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> |
4b4e25f2 | 42 | #include "compat.h" |
39279cc3 CM |
43 | #include "ctree.h" |
44 | #include "disk-io.h" | |
45 | #include "transaction.h" | |
46 | #include "btrfs_inode.h" | |
47 | #include "ioctl.h" | |
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 | ||
a41ad394 JB |
91 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
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); | |
7b128766 | 98 | |
f34f57a3 | 99 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
100 | struct inode *inode, struct inode *dir, |
101 | const struct qstr *qstr) | |
0279b4cd JO |
102 | { |
103 | int err; | |
104 | ||
f34f57a3 | 105 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 106 | if (!err) |
2a7dba39 | 107 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
108 | return err; |
109 | } | |
110 | ||
c8b97818 CM |
111 | /* |
112 | * this does all the hard work for inserting an inline extent into | |
113 | * the btree. The caller should have done a btrfs_drop_extents so that | |
114 | * no overlapping inline items exist in the btree | |
115 | */ | |
d397712b | 116 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
117 | struct btrfs_root *root, struct inode *inode, |
118 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 119 | int compress_type, |
c8b97818 CM |
120 | struct page **compressed_pages) |
121 | { | |
122 | struct btrfs_key key; | |
123 | struct btrfs_path *path; | |
124 | struct extent_buffer *leaf; | |
125 | struct page *page = NULL; | |
126 | char *kaddr; | |
127 | unsigned long ptr; | |
128 | struct btrfs_file_extent_item *ei; | |
129 | int err = 0; | |
130 | int ret; | |
131 | size_t cur_size = size; | |
132 | size_t datasize; | |
133 | unsigned long offset; | |
c8b97818 | 134 | |
fe3f566c | 135 | if (compressed_size && compressed_pages) |
c8b97818 | 136 | cur_size = compressed_size; |
c8b97818 | 137 | |
d397712b CM |
138 | path = btrfs_alloc_path(); |
139 | if (!path) | |
c8b97818 CM |
140 | return -ENOMEM; |
141 | ||
b9473439 | 142 | path->leave_spinning = 1; |
c8b97818 | 143 | |
33345d01 | 144 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
145 | key.offset = start; |
146 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
147 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
148 | ||
149 | inode_add_bytes(inode, size); | |
150 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
151 | datasize); | |
c8b97818 CM |
152 | if (ret) { |
153 | err = ret; | |
c8b97818 CM |
154 | goto fail; |
155 | } | |
156 | leaf = path->nodes[0]; | |
157 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
158 | struct btrfs_file_extent_item); | |
159 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
160 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
161 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
162 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
163 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
164 | ptr = btrfs_file_extent_inline_start(ei); | |
165 | ||
261507a0 | 166 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
167 | struct page *cpage; |
168 | int i = 0; | |
d397712b | 169 | while (compressed_size > 0) { |
c8b97818 | 170 | cpage = compressed_pages[i]; |
5b050f04 | 171 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
172 | PAGE_CACHE_SIZE); |
173 | ||
7ac687d9 | 174 | kaddr = kmap_atomic(cpage); |
c8b97818 | 175 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 176 | kunmap_atomic(kaddr); |
c8b97818 CM |
177 | |
178 | i++; | |
179 | ptr += cur_size; | |
180 | compressed_size -= cur_size; | |
181 | } | |
182 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 183 | compress_type); |
c8b97818 CM |
184 | } else { |
185 | page = find_get_page(inode->i_mapping, | |
186 | start >> PAGE_CACHE_SHIFT); | |
187 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 188 | kaddr = kmap_atomic(page); |
c8b97818 CM |
189 | offset = start & (PAGE_CACHE_SIZE - 1); |
190 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 191 | kunmap_atomic(kaddr); |
c8b97818 CM |
192 | page_cache_release(page); |
193 | } | |
194 | btrfs_mark_buffer_dirty(leaf); | |
195 | btrfs_free_path(path); | |
196 | ||
c2167754 YZ |
197 | /* |
198 | * we're an inline extent, so nobody can | |
199 | * extend the file past i_size without locking | |
200 | * a page we already have locked. | |
201 | * | |
202 | * We must do any isize and inode updates | |
203 | * before we unlock the pages. Otherwise we | |
204 | * could end up racing with unlink. | |
205 | */ | |
c8b97818 | 206 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 207 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 208 | |
79787eaa | 209 | return ret; |
c8b97818 CM |
210 | fail: |
211 | btrfs_free_path(path); | |
212 | return err; | |
213 | } | |
214 | ||
215 | ||
216 | /* | |
217 | * conditionally insert an inline extent into the file. This | |
218 | * does the checks required to make sure the data is small enough | |
219 | * to fit as an inline extent. | |
220 | */ | |
7f366cfe | 221 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
222 | struct btrfs_root *root, |
223 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 224 | size_t compressed_size, int compress_type, |
c8b97818 CM |
225 | struct page **compressed_pages) |
226 | { | |
227 | u64 isize = i_size_read(inode); | |
228 | u64 actual_end = min(end + 1, isize); | |
229 | u64 inline_len = actual_end - start; | |
230 | u64 aligned_end = (end + root->sectorsize - 1) & | |
231 | ~((u64)root->sectorsize - 1); | |
c8b97818 CM |
232 | u64 data_len = inline_len; |
233 | int ret; | |
234 | ||
235 | if (compressed_size) | |
236 | data_len = compressed_size; | |
237 | ||
238 | if (start > 0 || | |
70b99e69 | 239 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
240 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
241 | (!compressed_size && | |
242 | (actual_end & (root->sectorsize - 1)) == 0) || | |
243 | end + 1 < isize || | |
244 | data_len > root->fs_info->max_inline) { | |
245 | return 1; | |
246 | } | |
247 | ||
2671485d | 248 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
249 | if (ret) |
250 | return ret; | |
c8b97818 CM |
251 | |
252 | if (isize > actual_end) | |
253 | inline_len = min_t(u64, isize, actual_end); | |
254 | ret = insert_inline_extent(trans, root, inode, start, | |
255 | inline_len, compressed_size, | |
fe3f566c | 256 | compress_type, compressed_pages); |
2adcac1a | 257 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
258 | btrfs_abort_transaction(trans, root, ret); |
259 | return ret; | |
2adcac1a JB |
260 | } else if (ret == -ENOSPC) { |
261 | return 1; | |
79787eaa | 262 | } |
2adcac1a | 263 | |
0ca1f7ce | 264 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 265 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
266 | return 0; |
267 | } | |
268 | ||
771ed689 CM |
269 | struct async_extent { |
270 | u64 start; | |
271 | u64 ram_size; | |
272 | u64 compressed_size; | |
273 | struct page **pages; | |
274 | unsigned long nr_pages; | |
261507a0 | 275 | int compress_type; |
771ed689 CM |
276 | struct list_head list; |
277 | }; | |
278 | ||
279 | struct async_cow { | |
280 | struct inode *inode; | |
281 | struct btrfs_root *root; | |
282 | struct page *locked_page; | |
283 | u64 start; | |
284 | u64 end; | |
285 | struct list_head extents; | |
286 | struct btrfs_work work; | |
287 | }; | |
288 | ||
289 | static noinline int add_async_extent(struct async_cow *cow, | |
290 | u64 start, u64 ram_size, | |
291 | u64 compressed_size, | |
292 | struct page **pages, | |
261507a0 LZ |
293 | unsigned long nr_pages, |
294 | int compress_type) | |
771ed689 CM |
295 | { |
296 | struct async_extent *async_extent; | |
297 | ||
298 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 299 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
300 | async_extent->start = start; |
301 | async_extent->ram_size = ram_size; | |
302 | async_extent->compressed_size = compressed_size; | |
303 | async_extent->pages = pages; | |
304 | async_extent->nr_pages = nr_pages; | |
261507a0 | 305 | async_extent->compress_type = compress_type; |
771ed689 CM |
306 | list_add_tail(&async_extent->list, &cow->extents); |
307 | return 0; | |
308 | } | |
309 | ||
d352ac68 | 310 | /* |
771ed689 CM |
311 | * we create compressed extents in two phases. The first |
312 | * phase compresses a range of pages that have already been | |
313 | * locked (both pages and state bits are locked). | |
c8b97818 | 314 | * |
771ed689 CM |
315 | * This is done inside an ordered work queue, and the compression |
316 | * is spread across many cpus. The actual IO submission is step | |
317 | * two, and the ordered work queue takes care of making sure that | |
318 | * happens in the same order things were put onto the queue by | |
319 | * writepages and friends. | |
c8b97818 | 320 | * |
771ed689 CM |
321 | * If this code finds it can't get good compression, it puts an |
322 | * entry onto the work queue to write the uncompressed bytes. This | |
323 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
324 | * are written in the same order that the flusher thread sent them |
325 | * down. | |
d352ac68 | 326 | */ |
771ed689 CM |
327 | static noinline int compress_file_range(struct inode *inode, |
328 | struct page *locked_page, | |
329 | u64 start, u64 end, | |
330 | struct async_cow *async_cow, | |
331 | int *num_added) | |
b888db2b CM |
332 | { |
333 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
334 | struct btrfs_trans_handle *trans; | |
db94535d | 335 | u64 num_bytes; |
db94535d | 336 | u64 blocksize = root->sectorsize; |
c8b97818 | 337 | u64 actual_end; |
42dc7bab | 338 | u64 isize = i_size_read(inode); |
e6dcd2dc | 339 | int ret = 0; |
c8b97818 CM |
340 | struct page **pages = NULL; |
341 | unsigned long nr_pages; | |
342 | unsigned long nr_pages_ret = 0; | |
343 | unsigned long total_compressed = 0; | |
344 | unsigned long total_in = 0; | |
345 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 346 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
347 | int i; |
348 | int will_compress; | |
261507a0 | 349 | int compress_type = root->fs_info->compress_type; |
b888db2b | 350 | |
4cb13e5d LB |
351 | /* if this is a small write inside eof, kick off a defrag */ |
352 | if ((end - start + 1) < 16 * 1024 && | |
353 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
354 | btrfs_add_inode_defrag(NULL, inode); |
355 | ||
42dc7bab | 356 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
357 | again: |
358 | will_compress = 0; | |
359 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
360 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 361 | |
f03d9301 CM |
362 | /* |
363 | * we don't want to send crud past the end of i_size through | |
364 | * compression, that's just a waste of CPU time. So, if the | |
365 | * end of the file is before the start of our current | |
366 | * requested range of bytes, we bail out to the uncompressed | |
367 | * cleanup code that can deal with all of this. | |
368 | * | |
369 | * It isn't really the fastest way to fix things, but this is a | |
370 | * very uncommon corner. | |
371 | */ | |
372 | if (actual_end <= start) | |
373 | goto cleanup_and_bail_uncompressed; | |
374 | ||
c8b97818 CM |
375 | total_compressed = actual_end - start; |
376 | ||
377 | /* we want to make sure that amount of ram required to uncompress | |
378 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
379 | * of a compressed extent to 128k. This is a crucial number |
380 | * because it also controls how easily we can spread reads across | |
381 | * cpus for decompression. | |
382 | * | |
383 | * We also want to make sure the amount of IO required to do | |
384 | * a random read is reasonably small, so we limit the size of | |
385 | * a compressed extent to 128k. | |
c8b97818 CM |
386 | */ |
387 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 388 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 389 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
390 | total_in = 0; |
391 | ret = 0; | |
db94535d | 392 | |
771ed689 CM |
393 | /* |
394 | * we do compression for mount -o compress and when the | |
395 | * inode has not been flagged as nocompress. This flag can | |
396 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 397 | */ |
6cbff00f | 398 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 399 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
400 | (BTRFS_I(inode)->force_compress) || |
401 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 402 | WARN_ON(pages); |
cfbc246e | 403 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
404 | if (!pages) { |
405 | /* just bail out to the uncompressed code */ | |
406 | goto cont; | |
407 | } | |
c8b97818 | 408 | |
261507a0 LZ |
409 | if (BTRFS_I(inode)->force_compress) |
410 | compress_type = BTRFS_I(inode)->force_compress; | |
411 | ||
412 | ret = btrfs_compress_pages(compress_type, | |
413 | inode->i_mapping, start, | |
414 | total_compressed, pages, | |
415 | nr_pages, &nr_pages_ret, | |
416 | &total_in, | |
417 | &total_compressed, | |
418 | max_compressed); | |
c8b97818 CM |
419 | |
420 | if (!ret) { | |
421 | unsigned long offset = total_compressed & | |
422 | (PAGE_CACHE_SIZE - 1); | |
423 | struct page *page = pages[nr_pages_ret - 1]; | |
424 | char *kaddr; | |
425 | ||
426 | /* zero the tail end of the last page, we might be | |
427 | * sending it down to disk | |
428 | */ | |
429 | if (offset) { | |
7ac687d9 | 430 | kaddr = kmap_atomic(page); |
c8b97818 CM |
431 | memset(kaddr + offset, 0, |
432 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 433 | kunmap_atomic(kaddr); |
c8b97818 CM |
434 | } |
435 | will_compress = 1; | |
436 | } | |
437 | } | |
560f7d75 | 438 | cont: |
c8b97818 | 439 | if (start == 0) { |
7a7eaa40 | 440 | trans = btrfs_join_transaction(root); |
79787eaa JM |
441 | if (IS_ERR(trans)) { |
442 | ret = PTR_ERR(trans); | |
443 | trans = NULL; | |
444 | goto cleanup_and_out; | |
445 | } | |
0ca1f7ce | 446 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 447 | |
c8b97818 | 448 | /* lets try to make an inline extent */ |
771ed689 | 449 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 450 | /* we didn't compress the entire range, try |
771ed689 | 451 | * to make an uncompressed inline extent. |
c8b97818 CM |
452 | */ |
453 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 454 | start, end, 0, 0, NULL); |
c8b97818 | 455 | } else { |
771ed689 | 456 | /* try making a compressed inline extent */ |
c8b97818 CM |
457 | ret = cow_file_range_inline(trans, root, inode, |
458 | start, end, | |
fe3f566c LZ |
459 | total_compressed, |
460 | compress_type, pages); | |
c8b97818 | 461 | } |
79787eaa | 462 | if (ret <= 0) { |
771ed689 | 463 | /* |
79787eaa JM |
464 | * inline extent creation worked or returned error, |
465 | * we don't need to create any more async work items. | |
466 | * Unlock and free up our temp pages. | |
771ed689 | 467 | */ |
c8b97818 | 468 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
469 | &BTRFS_I(inode)->io_tree, |
470 | start, end, NULL, | |
471 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 472 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 473 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
474 | |
475 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
476 | goto free_pages_out; |
477 | } | |
c2167754 | 478 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
479 | } |
480 | ||
481 | if (will_compress) { | |
482 | /* | |
483 | * we aren't doing an inline extent round the compressed size | |
484 | * up to a block size boundary so the allocator does sane | |
485 | * things | |
486 | */ | |
487 | total_compressed = (total_compressed + blocksize - 1) & | |
488 | ~(blocksize - 1); | |
489 | ||
490 | /* | |
491 | * one last check to make sure the compression is really a | |
492 | * win, compare the page count read with the blocks on disk | |
493 | */ | |
494 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
495 | ~(PAGE_CACHE_SIZE - 1); | |
496 | if (total_compressed >= total_in) { | |
497 | will_compress = 0; | |
498 | } else { | |
c8b97818 CM |
499 | num_bytes = total_in; |
500 | } | |
501 | } | |
502 | if (!will_compress && pages) { | |
503 | /* | |
504 | * the compression code ran but failed to make things smaller, | |
505 | * free any pages it allocated and our page pointer array | |
506 | */ | |
507 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 508 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
509 | page_cache_release(pages[i]); |
510 | } | |
511 | kfree(pages); | |
512 | pages = NULL; | |
513 | total_compressed = 0; | |
514 | nr_pages_ret = 0; | |
515 | ||
516 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
517 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
518 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 519 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 520 | } |
c8b97818 | 521 | } |
771ed689 CM |
522 | if (will_compress) { |
523 | *num_added += 1; | |
c8b97818 | 524 | |
771ed689 CM |
525 | /* the async work queues will take care of doing actual |
526 | * allocation on disk for these compressed pages, | |
527 | * and will submit them to the elevator. | |
528 | */ | |
529 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
530 | total_compressed, pages, nr_pages_ret, |
531 | compress_type); | |
179e29e4 | 532 | |
24ae6365 | 533 | if (start + num_bytes < end) { |
771ed689 CM |
534 | start += num_bytes; |
535 | pages = NULL; | |
536 | cond_resched(); | |
537 | goto again; | |
538 | } | |
539 | } else { | |
f03d9301 | 540 | cleanup_and_bail_uncompressed: |
771ed689 CM |
541 | /* |
542 | * No compression, but we still need to write the pages in | |
543 | * the file we've been given so far. redirty the locked | |
544 | * page if it corresponds to our extent and set things up | |
545 | * for the async work queue to run cow_file_range to do | |
546 | * the normal delalloc dance | |
547 | */ | |
548 | if (page_offset(locked_page) >= start && | |
549 | page_offset(locked_page) <= end) { | |
550 | __set_page_dirty_nobuffers(locked_page); | |
551 | /* unlocked later on in the async handlers */ | |
552 | } | |
261507a0 LZ |
553 | add_async_extent(async_cow, start, end - start + 1, |
554 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
555 | *num_added += 1; |
556 | } | |
3b951516 | 557 | |
771ed689 | 558 | out: |
79787eaa | 559 | return ret; |
771ed689 CM |
560 | |
561 | free_pages_out: | |
562 | for (i = 0; i < nr_pages_ret; i++) { | |
563 | WARN_ON(pages[i]->mapping); | |
564 | page_cache_release(pages[i]); | |
565 | } | |
d397712b | 566 | kfree(pages); |
771ed689 CM |
567 | |
568 | goto out; | |
79787eaa JM |
569 | |
570 | cleanup_and_out: | |
571 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
572 | start, end, NULL, | |
573 | EXTENT_CLEAR_UNLOCK_PAGE | | |
574 | EXTENT_CLEAR_DIRTY | | |
575 | EXTENT_CLEAR_DELALLOC | | |
576 | EXTENT_SET_WRITEBACK | | |
577 | EXTENT_END_WRITEBACK); | |
578 | if (!trans || IS_ERR(trans)) | |
579 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
580 | else | |
581 | btrfs_abort_transaction(trans, root, ret); | |
582 | goto free_pages_out; | |
771ed689 CM |
583 | } |
584 | ||
585 | /* | |
586 | * phase two of compressed writeback. This is the ordered portion | |
587 | * of the code, which only gets called in the order the work was | |
588 | * queued. We walk all the async extents created by compress_file_range | |
589 | * and send them down to the disk. | |
590 | */ | |
591 | static noinline int submit_compressed_extents(struct inode *inode, | |
592 | struct async_cow *async_cow) | |
593 | { | |
594 | struct async_extent *async_extent; | |
595 | u64 alloc_hint = 0; | |
596 | struct btrfs_trans_handle *trans; | |
597 | struct btrfs_key ins; | |
598 | struct extent_map *em; | |
599 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
600 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
601 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 602 | int ret = 0; |
771ed689 CM |
603 | |
604 | if (list_empty(&async_cow->extents)) | |
605 | return 0; | |
606 | ||
771ed689 | 607 | |
d397712b | 608 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
609 | async_extent = list_entry(async_cow->extents.next, |
610 | struct async_extent, list); | |
611 | list_del(&async_extent->list); | |
c8b97818 | 612 | |
771ed689 CM |
613 | io_tree = &BTRFS_I(inode)->io_tree; |
614 | ||
f5a84ee3 | 615 | retry: |
771ed689 CM |
616 | /* did the compression code fall back to uncompressed IO? */ |
617 | if (!async_extent->pages) { | |
618 | int page_started = 0; | |
619 | unsigned long nr_written = 0; | |
620 | ||
621 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 622 | async_extent->start + |
d0082371 | 623 | async_extent->ram_size - 1); |
771ed689 CM |
624 | |
625 | /* allocate blocks */ | |
f5a84ee3 JB |
626 | ret = cow_file_range(inode, async_cow->locked_page, |
627 | async_extent->start, | |
628 | async_extent->start + | |
629 | async_extent->ram_size - 1, | |
630 | &page_started, &nr_written, 0); | |
771ed689 | 631 | |
79787eaa JM |
632 | /* JDM XXX */ |
633 | ||
771ed689 CM |
634 | /* |
635 | * if page_started, cow_file_range inserted an | |
636 | * inline extent and took care of all the unlocking | |
637 | * and IO for us. Otherwise, we need to submit | |
638 | * all those pages down to the drive. | |
639 | */ | |
f5a84ee3 | 640 | if (!page_started && !ret) |
771ed689 CM |
641 | extent_write_locked_range(io_tree, |
642 | inode, async_extent->start, | |
d397712b | 643 | async_extent->start + |
771ed689 CM |
644 | async_extent->ram_size - 1, |
645 | btrfs_get_extent, | |
646 | WB_SYNC_ALL); | |
647 | kfree(async_extent); | |
648 | cond_resched(); | |
649 | continue; | |
650 | } | |
651 | ||
652 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 653 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 654 | |
7a7eaa40 | 655 | trans = btrfs_join_transaction(root); |
79787eaa JM |
656 | if (IS_ERR(trans)) { |
657 | ret = PTR_ERR(trans); | |
658 | } else { | |
659 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
660 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
661 | async_extent->compressed_size, |
662 | async_extent->compressed_size, | |
81c9ad23 | 663 | 0, alloc_hint, &ins, 1); |
962197ba | 664 | if (ret && ret != -ENOSPC) |
79787eaa JM |
665 | btrfs_abort_transaction(trans, root, ret); |
666 | btrfs_end_transaction(trans, root); | |
667 | } | |
c2167754 | 668 | |
f5a84ee3 JB |
669 | if (ret) { |
670 | int i; | |
671 | for (i = 0; i < async_extent->nr_pages; i++) { | |
672 | WARN_ON(async_extent->pages[i]->mapping); | |
673 | page_cache_release(async_extent->pages[i]); | |
674 | } | |
675 | kfree(async_extent->pages); | |
676 | async_extent->nr_pages = 0; | |
677 | async_extent->pages = NULL; | |
678 | unlock_extent(io_tree, async_extent->start, | |
679 | async_extent->start + | |
d0082371 | 680 | async_extent->ram_size - 1); |
79787eaa JM |
681 | if (ret == -ENOSPC) |
682 | goto retry; | |
683 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
684 | } |
685 | ||
c2167754 YZ |
686 | /* |
687 | * here we're doing allocation and writeback of the | |
688 | * compressed pages | |
689 | */ | |
690 | btrfs_drop_extent_cache(inode, async_extent->start, | |
691 | async_extent->start + | |
692 | async_extent->ram_size - 1, 0); | |
693 | ||
172ddd60 | 694 | em = alloc_extent_map(); |
79787eaa | 695 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
696 | em->start = async_extent->start; |
697 | em->len = async_extent->ram_size; | |
445a6944 | 698 | em->orig_start = em->start; |
c8b97818 | 699 | |
771ed689 CM |
700 | em->block_start = ins.objectid; |
701 | em->block_len = ins.offset; | |
702 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 703 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
704 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
705 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
706 | ||
d397712b | 707 | while (1) { |
890871be | 708 | write_lock(&em_tree->lock); |
771ed689 | 709 | ret = add_extent_mapping(em_tree, em); |
890871be | 710 | write_unlock(&em_tree->lock); |
771ed689 CM |
711 | if (ret != -EEXIST) { |
712 | free_extent_map(em); | |
713 | break; | |
714 | } | |
715 | btrfs_drop_extent_cache(inode, async_extent->start, | |
716 | async_extent->start + | |
717 | async_extent->ram_size - 1, 0); | |
718 | } | |
719 | ||
261507a0 LZ |
720 | ret = btrfs_add_ordered_extent_compress(inode, |
721 | async_extent->start, | |
722 | ins.objectid, | |
723 | async_extent->ram_size, | |
724 | ins.offset, | |
725 | BTRFS_ORDERED_COMPRESSED, | |
726 | async_extent->compress_type); | |
79787eaa | 727 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 728 | |
771ed689 CM |
729 | /* |
730 | * clear dirty, set writeback and unlock the pages. | |
731 | */ | |
732 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
733 | &BTRFS_I(inode)->io_tree, |
734 | async_extent->start, | |
735 | async_extent->start + | |
736 | async_extent->ram_size - 1, | |
737 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
738 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 739 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 740 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
741 | |
742 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
743 | async_extent->start, |
744 | async_extent->ram_size, | |
745 | ins.objectid, | |
746 | ins.offset, async_extent->pages, | |
747 | async_extent->nr_pages); | |
771ed689 | 748 | |
79787eaa | 749 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
750 | alloc_hint = ins.objectid + ins.offset; |
751 | kfree(async_extent); | |
752 | cond_resched(); | |
753 | } | |
79787eaa JM |
754 | ret = 0; |
755 | out: | |
756 | return ret; | |
757 | out_free: | |
758 | kfree(async_extent); | |
759 | goto out; | |
771ed689 CM |
760 | } |
761 | ||
4b46fce2 JB |
762 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
763 | u64 num_bytes) | |
764 | { | |
765 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
766 | struct extent_map *em; | |
767 | u64 alloc_hint = 0; | |
768 | ||
769 | read_lock(&em_tree->lock); | |
770 | em = search_extent_mapping(em_tree, start, num_bytes); | |
771 | if (em) { | |
772 | /* | |
773 | * if block start isn't an actual block number then find the | |
774 | * first block in this inode and use that as a hint. If that | |
775 | * block is also bogus then just don't worry about it. | |
776 | */ | |
777 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
778 | free_extent_map(em); | |
779 | em = search_extent_mapping(em_tree, 0, 0); | |
780 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
781 | alloc_hint = em->block_start; | |
782 | if (em) | |
783 | free_extent_map(em); | |
784 | } else { | |
785 | alloc_hint = em->block_start; | |
786 | free_extent_map(em); | |
787 | } | |
788 | } | |
789 | read_unlock(&em_tree->lock); | |
790 | ||
791 | return alloc_hint; | |
792 | } | |
793 | ||
771ed689 CM |
794 | /* |
795 | * when extent_io.c finds a delayed allocation range in the file, | |
796 | * the call backs end up in this code. The basic idea is to | |
797 | * allocate extents on disk for the range, and create ordered data structs | |
798 | * in ram to track those extents. | |
799 | * | |
800 | * locked_page is the page that writepage had locked already. We use | |
801 | * it to make sure we don't do extra locks or unlocks. | |
802 | * | |
803 | * *page_started is set to one if we unlock locked_page and do everything | |
804 | * required to start IO on it. It may be clean and already done with | |
805 | * IO when we return. | |
806 | */ | |
b7d5b0a8 MX |
807 | static noinline int __cow_file_range(struct btrfs_trans_handle *trans, |
808 | struct inode *inode, | |
809 | struct btrfs_root *root, | |
810 | struct page *locked_page, | |
811 | u64 start, u64 end, int *page_started, | |
812 | unsigned long *nr_written, | |
813 | int unlock) | |
771ed689 | 814 | { |
771ed689 CM |
815 | u64 alloc_hint = 0; |
816 | u64 num_bytes; | |
817 | unsigned long ram_size; | |
818 | u64 disk_num_bytes; | |
819 | u64 cur_alloc_size; | |
820 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
821 | struct btrfs_key ins; |
822 | struct extent_map *em; | |
823 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
824 | int ret = 0; | |
825 | ||
83eea1f1 | 826 | BUG_ON(btrfs_is_free_space_inode(inode)); |
771ed689 | 827 | |
771ed689 CM |
828 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
829 | num_bytes = max(blocksize, num_bytes); | |
830 | disk_num_bytes = num_bytes; | |
771ed689 | 831 | |
4cb5300b | 832 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
833 | if (num_bytes < 64 * 1024 && |
834 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
835 | btrfs_add_inode_defrag(trans, inode); |
836 | ||
771ed689 CM |
837 | if (start == 0) { |
838 | /* lets try to make an inline extent */ | |
839 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 840 | start, end, 0, 0, NULL); |
771ed689 CM |
841 | if (ret == 0) { |
842 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
843 | &BTRFS_I(inode)->io_tree, |
844 | start, end, NULL, | |
845 | EXTENT_CLEAR_UNLOCK_PAGE | | |
846 | EXTENT_CLEAR_UNLOCK | | |
847 | EXTENT_CLEAR_DELALLOC | | |
848 | EXTENT_CLEAR_DIRTY | | |
849 | EXTENT_SET_WRITEBACK | | |
850 | EXTENT_END_WRITEBACK); | |
c2167754 | 851 | |
771ed689 CM |
852 | *nr_written = *nr_written + |
853 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
854 | *page_started = 1; | |
771ed689 | 855 | goto out; |
79787eaa JM |
856 | } else if (ret < 0) { |
857 | btrfs_abort_transaction(trans, root, ret); | |
858 | goto out_unlock; | |
771ed689 CM |
859 | } |
860 | } | |
861 | ||
862 | BUG_ON(disk_num_bytes > | |
6c41761f | 863 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 864 | |
4b46fce2 | 865 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
866 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
867 | ||
d397712b | 868 | while (disk_num_bytes > 0) { |
a791e35e CM |
869 | unsigned long op; |
870 | ||
287a0ab9 | 871 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 872 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 873 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 874 | &ins, 1); |
79787eaa JM |
875 | if (ret < 0) { |
876 | btrfs_abort_transaction(trans, root, ret); | |
877 | goto out_unlock; | |
878 | } | |
d397712b | 879 | |
172ddd60 | 880 | em = alloc_extent_map(); |
79787eaa | 881 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 882 | em->start = start; |
445a6944 | 883 | em->orig_start = em->start; |
771ed689 CM |
884 | ram_size = ins.offset; |
885 | em->len = ins.offset; | |
c8b97818 | 886 | |
e6dcd2dc | 887 | em->block_start = ins.objectid; |
c8b97818 | 888 | em->block_len = ins.offset; |
e6dcd2dc | 889 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 890 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 891 | |
d397712b | 892 | while (1) { |
890871be | 893 | write_lock(&em_tree->lock); |
e6dcd2dc | 894 | ret = add_extent_mapping(em_tree, em); |
890871be | 895 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
896 | if (ret != -EEXIST) { |
897 | free_extent_map(em); | |
898 | break; | |
899 | } | |
900 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 901 | start + ram_size - 1, 0); |
e6dcd2dc CM |
902 | } |
903 | ||
98d20f67 | 904 | cur_alloc_size = ins.offset; |
e6dcd2dc | 905 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 906 | ram_size, cur_alloc_size, 0); |
79787eaa | 907 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 908 | |
17d217fe YZ |
909 | if (root->root_key.objectid == |
910 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
911 | ret = btrfs_reloc_clone_csums(inode, start, | |
912 | cur_alloc_size); | |
79787eaa JM |
913 | if (ret) { |
914 | btrfs_abort_transaction(trans, root, ret); | |
915 | goto out_unlock; | |
916 | } | |
17d217fe YZ |
917 | } |
918 | ||
d397712b | 919 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 920 | break; |
d397712b | 921 | |
c8b97818 CM |
922 | /* we're not doing compressed IO, don't unlock the first |
923 | * page (which the caller expects to stay locked), don't | |
924 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
925 | * |
926 | * Do set the Private2 bit so we know this page was properly | |
927 | * setup for writepage | |
c8b97818 | 928 | */ |
a791e35e CM |
929 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
930 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
931 | EXTENT_SET_PRIVATE2; | |
932 | ||
c8b97818 CM |
933 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
934 | start, start + ram_size - 1, | |
a791e35e | 935 | locked_page, op); |
c8b97818 | 936 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
937 | num_bytes -= cur_alloc_size; |
938 | alloc_hint = ins.objectid + ins.offset; | |
939 | start += cur_alloc_size; | |
b888db2b | 940 | } |
79787eaa | 941 | out: |
be20aa9d | 942 | return ret; |
b7d5b0a8 | 943 | |
79787eaa JM |
944 | out_unlock: |
945 | extent_clear_unlock_delalloc(inode, | |
946 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 947 | start, end, locked_page, |
79787eaa JM |
948 | EXTENT_CLEAR_UNLOCK_PAGE | |
949 | EXTENT_CLEAR_UNLOCK | | |
950 | EXTENT_CLEAR_DELALLOC | | |
951 | EXTENT_CLEAR_DIRTY | | |
952 | EXTENT_SET_WRITEBACK | | |
953 | EXTENT_END_WRITEBACK); | |
954 | ||
955 | goto out; | |
771ed689 | 956 | } |
c8b97818 | 957 | |
b7d5b0a8 MX |
958 | static noinline int cow_file_range(struct inode *inode, |
959 | struct page *locked_page, | |
960 | u64 start, u64 end, int *page_started, | |
961 | unsigned long *nr_written, | |
962 | int unlock) | |
963 | { | |
964 | struct btrfs_trans_handle *trans; | |
965 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
966 | int ret; | |
967 | ||
968 | trans = btrfs_join_transaction(root); | |
969 | if (IS_ERR(trans)) { | |
970 | extent_clear_unlock_delalloc(inode, | |
971 | &BTRFS_I(inode)->io_tree, | |
972 | start, end, locked_page, | |
973 | EXTENT_CLEAR_UNLOCK_PAGE | | |
974 | EXTENT_CLEAR_UNLOCK | | |
975 | EXTENT_CLEAR_DELALLOC | | |
976 | EXTENT_CLEAR_DIRTY | | |
977 | EXTENT_SET_WRITEBACK | | |
978 | EXTENT_END_WRITEBACK); | |
979 | return PTR_ERR(trans); | |
980 | } | |
981 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
982 | ||
983 | ret = __cow_file_range(trans, inode, root, locked_page, start, end, | |
984 | page_started, nr_written, unlock); | |
985 | ||
986 | btrfs_end_transaction(trans, root); | |
987 | ||
988 | return ret; | |
989 | } | |
990 | ||
771ed689 CM |
991 | /* |
992 | * work queue call back to started compression on a file and pages | |
993 | */ | |
994 | static noinline void async_cow_start(struct btrfs_work *work) | |
995 | { | |
996 | struct async_cow *async_cow; | |
997 | int num_added = 0; | |
998 | async_cow = container_of(work, struct async_cow, work); | |
999 | ||
1000 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1001 | async_cow->start, async_cow->end, async_cow, | |
1002 | &num_added); | |
8180ef88 | 1003 | if (num_added == 0) { |
cb77fcd8 | 1004 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1005 | async_cow->inode = NULL; |
8180ef88 | 1006 | } |
771ed689 CM |
1007 | } |
1008 | ||
1009 | /* | |
1010 | * work queue call back to submit previously compressed pages | |
1011 | */ | |
1012 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1013 | { | |
1014 | struct async_cow *async_cow; | |
1015 | struct btrfs_root *root; | |
1016 | unsigned long nr_pages; | |
1017 | ||
1018 | async_cow = container_of(work, struct async_cow, work); | |
1019 | ||
1020 | root = async_cow->root; | |
1021 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1022 | PAGE_CACHE_SHIFT; | |
1023 | ||
66657b31 | 1024 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1025 | 5 * 1024 * 1024 && |
771ed689 CM |
1026 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1027 | wake_up(&root->fs_info->async_submit_wait); | |
1028 | ||
d397712b | 1029 | if (async_cow->inode) |
771ed689 | 1030 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1031 | } |
c8b97818 | 1032 | |
771ed689 CM |
1033 | static noinline void async_cow_free(struct btrfs_work *work) |
1034 | { | |
1035 | struct async_cow *async_cow; | |
1036 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1037 | if (async_cow->inode) |
cb77fcd8 | 1038 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1039 | kfree(async_cow); |
1040 | } | |
1041 | ||
1042 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1043 | u64 start, u64 end, int *page_started, | |
1044 | unsigned long *nr_written) | |
1045 | { | |
1046 | struct async_cow *async_cow; | |
1047 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1048 | unsigned long nr_pages; | |
1049 | u64 cur_end; | |
287082b0 | 1050 | int limit = 10 * 1024 * 1024; |
771ed689 | 1051 | |
a3429ab7 CM |
1052 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1053 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1054 | while (start < end) { |
771ed689 | 1055 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1056 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1057 | async_cow->inode = igrab(inode); |
771ed689 CM |
1058 | async_cow->root = root; |
1059 | async_cow->locked_page = locked_page; | |
1060 | async_cow->start = start; | |
1061 | ||
6cbff00f | 1062 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1063 | cur_end = end; |
1064 | else | |
1065 | cur_end = min(end, start + 512 * 1024 - 1); | |
1066 | ||
1067 | async_cow->end = cur_end; | |
1068 | INIT_LIST_HEAD(&async_cow->extents); | |
1069 | ||
1070 | async_cow->work.func = async_cow_start; | |
1071 | async_cow->work.ordered_func = async_cow_submit; | |
1072 | async_cow->work.ordered_free = async_cow_free; | |
1073 | async_cow->work.flags = 0; | |
1074 | ||
771ed689 CM |
1075 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1076 | PAGE_CACHE_SHIFT; | |
1077 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1078 | ||
1079 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1080 | &async_cow->work); | |
1081 | ||
1082 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1083 | wait_event(root->fs_info->async_submit_wait, | |
1084 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1085 | limit)); | |
1086 | } | |
1087 | ||
d397712b | 1088 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1089 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1090 | wait_event(root->fs_info->async_submit_wait, | |
1091 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1092 | 0)); | |
1093 | } | |
1094 | ||
1095 | *nr_written += nr_pages; | |
1096 | start = cur_end + 1; | |
1097 | } | |
1098 | *page_started = 1; | |
1099 | return 0; | |
be20aa9d CM |
1100 | } |
1101 | ||
d397712b | 1102 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1103 | u64 bytenr, u64 num_bytes) |
1104 | { | |
1105 | int ret; | |
1106 | struct btrfs_ordered_sum *sums; | |
1107 | LIST_HEAD(list); | |
1108 | ||
07d400a6 | 1109 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1110 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1111 | if (ret == 0 && list_empty(&list)) |
1112 | return 0; | |
1113 | ||
1114 | while (!list_empty(&list)) { | |
1115 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1116 | list_del(&sums->list); | |
1117 | kfree(sums); | |
1118 | } | |
1119 | return 1; | |
1120 | } | |
1121 | ||
d352ac68 CM |
1122 | /* |
1123 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1124 | * of the extents that exist in the file, and COWs the file as required. | |
1125 | * | |
1126 | * If no cow copies or snapshots exist, we write directly to the existing | |
1127 | * blocks on disk | |
1128 | */ | |
7f366cfe CM |
1129 | static noinline int run_delalloc_nocow(struct inode *inode, |
1130 | struct page *locked_page, | |
771ed689 CM |
1131 | u64 start, u64 end, int *page_started, int force, |
1132 | unsigned long *nr_written) | |
be20aa9d | 1133 | { |
be20aa9d | 1134 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1135 | struct btrfs_trans_handle *trans; |
be20aa9d | 1136 | struct extent_buffer *leaf; |
be20aa9d | 1137 | struct btrfs_path *path; |
80ff3856 | 1138 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1139 | struct btrfs_key found_key; |
80ff3856 YZ |
1140 | u64 cow_start; |
1141 | u64 cur_offset; | |
1142 | u64 extent_end; | |
5d4f98a2 | 1143 | u64 extent_offset; |
80ff3856 YZ |
1144 | u64 disk_bytenr; |
1145 | u64 num_bytes; | |
1146 | int extent_type; | |
79787eaa | 1147 | int ret, err; |
d899e052 | 1148 | int type; |
80ff3856 YZ |
1149 | int nocow; |
1150 | int check_prev = 1; | |
82d5902d | 1151 | bool nolock; |
33345d01 | 1152 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1153 | |
1154 | path = btrfs_alloc_path(); | |
17ca04af JB |
1155 | if (!path) { |
1156 | extent_clear_unlock_delalloc(inode, | |
1157 | &BTRFS_I(inode)->io_tree, | |
1158 | start, end, locked_page, | |
1159 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1160 | EXTENT_CLEAR_UNLOCK | | |
1161 | EXTENT_CLEAR_DELALLOC | | |
1162 | EXTENT_CLEAR_DIRTY | | |
1163 | EXTENT_SET_WRITEBACK | | |
1164 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1165 | return -ENOMEM; |
17ca04af | 1166 | } |
82d5902d | 1167 | |
83eea1f1 | 1168 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1169 | |
1170 | if (nolock) | |
7a7eaa40 | 1171 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1172 | else |
7a7eaa40 | 1173 | trans = btrfs_join_transaction(root); |
ff5714cc | 1174 | |
79787eaa | 1175 | if (IS_ERR(trans)) { |
17ca04af JB |
1176 | extent_clear_unlock_delalloc(inode, |
1177 | &BTRFS_I(inode)->io_tree, | |
1178 | start, end, locked_page, | |
1179 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1180 | EXTENT_CLEAR_UNLOCK | | |
1181 | EXTENT_CLEAR_DELALLOC | | |
1182 | EXTENT_CLEAR_DIRTY | | |
1183 | EXTENT_SET_WRITEBACK | | |
1184 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1185 | btrfs_free_path(path); |
1186 | return PTR_ERR(trans); | |
1187 | } | |
1188 | ||
74b21075 | 1189 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1190 | |
80ff3856 YZ |
1191 | cow_start = (u64)-1; |
1192 | cur_offset = start; | |
1193 | while (1) { | |
33345d01 | 1194 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1195 | cur_offset, 0); |
79787eaa JM |
1196 | if (ret < 0) { |
1197 | btrfs_abort_transaction(trans, root, ret); | |
1198 | goto error; | |
1199 | } | |
80ff3856 YZ |
1200 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1201 | leaf = path->nodes[0]; | |
1202 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1203 | path->slots[0] - 1); | |
33345d01 | 1204 | if (found_key.objectid == ino && |
80ff3856 YZ |
1205 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1206 | path->slots[0]--; | |
1207 | } | |
1208 | check_prev = 0; | |
1209 | next_slot: | |
1210 | leaf = path->nodes[0]; | |
1211 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1212 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1213 | if (ret < 0) { |
1214 | btrfs_abort_transaction(trans, root, ret); | |
1215 | goto error; | |
1216 | } | |
80ff3856 YZ |
1217 | if (ret > 0) |
1218 | break; | |
1219 | leaf = path->nodes[0]; | |
1220 | } | |
be20aa9d | 1221 | |
80ff3856 YZ |
1222 | nocow = 0; |
1223 | disk_bytenr = 0; | |
17d217fe | 1224 | num_bytes = 0; |
80ff3856 YZ |
1225 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1226 | ||
33345d01 | 1227 | if (found_key.objectid > ino || |
80ff3856 YZ |
1228 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1229 | found_key.offset > end) | |
1230 | break; | |
1231 | ||
1232 | if (found_key.offset > cur_offset) { | |
1233 | extent_end = found_key.offset; | |
e9061e21 | 1234 | extent_type = 0; |
80ff3856 YZ |
1235 | goto out_check; |
1236 | } | |
1237 | ||
1238 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1239 | struct btrfs_file_extent_item); | |
1240 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1241 | ||
d899e052 YZ |
1242 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1243 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1244 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1245 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1246 | extent_end = found_key.offset + |
1247 | btrfs_file_extent_num_bytes(leaf, fi); | |
1248 | if (extent_end <= start) { | |
1249 | path->slots[0]++; | |
1250 | goto next_slot; | |
1251 | } | |
17d217fe YZ |
1252 | if (disk_bytenr == 0) |
1253 | goto out_check; | |
80ff3856 YZ |
1254 | if (btrfs_file_extent_compression(leaf, fi) || |
1255 | btrfs_file_extent_encryption(leaf, fi) || | |
1256 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1257 | goto out_check; | |
d899e052 YZ |
1258 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1259 | goto out_check; | |
d2fb3437 | 1260 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1261 | goto out_check; |
33345d01 | 1262 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1263 | found_key.offset - |
1264 | extent_offset, disk_bytenr)) | |
17d217fe | 1265 | goto out_check; |
5d4f98a2 | 1266 | disk_bytenr += extent_offset; |
17d217fe YZ |
1267 | disk_bytenr += cur_offset - found_key.offset; |
1268 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1269 | /* | |
1270 | * force cow if csum exists in the range. | |
1271 | * this ensure that csum for a given extent are | |
1272 | * either valid or do not exist. | |
1273 | */ | |
1274 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1275 | goto out_check; | |
80ff3856 YZ |
1276 | nocow = 1; |
1277 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1278 | extent_end = found_key.offset + | |
1279 | btrfs_file_extent_inline_len(leaf, fi); | |
1280 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1281 | } else { | |
1282 | BUG_ON(1); | |
1283 | } | |
1284 | out_check: | |
1285 | if (extent_end <= start) { | |
1286 | path->slots[0]++; | |
1287 | goto next_slot; | |
1288 | } | |
1289 | if (!nocow) { | |
1290 | if (cow_start == (u64)-1) | |
1291 | cow_start = cur_offset; | |
1292 | cur_offset = extent_end; | |
1293 | if (cur_offset > end) | |
1294 | break; | |
1295 | path->slots[0]++; | |
1296 | goto next_slot; | |
7ea394f1 YZ |
1297 | } |
1298 | ||
b3b4aa74 | 1299 | btrfs_release_path(path); |
80ff3856 | 1300 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1301 | ret = __cow_file_range(trans, inode, root, locked_page, |
1302 | cow_start, found_key.offset - 1, | |
1303 | page_started, nr_written, 1); | |
79787eaa JM |
1304 | if (ret) { |
1305 | btrfs_abort_transaction(trans, root, ret); | |
1306 | goto error; | |
1307 | } | |
80ff3856 | 1308 | cow_start = (u64)-1; |
7ea394f1 | 1309 | } |
80ff3856 | 1310 | |
d899e052 YZ |
1311 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1312 | struct extent_map *em; | |
1313 | struct extent_map_tree *em_tree; | |
1314 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1315 | em = alloc_extent_map(); |
79787eaa | 1316 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1317 | em->start = cur_offset; |
445a6944 | 1318 | em->orig_start = em->start; |
d899e052 YZ |
1319 | em->len = num_bytes; |
1320 | em->block_len = num_bytes; | |
1321 | em->block_start = disk_bytenr; | |
1322 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1323 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
4e2f84e6 | 1324 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
d899e052 | 1325 | while (1) { |
890871be | 1326 | write_lock(&em_tree->lock); |
d899e052 | 1327 | ret = add_extent_mapping(em_tree, em); |
890871be | 1328 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1329 | if (ret != -EEXIST) { |
1330 | free_extent_map(em); | |
1331 | break; | |
1332 | } | |
1333 | btrfs_drop_extent_cache(inode, em->start, | |
1334 | em->start + em->len - 1, 0); | |
1335 | } | |
1336 | type = BTRFS_ORDERED_PREALLOC; | |
1337 | } else { | |
1338 | type = BTRFS_ORDERED_NOCOW; | |
1339 | } | |
80ff3856 YZ |
1340 | |
1341 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1342 | num_bytes, num_bytes, type); |
79787eaa | 1343 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1344 | |
efa56464 YZ |
1345 | if (root->root_key.objectid == |
1346 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1347 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1348 | num_bytes); | |
79787eaa JM |
1349 | if (ret) { |
1350 | btrfs_abort_transaction(trans, root, ret); | |
1351 | goto error; | |
1352 | } | |
efa56464 YZ |
1353 | } |
1354 | ||
d899e052 | 1355 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1356 | cur_offset, cur_offset + num_bytes - 1, |
1357 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1358 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1359 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1360 | cur_offset = extent_end; |
1361 | if (cur_offset > end) | |
1362 | break; | |
be20aa9d | 1363 | } |
b3b4aa74 | 1364 | btrfs_release_path(path); |
80ff3856 | 1365 | |
17ca04af | 1366 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1367 | cow_start = cur_offset; |
17ca04af JB |
1368 | cur_offset = end; |
1369 | } | |
1370 | ||
80ff3856 | 1371 | if (cow_start != (u64)-1) { |
b7d5b0a8 MX |
1372 | ret = __cow_file_range(trans, inode, root, locked_page, |
1373 | cow_start, end, | |
1374 | page_started, nr_written, 1); | |
79787eaa JM |
1375 | if (ret) { |
1376 | btrfs_abort_transaction(trans, root, ret); | |
1377 | goto error; | |
1378 | } | |
80ff3856 YZ |
1379 | } |
1380 | ||
79787eaa | 1381 | error: |
a698d075 | 1382 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1383 | if (!ret) |
1384 | ret = err; | |
1385 | ||
17ca04af JB |
1386 | if (ret && cur_offset < end) |
1387 | extent_clear_unlock_delalloc(inode, | |
1388 | &BTRFS_I(inode)->io_tree, | |
1389 | cur_offset, end, locked_page, | |
1390 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1391 | EXTENT_CLEAR_UNLOCK | | |
1392 | EXTENT_CLEAR_DELALLOC | | |
1393 | EXTENT_CLEAR_DIRTY | | |
1394 | EXTENT_SET_WRITEBACK | | |
1395 | EXTENT_END_WRITEBACK); | |
1396 | ||
7ea394f1 | 1397 | btrfs_free_path(path); |
79787eaa | 1398 | return ret; |
be20aa9d CM |
1399 | } |
1400 | ||
d352ac68 CM |
1401 | /* |
1402 | * extent_io.c call back to do delayed allocation processing | |
1403 | */ | |
c8b97818 | 1404 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1405 | u64 start, u64 end, int *page_started, |
1406 | unsigned long *nr_written) | |
be20aa9d | 1407 | { |
be20aa9d | 1408 | int ret; |
7f366cfe | 1409 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1410 | |
7ddf5a42 | 1411 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1412 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1413 | page_started, 1, nr_written); |
7ddf5a42 | 1414 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1415 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1416 | page_started, 0, nr_written); |
7ddf5a42 JB |
1417 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1418 | !(BTRFS_I(inode)->force_compress) && | |
1419 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1420 | ret = cow_file_range(inode, locked_page, start, end, |
1421 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1422 | } else { |
1423 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1424 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1425 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1426 | page_started, nr_written); |
7ddf5a42 | 1427 | } |
b888db2b CM |
1428 | return ret; |
1429 | } | |
1430 | ||
1bf85046 JM |
1431 | static void btrfs_split_extent_hook(struct inode *inode, |
1432 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1433 | { |
0ca1f7ce | 1434 | /* not delalloc, ignore it */ |
9ed74f2d | 1435 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1436 | return; |
9ed74f2d | 1437 | |
9e0baf60 JB |
1438 | spin_lock(&BTRFS_I(inode)->lock); |
1439 | BTRFS_I(inode)->outstanding_extents++; | |
1440 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1441 | } |
1442 | ||
1443 | /* | |
1444 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1445 | * extents so we can keep track of new extents that are just merged onto old | |
1446 | * extents, such as when we are doing sequential writes, so we can properly | |
1447 | * account for the metadata space we'll need. | |
1448 | */ | |
1bf85046 JM |
1449 | static void btrfs_merge_extent_hook(struct inode *inode, |
1450 | struct extent_state *new, | |
1451 | struct extent_state *other) | |
9ed74f2d | 1452 | { |
9ed74f2d JB |
1453 | /* not delalloc, ignore it */ |
1454 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1455 | return; |
9ed74f2d | 1456 | |
9e0baf60 JB |
1457 | spin_lock(&BTRFS_I(inode)->lock); |
1458 | BTRFS_I(inode)->outstanding_extents--; | |
1459 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1460 | } |
1461 | ||
d352ac68 CM |
1462 | /* |
1463 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1464 | * bytes in this file, and to maintain the list of inodes that | |
1465 | * have pending delalloc work to be done. | |
1466 | */ | |
1bf85046 JM |
1467 | static void btrfs_set_bit_hook(struct inode *inode, |
1468 | struct extent_state *state, int *bits) | |
291d673e | 1469 | { |
9ed74f2d | 1470 | |
75eff68e CM |
1471 | /* |
1472 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1473 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1474 | * bit, which is only set or cleared with irqs on |
1475 | */ | |
0ca1f7ce | 1476 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1477 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1478 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1479 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1480 | |
9e0baf60 | 1481 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1482 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1483 | } else { |
1484 | spin_lock(&BTRFS_I(inode)->lock); | |
1485 | BTRFS_I(inode)->outstanding_extents++; | |
1486 | spin_unlock(&BTRFS_I(inode)->lock); | |
1487 | } | |
287a0ab9 | 1488 | |
75eff68e | 1489 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1490 | BTRFS_I(inode)->delalloc_bytes += len; |
1491 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1492 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1493 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1494 | &root->fs_info->delalloc_inodes); | |
1495 | } | |
75eff68e | 1496 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1497 | } |
291d673e CM |
1498 | } |
1499 | ||
d352ac68 CM |
1500 | /* |
1501 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1502 | */ | |
1bf85046 JM |
1503 | static void btrfs_clear_bit_hook(struct inode *inode, |
1504 | struct extent_state *state, int *bits) | |
291d673e | 1505 | { |
75eff68e CM |
1506 | /* |
1507 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1508 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1509 | * bit, which is only set or cleared with irqs on |
1510 | */ | |
0ca1f7ce | 1511 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1512 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1513 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1514 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1515 | |
9e0baf60 | 1516 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1517 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1518 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1519 | spin_lock(&BTRFS_I(inode)->lock); | |
1520 | BTRFS_I(inode)->outstanding_extents--; | |
1521 | spin_unlock(&BTRFS_I(inode)->lock); | |
1522 | } | |
0ca1f7ce YZ |
1523 | |
1524 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1525 | btrfs_delalloc_release_metadata(inode, len); | |
1526 | ||
0cb59c99 JB |
1527 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1528 | && do_list) | |
0ca1f7ce | 1529 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1530 | |
75eff68e | 1531 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1532 | root->fs_info->delalloc_bytes -= len; |
1533 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1534 | ||
0cb59c99 | 1535 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1536 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1537 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1538 | } | |
75eff68e | 1539 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1540 | } |
291d673e CM |
1541 | } |
1542 | ||
d352ac68 CM |
1543 | /* |
1544 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1545 | * we don't create bios that span stripes or chunks | |
1546 | */ | |
239b14b3 | 1547 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1548 | size_t size, struct bio *bio, |
1549 | unsigned long bio_flags) | |
239b14b3 CM |
1550 | { |
1551 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
a62b9401 | 1552 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1553 | u64 length = 0; |
1554 | u64 map_length; | |
239b14b3 CM |
1555 | int ret; |
1556 | ||
771ed689 CM |
1557 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1558 | return 0; | |
1559 | ||
f2d8d74d | 1560 | length = bio->bi_size; |
239b14b3 | 1561 | map_length = length; |
3ec706c8 | 1562 | ret = btrfs_map_block(root->fs_info, READ, logical, |
f188591e | 1563 | &map_length, NULL, 0); |
3ec706c8 | 1564 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1565 | BUG_ON(ret < 0); |
d397712b | 1566 | if (map_length < length + size) |
239b14b3 | 1567 | return 1; |
3444a972 | 1568 | return 0; |
239b14b3 CM |
1569 | } |
1570 | ||
d352ac68 CM |
1571 | /* |
1572 | * in order to insert checksums into the metadata in large chunks, | |
1573 | * we wait until bio submission time. All the pages in the bio are | |
1574 | * checksummed and sums are attached onto the ordered extent record. | |
1575 | * | |
1576 | * At IO completion time the cums attached on the ordered extent record | |
1577 | * are inserted into the btree | |
1578 | */ | |
d397712b CM |
1579 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1580 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1581 | unsigned long bio_flags, |
1582 | u64 bio_offset) | |
065631f6 | 1583 | { |
065631f6 | 1584 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1585 | int ret = 0; |
e015640f | 1586 | |
d20f7043 | 1587 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1588 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1589 | return 0; |
1590 | } | |
e015640f | 1591 | |
4a69a410 CM |
1592 | /* |
1593 | * in order to insert checksums into the metadata in large chunks, | |
1594 | * we wait until bio submission time. All the pages in the bio are | |
1595 | * checksummed and sums are attached onto the ordered extent record. | |
1596 | * | |
1597 | * At IO completion time the cums attached on the ordered extent record | |
1598 | * are inserted into the btree | |
1599 | */ | |
b2950863 | 1600 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1601 | int mirror_num, unsigned long bio_flags, |
1602 | u64 bio_offset) | |
4a69a410 CM |
1603 | { |
1604 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1605 | int ret; |
1606 | ||
1607 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
1608 | if (ret) | |
1609 | bio_endio(bio, ret); | |
1610 | return ret; | |
44b8bd7e CM |
1611 | } |
1612 | ||
d352ac68 | 1613 | /* |
cad321ad CM |
1614 | * extent_io.c submission hook. This does the right thing for csum calculation |
1615 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1616 | */ |
b2950863 | 1617 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1618 | int mirror_num, unsigned long bio_flags, |
1619 | u64 bio_offset) | |
44b8bd7e CM |
1620 | { |
1621 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1622 | int ret = 0; | |
19b9bdb0 | 1623 | int skip_sum; |
0417341e | 1624 | int metadata = 0; |
44b8bd7e | 1625 | |
6cbff00f | 1626 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1627 | |
83eea1f1 | 1628 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1629 | metadata = 2; |
1630 | ||
7b6d91da | 1631 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1632 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1633 | if (ret) | |
61891923 | 1634 | goto out; |
5fd02043 | 1635 | |
d20f7043 | 1636 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1637 | ret = btrfs_submit_compressed_read(inode, bio, |
1638 | mirror_num, | |
1639 | bio_flags); | |
1640 | goto out; | |
c2db1073 TI |
1641 | } else if (!skip_sum) { |
1642 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1643 | if (ret) | |
61891923 | 1644 | goto out; |
c2db1073 | 1645 | } |
4d1b5fb4 | 1646 | goto mapit; |
19b9bdb0 | 1647 | } else if (!skip_sum) { |
17d217fe YZ |
1648 | /* csum items have already been cloned */ |
1649 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1650 | goto mapit; | |
19b9bdb0 | 1651 | /* we're doing a write, do the async checksumming */ |
61891923 | 1652 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1653 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1654 | bio_flags, bio_offset, |
1655 | __btrfs_submit_bio_start, | |
4a69a410 | 1656 | __btrfs_submit_bio_done); |
61891923 | 1657 | goto out; |
19b9bdb0 CM |
1658 | } |
1659 | ||
0b86a832 | 1660 | mapit: |
61891923 SB |
1661 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1662 | ||
1663 | out: | |
1664 | if (ret < 0) | |
1665 | bio_endio(bio, ret); | |
1666 | return ret; | |
065631f6 | 1667 | } |
6885f308 | 1668 | |
d352ac68 CM |
1669 | /* |
1670 | * given a list of ordered sums record them in the inode. This happens | |
1671 | * at IO completion time based on sums calculated at bio submission time. | |
1672 | */ | |
ba1da2f4 | 1673 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1674 | struct inode *inode, u64 file_offset, |
1675 | struct list_head *list) | |
1676 | { | |
e6dcd2dc CM |
1677 | struct btrfs_ordered_sum *sum; |
1678 | ||
c6e30871 | 1679 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1680 | btrfs_csum_file_blocks(trans, |
1681 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1682 | } |
1683 | return 0; | |
1684 | } | |
1685 | ||
2ac55d41 JB |
1686 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1687 | struct extent_state **cached_state) | |
ea8c2819 | 1688 | { |
6c1500f2 | 1689 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1690 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1691 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1692 | } |
1693 | ||
d352ac68 | 1694 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1695 | struct btrfs_writepage_fixup { |
1696 | struct page *page; | |
1697 | struct btrfs_work work; | |
1698 | }; | |
1699 | ||
b2950863 | 1700 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1701 | { |
1702 | struct btrfs_writepage_fixup *fixup; | |
1703 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1704 | struct extent_state *cached_state = NULL; |
247e743c CM |
1705 | struct page *page; |
1706 | struct inode *inode; | |
1707 | u64 page_start; | |
1708 | u64 page_end; | |
87826df0 | 1709 | int ret; |
247e743c CM |
1710 | |
1711 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1712 | page = fixup->page; | |
4a096752 | 1713 | again: |
247e743c CM |
1714 | lock_page(page); |
1715 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1716 | ClearPageChecked(page); | |
1717 | goto out_page; | |
1718 | } | |
1719 | ||
1720 | inode = page->mapping->host; | |
1721 | page_start = page_offset(page); | |
1722 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1723 | ||
2ac55d41 | 1724 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1725 | &cached_state); |
4a096752 CM |
1726 | |
1727 | /* already ordered? We're done */ | |
8b62b72b | 1728 | if (PagePrivate2(page)) |
247e743c | 1729 | goto out; |
4a096752 CM |
1730 | |
1731 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1732 | if (ordered) { | |
2ac55d41 JB |
1733 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1734 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1735 | unlock_page(page); |
1736 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1737 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1738 | goto again; |
1739 | } | |
247e743c | 1740 | |
87826df0 JM |
1741 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1742 | if (ret) { | |
1743 | mapping_set_error(page->mapping, ret); | |
1744 | end_extent_writepage(page, ret, page_start, page_end); | |
1745 | ClearPageChecked(page); | |
1746 | goto out; | |
1747 | } | |
1748 | ||
2ac55d41 | 1749 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1750 | ClearPageChecked(page); |
87826df0 | 1751 | set_page_dirty(page); |
247e743c | 1752 | out: |
2ac55d41 JB |
1753 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1754 | &cached_state, GFP_NOFS); | |
247e743c CM |
1755 | out_page: |
1756 | unlock_page(page); | |
1757 | page_cache_release(page); | |
b897abec | 1758 | kfree(fixup); |
247e743c CM |
1759 | } |
1760 | ||
1761 | /* | |
1762 | * There are a few paths in the higher layers of the kernel that directly | |
1763 | * set the page dirty bit without asking the filesystem if it is a | |
1764 | * good idea. This causes problems because we want to make sure COW | |
1765 | * properly happens and the data=ordered rules are followed. | |
1766 | * | |
c8b97818 | 1767 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1768 | * hasn't been properly setup for IO. We kick off an async process |
1769 | * to fix it up. The async helper will wait for ordered extents, set | |
1770 | * the delalloc bit and make it safe to write the page. | |
1771 | */ | |
b2950863 | 1772 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1773 | { |
1774 | struct inode *inode = page->mapping->host; | |
1775 | struct btrfs_writepage_fixup *fixup; | |
1776 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1777 | |
8b62b72b CM |
1778 | /* this page is properly in the ordered list */ |
1779 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1780 | return 0; |
1781 | ||
1782 | if (PageChecked(page)) | |
1783 | return -EAGAIN; | |
1784 | ||
1785 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1786 | if (!fixup) | |
1787 | return -EAGAIN; | |
f421950f | 1788 | |
247e743c CM |
1789 | SetPageChecked(page); |
1790 | page_cache_get(page); | |
1791 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1792 | fixup->page = page; | |
1793 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1794 | return -EBUSY; |
247e743c CM |
1795 | } |
1796 | ||
d899e052 YZ |
1797 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1798 | struct inode *inode, u64 file_pos, | |
1799 | u64 disk_bytenr, u64 disk_num_bytes, | |
1800 | u64 num_bytes, u64 ram_bytes, | |
1801 | u8 compression, u8 encryption, | |
1802 | u16 other_encoding, int extent_type) | |
1803 | { | |
1804 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1805 | struct btrfs_file_extent_item *fi; | |
1806 | struct btrfs_path *path; | |
1807 | struct extent_buffer *leaf; | |
1808 | struct btrfs_key ins; | |
d899e052 YZ |
1809 | int ret; |
1810 | ||
1811 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1812 | if (!path) |
1813 | return -ENOMEM; | |
d899e052 | 1814 | |
b9473439 | 1815 | path->leave_spinning = 1; |
a1ed835e CM |
1816 | |
1817 | /* | |
1818 | * we may be replacing one extent in the tree with another. | |
1819 | * The new extent is pinned in the extent map, and we don't want | |
1820 | * to drop it from the cache until it is completely in the btree. | |
1821 | * | |
1822 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1823 | * the caller is expected to unpin it and allow it to be merged | |
1824 | * with the others. | |
1825 | */ | |
5dc562c5 | 1826 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1827 | file_pos + num_bytes, 0); |
79787eaa JM |
1828 | if (ret) |
1829 | goto out; | |
d899e052 | 1830 | |
33345d01 | 1831 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1832 | ins.offset = file_pos; |
1833 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1834 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1835 | if (ret) |
1836 | goto out; | |
d899e052 YZ |
1837 | leaf = path->nodes[0]; |
1838 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1839 | struct btrfs_file_extent_item); | |
1840 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1841 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1842 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1843 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1844 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1845 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1846 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1847 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1848 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1849 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 1850 | |
d899e052 | 1851 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 1852 | btrfs_release_path(path); |
d899e052 YZ |
1853 | |
1854 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1855 | |
1856 | ins.objectid = disk_bytenr; | |
1857 | ins.offset = disk_num_bytes; | |
1858 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1859 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1860 | root->root_key.objectid, | |
33345d01 | 1861 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1862 | out: |
d899e052 | 1863 | btrfs_free_path(path); |
b9473439 | 1864 | |
79787eaa | 1865 | return ret; |
d899e052 YZ |
1866 | } |
1867 | ||
5d13a98f CM |
1868 | /* |
1869 | * helper function for btrfs_finish_ordered_io, this | |
1870 | * just reads in some of the csum leaves to prime them into ram | |
1871 | * before we start the transaction. It limits the amount of btree | |
1872 | * reads required while inside the transaction. | |
1873 | */ | |
d352ac68 CM |
1874 | /* as ordered data IO finishes, this gets called so we can finish |
1875 | * an ordered extent if the range of bytes in the file it covers are | |
1876 | * fully written. | |
1877 | */ | |
5fd02043 | 1878 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1879 | { |
5fd02043 | 1880 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1881 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1882 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1883 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1884 | struct extent_state *cached_state = NULL; |
261507a0 | 1885 | int compress_type = 0; |
e6dcd2dc | 1886 | int ret; |
82d5902d | 1887 | bool nolock; |
e6dcd2dc | 1888 | |
83eea1f1 | 1889 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 1890 | |
5fd02043 JB |
1891 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1892 | ret = -EIO; | |
1893 | goto out; | |
1894 | } | |
1895 | ||
c2167754 | 1896 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1897 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
c2167754 YZ |
1898 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1899 | if (!ret) { | |
0cb59c99 | 1900 | if (nolock) |
7a7eaa40 | 1901 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1902 | else |
7a7eaa40 | 1903 | trans = btrfs_join_transaction(root); |
d280e5be LB |
1904 | if (IS_ERR(trans)) { |
1905 | ret = PTR_ERR(trans); | |
1906 | trans = NULL; | |
1907 | goto out; | |
1908 | } | |
0ca1f7ce | 1909 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1910 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1911 | if (ret) /* -ENOMEM or corruption */ |
1912 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1913 | } |
1914 | goto out; | |
1915 | } | |
e6dcd2dc | 1916 | |
2ac55d41 JB |
1917 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1918 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1919 | 0, &cached_state); |
e6dcd2dc | 1920 | |
0cb59c99 | 1921 | if (nolock) |
7a7eaa40 | 1922 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1923 | else |
7a7eaa40 | 1924 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1925 | if (IS_ERR(trans)) { |
1926 | ret = PTR_ERR(trans); | |
1927 | trans = NULL; | |
1928 | goto out_unlock; | |
1929 | } | |
0ca1f7ce | 1930 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1931 | |
c8b97818 | 1932 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1933 | compress_type = ordered_extent->compress_type; |
d899e052 | 1934 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1935 | BUG_ON(compress_type); |
920bbbfb | 1936 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1937 | ordered_extent->file_offset, |
1938 | ordered_extent->file_offset + | |
1939 | ordered_extent->len); | |
d899e052 | 1940 | } else { |
0af3d00b | 1941 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1942 | ret = insert_reserved_file_extent(trans, inode, |
1943 | ordered_extent->file_offset, | |
1944 | ordered_extent->start, | |
1945 | ordered_extent->disk_len, | |
1946 | ordered_extent->len, | |
1947 | ordered_extent->len, | |
261507a0 | 1948 | compress_type, 0, 0, |
d899e052 | 1949 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 1950 | } |
5dc562c5 JB |
1951 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1952 | ordered_extent->file_offset, ordered_extent->len, | |
1953 | trans->transid); | |
79787eaa JM |
1954 | if (ret < 0) { |
1955 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1956 | goto out_unlock; |
79787eaa | 1957 | } |
2ac55d41 | 1958 | |
e6dcd2dc CM |
1959 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1960 | &ordered_extent->list); | |
1961 | ||
1ef30be1 | 1962 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1963 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1964 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1965 | if (ret) { /* -ENOMEM or corruption */ |
1966 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1967 | goto out_unlock; |
79787eaa | 1968 | } |
7c735313 JB |
1969 | } else { |
1970 | btrfs_set_inode_last_trans(trans, inode); | |
1ef30be1 JB |
1971 | } |
1972 | ret = 0; | |
5fd02043 JB |
1973 | out_unlock: |
1974 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
1975 | ordered_extent->file_offset + | |
1976 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 1977 | out: |
5b0e95bf | 1978 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1979 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
1980 | if (trans) |
1981 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 1982 | |
5fd02043 JB |
1983 | if (ret) |
1984 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, | |
1985 | ordered_extent->file_offset + | |
1986 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
1987 | ||
1988 | /* | |
8bad3c02 LB |
1989 | * This needs to be done to make sure anybody waiting knows we are done |
1990 | * updating everything for this ordered extent. | |
5fd02043 JB |
1991 | */ |
1992 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
1993 | ||
e6dcd2dc CM |
1994 | /* once for us */ |
1995 | btrfs_put_ordered_extent(ordered_extent); | |
1996 | /* once for the tree */ | |
1997 | btrfs_put_ordered_extent(ordered_extent); | |
1998 | ||
5fd02043 JB |
1999 | return ret; |
2000 | } | |
2001 | ||
2002 | static void finish_ordered_fn(struct btrfs_work *work) | |
2003 | { | |
2004 | struct btrfs_ordered_extent *ordered_extent; | |
2005 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2006 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2007 | } |
2008 | ||
b2950863 | 2009 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2010 | struct extent_state *state, int uptodate) |
2011 | { | |
5fd02043 JB |
2012 | struct inode *inode = page->mapping->host; |
2013 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2014 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
2015 | struct btrfs_workers *workers; | |
2016 | ||
1abe9b8a | 2017 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2018 | ||
8b62b72b | 2019 | ClearPagePrivate2(page); |
5fd02043 JB |
2020 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2021 | end - start + 1, uptodate)) | |
2022 | return 0; | |
2023 | ||
2024 | ordered_extent->work.func = finish_ordered_fn; | |
2025 | ordered_extent->work.flags = 0; | |
2026 | ||
83eea1f1 | 2027 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2028 | workers = &root->fs_info->endio_freespace_worker; |
2029 | else | |
2030 | workers = &root->fs_info->endio_write_workers; | |
2031 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2032 | ||
2033 | return 0; | |
211f90e6 CM |
2034 | } |
2035 | ||
d352ac68 CM |
2036 | /* |
2037 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2038 | * if there's a match, we allow the bio to finish. If not, the code in |
2039 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2040 | */ |
b2950863 | 2041 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2042 | struct extent_state *state, int mirror) |
07157aac | 2043 | { |
35ebb934 | 2044 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 2045 | struct inode *inode = page->mapping->host; |
d1310b2e | 2046 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2047 | char *kaddr; |
aadfeb6e | 2048 | u64 private = ~(u32)0; |
07157aac | 2049 | int ret; |
ff79f819 CM |
2050 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2051 | u32 csum = ~(u32)0; | |
d1310b2e | 2052 | |
d20f7043 CM |
2053 | if (PageChecked(page)) { |
2054 | ClearPageChecked(page); | |
2055 | goto good; | |
2056 | } | |
6cbff00f CH |
2057 | |
2058 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2059 | goto good; |
17d217fe YZ |
2060 | |
2061 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2062 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2063 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2064 | GFP_NOFS); | |
b6cda9bc | 2065 | return 0; |
17d217fe | 2066 | } |
d20f7043 | 2067 | |
c2e639f0 | 2068 | if (state && state->start == start) { |
70dec807 CM |
2069 | private = state->private; |
2070 | ret = 0; | |
2071 | } else { | |
2072 | ret = get_state_private(io_tree, start, &private); | |
2073 | } | |
7ac687d9 | 2074 | kaddr = kmap_atomic(page); |
d397712b | 2075 | if (ret) |
07157aac | 2076 | goto zeroit; |
d397712b | 2077 | |
ff79f819 CM |
2078 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2079 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2080 | if (csum != private) |
07157aac | 2081 | goto zeroit; |
d397712b | 2082 | |
7ac687d9 | 2083 | kunmap_atomic(kaddr); |
d20f7043 | 2084 | good: |
07157aac CM |
2085 | return 0; |
2086 | ||
2087 | zeroit: | |
945d8962 | 2088 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2089 | "private %llu\n", |
2090 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2091 | (unsigned long long)start, csum, |
2092 | (unsigned long long)private); | |
db94535d CM |
2093 | memset(kaddr + offset, 1, end - start + 1); |
2094 | flush_dcache_page(page); | |
7ac687d9 | 2095 | kunmap_atomic(kaddr); |
3b951516 CM |
2096 | if (private == 0) |
2097 | return 0; | |
7e38326f | 2098 | return -EIO; |
07157aac | 2099 | } |
b888db2b | 2100 | |
24bbcf04 YZ |
2101 | struct delayed_iput { |
2102 | struct list_head list; | |
2103 | struct inode *inode; | |
2104 | }; | |
2105 | ||
79787eaa JM |
2106 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2107 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2108 | void btrfs_add_delayed_iput(struct inode *inode) |
2109 | { | |
2110 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2111 | struct delayed_iput *delayed; | |
2112 | ||
2113 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2114 | return; | |
2115 | ||
2116 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2117 | delayed->inode = inode; | |
2118 | ||
2119 | spin_lock(&fs_info->delayed_iput_lock); | |
2120 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2121 | spin_unlock(&fs_info->delayed_iput_lock); | |
2122 | } | |
2123 | ||
2124 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2125 | { | |
2126 | LIST_HEAD(list); | |
2127 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2128 | struct delayed_iput *delayed; | |
2129 | int empty; | |
2130 | ||
2131 | spin_lock(&fs_info->delayed_iput_lock); | |
2132 | empty = list_empty(&fs_info->delayed_iputs); | |
2133 | spin_unlock(&fs_info->delayed_iput_lock); | |
2134 | if (empty) | |
2135 | return; | |
2136 | ||
24bbcf04 YZ |
2137 | spin_lock(&fs_info->delayed_iput_lock); |
2138 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2139 | spin_unlock(&fs_info->delayed_iput_lock); | |
2140 | ||
2141 | while (!list_empty(&list)) { | |
2142 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2143 | list_del(&delayed->list); | |
2144 | iput(delayed->inode); | |
2145 | kfree(delayed); | |
2146 | } | |
24bbcf04 YZ |
2147 | } |
2148 | ||
d68fc57b YZ |
2149 | enum btrfs_orphan_cleanup_state { |
2150 | ORPHAN_CLEANUP_STARTED = 1, | |
2151 | ORPHAN_CLEANUP_DONE = 2, | |
2152 | }; | |
2153 | ||
2154 | /* | |
42b2aa86 | 2155 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2156 | * files in the subvolume, it removes orphan item and frees block_rsv |
2157 | * structure. | |
2158 | */ | |
2159 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2160 | struct btrfs_root *root) | |
2161 | { | |
90290e19 | 2162 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2163 | int ret; |
2164 | ||
8a35d95f | 2165 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2166 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2167 | return; | |
2168 | ||
90290e19 | 2169 | spin_lock(&root->orphan_lock); |
8a35d95f | 2170 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2171 | spin_unlock(&root->orphan_lock); |
2172 | return; | |
2173 | } | |
2174 | ||
2175 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2176 | spin_unlock(&root->orphan_lock); | |
2177 | return; | |
2178 | } | |
2179 | ||
2180 | block_rsv = root->orphan_block_rsv; | |
2181 | root->orphan_block_rsv = NULL; | |
2182 | spin_unlock(&root->orphan_lock); | |
2183 | ||
d68fc57b YZ |
2184 | if (root->orphan_item_inserted && |
2185 | btrfs_root_refs(&root->root_item) > 0) { | |
2186 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2187 | root->root_key.objectid); | |
2188 | BUG_ON(ret); | |
2189 | root->orphan_item_inserted = 0; | |
2190 | } | |
2191 | ||
90290e19 JB |
2192 | if (block_rsv) { |
2193 | WARN_ON(block_rsv->size > 0); | |
2194 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2195 | } |
2196 | } | |
2197 | ||
7b128766 JB |
2198 | /* |
2199 | * This creates an orphan entry for the given inode in case something goes | |
2200 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2201 | * |
2202 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2203 | * this function. | |
7b128766 JB |
2204 | */ |
2205 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2206 | { | |
2207 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2208 | struct btrfs_block_rsv *block_rsv = NULL; |
2209 | int reserve = 0; | |
2210 | int insert = 0; | |
2211 | int ret; | |
7b128766 | 2212 | |
d68fc57b | 2213 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2214 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2215 | if (!block_rsv) |
2216 | return -ENOMEM; | |
d68fc57b | 2217 | } |
7b128766 | 2218 | |
d68fc57b YZ |
2219 | spin_lock(&root->orphan_lock); |
2220 | if (!root->orphan_block_rsv) { | |
2221 | root->orphan_block_rsv = block_rsv; | |
2222 | } else if (block_rsv) { | |
2223 | btrfs_free_block_rsv(root, block_rsv); | |
2224 | block_rsv = NULL; | |
7b128766 | 2225 | } |
7b128766 | 2226 | |
8a35d95f JB |
2227 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2228 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2229 | #if 0 |
2230 | /* | |
2231 | * For proper ENOSPC handling, we should do orphan | |
2232 | * cleanup when mounting. But this introduces backward | |
2233 | * compatibility issue. | |
2234 | */ | |
2235 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2236 | insert = 2; | |
2237 | else | |
2238 | insert = 1; | |
2239 | #endif | |
2240 | insert = 1; | |
321f0e70 | 2241 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2242 | } |
2243 | ||
72ac3c0d JB |
2244 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2245 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2246 | reserve = 1; |
d68fc57b | 2247 | spin_unlock(&root->orphan_lock); |
7b128766 | 2248 | |
d68fc57b YZ |
2249 | /* grab metadata reservation from transaction handle */ |
2250 | if (reserve) { | |
2251 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2252 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2253 | } |
7b128766 | 2254 | |
d68fc57b YZ |
2255 | /* insert an orphan item to track this unlinked/truncated file */ |
2256 | if (insert >= 1) { | |
33345d01 | 2257 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2258 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2259 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2260 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2261 | btrfs_abort_transaction(trans, root, ret); |
2262 | return ret; | |
2263 | } | |
2264 | ret = 0; | |
d68fc57b YZ |
2265 | } |
2266 | ||
2267 | /* insert an orphan item to track subvolume contains orphan files */ | |
2268 | if (insert >= 2) { | |
2269 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2270 | root->root_key.objectid); | |
79787eaa JM |
2271 | if (ret && ret != -EEXIST) { |
2272 | btrfs_abort_transaction(trans, root, ret); | |
2273 | return ret; | |
2274 | } | |
d68fc57b YZ |
2275 | } |
2276 | return 0; | |
7b128766 JB |
2277 | } |
2278 | ||
2279 | /* | |
2280 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2281 | * item for this particular inode. | |
2282 | */ | |
2283 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2284 | { | |
2285 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2286 | int delete_item = 0; |
2287 | int release_rsv = 0; | |
7b128766 JB |
2288 | int ret = 0; |
2289 | ||
d68fc57b | 2290 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2291 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2292 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2293 | delete_item = 1; |
7b128766 | 2294 | |
72ac3c0d JB |
2295 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2296 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2297 | release_rsv = 1; |
d68fc57b | 2298 | spin_unlock(&root->orphan_lock); |
7b128766 | 2299 | |
d68fc57b | 2300 | if (trans && delete_item) { |
33345d01 | 2301 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2302 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2303 | } |
7b128766 | 2304 | |
8a35d95f | 2305 | if (release_rsv) { |
d68fc57b | 2306 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2307 | atomic_dec(&root->orphan_inodes); |
2308 | } | |
7b128766 | 2309 | |
d68fc57b | 2310 | return 0; |
7b128766 JB |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * this cleans up any orphans that may be left on the list from the last use | |
2315 | * of this root. | |
2316 | */ | |
66b4ffd1 | 2317 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2318 | { |
2319 | struct btrfs_path *path; | |
2320 | struct extent_buffer *leaf; | |
7b128766 JB |
2321 | struct btrfs_key key, found_key; |
2322 | struct btrfs_trans_handle *trans; | |
2323 | struct inode *inode; | |
8f6d7f4f | 2324 | u64 last_objectid = 0; |
7b128766 JB |
2325 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2326 | ||
d68fc57b | 2327 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2328 | return 0; |
c71bf099 YZ |
2329 | |
2330 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2331 | if (!path) { |
2332 | ret = -ENOMEM; | |
2333 | goto out; | |
2334 | } | |
7b128766 JB |
2335 | path->reada = -1; |
2336 | ||
2337 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2338 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2339 | key.offset = (u64)-1; | |
2340 | ||
7b128766 JB |
2341 | while (1) { |
2342 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2343 | if (ret < 0) |
2344 | goto out; | |
7b128766 JB |
2345 | |
2346 | /* | |
2347 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2348 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2349 | * find the key and see if we have stuff that matches |
2350 | */ | |
2351 | if (ret > 0) { | |
66b4ffd1 | 2352 | ret = 0; |
7b128766 JB |
2353 | if (path->slots[0] == 0) |
2354 | break; | |
2355 | path->slots[0]--; | |
2356 | } | |
2357 | ||
2358 | /* pull out the item */ | |
2359 | leaf = path->nodes[0]; | |
7b128766 JB |
2360 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2361 | ||
2362 | /* make sure the item matches what we want */ | |
2363 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2364 | break; | |
2365 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2366 | break; | |
2367 | ||
2368 | /* release the path since we're done with it */ | |
b3b4aa74 | 2369 | btrfs_release_path(path); |
7b128766 JB |
2370 | |
2371 | /* | |
2372 | * this is where we are basically btrfs_lookup, without the | |
2373 | * crossing root thing. we store the inode number in the | |
2374 | * offset of the orphan item. | |
2375 | */ | |
8f6d7f4f JB |
2376 | |
2377 | if (found_key.offset == last_objectid) { | |
2378 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2379 | "stopping orphan cleanup\n"); | |
2380 | ret = -EINVAL; | |
2381 | goto out; | |
2382 | } | |
2383 | ||
2384 | last_objectid = found_key.offset; | |
2385 | ||
5d4f98a2 YZ |
2386 | found_key.objectid = found_key.offset; |
2387 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2388 | found_key.offset = 0; | |
73f73415 | 2389 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2390 | ret = PTR_RET(inode); |
2391 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2392 | goto out; |
7b128766 | 2393 | |
f8e9e0b0 AJ |
2394 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2395 | struct btrfs_root *dead_root; | |
2396 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2397 | int is_dead_root = 0; | |
2398 | ||
2399 | /* | |
2400 | * this is an orphan in the tree root. Currently these | |
2401 | * could come from 2 sources: | |
2402 | * a) a snapshot deletion in progress | |
2403 | * b) a free space cache inode | |
2404 | * We need to distinguish those two, as the snapshot | |
2405 | * orphan must not get deleted. | |
2406 | * find_dead_roots already ran before us, so if this | |
2407 | * is a snapshot deletion, we should find the root | |
2408 | * in the dead_roots list | |
2409 | */ | |
2410 | spin_lock(&fs_info->trans_lock); | |
2411 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2412 | root_list) { | |
2413 | if (dead_root->root_key.objectid == | |
2414 | found_key.objectid) { | |
2415 | is_dead_root = 1; | |
2416 | break; | |
2417 | } | |
2418 | } | |
2419 | spin_unlock(&fs_info->trans_lock); | |
2420 | if (is_dead_root) { | |
2421 | /* prevent this orphan from being found again */ | |
2422 | key.offset = found_key.objectid - 1; | |
2423 | continue; | |
2424 | } | |
2425 | } | |
7b128766 | 2426 | /* |
a8c9e576 JB |
2427 | * Inode is already gone but the orphan item is still there, |
2428 | * kill the orphan item. | |
7b128766 | 2429 | */ |
a8c9e576 JB |
2430 | if (ret == -ESTALE) { |
2431 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2432 | if (IS_ERR(trans)) { |
2433 | ret = PTR_ERR(trans); | |
2434 | goto out; | |
2435 | } | |
8a35d95f JB |
2436 | printk(KERN_ERR "auto deleting %Lu\n", |
2437 | found_key.objectid); | |
a8c9e576 JB |
2438 | ret = btrfs_del_orphan_item(trans, root, |
2439 | found_key.objectid); | |
79787eaa | 2440 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2441 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2442 | continue; |
2443 | } | |
2444 | ||
a8c9e576 JB |
2445 | /* |
2446 | * add this inode to the orphan list so btrfs_orphan_del does | |
2447 | * the proper thing when we hit it | |
2448 | */ | |
8a35d95f JB |
2449 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2450 | &BTRFS_I(inode)->runtime_flags); | |
a8c9e576 | 2451 | |
7b128766 JB |
2452 | /* if we have links, this was a truncate, lets do that */ |
2453 | if (inode->i_nlink) { | |
a41ad394 JB |
2454 | if (!S_ISREG(inode->i_mode)) { |
2455 | WARN_ON(1); | |
2456 | iput(inode); | |
2457 | continue; | |
2458 | } | |
7b128766 | 2459 | nr_truncate++; |
66b4ffd1 | 2460 | ret = btrfs_truncate(inode); |
7b128766 JB |
2461 | } else { |
2462 | nr_unlink++; | |
2463 | } | |
2464 | ||
2465 | /* this will do delete_inode and everything for us */ | |
2466 | iput(inode); | |
66b4ffd1 JB |
2467 | if (ret) |
2468 | goto out; | |
7b128766 | 2469 | } |
3254c876 MX |
2470 | /* release the path since we're done with it */ |
2471 | btrfs_release_path(path); | |
2472 | ||
d68fc57b YZ |
2473 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2474 | ||
2475 | if (root->orphan_block_rsv) | |
2476 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2477 | (u64)-1); | |
2478 | ||
2479 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2480 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2481 | if (!IS_ERR(trans)) |
2482 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2483 | } |
7b128766 JB |
2484 | |
2485 | if (nr_unlink) | |
2486 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2487 | if (nr_truncate) | |
2488 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2489 | |
2490 | out: | |
2491 | if (ret) | |
2492 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2493 | btrfs_free_path(path); | |
2494 | return ret; | |
7b128766 JB |
2495 | } |
2496 | ||
46a53cca CM |
2497 | /* |
2498 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2499 | * don't find any xattrs, we know there can't be any acls. | |
2500 | * | |
2501 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2502 | */ | |
2503 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2504 | int slot, u64 objectid) | |
2505 | { | |
2506 | u32 nritems = btrfs_header_nritems(leaf); | |
2507 | struct btrfs_key found_key; | |
2508 | int scanned = 0; | |
2509 | ||
2510 | slot++; | |
2511 | while (slot < nritems) { | |
2512 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2513 | ||
2514 | /* we found a different objectid, there must not be acls */ | |
2515 | if (found_key.objectid != objectid) | |
2516 | return 0; | |
2517 | ||
2518 | /* we found an xattr, assume we've got an acl */ | |
2519 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2520 | return 1; | |
2521 | ||
2522 | /* | |
2523 | * we found a key greater than an xattr key, there can't | |
2524 | * be any acls later on | |
2525 | */ | |
2526 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2527 | return 0; | |
2528 | ||
2529 | slot++; | |
2530 | scanned++; | |
2531 | ||
2532 | /* | |
2533 | * it goes inode, inode backrefs, xattrs, extents, | |
2534 | * so if there are a ton of hard links to an inode there can | |
2535 | * be a lot of backrefs. Don't waste time searching too hard, | |
2536 | * this is just an optimization | |
2537 | */ | |
2538 | if (scanned >= 8) | |
2539 | break; | |
2540 | } | |
2541 | /* we hit the end of the leaf before we found an xattr or | |
2542 | * something larger than an xattr. We have to assume the inode | |
2543 | * has acls | |
2544 | */ | |
2545 | return 1; | |
2546 | } | |
2547 | ||
d352ac68 CM |
2548 | /* |
2549 | * read an inode from the btree into the in-memory inode | |
2550 | */ | |
5d4f98a2 | 2551 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2552 | { |
2553 | struct btrfs_path *path; | |
5f39d397 | 2554 | struct extent_buffer *leaf; |
39279cc3 | 2555 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2556 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2557 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2558 | struct btrfs_key location; | |
46a53cca | 2559 | int maybe_acls; |
618e21d5 | 2560 | u32 rdev; |
39279cc3 | 2561 | int ret; |
2f7e33d4 MX |
2562 | bool filled = false; |
2563 | ||
2564 | ret = btrfs_fill_inode(inode, &rdev); | |
2565 | if (!ret) | |
2566 | filled = true; | |
39279cc3 CM |
2567 | |
2568 | path = btrfs_alloc_path(); | |
1748f843 MF |
2569 | if (!path) |
2570 | goto make_bad; | |
2571 | ||
d90c7321 | 2572 | path->leave_spinning = 1; |
39279cc3 | 2573 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2574 | |
39279cc3 | 2575 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2576 | if (ret) |
39279cc3 | 2577 | goto make_bad; |
39279cc3 | 2578 | |
5f39d397 | 2579 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2580 | |
2581 | if (filled) | |
2582 | goto cache_acl; | |
2583 | ||
5f39d397 CM |
2584 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2585 | struct btrfs_inode_item); | |
5f39d397 | 2586 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2587 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
2588 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
2589 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 2590 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2591 | |
2592 | tspec = btrfs_inode_atime(inode_item); | |
2593 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2594 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2595 | ||
2596 | tspec = btrfs_inode_mtime(inode_item); | |
2597 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2598 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2599 | ||
2600 | tspec = btrfs_inode_ctime(inode_item); | |
2601 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2602 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2603 | ||
a76a3cd4 | 2604 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2605 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
2606 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
2607 | ||
2608 | /* | |
2609 | * If we were modified in the current generation and evicted from memory | |
2610 | * and then re-read we need to do a full sync since we don't have any | |
2611 | * idea about which extents were modified before we were evicted from | |
2612 | * cache. | |
2613 | */ | |
2614 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
2615 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2616 | &BTRFS_I(inode)->runtime_flags); | |
2617 | ||
0c4d2d95 | 2618 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2619 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2620 | inode->i_rdev = 0; |
5f39d397 CM |
2621 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2622 | ||
aec7477b | 2623 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2624 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2625 | cache_acl: |
46a53cca CM |
2626 | /* |
2627 | * try to precache a NULL acl entry for files that don't have | |
2628 | * any xattrs or acls | |
2629 | */ | |
33345d01 LZ |
2630 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2631 | btrfs_ino(inode)); | |
72c04902 AV |
2632 | if (!maybe_acls) |
2633 | cache_no_acl(inode); | |
46a53cca | 2634 | |
39279cc3 | 2635 | btrfs_free_path(path); |
39279cc3 | 2636 | |
39279cc3 | 2637 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2638 | case S_IFREG: |
2639 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2640 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2641 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2642 | inode->i_fop = &btrfs_file_operations; |
2643 | inode->i_op = &btrfs_file_inode_operations; | |
2644 | break; | |
2645 | case S_IFDIR: | |
2646 | inode->i_fop = &btrfs_dir_file_operations; | |
2647 | if (root == root->fs_info->tree_root) | |
2648 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2649 | else | |
2650 | inode->i_op = &btrfs_dir_inode_operations; | |
2651 | break; | |
2652 | case S_IFLNK: | |
2653 | inode->i_op = &btrfs_symlink_inode_operations; | |
2654 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2655 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2656 | break; |
618e21d5 | 2657 | default: |
0279b4cd | 2658 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2659 | init_special_inode(inode, inode->i_mode, rdev); |
2660 | break; | |
39279cc3 | 2661 | } |
6cbff00f CH |
2662 | |
2663 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2664 | return; |
2665 | ||
2666 | make_bad: | |
39279cc3 | 2667 | btrfs_free_path(path); |
39279cc3 CM |
2668 | make_bad_inode(inode); |
2669 | } | |
2670 | ||
d352ac68 CM |
2671 | /* |
2672 | * given a leaf and an inode, copy the inode fields into the leaf | |
2673 | */ | |
e02119d5 CM |
2674 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2675 | struct extent_buffer *leaf, | |
5f39d397 | 2676 | struct btrfs_inode_item *item, |
39279cc3 CM |
2677 | struct inode *inode) |
2678 | { | |
2f2f43d3 EB |
2679 | btrfs_set_inode_uid(leaf, item, i_uid_read(inode)); |
2680 | btrfs_set_inode_gid(leaf, item, i_gid_read(inode)); | |
dbe674a9 | 2681 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2682 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2683 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2684 | ||
2685 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2686 | inode->i_atime.tv_sec); | |
2687 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2688 | inode->i_atime.tv_nsec); | |
2689 | ||
2690 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2691 | inode->i_mtime.tv_sec); | |
2692 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2693 | inode->i_mtime.tv_nsec); | |
2694 | ||
2695 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2696 | inode->i_ctime.tv_sec); | |
2697 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2698 | inode->i_ctime.tv_nsec); | |
2699 | ||
a76a3cd4 | 2700 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2701 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
0c4d2d95 | 2702 | btrfs_set_inode_sequence(leaf, item, inode->i_version); |
e02119d5 | 2703 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2704 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2705 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2706 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2707 | } |
2708 | ||
d352ac68 CM |
2709 | /* |
2710 | * copy everything in the in-memory inode into the btree. | |
2711 | */ | |
2115133f | 2712 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2713 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2714 | { |
2715 | struct btrfs_inode_item *inode_item; | |
2716 | struct btrfs_path *path; | |
5f39d397 | 2717 | struct extent_buffer *leaf; |
39279cc3 CM |
2718 | int ret; |
2719 | ||
2720 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2721 | if (!path) |
2722 | return -ENOMEM; | |
2723 | ||
b9473439 | 2724 | path->leave_spinning = 1; |
16cdcec7 MX |
2725 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2726 | 1); | |
39279cc3 CM |
2727 | if (ret) { |
2728 | if (ret > 0) | |
2729 | ret = -ENOENT; | |
2730 | goto failed; | |
2731 | } | |
2732 | ||
b4ce94de | 2733 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2734 | leaf = path->nodes[0]; |
2735 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2736 | struct btrfs_inode_item); |
39279cc3 | 2737 | |
e02119d5 | 2738 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2739 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2740 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2741 | ret = 0; |
2742 | failed: | |
39279cc3 CM |
2743 | btrfs_free_path(path); |
2744 | return ret; | |
2745 | } | |
2746 | ||
2115133f CM |
2747 | /* |
2748 | * copy everything in the in-memory inode into the btree. | |
2749 | */ | |
2750 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2751 | struct btrfs_root *root, struct inode *inode) | |
2752 | { | |
2753 | int ret; | |
2754 | ||
2755 | /* | |
2756 | * If the inode is a free space inode, we can deadlock during commit | |
2757 | * if we put it into the delayed code. | |
2758 | * | |
2759 | * The data relocation inode should also be directly updated | |
2760 | * without delay | |
2761 | */ | |
83eea1f1 | 2762 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 2763 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
2764 | btrfs_update_root_times(trans, root); |
2765 | ||
2115133f CM |
2766 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2767 | if (!ret) | |
2768 | btrfs_set_inode_last_trans(trans, inode); | |
2769 | return ret; | |
2770 | } | |
2771 | ||
2772 | return btrfs_update_inode_item(trans, root, inode); | |
2773 | } | |
2774 | ||
be6aef60 JB |
2775 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
2776 | struct btrfs_root *root, | |
2777 | struct inode *inode) | |
2115133f CM |
2778 | { |
2779 | int ret; | |
2780 | ||
2781 | ret = btrfs_update_inode(trans, root, inode); | |
2782 | if (ret == -ENOSPC) | |
2783 | return btrfs_update_inode_item(trans, root, inode); | |
2784 | return ret; | |
2785 | } | |
2786 | ||
d352ac68 CM |
2787 | /* |
2788 | * unlink helper that gets used here in inode.c and in the tree logging | |
2789 | * recovery code. It remove a link in a directory with a given name, and | |
2790 | * also drops the back refs in the inode to the directory | |
2791 | */ | |
92986796 AV |
2792 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2793 | struct btrfs_root *root, | |
2794 | struct inode *dir, struct inode *inode, | |
2795 | const char *name, int name_len) | |
39279cc3 CM |
2796 | { |
2797 | struct btrfs_path *path; | |
39279cc3 | 2798 | int ret = 0; |
5f39d397 | 2799 | struct extent_buffer *leaf; |
39279cc3 | 2800 | struct btrfs_dir_item *di; |
5f39d397 | 2801 | struct btrfs_key key; |
aec7477b | 2802 | u64 index; |
33345d01 LZ |
2803 | u64 ino = btrfs_ino(inode); |
2804 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2805 | |
2806 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2807 | if (!path) { |
2808 | ret = -ENOMEM; | |
554233a6 | 2809 | goto out; |
54aa1f4d CM |
2810 | } |
2811 | ||
b9473439 | 2812 | path->leave_spinning = 1; |
33345d01 | 2813 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2814 | name, name_len, -1); |
2815 | if (IS_ERR(di)) { | |
2816 | ret = PTR_ERR(di); | |
2817 | goto err; | |
2818 | } | |
2819 | if (!di) { | |
2820 | ret = -ENOENT; | |
2821 | goto err; | |
2822 | } | |
5f39d397 CM |
2823 | leaf = path->nodes[0]; |
2824 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2825 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2826 | if (ret) |
2827 | goto err; | |
b3b4aa74 | 2828 | btrfs_release_path(path); |
39279cc3 | 2829 | |
33345d01 LZ |
2830 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2831 | dir_ino, &index); | |
aec7477b | 2832 | if (ret) { |
d397712b | 2833 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2834 | "inode %llu parent %llu\n", name_len, name, |
2835 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2836 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2837 | goto err; |
2838 | } | |
2839 | ||
16cdcec7 | 2840 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2841 | if (ret) { |
2842 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2843 | goto err; |
79787eaa | 2844 | } |
39279cc3 | 2845 | |
e02119d5 | 2846 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2847 | inode, dir_ino); |
79787eaa JM |
2848 | if (ret != 0 && ret != -ENOENT) { |
2849 | btrfs_abort_transaction(trans, root, ret); | |
2850 | goto err; | |
2851 | } | |
e02119d5 CM |
2852 | |
2853 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2854 | dir, index); | |
6418c961 CM |
2855 | if (ret == -ENOENT) |
2856 | ret = 0; | |
39279cc3 CM |
2857 | err: |
2858 | btrfs_free_path(path); | |
e02119d5 CM |
2859 | if (ret) |
2860 | goto out; | |
2861 | ||
2862 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2863 | inode_inc_iversion(inode); |
2864 | inode_inc_iversion(dir); | |
e02119d5 | 2865 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 2866 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 2867 | out: |
39279cc3 CM |
2868 | return ret; |
2869 | } | |
2870 | ||
92986796 AV |
2871 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2872 | struct btrfs_root *root, | |
2873 | struct inode *dir, struct inode *inode, | |
2874 | const char *name, int name_len) | |
2875 | { | |
2876 | int ret; | |
2877 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2878 | if (!ret) { | |
2879 | btrfs_drop_nlink(inode); | |
2880 | ret = btrfs_update_inode(trans, root, inode); | |
2881 | } | |
2882 | return ret; | |
2883 | } | |
2884 | ||
2885 | ||
a22285a6 YZ |
2886 | /* helper to check if there is any shared block in the path */ |
2887 | static int check_path_shared(struct btrfs_root *root, | |
2888 | struct btrfs_path *path) | |
39279cc3 | 2889 | { |
a22285a6 YZ |
2890 | struct extent_buffer *eb; |
2891 | int level; | |
0e4dcbef | 2892 | u64 refs = 1; |
5df6a9f6 | 2893 | |
a22285a6 | 2894 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2895 | int ret; |
2896 | ||
a22285a6 YZ |
2897 | if (!path->nodes[level]) |
2898 | break; | |
2899 | eb = path->nodes[level]; | |
2900 | if (!btrfs_block_can_be_shared(root, eb)) | |
2901 | continue; | |
2902 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2903 | &refs, NULL); | |
2904 | if (refs > 1) | |
2905 | return 1; | |
5df6a9f6 | 2906 | } |
dedefd72 | 2907 | return 0; |
39279cc3 CM |
2908 | } |
2909 | ||
a22285a6 YZ |
2910 | /* |
2911 | * helper to start transaction for unlink and rmdir. | |
2912 | * | |
2913 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2914 | * so in enospc case, we should make sure they will free space before | |
2915 | * allowing them to use the global metadata reservation. | |
2916 | */ | |
2917 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2918 | struct dentry *dentry) | |
4df27c4d | 2919 | { |
39279cc3 | 2920 | struct btrfs_trans_handle *trans; |
a22285a6 | 2921 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2922 | struct btrfs_path *path; |
4df27c4d | 2923 | struct btrfs_dir_item *di; |
7b128766 | 2924 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2925 | u64 index; |
a22285a6 YZ |
2926 | int check_link = 1; |
2927 | int err = -ENOSPC; | |
4df27c4d | 2928 | int ret; |
33345d01 LZ |
2929 | u64 ino = btrfs_ino(inode); |
2930 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2931 | |
e70bea5f JB |
2932 | /* |
2933 | * 1 for the possible orphan item | |
2934 | * 1 for the dir item | |
2935 | * 1 for the dir index | |
2936 | * 1 for the inode ref | |
2937 | * 1 for the inode ref in the tree log | |
2938 | * 2 for the dir entries in the log | |
2939 | * 1 for the inode | |
2940 | */ | |
2941 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2942 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2943 | return trans; | |
4df27c4d | 2944 | |
33345d01 | 2945 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2946 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2947 | |
a22285a6 YZ |
2948 | /* check if there is someone else holds reference */ |
2949 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2950 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2951 | |
a22285a6 YZ |
2952 | if (atomic_read(&inode->i_count) > 2) |
2953 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2954 | |
a22285a6 YZ |
2955 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2956 | return ERR_PTR(-ENOSPC); | |
2957 | ||
2958 | path = btrfs_alloc_path(); | |
2959 | if (!path) { | |
2960 | root->fs_info->enospc_unlink = 0; | |
2961 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2962 | } |
2963 | ||
3880a1b4 JB |
2964 | /* 1 for the orphan item */ |
2965 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2966 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2967 | btrfs_free_path(path); |
2968 | root->fs_info->enospc_unlink = 0; | |
2969 | return trans; | |
2970 | } | |
4df27c4d | 2971 | |
a22285a6 YZ |
2972 | path->skip_locking = 1; |
2973 | path->search_commit_root = 1; | |
4df27c4d | 2974 | |
a22285a6 YZ |
2975 | ret = btrfs_lookup_inode(trans, root, path, |
2976 | &BTRFS_I(dir)->location, 0); | |
2977 | if (ret < 0) { | |
2978 | err = ret; | |
2979 | goto out; | |
2980 | } | |
2981 | if (ret == 0) { | |
2982 | if (check_path_shared(root, path)) | |
2983 | goto out; | |
2984 | } else { | |
2985 | check_link = 0; | |
5df6a9f6 | 2986 | } |
b3b4aa74 | 2987 | btrfs_release_path(path); |
a22285a6 YZ |
2988 | |
2989 | ret = btrfs_lookup_inode(trans, root, path, | |
2990 | &BTRFS_I(inode)->location, 0); | |
2991 | if (ret < 0) { | |
2992 | err = ret; | |
2993 | goto out; | |
2994 | } | |
2995 | if (ret == 0) { | |
2996 | if (check_path_shared(root, path)) | |
2997 | goto out; | |
2998 | } else { | |
2999 | check_link = 0; | |
3000 | } | |
b3b4aa74 | 3001 | btrfs_release_path(path); |
a22285a6 YZ |
3002 | |
3003 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
3004 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 3005 | ino, (u64)-1, 0); |
a22285a6 YZ |
3006 | if (ret < 0) { |
3007 | err = ret; | |
3008 | goto out; | |
3009 | } | |
79787eaa | 3010 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
3011 | if (check_path_shared(root, path)) |
3012 | goto out; | |
b3b4aa74 | 3013 | btrfs_release_path(path); |
a22285a6 YZ |
3014 | } |
3015 | ||
3016 | if (!check_link) { | |
3017 | err = 0; | |
3018 | goto out; | |
3019 | } | |
3020 | ||
33345d01 | 3021 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
3022 | dentry->d_name.name, dentry->d_name.len, 0); |
3023 | if (IS_ERR(di)) { | |
3024 | err = PTR_ERR(di); | |
3025 | goto out; | |
3026 | } | |
3027 | if (di) { | |
3028 | if (check_path_shared(root, path)) | |
3029 | goto out; | |
3030 | } else { | |
3031 | err = 0; | |
3032 | goto out; | |
3033 | } | |
b3b4aa74 | 3034 | btrfs_release_path(path); |
a22285a6 | 3035 | |
f186373f MF |
3036 | ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name, |
3037 | dentry->d_name.len, ino, dir_ino, 0, | |
3038 | &index); | |
3039 | if (ret) { | |
3040 | err = ret; | |
a22285a6 YZ |
3041 | goto out; |
3042 | } | |
f186373f | 3043 | |
a22285a6 YZ |
3044 | if (check_path_shared(root, path)) |
3045 | goto out; | |
f186373f | 3046 | |
b3b4aa74 | 3047 | btrfs_release_path(path); |
a22285a6 | 3048 | |
16cdcec7 MX |
3049 | /* |
3050 | * This is a commit root search, if we can lookup inode item and other | |
3051 | * relative items in the commit root, it means the transaction of | |
3052 | * dir/file creation has been committed, and the dir index item that we | |
3053 | * delay to insert has also been inserted into the commit root. So | |
3054 | * we needn't worry about the delayed insertion of the dir index item | |
3055 | * here. | |
3056 | */ | |
33345d01 | 3057 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
3058 | dentry->d_name.name, dentry->d_name.len, 0); |
3059 | if (IS_ERR(di)) { | |
3060 | err = PTR_ERR(di); | |
3061 | goto out; | |
3062 | } | |
3063 | BUG_ON(ret == -ENOENT); | |
3064 | if (check_path_shared(root, path)) | |
3065 | goto out; | |
3066 | ||
3067 | err = 0; | |
3068 | out: | |
3069 | btrfs_free_path(path); | |
3880a1b4 JB |
3070 | /* Migrate the orphan reservation over */ |
3071 | if (!err) | |
3072 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3073 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3074 | trans->bytes_reserved); |
3880a1b4 | 3075 | |
a22285a6 YZ |
3076 | if (err) { |
3077 | btrfs_end_transaction(trans, root); | |
3078 | root->fs_info->enospc_unlink = 0; | |
3079 | return ERR_PTR(err); | |
3080 | } | |
3081 | ||
3082 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3083 | return trans; | |
3084 | } | |
3085 | ||
3086 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3087 | struct btrfs_root *root) | |
3088 | { | |
66d8f3dd | 3089 | if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) { |
5a77d76c JB |
3090 | btrfs_block_rsv_release(root, trans->block_rsv, |
3091 | trans->bytes_reserved); | |
3092 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3093 | BUG_ON(!root->fs_info->enospc_unlink); |
3094 | root->fs_info->enospc_unlink = 0; | |
3095 | } | |
7ad85bb7 | 3096 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3097 | } |
3098 | ||
3099 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3100 | { | |
3101 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3102 | struct btrfs_trans_handle *trans; | |
3103 | struct inode *inode = dentry->d_inode; | |
3104 | int ret; | |
a22285a6 YZ |
3105 | |
3106 | trans = __unlink_start_trans(dir, dentry); | |
3107 | if (IS_ERR(trans)) | |
3108 | return PTR_ERR(trans); | |
5f39d397 | 3109 | |
12fcfd22 CM |
3110 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3111 | ||
e02119d5 CM |
3112 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3113 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3114 | if (ret) |
3115 | goto out; | |
7b128766 | 3116 | |
a22285a6 | 3117 | if (inode->i_nlink == 0) { |
7b128766 | 3118 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3119 | if (ret) |
3120 | goto out; | |
a22285a6 | 3121 | } |
7b128766 | 3122 | |
b532402e | 3123 | out: |
a22285a6 | 3124 | __unlink_end_trans(trans, root); |
b53d3f5d | 3125 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
3126 | return ret; |
3127 | } | |
3128 | ||
4df27c4d YZ |
3129 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3130 | struct btrfs_root *root, | |
3131 | struct inode *dir, u64 objectid, | |
3132 | const char *name, int name_len) | |
3133 | { | |
3134 | struct btrfs_path *path; | |
3135 | struct extent_buffer *leaf; | |
3136 | struct btrfs_dir_item *di; | |
3137 | struct btrfs_key key; | |
3138 | u64 index; | |
3139 | int ret; | |
33345d01 | 3140 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3141 | |
3142 | path = btrfs_alloc_path(); | |
3143 | if (!path) | |
3144 | return -ENOMEM; | |
3145 | ||
33345d01 | 3146 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3147 | name, name_len, -1); |
79787eaa JM |
3148 | if (IS_ERR_OR_NULL(di)) { |
3149 | if (!di) | |
3150 | ret = -ENOENT; | |
3151 | else | |
3152 | ret = PTR_ERR(di); | |
3153 | goto out; | |
3154 | } | |
4df27c4d YZ |
3155 | |
3156 | leaf = path->nodes[0]; | |
3157 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3158 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3159 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3160 | if (ret) { |
3161 | btrfs_abort_transaction(trans, root, ret); | |
3162 | goto out; | |
3163 | } | |
b3b4aa74 | 3164 | btrfs_release_path(path); |
4df27c4d YZ |
3165 | |
3166 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3167 | objectid, root->root_key.objectid, | |
33345d01 | 3168 | dir_ino, &index, name, name_len); |
4df27c4d | 3169 | if (ret < 0) { |
79787eaa JM |
3170 | if (ret != -ENOENT) { |
3171 | btrfs_abort_transaction(trans, root, ret); | |
3172 | goto out; | |
3173 | } | |
33345d01 | 3174 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3175 | name, name_len); |
79787eaa JM |
3176 | if (IS_ERR_OR_NULL(di)) { |
3177 | if (!di) | |
3178 | ret = -ENOENT; | |
3179 | else | |
3180 | ret = PTR_ERR(di); | |
3181 | btrfs_abort_transaction(trans, root, ret); | |
3182 | goto out; | |
3183 | } | |
4df27c4d YZ |
3184 | |
3185 | leaf = path->nodes[0]; | |
3186 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3187 | btrfs_release_path(path); |
4df27c4d YZ |
3188 | index = key.offset; |
3189 | } | |
945d8962 | 3190 | btrfs_release_path(path); |
4df27c4d | 3191 | |
16cdcec7 | 3192 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3193 | if (ret) { |
3194 | btrfs_abort_transaction(trans, root, ret); | |
3195 | goto out; | |
3196 | } | |
4df27c4d YZ |
3197 | |
3198 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3199 | inode_inc_iversion(dir); |
4df27c4d | 3200 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3201 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3202 | if (ret) |
3203 | btrfs_abort_transaction(trans, root, ret); | |
3204 | out: | |
71d7aed0 | 3205 | btrfs_free_path(path); |
79787eaa | 3206 | return ret; |
4df27c4d YZ |
3207 | } |
3208 | ||
39279cc3 CM |
3209 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3210 | { | |
3211 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3212 | int err = 0; |
39279cc3 | 3213 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3214 | struct btrfs_trans_handle *trans; |
39279cc3 | 3215 | |
b3ae244e | 3216 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3217 | return -ENOTEMPTY; |
b3ae244e DS |
3218 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3219 | return -EPERM; | |
134d4512 | 3220 | |
a22285a6 YZ |
3221 | trans = __unlink_start_trans(dir, dentry); |
3222 | if (IS_ERR(trans)) | |
5df6a9f6 | 3223 | return PTR_ERR(trans); |
5df6a9f6 | 3224 | |
33345d01 | 3225 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3226 | err = btrfs_unlink_subvol(trans, root, dir, |
3227 | BTRFS_I(inode)->location.objectid, | |
3228 | dentry->d_name.name, | |
3229 | dentry->d_name.len); | |
3230 | goto out; | |
3231 | } | |
3232 | ||
7b128766 JB |
3233 | err = btrfs_orphan_add(trans, inode); |
3234 | if (err) | |
4df27c4d | 3235 | goto out; |
7b128766 | 3236 | |
39279cc3 | 3237 | /* now the directory is empty */ |
e02119d5 CM |
3238 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3239 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3240 | if (!err) |
dbe674a9 | 3241 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3242 | out: |
a22285a6 | 3243 | __unlink_end_trans(trans, root); |
b53d3f5d | 3244 | btrfs_btree_balance_dirty(root); |
3954401f | 3245 | |
39279cc3 CM |
3246 | return err; |
3247 | } | |
3248 | ||
39279cc3 CM |
3249 | /* |
3250 | * this can truncate away extent items, csum items and directory items. | |
3251 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3252 | * any higher than new_size |
39279cc3 CM |
3253 | * |
3254 | * csum items that cross the new i_size are truncated to the new size | |
3255 | * as well. | |
7b128766 JB |
3256 | * |
3257 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3258 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3259 | */ |
8082510e YZ |
3260 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3261 | struct btrfs_root *root, | |
3262 | struct inode *inode, | |
3263 | u64 new_size, u32 min_type) | |
39279cc3 | 3264 | { |
39279cc3 | 3265 | struct btrfs_path *path; |
5f39d397 | 3266 | struct extent_buffer *leaf; |
39279cc3 | 3267 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3268 | struct btrfs_key key; |
3269 | struct btrfs_key found_key; | |
39279cc3 | 3270 | u64 extent_start = 0; |
db94535d | 3271 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3272 | u64 extent_offset = 0; |
39279cc3 | 3273 | u64 item_end = 0; |
8082510e YZ |
3274 | u64 mask = root->sectorsize - 1; |
3275 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3276 | int found_extent; |
3277 | int del_item; | |
85e21bac CM |
3278 | int pending_del_nr = 0; |
3279 | int pending_del_slot = 0; | |
179e29e4 | 3280 | int extent_type = -1; |
8082510e YZ |
3281 | int ret; |
3282 | int err = 0; | |
33345d01 | 3283 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3284 | |
3285 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3286 | |
0eb0e19c MF |
3287 | path = btrfs_alloc_path(); |
3288 | if (!path) | |
3289 | return -ENOMEM; | |
3290 | path->reada = -1; | |
3291 | ||
5dc562c5 JB |
3292 | /* |
3293 | * We want to drop from the next block forward in case this new size is | |
3294 | * not block aligned since we will be keeping the last block of the | |
3295 | * extent just the way it is. | |
3296 | */ | |
0af3d00b | 3297 | if (root->ref_cows || root == root->fs_info->tree_root) |
5dc562c5 | 3298 | btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0); |
8082510e | 3299 | |
16cdcec7 MX |
3300 | /* |
3301 | * This function is also used to drop the items in the log tree before | |
3302 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3303 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3304 | * items. | |
3305 | */ | |
3306 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3307 | btrfs_kill_delayed_inode_items(inode); | |
3308 | ||
33345d01 | 3309 | key.objectid = ino; |
39279cc3 | 3310 | key.offset = (u64)-1; |
5f39d397 CM |
3311 | key.type = (u8)-1; |
3312 | ||
85e21bac | 3313 | search_again: |
b9473439 | 3314 | path->leave_spinning = 1; |
85e21bac | 3315 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3316 | if (ret < 0) { |
3317 | err = ret; | |
3318 | goto out; | |
3319 | } | |
d397712b | 3320 | |
85e21bac | 3321 | if (ret > 0) { |
e02119d5 CM |
3322 | /* there are no items in the tree for us to truncate, we're |
3323 | * done | |
3324 | */ | |
8082510e YZ |
3325 | if (path->slots[0] == 0) |
3326 | goto out; | |
85e21bac CM |
3327 | path->slots[0]--; |
3328 | } | |
3329 | ||
d397712b | 3330 | while (1) { |
39279cc3 | 3331 | fi = NULL; |
5f39d397 CM |
3332 | leaf = path->nodes[0]; |
3333 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3334 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3335 | |
33345d01 | 3336 | if (found_key.objectid != ino) |
39279cc3 | 3337 | break; |
5f39d397 | 3338 | |
85e21bac | 3339 | if (found_type < min_type) |
39279cc3 CM |
3340 | break; |
3341 | ||
5f39d397 | 3342 | item_end = found_key.offset; |
39279cc3 | 3343 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3344 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3345 | struct btrfs_file_extent_item); |
179e29e4 CM |
3346 | extent_type = btrfs_file_extent_type(leaf, fi); |
3347 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3348 | item_end += |
db94535d | 3349 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3350 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3351 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3352 | fi); |
39279cc3 | 3353 | } |
008630c1 | 3354 | item_end--; |
39279cc3 | 3355 | } |
8082510e YZ |
3356 | if (found_type > min_type) { |
3357 | del_item = 1; | |
3358 | } else { | |
3359 | if (item_end < new_size) | |
b888db2b | 3360 | break; |
8082510e YZ |
3361 | if (found_key.offset >= new_size) |
3362 | del_item = 1; | |
3363 | else | |
3364 | del_item = 0; | |
39279cc3 | 3365 | } |
39279cc3 | 3366 | found_extent = 0; |
39279cc3 | 3367 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3368 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3369 | goto delete; | |
3370 | ||
3371 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3372 | u64 num_dec; |
db94535d | 3373 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3374 | if (!del_item) { |
db94535d CM |
3375 | u64 orig_num_bytes = |
3376 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3377 | extent_num_bytes = new_size - |
5f39d397 | 3378 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3379 | extent_num_bytes = extent_num_bytes & |
3380 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3381 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3382 | extent_num_bytes); | |
3383 | num_dec = (orig_num_bytes - | |
9069218d | 3384 | extent_num_bytes); |
e02119d5 | 3385 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3386 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3387 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3388 | } else { |
db94535d CM |
3389 | extent_num_bytes = |
3390 | btrfs_file_extent_disk_num_bytes(leaf, | |
3391 | fi); | |
5d4f98a2 YZ |
3392 | extent_offset = found_key.offset - |
3393 | btrfs_file_extent_offset(leaf, fi); | |
3394 | ||
39279cc3 | 3395 | /* FIXME blocksize != 4096 */ |
9069218d | 3396 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3397 | if (extent_start != 0) { |
3398 | found_extent = 1; | |
e02119d5 | 3399 | if (root->ref_cows) |
a76a3cd4 | 3400 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3401 | } |
39279cc3 | 3402 | } |
9069218d | 3403 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3404 | /* |
3405 | * we can't truncate inline items that have had | |
3406 | * special encodings | |
3407 | */ | |
3408 | if (!del_item && | |
3409 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3410 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3411 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3412 | u32 size = new_size - found_key.offset; |
3413 | ||
3414 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3415 | inode_sub_bytes(inode, item_end + 1 - |
3416 | new_size); | |
e02119d5 CM |
3417 | } |
3418 | size = | |
3419 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3420 | btrfs_truncate_item(trans, root, path, |
3421 | size, 1); | |
e02119d5 | 3422 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3423 | inode_sub_bytes(inode, item_end + 1 - |
3424 | found_key.offset); | |
9069218d | 3425 | } |
39279cc3 | 3426 | } |
179e29e4 | 3427 | delete: |
39279cc3 | 3428 | if (del_item) { |
85e21bac CM |
3429 | if (!pending_del_nr) { |
3430 | /* no pending yet, add ourselves */ | |
3431 | pending_del_slot = path->slots[0]; | |
3432 | pending_del_nr = 1; | |
3433 | } else if (pending_del_nr && | |
3434 | path->slots[0] + 1 == pending_del_slot) { | |
3435 | /* hop on the pending chunk */ | |
3436 | pending_del_nr++; | |
3437 | pending_del_slot = path->slots[0]; | |
3438 | } else { | |
d397712b | 3439 | BUG(); |
85e21bac | 3440 | } |
39279cc3 CM |
3441 | } else { |
3442 | break; | |
3443 | } | |
0af3d00b JB |
3444 | if (found_extent && (root->ref_cows || |
3445 | root == root->fs_info->tree_root)) { | |
b9473439 | 3446 | btrfs_set_path_blocking(path); |
39279cc3 | 3447 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3448 | extent_num_bytes, 0, |
3449 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3450 | ino, extent_offset, 0); |
39279cc3 CM |
3451 | BUG_ON(ret); |
3452 | } | |
85e21bac | 3453 | |
8082510e YZ |
3454 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3455 | break; | |
3456 | ||
3457 | if (path->slots[0] == 0 || | |
3458 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
3459 | if (pending_del_nr) { |
3460 | ret = btrfs_del_items(trans, root, path, | |
3461 | pending_del_slot, | |
3462 | pending_del_nr); | |
79787eaa JM |
3463 | if (ret) { |
3464 | btrfs_abort_transaction(trans, | |
3465 | root, ret); | |
3466 | goto error; | |
3467 | } | |
8082510e YZ |
3468 | pending_del_nr = 0; |
3469 | } | |
b3b4aa74 | 3470 | btrfs_release_path(path); |
85e21bac | 3471 | goto search_again; |
8082510e YZ |
3472 | } else { |
3473 | path->slots[0]--; | |
85e21bac | 3474 | } |
39279cc3 | 3475 | } |
8082510e | 3476 | out: |
85e21bac CM |
3477 | if (pending_del_nr) { |
3478 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3479 | pending_del_nr); | |
79787eaa JM |
3480 | if (ret) |
3481 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3482 | } |
79787eaa | 3483 | error: |
39279cc3 | 3484 | btrfs_free_path(path); |
8082510e | 3485 | return err; |
39279cc3 CM |
3486 | } |
3487 | ||
3488 | /* | |
2aaa6655 JB |
3489 | * btrfs_truncate_page - read, zero a chunk and write a page |
3490 | * @inode - inode that we're zeroing | |
3491 | * @from - the offset to start zeroing | |
3492 | * @len - the length to zero, 0 to zero the entire range respective to the | |
3493 | * offset | |
3494 | * @front - zero up to the offset instead of from the offset on | |
3495 | * | |
3496 | * This will find the page for the "from" offset and cow the page and zero the | |
3497 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 3498 | */ |
2aaa6655 JB |
3499 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
3500 | int front) | |
39279cc3 | 3501 | { |
2aaa6655 | 3502 | struct address_space *mapping = inode->i_mapping; |
db94535d | 3503 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3504 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3505 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3506 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3507 | char *kaddr; |
db94535d | 3508 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3509 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3510 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3511 | struct page *page; | |
3b16a4e3 | 3512 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3513 | int ret = 0; |
a52d9a80 | 3514 | u64 page_start; |
e6dcd2dc | 3515 | u64 page_end; |
39279cc3 | 3516 | |
2aaa6655 JB |
3517 | if ((offset & (blocksize - 1)) == 0 && |
3518 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 3519 | goto out; |
0ca1f7ce | 3520 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3521 | if (ret) |
3522 | goto out; | |
39279cc3 CM |
3523 | |
3524 | ret = -ENOMEM; | |
211c17f5 | 3525 | again: |
3b16a4e3 | 3526 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3527 | if (!page) { |
0ca1f7ce | 3528 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3529 | goto out; |
5d5e103a | 3530 | } |
e6dcd2dc CM |
3531 | |
3532 | page_start = page_offset(page); | |
3533 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3534 | ||
39279cc3 | 3535 | if (!PageUptodate(page)) { |
9ebefb18 | 3536 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3537 | lock_page(page); |
211c17f5 CM |
3538 | if (page->mapping != mapping) { |
3539 | unlock_page(page); | |
3540 | page_cache_release(page); | |
3541 | goto again; | |
3542 | } | |
39279cc3 CM |
3543 | if (!PageUptodate(page)) { |
3544 | ret = -EIO; | |
89642229 | 3545 | goto out_unlock; |
39279cc3 CM |
3546 | } |
3547 | } | |
211c17f5 | 3548 | wait_on_page_writeback(page); |
e6dcd2dc | 3549 | |
d0082371 | 3550 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3551 | set_page_extent_mapped(page); |
3552 | ||
3553 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3554 | if (ordered) { | |
2ac55d41 JB |
3555 | unlock_extent_cached(io_tree, page_start, page_end, |
3556 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3557 | unlock_page(page); |
3558 | page_cache_release(page); | |
eb84ae03 | 3559 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3560 | btrfs_put_ordered_extent(ordered); |
3561 | goto again; | |
3562 | } | |
3563 | ||
2ac55d41 | 3564 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
3565 | EXTENT_DIRTY | EXTENT_DELALLOC | |
3566 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 3567 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3568 | |
2ac55d41 JB |
3569 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3570 | &cached_state); | |
9ed74f2d | 3571 | if (ret) { |
2ac55d41 JB |
3572 | unlock_extent_cached(io_tree, page_start, page_end, |
3573 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3574 | goto out_unlock; |
3575 | } | |
3576 | ||
e6dcd2dc CM |
3577 | ret = 0; |
3578 | if (offset != PAGE_CACHE_SIZE) { | |
2aaa6655 JB |
3579 | if (!len) |
3580 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 3581 | kaddr = kmap(page); |
2aaa6655 JB |
3582 | if (front) |
3583 | memset(kaddr, 0, offset); | |
3584 | else | |
3585 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
3586 | flush_dcache_page(page); |
3587 | kunmap(page); | |
3588 | } | |
247e743c | 3589 | ClearPageChecked(page); |
e6dcd2dc | 3590 | set_page_dirty(page); |
2ac55d41 JB |
3591 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3592 | GFP_NOFS); | |
39279cc3 | 3593 | |
89642229 | 3594 | out_unlock: |
5d5e103a | 3595 | if (ret) |
0ca1f7ce | 3596 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3597 | unlock_page(page); |
3598 | page_cache_release(page); | |
3599 | out: | |
3600 | return ret; | |
3601 | } | |
3602 | ||
695a0d0d JB |
3603 | /* |
3604 | * This function puts in dummy file extents for the area we're creating a hole | |
3605 | * for. So if we are truncating this file to a larger size we need to insert | |
3606 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3607 | * the range between oldsize and size | |
3608 | */ | |
a41ad394 | 3609 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3610 | { |
9036c102 YZ |
3611 | struct btrfs_trans_handle *trans; |
3612 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3613 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3614 | struct extent_map *em = NULL; |
2ac55d41 | 3615 | struct extent_state *cached_state = NULL; |
5dc562c5 | 3616 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9036c102 | 3617 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3618 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3619 | u64 block_end = (size + mask) & ~mask; |
3620 | u64 last_byte; | |
3621 | u64 cur_offset; | |
3622 | u64 hole_size; | |
9ed74f2d | 3623 | int err = 0; |
39279cc3 | 3624 | |
9036c102 YZ |
3625 | if (size <= hole_start) |
3626 | return 0; | |
3627 | ||
9036c102 YZ |
3628 | while (1) { |
3629 | struct btrfs_ordered_extent *ordered; | |
3630 | btrfs_wait_ordered_range(inode, hole_start, | |
3631 | block_end - hole_start); | |
2ac55d41 | 3632 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3633 | &cached_state); |
9036c102 YZ |
3634 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3635 | if (!ordered) | |
3636 | break; | |
2ac55d41 JB |
3637 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3638 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3639 | btrfs_put_ordered_extent(ordered); |
3640 | } | |
39279cc3 | 3641 | |
9036c102 YZ |
3642 | cur_offset = hole_start; |
3643 | while (1) { | |
3644 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3645 | block_end - cur_offset, 0); | |
79787eaa JM |
3646 | if (IS_ERR(em)) { |
3647 | err = PTR_ERR(em); | |
3648 | break; | |
3649 | } | |
9036c102 YZ |
3650 | last_byte = min(extent_map_end(em), block_end); |
3651 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3652 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 3653 | struct extent_map *hole_em; |
9036c102 | 3654 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3655 | |
3642320e | 3656 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3657 | if (IS_ERR(trans)) { |
3658 | err = PTR_ERR(trans); | |
9ed74f2d | 3659 | break; |
a22285a6 | 3660 | } |
8082510e | 3661 | |
5dc562c5 JB |
3662 | err = btrfs_drop_extents(trans, root, inode, |
3663 | cur_offset, | |
2671485d | 3664 | cur_offset + hole_size, 1); |
5b397377 | 3665 | if (err) { |
79787eaa | 3666 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3667 | btrfs_end_transaction(trans, root); |
3893e33b | 3668 | break; |
5b397377 | 3669 | } |
8082510e | 3670 | |
9036c102 | 3671 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3672 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3673 | 0, hole_size, 0, hole_size, |
3674 | 0, 0, 0); | |
5b397377 | 3675 | if (err) { |
79787eaa | 3676 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3677 | btrfs_end_transaction(trans, root); |
3893e33b | 3678 | break; |
5b397377 | 3679 | } |
8082510e | 3680 | |
5dc562c5 JB |
3681 | btrfs_drop_extent_cache(inode, cur_offset, |
3682 | cur_offset + hole_size - 1, 0); | |
3683 | hole_em = alloc_extent_map(); | |
3684 | if (!hole_em) { | |
3685 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3686 | &BTRFS_I(inode)->runtime_flags); | |
3687 | goto next; | |
3688 | } | |
3689 | hole_em->start = cur_offset; | |
3690 | hole_em->len = hole_size; | |
3691 | hole_em->orig_start = cur_offset; | |
8082510e | 3692 | |
5dc562c5 JB |
3693 | hole_em->block_start = EXTENT_MAP_HOLE; |
3694 | hole_em->block_len = 0; | |
3695 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; | |
3696 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
3697 | hole_em->generation = trans->transid; | |
8082510e | 3698 | |
5dc562c5 JB |
3699 | while (1) { |
3700 | write_lock(&em_tree->lock); | |
3701 | err = add_extent_mapping(em_tree, hole_em); | |
3702 | if (!err) | |
3703 | list_move(&hole_em->list, | |
3704 | &em_tree->modified_extents); | |
3705 | write_unlock(&em_tree->lock); | |
3706 | if (err != -EEXIST) | |
3707 | break; | |
3708 | btrfs_drop_extent_cache(inode, cur_offset, | |
3709 | cur_offset + | |
3710 | hole_size - 1, 0); | |
3711 | } | |
3712 | free_extent_map(hole_em); | |
3713 | next: | |
3642320e | 3714 | btrfs_update_inode(trans, root, inode); |
8082510e | 3715 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3716 | } |
3717 | free_extent_map(em); | |
a22285a6 | 3718 | em = NULL; |
9036c102 | 3719 | cur_offset = last_byte; |
8082510e | 3720 | if (cur_offset >= block_end) |
9036c102 YZ |
3721 | break; |
3722 | } | |
1832a6d5 | 3723 | |
a22285a6 | 3724 | free_extent_map(em); |
2ac55d41 JB |
3725 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3726 | GFP_NOFS); | |
9036c102 YZ |
3727 | return err; |
3728 | } | |
39279cc3 | 3729 | |
a41ad394 | 3730 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3731 | { |
f4a2f4c5 MX |
3732 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3733 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3734 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3735 | int ret; |
3736 | ||
a41ad394 | 3737 | if (newsize == oldsize) |
8082510e YZ |
3738 | return 0; |
3739 | ||
a41ad394 | 3740 | if (newsize > oldsize) { |
a41ad394 JB |
3741 | truncate_pagecache(inode, oldsize, newsize); |
3742 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3743 | if (ret) |
8082510e | 3744 | return ret; |
8082510e | 3745 | |
f4a2f4c5 MX |
3746 | trans = btrfs_start_transaction(root, 1); |
3747 | if (IS_ERR(trans)) | |
3748 | return PTR_ERR(trans); | |
3749 | ||
3750 | i_size_write(inode, newsize); | |
3751 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3752 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3753 | btrfs_end_transaction(trans, root); |
a41ad394 | 3754 | } else { |
8082510e | 3755 | |
a41ad394 JB |
3756 | /* |
3757 | * We're truncating a file that used to have good data down to | |
3758 | * zero. Make sure it gets into the ordered flush list so that | |
3759 | * any new writes get down to disk quickly. | |
3760 | */ | |
3761 | if (newsize == 0) | |
72ac3c0d JB |
3762 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3763 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3764 | |
a41ad394 JB |
3765 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3766 | truncate_setsize(inode, newsize); | |
3767 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3768 | } |
3769 | ||
a41ad394 | 3770 | return ret; |
8082510e YZ |
3771 | } |
3772 | ||
9036c102 YZ |
3773 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3774 | { | |
3775 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3776 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3777 | int err; |
39279cc3 | 3778 | |
b83cc969 LZ |
3779 | if (btrfs_root_readonly(root)) |
3780 | return -EROFS; | |
3781 | ||
9036c102 YZ |
3782 | err = inode_change_ok(inode, attr); |
3783 | if (err) | |
3784 | return err; | |
2bf5a725 | 3785 | |
5a3f23d5 | 3786 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3787 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3788 | if (err) |
3789 | return err; | |
39279cc3 | 3790 | } |
9036c102 | 3791 | |
1025774c CH |
3792 | if (attr->ia_valid) { |
3793 | setattr_copy(inode, attr); | |
0c4d2d95 | 3794 | inode_inc_iversion(inode); |
22c44fe6 | 3795 | err = btrfs_dirty_inode(inode); |
1025774c | 3796 | |
22c44fe6 | 3797 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3798 | err = btrfs_acl_chmod(inode); |
3799 | } | |
33268eaf | 3800 | |
39279cc3 CM |
3801 | return err; |
3802 | } | |
61295eb8 | 3803 | |
bd555975 | 3804 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3805 | { |
3806 | struct btrfs_trans_handle *trans; | |
3807 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3808 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3809 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
3810 | int ret; |
3811 | ||
1abe9b8a | 3812 | trace_btrfs_inode_evict(inode); |
3813 | ||
39279cc3 | 3814 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3815 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 3816 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
3817 | goto no_delete; |
3818 | ||
39279cc3 | 3819 | if (is_bad_inode(inode)) { |
7b128766 | 3820 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3821 | goto no_delete; |
3822 | } | |
bd555975 | 3823 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3824 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3825 | |
c71bf099 | 3826 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 3827 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 3828 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
3829 | goto no_delete; |
3830 | } | |
3831 | ||
76dda93c YZ |
3832 | if (inode->i_nlink > 0) { |
3833 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3834 | goto no_delete; | |
3835 | } | |
3836 | ||
66d8f3dd | 3837 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
3838 | if (!rsv) { |
3839 | btrfs_orphan_del(NULL, inode); | |
3840 | goto no_delete; | |
3841 | } | |
4a338542 | 3842 | rsv->size = min_size; |
ca7e70f5 | 3843 | rsv->failfast = 1; |
726c35fa | 3844 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3845 | |
dbe674a9 | 3846 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3847 | |
4289a667 | 3848 | /* |
8407aa46 MX |
3849 | * This is a bit simpler than btrfs_truncate since we've already |
3850 | * reserved our space for our orphan item in the unlink, so we just | |
3851 | * need to reserve some slack space in case we add bytes and update | |
3852 | * inode item when doing the truncate. | |
4289a667 | 3853 | */ |
8082510e | 3854 | while (1) { |
08e007d2 MX |
3855 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
3856 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
3857 | |
3858 | /* | |
3859 | * Try and steal from the global reserve since we will | |
3860 | * likely not use this space anyway, we want to try as | |
3861 | * hard as possible to get this to work. | |
3862 | */ | |
3863 | if (ret) | |
3864 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3865 | |
d68fc57b | 3866 | if (ret) { |
4289a667 | 3867 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3868 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3869 | btrfs_orphan_del(NULL, inode); |
3870 | btrfs_free_block_rsv(root, rsv); | |
3871 | goto no_delete; | |
d68fc57b | 3872 | } |
7b128766 | 3873 | |
08e007d2 | 3874 | trans = btrfs_start_transaction_lflush(root, 1); |
4289a667 JB |
3875 | if (IS_ERR(trans)) { |
3876 | btrfs_orphan_del(NULL, inode); | |
3877 | btrfs_free_block_rsv(root, rsv); | |
3878 | goto no_delete; | |
d68fc57b | 3879 | } |
7b128766 | 3880 | |
4289a667 JB |
3881 | trans->block_rsv = rsv; |
3882 | ||
d68fc57b | 3883 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 3884 | if (ret != -ENOSPC) |
8082510e | 3885 | break; |
85e21bac | 3886 | |
8407aa46 MX |
3887 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
3888 | ret = btrfs_update_inode(trans, root, inode); | |
3889 | BUG_ON(ret); | |
3890 | ||
8082510e YZ |
3891 | btrfs_end_transaction(trans, root); |
3892 | trans = NULL; | |
b53d3f5d | 3893 | btrfs_btree_balance_dirty(root); |
8082510e | 3894 | } |
5f39d397 | 3895 | |
4289a667 JB |
3896 | btrfs_free_block_rsv(root, rsv); |
3897 | ||
8082510e | 3898 | if (ret == 0) { |
4289a667 | 3899 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3900 | ret = btrfs_orphan_del(trans, inode); |
3901 | BUG_ON(ret); | |
3902 | } | |
54aa1f4d | 3903 | |
4289a667 | 3904 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3905 | if (!(root == root->fs_info->tree_root || |
3906 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3907 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3908 | |
54aa1f4d | 3909 | btrfs_end_transaction(trans, root); |
b53d3f5d | 3910 | btrfs_btree_balance_dirty(root); |
39279cc3 | 3911 | no_delete: |
dbd5768f | 3912 | clear_inode(inode); |
8082510e | 3913 | return; |
39279cc3 CM |
3914 | } |
3915 | ||
3916 | /* | |
3917 | * this returns the key found in the dir entry in the location pointer. | |
3918 | * If no dir entries were found, location->objectid is 0. | |
3919 | */ | |
3920 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3921 | struct btrfs_key *location) | |
3922 | { | |
3923 | const char *name = dentry->d_name.name; | |
3924 | int namelen = dentry->d_name.len; | |
3925 | struct btrfs_dir_item *di; | |
3926 | struct btrfs_path *path; | |
3927 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3928 | int ret = 0; |
39279cc3 CM |
3929 | |
3930 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3931 | if (!path) |
3932 | return -ENOMEM; | |
3954401f | 3933 | |
33345d01 | 3934 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3935 | namelen, 0); |
0d9f7f3e Y |
3936 | if (IS_ERR(di)) |
3937 | ret = PTR_ERR(di); | |
d397712b | 3938 | |
c704005d | 3939 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3940 | goto out_err; |
d397712b | 3941 | |
5f39d397 | 3942 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3943 | out: |
39279cc3 CM |
3944 | btrfs_free_path(path); |
3945 | return ret; | |
3954401f CM |
3946 | out_err: |
3947 | location->objectid = 0; | |
3948 | goto out; | |
39279cc3 CM |
3949 | } |
3950 | ||
3951 | /* | |
3952 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3953 | * needs to be changed to reflect the root directory of the tree root. This | |
3954 | * is kind of like crossing a mount point. | |
3955 | */ | |
3956 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3957 | struct inode *dir, |
3958 | struct dentry *dentry, | |
3959 | struct btrfs_key *location, | |
3960 | struct btrfs_root **sub_root) | |
39279cc3 | 3961 | { |
4df27c4d YZ |
3962 | struct btrfs_path *path; |
3963 | struct btrfs_root *new_root; | |
3964 | struct btrfs_root_ref *ref; | |
3965 | struct extent_buffer *leaf; | |
3966 | int ret; | |
3967 | int err = 0; | |
39279cc3 | 3968 | |
4df27c4d YZ |
3969 | path = btrfs_alloc_path(); |
3970 | if (!path) { | |
3971 | err = -ENOMEM; | |
3972 | goto out; | |
3973 | } | |
39279cc3 | 3974 | |
4df27c4d YZ |
3975 | err = -ENOENT; |
3976 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3977 | BTRFS_I(dir)->root->root_key.objectid, | |
3978 | location->objectid); | |
3979 | if (ret) { | |
3980 | if (ret < 0) | |
3981 | err = ret; | |
3982 | goto out; | |
3983 | } | |
39279cc3 | 3984 | |
4df27c4d YZ |
3985 | leaf = path->nodes[0]; |
3986 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3987 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3988 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3989 | goto out; | |
39279cc3 | 3990 | |
4df27c4d YZ |
3991 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3992 | (unsigned long)(ref + 1), | |
3993 | dentry->d_name.len); | |
3994 | if (ret) | |
3995 | goto out; | |
3996 | ||
b3b4aa74 | 3997 | btrfs_release_path(path); |
4df27c4d YZ |
3998 | |
3999 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
4000 | if (IS_ERR(new_root)) { | |
4001 | err = PTR_ERR(new_root); | |
4002 | goto out; | |
4003 | } | |
4004 | ||
4005 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
4006 | err = -ENOENT; | |
4007 | goto out; | |
4008 | } | |
4009 | ||
4010 | *sub_root = new_root; | |
4011 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
4012 | location->type = BTRFS_INODE_ITEM_KEY; | |
4013 | location->offset = 0; | |
4014 | err = 0; | |
4015 | out: | |
4016 | btrfs_free_path(path); | |
4017 | return err; | |
39279cc3 CM |
4018 | } |
4019 | ||
5d4f98a2 YZ |
4020 | static void inode_tree_add(struct inode *inode) |
4021 | { | |
4022 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4023 | struct btrfs_inode *entry; | |
03e860bd FNP |
4024 | struct rb_node **p; |
4025 | struct rb_node *parent; | |
33345d01 | 4026 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
4027 | again: |
4028 | p = &root->inode_tree.rb_node; | |
4029 | parent = NULL; | |
5d4f98a2 | 4030 | |
1d3382cb | 4031 | if (inode_unhashed(inode)) |
76dda93c YZ |
4032 | return; |
4033 | ||
5d4f98a2 YZ |
4034 | spin_lock(&root->inode_lock); |
4035 | while (*p) { | |
4036 | parent = *p; | |
4037 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4038 | ||
33345d01 | 4039 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4040 | p = &parent->rb_left; |
33345d01 | 4041 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4042 | p = &parent->rb_right; |
5d4f98a2 YZ |
4043 | else { |
4044 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4045 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4046 | rb_erase(parent, &root->inode_tree); |
4047 | RB_CLEAR_NODE(parent); | |
4048 | spin_unlock(&root->inode_lock); | |
4049 | goto again; | |
5d4f98a2 YZ |
4050 | } |
4051 | } | |
4052 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4053 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4054 | spin_unlock(&root->inode_lock); | |
4055 | } | |
4056 | ||
4057 | static void inode_tree_del(struct inode *inode) | |
4058 | { | |
4059 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4060 | int empty = 0; |
5d4f98a2 | 4061 | |
03e860bd | 4062 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4063 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4064 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4065 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4066 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4067 | } |
03e860bd | 4068 | spin_unlock(&root->inode_lock); |
76dda93c | 4069 | |
0af3d00b JB |
4070 | /* |
4071 | * Free space cache has inodes in the tree root, but the tree root has a | |
4072 | * root_refs of 0, so this could end up dropping the tree root as a | |
4073 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4074 | * make sure we don't drop it. | |
4075 | */ | |
4076 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4077 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4078 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4079 | spin_lock(&root->inode_lock); | |
4080 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4081 | spin_unlock(&root->inode_lock); | |
4082 | if (empty) | |
4083 | btrfs_add_dead_root(root); | |
4084 | } | |
4085 | } | |
4086 | ||
143bede5 | 4087 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4088 | { |
4089 | struct rb_node *node; | |
4090 | struct rb_node *prev; | |
4091 | struct btrfs_inode *entry; | |
4092 | struct inode *inode; | |
4093 | u64 objectid = 0; | |
4094 | ||
4095 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4096 | ||
4097 | spin_lock(&root->inode_lock); | |
4098 | again: | |
4099 | node = root->inode_tree.rb_node; | |
4100 | prev = NULL; | |
4101 | while (node) { | |
4102 | prev = node; | |
4103 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4104 | ||
33345d01 | 4105 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4106 | node = node->rb_left; |
33345d01 | 4107 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4108 | node = node->rb_right; |
4109 | else | |
4110 | break; | |
4111 | } | |
4112 | if (!node) { | |
4113 | while (prev) { | |
4114 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4115 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4116 | node = prev; |
4117 | break; | |
4118 | } | |
4119 | prev = rb_next(prev); | |
4120 | } | |
4121 | } | |
4122 | while (node) { | |
4123 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4124 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4125 | inode = igrab(&entry->vfs_inode); |
4126 | if (inode) { | |
4127 | spin_unlock(&root->inode_lock); | |
4128 | if (atomic_read(&inode->i_count) > 1) | |
4129 | d_prune_aliases(inode); | |
4130 | /* | |
45321ac5 | 4131 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4132 | * the inode cache when its usage count |
4133 | * hits zero. | |
4134 | */ | |
4135 | iput(inode); | |
4136 | cond_resched(); | |
4137 | spin_lock(&root->inode_lock); | |
4138 | goto again; | |
4139 | } | |
4140 | ||
4141 | if (cond_resched_lock(&root->inode_lock)) | |
4142 | goto again; | |
4143 | ||
4144 | node = rb_next(node); | |
4145 | } | |
4146 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4147 | } |
4148 | ||
e02119d5 CM |
4149 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4150 | { | |
4151 | struct btrfs_iget_args *args = p; | |
4152 | inode->i_ino = args->ino; | |
e02119d5 | 4153 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4154 | return 0; |
4155 | } | |
4156 | ||
4157 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4158 | { | |
4159 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4160 | return args->ino == btrfs_ino(inode) && |
d397712b | 4161 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4162 | } |
4163 | ||
5d4f98a2 YZ |
4164 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4165 | u64 objectid, | |
4166 | struct btrfs_root *root) | |
39279cc3 CM |
4167 | { |
4168 | struct inode *inode; | |
4169 | struct btrfs_iget_args args; | |
4170 | args.ino = objectid; | |
4171 | args.root = root; | |
4172 | ||
4173 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4174 | btrfs_init_locked_inode, | |
4175 | (void *)&args); | |
4176 | return inode; | |
4177 | } | |
4178 | ||
1a54ef8c BR |
4179 | /* Get an inode object given its location and corresponding root. |
4180 | * Returns in *is_new if the inode was read from disk | |
4181 | */ | |
4182 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4183 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4184 | { |
4185 | struct inode *inode; | |
4186 | ||
4187 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4188 | if (!inode) | |
5d4f98a2 | 4189 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4190 | |
4191 | if (inode->i_state & I_NEW) { | |
4192 | BTRFS_I(inode)->root = root; | |
4193 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4194 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4195 | if (!is_bad_inode(inode)) { |
4196 | inode_tree_add(inode); | |
4197 | unlock_new_inode(inode); | |
4198 | if (new) | |
4199 | *new = 1; | |
4200 | } else { | |
e0b6d65b ST |
4201 | unlock_new_inode(inode); |
4202 | iput(inode); | |
4203 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4204 | } |
4205 | } | |
4206 | ||
1a54ef8c BR |
4207 | return inode; |
4208 | } | |
4209 | ||
4df27c4d YZ |
4210 | static struct inode *new_simple_dir(struct super_block *s, |
4211 | struct btrfs_key *key, | |
4212 | struct btrfs_root *root) | |
4213 | { | |
4214 | struct inode *inode = new_inode(s); | |
4215 | ||
4216 | if (!inode) | |
4217 | return ERR_PTR(-ENOMEM); | |
4218 | ||
4df27c4d YZ |
4219 | BTRFS_I(inode)->root = root; |
4220 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4221 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4222 | |
4223 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4224 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4225 | inode->i_fop = &simple_dir_operations; |
4226 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4227 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4228 | ||
4229 | return inode; | |
4230 | } | |
4231 | ||
3de4586c | 4232 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4233 | { |
d397712b | 4234 | struct inode *inode; |
4df27c4d | 4235 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4236 | struct btrfs_root *sub_root = root; |
4237 | struct btrfs_key location; | |
76dda93c | 4238 | int index; |
b4aff1f8 | 4239 | int ret = 0; |
39279cc3 CM |
4240 | |
4241 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4242 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4243 | |
b4aff1f8 JB |
4244 | if (unlikely(d_need_lookup(dentry))) { |
4245 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4246 | kfree(dentry->d_fsdata); | |
4247 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4248 | /* This thing is hashed, drop it for now */ |
4249 | d_drop(dentry); | |
b4aff1f8 JB |
4250 | } else { |
4251 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4252 | } | |
5f39d397 | 4253 | |
39279cc3 CM |
4254 | if (ret < 0) |
4255 | return ERR_PTR(ret); | |
5f39d397 | 4256 | |
4df27c4d YZ |
4257 | if (location.objectid == 0) |
4258 | return NULL; | |
4259 | ||
4260 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4261 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4262 | return inode; |
4263 | } | |
4264 | ||
4265 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4266 | ||
76dda93c | 4267 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4268 | ret = fixup_tree_root_location(root, dir, dentry, |
4269 | &location, &sub_root); | |
4270 | if (ret < 0) { | |
4271 | if (ret != -ENOENT) | |
4272 | inode = ERR_PTR(ret); | |
4273 | else | |
4274 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4275 | } else { | |
73f73415 | 4276 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4277 | } |
76dda93c YZ |
4278 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4279 | ||
34d19bad | 4280 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4281 | down_read(&root->fs_info->cleanup_work_sem); |
4282 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4283 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4284 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4285 | if (ret) |
4286 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4287 | } |
4288 | ||
3de4586c CM |
4289 | return inode; |
4290 | } | |
4291 | ||
fe15ce44 | 4292 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4293 | { |
4294 | struct btrfs_root *root; | |
848cce0d | 4295 | struct inode *inode = dentry->d_inode; |
76dda93c | 4296 | |
848cce0d LZ |
4297 | if (!inode && !IS_ROOT(dentry)) |
4298 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4299 | |
848cce0d LZ |
4300 | if (inode) { |
4301 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4302 | if (btrfs_root_refs(&root->root_item) == 0) |
4303 | return 1; | |
848cce0d LZ |
4304 | |
4305 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4306 | return 1; | |
efefb143 | 4307 | } |
76dda93c YZ |
4308 | return 0; |
4309 | } | |
4310 | ||
b4aff1f8 JB |
4311 | static void btrfs_dentry_release(struct dentry *dentry) |
4312 | { | |
4313 | if (dentry->d_fsdata) | |
4314 | kfree(dentry->d_fsdata); | |
4315 | } | |
4316 | ||
3de4586c | 4317 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4318 | unsigned int flags) |
3de4586c | 4319 | { |
a66e7cc6 JB |
4320 | struct dentry *ret; |
4321 | ||
4322 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4323 | if (unlikely(d_need_lookup(dentry))) { | |
4324 | spin_lock(&dentry->d_lock); | |
4325 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4326 | spin_unlock(&dentry->d_lock); | |
4327 | } | |
4328 | return ret; | |
39279cc3 CM |
4329 | } |
4330 | ||
16cdcec7 | 4331 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4332 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4333 | }; | |
4334 | ||
cbdf5a24 DW |
4335 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4336 | filldir_t filldir) | |
39279cc3 | 4337 | { |
6da6abae | 4338 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4339 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4340 | struct btrfs_item *item; | |
4341 | struct btrfs_dir_item *di; | |
4342 | struct btrfs_key key; | |
5f39d397 | 4343 | struct btrfs_key found_key; |
39279cc3 | 4344 | struct btrfs_path *path; |
16cdcec7 MX |
4345 | struct list_head ins_list; |
4346 | struct list_head del_list; | |
39279cc3 | 4347 | int ret; |
5f39d397 | 4348 | struct extent_buffer *leaf; |
39279cc3 | 4349 | int slot; |
39279cc3 CM |
4350 | unsigned char d_type; |
4351 | int over = 0; | |
4352 | u32 di_cur; | |
4353 | u32 di_total; | |
4354 | u32 di_len; | |
4355 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4356 | char tmp_name[32]; |
4357 | char *name_ptr; | |
4358 | int name_len; | |
16cdcec7 | 4359 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4360 | |
4361 | /* FIXME, use a real flag for deciding about the key type */ | |
4362 | if (root->fs_info->tree_root == root) | |
4363 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4364 | |
3954401f CM |
4365 | /* special case for "." */ |
4366 | if (filp->f_pos == 0) { | |
3765fefa HS |
4367 | over = filldir(dirent, ".", 1, |
4368 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4369 | if (over) |
4370 | return 0; | |
4371 | filp->f_pos = 1; | |
4372 | } | |
3954401f CM |
4373 | /* special case for .., just use the back ref */ |
4374 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4375 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4376 | over = filldir(dirent, "..", 2, |
3765fefa | 4377 | filp->f_pos, pino, DT_DIR); |
3954401f | 4378 | if (over) |
49593bfa | 4379 | return 0; |
3954401f CM |
4380 | filp->f_pos = 2; |
4381 | } | |
49593bfa | 4382 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4383 | if (!path) |
4384 | return -ENOMEM; | |
ff5714cc | 4385 | |
026fd317 | 4386 | path->reada = 1; |
49593bfa | 4387 | |
16cdcec7 MX |
4388 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4389 | INIT_LIST_HEAD(&ins_list); | |
4390 | INIT_LIST_HEAD(&del_list); | |
4391 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4392 | } | |
4393 | ||
39279cc3 CM |
4394 | btrfs_set_key_type(&key, key_type); |
4395 | key.offset = filp->f_pos; | |
33345d01 | 4396 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4397 | |
39279cc3 CM |
4398 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4399 | if (ret < 0) | |
4400 | goto err; | |
49593bfa DW |
4401 | |
4402 | while (1) { | |
5f39d397 | 4403 | leaf = path->nodes[0]; |
39279cc3 | 4404 | slot = path->slots[0]; |
b9e03af0 LZ |
4405 | if (slot >= btrfs_header_nritems(leaf)) { |
4406 | ret = btrfs_next_leaf(root, path); | |
4407 | if (ret < 0) | |
4408 | goto err; | |
4409 | else if (ret > 0) | |
4410 | break; | |
4411 | continue; | |
39279cc3 | 4412 | } |
3de4586c | 4413 | |
5f39d397 CM |
4414 | item = btrfs_item_nr(leaf, slot); |
4415 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4416 | ||
4417 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4418 | break; |
5f39d397 | 4419 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4420 | break; |
5f39d397 | 4421 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4422 | goto next; |
16cdcec7 MX |
4423 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4424 | btrfs_should_delete_dir_index(&del_list, | |
4425 | found_key.offset)) | |
4426 | goto next; | |
5f39d397 CM |
4427 | |
4428 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4429 | is_curr = 1; |
49593bfa | 4430 | |
39279cc3 CM |
4431 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4432 | di_cur = 0; | |
5f39d397 | 4433 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4434 | |
4435 | while (di_cur < di_total) { | |
5f39d397 CM |
4436 | struct btrfs_key location; |
4437 | ||
22a94d44 JB |
4438 | if (verify_dir_item(root, leaf, di)) |
4439 | break; | |
4440 | ||
5f39d397 | 4441 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4442 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4443 | name_ptr = tmp_name; |
4444 | } else { | |
4445 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4446 | if (!name_ptr) { |
4447 | ret = -ENOMEM; | |
4448 | goto err; | |
4449 | } | |
5f39d397 CM |
4450 | } |
4451 | read_extent_buffer(leaf, name_ptr, | |
4452 | (unsigned long)(di + 1), name_len); | |
4453 | ||
4454 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4455 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4456 | |
fede766f | 4457 | |
3de4586c | 4458 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4459 | * skip it. |
4460 | * | |
4461 | * In contrast to old kernels, we insert the snapshot's | |
4462 | * dir item and dir index after it has been created, so | |
4463 | * we won't find a reference to our own snapshot. We | |
4464 | * still keep the following code for backward | |
4465 | * compatibility. | |
3de4586c CM |
4466 | */ |
4467 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4468 | location.objectid == root->root_key.objectid) { | |
4469 | over = 0; | |
4470 | goto skip; | |
4471 | } | |
5f39d397 | 4472 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4473 | found_key.offset, location.objectid, |
39279cc3 | 4474 | d_type); |
5f39d397 | 4475 | |
3de4586c | 4476 | skip: |
5f39d397 CM |
4477 | if (name_ptr != tmp_name) |
4478 | kfree(name_ptr); | |
4479 | ||
39279cc3 CM |
4480 | if (over) |
4481 | goto nopos; | |
5103e947 | 4482 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4483 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4484 | di_cur += di_len; |
4485 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4486 | } | |
b9e03af0 LZ |
4487 | next: |
4488 | path->slots[0]++; | |
39279cc3 | 4489 | } |
49593bfa | 4490 | |
16cdcec7 MX |
4491 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4492 | if (is_curr) | |
4493 | filp->f_pos++; | |
4494 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4495 | &ins_list); | |
4496 | if (ret) | |
4497 | goto nopos; | |
4498 | } | |
4499 | ||
49593bfa | 4500 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4501 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4502 | /* |
4503 | * 32-bit glibc will use getdents64, but then strtol - | |
4504 | * so the last number we can serve is this. | |
4505 | */ | |
4506 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4507 | else |
4508 | filp->f_pos++; | |
39279cc3 CM |
4509 | nopos: |
4510 | ret = 0; | |
4511 | err: | |
16cdcec7 MX |
4512 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4513 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4514 | btrfs_free_path(path); |
39279cc3 CM |
4515 | return ret; |
4516 | } | |
4517 | ||
a9185b41 | 4518 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4519 | { |
4520 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4521 | struct btrfs_trans_handle *trans; | |
4522 | int ret = 0; | |
0af3d00b | 4523 | bool nolock = false; |
39279cc3 | 4524 | |
72ac3c0d | 4525 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4526 | return 0; |
4527 | ||
83eea1f1 | 4528 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 4529 | nolock = true; |
0af3d00b | 4530 | |
a9185b41 | 4531 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4532 | if (nolock) |
7a7eaa40 | 4533 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4534 | else |
7a7eaa40 | 4535 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4536 | if (IS_ERR(trans)) |
4537 | return PTR_ERR(trans); | |
a698d075 | 4538 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
4539 | } |
4540 | return ret; | |
4541 | } | |
4542 | ||
4543 | /* | |
54aa1f4d | 4544 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4545 | * inode changes. But, it is most likely to find the inode in cache. |
4546 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4547 | * to keep or drop this code. | |
4548 | */ | |
22c44fe6 | 4549 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4550 | { |
4551 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4552 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4553 | int ret; |
4554 | ||
72ac3c0d | 4555 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4556 | return 0; |
39279cc3 | 4557 | |
7a7eaa40 | 4558 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4559 | if (IS_ERR(trans)) |
4560 | return PTR_ERR(trans); | |
8929ecfa YZ |
4561 | |
4562 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4563 | if (ret && ret == -ENOSPC) { |
4564 | /* whoops, lets try again with the full transaction */ | |
4565 | btrfs_end_transaction(trans, root); | |
4566 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4567 | if (IS_ERR(trans)) |
4568 | return PTR_ERR(trans); | |
8929ecfa | 4569 | |
94b60442 | 4570 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4571 | } |
39279cc3 | 4572 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4573 | if (BTRFS_I(inode)->delayed_node) |
4574 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4575 | |
4576 | return ret; | |
4577 | } | |
4578 | ||
4579 | /* | |
4580 | * This is a copy of file_update_time. We need this so we can return error on | |
4581 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4582 | */ | |
e41f941a JB |
4583 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
4584 | int flags) | |
22c44fe6 | 4585 | { |
2bc55652 AB |
4586 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4587 | ||
4588 | if (btrfs_root_readonly(root)) | |
4589 | return -EROFS; | |
4590 | ||
e41f941a | 4591 | if (flags & S_VERSION) |
22c44fe6 | 4592 | inode_inc_iversion(inode); |
e41f941a JB |
4593 | if (flags & S_CTIME) |
4594 | inode->i_ctime = *now; | |
4595 | if (flags & S_MTIME) | |
4596 | inode->i_mtime = *now; | |
4597 | if (flags & S_ATIME) | |
4598 | inode->i_atime = *now; | |
4599 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
4600 | } |
4601 | ||
d352ac68 CM |
4602 | /* |
4603 | * find the highest existing sequence number in a directory | |
4604 | * and then set the in-memory index_cnt variable to reflect | |
4605 | * free sequence numbers | |
4606 | */ | |
aec7477b JB |
4607 | static int btrfs_set_inode_index_count(struct inode *inode) |
4608 | { | |
4609 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4610 | struct btrfs_key key, found_key; | |
4611 | struct btrfs_path *path; | |
4612 | struct extent_buffer *leaf; | |
4613 | int ret; | |
4614 | ||
33345d01 | 4615 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4616 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4617 | key.offset = (u64)-1; | |
4618 | ||
4619 | path = btrfs_alloc_path(); | |
4620 | if (!path) | |
4621 | return -ENOMEM; | |
4622 | ||
4623 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4624 | if (ret < 0) | |
4625 | goto out; | |
4626 | /* FIXME: we should be able to handle this */ | |
4627 | if (ret == 0) | |
4628 | goto out; | |
4629 | ret = 0; | |
4630 | ||
4631 | /* | |
4632 | * MAGIC NUMBER EXPLANATION: | |
4633 | * since we search a directory based on f_pos we have to start at 2 | |
4634 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4635 | * else has to start at 2 | |
4636 | */ | |
4637 | if (path->slots[0] == 0) { | |
4638 | BTRFS_I(inode)->index_cnt = 2; | |
4639 | goto out; | |
4640 | } | |
4641 | ||
4642 | path->slots[0]--; | |
4643 | ||
4644 | leaf = path->nodes[0]; | |
4645 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4646 | ||
33345d01 | 4647 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4648 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4649 | BTRFS_I(inode)->index_cnt = 2; | |
4650 | goto out; | |
4651 | } | |
4652 | ||
4653 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4654 | out: | |
4655 | btrfs_free_path(path); | |
4656 | return ret; | |
4657 | } | |
4658 | ||
d352ac68 CM |
4659 | /* |
4660 | * helper to find a free sequence number in a given directory. This current | |
4661 | * code is very simple, later versions will do smarter things in the btree | |
4662 | */ | |
3de4586c | 4663 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4664 | { |
4665 | int ret = 0; | |
4666 | ||
4667 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4668 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4669 | if (ret) { | |
4670 | ret = btrfs_set_inode_index_count(dir); | |
4671 | if (ret) | |
4672 | return ret; | |
4673 | } | |
aec7477b JB |
4674 | } |
4675 | ||
00e4e6b3 | 4676 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4677 | BTRFS_I(dir)->index_cnt++; |
4678 | ||
4679 | return ret; | |
4680 | } | |
4681 | ||
39279cc3 CM |
4682 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4683 | struct btrfs_root *root, | |
aec7477b | 4684 | struct inode *dir, |
9c58309d | 4685 | const char *name, int name_len, |
175a4eb7 AV |
4686 | u64 ref_objectid, u64 objectid, |
4687 | umode_t mode, u64 *index) | |
39279cc3 CM |
4688 | { |
4689 | struct inode *inode; | |
5f39d397 | 4690 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4691 | struct btrfs_key *location; |
5f39d397 | 4692 | struct btrfs_path *path; |
9c58309d CM |
4693 | struct btrfs_inode_ref *ref; |
4694 | struct btrfs_key key[2]; | |
4695 | u32 sizes[2]; | |
4696 | unsigned long ptr; | |
39279cc3 CM |
4697 | int ret; |
4698 | int owner; | |
4699 | ||
5f39d397 | 4700 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4701 | if (!path) |
4702 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4703 | |
39279cc3 | 4704 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4705 | if (!inode) { |
4706 | btrfs_free_path(path); | |
39279cc3 | 4707 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4708 | } |
39279cc3 | 4709 | |
581bb050 LZ |
4710 | /* |
4711 | * we have to initialize this early, so we can reclaim the inode | |
4712 | * number if we fail afterwards in this function. | |
4713 | */ | |
4714 | inode->i_ino = objectid; | |
4715 | ||
aec7477b | 4716 | if (dir) { |
1abe9b8a | 4717 | trace_btrfs_inode_request(dir); |
4718 | ||
3de4586c | 4719 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4720 | if (ret) { |
8fb27640 | 4721 | btrfs_free_path(path); |
09771430 | 4722 | iput(inode); |
aec7477b | 4723 | return ERR_PTR(ret); |
09771430 | 4724 | } |
aec7477b JB |
4725 | } |
4726 | /* | |
4727 | * index_cnt is ignored for everything but a dir, | |
4728 | * btrfs_get_inode_index_count has an explanation for the magic | |
4729 | * number | |
4730 | */ | |
4731 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4732 | BTRFS_I(inode)->root = root; |
e02119d5 | 4733 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4734 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 4735 | |
5dc562c5 JB |
4736 | /* |
4737 | * We could have gotten an inode number from somebody who was fsynced | |
4738 | * and then removed in this same transaction, so let's just set full | |
4739 | * sync since it will be a full sync anyway and this will blow away the | |
4740 | * old info in the log. | |
4741 | */ | |
4742 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
4743 | ||
569254b0 | 4744 | if (S_ISDIR(mode)) |
39279cc3 CM |
4745 | owner = 0; |
4746 | else | |
4747 | owner = 1; | |
9c58309d CM |
4748 | |
4749 | key[0].objectid = objectid; | |
4750 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4751 | key[0].offset = 0; | |
4752 | ||
f186373f MF |
4753 | /* |
4754 | * Start new inodes with an inode_ref. This is slightly more | |
4755 | * efficient for small numbers of hard links since they will | |
4756 | * be packed into one item. Extended refs will kick in if we | |
4757 | * add more hard links than can fit in the ref item. | |
4758 | */ | |
9c58309d CM |
4759 | key[1].objectid = objectid; |
4760 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4761 | key[1].offset = ref_objectid; | |
4762 | ||
4763 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4764 | sizes[1] = name_len + sizeof(*ref); | |
4765 | ||
b9473439 | 4766 | path->leave_spinning = 1; |
9c58309d CM |
4767 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4768 | if (ret != 0) | |
5f39d397 CM |
4769 | goto fail; |
4770 | ||
ecc11fab | 4771 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4772 | inode_set_bytes(inode, 0); |
39279cc3 | 4773 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4774 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4775 | struct btrfs_inode_item); | |
293f7e07 LZ |
4776 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
4777 | sizeof(*inode_item)); | |
e02119d5 | 4778 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4779 | |
4780 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4781 | struct btrfs_inode_ref); | |
4782 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4783 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4784 | ptr = (unsigned long)(ref + 1); |
4785 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4786 | ||
5f39d397 CM |
4787 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4788 | btrfs_free_path(path); | |
4789 | ||
39279cc3 CM |
4790 | location = &BTRFS_I(inode)->location; |
4791 | location->objectid = objectid; | |
39279cc3 CM |
4792 | location->offset = 0; |
4793 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4794 | ||
6cbff00f CH |
4795 | btrfs_inherit_iflags(inode, dir); |
4796 | ||
569254b0 | 4797 | if (S_ISREG(mode)) { |
94272164 CM |
4798 | if (btrfs_test_opt(root, NODATASUM)) |
4799 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4800 | if (btrfs_test_opt(root, NODATACOW) || |
4801 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4802 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4803 | } | |
4804 | ||
39279cc3 | 4805 | insert_inode_hash(inode); |
5d4f98a2 | 4806 | inode_tree_add(inode); |
1abe9b8a | 4807 | |
4808 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4809 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4810 | |
8ea05e3a AB |
4811 | btrfs_update_root_times(trans, root); |
4812 | ||
39279cc3 | 4813 | return inode; |
5f39d397 | 4814 | fail: |
aec7477b JB |
4815 | if (dir) |
4816 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4817 | btrfs_free_path(path); |
09771430 | 4818 | iput(inode); |
5f39d397 | 4819 | return ERR_PTR(ret); |
39279cc3 CM |
4820 | } |
4821 | ||
4822 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4823 | { | |
4824 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4825 | } | |
4826 | ||
d352ac68 CM |
4827 | /* |
4828 | * utility function to add 'inode' into 'parent_inode' with | |
4829 | * a give name and a given sequence number. | |
4830 | * if 'add_backref' is true, also insert a backref from the | |
4831 | * inode to the parent directory. | |
4832 | */ | |
e02119d5 CM |
4833 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4834 | struct inode *parent_inode, struct inode *inode, | |
4835 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4836 | { |
4df27c4d | 4837 | int ret = 0; |
39279cc3 | 4838 | struct btrfs_key key; |
e02119d5 | 4839 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4840 | u64 ino = btrfs_ino(inode); |
4841 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4842 | |
33345d01 | 4843 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4844 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4845 | } else { | |
33345d01 | 4846 | key.objectid = ino; |
4df27c4d YZ |
4847 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4848 | key.offset = 0; | |
4849 | } | |
4850 | ||
33345d01 | 4851 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4852 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4853 | key.objectid, root->root_key.objectid, | |
33345d01 | 4854 | parent_ino, index, name, name_len); |
4df27c4d | 4855 | } else if (add_backref) { |
33345d01 LZ |
4856 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4857 | parent_ino, index); | |
4df27c4d | 4858 | } |
39279cc3 | 4859 | |
79787eaa JM |
4860 | /* Nothing to clean up yet */ |
4861 | if (ret) | |
4862 | return ret; | |
4df27c4d | 4863 | |
79787eaa JM |
4864 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4865 | parent_inode, &key, | |
4866 | btrfs_inode_type(inode), index); | |
4867 | if (ret == -EEXIST) | |
4868 | goto fail_dir_item; | |
4869 | else if (ret) { | |
4870 | btrfs_abort_transaction(trans, root, ret); | |
4871 | return ret; | |
39279cc3 | 4872 | } |
79787eaa JM |
4873 | |
4874 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4875 | name_len * 2); | |
0c4d2d95 | 4876 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4877 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4878 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4879 | if (ret) | |
4880 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4881 | return ret; |
fe66a05a CM |
4882 | |
4883 | fail_dir_item: | |
4884 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4885 | u64 local_index; | |
4886 | int err; | |
4887 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4888 | key.objectid, root->root_key.objectid, | |
4889 | parent_ino, &local_index, name, name_len); | |
4890 | ||
4891 | } else if (add_backref) { | |
4892 | u64 local_index; | |
4893 | int err; | |
4894 | ||
4895 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
4896 | ino, parent_ino, &local_index); | |
4897 | } | |
4898 | return ret; | |
39279cc3 CM |
4899 | } |
4900 | ||
4901 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4902 | struct inode *dir, struct dentry *dentry, |
4903 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4904 | { |
a1b075d2 JB |
4905 | int err = btrfs_add_link(trans, dir, inode, |
4906 | dentry->d_name.name, dentry->d_name.len, | |
4907 | backref, index); | |
39279cc3 CM |
4908 | if (err > 0) |
4909 | err = -EEXIST; | |
4910 | return err; | |
4911 | } | |
4912 | ||
618e21d5 | 4913 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 4914 | umode_t mode, dev_t rdev) |
618e21d5 JB |
4915 | { |
4916 | struct btrfs_trans_handle *trans; | |
4917 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4918 | struct inode *inode = NULL; |
618e21d5 JB |
4919 | int err; |
4920 | int drop_inode = 0; | |
4921 | u64 objectid; | |
00e4e6b3 | 4922 | u64 index = 0; |
618e21d5 JB |
4923 | |
4924 | if (!new_valid_dev(rdev)) | |
4925 | return -EINVAL; | |
4926 | ||
9ed74f2d JB |
4927 | /* |
4928 | * 2 for inode item and ref | |
4929 | * 2 for dir items | |
4930 | * 1 for xattr if selinux is on | |
4931 | */ | |
a22285a6 YZ |
4932 | trans = btrfs_start_transaction(root, 5); |
4933 | if (IS_ERR(trans)) | |
4934 | return PTR_ERR(trans); | |
1832a6d5 | 4935 | |
581bb050 LZ |
4936 | err = btrfs_find_free_ino(root, &objectid); |
4937 | if (err) | |
4938 | goto out_unlock; | |
4939 | ||
aec7477b | 4940 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4941 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4942 | mode, &index); |
7cf96da3 TI |
4943 | if (IS_ERR(inode)) { |
4944 | err = PTR_ERR(inode); | |
618e21d5 | 4945 | goto out_unlock; |
7cf96da3 | 4946 | } |
618e21d5 | 4947 | |
2a7dba39 | 4948 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4949 | if (err) { |
4950 | drop_inode = 1; | |
4951 | goto out_unlock; | |
4952 | } | |
4953 | ||
2794ed01 FB |
4954 | err = btrfs_update_inode(trans, root, inode); |
4955 | if (err) { | |
4956 | drop_inode = 1; | |
4957 | goto out_unlock; | |
4958 | } | |
4959 | ||
ad19db71 CS |
4960 | /* |
4961 | * If the active LSM wants to access the inode during | |
4962 | * d_instantiate it needs these. Smack checks to see | |
4963 | * if the filesystem supports xattrs by looking at the | |
4964 | * ops vector. | |
4965 | */ | |
4966 | ||
4967 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4968 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4969 | if (err) |
4970 | drop_inode = 1; | |
4971 | else { | |
618e21d5 | 4972 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4973 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4974 | d_instantiate(dentry, inode); |
618e21d5 | 4975 | } |
618e21d5 | 4976 | out_unlock: |
7ad85bb7 | 4977 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4978 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
4979 | if (drop_inode) { |
4980 | inode_dec_link_count(inode); | |
4981 | iput(inode); | |
4982 | } | |
618e21d5 JB |
4983 | return err; |
4984 | } | |
4985 | ||
39279cc3 | 4986 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 4987 | umode_t mode, bool excl) |
39279cc3 CM |
4988 | { |
4989 | struct btrfs_trans_handle *trans; | |
4990 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4991 | struct inode *inode = NULL; |
43baa579 | 4992 | int drop_inode_on_err = 0; |
a22285a6 | 4993 | int err; |
39279cc3 | 4994 | u64 objectid; |
00e4e6b3 | 4995 | u64 index = 0; |
39279cc3 | 4996 | |
9ed74f2d JB |
4997 | /* |
4998 | * 2 for inode item and ref | |
4999 | * 2 for dir items | |
5000 | * 1 for xattr if selinux is on | |
5001 | */ | |
a22285a6 YZ |
5002 | trans = btrfs_start_transaction(root, 5); |
5003 | if (IS_ERR(trans)) | |
5004 | return PTR_ERR(trans); | |
9ed74f2d | 5005 | |
581bb050 LZ |
5006 | err = btrfs_find_free_ino(root, &objectid); |
5007 | if (err) | |
5008 | goto out_unlock; | |
5009 | ||
aec7477b | 5010 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5011 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5012 | mode, &index); |
7cf96da3 TI |
5013 | if (IS_ERR(inode)) { |
5014 | err = PTR_ERR(inode); | |
39279cc3 | 5015 | goto out_unlock; |
7cf96da3 | 5016 | } |
43baa579 | 5017 | drop_inode_on_err = 1; |
39279cc3 | 5018 | |
2a7dba39 | 5019 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
43baa579 | 5020 | if (err) |
33268eaf | 5021 | goto out_unlock; |
33268eaf | 5022 | |
ad19db71 CS |
5023 | /* |
5024 | * If the active LSM wants to access the inode during | |
5025 | * d_instantiate it needs these. Smack checks to see | |
5026 | * if the filesystem supports xattrs by looking at the | |
5027 | * ops vector. | |
5028 | */ | |
5029 | inode->i_fop = &btrfs_file_operations; | |
5030 | inode->i_op = &btrfs_file_inode_operations; | |
5031 | ||
a1b075d2 | 5032 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 5033 | if (err) |
43baa579 FB |
5034 | goto out_unlock; |
5035 | ||
5036 | inode->i_mapping->a_ops = &btrfs_aops; | |
5037 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; | |
5038 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
5039 | d_instantiate(dentry, inode); | |
5040 | ||
39279cc3 | 5041 | out_unlock: |
7ad85bb7 | 5042 | btrfs_end_transaction(trans, root); |
43baa579 | 5043 | if (err && drop_inode_on_err) { |
39279cc3 CM |
5044 | inode_dec_link_count(inode); |
5045 | iput(inode); | |
5046 | } | |
b53d3f5d | 5047 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5048 | return err; |
5049 | } | |
5050 | ||
5051 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5052 | struct dentry *dentry) | |
5053 | { | |
5054 | struct btrfs_trans_handle *trans; | |
5055 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5056 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5057 | u64 index; |
39279cc3 CM |
5058 | int err; |
5059 | int drop_inode = 0; | |
5060 | ||
4a8be425 TH |
5061 | /* do not allow sys_link's with other subvols of the same device */ |
5062 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5063 | return -EXDEV; |
4a8be425 | 5064 | |
f186373f | 5065 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5066 | return -EMLINK; |
4a8be425 | 5067 | |
3de4586c | 5068 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5069 | if (err) |
5070 | goto fail; | |
5071 | ||
a22285a6 | 5072 | /* |
7e6b6465 | 5073 | * 2 items for inode and inode ref |
a22285a6 | 5074 | * 2 items for dir items |
7e6b6465 | 5075 | * 1 item for parent inode |
a22285a6 | 5076 | */ |
7e6b6465 | 5077 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5078 | if (IS_ERR(trans)) { |
5079 | err = PTR_ERR(trans); | |
5080 | goto fail; | |
5081 | } | |
5f39d397 | 5082 | |
3153495d | 5083 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5084 | inode_inc_iversion(inode); |
3153495d | 5085 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5086 | ihold(inode); |
aec7477b | 5087 | |
a1b075d2 | 5088 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5089 | |
a5719521 | 5090 | if (err) { |
54aa1f4d | 5091 | drop_inode = 1; |
a5719521 | 5092 | } else { |
10d9f309 | 5093 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5094 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5095 | if (err) |
5096 | goto fail; | |
08c422c2 | 5097 | d_instantiate(dentry, inode); |
6a912213 | 5098 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5099 | } |
39279cc3 | 5100 | |
7ad85bb7 | 5101 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5102 | fail: |
39279cc3 CM |
5103 | if (drop_inode) { |
5104 | inode_dec_link_count(inode); | |
5105 | iput(inode); | |
5106 | } | |
b53d3f5d | 5107 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5108 | return err; |
5109 | } | |
5110 | ||
18bb1db3 | 5111 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5112 | { |
b9d86667 | 5113 | struct inode *inode = NULL; |
39279cc3 CM |
5114 | struct btrfs_trans_handle *trans; |
5115 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5116 | int err = 0; | |
5117 | int drop_on_err = 0; | |
b9d86667 | 5118 | u64 objectid = 0; |
00e4e6b3 | 5119 | u64 index = 0; |
39279cc3 | 5120 | |
9ed74f2d JB |
5121 | /* |
5122 | * 2 items for inode and ref | |
5123 | * 2 items for dir items | |
5124 | * 1 for xattr if selinux is on | |
5125 | */ | |
a22285a6 YZ |
5126 | trans = btrfs_start_transaction(root, 5); |
5127 | if (IS_ERR(trans)) | |
5128 | return PTR_ERR(trans); | |
39279cc3 | 5129 | |
581bb050 LZ |
5130 | err = btrfs_find_free_ino(root, &objectid); |
5131 | if (err) | |
5132 | goto out_fail; | |
5133 | ||
aec7477b | 5134 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5135 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5136 | S_IFDIR | mode, &index); |
39279cc3 CM |
5137 | if (IS_ERR(inode)) { |
5138 | err = PTR_ERR(inode); | |
5139 | goto out_fail; | |
5140 | } | |
5f39d397 | 5141 | |
39279cc3 | 5142 | drop_on_err = 1; |
33268eaf | 5143 | |
2a7dba39 | 5144 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5145 | if (err) |
5146 | goto out_fail; | |
5147 | ||
39279cc3 CM |
5148 | inode->i_op = &btrfs_dir_inode_operations; |
5149 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5150 | |
dbe674a9 | 5151 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5152 | err = btrfs_update_inode(trans, root, inode); |
5153 | if (err) | |
5154 | goto out_fail; | |
5f39d397 | 5155 | |
a1b075d2 JB |
5156 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5157 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5158 | if (err) |
5159 | goto out_fail; | |
5f39d397 | 5160 | |
39279cc3 CM |
5161 | d_instantiate(dentry, inode); |
5162 | drop_on_err = 0; | |
39279cc3 CM |
5163 | |
5164 | out_fail: | |
7ad85bb7 | 5165 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5166 | if (drop_on_err) |
5167 | iput(inode); | |
b53d3f5d | 5168 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
5169 | return err; |
5170 | } | |
5171 | ||
d352ac68 CM |
5172 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5173 | * and an extent that you want to insert, deal with overlap and insert | |
5174 | * the new extent into the tree. | |
5175 | */ | |
3b951516 CM |
5176 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5177 | struct extent_map *existing, | |
e6dcd2dc CM |
5178 | struct extent_map *em, |
5179 | u64 map_start, u64 map_len) | |
3b951516 CM |
5180 | { |
5181 | u64 start_diff; | |
3b951516 | 5182 | |
e6dcd2dc CM |
5183 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5184 | start_diff = map_start - em->start; | |
5185 | em->start = map_start; | |
5186 | em->len = map_len; | |
c8b97818 CM |
5187 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5188 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5189 | em->block_start += start_diff; |
c8b97818 CM |
5190 | em->block_len -= start_diff; |
5191 | } | |
e6dcd2dc | 5192 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5193 | } |
5194 | ||
c8b97818 CM |
5195 | static noinline int uncompress_inline(struct btrfs_path *path, |
5196 | struct inode *inode, struct page *page, | |
5197 | size_t pg_offset, u64 extent_offset, | |
5198 | struct btrfs_file_extent_item *item) | |
5199 | { | |
5200 | int ret; | |
5201 | struct extent_buffer *leaf = path->nodes[0]; | |
5202 | char *tmp; | |
5203 | size_t max_size; | |
5204 | unsigned long inline_size; | |
5205 | unsigned long ptr; | |
261507a0 | 5206 | int compress_type; |
c8b97818 CM |
5207 | |
5208 | WARN_ON(pg_offset != 0); | |
261507a0 | 5209 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5210 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5211 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5212 | btrfs_item_nr(leaf, path->slots[0])); | |
5213 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5214 | if (!tmp) |
5215 | return -ENOMEM; | |
c8b97818 CM |
5216 | ptr = btrfs_file_extent_inline_start(item); |
5217 | ||
5218 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5219 | ||
5b050f04 | 5220 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5221 | ret = btrfs_decompress(compress_type, tmp, page, |
5222 | extent_offset, inline_size, max_size); | |
c8b97818 | 5223 | if (ret) { |
7ac687d9 | 5224 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5225 | unsigned long copy_size = min_t(u64, |
5226 | PAGE_CACHE_SIZE - pg_offset, | |
5227 | max_size - extent_offset); | |
5228 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5229 | kunmap_atomic(kaddr); |
c8b97818 CM |
5230 | } |
5231 | kfree(tmp); | |
5232 | return 0; | |
5233 | } | |
5234 | ||
d352ac68 CM |
5235 | /* |
5236 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5237 | * the ugly parts come from merging extents from the disk with the in-ram |
5238 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5239 | * where the in-ram extents might be locked pending data=ordered completion. |
5240 | * | |
5241 | * This also copies inline extents directly into the page. | |
5242 | */ | |
d397712b | 5243 | |
a52d9a80 | 5244 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5245 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5246 | int create) |
5247 | { | |
5248 | int ret; | |
5249 | int err = 0; | |
db94535d | 5250 | u64 bytenr; |
a52d9a80 CM |
5251 | u64 extent_start = 0; |
5252 | u64 extent_end = 0; | |
33345d01 | 5253 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5254 | u32 found_type; |
f421950f | 5255 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5256 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5257 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5258 | struct extent_buffer *leaf; |
5259 | struct btrfs_key found_key; | |
a52d9a80 CM |
5260 | struct extent_map *em = NULL; |
5261 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5262 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5263 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5264 | int compress_type; |
a52d9a80 | 5265 | |
a52d9a80 | 5266 | again: |
890871be | 5267 | read_lock(&em_tree->lock); |
d1310b2e | 5268 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5269 | if (em) |
5270 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5271 | read_unlock(&em_tree->lock); |
d1310b2e | 5272 | |
a52d9a80 | 5273 | if (em) { |
e1c4b745 CM |
5274 | if (em->start > start || em->start + em->len <= start) |
5275 | free_extent_map(em); | |
5276 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5277 | free_extent_map(em); |
5278 | else | |
5279 | goto out; | |
a52d9a80 | 5280 | } |
172ddd60 | 5281 | em = alloc_extent_map(); |
a52d9a80 | 5282 | if (!em) { |
d1310b2e CM |
5283 | err = -ENOMEM; |
5284 | goto out; | |
a52d9a80 | 5285 | } |
e6dcd2dc | 5286 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5287 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5288 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5289 | em->len = (u64)-1; |
c8b97818 | 5290 | em->block_len = (u64)-1; |
f421950f CM |
5291 | |
5292 | if (!path) { | |
5293 | path = btrfs_alloc_path(); | |
026fd317 JB |
5294 | if (!path) { |
5295 | err = -ENOMEM; | |
5296 | goto out; | |
5297 | } | |
5298 | /* | |
5299 | * Chances are we'll be called again, so go ahead and do | |
5300 | * readahead | |
5301 | */ | |
5302 | path->reada = 1; | |
f421950f CM |
5303 | } |
5304 | ||
179e29e4 CM |
5305 | ret = btrfs_lookup_file_extent(trans, root, path, |
5306 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5307 | if (ret < 0) { |
5308 | err = ret; | |
5309 | goto out; | |
5310 | } | |
5311 | ||
5312 | if (ret != 0) { | |
5313 | if (path->slots[0] == 0) | |
5314 | goto not_found; | |
5315 | path->slots[0]--; | |
5316 | } | |
5317 | ||
5f39d397 CM |
5318 | leaf = path->nodes[0]; |
5319 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5320 | struct btrfs_file_extent_item); |
a52d9a80 | 5321 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5322 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5323 | found_type = btrfs_key_type(&found_key); | |
5324 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5325 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5326 | goto not_found; | |
5327 | } | |
5328 | ||
5f39d397 CM |
5329 | found_type = btrfs_file_extent_type(leaf, item); |
5330 | extent_start = found_key.offset; | |
261507a0 | 5331 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5332 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5333 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5334 | extent_end = extent_start + |
db94535d | 5335 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5336 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5337 | size_t size; | |
5338 | size = btrfs_file_extent_inline_len(leaf, item); | |
5339 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5340 | ~((u64)root->sectorsize - 1); | |
5341 | } | |
5342 | ||
5343 | if (start >= extent_end) { | |
5344 | path->slots[0]++; | |
5345 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5346 | ret = btrfs_next_leaf(root, path); | |
5347 | if (ret < 0) { | |
5348 | err = ret; | |
5349 | goto out; | |
a52d9a80 | 5350 | } |
9036c102 YZ |
5351 | if (ret > 0) |
5352 | goto not_found; | |
5353 | leaf = path->nodes[0]; | |
a52d9a80 | 5354 | } |
9036c102 YZ |
5355 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5356 | if (found_key.objectid != objectid || | |
5357 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5358 | goto not_found; | |
5359 | if (start + len <= found_key.offset) | |
5360 | goto not_found; | |
5361 | em->start = start; | |
5362 | em->len = found_key.offset - start; | |
5363 | goto not_found_em; | |
5364 | } | |
5365 | ||
d899e052 YZ |
5366 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5367 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5368 | em->start = extent_start; |
5369 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5370 | em->orig_start = extent_start - |
5371 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5372 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5373 | if (bytenr == 0) { | |
5f39d397 | 5374 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5375 | goto insert; |
5376 | } | |
261507a0 | 5377 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5378 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5379 | em->compress_type = compress_type; |
c8b97818 CM |
5380 | em->block_start = bytenr; |
5381 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5382 | item); | |
5383 | } else { | |
5384 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5385 | em->block_start = bytenr; | |
5386 | em->block_len = em->len; | |
d899e052 YZ |
5387 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5388 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5389 | } |
a52d9a80 CM |
5390 | goto insert; |
5391 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5392 | unsigned long ptr; |
a52d9a80 | 5393 | char *map; |
3326d1b0 CM |
5394 | size_t size; |
5395 | size_t extent_offset; | |
5396 | size_t copy_size; | |
a52d9a80 | 5397 | |
689f9346 | 5398 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5399 | if (!page || create) { |
689f9346 | 5400 | em->start = extent_start; |
9036c102 | 5401 | em->len = extent_end - extent_start; |
689f9346 Y |
5402 | goto out; |
5403 | } | |
5f39d397 | 5404 | |
9036c102 YZ |
5405 | size = btrfs_file_extent_inline_len(leaf, item); |
5406 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5407 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5408 | size - extent_offset); |
3326d1b0 | 5409 | em->start = extent_start + extent_offset; |
70dec807 CM |
5410 | em->len = (copy_size + root->sectorsize - 1) & |
5411 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5412 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5413 | if (compress_type) { |
c8b97818 | 5414 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5415 | em->compress_type = compress_type; |
5416 | } | |
689f9346 | 5417 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5418 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5419 | if (btrfs_file_extent_compression(leaf, item) != |
5420 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5421 | ret = uncompress_inline(path, inode, page, |
5422 | pg_offset, | |
5423 | extent_offset, item); | |
79787eaa | 5424 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5425 | } else { |
5426 | map = kmap(page); | |
5427 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5428 | copy_size); | |
93c82d57 CM |
5429 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5430 | memset(map + pg_offset + copy_size, 0, | |
5431 | PAGE_CACHE_SIZE - pg_offset - | |
5432 | copy_size); | |
5433 | } | |
c8b97818 CM |
5434 | kunmap(page); |
5435 | } | |
179e29e4 CM |
5436 | flush_dcache_page(page); |
5437 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5438 | BUG(); |
179e29e4 CM |
5439 | if (!trans) { |
5440 | kunmap(page); | |
5441 | free_extent_map(em); | |
5442 | em = NULL; | |
ff5714cc | 5443 | |
b3b4aa74 | 5444 | btrfs_release_path(path); |
7a7eaa40 | 5445 | trans = btrfs_join_transaction(root); |
ff5714cc | 5446 | |
3612b495 TI |
5447 | if (IS_ERR(trans)) |
5448 | return ERR_CAST(trans); | |
179e29e4 CM |
5449 | goto again; |
5450 | } | |
c8b97818 | 5451 | map = kmap(page); |
70dec807 | 5452 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5453 | copy_size); |
c8b97818 | 5454 | kunmap(page); |
179e29e4 | 5455 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5456 | } |
d1310b2e | 5457 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5458 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5459 | goto insert; |
5460 | } else { | |
31b1a2bd | 5461 | WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5462 | } |
5463 | not_found: | |
5464 | em->start = start; | |
d1310b2e | 5465 | em->len = len; |
a52d9a80 | 5466 | not_found_em: |
5f39d397 | 5467 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5468 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5469 | insert: |
b3b4aa74 | 5470 | btrfs_release_path(path); |
d1310b2e | 5471 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5472 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5473 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5474 | (unsigned long long)em->len, | |
5475 | (unsigned long long)start, | |
5476 | (unsigned long long)len); | |
a52d9a80 CM |
5477 | err = -EIO; |
5478 | goto out; | |
5479 | } | |
d1310b2e CM |
5480 | |
5481 | err = 0; | |
890871be | 5482 | write_lock(&em_tree->lock); |
a52d9a80 | 5483 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5484 | /* it is possible that someone inserted the extent into the tree |
5485 | * while we had the lock dropped. It is also possible that | |
5486 | * an overlapping map exists in the tree | |
5487 | */ | |
a52d9a80 | 5488 | if (ret == -EEXIST) { |
3b951516 | 5489 | struct extent_map *existing; |
e6dcd2dc CM |
5490 | |
5491 | ret = 0; | |
5492 | ||
3b951516 | 5493 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5494 | if (existing && (existing->start > start || |
5495 | existing->start + existing->len <= start)) { | |
5496 | free_extent_map(existing); | |
5497 | existing = NULL; | |
5498 | } | |
3b951516 CM |
5499 | if (!existing) { |
5500 | existing = lookup_extent_mapping(em_tree, em->start, | |
5501 | em->len); | |
5502 | if (existing) { | |
5503 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5504 | em, start, |
5505 | root->sectorsize); | |
3b951516 CM |
5506 | free_extent_map(existing); |
5507 | if (err) { | |
5508 | free_extent_map(em); | |
5509 | em = NULL; | |
5510 | } | |
5511 | } else { | |
5512 | err = -EIO; | |
3b951516 CM |
5513 | free_extent_map(em); |
5514 | em = NULL; | |
5515 | } | |
5516 | } else { | |
5517 | free_extent_map(em); | |
5518 | em = existing; | |
e6dcd2dc | 5519 | err = 0; |
a52d9a80 | 5520 | } |
a52d9a80 | 5521 | } |
890871be | 5522 | write_unlock(&em_tree->lock); |
a52d9a80 | 5523 | out: |
1abe9b8a | 5524 | |
f0bd95ea TI |
5525 | if (em) |
5526 | trace_btrfs_get_extent(root, em); | |
1abe9b8a | 5527 | |
f421950f CM |
5528 | if (path) |
5529 | btrfs_free_path(path); | |
a52d9a80 CM |
5530 | if (trans) { |
5531 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5532 | if (!err) |
a52d9a80 CM |
5533 | err = ret; |
5534 | } | |
a52d9a80 CM |
5535 | if (err) { |
5536 | free_extent_map(em); | |
a52d9a80 CM |
5537 | return ERR_PTR(err); |
5538 | } | |
79787eaa | 5539 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5540 | return em; |
5541 | } | |
5542 | ||
ec29ed5b CM |
5543 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5544 | size_t pg_offset, u64 start, u64 len, | |
5545 | int create) | |
5546 | { | |
5547 | struct extent_map *em; | |
5548 | struct extent_map *hole_em = NULL; | |
5549 | u64 range_start = start; | |
5550 | u64 end; | |
5551 | u64 found; | |
5552 | u64 found_end; | |
5553 | int err = 0; | |
5554 | ||
5555 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5556 | if (IS_ERR(em)) | |
5557 | return em; | |
5558 | if (em) { | |
5559 | /* | |
5560 | * if our em maps to a hole, there might | |
5561 | * actually be delalloc bytes behind it | |
5562 | */ | |
5563 | if (em->block_start != EXTENT_MAP_HOLE) | |
5564 | return em; | |
5565 | else | |
5566 | hole_em = em; | |
5567 | } | |
5568 | ||
5569 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5570 | end = start + len; | |
5571 | if (end < start) | |
5572 | end = (u64)-1; | |
5573 | else | |
5574 | end -= 1; | |
5575 | ||
5576 | em = NULL; | |
5577 | ||
5578 | /* ok, we didn't find anything, lets look for delalloc */ | |
5579 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5580 | end, len, EXTENT_DELALLOC, 1); | |
5581 | found_end = range_start + found; | |
5582 | if (found_end < range_start) | |
5583 | found_end = (u64)-1; | |
5584 | ||
5585 | /* | |
5586 | * we didn't find anything useful, return | |
5587 | * the original results from get_extent() | |
5588 | */ | |
5589 | if (range_start > end || found_end <= start) { | |
5590 | em = hole_em; | |
5591 | hole_em = NULL; | |
5592 | goto out; | |
5593 | } | |
5594 | ||
5595 | /* adjust the range_start to make sure it doesn't | |
5596 | * go backwards from the start they passed in | |
5597 | */ | |
5598 | range_start = max(start,range_start); | |
5599 | found = found_end - range_start; | |
5600 | ||
5601 | if (found > 0) { | |
5602 | u64 hole_start = start; | |
5603 | u64 hole_len = len; | |
5604 | ||
172ddd60 | 5605 | em = alloc_extent_map(); |
ec29ed5b CM |
5606 | if (!em) { |
5607 | err = -ENOMEM; | |
5608 | goto out; | |
5609 | } | |
5610 | /* | |
5611 | * when btrfs_get_extent can't find anything it | |
5612 | * returns one huge hole | |
5613 | * | |
5614 | * make sure what it found really fits our range, and | |
5615 | * adjust to make sure it is based on the start from | |
5616 | * the caller | |
5617 | */ | |
5618 | if (hole_em) { | |
5619 | u64 calc_end = extent_map_end(hole_em); | |
5620 | ||
5621 | if (calc_end <= start || (hole_em->start > end)) { | |
5622 | free_extent_map(hole_em); | |
5623 | hole_em = NULL; | |
5624 | } else { | |
5625 | hole_start = max(hole_em->start, start); | |
5626 | hole_len = calc_end - hole_start; | |
5627 | } | |
5628 | } | |
5629 | em->bdev = NULL; | |
5630 | if (hole_em && range_start > hole_start) { | |
5631 | /* our hole starts before our delalloc, so we | |
5632 | * have to return just the parts of the hole | |
5633 | * that go until the delalloc starts | |
5634 | */ | |
5635 | em->len = min(hole_len, | |
5636 | range_start - hole_start); | |
5637 | em->start = hole_start; | |
5638 | em->orig_start = hole_start; | |
5639 | /* | |
5640 | * don't adjust block start at all, | |
5641 | * it is fixed at EXTENT_MAP_HOLE | |
5642 | */ | |
5643 | em->block_start = hole_em->block_start; | |
5644 | em->block_len = hole_len; | |
5645 | } else { | |
5646 | em->start = range_start; | |
5647 | em->len = found; | |
5648 | em->orig_start = range_start; | |
5649 | em->block_start = EXTENT_MAP_DELALLOC; | |
5650 | em->block_len = found; | |
5651 | } | |
5652 | } else if (hole_em) { | |
5653 | return hole_em; | |
5654 | } | |
5655 | out: | |
5656 | ||
5657 | free_extent_map(hole_em); | |
5658 | if (err) { | |
5659 | free_extent_map(em); | |
5660 | return ERR_PTR(err); | |
5661 | } | |
5662 | return em; | |
5663 | } | |
5664 | ||
4b46fce2 | 5665 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5666 | struct extent_map *em, |
4b46fce2 JB |
5667 | u64 start, u64 len) |
5668 | { | |
5669 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5670 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5671 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5672 | struct btrfs_key ins; | |
5673 | u64 alloc_hint; | |
5674 | int ret; | |
16d299ac | 5675 | bool insert = false; |
4b46fce2 | 5676 | |
16d299ac JB |
5677 | /* |
5678 | * Ok if the extent map we looked up is a hole and is for the exact | |
5679 | * range we want, there is no reason to allocate a new one, however if | |
5680 | * it is not right then we need to free this one and drop the cache for | |
5681 | * our range. | |
5682 | */ | |
5683 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5684 | em->len != len) { | |
5685 | free_extent_map(em); | |
5686 | em = NULL; | |
5687 | insert = true; | |
5688 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5689 | } | |
4b46fce2 | 5690 | |
7a7eaa40 | 5691 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5692 | if (IS_ERR(trans)) |
5693 | return ERR_CAST(trans); | |
4b46fce2 JB |
5694 | |
5695 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5696 | ||
5697 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5698 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5699 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5700 | if (ret) { |
5701 | em = ERR_PTR(ret); | |
5702 | goto out; | |
5703 | } | |
5704 | ||
4b46fce2 | 5705 | if (!em) { |
172ddd60 | 5706 | em = alloc_extent_map(); |
16d299ac JB |
5707 | if (!em) { |
5708 | em = ERR_PTR(-ENOMEM); | |
5709 | goto out; | |
5710 | } | |
4b46fce2 JB |
5711 | } |
5712 | ||
5713 | em->start = start; | |
5714 | em->orig_start = em->start; | |
5715 | em->len = ins.offset; | |
5716 | ||
5717 | em->block_start = ins.objectid; | |
5718 | em->block_len = ins.offset; | |
5719 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5720 | |
5721 | /* | |
5722 | * We need to do this because if we're using the original em we searched | |
5723 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5724 | */ | |
5725 | em->flags = 0; | |
4b46fce2 JB |
5726 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5727 | ||
16d299ac | 5728 | while (insert) { |
4b46fce2 JB |
5729 | write_lock(&em_tree->lock); |
5730 | ret = add_extent_mapping(em_tree, em); | |
5731 | write_unlock(&em_tree->lock); | |
5732 | if (ret != -EEXIST) | |
5733 | break; | |
5734 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5735 | } | |
5736 | ||
5737 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5738 | ins.offset, ins.offset, 0); | |
5739 | if (ret) { | |
5740 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5741 | em = ERR_PTR(ret); | |
5742 | } | |
5743 | out: | |
5744 | btrfs_end_transaction(trans, root); | |
5745 | return em; | |
5746 | } | |
5747 | ||
46bfbb5c CM |
5748 | /* |
5749 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5750 | * block must be cow'd | |
5751 | */ | |
5752 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5753 | struct inode *inode, u64 offset, u64 len) | |
5754 | { | |
5755 | struct btrfs_path *path; | |
5756 | int ret; | |
5757 | struct extent_buffer *leaf; | |
5758 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5759 | struct btrfs_file_extent_item *fi; | |
5760 | struct btrfs_key key; | |
5761 | u64 disk_bytenr; | |
5762 | u64 backref_offset; | |
5763 | u64 extent_end; | |
5764 | u64 num_bytes; | |
5765 | int slot; | |
5766 | int found_type; | |
5767 | ||
5768 | path = btrfs_alloc_path(); | |
5769 | if (!path) | |
5770 | return -ENOMEM; | |
5771 | ||
33345d01 | 5772 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5773 | offset, 0); |
5774 | if (ret < 0) | |
5775 | goto out; | |
5776 | ||
5777 | slot = path->slots[0]; | |
5778 | if (ret == 1) { | |
5779 | if (slot == 0) { | |
5780 | /* can't find the item, must cow */ | |
5781 | ret = 0; | |
5782 | goto out; | |
5783 | } | |
5784 | slot--; | |
5785 | } | |
5786 | ret = 0; | |
5787 | leaf = path->nodes[0]; | |
5788 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5789 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5790 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5791 | /* not our file or wrong item type, must cow */ | |
5792 | goto out; | |
5793 | } | |
5794 | ||
5795 | if (key.offset > offset) { | |
5796 | /* Wrong offset, must cow */ | |
5797 | goto out; | |
5798 | } | |
5799 | ||
5800 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5801 | found_type = btrfs_file_extent_type(leaf, fi); | |
5802 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5803 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5804 | /* not a regular extent, must cow */ | |
5805 | goto out; | |
5806 | } | |
5807 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5808 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5809 | ||
5810 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5811 | if (extent_end < offset + len) { | |
5812 | /* extent doesn't include our full range, must cow */ | |
5813 | goto out; | |
5814 | } | |
5815 | ||
5816 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5817 | goto out; | |
5818 | ||
5819 | /* | |
5820 | * look for other files referencing this extent, if we | |
5821 | * find any we must cow | |
5822 | */ | |
33345d01 | 5823 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5824 | key.offset - backref_offset, disk_bytenr)) |
5825 | goto out; | |
5826 | ||
5827 | /* | |
5828 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5829 | * in this extent we are about to write. If there | |
5830 | * are any csums in that range we have to cow in order | |
5831 | * to keep the csums correct | |
5832 | */ | |
5833 | disk_bytenr += backref_offset; | |
5834 | disk_bytenr += offset - key.offset; | |
5835 | num_bytes = min(offset + len, extent_end) - offset; | |
5836 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5837 | goto out; | |
5838 | /* | |
5839 | * all of the above have passed, it is safe to overwrite this extent | |
5840 | * without cow | |
5841 | */ | |
5842 | ret = 1; | |
5843 | out: | |
5844 | btrfs_free_path(path); | |
5845 | return ret; | |
5846 | } | |
5847 | ||
eb838e73 JB |
5848 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
5849 | struct extent_state **cached_state, int writing) | |
5850 | { | |
5851 | struct btrfs_ordered_extent *ordered; | |
5852 | int ret = 0; | |
5853 | ||
5854 | while (1) { | |
5855 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5856 | 0, cached_state); | |
5857 | /* | |
5858 | * We're concerned with the entire range that we're going to be | |
5859 | * doing DIO to, so we need to make sure theres no ordered | |
5860 | * extents in this range. | |
5861 | */ | |
5862 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
5863 | lockend - lockstart + 1); | |
5864 | ||
5865 | /* | |
5866 | * We need to make sure there are no buffered pages in this | |
5867 | * range either, we could have raced between the invalidate in | |
5868 | * generic_file_direct_write and locking the extent. The | |
5869 | * invalidate needs to happen so that reads after a write do not | |
5870 | * get stale data. | |
5871 | */ | |
5872 | if (!ordered && (!writing || | |
5873 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
5874 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
5875 | *cached_state))) | |
5876 | break; | |
5877 | ||
5878 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5879 | cached_state, GFP_NOFS); | |
5880 | ||
5881 | if (ordered) { | |
5882 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5883 | btrfs_put_ordered_extent(ordered); | |
5884 | } else { | |
5885 | /* Screw you mmap */ | |
5886 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
5887 | lockstart, | |
5888 | lockend); | |
5889 | if (ret) | |
5890 | break; | |
5891 | ||
5892 | /* | |
5893 | * If we found a page that couldn't be invalidated just | |
5894 | * fall back to buffered. | |
5895 | */ | |
5896 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5897 | lockstart >> PAGE_CACHE_SHIFT, | |
5898 | lockend >> PAGE_CACHE_SHIFT); | |
5899 | if (ret) | |
5900 | break; | |
5901 | } | |
5902 | ||
5903 | cond_resched(); | |
5904 | } | |
5905 | ||
5906 | return ret; | |
5907 | } | |
5908 | ||
69ffb543 JB |
5909 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
5910 | u64 len, u64 orig_start, | |
5911 | u64 block_start, u64 block_len, | |
5912 | int type) | |
5913 | { | |
5914 | struct extent_map_tree *em_tree; | |
5915 | struct extent_map *em; | |
5916 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5917 | int ret; | |
5918 | ||
5919 | em_tree = &BTRFS_I(inode)->extent_tree; | |
5920 | em = alloc_extent_map(); | |
5921 | if (!em) | |
5922 | return ERR_PTR(-ENOMEM); | |
5923 | ||
5924 | em->start = start; | |
5925 | em->orig_start = orig_start; | |
5926 | em->len = len; | |
5927 | em->block_len = block_len; | |
5928 | em->block_start = block_start; | |
5929 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5930 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5931 | if (type == BTRFS_ORDERED_PREALLOC) | |
5932 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
5933 | ||
5934 | do { | |
5935 | btrfs_drop_extent_cache(inode, em->start, | |
5936 | em->start + em->len - 1, 0); | |
5937 | write_lock(&em_tree->lock); | |
5938 | ret = add_extent_mapping(em_tree, em); | |
5939 | write_unlock(&em_tree->lock); | |
5940 | } while (ret == -EEXIST); | |
5941 | ||
5942 | if (ret) { | |
5943 | free_extent_map(em); | |
5944 | return ERR_PTR(ret); | |
5945 | } | |
5946 | ||
5947 | return em; | |
5948 | } | |
5949 | ||
5950 | ||
4b46fce2 JB |
5951 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5952 | struct buffer_head *bh_result, int create) | |
5953 | { | |
5954 | struct extent_map *em; | |
5955 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 5956 | struct extent_state *cached_state = NULL; |
4b46fce2 | 5957 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 5958 | u64 lockstart, lockend; |
4b46fce2 | 5959 | u64 len = bh_result->b_size; |
46bfbb5c | 5960 | struct btrfs_trans_handle *trans; |
eb838e73 JB |
5961 | int unlock_bits = EXTENT_LOCKED; |
5962 | int ret; | |
5963 | ||
eb838e73 JB |
5964 | if (create) { |
5965 | ret = btrfs_delalloc_reserve_space(inode, len); | |
5966 | if (ret) | |
5967 | return ret; | |
5968 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; | |
c329861d JB |
5969 | } else { |
5970 | len = min_t(u64, len, root->sectorsize); | |
eb838e73 JB |
5971 | } |
5972 | ||
c329861d JB |
5973 | lockstart = start; |
5974 | lockend = start + len - 1; | |
5975 | ||
eb838e73 JB |
5976 | /* |
5977 | * If this errors out it's because we couldn't invalidate pagecache for | |
5978 | * this range and we need to fallback to buffered. | |
5979 | */ | |
5980 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
5981 | return -ENOTBLK; | |
5982 | ||
5983 | if (create) { | |
5984 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
5985 | lockend, EXTENT_DELALLOC, NULL, | |
5986 | &cached_state, GFP_NOFS); | |
5987 | if (ret) | |
5988 | goto unlock_err; | |
5989 | } | |
4b46fce2 JB |
5990 | |
5991 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
eb838e73 JB |
5992 | if (IS_ERR(em)) { |
5993 | ret = PTR_ERR(em); | |
5994 | goto unlock_err; | |
5995 | } | |
4b46fce2 JB |
5996 | |
5997 | /* | |
5998 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5999 | * io. INLINE is special, and we could probably kludge it in here, but | |
6000 | * it's still buffered so for safety lets just fall back to the generic | |
6001 | * buffered path. | |
6002 | * | |
6003 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
6004 | * decompress it, so there will be buffering required no matter what we | |
6005 | * do, so go ahead and fallback to buffered. | |
6006 | * | |
6007 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
6008 | * to buffered IO. Don't blame me, this is the price we pay for using | |
6009 | * the generic code. | |
6010 | */ | |
6011 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
6012 | em->block_start == EXTENT_MAP_INLINE) { | |
6013 | free_extent_map(em); | |
eb838e73 JB |
6014 | ret = -ENOTBLK; |
6015 | goto unlock_err; | |
4b46fce2 JB |
6016 | } |
6017 | ||
6018 | /* Just a good old fashioned hole, return */ | |
6019 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6020 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6021 | free_extent_map(em); | |
eb838e73 JB |
6022 | ret = 0; |
6023 | goto unlock_err; | |
4b46fce2 JB |
6024 | } |
6025 | ||
6026 | /* | |
6027 | * We don't allocate a new extent in the following cases | |
6028 | * | |
6029 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6030 | * existing extent. | |
6031 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6032 | * just use the extent. | |
6033 | * | |
6034 | */ | |
46bfbb5c | 6035 | if (!create) { |
eb838e73 JB |
6036 | len = min(len, em->len - (start - em->start)); |
6037 | lockstart = start + len; | |
6038 | goto unlock; | |
46bfbb5c | 6039 | } |
4b46fce2 JB |
6040 | |
6041 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6042 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6043 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6044 | int type; |
6045 | int ret; | |
46bfbb5c | 6046 | u64 block_start; |
4b46fce2 JB |
6047 | |
6048 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6049 | type = BTRFS_ORDERED_PREALLOC; | |
6050 | else | |
6051 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6052 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6053 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6054 | |
6055 | /* | |
6056 | * we're not going to log anything, but we do need | |
6057 | * to make sure the current transaction stays open | |
6058 | * while we look for nocow cross refs | |
6059 | */ | |
7a7eaa40 | 6060 | trans = btrfs_join_transaction(root); |
3612b495 | 6061 | if (IS_ERR(trans)) |
46bfbb5c CM |
6062 | goto must_cow; |
6063 | ||
6064 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
69ffb543 JB |
6065 | u64 orig_start = em->start; |
6066 | ||
6067 | if (type == BTRFS_ORDERED_PREALLOC) { | |
6068 | free_extent_map(em); | |
6069 | em = create_pinned_em(inode, start, len, | |
6070 | orig_start, | |
6071 | block_start, len, type); | |
6072 | if (IS_ERR(em)) { | |
6073 | btrfs_end_transaction(trans, root); | |
6074 | goto unlock_err; | |
6075 | } | |
6076 | } | |
6077 | ||
46bfbb5c CM |
6078 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6079 | block_start, len, len, type); | |
6080 | btrfs_end_transaction(trans, root); | |
6081 | if (ret) { | |
6082 | free_extent_map(em); | |
eb838e73 | 6083 | goto unlock_err; |
46bfbb5c CM |
6084 | } |
6085 | goto unlock; | |
4b46fce2 | 6086 | } |
46bfbb5c | 6087 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6088 | } |
46bfbb5c CM |
6089 | must_cow: |
6090 | /* | |
6091 | * this will cow the extent, reset the len in case we changed | |
6092 | * it above | |
6093 | */ | |
6094 | len = bh_result->b_size; | |
16d299ac | 6095 | em = btrfs_new_extent_direct(inode, em, start, len); |
eb838e73 JB |
6096 | if (IS_ERR(em)) { |
6097 | ret = PTR_ERR(em); | |
6098 | goto unlock_err; | |
6099 | } | |
46bfbb5c CM |
6100 | len = min(len, em->len - (start - em->start)); |
6101 | unlock: | |
4b46fce2 JB |
6102 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6103 | inode->i_blkbits; | |
46bfbb5c | 6104 | bh_result->b_size = len; |
4b46fce2 JB |
6105 | bh_result->b_bdev = em->bdev; |
6106 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6107 | if (create) { |
6108 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6109 | set_buffer_new(bh_result); | |
6110 | ||
6111 | /* | |
6112 | * Need to update the i_size under the extent lock so buffered | |
6113 | * readers will get the updated i_size when we unlock. | |
6114 | */ | |
6115 | if (start + len > i_size_read(inode)) | |
6116 | i_size_write(inode, start + len); | |
6117 | } | |
4b46fce2 | 6118 | |
eb838e73 JB |
6119 | /* |
6120 | * In the case of write we need to clear and unlock the entire range, | |
6121 | * in the case of read we need to unlock only the end area that we | |
6122 | * aren't using if there is any left over space. | |
6123 | */ | |
24c03fa5 LB |
6124 | if (lockstart < lockend) { |
6125 | if (create && len < lockend - lockstart) { | |
6126 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
9e8a4a8b LB |
6127 | lockstart + len - 1, |
6128 | unlock_bits | EXTENT_DEFRAG, 1, 0, | |
24c03fa5 LB |
6129 | &cached_state, GFP_NOFS); |
6130 | /* | |
6131 | * Beside unlock, we also need to cleanup reserved space | |
6132 | * for the left range by attaching EXTENT_DO_ACCOUNTING. | |
6133 | */ | |
6134 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
6135 | lockstart + len, lockend, | |
9e8a4a8b LB |
6136 | unlock_bits | EXTENT_DO_ACCOUNTING | |
6137 | EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS); | |
24c03fa5 LB |
6138 | } else { |
6139 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6140 | lockend, unlock_bits, 1, 0, | |
6141 | &cached_state, GFP_NOFS); | |
6142 | } | |
6143 | } else { | |
eb838e73 | 6144 | free_extent_state(cached_state); |
24c03fa5 | 6145 | } |
eb838e73 | 6146 | |
4b46fce2 JB |
6147 | free_extent_map(em); |
6148 | ||
6149 | return 0; | |
eb838e73 JB |
6150 | |
6151 | unlock_err: | |
6152 | if (create) | |
6153 | unlock_bits |= EXTENT_DO_ACCOUNTING; | |
6154 | ||
6155 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6156 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6157 | return ret; | |
4b46fce2 JB |
6158 | } |
6159 | ||
6160 | struct btrfs_dio_private { | |
6161 | struct inode *inode; | |
6162 | u64 logical_offset; | |
6163 | u64 disk_bytenr; | |
6164 | u64 bytes; | |
4b46fce2 | 6165 | void *private; |
e65e1535 MX |
6166 | |
6167 | /* number of bios pending for this dio */ | |
6168 | atomic_t pending_bios; | |
6169 | ||
6170 | /* IO errors */ | |
6171 | int errors; | |
6172 | ||
6173 | struct bio *orig_bio; | |
4b46fce2 JB |
6174 | }; |
6175 | ||
6176 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6177 | { | |
e65e1535 | 6178 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6179 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6180 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6181 | struct inode *inode = dip->inode; |
6182 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6183 | u64 start; | |
4b46fce2 JB |
6184 | |
6185 | start = dip->logical_offset; | |
6186 | do { | |
6187 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6188 | struct page *page = bvec->bv_page; | |
6189 | char *kaddr; | |
6190 | u32 csum = ~(u32)0; | |
c329861d | 6191 | u64 private = ~(u32)0; |
4b46fce2 JB |
6192 | unsigned long flags; |
6193 | ||
c329861d JB |
6194 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6195 | start, &private)) | |
6196 | goto failed; | |
4b46fce2 | 6197 | local_irq_save(flags); |
7ac687d9 | 6198 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
6199 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
6200 | csum, bvec->bv_len); | |
6201 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6202 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6203 | local_irq_restore(flags); |
6204 | ||
6205 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6206 | if (csum != private) { |
6207 | failed: | |
33345d01 | 6208 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 6209 | " %llu csum %u private %u\n", |
33345d01 LZ |
6210 | (unsigned long long)btrfs_ino(inode), |
6211 | (unsigned long long)start, | |
c329861d | 6212 | csum, (unsigned)private); |
4b46fce2 JB |
6213 | err = -EIO; |
6214 | } | |
6215 | } | |
6216 | ||
6217 | start += bvec->bv_len; | |
4b46fce2 JB |
6218 | bvec++; |
6219 | } while (bvec <= bvec_end); | |
6220 | ||
6221 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6222 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
6223 | bio->bi_private = dip->private; |
6224 | ||
4b46fce2 | 6225 | kfree(dip); |
c0da7aa1 JB |
6226 | |
6227 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6228 | if (err) | |
6229 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6230 | dio_end_io(bio, err); |
6231 | } | |
6232 | ||
6233 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6234 | { | |
6235 | struct btrfs_dio_private *dip = bio->bi_private; | |
6236 | struct inode *inode = dip->inode; | |
6237 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6238 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6239 | u64 ordered_offset = dip->logical_offset; |
6240 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
6241 | int ret; |
6242 | ||
6243 | if (err) | |
6244 | goto out_done; | |
163cf09c CM |
6245 | again: |
6246 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6247 | &ordered_offset, | |
5fd02043 | 6248 | ordered_bytes, !err); |
4b46fce2 | 6249 | if (!ret) |
163cf09c | 6250 | goto out_test; |
4b46fce2 | 6251 | |
5fd02043 JB |
6252 | ordered->work.func = finish_ordered_fn; |
6253 | ordered->work.flags = 0; | |
6254 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6255 | &ordered->work); | |
163cf09c CM |
6256 | out_test: |
6257 | /* | |
6258 | * our bio might span multiple ordered extents. If we haven't | |
6259 | * completed the accounting for the whole dio, go back and try again | |
6260 | */ | |
6261 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6262 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6263 | ordered_offset; | |
5fd02043 | 6264 | ordered = NULL; |
163cf09c CM |
6265 | goto again; |
6266 | } | |
4b46fce2 JB |
6267 | out_done: |
6268 | bio->bi_private = dip->private; | |
6269 | ||
4b46fce2 | 6270 | kfree(dip); |
c0da7aa1 JB |
6271 | |
6272 | /* If we had an error make sure to clear the uptodate flag */ | |
6273 | if (err) | |
6274 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6275 | dio_end_io(bio, err); |
6276 | } | |
6277 | ||
eaf25d93 CM |
6278 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6279 | struct bio *bio, int mirror_num, | |
6280 | unsigned long bio_flags, u64 offset) | |
6281 | { | |
6282 | int ret; | |
6283 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6284 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6285 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6286 | return 0; |
6287 | } | |
6288 | ||
e65e1535 MX |
6289 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6290 | { | |
6291 | struct btrfs_dio_private *dip = bio->bi_private; | |
6292 | ||
6293 | if (err) { | |
33345d01 | 6294 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6295 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6296 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6297 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6298 | dip->errors = 1; |
6299 | ||
6300 | /* | |
6301 | * before atomic variable goto zero, we must make sure | |
6302 | * dip->errors is perceived to be set. | |
6303 | */ | |
6304 | smp_mb__before_atomic_dec(); | |
6305 | } | |
6306 | ||
6307 | /* if there are more bios still pending for this dio, just exit */ | |
6308 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6309 | goto out; | |
6310 | ||
6311 | if (dip->errors) | |
6312 | bio_io_error(dip->orig_bio); | |
6313 | else { | |
6314 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6315 | bio_endio(dip->orig_bio, 0); | |
6316 | } | |
6317 | out: | |
6318 | bio_put(bio); | |
6319 | } | |
6320 | ||
6321 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6322 | u64 first_sector, gfp_t gfp_flags) | |
6323 | { | |
6324 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6325 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6326 | } | |
6327 | ||
6328 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6329 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6330 | int async_submit) |
e65e1535 MX |
6331 | { |
6332 | int write = rw & REQ_WRITE; | |
6333 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6334 | int ret; | |
6335 | ||
6336 | bio_get(bio); | |
5fd02043 JB |
6337 | |
6338 | if (!write) { | |
6339 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6340 | if (ret) | |
6341 | goto err; | |
6342 | } | |
e65e1535 | 6343 | |
1ae39938 JB |
6344 | if (skip_sum) |
6345 | goto map; | |
6346 | ||
6347 | if (write && async_submit) { | |
e65e1535 MX |
6348 | ret = btrfs_wq_submit_bio(root->fs_info, |
6349 | inode, rw, bio, 0, 0, | |
6350 | file_offset, | |
6351 | __btrfs_submit_bio_start_direct_io, | |
6352 | __btrfs_submit_bio_done); | |
6353 | goto err; | |
1ae39938 JB |
6354 | } else if (write) { |
6355 | /* | |
6356 | * If we aren't doing async submit, calculate the csum of the | |
6357 | * bio now. | |
6358 | */ | |
6359 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6360 | if (ret) | |
6361 | goto err; | |
c2db1073 | 6362 | } else if (!skip_sum) { |
c329861d | 6363 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6364 | if (ret) |
6365 | goto err; | |
6366 | } | |
e65e1535 | 6367 | |
1ae39938 JB |
6368 | map: |
6369 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6370 | err: |
6371 | bio_put(bio); | |
6372 | return ret; | |
6373 | } | |
6374 | ||
6375 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6376 | int skip_sum) | |
6377 | { | |
6378 | struct inode *inode = dip->inode; | |
6379 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
6380 | struct bio *bio; |
6381 | struct bio *orig_bio = dip->orig_bio; | |
6382 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6383 | u64 start_sector = orig_bio->bi_sector; | |
6384 | u64 file_offset = dip->logical_offset; | |
6385 | u64 submit_len = 0; | |
6386 | u64 map_length; | |
6387 | int nr_pages = 0; | |
e65e1535 | 6388 | int ret = 0; |
1ae39938 | 6389 | int async_submit = 0; |
e65e1535 | 6390 | |
e65e1535 | 6391 | map_length = orig_bio->bi_size; |
3ec706c8 | 6392 | ret = btrfs_map_block(root->fs_info, READ, start_sector << 9, |
e65e1535 MX |
6393 | &map_length, NULL, 0); |
6394 | if (ret) { | |
64728bbb | 6395 | bio_put(orig_bio); |
e65e1535 MX |
6396 | return -EIO; |
6397 | } | |
6398 | ||
02f57c7a JB |
6399 | if (map_length >= orig_bio->bi_size) { |
6400 | bio = orig_bio; | |
6401 | goto submit; | |
6402 | } | |
6403 | ||
1ae39938 | 6404 | async_submit = 1; |
02f57c7a JB |
6405 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6406 | if (!bio) | |
6407 | return -ENOMEM; | |
6408 | bio->bi_private = dip; | |
6409 | bio->bi_end_io = btrfs_end_dio_bio; | |
6410 | atomic_inc(&dip->pending_bios); | |
6411 | ||
e65e1535 MX |
6412 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6413 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6414 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6415 | bvec->bv_offset) < bvec->bv_len)) { | |
6416 | /* | |
6417 | * inc the count before we submit the bio so | |
6418 | * we know the end IO handler won't happen before | |
6419 | * we inc the count. Otherwise, the dip might get freed | |
6420 | * before we're done setting it up | |
6421 | */ | |
6422 | atomic_inc(&dip->pending_bios); | |
6423 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6424 | file_offset, skip_sum, | |
c329861d | 6425 | async_submit); |
e65e1535 MX |
6426 | if (ret) { |
6427 | bio_put(bio); | |
6428 | atomic_dec(&dip->pending_bios); | |
6429 | goto out_err; | |
6430 | } | |
6431 | ||
e65e1535 MX |
6432 | start_sector += submit_len >> 9; |
6433 | file_offset += submit_len; | |
6434 | ||
6435 | submit_len = 0; | |
6436 | nr_pages = 0; | |
6437 | ||
6438 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6439 | start_sector, GFP_NOFS); | |
6440 | if (!bio) | |
6441 | goto out_err; | |
6442 | bio->bi_private = dip; | |
6443 | bio->bi_end_io = btrfs_end_dio_bio; | |
6444 | ||
6445 | map_length = orig_bio->bi_size; | |
3ec706c8 SB |
6446 | ret = btrfs_map_block(root->fs_info, READ, |
6447 | start_sector << 9, | |
e65e1535 MX |
6448 | &map_length, NULL, 0); |
6449 | if (ret) { | |
6450 | bio_put(bio); | |
6451 | goto out_err; | |
6452 | } | |
6453 | } else { | |
6454 | submit_len += bvec->bv_len; | |
6455 | nr_pages ++; | |
6456 | bvec++; | |
6457 | } | |
6458 | } | |
6459 | ||
02f57c7a | 6460 | submit: |
e65e1535 | 6461 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 6462 | async_submit); |
e65e1535 MX |
6463 | if (!ret) |
6464 | return 0; | |
6465 | ||
6466 | bio_put(bio); | |
6467 | out_err: | |
6468 | dip->errors = 1; | |
6469 | /* | |
6470 | * before atomic variable goto zero, we must | |
6471 | * make sure dip->errors is perceived to be set. | |
6472 | */ | |
6473 | smp_mb__before_atomic_dec(); | |
6474 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6475 | bio_io_error(dip->orig_bio); | |
6476 | ||
6477 | /* bio_end_io() will handle error, so we needn't return it */ | |
6478 | return 0; | |
6479 | } | |
6480 | ||
4b46fce2 JB |
6481 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6482 | loff_t file_offset) | |
6483 | { | |
6484 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6485 | struct btrfs_dio_private *dip; | |
6486 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6487 | int skip_sum; |
7b6d91da | 6488 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6489 | int ret = 0; |
6490 | ||
6491 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6492 | ||
6493 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6494 | if (!dip) { | |
6495 | ret = -ENOMEM; | |
6496 | goto free_ordered; | |
6497 | } | |
4b46fce2 JB |
6498 | |
6499 | dip->private = bio->bi_private; | |
6500 | dip->inode = inode; | |
6501 | dip->logical_offset = file_offset; | |
6502 | ||
4b46fce2 JB |
6503 | dip->bytes = 0; |
6504 | do { | |
6505 | dip->bytes += bvec->bv_len; | |
6506 | bvec++; | |
6507 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6508 | ||
46bfbb5c | 6509 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6510 | bio->bi_private = dip; |
e65e1535 MX |
6511 | dip->errors = 0; |
6512 | dip->orig_bio = bio; | |
6513 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6514 | |
6515 | if (write) | |
6516 | bio->bi_end_io = btrfs_endio_direct_write; | |
6517 | else | |
6518 | bio->bi_end_io = btrfs_endio_direct_read; | |
6519 | ||
e65e1535 MX |
6520 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6521 | if (!ret) | |
eaf25d93 | 6522 | return; |
4b46fce2 JB |
6523 | free_ordered: |
6524 | /* | |
6525 | * If this is a write, we need to clean up the reserved space and kill | |
6526 | * the ordered extent. | |
6527 | */ | |
6528 | if (write) { | |
6529 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6530 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6531 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6532 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6533 | btrfs_free_reserved_extent(root, ordered->start, | |
6534 | ordered->disk_len); | |
6535 | btrfs_put_ordered_extent(ordered); | |
6536 | btrfs_put_ordered_extent(ordered); | |
6537 | } | |
6538 | bio_endio(bio, ret); | |
6539 | } | |
6540 | ||
5a5f79b5 CM |
6541 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6542 | const struct iovec *iov, loff_t offset, | |
6543 | unsigned long nr_segs) | |
6544 | { | |
6545 | int seg; | |
a1b75f7d | 6546 | int i; |
5a5f79b5 CM |
6547 | size_t size; |
6548 | unsigned long addr; | |
6549 | unsigned blocksize_mask = root->sectorsize - 1; | |
6550 | ssize_t retval = -EINVAL; | |
6551 | loff_t end = offset; | |
6552 | ||
6553 | if (offset & blocksize_mask) | |
6554 | goto out; | |
6555 | ||
6556 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6557 | for (seg = 0; seg < nr_segs; seg++) { | |
6558 | addr = (unsigned long)iov[seg].iov_base; | |
6559 | size = iov[seg].iov_len; | |
6560 | end += size; | |
a1b75f7d | 6561 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6562 | goto out; |
a1b75f7d JB |
6563 | |
6564 | /* If this is a write we don't need to check anymore */ | |
6565 | if (rw & WRITE) | |
6566 | continue; | |
6567 | ||
6568 | /* | |
6569 | * Check to make sure we don't have duplicate iov_base's in this | |
6570 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6571 | * when reading back. | |
6572 | */ | |
6573 | for (i = seg + 1; i < nr_segs; i++) { | |
6574 | if (iov[seg].iov_base == iov[i].iov_base) | |
6575 | goto out; | |
6576 | } | |
5a5f79b5 CM |
6577 | } |
6578 | retval = 0; | |
6579 | out: | |
6580 | return retval; | |
6581 | } | |
eb838e73 | 6582 | |
16432985 CM |
6583 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6584 | const struct iovec *iov, loff_t offset, | |
6585 | unsigned long nr_segs) | |
6586 | { | |
4b46fce2 JB |
6587 | struct file *file = iocb->ki_filp; |
6588 | struct inode *inode = file->f_mapping->host; | |
4b46fce2 | 6589 | |
5a5f79b5 | 6590 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 6591 | offset, nr_segs)) |
5a5f79b5 | 6592 | return 0; |
3f7c579c | 6593 | |
eb838e73 | 6594 | return __blockdev_direct_IO(rw, iocb, inode, |
5a5f79b5 CM |
6595 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, |
6596 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6597 | btrfs_submit_direct, 0); | |
16432985 CM |
6598 | } |
6599 | ||
05dadc09 TI |
6600 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
6601 | ||
1506fcc8 YS |
6602 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6603 | __u64 start, __u64 len) | |
6604 | { | |
05dadc09 TI |
6605 | int ret; |
6606 | ||
6607 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
6608 | if (ret) | |
6609 | return ret; | |
6610 | ||
ec29ed5b | 6611 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6612 | } |
6613 | ||
a52d9a80 | 6614 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6615 | { |
d1310b2e CM |
6616 | struct extent_io_tree *tree; |
6617 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6618 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6619 | } |
1832a6d5 | 6620 | |
a52d9a80 | 6621 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6622 | { |
d1310b2e | 6623 | struct extent_io_tree *tree; |
b888db2b CM |
6624 | |
6625 | ||
6626 | if (current->flags & PF_MEMALLOC) { | |
6627 | redirty_page_for_writepage(wbc, page); | |
6628 | unlock_page(page); | |
6629 | return 0; | |
6630 | } | |
d1310b2e | 6631 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6632 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6633 | } |
6634 | ||
f421950f CM |
6635 | int btrfs_writepages(struct address_space *mapping, |
6636 | struct writeback_control *wbc) | |
b293f02e | 6637 | { |
d1310b2e | 6638 | struct extent_io_tree *tree; |
771ed689 | 6639 | |
d1310b2e | 6640 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6641 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6642 | } | |
6643 | ||
3ab2fb5a CM |
6644 | static int |
6645 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6646 | struct list_head *pages, unsigned nr_pages) | |
6647 | { | |
d1310b2e CM |
6648 | struct extent_io_tree *tree; |
6649 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6650 | return extent_readpages(tree, mapping, pages, nr_pages, |
6651 | btrfs_get_extent); | |
6652 | } | |
e6dcd2dc | 6653 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6654 | { |
d1310b2e CM |
6655 | struct extent_io_tree *tree; |
6656 | struct extent_map_tree *map; | |
a52d9a80 | 6657 | int ret; |
8c2383c3 | 6658 | |
d1310b2e CM |
6659 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6660 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6661 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6662 | if (ret == 1) { |
6663 | ClearPagePrivate(page); | |
6664 | set_page_private(page, 0); | |
6665 | page_cache_release(page); | |
39279cc3 | 6666 | } |
a52d9a80 | 6667 | return ret; |
39279cc3 CM |
6668 | } |
6669 | ||
e6dcd2dc CM |
6670 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6671 | { | |
98509cfc CM |
6672 | if (PageWriteback(page) || PageDirty(page)) |
6673 | return 0; | |
b335b003 | 6674 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6675 | } |
6676 | ||
a52d9a80 | 6677 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6678 | { |
5fd02043 | 6679 | struct inode *inode = page->mapping->host; |
d1310b2e | 6680 | struct extent_io_tree *tree; |
e6dcd2dc | 6681 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6682 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6683 | u64 page_start = page_offset(page); |
6684 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6685 | |
8b62b72b CM |
6686 | /* |
6687 | * we have the page locked, so new writeback can't start, | |
6688 | * and the dirty bit won't be cleared while we are here. | |
6689 | * | |
6690 | * Wait for IO on this page so that we can safely clear | |
6691 | * the PagePrivate2 bit and do ordered accounting | |
6692 | */ | |
e6dcd2dc | 6693 | wait_on_page_writeback(page); |
8b62b72b | 6694 | |
5fd02043 | 6695 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6696 | if (offset) { |
6697 | btrfs_releasepage(page, GFP_NOFS); | |
6698 | return; | |
6699 | } | |
d0082371 | 6700 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
5fd02043 | 6701 | ordered = btrfs_lookup_ordered_extent(inode, |
e6dcd2dc CM |
6702 | page_offset(page)); |
6703 | if (ordered) { | |
eb84ae03 CM |
6704 | /* |
6705 | * IO on this page will never be started, so we need | |
6706 | * to account for any ordered extents now | |
6707 | */ | |
e6dcd2dc CM |
6708 | clear_extent_bit(tree, page_start, page_end, |
6709 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
6710 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
6711 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
6712 | /* |
6713 | * whoever cleared the private bit is responsible | |
6714 | * for the finish_ordered_io | |
6715 | */ | |
5fd02043 JB |
6716 | if (TestClearPagePrivate2(page) && |
6717 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6718 | PAGE_CACHE_SIZE, 1)) { | |
6719 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6720 | } |
e6dcd2dc | 6721 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6722 | cached_state = NULL; |
d0082371 | 6723 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6724 | } |
6725 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6726 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
6727 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
6728 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
6729 | __btrfs_releasepage(page, GFP_NOFS); |
6730 | ||
4a096752 | 6731 | ClearPageChecked(page); |
9ad6b7bc | 6732 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6733 | ClearPagePrivate(page); |
6734 | set_page_private(page, 0); | |
6735 | page_cache_release(page); | |
6736 | } | |
39279cc3 CM |
6737 | } |
6738 | ||
9ebefb18 CM |
6739 | /* |
6740 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6741 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6742 | * be careful to check for EOF conditions here. We set the page up correctly | |
6743 | * for a written page which means we get ENOSPC checking when writing into | |
6744 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6745 | * support these features. | |
6746 | * | |
6747 | * We are not allowed to take the i_mutex here so we have to play games to | |
6748 | * protect against truncate races as the page could now be beyond EOF. Because | |
6749 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6750 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6751 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6752 | * unlock the page. | |
6753 | */ | |
c2ec175c | 6754 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6755 | { |
c2ec175c | 6756 | struct page *page = vmf->page; |
6da6abae | 6757 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6758 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6759 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6760 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6761 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6762 | char *kaddr; |
6763 | unsigned long zero_start; | |
9ebefb18 | 6764 | loff_t size; |
1832a6d5 | 6765 | int ret; |
9998eb70 | 6766 | int reserved = 0; |
a52d9a80 | 6767 | u64 page_start; |
e6dcd2dc | 6768 | u64 page_end; |
9ebefb18 | 6769 | |
b2b5ef5c | 6770 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 6771 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6772 | if (!ret) { |
e41f941a | 6773 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
6774 | reserved = 1; |
6775 | } | |
56a76f82 NP |
6776 | if (ret) { |
6777 | if (ret == -ENOMEM) | |
6778 | ret = VM_FAULT_OOM; | |
6779 | else /* -ENOSPC, -EIO, etc */ | |
6780 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6781 | if (reserved) |
6782 | goto out; | |
6783 | goto out_noreserve; | |
56a76f82 | 6784 | } |
1832a6d5 | 6785 | |
56a76f82 | 6786 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6787 | again: |
9ebefb18 | 6788 | lock_page(page); |
9ebefb18 | 6789 | size = i_size_read(inode); |
e6dcd2dc CM |
6790 | page_start = page_offset(page); |
6791 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6792 | |
9ebefb18 | 6793 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6794 | (page_start >= size)) { |
9ebefb18 CM |
6795 | /* page got truncated out from underneath us */ |
6796 | goto out_unlock; | |
6797 | } | |
e6dcd2dc CM |
6798 | wait_on_page_writeback(page); |
6799 | ||
d0082371 | 6800 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6801 | set_page_extent_mapped(page); |
6802 | ||
eb84ae03 CM |
6803 | /* |
6804 | * we can't set the delalloc bits if there are pending ordered | |
6805 | * extents. Drop our locks and wait for them to finish | |
6806 | */ | |
e6dcd2dc CM |
6807 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6808 | if (ordered) { | |
2ac55d41 JB |
6809 | unlock_extent_cached(io_tree, page_start, page_end, |
6810 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6811 | unlock_page(page); |
eb84ae03 | 6812 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6813 | btrfs_put_ordered_extent(ordered); |
6814 | goto again; | |
6815 | } | |
6816 | ||
fbf19087 JB |
6817 | /* |
6818 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6819 | * if it was already dirty, so for space accounting reasons we need to | |
6820 | * clear any delalloc bits for the range we are fixing to save. There | |
6821 | * is probably a better way to do this, but for now keep consistent with | |
6822 | * prepare_pages in the normal write path. | |
6823 | */ | |
2ac55d41 | 6824 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
6825 | EXTENT_DIRTY | EXTENT_DELALLOC | |
6826 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 6827 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6828 | |
2ac55d41 JB |
6829 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6830 | &cached_state); | |
9ed74f2d | 6831 | if (ret) { |
2ac55d41 JB |
6832 | unlock_extent_cached(io_tree, page_start, page_end, |
6833 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6834 | ret = VM_FAULT_SIGBUS; |
6835 | goto out_unlock; | |
6836 | } | |
e6dcd2dc | 6837 | ret = 0; |
9ebefb18 CM |
6838 | |
6839 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6840 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6841 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6842 | else |
e6dcd2dc | 6843 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6844 | |
e6dcd2dc CM |
6845 | if (zero_start != PAGE_CACHE_SIZE) { |
6846 | kaddr = kmap(page); | |
6847 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6848 | flush_dcache_page(page); | |
6849 | kunmap(page); | |
6850 | } | |
247e743c | 6851 | ClearPageChecked(page); |
e6dcd2dc | 6852 | set_page_dirty(page); |
50a9b214 | 6853 | SetPageUptodate(page); |
5a3f23d5 | 6854 | |
257c62e1 CM |
6855 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6856 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 6857 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 6858 | |
2ac55d41 | 6859 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6860 | |
6861 | out_unlock: | |
b2b5ef5c JK |
6862 | if (!ret) { |
6863 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 6864 | return VM_FAULT_LOCKED; |
b2b5ef5c | 6865 | } |
9ebefb18 | 6866 | unlock_page(page); |
1832a6d5 | 6867 | out: |
ec39e180 | 6868 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6869 | out_noreserve: |
b2b5ef5c | 6870 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
6871 | return ret; |
6872 | } | |
6873 | ||
a41ad394 | 6874 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6875 | { |
6876 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6877 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6878 | int ret; |
3893e33b | 6879 | int err = 0; |
39279cc3 | 6880 | struct btrfs_trans_handle *trans; |
dbe674a9 | 6881 | u64 mask = root->sectorsize - 1; |
07127184 | 6882 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6883 | |
2aaa6655 | 6884 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
5d5e103a | 6885 | if (ret) |
a41ad394 | 6886 | return ret; |
8082510e | 6887 | |
4a096752 | 6888 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6889 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6890 | |
fcb80c2a JB |
6891 | /* |
6892 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6893 | * 3 things going on here | |
6894 | * | |
6895 | * 1) We need to reserve space for our orphan item and the space to | |
6896 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6897 | * orphan item because we didn't reserve space to remove it. | |
6898 | * | |
6899 | * 2) We need to reserve space to update our inode. | |
6900 | * | |
6901 | * 3) We need to have something to cache all the space that is going to | |
6902 | * be free'd up by the truncate operation, but also have some slack | |
6903 | * space reserved in case it uses space during the truncate (thank you | |
6904 | * very much snapshotting). | |
6905 | * | |
6906 | * And we need these to all be seperate. The fact is we can use alot of | |
6907 | * space doing the truncate, and we have no earthly idea how much space | |
6908 | * we will use, so we need the truncate reservation to be seperate so it | |
6909 | * doesn't end up using space reserved for updating the inode or | |
6910 | * removing the orphan item. We also need to be able to stop the | |
6911 | * transaction and start a new one, which means we need to be able to | |
6912 | * update the inode several times, and we have no idea of knowing how | |
6913 | * many times that will be, so we can't just reserve 1 item for the | |
6914 | * entirety of the opration, so that has to be done seperately as well. | |
6915 | * Then there is the orphan item, which does indeed need to be held on | |
6916 | * to for the whole operation, and we need nobody to touch this reserved | |
6917 | * space except the orphan code. | |
6918 | * | |
6919 | * So that leaves us with | |
6920 | * | |
6921 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6922 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6923 | * transaction reservation. | |
6924 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6925 | * updating the inode. | |
6926 | */ | |
66d8f3dd | 6927 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
6928 | if (!rsv) |
6929 | return -ENOMEM; | |
4a338542 | 6930 | rsv->size = min_size; |
ca7e70f5 | 6931 | rsv->failfast = 1; |
f0cd846e | 6932 | |
907cbceb | 6933 | /* |
07127184 | 6934 | * 1 for the truncate slack space |
907cbceb JB |
6935 | * 1 for the orphan item we're going to add |
6936 | * 1 for the orphan item deletion | |
6937 | * 1 for updating the inode. | |
6938 | */ | |
fcb80c2a JB |
6939 | trans = btrfs_start_transaction(root, 4); |
6940 | if (IS_ERR(trans)) { | |
6941 | err = PTR_ERR(trans); | |
6942 | goto out; | |
6943 | } | |
f0cd846e | 6944 | |
907cbceb JB |
6945 | /* Migrate the slack space for the truncate to our reserve */ |
6946 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6947 | min_size); | |
fcb80c2a | 6948 | BUG_ON(ret); |
f0cd846e JB |
6949 | |
6950 | ret = btrfs_orphan_add(trans, inode); | |
6951 | if (ret) { | |
6952 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6953 | goto out; |
f0cd846e JB |
6954 | } |
6955 | ||
5a3f23d5 CM |
6956 | /* |
6957 | * setattr is responsible for setting the ordered_data_close flag, | |
6958 | * but that is only tested during the last file release. That | |
6959 | * could happen well after the next commit, leaving a great big | |
6960 | * window where new writes may get lost if someone chooses to write | |
6961 | * to this file after truncating to zero | |
6962 | * | |
6963 | * The inode doesn't have any dirty data here, and so if we commit | |
6964 | * this is a noop. If someone immediately starts writing to the inode | |
6965 | * it is very likely we'll catch some of their writes in this | |
6966 | * transaction, and the commit will find this file on the ordered | |
6967 | * data list with good things to send down. | |
6968 | * | |
6969 | * This is a best effort solution, there is still a window where | |
6970 | * using truncate to replace the contents of the file will | |
6971 | * end up with a zero length file after a crash. | |
6972 | */ | |
72ac3c0d JB |
6973 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
6974 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
6975 | btrfs_add_ordered_operation(trans, root, inode); |
6976 | ||
5dc562c5 JB |
6977 | /* |
6978 | * So if we truncate and then write and fsync we normally would just | |
6979 | * write the extents that changed, which is a problem if we need to | |
6980 | * first truncate that entire inode. So set this flag so we write out | |
6981 | * all of the extents in the inode to the sync log so we're completely | |
6982 | * safe. | |
6983 | */ | |
6984 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 6985 | trans->block_rsv = rsv; |
907cbceb | 6986 | |
8082510e YZ |
6987 | while (1) { |
6988 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
6989 | inode->i_size, | |
6990 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 6991 | if (ret != -ENOSPC) { |
3893e33b | 6992 | err = ret; |
8082510e | 6993 | break; |
3893e33b | 6994 | } |
39279cc3 | 6995 | |
fcb80c2a | 6996 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6997 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6998 | if (ret) { |
6999 | err = ret; | |
7000 | break; | |
7001 | } | |
ca7e70f5 | 7002 | |
8082510e | 7003 | btrfs_end_transaction(trans, root); |
b53d3f5d | 7004 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
7005 | |
7006 | trans = btrfs_start_transaction(root, 2); | |
7007 | if (IS_ERR(trans)) { | |
7008 | ret = err = PTR_ERR(trans); | |
7009 | trans = NULL; | |
7010 | break; | |
7011 | } | |
7012 | ||
7013 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
7014 | rsv, min_size); | |
7015 | BUG_ON(ret); /* shouldn't happen */ | |
7016 | trans->block_rsv = rsv; | |
8082510e YZ |
7017 | } |
7018 | ||
7019 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7020 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7021 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7022 | if (ret) |
7023 | err = ret; | |
ded5db9d JB |
7024 | } else if (ret && inode->i_nlink > 0) { |
7025 | /* | |
7026 | * Failed to do the truncate, remove us from the in memory | |
7027 | * orphan list. | |
7028 | */ | |
7029 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
7030 | } |
7031 | ||
917c16b2 CM |
7032 | if (trans) { |
7033 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7034 | ret = btrfs_update_inode(trans, root, inode); | |
7035 | if (ret && !err) | |
7036 | err = ret; | |
7b128766 | 7037 | |
7ad85bb7 | 7038 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 7039 | btrfs_btree_balance_dirty(root); |
917c16b2 | 7040 | } |
fcb80c2a JB |
7041 | |
7042 | out: | |
7043 | btrfs_free_block_rsv(root, rsv); | |
7044 | ||
3893e33b JB |
7045 | if (ret && !err) |
7046 | err = ret; | |
a41ad394 | 7047 | |
3893e33b | 7048 | return err; |
39279cc3 CM |
7049 | } |
7050 | ||
d352ac68 CM |
7051 | /* |
7052 | * create a new subvolume directory/inode (helper for the ioctl). | |
7053 | */ | |
d2fb3437 | 7054 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7055 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7056 | { |
39279cc3 | 7057 | struct inode *inode; |
76dda93c | 7058 | int err; |
00e4e6b3 | 7059 | u64 index = 0; |
39279cc3 | 7060 | |
12fc9d09 FA |
7061 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7062 | new_dirid, new_dirid, | |
7063 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7064 | &index); | |
54aa1f4d | 7065 | if (IS_ERR(inode)) |
f46b5a66 | 7066 | return PTR_ERR(inode); |
39279cc3 CM |
7067 | inode->i_op = &btrfs_dir_inode_operations; |
7068 | inode->i_fop = &btrfs_dir_file_operations; | |
7069 | ||
bfe86848 | 7070 | set_nlink(inode, 1); |
dbe674a9 | 7071 | btrfs_i_size_write(inode, 0); |
3b96362c | 7072 | |
76dda93c | 7073 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7074 | |
76dda93c | 7075 | iput(inode); |
ce598979 | 7076 | return err; |
39279cc3 CM |
7077 | } |
7078 | ||
39279cc3 CM |
7079 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7080 | { | |
7081 | struct btrfs_inode *ei; | |
2ead6ae7 | 7082 | struct inode *inode; |
39279cc3 CM |
7083 | |
7084 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7085 | if (!ei) | |
7086 | return NULL; | |
2ead6ae7 YZ |
7087 | |
7088 | ei->root = NULL; | |
2ead6ae7 | 7089 | ei->generation = 0; |
15ee9bc7 | 7090 | ei->last_trans = 0; |
257c62e1 | 7091 | ei->last_sub_trans = 0; |
e02119d5 | 7092 | ei->logged_trans = 0; |
2ead6ae7 | 7093 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7094 | ei->disk_i_size = 0; |
7095 | ei->flags = 0; | |
7709cde3 | 7096 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7097 | ei->index_cnt = (u64)-1; |
7098 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7099 | ei->last_log_commit = 0; |
2ead6ae7 | 7100 | |
9e0baf60 JB |
7101 | spin_lock_init(&ei->lock); |
7102 | ei->outstanding_extents = 0; | |
7103 | ei->reserved_extents = 0; | |
2ead6ae7 | 7104 | |
72ac3c0d | 7105 | ei->runtime_flags = 0; |
261507a0 | 7106 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7107 | |
16cdcec7 MX |
7108 | ei->delayed_node = NULL; |
7109 | ||
2ead6ae7 | 7110 | inode = &ei->vfs_inode; |
a8067e02 | 7111 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7112 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7113 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7114 | ei->io_tree.track_uptodate = 1; |
7115 | ei->io_failure_tree.track_uptodate = 1; | |
2ead6ae7 | 7116 | mutex_init(&ei->log_mutex); |
f248679e | 7117 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7118 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7119 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7120 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7121 | RB_CLEAR_NODE(&ei->rb_node); |
7122 | ||
7123 | return inode; | |
39279cc3 CM |
7124 | } |
7125 | ||
fa0d7e3d NP |
7126 | static void btrfs_i_callback(struct rcu_head *head) |
7127 | { | |
7128 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7129 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7130 | } | |
7131 | ||
39279cc3 CM |
7132 | void btrfs_destroy_inode(struct inode *inode) |
7133 | { | |
e6dcd2dc | 7134 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7135 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7136 | ||
b3d9b7a3 | 7137 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7138 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7139 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7140 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7141 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7142 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7143 | |
a6dbd429 JB |
7144 | /* |
7145 | * This can happen where we create an inode, but somebody else also | |
7146 | * created the same inode and we need to destroy the one we already | |
7147 | * created. | |
7148 | */ | |
7149 | if (!root) | |
7150 | goto free; | |
7151 | ||
5a3f23d5 CM |
7152 | /* |
7153 | * Make sure we're properly removed from the ordered operation | |
7154 | * lists. | |
7155 | */ | |
7156 | smp_mb(); | |
7157 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
7158 | spin_lock(&root->fs_info->ordered_extent_lock); | |
7159 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
7160 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
7161 | } | |
7162 | ||
8a35d95f JB |
7163 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7164 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
7165 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
7166 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7167 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7168 | } |
7b128766 | 7169 | |
d397712b | 7170 | while (1) { |
e6dcd2dc CM |
7171 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7172 | if (!ordered) | |
7173 | break; | |
7174 | else { | |
d397712b CM |
7175 | printk(KERN_ERR "btrfs found ordered " |
7176 | "extent %llu %llu on inode cleanup\n", | |
7177 | (unsigned long long)ordered->file_offset, | |
7178 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7179 | btrfs_remove_ordered_extent(inode, ordered); |
7180 | btrfs_put_ordered_extent(ordered); | |
7181 | btrfs_put_ordered_extent(ordered); | |
7182 | } | |
7183 | } | |
5d4f98a2 | 7184 | inode_tree_del(inode); |
5b21f2ed | 7185 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7186 | free: |
16cdcec7 | 7187 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 7188 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7189 | } |
7190 | ||
45321ac5 | 7191 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7192 | { |
7193 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7194 | |
0af3d00b | 7195 | if (btrfs_root_refs(&root->root_item) == 0 && |
83eea1f1 | 7196 | !btrfs_is_free_space_inode(inode)) |
45321ac5 | 7197 | return 1; |
76dda93c | 7198 | else |
45321ac5 | 7199 | return generic_drop_inode(inode); |
76dda93c YZ |
7200 | } |
7201 | ||
0ee0fda0 | 7202 | static void init_once(void *foo) |
39279cc3 CM |
7203 | { |
7204 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7205 | ||
7206 | inode_init_once(&ei->vfs_inode); | |
7207 | } | |
7208 | ||
7209 | void btrfs_destroy_cachep(void) | |
7210 | { | |
8c0a8537 KS |
7211 | /* |
7212 | * Make sure all delayed rcu free inodes are flushed before we | |
7213 | * destroy cache. | |
7214 | */ | |
7215 | rcu_barrier(); | |
39279cc3 CM |
7216 | if (btrfs_inode_cachep) |
7217 | kmem_cache_destroy(btrfs_inode_cachep); | |
7218 | if (btrfs_trans_handle_cachep) | |
7219 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7220 | if (btrfs_transaction_cachep) | |
7221 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7222 | if (btrfs_path_cachep) |
7223 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7224 | if (btrfs_free_space_cachep) |
7225 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
7226 | if (btrfs_delalloc_work_cachep) |
7227 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
7228 | } |
7229 | ||
7230 | int btrfs_init_cachep(void) | |
7231 | { | |
837e1972 | 7232 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7233 | sizeof(struct btrfs_inode), 0, |
7234 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7235 | if (!btrfs_inode_cachep) |
7236 | goto fail; | |
9601e3f6 | 7237 | |
837e1972 | 7238 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7239 | sizeof(struct btrfs_trans_handle), 0, |
7240 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7241 | if (!btrfs_trans_handle_cachep) |
7242 | goto fail; | |
9601e3f6 | 7243 | |
837e1972 | 7244 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7245 | sizeof(struct btrfs_transaction), 0, |
7246 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7247 | if (!btrfs_transaction_cachep) |
7248 | goto fail; | |
9601e3f6 | 7249 | |
837e1972 | 7250 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7251 | sizeof(struct btrfs_path), 0, |
7252 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7253 | if (!btrfs_path_cachep) |
7254 | goto fail; | |
9601e3f6 | 7255 | |
837e1972 | 7256 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7257 | sizeof(struct btrfs_free_space), 0, |
7258 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7259 | if (!btrfs_free_space_cachep) | |
7260 | goto fail; | |
7261 | ||
8ccf6f19 MX |
7262 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
7263 | sizeof(struct btrfs_delalloc_work), 0, | |
7264 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
7265 | NULL); | |
7266 | if (!btrfs_delalloc_work_cachep) | |
7267 | goto fail; | |
7268 | ||
39279cc3 CM |
7269 | return 0; |
7270 | fail: | |
7271 | btrfs_destroy_cachep(); | |
7272 | return -ENOMEM; | |
7273 | } | |
7274 | ||
7275 | static int btrfs_getattr(struct vfsmount *mnt, | |
7276 | struct dentry *dentry, struct kstat *stat) | |
7277 | { | |
7278 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
7279 | u32 blocksize = inode->i_sb->s_blocksize; |
7280 | ||
39279cc3 | 7281 | generic_fillattr(inode, stat); |
0ee5dc67 | 7282 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7283 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
7284 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
7285 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
7286 | return 0; |
7287 | } | |
7288 | ||
75e7cb7f LB |
7289 | /* |
7290 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7291 | * directory. | |
7292 | */ | |
7293 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7294 | { | |
7295 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7296 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7297 | ||
7298 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7299 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7300 | else | |
7301 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7302 | ||
bc178237 | 7303 | if (b_dir->flags & BTRFS_INODE_COMPRESS) { |
75e7cb7f | 7304 | b_inode->flags |= BTRFS_INODE_COMPRESS; |
bc178237 LB |
7305 | b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
7306 | } else { | |
7307 | b_inode->flags &= ~(BTRFS_INODE_COMPRESS | | |
7308 | BTRFS_INODE_NOCOMPRESS); | |
7309 | } | |
75e7cb7f LB |
7310 | } |
7311 | ||
d397712b CM |
7312 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7313 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7314 | { |
7315 | struct btrfs_trans_handle *trans; | |
7316 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7317 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7318 | struct inode *new_inode = new_dentry->d_inode; |
7319 | struct inode *old_inode = old_dentry->d_inode; | |
7320 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7321 | u64 index = 0; |
4df27c4d | 7322 | u64 root_objectid; |
39279cc3 | 7323 | int ret; |
33345d01 | 7324 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7325 | |
33345d01 | 7326 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7327 | return -EPERM; |
7328 | ||
4df27c4d | 7329 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7330 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7331 | return -EXDEV; |
7332 | ||
33345d01 LZ |
7333 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7334 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7335 | return -ENOTEMPTY; |
5f39d397 | 7336 | |
4df27c4d YZ |
7337 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7338 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7339 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7340 | /* |
7341 | * we're using rename to replace one file with another. | |
7342 | * and the replacement file is large. Start IO on it now so | |
7343 | * we don't add too much work to the end of the transaction | |
7344 | */ | |
4baf8c92 | 7345 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7346 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7347 | filemap_flush(old_inode->i_mapping); | |
7348 | ||
76dda93c | 7349 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7350 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7351 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7352 | /* |
7353 | * We want to reserve the absolute worst case amount of items. So if | |
7354 | * both inodes are subvols and we need to unlink them then that would | |
7355 | * require 4 item modifications, but if they are both normal inodes it | |
7356 | * would require 5 item modifications, so we'll assume their normal | |
7357 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7358 | * should cover the worst case number of items we'll modify. | |
7359 | */ | |
7360 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7361 | if (IS_ERR(trans)) { |
7362 | ret = PTR_ERR(trans); | |
7363 | goto out_notrans; | |
7364 | } | |
76dda93c | 7365 | |
4df27c4d YZ |
7366 | if (dest != root) |
7367 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7368 | |
a5719521 YZ |
7369 | ret = btrfs_set_inode_index(new_dir, &index); |
7370 | if (ret) | |
7371 | goto out_fail; | |
5a3f23d5 | 7372 | |
33345d01 | 7373 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7374 | /* force full log commit if subvolume involved. */ |
7375 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7376 | } else { | |
a5719521 YZ |
7377 | ret = btrfs_insert_inode_ref(trans, dest, |
7378 | new_dentry->d_name.name, | |
7379 | new_dentry->d_name.len, | |
33345d01 LZ |
7380 | old_ino, |
7381 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7382 | if (ret) |
7383 | goto out_fail; | |
4df27c4d YZ |
7384 | /* |
7385 | * this is an ugly little race, but the rename is required | |
7386 | * to make sure that if we crash, the inode is either at the | |
7387 | * old name or the new one. pinning the log transaction lets | |
7388 | * us make sure we don't allow a log commit to come in after | |
7389 | * we unlink the name but before we add the new name back in. | |
7390 | */ | |
7391 | btrfs_pin_log_trans(root); | |
7392 | } | |
5a3f23d5 CM |
7393 | /* |
7394 | * make sure the inode gets flushed if it is replacing | |
7395 | * something. | |
7396 | */ | |
33345d01 | 7397 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7398 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7399 | |
0c4d2d95 JB |
7400 | inode_inc_iversion(old_dir); |
7401 | inode_inc_iversion(new_dir); | |
7402 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7403 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7404 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7405 | old_inode->i_ctime = ctime; | |
5f39d397 | 7406 | |
12fcfd22 CM |
7407 | if (old_dentry->d_parent != new_dentry->d_parent) |
7408 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7409 | ||
33345d01 | 7410 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7411 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7412 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7413 | old_dentry->d_name.name, | |
7414 | old_dentry->d_name.len); | |
7415 | } else { | |
92986796 AV |
7416 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7417 | old_dentry->d_inode, | |
7418 | old_dentry->d_name.name, | |
7419 | old_dentry->d_name.len); | |
7420 | if (!ret) | |
7421 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7422 | } |
79787eaa JM |
7423 | if (ret) { |
7424 | btrfs_abort_transaction(trans, root, ret); | |
7425 | goto out_fail; | |
7426 | } | |
39279cc3 CM |
7427 | |
7428 | if (new_inode) { | |
0c4d2d95 | 7429 | inode_inc_iversion(new_inode); |
39279cc3 | 7430 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7431 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7432 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7433 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7434 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7435 | root_objectid, | |
7436 | new_dentry->d_name.name, | |
7437 | new_dentry->d_name.len); | |
7438 | BUG_ON(new_inode->i_nlink == 0); | |
7439 | } else { | |
7440 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7441 | new_dentry->d_inode, | |
7442 | new_dentry->d_name.name, | |
7443 | new_dentry->d_name.len); | |
7444 | } | |
79787eaa | 7445 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7446 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7447 | BUG_ON(ret); |
7b128766 | 7448 | } |
79787eaa JM |
7449 | if (ret) { |
7450 | btrfs_abort_transaction(trans, root, ret); | |
7451 | goto out_fail; | |
7452 | } | |
39279cc3 | 7453 | } |
aec7477b | 7454 | |
75e7cb7f LB |
7455 | fixup_inode_flags(new_dir, old_inode); |
7456 | ||
4df27c4d YZ |
7457 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7458 | new_dentry->d_name.name, | |
a5719521 | 7459 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7460 | if (ret) { |
7461 | btrfs_abort_transaction(trans, root, ret); | |
7462 | goto out_fail; | |
7463 | } | |
39279cc3 | 7464 | |
33345d01 | 7465 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7466 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7467 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7468 | btrfs_end_log_trans(root); |
7469 | } | |
39279cc3 | 7470 | out_fail: |
7ad85bb7 | 7471 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7472 | out_notrans: |
33345d01 | 7473 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7474 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7475 | |
39279cc3 CM |
7476 | return ret; |
7477 | } | |
7478 | ||
8ccf6f19 MX |
7479 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
7480 | { | |
7481 | struct btrfs_delalloc_work *delalloc_work; | |
7482 | ||
7483 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
7484 | work); | |
7485 | if (delalloc_work->wait) | |
7486 | btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1); | |
7487 | else | |
7488 | filemap_flush(delalloc_work->inode->i_mapping); | |
7489 | ||
7490 | if (delalloc_work->delay_iput) | |
7491 | btrfs_add_delayed_iput(delalloc_work->inode); | |
7492 | else | |
7493 | iput(delalloc_work->inode); | |
7494 | complete(&delalloc_work->completion); | |
7495 | } | |
7496 | ||
7497 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
7498 | int wait, int delay_iput) | |
7499 | { | |
7500 | struct btrfs_delalloc_work *work; | |
7501 | ||
7502 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
7503 | if (!work) | |
7504 | return NULL; | |
7505 | ||
7506 | init_completion(&work->completion); | |
7507 | INIT_LIST_HEAD(&work->list); | |
7508 | work->inode = inode; | |
7509 | work->wait = wait; | |
7510 | work->delay_iput = delay_iput; | |
7511 | work->work.func = btrfs_run_delalloc_work; | |
7512 | ||
7513 | return work; | |
7514 | } | |
7515 | ||
7516 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
7517 | { | |
7518 | wait_for_completion(&work->completion); | |
7519 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
7520 | } | |
7521 | ||
d352ac68 CM |
7522 | /* |
7523 | * some fairly slow code that needs optimization. This walks the list | |
7524 | * of all the inodes with pending delalloc and forces them to disk. | |
7525 | */ | |
24bbcf04 | 7526 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7527 | { |
7528 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7529 | struct btrfs_inode *binode; | |
5b21f2ed | 7530 | struct inode *inode; |
8ccf6f19 MX |
7531 | struct btrfs_delalloc_work *work, *next; |
7532 | struct list_head works; | |
7533 | int ret = 0; | |
ea8c2819 | 7534 | |
c146afad YZ |
7535 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7536 | return -EROFS; | |
7537 | ||
8ccf6f19 MX |
7538 | INIT_LIST_HEAD(&works); |
7539 | ||
75eff68e | 7540 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7541 | while (!list_empty(head)) { |
ea8c2819 CM |
7542 | binode = list_entry(head->next, struct btrfs_inode, |
7543 | delalloc_inodes); | |
5b21f2ed ZY |
7544 | inode = igrab(&binode->vfs_inode); |
7545 | if (!inode) | |
7546 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7547 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7548 | if (inode) { |
8ccf6f19 MX |
7549 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); |
7550 | if (!work) { | |
7551 | ret = -ENOMEM; | |
7552 | goto out; | |
7553 | } | |
7554 | list_add_tail(&work->list, &works); | |
7555 | btrfs_queue_worker(&root->fs_info->flush_workers, | |
7556 | &work->work); | |
5b21f2ed ZY |
7557 | } |
7558 | cond_resched(); | |
75eff68e | 7559 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7560 | } |
75eff68e | 7561 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7562 | |
7563 | /* the filemap_flush will queue IO into the worker threads, but | |
7564 | * we have to make sure the IO is actually started and that | |
7565 | * ordered extents get created before we return | |
7566 | */ | |
7567 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7568 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7569 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7570 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7571 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7572 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7573 | } |
7574 | atomic_dec(&root->fs_info->async_submit_draining); | |
8ccf6f19 MX |
7575 | out: |
7576 | list_for_each_entry_safe(work, next, &works, list) { | |
7577 | list_del_init(&work->list); | |
7578 | btrfs_wait_and_free_delalloc_work(work); | |
7579 | } | |
7580 | return ret; | |
ea8c2819 CM |
7581 | } |
7582 | ||
39279cc3 CM |
7583 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7584 | const char *symname) | |
7585 | { | |
7586 | struct btrfs_trans_handle *trans; | |
7587 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7588 | struct btrfs_path *path; | |
7589 | struct btrfs_key key; | |
1832a6d5 | 7590 | struct inode *inode = NULL; |
39279cc3 CM |
7591 | int err; |
7592 | int drop_inode = 0; | |
7593 | u64 objectid; | |
00e4e6b3 | 7594 | u64 index = 0 ; |
39279cc3 CM |
7595 | int name_len; |
7596 | int datasize; | |
5f39d397 | 7597 | unsigned long ptr; |
39279cc3 | 7598 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7599 | struct extent_buffer *leaf; |
39279cc3 CM |
7600 | |
7601 | name_len = strlen(symname) + 1; | |
7602 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7603 | return -ENAMETOOLONG; | |
1832a6d5 | 7604 | |
9ed74f2d JB |
7605 | /* |
7606 | * 2 items for inode item and ref | |
7607 | * 2 items for dir items | |
7608 | * 1 item for xattr if selinux is on | |
7609 | */ | |
a22285a6 YZ |
7610 | trans = btrfs_start_transaction(root, 5); |
7611 | if (IS_ERR(trans)) | |
7612 | return PTR_ERR(trans); | |
1832a6d5 | 7613 | |
581bb050 LZ |
7614 | err = btrfs_find_free_ino(root, &objectid); |
7615 | if (err) | |
7616 | goto out_unlock; | |
7617 | ||
aec7477b | 7618 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7619 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7620 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7621 | if (IS_ERR(inode)) { |
7622 | err = PTR_ERR(inode); | |
39279cc3 | 7623 | goto out_unlock; |
7cf96da3 | 7624 | } |
39279cc3 | 7625 | |
2a7dba39 | 7626 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7627 | if (err) { |
7628 | drop_inode = 1; | |
7629 | goto out_unlock; | |
7630 | } | |
7631 | ||
ad19db71 CS |
7632 | /* |
7633 | * If the active LSM wants to access the inode during | |
7634 | * d_instantiate it needs these. Smack checks to see | |
7635 | * if the filesystem supports xattrs by looking at the | |
7636 | * ops vector. | |
7637 | */ | |
7638 | inode->i_fop = &btrfs_file_operations; | |
7639 | inode->i_op = &btrfs_file_inode_operations; | |
7640 | ||
a1b075d2 | 7641 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7642 | if (err) |
7643 | drop_inode = 1; | |
7644 | else { | |
7645 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7646 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7647 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7648 | } |
39279cc3 CM |
7649 | if (drop_inode) |
7650 | goto out_unlock; | |
7651 | ||
7652 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7653 | if (!path) { |
7654 | err = -ENOMEM; | |
7655 | drop_inode = 1; | |
7656 | goto out_unlock; | |
7657 | } | |
33345d01 | 7658 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7659 | key.offset = 0; |
39279cc3 CM |
7660 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7661 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7662 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7663 | datasize); | |
54aa1f4d CM |
7664 | if (err) { |
7665 | drop_inode = 1; | |
b0839166 | 7666 | btrfs_free_path(path); |
54aa1f4d CM |
7667 | goto out_unlock; |
7668 | } | |
5f39d397 CM |
7669 | leaf = path->nodes[0]; |
7670 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7671 | struct btrfs_file_extent_item); | |
7672 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7673 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7674 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7675 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7676 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7677 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7678 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7679 | ||
39279cc3 | 7680 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7681 | write_extent_buffer(leaf, symname, ptr, name_len); |
7682 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7683 | btrfs_free_path(path); |
5f39d397 | 7684 | |
39279cc3 CM |
7685 | inode->i_op = &btrfs_symlink_inode_operations; |
7686 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7687 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7688 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7689 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7690 | err = btrfs_update_inode(trans, root, inode); |
7691 | if (err) | |
7692 | drop_inode = 1; | |
39279cc3 CM |
7693 | |
7694 | out_unlock: | |
08c422c2 AV |
7695 | if (!err) |
7696 | d_instantiate(dentry, inode); | |
7ad85bb7 | 7697 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7698 | if (drop_inode) { |
7699 | inode_dec_link_count(inode); | |
7700 | iput(inode); | |
7701 | } | |
b53d3f5d | 7702 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
7703 | return err; |
7704 | } | |
16432985 | 7705 | |
0af3d00b JB |
7706 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7707 | u64 start, u64 num_bytes, u64 min_size, | |
7708 | loff_t actual_len, u64 *alloc_hint, | |
7709 | struct btrfs_trans_handle *trans) | |
d899e052 | 7710 | { |
5dc562c5 JB |
7711 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7712 | struct extent_map *em; | |
d899e052 YZ |
7713 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7714 | struct btrfs_key ins; | |
d899e052 | 7715 | u64 cur_offset = start; |
55a61d1d | 7716 | u64 i_size; |
d899e052 | 7717 | int ret = 0; |
0af3d00b | 7718 | bool own_trans = true; |
d899e052 | 7719 | |
0af3d00b JB |
7720 | if (trans) |
7721 | own_trans = false; | |
d899e052 | 7722 | while (num_bytes > 0) { |
0af3d00b JB |
7723 | if (own_trans) { |
7724 | trans = btrfs_start_transaction(root, 3); | |
7725 | if (IS_ERR(trans)) { | |
7726 | ret = PTR_ERR(trans); | |
7727 | break; | |
7728 | } | |
5a303d5d YZ |
7729 | } |
7730 | ||
efa56464 | 7731 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7732 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7733 | if (ret) { |
0af3d00b JB |
7734 | if (own_trans) |
7735 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7736 | break; |
d899e052 | 7737 | } |
5a303d5d | 7738 | |
d899e052 YZ |
7739 | ret = insert_reserved_file_extent(trans, inode, |
7740 | cur_offset, ins.objectid, | |
7741 | ins.offset, ins.offset, | |
920bbbfb | 7742 | ins.offset, 0, 0, 0, |
d899e052 | 7743 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7744 | if (ret) { |
7745 | btrfs_abort_transaction(trans, root, ret); | |
7746 | if (own_trans) | |
7747 | btrfs_end_transaction(trans, root); | |
7748 | break; | |
7749 | } | |
a1ed835e CM |
7750 | btrfs_drop_extent_cache(inode, cur_offset, |
7751 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7752 | |
5dc562c5 JB |
7753 | em = alloc_extent_map(); |
7754 | if (!em) { | |
7755 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
7756 | &BTRFS_I(inode)->runtime_flags); | |
7757 | goto next; | |
7758 | } | |
7759 | ||
7760 | em->start = cur_offset; | |
7761 | em->orig_start = cur_offset; | |
7762 | em->len = ins.offset; | |
7763 | em->block_start = ins.objectid; | |
7764 | em->block_len = ins.offset; | |
7765 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
7766 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
7767 | em->generation = trans->transid; | |
7768 | ||
7769 | while (1) { | |
7770 | write_lock(&em_tree->lock); | |
7771 | ret = add_extent_mapping(em_tree, em); | |
7772 | if (!ret) | |
7773 | list_move(&em->list, | |
7774 | &em_tree->modified_extents); | |
7775 | write_unlock(&em_tree->lock); | |
7776 | if (ret != -EEXIST) | |
7777 | break; | |
7778 | btrfs_drop_extent_cache(inode, cur_offset, | |
7779 | cur_offset + ins.offset - 1, | |
7780 | 0); | |
7781 | } | |
7782 | free_extent_map(em); | |
7783 | next: | |
d899e052 YZ |
7784 | num_bytes -= ins.offset; |
7785 | cur_offset += ins.offset; | |
efa56464 | 7786 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7787 | |
0c4d2d95 | 7788 | inode_inc_iversion(inode); |
d899e052 | 7789 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7790 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7791 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7792 | (actual_len > inode->i_size) && |
7793 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7794 | if (cur_offset > actual_len) |
55a61d1d | 7795 | i_size = actual_len; |
d1ea6a61 | 7796 | else |
55a61d1d JB |
7797 | i_size = cur_offset; |
7798 | i_size_write(inode, i_size); | |
7799 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7800 | } |
7801 | ||
d899e052 | 7802 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7803 | |
7804 | if (ret) { | |
7805 | btrfs_abort_transaction(trans, root, ret); | |
7806 | if (own_trans) | |
7807 | btrfs_end_transaction(trans, root); | |
7808 | break; | |
7809 | } | |
d899e052 | 7810 | |
0af3d00b JB |
7811 | if (own_trans) |
7812 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7813 | } |
d899e052 YZ |
7814 | return ret; |
7815 | } | |
7816 | ||
0af3d00b JB |
7817 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7818 | u64 start, u64 num_bytes, u64 min_size, | |
7819 | loff_t actual_len, u64 *alloc_hint) | |
7820 | { | |
7821 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7822 | min_size, actual_len, alloc_hint, | |
7823 | NULL); | |
7824 | } | |
7825 | ||
7826 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7827 | struct btrfs_trans_handle *trans, int mode, | |
7828 | u64 start, u64 num_bytes, u64 min_size, | |
7829 | loff_t actual_len, u64 *alloc_hint) | |
7830 | { | |
7831 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7832 | min_size, actual_len, alloc_hint, trans); | |
7833 | } | |
7834 | ||
e6dcd2dc CM |
7835 | static int btrfs_set_page_dirty(struct page *page) |
7836 | { | |
e6dcd2dc CM |
7837 | return __set_page_dirty_nobuffers(page); |
7838 | } | |
7839 | ||
10556cb2 | 7840 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7841 | { |
b83cc969 | 7842 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7843 | umode_t mode = inode->i_mode; |
b83cc969 | 7844 | |
cb6db4e5 JM |
7845 | if (mask & MAY_WRITE && |
7846 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7847 | if (btrfs_root_readonly(root)) | |
7848 | return -EROFS; | |
7849 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7850 | return -EACCES; | |
7851 | } | |
2830ba7f | 7852 | return generic_permission(inode, mask); |
fdebe2bd | 7853 | } |
39279cc3 | 7854 | |
6e1d5dcc | 7855 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7856 | .getattr = btrfs_getattr, |
39279cc3 CM |
7857 | .lookup = btrfs_lookup, |
7858 | .create = btrfs_create, | |
7859 | .unlink = btrfs_unlink, | |
7860 | .link = btrfs_link, | |
7861 | .mkdir = btrfs_mkdir, | |
7862 | .rmdir = btrfs_rmdir, | |
7863 | .rename = btrfs_rename, | |
7864 | .symlink = btrfs_symlink, | |
7865 | .setattr = btrfs_setattr, | |
618e21d5 | 7866 | .mknod = btrfs_mknod, |
95819c05 CH |
7867 | .setxattr = btrfs_setxattr, |
7868 | .getxattr = btrfs_getxattr, | |
5103e947 | 7869 | .listxattr = btrfs_listxattr, |
95819c05 | 7870 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7871 | .permission = btrfs_permission, |
4e34e719 | 7872 | .get_acl = btrfs_get_acl, |
39279cc3 | 7873 | }; |
6e1d5dcc | 7874 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7875 | .lookup = btrfs_lookup, |
fdebe2bd | 7876 | .permission = btrfs_permission, |
4e34e719 | 7877 | .get_acl = btrfs_get_acl, |
39279cc3 | 7878 | }; |
76dda93c | 7879 | |
828c0950 | 7880 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7881 | .llseek = generic_file_llseek, |
7882 | .read = generic_read_dir, | |
cbdf5a24 | 7883 | .readdir = btrfs_real_readdir, |
34287aa3 | 7884 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7885 | #ifdef CONFIG_COMPAT |
34287aa3 | 7886 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7887 | #endif |
6bf13c0c | 7888 | .release = btrfs_release_file, |
e02119d5 | 7889 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7890 | }; |
7891 | ||
d1310b2e | 7892 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7893 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7894 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7895 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7896 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7897 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7898 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7899 | .set_bit_hook = btrfs_set_bit_hook, |
7900 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7901 | .merge_extent_hook = btrfs_merge_extent_hook, |
7902 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7903 | }; |
7904 | ||
35054394 CM |
7905 | /* |
7906 | * btrfs doesn't support the bmap operation because swapfiles | |
7907 | * use bmap to make a mapping of extents in the file. They assume | |
7908 | * these extents won't change over the life of the file and they | |
7909 | * use the bmap result to do IO directly to the drive. | |
7910 | * | |
7911 | * the btrfs bmap call would return logical addresses that aren't | |
7912 | * suitable for IO and they also will change frequently as COW | |
7913 | * operations happen. So, swapfile + btrfs == corruption. | |
7914 | * | |
7915 | * For now we're avoiding this by dropping bmap. | |
7916 | */ | |
7f09410b | 7917 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7918 | .readpage = btrfs_readpage, |
7919 | .writepage = btrfs_writepage, | |
b293f02e | 7920 | .writepages = btrfs_writepages, |
3ab2fb5a | 7921 | .readpages = btrfs_readpages, |
16432985 | 7922 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7923 | .invalidatepage = btrfs_invalidatepage, |
7924 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7925 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7926 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7927 | }; |
7928 | ||
7f09410b | 7929 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7930 | .readpage = btrfs_readpage, |
7931 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7932 | .invalidatepage = btrfs_invalidatepage, |
7933 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7934 | }; |
7935 | ||
6e1d5dcc | 7936 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7937 | .getattr = btrfs_getattr, |
7938 | .setattr = btrfs_setattr, | |
95819c05 CH |
7939 | .setxattr = btrfs_setxattr, |
7940 | .getxattr = btrfs_getxattr, | |
5103e947 | 7941 | .listxattr = btrfs_listxattr, |
95819c05 | 7942 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7943 | .permission = btrfs_permission, |
1506fcc8 | 7944 | .fiemap = btrfs_fiemap, |
4e34e719 | 7945 | .get_acl = btrfs_get_acl, |
e41f941a | 7946 | .update_time = btrfs_update_time, |
39279cc3 | 7947 | }; |
6e1d5dcc | 7948 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7949 | .getattr = btrfs_getattr, |
7950 | .setattr = btrfs_setattr, | |
fdebe2bd | 7951 | .permission = btrfs_permission, |
95819c05 CH |
7952 | .setxattr = btrfs_setxattr, |
7953 | .getxattr = btrfs_getxattr, | |
33268eaf | 7954 | .listxattr = btrfs_listxattr, |
95819c05 | 7955 | .removexattr = btrfs_removexattr, |
4e34e719 | 7956 | .get_acl = btrfs_get_acl, |
e41f941a | 7957 | .update_time = btrfs_update_time, |
618e21d5 | 7958 | }; |
6e1d5dcc | 7959 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7960 | .readlink = generic_readlink, |
7961 | .follow_link = page_follow_link_light, | |
7962 | .put_link = page_put_link, | |
f209561a | 7963 | .getattr = btrfs_getattr, |
22c44fe6 | 7964 | .setattr = btrfs_setattr, |
fdebe2bd | 7965 | .permission = btrfs_permission, |
0279b4cd JO |
7966 | .setxattr = btrfs_setxattr, |
7967 | .getxattr = btrfs_getxattr, | |
7968 | .listxattr = btrfs_listxattr, | |
7969 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7970 | .get_acl = btrfs_get_acl, |
e41f941a | 7971 | .update_time = btrfs_update_time, |
39279cc3 | 7972 | }; |
76dda93c | 7973 | |
82d339d9 | 7974 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7975 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7976 | .d_release = btrfs_dentry_release, |
76dda93c | 7977 | }; |