |
d0082371 | 684 | async_extent->ram_size - 1); |
79787eaa JM |
685 | if (ret == -ENOSPC) |
686 | goto retry; | |
687 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
688 | } |
689 | ||
c2167754 YZ |
690 | /* |
691 | * here we're doing allocation and writeback of the | |
692 | * compressed pages | |
693 | */ | |
694 | btrfs_drop_extent_cache(inode, async_extent->start, | |
695 | async_extent->start + | |
696 | async_extent->ram_size - 1, 0); | |
697 | ||
172ddd60 | 698 | em = alloc_extent_map(); |
79787eaa | 699 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
700 | em->start = async_extent->start; |
701 | em->len = async_extent->ram_size; | |
445a6944 | 702 | em->orig_start = em->start; |
2ab28f32 JB |
703 | em->mod_start = em->start; |
704 | em->mod_len = em->len; | |
c8b97818 | 705 | |
771ed689 CM |
706 | em->block_start = ins.objectid; |
707 | em->block_len = ins.offset; | |
b4939680 | 708 | em->orig_block_len = ins.offset; |
771ed689 | 709 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 710 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
711 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
712 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 713 | em->generation = -1; |
771ed689 | 714 | |
d397712b | 715 | while (1) { |
890871be | 716 | write_lock(&em_tree->lock); |
771ed689 | 717 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
718 | if (!ret) |
719 | list_move(&em->list, | |
720 | &em_tree->modified_extents); | |
890871be | 721 | write_unlock(&em_tree->lock); |
771ed689 CM |
722 | if (ret != -EEXIST) { |
723 | free_extent_map(em); | |
724 | break; | |
725 | } | |
726 | btrfs_drop_extent_cache(inode, async_extent->start, | |
727 | async_extent->start + | |
728 | async_extent->ram_size - 1, 0); | |
729 | } | |
730 | ||
261507a0 LZ |
731 | ret = btrfs_add_ordered_extent_compress(inode, |
732 | async_extent->start, | |
733 | ins.objectid, | |
734 | async_extent->ram_size, | |
735 | ins.offset, | |
736 | BTRFS_ORDERED_COMPRESSED, | |
737 | async_extent->compress_type); | |
79787eaa | 738 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 739 | |
771ed689 CM |
740 | /* |
741 | * clear dirty, set writeback and unlock the pages. | |
742 | */ | |
743 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
744 | &BTRFS_I(inode)->io_tree, |
745 | async_extent->start, | |
746 | async_extent->start + | |
747 | async_extent->ram_size - 1, | |
748 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
749 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 750 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 751 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
752 | |
753 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
754 | async_extent->start, |
755 | async_extent->ram_size, | |
756 | ins.objectid, | |
757 | ins.offset, async_extent->pages, | |
758 | async_extent->nr_pages); | |
771ed689 | 759 | |
79787eaa | 760 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
761 | alloc_hint = ins.objectid + ins.offset; |
762 | kfree(async_extent); | |
763 | cond_resched(); | |
764 | } | |
79787eaa JM |
765 | ret = 0; |
766 | out: | |
767 | return ret; | |
768 | out_free: | |
769 | kfree(async_extent); | |
770 | goto out; | |
771ed689 CM |
771 | } |
772 | ||
4b46fce2 JB |
773 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
774 | u64 num_bytes) | |
775 | { | |
776 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
777 | struct extent_map *em; | |
778 | u64 alloc_hint = 0; | |
779 | ||
780 | read_lock(&em_tree->lock); | |
781 | em = search_extent_mapping(em_tree, start, num_bytes); | |
782 | if (em) { | |
783 | /* | |
784 | * if block start isn't an actual block number then find the | |
785 | * first block in this inode and use that as a hint. If that | |
786 | * block is also bogus then just don't worry about it. | |
787 | */ | |
788 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
789 | free_extent_map(em); | |
790 | em = search_extent_mapping(em_tree, 0, 0); | |
791 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
792 | alloc_hint = em->block_start; | |
793 | if (em) | |
794 | free_extent_map(em); | |
795 | } else { | |
796 | alloc_hint = em->block_start; | |
797 | free_extent_map(em); | |
798 | } | |
799 | } | |
800 | read_unlock(&em_tree->lock); | |
801 | ||
802 | return alloc_hint; | |
803 | } | |
804 | ||
771ed689 CM |
805 | /* |
806 | * when extent_io.c finds a delayed allocation range in the file, | |
807 | * the call backs end up in this code. The basic idea is to | |
808 | * allocate extents on disk for the range, and create ordered data structs | |
809 | * in ram to track those extents. | |
810 | * | |
811 | * locked_page is the page that writepage had locked already. We use | |
812 | * it to make sure we don't do extra locks or unlocks. | |
813 | * | |
814 | * *page_started is set to one if we unlock locked_page and do everything | |
815 | * required to start IO on it. It may be clean and already done with | |
816 | * IO when we return. | |
817 | */ | |
b7d5b0a8 MX |
818 | static noinline int __cow_file_range(struct btrfs_trans_handle *trans, |
819 | struct inode *inode, | |
820 | struct btrfs_root *root, | |
821 | struct page *locked_page, | |
822 | u64 start, u64 end, int *page_started, | |
823 | unsigned long *nr_written, | |
824 | int unlock) | |
771ed689 | 825 | { |
771ed689 CM |
826 | u64 alloc_hint = 0; |
827 | u64 num_bytes; | |
828 | unsigned long ram_size; | |
829 | u64 disk_num_bytes; | |
830 | u64 cur_alloc_size; | |
831 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
832 | struct btrfs_key ins; |
833 | struct extent_map *em; | |
834 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
835 | int ret = 0; | |
836 | ||
83eea1f1 | 837 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 838 | |
771ed689 CM |
839 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
840 | num_bytes = max(blocksize, num_bytes); | |
841 | disk_num_bytes = num_bytes; | |
771ed689 | 842 | |
4cb5300b | 843 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
844 | if (num_bytes < 64 * 1024 && |
845 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
846 | btrfs_add_inode_defrag(trans, inode); |
847 | ||
771ed689 CM |
848 | if (start == 0) { |
849 | /* lets try to make an inline extent */ | |
850 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 851 | start, end, 0, 0, NULL); |
771ed689 CM |
852 | if (ret == 0) { |
853 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
854 | &BTRFS_I(inode)->io_tree, |
855 | start, end, NULL, | |
856 | EXTENT_CLEAR_UNLOCK_PAGE | | |
857 | EXTENT_CLEAR_UNLOCK | | |
858 | EXTENT_CLEAR_DELALLOC | | |
859 | EXTENT_CLEAR_DIRTY | | |
860 | EXTENT_SET_WRITEBACK | | |
861 | EXTENT_END_WRITEBACK); | |
c2167754 | 862 | |
771ed689 CM |
863 | *nr_written = *nr_written + |
864 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
865 | *page_started = 1; | |
771ed689 | 866 | goto out; |
79787eaa JM |
867 | } else if (ret < 0) { |
868 | btrfs_abort_transaction(trans, root, ret); | |
869 | goto out_unlock; | |
771ed689 CM |
870 | } |
871 | } | |
872 | ||
873 | BUG_ON(disk_num_bytes > | |
6c41761f | 874 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 875 | |
4b46fce2 | 876 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
877 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
878 | ||
d397712b | 879 | while (disk_num_bytes > 0) { |
a791e35e CM |
880 | unsigned long op; |
881 | ||
287a0ab9 | 882 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 883 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 884 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 885 | &ins, 1); |
79787eaa JM |
886 | if (ret < 0) { |
887 | btrfs_abort_transaction(trans, root, ret); | |
888 | goto out_unlock; | |
889 | } | |
d397712b | 890 | |
172ddd60 | 891 | em = alloc_extent_map(); |
79787eaa | 892 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 893 | em->start = start; |
445a6944 | 894 | em->orig_start = em->start; |
771ed689 CM |
895 | ram_size = ins.offset; |
896 | em->len = ins.offset; | |
2ab28f32 JB |
897 | em->mod_start = em->start; |
898 | em->mod_len = em->len; | |
c8b97818 | 899 | |
e6dcd2dc | 900 | em->block_start = ins.objectid; |
c8b97818 | 901 | em->block_len = ins.offset; |
b4939680 | 902 | em->orig_block_len = ins.offset; |
e6dcd2dc | 903 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 904 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 905 | em->generation = -1; |
c8b97818 | 906 | |
d397712b | 907 | while (1) { |
890871be | 908 | write_lock(&em_tree->lock); |
e6dcd2dc | 909 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
910 | if (!ret) |
911 | list_move(&em->list, | |
912 | &em_tree->modified_extents); | |
890871be | 913 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
914 | if (ret != -EEXIST) { |
915 | free_extent_map(em); | |
916 | break; | |
917 | } | |
918 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 919 | start + ram_size - 1, 0); |
e6dcd2dc CM |
920 | } |
921 | ||
98d20f67 | 922 | cur_alloc_size = ins.offset; |
e6dcd2dc | 923 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 924 | ram_size, cur_alloc_size, 0); |
79787eaa | 925 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 926 | |
17d217fe YZ |
927 | if (root->root_key.objectid == |
928 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
929 | ret = btrfs_reloc_clone_csums(inode, start, | |
930 | cur_alloc_size); | |
79787eaa JM |
931 | if (ret) { |
932 | btrfs_abort_transaction(trans, root, ret); | |
933 | goto out_unlock; | |
934 | } | |
17d217fe YZ |
935 | } |
936 | ||
d397712b | 937 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 938 | break; |
d397712b | 939 | |
c8b97818 CM |
940 | /* we're not doing compressed IO, don't unlock the first |
941 | * page (which the caller expects to stay locked), don't | |
942 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
943 | * |
944 | * Do set the Private2 bit so we know this page was properly | |
945 | * setup for writepage | |
c8b97818 | 946 | */ |
a791e35e CM |
947 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
948 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
949 | EXTENT_SET_PRIVATE2; | |
950 | ||
c8b97818 CM |
951 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
952 | start, start + ram_size - 1, | |
a791e35e | 953 | locked_page, op); |
c8b97818 | 954 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
955 | num_bytes -= cur_alloc_size; |
956 | alloc_hint = ins.objectid + ins.offset; | |
957 | start += cur_alloc_size; | |
b888db2b | 958 | } |
79787eaa | 959 | out: |
be20aa9d | 960 | return ret; |
b7d5b0a8 | 961 | |
79787eaa JM |
962 | out_unlock: |
963 | extent_clear_unlock_delalloc(inode, | |
964 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 965 | start, end, locked_page, |
79787eaa JM |
966 | EXTENT_CLEAR_UNLOCK_PAGE | |
967 | EXTENT_CLEAR_UNLOCK | | |
968 | EXTENT_CLEAR_DELALLOC | | |
969 | EXTENT_CLEAR_DIRTY | | |
970 | EXTENT_SET_WRITEBACK | | |
971 | EXTENT_END_WRITEBACK); | |
972 | ||
973 | goto out; | |
771ed689 | 974 | } |
c8b97818 | 975 | |
b7d5b0a8 MX |
976 | static noinline int cow_file_range(struct inode *inode, |
977 | struct page *locked_page, | |
978 | u64 start, u64 end, int *page_started, | |
979 | unsigned long *nr_written, | |
980 | int unlock) | |
981 | { | |
982 | struct btrfs_trans_handle *trans; | |
983 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
984 | int ret; | |
985 | ||
986 | trans = btrfs_join_transaction(root); | |
987 | if (IS_ERR(trans)) { | |
988 | extent_clear_unlock_delalloc(inode, | |
989 | &BTRFS_I(inode)->io_tree, | |
990 | start, end, locked_page, | |
991 | EXTENT_CLEAR_UNLOCK_PAGE | | |
992 | EXTENT_CLEAR_UNLOCK | | |
993 | EXTENT_CLEAR_DELALLOC | | |
994 | EXTENT_CLEAR_DIRTY | | |
995 | EXTENT_SET_WRITEBACK | | |
996 | EXTENT_END_WRITEBACK); | |
997 | return PTR_ERR(trans); | |
998 | } | |
999 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
1000 | ||
1001 | ret = __cow_file_range(trans, inode, root, locked_page, start, end, | |
1002 | page_started, nr_written, unlock); | |
1003 | ||
1004 | btrfs_end_transaction(trans, root); | |
1005 | ||
1006 | return ret; | |
1007 | } | |
1008 | ||
771ed689 CM |
1009 | /* |
1010 | * work queue call back to started compression on a file and pages | |
1011 | */ | |
1012 | static noinline void async_cow_start(struct btrfs_work *work) | |
1013 | { | |
1014 | struct async_cow *async_cow; | |
1015 | int num_added = 0; | |
1016 | async_cow = container_of(work, struct async_cow, work); | |
1017 | ||
1018 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1019 | async_cow->start, async_cow->end, async_cow, | |
1020 | &num_added); | |
8180ef88 | 1021 | if (num_added == 0) { |
cb77fcd8 | 1022 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1023 | async_cow->inode = NULL; |
8180ef88 | 1024 | } |
771ed689 CM |
1025 | } |
1026 | ||
1027 | /* | |
1028 | * work queue call back to submit previously compressed pages | |
1029 | */ | |
1030 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1031 | { | |
1032 | struct async_cow *async_cow; | |
1033 | struct btrfs_root *root; | |
1034 | unsigned long nr_pages; | |
1035 | ||
1036 | async_cow = container_of(work, struct async_cow, work); | |
1037 | ||
1038 | root = async_cow->root; | |
1039 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1040 | PAGE_CACHE_SHIFT; | |
1041 | ||
66657b31 | 1042 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1043 | 5 * 1024 * 1024 && |
771ed689 CM |
1044 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1045 | wake_up(&root->fs_info->async_submit_wait); | |
1046 | ||
d397712b | 1047 | if (async_cow->inode) |
771ed689 | 1048 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1049 | } |
c8b97818 | 1050 | |
771ed689 CM |
1051 | static noinline void async_cow_free(struct btrfs_work *work) |
1052 | { | |
1053 | struct async_cow *async_cow; | |
1054 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1055 | if (async_cow->inode) |
cb77fcd8 | 1056 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1057 | kfree(async_cow); |
1058 | } | |
1059 | ||
1060 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1061 | u64 start, u64 end, int *page_started, | |
1062 | unsigned long *nr_written) | |
1063 | { | |
1064 | struct async_cow *async_cow; | |
1065 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1066 | unsigned long nr_pages; | |
1067 | u64 cur_end; | |
287082b0 | 1068 | int limit = 10 * 1024 * 1024; |
771ed689 | 1069 | |
a3429ab7 CM |
1070 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1071 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1072 | while (start < end) { |
771ed689 | 1073 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1074 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1075 | async_cow->inode = igrab(inode); |
771ed689 CM |
1076 | async_cow->root = root; |
1077 | async_cow->locked_page = locked_page; | |
1078 | async_cow->start = start; | |
1079 | ||
6cbff00f | 1080 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1081 | cur_end = end; |
1082 | else | |
1083 | cur_end = min(end, start + 512 * 1024 - 1); | |
1084 | ||
1085 | async_cow->end = cur_end; | |
1086 | INIT_LIST_HEAD(&async_cow->extents); | |
1087 | ||
1088 | async_cow->work.func = async_cow_start; | |
1089 | async_cow->work.ordered_func = async_cow_submit; | |
1090 | async_cow->work.ordered_free = async_cow_free; | |
1091 | async_cow->work.flags = 0; | |
1092 | ||
771ed689 CM |
1093 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1094 | PAGE_CACHE_SHIFT; | |
1095 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1096 | ||
1097 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1098 | &async_cow->work); | |
1099 | ||
1100 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1101 | wait_event(root->fs_info->async_submit_wait, | |
1102 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1103 | limit)); | |
1104 | } | |
1105 | ||
d397712b | 1106 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1107 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1108 | wait_event(root->fs_info->async_submit_wait, | |
1109 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1110 | 0)); | |
1111 | } | |
1112 | ||
1113 | *nr_written += nr_pages; | |
1114 | start = cur_end + 1; | |
1115 | } | |
1116 | *page_started = 1; | |
1117 | return 0; | |
be20aa9d CM |
1118 | } |
1119 | ||
d397712b | 1120 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1121 | u64 bytenr, u64 num_bytes) |
1122 | { | |
1123 | int ret; | |
1124 | struct btrfs_ordered_sum *sums; | |
1125 | LIST_HEAD(list); | |
1126 | ||
07d400a6 | 1127 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1128 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1129 | if (ret == 0 && list_empty(&list)) |
1130 | return 0; | |
1131 | ||
1132 | while (!list_empty(&list)) { | |
1133 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1134 | list_del(&sums->list); | |
1135 | kfree(sums); | |
1136 | } | |
1137 | return 1; | |
1138 | } | |
1139 | ||
d352ac68 CM |
1140 | /* |
1141 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1142 | * of the extents that exist in the file, and COWs the file as required. | |
1143 | * | |
1144 | * If no cow copies or snapshots exist, we write directly to the existing | |
1145 | * blocks on disk | |
1146 | */ | |
7f366cfe CM |
1147 | static noinline int run_delalloc_nocow(struct inode *inode, |
1148 | struct page *locked_page, | |
771ed689 CM |
1149 | u64 start, u64 end, int *page_started, int force, |
1150 | unsigned long *nr_written) | |
be20aa9d | 1151 | { |
be20aa9d | 1152 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1153 | struct btrfs_trans_handle *trans; |
be20aa9d | 1154 | struct extent_buffer *leaf; |
be20aa9d | 1155 | struct btrfs_path *path; |
80ff3856 | 1156 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1157 | struct btrfs_key found_key; |
80ff3856 YZ |
1158 | u64 cow_start; |
1159 | u64 cur_offset; | |
1160 | u64 extent_end; | |
5d4f98a2 | 1161 | u64 extent_offset; |
80ff3856 YZ |
1162 | u64 disk_bytenr; |
1163 | u64 num_bytes; | |
b4939680 | 1164 | u64 disk_num_bytes; |
80ff3856 | 1165 | int extent_type; |
79787eaa | 1166 | int ret, err; |
d899e052 | 1167 | int type; |
80ff3856 YZ |
1168 | int nocow; |
1169 | int check_prev = 1; | |
82d5902d | 1170 | bool nolock; |
33345d01 | 1171 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1172 | |
1173 | path = btrfs_alloc_path(); | |
17ca04af JB |
1174 | if (!path) { |
1175 | extent_clear_unlock_delalloc(inode, | |
1176 | &BTRFS_I(inode)->io_tree, | |
1177 | start, end, locked_page, | |
1178 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1179 | EXTENT_CLEAR_UNLOCK | | |
1180 | EXTENT_CLEAR_DELALLOC | | |
1181 | EXTENT_CLEAR_DIRTY | | |
1182 | EXTENT_SET_WRITEBACK | | |
1183 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1184 | return -ENOMEM; |
17ca04af | 1185 | } |
82d5902d | 1186 | |
83eea1f1 | 1187 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1188 | |
1189 | if (nolock) | |
7a7eaa40 | 1190 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1191 | else |
7a7eaa40 | 1192 | trans = btrfs_join_transaction(root); |
ff5714cc | 1193 | |
79787eaa | 1194 | if (IS_ERR(trans)) { |
17ca04af JB |
1195 | extent_clear_unlock_delalloc(inode, |
1196 | &BTRFS_I(inode)->io_tree, | |
1197 | start, end, locked_page, | |
1198 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1199 | EXTENT_CLEAR_UNLOCK | | |
1200 | EXTENT_CLEAR_DELALLOC | | |
1201 | EXTENT_CLEAR_DIRTY | | |
1202 | EXTENT_SET_WRITEBACK | | |
1203 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1204 | btrfs_free_path(path); |
1205 | return PTR_ERR(trans); | |
1206 | } | |
1207 | ||
74b21075 | 1208 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1209 | |
80ff3856 YZ |
1210 | cow_start = (u64)-1; |
1211 | cur_offset = start; | |
1212 | while (1) { | |
33345d01 | 1213 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1214 | cur_offset, 0); |
79787eaa JM |
1215 | if (ret < 0) { |
1216 | btrfs_abort_transaction(trans, root, ret); | |
1217 | goto error; | |
1218 | } | |
80ff3856 YZ |
1219 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1220 | leaf = path->nodes[0]; | |
1221 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1222 | path->slots[0] - 1); | |
33345d01 | 1223 | if (found_key.objectid == ino && |
80ff3856 YZ |
1224 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1225 | path->slots[0]--; | |
1226 | } | |
1227 | check_prev = 0; | |
1228 | next_slot: | |
1229 | leaf = path->nodes[0]; | |
1230 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1231 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1232 | if (ret < 0) { |
1233 | btrfs_abort_transaction(trans, root, ret); | |
1234 | goto error; | |
1235 | } | |
80ff3856 YZ |
1236 | if (ret > 0) |
1237 | break; | |
1238 | leaf = path->nodes[0]; | |
1239 | } | |
be20aa9d | 1240 | |
80ff3856 YZ |
1241 | nocow = 0; |
1242 | disk_bytenr = 0; | |
17d217fe | 1243 | num_bytes = 0; |
80ff3856 YZ |
1244 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1245 | ||
33345d01 | 1246 | if (found_key.objectid > ino || |
80ff3856 YZ |
1247 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1248 | found_key.offset > end) | |
1249 | break; | |
1250 | ||
1251 | if (found_key.offset > cur_offset) { | |
1252 | extent_end = found_key.offset; | |
e9061e21 | 1253 | extent_type = 0; |
80ff3856 YZ |
1254 | goto out_check; |
1255 | } | |
1256 | ||
1257 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_file_extent_item); | |
1259 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1260 | ||
d899e052 YZ |
1261 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1262 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1263 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1264 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1265 | extent_end = found_key.offset + |
1266 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1267 | disk_num_bytes = |
1268 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1269 | if (extent_end <= start) { |
1270 | path->slots[0]++; | |
1271 | goto next_slot; | |
1272 | } | |
17d217fe YZ |
1273 | if (disk_bytenr == 0) |
1274 | goto out_check; | |
80ff3856 YZ |
1275 | if (btrfs_file_extent_compression(leaf, fi) || |
1276 | btrfs_file_extent_encryption(leaf, fi) || | |
1277 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1278 | goto out_check; | |
d899e052 YZ |
1279 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1280 | goto out_check; | |
d2fb3437 | 1281 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1282 | goto out_check; |
33345d01 | 1283 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1284 | found_key.offset - |
1285 | extent_offset, disk_bytenr)) | |
17d217fe | 1286 | goto out_check; |
5d4f98a2 | 1287 | disk_bytenr += extent_offset; |
17d217fe YZ |
1288 | disk_bytenr += cur_offset - found_key.offset; |
1289 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1290 | /* | |
1291 | * force cow if csum exists in the range. | |
1292 | * this ensure that csum for a given extent are | |
1293 | * either valid or do not exist. | |
1294 | */ | |
1295 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1296 | goto out_check; | |
80ff3856 YZ |
1297 | nocow = 1; |
1298 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1299 | extent_end = found_key.offset + | |
1300 | btrfs_file_extent_inline_len(leaf, fi); | |
1301 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1302 | } else { | |
1303 | BUG_ON(1); | |
1304 | } | |
1305 | out_check: | |
1306 | if (extent_end <= start) { | |
1307 | path->slots[0]++; | |
1308 | goto next_slot; | |
1309 | } | |
1310 | if (!nocow) { | |
1311 | if (cow_start == (u64)-1) | |
1312 | cow_start = cur_offset; | |
1313 | cur_offset = extent_end; | |
1314 | if (cur_offset > end) | |
1315 | break; | |
1316 | path->slots[0]++; | |
1317 | goto next_slot; | |
7ea394f1 YZ |
1318 | } |
1319 | ||
b3b4aa74 | 1320 | btrfs_release_path(path); |
80ff3856 | 1321 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1322 | ret = __cow_file_range(trans, inode, root, locked_page, |
1323 | cow_start, found_key.offset - 1, | |
1324 | page_started, nr_written, 1); | |
79787eaa JM |
1325 | if (ret) { |
1326 | btrfs_abort_transaction(trans, root, ret); | |
1327 | goto error; | |
1328 | } | |
80ff3856 | 1329 | cow_start = (u64)-1; |
7ea394f1 | 1330 | } |
80ff3856 | 1331 | |
d899e052 YZ |
1332 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1333 | struct extent_map *em; | |
1334 | struct extent_map_tree *em_tree; | |
1335 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1336 | em = alloc_extent_map(); |
79787eaa | 1337 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1338 | em->start = cur_offset; |
70c8a91c | 1339 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1340 | em->len = num_bytes; |
1341 | em->block_len = num_bytes; | |
1342 | em->block_start = disk_bytenr; | |
b4939680 | 1343 | em->orig_block_len = disk_num_bytes; |
d899e052 | 1344 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1345 | em->mod_start = em->start; |
1346 | em->mod_len = em->len; | |
d899e052 | 1347 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1348 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1349 | em->generation = -1; |
d899e052 | 1350 | while (1) { |
890871be | 1351 | write_lock(&em_tree->lock); |
d899e052 | 1352 | ret = add_extent_mapping(em_tree, em); |
70c8a91c JB |
1353 | if (!ret) |
1354 | list_move(&em->list, | |
1355 | &em_tree->modified_extents); | |
890871be | 1356 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1357 | if (ret != -EEXIST) { |
1358 | free_extent_map(em); | |
1359 | break; | |
1360 | } | |
1361 | btrfs_drop_extent_cache(inode, em->start, | |
1362 | em->start + em->len - 1, 0); | |
1363 | } | |
1364 | type = BTRFS_ORDERED_PREALLOC; | |
1365 | } else { | |
1366 | type = BTRFS_ORDERED_NOCOW; | |
1367 | } | |
80ff3856 YZ |
1368 | |
1369 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1370 | num_bytes, num_bytes, type); |
79787eaa | 1371 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1372 | |
efa56464 YZ |
1373 | if (root->root_key.objectid == |
1374 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1375 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1376 | num_bytes); | |
79787eaa JM |
1377 | if (ret) { |
1378 | btrfs_abort_transaction(trans, root, ret); | |
1379 | goto error; | |
1380 | } | |
efa56464 YZ |
1381 | } |
1382 | ||
d899e052 | 1383 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1384 | cur_offset, cur_offset + num_bytes - 1, |
1385 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1386 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1387 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1388 | cur_offset = extent_end; |
1389 | if (cur_offset > end) | |
1390 | break; | |
be20aa9d | 1391 | } |
b3b4aa74 | 1392 | btrfs_release_path(path); |
80ff3856 | 1393 | |
17ca04af | 1394 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1395 | cow_start = cur_offset; |
17ca04af JB |
1396 | cur_offset = end; |
1397 | } | |
1398 | ||
80ff3856 | 1399 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1400 | ret = __cow_file_range(trans, inode, root, locked_page, |
1401 | cow_start, end, | |
1402 | page_started, nr_written, 1); | |
79787eaa JM |
1403 | if (ret) { |
1404 | btrfs_abort_transaction(trans, root, ret); | |
1405 | goto error; | |
1406 | } | |
80ff3856 YZ |
1407 | } |
1408 | ||
79787eaa | 1409 | error: |
a698d075 | 1410 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1411 | if (!ret) |
1412 | ret = err; | |
1413 | ||
17ca04af JB |
1414 | if (ret && cur_offset < end) |
1415 | extent_clear_unlock_delalloc(inode, | |
1416 | &BTRFS_I(inode)->io_tree, | |
1417 | cur_offset, end, locked_page, | |
1418 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1419 | EXTENT_CLEAR_UNLOCK | | |
1420 | EXTENT_CLEAR_DELALLOC | | |
1421 | EXTENT_CLEAR_DIRTY | | |
1422 | EXTENT_SET_WRITEBACK | | |
1423 | EXTENT_END_WRITEBACK); | |
1424 | ||
7ea394f1 | 1425 | btrfs_free_path(path); |
79787eaa | 1426 | return ret; |
be20aa9d CM |
1427 | } |
1428 | ||
d352ac68 CM |
1429 | /* |
1430 | * extent_io.c call back to do delayed allocation processing | |
1431 | */ | |
c8b97818 | 1432 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1433 | u64 start, u64 end, int *page_started, |
1434 | unsigned long *nr_written) | |
be20aa9d | 1435 | { |
be20aa9d | 1436 | int ret; |
7f366cfe | 1437 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1438 | |
7ddf5a42 | 1439 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1440 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1441 | page_started, 1, nr_written); |
7ddf5a42 | 1442 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1443 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1444 | page_started, 0, nr_written); |
7ddf5a42 JB |
1445 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1446 | !(BTRFS_I(inode)->force_compress) && | |
1447 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1448 | ret = cow_file_range(inode, locked_page, start, end, |
1449 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1450 | } else { |
1451 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1452 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1453 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1454 | page_started, nr_written); |
7ddf5a42 | 1455 | } |
b888db2b CM |
1456 | return ret; |
1457 | } | |
1458 | ||
1bf85046 JM |
1459 | static void btrfs_split_extent_hook(struct inode *inode, |
1460 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1461 | { |
0ca1f7ce | 1462 | /* not delalloc, ignore it */ |
9ed74f2d | 1463 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1464 | return; |
9ed74f2d | 1465 | |
9e0baf60 JB |
1466 | spin_lock(&BTRFS_I(inode)->lock); |
1467 | BTRFS_I(inode)->outstanding_extents++; | |
1468 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1469 | } |
1470 | ||
1471 | /* | |
1472 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1473 | * extents so we can keep track of new extents that are just merged onto old | |
1474 | * extents, such as when we are doing sequential writes, so we can properly | |
1475 | * account for the metadata space we'll need. | |
1476 | */ | |
1bf85046 JM |
1477 | static void btrfs_merge_extent_hook(struct inode *inode, |
1478 | struct extent_state *new, | |
1479 | struct extent_state *other) | |
9ed74f2d | 1480 | { |
9ed74f2d JB |
1481 | /* not delalloc, ignore it */ |
1482 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1483 | return; |
9ed74f2d | 1484 | |
9e0baf60 JB |
1485 | spin_lock(&BTRFS_I(inode)->lock); |
1486 | BTRFS_I(inode)->outstanding_extents--; | |
1487 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1488 | } |
1489 | ||
d352ac68 CM |
1490 | /* |
1491 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1492 | * bytes in this file, and to maintain the list of inodes that | |
1493 | * have pending delalloc work to be done. | |
1494 | */ | |
1bf85046 JM |
1495 | static void btrfs_set_bit_hook(struct inode *inode, |
1496 | struct extent_state *state, int *bits) | |
291d673e | 1497 | { |
9ed74f2d | 1498 | |
75eff68e CM |
1499 | /* |
1500 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1501 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1502 | * bit, which is only set or cleared with irqs on |
1503 | */ | |
0ca1f7ce | 1504 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1505 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1506 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1507 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1508 | |
9e0baf60 | 1509 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1510 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1511 | } else { |
1512 | spin_lock(&BTRFS_I(inode)->lock); | |
1513 | BTRFS_I(inode)->outstanding_extents++; | |
1514 | spin_unlock(&BTRFS_I(inode)->lock); | |
1515 | } | |
287a0ab9 | 1516 | |
963d678b MX |
1517 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1518 | root->fs_info->delalloc_batch); | |
df0af1a5 MX |
1519 | spin_lock(&BTRFS_I(inode)->lock); |
1520 | BTRFS_I(inode)->delalloc_bytes += len; | |
1521 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1522 | &BTRFS_I(inode)->runtime_flags)) { | |
1523 | spin_lock(&root->fs_info->delalloc_lock); | |
1524 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1525 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1526 | &root->fs_info->delalloc_inodes); | |
1527 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1528 | &BTRFS_I(inode)->runtime_flags); | |
1529 | } | |
1530 | spin_unlock(&root->fs_info->delalloc_lock); | |
ea8c2819 | 1531 | } |
df0af1a5 | 1532 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1533 | } |
291d673e CM |
1534 | } |
1535 | ||
d352ac68 CM |
1536 | /* |
1537 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1538 | */ | |
1bf85046 JM |
1539 | static void btrfs_clear_bit_hook(struct inode *inode, |
1540 | struct extent_state *state, int *bits) | |
291d673e | 1541 | { |
75eff68e CM |
1542 | /* |
1543 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1544 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1545 | * bit, which is only set or cleared with irqs on |
1546 | */ | |
0ca1f7ce | 1547 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1548 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1549 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1550 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1551 | |
9e0baf60 | 1552 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1553 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1554 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1555 | spin_lock(&BTRFS_I(inode)->lock); | |
1556 | BTRFS_I(inode)->outstanding_extents--; | |
1557 | spin_unlock(&BTRFS_I(inode)->lock); | |
1558 | } | |
0ca1f7ce YZ |
1559 | |
1560 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1561 | btrfs_delalloc_release_metadata(inode, len); | |
1562 | ||
0cb59c99 JB |
1563 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1564 | && do_list) | |
0ca1f7ce | 1565 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1566 | |
963d678b MX |
1567 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1568 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1569 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1570 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1571 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 MX |
1572 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
1573 | &BTRFS_I(inode)->runtime_flags)) { | |
1574 | spin_lock(&root->fs_info->delalloc_lock); | |
1575 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1576 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1577 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1578 | &BTRFS_I(inode)->runtime_flags); | |
1579 | } | |
1580 | spin_unlock(&root->fs_info->delalloc_lock); | |
ea8c2819 | 1581 | } |
df0af1a5 | 1582 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1583 | } |
291d673e CM |
1584 | } |
1585 | ||
d352ac68 CM |
1586 | /* |
1587 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1588 | * we don't create bios that span stripes or chunks | |
1589 | */ | |
239b14b3 | 1590 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1591 | size_t size, struct bio *bio, |
1592 | unsigned long bio_flags) | |
239b14b3 CM |
1593 | { |
1594 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1595 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1596 | u64 length = 0; |
1597 | u64 map_length; | |
239b14b3 CM |
1598 | int ret; |
1599 | ||
771ed689 CM |
1600 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1601 | return 0; | |
1602 | ||
f2d8d74d | 1603 | length = bio->bi_size; |
239b14b3 | 1604 | map_length = length; |
3ec706c8 | 1605 | ret = btrfs_map_block(root->fs_info, READ, logical, |
f188591e | 1606 | &map_length, NULL, 0); |
3ec706c8 | 1607 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1608 | BUG_ON(ret < 0); |
d397712b | 1609 | if (map_length < length + size) |
239b14b3 | 1610 | return 1; |
3444a972 | 1611 | return 0; |
239b14b3 CM |
1612 | } |
1613 | ||
d352ac68 CM |
1614 | /* |
1615 | * in order to insert checksums into the metadata in large chunks, | |
1616 | * we wait until bio submission time. All the pages in the bio are | |
1617 | * checksummed and sums are attached onto the ordered extent record. | |
1618 | * | |
1619 | * At IO completion time the cums attached on the ordered extent record | |
1620 | * are inserted into the btree | |
1621 | */ | |
d397712b CM |
1622 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1623 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1624 | unsigned long bio_flags, |
1625 | u64 bio_offset) | |
065631f6 | 1626 | { |
065631f6 | 1627 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1628 | int ret = 0; |
e015640f | 1629 | |
d20f7043 | 1630 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1631 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1632 | return 0; |
1633 | } | |
e015640f | 1634 | |
4a69a410 CM |
1635 | /* |
1636 | * in order to insert checksums into the metadata in large chunks, | |
1637 | * we wait until bio submission time. All the pages in the bio are | |
1638 | * checksummed and sums are attached onto the ordered extent record. | |
1639 | * | |
1640 | * At IO completion time the cums attached on the ordered extent record | |
1641 | * are inserted into the btree | |
1642 | */ | |
b2950863 | 1643 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1644 | int mirror_num, unsigned long bio_flags, |
1645 | u64 bio_offset) | |
4a69a410 CM |
1646 | { |
1647 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1648 | int ret; |
1649 | ||
1650 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1651 | if (ret) | |
1652 | bio_endio(bio, ret); | |
1653 | return ret; | |
44b8bd7e CM |
1654 | } |
1655 | ||
d352ac68 | 1656 | /* |
cad321ad CM |
1657 | * extent_io.c submission hook. This does the right thing for csum calculation |
1658 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1659 | */ |
b2950863 | 1660 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1661 | int mirror_num, unsigned long bio_flags, |
1662 | u64 bio_offset) | |
44b8bd7e CM |
1663 | { |
1664 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1665 | int ret = 0; | |
19b9bdb0 | 1666 | int skip_sum; |
0417341e | 1667 | int metadata = 0; |
b812ce28 | 1668 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1669 | |
6cbff00f | 1670 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1671 | |
83eea1f1 | 1672 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1673 | metadata = 2; |
1674 | ||
7b6d91da | 1675 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1676 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1677 | if (ret) | |
61891923 | 1678 | goto out; |
5fd02043 | 1679 | |
d20f7043 | 1680 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1681 | ret = btrfs_submit_compressed_read(inode, bio, |
1682 | mirror_num, | |
1683 | bio_flags); | |
1684 | goto out; | |
c2db1073 TI |
1685 | } else if (!skip_sum) { |
1686 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1687 | if (ret) | |
61891923 | 1688 | goto out; |
c2db1073 | 1689 | } |
4d1b5fb4 | 1690 | goto mapit; |
b812ce28 | 1691 | } else if (async && !skip_sum) { |
17d217fe YZ |
1692 | /* csum items have already been cloned */ |
1693 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1694 | goto mapit; | |
19b9bdb0 | 1695 | /* we're doing a write, do the async checksumming */ |
61891923 | 1696 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1697 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1698 | bio_flags, bio_offset, |
1699 | __btrfs_submit_bio_start, | |
4a69a410 | 1700 | __btrfs_submit_bio_done); |
61891923 | 1701 | goto out; |
b812ce28 JB |
1702 | } else if (!skip_sum) { |
1703 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1704 | if (ret) | |
1705 | goto out; | |
19b9bdb0 CM |
1706 | } |
1707 | ||
0b86a832 | 1708 | mapit: |
61891923 SB |
1709 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1710 | ||
1711 | out: | |
1712 | if (ret < 0) | |
1713 | bio_endio(bio, ret); | |
1714 | return ret; | |
065631f6 | 1715 | } |
6885f308 | 1716 | |
d352ac68 CM |
1717 | /* |
1718 | * given a list of ordered sums record them in the inode. This happens | |
1719 | * at IO completion time based on sums calculated at bio submission time. | |
1720 | */ | |
ba1da2f4 | 1721 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1722 | struct inode *inode, u64 file_offset, |
1723 | struct list_head *list) | |
1724 | { | |
e6dcd2dc CM |
1725 | struct btrfs_ordered_sum *sum; |
1726 | ||
c6e30871 | 1727 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1728 | btrfs_csum_file_blocks(trans, |
1729 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1730 | } |
1731 | return 0; | |
1732 | } | |
1733 | ||
2ac55d41 JB |
1734 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1735 | struct extent_state **cached_state) | |
ea8c2819 | 1736 | { |
6c1500f2 | 1737 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1738 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1739 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1740 | } |
1741 | ||
d352ac68 | 1742 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1743 | struct btrfs_writepage_fixup { |
1744 | struct page *page; | |
1745 | struct btrfs_work work; | |
1746 | }; | |
1747 | ||
b2950863 | 1748 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1749 | { |
1750 | struct btrfs_writepage_fixup *fixup; | |
1751 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1752 | struct extent_state *cached_state = NULL; |
247e743c CM |
1753 | struct page *page; |
1754 | struct inode *inode; | |
1755 | u64 page_start; | |
1756 | u64 page_end; | |
87826df0 | 1757 | int ret; |
247e743c CM |
1758 | |
1759 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1760 | page = fixup->page; | |
4a096752 | 1761 | again: |
247e743c CM |
1762 | lock_page(page); |
1763 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1764 | ClearPageChecked(page); | |
1765 | goto out_page; | |
1766 | } | |
1767 | ||
1768 | inode = page->mapping->host; | |
1769 | page_start = page_offset(page); | |
1770 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1771 | ||
2ac55d41 | 1772 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1773 | &cached_state); |
4a096752 CM |
1774 | |
1775 | /* already ordered? We're done */ | |
8b62b72b | 1776 | if (PagePrivate2(page)) |
247e743c | 1777 | goto out; |
4a096752 CM |
1778 | |
1779 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1780 | if (ordered) { | |
2ac55d41 JB |
1781 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1782 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1783 | unlock_page(page); |
1784 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1785 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1786 | goto again; |
1787 | } | |
247e743c | 1788 | |
87826df0 JM |
1789 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1790 | if (ret) { | |
1791 | mapping_set_error(page->mapping, ret); | |
1792 | end_extent_writepage(page, ret, page_start, page_end); | |
1793 | ClearPageChecked(page); | |
1794 | goto out; | |
1795 | } | |
1796 | ||
2ac55d41 | 1797 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1798 | ClearPageChecked(page); |
87826df0 | 1799 | set_page_dirty(page); |
247e743c | 1800 | out: |
2ac55d41 JB |
1801 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1802 | &cached_state, GFP_NOFS); | |
247e743c CM |
1803 | out_page: |
1804 | unlock_page(page); | |
1805 | page_cache_release(page); | |
b897abec | 1806 | kfree(fixup); |
247e743c CM |
1807 | } |
1808 | ||
1809 | /* | |
1810 | * There are a few paths in the higher layers of the kernel that directly | |
1811 | * set the page dirty bit without asking the filesystem if it is a | |
1812 | * good idea. This causes problems because we want to make sure COW | |
1813 | * properly happens and the data=ordered rules are followed. | |
1814 | * | |
c8b97818 | 1815 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1816 | * hasn't been properly setup for IO. We kick off an async process |
1817 | * to fix it up. The async helper will wait for ordered extents, set | |
1818 | * the delalloc bit and make it safe to write the page. | |
1819 | */ | |
b2950863 | 1820 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1821 | { |
1822 | struct inode *inode = page->mapping->host; | |
1823 | struct btrfs_writepage_fixup *fixup; | |
1824 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1825 | |
8b62b72b CM |
1826 | /* this page is properly in the ordered list */ |
1827 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1828 | return 0; |
1829 | ||
1830 | if (PageChecked(page)) | |
1831 | return -EAGAIN; | |
1832 | ||
1833 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1834 | if (!fixup) | |
1835 | return -EAGAIN; | |
f421950f | 1836 | |
247e743c CM |
1837 | SetPageChecked(page); |
1838 | page_cache_get(page); | |
1839 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1840 | fixup->page = page; | |
1841 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1842 | return -EBUSY; |
247e743c CM |
1843 | } |
1844 | ||
d899e052 YZ |
1845 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1846 | struct inode *inode, u64 file_pos, | |
1847 | u64 disk_bytenr, u64 disk_num_bytes, | |
1848 | u64 num_bytes, u64 ram_bytes, | |
1849 | u8 compression, u8 encryption, | |
1850 | u16 other_encoding, int extent_type) | |
1851 | { | |
1852 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1853 | struct btrfs_file_extent_item *fi; | |
1854 | struct btrfs_path *path; | |
1855 | struct extent_buffer *leaf; | |
1856 | struct btrfs_key ins; | |
d899e052 YZ |
1857 | int ret; |
1858 | ||
1859 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1860 | if (!path) |
1861 | return -ENOMEM; | |
d899e052 | 1862 | |
b9473439 | 1863 | path->leave_spinning = 1; |
a1ed835e CM |
1864 | |
1865 | /* | |
1866 | * we may be replacing one extent in the tree with another. | |
1867 | * The new extent is pinned in the extent map, and we don't want | |
1868 | * to drop it from the cache until it is completely in the btree. | |
1869 | * | |
1870 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1871 | * the caller is expected to unpin it and allow it to be merged | |
1872 | * with the others. | |
1873 | */ | |
5dc562c5 | 1874 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1875 | file_pos + num_bytes, 0); |
79787eaa JM |
1876 | if (ret) |
1877 | goto out; | |
d899e052 | 1878 | |
33345d01 | 1879 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1880 | ins.offset = file_pos; |
1881 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1882 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1883 | if (ret) |
1884 | goto out; | |
d899e052 YZ |
1885 | leaf = path->nodes[0]; |
1886 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1887 | struct btrfs_file_extent_item); | |
1888 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1889 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1890 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1891 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1892 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1893 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1894 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1895 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1896 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1897 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1898 | |
d899e052 | 1899 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1900 | btrfs_release_path(path); |
d899e052 YZ |
1901 | |
1902 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1903 | |
1904 | ins.objectid = disk_bytenr; | |
1905 | ins.offset = disk_num_bytes; | |
1906 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1907 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1908 | root->root_key.objectid, | |
33345d01 | 1909 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1910 | out: |
d899e052 | 1911 | btrfs_free_path(path); |
b9473439 | 1912 | |
79787eaa | 1913 | return ret; |
d899e052 YZ |
1914 | } |
1915 | ||
5d13a98f CM |
1916 | /* |
1917 | * helper function for btrfs_finish_ordered_io, this | |
1918 | * just reads in some of the csum leaves to prime them into ram | |
1919 | * before we start the transaction. It limits the amount of btree | |
1920 | * reads required while inside the transaction. | |
1921 | */ | |
d352ac68 CM |
1922 | /* as ordered data IO finishes, this gets called so we can finish |
1923 | * an ordered extent if the range of bytes in the file it covers are | |
1924 | * fully written. | |
1925 | */ | |
5fd02043 | 1926 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1927 | { |
5fd02043 | 1928 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1929 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1930 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1931 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1932 | struct extent_state *cached_state = NULL; |
261507a0 | 1933 | int compress_type = 0; |
e6dcd2dc | 1934 | int ret; |
82d5902d | 1935 | bool nolock; |
e6dcd2dc | 1936 | |
83eea1f1 | 1937 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 1938 | |
5fd02043 JB |
1939 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1940 | ret = -EIO; | |
1941 | goto out; | |
1942 | } | |
1943 | ||
c2167754 | 1944 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1945 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
1946 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1947 | if (nolock) | |
1948 | trans = btrfs_join_transaction_nolock(root); | |
1949 | else | |
1950 | trans = btrfs_join_transaction(root); | |
1951 | if (IS_ERR(trans)) { | |
1952 | ret = PTR_ERR(trans); | |
1953 | trans = NULL; | |
1954 | goto out; | |
c2167754 | 1955 | } |
6c760c07 JB |
1956 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
1957 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
1958 | if (ret) /* -ENOMEM or corruption */ | |
1959 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1960 | goto out; |
1961 | } | |
e6dcd2dc | 1962 | |
2ac55d41 JB |
1963 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1964 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1965 | 0, &cached_state); |
e6dcd2dc | 1966 | |
0cb59c99 | 1967 | if (nolock) |
7a7eaa40 | 1968 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1969 | else |
7a7eaa40 | 1970 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1971 | if (IS_ERR(trans)) { |
1972 | ret = PTR_ERR(trans); | |
1973 | trans = NULL; | |
1974 | goto out_unlock; | |
1975 | } | |
0ca1f7ce | 1976 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1977 | |
c8b97818 | 1978 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1979 | compress_type = ordered_extent->compress_type; |
d899e052 | 1980 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1981 | BUG_ON(compress_type); |
920bbbfb | 1982 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1983 | ordered_extent->file_offset, |
1984 | ordered_extent->file_offset + | |
1985 | ordered_extent->len); | |
d899e052 | 1986 | } else { |
0af3d00b | 1987 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1988 | ret = insert_reserved_file_extent(trans, inode, |
1989 | ordered_extent->file_offset, | |
1990 | ordered_extent->start, | |
1991 | ordered_extent->disk_len, | |
1992 | ordered_extent->len, | |
1993 | ordered_extent->len, | |
261507a0 | 1994 | compress_type, 0, 0, |
d899e052 | 1995 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 1996 | } |
5dc562c5 JB |
1997 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1998 | ordered_extent->file_offset, ordered_extent->len, | |
1999 | trans->transid); | |
79787eaa JM |
2000 | if (ret < 0) { |
2001 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2002 | goto out_unlock; |
79787eaa | 2003 | } |
2ac55d41 | 2004 | |
e6dcd2dc CM |
2005 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2006 | &ordered_extent->list); | |
2007 | ||
6c760c07 JB |
2008 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2009 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2010 | if (ret) { /* -ENOMEM or corruption */ | |
2011 | btrfs_abort_transaction(trans, root, ret); | |
2012 | goto out_unlock; | |
1ef30be1 JB |
2013 | } |
2014 | ret = 0; | |
5fd02043 JB |
2015 | out_unlock: |
2016 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2017 | ordered_extent->file_offset + | |
2018 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2019 | out: |
5b0e95bf | 2020 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2021 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2022 | if (trans) |
2023 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2024 | |
0bec9ef5 | 2025 | if (ret) { |
5fd02043 JB |
2026 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, |
2027 | ordered_extent->file_offset + | |
2028 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
2029 | ||
0bec9ef5 JB |
2030 | /* |
2031 | * If the ordered extent had an IOERR or something else went | |
2032 | * wrong we need to return the space for this ordered extent | |
2033 | * back to the allocator. | |
2034 | */ | |
2035 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2036 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) | |
2037 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
2038 | ordered_extent->disk_len); | |
2039 | } | |
2040 | ||
2041 | ||
5fd02043 | 2042 | /* |
8bad3c02 LB |
2043 | * This needs to be done to make sure anybody waiting knows we are done |
2044 | * updating everything for this ordered extent. | |
5fd02043 JB |
2045 | */ |
2046 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2047 | ||
e6dcd2dc CM |
2048 | /* once for us */ |
2049 | btrfs_put_ordered_extent(ordered_extent); | |
2050 | /* once for the tree */ | |
2051 | btrfs_put_ordered_extent(ordered_extent); | |
2052 | ||
5fd02043 JB |
2053 | return ret; |
2054 | } | |
2055 | ||
2056 | static void finish_ordered_fn(struct btrfs_work *work) | |
2057 | { | |
2058 | struct btrfs_ordered_extent *ordered_extent; | |
2059 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2060 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2061 | } |
2062 | ||
b2950863 | 2063 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2064 | struct extent_state *state, int uptodate) |
2065 | { | |
5fd02043 JB |
2066 | struct inode *inode = page->mapping->host; |
2067 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2068 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2069 | struct btrfs_workers *workers; | |
2070 | ||
1abe9b8a | 2071 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2072 | ||
8b62b72b | 2073 | ClearPagePrivate2(page); |
5fd02043 JB |
2074 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2075 | end - start + 1, uptodate)) | |
2076 | return 0; | |
2077 | ||
2078 | ordered_extent->work.func = finish_ordered_fn; | |
2079 | ordered_extent->work.flags = 0; | |
2080 | ||
83eea1f1 | 2081 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2082 | workers = &root->fs_info->endio_freespace_worker; |
2083 | else | |
2084 | workers = &root->fs_info->endio_write_workers; | |
2085 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2086 | ||
2087 | return 0; | |
211f90e6 CM |
2088 | } |
2089 | ||
d352ac68 CM |
2090 | /* |
2091 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2092 | * if there's a match, we allow the bio to finish. If not, the code in |
2093 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2094 | */ |
b2950863 | 2095 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2096 | struct extent_state *state, int mirror) |
07157aac | 2097 | { |
4eee4fa4 | 2098 | size_t offset = start - page_offset(page); |
07157aac | 2099 | struct inode *inode = page->mapping->host; |
d1310b2e | 2100 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2101 | char *kaddr; |
aadfeb6e | 2102 | u64 private = ~(u32)0; |
07157aac | 2103 | int ret; |
ff79f819 CM |
2104 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2105 | u32 csum = ~(u32)0; | |
d1310b2e | 2106 | |
d20f7043 CM |
2107 | if (PageChecked(page)) { |
2108 | ClearPageChecked(page); | |
2109 | goto good; | |
2110 | } | |
6cbff00f CH |
2111 | |
2112 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2113 | goto good; |
17d217fe YZ |
2114 | |
2115 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2116 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2117 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2118 | GFP_NOFS); | |
b6cda9bc | 2119 | return 0; |
17d217fe | 2120 | } |
d20f7043 | 2121 | |
c2e639f0 | 2122 | if (state && state->start == start) { |
70dec807 CM |
2123 | private = state->private; |
2124 | ret = 0; | |
2125 | } else { | |
2126 | ret = get_state_private(io_tree, start, &private); | |
2127 | } | |
7ac687d9 | 2128 | kaddr = kmap_atomic(page); |
d397712b | 2129 | if (ret) |
07157aac | 2130 | goto zeroit; |
d397712b | 2131 | |
ff79f819 CM |
2132 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2133 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2134 | if (csum != private) |
07157aac | 2135 | goto zeroit; |
d397712b | 2136 | |
7ac687d9 | 2137 | kunmap_atomic(kaddr); |
d20f7043 | 2138 | good: |
07157aac CM |
2139 | return 0; |
2140 | ||
2141 | zeroit: | |
945d8962 | 2142 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2143 | "private %llu\n", |
2144 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2145 | (unsigned long long)start, csum, |
2146 | (unsigned long long)private); | |
db94535d CM |
2147 | memset(kaddr + offset, 1, end - start + 1); |
2148 | flush_dcache_page(page); | |
7ac687d9 | 2149 | kunmap_atomic(kaddr); |
3b951516 CM |
2150 | if (private == 0) |
2151 | return 0; | |
7e38326f | 2152 | return -EIO; |
07157aac | 2153 | } |
b888db2b | 2154 | |
24bbcf04 YZ |
2155 | struct delayed_iput { |
2156 | struct list_head list; | |
2157 | struct inode *inode; | |
2158 | }; | |
2159 | ||
79787eaa JM |
2160 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2161 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2162 | void btrfs_add_delayed_iput(struct inode *inode) |
2163 | { | |
2164 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2165 | struct delayed_iput *delayed; | |
2166 | ||
2167 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2168 | return; | |
2169 | ||
2170 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2171 | delayed->inode = inode; | |
2172 | ||
2173 | spin_lock(&fs_info->delayed_iput_lock); | |
2174 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2175 | spin_unlock(&fs_info->delayed_iput_lock); | |
2176 | } | |
2177 | ||
2178 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2179 | { | |
2180 | LIST_HEAD(list); | |
2181 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2182 | struct delayed_iput *delayed; | |
2183 | int empty; | |
2184 | ||
2185 | spin_lock(&fs_info->delayed_iput_lock); | |
2186 | empty = list_empty(&fs_info->delayed_iputs); | |
2187 | spin_unlock(&fs_info->delayed_iput_lock); | |
2188 | if (empty) | |
2189 | return; | |
2190 | ||
24bbcf04 YZ |
2191 | spin_lock(&fs_info->delayed_iput_lock); |
2192 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2193 | spin_unlock(&fs_info->delayed_iput_lock); | |
2194 | ||
2195 | while (!list_empty(&list)) { | |
2196 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2197 | list_del(&delayed->list); | |
2198 | iput(delayed->inode); | |
2199 | kfree(delayed); | |
2200 | } | |
24bbcf04 YZ |
2201 | } |
2202 | ||
d68fc57b YZ |
2203 | enum btrfs_orphan_cleanup_state { |
2204 | ORPHAN_CLEANUP_STARTED = 1, | |
2205 | ORPHAN_CLEANUP_DONE = 2, | |
2206 | }; | |
2207 | ||
2208 | /* | |
42b2aa86 | 2209 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2210 | * files in the subvolume, it removes orphan item and frees block_rsv |
2211 | * structure. | |
2212 | */ | |
2213 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2214 | struct btrfs_root *root) | |
2215 | { | |
90290e19 | 2216 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2217 | int ret; |
2218 | ||
8a35d95f | 2219 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2220 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2221 | return; | |
2222 | ||
90290e19 | 2223 | spin_lock(&root->orphan_lock); |
8a35d95f | 2224 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2225 | spin_unlock(&root->orphan_lock); |
2226 | return; | |
2227 | } | |
2228 | ||
2229 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2230 | spin_unlock(&root->orphan_lock); | |
2231 | return; | |
2232 | } | |
2233 | ||
2234 | block_rsv = root->orphan_block_rsv; | |
2235 | root->orphan_block_rsv = NULL; | |
2236 | spin_unlock(&root->orphan_lock); | |
2237 | ||
d68fc57b YZ |
2238 | if (root->orphan_item_inserted && |
2239 | btrfs_root_refs(&root->root_item) > 0) { | |
2240 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2241 | root->root_key.objectid); | |
2242 | BUG_ON(ret); | |
2243 | root->orphan_item_inserted = 0; | |
2244 | } | |
2245 | ||
90290e19 JB |
2246 | if (block_rsv) { |
2247 | WARN_ON(block_rsv->size > 0); | |
2248 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2249 | } |
2250 | } | |
2251 | ||
7b128766 JB |
2252 | /* |
2253 | * This creates an orphan entry for the given inode in case something goes | |
2254 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2255 | * |
2256 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2257 | * this function. | |
7b128766 JB |
2258 | */ |
2259 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2260 | { | |
2261 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2262 | struct btrfs_block_rsv *block_rsv = NULL; |
2263 | int reserve = 0; | |
2264 | int insert = 0; | |
2265 | int ret; | |
7b128766 | 2266 | |
d68fc57b | 2267 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2268 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2269 | if (!block_rsv) |
2270 | return -ENOMEM; | |
d68fc57b | 2271 | } |
7b128766 | 2272 | |
d68fc57b YZ |
2273 | spin_lock(&root->orphan_lock); |
2274 | if (!root->orphan_block_rsv) { | |
2275 | root->orphan_block_rsv = block_rsv; | |
2276 | } else if (block_rsv) { | |
2277 | btrfs_free_block_rsv(root, block_rsv); | |
2278 | block_rsv = NULL; | |
7b128766 | 2279 | } |
7b128766 | 2280 | |
8a35d95f JB |
2281 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2282 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2283 | #if 0 |
2284 | /* | |
2285 | * For proper ENOSPC handling, we should do orphan | |
2286 | * cleanup when mounting. But this introduces backward | |
2287 | * compatibility issue. | |
2288 | */ | |
2289 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2290 | insert = 2; | |
2291 | else | |
2292 | insert = 1; | |
2293 | #endif | |
2294 | insert = 1; | |
321f0e70 | 2295 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2296 | } |
2297 | ||
72ac3c0d JB |
2298 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2299 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2300 | reserve = 1; |
d68fc57b | 2301 | spin_unlock(&root->orphan_lock); |
7b128766 | 2302 | |
d68fc57b YZ |
2303 | /* grab metadata reservation from transaction handle */ |
2304 | if (reserve) { | |
2305 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2306 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2307 | } |
7b128766 | 2308 | |
d68fc57b YZ |
2309 | /* insert an orphan item to track this unlinked/truncated file */ |
2310 | if (insert >= 1) { | |
33345d01 | 2311 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2312 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2313 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2314 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2315 | btrfs_abort_transaction(trans, root, ret); |
2316 | return ret; | |
2317 | } | |
2318 | ret = 0; | |
d68fc57b YZ |
2319 | } |
2320 | ||
2321 | /* insert an orphan item to track subvolume contains orphan files */ | |
2322 | if (insert >= 2) { | |
2323 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2324 | root->root_key.objectid); | |
79787eaa JM |
2325 | if (ret && ret != -EEXIST) { |
2326 | btrfs_abort_transaction(trans, root, ret); | |
2327 | return ret; | |
2328 | } | |
d68fc57b YZ |
2329 | } |
2330 | return 0; | |
7b128766 JB |
2331 | } |
2332 | ||
2333 | /* | |
2334 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2335 | * item for this particular inode. | |
2336 | */ | |
2337 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2338 | { | |
2339 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2340 | int delete_item = 0; |
2341 | int release_rsv = 0; | |
7b128766 JB |
2342 | int ret = 0; |
2343 | ||
d68fc57b | 2344 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2345 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2346 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2347 | delete_item = 1; |
7b128766 | 2348 | |
72ac3c0d JB |
2349 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2350 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2351 | release_rsv = 1; |
d68fc57b | 2352 | spin_unlock(&root->orphan_lock); |
7b128766 | 2353 | |
d68fc57b | 2354 | if (trans && delete_item) { |
33345d01 | 2355 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2356 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2357 | } |
7b128766 | 2358 | |
8a35d95f | 2359 | if (release_rsv) { |
d68fc57b | 2360 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2361 | atomic_dec(&root->orphan_inodes); |
2362 | } | |
7b128766 | 2363 | |
d68fc57b | 2364 | return 0; |
7b128766 JB |
2365 | } |
2366 | ||
2367 | /* | |
2368 | * this cleans up any orphans that may be left on the list from the last use | |
2369 | * of this root. | |
2370 | */ | |
66b4ffd1 | 2371 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2372 | { |
2373 | struct btrfs_path *path; | |
2374 | struct extent_buffer *leaf; | |
7b128766 JB |
2375 | struct btrfs_key key, found_key; |
2376 | struct btrfs_trans_handle *trans; | |
2377 | struct inode *inode; | |
8f6d7f4f | 2378 | u64 last_objectid = 0; |
7b128766 JB |
2379 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2380 | ||
d68fc57b | 2381 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2382 | return 0; |
c71bf099 YZ |
2383 | |
2384 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2385 | if (!path) { |
2386 | ret = -ENOMEM; | |
2387 | goto out; | |
2388 | } | |
7b128766 JB |
2389 | path->reada = -1; |
2390 | ||
2391 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2392 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2393 | key.offset = (u64)-1; | |
2394 | ||
7b128766 JB |
2395 | while (1) { |
2396 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2397 | if (ret < 0) |
2398 | goto out; | |
7b128766 JB |
2399 | |
2400 | /* | |
2401 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2402 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2403 | * find the key and see if we have stuff that matches |
2404 | */ | |
2405 | if (ret > 0) { | |
66b4ffd1 | 2406 | ret = 0; |
7b128766 JB |
2407 | if (path->slots[0] == 0) |
2408 | break; | |
2409 | path->slots[0]--; | |
2410 | } | |
2411 | ||
2412 | /* pull out the item */ | |
2413 | leaf = path->nodes[0]; | |
7b128766 JB |
2414 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2415 | ||
2416 | /* make sure the item matches what we want */ | |
2417 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2418 | break; | |
2419 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2420 | break; | |
2421 | ||
2422 | /* release the path since we're done with it */ | |
b3b4aa74 | 2423 | btrfs_release_path(path); |
7b128766 JB |
2424 | |
2425 | /* | |
2426 | * this is where we are basically btrfs_lookup, without the | |
2427 | * crossing root thing. we store the inode number in the | |
2428 | * offset of the orphan item. | |
2429 | */ | |
8f6d7f4f JB |
2430 | |
2431 | if (found_key.offset == last_objectid) { | |
2432 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2433 | "stopping orphan cleanup\n"); | |
2434 | ret = -EINVAL; | |
2435 | goto out; | |
2436 | } | |
2437 | ||
2438 | last_objectid = found_key.offset; | |
2439 | ||
5d4f98a2 YZ |
2440 | found_key.objectid = found_key.offset; |
2441 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2442 | found_key.offset = 0; | |
73f73415 | 2443 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2444 | ret = PTR_RET(inode); |
2445 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2446 | goto out; |
7b128766 | 2447 | |
f8e9e0b0 AJ |
2448 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2449 | struct btrfs_root *dead_root; | |
2450 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2451 | int is_dead_root = 0; | |
2452 | ||
2453 | /* | |
2454 | * this is an orphan in the tree root. Currently these | |
2455 | * could come from 2 sources: | |
2456 | * a) a snapshot deletion in progress | |
2457 | * b) a free space cache inode | |
2458 | * We need to distinguish those two, as the snapshot | |
2459 | * orphan must not get deleted. | |
2460 | * find_dead_roots already ran before us, so if this | |
2461 | * is a snapshot deletion, we should find the root | |
2462 | * in the dead_roots list | |
2463 | */ | |
2464 | spin_lock(&fs_info->trans_lock); | |
2465 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2466 | root_list) { | |
2467 | if (dead_root->root_key.objectid == | |
2468 | found_key.objectid) { | |
2469 | is_dead_root = 1; | |
2470 | break; | |
2471 | } | |
2472 | } | |
2473 | spin_unlock(&fs_info->trans_lock); | |
2474 | if (is_dead_root) { | |
2475 | /* prevent this orphan from being found again */ | |
2476 | key.offset = found_key.objectid - 1; | |
2477 | continue; | |
2478 | } | |
2479 | } | |
7b128766 | 2480 | /* |
a8c9e576 JB |
2481 | * Inode is already gone but the orphan item is still there, |
2482 | * kill the orphan item. | |
7b128766 | 2483 | */ |
a8c9e576 JB |
2484 | if (ret == -ESTALE) { |
2485 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2486 | if (IS_ERR(trans)) { |
2487 | ret = PTR_ERR(trans); | |
2488 | goto out; | |
2489 | } | |
8a35d95f JB |
2490 | printk(KERN_ERR "auto deleting %Lu\n", |
2491 | found_key.objectid); | |
a8c9e576 JB |
2492 | ret = btrfs_del_orphan_item(trans, root, |
2493 | found_key.objectid); | |
79787eaa | 2494 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2495 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2496 | continue; |
2497 | } | |
2498 | ||
a8c9e576 JB |
2499 | /* |
2500 | * add this inode to the orphan list so btrfs_orphan_del does | |
2501 | * the proper thing when we hit it | |
2502 | */ | |
8a35d95f JB |
2503 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2504 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 2505 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 2506 | |
7b128766 JB |
2507 | /* if we have links, this was a truncate, lets do that */ |
2508 | if (inode->i_nlink) { | |
a41ad394 JB |
2509 | if (!S_ISREG(inode->i_mode)) { |
2510 | WARN_ON(1); | |
2511 | iput(inode); | |
2512 | continue; | |
2513 | } | |
7b128766 | 2514 | nr_truncate++; |
f3fe820c JB |
2515 | |
2516 | /* 1 for the orphan item deletion. */ | |
2517 | trans = btrfs_start_transaction(root, 1); | |
2518 | if (IS_ERR(trans)) { | |
2519 | ret = PTR_ERR(trans); | |
2520 | goto out; | |
2521 | } | |
2522 | ret = btrfs_orphan_add(trans, inode); | |
2523 | btrfs_end_transaction(trans, root); | |
2524 | if (ret) | |
2525 | goto out; | |
2526 | ||
66b4ffd1 | 2527 | ret = btrfs_truncate(inode); |
7b128766 JB |
2528 | } else { |
2529 | nr_unlink++; | |
2530 | } | |
2531 | ||
2532 | /* this will do delete_inode and everything for us */ | |
2533 | iput(inode); | |
66b4ffd1 JB |
2534 | if (ret) |
2535 | goto out; | |
7b128766 | 2536 | } |
3254c876 MX |
2537 | /* release the path since we're done with it */ |
2538 | btrfs_release_path(path); | |
2539 | ||
d68fc57b YZ |
2540 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2541 | ||
2542 | if (root->orphan_block_rsv) | |
2543 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2544 | (u64)-1); | |
2545 | ||
2546 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2547 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2548 | if (!IS_ERR(trans)) |
2549 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2550 | } |
7b128766 JB |
2551 | |
2552 | if (nr_unlink) | |
2553 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2554 | if (nr_truncate) | |
2555 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2556 | |
2557 | out: | |
2558 | if (ret) | |
2559 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2560 | btrfs_free_path(path); | |
2561 | return ret; | |
7b128766 JB |
2562 | } |
2563 | ||
46a53cca CM |
2564 | /* |
2565 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2566 | * don't find any xattrs, we know there can't be any acls. | |
2567 | * | |
2568 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2569 | */ | |
2570 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2571 | int slot, u64 objectid) | |
2572 | { | |
2573 | u32 nritems = btrfs_header_nritems(leaf); | |
2574 | struct btrfs_key found_key; | |
2575 | int scanned = 0; | |
2576 | ||
2577 | slot++; | |
2578 | while (slot < nritems) { | |
2579 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2580 | ||
2581 | /* we found a different objectid, there must not be acls */ | |
2582 | if (found_key.objectid != objectid) | |
2583 | return 0; | |
2584 | ||
2585 | /* we found an xattr, assume we've got an acl */ | |
2586 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2587 | return 1; | |
2588 | ||
2589 | /* | |
2590 | * we found a key greater than an xattr key, there can't | |
2591 | * be any acls later on | |
2592 | */ | |
2593 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2594 | return 0; | |
2595 | ||
2596 | slot++; | |
2597 | scanned++; | |
2598 | ||
2599 | /* | |
2600 | * it goes inode, inode backrefs, xattrs, extents, | |
2601 | * so if there are a ton of hard links to an inode there can | |
2602 | * be a lot of backrefs. Don't waste time searching too hard, | |
2603 | * this is just an optimization | |
2604 | */ | |
2605 | if (scanned >= 8) | |
2606 | break; | |
2607 | } | |
2608 | /* we hit the end of the leaf before we found an xattr or | |
2609 | * something larger than an xattr. We have to assume the inode | |
2610 | * has acls | |
2611 | */ | |
2612 | return 1; | |
2613 | } | |
2614 | ||
d352ac68 CM |
2615 | /* |
2616 | * read an inode from the btree into the in-memory inode | |
2617 | */ | |
5d4f98a2 | 2618 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2619 | { |
2620 | struct btrfs_path *path; | |
5f39d397 | 2621 | struct extent_buffer *leaf; |
39279cc3 | 2622 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2623 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2624 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2625 | struct btrfs_key location; | |
46a53cca | 2626 | int maybe_acls; |
618e21d5 | 2627 | u32 rdev; |
39279cc3 | 2628 | int ret; |
2f7e33d4 MX |
2629 | bool filled = false; |
2630 | ||
2631 | ret = btrfs_fill_inode(inode, &rdev); | |
2632 | if (!ret) | |
2633 | filled = true; | |
39279cc3 CM |
2634 | |
2635 | path = btrfs_alloc_path(); | |
1748f843 MF |
2636 | if (!path) |
2637 | goto make_bad; | |
2638 | ||
d90c7321 | 2639 | path->leave_spinning = 1; |
39279cc3 | 2640 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2641 | |
39279cc3 | 2642 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2643 | if (ret) |
39279cc3 | 2644 | goto make_bad; |
39279cc3 | 2645 | |
5f39d397 | 2646 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2647 | |
2648 | if (filled) | |
2649 | goto cache_acl; | |
2650 | ||
5f39d397 CM |
2651 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2652 | struct btrfs_inode_item); | |
5f39d397 | 2653 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2654 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
2655 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
2656 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 2657 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2658 | |
2659 | tspec = btrfs_inode_atime(inode_item); | |
2660 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2661 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2662 | ||
2663 | tspec = btrfs_inode_mtime(inode_item); | |
2664 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2665 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2666 | ||
2667 | tspec = btrfs_inode_ctime(inode_item); | |
2668 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2669 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2670 | ||
a76a3cd4 | 2671 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2672 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
2673 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
2674 | ||
2675 | /* | |
2676 | * If we were modified in the current generation and evicted from memory | |
2677 | * and then re-read we need to do a full sync since we don't have any | |
2678 | * idea about which extents were modified before we were evicted from | |
2679 | * cache. | |
2680 | */ | |
2681 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
2682 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2683 | &BTRFS_I(inode)->runtime_flags); | |
2684 | ||
0c4d2d95 | 2685 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2686 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2687 | inode->i_rdev = 0; |
5f39d397 CM |
2688 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2689 | ||
aec7477b | 2690 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2691 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2692 | cache_acl: |
46a53cca CM |
2693 | /* |
2694 | * try to precache a NULL acl entry for files that don't have | |
2695 | * any xattrs or acls | |
2696 | */ | |
33345d01 LZ |
2697 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2698 | btrfs_ino(inode)); | |
72c04902 AV |
2699 | if (!maybe_acls) |
2700 | cache_no_acl(inode); | |
46a53cca | 2701 | |
39279cc3 | 2702 | btrfs_free_path(path); |
39279cc3 | 2703 | |
39279cc3 | 2704 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2705 | case S_IFREG: |
2706 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2707 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2708 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2709 | inode->i_fop = &btrfs_file_operations; |
2710 | inode->i_op = &btrfs_file_inode_operations; | |
2711 | break; | |
2712 | case S_IFDIR: | |
2713 | inode->i_fop = &btrfs_dir_file_operations; | |
2714 | if (root == root->fs_info->tree_root) | |
2715 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2716 | else | |
2717 | inode->i_op = &btrfs_dir_inode_operations; | |
2718 | break; | |
2719 | case S_IFLNK: | |
2720 | inode->i_op = &btrfs_symlink_inode_operations; | |
2721 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2722 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2723 | break; |
618e21d5 | 2724 | default: |
0279b4cd | 2725 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2726 | init_special_inode(inode, inode->i_mode, rdev); |
2727 | break; | |
39279cc3 | 2728 | } |
6cbff00f CH |
2729 | |
2730 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2731 | return; |
2732 | ||
2733 | make_bad: | |
39279cc3 | 2734 | btrfs_free_path(path); |
39279cc3 CM |
2735 | make_bad_inode(inode); |
2736 | } | |
2737 | ||
d352ac68 CM |
2738 | /* |
2739 | * given a leaf and an inode, copy the inode fields into the leaf | |
2740 | */ | |
e02119d5 CM |
2741 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2742 | struct extent_buffer *leaf, | |
5f39d397 | 2743 | struct btrfs_inode_item *item, |
39279cc3 CM |
2744 | struct inode *inode) |
2745 | { | |
51fab693 LB |
2746 | struct btrfs_map_token token; |
2747 | ||
2748 | btrfs_init_map_token(&token); | |
2749 | ||
2750 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); | |
2751 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
2752 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
2753 | &token); | |
2754 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
2755 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
2756 | ||
2757 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item), | |
2758 | inode->i_atime.tv_sec, &token); | |
2759 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2760 | inode->i_atime.tv_nsec, &token); | |
2761 | ||
2762 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2763 | inode->i_mtime.tv_sec, &token); | |
2764 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2765 | inode->i_mtime.tv_nsec, &token); | |
2766 | ||
2767 | btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2768 | inode->i_ctime.tv_sec, &token); | |
2769 | btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2770 | inode->i_ctime.tv_nsec, &token); | |
2771 | ||
2772 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), | |
2773 | &token); | |
2774 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
2775 | &token); | |
2776 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
2777 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
2778 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
2779 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
2780 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
2781 | } |
2782 | ||
d352ac68 CM |
2783 | /* |
2784 | * copy everything in the in-memory inode into the btree. | |
2785 | */ | |
2115133f | 2786 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2787 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2788 | { |
2789 | struct btrfs_inode_item *inode_item; | |
2790 | struct btrfs_path *path; | |
5f39d397 | 2791 | struct extent_buffer *leaf; |
39279cc3 CM |
2792 | int ret; |
2793 | ||
2794 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2795 | if (!path) |
2796 | return -ENOMEM; | |
2797 | ||
b9473439 | 2798 | path->leave_spinning = 1; |
16cdcec7 MX |
2799 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2800 | 1); | |
39279cc3 CM |
2801 | if (ret) { |
2802 | if (ret > 0) | |
2803 | ret = -ENOENT; | |
2804 | goto failed; | |
2805 | } | |
2806 | ||
b4ce94de | 2807 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2808 | leaf = path->nodes[0]; |
2809 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2810 | struct btrfs_inode_item); |
39279cc3 | 2811 | |
e02119d5 | 2812 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2813 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2814 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2815 | ret = 0; |
2816 | failed: | |
39279cc3 CM |
2817 | btrfs_free_path(path); |
2818 | return ret; | |
2819 | } | |
2820 | ||
2115133f CM |
2821 | /* |
2822 | * copy everything in the in-memory inode into the btree. | |
2823 | */ | |
2824 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2825 | struct btrfs_root *root, struct inode *inode) | |
2826 | { | |
2827 | int ret; | |
2828 | ||
2829 | /* | |
2830 | * If the inode is a free space inode, we can deadlock during commit | |
2831 | * if we put it into the delayed code. | |
2832 | * | |
2833 | * The data relocation inode should also be directly updated | |
2834 | * without delay | |
2835 | */ | |
83eea1f1 | 2836 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 2837 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
2838 | btrfs_update_root_times(trans, root); |
2839 | ||
2115133f CM |
2840 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2841 | if (!ret) | |
2842 | btrfs_set_inode_last_trans(trans, inode); | |
2843 | return ret; | |
2844 | } | |
2845 | ||
2846 | return btrfs_update_inode_item(trans, root, inode); | |
2847 | } | |
2848 | ||
be6aef60 JB |
2849 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
2850 | struct btrfs_root *root, | |
2851 | struct inode *inode) | |
2115133f CM |
2852 | { |
2853 | int ret; | |
2854 | ||
2855 | ret = btrfs_update_inode(trans, root, inode); | |
2856 | if (ret == -ENOSPC) | |
2857 | return btrfs_update_inode_item(trans, root, inode); | |
2858 | return ret; | |
2859 | } | |
2860 | ||
d352ac68 CM |
2861 | /* |
2862 | * unlink helper that gets used here in inode.c and in the tree logging | |
2863 | * recovery code. It remove a link in a directory with a given name, and | |
2864 | * also drops the back refs in the inode to the directory | |
2865 | */ | |
92986796 AV |
2866 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2867 | struct btrfs_root *root, | |
2868 | struct inode *dir, struct inode *inode, | |
2869 | const char *name, int name_len) | |
39279cc3 CM |
2870 | { |
2871 | struct btrfs_path *path; | |
39279cc3 | 2872 | int ret = 0; |
5f39d397 | 2873 | struct extent_buffer *leaf; |
39279cc3 | 2874 | struct btrfs_dir_item *di; |
5f39d397 | 2875 | struct btrfs_key key; |
aec7477b | 2876 | u64 index; |
33345d01 LZ |
2877 | u64 ino = btrfs_ino(inode); |
2878 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2879 | |
2880 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2881 | if (!path) { |
2882 | ret = -ENOMEM; | |
554233a6 | 2883 | goto out; |
54aa1f4d CM |
2884 | } |
2885 | ||
b9473439 | 2886 | path->leave_spinning = 1; |
33345d01 | 2887 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2888 | name, name_len, -1); |
2889 | if (IS_ERR(di)) { | |
2890 | ret = PTR_ERR(di); | |
2891 | goto err; | |
2892 | } | |
2893 | if (!di) { | |
2894 | ret = -ENOENT; | |
2895 | goto err; | |
2896 | } | |
5f39d397 CM |
2897 | leaf = path->nodes[0]; |
2898 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2899 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2900 | if (ret) |
2901 | goto err; | |
b3b4aa74 | 2902 | btrfs_release_path(path); |
39279cc3 | 2903 | |
33345d01 LZ |
2904 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2905 | dir_ino, &index); | |
aec7477b | 2906 | if (ret) { |
d397712b | 2907 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2908 | "inode %llu parent %llu\n", name_len, name, |
2909 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2910 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2911 | goto err; |
2912 | } | |
2913 | ||
16cdcec7 | 2914 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2915 | if (ret) { |
2916 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2917 | goto err; |
79787eaa | 2918 | } |
39279cc3 | 2919 | |
e02119d5 | 2920 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2921 | inode, dir_ino); |
79787eaa JM |
2922 | if (ret != 0 && ret != -ENOENT) { |
2923 | btrfs_abort_transaction(trans, root, ret); | |
2924 | goto err; | |
2925 | } | |
e02119d5 CM |
2926 | |
2927 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2928 | dir, index); | |
6418c961 CM |
2929 | if (ret == -ENOENT) |
2930 | ret = 0; | |
39279cc3 CM |
2931 | err: |
2932 | btrfs_free_path(path); | |
e02119d5 CM |
2933 | if (ret) |
2934 | goto out; | |
2935 | ||
2936 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2937 | inode_inc_iversion(inode); |
2938 | inode_inc_iversion(dir); | |
e02119d5 | 2939 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 2940 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 2941 | out: |
39279cc3 CM |
2942 | return ret; |
2943 | } | |
2944 | ||
92986796 AV |
2945 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2946 | struct btrfs_root *root, | |
2947 | struct inode *dir, struct inode *inode, | |
2948 | const char *name, int name_len) | |
2949 | { | |
2950 | int ret; | |
2951 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2952 | if (!ret) { | |
2953 | btrfs_drop_nlink(inode); | |
2954 | ret = btrfs_update_inode(trans, root, inode); | |
2955 | } | |
2956 | return ret; | |
2957 | } | |
2958 | ||
2959 | ||
a22285a6 YZ |
2960 | /* helper to check if there is any shared block in the path */ |
2961 | static int check_path_shared(struct btrfs_root *root, | |
2962 | struct btrfs_path *path) | |
39279cc3 | 2963 | { |
a22285a6 YZ |
2964 | struct extent_buffer *eb; |
2965 | int level; | |
0e4dcbef | 2966 | u64 refs = 1; |
5df6a9f6 | 2967 | |
a22285a6 | 2968 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2969 | int ret; |
2970 | ||
a22285a6 YZ |
2971 | if (!path->nodes[level]) |
2972 | break; | |
2973 | eb = path->nodes[level]; | |
2974 | if (!btrfs_block_can_be_shared(root, eb)) | |
2975 | continue; | |
2976 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2977 | &refs, NULL); | |
2978 | if (refs > 1) | |
2979 | return 1; | |
5df6a9f6 | 2980 | } |
dedefd72 | 2981 | return 0; |
39279cc3 CM |
2982 | } |
2983 | ||
a22285a6 YZ |
2984 | /* |
2985 | * helper to start transaction for unlink and rmdir. | |
2986 | * | |
2987 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2988 | * so in enospc case, we should make sure they will free space before | |
2989 | * allowing them to use the global metadata reservation. | |
2990 | */ | |
2991 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2992 | struct dentry *dentry) | |
4df27c4d | 2993 | { |
39279cc3 | 2994 | struct btrfs_trans_handle *trans; |
a22285a6 | 2995 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2996 | struct btrfs_path *path; |
4df27c4d | 2997 | struct btrfs_dir_item *di; |
7b128766 | 2998 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2999 | u64 index; |
a22285a6 YZ |
3000 | int check_link = 1; |
3001 | int err = -ENOSPC; | |
4df27c4d | 3002 | int ret; |
33345d01 LZ |
3003 | u64 ino = btrfs_ino(inode); |
3004 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 3005 | |
e70bea5f JB |
3006 | /* |
3007 | * 1 for the possible orphan item | |
3008 | * 1 for the dir item | |
3009 | * 1 for the dir index | |
3010 | * 1 for the inode ref | |
3011 | * 1 for the inode ref in the tree log | |
3012 | * 2 for the dir entries in the log | |
3013 | * 1 for the inode | |
3014 | */ | |
3015 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
3016 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
3017 | return trans; | |
4df27c4d | 3018 | |
33345d01 | 3019 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 3020 | return ERR_PTR(-ENOSPC); |
4df27c4d | 3021 | |
a22285a6 YZ |
3022 | /* check if there is someone else holds reference */ |
3023 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
3024 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 3025 | |
a22285a6 YZ |
3026 | if (atomic_read(&inode->i_count) > 2) |
3027 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 3028 | |
a22285a6 YZ |
3029 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
3030 | return ERR_PTR(-ENOSPC); | |
3031 | ||
3032 | path = btrfs_alloc_path(); | |
3033 | if (!path) { | |
3034 | root->fs_info->enospc_unlink = 0; | |
3035 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
3036 | } |
3037 | ||
3880a1b4 JB |
3038 | /* 1 for the orphan item */ |
3039 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 3040 | if (IS_ERR(trans)) { |
a22285a6 YZ |
3041 | btrfs_free_path(path); |
3042 | root->fs_info->enospc_unlink = 0; | |
3043 | return trans; | |
3044 | } | |
4df27c4d | 3045 | |
a22285a6 YZ |
3046 | path->skip_locking = 1; |
3047 | path->search_commit_root = 1; | |
4df27c4d | 3048 | |
a22285a6 YZ |
3049 | ret = btrfs_lookup_inode(trans, root, path, |
3050 | &BTRFS_I(dir)->location, 0); | |
3051 | if (ret < 0) { | |
3052 | err = ret; | |
3053 | goto out; | |
3054 | } | |
3055 | if (ret == 0) { | |
3056 | if (check_path_shared(root, path)) | |
3057 | goto out; | |
3058 | } else { | |
3059 | check_link = 0; | |
5df6a9f6 | 3060 | } |
b3b4aa74 | 3061 | btrfs_release_path(path); |
a22285a6 YZ |
3062 | |
3063 | ret = btrfs_lookup_inode(trans, root, path, | |
3064 | &BTRFS_I(inode)->location, 0); | |
3065 | if (ret < 0) { | |
3066 | err = ret; | |
3067 | goto out; | |
3068 | } | |
3069 | if (ret == 0) { | |
3070 | if (check_path_shared(root, path)) | |
3071 | goto out; | |
3072 | } else { | |
3073 | check_link = 0; | |
3074 | } | |
b3b4aa74 | 3075 | btrfs_release_path(path); |
a22285a6 YZ |
3076 | |
3077 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
3078 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 3079 | ino, (u64)-1, 0); |
a22285a6 YZ |
3080 | if (ret < 0) { |
3081 | err = ret; | |
3082 | goto out; | |
3083 | } | |
79787eaa | 3084 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
3085 | if (check_path_shared(root, path)) |
3086 | goto out; | |
b3b4aa74 | 3087 | btrfs_release_path(path); |
a22285a6 YZ |
3088 | } |
3089 | ||
3090 | if (!check_link) { | |
3091 | err = 0; | |
3092 | goto out; | |
3093 | } | |
3094 | ||
33345d01 | 3095 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
3096 | dentry->d_name.name, dentry->d_name.len, 0); |
3097 | if (IS_ERR(di)) { | |
3098 | err = PTR_ERR(di); | |
3099 | goto out; | |
3100 | } | |
3101 | if (di) { | |
3102 | if (check_path_shared(root, path)) | |
3103 | goto out; | |
3104 | } else { | |
3105 | err = 0; | |
3106 | goto out; | |
3107 | } | |
b3b4aa74 | 3108 | btrfs_release_path(path); |
a22285a6 | 3109 | |
f186373f MF |
3110 | ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name, |
3111 | dentry->d_name.len, ino, dir_ino, 0, | |
3112 | &index); | |
3113 | if (ret) { | |
3114 | err = ret; | |
a22285a6 YZ |
3115 | goto out; |
3116 | } | |
f186373f | 3117 | |
a22285a6 YZ |
3118 | if (check_path_shared(root, path)) |
3119 | goto out; | |
f186373f | 3120 | |
b3b4aa74 | 3121 | btrfs_release_path(path); |
a22285a6 | 3122 | |
16cdcec7 MX |
3123 | /* |
3124 | * This is a commit root search, if we can lookup inode item and other | |
3125 | * relative items in the commit root, it means the transaction of | |
3126 | * dir/file creation has been committed, and the dir index item that we | |
3127 | * delay to insert has also been inserted into the commit root. So | |
3128 | * we needn't worry about the delayed insertion of the dir index item | |
3129 | * here. | |
3130 | */ | |
33345d01 | 3131 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
3132 | dentry->d_name.name, dentry->d_name.len, 0); |
3133 | if (IS_ERR(di)) { | |
3134 | err = PTR_ERR(di); | |
3135 | goto out; | |
3136 | } | |
3137 | BUG_ON(ret == -ENOENT); | |
3138 | if (check_path_shared(root, path)) | |
3139 | goto out; | |
3140 | ||
3141 | err = 0; | |
3142 | out: | |
3143 | btrfs_free_path(path); | |
3880a1b4 JB |
3144 | /* Migrate the orphan reservation over */ |
3145 | if (!err) | |
3146 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3147 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3148 | trans->bytes_reserved); |
3880a1b4 | 3149 | |
a22285a6 YZ |
3150 | if (err) { |
3151 | btrfs_end_transaction(trans, root); | |
3152 | root->fs_info->enospc_unlink = 0; | |
3153 | return ERR_PTR(err); | |
3154 | } | |
3155 | ||
3156 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3157 | return trans; | |
3158 | } | |
3159 | ||
3160 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3161 | struct btrfs_root *root) | |
3162 | { | |
66d8f3dd | 3163 | if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) { |
5a77d76c JB |
3164 | btrfs_block_rsv_release(root, trans->block_rsv, |
3165 | trans->bytes_reserved); | |
3166 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3167 | BUG_ON(!root->fs_info->enospc_unlink); |
3168 | root->fs_info->enospc_unlink = 0; | |
3169 | } | |
7ad85bb7 | 3170 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3171 | } |
3172 | ||
3173 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3174 | { | |
3175 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3176 | struct btrfs_trans_handle *trans; | |
3177 | struct inode *inode = dentry->d_inode; | |
3178 | int ret; | |
a22285a6 YZ |
3179 | |
3180 | trans = __unlink_start_trans(dir, dentry); | |
3181 | if (IS_ERR(trans)) | |
3182 | return PTR_ERR(trans); | |
5f39d397 | 3183 | |
12fcfd22 CM |
3184 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3185 | ||
e02119d5 CM |
3186 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3187 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3188 | if (ret) |
3189 | goto out; | |
7b128766 | 3190 | |
a22285a6 | 3191 | if (inode->i_nlink == 0) { |
7b128766 | 3192 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3193 | if (ret) |
3194 | goto out; | |
a22285a6 | 3195 | } |
7b128766 | 3196 | |
b532402e | 3197 | out: |
a22285a6 | 3198 | __unlink_end_trans(trans, root); |
b53d3f5d | 3199 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3200 | return ret; |
3201 | } | |
3202 | ||
4df27c4d YZ |
3203 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3204 | struct btrfs_root *root, | |
3205 | struct inode *dir, u64 objectid, | |
3206 | const char *name, int name_len) | |
3207 | { | |
3208 | struct btrfs_path *path; | |
3209 | struct extent_buffer *leaf; | |
3210 | struct btrfs_dir_item *di; | |
3211 | struct btrfs_key key; | |
3212 | u64 index; | |
3213 | int ret; | |
33345d01 | 3214 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3215 | |
3216 | path = btrfs_alloc_path(); | |
3217 | if (!path) | |
3218 | return -ENOMEM; | |
3219 | ||
33345d01 | 3220 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3221 | name, name_len, -1); |
79787eaa JM |
3222 | if (IS_ERR_OR_NULL(di)) { |
3223 | if (!di) | |
3224 | ret = -ENOENT; | |
3225 | else | |
3226 | ret = PTR_ERR(di); | |
3227 | goto out; | |
3228 | } | |
4df27c4d YZ |
3229 | |
3230 | leaf = path->nodes[0]; | |
3231 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3232 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3233 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3234 | if (ret) { |
3235 | btrfs_abort_transaction(trans, root, ret); | |
3236 | goto out; | |
3237 | } | |
b3b4aa74 | 3238 | btrfs_release_path(path); |
4df27c4d YZ |
3239 | |
3240 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3241 | objectid, root->root_key.objectid, | |
33345d01 | 3242 | dir_ino, &index, name, name_len); |
4df27c4d | 3243 | if (ret < 0) { |
79787eaa JM |
3244 | if (ret != -ENOENT) { |
3245 | btrfs_abort_transaction(trans, root, ret); | |
3246 | goto out; | |
3247 | } | |
33345d01 | 3248 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3249 | name, name_len); |
79787eaa JM |
3250 | if (IS_ERR_OR_NULL(di)) { |
3251 | if (!di) | |
3252 | ret = -ENOENT; | |
3253 | else | |
3254 | ret = PTR_ERR(di); | |
3255 | btrfs_abort_transaction(trans, root, ret); | |
3256 | goto out; | |
3257 | } | |
4df27c4d YZ |
3258 | |
3259 | leaf = path->nodes[0]; | |
3260 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3261 | btrfs_release_path(path); |
4df27c4d YZ |
3262 | index = key.offset; |
3263 | } | |
945d8962 | 3264 | btrfs_release_path(path); |
4df27c4d | 3265 | |
16cdcec7 | 3266 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3267 | if (ret) { |
3268 | btrfs_abort_transaction(trans, root, ret); | |
3269 | goto out; | |
3270 | } | |
4df27c4d YZ |
3271 | |
3272 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3273 | inode_inc_iversion(dir); |
4df27c4d | 3274 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3275 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3276 | if (ret) |
3277 | btrfs_abort_transaction(trans, root, ret); | |
3278 | out: | |
71d7aed0 | 3279 | btrfs_free_path(path); |
79787eaa | 3280 | return ret; |
4df27c4d YZ |
3281 | } |
3282 | ||
39279cc3 CM |
3283 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3284 | { | |
3285 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3286 | int err = 0; |
39279cc3 | 3287 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3288 | struct btrfs_trans_handle *trans; |
39279cc3 | 3289 | |
b3ae244e | 3290 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3291 | return -ENOTEMPTY; |
b3ae244e DS |
3292 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3293 | return -EPERM; | |
134d4512 | 3294 | |
a22285a6 YZ |
3295 | trans = __unlink_start_trans(dir, dentry); |
3296 | if (IS_ERR(trans)) | |
5df6a9f6 | 3297 | return PTR_ERR(trans); |
5df6a9f6 | 3298 | |
33345d01 | 3299 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3300 | err = btrfs_unlink_subvol(trans, root, dir, |
3301 | BTRFS_I(inode)->location.objectid, | |
3302 | dentry->d_name.name, | |
3303 | dentry->d_name.len); | |
3304 | goto out; | |
3305 | } | |
3306 | ||
7b128766 JB |
3307 | err = btrfs_orphan_add(trans, inode); |
3308 | if (err) | |
4df27c4d | 3309 | goto out; |
7b128766 | 3310 | |
39279cc3 | 3311 | /* now the directory is empty */ |
e02119d5 CM |
3312 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3313 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3314 | if (!err) |
dbe674a9 | 3315 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3316 | out: |
a22285a6 | 3317 | __unlink_end_trans(trans, root); |
b53d3f5d | 3318 | btrfs_btree_balance_dirty(root); |
3954401f | 3319 | |
39279cc3 CM |
3320 | return err; |
3321 | } | |
3322 | ||
39279cc3 CM |
3323 | /* |
3324 | * this can truncate away extent items, csum items and directory items. | |
3325 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3326 | * any higher than new_size |
39279cc3 CM |
3327 | * |
3328 | * csum items that cross the new i_size are truncated to the new size | |
3329 | * as well. | |
7b128766 JB |
3330 | * |
3331 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3332 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3333 | */ |
8082510e YZ |
3334 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3335 | struct btrfs_root *root, | |
3336 | struct inode *inode, | |
3337 | u64 new_size, u32 min_type) | |
39279cc3 | 3338 | { |
39279cc3 | 3339 | struct btrfs_path *path; |
5f39d397 | 3340 | struct extent_buffer *leaf; |
39279cc3 | 3341 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3342 | struct btrfs_key key; |
3343 | struct btrfs_key found_key; | |
39279cc3 | 3344 | u64 extent_start = 0; |
db94535d | 3345 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3346 | u64 extent_offset = 0; |
39279cc3 | 3347 | u64 item_end = 0; |
8082510e YZ |
3348 | u64 mask = root->sectorsize - 1; |
3349 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3350 | int found_extent; |
3351 | int del_item; | |
85e21bac CM |
3352 | int pending_del_nr = 0; |
3353 | int pending_del_slot = 0; | |
179e29e4 | 3354 | int extent_type = -1; |
8082510e YZ |
3355 | int ret; |
3356 | int err = 0; | |
33345d01 | 3357 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3358 | |
3359 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3360 | |
0eb0e19c MF |
3361 | path = btrfs_alloc_path(); |
3362 | if (!path) | |
3363 | return -ENOMEM; | |
3364 | path->reada = -1; | |
3365 | ||
5dc562c5 JB |
3366 | /* |
3367 | * We want to drop from the next block forward in case this new size is | |
3368 | * not block aligned since we will be keeping the last block of the | |
3369 | * extent just the way it is. | |
3370 | */ | |
0af3d00b | 3371 | if (root->ref_cows || root == root->fs_info->tree_root) |
5dc562c5 | 3372 | btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0); |
8082510e | 3373 | |
16cdcec7 MX |
3374 | /* |
3375 | * This function is also used to drop the items in the log tree before | |
3376 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3377 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3378 | * items. | |
3379 | */ | |
3380 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3381 | btrfs_kill_delayed_inode_items(inode); | |
3382 | ||
33345d01 | 3383 | key.objectid = ino; |
39279cc3 | 3384 | key.offset = (u64)-1; |
5f39d397 CM |
3385 | key.type = (u8)-1; |
3386 | ||
85e21bac | 3387 | search_again: |
b9473439 | 3388 | path->leave_spinning = 1; |
85e21bac | 3389 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3390 | if (ret < 0) { |
3391 | err = ret; | |
3392 | goto out; | |
3393 | } | |
d397712b | 3394 | |
85e21bac | 3395 | if (ret > 0) { |
e02119d5 CM |
3396 | /* there are no items in the tree for us to truncate, we're |
3397 | * done | |
3398 | */ | |
8082510e YZ |
3399 | if (path->slots[0] == 0) |
3400 | goto out; | |
85e21bac CM |
3401 | path->slots[0]--; |
3402 | } | |
3403 | ||
d397712b | 3404 | while (1) { |
39279cc3 | 3405 | fi = NULL; |
5f39d397 CM |
3406 | leaf = path->nodes[0]; |
3407 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3408 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3409 | |
33345d01 | 3410 | if (found_key.objectid != ino) |
39279cc3 | 3411 | break; |
5f39d397 | 3412 | |
85e21bac | 3413 | if (found_type < min_type) |
39279cc3 CM |
3414 | break; |
3415 | ||
5f39d397 | 3416 | item_end = found_key.offset; |
39279cc3 | 3417 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3418 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3419 | struct btrfs_file_extent_item); |
179e29e4 CM |
3420 | extent_type = btrfs_file_extent_type(leaf, fi); |
3421 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3422 | item_end += |
db94535d | 3423 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3424 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3425 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3426 | fi); |
39279cc3 | 3427 | } |
008630c1 | 3428 | item_end--; |
39279cc3 | 3429 | } |
8082510e YZ |
3430 | if (found_type > min_type) { |
3431 | del_item = 1; | |
3432 | } else { | |
3433 | if (item_end < new_size) | |
b888db2b | 3434 | break; |
8082510e YZ |
3435 | if (found_key.offset >= new_size) |
3436 | del_item = 1; | |
3437 | else | |
3438 | del_item = 0; | |
39279cc3 | 3439 | } |
39279cc3 | 3440 | found_extent = 0; |
39279cc3 | 3441 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3442 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3443 | goto delete; | |
3444 | ||
3445 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3446 | u64 num_dec; |
db94535d | 3447 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3448 | if (!del_item) { |
db94535d CM |
3449 | u64 orig_num_bytes = |
3450 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3451 | extent_num_bytes = new_size - |
5f39d397 | 3452 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3453 | extent_num_bytes = extent_num_bytes & |
3454 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3455 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3456 | extent_num_bytes); | |
3457 | num_dec = (orig_num_bytes - | |
9069218d | 3458 | extent_num_bytes); |
e02119d5 | 3459 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3460 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3461 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3462 | } else { |
db94535d CM |
3463 | extent_num_bytes = |
3464 | btrfs_file_extent_disk_num_bytes(leaf, | |
3465 | fi); | |
5d4f98a2 YZ |
3466 | extent_offset = found_key.offset - |
3467 | btrfs_file_extent_offset(leaf, fi); | |
3468 | ||
39279cc3 | 3469 | /* FIXME blocksize != 4096 */ |
9069218d | 3470 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3471 | if (extent_start != 0) { |
3472 | found_extent = 1; | |
e02119d5 | 3473 | if (root->ref_cows) |
a76a3cd4 | 3474 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3475 | } |
39279cc3 | 3476 | } |
9069218d | 3477 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3478 | /* |
3479 | * we can't truncate inline items that have had | |
3480 | * special encodings | |
3481 | */ | |
3482 | if (!del_item && | |
3483 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3484 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3485 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3486 | u32 size = new_size - found_key.offset; |
3487 | ||
3488 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3489 | inode_sub_bytes(inode, item_end + 1 - |
3490 | new_size); | |
e02119d5 CM |
3491 | } |
3492 | size = | |
3493 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3494 | btrfs_truncate_item(trans, root, path, |
3495 | size, 1); | |
e02119d5 | 3496 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3497 | inode_sub_bytes(inode, item_end + 1 - |
3498 | found_key.offset); | |
9069218d | 3499 | } |
39279cc3 | 3500 | } |
179e29e4 | 3501 | delete: |
39279cc3 | 3502 | if (del_item) { |
85e21bac CM |
3503 | if (!pending_del_nr) { |
3504 | /* no pending yet, add ourselves */ | |
3505 | pending_del_slot = path->slots[0]; | |
3506 | pending_del_nr = 1; | |
3507 | } else if (pending_del_nr && | |
3508 | path->slots[0] + 1 == pending_del_slot) { | |
3509 | /* hop on the pending chunk */ | |
3510 | pending_del_nr++; | |
3511 | pending_del_slot = path->slots[0]; | |
3512 | } else { | |
d397712b | 3513 | BUG(); |
85e21bac | 3514 | } |
39279cc3 CM |
3515 | } else { |
3516 | break; | |
3517 | } | |
0af3d00b JB |
3518 | if (found_extent && (root->ref_cows || |
3519 | root == root->fs_info->tree_root)) { | |
b9473439 | 3520 | btrfs_set_path_blocking(path); |
39279cc3 | 3521 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3522 | extent_num_bytes, 0, |
3523 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3524 | ino, extent_offset, 0); |
39279cc3 CM |
3525 | BUG_ON(ret); |
3526 | } | |
85e21bac | 3527 | |
8082510e YZ |
3528 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3529 | break; | |
3530 | ||
3531 | if (path->slots[0] == 0 || | |
3532 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
3533 | if (pending_del_nr) { |
3534 | ret = btrfs_del_items(trans, root, path, | |
3535 | pending_del_slot, | |
3536 | pending_del_nr); | |
79787eaa JM |
3537 | if (ret) { |
3538 | btrfs_abort_transaction(trans, | |
3539 | root, ret); | |
3540 | goto error; | |
3541 | } | |
8082510e YZ |
3542 | pending_del_nr = 0; |
3543 | } | |
b3b4aa74 | 3544 | btrfs_release_path(path); |
85e21bac | 3545 | goto search_again; |
8082510e YZ |
3546 | } else { |
3547 | path->slots[0]--; | |
85e21bac | 3548 | } |
39279cc3 | 3549 | } |
8082510e | 3550 | out: |
85e21bac CM |
3551 | if (pending_del_nr) { |
3552 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3553 | pending_del_nr); | |
79787eaa JM |
3554 | if (ret) |
3555 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3556 | } |
79787eaa | 3557 | error: |
39279cc3 | 3558 | btrfs_free_path(path); |
8082510e | 3559 | return err; |
39279cc3 CM |
3560 | } |
3561 | ||
3562 | /* | |
2aaa6655 JB |
3563 | * btrfs_truncate_page - read, zero a chunk and write a page |
3564 | * @inode - inode that we're zeroing | |
3565 | * @from - the offset to start zeroing | |
3566 | * @len - the length to zero, 0 to zero the entire range respective to the | |
3567 | * offset | |
3568 | * @front - zero up to the offset instead of from the offset on | |
3569 | * | |
3570 | * This will find the page for the "from" offset and cow the page and zero the | |
3571 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 3572 | */ |
2aaa6655 JB |
3573 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
3574 | int front) | |
39279cc3 | 3575 | { |
2aaa6655 | 3576 | struct address_space *mapping = inode->i_mapping; |
db94535d | 3577 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3578 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3579 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3580 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3581 | char *kaddr; |
db94535d | 3582 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3583 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3584 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3585 | struct page *page; | |
3b16a4e3 | 3586 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3587 | int ret = 0; |
a52d9a80 | 3588 | u64 page_start; |
e6dcd2dc | 3589 | u64 page_end; |
39279cc3 | 3590 | |
2aaa6655 JB |
3591 | if ((offset & (blocksize - 1)) == 0 && |
3592 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 3593 | goto out; |
0ca1f7ce | 3594 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3595 | if (ret) |
3596 | goto out; | |
39279cc3 | 3597 | |
211c17f5 | 3598 | again: |
3b16a4e3 | 3599 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3600 | if (!page) { |
0ca1f7ce | 3601 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 3602 | ret = -ENOMEM; |
39279cc3 | 3603 | goto out; |
5d5e103a | 3604 | } |
e6dcd2dc CM |
3605 | |
3606 | page_start = page_offset(page); | |
3607 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3608 | ||
39279cc3 | 3609 | if (!PageUptodate(page)) { |
9ebefb18 | 3610 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3611 | lock_page(page); |
211c17f5 CM |
3612 | if (page->mapping != mapping) { |
3613 | unlock_page(page); | |
3614 | page_cache_release(page); | |
3615 | goto again; | |
3616 | } | |
39279cc3 CM |
3617 | if (!PageUptodate(page)) { |
3618 | ret = -EIO; | |
89642229 | 3619 | goto out_unlock; |
39279cc3 CM |
3620 | } |
3621 | } | |
211c17f5 | 3622 | wait_on_page_writeback(page); |
e6dcd2dc | 3623 | |
d0082371 | 3624 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3625 | set_page_extent_mapped(page); |
3626 | ||
3627 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3628 | if (ordered) { | |
2ac55d41 JB |
3629 | unlock_extent_cached(io_tree, page_start, page_end, |
3630 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3631 | unlock_page(page); |
3632 | page_cache_release(page); | |
eb84ae03 | 3633 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3634 | btrfs_put_ordered_extent(ordered); |
3635 | goto again; | |
3636 | } | |
3637 | ||
2ac55d41 | 3638 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
3639 | EXTENT_DIRTY | EXTENT_DELALLOC | |
3640 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 3641 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3642 | |
2ac55d41 JB |
3643 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3644 | &cached_state); | |
9ed74f2d | 3645 | if (ret) { |
2ac55d41 JB |
3646 | unlock_extent_cached(io_tree, page_start, page_end, |
3647 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3648 | goto out_unlock; |
3649 | } | |
3650 | ||
e6dcd2dc | 3651 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
3652 | if (!len) |
3653 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 3654 | kaddr = kmap(page); |
2aaa6655 JB |
3655 | if (front) |
3656 | memset(kaddr, 0, offset); | |
3657 | else | |
3658 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
3659 | flush_dcache_page(page); |
3660 | kunmap(page); | |
3661 | } | |
247e743c | 3662 | ClearPageChecked(page); |
e6dcd2dc | 3663 | set_page_dirty(page); |
2ac55d41 JB |
3664 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3665 | GFP_NOFS); | |
39279cc3 | 3666 | |
89642229 | 3667 | out_unlock: |
5d5e103a | 3668 | if (ret) |
0ca1f7ce | 3669 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3670 | unlock_page(page); |
3671 | page_cache_release(page); | |
3672 | out: | |
3673 | return ret; | |
3674 | } | |
3675 | ||
695a0d0d JB |
3676 | /* |
3677 | * This function puts in dummy file extents for the area we're creating a hole | |
3678 | * for. So if we are truncating this file to a larger size we need to insert | |
3679 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3680 | * the range between oldsize and size | |
3681 | */ | |
a41ad394 | 3682 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3683 | { |
9036c102 YZ |
3684 | struct btrfs_trans_handle *trans; |
3685 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3686 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3687 | struct extent_map *em = NULL; |
2ac55d41 | 3688 | struct extent_state *cached_state = NULL; |
5dc562c5 | 3689 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9036c102 | 3690 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3691 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3692 | u64 block_end = (size + mask) & ~mask; |
3693 | u64 last_byte; | |
3694 | u64 cur_offset; | |
3695 | u64 hole_size; | |
9ed74f2d | 3696 | int err = 0; |
39279cc3 | 3697 | |
9036c102 YZ |
3698 | if (size <= hole_start) |
3699 | return 0; | |
3700 | ||
9036c102 YZ |
3701 | while (1) { |
3702 | struct btrfs_ordered_extent *ordered; | |
3703 | btrfs_wait_ordered_range(inode, hole_start, | |
3704 | block_end - hole_start); | |
2ac55d41 | 3705 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3706 | &cached_state); |
9036c102 YZ |
3707 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3708 | if (!ordered) | |
3709 | break; | |
2ac55d41 JB |
3710 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3711 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3712 | btrfs_put_ordered_extent(ordered); |
3713 | } | |
39279cc3 | 3714 | |
9036c102 YZ |
3715 | cur_offset = hole_start; |
3716 | while (1) { | |
3717 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3718 | block_end - cur_offset, 0); | |
79787eaa JM |
3719 | if (IS_ERR(em)) { |
3720 | err = PTR_ERR(em); | |
f2767956 | 3721 | em = NULL; |
79787eaa JM |
3722 | break; |
3723 | } | |
9036c102 YZ |
3724 | last_byte = min(extent_map_end(em), block_end); |
3725 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3726 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 3727 | struct extent_map *hole_em; |
9036c102 | 3728 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3729 | |
3642320e | 3730 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3731 | if (IS_ERR(trans)) { |
3732 | err = PTR_ERR(trans); | |
9ed74f2d | 3733 | break; |
a22285a6 | 3734 | } |
8082510e | 3735 | |
5dc562c5 JB |
3736 | err = btrfs_drop_extents(trans, root, inode, |
3737 | cur_offset, | |
2671485d | 3738 | cur_offset + hole_size, 1); |
5b397377 | 3739 | if (err) { |
79787eaa | 3740 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3741 | btrfs_end_transaction(trans, root); |
3893e33b | 3742 | break; |
5b397377 | 3743 | } |
8082510e | 3744 | |
9036c102 | 3745 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3746 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3747 | 0, hole_size, 0, hole_size, |
3748 | 0, 0, 0); | |
5b397377 | 3749 | if (err) { |
79787eaa | 3750 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3751 | btrfs_end_transaction(trans, root); |
3893e33b | 3752 | break; |
5b397377 | 3753 | } |
8082510e | 3754 | |
5dc562c5 JB |
3755 | btrfs_drop_extent_cache(inode, cur_offset, |
3756 | cur_offset + hole_size - 1, 0); | |
3757 | hole_em = alloc_extent_map(); | |
3758 | if (!hole_em) { | |
3759 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3760 | &BTRFS_I(inode)->runtime_flags); | |
3761 | goto next; | |
3762 | } | |
3763 | hole_em->start = cur_offset; | |
3764 | hole_em->len = hole_size; | |
3765 | hole_em->orig_start = cur_offset; | |
8082510e | 3766 | |
5dc562c5 JB |
3767 | hole_em->block_start = EXTENT_MAP_HOLE; |
3768 | hole_em->block_len = 0; | |
b4939680 | 3769 | hole_em->orig_block_len = 0; |
5dc562c5 JB |
3770 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
3771 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
3772 | hole_em->generation = trans->transid; | |
8082510e | 3773 | |
5dc562c5 JB |
3774 | while (1) { |
3775 | write_lock(&em_tree->lock); | |
3776 | err = add_extent_mapping(em_tree, hole_em); | |
3777 | if (!err) | |
3778 | list_move(&hole_em->list, | |
3779 | &em_tree->modified_extents); | |
3780 | write_unlock(&em_tree->lock); | |
3781 | if (err != -EEXIST) | |
3782 | break; | |
3783 | btrfs_drop_extent_cache(inode, cur_offset, | |
3784 | cur_offset + | |
3785 | hole_size - 1, 0); | |
3786 | } | |
3787 | free_extent_map(hole_em); | |
3788 | next: | |
3642320e | 3789 | btrfs_update_inode(trans, root, inode); |
8082510e | 3790 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3791 | } |
3792 | free_extent_map(em); | |
a22285a6 | 3793 | em = NULL; |
9036c102 | 3794 | cur_offset = last_byte; |
8082510e | 3795 | if (cur_offset >= block_end) |
9036c102 YZ |
3796 | break; |
3797 | } | |
1832a6d5 | 3798 | |
a22285a6 | 3799 | free_extent_map(em); |
2ac55d41 JB |
3800 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3801 | GFP_NOFS); | |
9036c102 YZ |
3802 | return err; |
3803 | } | |
39279cc3 | 3804 | |
3972f260 | 3805 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 3806 | { |
f4a2f4c5 MX |
3807 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3808 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3809 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
3810 | loff_t newsize = attr->ia_size; |
3811 | int mask = attr->ia_valid; | |
8082510e YZ |
3812 | int ret; |
3813 | ||
a41ad394 | 3814 | if (newsize == oldsize) |
8082510e YZ |
3815 | return 0; |
3816 | ||
3972f260 ES |
3817 | /* |
3818 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
3819 | * special case where we need to update the times despite not having | |
3820 | * these flags set. For all other operations the VFS set these flags | |
3821 | * explicitly if it wants a timestamp update. | |
3822 | */ | |
3823 | if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) | |
3824 | inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb); | |
3825 | ||
a41ad394 | 3826 | if (newsize > oldsize) { |
a41ad394 JB |
3827 | truncate_pagecache(inode, oldsize, newsize); |
3828 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3829 | if (ret) |
8082510e | 3830 | return ret; |
8082510e | 3831 | |
f4a2f4c5 MX |
3832 | trans = btrfs_start_transaction(root, 1); |
3833 | if (IS_ERR(trans)) | |
3834 | return PTR_ERR(trans); | |
3835 | ||
3836 | i_size_write(inode, newsize); | |
3837 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3838 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3839 | btrfs_end_transaction(trans, root); |
a41ad394 | 3840 | } else { |
8082510e | 3841 | |
a41ad394 JB |
3842 | /* |
3843 | * We're truncating a file that used to have good data down to | |
3844 | * zero. Make sure it gets into the ordered flush list so that | |
3845 | * any new writes get down to disk quickly. | |
3846 | */ | |
3847 | if (newsize == 0) | |
72ac3c0d JB |
3848 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3849 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3850 | |
f3fe820c JB |
3851 | /* |
3852 | * 1 for the orphan item we're going to add | |
3853 | * 1 for the orphan item deletion. | |
3854 | */ | |
3855 | trans = btrfs_start_transaction(root, 2); | |
3856 | if (IS_ERR(trans)) | |
3857 | return PTR_ERR(trans); | |
3858 | ||
3859 | /* | |
3860 | * We need to do this in case we fail at _any_ point during the | |
3861 | * actual truncate. Once we do the truncate_setsize we could | |
3862 | * invalidate pages which forces any outstanding ordered io to | |
3863 | * be instantly completed which will give us extents that need | |
3864 | * to be truncated. If we fail to get an orphan inode down we | |
3865 | * could have left over extents that were never meant to live, | |
3866 | * so we need to garuntee from this point on that everything | |
3867 | * will be consistent. | |
3868 | */ | |
3869 | ret = btrfs_orphan_add(trans, inode); | |
3870 | btrfs_end_transaction(trans, root); | |
3871 | if (ret) | |
3872 | return ret; | |
3873 | ||
a41ad394 JB |
3874 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3875 | truncate_setsize(inode, newsize); | |
3876 | ret = btrfs_truncate(inode); | |
f3fe820c JB |
3877 | if (ret && inode->i_nlink) |
3878 | btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
3879 | } |
3880 | ||
a41ad394 | 3881 | return ret; |
8082510e YZ |
3882 | } |
3883 | ||
9036c102 YZ |
3884 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3885 | { | |
3886 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3887 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3888 | int err; |
39279cc3 | 3889 | |
b83cc969 LZ |
3890 | if (btrfs_root_readonly(root)) |
3891 | return -EROFS; | |
3892 | ||
9036c102 YZ |
3893 | err = inode_change_ok(inode, attr); |
3894 | if (err) | |
3895 | return err; | |
2bf5a725 | 3896 | |
5a3f23d5 | 3897 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 3898 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
3899 | if (err) |
3900 | return err; | |
39279cc3 | 3901 | } |
9036c102 | 3902 | |
1025774c CH |
3903 | if (attr->ia_valid) { |
3904 | setattr_copy(inode, attr); | |
0c4d2d95 | 3905 | inode_inc_iversion(inode); |
22c44fe6 | 3906 | err = btrfs_dirty_inode(inode); |
1025774c | 3907 | |
22c44fe6 | 3908 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3909 | err = btrfs_acl_chmod(inode); |
3910 | } | |
33268eaf | 3911 | |
39279cc3 CM |
3912 | return err; |
3913 | } | |
61295eb8 | 3914 | |
bd555975 | 3915 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3916 | { |
3917 | struct btrfs_trans_handle *trans; | |
3918 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3919 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3920 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
3921 | int ret; |
3922 | ||
1abe9b8a | 3923 | trace_btrfs_inode_evict(inode); |
3924 | ||
39279cc3 | 3925 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3926 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 3927 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
3928 | goto no_delete; |
3929 | ||
39279cc3 | 3930 | if (is_bad_inode(inode)) { |
7b128766 | 3931 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3932 | goto no_delete; |
3933 | } | |
bd555975 | 3934 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3935 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3936 | |
c71bf099 | 3937 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 3938 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 3939 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
3940 | goto no_delete; |
3941 | } | |
3942 | ||
76dda93c YZ |
3943 | if (inode->i_nlink > 0) { |
3944 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3945 | goto no_delete; | |
3946 | } | |
3947 | ||
0e8c36a9 MX |
3948 | ret = btrfs_commit_inode_delayed_inode(inode); |
3949 | if (ret) { | |
3950 | btrfs_orphan_del(NULL, inode); | |
3951 | goto no_delete; | |
3952 | } | |
3953 | ||
66d8f3dd | 3954 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
3955 | if (!rsv) { |
3956 | btrfs_orphan_del(NULL, inode); | |
3957 | goto no_delete; | |
3958 | } | |
4a338542 | 3959 | rsv->size = min_size; |
ca7e70f5 | 3960 | rsv->failfast = 1; |
726c35fa | 3961 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3962 | |
dbe674a9 | 3963 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3964 | |
4289a667 | 3965 | /* |
8407aa46 MX |
3966 | * This is a bit simpler than btrfs_truncate since we've already |
3967 | * reserved our space for our orphan item in the unlink, so we just | |
3968 | * need to reserve some slack space in case we add bytes and update | |
3969 | * inode item when doing the truncate. | |
4289a667 | 3970 | */ |
8082510e | 3971 | while (1) { |
08e007d2 MX |
3972 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
3973 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
3974 | |
3975 | /* | |
3976 | * Try and steal from the global reserve since we will | |
3977 | * likely not use this space anyway, we want to try as | |
3978 | * hard as possible to get this to work. | |
3979 | */ | |
3980 | if (ret) | |
3981 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3982 | |
d68fc57b | 3983 | if (ret) { |
4289a667 | 3984 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3985 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3986 | btrfs_orphan_del(NULL, inode); |
3987 | btrfs_free_block_rsv(root, rsv); | |
3988 | goto no_delete; | |
d68fc57b | 3989 | } |
7b128766 | 3990 | |
0e8c36a9 | 3991 | trans = btrfs_join_transaction(root); |
4289a667 JB |
3992 | if (IS_ERR(trans)) { |
3993 | btrfs_orphan_del(NULL, inode); | |
3994 | btrfs_free_block_rsv(root, rsv); | |
3995 | goto no_delete; | |
d68fc57b | 3996 | } |
7b128766 | 3997 | |
4289a667 JB |
3998 | trans->block_rsv = rsv; |
3999 | ||
d68fc57b | 4000 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 4001 | if (ret != -ENOSPC) |
8082510e | 4002 | break; |
85e21bac | 4003 | |
8407aa46 | 4004 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
4005 | btrfs_end_transaction(trans, root); |
4006 | trans = NULL; | |
b53d3f5d | 4007 | btrfs_btree_balance_dirty(root); |
8082510e | 4008 | } |
5f39d397 | 4009 | |
4289a667 JB |
4010 | btrfs_free_block_rsv(root, rsv); |
4011 | ||
8082510e | 4012 | if (ret == 0) { |
4289a667 | 4013 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
4014 | ret = btrfs_orphan_del(trans, inode); |
4015 | BUG_ON(ret); | |
4016 | } | |
54aa1f4d | 4017 | |
4289a667 | 4018 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
4019 | if (!(root == root->fs_info->tree_root || |
4020 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 4021 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 4022 | |
54aa1f4d | 4023 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4024 | btrfs_btree_balance_dirty(root); |
39279cc3 | 4025 | no_delete: |
dbd5768f | 4026 | clear_inode(inode); |
8082510e | 4027 | return; |
39279cc3 CM |
4028 | } |
4029 | ||
4030 | /* | |
4031 | * this returns the key found in the dir entry in the location pointer. | |
4032 | * If no dir entries were found, location->objectid is 0. | |
4033 | */ | |
4034 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
4035 | struct btrfs_key *location) | |
4036 | { | |
4037 | const char *name = dentry->d_name.name; | |
4038 | int namelen = dentry->d_name.len; | |
4039 | struct btrfs_dir_item *di; | |
4040 | struct btrfs_path *path; | |
4041 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 4042 | int ret = 0; |
39279cc3 CM |
4043 | |
4044 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
4045 | if (!path) |
4046 | return -ENOMEM; | |
3954401f | 4047 | |
33345d01 | 4048 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 4049 | namelen, 0); |
0d9f7f3e Y |
4050 | if (IS_ERR(di)) |
4051 | ret = PTR_ERR(di); | |
d397712b | 4052 | |
c704005d | 4053 | if (IS_ERR_OR_NULL(di)) |
3954401f | 4054 | goto out_err; |
d397712b | 4055 | |
5f39d397 | 4056 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 4057 | out: |
39279cc3 CM |
4058 | btrfs_free_path(path); |
4059 | return ret; | |
3954401f CM |
4060 | out_err: |
4061 | location->objectid = 0; | |
4062 | goto out; | |
39279cc3 CM |
4063 | } |
4064 | ||
4065 | /* | |
4066 | * when we hit a tree root in a directory, the btrfs part of the inode | |
4067 | * needs to be changed to reflect the root directory of the tree root. This | |
4068 | * is kind of like crossing a mount point. | |
4069 | */ | |
4070 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
4071 | struct inode *dir, |
4072 | struct dentry *dentry, | |
4073 | struct btrfs_key *location, | |
4074 | struct btrfs_root **sub_root) | |
39279cc3 | 4075 | { |
4df27c4d YZ |
4076 | struct btrfs_path *path; |
4077 | struct btrfs_root *new_root; | |
4078 | struct btrfs_root_ref *ref; | |
4079 | struct extent_buffer *leaf; | |
4080 | int ret; | |
4081 | int err = 0; | |
39279cc3 | 4082 | |
4df27c4d YZ |
4083 | path = btrfs_alloc_path(); |
4084 | if (!path) { | |
4085 | err = -ENOMEM; | |
4086 | goto out; | |
4087 | } | |
39279cc3 | 4088 | |
4df27c4d YZ |
4089 | err = -ENOENT; |
4090 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
4091 | BTRFS_I(dir)->root->root_key.objectid, | |
4092 | location->objectid); | |
4093 | if (ret) { | |
4094 | if (ret < 0) | |
4095 | err = ret; | |
4096 | goto out; | |
4097 | } | |
39279cc3 | 4098 | |
4df27c4d YZ |
4099 | leaf = path->nodes[0]; |
4100 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 4101 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
4102 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
4103 | goto out; | |
39279cc3 | 4104 | |
4df27c4d YZ |
4105 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
4106 | (unsigned long)(ref + 1), | |
4107 | dentry->d_name.len); | |
4108 | if (ret) | |
4109 | goto out; | |
4110 | ||
b3b4aa74 | 4111 | btrfs_release_path(path); |
4df27c4d YZ |
4112 | |
4113 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4114 | if (IS_ERR(new_root)) { | |
4115 | err = PTR_ERR(new_root); | |
4116 | goto out; | |
4117 | } | |
4118 | ||
4119 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
4120 | err = -ENOENT; | |
4121 | goto out; | |
4122 | } | |
4123 | ||
4124 | *sub_root = new_root; | |
4125 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4126 | location->type = BTRFS_INODE_ITEM_KEY; | |
4127 | location->offset = 0; | |
4128 | err = 0; | |
4129 | out: | |
4130 | btrfs_free_path(path); | |
4131 | return err; | |
39279cc3 CM |
4132 | } |
4133 | ||
5d4f98a2 YZ |
4134 | static void inode_tree_add(struct inode *inode) |
4135 | { | |
4136 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4137 | struct btrfs_inode *entry; | |
03e860bd FNP |
4138 | struct rb_node **p; |
4139 | struct rb_node *parent; | |
33345d01 | 4140 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
4141 | again: |
4142 | p = &root->inode_tree.rb_node; | |
4143 | parent = NULL; | |
5d4f98a2 | 4144 | |
1d3382cb | 4145 | if (inode_unhashed(inode)) |
76dda93c YZ |
4146 | return; |
4147 | ||
5d4f98a2 YZ |
4148 | spin_lock(&root->inode_lock); |
4149 | while (*p) { | |
4150 | parent = *p; | |
4151 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4152 | ||
33345d01 | 4153 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4154 | p = &parent->rb_left; |
33345d01 | 4155 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4156 | p = &parent->rb_right; |
5d4f98a2 YZ |
4157 | else { |
4158 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4159 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4160 | rb_erase(parent, &root->inode_tree); |
4161 | RB_CLEAR_NODE(parent); | |
4162 | spin_unlock(&root->inode_lock); | |
4163 | goto again; | |
5d4f98a2 YZ |
4164 | } |
4165 | } | |
4166 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4167 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4168 | spin_unlock(&root->inode_lock); | |
4169 | } | |
4170 | ||
4171 | static void inode_tree_del(struct inode *inode) | |
4172 | { | |
4173 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4174 | int empty = 0; |
5d4f98a2 | 4175 | |
03e860bd | 4176 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4177 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4178 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4179 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4180 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4181 | } |
03e860bd | 4182 | spin_unlock(&root->inode_lock); |
76dda93c | 4183 | |
0af3d00b JB |
4184 | /* |
4185 | * Free space cache has inodes in the tree root, but the tree root has a | |
4186 | * root_refs of 0, so this could end up dropping the tree root as a | |
4187 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4188 | * make sure we don't drop it. | |
4189 | */ | |
4190 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4191 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4192 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4193 | spin_lock(&root->inode_lock); | |
4194 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4195 | spin_unlock(&root->inode_lock); | |
4196 | if (empty) | |
4197 | btrfs_add_dead_root(root); | |
4198 | } | |
4199 | } | |
4200 | ||
143bede5 | 4201 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4202 | { |
4203 | struct rb_node *node; | |
4204 | struct rb_node *prev; | |
4205 | struct btrfs_inode *entry; | |
4206 | struct inode *inode; | |
4207 | u64 objectid = 0; | |
4208 | ||
4209 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4210 | ||
4211 | spin_lock(&root->inode_lock); | |
4212 | again: | |
4213 | node = root->inode_tree.rb_node; | |
4214 | prev = NULL; | |
4215 | while (node) { | |
4216 | prev = node; | |
4217 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4218 | ||
33345d01 | 4219 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4220 | node = node->rb_left; |
33345d01 | 4221 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4222 | node = node->rb_right; |
4223 | else | |
4224 | break; | |
4225 | } | |
4226 | if (!node) { | |
4227 | while (prev) { | |
4228 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4229 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4230 | node = prev; |
4231 | break; | |
4232 | } | |
4233 | prev = rb_next(prev); | |
4234 | } | |
4235 | } | |
4236 | while (node) { | |
4237 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4238 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4239 | inode = igrab(&entry->vfs_inode); |
4240 | if (inode) { | |
4241 | spin_unlock(&root->inode_lock); | |
4242 | if (atomic_read(&inode->i_count) > 1) | |
4243 | d_prune_aliases(inode); | |
4244 | /* | |
45321ac5 | 4245 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4246 | * the inode cache when its usage count |
4247 | * hits zero. | |
4248 | */ | |
4249 | iput(inode); | |
4250 | cond_resched(); | |
4251 | spin_lock(&root->inode_lock); | |
4252 | goto again; | |
4253 | } | |
4254 | ||
4255 | if (cond_resched_lock(&root->inode_lock)) | |
4256 | goto again; | |
4257 | ||
4258 | node = rb_next(node); | |
4259 | } | |
4260 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4261 | } |
4262 | ||
e02119d5 CM |
4263 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4264 | { | |
4265 | struct btrfs_iget_args *args = p; | |
4266 | inode->i_ino = args->ino; | |
e02119d5 | 4267 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4268 | return 0; |
4269 | } | |
4270 | ||
4271 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4272 | { | |
4273 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4274 | return args->ino == btrfs_ino(inode) && |
d397712b | 4275 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4276 | } |
4277 | ||
5d4f98a2 YZ |
4278 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4279 | u64 objectid, | |
4280 | struct btrfs_root *root) | |
39279cc3 CM |
4281 | { |
4282 | struct inode *inode; | |
4283 | struct btrfs_iget_args args; | |
4284 | args.ino = objectid; | |
4285 | args.root = root; | |
4286 | ||
4287 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4288 | btrfs_init_locked_inode, | |
4289 | (void *)&args); | |
4290 | return inode; | |
4291 | } | |
4292 | ||
1a54ef8c BR |
4293 | /* Get an inode object given its location and corresponding root. |
4294 | * Returns in *is_new if the inode was read from disk | |
4295 | */ | |
4296 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4297 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4298 | { |
4299 | struct inode *inode; | |
4300 | ||
4301 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4302 | if (!inode) | |
5d4f98a2 | 4303 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4304 | |
4305 | if (inode->i_state & I_NEW) { | |
4306 | BTRFS_I(inode)->root = root; | |
4307 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4308 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4309 | if (!is_bad_inode(inode)) { |
4310 | inode_tree_add(inode); | |
4311 | unlock_new_inode(inode); | |
4312 | if (new) | |
4313 | *new = 1; | |
4314 | } else { | |
e0b6d65b ST |
4315 | unlock_new_inode(inode); |
4316 | iput(inode); | |
4317 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4318 | } |
4319 | } | |
4320 | ||
1a54ef8c BR |
4321 | return inode; |
4322 | } | |
4323 | ||
4df27c4d YZ |
4324 | static struct inode *new_simple_dir(struct super_block *s, |
4325 | struct btrfs_key *key, | |
4326 | struct btrfs_root *root) | |
4327 | { | |
4328 | struct inode *inode = new_inode(s); | |
4329 | ||
4330 | if (!inode) | |
4331 | return ERR_PTR(-ENOMEM); | |
4332 | ||
4df27c4d YZ |
4333 | BTRFS_I(inode)->root = root; |
4334 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4335 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4336 | |
4337 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4338 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4339 | inode->i_fop = &simple_dir_operations; |
4340 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4341 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4342 | ||
4343 | return inode; | |
4344 | } | |
4345 | ||
3de4586c | 4346 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4347 | { |
d397712b | 4348 | struct inode *inode; |
4df27c4d | 4349 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4350 | struct btrfs_root *sub_root = root; |
4351 | struct btrfs_key location; | |
76dda93c | 4352 | int index; |
b4aff1f8 | 4353 | int ret = 0; |
39279cc3 CM |
4354 | |
4355 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4356 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4357 | |
b4aff1f8 JB |
4358 | if (unlikely(d_need_lookup(dentry))) { |
4359 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4360 | kfree(dentry->d_fsdata); | |
4361 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4362 | /* This thing is hashed, drop it for now */ |
4363 | d_drop(dentry); | |
b4aff1f8 JB |
4364 | } else { |
4365 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4366 | } | |
5f39d397 | 4367 | |
39279cc3 CM |
4368 | if (ret < 0) |
4369 | return ERR_PTR(ret); | |
5f39d397 | 4370 | |
4df27c4d YZ |
4371 | if (location.objectid == 0) |
4372 | return NULL; | |
4373 | ||
4374 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4375 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4376 | return inode; |
4377 | } | |
4378 | ||
4379 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4380 | ||
76dda93c | 4381 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4382 | ret = fixup_tree_root_location(root, dir, dentry, |
4383 | &location, &sub_root); | |
4384 | if (ret < 0) { | |
4385 | if (ret != -ENOENT) | |
4386 | inode = ERR_PTR(ret); | |
4387 | else | |
4388 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4389 | } else { | |
73f73415 | 4390 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4391 | } |
76dda93c YZ |
4392 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4393 | ||
34d19bad | 4394 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4395 | down_read(&root->fs_info->cleanup_work_sem); |
4396 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4397 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4398 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4399 | if (ret) |
4400 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4401 | } |
4402 | ||
3de4586c CM |
4403 | return inode; |
4404 | } | |
4405 | ||
fe15ce44 | 4406 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4407 | { |
4408 | struct btrfs_root *root; | |
848cce0d | 4409 | struct inode *inode = dentry->d_inode; |
76dda93c | 4410 | |
848cce0d LZ |
4411 | if (!inode && !IS_ROOT(dentry)) |
4412 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4413 | |
848cce0d LZ |
4414 | if (inode) { |
4415 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4416 | if (btrfs_root_refs(&root->root_item) == 0) |
4417 | return 1; | |
848cce0d LZ |
4418 | |
4419 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4420 | return 1; | |
efefb143 | 4421 | } |
76dda93c YZ |
4422 | return 0; |
4423 | } | |
4424 | ||
b4aff1f8 JB |
4425 | static void btrfs_dentry_release(struct dentry *dentry) |
4426 | { | |
4427 | if (dentry->d_fsdata) | |
4428 | kfree(dentry->d_fsdata); | |
4429 | } | |
4430 | ||
3de4586c | 4431 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4432 | unsigned int flags) |
3de4586c | 4433 | { |
a66e7cc6 JB |
4434 | struct dentry *ret; |
4435 | ||
4436 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4437 | if (unlikely(d_need_lookup(dentry))) { | |
4438 | spin_lock(&dentry->d_lock); | |
4439 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4440 | spin_unlock(&dentry->d_lock); | |
4441 | } | |
4442 | return ret; | |
39279cc3 CM |
4443 | } |
4444 | ||
16cdcec7 | 4445 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4446 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4447 | }; | |
4448 | ||
cbdf5a24 DW |
4449 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4450 | filldir_t filldir) | |
39279cc3 | 4451 | { |
6da6abae | 4452 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4453 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4454 | struct btrfs_item *item; | |
4455 | struct btrfs_dir_item *di; | |
4456 | struct btrfs_key key; | |
5f39d397 | 4457 | struct btrfs_key found_key; |
39279cc3 | 4458 | struct btrfs_path *path; |
16cdcec7 MX |
4459 | struct list_head ins_list; |
4460 | struct list_head del_list; | |
39279cc3 | 4461 | int ret; |
5f39d397 | 4462 | struct extent_buffer *leaf; |
39279cc3 | 4463 | int slot; |
39279cc3 CM |
4464 | unsigned char d_type; |
4465 | int over = 0; | |
4466 | u32 di_cur; | |
4467 | u32 di_total; | |
4468 | u32 di_len; | |
4469 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4470 | char tmp_name[32]; |
4471 | char *name_ptr; | |
4472 | int name_len; | |
16cdcec7 | 4473 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4474 | |
4475 | /* FIXME, use a real flag for deciding about the key type */ | |
4476 | if (root->fs_info->tree_root == root) | |
4477 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4478 | |
3954401f CM |
4479 | /* special case for "." */ |
4480 | if (filp->f_pos == 0) { | |
3765fefa HS |
4481 | over = filldir(dirent, ".", 1, |
4482 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4483 | if (over) |
4484 | return 0; | |
4485 | filp->f_pos = 1; | |
4486 | } | |
3954401f CM |
4487 | /* special case for .., just use the back ref */ |
4488 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4489 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4490 | over = filldir(dirent, "..", 2, |
3765fefa | 4491 | filp->f_pos, pino, DT_DIR); |
3954401f | 4492 | if (over) |
49593bfa | 4493 | return 0; |
3954401f CM |
4494 | filp->f_pos = 2; |
4495 | } | |
49593bfa | 4496 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4497 | if (!path) |
4498 | return -ENOMEM; | |
ff5714cc | 4499 | |
026fd317 | 4500 | path->reada = 1; |
49593bfa | 4501 | |
16cdcec7 MX |
4502 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4503 | INIT_LIST_HEAD(&ins_list); | |
4504 | INIT_LIST_HEAD(&del_list); | |
4505 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4506 | } | |
4507 | ||
39279cc3 CM |
4508 | btrfs_set_key_type(&key, key_type); |
4509 | key.offset = filp->f_pos; | |
33345d01 | 4510 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4511 | |
39279cc3 CM |
4512 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4513 | if (ret < 0) | |
4514 | goto err; | |
49593bfa DW |
4515 | |
4516 | while (1) { | |
5f39d397 | 4517 | leaf = path->nodes[0]; |
39279cc3 | 4518 | slot = path->slots[0]; |
b9e03af0 LZ |
4519 | if (slot >= btrfs_header_nritems(leaf)) { |
4520 | ret = btrfs_next_leaf(root, path); | |
4521 | if (ret < 0) | |
4522 | goto err; | |
4523 | else if (ret > 0) | |
4524 | break; | |
4525 | continue; | |
39279cc3 | 4526 | } |
3de4586c | 4527 | |
5f39d397 CM |
4528 | item = btrfs_item_nr(leaf, slot); |
4529 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4530 | ||
4531 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4532 | break; |
5f39d397 | 4533 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4534 | break; |
5f39d397 | 4535 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4536 | goto next; |
16cdcec7 MX |
4537 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4538 | btrfs_should_delete_dir_index(&del_list, | |
4539 | found_key.offset)) | |
4540 | goto next; | |
5f39d397 CM |
4541 | |
4542 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4543 | is_curr = 1; |
49593bfa | 4544 | |
39279cc3 CM |
4545 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4546 | di_cur = 0; | |
5f39d397 | 4547 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4548 | |
4549 | while (di_cur < di_total) { | |
5f39d397 CM |
4550 | struct btrfs_key location; |
4551 | ||
22a94d44 JB |
4552 | if (verify_dir_item(root, leaf, di)) |
4553 | break; | |
4554 | ||
5f39d397 | 4555 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4556 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4557 | name_ptr = tmp_name; |
4558 | } else { | |
4559 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4560 | if (!name_ptr) { |
4561 | ret = -ENOMEM; | |
4562 | goto err; | |
4563 | } | |
5f39d397 CM |
4564 | } |
4565 | read_extent_buffer(leaf, name_ptr, | |
4566 | (unsigned long)(di + 1), name_len); | |
4567 | ||
4568 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4569 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4570 | |
fede766f | 4571 | |
3de4586c | 4572 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4573 | * skip it. |
4574 | * | |
4575 | * In contrast to old kernels, we insert the snapshot's | |
4576 | * dir item and dir index after it has been created, so | |
4577 | * we won't find a reference to our own snapshot. We | |
4578 | * still keep the following code for backward | |
4579 | * compatibility. | |
3de4586c CM |
4580 | */ |
4581 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4582 | location.objectid == root->root_key.objectid) { | |
4583 | over = 0; | |
4584 | goto skip; | |
4585 | } | |
5f39d397 | 4586 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4587 | found_key.offset, location.objectid, |
39279cc3 | 4588 | d_type); |
5f39d397 | 4589 | |
3de4586c | 4590 | skip: |
5f39d397 CM |
4591 | if (name_ptr != tmp_name) |
4592 | kfree(name_ptr); | |
4593 | ||
39279cc3 CM |
4594 | if (over) |
4595 | goto nopos; | |
5103e947 | 4596 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4597 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4598 | di_cur += di_len; |
4599 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4600 | } | |
b9e03af0 LZ |
4601 | next: |
4602 | path->slots[0]++; | |
39279cc3 | 4603 | } |
49593bfa | 4604 | |
16cdcec7 MX |
4605 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4606 | if (is_curr) | |
4607 | filp->f_pos++; | |
4608 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4609 | &ins_list); | |
4610 | if (ret) | |
4611 | goto nopos; | |
4612 | } | |
4613 | ||
49593bfa | 4614 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4615 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4616 | /* |
4617 | * 32-bit glibc will use getdents64, but then strtol - | |
4618 | * so the last number we can serve is this. | |
4619 | */ | |
4620 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4621 | else |
4622 | filp->f_pos++; | |
39279cc3 CM |
4623 | nopos: |
4624 | ret = 0; | |
4625 | err: | |
16cdcec7 MX |
4626 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4627 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4628 | btrfs_free_path(path); |
39279cc3 CM |
4629 | return ret; |
4630 | } | |
4631 | ||
a9185b41 | 4632 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4633 | { |
4634 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4635 | struct btrfs_trans_handle *trans; | |
4636 | int ret = 0; | |
0af3d00b | 4637 | bool nolock = false; |
39279cc3 | 4638 | |
72ac3c0d | 4639 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4640 | return 0; |
4641 | ||
83eea1f1 | 4642 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 4643 | nolock = true; |
0af3d00b | 4644 | |
a9185b41 | 4645 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4646 | if (nolock) |
7a7eaa40 | 4647 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4648 | else |
7a7eaa40 | 4649 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4650 | if (IS_ERR(trans)) |
4651 | return PTR_ERR(trans); | |
a698d075 | 4652 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
4653 | } |
4654 | return ret; | |
4655 | } | |
4656 | ||
4657 | /* | |
54aa1f4d | 4658 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4659 | * inode changes. But, it is most likely to find the inode in cache. |
4660 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4661 | * to keep or drop this code. | |
4662 | */ | |
22c44fe6 | 4663 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4664 | { |
4665 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4666 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4667 | int ret; |
4668 | ||
72ac3c0d | 4669 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4670 | return 0; |
39279cc3 | 4671 | |
7a7eaa40 | 4672 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4673 | if (IS_ERR(trans)) |
4674 | return PTR_ERR(trans); | |
8929ecfa YZ |
4675 | |
4676 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4677 | if (ret && ret == -ENOSPC) { |
4678 | /* whoops, lets try again with the full transaction */ | |
4679 | btrfs_end_transaction(trans, root); | |
4680 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4681 | if (IS_ERR(trans)) |
4682 | return PTR_ERR(trans); | |
8929ecfa | 4683 | |
94b60442 | 4684 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4685 | } |
39279cc3 | 4686 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4687 | if (BTRFS_I(inode)->delayed_node) |
4688 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4689 | |
4690 | return ret; | |
4691 | } | |
4692 | ||
4693 | /* | |
4694 | * This is a copy of file_update_time. We need this so we can return error on | |
4695 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4696 | */ | |
e41f941a JB |
4697 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
4698 | int flags) | |
22c44fe6 | 4699 | { |
2bc55652 AB |
4700 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4701 | ||
4702 | if (btrfs_root_readonly(root)) | |
4703 | return -EROFS; | |
4704 | ||
e41f941a | 4705 | if (flags & S_VERSION) |
22c44fe6 | 4706 | inode_inc_iversion(inode); |
e41f941a JB |
4707 | if (flags & S_CTIME) |
4708 | inode->i_ctime = *now; | |
4709 | if (flags & S_MTIME) | |
4710 | inode->i_mtime = *now; | |
4711 | if (flags & S_ATIME) | |
4712 | inode->i_atime = *now; | |
4713 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
4714 | } |
4715 | ||
d352ac68 CM |
4716 | /* |
4717 | * find the highest existing sequence number in a directory | |
4718 | * and then set the in-memory index_cnt variable to reflect | |
4719 | * free sequence numbers | |
4720 | */ | |
aec7477b JB |
4721 | static int btrfs_set_inode_index_count(struct inode *inode) |
4722 | { | |
4723 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4724 | struct btrfs_key key, found_key; | |
4725 | struct btrfs_path *path; | |
4726 | struct extent_buffer *leaf; | |
4727 | int ret; | |
4728 | ||
33345d01 | 4729 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4730 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4731 | key.offset = (u64)-1; | |
4732 | ||
4733 | path = btrfs_alloc_path(); | |
4734 | if (!path) | |
4735 | return -ENOMEM; | |
4736 | ||
4737 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4738 | if (ret < 0) | |
4739 | goto out; | |
4740 | /* FIXME: we should be able to handle this */ | |
4741 | if (ret == 0) | |
4742 | goto out; | |
4743 | ret = 0; | |
4744 | ||
4745 | /* | |
4746 | * MAGIC NUMBER EXPLANATION: | |
4747 | * since we search a directory based on f_pos we have to start at 2 | |
4748 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4749 | * else has to start at 2 | |
4750 | */ | |
4751 | if (path->slots[0] == 0) { | |
4752 | BTRFS_I(inode)->index_cnt = 2; | |
4753 | goto out; | |
4754 | } | |
4755 | ||
4756 | path->slots[0]--; | |
4757 | ||
4758 | leaf = path->nodes[0]; | |
4759 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4760 | ||
33345d01 | 4761 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4762 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4763 | BTRFS_I(inode)->index_cnt = 2; | |
4764 | goto out; | |
4765 | } | |
4766 | ||
4767 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4768 | out: | |
4769 | btrfs_free_path(path); | |
4770 | return ret; | |
4771 | } | |
4772 | ||
d352ac68 CM |
4773 | /* |
4774 | * helper to find a free sequence number in a given directory. This current | |
4775 | * code is very simple, later versions will do smarter things in the btree | |
4776 | */ | |
3de4586c | 4777 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4778 | { |
4779 | int ret = 0; | |
4780 | ||
4781 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4782 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4783 | if (ret) { | |
4784 | ret = btrfs_set_inode_index_count(dir); | |
4785 | if (ret) | |
4786 | return ret; | |
4787 | } | |
aec7477b JB |
4788 | } |
4789 | ||
00e4e6b3 | 4790 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4791 | BTRFS_I(dir)->index_cnt++; |
4792 | ||
4793 | return ret; | |
4794 | } | |
4795 | ||
39279cc3 CM |
4796 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4797 | struct btrfs_root *root, | |
aec7477b | 4798 | struct inode *dir, |
9c58309d | 4799 | const char *name, int name_len, |
175a4eb7 AV |
4800 | u64 ref_objectid, u64 objectid, |
4801 | umode_t mode, u64 *index) | |
39279cc3 CM |
4802 | { |
4803 | struct inode *inode; | |
5f39d397 | 4804 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4805 | struct btrfs_key *location; |
5f39d397 | 4806 | struct btrfs_path *path; |
9c58309d CM |
4807 | struct btrfs_inode_ref *ref; |
4808 | struct btrfs_key key[2]; | |
4809 | u32 sizes[2]; | |
4810 | unsigned long ptr; | |
39279cc3 CM |
4811 | int ret; |
4812 | int owner; | |
4813 | ||
5f39d397 | 4814 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4815 | if (!path) |
4816 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4817 | |
39279cc3 | 4818 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4819 | if (!inode) { |
4820 | btrfs_free_path(path); | |
39279cc3 | 4821 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4822 | } |
39279cc3 | 4823 | |
581bb050 LZ |
4824 | /* |
4825 | * we have to initialize this early, so we can reclaim the inode | |
4826 | * number if we fail afterwards in this function. | |
4827 | */ | |
4828 | inode->i_ino = objectid; | |
4829 | ||
aec7477b | 4830 | if (dir) { |
1abe9b8a | 4831 | trace_btrfs_inode_request(dir); |
4832 | ||
3de4586c | 4833 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4834 | if (ret) { |
8fb27640 | 4835 | btrfs_free_path(path); |
09771430 | 4836 | iput(inode); |
aec7477b | 4837 | return ERR_PTR(ret); |
09771430 | 4838 | } |
aec7477b JB |
4839 | } |
4840 | /* | |
4841 | * index_cnt is ignored for everything but a dir, | |
4842 | * btrfs_get_inode_index_count has an explanation for the magic | |
4843 | * number | |
4844 | */ | |
4845 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4846 | BTRFS_I(inode)->root = root; |
e02119d5 | 4847 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4848 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 4849 | |
5dc562c5 JB |
4850 | /* |
4851 | * We could have gotten an inode number from somebody who was fsynced | |
4852 | * and then removed in this same transaction, so let's just set full | |
4853 | * sync since it will be a full sync anyway and this will blow away the | |
4854 | * old info in the log. | |
4855 | */ | |
4856 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
4857 | ||
569254b0 | 4858 | if (S_ISDIR(mode)) |
39279cc3 CM |
4859 | owner = 0; |
4860 | else | |
4861 | owner = 1; | |
9c58309d CM |
4862 | |
4863 | key[0].objectid = objectid; | |
4864 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4865 | key[0].offset = 0; | |
4866 | ||
f186373f MF |
4867 | /* |
4868 | * Start new inodes with an inode_ref. This is slightly more | |
4869 | * efficient for small numbers of hard links since they will | |
4870 | * be packed into one item. Extended refs will kick in if we | |
4871 | * add more hard links than can fit in the ref item. | |
4872 | */ | |
9c58309d CM |
4873 | key[1].objectid = objectid; |
4874 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4875 | key[1].offset = ref_objectid; | |
4876 | ||
4877 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4878 | sizes[1] = name_len + sizeof(*ref); | |
4879 | ||
b9473439 | 4880 | path->leave_spinning = 1; |
9c58309d CM |
4881 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4882 | if (ret != 0) | |
5f39d397 CM |
4883 | goto fail; |
4884 | ||
ecc11fab | 4885 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4886 | inode_set_bytes(inode, 0); |
39279cc3 | 4887 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4888 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4889 | struct btrfs_inode_item); | |
293f7e07 LZ |
4890 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
4891 | sizeof(*inode_item)); | |
e02119d5 | 4892 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4893 | |
4894 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4895 | struct btrfs_inode_ref); | |
4896 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4897 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4898 | ptr = (unsigned long)(ref + 1); |
4899 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4900 | ||
5f39d397 CM |
4901 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4902 | btrfs_free_path(path); | |
4903 | ||
39279cc3 CM |
4904 | location = &BTRFS_I(inode)->location; |
4905 | location->objectid = objectid; | |
39279cc3 CM |
4906 | location->offset = 0; |
4907 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4908 | ||
6cbff00f CH |
4909 | btrfs_inherit_iflags(inode, dir); |
4910 | ||
569254b0 | 4911 | if (S_ISREG(mode)) { |
94272164 CM |
4912 | if (btrfs_test_opt(root, NODATASUM)) |
4913 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 4914 | if (btrfs_test_opt(root, NODATACOW)) |
94272164 CM |
4915 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4916 | } | |
4917 | ||
39279cc3 | 4918 | insert_inode_hash(inode); |
5d4f98a2 | 4919 | inode_tree_add(inode); |
1abe9b8a | 4920 | |
4921 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4922 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4923 | |
8ea05e3a AB |
4924 | btrfs_update_root_times(trans, root); |
4925 | ||
39279cc3 | 4926 | return inode; |
5f39d397 | 4927 | fail: |
aec7477b JB |
4928 | if (dir) |
4929 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4930 | btrfs_free_path(path); |
09771430 | 4931 | iput(inode); |
5f39d397 | 4932 | return ERR_PTR(ret); |
39279cc3 CM |
4933 | } |
4934 | ||
4935 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4936 | { | |
4937 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4938 | } | |
4939 | ||
d352ac68 CM |
4940 | /* |
4941 | * utility function to add 'inode' into 'parent_inode' with | |
4942 | * a give name and a given sequence number. | |
4943 | * if 'add_backref' is true, also insert a backref from the | |
4944 | * inode to the parent directory. | |
4945 | */ | |
e02119d5 CM |
4946 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4947 | struct inode *parent_inode, struct inode *inode, | |
4948 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4949 | { |
4df27c4d | 4950 | int ret = 0; |
39279cc3 | 4951 | struct btrfs_key key; |
e02119d5 | 4952 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4953 | u64 ino = btrfs_ino(inode); |
4954 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4955 | |
33345d01 | 4956 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4957 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4958 | } else { | |
33345d01 | 4959 | key.objectid = ino; |
4df27c4d YZ |
4960 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4961 | key.offset = 0; | |
4962 | } | |
4963 | ||
33345d01 | 4964 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4965 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4966 | key.objectid, root->root_key.objectid, | |
33345d01 | 4967 | parent_ino, index, name, name_len); |
4df27c4d | 4968 | } else if (add_backref) { |
33345d01 LZ |
4969 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4970 | parent_ino, index); | |
4df27c4d | 4971 | } |
39279cc3 | 4972 | |
79787eaa JM |
4973 | /* Nothing to clean up yet */ |
4974 | if (ret) | |
4975 | return ret; | |
4df27c4d | 4976 | |
79787eaa JM |
4977 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4978 | parent_inode, &key, | |
4979 | btrfs_inode_type(inode), index); | |
9c52057c | 4980 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
4981 | goto fail_dir_item; |
4982 | else if (ret) { | |
4983 | btrfs_abort_transaction(trans, root, ret); | |
4984 | return ret; | |
39279cc3 | 4985 | } |
79787eaa JM |
4986 | |
4987 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4988 | name_len * 2); | |
0c4d2d95 | 4989 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4990 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4991 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4992 | if (ret) | |
4993 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4994 | return ret; |
fe66a05a CM |
4995 | |
4996 | fail_dir_item: | |
4997 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4998 | u64 local_index; | |
4999 | int err; | |
5000 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
5001 | key.objectid, root->root_key.objectid, | |
5002 | parent_ino, &local_index, name, name_len); | |
5003 | ||
5004 | } else if (add_backref) { | |
5005 | u64 local_index; | |
5006 | int err; | |
5007 | ||
5008 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
5009 | ino, parent_ino, &local_index); | |
5010 | } | |
5011 | return ret; | |
39279cc3 CM |
5012 | } |
5013 | ||
5014 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
5015 | struct inode *dir, struct dentry *dentry, |
5016 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 5017 | { |
a1b075d2 JB |
5018 | int err = btrfs_add_link(trans, dir, inode, |
5019 | dentry->d_name.name, dentry->d_name.len, | |
5020 | backref, index); | |
39279cc3 CM |
5021 | if (err > 0) |
5022 | err = -EEXIST; | |
5023 | return err; | |
5024 | } | |
5025 | ||
618e21d5 | 5026 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 5027 | umode_t mode, dev_t rdev) |
618e21d5 JB |
5028 | { |
5029 | struct btrfs_trans_handle *trans; | |
5030 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5031 | struct inode *inode = NULL; |
618e21d5 JB |
5032 | int err; |
5033 | int drop_inode = 0; | |
5034 | u64 objectid; | |
00e4e6b3 | 5035 | u64 index = 0; |
618e21d5 JB |
5036 | |
5037 | if (!new_valid_dev(rdev)) | |
5038 | return -EINVAL; | |
5039 | ||
9ed74f2d JB |
5040 | /* |
5041 | * 2 for inode item and ref | |
5042 | * 2 for dir items | |
5043 | * 1 for xattr if selinux is on | |
5044 | */ | |
a22285a6 YZ |
5045 | trans = btrfs_start_transaction(root, 5); |
5046 | if (IS_ERR(trans)) | |
5047 | return PTR_ERR(trans); | |
1832a6d5 | 5048 | |
581bb050 LZ |
5049 | err = btrfs_find_free_ino(root, &objectid); |
5050 | if (err) | |
5051 | goto out_unlock; | |
5052 | ||
aec7477b | 5053 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5054 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5055 | mode, &index); |
7cf96da3 TI |
5056 | if (IS_ERR(inode)) { |
5057 | err = PTR_ERR(inode); | |
618e21d5 | 5058 | goto out_unlock; |
7cf96da3 | 5059 | } |
618e21d5 | 5060 | |
2a7dba39 | 5061 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5062 | if (err) { |
5063 | drop_inode = 1; | |
5064 | goto out_unlock; | |
5065 | } | |
5066 | ||
ad19db71 CS |
5067 | /* |
5068 | * If the active LSM wants to access the inode during | |
5069 | * d_instantiate it needs these. Smack checks to see | |
5070 | * if the filesystem supports xattrs by looking at the | |
5071 | * ops vector. | |
5072 | */ | |
5073 | ||
5074 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 5075 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
5076 | if (err) |
5077 | drop_inode = 1; | |
5078 | else { | |
618e21d5 | 5079 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 5080 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 5081 | d_instantiate(dentry, inode); |
618e21d5 | 5082 | } |
618e21d5 | 5083 | out_unlock: |
7ad85bb7 | 5084 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5085 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
5086 | if (drop_inode) { |
5087 | inode_dec_link_count(inode); | |
5088 | iput(inode); | |
5089 | } | |
618e21d5 JB |
5090 | return err; |
5091 | } | |
5092 | ||
39279cc3 | 5093 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 5094 | umode_t mode, bool excl) |
39279cc3 CM |
5095 | { |
5096 | struct btrfs_trans_handle *trans; | |
5097 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 5098 | struct inode *inode = NULL; |
43baa579 | 5099 | int drop_inode_on_err = 0; |
a22285a6 | 5100 | int err; |
39279cc3 | 5101 | u64 objectid; |
00e4e6b3 | 5102 | u64 index = 0; |
39279cc3 | 5103 | |
9ed74f2d JB |
5104 | /* |
5105 | * 2 for inode item and ref | |
5106 | * 2 for dir items | |
5107 | * 1 for xattr if selinux is on | |
5108 | */ | |
a22285a6 YZ |
5109 | trans = btrfs_start_transaction(root, 5); |
5110 | if (IS_ERR(trans)) | |
5111 | return PTR_ERR(trans); | |
9ed74f2d | 5112 | |
581bb050 LZ |
5113 | err = btrfs_find_free_ino(root, &objectid); |
5114 | if (err) | |
5115 | goto out_unlock; | |
5116 | ||
aec7477b | 5117 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5118 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5119 | mode, &index); |
7cf96da3 TI |
5120 | if (IS_ERR(inode)) { |
5121 | err = PTR_ERR(inode); | |
39279cc3 | 5122 | goto out_unlock; |
7cf96da3 | 5123 | } |
43baa579 | 5124 | drop_inode_on_err = 1; |
39279cc3 | 5125 | |
2a7dba39 | 5126 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5127 | if (err) |
33268eaf | 5128 | goto out_unlock; |
33268eaf | 5129 | |
9185aa58 FB |
5130 | err = btrfs_update_inode(trans, root, inode); |
5131 | if (err) | |
5132 | goto out_unlock; | |
5133 | ||
ad19db71 CS |
5134 | /* |
5135 | * If the active LSM wants to access the inode during | |
5136 | * d_instantiate it needs these. Smack checks to see | |
5137 | * if the filesystem supports xattrs by looking at the | |
5138 | * ops vector. | |
5139 | */ | |
5140 | inode->i_fop = &btrfs_file_operations; | |
5141 | inode->i_op = &btrfs_file_inode_operations; | |
5142 | ||
a1b075d2 | 5143 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5144 | if (err) |
43baa579 FB |
5145 | goto out_unlock; |
5146 | ||
5147 | inode->i_mapping->a_ops = &btrfs_aops; | |
5148 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5149 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5150 | d_instantiate(dentry, inode); | |
5151 | ||
39279cc3 | 5152 | out_unlock: |
7ad85bb7 | 5153 | btrfs_end_transaction(trans, root); |
43baa579 | 5154 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5155 | inode_dec_link_count(inode); |
5156 | iput(inode); | |
5157 | } | |
b53d3f5d | 5158 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5159 | return err; |
5160 | } | |
5161 | ||
5162 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5163 | struct dentry *dentry) | |
5164 | { | |
5165 | struct btrfs_trans_handle *trans; | |
5166 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5167 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5168 | u64 index; |
39279cc3 CM |
5169 | int err; |
5170 | int drop_inode = 0; | |
5171 | ||
4a8be425 TH |
5172 | /* do not allow sys_link's with other subvols of the same device */ |
5173 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5174 | return -EXDEV; |
4a8be425 | 5175 | |
f186373f | 5176 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5177 | return -EMLINK; |
4a8be425 | 5178 | |
3de4586c | 5179 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5180 | if (err) |
5181 | goto fail; | |
5182 | ||
a22285a6 | 5183 | /* |
7e6b6465 | 5184 | * 2 items for inode and inode ref |
a22285a6 | 5185 | * 2 items for dir items |
7e6b6465 | 5186 | * 1 item for parent inode |
a22285a6 | 5187 | */ |
7e6b6465 | 5188 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5189 | if (IS_ERR(trans)) { |
5190 | err = PTR_ERR(trans); | |
5191 | goto fail; | |
5192 | } | |
5f39d397 | 5193 | |
3153495d | 5194 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5195 | inode_inc_iversion(inode); |
3153495d | 5196 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5197 | ihold(inode); |
e9976151 | 5198 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 5199 | |
a1b075d2 | 5200 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5201 | |
a5719521 | 5202 | if (err) { |
54aa1f4d | 5203 | drop_inode = 1; |
a5719521 | 5204 | } else { |
10d9f309 | 5205 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5206 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5207 | if (err) |
5208 | goto fail; | |
08c422c2 | 5209 | d_instantiate(dentry, inode); |
6a912213 | 5210 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5211 | } |
39279cc3 | 5212 | |
7ad85bb7 | 5213 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5214 | fail: |
39279cc3 CM |
5215 | if (drop_inode) { |
5216 | inode_dec_link_count(inode); | |
5217 | iput(inode); | |
5218 | } | |
b53d3f5d | 5219 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5220 | return err; |
5221 | } | |
5222 | ||
18bb1db3 | 5223 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5224 | { |
b9d86667 | 5225 | struct inode *inode = NULL; |
39279cc3 CM |
5226 | struct btrfs_trans_handle *trans; |
5227 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5228 | int err = 0; | |
5229 | int drop_on_err = 0; | |
b9d86667 | 5230 | u64 objectid = 0; |
00e4e6b3 | 5231 | u64 index = 0; |
39279cc3 | 5232 | |
9ed74f2d JB |
5233 | /* |
5234 | * 2 items for inode and ref | |
5235 | * 2 items for dir items | |
5236 | * 1 for xattr if selinux is on | |
5237 | */ | |
a22285a6 YZ |
5238 | trans = btrfs_start_transaction(root, 5); |
5239 | if (IS_ERR(trans)) | |
5240 | return PTR_ERR(trans); | |
39279cc3 | 5241 | |
581bb050 LZ |
5242 | err = btrfs_find_free_ino(root, &objectid); |
5243 | if (err) | |
5244 | goto out_fail; | |
5245 | ||
aec7477b | 5246 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5247 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5248 | S_IFDIR | mode, &index); |
39279cc3 CM |
5249 | if (IS_ERR(inode)) { |
5250 | err = PTR_ERR(inode); | |
5251 | goto out_fail; | |
5252 | } | |
5f39d397 | 5253 | |
39279cc3 | 5254 | drop_on_err = 1; |
33268eaf | 5255 | |
2a7dba39 | 5256 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5257 | if (err) |
5258 | goto out_fail; | |
5259 | ||
39279cc3 CM |
5260 | inode->i_op = &btrfs_dir_inode_operations; |
5261 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5262 | |
dbe674a9 | 5263 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5264 | err = btrfs_update_inode(trans, root, inode); |
5265 | if (err) | |
5266 | goto out_fail; | |
5f39d397 | 5267 | |
a1b075d2 JB |
5268 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5269 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5270 | if (err) |
5271 | goto out_fail; | |
5f39d397 | 5272 | |
39279cc3 CM |
5273 | d_instantiate(dentry, inode); |
5274 | drop_on_err = 0; | |
39279cc3 CM |
5275 | |
5276 | out_fail: | |
7ad85bb7 | 5277 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5278 | if (drop_on_err) |
5279 | iput(inode); | |
b53d3f5d | 5280 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5281 | return err; |
5282 | } | |
5283 | ||
d352ac68 CM |
5284 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5285 | * and an extent that you want to insert, deal with overlap and insert | |
5286 | * the new extent into the tree. | |
5287 | */ | |
3b951516 CM |
5288 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5289 | struct extent_map *existing, | |
e6dcd2dc CM |
5290 | struct extent_map *em, |
5291 | u64 map_start, u64 map_len) | |
3b951516 CM |
5292 | { |
5293 | u64 start_diff; | |
3b951516 | 5294 | |
e6dcd2dc CM |
5295 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5296 | start_diff = map_start - em->start; | |
5297 | em->start = map_start; | |
5298 | em->len = map_len; | |
c8b97818 CM |
5299 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5300 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5301 | em->block_start += start_diff; |
c8b97818 CM |
5302 | em->block_len -= start_diff; |
5303 | } | |
e6dcd2dc | 5304 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5305 | } |
5306 | ||
c8b97818 CM |
5307 | static noinline int uncompress_inline(struct btrfs_path *path, |
5308 | struct inode *inode, struct page *page, | |
5309 | size_t pg_offset, u64 extent_offset, | |
5310 | struct btrfs_file_extent_item *item) | |
5311 | { | |
5312 | int ret; | |
5313 | struct extent_buffer *leaf = path->nodes[0]; | |
5314 | char *tmp; | |
5315 | size_t max_size; | |
5316 | unsigned long inline_size; | |
5317 | unsigned long ptr; | |
261507a0 | 5318 | int compress_type; |
c8b97818 CM |
5319 | |
5320 | WARN_ON(pg_offset != 0); | |
261507a0 | 5321 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5322 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5323 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5324 | btrfs_item_nr(leaf, path->slots[0])); | |
5325 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5326 | if (!tmp) |
5327 | return -ENOMEM; | |
c8b97818 CM |
5328 | ptr = btrfs_file_extent_inline_start(item); |
5329 | ||
5330 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5331 | ||
5b050f04 | 5332 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5333 | ret = btrfs_decompress(compress_type, tmp, page, |
5334 | extent_offset, inline_size, max_size); | |
c8b97818 | 5335 | if (ret) { |
7ac687d9 | 5336 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5337 | unsigned long copy_size = min_t(u64, |
5338 | PAGE_CACHE_SIZE - pg_offset, | |
5339 | max_size - extent_offset); | |
5340 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5341 | kunmap_atomic(kaddr); |
c8b97818 CM |
5342 | } |
5343 | kfree(tmp); | |
5344 | return 0; | |
5345 | } | |
5346 | ||
d352ac68 CM |
5347 | /* |
5348 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5349 | * the ugly parts come from merging extents from the disk with the in-ram |
5350 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5351 | * where the in-ram extents might be locked pending data=ordered completion. |
5352 | * | |
5353 | * This also copies inline extents directly into the page. | |
5354 | */ | |
d397712b | 5355 | |
a52d9a80 | 5356 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5357 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5358 | int create) |
5359 | { | |
5360 | int ret; | |
5361 | int err = 0; | |
db94535d | 5362 | u64 bytenr; |
a52d9a80 CM |
5363 | u64 extent_start = 0; |
5364 | u64 extent_end = 0; | |
33345d01 | 5365 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5366 | u32 found_type; |
f421950f | 5367 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5368 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5369 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5370 | struct extent_buffer *leaf; |
5371 | struct btrfs_key found_key; | |
a52d9a80 CM |
5372 | struct extent_map *em = NULL; |
5373 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5374 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5375 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5376 | int compress_type; |
a52d9a80 | 5377 | |
a52d9a80 | 5378 | again: |
890871be | 5379 | read_lock(&em_tree->lock); |
d1310b2e | 5380 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5381 | if (em) |
5382 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5383 | read_unlock(&em_tree->lock); |
d1310b2e | 5384 | |
a52d9a80 | 5385 | if (em) { |
e1c4b745 CM |
5386 | if (em->start > start || em->start + em->len <= start) |
5387 | free_extent_map(em); | |
5388 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5389 | free_extent_map(em); |
5390 | else | |
5391 | goto out; | |
a52d9a80 | 5392 | } |
172ddd60 | 5393 | em = alloc_extent_map(); |
a52d9a80 | 5394 | if (!em) { |
d1310b2e CM |
5395 | err = -ENOMEM; |
5396 | goto out; | |
a52d9a80 | 5397 | } |
e6dcd2dc | 5398 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5399 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5400 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5401 | em->len = (u64)-1; |
c8b97818 | 5402 | em->block_len = (u64)-1; |
f421950f CM |
5403 | |
5404 | if (!path) { | |
5405 | path = btrfs_alloc_path(); | |
026fd317 JB |
5406 | if (!path) { |
5407 | err = -ENOMEM; | |
5408 | goto out; | |
5409 | } | |
5410 | /* | |
5411 | * Chances are we'll be called again, so go ahead and do | |
5412 | * readahead | |
5413 | */ | |
5414 | path->reada = 1; | |
f421950f CM |
5415 | } |
5416 | ||
179e29e4 CM |
5417 | ret = btrfs_lookup_file_extent(trans, root, path, |
5418 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5419 | if (ret < 0) { |
5420 | err = ret; | |
5421 | goto out; | |
5422 | } | |
5423 | ||
5424 | if (ret != 0) { | |
5425 | if (path->slots[0] == 0) | |
5426 | goto not_found; | |
5427 | path->slots[0]--; | |
5428 | } | |
5429 | ||
5f39d397 CM |
5430 | leaf = path->nodes[0]; |
5431 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5432 | struct btrfs_file_extent_item); |
a52d9a80 | 5433 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5434 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5435 | found_type = btrfs_key_type(&found_key); | |
5436 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5437 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5438 | goto not_found; | |
5439 | } | |
5440 | ||
5f39d397 CM |
5441 | found_type = btrfs_file_extent_type(leaf, item); |
5442 | extent_start = found_key.offset; | |
261507a0 | 5443 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5444 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5445 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5446 | extent_end = extent_start + |
db94535d | 5447 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5448 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5449 | size_t size; | |
5450 | size = btrfs_file_extent_inline_len(leaf, item); | |
5451 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5452 | ~((u64)root->sectorsize - 1); | |
5453 | } | |
5454 | ||
5455 | if (start >= extent_end) { | |
5456 | path->slots[0]++; | |
5457 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5458 | ret = btrfs_next_leaf(root, path); | |
5459 | if (ret < 0) { | |
5460 | err = ret; | |
5461 | goto out; | |
a52d9a80 | 5462 | } |
9036c102 YZ |
5463 | if (ret > 0) |
5464 | goto not_found; | |
5465 | leaf = path->nodes[0]; | |
a52d9a80 | 5466 | } |
9036c102 YZ |
5467 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5468 | if (found_key.objectid != objectid || | |
5469 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5470 | goto not_found; | |
5471 | if (start + len <= found_key.offset) | |
5472 | goto not_found; | |
5473 | em->start = start; | |
70c8a91c | 5474 | em->orig_start = start; |
9036c102 YZ |
5475 | em->len = found_key.offset - start; |
5476 | goto not_found_em; | |
5477 | } | |
5478 | ||
d899e052 YZ |
5479 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5480 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5481 | em->start = extent_start; |
5482 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5483 | em->orig_start = extent_start - |
5484 | btrfs_file_extent_offset(leaf, item); | |
b4939680 JB |
5485 | em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, |
5486 | item); | |
db94535d CM |
5487 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5488 | if (bytenr == 0) { | |
5f39d397 | 5489 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5490 | goto insert; |
5491 | } | |
261507a0 | 5492 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5493 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5494 | em->compress_type = compress_type; |
c8b97818 | 5495 | em->block_start = bytenr; |
b4939680 | 5496 | em->block_len = em->orig_block_len; |
c8b97818 CM |
5497 | } else { |
5498 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5499 | em->block_start = bytenr; | |
5500 | em->block_len = em->len; | |
d899e052 YZ |
5501 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5502 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5503 | } |
a52d9a80 CM |
5504 | goto insert; |
5505 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5506 | unsigned long ptr; |
a52d9a80 | 5507 | char *map; |
3326d1b0 CM |
5508 | size_t size; |
5509 | size_t extent_offset; | |
5510 | size_t copy_size; | |
a52d9a80 | 5511 | |
689f9346 | 5512 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5513 | if (!page || create) { |
689f9346 | 5514 | em->start = extent_start; |
9036c102 | 5515 | em->len = extent_end - extent_start; |
689f9346 Y |
5516 | goto out; |
5517 | } | |
5f39d397 | 5518 | |
9036c102 YZ |
5519 | size = btrfs_file_extent_inline_len(leaf, item); |
5520 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5521 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5522 | size - extent_offset); |
3326d1b0 | 5523 | em->start = extent_start + extent_offset; |
70dec807 CM |
5524 | em->len = (copy_size + root->sectorsize - 1) & |
5525 | ~((u64)root->sectorsize - 1); | |
b4939680 | 5526 | em->orig_block_len = em->len; |
70c8a91c | 5527 | em->orig_start = em->start; |
261507a0 | 5528 | if (compress_type) { |
c8b97818 | 5529 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5530 | em->compress_type = compress_type; |
5531 | } | |
689f9346 | 5532 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5533 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5534 | if (btrfs_file_extent_compression(leaf, item) != |
5535 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5536 | ret = uncompress_inline(path, inode, page, |
5537 | pg_offset, | |
5538 | extent_offset, item); | |
79787eaa | 5539 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5540 | } else { |
5541 | map = kmap(page); | |
5542 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5543 | copy_size); | |
93c82d57 CM |
5544 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5545 | memset(map + pg_offset + copy_size, 0, | |
5546 | PAGE_CACHE_SIZE - pg_offset - | |
5547 | copy_size); | |
5548 | } | |
c8b97818 CM |
5549 | kunmap(page); |
5550 | } | |
179e29e4 CM |
5551 | flush_dcache_page(page); |
5552 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5553 | BUG(); |
179e29e4 CM |
5554 | if (!trans) { |
5555 | kunmap(page); | |
5556 | free_extent_map(em); | |
5557 | em = NULL; | |
ff5714cc | 5558 | |
b3b4aa74 | 5559 | btrfs_release_path(path); |
7a7eaa40 | 5560 | trans = btrfs_join_transaction(root); |
ff5714cc | 5561 | |
3612b495 TI |
5562 | if (IS_ERR(trans)) |
5563 | return ERR_CAST(trans); | |
179e29e4 CM |
5564 | goto again; |
5565 | } | |
c8b97818 | 5566 | map = kmap(page); |
70dec807 | 5567 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5568 | copy_size); |
c8b97818 | 5569 | kunmap(page); |
179e29e4 | 5570 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5571 | } |
d1310b2e | 5572 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5573 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5574 | goto insert; |
5575 | } else { | |
31b1a2bd | 5576 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5577 | } |
5578 | not_found: | |
5579 | em->start = start; | |
70c8a91c | 5580 | em->orig_start = start; |
d1310b2e | 5581 | em->len = len; |
a52d9a80 | 5582 | not_found_em: |
5f39d397 | 5583 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5584 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5585 | insert: |
b3b4aa74 | 5586 | btrfs_release_path(path); |
d1310b2e | 5587 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5588 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5589 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5590 | (unsigned long long)em->len, | |
5591 | (unsigned long long)start, | |
5592 | (unsigned long long)len); | |
a52d9a80 CM |
5593 | err = -EIO; |
5594 | goto out; | |
5595 | } | |
d1310b2e CM |
5596 | |
5597 | err = 0; | |
890871be | 5598 | write_lock(&em_tree->lock); |
a52d9a80 | 5599 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5600 | /* it is possible that someone inserted the extent into the tree |
5601 | * while we had the lock dropped. It is also possible that | |
5602 | * an overlapping map exists in the tree | |
5603 | */ | |
a52d9a80 | 5604 | if (ret == -EEXIST) { |
3b951516 | 5605 | struct extent_map *existing; |
e6dcd2dc CM |
5606 | |
5607 | ret = 0; | |
5608 | ||
3b951516 | 5609 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5610 | if (existing && (existing->start > start || |
5611 | existing->start + existing->len <= start)) { | |
5612 | free_extent_map(existing); | |
5613 | existing = NULL; | |
5614 | } | |
3b951516 CM |
5615 | if (!existing) { |
5616 | existing = lookup_extent_mapping(em_tree, em->start, | |
5617 | em->len); | |
5618 | if (existing) { | |
5619 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5620 | em, start, |
5621 | root->sectorsize); | |
3b951516 CM |
5622 | free_extent_map(existing); |
5623 | if (err) { | |
5624 | free_extent_map(em); | |
5625 | em = NULL; | |
5626 | } | |
5627 | } else { | |
5628 | err = -EIO; | |
3b951516 CM |
5629 | free_extent_map(em); |
5630 | em = NULL; | |
5631 | } | |
5632 | } else { | |
5633 | free_extent_map(em); | |
5634 | em = existing; | |
e6dcd2dc | 5635 | err = 0; |
a52d9a80 | 5636 | } |
a52d9a80 | 5637 | } |
890871be | 5638 | write_unlock(&em_tree->lock); |
a52d9a80 | 5639 | out: |
1abe9b8a | 5640 | |
f0bd95ea TI |
5641 | if (em) |
5642 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 5643 | |
f421950f CM |
5644 | if (path) |
5645 | btrfs_free_path(path); | |
a52d9a80 CM |
5646 | if (trans) { |
5647 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5648 | if (!err) |
a52d9a80 CM |
5649 | err = ret; |
5650 | } | |
a52d9a80 CM |
5651 | if (err) { |
5652 | free_extent_map(em); | |
a52d9a80 CM |
5653 | return ERR_PTR(err); |
5654 | } | |
79787eaa | 5655 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5656 | return em; |
5657 | } | |
5658 | ||
ec29ed5b CM |
5659 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5660 | size_t pg_offset, u64 start, u64 len, | |
5661 | int create) | |
5662 | { | |
5663 | struct extent_map *em; | |
5664 | struct extent_map *hole_em = NULL; | |
5665 | u64 range_start = start; | |
5666 | u64 end; | |
5667 | u64 found; | |
5668 | u64 found_end; | |
5669 | int err = 0; | |
5670 | ||
5671 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5672 | if (IS_ERR(em)) | |
5673 | return em; | |
5674 | if (em) { | |
5675 | /* | |
f9e4fb53 LB |
5676 | * if our em maps to |
5677 | * - a hole or | |
5678 | * - a pre-alloc extent, | |
5679 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 5680 | */ |
f9e4fb53 LB |
5681 | if (em->block_start != EXTENT_MAP_HOLE && |
5682 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
5683 | return em; |
5684 | else | |
5685 | hole_em = em; | |
5686 | } | |
5687 | ||
5688 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5689 | end = start + len; | |
5690 | if (end < start) | |
5691 | end = (u64)-1; | |
5692 | else | |
5693 | end -= 1; | |
5694 | ||
5695 | em = NULL; | |
5696 | ||
5697 | /* ok, we didn't find anything, lets look for delalloc */ | |
5698 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5699 | end, len, EXTENT_DELALLOC, 1); | |
5700 | found_end = range_start + found; | |
5701 | if (found_end < range_start) | |
5702 | found_end = (u64)-1; | |
5703 | ||
5704 | /* | |
5705 | * we didn't find anything useful, return | |
5706 | * the original results from get_extent() | |
5707 | */ | |
5708 | if (range_start > end || found_end <= start) { | |
5709 | em = hole_em; | |
5710 | hole_em = NULL; | |
5711 | goto out; | |
5712 | } | |
5713 | ||
5714 | /* adjust the range_start to make sure it doesn't | |
5715 | * go backwards from the start they passed in | |
5716 | */ | |
5717 | range_start = max(start,range_start); | |
5718 | found = found_end - range_start; | |
5719 | ||
5720 | if (found > 0) { | |
5721 | u64 hole_start = start; | |
5722 | u64 hole_len = len; | |
5723 | ||
172ddd60 | 5724 | em = alloc_extent_map(); |
ec29ed5b CM |
5725 | if (!em) { |
5726 | err = -ENOMEM; | |
5727 | goto out; | |
5728 | } | |
5729 | /* | |
5730 | * when btrfs_get_extent can't find anything it | |
5731 | * returns one huge hole | |
5732 | * | |
5733 | * make sure what it found really fits our range, and | |
5734 | * adjust to make sure it is based on the start from | |
5735 | * the caller | |
5736 | */ | |
5737 | if (hole_em) { | |
5738 | u64 calc_end = extent_map_end(hole_em); | |
5739 | ||
5740 | if (calc_end <= start || (hole_em->start > end)) { | |
5741 | free_extent_map(hole_em); | |
5742 | hole_em = NULL; | |
5743 | } else { | |
5744 | hole_start = max(hole_em->start, start); | |
5745 | hole_len = calc_end - hole_start; | |
5746 | } | |
5747 | } | |
5748 | em->bdev = NULL; | |
5749 | if (hole_em && range_start > hole_start) { | |
5750 | /* our hole starts before our delalloc, so we | |
5751 | * have to return just the parts of the hole | |
5752 | * that go until the delalloc starts | |
5753 | */ | |
5754 | em->len = min(hole_len, | |
5755 | range_start - hole_start); | |
5756 | em->start = hole_start; | |
5757 | em->orig_start = hole_start; | |
5758 | /* | |
5759 | * don't adjust block start at all, | |
5760 | * it is fixed at EXTENT_MAP_HOLE | |
5761 | */ | |
5762 | em->block_start = hole_em->block_start; | |
5763 | em->block_len = hole_len; | |
f9e4fb53 LB |
5764 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
5765 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
5766 | } else { |
5767 | em->start = range_start; | |
5768 | em->len = found; | |
5769 | em->orig_start = range_start; | |
5770 | em->block_start = EXTENT_MAP_DELALLOC; | |
5771 | em->block_len = found; | |
5772 | } | |
5773 | } else if (hole_em) { | |
5774 | return hole_em; | |
5775 | } | |
5776 | out: | |
5777 | ||
5778 | free_extent_map(hole_em); | |
5779 | if (err) { | |
5780 | free_extent_map(em); | |
5781 | return ERR_PTR(err); | |
5782 | } | |
5783 | return em; | |
5784 | } | |
5785 | ||
4b46fce2 JB |
5786 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
5787 | u64 start, u64 len) | |
5788 | { | |
5789 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5790 | struct btrfs_trans_handle *trans; | |
70c8a91c | 5791 | struct extent_map *em; |
4b46fce2 JB |
5792 | struct btrfs_key ins; |
5793 | u64 alloc_hint; | |
5794 | int ret; | |
4b46fce2 | 5795 | |
7a7eaa40 | 5796 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5797 | if (IS_ERR(trans)) |
5798 | return ERR_CAST(trans); | |
4b46fce2 JB |
5799 | |
5800 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5801 | ||
5802 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5803 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5804 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5805 | if (ret) { |
5806 | em = ERR_PTR(ret); | |
5807 | goto out; | |
5808 | } | |
5809 | ||
70c8a91c JB |
5810 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
5811 | ins.offset, ins.offset, 0); | |
5812 | if (IS_ERR(em)) | |
5813 | goto out; | |
4b46fce2 JB |
5814 | |
5815 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5816 | ins.offset, ins.offset, 0); | |
5817 | if (ret) { | |
5818 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5819 | em = ERR_PTR(ret); | |
5820 | } | |
5821 | out: | |
5822 | btrfs_end_transaction(trans, root); | |
5823 | return em; | |
5824 | } | |
5825 | ||
46bfbb5c CM |
5826 | /* |
5827 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5828 | * block must be cow'd | |
5829 | */ | |
5830 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5831 | struct inode *inode, u64 offset, u64 len) | |
5832 | { | |
5833 | struct btrfs_path *path; | |
5834 | int ret; | |
5835 | struct extent_buffer *leaf; | |
5836 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5837 | struct btrfs_file_extent_item *fi; | |
5838 | struct btrfs_key key; | |
5839 | u64 disk_bytenr; | |
5840 | u64 backref_offset; | |
5841 | u64 extent_end; | |
5842 | u64 num_bytes; | |
5843 | int slot; | |
5844 | int found_type; | |
5845 | ||
5846 | path = btrfs_alloc_path(); | |
5847 | if (!path) | |
5848 | return -ENOMEM; | |
5849 | ||
33345d01 | 5850 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5851 | offset, 0); |
5852 | if (ret < 0) | |
5853 | goto out; | |
5854 | ||
5855 | slot = path->slots[0]; | |
5856 | if (ret == 1) { | |
5857 | if (slot == 0) { | |
5858 | /* can't find the item, must cow */ | |
5859 | ret = 0; | |
5860 | goto out; | |
5861 | } | |
5862 | slot--; | |
5863 | } | |
5864 | ret = 0; | |
5865 | leaf = path->nodes[0]; | |
5866 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5867 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5868 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5869 | /* not our file or wrong item type, must cow */ | |
5870 | goto out; | |
5871 | } | |
5872 | ||
5873 | if (key.offset > offset) { | |
5874 | /* Wrong offset, must cow */ | |
5875 | goto out; | |
5876 | } | |
5877 | ||
5878 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5879 | found_type = btrfs_file_extent_type(leaf, fi); | |
5880 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5881 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5882 | /* not a regular extent, must cow */ | |
5883 | goto out; | |
5884 | } | |
5885 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5886 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5887 | ||
5888 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5889 | if (extent_end < offset + len) { | |
5890 | /* extent doesn't include our full range, must cow */ | |
5891 | goto out; | |
5892 | } | |
5893 | ||
5894 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5895 | goto out; | |
5896 | ||
5897 | /* | |
5898 | * look for other files referencing this extent, if we | |
5899 | * find any we must cow | |
5900 | */ | |
33345d01 | 5901 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5902 | key.offset - backref_offset, disk_bytenr)) |
5903 | goto out; | |
5904 | ||
5905 | /* | |
5906 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5907 | * in this extent we are about to write. If there | |
5908 | * are any csums in that range we have to cow in order | |
5909 | * to keep the csums correct | |
5910 | */ | |
5911 | disk_bytenr += backref_offset; | |
5912 | disk_bytenr += offset - key.offset; | |
5913 | num_bytes = min(offset + len, extent_end) - offset; | |
5914 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5915 | goto out; | |
5916 | /* | |
5917 | * all of the above have passed, it is safe to overwrite this extent | |
5918 | * without cow | |
5919 | */ | |
5920 | ret = 1; | |
5921 | out: | |
5922 | btrfs_free_path(path); | |
5923 | return ret; | |
5924 | } | |
5925 | ||
eb838e73 JB |
5926 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
5927 | struct extent_state **cached_state, int writing) | |
5928 | { | |
5929 | struct btrfs_ordered_extent *ordered; | |
5930 | int ret = 0; | |
5931 | ||
5932 | while (1) { | |
5933 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5934 | 0, cached_state); | |
5935 | /* | |
5936 | * We're concerned with the entire range that we're going to be | |
5937 | * doing DIO to, so we need to make sure theres no ordered | |
5938 | * extents in this range. | |
5939 | */ | |
5940 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
5941 | lockend - lockstart + 1); | |
5942 | ||
5943 | /* | |
5944 | * We need to make sure there are no buffered pages in this | |
5945 | * range either, we could have raced between the invalidate in | |
5946 | * generic_file_direct_write and locking the extent. The | |
5947 | * invalidate needs to happen so that reads after a write do not | |
5948 | * get stale data. | |
5949 | */ | |
5950 | if (!ordered && (!writing || | |
5951 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
5952 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
5953 | *cached_state))) | |
5954 | break; | |
5955 | ||
5956 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5957 | cached_state, GFP_NOFS); | |
5958 | ||
5959 | if (ordered) { | |
5960 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5961 | btrfs_put_ordered_extent(ordered); | |
5962 | } else { | |
5963 | /* Screw you mmap */ | |
5964 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
5965 | lockstart, | |
5966 | lockend); | |
5967 | if (ret) | |
5968 | break; | |
5969 | ||
5970 | /* | |
5971 | * If we found a page that couldn't be invalidated just | |
5972 | * fall back to buffered. | |
5973 | */ | |
5974 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5975 | lockstart >> PAGE_CACHE_SHIFT, | |
5976 | lockend >> PAGE_CACHE_SHIFT); | |
5977 | if (ret) | |
5978 | break; | |
5979 | } | |
5980 | ||
5981 | cond_resched(); | |
5982 | } | |
5983 | ||
5984 | return ret; | |
5985 | } | |
5986 | ||
69ffb543 JB |
5987 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
5988 | u64 len, u64 orig_start, | |
5989 | u64 block_start, u64 block_len, | |
b4939680 | 5990 | u64 orig_block_len, int type) |
69ffb543 JB |
5991 | { |
5992 | struct extent_map_tree *em_tree; | |
5993 | struct extent_map *em; | |
5994 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5995 | int ret; | |
5996 | ||
5997 | em_tree = &BTRFS_I(inode)->extent_tree; | |
5998 | em = alloc_extent_map(); | |
5999 | if (!em) | |
6000 | return ERR_PTR(-ENOMEM); | |
6001 | ||
6002 | em->start = start; | |
6003 | em->orig_start = orig_start; | |
2ab28f32 JB |
6004 | em->mod_start = start; |
6005 | em->mod_len = len; | |
69ffb543 JB |
6006 | em->len = len; |
6007 | em->block_len = block_len; | |
6008 | em->block_start = block_start; | |
6009 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 6010 | em->orig_block_len = orig_block_len; |
70c8a91c | 6011 | em->generation = -1; |
69ffb543 JB |
6012 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
6013 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 6014 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
6015 | |
6016 | do { | |
6017 | btrfs_drop_extent_cache(inode, em->start, | |
6018 | em->start + em->len - 1, 0); | |
6019 | write_lock(&em_tree->lock); | |
6020 | ret = add_extent_mapping(em_tree, em); | |
70c8a91c JB |
6021 | if (!ret) |
6022 | list_move(&em->list, | |
6023 | &em_tree->modified_extents); | |
69ffb543 JB |
6024 | write_unlock(&em_tree->lock); |
6025 | } while (ret == -EEXIST); | |
6026 | ||
6027 | if (ret) { | |
6028 | free_extent_map(em); | |
6029 | return ERR_PTR(ret); | |
6030 | } | |
6031 | ||
6032 | return em; | |
6033 | } | |
6034 | ||
6035 | ||
4b46fce2 JB |
6036 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
6037 | struct buffer_head *bh_result, int create) | |
6038 | { | |
6039 | struct extent_map *em; | |
6040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 6041 | struct extent_state *cached_state = NULL; |
4b46fce2 | 6042 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 6043 | u64 lockstart, lockend; |
4b46fce2 | 6044 | u64 len = bh_result->b_size; |
46bfbb5c | 6045 | struct btrfs_trans_handle *trans; |
eb838e73 JB |
6046 | int unlock_bits = EXTENT_LOCKED; |
6047 | int ret; | |
6048 | ||
eb838e73 JB |
6049 | if (create) { |
6050 | ret = btrfs_delalloc_reserve_space(inode, len); | |
6051 | if (ret) | |
6052 | return ret; | |
6053 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; | |
c329861d JB |
6054 | } else { |
6055 | len = min_t(u64, len, root->sectorsize); | |
eb838e73 JB |
6056 | } |
6057 | ||
c329861d JB |
6058 | lockstart = start; |
6059 | lockend = start + len - 1; | |
6060 | ||
eb838e73 JB |
6061 | /* |
6062 | * If this errors out it's because we couldn't invalidate pagecache for | |
6063 | * this range and we need to fallback to buffered. | |
6064 | */ | |
6065 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
6066 | return -ENOTBLK; | |
6067 | ||
6068 | if (create) { | |
6069 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6070 | lockend, EXTENT_DELALLOC, NULL, | |
6071 | &cached_state, GFP_NOFS); | |
6072 | if (ret) | |
6073 | goto unlock_err; | |
6074 | } | |
4b46fce2 JB |
6075 | |
6076 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
eb838e73 JB |
6077 | if (IS_ERR(em)) { |
6078 | ret = PTR_ERR(em); | |
6079 | goto unlock_err; | |
6080 | } | |
4b46fce2 JB |
6081 | |
6082 | /* | |
6083 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
6084 | * io. INLINE is special, and we could probably kludge it in here, but | |
6085 | * it's still buffered so for safety lets just fall back to the generic | |
6086 | * buffered path. | |
6087 | * | |
6088 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6089 | * decompress it, so there will be buffering required no matter what we | |
6090 | * do, so go ahead and fallback to buffered. | |
6091 | * | |
6092 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6093 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6094 | * the generic code. | |
6095 | */ | |
6096 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6097 | em->block_start == EXTENT_MAP_INLINE) { | |
6098 | free_extent_map(em); | |
eb838e73 JB |
6099 | ret = -ENOTBLK; |
6100 | goto unlock_err; | |
4b46fce2 JB |
6101 | } |
6102 | ||
6103 | /* Just a good old fashioned hole, return */ | |
6104 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6105 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6106 | free_extent_map(em); | |
eb838e73 JB |
6107 | ret = 0; |
6108 | goto unlock_err; | |
4b46fce2 JB |
6109 | } |
6110 | ||
6111 | /* | |
6112 | * We don't allocate a new extent in the following cases | |
6113 | * | |
6114 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6115 | * existing extent. | |
6116 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6117 | * just use the extent. | |
6118 | * | |
6119 | */ | |
46bfbb5c | 6120 | if (!create) { |
eb838e73 JB |
6121 | len = min(len, em->len - (start - em->start)); |
6122 | lockstart = start + len; | |
6123 | goto unlock; | |
46bfbb5c | 6124 | } |
4b46fce2 JB |
6125 | |
6126 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6127 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6128 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6129 | int type; |
6130 | int ret; | |
46bfbb5c | 6131 | u64 block_start; |
4b46fce2 JB |
6132 | |
6133 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6134 | type = BTRFS_ORDERED_PREALLOC; | |
6135 | else | |
6136 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6137 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6138 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6139 | |
6140 | /* | |
6141 | * we're not going to log anything, but we do need | |
6142 | * to make sure the current transaction stays open | |
6143 | * while we look for nocow cross refs | |
6144 | */ | |
7a7eaa40 | 6145 | trans = btrfs_join_transaction(root); |
3612b495 | 6146 | if (IS_ERR(trans)) |
46bfbb5c CM |
6147 | goto must_cow; |
6148 | ||
6149 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
70c8a91c | 6150 | u64 orig_start = em->orig_start; |
b4939680 | 6151 | u64 orig_block_len = em->orig_block_len; |
69ffb543 JB |
6152 | |
6153 | if (type == BTRFS_ORDERED_PREALLOC) { | |
6154 | free_extent_map(em); | |
6155 | em = create_pinned_em(inode, start, len, | |
6156 | orig_start, | |
b4939680 JB |
6157 | block_start, len, |
6158 | orig_block_len, type); | |
69ffb543 JB |
6159 | if (IS_ERR(em)) { |
6160 | btrfs_end_transaction(trans, root); | |
6161 | goto unlock_err; | |
6162 | } | |
6163 | } | |
6164 | ||
46bfbb5c CM |
6165 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6166 | block_start, len, len, type); | |
6167 | btrfs_end_transaction(trans, root); | |
6168 | if (ret) { | |
6169 | free_extent_map(em); | |
eb838e73 | 6170 | goto unlock_err; |
46bfbb5c CM |
6171 | } |
6172 | goto unlock; | |
4b46fce2 | 6173 | } |
46bfbb5c | 6174 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6175 | } |
46bfbb5c CM |
6176 | must_cow: |
6177 | /* | |
6178 | * this will cow the extent, reset the len in case we changed | |
6179 | * it above | |
6180 | */ | |
6181 | len = bh_result->b_size; | |
70c8a91c JB |
6182 | free_extent_map(em); |
6183 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
6184 | if (IS_ERR(em)) { |
6185 | ret = PTR_ERR(em); | |
6186 | goto unlock_err; | |
6187 | } | |
46bfbb5c CM |
6188 | len = min(len, em->len - (start - em->start)); |
6189 | unlock: | |
4b46fce2 JB |
6190 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6191 | inode->i_blkbits; | |
46bfbb5c | 6192 | bh_result->b_size = len; |
4b46fce2 JB |
6193 | bh_result->b_bdev = em->bdev; |
6194 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6195 | if (create) { |
6196 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6197 | set_buffer_new(bh_result); | |
6198 | ||
6199 | /* | |
6200 | * Need to update the i_size under the extent lock so buffered | |
6201 | * readers will get the updated i_size when we unlock. | |
6202 | */ | |
6203 | if (start + len > i_size_read(inode)) | |
6204 | i_size_write(inode, start + len); | |
6205 | } | |
4b46fce2 | 6206 | |
eb838e73 JB |
6207 | /* |
6208 | * In the case of write we need to clear and unlock the entire range, | |
6209 | * in the case of read we need to unlock only the end area that we | |
6210 | * aren't using if there is any left over space. | |
6211 | */ | |
24c03fa5 LB |
6212 | if (lockstart < lockend) { |
6213 | if (create && len < lockend - lockstart) { | |
6214 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
9e8a4a8b LB |
6215 | lockstart + len - 1, |
6216 | unlock_bits | EXTENT_DEFRAG, 1, 0, | |
24c03fa5 LB |
6217 | &cached_state, GFP_NOFS); |
6218 | /* | |
6219 | * Beside unlock, we also need to cleanup reserved space | |
6220 | * for the left range by attaching EXTENT_DO_ACCOUNTING. | |
6221 | */ | |
6222 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
6223 | lockstart + len, lockend, | |
9e8a4a8b LB |
6224 | unlock_bits | EXTENT_DO_ACCOUNTING | |
6225 | EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS); | |
24c03fa5 LB |
6226 | } else { |
6227 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6228 | lockend, unlock_bits, 1, 0, | |
6229 | &cached_state, GFP_NOFS); | |
6230 | } | |
6231 | } else { | |
eb838e73 | 6232 | free_extent_state(cached_state); |
24c03fa5 | 6233 | } |
eb838e73 | 6234 | |
4b46fce2 JB |
6235 | free_extent_map(em); |
6236 | ||
6237 | return 0; | |
eb838e73 JB |
6238 | |
6239 | unlock_err: | |
6240 | if (create) | |
6241 | unlock_bits |= EXTENT_DO_ACCOUNTING; | |
6242 | ||
6243 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6244 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6245 | return ret; | |
4b46fce2 JB |
6246 | } |
6247 | ||
6248 | struct btrfs_dio_private { | |
6249 | struct inode *inode; | |
6250 | u64 logical_offset; | |
6251 | u64 disk_bytenr; | |
6252 | u64 bytes; | |
4b46fce2 | 6253 | void *private; |
e65e1535 MX |
6254 | |
6255 | /* number of bios pending for this dio */ | |
6256 | atomic_t pending_bios; | |
6257 | ||
6258 | /* IO errors */ | |
6259 | int errors; | |
6260 | ||
6261 | struct bio *orig_bio; | |
4b46fce2 JB |
6262 | }; |
6263 | ||
6264 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6265 | { | |
e65e1535 | 6266 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6267 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6268 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6269 | struct inode *inode = dip->inode; |
6270 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6271 | u64 start; | |
4b46fce2 JB |
6272 | |
6273 | start = dip->logical_offset; | |
6274 | do { | |
6275 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6276 | struct page *page = bvec->bv_page; | |
6277 | char *kaddr; | |
6278 | u32 csum = ~(u32)0; | |
c329861d | 6279 | u64 private = ~(u32)0; |
4b46fce2 JB |
6280 | unsigned long flags; |
6281 | ||
c329861d JB |
6282 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6283 | start, &private)) | |
6284 | goto failed; | |
4b46fce2 | 6285 | local_irq_save(flags); |
7ac687d9 | 6286 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
6287 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
6288 | csum, bvec->bv_len); | |
6289 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6290 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6291 | local_irq_restore(flags); |
6292 | ||
6293 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6294 | if (csum != private) { |
6295 | failed: | |
33345d01 | 6296 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 6297 | " %llu csum %u private %u\n", |
33345d01 LZ |
6298 | (unsigned long long)btrfs_ino(inode), |
6299 | (unsigned long long)start, | |
c329861d | 6300 | csum, (unsigned)private); |
4b46fce2 JB |
6301 | err = -EIO; |
6302 | } | |
6303 | } | |
6304 | ||
6305 | start += bvec->bv_len; | |
4b46fce2 JB |
6306 | bvec++; |
6307 | } while (bvec <= bvec_end); | |
6308 | ||
6309 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6310 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
6311 | bio->bi_private = dip->private; |
6312 | ||
4b46fce2 | 6313 | kfree(dip); |
c0da7aa1 JB |
6314 | |
6315 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6316 | if (err) | |
6317 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6318 | dio_end_io(bio, err); |
6319 | } | |
6320 | ||
6321 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6322 | { | |
6323 | struct btrfs_dio_private *dip = bio->bi_private; | |
6324 | struct inode *inode = dip->inode; | |
6325 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6326 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6327 | u64 ordered_offset = dip->logical_offset; |
6328 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
6329 | int ret; |
6330 | ||
6331 | if (err) | |
6332 | goto out_done; | |
163cf09c CM |
6333 | again: |
6334 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6335 | &ordered_offset, | |
5fd02043 | 6336 | ordered_bytes, !err); |
4b46fce2 | 6337 | if (!ret) |
163cf09c | 6338 | goto out_test; |
4b46fce2 | 6339 | |
5fd02043 JB |
6340 | ordered->work.func = finish_ordered_fn; |
6341 | ordered->work.flags = 0; | |
6342 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6343 | &ordered->work); | |
163cf09c CM |
6344 | out_test: |
6345 | /* | |
6346 | * our bio might span multiple ordered extents. If we haven't | |
6347 | * completed the accounting for the whole dio, go back and try again | |
6348 | */ | |
6349 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6350 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6351 | ordered_offset; | |
5fd02043 | 6352 | ordered = NULL; |
163cf09c CM |
6353 | goto again; |
6354 | } | |
4b46fce2 JB |
6355 | out_done: |
6356 | bio->bi_private = dip->private; | |
6357 | ||
4b46fce2 | 6358 | kfree(dip); |
c0da7aa1 JB |
6359 | |
6360 | /* If we had an error make sure to clear the uptodate flag */ | |
6361 | if (err) | |
6362 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6363 | dio_end_io(bio, err); |
6364 | } | |
6365 | ||
eaf25d93 CM |
6366 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6367 | struct bio *bio, int mirror_num, | |
6368 | unsigned long bio_flags, u64 offset) | |
6369 | { | |
6370 | int ret; | |
6371 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6372 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6373 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6374 | return 0; |
6375 | } | |
6376 | ||
e65e1535 MX |
6377 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6378 | { | |
6379 | struct btrfs_dio_private *dip = bio->bi_private; | |
6380 | ||
6381 | if (err) { | |
33345d01 | 6382 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6383 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6384 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6385 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6386 | dip->errors = 1; |
6387 | ||
6388 | /* | |
6389 | * before atomic variable goto zero, we must make sure | |
6390 | * dip->errors is perceived to be set. | |
6391 | */ | |
6392 | smp_mb__before_atomic_dec(); | |
6393 | } | |
6394 | ||
6395 | /* if there are more bios still pending for this dio, just exit */ | |
6396 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6397 | goto out; | |
6398 | ||
6399 | if (dip->errors) | |
6400 | bio_io_error(dip->orig_bio); | |
6401 | else { | |
6402 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6403 | bio_endio(dip->orig_bio, 0); | |
6404 | } | |
6405 | out: | |
6406 | bio_put(bio); | |
6407 | } | |
6408 | ||
6409 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6410 | u64 first_sector, gfp_t gfp_flags) | |
6411 | { | |
6412 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6413 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6414 | } | |
6415 | ||
6416 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6417 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6418 | int async_submit) |
e65e1535 MX |
6419 | { |
6420 | int write = rw & REQ_WRITE; | |
6421 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6422 | int ret; | |
6423 | ||
b812ce28 JB |
6424 | if (async_submit) |
6425 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
6426 | ||
e65e1535 | 6427 | bio_get(bio); |
5fd02043 JB |
6428 | |
6429 | if (!write) { | |
6430 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6431 | if (ret) | |
6432 | goto err; | |
6433 | } | |
e65e1535 | 6434 | |
1ae39938 JB |
6435 | if (skip_sum) |
6436 | goto map; | |
6437 | ||
6438 | if (write && async_submit) { | |
e65e1535 MX |
6439 | ret = btrfs_wq_submit_bio(root->fs_info, |
6440 | inode, rw, bio, 0, 0, | |
6441 | file_offset, | |
6442 | __btrfs_submit_bio_start_direct_io, | |
6443 | __btrfs_submit_bio_done); | |
6444 | goto err; | |
1ae39938 JB |
6445 | } else if (write) { |
6446 | /* | |
6447 | * If we aren't doing async submit, calculate the csum of the | |
6448 | * bio now. | |
6449 | */ | |
6450 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6451 | if (ret) | |
6452 | goto err; | |
c2db1073 | 6453 | } else if (!skip_sum) { |
c329861d | 6454 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6455 | if (ret) |
6456 | goto err; | |
6457 | } | |
e65e1535 | 6458 | |
1ae39938 JB |
6459 | map: |
6460 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6461 | err: |
6462 | bio_put(bio); | |
6463 | return ret; | |
6464 | } | |
6465 | ||
6466 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6467 | int skip_sum) | |
6468 | { | |
6469 | struct inode *inode = dip->inode; | |
6470 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
6471 | struct bio *bio; |
6472 | struct bio *orig_bio = dip->orig_bio; | |
6473 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6474 | u64 start_sector = orig_bio->bi_sector; | |
6475 | u64 file_offset = dip->logical_offset; | |
6476 | u64 submit_len = 0; | |
6477 | u64 map_length; | |
6478 | int nr_pages = 0; | |
e65e1535 | 6479 | int ret = 0; |
1ae39938 | 6480 | int async_submit = 0; |
e65e1535 | 6481 | |
e65e1535 | 6482 | map_length = orig_bio->bi_size; |
3ec706c8 | 6483 | ret = btrfs_map_block(root->fs_info, READ, start_sector << 9, |
e65e1535 MX |
6484 | &map_length, NULL, 0); |
6485 | if (ret) { | |
64728bbb | 6486 | bio_put(orig_bio); |
e65e1535 MX |
6487 | return -EIO; |
6488 | } | |
6489 | ||
02f57c7a JB |
6490 | if (map_length >= orig_bio->bi_size) { |
6491 | bio = orig_bio; | |
6492 | goto submit; | |
6493 | } | |
6494 | ||
1ae39938 | 6495 | async_submit = 1; |
02f57c7a JB |
6496 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6497 | if (!bio) | |
6498 | return -ENOMEM; | |
6499 | bio->bi_private = dip; | |
6500 | bio->bi_end_io = btrfs_end_dio_bio; | |
6501 | atomic_inc(&dip->pending_bios); | |
6502 | ||
e65e1535 MX |
6503 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6504 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6505 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6506 | bvec->bv_offset) < bvec->bv_len)) { | |
6507 | /* | |
6508 | * inc the count before we submit the bio so | |
6509 | * we know the end IO handler won't happen before | |
6510 | * we inc the count. Otherwise, the dip might get freed | |
6511 | * before we're done setting it up | |
6512 | */ | |
6513 | atomic_inc(&dip->pending_bios); | |
6514 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6515 | file_offset, skip_sum, | |
c329861d | 6516 | async_submit); |
e65e1535 MX |
6517 | if (ret) { |
6518 | bio_put(bio); | |
6519 | atomic_dec(&dip->pending_bios); | |
6520 | goto out_err; | |
6521 | } | |
6522 | ||
e65e1535 MX |
6523 | start_sector += submit_len >> 9; |
6524 | file_offset += submit_len; | |
6525 | ||
6526 | submit_len = 0; | |
6527 | nr_pages = 0; | |
6528 | ||
6529 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6530 | start_sector, GFP_NOFS); | |
6531 | if (!bio) | |
6532 | goto out_err; | |
6533 | bio->bi_private = dip; | |
6534 | bio->bi_end_io = btrfs_end_dio_bio; | |
6535 | ||
6536 | map_length = orig_bio->bi_size; | |
3ec706c8 SB |
6537 | ret = btrfs_map_block(root->fs_info, READ, |
6538 | start_sector << 9, | |
e65e1535 MX |
6539 | &map_length, NULL, 0); |
6540 | if (ret) { | |
6541 | bio_put(bio); | |
6542 | goto out_err; | |
6543 | } | |
6544 | } else { | |
6545 | submit_len += bvec->bv_len; | |
6546 | nr_pages ++; | |
6547 | bvec++; | |
6548 | } | |
6549 | } | |
6550 | ||
02f57c7a | 6551 | submit: |
e65e1535 | 6552 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 6553 | async_submit); |
e65e1535 MX |
6554 | if (!ret) |
6555 | return 0; | |
6556 | ||
6557 | bio_put(bio); | |
6558 | out_err: | |
6559 | dip->errors = 1; | |
6560 | /* | |
6561 | * before atomic variable goto zero, we must | |
6562 | * make sure dip->errors is perceived to be set. | |
6563 | */ | |
6564 | smp_mb__before_atomic_dec(); | |
6565 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6566 | bio_io_error(dip->orig_bio); | |
6567 | ||
6568 | /* bio_end_io() will handle error, so we needn't return it */ | |
6569 | return 0; | |
6570 | } | |
6571 | ||
4b46fce2 JB |
6572 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6573 | loff_t file_offset) | |
6574 | { | |
6575 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6576 | struct btrfs_dio_private *dip; | |
6577 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6578 | int skip_sum; |
7b6d91da | 6579 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6580 | int ret = 0; |
6581 | ||
6582 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6583 | ||
6584 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6585 | if (!dip) { | |
6586 | ret = -ENOMEM; | |
6587 | goto free_ordered; | |
6588 | } | |
4b46fce2 JB |
6589 | |
6590 | dip->private = bio->bi_private; | |
6591 | dip->inode = inode; | |
6592 | dip->logical_offset = file_offset; | |
6593 | ||
4b46fce2 JB |
6594 | dip->bytes = 0; |
6595 | do { | |
6596 | dip->bytes += bvec->bv_len; | |
6597 | bvec++; | |
6598 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6599 | ||
46bfbb5c | 6600 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6601 | bio->bi_private = dip; |
e65e1535 MX |
6602 | dip->errors = 0; |
6603 | dip->orig_bio = bio; | |
6604 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6605 | |
6606 | if (write) | |
6607 | bio->bi_end_io = btrfs_endio_direct_write; | |
6608 | else | |
6609 | bio->bi_end_io = btrfs_endio_direct_read; | |
6610 | ||
e65e1535 MX |
6611 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6612 | if (!ret) | |
eaf25d93 | 6613 | return; |
4b46fce2 JB |
6614 | free_ordered: |
6615 | /* | |
6616 | * If this is a write, we need to clean up the reserved space and kill | |
6617 | * the ordered extent. | |
6618 | */ | |
6619 | if (write) { | |
6620 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6621 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6622 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6623 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6624 | btrfs_free_reserved_extent(root, ordered->start, | |
6625 | ordered->disk_len); | |
6626 | btrfs_put_ordered_extent(ordered); | |
6627 | btrfs_put_ordered_extent(ordered); | |
6628 | } | |
6629 | bio_endio(bio, ret); | |
6630 | } | |
6631 | ||
5a5f79b5 CM |
6632 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6633 | const struct iovec *iov, loff_t offset, | |
6634 | unsigned long nr_segs) | |
6635 | { | |
6636 | int seg; | |
a1b75f7d | 6637 | int i; |
5a5f79b5 CM |
6638 | size_t size; |
6639 | unsigned long addr; | |
6640 | unsigned blocksize_mask = root->sectorsize - 1; | |
6641 | ssize_t retval = -EINVAL; | |
6642 | loff_t end = offset; | |
6643 | ||
6644 | if (offset & blocksize_mask) | |
6645 | goto out; | |
6646 | ||
6647 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6648 | for (seg = 0; seg < nr_segs; seg++) { | |
6649 | addr = (unsigned long)iov[seg].iov_base; | |
6650 | size = iov[seg].iov_len; | |
6651 | end += size; | |
a1b75f7d | 6652 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6653 | goto out; |
a1b75f7d JB |
6654 | |
6655 | /* If this is a write we don't need to check anymore */ | |
6656 | if (rw & WRITE) | |
6657 | continue; | |
6658 | ||
6659 | /* | |
6660 | * Check to make sure we don't have duplicate iov_base's in this | |
6661 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6662 | * when reading back. | |
6663 | */ | |
6664 | for (i = seg + 1; i < nr_segs; i++) { | |
6665 | if (iov[seg].iov_base == iov[i].iov_base) | |
6666 | goto out; | |
6667 | } | |
5a5f79b5 CM |
6668 | } |
6669 | retval = 0; | |
6670 | out: | |
6671 | return retval; | |
6672 | } | |
eb838e73 | 6673 | |
16432985 CM |
6674 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6675 | const struct iovec *iov, loff_t offset, | |
6676 | unsigned long nr_segs) | |
6677 | { | |
4b46fce2 JB |
6678 | struct file *file = iocb->ki_filp; |
6679 | struct inode *inode = file->f_mapping->host; | |
4b46fce2 | 6680 | |
5a5f79b5 | 6681 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 6682 | offset, nr_segs)) |
5a5f79b5 | 6683 | return 0; |
3f7c579c | 6684 | |
eb838e73 | 6685 | return __blockdev_direct_IO(rw, iocb, inode, |
5a5f79b5 CM |
6686 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, |
6687 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6688 | btrfs_submit_direct, 0); | |
16432985 CM |
6689 | } |
6690 | ||
05dadc09 TI |
6691 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
6692 | ||
1506fcc8 YS |
6693 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6694 | __u64 start, __u64 len) | |
6695 | { | |
05dadc09 TI |
6696 | int ret; |
6697 | ||
6698 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
6699 | if (ret) | |
6700 | return ret; | |
6701 | ||
ec29ed5b | 6702 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6703 | } |
6704 | ||
a52d9a80 | 6705 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6706 | { |
d1310b2e CM |
6707 | struct extent_io_tree *tree; |
6708 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6709 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6710 | } |
1832a6d5 | 6711 | |
a52d9a80 | 6712 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6713 | { |
d1310b2e | 6714 | struct extent_io_tree *tree; |
b888db2b CM |
6715 | |
6716 | ||
6717 | if (current->flags & PF_MEMALLOC) { | |
6718 | redirty_page_for_writepage(wbc, page); | |
6719 | unlock_page(page); | |
6720 | return 0; | |
6721 | } | |
d1310b2e | 6722 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6723 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6724 | } |
6725 | ||
f421950f CM |
6726 | int btrfs_writepages(struct address_space *mapping, |
6727 | struct writeback_control *wbc) | |
b293f02e | 6728 | { |
d1310b2e | 6729 | struct extent_io_tree *tree; |
771ed689 | 6730 | |
d1310b2e | 6731 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6732 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6733 | } | |
6734 | ||
3ab2fb5a CM |
6735 | static int |
6736 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6737 | struct list_head *pages, unsigned nr_pages) | |
6738 | { | |
d1310b2e CM |
6739 | struct extent_io_tree *tree; |
6740 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6741 | return extent_readpages(tree, mapping, pages, nr_pages, |
6742 | btrfs_get_extent); | |
6743 | } | |
e6dcd2dc | 6744 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6745 | { |
d1310b2e CM |
6746 | struct extent_io_tree *tree; |
6747 | struct extent_map_tree *map; | |
a52d9a80 | 6748 | int ret; |
8c2383c3 | 6749 | |
d1310b2e CM |
6750 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6751 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6752 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6753 | if (ret == 1) { |
6754 | ClearPagePrivate(page); | |
6755 | set_page_private(page, 0); | |
6756 | page_cache_release(page); | |
39279cc3 | 6757 | } |
a52d9a80 | 6758 | return ret; |
39279cc3 CM |
6759 | } |
6760 | ||
e6dcd2dc CM |
6761 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6762 | { | |
98509cfc CM |
6763 | if (PageWriteback(page) || PageDirty(page)) |
6764 | return 0; | |
b335b003 | 6765 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6766 | } |
6767 | ||
a52d9a80 | 6768 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6769 | { |
5fd02043 | 6770 | struct inode *inode = page->mapping->host; |
d1310b2e | 6771 | struct extent_io_tree *tree; |
e6dcd2dc | 6772 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6773 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6774 | u64 page_start = page_offset(page); |
6775 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6776 | |
8b62b72b CM |
6777 | /* |
6778 | * we have the page locked, so new writeback can't start, | |
6779 | * and the dirty bit won't be cleared while we are here. | |
6780 | * | |
6781 | * Wait for IO on this page so that we can safely clear | |
6782 | * the PagePrivate2 bit and do ordered accounting | |
6783 | */ | |
e6dcd2dc | 6784 | wait_on_page_writeback(page); |
8b62b72b | 6785 | |
5fd02043 | 6786 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6787 | if (offset) { |
6788 | btrfs_releasepage(page, GFP_NOFS); | |
6789 | return; | |
6790 | } | |
d0082371 | 6791 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
4eee4fa4 | 6792 | ordered = btrfs_lookup_ordered_extent(inode, page_offset(page)); |
e6dcd2dc | 6793 | if (ordered) { |
eb84ae03 CM |
6794 | /* |
6795 | * IO on this page will never be started, so we need | |
6796 | * to account for any ordered extents now | |
6797 | */ | |
e6dcd2dc CM |
6798 | clear_extent_bit(tree, page_start, page_end, |
6799 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
6800 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
6801 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
6802 | /* |
6803 | * whoever cleared the private bit is responsible | |
6804 | * for the finish_ordered_io | |
6805 | */ | |
5fd02043 JB |
6806 | if (TestClearPagePrivate2(page) && |
6807 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6808 | PAGE_CACHE_SIZE, 1)) { | |
6809 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6810 | } |
e6dcd2dc | 6811 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6812 | cached_state = NULL; |
d0082371 | 6813 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6814 | } |
6815 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6816 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
6817 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
6818 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
6819 | __btrfs_releasepage(page, GFP_NOFS); |
6820 | ||
4a096752 | 6821 | ClearPageChecked(page); |
9ad6b7bc | 6822 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6823 | ClearPagePrivate(page); |
6824 | set_page_private(page, 0); | |
6825 | page_cache_release(page); | |
6826 | } | |
39279cc3 CM |
6827 | } |
6828 | ||
9ebefb18 CM |
6829 | /* |
6830 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6831 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6832 | * be careful to check for EOF conditions here. We set the page up correctly | |
6833 | * for a written page which means we get ENOSPC checking when writing into | |
6834 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6835 | * support these features. | |
6836 | * | |
6837 | * We are not allowed to take the i_mutex here so we have to play games to | |
6838 | * protect against truncate races as the page could now be beyond EOF. Because | |
6839 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6840 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6841 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6842 | * unlock the page. | |
6843 | */ | |
c2ec175c | 6844 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6845 | { |
c2ec175c | 6846 | struct page *page = vmf->page; |
6da6abae | 6847 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6848 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6849 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6850 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6851 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6852 | char *kaddr; |
6853 | unsigned long zero_start; | |
9ebefb18 | 6854 | loff_t size; |
1832a6d5 | 6855 | int ret; |
9998eb70 | 6856 | int reserved = 0; |
a52d9a80 | 6857 | u64 page_start; |
e6dcd2dc | 6858 | u64 page_end; |
9ebefb18 | 6859 | |
b2b5ef5c | 6860 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 6861 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6862 | if (!ret) { |
e41f941a | 6863 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
6864 | reserved = 1; |
6865 | } | |
56a76f82 NP |
6866 | if (ret) { |
6867 | if (ret == -ENOMEM) | |
6868 | ret = VM_FAULT_OOM; | |
6869 | else /* -ENOSPC, -EIO, etc */ | |
6870 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6871 | if (reserved) |
6872 | goto out; | |
6873 | goto out_noreserve; | |
56a76f82 | 6874 | } |
1832a6d5 | 6875 | |
56a76f82 | 6876 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6877 | again: |
9ebefb18 | 6878 | lock_page(page); |
9ebefb18 | 6879 | size = i_size_read(inode); |
e6dcd2dc CM |
6880 | page_start = page_offset(page); |
6881 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6882 | |
9ebefb18 | 6883 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6884 | (page_start >= size)) { |
9ebefb18 CM |
6885 | /* page got truncated out from underneath us */ |
6886 | goto out_unlock; | |
6887 | } | |
e6dcd2dc CM |
6888 | wait_on_page_writeback(page); |
6889 | ||
d0082371 | 6890 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6891 | set_page_extent_mapped(page); |
6892 | ||
eb84ae03 CM |
6893 | /* |
6894 | * we can't set the delalloc bits if there are pending ordered | |
6895 | * extents. Drop our locks and wait for them to finish | |
6896 | */ | |
e6dcd2dc CM |
6897 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6898 | if (ordered) { | |
2ac55d41 JB |
6899 | unlock_extent_cached(io_tree, page_start, page_end, |
6900 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6901 | unlock_page(page); |
eb84ae03 | 6902 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6903 | btrfs_put_ordered_extent(ordered); |
6904 | goto again; | |
6905 | } | |
6906 | ||
fbf19087 JB |
6907 | /* |
6908 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6909 | * if it was already dirty, so for space accounting reasons we need to | |
6910 | * clear any delalloc bits for the range we are fixing to save. There | |
6911 | * is probably a better way to do this, but for now keep consistent with | |
6912 | * prepare_pages in the normal write path. | |
6913 | */ | |
2ac55d41 | 6914 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
6915 | EXTENT_DIRTY | EXTENT_DELALLOC | |
6916 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 6917 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6918 | |
2ac55d41 JB |
6919 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6920 | &cached_state); | |
9ed74f2d | 6921 | if (ret) { |
2ac55d41 JB |
6922 | unlock_extent_cached(io_tree, page_start, page_end, |
6923 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6924 | ret = VM_FAULT_SIGBUS; |
6925 | goto out_unlock; | |
6926 | } | |
e6dcd2dc | 6927 | ret = 0; |
9ebefb18 CM |
6928 | |
6929 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6930 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6931 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6932 | else |
e6dcd2dc | 6933 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6934 | |
e6dcd2dc CM |
6935 | if (zero_start != PAGE_CACHE_SIZE) { |
6936 | kaddr = kmap(page); | |
6937 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6938 | flush_dcache_page(page); | |
6939 | kunmap(page); | |
6940 | } | |
247e743c | 6941 | ClearPageChecked(page); |
e6dcd2dc | 6942 | set_page_dirty(page); |
50a9b214 | 6943 | SetPageUptodate(page); |
5a3f23d5 | 6944 | |
257c62e1 CM |
6945 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6946 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 6947 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 6948 | |
2ac55d41 | 6949 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6950 | |
6951 | out_unlock: | |
b2b5ef5c JK |
6952 | if (!ret) { |
6953 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 6954 | return VM_FAULT_LOCKED; |
b2b5ef5c | 6955 | } |
9ebefb18 | 6956 | unlock_page(page); |
1832a6d5 | 6957 | out: |
ec39e180 | 6958 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6959 | out_noreserve: |
b2b5ef5c | 6960 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
6961 | return ret; |
6962 | } | |
6963 | ||
a41ad394 | 6964 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6965 | { |
6966 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6967 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6968 | int ret; |
3893e33b | 6969 | int err = 0; |
39279cc3 | 6970 | struct btrfs_trans_handle *trans; |
dbe674a9 | 6971 | u64 mask = root->sectorsize - 1; |
07127184 | 6972 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6973 | |
2aaa6655 | 6974 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
5d5e103a | 6975 | if (ret) |
a41ad394 | 6976 | return ret; |
8082510e | 6977 | |
4a096752 | 6978 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6979 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6980 | |
fcb80c2a JB |
6981 | /* |
6982 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6983 | * 3 things going on here | |
6984 | * | |
6985 | * 1) We need to reserve space for our orphan item and the space to | |
6986 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6987 | * orphan item because we didn't reserve space to remove it. | |
6988 | * | |
6989 | * 2) We need to reserve space to update our inode. | |
6990 | * | |
6991 | * 3) We need to have something to cache all the space that is going to | |
6992 | * be free'd up by the truncate operation, but also have some slack | |
6993 | * space reserved in case it uses space during the truncate (thank you | |
6994 | * very much snapshotting). | |
6995 | * | |
6996 | * And we need these to all be seperate. The fact is we can use alot of | |
6997 | * space doing the truncate, and we have no earthly idea how much space | |
6998 | * we will use, so we need the truncate reservation to be seperate so it | |
6999 | * doesn't end up using space reserved for updating the inode or | |
7000 | * removing the orphan item. We also need to be able to stop the | |
7001 | * transaction and start a new one, which means we need to be able to | |
7002 | * update the inode several times, and we have no idea of knowing how | |
7003 | * many times that will be, so we can't just reserve 1 item for the | |
7004 | * entirety of the opration, so that has to be done seperately as well. | |
7005 | * Then there is the orphan item, which does indeed need to be held on | |
7006 | * to for the whole operation, and we need nobody to touch this reserved | |
7007 | * space except the orphan code. | |
7008 | * | |
7009 | * So that leaves us with | |
7010 | * | |
7011 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
7012 | * 2) rsv - for the truncate reservation, which we will steal from the | |
7013 | * transaction reservation. | |
7014 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
7015 | * updating the inode. | |
7016 | */ | |
66d8f3dd | 7017 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
7018 | if (!rsv) |
7019 | return -ENOMEM; | |
4a338542 | 7020 | rsv->size = min_size; |
ca7e70f5 | 7021 | rsv->failfast = 1; |
f0cd846e | 7022 | |
907cbceb | 7023 | /* |
07127184 | 7024 | * 1 for the truncate slack space |
907cbceb JB |
7025 | * 1 for updating the inode. |
7026 | */ | |
f3fe820c | 7027 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
7028 | if (IS_ERR(trans)) { |
7029 | err = PTR_ERR(trans); | |
7030 | goto out; | |
7031 | } | |
f0cd846e | 7032 | |
907cbceb JB |
7033 | /* Migrate the slack space for the truncate to our reserve */ |
7034 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
7035 | min_size); | |
fcb80c2a | 7036 | BUG_ON(ret); |
f0cd846e | 7037 | |
5a3f23d5 CM |
7038 | /* |
7039 | * setattr is responsible for setting the ordered_data_close flag, | |
7040 | * but that is only tested during the last file release. That | |
7041 | * could happen well after the next commit, leaving a great big | |
7042 | * window where new writes may get lost if someone chooses to write | |
7043 | * to this file after truncating to zero | |
7044 | * | |
7045 | * The inode doesn't have any dirty data here, and so if we commit | |
7046 | * this is a noop. If someone immediately starts writing to the inode | |
7047 | * it is very likely we'll catch some of their writes in this | |
7048 | * transaction, and the commit will find this file on the ordered | |
7049 | * data list with good things to send down. | |
7050 | * | |
7051 | * This is a best effort solution, there is still a window where | |
7052 | * using truncate to replace the contents of the file will | |
7053 | * end up with a zero length file after a crash. | |
7054 | */ | |
72ac3c0d JB |
7055 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
7056 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
7057 | btrfs_add_ordered_operation(trans, root, inode); |
7058 | ||
5dc562c5 JB |
7059 | /* |
7060 | * So if we truncate and then write and fsync we normally would just | |
7061 | * write the extents that changed, which is a problem if we need to | |
7062 | * first truncate that entire inode. So set this flag so we write out | |
7063 | * all of the extents in the inode to the sync log so we're completely | |
7064 | * safe. | |
7065 | */ | |
7066 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 7067 | trans->block_rsv = rsv; |
907cbceb | 7068 | |
8082510e YZ |
7069 | while (1) { |
7070 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
7071 | inode->i_size, | |
7072 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 7073 | if (ret != -ENOSPC) { |
3893e33b | 7074 | err = ret; |
8082510e | 7075 | break; |
3893e33b | 7076 | } |
39279cc3 | 7077 | |
fcb80c2a | 7078 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 7079 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
7080 | if (ret) { |
7081 | err = ret; | |
7082 | break; | |
7083 | } | |
ca7e70f5 | 7084 | |
8082510e | 7085 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7086 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7087 | |
7088 | trans = btrfs_start_transaction(root, 2); | |
7089 | if (IS_ERR(trans)) { | |
7090 | ret = err = PTR_ERR(trans); | |
7091 | trans = NULL; | |
7092 | break; | |
7093 | } | |
7094 | ||
7095 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7096 | rsv, min_size); | |
7097 | BUG_ON(ret); /* shouldn't happen */ | |
7098 | trans->block_rsv = rsv; | |
8082510e YZ |
7099 | } |
7100 | ||
7101 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7102 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7103 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7104 | if (ret) |
7105 | err = ret; | |
8082510e YZ |
7106 | } |
7107 | ||
917c16b2 CM |
7108 | if (trans) { |
7109 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7110 | ret = btrfs_update_inode(trans, root, inode); | |
7111 | if (ret && !err) | |
7112 | err = ret; | |
7b128766 | 7113 | |
7ad85bb7 | 7114 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7115 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7116 | } |
fcb80c2a JB |
7117 | |
7118 | out: | |
7119 | btrfs_free_block_rsv(root, rsv); | |
7120 | ||
3893e33b JB |
7121 | if (ret && !err) |
7122 | err = ret; | |
a41ad394 | 7123 | |
3893e33b | 7124 | return err; |
39279cc3 CM |
7125 | } |
7126 | ||
d352ac68 CM |
7127 | /* |
7128 | * create a new subvolume directory/inode (helper for the ioctl). | |
7129 | */ | |
d2fb3437 | 7130 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7131 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7132 | { |
39279cc3 | 7133 | struct inode *inode; |
76dda93c | 7134 | int err; |
00e4e6b3 | 7135 | u64 index = 0; |
39279cc3 | 7136 | |
12fc9d09 FA |
7137 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7138 | new_dirid, new_dirid, | |
7139 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7140 | &index); | |
54aa1f4d | 7141 | if (IS_ERR(inode)) |
f46b5a66 | 7142 | return PTR_ERR(inode); |
39279cc3 CM |
7143 | inode->i_op = &btrfs_dir_inode_operations; |
7144 | inode->i_fop = &btrfs_dir_file_operations; | |
7145 | ||
bfe86848 | 7146 | set_nlink(inode, 1); |
dbe674a9 | 7147 | btrfs_i_size_write(inode, 0); |
3b96362c | 7148 | |
76dda93c | 7149 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7150 | |
76dda93c | 7151 | iput(inode); |
ce598979 | 7152 | return err; |
39279cc3 CM |
7153 | } |
7154 | ||
39279cc3 CM |
7155 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7156 | { | |
7157 | struct btrfs_inode *ei; | |
2ead6ae7 | 7158 | struct inode *inode; |
39279cc3 CM |
7159 | |
7160 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7161 | if (!ei) | |
7162 | return NULL; | |
2ead6ae7 YZ |
7163 | |
7164 | ei->root = NULL; | |
2ead6ae7 | 7165 | ei->generation = 0; |
15ee9bc7 | 7166 | ei->last_trans = 0; |
257c62e1 | 7167 | ei->last_sub_trans = 0; |
e02119d5 | 7168 | ei->logged_trans = 0; |
2ead6ae7 | 7169 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7170 | ei->disk_i_size = 0; |
7171 | ei->flags = 0; | |
7709cde3 | 7172 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7173 | ei->index_cnt = (u64)-1; |
7174 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7175 | ei->last_log_commit = 0; |
2ead6ae7 | 7176 | |
9e0baf60 JB |
7177 | spin_lock_init(&ei->lock); |
7178 | ei->outstanding_extents = 0; | |
7179 | ei->reserved_extents = 0; | |
2ead6ae7 | 7180 | |
72ac3c0d | 7181 | ei->runtime_flags = 0; |
261507a0 | 7182 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7183 | |
16cdcec7 MX |
7184 | ei->delayed_node = NULL; |
7185 | ||
2ead6ae7 | 7186 | inode = &ei->vfs_inode; |
a8067e02 | 7187 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7188 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7189 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7190 | ei->io_tree.track_uptodate = 1; |
7191 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 7192 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 7193 | mutex_init(&ei->log_mutex); |
f248679e | 7194 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7195 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7196 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7197 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7198 | RB_CLEAR_NODE(&ei->rb_node); |
7199 | ||
7200 | return inode; | |
39279cc3 CM |
7201 | } |
7202 | ||
fa0d7e3d NP |
7203 | static void btrfs_i_callback(struct rcu_head *head) |
7204 | { | |
7205 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7206 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7207 | } | |
7208 | ||
39279cc3 CM |
7209 | void btrfs_destroy_inode(struct inode *inode) |
7210 | { | |
e6dcd2dc | 7211 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7212 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7213 | ||
b3d9b7a3 | 7214 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7215 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7216 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7217 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7218 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7219 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7220 | |
a6dbd429 JB |
7221 | /* |
7222 | * This can happen where we create an inode, but somebody else also | |
7223 | * created the same inode and we need to destroy the one we already | |
7224 | * created. | |
7225 | */ | |
7226 | if (!root) | |
7227 | goto free; | |
7228 | ||
5a3f23d5 CM |
7229 | /* |
7230 | * Make sure we're properly removed from the ordered operation | |
7231 | * lists. | |
7232 | */ | |
7233 | smp_mb(); | |
7234 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
7235 | spin_lock(&root->fs_info->ordered_extent_lock); | |
7236 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
7237 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
7238 | } | |
7239 | ||
8a35d95f JB |
7240 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7241 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
7242 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
7243 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7244 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7245 | } |
7b128766 | 7246 | |
d397712b | 7247 | while (1) { |
e6dcd2dc CM |
7248 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7249 | if (!ordered) | |
7250 | break; | |
7251 | else { | |
d397712b CM |
7252 | printk(KERN_ERR "btrfs found ordered " |
7253 | "extent %llu %llu on inode cleanup\n", | |
7254 | (unsigned long long)ordered->file_offset, | |
7255 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7256 | btrfs_remove_ordered_extent(inode, ordered); |
7257 | btrfs_put_ordered_extent(ordered); | |
7258 | btrfs_put_ordered_extent(ordered); | |
7259 | } | |
7260 | } | |
5d4f98a2 | 7261 | inode_tree_del(inode); |
5b21f2ed | 7262 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7263 | free: |
16cdcec7 | 7264 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 7265 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7266 | } |
7267 | ||
45321ac5 | 7268 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7269 | { |
7270 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7271 | |
0af3d00b | 7272 | if (btrfs_root_refs(&root->root_item) == 0 && |
83eea1f1 | 7273 | !btrfs_is_free_space_inode(inode)) |
45321ac5 | 7274 | return 1; |
76dda93c | 7275 | else |
45321ac5 | 7276 | return generic_drop_inode(inode); |
76dda93c YZ |
7277 | } |
7278 | ||
0ee0fda0 | 7279 | static void init_once(void *foo) |
39279cc3 CM |
7280 | { |
7281 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7282 | ||
7283 | inode_init_once(&ei->vfs_inode); | |
7284 | } | |
7285 | ||
7286 | void btrfs_destroy_cachep(void) | |
7287 | { | |
8c0a8537 KS |
7288 | /* |
7289 | * Make sure all delayed rcu free inodes are flushed before we | |
7290 | * destroy cache. | |
7291 | */ | |
7292 | rcu_barrier(); | |
39279cc3 CM |
7293 | if (btrfs_inode_cachep) |
7294 | kmem_cache_destroy(btrfs_inode_cachep); | |
7295 | if (btrfs_trans_handle_cachep) | |
7296 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7297 | if (btrfs_transaction_cachep) | |
7298 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7299 | if (btrfs_path_cachep) |
7300 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7301 | if (btrfs_free_space_cachep) |
7302 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7303 | if (btrfs_delalloc_work_cachep) |
7304 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7305 | } |
7306 | ||
7307 | int btrfs_init_cachep(void) | |
7308 | { | |
837e1972 | 7309 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7310 | sizeof(struct btrfs_inode), 0, |
7311 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7312 | if (!btrfs_inode_cachep) |
7313 | goto fail; | |
9601e3f6 | 7314 | |
837e1972 | 7315 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7316 | sizeof(struct btrfs_trans_handle), 0, |
7317 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7318 | if (!btrfs_trans_handle_cachep) |
7319 | goto fail; | |
9601e3f6 | 7320 | |
837e1972 | 7321 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7322 | sizeof(struct btrfs_transaction), 0, |
7323 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7324 | if (!btrfs_transaction_cachep) |
7325 | goto fail; | |
9601e3f6 | 7326 | |
837e1972 | 7327 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7328 | sizeof(struct btrfs_path), 0, |
7329 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7330 | if (!btrfs_path_cachep) |
7331 | goto fail; | |
9601e3f6 | 7332 | |
837e1972 | 7333 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7334 | sizeof(struct btrfs_free_space), 0, |
7335 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7336 | if (!btrfs_free_space_cachep) | |
7337 | goto fail; | |
7338 | ||
8ccf6f19 MX |
7339 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7340 | sizeof(struct btrfs_delalloc_work), 0, | |
7341 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7342 | NULL); | |
7343 | if (!btrfs_delalloc_work_cachep) | |
7344 | goto fail; | |
7345 | ||
39279cc3 CM |
7346 | return 0; |
7347 | fail: | |
7348 | btrfs_destroy_cachep(); | |
7349 | return -ENOMEM; | |
7350 | } | |
7351 | ||
7352 | static int btrfs_getattr(struct vfsmount *mnt, | |
7353 | struct dentry *dentry, struct kstat *stat) | |
7354 | { | |
df0af1a5 | 7355 | u64 delalloc_bytes; |
39279cc3 | 7356 | struct inode *inode = dentry->d_inode; |
fadc0d8b DS |
7357 | u32 blocksize = inode->i_sb->s_blocksize; |
7358 | ||
39279cc3 | 7359 | generic_fillattr(inode, stat); |
0ee5dc67 | 7360 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7361 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
7362 | |
7363 | spin_lock(&BTRFS_I(inode)->lock); | |
7364 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
7365 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 7366 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 7367 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
7368 | return 0; |
7369 | } | |
7370 | ||
75e7cb7f LB |
7371 | /* |
7372 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7373 | * directory. | |
7374 | */ | |
7375 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7376 | { | |
7377 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7378 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7379 | ||
7380 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7381 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7382 | else | |
7383 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7384 | ||
bc178237 | 7385 | if (b_dir->flags & BTRFS_INODE_COMPRESS) { |
75e7cb7f | 7386 | b_inode->flags |= BTRFS_INODE_COMPRESS; |
bc178237 LB |
7387 | b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
7388 | } else { | |
7389 | b_inode->flags &= ~(BTRFS_INODE_COMPRESS | | |
7390 | BTRFS_INODE_NOCOMPRESS); | |
7391 | } | |
75e7cb7f LB |
7392 | } |
7393 | ||
d397712b CM |
7394 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7395 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7396 | { |
7397 | struct btrfs_trans_handle *trans; | |
7398 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7399 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7400 | struct inode *new_inode = new_dentry->d_inode; |
7401 | struct inode *old_inode = old_dentry->d_inode; | |
7402 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7403 | u64 index = 0; |
4df27c4d | 7404 | u64 root_objectid; |
39279cc3 | 7405 | int ret; |
33345d01 | 7406 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7407 | |
33345d01 | 7408 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7409 | return -EPERM; |
7410 | ||
4df27c4d | 7411 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7412 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7413 | return -EXDEV; |
7414 | ||
33345d01 LZ |
7415 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7416 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7417 | return -ENOTEMPTY; |
5f39d397 | 7418 | |
4df27c4d YZ |
7419 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7420 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7421 | return -ENOTEMPTY; | |
9c52057c CM |
7422 | |
7423 | ||
7424 | /* check for collisions, even if the name isn't there */ | |
7425 | ret = btrfs_check_dir_item_collision(root, new_dir->i_ino, | |
7426 | new_dentry->d_name.name, | |
7427 | new_dentry->d_name.len); | |
7428 | ||
7429 | if (ret) { | |
7430 | if (ret == -EEXIST) { | |
7431 | /* we shouldn't get | |
7432 | * eexist without a new_inode */ | |
7433 | if (!new_inode) { | |
7434 | WARN_ON(1); | |
7435 | return ret; | |
7436 | } | |
7437 | } else { | |
7438 | /* maybe -EOVERFLOW */ | |
7439 | return ret; | |
7440 | } | |
7441 | } | |
7442 | ret = 0; | |
7443 | ||
5a3f23d5 CM |
7444 | /* |
7445 | * we're using rename to replace one file with another. | |
7446 | * and the replacement file is large. Start IO on it now so | |
7447 | * we don't add too much work to the end of the transaction | |
7448 | */ | |
4baf8c92 | 7449 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7450 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7451 | filemap_flush(old_inode->i_mapping); | |
7452 | ||
76dda93c | 7453 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7454 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7455 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7456 | /* |
7457 | * We want to reserve the absolute worst case amount of items. So if | |
7458 | * both inodes are subvols and we need to unlink them then that would | |
7459 | * require 4 item modifications, but if they are both normal inodes it | |
7460 | * would require 5 item modifications, so we'll assume their normal | |
7461 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7462 | * should cover the worst case number of items we'll modify. | |
7463 | */ | |
7464 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7465 | if (IS_ERR(trans)) { |
7466 | ret = PTR_ERR(trans); | |
7467 | goto out_notrans; | |
7468 | } | |
76dda93c | 7469 | |
4df27c4d YZ |
7470 | if (dest != root) |
7471 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7472 | |
a5719521 YZ |
7473 | ret = btrfs_set_inode_index(new_dir, &index); |
7474 | if (ret) | |
7475 | goto out_fail; | |
5a3f23d5 | 7476 | |
33345d01 | 7477 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7478 | /* force full log commit if subvolume involved. */ |
7479 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7480 | } else { | |
a5719521 YZ |
7481 | ret = btrfs_insert_inode_ref(trans, dest, |
7482 | new_dentry->d_name.name, | |
7483 | new_dentry->d_name.len, | |
33345d01 LZ |
7484 | old_ino, |
7485 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7486 | if (ret) |
7487 | goto out_fail; | |
4df27c4d YZ |
7488 | /* |
7489 | * this is an ugly little race, but the rename is required | |
7490 | * to make sure that if we crash, the inode is either at the | |
7491 | * old name or the new one. pinning the log transaction lets | |
7492 | * us make sure we don't allow a log commit to come in after | |
7493 | * we unlink the name but before we add the new name back in. | |
7494 | */ | |
7495 | btrfs_pin_log_trans(root); | |
7496 | } | |
5a3f23d5 CM |
7497 | /* |
7498 | * make sure the inode gets flushed if it is replacing | |
7499 | * something. | |
7500 | */ | |
33345d01 | 7501 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7502 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7503 | |
0c4d2d95 JB |
7504 | inode_inc_iversion(old_dir); |
7505 | inode_inc_iversion(new_dir); | |
7506 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7507 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7508 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7509 | old_inode->i_ctime = ctime; | |
5f39d397 | 7510 | |
12fcfd22 CM |
7511 | if (old_dentry->d_parent != new_dentry->d_parent) |
7512 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7513 | ||
33345d01 | 7514 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7515 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7516 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7517 | old_dentry->d_name.name, | |
7518 | old_dentry->d_name.len); | |
7519 | } else { | |
92986796 AV |
7520 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7521 | old_dentry->d_inode, | |
7522 | old_dentry->d_name.name, | |
7523 | old_dentry->d_name.len); | |
7524 | if (!ret) | |
7525 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7526 | } |
79787eaa JM |
7527 | if (ret) { |
7528 | btrfs_abort_transaction(trans, root, ret); | |
7529 | goto out_fail; | |
7530 | } | |
39279cc3 CM |
7531 | |
7532 | if (new_inode) { | |
0c4d2d95 | 7533 | inode_inc_iversion(new_inode); |
39279cc3 | 7534 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7535 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7536 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7537 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7538 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7539 | root_objectid, | |
7540 | new_dentry->d_name.name, | |
7541 | new_dentry->d_name.len); | |
7542 | BUG_ON(new_inode->i_nlink == 0); | |
7543 | } else { | |
7544 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7545 | new_dentry->d_inode, | |
7546 | new_dentry->d_name.name, | |
7547 | new_dentry->d_name.len); | |
7548 | } | |
79787eaa | 7549 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7550 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7551 | BUG_ON(ret); |
7b128766 | 7552 | } |
79787eaa JM |
7553 | if (ret) { |
7554 | btrfs_abort_transaction(trans, root, ret); | |
7555 | goto out_fail; | |
7556 | } | |
39279cc3 | 7557 | } |
aec7477b | 7558 | |
75e7cb7f LB |
7559 | fixup_inode_flags(new_dir, old_inode); |
7560 | ||
4df27c4d YZ |
7561 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7562 | new_dentry->d_name.name, | |
a5719521 | 7563 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7564 | if (ret) { |
7565 | btrfs_abort_transaction(trans, root, ret); | |
7566 | goto out_fail; | |
7567 | } | |
39279cc3 | 7568 | |
33345d01 | 7569 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7570 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7571 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7572 | btrfs_end_log_trans(root); |
7573 | } | |
39279cc3 | 7574 | out_fail: |
7ad85bb7 | 7575 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7576 | out_notrans: |
33345d01 | 7577 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7578 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7579 | |
39279cc3 CM |
7580 | return ret; |
7581 | } | |
7582 | ||
8ccf6f19 MX |
7583 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
7584 | { | |
7585 | struct btrfs_delalloc_work *delalloc_work; | |
7586 | ||
7587 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
7588 | work); | |
7589 | if (delalloc_work->wait) | |
7590 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
7591 | else | |
7592 | filemap_flush(delalloc_work->inode->i_mapping); | |
7593 | ||
7594 | if (delalloc_work->delay_iput) | |
7595 | btrfs_add_delayed_iput(delalloc_work->inode); | |
7596 | else | |
7597 | iput(delalloc_work->inode); | |
7598 | complete(&delalloc_work->completion); | |
7599 | } | |
7600 | ||
7601 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
7602 | int wait, int delay_iput) | |
7603 | { | |
7604 | struct btrfs_delalloc_work *work; | |
7605 | ||
7606 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
7607 | if (!work) | |
7608 | return NULL; | |
7609 | ||
7610 | init_completion(&work->completion); | |
7611 | INIT_LIST_HEAD(&work->list); | |
7612 | work->inode = inode; | |
7613 | work->wait = wait; | |
7614 | work->delay_iput = delay_iput; | |
7615 | work->work.func = btrfs_run_delalloc_work; | |
7616 | ||
7617 | return work; | |
7618 | } | |
7619 | ||
7620 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
7621 | { | |
7622 | wait_for_completion(&work->completion); | |
7623 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
7624 | } | |
7625 | ||
d352ac68 CM |
7626 | /* |
7627 | * some fairly slow code that needs optimization. This walks the list | |
7628 | * of all the inodes with pending delalloc and forces them to disk. | |
7629 | */ | |
24bbcf04 | 7630 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 | 7631 | { |
ea8c2819 | 7632 | struct btrfs_inode *binode; |
5b21f2ed | 7633 | struct inode *inode; |
8ccf6f19 MX |
7634 | struct btrfs_delalloc_work *work, *next; |
7635 | struct list_head works; | |
1eafa6c7 | 7636 | struct list_head splice; |
8ccf6f19 | 7637 | int ret = 0; |
ea8c2819 | 7638 | |
c146afad YZ |
7639 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7640 | return -EROFS; | |
7641 | ||
8ccf6f19 | 7642 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 7643 | INIT_LIST_HEAD(&splice); |
63607cc8 | 7644 | |
75eff68e | 7645 | spin_lock(&root->fs_info->delalloc_lock); |
1eafa6c7 MX |
7646 | list_splice_init(&root->fs_info->delalloc_inodes, &splice); |
7647 | while (!list_empty(&splice)) { | |
7648 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 7649 | delalloc_inodes); |
1eafa6c7 MX |
7650 | |
7651 | list_del_init(&binode->delalloc_inodes); | |
7652 | ||
5b21f2ed | 7653 | inode = igrab(&binode->vfs_inode); |
df0af1a5 MX |
7654 | if (!inode) { |
7655 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
7656 | &binode->runtime_flags); | |
1eafa6c7 | 7657 | continue; |
df0af1a5 | 7658 | } |
1eafa6c7 MX |
7659 | |
7660 | list_add_tail(&binode->delalloc_inodes, | |
7661 | &root->fs_info->delalloc_inodes); | |
75eff68e | 7662 | spin_unlock(&root->fs_info->delalloc_lock); |
1eafa6c7 MX |
7663 | |
7664 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
7665 | if (unlikely(!work)) { | |
7666 | ret = -ENOMEM; | |
7667 | goto out; | |
5b21f2ed | 7668 | } |
1eafa6c7 MX |
7669 | list_add_tail(&work->list, &works); |
7670 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
7671 | &work->work); | |
7672 | ||
5b21f2ed | 7673 | cond_resched(); |
75eff68e | 7674 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7675 | } |
75eff68e | 7676 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d | 7677 | |
1eafa6c7 MX |
7678 | list_for_each_entry_safe(work, next, &works, list) { |
7679 | list_del_init(&work->list); | |
7680 | btrfs_wait_and_free_delalloc_work(work); | |
7681 | } | |
7682 | ||
8c8bee1d CM |
7683 | /* the filemap_flush will queue IO into the worker threads, but |
7684 | * we have to make sure the IO is actually started and that | |
7685 | * ordered extents get created before we return | |
7686 | */ | |
7687 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7688 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7689 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7690 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7691 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7692 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7693 | } |
7694 | atomic_dec(&root->fs_info->async_submit_draining); | |
1eafa6c7 | 7695 | return 0; |
8ccf6f19 MX |
7696 | out: |
7697 | list_for_each_entry_safe(work, next, &works, list) { | |
7698 | list_del_init(&work->list); | |
7699 | btrfs_wait_and_free_delalloc_work(work); | |
7700 | } | |
1eafa6c7 MX |
7701 | |
7702 | if (!list_empty_careful(&splice)) { | |
7703 | spin_lock(&root->fs_info->delalloc_lock); | |
7704 | list_splice_tail(&splice, &root->fs_info->delalloc_inodes); | |
7705 | spin_unlock(&root->fs_info->delalloc_lock); | |
7706 | } | |
8ccf6f19 | 7707 | return ret; |
ea8c2819 CM |
7708 | } |
7709 | ||
39279cc3 CM |
7710 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7711 | const char *symname) | |
7712 | { | |
7713 | struct btrfs_trans_handle *trans; | |
7714 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7715 | struct btrfs_path *path; | |
7716 | struct btrfs_key key; | |
1832a6d5 | 7717 | struct inode *inode = NULL; |
39279cc3 CM |
7718 | int err; |
7719 | int drop_inode = 0; | |
7720 | u64 objectid; | |
00e4e6b3 | 7721 | u64 index = 0 ; |
39279cc3 CM |
7722 | int name_len; |
7723 | int datasize; | |
5f39d397 | 7724 | unsigned long ptr; |
39279cc3 | 7725 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7726 | struct extent_buffer *leaf; |
39279cc3 CM |
7727 | |
7728 | name_len = strlen(symname) + 1; | |
7729 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7730 | return -ENAMETOOLONG; | |
1832a6d5 | 7731 | |
9ed74f2d JB |
7732 | /* |
7733 | * 2 items for inode item and ref | |
7734 | * 2 items for dir items | |
7735 | * 1 item for xattr if selinux is on | |
7736 | */ | |
a22285a6 YZ |
7737 | trans = btrfs_start_transaction(root, 5); |
7738 | if (IS_ERR(trans)) | |
7739 | return PTR_ERR(trans); | |
1832a6d5 | 7740 | |
581bb050 LZ |
7741 | err = btrfs_find_free_ino(root, &objectid); |
7742 | if (err) | |
7743 | goto out_unlock; | |
7744 | ||
aec7477b | 7745 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7746 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7747 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7748 | if (IS_ERR(inode)) { |
7749 | err = PTR_ERR(inode); | |
39279cc3 | 7750 | goto out_unlock; |
7cf96da3 | 7751 | } |
39279cc3 | 7752 | |
2a7dba39 | 7753 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7754 | if (err) { |
7755 | drop_inode = 1; | |
7756 | goto out_unlock; | |
7757 | } | |
7758 | ||
ad19db71 CS |
7759 | /* |
7760 | * If the active LSM wants to access the inode during | |
7761 | * d_instantiate it needs these. Smack checks to see | |
7762 | * if the filesystem supports xattrs by looking at the | |
7763 | * ops vector. | |
7764 | */ | |
7765 | inode->i_fop = &btrfs_file_operations; | |
7766 | inode->i_op = &btrfs_file_inode_operations; | |
7767 | ||
a1b075d2 | 7768 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7769 | if (err) |
7770 | drop_inode = 1; | |
7771 | else { | |
7772 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7773 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7774 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7775 | } |
39279cc3 CM |
7776 | if (drop_inode) |
7777 | goto out_unlock; | |
7778 | ||
7779 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7780 | if (!path) { |
7781 | err = -ENOMEM; | |
7782 | drop_inode = 1; | |
7783 | goto out_unlock; | |
7784 | } | |
33345d01 | 7785 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7786 | key.offset = 0; |
39279cc3 CM |
7787 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7788 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7789 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7790 | datasize); | |
54aa1f4d CM |
7791 | if (err) { |
7792 | drop_inode = 1; | |
b0839166 | 7793 | btrfs_free_path(path); |
54aa1f4d CM |
7794 | goto out_unlock; |
7795 | } | |
5f39d397 CM |
7796 | leaf = path->nodes[0]; |
7797 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7798 | struct btrfs_file_extent_item); | |
7799 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7800 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7801 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7802 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7803 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7804 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7805 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7806 | ||
39279cc3 | 7807 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7808 | write_extent_buffer(leaf, symname, ptr, name_len); |
7809 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7810 | btrfs_free_path(path); |
5f39d397 | 7811 | |
39279cc3 CM |
7812 | inode->i_op = &btrfs_symlink_inode_operations; |
7813 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7814 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7815 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7816 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7817 | err = btrfs_update_inode(trans, root, inode); |
7818 | if (err) | |
7819 | drop_inode = 1; | |
39279cc3 CM |
7820 | |
7821 | out_unlock: | |
08c422c2 AV |
7822 | if (!err) |
7823 | d_instantiate(dentry, inode); | |
7ad85bb7 | 7824 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7825 | if (drop_inode) { |
7826 | inode_dec_link_count(inode); | |
7827 | iput(inode); | |
7828 | } | |
b53d3f5d | 7829 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
7830 | return err; |
7831 | } | |
16432985 | 7832 | |
0af3d00b JB |
7833 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7834 | u64 start, u64 num_bytes, u64 min_size, | |
7835 | loff_t actual_len, u64 *alloc_hint, | |
7836 | struct btrfs_trans_handle *trans) | |
d899e052 | 7837 | { |
5dc562c5 JB |
7838 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7839 | struct extent_map *em; | |
d899e052 YZ |
7840 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7841 | struct btrfs_key ins; | |
d899e052 | 7842 | u64 cur_offset = start; |
55a61d1d | 7843 | u64 i_size; |
d899e052 | 7844 | int ret = 0; |
0af3d00b | 7845 | bool own_trans = true; |
d899e052 | 7846 | |
0af3d00b JB |
7847 | if (trans) |
7848 | own_trans = false; | |
d899e052 | 7849 | while (num_bytes > 0) { |
0af3d00b JB |
7850 | if (own_trans) { |
7851 | trans = btrfs_start_transaction(root, 3); | |
7852 | if (IS_ERR(trans)) { | |
7853 | ret = PTR_ERR(trans); | |
7854 | break; | |
7855 | } | |
5a303d5d YZ |
7856 | } |
7857 | ||
efa56464 | 7858 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7859 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7860 | if (ret) { |
0af3d00b JB |
7861 | if (own_trans) |
7862 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7863 | break; |
d899e052 | 7864 | } |
5a303d5d | 7865 | |
d899e052 YZ |
7866 | ret = insert_reserved_file_extent(trans, inode, |
7867 | cur_offset, ins.objectid, | |
7868 | ins.offset, ins.offset, | |
920bbbfb | 7869 | ins.offset, 0, 0, 0, |
d899e052 | 7870 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7871 | if (ret) { |
7872 | btrfs_abort_transaction(trans, root, ret); | |
7873 | if (own_trans) | |
7874 | btrfs_end_transaction(trans, root); | |
7875 | break; | |
7876 | } | |
a1ed835e CM |
7877 | btrfs_drop_extent_cache(inode, cur_offset, |
7878 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7879 | |
5dc562c5 JB |
7880 | em = alloc_extent_map(); |
7881 | if (!em) { | |
7882 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
7883 | &BTRFS_I(inode)->runtime_flags); | |
7884 | goto next; | |
7885 | } | |
7886 | ||
7887 | em->start = cur_offset; | |
7888 | em->orig_start = cur_offset; | |
7889 | em->len = ins.offset; | |
7890 | em->block_start = ins.objectid; | |
7891 | em->block_len = ins.offset; | |
b4939680 | 7892 | em->orig_block_len = ins.offset; |
5dc562c5 JB |
7893 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7894 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
7895 | em->generation = trans->transid; | |
7896 | ||
7897 | while (1) { | |
7898 | write_lock(&em_tree->lock); | |
7899 | ret = add_extent_mapping(em_tree, em); | |
7900 | if (!ret) | |
7901 | list_move(&em->list, | |
7902 | &em_tree->modified_extents); | |
7903 | write_unlock(&em_tree->lock); | |
7904 | if (ret != -EEXIST) | |
7905 | break; | |
7906 | btrfs_drop_extent_cache(inode, cur_offset, | |
7907 | cur_offset + ins.offset - 1, | |
7908 | 0); | |
7909 | } | |
7910 | free_extent_map(em); | |
7911 | next: | |
d899e052 YZ |
7912 | num_bytes -= ins.offset; |
7913 | cur_offset += ins.offset; | |
efa56464 | 7914 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7915 | |
0c4d2d95 | 7916 | inode_inc_iversion(inode); |
d899e052 | 7917 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7918 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7919 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7920 | (actual_len > inode->i_size) && |
7921 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7922 | if (cur_offset > actual_len) |
55a61d1d | 7923 | i_size = actual_len; |
d1ea6a61 | 7924 | else |
55a61d1d JB |
7925 | i_size = cur_offset; |
7926 | i_size_write(inode, i_size); | |
7927 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7928 | } |
7929 | ||
d899e052 | 7930 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7931 | |
7932 | if (ret) { | |
7933 | btrfs_abort_transaction(trans, root, ret); | |
7934 | if (own_trans) | |
7935 | btrfs_end_transaction(trans, root); | |
7936 | break; | |
7937 | } | |
d899e052 | 7938 | |
0af3d00b JB |
7939 | if (own_trans) |
7940 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7941 | } |
d899e052 YZ |
7942 | return ret; |
7943 | } | |
7944 | ||
0af3d00b JB |
7945 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7946 | u64 start, u64 num_bytes, u64 min_size, | |
7947 | loff_t actual_len, u64 *alloc_hint) | |
7948 | { | |
7949 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7950 | min_size, actual_len, alloc_hint, | |
7951 | NULL); | |
7952 | } | |
7953 | ||
7954 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7955 | struct btrfs_trans_handle *trans, int mode, | |
7956 | u64 start, u64 num_bytes, u64 min_size, | |
7957 | loff_t actual_len, u64 *alloc_hint) | |
7958 | { | |
7959 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7960 | min_size, actual_len, alloc_hint, trans); | |
7961 | } | |
7962 | ||
e6dcd2dc CM |
7963 | static int btrfs_set_page_dirty(struct page *page) |
7964 | { | |
e6dcd2dc CM |
7965 | return __set_page_dirty_nobuffers(page); |
7966 | } | |
7967 | ||
10556cb2 | 7968 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7969 | { |
b83cc969 | 7970 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7971 | umode_t mode = inode->i_mode; |
b83cc969 | 7972 | |
cb6db4e5 JM |
7973 | if (mask & MAY_WRITE && |
7974 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7975 | if (btrfs_root_readonly(root)) | |
7976 | return -EROFS; | |
7977 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7978 | return -EACCES; | |
7979 | } | |
2830ba7f | 7980 | return generic_permission(inode, mask); |
fdebe2bd | 7981 | } |
39279cc3 | 7982 | |
6e1d5dcc | 7983 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7984 | .getattr = btrfs_getattr, |
39279cc3 CM |
7985 | .lookup = btrfs_lookup, |
7986 | .create = btrfs_create, | |
7987 | .unlink = btrfs_unlink, | |
7988 | .link = btrfs_link, | |
7989 | .mkdir = btrfs_mkdir, | |
7990 | .rmdir = btrfs_rmdir, | |
7991 | .rename = btrfs_rename, | |
7992 | .symlink = btrfs_symlink, | |
7993 | .setattr = btrfs_setattr, | |
618e21d5 | 7994 | .mknod = btrfs_mknod, |
95819c05 CH |
7995 | .setxattr = btrfs_setxattr, |
7996 | .getxattr = btrfs_getxattr, | |
5103e947 | 7997 | .listxattr = btrfs_listxattr, |
95819c05 | 7998 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7999 | .permission = btrfs_permission, |
4e34e719 | 8000 | .get_acl = btrfs_get_acl, |
39279cc3 | 8001 | }; |
6e1d5dcc | 8002 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 8003 | .lookup = btrfs_lookup, |
fdebe2bd | 8004 | .permission = btrfs_permission, |
4e34e719 | 8005 | .get_acl = btrfs_get_acl, |
39279cc3 | 8006 | }; |
76dda93c | 8007 | |
828c0950 | 8008 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
8009 | .llseek = generic_file_llseek, |
8010 | .read = generic_read_dir, | |
cbdf5a24 | 8011 | .readdir = btrfs_real_readdir, |
34287aa3 | 8012 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 8013 | #ifdef CONFIG_COMPAT |
34287aa3 | 8014 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 8015 | #endif |
6bf13c0c | 8016 | .release = btrfs_release_file, |
e02119d5 | 8017 | .fsync = btrfs_sync_file, |
39279cc3 CM |
8018 | }; |
8019 | ||
d1310b2e | 8020 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 8021 | .fill_delalloc = run_delalloc_range, |
065631f6 | 8022 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 8023 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 8024 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 8025 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 8026 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
8027 | .set_bit_hook = btrfs_set_bit_hook, |
8028 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
8029 | .merge_extent_hook = btrfs_merge_extent_hook, |
8030 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
8031 | }; |
8032 | ||
35054394 CM |
8033 | /* |
8034 | * btrfs doesn't support the bmap operation because swapfiles | |
8035 | * use bmap to make a mapping of extents in the file. They assume | |
8036 | * these extents won't change over the life of the file and they | |
8037 | * use the bmap result to do IO directly to the drive. | |
8038 | * | |
8039 | * the btrfs bmap call would return logical addresses that aren't | |
8040 | * suitable for IO and they also will change frequently as COW | |
8041 | * operations happen. So, swapfile + btrfs == corruption. | |
8042 | * | |
8043 | * For now we're avoiding this by dropping bmap. | |
8044 | */ | |
7f09410b | 8045 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
8046 | .readpage = btrfs_readpage, |
8047 | .writepage = btrfs_writepage, | |
b293f02e | 8048 | .writepages = btrfs_writepages, |
3ab2fb5a | 8049 | .readpages = btrfs_readpages, |
16432985 | 8050 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
8051 | .invalidatepage = btrfs_invalidatepage, |
8052 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 8053 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 8054 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
8055 | }; |
8056 | ||
7f09410b | 8057 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
8058 | .readpage = btrfs_readpage, |
8059 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
8060 | .invalidatepage = btrfs_invalidatepage, |
8061 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
8062 | }; |
8063 | ||
6e1d5dcc | 8064 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
8065 | .getattr = btrfs_getattr, |
8066 | .setattr = btrfs_setattr, | |
95819c05 CH |
8067 | .setxattr = btrfs_setxattr, |
8068 | .getxattr = btrfs_getxattr, | |
5103e947 | 8069 | .listxattr = btrfs_listxattr, |
95819c05 | 8070 | .removexattr = btrfs_removexattr, |
fdebe2bd | 8071 | .permission = btrfs_permission, |
1506fcc8 | 8072 | .fiemap = btrfs_fiemap, |
4e34e719 | 8073 | .get_acl = btrfs_get_acl, |
e41f941a | 8074 | .update_time = btrfs_update_time, |
39279cc3 | 8075 | }; |
6e1d5dcc | 8076 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
8077 | .getattr = btrfs_getattr, |
8078 | .setattr = btrfs_setattr, | |
fdebe2bd | 8079 | .permission = btrfs_permission, |
95819c05 CH |
8080 | .setxattr = btrfs_setxattr, |
8081 | .getxattr = btrfs_getxattr, | |
33268eaf | 8082 | .listxattr = btrfs_listxattr, |
95819c05 | 8083 | .removexattr = btrfs_removexattr, |
4e34e719 | 8084 | .get_acl = btrfs_get_acl, |
e41f941a | 8085 | .update_time = btrfs_update_time, |
618e21d5 | 8086 | }; |
6e1d5dcc | 8087 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
8088 | .readlink = generic_readlink, |
8089 | .follow_link = page_follow_link_light, | |
8090 | .put_link = page_put_link, | |
f209561a | 8091 | .getattr = btrfs_getattr, |
22c44fe6 | 8092 | .setattr = btrfs_setattr, |
fdebe2bd | 8093 | .permission = btrfs_permission, |
0279b4cd JO |
8094 | .setxattr = btrfs_setxattr, |
8095 | .getxattr = btrfs_getxattr, | |
8096 | .listxattr = btrfs_listxattr, | |
8097 | .removexattr = btrfs_removexattr, | |
4e34e719 | 8098 | .get_acl = btrfs_get_acl, |
e41f941a | 8099 | .update_time = btrfs_update_time, |
39279cc3 | 8100 | }; |
76dda93c | 8101 | |
82d339d9 | 8102 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 8103 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 8104 | .d_release = btrfs_dentry_release, |
76dda93c | 8105 | }; |