]>
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; | |
74 | struct kmem_cache *btrfs_trans_handle_cachep; | |
75 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 76 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 77 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
78 | |
79 | #define S_SHIFT 12 | |
80 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
81 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
82 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
83 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
84 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
85 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
86 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
87 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
88 | }; | |
89 | ||
a41ad394 JB |
90 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
91 | static int btrfs_truncate(struct inode *inode); | |
5fd02043 | 92 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
93 | static noinline int cow_file_range(struct inode *inode, |
94 | struct page *locked_page, | |
95 | u64 start, u64 end, int *page_started, | |
96 | unsigned long *nr_written, int unlock); | |
2115133f CM |
97 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
98 | struct btrfs_root *root, struct inode *inode); | |
7b128766 | 99 | |
f34f57a3 | 100 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
101 | struct inode *inode, struct inode *dir, |
102 | const struct qstr *qstr) | |
0279b4cd JO |
103 | { |
104 | int err; | |
105 | ||
f34f57a3 | 106 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 107 | if (!err) |
2a7dba39 | 108 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
109 | return err; |
110 | } | |
111 | ||
c8b97818 CM |
112 | /* |
113 | * this does all the hard work for inserting an inline extent into | |
114 | * the btree. The caller should have done a btrfs_drop_extents so that | |
115 | * no overlapping inline items exist in the btree | |
116 | */ | |
d397712b | 117 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
118 | struct btrfs_root *root, struct inode *inode, |
119 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 120 | int compress_type, |
c8b97818 CM |
121 | struct page **compressed_pages) |
122 | { | |
123 | struct btrfs_key key; | |
124 | struct btrfs_path *path; | |
125 | struct extent_buffer *leaf; | |
126 | struct page *page = NULL; | |
127 | char *kaddr; | |
128 | unsigned long ptr; | |
129 | struct btrfs_file_extent_item *ei; | |
130 | int err = 0; | |
131 | int ret; | |
132 | size_t cur_size = size; | |
133 | size_t datasize; | |
134 | unsigned long offset; | |
c8b97818 | 135 | |
fe3f566c | 136 | if (compressed_size && compressed_pages) |
c8b97818 | 137 | cur_size = compressed_size; |
c8b97818 | 138 | |
d397712b CM |
139 | path = btrfs_alloc_path(); |
140 | if (!path) | |
c8b97818 CM |
141 | return -ENOMEM; |
142 | ||
b9473439 | 143 | path->leave_spinning = 1; |
c8b97818 | 144 | |
33345d01 | 145 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
146 | key.offset = start; |
147 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
148 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
149 | ||
150 | inode_add_bytes(inode, size); | |
151 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
152 | datasize); | |
c8b97818 CM |
153 | if (ret) { |
154 | err = ret; | |
c8b97818 CM |
155 | goto fail; |
156 | } | |
157 | leaf = path->nodes[0]; | |
158 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
159 | struct btrfs_file_extent_item); | |
160 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
161 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
162 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
163 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
164 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
165 | ptr = btrfs_file_extent_inline_start(ei); | |
166 | ||
261507a0 | 167 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
168 | struct page *cpage; |
169 | int i = 0; | |
d397712b | 170 | while (compressed_size > 0) { |
c8b97818 | 171 | cpage = compressed_pages[i]; |
5b050f04 | 172 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
173 | PAGE_CACHE_SIZE); |
174 | ||
7ac687d9 | 175 | kaddr = kmap_atomic(cpage); |
c8b97818 | 176 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 177 | kunmap_atomic(kaddr); |
c8b97818 CM |
178 | |
179 | i++; | |
180 | ptr += cur_size; | |
181 | compressed_size -= cur_size; | |
182 | } | |
183 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 184 | compress_type); |
c8b97818 CM |
185 | } else { |
186 | page = find_get_page(inode->i_mapping, | |
187 | start >> PAGE_CACHE_SHIFT); | |
188 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 189 | kaddr = kmap_atomic(page); |
c8b97818 CM |
190 | offset = start & (PAGE_CACHE_SIZE - 1); |
191 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 192 | kunmap_atomic(kaddr); |
c8b97818 CM |
193 | page_cache_release(page); |
194 | } | |
195 | btrfs_mark_buffer_dirty(leaf); | |
196 | btrfs_free_path(path); | |
197 | ||
c2167754 YZ |
198 | /* |
199 | * we're an inline extent, so nobody can | |
200 | * extend the file past i_size without locking | |
201 | * a page we already have locked. | |
202 | * | |
203 | * We must do any isize and inode updates | |
204 | * before we unlock the pages. Otherwise we | |
205 | * could end up racing with unlink. | |
206 | */ | |
c8b97818 | 207 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 208 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 209 | |
79787eaa | 210 | return ret; |
c8b97818 CM |
211 | fail: |
212 | btrfs_free_path(path); | |
213 | return err; | |
214 | } | |
215 | ||
216 | ||
217 | /* | |
218 | * conditionally insert an inline extent into the file. This | |
219 | * does the checks required to make sure the data is small enough | |
220 | * to fit as an inline extent. | |
221 | */ | |
7f366cfe | 222 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
223 | struct btrfs_root *root, |
224 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 225 | size_t compressed_size, int compress_type, |
c8b97818 CM |
226 | struct page **compressed_pages) |
227 | { | |
228 | u64 isize = i_size_read(inode); | |
229 | u64 actual_end = min(end + 1, isize); | |
230 | u64 inline_len = actual_end - start; | |
231 | u64 aligned_end = (end + root->sectorsize - 1) & | |
232 | ~((u64)root->sectorsize - 1); | |
c8b97818 CM |
233 | u64 data_len = inline_len; |
234 | int ret; | |
235 | ||
236 | if (compressed_size) | |
237 | data_len = compressed_size; | |
238 | ||
239 | if (start > 0 || | |
70b99e69 | 240 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
241 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
242 | (!compressed_size && | |
243 | (actual_end & (root->sectorsize - 1)) == 0) || | |
244 | end + 1 < isize || | |
245 | data_len > root->fs_info->max_inline) { | |
246 | return 1; | |
247 | } | |
248 | ||
2671485d | 249 | ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1); |
79787eaa JM |
250 | if (ret) |
251 | return ret; | |
c8b97818 CM |
252 | |
253 | if (isize > actual_end) | |
254 | inline_len = min_t(u64, isize, actual_end); | |
255 | ret = insert_inline_extent(trans, root, inode, start, | |
256 | inline_len, compressed_size, | |
fe3f566c | 257 | compress_type, compressed_pages); |
2adcac1a | 258 | if (ret && ret != -ENOSPC) { |
79787eaa JM |
259 | btrfs_abort_transaction(trans, root, ret); |
260 | return ret; | |
2adcac1a JB |
261 | } else if (ret == -ENOSPC) { |
262 | return 1; | |
79787eaa | 263 | } |
2adcac1a | 264 | |
0ca1f7ce | 265 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 266 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
267 | return 0; |
268 | } | |
269 | ||
771ed689 CM |
270 | struct async_extent { |
271 | u64 start; | |
272 | u64 ram_size; | |
273 | u64 compressed_size; | |
274 | struct page **pages; | |
275 | unsigned long nr_pages; | |
261507a0 | 276 | int compress_type; |
771ed689 CM |
277 | struct list_head list; |
278 | }; | |
279 | ||
280 | struct async_cow { | |
281 | struct inode *inode; | |
282 | struct btrfs_root *root; | |
283 | struct page *locked_page; | |
284 | u64 start; | |
285 | u64 end; | |
286 | struct list_head extents; | |
287 | struct btrfs_work work; | |
288 | }; | |
289 | ||
290 | static noinline int add_async_extent(struct async_cow *cow, | |
291 | u64 start, u64 ram_size, | |
292 | u64 compressed_size, | |
293 | struct page **pages, | |
261507a0 LZ |
294 | unsigned long nr_pages, |
295 | int compress_type) | |
771ed689 CM |
296 | { |
297 | struct async_extent *async_extent; | |
298 | ||
299 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 300 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
301 | async_extent->start = start; |
302 | async_extent->ram_size = ram_size; | |
303 | async_extent->compressed_size = compressed_size; | |
304 | async_extent->pages = pages; | |
305 | async_extent->nr_pages = nr_pages; | |
261507a0 | 306 | async_extent->compress_type = compress_type; |
771ed689 CM |
307 | list_add_tail(&async_extent->list, &cow->extents); |
308 | return 0; | |
309 | } | |
310 | ||
d352ac68 | 311 | /* |
771ed689 CM |
312 | * we create compressed extents in two phases. The first |
313 | * phase compresses a range of pages that have already been | |
314 | * locked (both pages and state bits are locked). | |
c8b97818 | 315 | * |
771ed689 CM |
316 | * This is done inside an ordered work queue, and the compression |
317 | * is spread across many cpus. The actual IO submission is step | |
318 | * two, and the ordered work queue takes care of making sure that | |
319 | * happens in the same order things were put onto the queue by | |
320 | * writepages and friends. | |
c8b97818 | 321 | * |
771ed689 CM |
322 | * If this code finds it can't get good compression, it puts an |
323 | * entry onto the work queue to write the uncompressed bytes. This | |
324 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
325 | * are written in the same order that the flusher thread sent them |
326 | * down. | |
d352ac68 | 327 | */ |
771ed689 CM |
328 | static noinline int compress_file_range(struct inode *inode, |
329 | struct page *locked_page, | |
330 | u64 start, u64 end, | |
331 | struct async_cow *async_cow, | |
332 | int *num_added) | |
b888db2b CM |
333 | { |
334 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
335 | struct btrfs_trans_handle *trans; | |
db94535d | 336 | u64 num_bytes; |
db94535d | 337 | u64 blocksize = root->sectorsize; |
c8b97818 | 338 | u64 actual_end; |
42dc7bab | 339 | u64 isize = i_size_read(inode); |
e6dcd2dc | 340 | int ret = 0; |
c8b97818 CM |
341 | struct page **pages = NULL; |
342 | unsigned long nr_pages; | |
343 | unsigned long nr_pages_ret = 0; | |
344 | unsigned long total_compressed = 0; | |
345 | unsigned long total_in = 0; | |
346 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 347 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
348 | int i; |
349 | int will_compress; | |
261507a0 | 350 | int compress_type = root->fs_info->compress_type; |
b888db2b | 351 | |
4cb13e5d LB |
352 | /* if this is a small write inside eof, kick off a defrag */ |
353 | if ((end - start + 1) < 16 * 1024 && | |
354 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
355 | btrfs_add_inode_defrag(NULL, inode); |
356 | ||
42dc7bab | 357 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
358 | again: |
359 | will_compress = 0; | |
360 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
361 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 362 | |
f03d9301 CM |
363 | /* |
364 | * we don't want to send crud past the end of i_size through | |
365 | * compression, that's just a waste of CPU time. So, if the | |
366 | * end of the file is before the start of our current | |
367 | * requested range of bytes, we bail out to the uncompressed | |
368 | * cleanup code that can deal with all of this. | |
369 | * | |
370 | * It isn't really the fastest way to fix things, but this is a | |
371 | * very uncommon corner. | |
372 | */ | |
373 | if (actual_end <= start) | |
374 | goto cleanup_and_bail_uncompressed; | |
375 | ||
c8b97818 CM |
376 | total_compressed = actual_end - start; |
377 | ||
378 | /* we want to make sure that amount of ram required to uncompress | |
379 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
380 | * of a compressed extent to 128k. This is a crucial number |
381 | * because it also controls how easily we can spread reads across | |
382 | * cpus for decompression. | |
383 | * | |
384 | * We also want to make sure the amount of IO required to do | |
385 | * a random read is reasonably small, so we limit the size of | |
386 | * a compressed extent to 128k. | |
c8b97818 CM |
387 | */ |
388 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 389 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 390 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
391 | total_in = 0; |
392 | ret = 0; | |
db94535d | 393 | |
771ed689 CM |
394 | /* |
395 | * we do compression for mount -o compress and when the | |
396 | * inode has not been flagged as nocompress. This flag can | |
397 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 398 | */ |
6cbff00f | 399 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 400 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
401 | (BTRFS_I(inode)->force_compress) || |
402 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 403 | WARN_ON(pages); |
cfbc246e | 404 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
405 | if (!pages) { |
406 | /* just bail out to the uncompressed code */ | |
407 | goto cont; | |
408 | } | |
c8b97818 | 409 | |
261507a0 LZ |
410 | if (BTRFS_I(inode)->force_compress) |
411 | compress_type = BTRFS_I(inode)->force_compress; | |
412 | ||
413 | ret = btrfs_compress_pages(compress_type, | |
414 | inode->i_mapping, start, | |
415 | total_compressed, pages, | |
416 | nr_pages, &nr_pages_ret, | |
417 | &total_in, | |
418 | &total_compressed, | |
419 | max_compressed); | |
c8b97818 CM |
420 | |
421 | if (!ret) { | |
422 | unsigned long offset = total_compressed & | |
423 | (PAGE_CACHE_SIZE - 1); | |
424 | struct page *page = pages[nr_pages_ret - 1]; | |
425 | char *kaddr; | |
426 | ||
427 | /* zero the tail end of the last page, we might be | |
428 | * sending it down to disk | |
429 | */ | |
430 | if (offset) { | |
7ac687d9 | 431 | kaddr = kmap_atomic(page); |
c8b97818 CM |
432 | memset(kaddr + offset, 0, |
433 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 434 | kunmap_atomic(kaddr); |
c8b97818 CM |
435 | } |
436 | will_compress = 1; | |
437 | } | |
438 | } | |
560f7d75 | 439 | cont: |
c8b97818 | 440 | if (start == 0) { |
7a7eaa40 | 441 | trans = btrfs_join_transaction(root); |
79787eaa JM |
442 | if (IS_ERR(trans)) { |
443 | ret = PTR_ERR(trans); | |
444 | trans = NULL; | |
445 | goto cleanup_and_out; | |
446 | } | |
0ca1f7ce | 447 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 448 | |
c8b97818 | 449 | /* lets try to make an inline extent */ |
771ed689 | 450 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 451 | /* we didn't compress the entire range, try |
771ed689 | 452 | * to make an uncompressed inline extent. |
c8b97818 CM |
453 | */ |
454 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 455 | start, end, 0, 0, NULL); |
c8b97818 | 456 | } else { |
771ed689 | 457 | /* try making a compressed inline extent */ |
c8b97818 CM |
458 | ret = cow_file_range_inline(trans, root, inode, |
459 | start, end, | |
fe3f566c LZ |
460 | total_compressed, |
461 | compress_type, pages); | |
c8b97818 | 462 | } |
79787eaa | 463 | if (ret <= 0) { |
771ed689 | 464 | /* |
79787eaa JM |
465 | * inline extent creation worked or returned error, |
466 | * we don't need to create any more async work items. | |
467 | * Unlock and free up our temp pages. | |
771ed689 | 468 | */ |
c8b97818 | 469 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
470 | &BTRFS_I(inode)->io_tree, |
471 | start, end, NULL, | |
472 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 473 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 474 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
475 | |
476 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
477 | goto free_pages_out; |
478 | } | |
c2167754 | 479 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
480 | } |
481 | ||
482 | if (will_compress) { | |
483 | /* | |
484 | * we aren't doing an inline extent round the compressed size | |
485 | * up to a block size boundary so the allocator does sane | |
486 | * things | |
487 | */ | |
488 | total_compressed = (total_compressed + blocksize - 1) & | |
489 | ~(blocksize - 1); | |
490 | ||
491 | /* | |
492 | * one last check to make sure the compression is really a | |
493 | * win, compare the page count read with the blocks on disk | |
494 | */ | |
495 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
496 | ~(PAGE_CACHE_SIZE - 1); | |
497 | if (total_compressed >= total_in) { | |
498 | will_compress = 0; | |
499 | } else { | |
c8b97818 CM |
500 | num_bytes = total_in; |
501 | } | |
502 | } | |
503 | if (!will_compress && pages) { | |
504 | /* | |
505 | * the compression code ran but failed to make things smaller, | |
506 | * free any pages it allocated and our page pointer array | |
507 | */ | |
508 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 509 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
510 | page_cache_release(pages[i]); |
511 | } | |
512 | kfree(pages); | |
513 | pages = NULL; | |
514 | total_compressed = 0; | |
515 | nr_pages_ret = 0; | |
516 | ||
517 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
518 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
519 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 520 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 521 | } |
c8b97818 | 522 | } |
771ed689 CM |
523 | if (will_compress) { |
524 | *num_added += 1; | |
c8b97818 | 525 | |
771ed689 CM |
526 | /* the async work queues will take care of doing actual |
527 | * allocation on disk for these compressed pages, | |
528 | * and will submit them to the elevator. | |
529 | */ | |
530 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
531 | total_compressed, pages, nr_pages_ret, |
532 | compress_type); | |
179e29e4 | 533 | |
24ae6365 | 534 | if (start + num_bytes < end) { |
771ed689 CM |
535 | start += num_bytes; |
536 | pages = NULL; | |
537 | cond_resched(); | |
538 | goto again; | |
539 | } | |
540 | } else { | |
f03d9301 | 541 | cleanup_and_bail_uncompressed: |
771ed689 CM |
542 | /* |
543 | * No compression, but we still need to write the pages in | |
544 | * the file we've been given so far. redirty the locked | |
545 | * page if it corresponds to our extent and set things up | |
546 | * for the async work queue to run cow_file_range to do | |
547 | * the normal delalloc dance | |
548 | */ | |
549 | if (page_offset(locked_page) >= start && | |
550 | page_offset(locked_page) <= end) { | |
551 | __set_page_dirty_nobuffers(locked_page); | |
552 | /* unlocked later on in the async handlers */ | |
553 | } | |
261507a0 LZ |
554 | add_async_extent(async_cow, start, end - start + 1, |
555 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
556 | *num_added += 1; |
557 | } | |
3b951516 | 558 | |
771ed689 | 559 | out: |
79787eaa | 560 | return ret; |
771ed689 CM |
561 | |
562 | free_pages_out: | |
563 | for (i = 0; i < nr_pages_ret; i++) { | |
564 | WARN_ON(pages[i]->mapping); | |
565 | page_cache_release(pages[i]); | |
566 | } | |
d397712b | 567 | kfree(pages); |
771ed689 CM |
568 | |
569 | goto out; | |
79787eaa JM |
570 | |
571 | cleanup_and_out: | |
572 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, | |
573 | start, end, NULL, | |
574 | EXTENT_CLEAR_UNLOCK_PAGE | | |
575 | EXTENT_CLEAR_DIRTY | | |
576 | EXTENT_CLEAR_DELALLOC | | |
577 | EXTENT_SET_WRITEBACK | | |
578 | EXTENT_END_WRITEBACK); | |
579 | if (!trans || IS_ERR(trans)) | |
580 | btrfs_error(root->fs_info, ret, "Failed to join transaction"); | |
581 | else | |
582 | btrfs_abort_transaction(trans, root, ret); | |
583 | goto free_pages_out; | |
771ed689 CM |
584 | } |
585 | ||
586 | /* | |
587 | * phase two of compressed writeback. This is the ordered portion | |
588 | * of the code, which only gets called in the order the work was | |
589 | * queued. We walk all the async extents created by compress_file_range | |
590 | * and send them down to the disk. | |
591 | */ | |
592 | static noinline int submit_compressed_extents(struct inode *inode, | |
593 | struct async_cow *async_cow) | |
594 | { | |
595 | struct async_extent *async_extent; | |
596 | u64 alloc_hint = 0; | |
597 | struct btrfs_trans_handle *trans; | |
598 | struct btrfs_key ins; | |
599 | struct extent_map *em; | |
600 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
601 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
602 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 603 | int ret = 0; |
771ed689 CM |
604 | |
605 | if (list_empty(&async_cow->extents)) | |
606 | return 0; | |
607 | ||
771ed689 | 608 | |
d397712b | 609 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
610 | async_extent = list_entry(async_cow->extents.next, |
611 | struct async_extent, list); | |
612 | list_del(&async_extent->list); | |
c8b97818 | 613 | |
771ed689 CM |
614 | io_tree = &BTRFS_I(inode)->io_tree; |
615 | ||
f5a84ee3 | 616 | retry: |
771ed689 CM |
617 | /* did the compression code fall back to uncompressed IO? */ |
618 | if (!async_extent->pages) { | |
619 | int page_started = 0; | |
620 | unsigned long nr_written = 0; | |
621 | ||
622 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 623 | async_extent->start + |
d0082371 | 624 | async_extent->ram_size - 1); |
771ed689 CM |
625 | |
626 | /* allocate blocks */ | |
f5a84ee3 JB |
627 | ret = cow_file_range(inode, async_cow->locked_page, |
628 | async_extent->start, | |
629 | async_extent->start + | |
630 | async_extent->ram_size - 1, | |
631 | &page_started, &nr_written, 0); | |
771ed689 | 632 | |
79787eaa JM |
633 | /* JDM XXX */ |
634 | ||
771ed689 CM |
635 | /* |
636 | * if page_started, cow_file_range inserted an | |
637 | * inline extent and took care of all the unlocking | |
638 | * and IO for us. Otherwise, we need to submit | |
639 | * all those pages down to the drive. | |
640 | */ | |
f5a84ee3 | 641 | if (!page_started && !ret) |
771ed689 CM |
642 | extent_write_locked_range(io_tree, |
643 | inode, async_extent->start, | |
d397712b | 644 | async_extent->start + |
771ed689 CM |
645 | async_extent->ram_size - 1, |
646 | btrfs_get_extent, | |
647 | WB_SYNC_ALL); | |
648 | kfree(async_extent); | |
649 | cond_resched(); | |
650 | continue; | |
651 | } | |
652 | ||
653 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 654 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 655 | |
7a7eaa40 | 656 | trans = btrfs_join_transaction(root); |
79787eaa JM |
657 | if (IS_ERR(trans)) { |
658 | ret = PTR_ERR(trans); | |
659 | } else { | |
660 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
661 | ret = btrfs_reserve_extent(trans, root, | |
771ed689 CM |
662 | async_extent->compressed_size, |
663 | async_extent->compressed_size, | |
81c9ad23 | 664 | 0, alloc_hint, &ins, 1); |
962197ba | 665 | if (ret && ret != -ENOSPC) |
79787eaa JM |
666 | btrfs_abort_transaction(trans, root, ret); |
667 | btrfs_end_transaction(trans, root); | |
668 | } | |
c2167754 | 669 | |
f5a84ee3 JB |
670 | if (ret) { |
671 | int i; | |
672 | for (i = 0; i < async_extent->nr_pages; i++) { | |
673 | WARN_ON(async_extent->pages[i]->mapping); | |
674 | page_cache_release(async_extent->pages[i]); | |
675 | } | |
676 | kfree(async_extent->pages); | |
677 | async_extent->nr_pages = 0; | |
678 | async_extent->pages = NULL; | |
679 | unlock_extent(io_tree, async_extent->start, | |
680 | async_extent->start + | |
d0082371 | 681 | async_extent->ram_size - 1); |
79787eaa JM |
682 | if (ret == -ENOSPC) |
683 | goto retry; | |
684 | goto out_free; /* JDM: Requeue? */ | |
f5a84ee3 JB |
685 | } |
686 | ||
c2167754 YZ |
687 | /* |
688 | * here we're doing allocation and writeback of the | |
689 | * compressed pages | |
690 | */ | |
691 | btrfs_drop_extent_cache(inode, async_extent->start, | |
692 | async_extent->start + | |
693 | async_extent->ram_size - 1, 0); | |
694 | ||
172ddd60 | 695 | em = alloc_extent_map(); |
79787eaa | 696 | BUG_ON(!em); /* -ENOMEM */ |
771ed689 CM |
697 | em->start = async_extent->start; |
698 | em->len = async_extent->ram_size; | |
445a6944 | 699 | em->orig_start = em->start; |
c8b97818 | 700 | |
771ed689 CM |
701 | em->block_start = ins.objectid; |
702 | em->block_len = ins.offset; | |
703 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 704 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
705 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
706 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
707 | ||
d397712b | 708 | while (1) { |
890871be | 709 | write_lock(&em_tree->lock); |
771ed689 | 710 | ret = add_extent_mapping(em_tree, em); |
890871be | 711 | write_unlock(&em_tree->lock); |
771ed689 CM |
712 | if (ret != -EEXIST) { |
713 | free_extent_map(em); | |
714 | break; | |
715 | } | |
716 | btrfs_drop_extent_cache(inode, async_extent->start, | |
717 | async_extent->start + | |
718 | async_extent->ram_size - 1, 0); | |
719 | } | |
720 | ||
261507a0 LZ |
721 | ret = btrfs_add_ordered_extent_compress(inode, |
722 | async_extent->start, | |
723 | ins.objectid, | |
724 | async_extent->ram_size, | |
725 | ins.offset, | |
726 | BTRFS_ORDERED_COMPRESSED, | |
727 | async_extent->compress_type); | |
79787eaa | 728 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 729 | |
771ed689 CM |
730 | /* |
731 | * clear dirty, set writeback and unlock the pages. | |
732 | */ | |
733 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
734 | &BTRFS_I(inode)->io_tree, |
735 | async_extent->start, | |
736 | async_extent->start + | |
737 | async_extent->ram_size - 1, | |
738 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
739 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 740 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 741 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
742 | |
743 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
744 | async_extent->start, |
745 | async_extent->ram_size, | |
746 | ins.objectid, | |
747 | ins.offset, async_extent->pages, | |
748 | async_extent->nr_pages); | |
771ed689 | 749 | |
79787eaa | 750 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 CM |
751 | alloc_hint = ins.objectid + ins.offset; |
752 | kfree(async_extent); | |
753 | cond_resched(); | |
754 | } | |
79787eaa JM |
755 | ret = 0; |
756 | out: | |
757 | return ret; | |
758 | out_free: | |
759 | kfree(async_extent); | |
760 | goto out; | |
771ed689 CM |
761 | } |
762 | ||
4b46fce2 JB |
763 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
764 | u64 num_bytes) | |
765 | { | |
766 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
767 | struct extent_map *em; | |
768 | u64 alloc_hint = 0; | |
769 | ||
770 | read_lock(&em_tree->lock); | |
771 | em = search_extent_mapping(em_tree, start, num_bytes); | |
772 | if (em) { | |
773 | /* | |
774 | * if block start isn't an actual block number then find the | |
775 | * first block in this inode and use that as a hint. If that | |
776 | * block is also bogus then just don't worry about it. | |
777 | */ | |
778 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
779 | free_extent_map(em); | |
780 | em = search_extent_mapping(em_tree, 0, 0); | |
781 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
782 | alloc_hint = em->block_start; | |
783 | if (em) | |
784 | free_extent_map(em); | |
785 | } else { | |
786 | alloc_hint = em->block_start; | |
787 | free_extent_map(em); | |
788 | } | |
789 | } | |
790 | read_unlock(&em_tree->lock); | |
791 | ||
792 | return alloc_hint; | |
793 | } | |
794 | ||
771ed689 CM |
795 | /* |
796 | * when extent_io.c finds a delayed allocation range in the file, | |
797 | * the call backs end up in this code. The basic idea is to | |
798 | * allocate extents on disk for the range, and create ordered data structs | |
799 | * in ram to track those extents. | |
800 | * | |
801 | * locked_page is the page that writepage had locked already. We use | |
802 | * it to make sure we don't do extra locks or unlocks. | |
803 | * | |
804 | * *page_started is set to one if we unlock locked_page and do everything | |
805 | * required to start IO on it. It may be clean and already done with | |
806 | * IO when we return. | |
807 | */ | |
808 | static noinline int cow_file_range(struct inode *inode, | |
809 | struct page *locked_page, | |
810 | u64 start, u64 end, int *page_started, | |
811 | unsigned long *nr_written, | |
812 | int unlock) | |
813 | { | |
814 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
815 | struct btrfs_trans_handle *trans; | |
816 | u64 alloc_hint = 0; | |
817 | u64 num_bytes; | |
818 | unsigned long ram_size; | |
819 | u64 disk_num_bytes; | |
820 | u64 cur_alloc_size; | |
821 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
822 | struct btrfs_key ins; |
823 | struct extent_map *em; | |
824 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
825 | int ret = 0; | |
826 | ||
83eea1f1 | 827 | BUG_ON(btrfs_is_free_space_inode(inode)); |
7a7eaa40 | 828 | trans = btrfs_join_transaction(root); |
79787eaa JM |
829 | if (IS_ERR(trans)) { |
830 | extent_clear_unlock_delalloc(inode, | |
831 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 832 | start, end, locked_page, |
79787eaa JM |
833 | EXTENT_CLEAR_UNLOCK_PAGE | |
834 | EXTENT_CLEAR_UNLOCK | | |
835 | EXTENT_CLEAR_DELALLOC | | |
836 | EXTENT_CLEAR_DIRTY | | |
837 | EXTENT_SET_WRITEBACK | | |
838 | EXTENT_END_WRITEBACK); | |
839 | return PTR_ERR(trans); | |
840 | } | |
0ca1f7ce | 841 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 842 | |
771ed689 CM |
843 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
844 | num_bytes = max(blocksize, num_bytes); | |
845 | disk_num_bytes = num_bytes; | |
846 | ret = 0; | |
847 | ||
4cb5300b | 848 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
849 | if (num_bytes < 64 * 1024 && |
850 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
851 | btrfs_add_inode_defrag(trans, inode); |
852 | ||
771ed689 CM |
853 | if (start == 0) { |
854 | /* lets try to make an inline extent */ | |
855 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 856 | start, end, 0, 0, NULL); |
771ed689 CM |
857 | if (ret == 0) { |
858 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
859 | &BTRFS_I(inode)->io_tree, |
860 | start, end, NULL, | |
861 | EXTENT_CLEAR_UNLOCK_PAGE | | |
862 | EXTENT_CLEAR_UNLOCK | | |
863 | EXTENT_CLEAR_DELALLOC | | |
864 | EXTENT_CLEAR_DIRTY | | |
865 | EXTENT_SET_WRITEBACK | | |
866 | EXTENT_END_WRITEBACK); | |
c2167754 | 867 | |
771ed689 CM |
868 | *nr_written = *nr_written + |
869 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
870 | *page_started = 1; | |
771ed689 | 871 | goto out; |
79787eaa JM |
872 | } else if (ret < 0) { |
873 | btrfs_abort_transaction(trans, root, ret); | |
874 | goto out_unlock; | |
771ed689 CM |
875 | } |
876 | } | |
877 | ||
878 | BUG_ON(disk_num_bytes > | |
6c41761f | 879 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 880 | |
4b46fce2 | 881 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
882 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
883 | ||
d397712b | 884 | while (disk_num_bytes > 0) { |
a791e35e CM |
885 | unsigned long op; |
886 | ||
287a0ab9 | 887 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 888 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 889 | root->sectorsize, 0, alloc_hint, |
81c9ad23 | 890 | &ins, 1); |
79787eaa JM |
891 | if (ret < 0) { |
892 | btrfs_abort_transaction(trans, root, ret); | |
893 | goto out_unlock; | |
894 | } | |
d397712b | 895 | |
172ddd60 | 896 | em = alloc_extent_map(); |
79787eaa | 897 | BUG_ON(!em); /* -ENOMEM */ |
e6dcd2dc | 898 | em->start = start; |
445a6944 | 899 | em->orig_start = em->start; |
771ed689 CM |
900 | ram_size = ins.offset; |
901 | em->len = ins.offset; | |
c8b97818 | 902 | |
e6dcd2dc | 903 | em->block_start = ins.objectid; |
c8b97818 | 904 | em->block_len = ins.offset; |
e6dcd2dc | 905 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 906 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 907 | |
d397712b | 908 | while (1) { |
890871be | 909 | write_lock(&em_tree->lock); |
e6dcd2dc | 910 | ret = add_extent_mapping(em_tree, em); |
890871be | 911 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
912 | if (ret != -EEXIST) { |
913 | free_extent_map(em); | |
914 | break; | |
915 | } | |
916 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 917 | start + ram_size - 1, 0); |
e6dcd2dc CM |
918 | } |
919 | ||
98d20f67 | 920 | cur_alloc_size = ins.offset; |
e6dcd2dc | 921 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 922 | ram_size, cur_alloc_size, 0); |
79787eaa | 923 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 924 | |
17d217fe YZ |
925 | if (root->root_key.objectid == |
926 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
927 | ret = btrfs_reloc_clone_csums(inode, start, | |
928 | cur_alloc_size); | |
79787eaa JM |
929 | if (ret) { |
930 | btrfs_abort_transaction(trans, root, ret); | |
931 | goto out_unlock; | |
932 | } | |
17d217fe YZ |
933 | } |
934 | ||
d397712b | 935 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 936 | break; |
d397712b | 937 | |
c8b97818 CM |
938 | /* we're not doing compressed IO, don't unlock the first |
939 | * page (which the caller expects to stay locked), don't | |
940 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
941 | * |
942 | * Do set the Private2 bit so we know this page was properly | |
943 | * setup for writepage | |
c8b97818 | 944 | */ |
a791e35e CM |
945 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
946 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
947 | EXTENT_SET_PRIVATE2; | |
948 | ||
c8b97818 CM |
949 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
950 | start, start + ram_size - 1, | |
a791e35e | 951 | locked_page, op); |
c8b97818 | 952 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
953 | num_bytes -= cur_alloc_size; |
954 | alloc_hint = ins.objectid + ins.offset; | |
955 | start += cur_alloc_size; | |
b888db2b | 956 | } |
771ed689 | 957 | ret = 0; |
79787eaa | 958 | out: |
b888db2b | 959 | btrfs_end_transaction(trans, root); |
c8b97818 | 960 | |
be20aa9d | 961 | return ret; |
79787eaa JM |
962 | out_unlock: |
963 | extent_clear_unlock_delalloc(inode, | |
964 | &BTRFS_I(inode)->io_tree, | |
beb42dd7 | 965 | start, end, locked_page, |
79787eaa JM |
966 | EXTENT_CLEAR_UNLOCK_PAGE | |
967 | EXTENT_CLEAR_UNLOCK | | |
968 | EXTENT_CLEAR_DELALLOC | | |
969 | EXTENT_CLEAR_DIRTY | | |
970 | EXTENT_SET_WRITEBACK | | |
971 | EXTENT_END_WRITEBACK); | |
972 | ||
973 | goto out; | |
771ed689 | 974 | } |
c8b97818 | 975 | |
771ed689 CM |
976 | /* |
977 | * work queue call back to started compression on a file and pages | |
978 | */ | |
979 | static noinline void async_cow_start(struct btrfs_work *work) | |
980 | { | |
981 | struct async_cow *async_cow; | |
982 | int num_added = 0; | |
983 | async_cow = container_of(work, struct async_cow, work); | |
984 | ||
985 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
986 | async_cow->start, async_cow->end, async_cow, | |
987 | &num_added); | |
8180ef88 | 988 | if (num_added == 0) { |
cb77fcd8 | 989 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 990 | async_cow->inode = NULL; |
8180ef88 | 991 | } |
771ed689 CM |
992 | } |
993 | ||
994 | /* | |
995 | * work queue call back to submit previously compressed pages | |
996 | */ | |
997 | static noinline void async_cow_submit(struct btrfs_work *work) | |
998 | { | |
999 | struct async_cow *async_cow; | |
1000 | struct btrfs_root *root; | |
1001 | unsigned long nr_pages; | |
1002 | ||
1003 | async_cow = container_of(work, struct async_cow, work); | |
1004 | ||
1005 | root = async_cow->root; | |
1006 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1007 | PAGE_CACHE_SHIFT; | |
1008 | ||
66657b31 | 1009 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1010 | 5 * 1024 * 1024 && |
771ed689 CM |
1011 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1012 | wake_up(&root->fs_info->async_submit_wait); | |
1013 | ||
d397712b | 1014 | if (async_cow->inode) |
771ed689 | 1015 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1016 | } |
c8b97818 | 1017 | |
771ed689 CM |
1018 | static noinline void async_cow_free(struct btrfs_work *work) |
1019 | { | |
1020 | struct async_cow *async_cow; | |
1021 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1022 | if (async_cow->inode) |
cb77fcd8 | 1023 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1024 | kfree(async_cow); |
1025 | } | |
1026 | ||
1027 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1028 | u64 start, u64 end, int *page_started, | |
1029 | unsigned long *nr_written) | |
1030 | { | |
1031 | struct async_cow *async_cow; | |
1032 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1033 | unsigned long nr_pages; | |
1034 | u64 cur_end; | |
287082b0 | 1035 | int limit = 10 * 1024 * 1024; |
771ed689 | 1036 | |
a3429ab7 CM |
1037 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1038 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1039 | while (start < end) { |
771ed689 | 1040 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1041 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1042 | async_cow->inode = igrab(inode); |
771ed689 CM |
1043 | async_cow->root = root; |
1044 | async_cow->locked_page = locked_page; | |
1045 | async_cow->start = start; | |
1046 | ||
6cbff00f | 1047 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
1048 | cur_end = end; |
1049 | else | |
1050 | cur_end = min(end, start + 512 * 1024 - 1); | |
1051 | ||
1052 | async_cow->end = cur_end; | |
1053 | INIT_LIST_HEAD(&async_cow->extents); | |
1054 | ||
1055 | async_cow->work.func = async_cow_start; | |
1056 | async_cow->work.ordered_func = async_cow_submit; | |
1057 | async_cow->work.ordered_free = async_cow_free; | |
1058 | async_cow->work.flags = 0; | |
1059 | ||
771ed689 CM |
1060 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1061 | PAGE_CACHE_SHIFT; | |
1062 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1063 | ||
1064 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
1065 | &async_cow->work); | |
1066 | ||
1067 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1068 | wait_event(root->fs_info->async_submit_wait, | |
1069 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1070 | limit)); | |
1071 | } | |
1072 | ||
d397712b | 1073 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1074 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1075 | wait_event(root->fs_info->async_submit_wait, | |
1076 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1077 | 0)); | |
1078 | } | |
1079 | ||
1080 | *nr_written += nr_pages; | |
1081 | start = cur_end + 1; | |
1082 | } | |
1083 | *page_started = 1; | |
1084 | return 0; | |
be20aa9d CM |
1085 | } |
1086 | ||
d397712b | 1087 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1088 | u64 bytenr, u64 num_bytes) |
1089 | { | |
1090 | int ret; | |
1091 | struct btrfs_ordered_sum *sums; | |
1092 | LIST_HEAD(list); | |
1093 | ||
07d400a6 | 1094 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1095 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1096 | if (ret == 0 && list_empty(&list)) |
1097 | return 0; | |
1098 | ||
1099 | while (!list_empty(&list)) { | |
1100 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1101 | list_del(&sums->list); | |
1102 | kfree(sums); | |
1103 | } | |
1104 | return 1; | |
1105 | } | |
1106 | ||
d352ac68 CM |
1107 | /* |
1108 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1109 | * of the extents that exist in the file, and COWs the file as required. | |
1110 | * | |
1111 | * If no cow copies or snapshots exist, we write directly to the existing | |
1112 | * blocks on disk | |
1113 | */ | |
7f366cfe CM |
1114 | static noinline int run_delalloc_nocow(struct inode *inode, |
1115 | struct page *locked_page, | |
771ed689 CM |
1116 | u64 start, u64 end, int *page_started, int force, |
1117 | unsigned long *nr_written) | |
be20aa9d | 1118 | { |
be20aa9d | 1119 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1120 | struct btrfs_trans_handle *trans; |
be20aa9d | 1121 | struct extent_buffer *leaf; |
be20aa9d | 1122 | struct btrfs_path *path; |
80ff3856 | 1123 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1124 | struct btrfs_key found_key; |
80ff3856 YZ |
1125 | u64 cow_start; |
1126 | u64 cur_offset; | |
1127 | u64 extent_end; | |
5d4f98a2 | 1128 | u64 extent_offset; |
80ff3856 YZ |
1129 | u64 disk_bytenr; |
1130 | u64 num_bytes; | |
1131 | int extent_type; | |
79787eaa | 1132 | int ret, err; |
d899e052 | 1133 | int type; |
80ff3856 YZ |
1134 | int nocow; |
1135 | int check_prev = 1; | |
82d5902d | 1136 | bool nolock; |
33345d01 | 1137 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1138 | |
1139 | path = btrfs_alloc_path(); | |
17ca04af JB |
1140 | if (!path) { |
1141 | extent_clear_unlock_delalloc(inode, | |
1142 | &BTRFS_I(inode)->io_tree, | |
1143 | start, end, locked_page, | |
1144 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1145 | EXTENT_CLEAR_UNLOCK | | |
1146 | EXTENT_CLEAR_DELALLOC | | |
1147 | EXTENT_CLEAR_DIRTY | | |
1148 | EXTENT_SET_WRITEBACK | | |
1149 | EXTENT_END_WRITEBACK); | |
d8926bb3 | 1150 | return -ENOMEM; |
17ca04af | 1151 | } |
82d5902d | 1152 | |
83eea1f1 | 1153 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1154 | |
1155 | if (nolock) | |
7a7eaa40 | 1156 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1157 | else |
7a7eaa40 | 1158 | trans = btrfs_join_transaction(root); |
ff5714cc | 1159 | |
79787eaa | 1160 | if (IS_ERR(trans)) { |
17ca04af JB |
1161 | extent_clear_unlock_delalloc(inode, |
1162 | &BTRFS_I(inode)->io_tree, | |
1163 | start, end, locked_page, | |
1164 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1165 | EXTENT_CLEAR_UNLOCK | | |
1166 | EXTENT_CLEAR_DELALLOC | | |
1167 | EXTENT_CLEAR_DIRTY | | |
1168 | EXTENT_SET_WRITEBACK | | |
1169 | EXTENT_END_WRITEBACK); | |
79787eaa JM |
1170 | btrfs_free_path(path); |
1171 | return PTR_ERR(trans); | |
1172 | } | |
1173 | ||
74b21075 | 1174 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1175 | |
80ff3856 YZ |
1176 | cow_start = (u64)-1; |
1177 | cur_offset = start; | |
1178 | while (1) { | |
33345d01 | 1179 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1180 | cur_offset, 0); |
79787eaa JM |
1181 | if (ret < 0) { |
1182 | btrfs_abort_transaction(trans, root, ret); | |
1183 | goto error; | |
1184 | } | |
80ff3856 YZ |
1185 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1186 | leaf = path->nodes[0]; | |
1187 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1188 | path->slots[0] - 1); | |
33345d01 | 1189 | if (found_key.objectid == ino && |
80ff3856 YZ |
1190 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1191 | path->slots[0]--; | |
1192 | } | |
1193 | check_prev = 0; | |
1194 | next_slot: | |
1195 | leaf = path->nodes[0]; | |
1196 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1197 | ret = btrfs_next_leaf(root, path); | |
79787eaa JM |
1198 | if (ret < 0) { |
1199 | btrfs_abort_transaction(trans, root, ret); | |
1200 | goto error; | |
1201 | } | |
80ff3856 YZ |
1202 | if (ret > 0) |
1203 | break; | |
1204 | leaf = path->nodes[0]; | |
1205 | } | |
be20aa9d | 1206 | |
80ff3856 YZ |
1207 | nocow = 0; |
1208 | disk_bytenr = 0; | |
17d217fe | 1209 | num_bytes = 0; |
80ff3856 YZ |
1210 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1211 | ||
33345d01 | 1212 | if (found_key.objectid > ino || |
80ff3856 YZ |
1213 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1214 | found_key.offset > end) | |
1215 | break; | |
1216 | ||
1217 | if (found_key.offset > cur_offset) { | |
1218 | extent_end = found_key.offset; | |
e9061e21 | 1219 | extent_type = 0; |
80ff3856 YZ |
1220 | goto out_check; |
1221 | } | |
1222 | ||
1223 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1224 | struct btrfs_file_extent_item); | |
1225 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1226 | ||
d899e052 YZ |
1227 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1228 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1229 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1230 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1231 | extent_end = found_key.offset + |
1232 | btrfs_file_extent_num_bytes(leaf, fi); | |
1233 | if (extent_end <= start) { | |
1234 | path->slots[0]++; | |
1235 | goto next_slot; | |
1236 | } | |
17d217fe YZ |
1237 | if (disk_bytenr == 0) |
1238 | goto out_check; | |
80ff3856 YZ |
1239 | if (btrfs_file_extent_compression(leaf, fi) || |
1240 | btrfs_file_extent_encryption(leaf, fi) || | |
1241 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1242 | goto out_check; | |
d899e052 YZ |
1243 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1244 | goto out_check; | |
d2fb3437 | 1245 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1246 | goto out_check; |
33345d01 | 1247 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1248 | found_key.offset - |
1249 | extent_offset, disk_bytenr)) | |
17d217fe | 1250 | goto out_check; |
5d4f98a2 | 1251 | disk_bytenr += extent_offset; |
17d217fe YZ |
1252 | disk_bytenr += cur_offset - found_key.offset; |
1253 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1254 | /* | |
1255 | * force cow if csum exists in the range. | |
1256 | * this ensure that csum for a given extent are | |
1257 | * either valid or do not exist. | |
1258 | */ | |
1259 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1260 | goto out_check; | |
80ff3856 YZ |
1261 | nocow = 1; |
1262 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1263 | extent_end = found_key.offset + | |
1264 | btrfs_file_extent_inline_len(leaf, fi); | |
1265 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1266 | } else { | |
1267 | BUG_ON(1); | |
1268 | } | |
1269 | out_check: | |
1270 | if (extent_end <= start) { | |
1271 | path->slots[0]++; | |
1272 | goto next_slot; | |
1273 | } | |
1274 | if (!nocow) { | |
1275 | if (cow_start == (u64)-1) | |
1276 | cow_start = cur_offset; | |
1277 | cur_offset = extent_end; | |
1278 | if (cur_offset > end) | |
1279 | break; | |
1280 | path->slots[0]++; | |
1281 | goto next_slot; | |
7ea394f1 YZ |
1282 | } |
1283 | ||
b3b4aa74 | 1284 | btrfs_release_path(path); |
80ff3856 YZ |
1285 | if (cow_start != (u64)-1) { |
1286 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1287 | found_key.offset - 1, page_started, |
1288 | nr_written, 1); | |
79787eaa JM |
1289 | if (ret) { |
1290 | btrfs_abort_transaction(trans, root, ret); | |
1291 | goto error; | |
1292 | } | |
80ff3856 | 1293 | cow_start = (u64)-1; |
7ea394f1 | 1294 | } |
80ff3856 | 1295 | |
d899e052 YZ |
1296 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1297 | struct extent_map *em; | |
1298 | struct extent_map_tree *em_tree; | |
1299 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1300 | em = alloc_extent_map(); |
79787eaa | 1301 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1302 | em->start = cur_offset; |
445a6944 | 1303 | em->orig_start = em->start; |
d899e052 YZ |
1304 | em->len = num_bytes; |
1305 | em->block_len = num_bytes; | |
1306 | em->block_start = disk_bytenr; | |
1307 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1308 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
4e2f84e6 | 1309 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
d899e052 | 1310 | while (1) { |
890871be | 1311 | write_lock(&em_tree->lock); |
d899e052 | 1312 | ret = add_extent_mapping(em_tree, em); |
890871be | 1313 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1314 | if (ret != -EEXIST) { |
1315 | free_extent_map(em); | |
1316 | break; | |
1317 | } | |
1318 | btrfs_drop_extent_cache(inode, em->start, | |
1319 | em->start + em->len - 1, 0); | |
1320 | } | |
1321 | type = BTRFS_ORDERED_PREALLOC; | |
1322 | } else { | |
1323 | type = BTRFS_ORDERED_NOCOW; | |
1324 | } | |
80ff3856 YZ |
1325 | |
1326 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1327 | num_bytes, num_bytes, type); |
79787eaa | 1328 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1329 | |
efa56464 YZ |
1330 | if (root->root_key.objectid == |
1331 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1332 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1333 | num_bytes); | |
79787eaa JM |
1334 | if (ret) { |
1335 | btrfs_abort_transaction(trans, root, ret); | |
1336 | goto error; | |
1337 | } | |
efa56464 YZ |
1338 | } |
1339 | ||
d899e052 | 1340 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1341 | cur_offset, cur_offset + num_bytes - 1, |
1342 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1343 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1344 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1345 | cur_offset = extent_end; |
1346 | if (cur_offset > end) | |
1347 | break; | |
be20aa9d | 1348 | } |
b3b4aa74 | 1349 | btrfs_release_path(path); |
80ff3856 | 1350 | |
17ca04af | 1351 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1352 | cow_start = cur_offset; |
17ca04af JB |
1353 | cur_offset = end; |
1354 | } | |
1355 | ||
80ff3856 YZ |
1356 | if (cow_start != (u64)-1) { |
1357 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1358 | page_started, nr_written, 1); |
79787eaa JM |
1359 | if (ret) { |
1360 | btrfs_abort_transaction(trans, root, ret); | |
1361 | goto error; | |
1362 | } | |
80ff3856 YZ |
1363 | } |
1364 | ||
79787eaa | 1365 | error: |
a698d075 | 1366 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1367 | if (!ret) |
1368 | ret = err; | |
1369 | ||
17ca04af JB |
1370 | if (ret && cur_offset < end) |
1371 | extent_clear_unlock_delalloc(inode, | |
1372 | &BTRFS_I(inode)->io_tree, | |
1373 | cur_offset, end, locked_page, | |
1374 | EXTENT_CLEAR_UNLOCK_PAGE | | |
1375 | EXTENT_CLEAR_UNLOCK | | |
1376 | EXTENT_CLEAR_DELALLOC | | |
1377 | EXTENT_CLEAR_DIRTY | | |
1378 | EXTENT_SET_WRITEBACK | | |
1379 | EXTENT_END_WRITEBACK); | |
1380 | ||
7ea394f1 | 1381 | btrfs_free_path(path); |
79787eaa | 1382 | return ret; |
be20aa9d CM |
1383 | } |
1384 | ||
d352ac68 CM |
1385 | /* |
1386 | * extent_io.c call back to do delayed allocation processing | |
1387 | */ | |
c8b97818 | 1388 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1389 | u64 start, u64 end, int *page_started, |
1390 | unsigned long *nr_written) | |
be20aa9d | 1391 | { |
be20aa9d | 1392 | int ret; |
7f366cfe | 1393 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1394 | |
7ddf5a42 | 1395 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) { |
c8b97818 | 1396 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1397 | page_started, 1, nr_written); |
7ddf5a42 | 1398 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) { |
d899e052 | 1399 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1400 | page_started, 0, nr_written); |
7ddf5a42 JB |
1401 | } else if (!btrfs_test_opt(root, COMPRESS) && |
1402 | !(BTRFS_I(inode)->force_compress) && | |
1403 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) { | |
7f366cfe CM |
1404 | ret = cow_file_range(inode, locked_page, start, end, |
1405 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1406 | } else { |
1407 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1408 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1409 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1410 | page_started, nr_written); |
7ddf5a42 | 1411 | } |
b888db2b CM |
1412 | return ret; |
1413 | } | |
1414 | ||
1bf85046 JM |
1415 | static void btrfs_split_extent_hook(struct inode *inode, |
1416 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1417 | { |
0ca1f7ce | 1418 | /* not delalloc, ignore it */ |
9ed74f2d | 1419 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1420 | return; |
9ed74f2d | 1421 | |
9e0baf60 JB |
1422 | spin_lock(&BTRFS_I(inode)->lock); |
1423 | BTRFS_I(inode)->outstanding_extents++; | |
1424 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1425 | } |
1426 | ||
1427 | /* | |
1428 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1429 | * extents so we can keep track of new extents that are just merged onto old | |
1430 | * extents, such as when we are doing sequential writes, so we can properly | |
1431 | * account for the metadata space we'll need. | |
1432 | */ | |
1bf85046 JM |
1433 | static void btrfs_merge_extent_hook(struct inode *inode, |
1434 | struct extent_state *new, | |
1435 | struct extent_state *other) | |
9ed74f2d | 1436 | { |
9ed74f2d JB |
1437 | /* not delalloc, ignore it */ |
1438 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1439 | return; |
9ed74f2d | 1440 | |
9e0baf60 JB |
1441 | spin_lock(&BTRFS_I(inode)->lock); |
1442 | BTRFS_I(inode)->outstanding_extents--; | |
1443 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1444 | } |
1445 | ||
d352ac68 CM |
1446 | /* |
1447 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1448 | * bytes in this file, and to maintain the list of inodes that | |
1449 | * have pending delalloc work to be done. | |
1450 | */ | |
1bf85046 JM |
1451 | static void btrfs_set_bit_hook(struct inode *inode, |
1452 | struct extent_state *state, int *bits) | |
291d673e | 1453 | { |
9ed74f2d | 1454 | |
75eff68e CM |
1455 | /* |
1456 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1457 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1458 | * bit, which is only set or cleared with irqs on |
1459 | */ | |
0ca1f7ce | 1460 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1461 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1462 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1463 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1464 | |
9e0baf60 | 1465 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1466 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1467 | } else { |
1468 | spin_lock(&BTRFS_I(inode)->lock); | |
1469 | BTRFS_I(inode)->outstanding_extents++; | |
1470 | spin_unlock(&BTRFS_I(inode)->lock); | |
1471 | } | |
287a0ab9 | 1472 | |
75eff68e | 1473 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1474 | BTRFS_I(inode)->delalloc_bytes += len; |
1475 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1476 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1477 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1478 | &root->fs_info->delalloc_inodes); | |
1479 | } | |
75eff68e | 1480 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1481 | } |
291d673e CM |
1482 | } |
1483 | ||
d352ac68 CM |
1484 | /* |
1485 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1486 | */ | |
1bf85046 JM |
1487 | static void btrfs_clear_bit_hook(struct inode *inode, |
1488 | struct extent_state *state, int *bits) | |
291d673e | 1489 | { |
75eff68e CM |
1490 | /* |
1491 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1492 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1493 | * bit, which is only set or cleared with irqs on |
1494 | */ | |
0ca1f7ce | 1495 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1496 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1497 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1498 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1499 | |
9e0baf60 | 1500 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1501 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1502 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1503 | spin_lock(&BTRFS_I(inode)->lock); | |
1504 | BTRFS_I(inode)->outstanding_extents--; | |
1505 | spin_unlock(&BTRFS_I(inode)->lock); | |
1506 | } | |
0ca1f7ce YZ |
1507 | |
1508 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1509 | btrfs_delalloc_release_metadata(inode, len); | |
1510 | ||
0cb59c99 JB |
1511 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1512 | && do_list) | |
0ca1f7ce | 1513 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1514 | |
75eff68e | 1515 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1516 | root->fs_info->delalloc_bytes -= len; |
1517 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1518 | ||
0cb59c99 | 1519 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1520 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1521 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1522 | } | |
75eff68e | 1523 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1524 | } |
291d673e CM |
1525 | } |
1526 | ||
d352ac68 CM |
1527 | /* |
1528 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1529 | * we don't create bios that span stripes or chunks | |
1530 | */ | |
239b14b3 | 1531 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1532 | size_t size, struct bio *bio, |
1533 | unsigned long bio_flags) | |
239b14b3 CM |
1534 | { |
1535 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1536 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1537 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1538 | u64 length = 0; |
1539 | u64 map_length; | |
239b14b3 CM |
1540 | int ret; |
1541 | ||
771ed689 CM |
1542 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1543 | return 0; | |
1544 | ||
f2d8d74d | 1545 | length = bio->bi_size; |
239b14b3 CM |
1546 | map_tree = &root->fs_info->mapping_tree; |
1547 | map_length = length; | |
cea9e445 | 1548 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1549 | &map_length, NULL, 0); |
3444a972 JM |
1550 | /* Will always return 0 or 1 with map_multi == NULL */ |
1551 | BUG_ON(ret < 0); | |
d397712b | 1552 | if (map_length < length + size) |
239b14b3 | 1553 | return 1; |
3444a972 | 1554 | return 0; |
239b14b3 CM |
1555 | } |
1556 | ||
d352ac68 CM |
1557 | /* |
1558 | * in order to insert checksums into the metadata in large chunks, | |
1559 | * we wait until bio submission time. All the pages in the bio are | |
1560 | * checksummed and sums are attached onto the ordered extent record. | |
1561 | * | |
1562 | * At IO completion time the cums attached on the ordered extent record | |
1563 | * are inserted into the btree | |
1564 | */ | |
d397712b CM |
1565 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1566 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1567 | unsigned long bio_flags, |
1568 | u64 bio_offset) | |
065631f6 | 1569 | { |
065631f6 | 1570 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1571 | int ret = 0; |
e015640f | 1572 | |
d20f7043 | 1573 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1574 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1575 | return 0; |
1576 | } | |
e015640f | 1577 | |
4a69a410 CM |
1578 | /* |
1579 | * in order to insert checksums into the metadata in large chunks, | |
1580 | * we wait until bio submission time. All the pages in the bio are | |
1581 | * checksummed and sums are attached onto the ordered extent record. | |
1582 | * | |
1583 | * At IO completion time the cums attached on the ordered extent record | |
1584 | * are inserted into the btree | |
1585 | */ | |
b2950863 | 1586 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1587 | int mirror_num, unsigned long bio_flags, |
1588 | u64 bio_offset) | |
4a69a410 CM |
1589 | { |
1590 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1591 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1592 | } |
1593 | ||
d352ac68 | 1594 | /* |
cad321ad CM |
1595 | * extent_io.c submission hook. This does the right thing for csum calculation |
1596 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1597 | */ |
b2950863 | 1598 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1599 | int mirror_num, unsigned long bio_flags, |
1600 | u64 bio_offset) | |
44b8bd7e CM |
1601 | { |
1602 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1603 | int ret = 0; | |
19b9bdb0 | 1604 | int skip_sum; |
0417341e | 1605 | int metadata = 0; |
44b8bd7e | 1606 | |
6cbff00f | 1607 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1608 | |
83eea1f1 | 1609 | if (btrfs_is_free_space_inode(inode)) |
0417341e JM |
1610 | metadata = 2; |
1611 | ||
7b6d91da | 1612 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1613 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1614 | if (ret) | |
1615 | return ret; | |
1616 | ||
d20f7043 | 1617 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1618 | return btrfs_submit_compressed_read(inode, bio, |
1619 | mirror_num, bio_flags); | |
c2db1073 TI |
1620 | } else if (!skip_sum) { |
1621 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1622 | if (ret) | |
1623 | return ret; | |
1624 | } | |
4d1b5fb4 | 1625 | goto mapit; |
19b9bdb0 | 1626 | } else if (!skip_sum) { |
17d217fe YZ |
1627 | /* csum items have already been cloned */ |
1628 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1629 | goto mapit; | |
19b9bdb0 CM |
1630 | /* we're doing a write, do the async checksumming */ |
1631 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1632 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1633 | bio_flags, bio_offset, |
1634 | __btrfs_submit_bio_start, | |
4a69a410 | 1635 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1636 | } |
1637 | ||
0b86a832 | 1638 | mapit: |
8b712842 | 1639 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1640 | } |
6885f308 | 1641 | |
d352ac68 CM |
1642 | /* |
1643 | * given a list of ordered sums record them in the inode. This happens | |
1644 | * at IO completion time based on sums calculated at bio submission time. | |
1645 | */ | |
ba1da2f4 | 1646 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1647 | struct inode *inode, u64 file_offset, |
1648 | struct list_head *list) | |
1649 | { | |
e6dcd2dc CM |
1650 | struct btrfs_ordered_sum *sum; |
1651 | ||
c6e30871 | 1652 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1653 | btrfs_csum_file_blocks(trans, |
1654 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1655 | } |
1656 | return 0; | |
1657 | } | |
1658 | ||
2ac55d41 JB |
1659 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1660 | struct extent_state **cached_state) | |
ea8c2819 | 1661 | { |
d397712b | 1662 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1663 | WARN_ON(1); |
ea8c2819 | 1664 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1665 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1666 | } |
1667 | ||
d352ac68 | 1668 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1669 | struct btrfs_writepage_fixup { |
1670 | struct page *page; | |
1671 | struct btrfs_work work; | |
1672 | }; | |
1673 | ||
b2950863 | 1674 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1675 | { |
1676 | struct btrfs_writepage_fixup *fixup; | |
1677 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1678 | struct extent_state *cached_state = NULL; |
247e743c CM |
1679 | struct page *page; |
1680 | struct inode *inode; | |
1681 | u64 page_start; | |
1682 | u64 page_end; | |
87826df0 | 1683 | int ret; |
247e743c CM |
1684 | |
1685 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1686 | page = fixup->page; | |
4a096752 | 1687 | again: |
247e743c CM |
1688 | lock_page(page); |
1689 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1690 | ClearPageChecked(page); | |
1691 | goto out_page; | |
1692 | } | |
1693 | ||
1694 | inode = page->mapping->host; | |
1695 | page_start = page_offset(page); | |
1696 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1697 | ||
2ac55d41 | 1698 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1699 | &cached_state); |
4a096752 CM |
1700 | |
1701 | /* already ordered? We're done */ | |
8b62b72b | 1702 | if (PagePrivate2(page)) |
247e743c | 1703 | goto out; |
4a096752 CM |
1704 | |
1705 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1706 | if (ordered) { | |
2ac55d41 JB |
1707 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1708 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1709 | unlock_page(page); |
1710 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1711 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1712 | goto again; |
1713 | } | |
247e743c | 1714 | |
87826df0 JM |
1715 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1716 | if (ret) { | |
1717 | mapping_set_error(page->mapping, ret); | |
1718 | end_extent_writepage(page, ret, page_start, page_end); | |
1719 | ClearPageChecked(page); | |
1720 | goto out; | |
1721 | } | |
1722 | ||
2ac55d41 | 1723 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1724 | ClearPageChecked(page); |
87826df0 | 1725 | set_page_dirty(page); |
247e743c | 1726 | out: |
2ac55d41 JB |
1727 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1728 | &cached_state, GFP_NOFS); | |
247e743c CM |
1729 | out_page: |
1730 | unlock_page(page); | |
1731 | page_cache_release(page); | |
b897abec | 1732 | kfree(fixup); |
247e743c CM |
1733 | } |
1734 | ||
1735 | /* | |
1736 | * There are a few paths in the higher layers of the kernel that directly | |
1737 | * set the page dirty bit without asking the filesystem if it is a | |
1738 | * good idea. This causes problems because we want to make sure COW | |
1739 | * properly happens and the data=ordered rules are followed. | |
1740 | * | |
c8b97818 | 1741 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1742 | * hasn't been properly setup for IO. We kick off an async process |
1743 | * to fix it up. The async helper will wait for ordered extents, set | |
1744 | * the delalloc bit and make it safe to write the page. | |
1745 | */ | |
b2950863 | 1746 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1747 | { |
1748 | struct inode *inode = page->mapping->host; | |
1749 | struct btrfs_writepage_fixup *fixup; | |
1750 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1751 | |
8b62b72b CM |
1752 | /* this page is properly in the ordered list */ |
1753 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1754 | return 0; |
1755 | ||
1756 | if (PageChecked(page)) | |
1757 | return -EAGAIN; | |
1758 | ||
1759 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1760 | if (!fixup) | |
1761 | return -EAGAIN; | |
f421950f | 1762 | |
247e743c CM |
1763 | SetPageChecked(page); |
1764 | page_cache_get(page); | |
1765 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1766 | fixup->page = page; | |
1767 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1768 | return -EBUSY; |
247e743c CM |
1769 | } |
1770 | ||
d899e052 YZ |
1771 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1772 | struct inode *inode, u64 file_pos, | |
1773 | u64 disk_bytenr, u64 disk_num_bytes, | |
1774 | u64 num_bytes, u64 ram_bytes, | |
1775 | u8 compression, u8 encryption, | |
1776 | u16 other_encoding, int extent_type) | |
1777 | { | |
1778 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1779 | struct btrfs_file_extent_item *fi; | |
1780 | struct btrfs_path *path; | |
1781 | struct extent_buffer *leaf; | |
1782 | struct btrfs_key ins; | |
d899e052 YZ |
1783 | int ret; |
1784 | ||
1785 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1786 | if (!path) |
1787 | return -ENOMEM; | |
d899e052 | 1788 | |
b9473439 | 1789 | path->leave_spinning = 1; |
a1ed835e CM |
1790 | |
1791 | /* | |
1792 | * we may be replacing one extent in the tree with another. | |
1793 | * The new extent is pinned in the extent map, and we don't want | |
1794 | * to drop it from the cache until it is completely in the btree. | |
1795 | * | |
1796 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1797 | * the caller is expected to unpin it and allow it to be merged | |
1798 | * with the others. | |
1799 | */ | |
5dc562c5 | 1800 | ret = btrfs_drop_extents(trans, root, inode, file_pos, |
2671485d | 1801 | file_pos + num_bytes, 0); |
79787eaa JM |
1802 | if (ret) |
1803 | goto out; | |
d899e052 | 1804 | |
33345d01 | 1805 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1806 | ins.offset = file_pos; |
1807 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1808 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
79787eaa JM |
1809 | if (ret) |
1810 | goto out; | |
d899e052 YZ |
1811 | leaf = path->nodes[0]; |
1812 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1813 | struct btrfs_file_extent_item); | |
1814 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1815 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1816 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1817 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1818 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1819 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1820 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1821 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1822 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1823 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1824 | |
1825 | btrfs_unlock_up_safe(path, 1); | |
1826 | btrfs_set_lock_blocking(leaf); | |
1827 | ||
d899e052 YZ |
1828 | btrfs_mark_buffer_dirty(leaf); |
1829 | ||
1830 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1831 | |
1832 | ins.objectid = disk_bytenr; | |
1833 | ins.offset = disk_num_bytes; | |
1834 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1835 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1836 | root->root_key.objectid, | |
33345d01 | 1837 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 1838 | out: |
d899e052 | 1839 | btrfs_free_path(path); |
b9473439 | 1840 | |
79787eaa | 1841 | return ret; |
d899e052 YZ |
1842 | } |
1843 | ||
5d13a98f CM |
1844 | /* |
1845 | * helper function for btrfs_finish_ordered_io, this | |
1846 | * just reads in some of the csum leaves to prime them into ram | |
1847 | * before we start the transaction. It limits the amount of btree | |
1848 | * reads required while inside the transaction. | |
1849 | */ | |
d352ac68 CM |
1850 | /* as ordered data IO finishes, this gets called so we can finish |
1851 | * an ordered extent if the range of bytes in the file it covers are | |
1852 | * fully written. | |
1853 | */ | |
5fd02043 | 1854 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 1855 | { |
5fd02043 | 1856 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 1857 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1858 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 1859 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1860 | struct extent_state *cached_state = NULL; |
261507a0 | 1861 | int compress_type = 0; |
e6dcd2dc | 1862 | int ret; |
82d5902d | 1863 | bool nolock; |
e6dcd2dc | 1864 | |
83eea1f1 | 1865 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 1866 | |
5fd02043 JB |
1867 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
1868 | ret = -EIO; | |
1869 | goto out; | |
1870 | } | |
1871 | ||
c2167754 | 1872 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 1873 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
c2167754 YZ |
1874 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
1875 | if (!ret) { | |
0cb59c99 | 1876 | if (nolock) |
7a7eaa40 | 1877 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1878 | else |
7a7eaa40 | 1879 | trans = btrfs_join_transaction(root); |
d280e5be LB |
1880 | if (IS_ERR(trans)) { |
1881 | ret = PTR_ERR(trans); | |
1882 | trans = NULL; | |
1883 | goto out; | |
1884 | } | |
0ca1f7ce | 1885 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1886 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1887 | if (ret) /* -ENOMEM or corruption */ |
1888 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
1889 | } |
1890 | goto out; | |
1891 | } | |
e6dcd2dc | 1892 | |
2ac55d41 JB |
1893 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1894 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 1895 | 0, &cached_state); |
e6dcd2dc | 1896 | |
0cb59c99 | 1897 | if (nolock) |
7a7eaa40 | 1898 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1899 | else |
7a7eaa40 | 1900 | trans = btrfs_join_transaction(root); |
79787eaa JM |
1901 | if (IS_ERR(trans)) { |
1902 | ret = PTR_ERR(trans); | |
1903 | trans = NULL; | |
1904 | goto out_unlock; | |
1905 | } | |
0ca1f7ce | 1906 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1907 | |
c8b97818 | 1908 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1909 | compress_type = ordered_extent->compress_type; |
d899e052 | 1910 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1911 | BUG_ON(compress_type); |
920bbbfb | 1912 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1913 | ordered_extent->file_offset, |
1914 | ordered_extent->file_offset + | |
1915 | ordered_extent->len); | |
d899e052 | 1916 | } else { |
0af3d00b | 1917 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1918 | ret = insert_reserved_file_extent(trans, inode, |
1919 | ordered_extent->file_offset, | |
1920 | ordered_extent->start, | |
1921 | ordered_extent->disk_len, | |
1922 | ordered_extent->len, | |
1923 | ordered_extent->len, | |
261507a0 | 1924 | compress_type, 0, 0, |
d899e052 | 1925 | BTRFS_FILE_EXTENT_REG); |
d899e052 | 1926 | } |
5dc562c5 JB |
1927 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1928 | ordered_extent->file_offset, ordered_extent->len, | |
1929 | trans->transid); | |
79787eaa JM |
1930 | if (ret < 0) { |
1931 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1932 | goto out_unlock; |
79787eaa | 1933 | } |
2ac55d41 | 1934 | |
e6dcd2dc CM |
1935 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1936 | &ordered_extent->list); | |
1937 | ||
1ef30be1 | 1938 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1939 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1940 | ret = btrfs_update_inode_fallback(trans, root, inode); |
79787eaa JM |
1941 | if (ret) { /* -ENOMEM or corruption */ |
1942 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 1943 | goto out_unlock; |
79787eaa | 1944 | } |
7c735313 JB |
1945 | } else { |
1946 | btrfs_set_inode_last_trans(trans, inode); | |
1ef30be1 JB |
1947 | } |
1948 | ret = 0; | |
5fd02043 JB |
1949 | out_unlock: |
1950 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
1951 | ordered_extent->file_offset + | |
1952 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 1953 | out: |
5b0e95bf | 1954 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1955 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
1956 | if (trans) |
1957 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 1958 | |
5fd02043 JB |
1959 | if (ret) |
1960 | clear_extent_uptodate(io_tree, ordered_extent->file_offset, | |
1961 | ordered_extent->file_offset + | |
1962 | ordered_extent->len - 1, NULL, GFP_NOFS); | |
1963 | ||
1964 | /* | |
1965 | * This needs to be dont to make sure anybody waiting knows we are done | |
1966 | * upating everything for this ordered extent. | |
1967 | */ | |
1968 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
1969 | ||
e6dcd2dc CM |
1970 | /* once for us */ |
1971 | btrfs_put_ordered_extent(ordered_extent); | |
1972 | /* once for the tree */ | |
1973 | btrfs_put_ordered_extent(ordered_extent); | |
1974 | ||
5fd02043 JB |
1975 | return ret; |
1976 | } | |
1977 | ||
1978 | static void finish_ordered_fn(struct btrfs_work *work) | |
1979 | { | |
1980 | struct btrfs_ordered_extent *ordered_extent; | |
1981 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
1982 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
1983 | } |
1984 | ||
b2950863 | 1985 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1986 | struct extent_state *state, int uptodate) |
1987 | { | |
5fd02043 JB |
1988 | struct inode *inode = page->mapping->host; |
1989 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1990 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
1991 | struct btrfs_workers *workers; | |
1992 | ||
1abe9b8a | 1993 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1994 | ||
8b62b72b | 1995 | ClearPagePrivate2(page); |
5fd02043 JB |
1996 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1997 | end - start + 1, uptodate)) | |
1998 | return 0; | |
1999 | ||
2000 | ordered_extent->work.func = finish_ordered_fn; | |
2001 | ordered_extent->work.flags = 0; | |
2002 | ||
83eea1f1 | 2003 | if (btrfs_is_free_space_inode(inode)) |
5fd02043 JB |
2004 | workers = &root->fs_info->endio_freespace_worker; |
2005 | else | |
2006 | workers = &root->fs_info->endio_write_workers; | |
2007 | btrfs_queue_worker(workers, &ordered_extent->work); | |
2008 | ||
2009 | return 0; | |
211f90e6 CM |
2010 | } |
2011 | ||
d352ac68 CM |
2012 | /* |
2013 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
2014 | * if there's a match, we allow the bio to finish. If not, the code in |
2015 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 2016 | */ |
b2950863 | 2017 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
5cf1ab56 | 2018 | struct extent_state *state, int mirror) |
07157aac | 2019 | { |
35ebb934 | 2020 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 2021 | struct inode *inode = page->mapping->host; |
d1310b2e | 2022 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 2023 | char *kaddr; |
aadfeb6e | 2024 | u64 private = ~(u32)0; |
07157aac | 2025 | int ret; |
ff79f819 CM |
2026 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2027 | u32 csum = ~(u32)0; | |
d1310b2e | 2028 | |
d20f7043 CM |
2029 | if (PageChecked(page)) { |
2030 | ClearPageChecked(page); | |
2031 | goto good; | |
2032 | } | |
6cbff00f CH |
2033 | |
2034 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 2035 | goto good; |
17d217fe YZ |
2036 | |
2037 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 2038 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
2039 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
2040 | GFP_NOFS); | |
b6cda9bc | 2041 | return 0; |
17d217fe | 2042 | } |
d20f7043 | 2043 | |
c2e639f0 | 2044 | if (state && state->start == start) { |
70dec807 CM |
2045 | private = state->private; |
2046 | ret = 0; | |
2047 | } else { | |
2048 | ret = get_state_private(io_tree, start, &private); | |
2049 | } | |
7ac687d9 | 2050 | kaddr = kmap_atomic(page); |
d397712b | 2051 | if (ret) |
07157aac | 2052 | goto zeroit; |
d397712b | 2053 | |
ff79f819 CM |
2054 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
2055 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 2056 | if (csum != private) |
07157aac | 2057 | goto zeroit; |
d397712b | 2058 | |
7ac687d9 | 2059 | kunmap_atomic(kaddr); |
d20f7043 | 2060 | good: |
07157aac CM |
2061 | return 0; |
2062 | ||
2063 | zeroit: | |
945d8962 | 2064 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
2065 | "private %llu\n", |
2066 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
2067 | (unsigned long long)start, csum, |
2068 | (unsigned long long)private); | |
db94535d CM |
2069 | memset(kaddr + offset, 1, end - start + 1); |
2070 | flush_dcache_page(page); | |
7ac687d9 | 2071 | kunmap_atomic(kaddr); |
3b951516 CM |
2072 | if (private == 0) |
2073 | return 0; | |
7e38326f | 2074 | return -EIO; |
07157aac | 2075 | } |
b888db2b | 2076 | |
24bbcf04 YZ |
2077 | struct delayed_iput { |
2078 | struct list_head list; | |
2079 | struct inode *inode; | |
2080 | }; | |
2081 | ||
79787eaa JM |
2082 | /* JDM: If this is fs-wide, why can't we add a pointer to |
2083 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
2084 | void btrfs_add_delayed_iput(struct inode *inode) |
2085 | { | |
2086 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
2087 | struct delayed_iput *delayed; | |
2088 | ||
2089 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
2090 | return; | |
2091 | ||
2092 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
2093 | delayed->inode = inode; | |
2094 | ||
2095 | spin_lock(&fs_info->delayed_iput_lock); | |
2096 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
2097 | spin_unlock(&fs_info->delayed_iput_lock); | |
2098 | } | |
2099 | ||
2100 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
2101 | { | |
2102 | LIST_HEAD(list); | |
2103 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2104 | struct delayed_iput *delayed; | |
2105 | int empty; | |
2106 | ||
2107 | spin_lock(&fs_info->delayed_iput_lock); | |
2108 | empty = list_empty(&fs_info->delayed_iputs); | |
2109 | spin_unlock(&fs_info->delayed_iput_lock); | |
2110 | if (empty) | |
2111 | return; | |
2112 | ||
24bbcf04 YZ |
2113 | spin_lock(&fs_info->delayed_iput_lock); |
2114 | list_splice_init(&fs_info->delayed_iputs, &list); | |
2115 | spin_unlock(&fs_info->delayed_iput_lock); | |
2116 | ||
2117 | while (!list_empty(&list)) { | |
2118 | delayed = list_entry(list.next, struct delayed_iput, list); | |
2119 | list_del(&delayed->list); | |
2120 | iput(delayed->inode); | |
2121 | kfree(delayed); | |
2122 | } | |
24bbcf04 YZ |
2123 | } |
2124 | ||
d68fc57b YZ |
2125 | enum btrfs_orphan_cleanup_state { |
2126 | ORPHAN_CLEANUP_STARTED = 1, | |
2127 | ORPHAN_CLEANUP_DONE = 2, | |
2128 | }; | |
2129 | ||
2130 | /* | |
42b2aa86 | 2131 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
2132 | * files in the subvolume, it removes orphan item and frees block_rsv |
2133 | * structure. | |
2134 | */ | |
2135 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
2136 | struct btrfs_root *root) | |
2137 | { | |
90290e19 | 2138 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
2139 | int ret; |
2140 | ||
8a35d95f | 2141 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
2142 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
2143 | return; | |
2144 | ||
90290e19 | 2145 | spin_lock(&root->orphan_lock); |
8a35d95f | 2146 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
2147 | spin_unlock(&root->orphan_lock); |
2148 | return; | |
2149 | } | |
2150 | ||
2151 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
2152 | spin_unlock(&root->orphan_lock); | |
2153 | return; | |
2154 | } | |
2155 | ||
2156 | block_rsv = root->orphan_block_rsv; | |
2157 | root->orphan_block_rsv = NULL; | |
2158 | spin_unlock(&root->orphan_lock); | |
2159 | ||
d68fc57b YZ |
2160 | if (root->orphan_item_inserted && |
2161 | btrfs_root_refs(&root->root_item) > 0) { | |
2162 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
2163 | root->root_key.objectid); | |
2164 | BUG_ON(ret); | |
2165 | root->orphan_item_inserted = 0; | |
2166 | } | |
2167 | ||
90290e19 JB |
2168 | if (block_rsv) { |
2169 | WARN_ON(block_rsv->size > 0); | |
2170 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
2171 | } |
2172 | } | |
2173 | ||
7b128766 JB |
2174 | /* |
2175 | * This creates an orphan entry for the given inode in case something goes | |
2176 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2177 | * |
2178 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2179 | * this function. | |
7b128766 JB |
2180 | */ |
2181 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2182 | { | |
2183 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2184 | struct btrfs_block_rsv *block_rsv = NULL; |
2185 | int reserve = 0; | |
2186 | int insert = 0; | |
2187 | int ret; | |
7b128766 | 2188 | |
d68fc57b | 2189 | if (!root->orphan_block_rsv) { |
66d8f3dd | 2190 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
2191 | if (!block_rsv) |
2192 | return -ENOMEM; | |
d68fc57b | 2193 | } |
7b128766 | 2194 | |
d68fc57b YZ |
2195 | spin_lock(&root->orphan_lock); |
2196 | if (!root->orphan_block_rsv) { | |
2197 | root->orphan_block_rsv = block_rsv; | |
2198 | } else if (block_rsv) { | |
2199 | btrfs_free_block_rsv(root, block_rsv); | |
2200 | block_rsv = NULL; | |
7b128766 | 2201 | } |
7b128766 | 2202 | |
8a35d95f JB |
2203 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2204 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
2205 | #if 0 |
2206 | /* | |
2207 | * For proper ENOSPC handling, we should do orphan | |
2208 | * cleanup when mounting. But this introduces backward | |
2209 | * compatibility issue. | |
2210 | */ | |
2211 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2212 | insert = 2; | |
2213 | else | |
2214 | insert = 1; | |
2215 | #endif | |
2216 | insert = 1; | |
321f0e70 | 2217 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
2218 | } |
2219 | ||
72ac3c0d JB |
2220 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2221 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2222 | reserve = 1; |
d68fc57b | 2223 | spin_unlock(&root->orphan_lock); |
7b128766 | 2224 | |
d68fc57b YZ |
2225 | /* grab metadata reservation from transaction handle */ |
2226 | if (reserve) { | |
2227 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 2228 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 2229 | } |
7b128766 | 2230 | |
d68fc57b YZ |
2231 | /* insert an orphan item to track this unlinked/truncated file */ |
2232 | if (insert >= 1) { | |
33345d01 | 2233 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2234 | if (ret && ret != -EEXIST) { |
8a35d95f JB |
2235 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2236 | &BTRFS_I(inode)->runtime_flags); | |
79787eaa JM |
2237 | btrfs_abort_transaction(trans, root, ret); |
2238 | return ret; | |
2239 | } | |
2240 | ret = 0; | |
d68fc57b YZ |
2241 | } |
2242 | ||
2243 | /* insert an orphan item to track subvolume contains orphan files */ | |
2244 | if (insert >= 2) { | |
2245 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2246 | root->root_key.objectid); | |
79787eaa JM |
2247 | if (ret && ret != -EEXIST) { |
2248 | btrfs_abort_transaction(trans, root, ret); | |
2249 | return ret; | |
2250 | } | |
d68fc57b YZ |
2251 | } |
2252 | return 0; | |
7b128766 JB |
2253 | } |
2254 | ||
2255 | /* | |
2256 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2257 | * item for this particular inode. | |
2258 | */ | |
2259 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2260 | { | |
2261 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2262 | int delete_item = 0; |
2263 | int release_rsv = 0; | |
7b128766 JB |
2264 | int ret = 0; |
2265 | ||
d68fc57b | 2266 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
2267 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2268 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2269 | delete_item = 1; |
7b128766 | 2270 | |
72ac3c0d JB |
2271 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
2272 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 2273 | release_rsv = 1; |
d68fc57b | 2274 | spin_unlock(&root->orphan_lock); |
7b128766 | 2275 | |
d68fc57b | 2276 | if (trans && delete_item) { |
33345d01 | 2277 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
79787eaa | 2278 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
d68fc57b | 2279 | } |
7b128766 | 2280 | |
8a35d95f | 2281 | if (release_rsv) { |
d68fc57b | 2282 | btrfs_orphan_release_metadata(inode); |
8a35d95f JB |
2283 | atomic_dec(&root->orphan_inodes); |
2284 | } | |
7b128766 | 2285 | |
d68fc57b | 2286 | return 0; |
7b128766 JB |
2287 | } |
2288 | ||
2289 | /* | |
2290 | * this cleans up any orphans that may be left on the list from the last use | |
2291 | * of this root. | |
2292 | */ | |
66b4ffd1 | 2293 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2294 | { |
2295 | struct btrfs_path *path; | |
2296 | struct extent_buffer *leaf; | |
7b128766 JB |
2297 | struct btrfs_key key, found_key; |
2298 | struct btrfs_trans_handle *trans; | |
2299 | struct inode *inode; | |
8f6d7f4f | 2300 | u64 last_objectid = 0; |
7b128766 JB |
2301 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2302 | ||
d68fc57b | 2303 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2304 | return 0; |
c71bf099 YZ |
2305 | |
2306 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2307 | if (!path) { |
2308 | ret = -ENOMEM; | |
2309 | goto out; | |
2310 | } | |
7b128766 JB |
2311 | path->reada = -1; |
2312 | ||
2313 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2314 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2315 | key.offset = (u64)-1; | |
2316 | ||
7b128766 JB |
2317 | while (1) { |
2318 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2319 | if (ret < 0) |
2320 | goto out; | |
7b128766 JB |
2321 | |
2322 | /* | |
2323 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2324 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2325 | * find the key and see if we have stuff that matches |
2326 | */ | |
2327 | if (ret > 0) { | |
66b4ffd1 | 2328 | ret = 0; |
7b128766 JB |
2329 | if (path->slots[0] == 0) |
2330 | break; | |
2331 | path->slots[0]--; | |
2332 | } | |
2333 | ||
2334 | /* pull out the item */ | |
2335 | leaf = path->nodes[0]; | |
7b128766 JB |
2336 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2337 | ||
2338 | /* make sure the item matches what we want */ | |
2339 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2340 | break; | |
2341 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2342 | break; | |
2343 | ||
2344 | /* release the path since we're done with it */ | |
b3b4aa74 | 2345 | btrfs_release_path(path); |
7b128766 JB |
2346 | |
2347 | /* | |
2348 | * this is where we are basically btrfs_lookup, without the | |
2349 | * crossing root thing. we store the inode number in the | |
2350 | * offset of the orphan item. | |
2351 | */ | |
8f6d7f4f JB |
2352 | |
2353 | if (found_key.offset == last_objectid) { | |
2354 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2355 | "stopping orphan cleanup\n"); | |
2356 | ret = -EINVAL; | |
2357 | goto out; | |
2358 | } | |
2359 | ||
2360 | last_objectid = found_key.offset; | |
2361 | ||
5d4f98a2 YZ |
2362 | found_key.objectid = found_key.offset; |
2363 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2364 | found_key.offset = 0; | |
73f73415 | 2365 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2366 | ret = PTR_RET(inode); |
2367 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2368 | goto out; |
7b128766 | 2369 | |
f8e9e0b0 AJ |
2370 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2371 | struct btrfs_root *dead_root; | |
2372 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2373 | int is_dead_root = 0; | |
2374 | ||
2375 | /* | |
2376 | * this is an orphan in the tree root. Currently these | |
2377 | * could come from 2 sources: | |
2378 | * a) a snapshot deletion in progress | |
2379 | * b) a free space cache inode | |
2380 | * We need to distinguish those two, as the snapshot | |
2381 | * orphan must not get deleted. | |
2382 | * find_dead_roots already ran before us, so if this | |
2383 | * is a snapshot deletion, we should find the root | |
2384 | * in the dead_roots list | |
2385 | */ | |
2386 | spin_lock(&fs_info->trans_lock); | |
2387 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2388 | root_list) { | |
2389 | if (dead_root->root_key.objectid == | |
2390 | found_key.objectid) { | |
2391 | is_dead_root = 1; | |
2392 | break; | |
2393 | } | |
2394 | } | |
2395 | spin_unlock(&fs_info->trans_lock); | |
2396 | if (is_dead_root) { | |
2397 | /* prevent this orphan from being found again */ | |
2398 | key.offset = found_key.objectid - 1; | |
2399 | continue; | |
2400 | } | |
2401 | } | |
7b128766 | 2402 | /* |
a8c9e576 JB |
2403 | * Inode is already gone but the orphan item is still there, |
2404 | * kill the orphan item. | |
7b128766 | 2405 | */ |
a8c9e576 JB |
2406 | if (ret == -ESTALE) { |
2407 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2408 | if (IS_ERR(trans)) { |
2409 | ret = PTR_ERR(trans); | |
2410 | goto out; | |
2411 | } | |
8a35d95f JB |
2412 | printk(KERN_ERR "auto deleting %Lu\n", |
2413 | found_key.objectid); | |
a8c9e576 JB |
2414 | ret = btrfs_del_orphan_item(trans, root, |
2415 | found_key.objectid); | |
79787eaa | 2416 | BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */ |
5b21f2ed | 2417 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2418 | continue; |
2419 | } | |
2420 | ||
a8c9e576 JB |
2421 | /* |
2422 | * add this inode to the orphan list so btrfs_orphan_del does | |
2423 | * the proper thing when we hit it | |
2424 | */ | |
8a35d95f JB |
2425 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
2426 | &BTRFS_I(inode)->runtime_flags); | |
a8c9e576 | 2427 | |
7b128766 JB |
2428 | /* if we have links, this was a truncate, lets do that */ |
2429 | if (inode->i_nlink) { | |
a41ad394 JB |
2430 | if (!S_ISREG(inode->i_mode)) { |
2431 | WARN_ON(1); | |
2432 | iput(inode); | |
2433 | continue; | |
2434 | } | |
7b128766 | 2435 | nr_truncate++; |
66b4ffd1 | 2436 | ret = btrfs_truncate(inode); |
7b128766 JB |
2437 | } else { |
2438 | nr_unlink++; | |
2439 | } | |
2440 | ||
2441 | /* this will do delete_inode and everything for us */ | |
2442 | iput(inode); | |
66b4ffd1 JB |
2443 | if (ret) |
2444 | goto out; | |
7b128766 | 2445 | } |
3254c876 MX |
2446 | /* release the path since we're done with it */ |
2447 | btrfs_release_path(path); | |
2448 | ||
d68fc57b YZ |
2449 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2450 | ||
2451 | if (root->orphan_block_rsv) | |
2452 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2453 | (u64)-1); | |
2454 | ||
2455 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2456 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2457 | if (!IS_ERR(trans)) |
2458 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2459 | } |
7b128766 JB |
2460 | |
2461 | if (nr_unlink) | |
2462 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2463 | if (nr_truncate) | |
2464 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2465 | |
2466 | out: | |
2467 | if (ret) | |
2468 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2469 | btrfs_free_path(path); | |
2470 | return ret; | |
7b128766 JB |
2471 | } |
2472 | ||
46a53cca CM |
2473 | /* |
2474 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2475 | * don't find any xattrs, we know there can't be any acls. | |
2476 | * | |
2477 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2478 | */ | |
2479 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2480 | int slot, u64 objectid) | |
2481 | { | |
2482 | u32 nritems = btrfs_header_nritems(leaf); | |
2483 | struct btrfs_key found_key; | |
2484 | int scanned = 0; | |
2485 | ||
2486 | slot++; | |
2487 | while (slot < nritems) { | |
2488 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2489 | ||
2490 | /* we found a different objectid, there must not be acls */ | |
2491 | if (found_key.objectid != objectid) | |
2492 | return 0; | |
2493 | ||
2494 | /* we found an xattr, assume we've got an acl */ | |
2495 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2496 | return 1; | |
2497 | ||
2498 | /* | |
2499 | * we found a key greater than an xattr key, there can't | |
2500 | * be any acls later on | |
2501 | */ | |
2502 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2503 | return 0; | |
2504 | ||
2505 | slot++; | |
2506 | scanned++; | |
2507 | ||
2508 | /* | |
2509 | * it goes inode, inode backrefs, xattrs, extents, | |
2510 | * so if there are a ton of hard links to an inode there can | |
2511 | * be a lot of backrefs. Don't waste time searching too hard, | |
2512 | * this is just an optimization | |
2513 | */ | |
2514 | if (scanned >= 8) | |
2515 | break; | |
2516 | } | |
2517 | /* we hit the end of the leaf before we found an xattr or | |
2518 | * something larger than an xattr. We have to assume the inode | |
2519 | * has acls | |
2520 | */ | |
2521 | return 1; | |
2522 | } | |
2523 | ||
d352ac68 CM |
2524 | /* |
2525 | * read an inode from the btree into the in-memory inode | |
2526 | */ | |
5d4f98a2 | 2527 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2528 | { |
2529 | struct btrfs_path *path; | |
5f39d397 | 2530 | struct extent_buffer *leaf; |
39279cc3 | 2531 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2532 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2533 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2534 | struct btrfs_key location; | |
46a53cca | 2535 | int maybe_acls; |
618e21d5 | 2536 | u32 rdev; |
39279cc3 | 2537 | int ret; |
2f7e33d4 MX |
2538 | bool filled = false; |
2539 | ||
2540 | ret = btrfs_fill_inode(inode, &rdev); | |
2541 | if (!ret) | |
2542 | filled = true; | |
39279cc3 CM |
2543 | |
2544 | path = btrfs_alloc_path(); | |
1748f843 MF |
2545 | if (!path) |
2546 | goto make_bad; | |
2547 | ||
d90c7321 | 2548 | path->leave_spinning = 1; |
39279cc3 | 2549 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2550 | |
39279cc3 | 2551 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2552 | if (ret) |
39279cc3 | 2553 | goto make_bad; |
39279cc3 | 2554 | |
5f39d397 | 2555 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2556 | |
2557 | if (filled) | |
2558 | goto cache_acl; | |
2559 | ||
5f39d397 CM |
2560 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2561 | struct btrfs_inode_item); | |
5f39d397 | 2562 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2563 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
5f39d397 CM |
2564 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); |
2565 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2566 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2567 | |
2568 | tspec = btrfs_inode_atime(inode_item); | |
2569 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2570 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2571 | ||
2572 | tspec = btrfs_inode_mtime(inode_item); | |
2573 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2574 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2575 | ||
2576 | tspec = btrfs_inode_ctime(inode_item); | |
2577 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2578 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2579 | ||
a76a3cd4 | 2580 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2581 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
2582 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
2583 | ||
2584 | /* | |
2585 | * If we were modified in the current generation and evicted from memory | |
2586 | * and then re-read we need to do a full sync since we don't have any | |
2587 | * idea about which extents were modified before we were evicted from | |
2588 | * cache. | |
2589 | */ | |
2590 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
2591 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
2592 | &BTRFS_I(inode)->runtime_flags); | |
2593 | ||
0c4d2d95 | 2594 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2595 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2596 | inode->i_rdev = 0; |
5f39d397 CM |
2597 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2598 | ||
aec7477b | 2599 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2600 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2601 | cache_acl: |
46a53cca CM |
2602 | /* |
2603 | * try to precache a NULL acl entry for files that don't have | |
2604 | * any xattrs or acls | |
2605 | */ | |
33345d01 LZ |
2606 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2607 | btrfs_ino(inode)); | |
72c04902 AV |
2608 | if (!maybe_acls) |
2609 | cache_no_acl(inode); | |
46a53cca | 2610 | |
39279cc3 | 2611 | btrfs_free_path(path); |
39279cc3 | 2612 | |
39279cc3 | 2613 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2614 | case S_IFREG: |
2615 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2616 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2617 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2618 | inode->i_fop = &btrfs_file_operations; |
2619 | inode->i_op = &btrfs_file_inode_operations; | |
2620 | break; | |
2621 | case S_IFDIR: | |
2622 | inode->i_fop = &btrfs_dir_file_operations; | |
2623 | if (root == root->fs_info->tree_root) | |
2624 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2625 | else | |
2626 | inode->i_op = &btrfs_dir_inode_operations; | |
2627 | break; | |
2628 | case S_IFLNK: | |
2629 | inode->i_op = &btrfs_symlink_inode_operations; | |
2630 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2631 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2632 | break; |
618e21d5 | 2633 | default: |
0279b4cd | 2634 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2635 | init_special_inode(inode, inode->i_mode, rdev); |
2636 | break; | |
39279cc3 | 2637 | } |
6cbff00f CH |
2638 | |
2639 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2640 | return; |
2641 | ||
2642 | make_bad: | |
39279cc3 | 2643 | btrfs_free_path(path); |
39279cc3 CM |
2644 | make_bad_inode(inode); |
2645 | } | |
2646 | ||
d352ac68 CM |
2647 | /* |
2648 | * given a leaf and an inode, copy the inode fields into the leaf | |
2649 | */ | |
e02119d5 CM |
2650 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2651 | struct extent_buffer *leaf, | |
5f39d397 | 2652 | struct btrfs_inode_item *item, |
39279cc3 CM |
2653 | struct inode *inode) |
2654 | { | |
5f39d397 CM |
2655 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2656 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2657 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2658 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2659 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2660 | ||
2661 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2662 | inode->i_atime.tv_sec); | |
2663 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2664 | inode->i_atime.tv_nsec); | |
2665 | ||
2666 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2667 | inode->i_mtime.tv_sec); | |
2668 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2669 | inode->i_mtime.tv_nsec); | |
2670 | ||
2671 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2672 | inode->i_ctime.tv_sec); | |
2673 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2674 | inode->i_ctime.tv_nsec); | |
2675 | ||
a76a3cd4 | 2676 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2677 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
0c4d2d95 | 2678 | btrfs_set_inode_sequence(leaf, item, inode->i_version); |
e02119d5 | 2679 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2680 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2681 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2682 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2683 | } |
2684 | ||
d352ac68 CM |
2685 | /* |
2686 | * copy everything in the in-memory inode into the btree. | |
2687 | */ | |
2115133f | 2688 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2689 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2690 | { |
2691 | struct btrfs_inode_item *inode_item; | |
2692 | struct btrfs_path *path; | |
5f39d397 | 2693 | struct extent_buffer *leaf; |
39279cc3 CM |
2694 | int ret; |
2695 | ||
2696 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2697 | if (!path) |
2698 | return -ENOMEM; | |
2699 | ||
b9473439 | 2700 | path->leave_spinning = 1; |
16cdcec7 MX |
2701 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2702 | 1); | |
39279cc3 CM |
2703 | if (ret) { |
2704 | if (ret > 0) | |
2705 | ret = -ENOENT; | |
2706 | goto failed; | |
2707 | } | |
2708 | ||
b4ce94de | 2709 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2710 | leaf = path->nodes[0]; |
2711 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2712 | struct btrfs_inode_item); |
39279cc3 | 2713 | |
e02119d5 | 2714 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2715 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2716 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2717 | ret = 0; |
2718 | failed: | |
39279cc3 CM |
2719 | btrfs_free_path(path); |
2720 | return ret; | |
2721 | } | |
2722 | ||
2115133f CM |
2723 | /* |
2724 | * copy everything in the in-memory inode into the btree. | |
2725 | */ | |
2726 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2727 | struct btrfs_root *root, struct inode *inode) | |
2728 | { | |
2729 | int ret; | |
2730 | ||
2731 | /* | |
2732 | * If the inode is a free space inode, we can deadlock during commit | |
2733 | * if we put it into the delayed code. | |
2734 | * | |
2735 | * The data relocation inode should also be directly updated | |
2736 | * without delay | |
2737 | */ | |
83eea1f1 | 2738 | if (!btrfs_is_free_space_inode(inode) |
2115133f | 2739 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { |
8ea05e3a AB |
2740 | btrfs_update_root_times(trans, root); |
2741 | ||
2115133f CM |
2742 | ret = btrfs_delayed_update_inode(trans, root, inode); |
2743 | if (!ret) | |
2744 | btrfs_set_inode_last_trans(trans, inode); | |
2745 | return ret; | |
2746 | } | |
2747 | ||
2748 | return btrfs_update_inode_item(trans, root, inode); | |
2749 | } | |
2750 | ||
2751 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, | |
2752 | struct btrfs_root *root, struct inode *inode) | |
2753 | { | |
2754 | int ret; | |
2755 | ||
2756 | ret = btrfs_update_inode(trans, root, inode); | |
2757 | if (ret == -ENOSPC) | |
2758 | return btrfs_update_inode_item(trans, root, inode); | |
2759 | return ret; | |
2760 | } | |
2761 | ||
d352ac68 CM |
2762 | /* |
2763 | * unlink helper that gets used here in inode.c and in the tree logging | |
2764 | * recovery code. It remove a link in a directory with a given name, and | |
2765 | * also drops the back refs in the inode to the directory | |
2766 | */ | |
92986796 AV |
2767 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2768 | struct btrfs_root *root, | |
2769 | struct inode *dir, struct inode *inode, | |
2770 | const char *name, int name_len) | |
39279cc3 CM |
2771 | { |
2772 | struct btrfs_path *path; | |
39279cc3 | 2773 | int ret = 0; |
5f39d397 | 2774 | struct extent_buffer *leaf; |
39279cc3 | 2775 | struct btrfs_dir_item *di; |
5f39d397 | 2776 | struct btrfs_key key; |
aec7477b | 2777 | u64 index; |
33345d01 LZ |
2778 | u64 ino = btrfs_ino(inode); |
2779 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2780 | |
2781 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2782 | if (!path) { |
2783 | ret = -ENOMEM; | |
554233a6 | 2784 | goto out; |
54aa1f4d CM |
2785 | } |
2786 | ||
b9473439 | 2787 | path->leave_spinning = 1; |
33345d01 | 2788 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2789 | name, name_len, -1); |
2790 | if (IS_ERR(di)) { | |
2791 | ret = PTR_ERR(di); | |
2792 | goto err; | |
2793 | } | |
2794 | if (!di) { | |
2795 | ret = -ENOENT; | |
2796 | goto err; | |
2797 | } | |
5f39d397 CM |
2798 | leaf = path->nodes[0]; |
2799 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2800 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2801 | if (ret) |
2802 | goto err; | |
b3b4aa74 | 2803 | btrfs_release_path(path); |
39279cc3 | 2804 | |
33345d01 LZ |
2805 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2806 | dir_ino, &index); | |
aec7477b | 2807 | if (ret) { |
d397712b | 2808 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2809 | "inode %llu parent %llu\n", name_len, name, |
2810 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
79787eaa | 2811 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
2812 | goto err; |
2813 | } | |
2814 | ||
16cdcec7 | 2815 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
2816 | if (ret) { |
2817 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 2818 | goto err; |
79787eaa | 2819 | } |
39279cc3 | 2820 | |
e02119d5 | 2821 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2822 | inode, dir_ino); |
79787eaa JM |
2823 | if (ret != 0 && ret != -ENOENT) { |
2824 | btrfs_abort_transaction(trans, root, ret); | |
2825 | goto err; | |
2826 | } | |
e02119d5 CM |
2827 | |
2828 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2829 | dir, index); | |
6418c961 CM |
2830 | if (ret == -ENOENT) |
2831 | ret = 0; | |
39279cc3 CM |
2832 | err: |
2833 | btrfs_free_path(path); | |
e02119d5 CM |
2834 | if (ret) |
2835 | goto out; | |
2836 | ||
2837 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
2838 | inode_inc_iversion(inode); |
2839 | inode_inc_iversion(dir); | |
e02119d5 | 2840 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 2841 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 2842 | out: |
39279cc3 CM |
2843 | return ret; |
2844 | } | |
2845 | ||
92986796 AV |
2846 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2847 | struct btrfs_root *root, | |
2848 | struct inode *dir, struct inode *inode, | |
2849 | const char *name, int name_len) | |
2850 | { | |
2851 | int ret; | |
2852 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2853 | if (!ret) { | |
2854 | btrfs_drop_nlink(inode); | |
2855 | ret = btrfs_update_inode(trans, root, inode); | |
2856 | } | |
2857 | return ret; | |
2858 | } | |
2859 | ||
2860 | ||
a22285a6 YZ |
2861 | /* helper to check if there is any shared block in the path */ |
2862 | static int check_path_shared(struct btrfs_root *root, | |
2863 | struct btrfs_path *path) | |
39279cc3 | 2864 | { |
a22285a6 YZ |
2865 | struct extent_buffer *eb; |
2866 | int level; | |
0e4dcbef | 2867 | u64 refs = 1; |
5df6a9f6 | 2868 | |
a22285a6 | 2869 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2870 | int ret; |
2871 | ||
a22285a6 YZ |
2872 | if (!path->nodes[level]) |
2873 | break; | |
2874 | eb = path->nodes[level]; | |
2875 | if (!btrfs_block_can_be_shared(root, eb)) | |
2876 | continue; | |
2877 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2878 | &refs, NULL); | |
2879 | if (refs > 1) | |
2880 | return 1; | |
5df6a9f6 | 2881 | } |
dedefd72 | 2882 | return 0; |
39279cc3 CM |
2883 | } |
2884 | ||
a22285a6 YZ |
2885 | /* |
2886 | * helper to start transaction for unlink and rmdir. | |
2887 | * | |
2888 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2889 | * so in enospc case, we should make sure they will free space before | |
2890 | * allowing them to use the global metadata reservation. | |
2891 | */ | |
2892 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2893 | struct dentry *dentry) | |
4df27c4d | 2894 | { |
39279cc3 | 2895 | struct btrfs_trans_handle *trans; |
a22285a6 | 2896 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2897 | struct btrfs_path *path; |
4df27c4d | 2898 | struct btrfs_dir_item *di; |
7b128766 | 2899 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2900 | u64 index; |
a22285a6 YZ |
2901 | int check_link = 1; |
2902 | int err = -ENOSPC; | |
4df27c4d | 2903 | int ret; |
33345d01 LZ |
2904 | u64 ino = btrfs_ino(inode); |
2905 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2906 | |
e70bea5f JB |
2907 | /* |
2908 | * 1 for the possible orphan item | |
2909 | * 1 for the dir item | |
2910 | * 1 for the dir index | |
2911 | * 1 for the inode ref | |
2912 | * 1 for the inode ref in the tree log | |
2913 | * 2 for the dir entries in the log | |
2914 | * 1 for the inode | |
2915 | */ | |
2916 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2917 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2918 | return trans; | |
4df27c4d | 2919 | |
33345d01 | 2920 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2921 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2922 | |
a22285a6 YZ |
2923 | /* check if there is someone else holds reference */ |
2924 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2925 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2926 | |
a22285a6 YZ |
2927 | if (atomic_read(&inode->i_count) > 2) |
2928 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2929 | |
a22285a6 YZ |
2930 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2931 | return ERR_PTR(-ENOSPC); | |
2932 | ||
2933 | path = btrfs_alloc_path(); | |
2934 | if (!path) { | |
2935 | root->fs_info->enospc_unlink = 0; | |
2936 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2937 | } |
2938 | ||
3880a1b4 JB |
2939 | /* 1 for the orphan item */ |
2940 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2941 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2942 | btrfs_free_path(path); |
2943 | root->fs_info->enospc_unlink = 0; | |
2944 | return trans; | |
2945 | } | |
4df27c4d | 2946 | |
a22285a6 YZ |
2947 | path->skip_locking = 1; |
2948 | path->search_commit_root = 1; | |
4df27c4d | 2949 | |
a22285a6 YZ |
2950 | ret = btrfs_lookup_inode(trans, root, path, |
2951 | &BTRFS_I(dir)->location, 0); | |
2952 | if (ret < 0) { | |
2953 | err = ret; | |
2954 | goto out; | |
2955 | } | |
2956 | if (ret == 0) { | |
2957 | if (check_path_shared(root, path)) | |
2958 | goto out; | |
2959 | } else { | |
2960 | check_link = 0; | |
5df6a9f6 | 2961 | } |
b3b4aa74 | 2962 | btrfs_release_path(path); |
a22285a6 YZ |
2963 | |
2964 | ret = btrfs_lookup_inode(trans, root, path, | |
2965 | &BTRFS_I(inode)->location, 0); | |
2966 | if (ret < 0) { | |
2967 | err = ret; | |
2968 | goto out; | |
2969 | } | |
2970 | if (ret == 0) { | |
2971 | if (check_path_shared(root, path)) | |
2972 | goto out; | |
2973 | } else { | |
2974 | check_link = 0; | |
2975 | } | |
b3b4aa74 | 2976 | btrfs_release_path(path); |
a22285a6 YZ |
2977 | |
2978 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2979 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2980 | ino, (u64)-1, 0); |
a22285a6 YZ |
2981 | if (ret < 0) { |
2982 | err = ret; | |
2983 | goto out; | |
2984 | } | |
79787eaa | 2985 | BUG_ON(ret == 0); /* Corruption */ |
a22285a6 YZ |
2986 | if (check_path_shared(root, path)) |
2987 | goto out; | |
b3b4aa74 | 2988 | btrfs_release_path(path); |
a22285a6 YZ |
2989 | } |
2990 | ||
2991 | if (!check_link) { | |
2992 | err = 0; | |
2993 | goto out; | |
2994 | } | |
2995 | ||
33345d01 | 2996 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2997 | dentry->d_name.name, dentry->d_name.len, 0); |
2998 | if (IS_ERR(di)) { | |
2999 | err = PTR_ERR(di); | |
3000 | goto out; | |
3001 | } | |
3002 | if (di) { | |
3003 | if (check_path_shared(root, path)) | |
3004 | goto out; | |
3005 | } else { | |
3006 | err = 0; | |
3007 | goto out; | |
3008 | } | |
b3b4aa74 | 3009 | btrfs_release_path(path); |
a22285a6 | 3010 | |
f186373f MF |
3011 | ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name, |
3012 | dentry->d_name.len, ino, dir_ino, 0, | |
3013 | &index); | |
3014 | if (ret) { | |
3015 | err = ret; | |
a22285a6 YZ |
3016 | goto out; |
3017 | } | |
f186373f | 3018 | |
a22285a6 YZ |
3019 | if (check_path_shared(root, path)) |
3020 | goto out; | |
f186373f | 3021 | |
b3b4aa74 | 3022 | btrfs_release_path(path); |
a22285a6 | 3023 | |
16cdcec7 MX |
3024 | /* |
3025 | * This is a commit root search, if we can lookup inode item and other | |
3026 | * relative items in the commit root, it means the transaction of | |
3027 | * dir/file creation has been committed, and the dir index item that we | |
3028 | * delay to insert has also been inserted into the commit root. So | |
3029 | * we needn't worry about the delayed insertion of the dir index item | |
3030 | * here. | |
3031 | */ | |
33345d01 | 3032 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
3033 | dentry->d_name.name, dentry->d_name.len, 0); |
3034 | if (IS_ERR(di)) { | |
3035 | err = PTR_ERR(di); | |
3036 | goto out; | |
3037 | } | |
3038 | BUG_ON(ret == -ENOENT); | |
3039 | if (check_path_shared(root, path)) | |
3040 | goto out; | |
3041 | ||
3042 | err = 0; | |
3043 | out: | |
3044 | btrfs_free_path(path); | |
3880a1b4 JB |
3045 | /* Migrate the orphan reservation over */ |
3046 | if (!err) | |
3047 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
3048 | &root->fs_info->global_block_rsv, | |
5a77d76c | 3049 | trans->bytes_reserved); |
3880a1b4 | 3050 | |
a22285a6 YZ |
3051 | if (err) { |
3052 | btrfs_end_transaction(trans, root); | |
3053 | root->fs_info->enospc_unlink = 0; | |
3054 | return ERR_PTR(err); | |
3055 | } | |
3056 | ||
3057 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
3058 | return trans; | |
3059 | } | |
3060 | ||
3061 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
3062 | struct btrfs_root *root) | |
3063 | { | |
66d8f3dd | 3064 | if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) { |
5a77d76c JB |
3065 | btrfs_block_rsv_release(root, trans->block_rsv, |
3066 | trans->bytes_reserved); | |
3067 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
3068 | BUG_ON(!root->fs_info->enospc_unlink); |
3069 | root->fs_info->enospc_unlink = 0; | |
3070 | } | |
7ad85bb7 | 3071 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
3072 | } |
3073 | ||
3074 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
3075 | { | |
3076 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
3077 | struct btrfs_trans_handle *trans; | |
3078 | struct inode *inode = dentry->d_inode; | |
3079 | int ret; | |
3080 | unsigned long nr = 0; | |
3081 | ||
3082 | trans = __unlink_start_trans(dir, dentry); | |
3083 | if (IS_ERR(trans)) | |
3084 | return PTR_ERR(trans); | |
5f39d397 | 3085 | |
12fcfd22 CM |
3086 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
3087 | ||
e02119d5 CM |
3088 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3089 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
3090 | if (ret) |
3091 | goto out; | |
7b128766 | 3092 | |
a22285a6 | 3093 | if (inode->i_nlink == 0) { |
7b128766 | 3094 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
3095 | if (ret) |
3096 | goto out; | |
a22285a6 | 3097 | } |
7b128766 | 3098 | |
b532402e | 3099 | out: |
d3c2fdcf | 3100 | nr = trans->blocks_used; |
a22285a6 | 3101 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3102 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
3103 | return ret; |
3104 | } | |
3105 | ||
4df27c4d YZ |
3106 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
3107 | struct btrfs_root *root, | |
3108 | struct inode *dir, u64 objectid, | |
3109 | const char *name, int name_len) | |
3110 | { | |
3111 | struct btrfs_path *path; | |
3112 | struct extent_buffer *leaf; | |
3113 | struct btrfs_dir_item *di; | |
3114 | struct btrfs_key key; | |
3115 | u64 index; | |
3116 | int ret; | |
33345d01 | 3117 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
3118 | |
3119 | path = btrfs_alloc_path(); | |
3120 | if (!path) | |
3121 | return -ENOMEM; | |
3122 | ||
33345d01 | 3123 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 3124 | name, name_len, -1); |
79787eaa JM |
3125 | if (IS_ERR_OR_NULL(di)) { |
3126 | if (!di) | |
3127 | ret = -ENOENT; | |
3128 | else | |
3129 | ret = PTR_ERR(di); | |
3130 | goto out; | |
3131 | } | |
4df27c4d YZ |
3132 | |
3133 | leaf = path->nodes[0]; | |
3134 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
3135 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
3136 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
3137 | if (ret) { |
3138 | btrfs_abort_transaction(trans, root, ret); | |
3139 | goto out; | |
3140 | } | |
b3b4aa74 | 3141 | btrfs_release_path(path); |
4df27c4d YZ |
3142 | |
3143 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
3144 | objectid, root->root_key.objectid, | |
33345d01 | 3145 | dir_ino, &index, name, name_len); |
4df27c4d | 3146 | if (ret < 0) { |
79787eaa JM |
3147 | if (ret != -ENOENT) { |
3148 | btrfs_abort_transaction(trans, root, ret); | |
3149 | goto out; | |
3150 | } | |
33345d01 | 3151 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 3152 | name, name_len); |
79787eaa JM |
3153 | if (IS_ERR_OR_NULL(di)) { |
3154 | if (!di) | |
3155 | ret = -ENOENT; | |
3156 | else | |
3157 | ret = PTR_ERR(di); | |
3158 | btrfs_abort_transaction(trans, root, ret); | |
3159 | goto out; | |
3160 | } | |
4df27c4d YZ |
3161 | |
3162 | leaf = path->nodes[0]; | |
3163 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 3164 | btrfs_release_path(path); |
4df27c4d YZ |
3165 | index = key.offset; |
3166 | } | |
945d8962 | 3167 | btrfs_release_path(path); |
4df27c4d | 3168 | |
16cdcec7 | 3169 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3170 | if (ret) { |
3171 | btrfs_abort_transaction(trans, root, ret); | |
3172 | goto out; | |
3173 | } | |
4df27c4d YZ |
3174 | |
3175 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 3176 | inode_inc_iversion(dir); |
4df27c4d | 3177 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 3178 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
3179 | if (ret) |
3180 | btrfs_abort_transaction(trans, root, ret); | |
3181 | out: | |
71d7aed0 | 3182 | btrfs_free_path(path); |
79787eaa | 3183 | return ret; |
4df27c4d YZ |
3184 | } |
3185 | ||
39279cc3 CM |
3186 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
3187 | { | |
3188 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 3189 | int err = 0; |
39279cc3 | 3190 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 3191 | struct btrfs_trans_handle *trans; |
1832a6d5 | 3192 | unsigned long nr = 0; |
39279cc3 | 3193 | |
b3ae244e | 3194 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 3195 | return -ENOTEMPTY; |
b3ae244e DS |
3196 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
3197 | return -EPERM; | |
134d4512 | 3198 | |
a22285a6 YZ |
3199 | trans = __unlink_start_trans(dir, dentry); |
3200 | if (IS_ERR(trans)) | |
5df6a9f6 | 3201 | return PTR_ERR(trans); |
5df6a9f6 | 3202 | |
33345d01 | 3203 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
3204 | err = btrfs_unlink_subvol(trans, root, dir, |
3205 | BTRFS_I(inode)->location.objectid, | |
3206 | dentry->d_name.name, | |
3207 | dentry->d_name.len); | |
3208 | goto out; | |
3209 | } | |
3210 | ||
7b128766 JB |
3211 | err = btrfs_orphan_add(trans, inode); |
3212 | if (err) | |
4df27c4d | 3213 | goto out; |
7b128766 | 3214 | |
39279cc3 | 3215 | /* now the directory is empty */ |
e02119d5 CM |
3216 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
3217 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 3218 | if (!err) |
dbe674a9 | 3219 | btrfs_i_size_write(inode, 0); |
4df27c4d | 3220 | out: |
d3c2fdcf | 3221 | nr = trans->blocks_used; |
a22285a6 | 3222 | __unlink_end_trans(trans, root); |
d3c2fdcf | 3223 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 3224 | |
39279cc3 CM |
3225 | return err; |
3226 | } | |
3227 | ||
39279cc3 CM |
3228 | /* |
3229 | * this can truncate away extent items, csum items and directory items. | |
3230 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3231 | * any higher than new_size |
39279cc3 CM |
3232 | * |
3233 | * csum items that cross the new i_size are truncated to the new size | |
3234 | * as well. | |
7b128766 JB |
3235 | * |
3236 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3237 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3238 | */ |
8082510e YZ |
3239 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3240 | struct btrfs_root *root, | |
3241 | struct inode *inode, | |
3242 | u64 new_size, u32 min_type) | |
39279cc3 | 3243 | { |
39279cc3 | 3244 | struct btrfs_path *path; |
5f39d397 | 3245 | struct extent_buffer *leaf; |
39279cc3 | 3246 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3247 | struct btrfs_key key; |
3248 | struct btrfs_key found_key; | |
39279cc3 | 3249 | u64 extent_start = 0; |
db94535d | 3250 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3251 | u64 extent_offset = 0; |
39279cc3 | 3252 | u64 item_end = 0; |
8082510e YZ |
3253 | u64 mask = root->sectorsize - 1; |
3254 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3255 | int found_extent; |
3256 | int del_item; | |
85e21bac CM |
3257 | int pending_del_nr = 0; |
3258 | int pending_del_slot = 0; | |
179e29e4 | 3259 | int extent_type = -1; |
8082510e YZ |
3260 | int ret; |
3261 | int err = 0; | |
33345d01 | 3262 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3263 | |
3264 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3265 | |
0eb0e19c MF |
3266 | path = btrfs_alloc_path(); |
3267 | if (!path) | |
3268 | return -ENOMEM; | |
3269 | path->reada = -1; | |
3270 | ||
5dc562c5 JB |
3271 | /* |
3272 | * We want to drop from the next block forward in case this new size is | |
3273 | * not block aligned since we will be keeping the last block of the | |
3274 | * extent just the way it is. | |
3275 | */ | |
0af3d00b | 3276 | if (root->ref_cows || root == root->fs_info->tree_root) |
5dc562c5 | 3277 | btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0); |
8082510e | 3278 | |
16cdcec7 MX |
3279 | /* |
3280 | * This function is also used to drop the items in the log tree before | |
3281 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3282 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3283 | * items. | |
3284 | */ | |
3285 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3286 | btrfs_kill_delayed_inode_items(inode); | |
3287 | ||
33345d01 | 3288 | key.objectid = ino; |
39279cc3 | 3289 | key.offset = (u64)-1; |
5f39d397 CM |
3290 | key.type = (u8)-1; |
3291 | ||
85e21bac | 3292 | search_again: |
b9473439 | 3293 | path->leave_spinning = 1; |
85e21bac | 3294 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3295 | if (ret < 0) { |
3296 | err = ret; | |
3297 | goto out; | |
3298 | } | |
d397712b | 3299 | |
85e21bac | 3300 | if (ret > 0) { |
e02119d5 CM |
3301 | /* there are no items in the tree for us to truncate, we're |
3302 | * done | |
3303 | */ | |
8082510e YZ |
3304 | if (path->slots[0] == 0) |
3305 | goto out; | |
85e21bac CM |
3306 | path->slots[0]--; |
3307 | } | |
3308 | ||
d397712b | 3309 | while (1) { |
39279cc3 | 3310 | fi = NULL; |
5f39d397 CM |
3311 | leaf = path->nodes[0]; |
3312 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3313 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3314 | |
33345d01 | 3315 | if (found_key.objectid != ino) |
39279cc3 | 3316 | break; |
5f39d397 | 3317 | |
85e21bac | 3318 | if (found_type < min_type) |
39279cc3 CM |
3319 | break; |
3320 | ||
5f39d397 | 3321 | item_end = found_key.offset; |
39279cc3 | 3322 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3323 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3324 | struct btrfs_file_extent_item); |
179e29e4 CM |
3325 | extent_type = btrfs_file_extent_type(leaf, fi); |
3326 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3327 | item_end += |
db94535d | 3328 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3329 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3330 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3331 | fi); |
39279cc3 | 3332 | } |
008630c1 | 3333 | item_end--; |
39279cc3 | 3334 | } |
8082510e YZ |
3335 | if (found_type > min_type) { |
3336 | del_item = 1; | |
3337 | } else { | |
3338 | if (item_end < new_size) | |
b888db2b | 3339 | break; |
8082510e YZ |
3340 | if (found_key.offset >= new_size) |
3341 | del_item = 1; | |
3342 | else | |
3343 | del_item = 0; | |
39279cc3 | 3344 | } |
39279cc3 | 3345 | found_extent = 0; |
39279cc3 | 3346 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3347 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3348 | goto delete; | |
3349 | ||
3350 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3351 | u64 num_dec; |
db94535d | 3352 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3353 | if (!del_item) { |
db94535d CM |
3354 | u64 orig_num_bytes = |
3355 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3356 | extent_num_bytes = new_size - |
5f39d397 | 3357 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3358 | extent_num_bytes = extent_num_bytes & |
3359 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3360 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3361 | extent_num_bytes); | |
3362 | num_dec = (orig_num_bytes - | |
9069218d | 3363 | extent_num_bytes); |
e02119d5 | 3364 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3365 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3366 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3367 | } else { |
db94535d CM |
3368 | extent_num_bytes = |
3369 | btrfs_file_extent_disk_num_bytes(leaf, | |
3370 | fi); | |
5d4f98a2 YZ |
3371 | extent_offset = found_key.offset - |
3372 | btrfs_file_extent_offset(leaf, fi); | |
3373 | ||
39279cc3 | 3374 | /* FIXME blocksize != 4096 */ |
9069218d | 3375 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3376 | if (extent_start != 0) { |
3377 | found_extent = 1; | |
e02119d5 | 3378 | if (root->ref_cows) |
a76a3cd4 | 3379 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3380 | } |
39279cc3 | 3381 | } |
9069218d | 3382 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3383 | /* |
3384 | * we can't truncate inline items that have had | |
3385 | * special encodings | |
3386 | */ | |
3387 | if (!del_item && | |
3388 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3389 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3390 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3391 | u32 size = new_size - found_key.offset; |
3392 | ||
3393 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3394 | inode_sub_bytes(inode, item_end + 1 - |
3395 | new_size); | |
e02119d5 CM |
3396 | } |
3397 | size = | |
3398 | btrfs_file_extent_calc_inline_size(size); | |
143bede5 JM |
3399 | btrfs_truncate_item(trans, root, path, |
3400 | size, 1); | |
e02119d5 | 3401 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3402 | inode_sub_bytes(inode, item_end + 1 - |
3403 | found_key.offset); | |
9069218d | 3404 | } |
39279cc3 | 3405 | } |
179e29e4 | 3406 | delete: |
39279cc3 | 3407 | if (del_item) { |
85e21bac CM |
3408 | if (!pending_del_nr) { |
3409 | /* no pending yet, add ourselves */ | |
3410 | pending_del_slot = path->slots[0]; | |
3411 | pending_del_nr = 1; | |
3412 | } else if (pending_del_nr && | |
3413 | path->slots[0] + 1 == pending_del_slot) { | |
3414 | /* hop on the pending chunk */ | |
3415 | pending_del_nr++; | |
3416 | pending_del_slot = path->slots[0]; | |
3417 | } else { | |
d397712b | 3418 | BUG(); |
85e21bac | 3419 | } |
39279cc3 CM |
3420 | } else { |
3421 | break; | |
3422 | } | |
0af3d00b JB |
3423 | if (found_extent && (root->ref_cows || |
3424 | root == root->fs_info->tree_root)) { | |
b9473439 | 3425 | btrfs_set_path_blocking(path); |
39279cc3 | 3426 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3427 | extent_num_bytes, 0, |
3428 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3429 | ino, extent_offset, 0); |
39279cc3 CM |
3430 | BUG_ON(ret); |
3431 | } | |
85e21bac | 3432 | |
8082510e YZ |
3433 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3434 | break; | |
3435 | ||
3436 | if (path->slots[0] == 0 || | |
3437 | path->slots[0] != pending_del_slot) { | |
8082510e YZ |
3438 | if (pending_del_nr) { |
3439 | ret = btrfs_del_items(trans, root, path, | |
3440 | pending_del_slot, | |
3441 | pending_del_nr); | |
79787eaa JM |
3442 | if (ret) { |
3443 | btrfs_abort_transaction(trans, | |
3444 | root, ret); | |
3445 | goto error; | |
3446 | } | |
8082510e YZ |
3447 | pending_del_nr = 0; |
3448 | } | |
b3b4aa74 | 3449 | btrfs_release_path(path); |
85e21bac | 3450 | goto search_again; |
8082510e YZ |
3451 | } else { |
3452 | path->slots[0]--; | |
85e21bac | 3453 | } |
39279cc3 | 3454 | } |
8082510e | 3455 | out: |
85e21bac CM |
3456 | if (pending_del_nr) { |
3457 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3458 | pending_del_nr); | |
79787eaa JM |
3459 | if (ret) |
3460 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 3461 | } |
79787eaa | 3462 | error: |
39279cc3 | 3463 | btrfs_free_path(path); |
8082510e | 3464 | return err; |
39279cc3 CM |
3465 | } |
3466 | ||
3467 | /* | |
2aaa6655 JB |
3468 | * btrfs_truncate_page - read, zero a chunk and write a page |
3469 | * @inode - inode that we're zeroing | |
3470 | * @from - the offset to start zeroing | |
3471 | * @len - the length to zero, 0 to zero the entire range respective to the | |
3472 | * offset | |
3473 | * @front - zero up to the offset instead of from the offset on | |
3474 | * | |
3475 | * This will find the page for the "from" offset and cow the page and zero the | |
3476 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 3477 | */ |
2aaa6655 JB |
3478 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
3479 | int front) | |
39279cc3 | 3480 | { |
2aaa6655 | 3481 | struct address_space *mapping = inode->i_mapping; |
db94535d | 3482 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3483 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3484 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3485 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3486 | char *kaddr; |
db94535d | 3487 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3488 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3489 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3490 | struct page *page; | |
3b16a4e3 | 3491 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3492 | int ret = 0; |
a52d9a80 | 3493 | u64 page_start; |
e6dcd2dc | 3494 | u64 page_end; |
39279cc3 | 3495 | |
2aaa6655 JB |
3496 | if ((offset & (blocksize - 1)) == 0 && |
3497 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 3498 | goto out; |
0ca1f7ce | 3499 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3500 | if (ret) |
3501 | goto out; | |
39279cc3 CM |
3502 | |
3503 | ret = -ENOMEM; | |
211c17f5 | 3504 | again: |
3b16a4e3 | 3505 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3506 | if (!page) { |
0ca1f7ce | 3507 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3508 | goto out; |
5d5e103a | 3509 | } |
e6dcd2dc CM |
3510 | |
3511 | page_start = page_offset(page); | |
3512 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3513 | ||
39279cc3 | 3514 | if (!PageUptodate(page)) { |
9ebefb18 | 3515 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3516 | lock_page(page); |
211c17f5 CM |
3517 | if (page->mapping != mapping) { |
3518 | unlock_page(page); | |
3519 | page_cache_release(page); | |
3520 | goto again; | |
3521 | } | |
39279cc3 CM |
3522 | if (!PageUptodate(page)) { |
3523 | ret = -EIO; | |
89642229 | 3524 | goto out_unlock; |
39279cc3 CM |
3525 | } |
3526 | } | |
211c17f5 | 3527 | wait_on_page_writeback(page); |
e6dcd2dc | 3528 | |
d0082371 | 3529 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
3530 | set_page_extent_mapped(page); |
3531 | ||
3532 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3533 | if (ordered) { | |
2ac55d41 JB |
3534 | unlock_extent_cached(io_tree, page_start, page_end, |
3535 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3536 | unlock_page(page); |
3537 | page_cache_release(page); | |
eb84ae03 | 3538 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3539 | btrfs_put_ordered_extent(ordered); |
3540 | goto again; | |
3541 | } | |
3542 | ||
2ac55d41 | 3543 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
3544 | EXTENT_DIRTY | EXTENT_DELALLOC | |
3545 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 3546 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3547 | |
2ac55d41 JB |
3548 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3549 | &cached_state); | |
9ed74f2d | 3550 | if (ret) { |
2ac55d41 JB |
3551 | unlock_extent_cached(io_tree, page_start, page_end, |
3552 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3553 | goto out_unlock; |
3554 | } | |
3555 | ||
e6dcd2dc CM |
3556 | ret = 0; |
3557 | if (offset != PAGE_CACHE_SIZE) { | |
2aaa6655 JB |
3558 | if (!len) |
3559 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 3560 | kaddr = kmap(page); |
2aaa6655 JB |
3561 | if (front) |
3562 | memset(kaddr, 0, offset); | |
3563 | else | |
3564 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
3565 | flush_dcache_page(page); |
3566 | kunmap(page); | |
3567 | } | |
247e743c | 3568 | ClearPageChecked(page); |
e6dcd2dc | 3569 | set_page_dirty(page); |
2ac55d41 JB |
3570 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3571 | GFP_NOFS); | |
39279cc3 | 3572 | |
89642229 | 3573 | out_unlock: |
5d5e103a | 3574 | if (ret) |
0ca1f7ce | 3575 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3576 | unlock_page(page); |
3577 | page_cache_release(page); | |
3578 | out: | |
3579 | return ret; | |
3580 | } | |
3581 | ||
695a0d0d JB |
3582 | /* |
3583 | * This function puts in dummy file extents for the area we're creating a hole | |
3584 | * for. So if we are truncating this file to a larger size we need to insert | |
3585 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3586 | * the range between oldsize and size | |
3587 | */ | |
a41ad394 | 3588 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3589 | { |
9036c102 YZ |
3590 | struct btrfs_trans_handle *trans; |
3591 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3592 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3593 | struct extent_map *em = NULL; |
2ac55d41 | 3594 | struct extent_state *cached_state = NULL; |
5dc562c5 | 3595 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9036c102 | 3596 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3597 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3598 | u64 block_end = (size + mask) & ~mask; |
3599 | u64 last_byte; | |
3600 | u64 cur_offset; | |
3601 | u64 hole_size; | |
9ed74f2d | 3602 | int err = 0; |
39279cc3 | 3603 | |
9036c102 YZ |
3604 | if (size <= hole_start) |
3605 | return 0; | |
3606 | ||
9036c102 YZ |
3607 | while (1) { |
3608 | struct btrfs_ordered_extent *ordered; | |
3609 | btrfs_wait_ordered_range(inode, hole_start, | |
3610 | block_end - hole_start); | |
2ac55d41 | 3611 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 3612 | &cached_state); |
9036c102 YZ |
3613 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3614 | if (!ordered) | |
3615 | break; | |
2ac55d41 JB |
3616 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3617 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3618 | btrfs_put_ordered_extent(ordered); |
3619 | } | |
39279cc3 | 3620 | |
9036c102 YZ |
3621 | cur_offset = hole_start; |
3622 | while (1) { | |
3623 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3624 | block_end - cur_offset, 0); | |
79787eaa JM |
3625 | if (IS_ERR(em)) { |
3626 | err = PTR_ERR(em); | |
3627 | break; | |
3628 | } | |
9036c102 YZ |
3629 | last_byte = min(extent_map_end(em), block_end); |
3630 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3631 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 3632 | struct extent_map *hole_em; |
9036c102 | 3633 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3634 | |
3642320e | 3635 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3636 | if (IS_ERR(trans)) { |
3637 | err = PTR_ERR(trans); | |
9ed74f2d | 3638 | break; |
a22285a6 | 3639 | } |
8082510e | 3640 | |
5dc562c5 JB |
3641 | err = btrfs_drop_extents(trans, root, inode, |
3642 | cur_offset, | |
2671485d | 3643 | cur_offset + hole_size, 1); |
5b397377 | 3644 | if (err) { |
79787eaa | 3645 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3646 | btrfs_end_transaction(trans, root); |
3893e33b | 3647 | break; |
5b397377 | 3648 | } |
8082510e | 3649 | |
9036c102 | 3650 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3651 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3652 | 0, hole_size, 0, hole_size, |
3653 | 0, 0, 0); | |
5b397377 | 3654 | if (err) { |
79787eaa | 3655 | btrfs_abort_transaction(trans, root, err); |
5b397377 | 3656 | btrfs_end_transaction(trans, root); |
3893e33b | 3657 | break; |
5b397377 | 3658 | } |
8082510e | 3659 | |
5dc562c5 JB |
3660 | btrfs_drop_extent_cache(inode, cur_offset, |
3661 | cur_offset + hole_size - 1, 0); | |
3662 | hole_em = alloc_extent_map(); | |
3663 | if (!hole_em) { | |
3664 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3665 | &BTRFS_I(inode)->runtime_flags); | |
3666 | goto next; | |
3667 | } | |
3668 | hole_em->start = cur_offset; | |
3669 | hole_em->len = hole_size; | |
3670 | hole_em->orig_start = cur_offset; | |
3671 | ||
3672 | hole_em->block_start = EXTENT_MAP_HOLE; | |
3673 | hole_em->block_len = 0; | |
3674 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; | |
3675 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
3676 | hole_em->generation = trans->transid; | |
8082510e | 3677 | |
5dc562c5 JB |
3678 | while (1) { |
3679 | write_lock(&em_tree->lock); | |
3680 | err = add_extent_mapping(em_tree, hole_em); | |
3681 | if (!err) | |
3682 | list_move(&hole_em->list, | |
3683 | &em_tree->modified_extents); | |
3684 | write_unlock(&em_tree->lock); | |
3685 | if (err != -EEXIST) | |
3686 | break; | |
3687 | btrfs_drop_extent_cache(inode, cur_offset, | |
3688 | cur_offset + | |
3689 | hole_size - 1, 0); | |
3690 | } | |
3691 | free_extent_map(hole_em); | |
3692 | next: | |
3642320e | 3693 | btrfs_update_inode(trans, root, inode); |
8082510e | 3694 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3695 | } |
3696 | free_extent_map(em); | |
a22285a6 | 3697 | em = NULL; |
9036c102 | 3698 | cur_offset = last_byte; |
8082510e | 3699 | if (cur_offset >= block_end) |
9036c102 YZ |
3700 | break; |
3701 | } | |
1832a6d5 | 3702 | |
a22285a6 | 3703 | free_extent_map(em); |
2ac55d41 JB |
3704 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3705 | GFP_NOFS); | |
9036c102 YZ |
3706 | return err; |
3707 | } | |
39279cc3 | 3708 | |
a41ad394 | 3709 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3710 | { |
f4a2f4c5 MX |
3711 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3712 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3713 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3714 | int ret; |
3715 | ||
a41ad394 | 3716 | if (newsize == oldsize) |
8082510e YZ |
3717 | return 0; |
3718 | ||
a41ad394 | 3719 | if (newsize > oldsize) { |
a41ad394 JB |
3720 | truncate_pagecache(inode, oldsize, newsize); |
3721 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3722 | if (ret) |
8082510e | 3723 | return ret; |
8082510e | 3724 | |
f4a2f4c5 MX |
3725 | trans = btrfs_start_transaction(root, 1); |
3726 | if (IS_ERR(trans)) | |
3727 | return PTR_ERR(trans); | |
3728 | ||
3729 | i_size_write(inode, newsize); | |
3730 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3731 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3732 | btrfs_end_transaction(trans, root); |
a41ad394 | 3733 | } else { |
8082510e | 3734 | |
a41ad394 JB |
3735 | /* |
3736 | * We're truncating a file that used to have good data down to | |
3737 | * zero. Make sure it gets into the ordered flush list so that | |
3738 | * any new writes get down to disk quickly. | |
3739 | */ | |
3740 | if (newsize == 0) | |
72ac3c0d JB |
3741 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
3742 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 3743 | |
a41ad394 JB |
3744 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3745 | truncate_setsize(inode, newsize); | |
3746 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3747 | } |
3748 | ||
a41ad394 | 3749 | return ret; |
8082510e YZ |
3750 | } |
3751 | ||
9036c102 YZ |
3752 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3753 | { | |
3754 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3755 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3756 | int err; |
39279cc3 | 3757 | |
b83cc969 LZ |
3758 | if (btrfs_root_readonly(root)) |
3759 | return -EROFS; | |
3760 | ||
9036c102 YZ |
3761 | err = inode_change_ok(inode, attr); |
3762 | if (err) | |
3763 | return err; | |
2bf5a725 | 3764 | |
5a3f23d5 | 3765 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3766 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3767 | if (err) |
3768 | return err; | |
39279cc3 | 3769 | } |
9036c102 | 3770 | |
1025774c CH |
3771 | if (attr->ia_valid) { |
3772 | setattr_copy(inode, attr); | |
0c4d2d95 | 3773 | inode_inc_iversion(inode); |
22c44fe6 | 3774 | err = btrfs_dirty_inode(inode); |
1025774c | 3775 | |
22c44fe6 | 3776 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3777 | err = btrfs_acl_chmod(inode); |
3778 | } | |
33268eaf | 3779 | |
39279cc3 CM |
3780 | return err; |
3781 | } | |
61295eb8 | 3782 | |
bd555975 | 3783 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3784 | { |
3785 | struct btrfs_trans_handle *trans; | |
3786 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3787 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3788 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3789 | unsigned long nr; |
39279cc3 CM |
3790 | int ret; |
3791 | ||
1abe9b8a | 3792 | trace_btrfs_inode_evict(inode); |
3793 | ||
39279cc3 | 3794 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3795 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
83eea1f1 | 3796 | btrfs_is_free_space_inode(inode))) |
bd555975 AV |
3797 | goto no_delete; |
3798 | ||
39279cc3 | 3799 | if (is_bad_inode(inode)) { |
7b128766 | 3800 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3801 | goto no_delete; |
3802 | } | |
bd555975 | 3803 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3804 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3805 | |
c71bf099 | 3806 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 3807 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 3808 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
3809 | goto no_delete; |
3810 | } | |
3811 | ||
76dda93c YZ |
3812 | if (inode->i_nlink > 0) { |
3813 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3814 | goto no_delete; | |
3815 | } | |
3816 | ||
66d8f3dd | 3817 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
3818 | if (!rsv) { |
3819 | btrfs_orphan_del(NULL, inode); | |
3820 | goto no_delete; | |
3821 | } | |
4a338542 | 3822 | rsv->size = min_size; |
ca7e70f5 | 3823 | rsv->failfast = 1; |
726c35fa | 3824 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3825 | |
dbe674a9 | 3826 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3827 | |
4289a667 | 3828 | /* |
8407aa46 MX |
3829 | * This is a bit simpler than btrfs_truncate since we've already |
3830 | * reserved our space for our orphan item in the unlink, so we just | |
3831 | * need to reserve some slack space in case we add bytes and update | |
3832 | * inode item when doing the truncate. | |
4289a667 | 3833 | */ |
8082510e | 3834 | while (1) { |
aa38a711 | 3835 | ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size); |
726c35fa JB |
3836 | |
3837 | /* | |
3838 | * Try and steal from the global reserve since we will | |
3839 | * likely not use this space anyway, we want to try as | |
3840 | * hard as possible to get this to work. | |
3841 | */ | |
3842 | if (ret) | |
3843 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3844 | |
d68fc57b | 3845 | if (ret) { |
4289a667 | 3846 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3847 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3848 | btrfs_orphan_del(NULL, inode); |
3849 | btrfs_free_block_rsv(root, rsv); | |
3850 | goto no_delete; | |
d68fc57b | 3851 | } |
7b128766 | 3852 | |
8407aa46 | 3853 | trans = btrfs_start_transaction_noflush(root, 1); |
4289a667 JB |
3854 | if (IS_ERR(trans)) { |
3855 | btrfs_orphan_del(NULL, inode); | |
3856 | btrfs_free_block_rsv(root, rsv); | |
3857 | goto no_delete; | |
d68fc57b | 3858 | } |
7b128766 | 3859 | |
4289a667 JB |
3860 | trans->block_rsv = rsv; |
3861 | ||
d68fc57b | 3862 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ca7e70f5 | 3863 | if (ret != -ENOSPC) |
8082510e | 3864 | break; |
85e21bac | 3865 | |
8407aa46 MX |
3866 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
3867 | ret = btrfs_update_inode(trans, root, inode); | |
3868 | BUG_ON(ret); | |
3869 | ||
8082510e YZ |
3870 | nr = trans->blocks_used; |
3871 | btrfs_end_transaction(trans, root); | |
3872 | trans = NULL; | |
3873 | btrfs_btree_balance_dirty(root, nr); | |
3874 | } | |
5f39d397 | 3875 | |
4289a667 JB |
3876 | btrfs_free_block_rsv(root, rsv); |
3877 | ||
8082510e | 3878 | if (ret == 0) { |
4289a667 | 3879 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3880 | ret = btrfs_orphan_del(trans, inode); |
3881 | BUG_ON(ret); | |
3882 | } | |
54aa1f4d | 3883 | |
4289a667 | 3884 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3885 | if (!(root == root->fs_info->tree_root || |
3886 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3887 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3888 | |
d3c2fdcf | 3889 | nr = trans->blocks_used; |
54aa1f4d | 3890 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3891 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3892 | no_delete: |
dbd5768f | 3893 | clear_inode(inode); |
8082510e | 3894 | return; |
39279cc3 CM |
3895 | } |
3896 | ||
3897 | /* | |
3898 | * this returns the key found in the dir entry in the location pointer. | |
3899 | * If no dir entries were found, location->objectid is 0. | |
3900 | */ | |
3901 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3902 | struct btrfs_key *location) | |
3903 | { | |
3904 | const char *name = dentry->d_name.name; | |
3905 | int namelen = dentry->d_name.len; | |
3906 | struct btrfs_dir_item *di; | |
3907 | struct btrfs_path *path; | |
3908 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3909 | int ret = 0; |
39279cc3 CM |
3910 | |
3911 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3912 | if (!path) |
3913 | return -ENOMEM; | |
3954401f | 3914 | |
33345d01 | 3915 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3916 | namelen, 0); |
0d9f7f3e Y |
3917 | if (IS_ERR(di)) |
3918 | ret = PTR_ERR(di); | |
d397712b | 3919 | |
c704005d | 3920 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3921 | goto out_err; |
d397712b | 3922 | |
5f39d397 | 3923 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3924 | out: |
39279cc3 CM |
3925 | btrfs_free_path(path); |
3926 | return ret; | |
3954401f CM |
3927 | out_err: |
3928 | location->objectid = 0; | |
3929 | goto out; | |
39279cc3 CM |
3930 | } |
3931 | ||
3932 | /* | |
3933 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3934 | * needs to be changed to reflect the root directory of the tree root. This | |
3935 | * is kind of like crossing a mount point. | |
3936 | */ | |
3937 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3938 | struct inode *dir, |
3939 | struct dentry *dentry, | |
3940 | struct btrfs_key *location, | |
3941 | struct btrfs_root **sub_root) | |
39279cc3 | 3942 | { |
4df27c4d YZ |
3943 | struct btrfs_path *path; |
3944 | struct btrfs_root *new_root; | |
3945 | struct btrfs_root_ref *ref; | |
3946 | struct extent_buffer *leaf; | |
3947 | int ret; | |
3948 | int err = 0; | |
39279cc3 | 3949 | |
4df27c4d YZ |
3950 | path = btrfs_alloc_path(); |
3951 | if (!path) { | |
3952 | err = -ENOMEM; | |
3953 | goto out; | |
3954 | } | |
39279cc3 | 3955 | |
4df27c4d YZ |
3956 | err = -ENOENT; |
3957 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3958 | BTRFS_I(dir)->root->root_key.objectid, | |
3959 | location->objectid); | |
3960 | if (ret) { | |
3961 | if (ret < 0) | |
3962 | err = ret; | |
3963 | goto out; | |
3964 | } | |
39279cc3 | 3965 | |
4df27c4d YZ |
3966 | leaf = path->nodes[0]; |
3967 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3968 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3969 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3970 | goto out; | |
39279cc3 | 3971 | |
4df27c4d YZ |
3972 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3973 | (unsigned long)(ref + 1), | |
3974 | dentry->d_name.len); | |
3975 | if (ret) | |
3976 | goto out; | |
3977 | ||
b3b4aa74 | 3978 | btrfs_release_path(path); |
4df27c4d YZ |
3979 | |
3980 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3981 | if (IS_ERR(new_root)) { | |
3982 | err = PTR_ERR(new_root); | |
3983 | goto out; | |
3984 | } | |
3985 | ||
3986 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3987 | err = -ENOENT; | |
3988 | goto out; | |
3989 | } | |
3990 | ||
3991 | *sub_root = new_root; | |
3992 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3993 | location->type = BTRFS_INODE_ITEM_KEY; | |
3994 | location->offset = 0; | |
3995 | err = 0; | |
3996 | out: | |
3997 | btrfs_free_path(path); | |
3998 | return err; | |
39279cc3 CM |
3999 | } |
4000 | ||
5d4f98a2 YZ |
4001 | static void inode_tree_add(struct inode *inode) |
4002 | { | |
4003 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4004 | struct btrfs_inode *entry; | |
03e860bd FNP |
4005 | struct rb_node **p; |
4006 | struct rb_node *parent; | |
33345d01 | 4007 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
4008 | again: |
4009 | p = &root->inode_tree.rb_node; | |
4010 | parent = NULL; | |
5d4f98a2 | 4011 | |
1d3382cb | 4012 | if (inode_unhashed(inode)) |
76dda93c YZ |
4013 | return; |
4014 | ||
5d4f98a2 YZ |
4015 | spin_lock(&root->inode_lock); |
4016 | while (*p) { | |
4017 | parent = *p; | |
4018 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
4019 | ||
33345d01 | 4020 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4021 | p = &parent->rb_left; |
33345d01 | 4022 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 4023 | p = &parent->rb_right; |
5d4f98a2 YZ |
4024 | else { |
4025 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 4026 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
4027 | rb_erase(parent, &root->inode_tree); |
4028 | RB_CLEAR_NODE(parent); | |
4029 | spin_unlock(&root->inode_lock); | |
4030 | goto again; | |
5d4f98a2 YZ |
4031 | } |
4032 | } | |
4033 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
4034 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
4035 | spin_unlock(&root->inode_lock); | |
4036 | } | |
4037 | ||
4038 | static void inode_tree_del(struct inode *inode) | |
4039 | { | |
4040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 4041 | int empty = 0; |
5d4f98a2 | 4042 | |
03e860bd | 4043 | spin_lock(&root->inode_lock); |
5d4f98a2 | 4044 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 4045 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 4046 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 4047 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 4048 | } |
03e860bd | 4049 | spin_unlock(&root->inode_lock); |
76dda93c | 4050 | |
0af3d00b JB |
4051 | /* |
4052 | * Free space cache has inodes in the tree root, but the tree root has a | |
4053 | * root_refs of 0, so this could end up dropping the tree root as a | |
4054 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
4055 | * make sure we don't drop it. | |
4056 | */ | |
4057 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
4058 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
4059 | synchronize_srcu(&root->fs_info->subvol_srcu); |
4060 | spin_lock(&root->inode_lock); | |
4061 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
4062 | spin_unlock(&root->inode_lock); | |
4063 | if (empty) | |
4064 | btrfs_add_dead_root(root); | |
4065 | } | |
4066 | } | |
4067 | ||
143bede5 | 4068 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
4069 | { |
4070 | struct rb_node *node; | |
4071 | struct rb_node *prev; | |
4072 | struct btrfs_inode *entry; | |
4073 | struct inode *inode; | |
4074 | u64 objectid = 0; | |
4075 | ||
4076 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
4077 | ||
4078 | spin_lock(&root->inode_lock); | |
4079 | again: | |
4080 | node = root->inode_tree.rb_node; | |
4081 | prev = NULL; | |
4082 | while (node) { | |
4083 | prev = node; | |
4084 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4085 | ||
33345d01 | 4086 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 4087 | node = node->rb_left; |
33345d01 | 4088 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
4089 | node = node->rb_right; |
4090 | else | |
4091 | break; | |
4092 | } | |
4093 | if (!node) { | |
4094 | while (prev) { | |
4095 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 4096 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
4097 | node = prev; |
4098 | break; | |
4099 | } | |
4100 | prev = rb_next(prev); | |
4101 | } | |
4102 | } | |
4103 | while (node) { | |
4104 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 4105 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
4106 | inode = igrab(&entry->vfs_inode); |
4107 | if (inode) { | |
4108 | spin_unlock(&root->inode_lock); | |
4109 | if (atomic_read(&inode->i_count) > 1) | |
4110 | d_prune_aliases(inode); | |
4111 | /* | |
45321ac5 | 4112 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
4113 | * the inode cache when its usage count |
4114 | * hits zero. | |
4115 | */ | |
4116 | iput(inode); | |
4117 | cond_resched(); | |
4118 | spin_lock(&root->inode_lock); | |
4119 | goto again; | |
4120 | } | |
4121 | ||
4122 | if (cond_resched_lock(&root->inode_lock)) | |
4123 | goto again; | |
4124 | ||
4125 | node = rb_next(node); | |
4126 | } | |
4127 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
4128 | } |
4129 | ||
e02119d5 CM |
4130 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
4131 | { | |
4132 | struct btrfs_iget_args *args = p; | |
4133 | inode->i_ino = args->ino; | |
e02119d5 | 4134 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
4135 | return 0; |
4136 | } | |
4137 | ||
4138 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
4139 | { | |
4140 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 4141 | return args->ino == btrfs_ino(inode) && |
d397712b | 4142 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
4143 | } |
4144 | ||
5d4f98a2 YZ |
4145 | static struct inode *btrfs_iget_locked(struct super_block *s, |
4146 | u64 objectid, | |
4147 | struct btrfs_root *root) | |
39279cc3 CM |
4148 | { |
4149 | struct inode *inode; | |
4150 | struct btrfs_iget_args args; | |
4151 | args.ino = objectid; | |
4152 | args.root = root; | |
4153 | ||
4154 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
4155 | btrfs_init_locked_inode, | |
4156 | (void *)&args); | |
4157 | return inode; | |
4158 | } | |
4159 | ||
1a54ef8c BR |
4160 | /* Get an inode object given its location and corresponding root. |
4161 | * Returns in *is_new if the inode was read from disk | |
4162 | */ | |
4163 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 4164 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
4165 | { |
4166 | struct inode *inode; | |
4167 | ||
4168 | inode = btrfs_iget_locked(s, location->objectid, root); | |
4169 | if (!inode) | |
5d4f98a2 | 4170 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
4171 | |
4172 | if (inode->i_state & I_NEW) { | |
4173 | BTRFS_I(inode)->root = root; | |
4174 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
4175 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
4176 | if (!is_bad_inode(inode)) { |
4177 | inode_tree_add(inode); | |
4178 | unlock_new_inode(inode); | |
4179 | if (new) | |
4180 | *new = 1; | |
4181 | } else { | |
e0b6d65b ST |
4182 | unlock_new_inode(inode); |
4183 | iput(inode); | |
4184 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
4185 | } |
4186 | } | |
4187 | ||
1a54ef8c BR |
4188 | return inode; |
4189 | } | |
4190 | ||
4df27c4d YZ |
4191 | static struct inode *new_simple_dir(struct super_block *s, |
4192 | struct btrfs_key *key, | |
4193 | struct btrfs_root *root) | |
4194 | { | |
4195 | struct inode *inode = new_inode(s); | |
4196 | ||
4197 | if (!inode) | |
4198 | return ERR_PTR(-ENOMEM); | |
4199 | ||
4df27c4d YZ |
4200 | BTRFS_I(inode)->root = root; |
4201 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 4202 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
4203 | |
4204 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 4205 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
4206 | inode->i_fop = &simple_dir_operations; |
4207 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
4208 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
4209 | ||
4210 | return inode; | |
4211 | } | |
4212 | ||
3de4586c | 4213 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 4214 | { |
d397712b | 4215 | struct inode *inode; |
4df27c4d | 4216 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
4217 | struct btrfs_root *sub_root = root; |
4218 | struct btrfs_key location; | |
76dda93c | 4219 | int index; |
b4aff1f8 | 4220 | int ret = 0; |
39279cc3 CM |
4221 | |
4222 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
4223 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 4224 | |
b4aff1f8 JB |
4225 | if (unlikely(d_need_lookup(dentry))) { |
4226 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
4227 | kfree(dentry->d_fsdata); | |
4228 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
4229 | /* This thing is hashed, drop it for now */ |
4230 | d_drop(dentry); | |
b4aff1f8 JB |
4231 | } else { |
4232 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
4233 | } | |
5f39d397 | 4234 | |
39279cc3 CM |
4235 | if (ret < 0) |
4236 | return ERR_PTR(ret); | |
5f39d397 | 4237 | |
4df27c4d YZ |
4238 | if (location.objectid == 0) |
4239 | return NULL; | |
4240 | ||
4241 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 4242 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
4243 | return inode; |
4244 | } | |
4245 | ||
4246 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
4247 | ||
76dda93c | 4248 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
4249 | ret = fixup_tree_root_location(root, dir, dentry, |
4250 | &location, &sub_root); | |
4251 | if (ret < 0) { | |
4252 | if (ret != -ENOENT) | |
4253 | inode = ERR_PTR(ret); | |
4254 | else | |
4255 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
4256 | } else { | |
73f73415 | 4257 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 4258 | } |
76dda93c YZ |
4259 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
4260 | ||
34d19bad | 4261 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
4262 | down_read(&root->fs_info->cleanup_work_sem); |
4263 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 4264 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 4265 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
4266 | if (ret) |
4267 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
4268 | } |
4269 | ||
3de4586c CM |
4270 | return inode; |
4271 | } | |
4272 | ||
fe15ce44 | 4273 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
4274 | { |
4275 | struct btrfs_root *root; | |
848cce0d | 4276 | struct inode *inode = dentry->d_inode; |
76dda93c | 4277 | |
848cce0d LZ |
4278 | if (!inode && !IS_ROOT(dentry)) |
4279 | inode = dentry->d_parent->d_inode; | |
76dda93c | 4280 | |
848cce0d LZ |
4281 | if (inode) { |
4282 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
4283 | if (btrfs_root_refs(&root->root_item) == 0) |
4284 | return 1; | |
848cce0d LZ |
4285 | |
4286 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
4287 | return 1; | |
efefb143 | 4288 | } |
76dda93c YZ |
4289 | return 0; |
4290 | } | |
4291 | ||
b4aff1f8 JB |
4292 | static void btrfs_dentry_release(struct dentry *dentry) |
4293 | { | |
4294 | if (dentry->d_fsdata) | |
4295 | kfree(dentry->d_fsdata); | |
4296 | } | |
4297 | ||
3de4586c | 4298 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 4299 | unsigned int flags) |
3de4586c | 4300 | { |
a66e7cc6 JB |
4301 | struct dentry *ret; |
4302 | ||
4303 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4304 | if (unlikely(d_need_lookup(dentry))) { | |
4305 | spin_lock(&dentry->d_lock); | |
4306 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4307 | spin_unlock(&dentry->d_lock); | |
4308 | } | |
4309 | return ret; | |
39279cc3 CM |
4310 | } |
4311 | ||
16cdcec7 | 4312 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4313 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4314 | }; | |
4315 | ||
cbdf5a24 DW |
4316 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4317 | filldir_t filldir) | |
39279cc3 | 4318 | { |
6da6abae | 4319 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4320 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4321 | struct btrfs_item *item; | |
4322 | struct btrfs_dir_item *di; | |
4323 | struct btrfs_key key; | |
5f39d397 | 4324 | struct btrfs_key found_key; |
39279cc3 | 4325 | struct btrfs_path *path; |
16cdcec7 MX |
4326 | struct list_head ins_list; |
4327 | struct list_head del_list; | |
39279cc3 | 4328 | int ret; |
5f39d397 | 4329 | struct extent_buffer *leaf; |
39279cc3 | 4330 | int slot; |
39279cc3 CM |
4331 | unsigned char d_type; |
4332 | int over = 0; | |
4333 | u32 di_cur; | |
4334 | u32 di_total; | |
4335 | u32 di_len; | |
4336 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4337 | char tmp_name[32]; |
4338 | char *name_ptr; | |
4339 | int name_len; | |
16cdcec7 | 4340 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4341 | |
4342 | /* FIXME, use a real flag for deciding about the key type */ | |
4343 | if (root->fs_info->tree_root == root) | |
4344 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4345 | |
3954401f CM |
4346 | /* special case for "." */ |
4347 | if (filp->f_pos == 0) { | |
3765fefa HS |
4348 | over = filldir(dirent, ".", 1, |
4349 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4350 | if (over) |
4351 | return 0; | |
4352 | filp->f_pos = 1; | |
4353 | } | |
3954401f CM |
4354 | /* special case for .., just use the back ref */ |
4355 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4356 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4357 | over = filldir(dirent, "..", 2, |
3765fefa | 4358 | filp->f_pos, pino, DT_DIR); |
3954401f | 4359 | if (over) |
49593bfa | 4360 | return 0; |
3954401f CM |
4361 | filp->f_pos = 2; |
4362 | } | |
49593bfa | 4363 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4364 | if (!path) |
4365 | return -ENOMEM; | |
ff5714cc | 4366 | |
026fd317 | 4367 | path->reada = 1; |
49593bfa | 4368 | |
16cdcec7 MX |
4369 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4370 | INIT_LIST_HEAD(&ins_list); | |
4371 | INIT_LIST_HEAD(&del_list); | |
4372 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4373 | } | |
4374 | ||
39279cc3 CM |
4375 | btrfs_set_key_type(&key, key_type); |
4376 | key.offset = filp->f_pos; | |
33345d01 | 4377 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4378 | |
39279cc3 CM |
4379 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4380 | if (ret < 0) | |
4381 | goto err; | |
49593bfa DW |
4382 | |
4383 | while (1) { | |
5f39d397 | 4384 | leaf = path->nodes[0]; |
39279cc3 | 4385 | slot = path->slots[0]; |
b9e03af0 LZ |
4386 | if (slot >= btrfs_header_nritems(leaf)) { |
4387 | ret = btrfs_next_leaf(root, path); | |
4388 | if (ret < 0) | |
4389 | goto err; | |
4390 | else if (ret > 0) | |
4391 | break; | |
4392 | continue; | |
39279cc3 | 4393 | } |
3de4586c | 4394 | |
5f39d397 CM |
4395 | item = btrfs_item_nr(leaf, slot); |
4396 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4397 | ||
4398 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4399 | break; |
5f39d397 | 4400 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4401 | break; |
5f39d397 | 4402 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4403 | goto next; |
16cdcec7 MX |
4404 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4405 | btrfs_should_delete_dir_index(&del_list, | |
4406 | found_key.offset)) | |
4407 | goto next; | |
5f39d397 CM |
4408 | |
4409 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4410 | is_curr = 1; |
49593bfa | 4411 | |
39279cc3 CM |
4412 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4413 | di_cur = 0; | |
5f39d397 | 4414 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4415 | |
4416 | while (di_cur < di_total) { | |
5f39d397 CM |
4417 | struct btrfs_key location; |
4418 | ||
22a94d44 JB |
4419 | if (verify_dir_item(root, leaf, di)) |
4420 | break; | |
4421 | ||
5f39d397 | 4422 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4423 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4424 | name_ptr = tmp_name; |
4425 | } else { | |
4426 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4427 | if (!name_ptr) { |
4428 | ret = -ENOMEM; | |
4429 | goto err; | |
4430 | } | |
5f39d397 CM |
4431 | } |
4432 | read_extent_buffer(leaf, name_ptr, | |
4433 | (unsigned long)(di + 1), name_len); | |
4434 | ||
4435 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4436 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4437 | |
fede766f | 4438 | |
3de4586c | 4439 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
4440 | * skip it. |
4441 | * | |
4442 | * In contrast to old kernels, we insert the snapshot's | |
4443 | * dir item and dir index after it has been created, so | |
4444 | * we won't find a reference to our own snapshot. We | |
4445 | * still keep the following code for backward | |
4446 | * compatibility. | |
3de4586c CM |
4447 | */ |
4448 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4449 | location.objectid == root->root_key.objectid) { | |
4450 | over = 0; | |
4451 | goto skip; | |
4452 | } | |
5f39d397 | 4453 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4454 | found_key.offset, location.objectid, |
39279cc3 | 4455 | d_type); |
5f39d397 | 4456 | |
3de4586c | 4457 | skip: |
5f39d397 CM |
4458 | if (name_ptr != tmp_name) |
4459 | kfree(name_ptr); | |
4460 | ||
39279cc3 CM |
4461 | if (over) |
4462 | goto nopos; | |
5103e947 | 4463 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4464 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4465 | di_cur += di_len; |
4466 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4467 | } | |
b9e03af0 LZ |
4468 | next: |
4469 | path->slots[0]++; | |
39279cc3 | 4470 | } |
49593bfa | 4471 | |
16cdcec7 MX |
4472 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4473 | if (is_curr) | |
4474 | filp->f_pos++; | |
4475 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4476 | &ins_list); | |
4477 | if (ret) | |
4478 | goto nopos; | |
4479 | } | |
4480 | ||
49593bfa | 4481 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4482 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4483 | /* |
4484 | * 32-bit glibc will use getdents64, but then strtol - | |
4485 | * so the last number we can serve is this. | |
4486 | */ | |
4487 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4488 | else |
4489 | filp->f_pos++; | |
39279cc3 CM |
4490 | nopos: |
4491 | ret = 0; | |
4492 | err: | |
16cdcec7 MX |
4493 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4494 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4495 | btrfs_free_path(path); |
39279cc3 CM |
4496 | return ret; |
4497 | } | |
4498 | ||
a9185b41 | 4499 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4500 | { |
4501 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4502 | struct btrfs_trans_handle *trans; | |
4503 | int ret = 0; | |
0af3d00b | 4504 | bool nolock = false; |
39279cc3 | 4505 | |
72ac3c0d | 4506 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
4507 | return 0; |
4508 | ||
83eea1f1 | 4509 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 4510 | nolock = true; |
0af3d00b | 4511 | |
a9185b41 | 4512 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4513 | if (nolock) |
7a7eaa40 | 4514 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4515 | else |
7a7eaa40 | 4516 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4517 | if (IS_ERR(trans)) |
4518 | return PTR_ERR(trans); | |
a698d075 | 4519 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
4520 | } |
4521 | return ret; | |
4522 | } | |
4523 | ||
4524 | /* | |
54aa1f4d | 4525 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4526 | * inode changes. But, it is most likely to find the inode in cache. |
4527 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4528 | * to keep or drop this code. | |
4529 | */ | |
22c44fe6 | 4530 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4531 | { |
4532 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4533 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4534 | int ret; |
4535 | ||
72ac3c0d | 4536 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 4537 | return 0; |
39279cc3 | 4538 | |
7a7eaa40 | 4539 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4540 | if (IS_ERR(trans)) |
4541 | return PTR_ERR(trans); | |
8929ecfa YZ |
4542 | |
4543 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4544 | if (ret && ret == -ENOSPC) { |
4545 | /* whoops, lets try again with the full transaction */ | |
4546 | btrfs_end_transaction(trans, root); | |
4547 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4548 | if (IS_ERR(trans)) |
4549 | return PTR_ERR(trans); | |
8929ecfa | 4550 | |
94b60442 | 4551 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4552 | } |
39279cc3 | 4553 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4554 | if (BTRFS_I(inode)->delayed_node) |
4555 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4556 | |
4557 | return ret; | |
4558 | } | |
4559 | ||
4560 | /* | |
4561 | * This is a copy of file_update_time. We need this so we can return error on | |
4562 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4563 | */ | |
e41f941a JB |
4564 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
4565 | int flags) | |
22c44fe6 | 4566 | { |
2bc55652 AB |
4567 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4568 | ||
4569 | if (btrfs_root_readonly(root)) | |
4570 | return -EROFS; | |
4571 | ||
e41f941a | 4572 | if (flags & S_VERSION) |
22c44fe6 | 4573 | inode_inc_iversion(inode); |
e41f941a JB |
4574 | if (flags & S_CTIME) |
4575 | inode->i_ctime = *now; | |
4576 | if (flags & S_MTIME) | |
4577 | inode->i_mtime = *now; | |
4578 | if (flags & S_ATIME) | |
4579 | inode->i_atime = *now; | |
4580 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
4581 | } |
4582 | ||
d352ac68 CM |
4583 | /* |
4584 | * find the highest existing sequence number in a directory | |
4585 | * and then set the in-memory index_cnt variable to reflect | |
4586 | * free sequence numbers | |
4587 | */ | |
aec7477b JB |
4588 | static int btrfs_set_inode_index_count(struct inode *inode) |
4589 | { | |
4590 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4591 | struct btrfs_key key, found_key; | |
4592 | struct btrfs_path *path; | |
4593 | struct extent_buffer *leaf; | |
4594 | int ret; | |
4595 | ||
33345d01 | 4596 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4597 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4598 | key.offset = (u64)-1; | |
4599 | ||
4600 | path = btrfs_alloc_path(); | |
4601 | if (!path) | |
4602 | return -ENOMEM; | |
4603 | ||
4604 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4605 | if (ret < 0) | |
4606 | goto out; | |
4607 | /* FIXME: we should be able to handle this */ | |
4608 | if (ret == 0) | |
4609 | goto out; | |
4610 | ret = 0; | |
4611 | ||
4612 | /* | |
4613 | * MAGIC NUMBER EXPLANATION: | |
4614 | * since we search a directory based on f_pos we have to start at 2 | |
4615 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4616 | * else has to start at 2 | |
4617 | */ | |
4618 | if (path->slots[0] == 0) { | |
4619 | BTRFS_I(inode)->index_cnt = 2; | |
4620 | goto out; | |
4621 | } | |
4622 | ||
4623 | path->slots[0]--; | |
4624 | ||
4625 | leaf = path->nodes[0]; | |
4626 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4627 | ||
33345d01 | 4628 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4629 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4630 | BTRFS_I(inode)->index_cnt = 2; | |
4631 | goto out; | |
4632 | } | |
4633 | ||
4634 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4635 | out: | |
4636 | btrfs_free_path(path); | |
4637 | return ret; | |
4638 | } | |
4639 | ||
d352ac68 CM |
4640 | /* |
4641 | * helper to find a free sequence number in a given directory. This current | |
4642 | * code is very simple, later versions will do smarter things in the btree | |
4643 | */ | |
3de4586c | 4644 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4645 | { |
4646 | int ret = 0; | |
4647 | ||
4648 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4649 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4650 | if (ret) { | |
4651 | ret = btrfs_set_inode_index_count(dir); | |
4652 | if (ret) | |
4653 | return ret; | |
4654 | } | |
aec7477b JB |
4655 | } |
4656 | ||
00e4e6b3 | 4657 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4658 | BTRFS_I(dir)->index_cnt++; |
4659 | ||
4660 | return ret; | |
4661 | } | |
4662 | ||
39279cc3 CM |
4663 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4664 | struct btrfs_root *root, | |
aec7477b | 4665 | struct inode *dir, |
9c58309d | 4666 | const char *name, int name_len, |
175a4eb7 AV |
4667 | u64 ref_objectid, u64 objectid, |
4668 | umode_t mode, u64 *index) | |
39279cc3 CM |
4669 | { |
4670 | struct inode *inode; | |
5f39d397 | 4671 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4672 | struct btrfs_key *location; |
5f39d397 | 4673 | struct btrfs_path *path; |
9c58309d CM |
4674 | struct btrfs_inode_ref *ref; |
4675 | struct btrfs_key key[2]; | |
4676 | u32 sizes[2]; | |
4677 | unsigned long ptr; | |
39279cc3 CM |
4678 | int ret; |
4679 | int owner; | |
4680 | ||
5f39d397 | 4681 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4682 | if (!path) |
4683 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4684 | |
39279cc3 | 4685 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4686 | if (!inode) { |
4687 | btrfs_free_path(path); | |
39279cc3 | 4688 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4689 | } |
39279cc3 | 4690 | |
581bb050 LZ |
4691 | /* |
4692 | * we have to initialize this early, so we can reclaim the inode | |
4693 | * number if we fail afterwards in this function. | |
4694 | */ | |
4695 | inode->i_ino = objectid; | |
4696 | ||
aec7477b | 4697 | if (dir) { |
1abe9b8a | 4698 | trace_btrfs_inode_request(dir); |
4699 | ||
3de4586c | 4700 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4701 | if (ret) { |
8fb27640 | 4702 | btrfs_free_path(path); |
09771430 | 4703 | iput(inode); |
aec7477b | 4704 | return ERR_PTR(ret); |
09771430 | 4705 | } |
aec7477b JB |
4706 | } |
4707 | /* | |
4708 | * index_cnt is ignored for everything but a dir, | |
4709 | * btrfs_get_inode_index_count has an explanation for the magic | |
4710 | * number | |
4711 | */ | |
4712 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4713 | BTRFS_I(inode)->root = root; |
e02119d5 | 4714 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4715 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 4716 | |
5dc562c5 JB |
4717 | /* |
4718 | * We could have gotten an inode number from somebody who was fsynced | |
4719 | * and then removed in this same transaction, so let's just set full | |
4720 | * sync since it will be a full sync anyway and this will blow away the | |
4721 | * old info in the log. | |
4722 | */ | |
4723 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
4724 | ||
569254b0 | 4725 | if (S_ISDIR(mode)) |
39279cc3 CM |
4726 | owner = 0; |
4727 | else | |
4728 | owner = 1; | |
9c58309d CM |
4729 | |
4730 | key[0].objectid = objectid; | |
4731 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4732 | key[0].offset = 0; | |
4733 | ||
f186373f MF |
4734 | /* |
4735 | * Start new inodes with an inode_ref. This is slightly more | |
4736 | * efficient for small numbers of hard links since they will | |
4737 | * be packed into one item. Extended refs will kick in if we | |
4738 | * add more hard links than can fit in the ref item. | |
4739 | */ | |
9c58309d CM |
4740 | key[1].objectid = objectid; |
4741 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4742 | key[1].offset = ref_objectid; | |
4743 | ||
4744 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4745 | sizes[1] = name_len + sizeof(*ref); | |
4746 | ||
b9473439 | 4747 | path->leave_spinning = 1; |
9c58309d CM |
4748 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4749 | if (ret != 0) | |
5f39d397 CM |
4750 | goto fail; |
4751 | ||
ecc11fab | 4752 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4753 | inode_set_bytes(inode, 0); |
39279cc3 | 4754 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4755 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4756 | struct btrfs_inode_item); | |
293f7e07 LZ |
4757 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
4758 | sizeof(*inode_item)); | |
e02119d5 | 4759 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4760 | |
4761 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4762 | struct btrfs_inode_ref); | |
4763 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4764 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4765 | ptr = (unsigned long)(ref + 1); |
4766 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4767 | ||
5f39d397 CM |
4768 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4769 | btrfs_free_path(path); | |
4770 | ||
39279cc3 CM |
4771 | location = &BTRFS_I(inode)->location; |
4772 | location->objectid = objectid; | |
39279cc3 CM |
4773 | location->offset = 0; |
4774 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4775 | ||
6cbff00f CH |
4776 | btrfs_inherit_iflags(inode, dir); |
4777 | ||
569254b0 | 4778 | if (S_ISREG(mode)) { |
94272164 CM |
4779 | if (btrfs_test_opt(root, NODATASUM)) |
4780 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4781 | if (btrfs_test_opt(root, NODATACOW) || |
4782 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4783 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4784 | } | |
4785 | ||
39279cc3 | 4786 | insert_inode_hash(inode); |
5d4f98a2 | 4787 | inode_tree_add(inode); |
1abe9b8a | 4788 | |
4789 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4790 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4791 | |
8ea05e3a AB |
4792 | btrfs_update_root_times(trans, root); |
4793 | ||
39279cc3 | 4794 | return inode; |
5f39d397 | 4795 | fail: |
aec7477b JB |
4796 | if (dir) |
4797 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4798 | btrfs_free_path(path); |
09771430 | 4799 | iput(inode); |
5f39d397 | 4800 | return ERR_PTR(ret); |
39279cc3 CM |
4801 | } |
4802 | ||
4803 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4804 | { | |
4805 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4806 | } | |
4807 | ||
d352ac68 CM |
4808 | /* |
4809 | * utility function to add 'inode' into 'parent_inode' with | |
4810 | * a give name and a given sequence number. | |
4811 | * if 'add_backref' is true, also insert a backref from the | |
4812 | * inode to the parent directory. | |
4813 | */ | |
e02119d5 CM |
4814 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4815 | struct inode *parent_inode, struct inode *inode, | |
4816 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4817 | { |
4df27c4d | 4818 | int ret = 0; |
39279cc3 | 4819 | struct btrfs_key key; |
e02119d5 | 4820 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4821 | u64 ino = btrfs_ino(inode); |
4822 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4823 | |
33345d01 | 4824 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4825 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4826 | } else { | |
33345d01 | 4827 | key.objectid = ino; |
4df27c4d YZ |
4828 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4829 | key.offset = 0; | |
4830 | } | |
4831 | ||
33345d01 | 4832 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4833 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4834 | key.objectid, root->root_key.objectid, | |
33345d01 | 4835 | parent_ino, index, name, name_len); |
4df27c4d | 4836 | } else if (add_backref) { |
33345d01 LZ |
4837 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4838 | parent_ino, index); | |
4df27c4d | 4839 | } |
39279cc3 | 4840 | |
79787eaa JM |
4841 | /* Nothing to clean up yet */ |
4842 | if (ret) | |
4843 | return ret; | |
4df27c4d | 4844 | |
79787eaa JM |
4845 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
4846 | parent_inode, &key, | |
4847 | btrfs_inode_type(inode), index); | |
4848 | if (ret == -EEXIST) | |
4849 | goto fail_dir_item; | |
4850 | else if (ret) { | |
4851 | btrfs_abort_transaction(trans, root, ret); | |
4852 | return ret; | |
39279cc3 | 4853 | } |
79787eaa JM |
4854 | |
4855 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
4856 | name_len * 2); | |
0c4d2d95 | 4857 | inode_inc_iversion(parent_inode); |
79787eaa JM |
4858 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
4859 | ret = btrfs_update_inode(trans, root, parent_inode); | |
4860 | if (ret) | |
4861 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 4862 | return ret; |
fe66a05a CM |
4863 | |
4864 | fail_dir_item: | |
4865 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
4866 | u64 local_index; | |
4867 | int err; | |
4868 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4869 | key.objectid, root->root_key.objectid, | |
4870 | parent_ino, &local_index, name, name_len); | |
4871 | ||
4872 | } else if (add_backref) { | |
4873 | u64 local_index; | |
4874 | int err; | |
4875 | ||
4876 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
4877 | ino, parent_ino, &local_index); | |
4878 | } | |
4879 | return ret; | |
39279cc3 CM |
4880 | } |
4881 | ||
4882 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4883 | struct inode *dir, struct dentry *dentry, |
4884 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4885 | { |
a1b075d2 JB |
4886 | int err = btrfs_add_link(trans, dir, inode, |
4887 | dentry->d_name.name, dentry->d_name.len, | |
4888 | backref, index); | |
39279cc3 CM |
4889 | if (err > 0) |
4890 | err = -EEXIST; | |
4891 | return err; | |
4892 | } | |
4893 | ||
618e21d5 | 4894 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 4895 | umode_t mode, dev_t rdev) |
618e21d5 JB |
4896 | { |
4897 | struct btrfs_trans_handle *trans; | |
4898 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4899 | struct inode *inode = NULL; |
618e21d5 JB |
4900 | int err; |
4901 | int drop_inode = 0; | |
4902 | u64 objectid; | |
1832a6d5 | 4903 | unsigned long nr = 0; |
00e4e6b3 | 4904 | u64 index = 0; |
618e21d5 JB |
4905 | |
4906 | if (!new_valid_dev(rdev)) | |
4907 | return -EINVAL; | |
4908 | ||
9ed74f2d JB |
4909 | /* |
4910 | * 2 for inode item and ref | |
4911 | * 2 for dir items | |
4912 | * 1 for xattr if selinux is on | |
4913 | */ | |
a22285a6 YZ |
4914 | trans = btrfs_start_transaction(root, 5); |
4915 | if (IS_ERR(trans)) | |
4916 | return PTR_ERR(trans); | |
1832a6d5 | 4917 | |
581bb050 LZ |
4918 | err = btrfs_find_free_ino(root, &objectid); |
4919 | if (err) | |
4920 | goto out_unlock; | |
4921 | ||
aec7477b | 4922 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4923 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4924 | mode, &index); |
7cf96da3 TI |
4925 | if (IS_ERR(inode)) { |
4926 | err = PTR_ERR(inode); | |
618e21d5 | 4927 | goto out_unlock; |
7cf96da3 | 4928 | } |
618e21d5 | 4929 | |
2a7dba39 | 4930 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4931 | if (err) { |
4932 | drop_inode = 1; | |
4933 | goto out_unlock; | |
4934 | } | |
4935 | ||
ad19db71 CS |
4936 | /* |
4937 | * If the active LSM wants to access the inode during | |
4938 | * d_instantiate it needs these. Smack checks to see | |
4939 | * if the filesystem supports xattrs by looking at the | |
4940 | * ops vector. | |
4941 | */ | |
4942 | ||
4943 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4944 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4945 | if (err) |
4946 | drop_inode = 1; | |
4947 | else { | |
618e21d5 | 4948 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4949 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4950 | d_instantiate(dentry, inode); |
618e21d5 | 4951 | } |
618e21d5 | 4952 | out_unlock: |
d3c2fdcf | 4953 | nr = trans->blocks_used; |
7ad85bb7 | 4954 | btrfs_end_transaction(trans, root); |
a22285a6 | 4955 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4956 | if (drop_inode) { |
4957 | inode_dec_link_count(inode); | |
4958 | iput(inode); | |
4959 | } | |
618e21d5 JB |
4960 | return err; |
4961 | } | |
4962 | ||
39279cc3 | 4963 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 4964 | umode_t mode, bool excl) |
39279cc3 CM |
4965 | { |
4966 | struct btrfs_trans_handle *trans; | |
4967 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4968 | struct inode *inode = NULL; |
39279cc3 | 4969 | int drop_inode = 0; |
a22285a6 | 4970 | int err; |
1832a6d5 | 4971 | unsigned long nr = 0; |
39279cc3 | 4972 | u64 objectid; |
00e4e6b3 | 4973 | u64 index = 0; |
39279cc3 | 4974 | |
9ed74f2d JB |
4975 | /* |
4976 | * 2 for inode item and ref | |
4977 | * 2 for dir items | |
4978 | * 1 for xattr if selinux is on | |
4979 | */ | |
a22285a6 YZ |
4980 | trans = btrfs_start_transaction(root, 5); |
4981 | if (IS_ERR(trans)) | |
4982 | return PTR_ERR(trans); | |
9ed74f2d | 4983 | |
581bb050 LZ |
4984 | err = btrfs_find_free_ino(root, &objectid); |
4985 | if (err) | |
4986 | goto out_unlock; | |
4987 | ||
aec7477b | 4988 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4989 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4990 | mode, &index); |
7cf96da3 TI |
4991 | if (IS_ERR(inode)) { |
4992 | err = PTR_ERR(inode); | |
39279cc3 | 4993 | goto out_unlock; |
7cf96da3 | 4994 | } |
39279cc3 | 4995 | |
2a7dba39 | 4996 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4997 | if (err) { |
4998 | drop_inode = 1; | |
4999 | goto out_unlock; | |
5000 | } | |
5001 | ||
ad19db71 CS |
5002 | /* |
5003 | * If the active LSM wants to access the inode during | |
5004 | * d_instantiate it needs these. Smack checks to see | |
5005 | * if the filesystem supports xattrs by looking at the | |
5006 | * ops vector. | |
5007 | */ | |
5008 | inode->i_fop = &btrfs_file_operations; | |
5009 | inode->i_op = &btrfs_file_inode_operations; | |
5010 | ||
a1b075d2 | 5011 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
5012 | if (err) |
5013 | drop_inode = 1; | |
5014 | else { | |
5015 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 5016 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 5017 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
08c422c2 | 5018 | d_instantiate(dentry, inode); |
39279cc3 | 5019 | } |
39279cc3 | 5020 | out_unlock: |
d3c2fdcf | 5021 | nr = trans->blocks_used; |
7ad85bb7 | 5022 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5023 | if (drop_inode) { |
5024 | inode_dec_link_count(inode); | |
5025 | iput(inode); | |
5026 | } | |
d3c2fdcf | 5027 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5028 | return err; |
5029 | } | |
5030 | ||
5031 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
5032 | struct dentry *dentry) | |
5033 | { | |
5034 | struct btrfs_trans_handle *trans; | |
5035 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5036 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 5037 | u64 index; |
1832a6d5 | 5038 | unsigned long nr = 0; |
39279cc3 CM |
5039 | int err; |
5040 | int drop_inode = 0; | |
5041 | ||
4a8be425 TH |
5042 | /* do not allow sys_link's with other subvols of the same device */ |
5043 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 5044 | return -EXDEV; |
4a8be425 | 5045 | |
f186373f | 5046 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 5047 | return -EMLINK; |
4a8be425 | 5048 | |
3de4586c | 5049 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
5050 | if (err) |
5051 | goto fail; | |
5052 | ||
a22285a6 | 5053 | /* |
7e6b6465 | 5054 | * 2 items for inode and inode ref |
a22285a6 | 5055 | * 2 items for dir items |
7e6b6465 | 5056 | * 1 item for parent inode |
a22285a6 | 5057 | */ |
7e6b6465 | 5058 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
5059 | if (IS_ERR(trans)) { |
5060 | err = PTR_ERR(trans); | |
5061 | goto fail; | |
5062 | } | |
5f39d397 | 5063 | |
3153495d | 5064 | btrfs_inc_nlink(inode); |
0c4d2d95 | 5065 | inode_inc_iversion(inode); |
3153495d | 5066 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 5067 | ihold(inode); |
aec7477b | 5068 | |
a1b075d2 | 5069 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 5070 | |
a5719521 | 5071 | if (err) { |
54aa1f4d | 5072 | drop_inode = 1; |
a5719521 | 5073 | } else { |
10d9f309 | 5074 | struct dentry *parent = dentry->d_parent; |
a5719521 | 5075 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
5076 | if (err) |
5077 | goto fail; | |
08c422c2 | 5078 | d_instantiate(dentry, inode); |
6a912213 | 5079 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 5080 | } |
39279cc3 | 5081 | |
d3c2fdcf | 5082 | nr = trans->blocks_used; |
7ad85bb7 | 5083 | btrfs_end_transaction(trans, root); |
1832a6d5 | 5084 | fail: |
39279cc3 CM |
5085 | if (drop_inode) { |
5086 | inode_dec_link_count(inode); | |
5087 | iput(inode); | |
5088 | } | |
d3c2fdcf | 5089 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5090 | return err; |
5091 | } | |
5092 | ||
18bb1db3 | 5093 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 5094 | { |
b9d86667 | 5095 | struct inode *inode = NULL; |
39279cc3 CM |
5096 | struct btrfs_trans_handle *trans; |
5097 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
5098 | int err = 0; | |
5099 | int drop_on_err = 0; | |
b9d86667 | 5100 | u64 objectid = 0; |
00e4e6b3 | 5101 | u64 index = 0; |
d3c2fdcf | 5102 | unsigned long nr = 1; |
39279cc3 | 5103 | |
9ed74f2d JB |
5104 | /* |
5105 | * 2 items for inode and ref | |
5106 | * 2 items for dir items | |
5107 | * 1 for xattr if selinux is on | |
5108 | */ | |
a22285a6 YZ |
5109 | trans = btrfs_start_transaction(root, 5); |
5110 | if (IS_ERR(trans)) | |
5111 | return PTR_ERR(trans); | |
39279cc3 | 5112 | |
581bb050 LZ |
5113 | err = btrfs_find_free_ino(root, &objectid); |
5114 | if (err) | |
5115 | goto out_fail; | |
5116 | ||
aec7477b | 5117 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 5118 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 5119 | S_IFDIR | mode, &index); |
39279cc3 CM |
5120 | if (IS_ERR(inode)) { |
5121 | err = PTR_ERR(inode); | |
5122 | goto out_fail; | |
5123 | } | |
5f39d397 | 5124 | |
39279cc3 | 5125 | drop_on_err = 1; |
33268eaf | 5126 | |
2a7dba39 | 5127 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
5128 | if (err) |
5129 | goto out_fail; | |
5130 | ||
39279cc3 CM |
5131 | inode->i_op = &btrfs_dir_inode_operations; |
5132 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 5133 | |
dbe674a9 | 5134 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
5135 | err = btrfs_update_inode(trans, root, inode); |
5136 | if (err) | |
5137 | goto out_fail; | |
5f39d397 | 5138 | |
a1b075d2 JB |
5139 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
5140 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
5141 | if (err) |
5142 | goto out_fail; | |
5f39d397 | 5143 | |
39279cc3 CM |
5144 | d_instantiate(dentry, inode); |
5145 | drop_on_err = 0; | |
39279cc3 CM |
5146 | |
5147 | out_fail: | |
d3c2fdcf | 5148 | nr = trans->blocks_used; |
7ad85bb7 | 5149 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
5150 | if (drop_on_err) |
5151 | iput(inode); | |
d3c2fdcf | 5152 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
5153 | return err; |
5154 | } | |
5155 | ||
d352ac68 CM |
5156 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
5157 | * and an extent that you want to insert, deal with overlap and insert | |
5158 | * the new extent into the tree. | |
5159 | */ | |
3b951516 CM |
5160 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
5161 | struct extent_map *existing, | |
e6dcd2dc CM |
5162 | struct extent_map *em, |
5163 | u64 map_start, u64 map_len) | |
3b951516 CM |
5164 | { |
5165 | u64 start_diff; | |
3b951516 | 5166 | |
e6dcd2dc CM |
5167 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
5168 | start_diff = map_start - em->start; | |
5169 | em->start = map_start; | |
5170 | em->len = map_len; | |
c8b97818 CM |
5171 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
5172 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 5173 | em->block_start += start_diff; |
c8b97818 CM |
5174 | em->block_len -= start_diff; |
5175 | } | |
e6dcd2dc | 5176 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
5177 | } |
5178 | ||
c8b97818 CM |
5179 | static noinline int uncompress_inline(struct btrfs_path *path, |
5180 | struct inode *inode, struct page *page, | |
5181 | size_t pg_offset, u64 extent_offset, | |
5182 | struct btrfs_file_extent_item *item) | |
5183 | { | |
5184 | int ret; | |
5185 | struct extent_buffer *leaf = path->nodes[0]; | |
5186 | char *tmp; | |
5187 | size_t max_size; | |
5188 | unsigned long inline_size; | |
5189 | unsigned long ptr; | |
261507a0 | 5190 | int compress_type; |
c8b97818 CM |
5191 | |
5192 | WARN_ON(pg_offset != 0); | |
261507a0 | 5193 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
5194 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
5195 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
5196 | btrfs_item_nr(leaf, path->slots[0])); | |
5197 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
5198 | if (!tmp) |
5199 | return -ENOMEM; | |
c8b97818 CM |
5200 | ptr = btrfs_file_extent_inline_start(item); |
5201 | ||
5202 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
5203 | ||
5b050f04 | 5204 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
5205 | ret = btrfs_decompress(compress_type, tmp, page, |
5206 | extent_offset, inline_size, max_size); | |
c8b97818 | 5207 | if (ret) { |
7ac687d9 | 5208 | char *kaddr = kmap_atomic(page); |
c8b97818 CM |
5209 | unsigned long copy_size = min_t(u64, |
5210 | PAGE_CACHE_SIZE - pg_offset, | |
5211 | max_size - extent_offset); | |
5212 | memset(kaddr + pg_offset, 0, copy_size); | |
7ac687d9 | 5213 | kunmap_atomic(kaddr); |
c8b97818 CM |
5214 | } |
5215 | kfree(tmp); | |
5216 | return 0; | |
5217 | } | |
5218 | ||
d352ac68 CM |
5219 | /* |
5220 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
5221 | * the ugly parts come from merging extents from the disk with the in-ram |
5222 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
5223 | * where the in-ram extents might be locked pending data=ordered completion. |
5224 | * | |
5225 | * This also copies inline extents directly into the page. | |
5226 | */ | |
d397712b | 5227 | |
a52d9a80 | 5228 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 5229 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
5230 | int create) |
5231 | { | |
5232 | int ret; | |
5233 | int err = 0; | |
db94535d | 5234 | u64 bytenr; |
a52d9a80 CM |
5235 | u64 extent_start = 0; |
5236 | u64 extent_end = 0; | |
33345d01 | 5237 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 5238 | u32 found_type; |
f421950f | 5239 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
5240 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5241 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
5242 | struct extent_buffer *leaf; |
5243 | struct btrfs_key found_key; | |
a52d9a80 CM |
5244 | struct extent_map *em = NULL; |
5245 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 5246 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 5247 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 5248 | int compress_type; |
a52d9a80 | 5249 | |
a52d9a80 | 5250 | again: |
890871be | 5251 | read_lock(&em_tree->lock); |
d1310b2e | 5252 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
5253 | if (em) |
5254 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 5255 | read_unlock(&em_tree->lock); |
d1310b2e | 5256 | |
a52d9a80 | 5257 | if (em) { |
e1c4b745 CM |
5258 | if (em->start > start || em->start + em->len <= start) |
5259 | free_extent_map(em); | |
5260 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
5261 | free_extent_map(em); |
5262 | else | |
5263 | goto out; | |
a52d9a80 | 5264 | } |
172ddd60 | 5265 | em = alloc_extent_map(); |
a52d9a80 | 5266 | if (!em) { |
d1310b2e CM |
5267 | err = -ENOMEM; |
5268 | goto out; | |
a52d9a80 | 5269 | } |
e6dcd2dc | 5270 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 5271 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 5272 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 5273 | em->len = (u64)-1; |
c8b97818 | 5274 | em->block_len = (u64)-1; |
f421950f CM |
5275 | |
5276 | if (!path) { | |
5277 | path = btrfs_alloc_path(); | |
026fd317 JB |
5278 | if (!path) { |
5279 | err = -ENOMEM; | |
5280 | goto out; | |
5281 | } | |
5282 | /* | |
5283 | * Chances are we'll be called again, so go ahead and do | |
5284 | * readahead | |
5285 | */ | |
5286 | path->reada = 1; | |
f421950f CM |
5287 | } |
5288 | ||
179e29e4 CM |
5289 | ret = btrfs_lookup_file_extent(trans, root, path, |
5290 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5291 | if (ret < 0) { |
5292 | err = ret; | |
5293 | goto out; | |
5294 | } | |
5295 | ||
5296 | if (ret != 0) { | |
5297 | if (path->slots[0] == 0) | |
5298 | goto not_found; | |
5299 | path->slots[0]--; | |
5300 | } | |
5301 | ||
5f39d397 CM |
5302 | leaf = path->nodes[0]; |
5303 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5304 | struct btrfs_file_extent_item); |
a52d9a80 | 5305 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5306 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5307 | found_type = btrfs_key_type(&found_key); | |
5308 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5309 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5310 | goto not_found; | |
5311 | } | |
5312 | ||
5f39d397 CM |
5313 | found_type = btrfs_file_extent_type(leaf, item); |
5314 | extent_start = found_key.offset; | |
261507a0 | 5315 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5316 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5317 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5318 | extent_end = extent_start + |
db94535d | 5319 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5320 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5321 | size_t size; | |
5322 | size = btrfs_file_extent_inline_len(leaf, item); | |
5323 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5324 | ~((u64)root->sectorsize - 1); | |
5325 | } | |
5326 | ||
5327 | if (start >= extent_end) { | |
5328 | path->slots[0]++; | |
5329 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5330 | ret = btrfs_next_leaf(root, path); | |
5331 | if (ret < 0) { | |
5332 | err = ret; | |
5333 | goto out; | |
a52d9a80 | 5334 | } |
9036c102 YZ |
5335 | if (ret > 0) |
5336 | goto not_found; | |
5337 | leaf = path->nodes[0]; | |
a52d9a80 | 5338 | } |
9036c102 YZ |
5339 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5340 | if (found_key.objectid != objectid || | |
5341 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5342 | goto not_found; | |
5343 | if (start + len <= found_key.offset) | |
5344 | goto not_found; | |
5345 | em->start = start; | |
5346 | em->len = found_key.offset - start; | |
5347 | goto not_found_em; | |
5348 | } | |
5349 | ||
d899e052 YZ |
5350 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5351 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5352 | em->start = extent_start; |
5353 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5354 | em->orig_start = extent_start - |
5355 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5356 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5357 | if (bytenr == 0) { | |
5f39d397 | 5358 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5359 | goto insert; |
5360 | } | |
261507a0 | 5361 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5362 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5363 | em->compress_type = compress_type; |
c8b97818 CM |
5364 | em->block_start = bytenr; |
5365 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5366 | item); | |
5367 | } else { | |
5368 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5369 | em->block_start = bytenr; | |
5370 | em->block_len = em->len; | |
d899e052 YZ |
5371 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5372 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5373 | } |
a52d9a80 CM |
5374 | goto insert; |
5375 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5376 | unsigned long ptr; |
a52d9a80 | 5377 | char *map; |
3326d1b0 CM |
5378 | size_t size; |
5379 | size_t extent_offset; | |
5380 | size_t copy_size; | |
a52d9a80 | 5381 | |
689f9346 | 5382 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5383 | if (!page || create) { |
689f9346 | 5384 | em->start = extent_start; |
9036c102 | 5385 | em->len = extent_end - extent_start; |
689f9346 Y |
5386 | goto out; |
5387 | } | |
5f39d397 | 5388 | |
9036c102 YZ |
5389 | size = btrfs_file_extent_inline_len(leaf, item); |
5390 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5391 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5392 | size - extent_offset); |
3326d1b0 | 5393 | em->start = extent_start + extent_offset; |
70dec807 CM |
5394 | em->len = (copy_size + root->sectorsize - 1) & |
5395 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5396 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5397 | if (compress_type) { |
c8b97818 | 5398 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5399 | em->compress_type = compress_type; |
5400 | } | |
689f9346 | 5401 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5402 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5403 | if (btrfs_file_extent_compression(leaf, item) != |
5404 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5405 | ret = uncompress_inline(path, inode, page, |
5406 | pg_offset, | |
5407 | extent_offset, item); | |
79787eaa | 5408 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
5409 | } else { |
5410 | map = kmap(page); | |
5411 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5412 | copy_size); | |
93c82d57 CM |
5413 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5414 | memset(map + pg_offset + copy_size, 0, | |
5415 | PAGE_CACHE_SIZE - pg_offset - | |
5416 | copy_size); | |
5417 | } | |
c8b97818 CM |
5418 | kunmap(page); |
5419 | } | |
179e29e4 CM |
5420 | flush_dcache_page(page); |
5421 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5422 | BUG(); |
179e29e4 CM |
5423 | if (!trans) { |
5424 | kunmap(page); | |
5425 | free_extent_map(em); | |
5426 | em = NULL; | |
ff5714cc | 5427 | |
b3b4aa74 | 5428 | btrfs_release_path(path); |
7a7eaa40 | 5429 | trans = btrfs_join_transaction(root); |
ff5714cc | 5430 | |
3612b495 TI |
5431 | if (IS_ERR(trans)) |
5432 | return ERR_CAST(trans); | |
179e29e4 CM |
5433 | goto again; |
5434 | } | |
c8b97818 | 5435 | map = kmap(page); |
70dec807 | 5436 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5437 | copy_size); |
c8b97818 | 5438 | kunmap(page); |
179e29e4 | 5439 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5440 | } |
d1310b2e | 5441 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5442 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5443 | goto insert; |
5444 | } else { | |
d397712b | 5445 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5446 | WARN_ON(1); |
5447 | } | |
5448 | not_found: | |
5449 | em->start = start; | |
d1310b2e | 5450 | em->len = len; |
a52d9a80 | 5451 | not_found_em: |
5f39d397 | 5452 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5453 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5454 | insert: |
b3b4aa74 | 5455 | btrfs_release_path(path); |
d1310b2e | 5456 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5457 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5458 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5459 | (unsigned long long)em->len, | |
5460 | (unsigned long long)start, | |
5461 | (unsigned long long)len); | |
a52d9a80 CM |
5462 | err = -EIO; |
5463 | goto out; | |
5464 | } | |
d1310b2e CM |
5465 | |
5466 | err = 0; | |
890871be | 5467 | write_lock(&em_tree->lock); |
a52d9a80 | 5468 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5469 | /* it is possible that someone inserted the extent into the tree |
5470 | * while we had the lock dropped. It is also possible that | |
5471 | * an overlapping map exists in the tree | |
5472 | */ | |
a52d9a80 | 5473 | if (ret == -EEXIST) { |
3b951516 | 5474 | struct extent_map *existing; |
e6dcd2dc CM |
5475 | |
5476 | ret = 0; | |
5477 | ||
3b951516 | 5478 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5479 | if (existing && (existing->start > start || |
5480 | existing->start + existing->len <= start)) { | |
5481 | free_extent_map(existing); | |
5482 | existing = NULL; | |
5483 | } | |
3b951516 CM |
5484 | if (!existing) { |
5485 | existing = lookup_extent_mapping(em_tree, em->start, | |
5486 | em->len); | |
5487 | if (existing) { | |
5488 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5489 | em, start, |
5490 | root->sectorsize); | |
3b951516 CM |
5491 | free_extent_map(existing); |
5492 | if (err) { | |
5493 | free_extent_map(em); | |
5494 | em = NULL; | |
5495 | } | |
5496 | } else { | |
5497 | err = -EIO; | |
3b951516 CM |
5498 | free_extent_map(em); |
5499 | em = NULL; | |
5500 | } | |
5501 | } else { | |
5502 | free_extent_map(em); | |
5503 | em = existing; | |
e6dcd2dc | 5504 | err = 0; |
a52d9a80 | 5505 | } |
a52d9a80 | 5506 | } |
890871be | 5507 | write_unlock(&em_tree->lock); |
a52d9a80 | 5508 | out: |
1abe9b8a | 5509 | |
5510 | trace_btrfs_get_extent(root, em); | |
5511 | ||
f421950f CM |
5512 | if (path) |
5513 | btrfs_free_path(path); | |
a52d9a80 CM |
5514 | if (trans) { |
5515 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5516 | if (!err) |
a52d9a80 CM |
5517 | err = ret; |
5518 | } | |
a52d9a80 CM |
5519 | if (err) { |
5520 | free_extent_map(em); | |
a52d9a80 CM |
5521 | return ERR_PTR(err); |
5522 | } | |
79787eaa | 5523 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
5524 | return em; |
5525 | } | |
5526 | ||
ec29ed5b CM |
5527 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5528 | size_t pg_offset, u64 start, u64 len, | |
5529 | int create) | |
5530 | { | |
5531 | struct extent_map *em; | |
5532 | struct extent_map *hole_em = NULL; | |
5533 | u64 range_start = start; | |
5534 | u64 end; | |
5535 | u64 found; | |
5536 | u64 found_end; | |
5537 | int err = 0; | |
5538 | ||
5539 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5540 | if (IS_ERR(em)) | |
5541 | return em; | |
5542 | if (em) { | |
5543 | /* | |
5544 | * if our em maps to a hole, there might | |
5545 | * actually be delalloc bytes behind it | |
5546 | */ | |
5547 | if (em->block_start != EXTENT_MAP_HOLE) | |
5548 | return em; | |
5549 | else | |
5550 | hole_em = em; | |
5551 | } | |
5552 | ||
5553 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5554 | end = start + len; | |
5555 | if (end < start) | |
5556 | end = (u64)-1; | |
5557 | else | |
5558 | end -= 1; | |
5559 | ||
5560 | em = NULL; | |
5561 | ||
5562 | /* ok, we didn't find anything, lets look for delalloc */ | |
5563 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5564 | end, len, EXTENT_DELALLOC, 1); | |
5565 | found_end = range_start + found; | |
5566 | if (found_end < range_start) | |
5567 | found_end = (u64)-1; | |
5568 | ||
5569 | /* | |
5570 | * we didn't find anything useful, return | |
5571 | * the original results from get_extent() | |
5572 | */ | |
5573 | if (range_start > end || found_end <= start) { | |
5574 | em = hole_em; | |
5575 | hole_em = NULL; | |
5576 | goto out; | |
5577 | } | |
5578 | ||
5579 | /* adjust the range_start to make sure it doesn't | |
5580 | * go backwards from the start they passed in | |
5581 | */ | |
5582 | range_start = max(start,range_start); | |
5583 | found = found_end - range_start; | |
5584 | ||
5585 | if (found > 0) { | |
5586 | u64 hole_start = start; | |
5587 | u64 hole_len = len; | |
5588 | ||
172ddd60 | 5589 | em = alloc_extent_map(); |
ec29ed5b CM |
5590 | if (!em) { |
5591 | err = -ENOMEM; | |
5592 | goto out; | |
5593 | } | |
5594 | /* | |
5595 | * when btrfs_get_extent can't find anything it | |
5596 | * returns one huge hole | |
5597 | * | |
5598 | * make sure what it found really fits our range, and | |
5599 | * adjust to make sure it is based on the start from | |
5600 | * the caller | |
5601 | */ | |
5602 | if (hole_em) { | |
5603 | u64 calc_end = extent_map_end(hole_em); | |
5604 | ||
5605 | if (calc_end <= start || (hole_em->start > end)) { | |
5606 | free_extent_map(hole_em); | |
5607 | hole_em = NULL; | |
5608 | } else { | |
5609 | hole_start = max(hole_em->start, start); | |
5610 | hole_len = calc_end - hole_start; | |
5611 | } | |
5612 | } | |
5613 | em->bdev = NULL; | |
5614 | if (hole_em && range_start > hole_start) { | |
5615 | /* our hole starts before our delalloc, so we | |
5616 | * have to return just the parts of the hole | |
5617 | * that go until the delalloc starts | |
5618 | */ | |
5619 | em->len = min(hole_len, | |
5620 | range_start - hole_start); | |
5621 | em->start = hole_start; | |
5622 | em->orig_start = hole_start; | |
5623 | /* | |
5624 | * don't adjust block start at all, | |
5625 | * it is fixed at EXTENT_MAP_HOLE | |
5626 | */ | |
5627 | em->block_start = hole_em->block_start; | |
5628 | em->block_len = hole_len; | |
5629 | } else { | |
5630 | em->start = range_start; | |
5631 | em->len = found; | |
5632 | em->orig_start = range_start; | |
5633 | em->block_start = EXTENT_MAP_DELALLOC; | |
5634 | em->block_len = found; | |
5635 | } | |
5636 | } else if (hole_em) { | |
5637 | return hole_em; | |
5638 | } | |
5639 | out: | |
5640 | ||
5641 | free_extent_map(hole_em); | |
5642 | if (err) { | |
5643 | free_extent_map(em); | |
5644 | return ERR_PTR(err); | |
5645 | } | |
5646 | return em; | |
5647 | } | |
5648 | ||
4b46fce2 | 5649 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5650 | struct extent_map *em, |
4b46fce2 JB |
5651 | u64 start, u64 len) |
5652 | { | |
5653 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5654 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5655 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5656 | struct btrfs_key ins; | |
5657 | u64 alloc_hint; | |
5658 | int ret; | |
16d299ac | 5659 | bool insert = false; |
4b46fce2 | 5660 | |
16d299ac JB |
5661 | /* |
5662 | * Ok if the extent map we looked up is a hole and is for the exact | |
5663 | * range we want, there is no reason to allocate a new one, however if | |
5664 | * it is not right then we need to free this one and drop the cache for | |
5665 | * our range. | |
5666 | */ | |
5667 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5668 | em->len != len) { | |
5669 | free_extent_map(em); | |
5670 | em = NULL; | |
5671 | insert = true; | |
5672 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5673 | } | |
4b46fce2 | 5674 | |
7a7eaa40 | 5675 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5676 | if (IS_ERR(trans)) |
5677 | return ERR_CAST(trans); | |
4b46fce2 | 5678 | |
4cb5300b CM |
5679 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5680 | btrfs_add_inode_defrag(trans, inode); | |
5681 | ||
4b46fce2 JB |
5682 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5683 | ||
5684 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5685 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
81c9ad23 | 5686 | alloc_hint, &ins, 1); |
4b46fce2 JB |
5687 | if (ret) { |
5688 | em = ERR_PTR(ret); | |
5689 | goto out; | |
5690 | } | |
5691 | ||
4b46fce2 | 5692 | if (!em) { |
172ddd60 | 5693 | em = alloc_extent_map(); |
16d299ac JB |
5694 | if (!em) { |
5695 | em = ERR_PTR(-ENOMEM); | |
5696 | goto out; | |
5697 | } | |
4b46fce2 JB |
5698 | } |
5699 | ||
5700 | em->start = start; | |
5701 | em->orig_start = em->start; | |
5702 | em->len = ins.offset; | |
5703 | ||
5704 | em->block_start = ins.objectid; | |
5705 | em->block_len = ins.offset; | |
5706 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5707 | |
5708 | /* | |
5709 | * We need to do this because if we're using the original em we searched | |
5710 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5711 | */ | |
5712 | em->flags = 0; | |
4b46fce2 JB |
5713 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5714 | ||
16d299ac | 5715 | while (insert) { |
4b46fce2 JB |
5716 | write_lock(&em_tree->lock); |
5717 | ret = add_extent_mapping(em_tree, em); | |
5718 | write_unlock(&em_tree->lock); | |
5719 | if (ret != -EEXIST) | |
5720 | break; | |
5721 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5722 | } | |
5723 | ||
5724 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5725 | ins.offset, ins.offset, 0); | |
5726 | if (ret) { | |
5727 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5728 | em = ERR_PTR(ret); | |
5729 | } | |
5730 | out: | |
5731 | btrfs_end_transaction(trans, root); | |
5732 | return em; | |
5733 | } | |
5734 | ||
46bfbb5c CM |
5735 | /* |
5736 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5737 | * block must be cow'd | |
5738 | */ | |
5739 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5740 | struct inode *inode, u64 offset, u64 len) | |
5741 | { | |
5742 | struct btrfs_path *path; | |
5743 | int ret; | |
5744 | struct extent_buffer *leaf; | |
5745 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5746 | struct btrfs_file_extent_item *fi; | |
5747 | struct btrfs_key key; | |
5748 | u64 disk_bytenr; | |
5749 | u64 backref_offset; | |
5750 | u64 extent_end; | |
5751 | u64 num_bytes; | |
5752 | int slot; | |
5753 | int found_type; | |
5754 | ||
5755 | path = btrfs_alloc_path(); | |
5756 | if (!path) | |
5757 | return -ENOMEM; | |
5758 | ||
33345d01 | 5759 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5760 | offset, 0); |
5761 | if (ret < 0) | |
5762 | goto out; | |
5763 | ||
5764 | slot = path->slots[0]; | |
5765 | if (ret == 1) { | |
5766 | if (slot == 0) { | |
5767 | /* can't find the item, must cow */ | |
5768 | ret = 0; | |
5769 | goto out; | |
5770 | } | |
5771 | slot--; | |
5772 | } | |
5773 | ret = 0; | |
5774 | leaf = path->nodes[0]; | |
5775 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5776 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5777 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5778 | /* not our file or wrong item type, must cow */ | |
5779 | goto out; | |
5780 | } | |
5781 | ||
5782 | if (key.offset > offset) { | |
5783 | /* Wrong offset, must cow */ | |
5784 | goto out; | |
5785 | } | |
5786 | ||
5787 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5788 | found_type = btrfs_file_extent_type(leaf, fi); | |
5789 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5790 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5791 | /* not a regular extent, must cow */ | |
5792 | goto out; | |
5793 | } | |
5794 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5795 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5796 | ||
5797 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5798 | if (extent_end < offset + len) { | |
5799 | /* extent doesn't include our full range, must cow */ | |
5800 | goto out; | |
5801 | } | |
5802 | ||
5803 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5804 | goto out; | |
5805 | ||
5806 | /* | |
5807 | * look for other files referencing this extent, if we | |
5808 | * find any we must cow | |
5809 | */ | |
33345d01 | 5810 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5811 | key.offset - backref_offset, disk_bytenr)) |
5812 | goto out; | |
5813 | ||
5814 | /* | |
5815 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5816 | * in this extent we are about to write. If there | |
5817 | * are any csums in that range we have to cow in order | |
5818 | * to keep the csums correct | |
5819 | */ | |
5820 | disk_bytenr += backref_offset; | |
5821 | disk_bytenr += offset - key.offset; | |
5822 | num_bytes = min(offset + len, extent_end) - offset; | |
5823 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5824 | goto out; | |
5825 | /* | |
5826 | * all of the above have passed, it is safe to overwrite this extent | |
5827 | * without cow | |
5828 | */ | |
5829 | ret = 1; | |
5830 | out: | |
5831 | btrfs_free_path(path); | |
5832 | return ret; | |
5833 | } | |
5834 | ||
eb838e73 JB |
5835 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
5836 | struct extent_state **cached_state, int writing) | |
5837 | { | |
5838 | struct btrfs_ordered_extent *ordered; | |
5839 | int ret = 0; | |
5840 | ||
5841 | while (1) { | |
5842 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5843 | 0, cached_state); | |
5844 | /* | |
5845 | * We're concerned with the entire range that we're going to be | |
5846 | * doing DIO to, so we need to make sure theres no ordered | |
5847 | * extents in this range. | |
5848 | */ | |
5849 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
5850 | lockend - lockstart + 1); | |
5851 | ||
5852 | /* | |
5853 | * We need to make sure there are no buffered pages in this | |
5854 | * range either, we could have raced between the invalidate in | |
5855 | * generic_file_direct_write and locking the extent. The | |
5856 | * invalidate needs to happen so that reads after a write do not | |
5857 | * get stale data. | |
5858 | */ | |
5859 | if (!ordered && (!writing || | |
5860 | !test_range_bit(&BTRFS_I(inode)->io_tree, | |
5861 | lockstart, lockend, EXTENT_UPTODATE, 0, | |
5862 | *cached_state))) | |
5863 | break; | |
5864 | ||
5865 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
5866 | cached_state, GFP_NOFS); | |
5867 | ||
5868 | if (ordered) { | |
5869 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5870 | btrfs_put_ordered_extent(ordered); | |
5871 | } else { | |
5872 | /* Screw you mmap */ | |
5873 | ret = filemap_write_and_wait_range(inode->i_mapping, | |
5874 | lockstart, | |
5875 | lockend); | |
5876 | if (ret) | |
5877 | break; | |
5878 | ||
5879 | /* | |
5880 | * If we found a page that couldn't be invalidated just | |
5881 | * fall back to buffered. | |
5882 | */ | |
5883 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5884 | lockstart >> PAGE_CACHE_SHIFT, | |
5885 | lockend >> PAGE_CACHE_SHIFT); | |
5886 | if (ret) | |
5887 | break; | |
5888 | } | |
5889 | ||
5890 | cond_resched(); | |
5891 | } | |
5892 | ||
5893 | return ret; | |
5894 | } | |
5895 | ||
69ffb543 JB |
5896 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
5897 | u64 len, u64 orig_start, | |
5898 | u64 block_start, u64 block_len, | |
5899 | int type) | |
5900 | { | |
5901 | struct extent_map_tree *em_tree; | |
5902 | struct extent_map *em; | |
5903 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5904 | int ret; | |
5905 | ||
5906 | em_tree = &BTRFS_I(inode)->extent_tree; | |
5907 | em = alloc_extent_map(); | |
5908 | if (!em) | |
5909 | return ERR_PTR(-ENOMEM); | |
5910 | ||
5911 | em->start = start; | |
5912 | em->orig_start = orig_start; | |
5913 | em->len = len; | |
5914 | em->block_len = block_len; | |
5915 | em->block_start = block_start; | |
5916 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5917 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5918 | if (type == BTRFS_ORDERED_PREALLOC) | |
5919 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
5920 | ||
5921 | do { | |
5922 | btrfs_drop_extent_cache(inode, em->start, | |
5923 | em->start + em->len - 1, 0); | |
5924 | write_lock(&em_tree->lock); | |
5925 | ret = add_extent_mapping(em_tree, em); | |
5926 | write_unlock(&em_tree->lock); | |
5927 | } while (ret == -EEXIST); | |
5928 | ||
5929 | if (ret) { | |
5930 | free_extent_map(em); | |
5931 | return ERR_PTR(ret); | |
5932 | } | |
5933 | ||
5934 | return em; | |
5935 | } | |
5936 | ||
5937 | ||
4b46fce2 JB |
5938 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5939 | struct buffer_head *bh_result, int create) | |
5940 | { | |
5941 | struct extent_map *em; | |
5942 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 5943 | struct extent_state *cached_state = NULL; |
4b46fce2 | 5944 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 5945 | u64 lockstart, lockend; |
4b46fce2 | 5946 | u64 len = bh_result->b_size; |
46bfbb5c | 5947 | struct btrfs_trans_handle *trans; |
eb838e73 JB |
5948 | int unlock_bits = EXTENT_LOCKED; |
5949 | int ret; | |
5950 | ||
eb838e73 JB |
5951 | if (create) { |
5952 | ret = btrfs_delalloc_reserve_space(inode, len); | |
5953 | if (ret) | |
5954 | return ret; | |
5955 | unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY; | |
c329861d JB |
5956 | } else { |
5957 | len = min_t(u64, len, root->sectorsize); | |
eb838e73 JB |
5958 | } |
5959 | ||
c329861d JB |
5960 | lockstart = start; |
5961 | lockend = start + len - 1; | |
5962 | ||
eb838e73 JB |
5963 | /* |
5964 | * If this errors out it's because we couldn't invalidate pagecache for | |
5965 | * this range and we need to fallback to buffered. | |
5966 | */ | |
5967 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
5968 | return -ENOTBLK; | |
5969 | ||
5970 | if (create) { | |
5971 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
5972 | lockend, EXTENT_DELALLOC, NULL, | |
5973 | &cached_state, GFP_NOFS); | |
5974 | if (ret) | |
5975 | goto unlock_err; | |
5976 | } | |
4b46fce2 JB |
5977 | |
5978 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
eb838e73 JB |
5979 | if (IS_ERR(em)) { |
5980 | ret = PTR_ERR(em); | |
5981 | goto unlock_err; | |
5982 | } | |
4b46fce2 JB |
5983 | |
5984 | /* | |
5985 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5986 | * io. INLINE is special, and we could probably kludge it in here, but | |
5987 | * it's still buffered so for safety lets just fall back to the generic | |
5988 | * buffered path. | |
5989 | * | |
5990 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5991 | * decompress it, so there will be buffering required no matter what we | |
5992 | * do, so go ahead and fallback to buffered. | |
5993 | * | |
5994 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5995 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5996 | * the generic code. | |
5997 | */ | |
5998 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5999 | em->block_start == EXTENT_MAP_INLINE) { | |
6000 | free_extent_map(em); | |
eb838e73 JB |
6001 | ret = -ENOTBLK; |
6002 | goto unlock_err; | |
4b46fce2 JB |
6003 | } |
6004 | ||
6005 | /* Just a good old fashioned hole, return */ | |
6006 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
6007 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
6008 | free_extent_map(em); | |
eb838e73 JB |
6009 | ret = 0; |
6010 | goto unlock_err; | |
4b46fce2 JB |
6011 | } |
6012 | ||
6013 | /* | |
6014 | * We don't allocate a new extent in the following cases | |
6015 | * | |
6016 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
6017 | * existing extent. | |
6018 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
6019 | * just use the extent. | |
6020 | * | |
6021 | */ | |
46bfbb5c | 6022 | if (!create) { |
eb838e73 JB |
6023 | len = min(len, em->len - (start - em->start)); |
6024 | lockstart = start + len; | |
6025 | goto unlock; | |
46bfbb5c | 6026 | } |
4b46fce2 JB |
6027 | |
6028 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
6029 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
6030 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
6031 | int type; |
6032 | int ret; | |
46bfbb5c | 6033 | u64 block_start; |
4b46fce2 JB |
6034 | |
6035 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6036 | type = BTRFS_ORDERED_PREALLOC; | |
6037 | else | |
6038 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 6039 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 6040 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
6041 | |
6042 | /* | |
6043 | * we're not going to log anything, but we do need | |
6044 | * to make sure the current transaction stays open | |
6045 | * while we look for nocow cross refs | |
6046 | */ | |
7a7eaa40 | 6047 | trans = btrfs_join_transaction(root); |
3612b495 | 6048 | if (IS_ERR(trans)) |
46bfbb5c CM |
6049 | goto must_cow; |
6050 | ||
6051 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
69ffb543 JB |
6052 | u64 orig_start = em->start; |
6053 | ||
6054 | if (type == BTRFS_ORDERED_PREALLOC) { | |
6055 | free_extent_map(em); | |
6056 | em = create_pinned_em(inode, start, len, | |
6057 | orig_start, | |
6058 | block_start, len, type); | |
6059 | if (IS_ERR(em)) { | |
6060 | btrfs_end_transaction(trans, root); | |
6061 | goto unlock_err; | |
6062 | } | |
6063 | } | |
6064 | ||
46bfbb5c CM |
6065 | ret = btrfs_add_ordered_extent_dio(inode, start, |
6066 | block_start, len, len, type); | |
6067 | btrfs_end_transaction(trans, root); | |
6068 | if (ret) { | |
6069 | free_extent_map(em); | |
eb838e73 | 6070 | goto unlock_err; |
46bfbb5c CM |
6071 | } |
6072 | goto unlock; | |
4b46fce2 | 6073 | } |
46bfbb5c | 6074 | btrfs_end_transaction(trans, root); |
4b46fce2 | 6075 | } |
46bfbb5c CM |
6076 | must_cow: |
6077 | /* | |
6078 | * this will cow the extent, reset the len in case we changed | |
6079 | * it above | |
6080 | */ | |
6081 | len = bh_result->b_size; | |
16d299ac | 6082 | em = btrfs_new_extent_direct(inode, em, start, len); |
eb838e73 JB |
6083 | if (IS_ERR(em)) { |
6084 | ret = PTR_ERR(em); | |
6085 | goto unlock_err; | |
6086 | } | |
46bfbb5c CM |
6087 | len = min(len, em->len - (start - em->start)); |
6088 | unlock: | |
4b46fce2 JB |
6089 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
6090 | inode->i_blkbits; | |
46bfbb5c | 6091 | bh_result->b_size = len; |
4b46fce2 JB |
6092 | bh_result->b_bdev = em->bdev; |
6093 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
6094 | if (create) { |
6095 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
6096 | set_buffer_new(bh_result); | |
6097 | ||
6098 | /* | |
6099 | * Need to update the i_size under the extent lock so buffered | |
6100 | * readers will get the updated i_size when we unlock. | |
6101 | */ | |
6102 | if (start + len > i_size_read(inode)) | |
6103 | i_size_write(inode, start + len); | |
6104 | } | |
4b46fce2 | 6105 | |
eb838e73 JB |
6106 | /* |
6107 | * In the case of write we need to clear and unlock the entire range, | |
6108 | * in the case of read we need to unlock only the end area that we | |
6109 | * aren't using if there is any left over space. | |
6110 | */ | |
24c03fa5 LB |
6111 | if (lockstart < lockend) { |
6112 | if (create && len < lockend - lockstart) { | |
6113 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
9e8a4a8b LB |
6114 | lockstart + len - 1, |
6115 | unlock_bits | EXTENT_DEFRAG, 1, 0, | |
24c03fa5 LB |
6116 | &cached_state, GFP_NOFS); |
6117 | /* | |
6118 | * Beside unlock, we also need to cleanup reserved space | |
6119 | * for the left range by attaching EXTENT_DO_ACCOUNTING. | |
6120 | */ | |
6121 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
6122 | lockstart + len, lockend, | |
9e8a4a8b LB |
6123 | unlock_bits | EXTENT_DO_ACCOUNTING | |
6124 | EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS); | |
24c03fa5 LB |
6125 | } else { |
6126 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6127 | lockend, unlock_bits, 1, 0, | |
6128 | &cached_state, GFP_NOFS); | |
6129 | } | |
6130 | } else { | |
eb838e73 | 6131 | free_extent_state(cached_state); |
24c03fa5 | 6132 | } |
eb838e73 | 6133 | |
4b46fce2 JB |
6134 | free_extent_map(em); |
6135 | ||
6136 | return 0; | |
eb838e73 JB |
6137 | |
6138 | unlock_err: | |
6139 | if (create) | |
6140 | unlock_bits |= EXTENT_DO_ACCOUNTING; | |
6141 | ||
6142 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6143 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
6144 | return ret; | |
4b46fce2 JB |
6145 | } |
6146 | ||
6147 | struct btrfs_dio_private { | |
6148 | struct inode *inode; | |
6149 | u64 logical_offset; | |
6150 | u64 disk_bytenr; | |
6151 | u64 bytes; | |
4b46fce2 | 6152 | void *private; |
e65e1535 MX |
6153 | |
6154 | /* number of bios pending for this dio */ | |
6155 | atomic_t pending_bios; | |
6156 | ||
6157 | /* IO errors */ | |
6158 | int errors; | |
6159 | ||
6160 | struct bio *orig_bio; | |
4b46fce2 JB |
6161 | }; |
6162 | ||
6163 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
6164 | { | |
e65e1535 | 6165 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
6166 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
6167 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
6168 | struct inode *inode = dip->inode; |
6169 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6170 | u64 start; | |
4b46fce2 JB |
6171 | |
6172 | start = dip->logical_offset; | |
6173 | do { | |
6174 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
6175 | struct page *page = bvec->bv_page; | |
6176 | char *kaddr; | |
6177 | u32 csum = ~(u32)0; | |
c329861d | 6178 | u64 private = ~(u32)0; |
4b46fce2 JB |
6179 | unsigned long flags; |
6180 | ||
c329861d JB |
6181 | if (get_state_private(&BTRFS_I(inode)->io_tree, |
6182 | start, &private)) | |
6183 | goto failed; | |
4b46fce2 | 6184 | local_irq_save(flags); |
7ac687d9 | 6185 | kaddr = kmap_atomic(page); |
4b46fce2 JB |
6186 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, |
6187 | csum, bvec->bv_len); | |
6188 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 6189 | kunmap_atomic(kaddr); |
4b46fce2 JB |
6190 | local_irq_restore(flags); |
6191 | ||
6192 | flush_dcache_page(bvec->bv_page); | |
c329861d JB |
6193 | if (csum != private) { |
6194 | failed: | |
33345d01 | 6195 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 6196 | " %llu csum %u private %u\n", |
33345d01 LZ |
6197 | (unsigned long long)btrfs_ino(inode), |
6198 | (unsigned long long)start, | |
c329861d | 6199 | csum, (unsigned)private); |
4b46fce2 JB |
6200 | err = -EIO; |
6201 | } | |
6202 | } | |
6203 | ||
6204 | start += bvec->bv_len; | |
4b46fce2 JB |
6205 | bvec++; |
6206 | } while (bvec <= bvec_end); | |
6207 | ||
6208 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
d0082371 | 6209 | dip->logical_offset + dip->bytes - 1); |
4b46fce2 JB |
6210 | bio->bi_private = dip->private; |
6211 | ||
4b46fce2 | 6212 | kfree(dip); |
c0da7aa1 JB |
6213 | |
6214 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
6215 | if (err) | |
6216 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6217 | dio_end_io(bio, err); |
6218 | } | |
6219 | ||
6220 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
6221 | { | |
6222 | struct btrfs_dio_private *dip = bio->bi_private; | |
6223 | struct inode *inode = dip->inode; | |
6224 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 6225 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
6226 | u64 ordered_offset = dip->logical_offset; |
6227 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
6228 | int ret; |
6229 | ||
6230 | if (err) | |
6231 | goto out_done; | |
163cf09c CM |
6232 | again: |
6233 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
6234 | &ordered_offset, | |
5fd02043 | 6235 | ordered_bytes, !err); |
4b46fce2 | 6236 | if (!ret) |
163cf09c | 6237 | goto out_test; |
4b46fce2 | 6238 | |
5fd02043 JB |
6239 | ordered->work.func = finish_ordered_fn; |
6240 | ordered->work.flags = 0; | |
6241 | btrfs_queue_worker(&root->fs_info->endio_write_workers, | |
6242 | &ordered->work); | |
163cf09c CM |
6243 | out_test: |
6244 | /* | |
6245 | * our bio might span multiple ordered extents. If we haven't | |
6246 | * completed the accounting for the whole dio, go back and try again | |
6247 | */ | |
6248 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
6249 | ordered_bytes = dip->logical_offset + dip->bytes - | |
6250 | ordered_offset; | |
5fd02043 | 6251 | ordered = NULL; |
163cf09c CM |
6252 | goto again; |
6253 | } | |
4b46fce2 JB |
6254 | out_done: |
6255 | bio->bi_private = dip->private; | |
6256 | ||
4b46fce2 | 6257 | kfree(dip); |
c0da7aa1 JB |
6258 | |
6259 | /* If we had an error make sure to clear the uptodate flag */ | |
6260 | if (err) | |
6261 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
6262 | dio_end_io(bio, err); |
6263 | } | |
6264 | ||
eaf25d93 CM |
6265 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
6266 | struct bio *bio, int mirror_num, | |
6267 | unsigned long bio_flags, u64 offset) | |
6268 | { | |
6269 | int ret; | |
6270 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6271 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 6272 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
6273 | return 0; |
6274 | } | |
6275 | ||
e65e1535 MX |
6276 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
6277 | { | |
6278 | struct btrfs_dio_private *dip = bio->bi_private; | |
6279 | ||
6280 | if (err) { | |
33345d01 | 6281 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 6282 | "sector %#Lx len %u err no %d\n", |
33345d01 | 6283 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 6284 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
6285 | dip->errors = 1; |
6286 | ||
6287 | /* | |
6288 | * before atomic variable goto zero, we must make sure | |
6289 | * dip->errors is perceived to be set. | |
6290 | */ | |
6291 | smp_mb__before_atomic_dec(); | |
6292 | } | |
6293 | ||
6294 | /* if there are more bios still pending for this dio, just exit */ | |
6295 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
6296 | goto out; | |
6297 | ||
6298 | if (dip->errors) | |
6299 | bio_io_error(dip->orig_bio); | |
6300 | else { | |
6301 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
6302 | bio_endio(dip->orig_bio, 0); | |
6303 | } | |
6304 | out: | |
6305 | bio_put(bio); | |
6306 | } | |
6307 | ||
6308 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
6309 | u64 first_sector, gfp_t gfp_flags) | |
6310 | { | |
6311 | int nr_vecs = bio_get_nr_vecs(bdev); | |
6312 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
6313 | } | |
6314 | ||
6315 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
6316 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 6317 | int async_submit) |
e65e1535 MX |
6318 | { |
6319 | int write = rw & REQ_WRITE; | |
6320 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6321 | int ret; | |
6322 | ||
6323 | bio_get(bio); | |
5fd02043 JB |
6324 | |
6325 | if (!write) { | |
6326 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
6327 | if (ret) | |
6328 | goto err; | |
6329 | } | |
e65e1535 | 6330 | |
1ae39938 JB |
6331 | if (skip_sum) |
6332 | goto map; | |
6333 | ||
6334 | if (write && async_submit) { | |
e65e1535 MX |
6335 | ret = btrfs_wq_submit_bio(root->fs_info, |
6336 | inode, rw, bio, 0, 0, | |
6337 | file_offset, | |
6338 | __btrfs_submit_bio_start_direct_io, | |
6339 | __btrfs_submit_bio_done); | |
6340 | goto err; | |
1ae39938 JB |
6341 | } else if (write) { |
6342 | /* | |
6343 | * If we aren't doing async submit, calculate the csum of the | |
6344 | * bio now. | |
6345 | */ | |
6346 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
6347 | if (ret) | |
6348 | goto err; | |
c2db1073 | 6349 | } else if (!skip_sum) { |
c329861d | 6350 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset); |
c2db1073 TI |
6351 | if (ret) |
6352 | goto err; | |
6353 | } | |
e65e1535 | 6354 | |
1ae39938 JB |
6355 | map: |
6356 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
6357 | err: |
6358 | bio_put(bio); | |
6359 | return ret; | |
6360 | } | |
6361 | ||
6362 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
6363 | int skip_sum) | |
6364 | { | |
6365 | struct inode *inode = dip->inode; | |
6366 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6367 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6368 | struct bio *bio; | |
6369 | struct bio *orig_bio = dip->orig_bio; | |
6370 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
6371 | u64 start_sector = orig_bio->bi_sector; | |
6372 | u64 file_offset = dip->logical_offset; | |
6373 | u64 submit_len = 0; | |
6374 | u64 map_length; | |
6375 | int nr_pages = 0; | |
e65e1535 | 6376 | int ret = 0; |
1ae39938 | 6377 | int async_submit = 0; |
e65e1535 | 6378 | |
e65e1535 MX |
6379 | map_length = orig_bio->bi_size; |
6380 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6381 | &map_length, NULL, 0); | |
6382 | if (ret) { | |
64728bbb | 6383 | bio_put(orig_bio); |
e65e1535 MX |
6384 | return -EIO; |
6385 | } | |
6386 | ||
02f57c7a JB |
6387 | if (map_length >= orig_bio->bi_size) { |
6388 | bio = orig_bio; | |
6389 | goto submit; | |
6390 | } | |
6391 | ||
1ae39938 | 6392 | async_submit = 1; |
02f57c7a JB |
6393 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
6394 | if (!bio) | |
6395 | return -ENOMEM; | |
6396 | bio->bi_private = dip; | |
6397 | bio->bi_end_io = btrfs_end_dio_bio; | |
6398 | atomic_inc(&dip->pending_bios); | |
6399 | ||
e65e1535 MX |
6400 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
6401 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
6402 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
6403 | bvec->bv_offset) < bvec->bv_len)) { | |
6404 | /* | |
6405 | * inc the count before we submit the bio so | |
6406 | * we know the end IO handler won't happen before | |
6407 | * we inc the count. Otherwise, the dip might get freed | |
6408 | * before we're done setting it up | |
6409 | */ | |
6410 | atomic_inc(&dip->pending_bios); | |
6411 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
6412 | file_offset, skip_sum, | |
c329861d | 6413 | async_submit); |
e65e1535 MX |
6414 | if (ret) { |
6415 | bio_put(bio); | |
6416 | atomic_dec(&dip->pending_bios); | |
6417 | goto out_err; | |
6418 | } | |
6419 | ||
e65e1535 MX |
6420 | start_sector += submit_len >> 9; |
6421 | file_offset += submit_len; | |
6422 | ||
6423 | submit_len = 0; | |
6424 | nr_pages = 0; | |
6425 | ||
6426 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6427 | start_sector, GFP_NOFS); | |
6428 | if (!bio) | |
6429 | goto out_err; | |
6430 | bio->bi_private = dip; | |
6431 | bio->bi_end_io = btrfs_end_dio_bio; | |
6432 | ||
6433 | map_length = orig_bio->bi_size; | |
6434 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6435 | &map_length, NULL, 0); | |
6436 | if (ret) { | |
6437 | bio_put(bio); | |
6438 | goto out_err; | |
6439 | } | |
6440 | } else { | |
6441 | submit_len += bvec->bv_len; | |
6442 | nr_pages ++; | |
6443 | bvec++; | |
6444 | } | |
6445 | } | |
6446 | ||
02f57c7a | 6447 | submit: |
e65e1535 | 6448 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 6449 | async_submit); |
e65e1535 MX |
6450 | if (!ret) |
6451 | return 0; | |
6452 | ||
6453 | bio_put(bio); | |
6454 | out_err: | |
6455 | dip->errors = 1; | |
6456 | /* | |
6457 | * before atomic variable goto zero, we must | |
6458 | * make sure dip->errors is perceived to be set. | |
6459 | */ | |
6460 | smp_mb__before_atomic_dec(); | |
6461 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6462 | bio_io_error(dip->orig_bio); | |
6463 | ||
6464 | /* bio_end_io() will handle error, so we needn't return it */ | |
6465 | return 0; | |
6466 | } | |
6467 | ||
4b46fce2 JB |
6468 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6469 | loff_t file_offset) | |
6470 | { | |
6471 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6472 | struct btrfs_dio_private *dip; | |
6473 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6474 | int skip_sum; |
7b6d91da | 6475 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6476 | int ret = 0; |
6477 | ||
6478 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6479 | ||
6480 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6481 | if (!dip) { | |
6482 | ret = -ENOMEM; | |
6483 | goto free_ordered; | |
6484 | } | |
4b46fce2 JB |
6485 | |
6486 | dip->private = bio->bi_private; | |
6487 | dip->inode = inode; | |
6488 | dip->logical_offset = file_offset; | |
6489 | ||
4b46fce2 JB |
6490 | dip->bytes = 0; |
6491 | do { | |
6492 | dip->bytes += bvec->bv_len; | |
6493 | bvec++; | |
6494 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6495 | ||
46bfbb5c | 6496 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6497 | bio->bi_private = dip; |
e65e1535 MX |
6498 | dip->errors = 0; |
6499 | dip->orig_bio = bio; | |
6500 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6501 | |
6502 | if (write) | |
6503 | bio->bi_end_io = btrfs_endio_direct_write; | |
6504 | else | |
6505 | bio->bi_end_io = btrfs_endio_direct_read; | |
6506 | ||
e65e1535 MX |
6507 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6508 | if (!ret) | |
eaf25d93 | 6509 | return; |
4b46fce2 JB |
6510 | free_ordered: |
6511 | /* | |
6512 | * If this is a write, we need to clean up the reserved space and kill | |
6513 | * the ordered extent. | |
6514 | */ | |
6515 | if (write) { | |
6516 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6517 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6518 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6519 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6520 | btrfs_free_reserved_extent(root, ordered->start, | |
6521 | ordered->disk_len); | |
6522 | btrfs_put_ordered_extent(ordered); | |
6523 | btrfs_put_ordered_extent(ordered); | |
6524 | } | |
6525 | bio_endio(bio, ret); | |
6526 | } | |
6527 | ||
5a5f79b5 CM |
6528 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6529 | const struct iovec *iov, loff_t offset, | |
6530 | unsigned long nr_segs) | |
6531 | { | |
6532 | int seg; | |
a1b75f7d | 6533 | int i; |
5a5f79b5 CM |
6534 | size_t size; |
6535 | unsigned long addr; | |
6536 | unsigned blocksize_mask = root->sectorsize - 1; | |
6537 | ssize_t retval = -EINVAL; | |
6538 | loff_t end = offset; | |
6539 | ||
6540 | if (offset & blocksize_mask) | |
6541 | goto out; | |
6542 | ||
6543 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6544 | for (seg = 0; seg < nr_segs; seg++) { | |
6545 | addr = (unsigned long)iov[seg].iov_base; | |
6546 | size = iov[seg].iov_len; | |
6547 | end += size; | |
a1b75f7d | 6548 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6549 | goto out; |
a1b75f7d JB |
6550 | |
6551 | /* If this is a write we don't need to check anymore */ | |
6552 | if (rw & WRITE) | |
6553 | continue; | |
6554 | ||
6555 | /* | |
6556 | * Check to make sure we don't have duplicate iov_base's in this | |
6557 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6558 | * when reading back. | |
6559 | */ | |
6560 | for (i = seg + 1; i < nr_segs; i++) { | |
6561 | if (iov[seg].iov_base == iov[i].iov_base) | |
6562 | goto out; | |
6563 | } | |
5a5f79b5 CM |
6564 | } |
6565 | retval = 0; | |
6566 | out: | |
6567 | return retval; | |
6568 | } | |
eb838e73 | 6569 | |
16432985 CM |
6570 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6571 | const struct iovec *iov, loff_t offset, | |
6572 | unsigned long nr_segs) | |
6573 | { | |
4b46fce2 JB |
6574 | struct file *file = iocb->ki_filp; |
6575 | struct inode *inode = file->f_mapping->host; | |
4b46fce2 | 6576 | |
5a5f79b5 | 6577 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
eb838e73 | 6578 | offset, nr_segs)) |
5a5f79b5 | 6579 | return 0; |
3f7c579c | 6580 | |
eb838e73 | 6581 | return __blockdev_direct_IO(rw, iocb, inode, |
5a5f79b5 CM |
6582 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, |
6583 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6584 | btrfs_submit_direct, 0); | |
16432985 CM |
6585 | } |
6586 | ||
1506fcc8 YS |
6587 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6588 | __u64 start, __u64 len) | |
6589 | { | |
ec29ed5b | 6590 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6591 | } |
6592 | ||
a52d9a80 | 6593 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6594 | { |
d1310b2e CM |
6595 | struct extent_io_tree *tree; |
6596 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6597 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6598 | } |
1832a6d5 | 6599 | |
a52d9a80 | 6600 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6601 | { |
d1310b2e | 6602 | struct extent_io_tree *tree; |
b888db2b CM |
6603 | |
6604 | ||
6605 | if (current->flags & PF_MEMALLOC) { | |
6606 | redirty_page_for_writepage(wbc, page); | |
6607 | unlock_page(page); | |
6608 | return 0; | |
6609 | } | |
d1310b2e | 6610 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6611 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6612 | } |
6613 | ||
f421950f CM |
6614 | int btrfs_writepages(struct address_space *mapping, |
6615 | struct writeback_control *wbc) | |
b293f02e | 6616 | { |
d1310b2e | 6617 | struct extent_io_tree *tree; |
771ed689 | 6618 | |
d1310b2e | 6619 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6620 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6621 | } | |
6622 | ||
3ab2fb5a CM |
6623 | static int |
6624 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6625 | struct list_head *pages, unsigned nr_pages) | |
6626 | { | |
d1310b2e CM |
6627 | struct extent_io_tree *tree; |
6628 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6629 | return extent_readpages(tree, mapping, pages, nr_pages, |
6630 | btrfs_get_extent); | |
6631 | } | |
e6dcd2dc | 6632 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6633 | { |
d1310b2e CM |
6634 | struct extent_io_tree *tree; |
6635 | struct extent_map_tree *map; | |
a52d9a80 | 6636 | int ret; |
8c2383c3 | 6637 | |
d1310b2e CM |
6638 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6639 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6640 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6641 | if (ret == 1) { |
6642 | ClearPagePrivate(page); | |
6643 | set_page_private(page, 0); | |
6644 | page_cache_release(page); | |
39279cc3 | 6645 | } |
a52d9a80 | 6646 | return ret; |
39279cc3 CM |
6647 | } |
6648 | ||
e6dcd2dc CM |
6649 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6650 | { | |
98509cfc CM |
6651 | if (PageWriteback(page) || PageDirty(page)) |
6652 | return 0; | |
b335b003 | 6653 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6654 | } |
6655 | ||
a52d9a80 | 6656 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6657 | { |
5fd02043 | 6658 | struct inode *inode = page->mapping->host; |
d1310b2e | 6659 | struct extent_io_tree *tree; |
e6dcd2dc | 6660 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6661 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6662 | u64 page_start = page_offset(page); |
6663 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6664 | |
8b62b72b CM |
6665 | /* |
6666 | * we have the page locked, so new writeback can't start, | |
6667 | * and the dirty bit won't be cleared while we are here. | |
6668 | * | |
6669 | * Wait for IO on this page so that we can safely clear | |
6670 | * the PagePrivate2 bit and do ordered accounting | |
6671 | */ | |
e6dcd2dc | 6672 | wait_on_page_writeback(page); |
8b62b72b | 6673 | |
5fd02043 | 6674 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
6675 | if (offset) { |
6676 | btrfs_releasepage(page, GFP_NOFS); | |
6677 | return; | |
6678 | } | |
d0082371 | 6679 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
5fd02043 | 6680 | ordered = btrfs_lookup_ordered_extent(inode, |
e6dcd2dc CM |
6681 | page_offset(page)); |
6682 | if (ordered) { | |
eb84ae03 CM |
6683 | /* |
6684 | * IO on this page will never be started, so we need | |
6685 | * to account for any ordered extents now | |
6686 | */ | |
e6dcd2dc CM |
6687 | clear_extent_bit(tree, page_start, page_end, |
6688 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
9e8a4a8b LB |
6689 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
6690 | EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); | |
8b62b72b CM |
6691 | /* |
6692 | * whoever cleared the private bit is responsible | |
6693 | * for the finish_ordered_io | |
6694 | */ | |
5fd02043 JB |
6695 | if (TestClearPagePrivate2(page) && |
6696 | btrfs_dec_test_ordered_pending(inode, &ordered, page_start, | |
6697 | PAGE_CACHE_SIZE, 1)) { | |
6698 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 6699 | } |
e6dcd2dc | 6700 | btrfs_put_ordered_extent(ordered); |
2ac55d41 | 6701 | cached_state = NULL; |
d0082371 | 6702 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6703 | } |
6704 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6705 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
9e8a4a8b LB |
6706 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, |
6707 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
6708 | __btrfs_releasepage(page, GFP_NOFS); |
6709 | ||
4a096752 | 6710 | ClearPageChecked(page); |
9ad6b7bc | 6711 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6712 | ClearPagePrivate(page); |
6713 | set_page_private(page, 0); | |
6714 | page_cache_release(page); | |
6715 | } | |
39279cc3 CM |
6716 | } |
6717 | ||
9ebefb18 CM |
6718 | /* |
6719 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6720 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6721 | * be careful to check for EOF conditions here. We set the page up correctly | |
6722 | * for a written page which means we get ENOSPC checking when writing into | |
6723 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6724 | * support these features. | |
6725 | * | |
6726 | * We are not allowed to take the i_mutex here so we have to play games to | |
6727 | * protect against truncate races as the page could now be beyond EOF. Because | |
6728 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6729 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6730 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6731 | * unlock the page. | |
6732 | */ | |
c2ec175c | 6733 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6734 | { |
c2ec175c | 6735 | struct page *page = vmf->page; |
6da6abae | 6736 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6737 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6738 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6739 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6740 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6741 | char *kaddr; |
6742 | unsigned long zero_start; | |
9ebefb18 | 6743 | loff_t size; |
1832a6d5 | 6744 | int ret; |
9998eb70 | 6745 | int reserved = 0; |
a52d9a80 | 6746 | u64 page_start; |
e6dcd2dc | 6747 | u64 page_end; |
9ebefb18 | 6748 | |
b2b5ef5c | 6749 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 6750 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6751 | if (!ret) { |
e41f941a | 6752 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
6753 | reserved = 1; |
6754 | } | |
56a76f82 NP |
6755 | if (ret) { |
6756 | if (ret == -ENOMEM) | |
6757 | ret = VM_FAULT_OOM; | |
6758 | else /* -ENOSPC, -EIO, etc */ | |
6759 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6760 | if (reserved) |
6761 | goto out; | |
6762 | goto out_noreserve; | |
56a76f82 | 6763 | } |
1832a6d5 | 6764 | |
56a76f82 | 6765 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6766 | again: |
9ebefb18 | 6767 | lock_page(page); |
9ebefb18 | 6768 | size = i_size_read(inode); |
e6dcd2dc CM |
6769 | page_start = page_offset(page); |
6770 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6771 | |
9ebefb18 | 6772 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6773 | (page_start >= size)) { |
9ebefb18 CM |
6774 | /* page got truncated out from underneath us */ |
6775 | goto out_unlock; | |
6776 | } | |
e6dcd2dc CM |
6777 | wait_on_page_writeback(page); |
6778 | ||
d0082371 | 6779 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
6780 | set_page_extent_mapped(page); |
6781 | ||
eb84ae03 CM |
6782 | /* |
6783 | * we can't set the delalloc bits if there are pending ordered | |
6784 | * extents. Drop our locks and wait for them to finish | |
6785 | */ | |
e6dcd2dc CM |
6786 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6787 | if (ordered) { | |
2ac55d41 JB |
6788 | unlock_extent_cached(io_tree, page_start, page_end, |
6789 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6790 | unlock_page(page); |
eb84ae03 | 6791 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6792 | btrfs_put_ordered_extent(ordered); |
6793 | goto again; | |
6794 | } | |
6795 | ||
fbf19087 JB |
6796 | /* |
6797 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6798 | * if it was already dirty, so for space accounting reasons we need to | |
6799 | * clear any delalloc bits for the range we are fixing to save. There | |
6800 | * is probably a better way to do this, but for now keep consistent with | |
6801 | * prepare_pages in the normal write path. | |
6802 | */ | |
2ac55d41 | 6803 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
6804 | EXTENT_DIRTY | EXTENT_DELALLOC | |
6805 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 6806 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6807 | |
2ac55d41 JB |
6808 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6809 | &cached_state); | |
9ed74f2d | 6810 | if (ret) { |
2ac55d41 JB |
6811 | unlock_extent_cached(io_tree, page_start, page_end, |
6812 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6813 | ret = VM_FAULT_SIGBUS; |
6814 | goto out_unlock; | |
6815 | } | |
e6dcd2dc | 6816 | ret = 0; |
9ebefb18 CM |
6817 | |
6818 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6819 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6820 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6821 | else |
e6dcd2dc | 6822 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6823 | |
e6dcd2dc CM |
6824 | if (zero_start != PAGE_CACHE_SIZE) { |
6825 | kaddr = kmap(page); | |
6826 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6827 | flush_dcache_page(page); | |
6828 | kunmap(page); | |
6829 | } | |
247e743c | 6830 | ClearPageChecked(page); |
e6dcd2dc | 6831 | set_page_dirty(page); |
50a9b214 | 6832 | SetPageUptodate(page); |
5a3f23d5 | 6833 | |
257c62e1 CM |
6834 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6835 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 6836 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 6837 | |
2ac55d41 | 6838 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6839 | |
6840 | out_unlock: | |
b2b5ef5c JK |
6841 | if (!ret) { |
6842 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 6843 | return VM_FAULT_LOCKED; |
b2b5ef5c | 6844 | } |
9ebefb18 | 6845 | unlock_page(page); |
1832a6d5 | 6846 | out: |
ec39e180 | 6847 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6848 | out_noreserve: |
b2b5ef5c | 6849 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
6850 | return ret; |
6851 | } | |
6852 | ||
a41ad394 | 6853 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6854 | { |
6855 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6856 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6857 | int ret; |
3893e33b | 6858 | int err = 0; |
39279cc3 | 6859 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6860 | unsigned long nr; |
dbe674a9 | 6861 | u64 mask = root->sectorsize - 1; |
07127184 | 6862 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6863 | |
2aaa6655 | 6864 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
5d5e103a | 6865 | if (ret) |
a41ad394 | 6866 | return ret; |
8082510e | 6867 | |
4a096752 | 6868 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6869 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6870 | |
fcb80c2a JB |
6871 | /* |
6872 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6873 | * 3 things going on here | |
6874 | * | |
6875 | * 1) We need to reserve space for our orphan item and the space to | |
6876 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6877 | * orphan item because we didn't reserve space to remove it. | |
6878 | * | |
6879 | * 2) We need to reserve space to update our inode. | |
6880 | * | |
6881 | * 3) We need to have something to cache all the space that is going to | |
6882 | * be free'd up by the truncate operation, but also have some slack | |
6883 | * space reserved in case it uses space during the truncate (thank you | |
6884 | * very much snapshotting). | |
6885 | * | |
6886 | * And we need these to all be seperate. The fact is we can use alot of | |
6887 | * space doing the truncate, and we have no earthly idea how much space | |
6888 | * we will use, so we need the truncate reservation to be seperate so it | |
6889 | * doesn't end up using space reserved for updating the inode or | |
6890 | * removing the orphan item. We also need to be able to stop the | |
6891 | * transaction and start a new one, which means we need to be able to | |
6892 | * update the inode several times, and we have no idea of knowing how | |
6893 | * many times that will be, so we can't just reserve 1 item for the | |
6894 | * entirety of the opration, so that has to be done seperately as well. | |
6895 | * Then there is the orphan item, which does indeed need to be held on | |
6896 | * to for the whole operation, and we need nobody to touch this reserved | |
6897 | * space except the orphan code. | |
6898 | * | |
6899 | * So that leaves us with | |
6900 | * | |
6901 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6902 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6903 | * transaction reservation. | |
6904 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6905 | * updating the inode. | |
6906 | */ | |
66d8f3dd | 6907 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
6908 | if (!rsv) |
6909 | return -ENOMEM; | |
4a338542 | 6910 | rsv->size = min_size; |
ca7e70f5 | 6911 | rsv->failfast = 1; |
f0cd846e | 6912 | |
907cbceb | 6913 | /* |
07127184 | 6914 | * 1 for the truncate slack space |
907cbceb JB |
6915 | * 1 for the orphan item we're going to add |
6916 | * 1 for the orphan item deletion | |
6917 | * 1 for updating the inode. | |
6918 | */ | |
fcb80c2a JB |
6919 | trans = btrfs_start_transaction(root, 4); |
6920 | if (IS_ERR(trans)) { | |
6921 | err = PTR_ERR(trans); | |
6922 | goto out; | |
6923 | } | |
f0cd846e | 6924 | |
907cbceb JB |
6925 | /* Migrate the slack space for the truncate to our reserve */ |
6926 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6927 | min_size); | |
fcb80c2a | 6928 | BUG_ON(ret); |
f0cd846e JB |
6929 | |
6930 | ret = btrfs_orphan_add(trans, inode); | |
6931 | if (ret) { | |
6932 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6933 | goto out; |
f0cd846e JB |
6934 | } |
6935 | ||
5a3f23d5 CM |
6936 | /* |
6937 | * setattr is responsible for setting the ordered_data_close flag, | |
6938 | * but that is only tested during the last file release. That | |
6939 | * could happen well after the next commit, leaving a great big | |
6940 | * window where new writes may get lost if someone chooses to write | |
6941 | * to this file after truncating to zero | |
6942 | * | |
6943 | * The inode doesn't have any dirty data here, and so if we commit | |
6944 | * this is a noop. If someone immediately starts writing to the inode | |
6945 | * it is very likely we'll catch some of their writes in this | |
6946 | * transaction, and the commit will find this file on the ordered | |
6947 | * data list with good things to send down. | |
6948 | * | |
6949 | * This is a best effort solution, there is still a window where | |
6950 | * using truncate to replace the contents of the file will | |
6951 | * end up with a zero length file after a crash. | |
6952 | */ | |
72ac3c0d JB |
6953 | if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
6954 | &BTRFS_I(inode)->runtime_flags)) | |
5a3f23d5 CM |
6955 | btrfs_add_ordered_operation(trans, root, inode); |
6956 | ||
5dc562c5 JB |
6957 | /* |
6958 | * So if we truncate and then write and fsync we normally would just | |
6959 | * write the extents that changed, which is a problem if we need to | |
6960 | * first truncate that entire inode. So set this flag so we write out | |
6961 | * all of the extents in the inode to the sync log so we're completely | |
6962 | * safe. | |
6963 | */ | |
6964 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 6965 | trans->block_rsv = rsv; |
5dc562c5 | 6966 | |
8082510e YZ |
6967 | while (1) { |
6968 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
6969 | inode->i_size, | |
6970 | BTRFS_EXTENT_DATA_KEY); | |
ca7e70f5 | 6971 | if (ret != -ENOSPC) { |
3893e33b | 6972 | err = ret; |
8082510e | 6973 | break; |
3893e33b | 6974 | } |
39279cc3 | 6975 | |
fcb80c2a | 6976 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6977 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6978 | if (ret) { |
6979 | err = ret; | |
6980 | break; | |
6981 | } | |
ca7e70f5 | 6982 | |
8082510e YZ |
6983 | nr = trans->blocks_used; |
6984 | btrfs_end_transaction(trans, root); | |
6985 | btrfs_btree_balance_dirty(root, nr); | |
ca7e70f5 JB |
6986 | |
6987 | trans = btrfs_start_transaction(root, 2); | |
6988 | if (IS_ERR(trans)) { | |
6989 | ret = err = PTR_ERR(trans); | |
6990 | trans = NULL; | |
6991 | break; | |
6992 | } | |
6993 | ||
6994 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
6995 | rsv, min_size); | |
6996 | BUG_ON(ret); /* shouldn't happen */ | |
6997 | trans->block_rsv = rsv; | |
8082510e YZ |
6998 | } |
6999 | ||
7000 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 7001 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 7002 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
7003 | if (ret) |
7004 | err = ret; | |
ded5db9d JB |
7005 | } else if (ret && inode->i_nlink > 0) { |
7006 | /* | |
7007 | * Failed to do the truncate, remove us from the in memory | |
7008 | * orphan list. | |
7009 | */ | |
7010 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
7011 | } |
7012 | ||
917c16b2 CM |
7013 | if (trans) { |
7014 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
7015 | ret = btrfs_update_inode(trans, root, inode); | |
7016 | if (ret && !err) | |
7017 | err = ret; | |
7b128766 | 7018 | |
917c16b2 | 7019 | nr = trans->blocks_used; |
7ad85bb7 | 7020 | ret = btrfs_end_transaction(trans, root); |
917c16b2 CM |
7021 | btrfs_btree_balance_dirty(root, nr); |
7022 | } | |
fcb80c2a JB |
7023 | |
7024 | out: | |
7025 | btrfs_free_block_rsv(root, rsv); | |
7026 | ||
3893e33b JB |
7027 | if (ret && !err) |
7028 | err = ret; | |
a41ad394 | 7029 | |
3893e33b | 7030 | return err; |
39279cc3 CM |
7031 | } |
7032 | ||
d352ac68 CM |
7033 | /* |
7034 | * create a new subvolume directory/inode (helper for the ioctl). | |
7035 | */ | |
d2fb3437 | 7036 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 7037 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 7038 | { |
39279cc3 | 7039 | struct inode *inode; |
76dda93c | 7040 | int err; |
00e4e6b3 | 7041 | u64 index = 0; |
39279cc3 | 7042 | |
12fc9d09 FA |
7043 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
7044 | new_dirid, new_dirid, | |
7045 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
7046 | &index); | |
54aa1f4d | 7047 | if (IS_ERR(inode)) |
f46b5a66 | 7048 | return PTR_ERR(inode); |
39279cc3 CM |
7049 | inode->i_op = &btrfs_dir_inode_operations; |
7050 | inode->i_fop = &btrfs_dir_file_operations; | |
7051 | ||
bfe86848 | 7052 | set_nlink(inode, 1); |
dbe674a9 | 7053 | btrfs_i_size_write(inode, 0); |
3b96362c | 7054 | |
76dda93c | 7055 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 7056 | |
76dda93c | 7057 | iput(inode); |
ce598979 | 7058 | return err; |
39279cc3 CM |
7059 | } |
7060 | ||
39279cc3 CM |
7061 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
7062 | { | |
7063 | struct btrfs_inode *ei; | |
2ead6ae7 | 7064 | struct inode *inode; |
39279cc3 CM |
7065 | |
7066 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
7067 | if (!ei) | |
7068 | return NULL; | |
2ead6ae7 YZ |
7069 | |
7070 | ei->root = NULL; | |
2ead6ae7 | 7071 | ei->generation = 0; |
15ee9bc7 | 7072 | ei->last_trans = 0; |
257c62e1 | 7073 | ei->last_sub_trans = 0; |
e02119d5 | 7074 | ei->logged_trans = 0; |
2ead6ae7 | 7075 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
7076 | ei->disk_i_size = 0; |
7077 | ei->flags = 0; | |
7709cde3 | 7078 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
7079 | ei->index_cnt = (u64)-1; |
7080 | ei->last_unlink_trans = 0; | |
46d8bc34 | 7081 | ei->last_log_commit = 0; |
2ead6ae7 | 7082 | |
9e0baf60 JB |
7083 | spin_lock_init(&ei->lock); |
7084 | ei->outstanding_extents = 0; | |
7085 | ei->reserved_extents = 0; | |
2ead6ae7 | 7086 | |
72ac3c0d | 7087 | ei->runtime_flags = 0; |
261507a0 | 7088 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 7089 | |
16cdcec7 MX |
7090 | ei->delayed_node = NULL; |
7091 | ||
2ead6ae7 | 7092 | inode = &ei->vfs_inode; |
a8067e02 | 7093 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
7094 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
7095 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
7096 | ei->io_tree.track_uptodate = 1; |
7097 | ei->io_failure_tree.track_uptodate = 1; | |
2ead6ae7 | 7098 | mutex_init(&ei->log_mutex); |
f248679e | 7099 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 7100 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 7101 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 7102 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
7103 | RB_CLEAR_NODE(&ei->rb_node); |
7104 | ||
7105 | return inode; | |
39279cc3 CM |
7106 | } |
7107 | ||
fa0d7e3d NP |
7108 | static void btrfs_i_callback(struct rcu_head *head) |
7109 | { | |
7110 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
7111 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
7112 | } | |
7113 | ||
39279cc3 CM |
7114 | void btrfs_destroy_inode(struct inode *inode) |
7115 | { | |
e6dcd2dc | 7116 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
7117 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7118 | ||
b3d9b7a3 | 7119 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 7120 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
7121 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7122 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
7123 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
7124 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 7125 | |
a6dbd429 JB |
7126 | /* |
7127 | * This can happen where we create an inode, but somebody else also | |
7128 | * created the same inode and we need to destroy the one we already | |
7129 | * created. | |
7130 | */ | |
7131 | if (!root) | |
7132 | goto free; | |
7133 | ||
5a3f23d5 CM |
7134 | /* |
7135 | * Make sure we're properly removed from the ordered operation | |
7136 | * lists. | |
7137 | */ | |
7138 | smp_mb(); | |
7139 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
7140 | spin_lock(&root->fs_info->ordered_extent_lock); | |
7141 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
7142 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
7143 | } | |
7144 | ||
8a35d95f JB |
7145 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
7146 | &BTRFS_I(inode)->runtime_flags)) { | |
33345d01 LZ |
7147 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
7148 | (unsigned long long)btrfs_ino(inode)); | |
8a35d95f | 7149 | atomic_dec(&root->orphan_inodes); |
7b128766 | 7150 | } |
7b128766 | 7151 | |
d397712b | 7152 | while (1) { |
e6dcd2dc CM |
7153 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
7154 | if (!ordered) | |
7155 | break; | |
7156 | else { | |
d397712b CM |
7157 | printk(KERN_ERR "btrfs found ordered " |
7158 | "extent %llu %llu on inode cleanup\n", | |
7159 | (unsigned long long)ordered->file_offset, | |
7160 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
7161 | btrfs_remove_ordered_extent(inode, ordered); |
7162 | btrfs_put_ordered_extent(ordered); | |
7163 | btrfs_put_ordered_extent(ordered); | |
7164 | } | |
7165 | } | |
5d4f98a2 | 7166 | inode_tree_del(inode); |
5b21f2ed | 7167 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 7168 | free: |
16cdcec7 | 7169 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 7170 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
7171 | } |
7172 | ||
45321ac5 | 7173 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
7174 | { |
7175 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 7176 | |
0af3d00b | 7177 | if (btrfs_root_refs(&root->root_item) == 0 && |
83eea1f1 | 7178 | !btrfs_is_free_space_inode(inode)) |
45321ac5 | 7179 | return 1; |
76dda93c | 7180 | else |
45321ac5 | 7181 | return generic_drop_inode(inode); |
76dda93c YZ |
7182 | } |
7183 | ||
0ee0fda0 | 7184 | static void init_once(void *foo) |
39279cc3 CM |
7185 | { |
7186 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
7187 | ||
7188 | inode_init_once(&ei->vfs_inode); | |
7189 | } | |
7190 | ||
7191 | void btrfs_destroy_cachep(void) | |
7192 | { | |
7193 | if (btrfs_inode_cachep) | |
7194 | kmem_cache_destroy(btrfs_inode_cachep); | |
7195 | if (btrfs_trans_handle_cachep) | |
7196 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
7197 | if (btrfs_transaction_cachep) | |
7198 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
7199 | if (btrfs_path_cachep) |
7200 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
7201 | if (btrfs_free_space_cachep) |
7202 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
7203 | } |
7204 | ||
7205 | int btrfs_init_cachep(void) | |
7206 | { | |
837e1972 | 7207 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
7208 | sizeof(struct btrfs_inode), 0, |
7209 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
7210 | if (!btrfs_inode_cachep) |
7211 | goto fail; | |
9601e3f6 | 7212 | |
837e1972 | 7213 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
7214 | sizeof(struct btrfs_trans_handle), 0, |
7215 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7216 | if (!btrfs_trans_handle_cachep) |
7217 | goto fail; | |
9601e3f6 | 7218 | |
837e1972 | 7219 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
7220 | sizeof(struct btrfs_transaction), 0, |
7221 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7222 | if (!btrfs_transaction_cachep) |
7223 | goto fail; | |
9601e3f6 | 7224 | |
837e1972 | 7225 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
7226 | sizeof(struct btrfs_path), 0, |
7227 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
7228 | if (!btrfs_path_cachep) |
7229 | goto fail; | |
9601e3f6 | 7230 | |
837e1972 | 7231 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
7232 | sizeof(struct btrfs_free_space), 0, |
7233 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
7234 | if (!btrfs_free_space_cachep) | |
7235 | goto fail; | |
7236 | ||
39279cc3 CM |
7237 | return 0; |
7238 | fail: | |
7239 | btrfs_destroy_cachep(); | |
7240 | return -ENOMEM; | |
7241 | } | |
7242 | ||
7243 | static int btrfs_getattr(struct vfsmount *mnt, | |
7244 | struct dentry *dentry, struct kstat *stat) | |
7245 | { | |
7246 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
7247 | u32 blocksize = inode->i_sb->s_blocksize; |
7248 | ||
39279cc3 | 7249 | generic_fillattr(inode, stat); |
0ee5dc67 | 7250 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 7251 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
7252 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
7253 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
7254 | return 0; |
7255 | } | |
7256 | ||
75e7cb7f LB |
7257 | /* |
7258 | * If a file is moved, it will inherit the cow and compression flags of the new | |
7259 | * directory. | |
7260 | */ | |
7261 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
7262 | { | |
7263 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
7264 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
7265 | ||
7266 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
7267 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
7268 | else | |
7269 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
7270 | ||
bc178237 | 7271 | if (b_dir->flags & BTRFS_INODE_COMPRESS) { |
75e7cb7f | 7272 | b_inode->flags |= BTRFS_INODE_COMPRESS; |
bc178237 LB |
7273 | b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS; |
7274 | } else { | |
7275 | b_inode->flags &= ~(BTRFS_INODE_COMPRESS | | |
7276 | BTRFS_INODE_NOCOMPRESS); | |
7277 | } | |
75e7cb7f LB |
7278 | } |
7279 | ||
d397712b CM |
7280 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
7281 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
7282 | { |
7283 | struct btrfs_trans_handle *trans; | |
7284 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 7285 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
7286 | struct inode *new_inode = new_dentry->d_inode; |
7287 | struct inode *old_inode = old_dentry->d_inode; | |
7288 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 7289 | u64 index = 0; |
4df27c4d | 7290 | u64 root_objectid; |
39279cc3 | 7291 | int ret; |
33345d01 | 7292 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 7293 | |
33345d01 | 7294 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
7295 | return -EPERM; |
7296 | ||
4df27c4d | 7297 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 7298 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
7299 | return -EXDEV; |
7300 | ||
33345d01 LZ |
7301 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
7302 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 7303 | return -ENOTEMPTY; |
5f39d397 | 7304 | |
4df27c4d YZ |
7305 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
7306 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
7307 | return -ENOTEMPTY; | |
5a3f23d5 CM |
7308 | /* |
7309 | * we're using rename to replace one file with another. | |
7310 | * and the replacement file is large. Start IO on it now so | |
7311 | * we don't add too much work to the end of the transaction | |
7312 | */ | |
4baf8c92 | 7313 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
7314 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
7315 | filemap_flush(old_inode->i_mapping); | |
7316 | ||
76dda93c | 7317 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 7318 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7319 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
7320 | /* |
7321 | * We want to reserve the absolute worst case amount of items. So if | |
7322 | * both inodes are subvols and we need to unlink them then that would | |
7323 | * require 4 item modifications, but if they are both normal inodes it | |
7324 | * would require 5 item modifications, so we'll assume their normal | |
7325 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
7326 | * should cover the worst case number of items we'll modify. | |
7327 | */ | |
7328 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
7329 | if (IS_ERR(trans)) { |
7330 | ret = PTR_ERR(trans); | |
7331 | goto out_notrans; | |
7332 | } | |
76dda93c | 7333 | |
4df27c4d YZ |
7334 | if (dest != root) |
7335 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7336 | |
a5719521 YZ |
7337 | ret = btrfs_set_inode_index(new_dir, &index); |
7338 | if (ret) | |
7339 | goto out_fail; | |
5a3f23d5 | 7340 | |
33345d01 | 7341 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7342 | /* force full log commit if subvolume involved. */ |
7343 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7344 | } else { | |
a5719521 YZ |
7345 | ret = btrfs_insert_inode_ref(trans, dest, |
7346 | new_dentry->d_name.name, | |
7347 | new_dentry->d_name.len, | |
33345d01 LZ |
7348 | old_ino, |
7349 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7350 | if (ret) |
7351 | goto out_fail; | |
4df27c4d YZ |
7352 | /* |
7353 | * this is an ugly little race, but the rename is required | |
7354 | * to make sure that if we crash, the inode is either at the | |
7355 | * old name or the new one. pinning the log transaction lets | |
7356 | * us make sure we don't allow a log commit to come in after | |
7357 | * we unlink the name but before we add the new name back in. | |
7358 | */ | |
7359 | btrfs_pin_log_trans(root); | |
7360 | } | |
5a3f23d5 CM |
7361 | /* |
7362 | * make sure the inode gets flushed if it is replacing | |
7363 | * something. | |
7364 | */ | |
33345d01 | 7365 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7366 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7367 | |
0c4d2d95 JB |
7368 | inode_inc_iversion(old_dir); |
7369 | inode_inc_iversion(new_dir); | |
7370 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
7371 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7372 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7373 | old_inode->i_ctime = ctime; | |
5f39d397 | 7374 | |
12fcfd22 CM |
7375 | if (old_dentry->d_parent != new_dentry->d_parent) |
7376 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7377 | ||
33345d01 | 7378 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7379 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7380 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7381 | old_dentry->d_name.name, | |
7382 | old_dentry->d_name.len); | |
7383 | } else { | |
92986796 AV |
7384 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7385 | old_dentry->d_inode, | |
7386 | old_dentry->d_name.name, | |
7387 | old_dentry->d_name.len); | |
7388 | if (!ret) | |
7389 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 7390 | } |
79787eaa JM |
7391 | if (ret) { |
7392 | btrfs_abort_transaction(trans, root, ret); | |
7393 | goto out_fail; | |
7394 | } | |
39279cc3 CM |
7395 | |
7396 | if (new_inode) { | |
0c4d2d95 | 7397 | inode_inc_iversion(new_inode); |
39279cc3 | 7398 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 7399 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7400 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7401 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7402 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7403 | root_objectid, | |
7404 | new_dentry->d_name.name, | |
7405 | new_dentry->d_name.len); | |
7406 | BUG_ON(new_inode->i_nlink == 0); | |
7407 | } else { | |
7408 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7409 | new_dentry->d_inode, | |
7410 | new_dentry->d_name.name, | |
7411 | new_dentry->d_name.len); | |
7412 | } | |
79787eaa | 7413 | if (!ret && new_inode->i_nlink == 0) { |
e02119d5 | 7414 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7415 | BUG_ON(ret); |
7b128766 | 7416 | } |
79787eaa JM |
7417 | if (ret) { |
7418 | btrfs_abort_transaction(trans, root, ret); | |
7419 | goto out_fail; | |
7420 | } | |
39279cc3 | 7421 | } |
aec7477b | 7422 | |
75e7cb7f LB |
7423 | fixup_inode_flags(new_dir, old_inode); |
7424 | ||
4df27c4d YZ |
7425 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7426 | new_dentry->d_name.name, | |
a5719521 | 7427 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
7428 | if (ret) { |
7429 | btrfs_abort_transaction(trans, root, ret); | |
7430 | goto out_fail; | |
7431 | } | |
39279cc3 | 7432 | |
33345d01 | 7433 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7434 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7435 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7436 | btrfs_end_log_trans(root); |
7437 | } | |
39279cc3 | 7438 | out_fail: |
7ad85bb7 | 7439 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7440 | out_notrans: |
33345d01 | 7441 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7442 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7443 | |
39279cc3 CM |
7444 | return ret; |
7445 | } | |
7446 | ||
d352ac68 CM |
7447 | /* |
7448 | * some fairly slow code that needs optimization. This walks the list | |
7449 | * of all the inodes with pending delalloc and forces them to disk. | |
7450 | */ | |
24bbcf04 | 7451 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7452 | { |
7453 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7454 | struct btrfs_inode *binode; | |
5b21f2ed | 7455 | struct inode *inode; |
ea8c2819 | 7456 | |
c146afad YZ |
7457 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7458 | return -EROFS; | |
7459 | ||
75eff68e | 7460 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7461 | while (!list_empty(head)) { |
ea8c2819 CM |
7462 | binode = list_entry(head->next, struct btrfs_inode, |
7463 | delalloc_inodes); | |
5b21f2ed ZY |
7464 | inode = igrab(&binode->vfs_inode); |
7465 | if (!inode) | |
7466 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7467 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7468 | if (inode) { |
8c8bee1d | 7469 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7470 | if (delay_iput) |
7471 | btrfs_add_delayed_iput(inode); | |
7472 | else | |
7473 | iput(inode); | |
5b21f2ed ZY |
7474 | } |
7475 | cond_resched(); | |
75eff68e | 7476 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7477 | } |
75eff68e | 7478 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7479 | |
7480 | /* the filemap_flush will queue IO into the worker threads, but | |
7481 | * we have to make sure the IO is actually started and that | |
7482 | * ordered extents get created before we return | |
7483 | */ | |
7484 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7485 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7486 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7487 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7488 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7489 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7490 | } |
7491 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7492 | return 0; |
7493 | } | |
7494 | ||
39279cc3 CM |
7495 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7496 | const char *symname) | |
7497 | { | |
7498 | struct btrfs_trans_handle *trans; | |
7499 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7500 | struct btrfs_path *path; | |
7501 | struct btrfs_key key; | |
1832a6d5 | 7502 | struct inode *inode = NULL; |
39279cc3 CM |
7503 | int err; |
7504 | int drop_inode = 0; | |
7505 | u64 objectid; | |
00e4e6b3 | 7506 | u64 index = 0 ; |
39279cc3 CM |
7507 | int name_len; |
7508 | int datasize; | |
5f39d397 | 7509 | unsigned long ptr; |
39279cc3 | 7510 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7511 | struct extent_buffer *leaf; |
1832a6d5 | 7512 | unsigned long nr = 0; |
39279cc3 CM |
7513 | |
7514 | name_len = strlen(symname) + 1; | |
7515 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7516 | return -ENAMETOOLONG; | |
1832a6d5 | 7517 | |
9ed74f2d JB |
7518 | /* |
7519 | * 2 items for inode item and ref | |
7520 | * 2 items for dir items | |
7521 | * 1 item for xattr if selinux is on | |
7522 | */ | |
a22285a6 YZ |
7523 | trans = btrfs_start_transaction(root, 5); |
7524 | if (IS_ERR(trans)) | |
7525 | return PTR_ERR(trans); | |
1832a6d5 | 7526 | |
581bb050 LZ |
7527 | err = btrfs_find_free_ino(root, &objectid); |
7528 | if (err) | |
7529 | goto out_unlock; | |
7530 | ||
aec7477b | 7531 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7532 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7533 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7534 | if (IS_ERR(inode)) { |
7535 | err = PTR_ERR(inode); | |
39279cc3 | 7536 | goto out_unlock; |
7cf96da3 | 7537 | } |
39279cc3 | 7538 | |
2a7dba39 | 7539 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7540 | if (err) { |
7541 | drop_inode = 1; | |
7542 | goto out_unlock; | |
7543 | } | |
7544 | ||
ad19db71 CS |
7545 | /* |
7546 | * If the active LSM wants to access the inode during | |
7547 | * d_instantiate it needs these. Smack checks to see | |
7548 | * if the filesystem supports xattrs by looking at the | |
7549 | * ops vector. | |
7550 | */ | |
7551 | inode->i_fop = &btrfs_file_operations; | |
7552 | inode->i_op = &btrfs_file_inode_operations; | |
7553 | ||
a1b075d2 | 7554 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7555 | if (err) |
7556 | drop_inode = 1; | |
7557 | else { | |
7558 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7559 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7560 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7561 | } |
39279cc3 CM |
7562 | if (drop_inode) |
7563 | goto out_unlock; | |
7564 | ||
7565 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7566 | if (!path) { |
7567 | err = -ENOMEM; | |
7568 | drop_inode = 1; | |
7569 | goto out_unlock; | |
7570 | } | |
33345d01 | 7571 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7572 | key.offset = 0; |
39279cc3 CM |
7573 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7574 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7575 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7576 | datasize); | |
54aa1f4d CM |
7577 | if (err) { |
7578 | drop_inode = 1; | |
b0839166 | 7579 | btrfs_free_path(path); |
54aa1f4d CM |
7580 | goto out_unlock; |
7581 | } | |
5f39d397 CM |
7582 | leaf = path->nodes[0]; |
7583 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7584 | struct btrfs_file_extent_item); | |
7585 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7586 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7587 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7588 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7589 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7590 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7591 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7592 | ||
39279cc3 | 7593 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7594 | write_extent_buffer(leaf, symname, ptr, name_len); |
7595 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7596 | btrfs_free_path(path); |
5f39d397 | 7597 | |
39279cc3 CM |
7598 | inode->i_op = &btrfs_symlink_inode_operations; |
7599 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7600 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7601 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7602 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7603 | err = btrfs_update_inode(trans, root, inode); |
7604 | if (err) | |
7605 | drop_inode = 1; | |
39279cc3 CM |
7606 | |
7607 | out_unlock: | |
08c422c2 AV |
7608 | if (!err) |
7609 | d_instantiate(dentry, inode); | |
d3c2fdcf | 7610 | nr = trans->blocks_used; |
7ad85bb7 | 7611 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7612 | if (drop_inode) { |
7613 | inode_dec_link_count(inode); | |
7614 | iput(inode); | |
7615 | } | |
d3c2fdcf | 7616 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7617 | return err; |
7618 | } | |
16432985 | 7619 | |
0af3d00b JB |
7620 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7621 | u64 start, u64 num_bytes, u64 min_size, | |
7622 | loff_t actual_len, u64 *alloc_hint, | |
7623 | struct btrfs_trans_handle *trans) | |
d899e052 | 7624 | { |
5dc562c5 JB |
7625 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
7626 | struct extent_map *em; | |
d899e052 YZ |
7627 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7628 | struct btrfs_key ins; | |
d899e052 | 7629 | u64 cur_offset = start; |
55a61d1d | 7630 | u64 i_size; |
d899e052 | 7631 | int ret = 0; |
0af3d00b | 7632 | bool own_trans = true; |
d899e052 | 7633 | |
0af3d00b JB |
7634 | if (trans) |
7635 | own_trans = false; | |
d899e052 | 7636 | while (num_bytes > 0) { |
0af3d00b JB |
7637 | if (own_trans) { |
7638 | trans = btrfs_start_transaction(root, 3); | |
7639 | if (IS_ERR(trans)) { | |
7640 | ret = PTR_ERR(trans); | |
7641 | break; | |
7642 | } | |
5a303d5d YZ |
7643 | } |
7644 | ||
efa56464 | 7645 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
81c9ad23 | 7646 | 0, *alloc_hint, &ins, 1); |
5a303d5d | 7647 | if (ret) { |
0af3d00b JB |
7648 | if (own_trans) |
7649 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7650 | break; |
d899e052 | 7651 | } |
5a303d5d | 7652 | |
d899e052 YZ |
7653 | ret = insert_reserved_file_extent(trans, inode, |
7654 | cur_offset, ins.objectid, | |
7655 | ins.offset, ins.offset, | |
920bbbfb | 7656 | ins.offset, 0, 0, 0, |
d899e052 | 7657 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa JM |
7658 | if (ret) { |
7659 | btrfs_abort_transaction(trans, root, ret); | |
7660 | if (own_trans) | |
7661 | btrfs_end_transaction(trans, root); | |
7662 | break; | |
7663 | } | |
a1ed835e CM |
7664 | btrfs_drop_extent_cache(inode, cur_offset, |
7665 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7666 | |
5dc562c5 JB |
7667 | em = alloc_extent_map(); |
7668 | if (!em) { | |
7669 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
7670 | &BTRFS_I(inode)->runtime_flags); | |
7671 | goto next; | |
7672 | } | |
7673 | ||
7674 | em->start = cur_offset; | |
7675 | em->orig_start = cur_offset; | |
7676 | em->len = ins.offset; | |
7677 | em->block_start = ins.objectid; | |
7678 | em->block_len = ins.offset; | |
7679 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
7680 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
7681 | em->generation = trans->transid; | |
7682 | ||
7683 | while (1) { | |
7684 | write_lock(&em_tree->lock); | |
7685 | ret = add_extent_mapping(em_tree, em); | |
7686 | if (!ret) | |
7687 | list_move(&em->list, | |
7688 | &em_tree->modified_extents); | |
7689 | write_unlock(&em_tree->lock); | |
7690 | if (ret != -EEXIST) | |
7691 | break; | |
7692 | btrfs_drop_extent_cache(inode, cur_offset, | |
7693 | cur_offset + ins.offset - 1, | |
7694 | 0); | |
7695 | } | |
7696 | free_extent_map(em); | |
7697 | next: | |
d899e052 YZ |
7698 | num_bytes -= ins.offset; |
7699 | cur_offset += ins.offset; | |
efa56464 | 7700 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7701 | |
0c4d2d95 | 7702 | inode_inc_iversion(inode); |
d899e052 | 7703 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7704 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7705 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7706 | (actual_len > inode->i_size) && |
7707 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7708 | if (cur_offset > actual_len) |
55a61d1d | 7709 | i_size = actual_len; |
d1ea6a61 | 7710 | else |
55a61d1d JB |
7711 | i_size = cur_offset; |
7712 | i_size_write(inode, i_size); | |
7713 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7714 | } |
7715 | ||
d899e052 | 7716 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
7717 | |
7718 | if (ret) { | |
7719 | btrfs_abort_transaction(trans, root, ret); | |
7720 | if (own_trans) | |
7721 | btrfs_end_transaction(trans, root); | |
7722 | break; | |
7723 | } | |
d899e052 | 7724 | |
0af3d00b JB |
7725 | if (own_trans) |
7726 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7727 | } |
d899e052 YZ |
7728 | return ret; |
7729 | } | |
7730 | ||
0af3d00b JB |
7731 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7732 | u64 start, u64 num_bytes, u64 min_size, | |
7733 | loff_t actual_len, u64 *alloc_hint) | |
7734 | { | |
7735 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7736 | min_size, actual_len, alloc_hint, | |
7737 | NULL); | |
7738 | } | |
7739 | ||
7740 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7741 | struct btrfs_trans_handle *trans, int mode, | |
7742 | u64 start, u64 num_bytes, u64 min_size, | |
7743 | loff_t actual_len, u64 *alloc_hint) | |
7744 | { | |
7745 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7746 | min_size, actual_len, alloc_hint, trans); | |
7747 | } | |
7748 | ||
e6dcd2dc CM |
7749 | static int btrfs_set_page_dirty(struct page *page) |
7750 | { | |
e6dcd2dc CM |
7751 | return __set_page_dirty_nobuffers(page); |
7752 | } | |
7753 | ||
10556cb2 | 7754 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7755 | { |
b83cc969 | 7756 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7757 | umode_t mode = inode->i_mode; |
b83cc969 | 7758 | |
cb6db4e5 JM |
7759 | if (mask & MAY_WRITE && |
7760 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7761 | if (btrfs_root_readonly(root)) | |
7762 | return -EROFS; | |
7763 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7764 | return -EACCES; | |
7765 | } | |
2830ba7f | 7766 | return generic_permission(inode, mask); |
fdebe2bd | 7767 | } |
39279cc3 | 7768 | |
6e1d5dcc | 7769 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7770 | .getattr = btrfs_getattr, |
39279cc3 CM |
7771 | .lookup = btrfs_lookup, |
7772 | .create = btrfs_create, | |
7773 | .unlink = btrfs_unlink, | |
7774 | .link = btrfs_link, | |
7775 | .mkdir = btrfs_mkdir, | |
7776 | .rmdir = btrfs_rmdir, | |
7777 | .rename = btrfs_rename, | |
7778 | .symlink = btrfs_symlink, | |
7779 | .setattr = btrfs_setattr, | |
618e21d5 | 7780 | .mknod = btrfs_mknod, |
95819c05 CH |
7781 | .setxattr = btrfs_setxattr, |
7782 | .getxattr = btrfs_getxattr, | |
5103e947 | 7783 | .listxattr = btrfs_listxattr, |
95819c05 | 7784 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7785 | .permission = btrfs_permission, |
4e34e719 | 7786 | .get_acl = btrfs_get_acl, |
39279cc3 | 7787 | }; |
6e1d5dcc | 7788 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7789 | .lookup = btrfs_lookup, |
fdebe2bd | 7790 | .permission = btrfs_permission, |
4e34e719 | 7791 | .get_acl = btrfs_get_acl, |
39279cc3 | 7792 | }; |
76dda93c | 7793 | |
828c0950 | 7794 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7795 | .llseek = generic_file_llseek, |
7796 | .read = generic_read_dir, | |
cbdf5a24 | 7797 | .readdir = btrfs_real_readdir, |
34287aa3 | 7798 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7799 | #ifdef CONFIG_COMPAT |
34287aa3 | 7800 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7801 | #endif |
6bf13c0c | 7802 | .release = btrfs_release_file, |
e02119d5 | 7803 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7804 | }; |
7805 | ||
d1310b2e | 7806 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7807 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7808 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7809 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7810 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7811 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7812 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7813 | .set_bit_hook = btrfs_set_bit_hook, |
7814 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7815 | .merge_extent_hook = btrfs_merge_extent_hook, |
7816 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7817 | }; |
7818 | ||
35054394 CM |
7819 | /* |
7820 | * btrfs doesn't support the bmap operation because swapfiles | |
7821 | * use bmap to make a mapping of extents in the file. They assume | |
7822 | * these extents won't change over the life of the file and they | |
7823 | * use the bmap result to do IO directly to the drive. | |
7824 | * | |
7825 | * the btrfs bmap call would return logical addresses that aren't | |
7826 | * suitable for IO and they also will change frequently as COW | |
7827 | * operations happen. So, swapfile + btrfs == corruption. | |
7828 | * | |
7829 | * For now we're avoiding this by dropping bmap. | |
7830 | */ | |
7f09410b | 7831 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7832 | .readpage = btrfs_readpage, |
7833 | .writepage = btrfs_writepage, | |
b293f02e | 7834 | .writepages = btrfs_writepages, |
3ab2fb5a | 7835 | .readpages = btrfs_readpages, |
16432985 | 7836 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7837 | .invalidatepage = btrfs_invalidatepage, |
7838 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7839 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7840 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7841 | }; |
7842 | ||
7f09410b | 7843 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7844 | .readpage = btrfs_readpage, |
7845 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7846 | .invalidatepage = btrfs_invalidatepage, |
7847 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7848 | }; |
7849 | ||
6e1d5dcc | 7850 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7851 | .getattr = btrfs_getattr, |
7852 | .setattr = btrfs_setattr, | |
95819c05 CH |
7853 | .setxattr = btrfs_setxattr, |
7854 | .getxattr = btrfs_getxattr, | |
5103e947 | 7855 | .listxattr = btrfs_listxattr, |
95819c05 | 7856 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7857 | .permission = btrfs_permission, |
1506fcc8 | 7858 | .fiemap = btrfs_fiemap, |
4e34e719 | 7859 | .get_acl = btrfs_get_acl, |
e41f941a | 7860 | .update_time = btrfs_update_time, |
39279cc3 | 7861 | }; |
6e1d5dcc | 7862 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7863 | .getattr = btrfs_getattr, |
7864 | .setattr = btrfs_setattr, | |
fdebe2bd | 7865 | .permission = btrfs_permission, |
95819c05 CH |
7866 | .setxattr = btrfs_setxattr, |
7867 | .getxattr = btrfs_getxattr, | |
33268eaf | 7868 | .listxattr = btrfs_listxattr, |
95819c05 | 7869 | .removexattr = btrfs_removexattr, |
4e34e719 | 7870 | .get_acl = btrfs_get_acl, |
e41f941a | 7871 | .update_time = btrfs_update_time, |
618e21d5 | 7872 | }; |
6e1d5dcc | 7873 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7874 | .readlink = generic_readlink, |
7875 | .follow_link = page_follow_link_light, | |
7876 | .put_link = page_put_link, | |
f209561a | 7877 | .getattr = btrfs_getattr, |
22c44fe6 | 7878 | .setattr = btrfs_setattr, |
fdebe2bd | 7879 | .permission = btrfs_permission, |
0279b4cd JO |
7880 | .setxattr = btrfs_setxattr, |
7881 | .getxattr = btrfs_getxattr, | |
7882 | .listxattr = btrfs_listxattr, | |
7883 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7884 | .get_acl = btrfs_get_acl, |
e41f941a | 7885 | .update_time = btrfs_update_time, |
39279cc3 | 7886 | }; |
76dda93c | 7887 | |
82d339d9 | 7888 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7889 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7890 | .d_release = btrfs_dentry_release, |
76dda93c | 7891 | }; |