Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README | |
3 | */ | |
4 | ||
5 | #include <linux/config.h> | |
6 | #include <linux/time.h> | |
7 | #include <linux/fs.h> | |
8 | #include <linux/reiserfs_fs.h> | |
9 | #include <linux/reiserfs_acl.h> | |
10 | #include <linux/reiserfs_xattr.h> | |
11 | #include <linux/smp_lock.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/highmem.h> | |
14 | #include <asm/uaccess.h> | |
15 | #include <asm/unaligned.h> | |
16 | #include <linux/buffer_head.h> | |
17 | #include <linux/mpage.h> | |
18 | #include <linux/writeback.h> | |
19 | #include <linux/quotaops.h> | |
20 | ||
21 | extern int reiserfs_default_io_size; /* default io size devuned in super.c */ | |
22 | ||
23 | static int reiserfs_commit_write(struct file *f, struct page *page, | |
24 | unsigned from, unsigned to); | |
25 | static int reiserfs_prepare_write(struct file *f, struct page *page, | |
26 | unsigned from, unsigned to); | |
27 | ||
28 | void reiserfs_delete_inode (struct inode * inode) | |
29 | { | |
30 | /* We need blocks for transaction + (user+group) quota update (possibly delete) */ | |
31 | int jbegin_count = JOURNAL_PER_BALANCE_CNT * 2 + 2 * REISERFS_QUOTA_INIT_BLOCKS; | |
32 | struct reiserfs_transaction_handle th ; | |
33 | ||
34 | reiserfs_write_lock(inode->i_sb); | |
35 | ||
36 | /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */ | |
37 | if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) { /* also handles bad_inode case */ | |
38 | down (&inode->i_sem); | |
39 | ||
40 | reiserfs_delete_xattrs (inode); | |
41 | ||
42 | if (journal_begin(&th, inode->i_sb, jbegin_count)) { | |
43 | up (&inode->i_sem); | |
44 | goto out; | |
45 | } | |
46 | reiserfs_update_inode_transaction(inode) ; | |
47 | ||
48 | if (reiserfs_delete_object (&th, inode)) { | |
49 | up (&inode->i_sem); | |
50 | goto out; | |
51 | } | |
52 | ||
53 | /* Do quota update inside a transaction for journaled quotas. We must do that | |
54 | * after delete_object so that quota updates go into the same transaction as | |
55 | * stat data deletion */ | |
56 | DQUOT_FREE_INODE(inode); | |
57 | ||
58 | if (journal_end(&th, inode->i_sb, jbegin_count)) { | |
59 | up (&inode->i_sem); | |
60 | goto out; | |
61 | } | |
62 | ||
63 | up (&inode->i_sem); | |
64 | ||
65 | /* all items of file are deleted, so we can remove "save" link */ | |
66 | remove_save_link (inode, 0/* not truncate */); /* we can't do anything | |
67 | * about an error here */ | |
68 | } else { | |
69 | /* no object items are in the tree */ | |
70 | ; | |
71 | } | |
72 | out: | |
73 | clear_inode (inode); /* note this must go after the journal_end to prevent deadlock */ | |
74 | inode->i_blocks = 0; | |
75 | reiserfs_write_unlock(inode->i_sb); | |
76 | } | |
77 | ||
78 | static void _make_cpu_key (struct cpu_key * key, int version, __u32 dirid, __u32 objectid, | |
79 | loff_t offset, int type, int length ) | |
80 | { | |
81 | key->version = version; | |
82 | ||
83 | key->on_disk_key.k_dir_id = dirid; | |
84 | key->on_disk_key.k_objectid = objectid; | |
85 | set_cpu_key_k_offset (key, offset); | |
86 | set_cpu_key_k_type (key, type); | |
87 | key->key_length = length; | |
88 | } | |
89 | ||
90 | ||
91 | /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set | |
92 | offset and type of key */ | |
93 | void make_cpu_key (struct cpu_key * key, struct inode * inode, loff_t offset, | |
94 | int type, int length ) | |
95 | { | |
96 | _make_cpu_key (key, get_inode_item_key_version (inode), le32_to_cpu (INODE_PKEY (inode)->k_dir_id), | |
97 | le32_to_cpu (INODE_PKEY (inode)->k_objectid), | |
98 | offset, type, length); | |
99 | } | |
100 | ||
101 | ||
102 | // | |
103 | // when key is 0, do not set version and short key | |
104 | // | |
105 | inline void make_le_item_head (struct item_head * ih, const struct cpu_key * key, | |
106 | int version, | |
107 | loff_t offset, int type, int length, | |
108 | int entry_count/*or ih_free_space*/) | |
109 | { | |
110 | if (key) { | |
111 | ih->ih_key.k_dir_id = cpu_to_le32 (key->on_disk_key.k_dir_id); | |
112 | ih->ih_key.k_objectid = cpu_to_le32 (key->on_disk_key.k_objectid); | |
113 | } | |
114 | put_ih_version( ih, version ); | |
115 | set_le_ih_k_offset (ih, offset); | |
116 | set_le_ih_k_type (ih, type); | |
117 | put_ih_item_len( ih, length ); | |
118 | /* set_ih_free_space (ih, 0);*/ | |
119 | // for directory items it is entry count, for directs and stat | |
120 | // datas - 0xffff, for indirects - 0 | |
121 | put_ih_entry_count( ih, entry_count ); | |
122 | } | |
123 | ||
124 | // | |
125 | // FIXME: we might cache recently accessed indirect item | |
126 | ||
127 | // Ugh. Not too eager for that.... | |
128 | // I cut the code until such time as I see a convincing argument (benchmark). | |
129 | // I don't want a bloated inode struct..., and I don't like code complexity.... | |
130 | ||
131 | /* cutting the code is fine, since it really isn't in use yet and is easy | |
132 | ** to add back in. But, Vladimir has a really good idea here. Think | |
133 | ** about what happens for reading a file. For each page, | |
134 | ** The VFS layer calls reiserfs_readpage, who searches the tree to find | |
135 | ** an indirect item. This indirect item has X number of pointers, where | |
136 | ** X is a big number if we've done the block allocation right. But, | |
137 | ** we only use one or two of these pointers during each call to readpage, | |
138 | ** needlessly researching again later on. | |
139 | ** | |
140 | ** The size of the cache could be dynamic based on the size of the file. | |
141 | ** | |
142 | ** I'd also like to see us cache the location the stat data item, since | |
143 | ** we are needlessly researching for that frequently. | |
144 | ** | |
145 | ** --chris | |
146 | */ | |
147 | ||
148 | /* If this page has a file tail in it, and | |
149 | ** it was read in by get_block_create_0, the page data is valid, | |
150 | ** but tail is still sitting in a direct item, and we can't write to | |
151 | ** it. So, look through this page, and check all the mapped buffers | |
152 | ** to make sure they have valid block numbers. Any that don't need | |
153 | ** to be unmapped, so that block_prepare_write will correctly call | |
154 | ** reiserfs_get_block to convert the tail into an unformatted node | |
155 | */ | |
156 | static inline void fix_tail_page_for_writing(struct page *page) { | |
157 | struct buffer_head *head, *next, *bh ; | |
158 | ||
159 | if (page && page_has_buffers(page)) { | |
160 | head = page_buffers(page) ; | |
161 | bh = head ; | |
162 | do { | |
163 | next = bh->b_this_page ; | |
164 | if (buffer_mapped(bh) && bh->b_blocknr == 0) { | |
165 | reiserfs_unmap_buffer(bh) ; | |
166 | } | |
167 | bh = next ; | |
168 | } while (bh != head) ; | |
169 | } | |
170 | } | |
171 | ||
172 | /* reiserfs_get_block does not need to allocate a block only if it has been | |
173 | done already or non-hole position has been found in the indirect item */ | |
174 | static inline int allocation_needed (int retval, b_blocknr_t allocated, | |
175 | struct item_head * ih, | |
176 | __u32 * item, int pos_in_item) | |
177 | { | |
178 | if (allocated) | |
179 | return 0; | |
180 | if (retval == POSITION_FOUND && is_indirect_le_ih (ih) && | |
181 | get_block_num(item, pos_in_item)) | |
182 | return 0; | |
183 | return 1; | |
184 | } | |
185 | ||
186 | static inline int indirect_item_found (int retval, struct item_head * ih) | |
187 | { | |
188 | return (retval == POSITION_FOUND) && is_indirect_le_ih (ih); | |
189 | } | |
190 | ||
191 | ||
192 | static inline void set_block_dev_mapped (struct buffer_head * bh, | |
193 | b_blocknr_t block, struct inode * inode) | |
194 | { | |
195 | map_bh(bh, inode->i_sb, block); | |
196 | } | |
197 | ||
198 | ||
199 | // | |
200 | // files which were created in the earlier version can not be longer, | |
201 | // than 2 gb | |
202 | // | |
203 | static int file_capable (struct inode * inode, long block) | |
204 | { | |
205 | if (get_inode_item_key_version (inode) != KEY_FORMAT_3_5 || // it is new file. | |
206 | block < (1 << (31 - inode->i_sb->s_blocksize_bits))) // old file, but 'block' is inside of 2gb | |
207 | return 1; | |
208 | ||
209 | return 0; | |
210 | } | |
211 | ||
212 | /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th, | |
213 | struct inode *inode, struct path *path) { | |
214 | struct super_block *s = th->t_super ; | |
215 | int len = th->t_blocks_allocated ; | |
216 | int err; | |
217 | ||
218 | BUG_ON (!th->t_trans_id); | |
219 | BUG_ON (!th->t_refcount); | |
220 | ||
221 | /* we cannot restart while nested */ | |
222 | if (th->t_refcount > 1) { | |
223 | return 0 ; | |
224 | } | |
225 | pathrelse(path) ; | |
226 | reiserfs_update_sd(th, inode) ; | |
227 | err = journal_end(th, s, len) ; | |
228 | if (!err) { | |
229 | err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6) ; | |
230 | if (!err) | |
231 | reiserfs_update_inode_transaction(inode) ; | |
232 | } | |
233 | return err; | |
234 | } | |
235 | ||
236 | // it is called by get_block when create == 0. Returns block number | |
237 | // for 'block'-th logical block of file. When it hits direct item it | |
238 | // returns 0 (being called from bmap) or read direct item into piece | |
239 | // of page (bh_result) | |
240 | ||
241 | // Please improve the english/clarity in the comment above, as it is | |
242 | // hard to understand. | |
243 | ||
244 | static int _get_block_create_0 (struct inode * inode, long block, | |
245 | struct buffer_head * bh_result, | |
246 | int args) | |
247 | { | |
248 | INITIALIZE_PATH (path); | |
249 | struct cpu_key key; | |
250 | struct buffer_head * bh; | |
251 | struct item_head * ih, tmp_ih; | |
252 | int fs_gen ; | |
253 | int blocknr; | |
254 | char * p = NULL; | |
255 | int chars; | |
256 | int ret ; | |
257 | int done = 0 ; | |
258 | unsigned long offset ; | |
259 | ||
260 | // prepare the key to look for the 'block'-th block of file | |
261 | make_cpu_key (&key, inode, | |
262 | (loff_t)block * inode->i_sb->s_blocksize + 1, TYPE_ANY, 3); | |
263 | ||
264 | research: | |
265 | if (search_for_position_by_key (inode->i_sb, &key, &path) != POSITION_FOUND) { | |
266 | pathrelse (&path); | |
267 | if (p) | |
268 | kunmap(bh_result->b_page) ; | |
269 | // We do not return -ENOENT if there is a hole but page is uptodate, because it means | |
270 | // That there is some MMAPED data associated with it that is yet to be written to disk. | |
271 | if ((args & GET_BLOCK_NO_HOLE) && !PageUptodate(bh_result->b_page) ) { | |
272 | return -ENOENT ; | |
273 | } | |
274 | return 0 ; | |
275 | } | |
276 | ||
277 | // | |
278 | bh = get_last_bh (&path); | |
279 | ih = get_ih (&path); | |
280 | if (is_indirect_le_ih (ih)) { | |
281 | __u32 * ind_item = (__u32 *)B_I_PITEM (bh, ih); | |
282 | ||
283 | /* FIXME: here we could cache indirect item or part of it in | |
284 | the inode to avoid search_by_key in case of subsequent | |
285 | access to file */ | |
286 | blocknr = get_block_num(ind_item, path.pos_in_item) ; | |
287 | ret = 0 ; | |
288 | if (blocknr) { | |
289 | map_bh(bh_result, inode->i_sb, blocknr); | |
290 | if (path.pos_in_item == ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) { | |
291 | set_buffer_boundary(bh_result); | |
292 | } | |
293 | } else | |
294 | // We do not return -ENOENT if there is a hole but page is uptodate, because it means | |
295 | // That there is some MMAPED data associated with it that is yet to be written to disk. | |
296 | if ((args & GET_BLOCK_NO_HOLE) && !PageUptodate(bh_result->b_page) ) { | |
297 | ret = -ENOENT ; | |
298 | } | |
299 | ||
300 | pathrelse (&path); | |
301 | if (p) | |
302 | kunmap(bh_result->b_page) ; | |
303 | return ret ; | |
304 | } | |
305 | ||
306 | // requested data are in direct item(s) | |
307 | if (!(args & GET_BLOCK_READ_DIRECT)) { | |
308 | // we are called by bmap. FIXME: we can not map block of file | |
309 | // when it is stored in direct item(s) | |
310 | pathrelse (&path); | |
311 | if (p) | |
312 | kunmap(bh_result->b_page) ; | |
313 | return -ENOENT; | |
314 | } | |
315 | ||
316 | /* if we've got a direct item, and the buffer or page was uptodate, | |
317 | ** we don't want to pull data off disk again. skip to the | |
318 | ** end, where we map the buffer and return | |
319 | */ | |
320 | if (buffer_uptodate(bh_result)) { | |
321 | goto finished ; | |
322 | } else | |
323 | /* | |
324 | ** grab_tail_page can trigger calls to reiserfs_get_block on up to date | |
325 | ** pages without any buffers. If the page is up to date, we don't want | |
326 | ** read old data off disk. Set the up to date bit on the buffer instead | |
327 | ** and jump to the end | |
328 | */ | |
329 | if (!bh_result->b_page || PageUptodate(bh_result->b_page)) { | |
330 | set_buffer_uptodate(bh_result); | |
331 | goto finished ; | |
332 | } | |
333 | ||
334 | // read file tail into part of page | |
335 | offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1) ; | |
336 | fs_gen = get_generation(inode->i_sb) ; | |
337 | copy_item_head (&tmp_ih, ih); | |
338 | ||
339 | /* we only want to kmap if we are reading the tail into the page. | |
340 | ** this is not the common case, so we don't kmap until we are | |
341 | ** sure we need to. But, this means the item might move if | |
342 | ** kmap schedules | |
343 | */ | |
344 | if (!p) { | |
345 | p = (char *)kmap(bh_result->b_page) ; | |
346 | if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) { | |
347 | goto research; | |
348 | } | |
349 | } | |
350 | p += offset ; | |
351 | memset (p, 0, inode->i_sb->s_blocksize); | |
352 | do { | |
353 | if (!is_direct_le_ih (ih)) { | |
354 | BUG (); | |
355 | } | |
356 | /* make sure we don't read more bytes than actually exist in | |
357 | ** the file. This can happen in odd cases where i_size isn't | |
358 | ** correct, and when direct item padding results in a few | |
359 | ** extra bytes at the end of the direct item | |
360 | */ | |
361 | if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size) | |
362 | break ; | |
363 | if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) { | |
364 | chars = inode->i_size - (le_ih_k_offset(ih) - 1) - path.pos_in_item; | |
365 | done = 1 ; | |
366 | } else { | |
367 | chars = ih_item_len(ih) - path.pos_in_item; | |
368 | } | |
369 | memcpy (p, B_I_PITEM (bh, ih) + path.pos_in_item, chars); | |
370 | ||
371 | if (done) | |
372 | break ; | |
373 | ||
374 | p += chars; | |
375 | ||
376 | if (PATH_LAST_POSITION (&path) != (B_NR_ITEMS (bh) - 1)) | |
377 | // we done, if read direct item is not the last item of | |
378 | // node FIXME: we could try to check right delimiting key | |
379 | // to see whether direct item continues in the right | |
380 | // neighbor or rely on i_size | |
381 | break; | |
382 | ||
383 | // update key to look for the next piece | |
384 | set_cpu_key_k_offset (&key, cpu_key_k_offset (&key) + chars); | |
385 | if (search_for_position_by_key (inode->i_sb, &key, &path) != POSITION_FOUND) | |
386 | // we read something from tail, even if now we got IO_ERROR | |
387 | break; | |
388 | bh = get_last_bh (&path); | |
389 | ih = get_ih (&path); | |
390 | } while (1); | |
391 | ||
392 | flush_dcache_page(bh_result->b_page) ; | |
393 | kunmap(bh_result->b_page) ; | |
394 | ||
395 | finished: | |
396 | pathrelse (&path); | |
397 | /* this buffer has valid data, but isn't valid for io. mapping it to | |
398 | * block #0 tells the rest of reiserfs it just has a tail in it | |
399 | */ | |
400 | map_bh(bh_result, inode->i_sb, 0); | |
401 | set_buffer_uptodate (bh_result); | |
402 | return 0; | |
403 | } | |
404 | ||
405 | ||
406 | // this is called to create file map. So, _get_block_create_0 will not | |
407 | // read direct item | |
408 | static int reiserfs_bmap (struct inode * inode, sector_t block, | |
409 | struct buffer_head * bh_result, int create) | |
410 | { | |
411 | if (!file_capable (inode, block)) | |
412 | return -EFBIG; | |
413 | ||
414 | reiserfs_write_lock(inode->i_sb); | |
415 | /* do not read the direct item */ | |
416 | _get_block_create_0 (inode, block, bh_result, 0) ; | |
417 | reiserfs_write_unlock(inode->i_sb); | |
418 | return 0; | |
419 | } | |
420 | ||
421 | /* special version of get_block that is only used by grab_tail_page right | |
422 | ** now. It is sent to block_prepare_write, and when you try to get a | |
423 | ** block past the end of the file (or a block from a hole) it returns | |
424 | ** -ENOENT instead of a valid buffer. block_prepare_write expects to | |
425 | ** be able to do i/o on the buffers returned, unless an error value | |
426 | ** is also returned. | |
427 | ** | |
428 | ** So, this allows block_prepare_write to be used for reading a single block | |
429 | ** in a page. Where it does not produce a valid page for holes, or past the | |
430 | ** end of the file. This turns out to be exactly what we need for reading | |
431 | ** tails for conversion. | |
432 | ** | |
433 | ** The point of the wrapper is forcing a certain value for create, even | |
434 | ** though the VFS layer is calling this function with create==1. If you | |
435 | ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block, | |
436 | ** don't use this function. | |
437 | */ | |
438 | static int reiserfs_get_block_create_0 (struct inode * inode, sector_t block, | |
439 | struct buffer_head * bh_result, int create) { | |
440 | return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE) ; | |
441 | } | |
442 | ||
443 | /* This is special helper for reiserfs_get_block in case we are executing | |
444 | direct_IO request. */ | |
445 | static int reiserfs_get_blocks_direct_io(struct inode *inode, | |
446 | sector_t iblock, | |
447 | unsigned long max_blocks, | |
448 | struct buffer_head *bh_result, | |
449 | int create) | |
450 | { | |
451 | int ret ; | |
452 | ||
453 | bh_result->b_page = NULL; | |
454 | ||
455 | /* We set the b_size before reiserfs_get_block call since it is | |
456 | referenced in convert_tail_for_hole() that may be called from | |
457 | reiserfs_get_block() */ | |
458 | bh_result->b_size = (1 << inode->i_blkbits); | |
459 | ||
460 | ret = reiserfs_get_block(inode, iblock, bh_result, | |
461 | create | GET_BLOCK_NO_DANGLE) ; | |
462 | if (ret) | |
463 | goto out; | |
464 | ||
465 | /* don't allow direct io onto tail pages */ | |
466 | if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) { | |
467 | /* make sure future calls to the direct io funcs for this offset | |
468 | ** in the file fail by unmapping the buffer | |
469 | */ | |
470 | clear_buffer_mapped(bh_result); | |
471 | ret = -EINVAL ; | |
472 | } | |
473 | /* Possible unpacked tail. Flush the data before pages have | |
474 | disappeared */ | |
475 | if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) { | |
476 | int err; | |
477 | lock_kernel(); | |
478 | err = reiserfs_commit_for_inode(inode); | |
479 | REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask; | |
480 | unlock_kernel(); | |
481 | if (err < 0) | |
482 | ret = err; | |
483 | } | |
484 | out: | |
485 | return ret ; | |
486 | } | |
487 | ||
488 | ||
489 | /* | |
490 | ** helper function for when reiserfs_get_block is called for a hole | |
491 | ** but the file tail is still in a direct item | |
492 | ** bh_result is the buffer head for the hole | |
493 | ** tail_offset is the offset of the start of the tail in the file | |
494 | ** | |
495 | ** This calls prepare_write, which will start a new transaction | |
496 | ** you should not be in a transaction, or have any paths held when you | |
497 | ** call this. | |
498 | */ | |
499 | static int convert_tail_for_hole(struct inode *inode, | |
500 | struct buffer_head *bh_result, | |
501 | loff_t tail_offset) { | |
502 | unsigned long index ; | |
503 | unsigned long tail_end ; | |
504 | unsigned long tail_start ; | |
505 | struct page * tail_page ; | |
506 | struct page * hole_page = bh_result->b_page ; | |
507 | int retval = 0 ; | |
508 | ||
509 | if ((tail_offset & (bh_result->b_size - 1)) != 1) | |
510 | return -EIO ; | |
511 | ||
512 | /* always try to read until the end of the block */ | |
513 | tail_start = tail_offset & (PAGE_CACHE_SIZE - 1) ; | |
514 | tail_end = (tail_start | (bh_result->b_size - 1)) + 1 ; | |
515 | ||
516 | index = tail_offset >> PAGE_CACHE_SHIFT ; | |
517 | /* hole_page can be zero in case of direct_io, we are sure | |
518 | that we cannot get here if we write with O_DIRECT into | |
519 | tail page */ | |
520 | if (!hole_page || index != hole_page->index) { | |
521 | tail_page = grab_cache_page(inode->i_mapping, index) ; | |
522 | retval = -ENOMEM; | |
523 | if (!tail_page) { | |
524 | goto out ; | |
525 | } | |
526 | } else { | |
527 | tail_page = hole_page ; | |
528 | } | |
529 | ||
530 | /* we don't have to make sure the conversion did not happen while | |
531 | ** we were locking the page because anyone that could convert | |
532 | ** must first take i_sem. | |
533 | ** | |
534 | ** We must fix the tail page for writing because it might have buffers | |
535 | ** that are mapped, but have a block number of 0. This indicates tail | |
536 | ** data that has been read directly into the page, and block_prepare_write | |
537 | ** won't trigger a get_block in this case. | |
538 | */ | |
539 | fix_tail_page_for_writing(tail_page) ; | |
540 | retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end); | |
541 | if (retval) | |
542 | goto unlock ; | |
543 | ||
544 | /* tail conversion might change the data in the page */ | |
545 | flush_dcache_page(tail_page) ; | |
546 | ||
547 | retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end) ; | |
548 | ||
549 | unlock: | |
550 | if (tail_page != hole_page) { | |
551 | unlock_page(tail_page) ; | |
552 | page_cache_release(tail_page) ; | |
553 | } | |
554 | out: | |
555 | return retval ; | |
556 | } | |
557 | ||
558 | static inline int _allocate_block(struct reiserfs_transaction_handle *th, | |
559 | long block, | |
560 | struct inode *inode, | |
561 | b_blocknr_t *allocated_block_nr, | |
562 | struct path * path, | |
563 | int flags) { | |
564 | BUG_ON (!th->t_trans_id); | |
565 | ||
566 | #ifdef REISERFS_PREALLOCATE | |
567 | if (!(flags & GET_BLOCK_NO_ISEM)) { | |
568 | return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr, path, block); | |
569 | } | |
570 | #endif | |
571 | return reiserfs_new_unf_blocknrs (th, inode, allocated_block_nr, path, block); | |
572 | } | |
573 | ||
574 | int reiserfs_get_block (struct inode * inode, sector_t block, | |
575 | struct buffer_head * bh_result, int create) | |
576 | { | |
577 | int repeat, retval = 0; | |
578 | b_blocknr_t allocated_block_nr = 0;// b_blocknr_t is (unsigned) 32 bit int | |
579 | INITIALIZE_PATH(path); | |
580 | int pos_in_item; | |
581 | struct cpu_key key; | |
582 | struct buffer_head * bh, * unbh = NULL; | |
583 | struct item_head * ih, tmp_ih; | |
584 | __u32 * item; | |
585 | int done; | |
586 | int fs_gen; | |
587 | struct reiserfs_transaction_handle *th = NULL; | |
588 | /* space reserved in transaction batch: | |
589 | . 3 balancings in direct->indirect conversion | |
590 | . 1 block involved into reiserfs_update_sd() | |
591 | XXX in practically impossible worst case direct2indirect() | |
592 | can incur (much) more than 3 balancings. | |
593 | quota update for user, group */ | |
594 | int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS; | |
595 | int version; | |
596 | int dangle = 1; | |
597 | loff_t new_offset = (((loff_t)block) << inode->i_sb->s_blocksize_bits) + 1 ; | |
598 | ||
599 | /* bad.... */ | |
600 | reiserfs_write_lock(inode->i_sb); | |
601 | version = get_inode_item_key_version (inode); | |
602 | ||
603 | if (block < 0) { | |
604 | reiserfs_write_unlock(inode->i_sb); | |
605 | return -EIO; | |
606 | } | |
607 | ||
608 | if (!file_capable (inode, block)) { | |
609 | reiserfs_write_unlock(inode->i_sb); | |
610 | return -EFBIG; | |
611 | } | |
612 | ||
613 | /* if !create, we aren't changing the FS, so we don't need to | |
614 | ** log anything, so we don't need to start a transaction | |
615 | */ | |
616 | if (!(create & GET_BLOCK_CREATE)) { | |
617 | int ret ; | |
618 | /* find number of block-th logical block of the file */ | |
619 | ret = _get_block_create_0 (inode, block, bh_result, | |
620 | create | GET_BLOCK_READ_DIRECT) ; | |
621 | reiserfs_write_unlock(inode->i_sb); | |
622 | return ret; | |
623 | } | |
624 | /* | |
625 | * if we're already in a transaction, make sure to close | |
626 | * any new transactions we start in this func | |
627 | */ | |
628 | if ((create & GET_BLOCK_NO_DANGLE) || | |
629 | reiserfs_transaction_running(inode->i_sb)) | |
630 | dangle = 0; | |
631 | ||
632 | /* If file is of such a size, that it might have a tail and tails are enabled | |
633 | ** we should mark it as possibly needing tail packing on close | |
634 | */ | |
635 | if ( (have_large_tails (inode->i_sb) && inode->i_size < i_block_size (inode)*4) || | |
636 | (have_small_tails (inode->i_sb) && inode->i_size < i_block_size(inode)) ) | |
637 | REISERFS_I(inode)->i_flags |= i_pack_on_close_mask ; | |
638 | ||
639 | /* set the key of the first byte in the 'block'-th block of file */ | |
640 | make_cpu_key (&key, inode, new_offset, | |
641 | TYPE_ANY, 3/*key length*/); | |
642 | if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) { | |
643 | start_trans: | |
644 | th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count); | |
645 | if (!th) { | |
646 | retval = -ENOMEM; | |
647 | goto failure; | |
648 | } | |
649 | reiserfs_update_inode_transaction(inode) ; | |
650 | } | |
651 | research: | |
652 | ||
653 | retval = search_for_position_by_key (inode->i_sb, &key, &path); | |
654 | if (retval == IO_ERROR) { | |
655 | retval = -EIO; | |
656 | goto failure; | |
657 | } | |
658 | ||
659 | bh = get_last_bh (&path); | |
660 | ih = get_ih (&path); | |
661 | item = get_item (&path); | |
662 | pos_in_item = path.pos_in_item; | |
663 | ||
664 | fs_gen = get_generation (inode->i_sb); | |
665 | copy_item_head (&tmp_ih, ih); | |
666 | ||
667 | if (allocation_needed (retval, allocated_block_nr, ih, item, pos_in_item)) { | |
668 | /* we have to allocate block for the unformatted node */ | |
669 | if (!th) { | |
670 | pathrelse(&path) ; | |
671 | goto start_trans; | |
672 | } | |
673 | ||
674 | repeat = _allocate_block(th, block, inode, &allocated_block_nr, &path, create); | |
675 | ||
676 | if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) { | |
677 | /* restart the transaction to give the journal a chance to free | |
678 | ** some blocks. releases the path, so we have to go back to | |
679 | ** research if we succeed on the second try | |
680 | */ | |
681 | SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1; | |
682 | retval = restart_transaction(th, inode, &path) ; | |
683 | if (retval) | |
684 | goto failure; | |
685 | repeat = _allocate_block(th, block, inode, &allocated_block_nr, NULL, create); | |
686 | ||
687 | if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) { | |
688 | goto research ; | |
689 | } | |
690 | if (repeat == QUOTA_EXCEEDED) | |
691 | retval = -EDQUOT; | |
692 | else | |
693 | retval = -ENOSPC; | |
694 | goto failure; | |
695 | } | |
696 | ||
697 | if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) { | |
698 | goto research; | |
699 | } | |
700 | } | |
701 | ||
702 | if (indirect_item_found (retval, ih)) { | |
703 | b_blocknr_t unfm_ptr; | |
704 | /* 'block'-th block is in the file already (there is | |
705 | corresponding cell in some indirect item). But it may be | |
706 | zero unformatted node pointer (hole) */ | |
707 | unfm_ptr = get_block_num (item, pos_in_item); | |
708 | if (unfm_ptr == 0) { | |
709 | /* use allocated block to plug the hole */ | |
710 | reiserfs_prepare_for_journal(inode->i_sb, bh, 1) ; | |
711 | if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) { | |
712 | reiserfs_restore_prepared_buffer(inode->i_sb, bh) ; | |
713 | goto research; | |
714 | } | |
715 | set_buffer_new(bh_result); | |
716 | if (buffer_dirty(bh_result) && reiserfs_data_ordered(inode->i_sb)) | |
717 | reiserfs_add_ordered_list(inode, bh_result); | |
718 | put_block_num(item, pos_in_item, allocated_block_nr) ; | |
719 | unfm_ptr = allocated_block_nr; | |
720 | journal_mark_dirty (th, inode->i_sb, bh); | |
721 | reiserfs_update_sd(th, inode) ; | |
722 | } | |
723 | set_block_dev_mapped(bh_result, unfm_ptr, inode); | |
724 | pathrelse (&path); | |
725 | retval = 0; | |
726 | if (!dangle && th) | |
727 | retval = reiserfs_end_persistent_transaction(th); | |
728 | ||
729 | reiserfs_write_unlock(inode->i_sb); | |
730 | ||
731 | /* the item was found, so new blocks were not added to the file | |
732 | ** there is no need to make sure the inode is updated with this | |
733 | ** transaction | |
734 | */ | |
735 | return retval; | |
736 | } | |
737 | ||
738 | if (!th) { | |
739 | pathrelse(&path) ; | |
740 | goto start_trans; | |
741 | } | |
742 | ||
743 | /* desired position is not found or is in the direct item. We have | |
744 | to append file with holes up to 'block'-th block converting | |
745 | direct items to indirect one if necessary */ | |
746 | done = 0; | |
747 | do { | |
748 | if (is_statdata_le_ih (ih)) { | |
749 | __u32 unp = 0; | |
750 | struct cpu_key tmp_key; | |
751 | ||
752 | /* indirect item has to be inserted */ | |
753 | make_le_item_head (&tmp_ih, &key, version, 1, TYPE_INDIRECT, | |
754 | UNFM_P_SIZE, 0/* free_space */); | |
755 | ||
756 | if (cpu_key_k_offset (&key) == 1) { | |
757 | /* we are going to add 'block'-th block to the file. Use | |
758 | allocated block for that */ | |
759 | unp = cpu_to_le32 (allocated_block_nr); | |
760 | set_block_dev_mapped (bh_result, allocated_block_nr, inode); | |
761 | set_buffer_new(bh_result); | |
762 | done = 1; | |
763 | } | |
764 | tmp_key = key; // ;) | |
765 | set_cpu_key_k_offset (&tmp_key, 1); | |
766 | PATH_LAST_POSITION(&path) ++; | |
767 | ||
768 | retval = reiserfs_insert_item (th, &path, &tmp_key, &tmp_ih, inode, (char *)&unp); | |
769 | if (retval) { | |
770 | reiserfs_free_block (th, inode, allocated_block_nr, 1); | |
771 | goto failure; // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST | |
772 | } | |
773 | //mark_tail_converted (inode); | |
774 | } else if (is_direct_le_ih (ih)) { | |
775 | /* direct item has to be converted */ | |
776 | loff_t tail_offset; | |
777 | ||
778 | tail_offset = ((le_ih_k_offset (ih) - 1) & ~(inode->i_sb->s_blocksize - 1)) + 1; | |
779 | if (tail_offset == cpu_key_k_offset (&key)) { | |
780 | /* direct item we just found fits into block we have | |
781 | to map. Convert it into unformatted node: use | |
782 | bh_result for the conversion */ | |
783 | set_block_dev_mapped (bh_result, allocated_block_nr, inode); | |
784 | unbh = bh_result; | |
785 | done = 1; | |
786 | } else { | |
787 | /* we have to padd file tail stored in direct item(s) | |
788 | up to block size and convert it to unformatted | |
789 | node. FIXME: this should also get into page cache */ | |
790 | ||
791 | pathrelse(&path) ; | |
792 | /* | |
793 | * ugly, but we can only end the transaction if | |
794 | * we aren't nested | |
795 | */ | |
796 | BUG_ON (!th->t_refcount); | |
797 | if (th->t_refcount == 1) { | |
798 | retval = reiserfs_end_persistent_transaction(th); | |
799 | th = NULL; | |
800 | if (retval) | |
801 | goto failure; | |
802 | } | |
803 | ||
804 | retval = convert_tail_for_hole(inode, bh_result, tail_offset) ; | |
805 | if (retval) { | |
806 | if ( retval != -ENOSPC ) | |
807 | reiserfs_warning (inode->i_sb, "clm-6004: convert tail failed inode %lu, error %d", inode->i_ino, retval) ; | |
808 | if (allocated_block_nr) { | |
809 | /* the bitmap, the super, and the stat data == 3 */ | |
810 | if (!th) | |
811 | th = reiserfs_persistent_transaction(inode->i_sb,3); | |
812 | if (th) | |
813 | reiserfs_free_block (th,inode,allocated_block_nr,1); | |
814 | } | |
815 | goto failure ; | |
816 | } | |
817 | goto research ; | |
818 | } | |
819 | retval = direct2indirect (th, inode, &path, unbh, tail_offset); | |
820 | if (retval) { | |
821 | reiserfs_unmap_buffer(unbh); | |
822 | reiserfs_free_block (th, inode, allocated_block_nr, 1); | |
823 | goto failure; | |
824 | } | |
825 | /* it is important the set_buffer_uptodate is done after | |
826 | ** the direct2indirect. The buffer might contain valid | |
827 | ** data newer than the data on disk (read by readpage, changed, | |
828 | ** and then sent here by writepage). direct2indirect needs | |
829 | ** to know if unbh was already up to date, so it can decide | |
830 | ** if the data in unbh needs to be replaced with data from | |
831 | ** the disk | |
832 | */ | |
833 | set_buffer_uptodate (unbh); | |
834 | ||
835 | /* unbh->b_page == NULL in case of DIRECT_IO request, this means | |
836 | buffer will disappear shortly, so it should not be added to | |
837 | */ | |
838 | if ( unbh->b_page ) { | |
839 | /* we've converted the tail, so we must | |
840 | ** flush unbh before the transaction commits | |
841 | */ | |
842 | reiserfs_add_tail_list(inode, unbh) ; | |
843 | ||
844 | /* mark it dirty now to prevent commit_write from adding | |
845 | ** this buffer to the inode's dirty buffer list | |
846 | */ | |
847 | /* | |
848 | * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty(). | |
849 | * It's still atomic, but it sets the page dirty too, | |
850 | * which makes it eligible for writeback at any time by the | |
851 | * VM (which was also the case with __mark_buffer_dirty()) | |
852 | */ | |
853 | mark_buffer_dirty(unbh) ; | |
854 | } | |
855 | } else { | |
856 | /* append indirect item with holes if needed, when appending | |
857 | pointer to 'block'-th block use block, which is already | |
858 | allocated */ | |
859 | struct cpu_key tmp_key; | |
860 | unp_t unf_single=0; // We use this in case we need to allocate only | |
861 | // one block which is a fastpath | |
862 | unp_t *un; | |
863 | __u64 max_to_insert=MAX_ITEM_LEN(inode->i_sb->s_blocksize)/UNFM_P_SIZE; | |
864 | __u64 blocks_needed; | |
865 | ||
866 | RFALSE( pos_in_item != ih_item_len(ih) / UNFM_P_SIZE, | |
867 | "vs-804: invalid position for append"); | |
868 | /* indirect item has to be appended, set up key of that position */ | |
869 | make_cpu_key (&tmp_key, inode, | |
870 | le_key_k_offset (version, &(ih->ih_key)) + op_bytes_number (ih, inode->i_sb->s_blocksize), | |
871 | //pos_in_item * inode->i_sb->s_blocksize, | |
872 | TYPE_INDIRECT, 3);// key type is unimportant | |
873 | ||
874 | blocks_needed = 1 + ((cpu_key_k_offset (&key) - cpu_key_k_offset (&tmp_key)) >> inode->i_sb->s_blocksize_bits); | |
875 | RFALSE( blocks_needed < 0, "green-805: invalid offset"); | |
876 | ||
877 | if ( blocks_needed == 1 ) { | |
878 | un = &unf_single; | |
879 | } else { | |
880 | un=kmalloc( min(blocks_needed,max_to_insert)*UNFM_P_SIZE, | |
881 | GFP_ATOMIC); // We need to avoid scheduling. | |
882 | if ( !un) { | |
883 | un = &unf_single; | |
884 | blocks_needed = 1; | |
885 | max_to_insert = 0; | |
886 | } else | |
887 | memset(un, 0, UNFM_P_SIZE * min(blocks_needed,max_to_insert)); | |
888 | } | |
889 | if ( blocks_needed <= max_to_insert) { | |
890 | /* we are going to add target block to the file. Use allocated | |
891 | block for that */ | |
892 | un[blocks_needed-1] = cpu_to_le32 (allocated_block_nr); | |
893 | set_block_dev_mapped (bh_result, allocated_block_nr, inode); | |
894 | set_buffer_new(bh_result); | |
895 | done = 1; | |
896 | } else { | |
897 | /* paste hole to the indirect item */ | |
898 | /* If kmalloc failed, max_to_insert becomes zero and it means we | |
899 | only have space for one block */ | |
900 | blocks_needed=max_to_insert?max_to_insert:1; | |
901 | } | |
902 | retval = reiserfs_paste_into_item (th, &path, &tmp_key, inode, (char *)un, UNFM_P_SIZE * blocks_needed); | |
903 | ||
904 | if (blocks_needed != 1) | |
905 | kfree(un); | |
906 | ||
907 | if (retval) { | |
908 | reiserfs_free_block (th, inode, allocated_block_nr, 1); | |
909 | goto failure; | |
910 | } | |
911 | if (!done) { | |
912 | /* We need to mark new file size in case this function will be | |
913 | interrupted/aborted later on. And we may do this only for | |
914 | holes. */ | |
915 | inode->i_size += inode->i_sb->s_blocksize * blocks_needed; | |
916 | } | |
917 | } | |
918 | ||
919 | if (done == 1) | |
920 | break; | |
921 | ||
922 | /* this loop could log more blocks than we had originally asked | |
923 | ** for. So, we have to allow the transaction to end if it is | |
924 | ** too big or too full. Update the inode so things are | |
925 | ** consistent if we crash before the function returns | |
926 | ** | |
927 | ** release the path so that anybody waiting on the path before | |
928 | ** ending their transaction will be able to continue. | |
929 | */ | |
930 | if (journal_transaction_should_end(th, th->t_blocks_allocated)) { | |
931 | retval = restart_transaction(th, inode, &path) ; | |
932 | if (retval) | |
933 | goto failure; | |
934 | } | |
935 | /* inserting indirect pointers for a hole can take a | |
936 | ** long time. reschedule if needed | |
937 | */ | |
938 | cond_resched(); | |
939 | ||
940 | retval = search_for_position_by_key (inode->i_sb, &key, &path); | |
941 | if (retval == IO_ERROR) { | |
942 | retval = -EIO; | |
943 | goto failure; | |
944 | } | |
945 | if (retval == POSITION_FOUND) { | |
946 | reiserfs_warning (inode->i_sb, "vs-825: reiserfs_get_block: " | |
947 | "%K should not be found", &key); | |
948 | retval = -EEXIST; | |
949 | if (allocated_block_nr) | |
950 | reiserfs_free_block (th, inode, allocated_block_nr, 1); | |
951 | pathrelse(&path) ; | |
952 | goto failure; | |
953 | } | |
954 | bh = get_last_bh (&path); | |
955 | ih = get_ih (&path); | |
956 | item = get_item (&path); | |
957 | pos_in_item = path.pos_in_item; | |
958 | } while (1); | |
959 | ||
960 | ||
961 | retval = 0; | |
962 | ||
963 | failure: | |
964 | if (th && (!dangle || (retval && !th->t_trans_id))) { | |
965 | int err; | |
966 | if (th->t_trans_id) | |
967 | reiserfs_update_sd(th, inode); | |
968 | err = reiserfs_end_persistent_transaction(th); | |
969 | if (err) | |
970 | retval = err; | |
971 | } | |
972 | ||
973 | reiserfs_write_unlock(inode->i_sb); | |
974 | reiserfs_check_path(&path) ; | |
975 | return retval; | |
976 | } | |
977 | ||
978 | static int | |
979 | reiserfs_readpages(struct file *file, struct address_space *mapping, | |
980 | struct list_head *pages, unsigned nr_pages) | |
981 | { | |
982 | return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block); | |
983 | } | |
984 | ||
985 | /* Compute real number of used bytes by file | |
986 | * Following three functions can go away when we'll have enough space in stat item | |
987 | */ | |
988 | static int real_space_diff(struct inode *inode, int sd_size) | |
989 | { | |
990 | int bytes; | |
991 | loff_t blocksize = inode->i_sb->s_blocksize ; | |
992 | ||
993 | if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) | |
994 | return sd_size ; | |
995 | ||
996 | /* End of file is also in full block with indirect reference, so round | |
997 | ** up to the next block. | |
998 | ** | |
999 | ** there is just no way to know if the tail is actually packed | |
1000 | ** on the file, so we have to assume it isn't. When we pack the | |
1001 | ** tail, we add 4 bytes to pretend there really is an unformatted | |
1002 | ** node pointer | |
1003 | */ | |
1004 | bytes = ((inode->i_size + (blocksize-1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE + sd_size; | |
1005 | return bytes ; | |
1006 | } | |
1007 | ||
1008 | static inline loff_t to_real_used_space(struct inode *inode, ulong blocks, | |
1009 | int sd_size) | |
1010 | { | |
1011 | if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) { | |
1012 | return inode->i_size + (loff_t)(real_space_diff(inode, sd_size)) ; | |
1013 | } | |
1014 | return ((loff_t)real_space_diff(inode, sd_size)) + (((loff_t)blocks) << 9); | |
1015 | } | |
1016 | ||
1017 | /* Compute number of blocks used by file in ReiserFS counting */ | |
1018 | static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size) | |
1019 | { | |
1020 | loff_t bytes = inode_get_bytes(inode) ; | |
1021 | loff_t real_space = real_space_diff(inode, sd_size) ; | |
1022 | ||
1023 | /* keeps fsck and non-quota versions of reiserfs happy */ | |
1024 | if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) { | |
1025 | bytes += (loff_t)511 ; | |
1026 | } | |
1027 | ||
1028 | /* files from before the quota patch might i_blocks such that | |
1029 | ** bytes < real_space. Deal with that here to prevent it from | |
1030 | ** going negative. | |
1031 | */ | |
1032 | if (bytes < real_space) | |
1033 | return 0 ; | |
1034 | return (bytes - real_space) >> 9; | |
1035 | } | |
1036 | ||
1037 | // | |
1038 | // BAD: new directories have stat data of new type and all other items | |
1039 | // of old type. Version stored in the inode says about body items, so | |
1040 | // in update_stat_data we can not rely on inode, but have to check | |
1041 | // item version directly | |
1042 | // | |
1043 | ||
1044 | // called by read_locked_inode | |
1045 | static void init_inode (struct inode * inode, struct path * path) | |
1046 | { | |
1047 | struct buffer_head * bh; | |
1048 | struct item_head * ih; | |
1049 | __u32 rdev; | |
1050 | //int version = ITEM_VERSION_1; | |
1051 | ||
1052 | bh = PATH_PLAST_BUFFER (path); | |
1053 | ih = PATH_PITEM_HEAD (path); | |
1054 | ||
1055 | ||
1056 | copy_key (INODE_PKEY (inode), &(ih->ih_key)); | |
1057 | inode->i_blksize = reiserfs_default_io_size; | |
1058 | ||
1059 | INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list )); | |
1060 | REISERFS_I(inode)->i_flags = 0; | |
1061 | REISERFS_I(inode)->i_prealloc_block = 0; | |
1062 | REISERFS_I(inode)->i_prealloc_count = 0; | |
1063 | REISERFS_I(inode)->i_trans_id = 0; | |
1064 | REISERFS_I(inode)->i_jl = NULL; | |
1065 | REISERFS_I(inode)->i_acl_access = NULL; | |
1066 | REISERFS_I(inode)->i_acl_default = NULL; | |
1067 | init_rwsem (&REISERFS_I(inode)->xattr_sem); | |
1068 | ||
1069 | if (stat_data_v1 (ih)) { | |
1070 | struct stat_data_v1 * sd = (struct stat_data_v1 *)B_I_PITEM (bh, ih); | |
1071 | unsigned long blocks; | |
1072 | ||
1073 | set_inode_item_key_version (inode, KEY_FORMAT_3_5); | |
1074 | set_inode_sd_version (inode, STAT_DATA_V1); | |
1075 | inode->i_mode = sd_v1_mode(sd); | |
1076 | inode->i_nlink = sd_v1_nlink(sd); | |
1077 | inode->i_uid = sd_v1_uid(sd); | |
1078 | inode->i_gid = sd_v1_gid(sd); | |
1079 | inode->i_size = sd_v1_size(sd); | |
1080 | inode->i_atime.tv_sec = sd_v1_atime(sd); | |
1081 | inode->i_mtime.tv_sec = sd_v1_mtime(sd); | |
1082 | inode->i_ctime.tv_sec = sd_v1_ctime(sd); | |
1083 | inode->i_atime.tv_nsec = 0; | |
1084 | inode->i_ctime.tv_nsec = 0; | |
1085 | inode->i_mtime.tv_nsec = 0; | |
1086 | ||
1087 | inode->i_blocks = sd_v1_blocks(sd); | |
1088 | inode->i_generation = le32_to_cpu (INODE_PKEY (inode)->k_dir_id); | |
1089 | blocks = (inode->i_size + 511) >> 9; | |
1090 | blocks = _ROUND_UP (blocks, inode->i_sb->s_blocksize >> 9); | |
1091 | if (inode->i_blocks > blocks) { | |
1092 | // there was a bug in <=3.5.23 when i_blocks could take negative | |
1093 | // values. Starting from 3.5.17 this value could even be stored in | |
1094 | // stat data. For such files we set i_blocks based on file | |
1095 | // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be | |
1096 | // only updated if file's inode will ever change | |
1097 | inode->i_blocks = blocks; | |
1098 | } | |
1099 | ||
1100 | rdev = sd_v1_rdev(sd); | |
1101 | REISERFS_I(inode)->i_first_direct_byte = sd_v1_first_direct_byte(sd); | |
1102 | /* an early bug in the quota code can give us an odd number for the | |
1103 | ** block count. This is incorrect, fix it here. | |
1104 | */ | |
1105 | if (inode->i_blocks & 1) { | |
1106 | inode->i_blocks++ ; | |
1107 | } | |
1108 | inode_set_bytes(inode, to_real_used_space(inode, inode->i_blocks, | |
1109 | SD_V1_SIZE)); | |
1110 | /* nopack is initially zero for v1 objects. For v2 objects, | |
1111 | nopack is initialised from sd_attrs */ | |
1112 | REISERFS_I(inode)->i_flags &= ~i_nopack_mask; | |
1113 | } else { | |
1114 | // new stat data found, but object may have old items | |
1115 | // (directories and symlinks) | |
1116 | struct stat_data * sd = (struct stat_data *)B_I_PITEM (bh, ih); | |
1117 | ||
1118 | inode->i_mode = sd_v2_mode(sd); | |
1119 | inode->i_nlink = sd_v2_nlink(sd); | |
1120 | inode->i_uid = sd_v2_uid(sd); | |
1121 | inode->i_size = sd_v2_size(sd); | |
1122 | inode->i_gid = sd_v2_gid(sd); | |
1123 | inode->i_mtime.tv_sec = sd_v2_mtime(sd); | |
1124 | inode->i_atime.tv_sec = sd_v2_atime(sd); | |
1125 | inode->i_ctime.tv_sec = sd_v2_ctime(sd); | |
1126 | inode->i_ctime.tv_nsec = 0; | |
1127 | inode->i_mtime.tv_nsec = 0; | |
1128 | inode->i_atime.tv_nsec = 0; | |
1129 | inode->i_blocks = sd_v2_blocks(sd); | |
1130 | rdev = sd_v2_rdev(sd); | |
1131 | if( S_ISCHR( inode -> i_mode ) || S_ISBLK( inode -> i_mode ) ) | |
1132 | inode->i_generation = le32_to_cpu (INODE_PKEY (inode)->k_dir_id); | |
1133 | else | |
1134 | inode->i_generation = sd_v2_generation(sd); | |
1135 | ||
1136 | if (S_ISDIR (inode->i_mode) || S_ISLNK (inode->i_mode)) | |
1137 | set_inode_item_key_version (inode, KEY_FORMAT_3_5); | |
1138 | else | |
1139 | set_inode_item_key_version (inode, KEY_FORMAT_3_6); | |
1140 | REISERFS_I(inode)->i_first_direct_byte = 0; | |
1141 | set_inode_sd_version (inode, STAT_DATA_V2); | |
1142 | inode_set_bytes(inode, to_real_used_space(inode, inode->i_blocks, | |
1143 | SD_V2_SIZE)); | |
1144 | /* read persistent inode attributes from sd and initalise | |
1145 | generic inode flags from them */ | |
1146 | REISERFS_I(inode)->i_attrs = sd_v2_attrs( sd ); | |
1147 | sd_attrs_to_i_attrs( sd_v2_attrs( sd ), inode ); | |
1148 | } | |
1149 | ||
1150 | pathrelse (path); | |
1151 | if (S_ISREG (inode->i_mode)) { | |
1152 | inode->i_op = &reiserfs_file_inode_operations; | |
1153 | inode->i_fop = &reiserfs_file_operations; | |
1154 | inode->i_mapping->a_ops = &reiserfs_address_space_operations ; | |
1155 | } else if (S_ISDIR (inode->i_mode)) { | |
1156 | inode->i_op = &reiserfs_dir_inode_operations; | |
1157 | inode->i_fop = &reiserfs_dir_operations; | |
1158 | } else if (S_ISLNK (inode->i_mode)) { | |
1159 | inode->i_op = &reiserfs_symlink_inode_operations; | |
1160 | inode->i_mapping->a_ops = &reiserfs_address_space_operations; | |
1161 | } else { | |
1162 | inode->i_blocks = 0; | |
1163 | inode->i_op = &reiserfs_special_inode_operations; | |
1164 | init_special_inode(inode, inode->i_mode, new_decode_dev(rdev)); | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | ||
1169 | // update new stat data with inode fields | |
1170 | static void inode2sd (void * sd, struct inode * inode, loff_t size) | |
1171 | { | |
1172 | struct stat_data * sd_v2 = (struct stat_data *)sd; | |
1173 | __u16 flags; | |
1174 | ||
1175 | set_sd_v2_mode(sd_v2, inode->i_mode ); | |
1176 | set_sd_v2_nlink(sd_v2, inode->i_nlink ); | |
1177 | set_sd_v2_uid(sd_v2, inode->i_uid ); | |
1178 | set_sd_v2_size(sd_v2, size ); | |
1179 | set_sd_v2_gid(sd_v2, inode->i_gid ); | |
1180 | set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec ); | |
1181 | set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec ); | |
1182 | set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec ); | |
1183 | set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE)); | |
1184 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) | |
1185 | set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev)); | |
1186 | else | |
1187 | set_sd_v2_generation(sd_v2, inode->i_generation); | |
1188 | flags = REISERFS_I(inode)->i_attrs; | |
1189 | i_attrs_to_sd_attrs( inode, &flags ); | |
1190 | set_sd_v2_attrs( sd_v2, flags ); | |
1191 | } | |
1192 | ||
1193 | ||
1194 | // used to copy inode's fields to old stat data | |
1195 | static void inode2sd_v1 (void * sd, struct inode * inode, loff_t size) | |
1196 | { | |
1197 | struct stat_data_v1 * sd_v1 = (struct stat_data_v1 *)sd; | |
1198 | ||
1199 | set_sd_v1_mode(sd_v1, inode->i_mode ); | |
1200 | set_sd_v1_uid(sd_v1, inode->i_uid ); | |
1201 | set_sd_v1_gid(sd_v1, inode->i_gid ); | |
1202 | set_sd_v1_nlink(sd_v1, inode->i_nlink ); | |
1203 | set_sd_v1_size(sd_v1, size ); | |
1204 | set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec ); | |
1205 | set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec ); | |
1206 | set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec ); | |
1207 | ||
1208 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) | |
1209 | set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev)); | |
1210 | else | |
1211 | set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE)); | |
1212 | ||
1213 | // Sigh. i_first_direct_byte is back | |
1214 | set_sd_v1_first_direct_byte(sd_v1, REISERFS_I(inode)->i_first_direct_byte); | |
1215 | } | |
1216 | ||
1217 | ||
1218 | /* NOTE, you must prepare the buffer head before sending it here, | |
1219 | ** and then log it after the call | |
1220 | */ | |
1221 | static void update_stat_data (struct path * path, struct inode * inode, | |
1222 | loff_t size) | |
1223 | { | |
1224 | struct buffer_head * bh; | |
1225 | struct item_head * ih; | |
1226 | ||
1227 | bh = PATH_PLAST_BUFFER (path); | |
1228 | ih = PATH_PITEM_HEAD (path); | |
1229 | ||
1230 | if (!is_statdata_le_ih (ih)) | |
1231 | reiserfs_panic (inode->i_sb, "vs-13065: update_stat_data: key %k, found item %h", | |
1232 | INODE_PKEY (inode), ih); | |
1233 | ||
1234 | if (stat_data_v1 (ih)) { | |
1235 | // path points to old stat data | |
1236 | inode2sd_v1 (B_I_PITEM (bh, ih), inode, size); | |
1237 | } else { | |
1238 | inode2sd (B_I_PITEM (bh, ih), inode, size); | |
1239 | } | |
1240 | ||
1241 | return; | |
1242 | } | |
1243 | ||
1244 | ||
1245 | void reiserfs_update_sd_size (struct reiserfs_transaction_handle *th, | |
1246 | struct inode * inode, loff_t size) | |
1247 | { | |
1248 | struct cpu_key key; | |
1249 | INITIALIZE_PATH(path); | |
1250 | struct buffer_head *bh ; | |
1251 | int fs_gen ; | |
1252 | struct item_head *ih, tmp_ih ; | |
1253 | int retval; | |
1254 | ||
1255 | BUG_ON (!th->t_trans_id); | |
1256 | ||
1257 | make_cpu_key (&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);//key type is unimportant | |
1258 | ||
1259 | for(;;) { | |
1260 | int pos; | |
1261 | /* look for the object's stat data */ | |
1262 | retval = search_item (inode->i_sb, &key, &path); | |
1263 | if (retval == IO_ERROR) { | |
1264 | reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: " | |
1265 | "i/o failure occurred trying to update %K stat data", | |
1266 | &key); | |
1267 | return; | |
1268 | } | |
1269 | if (retval == ITEM_NOT_FOUND) { | |
1270 | pos = PATH_LAST_POSITION (&path); | |
1271 | pathrelse(&path) ; | |
1272 | if (inode->i_nlink == 0) { | |
1273 | /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found");*/ | |
1274 | return; | |
1275 | } | |
1276 | reiserfs_warning (inode->i_sb, "vs-13060: reiserfs_update_sd: " | |
1277 | "stat data of object %k (nlink == %d) not found (pos %d)", | |
1278 | INODE_PKEY (inode), inode->i_nlink, pos); | |
1279 | reiserfs_check_path(&path) ; | |
1280 | return; | |
1281 | } | |
1282 | ||
1283 | /* sigh, prepare_for_journal might schedule. When it schedules the | |
1284 | ** FS might change. We have to detect that, and loop back to the | |
1285 | ** search if the stat data item has moved | |
1286 | */ | |
1287 | bh = get_last_bh(&path) ; | |
1288 | ih = get_ih(&path) ; | |
1289 | copy_item_head (&tmp_ih, ih); | |
1290 | fs_gen = get_generation (inode->i_sb); | |
1291 | reiserfs_prepare_for_journal(inode->i_sb, bh, 1) ; | |
1292 | if (fs_changed (fs_gen, inode->i_sb) && item_moved(&tmp_ih, &path)) { | |
1293 | reiserfs_restore_prepared_buffer(inode->i_sb, bh) ; | |
1294 | continue ; /* Stat_data item has been moved after scheduling. */ | |
1295 | } | |
1296 | break; | |
1297 | } | |
1298 | update_stat_data (&path, inode, size); | |
1299 | journal_mark_dirty(th, th->t_super, bh) ; | |
1300 | pathrelse (&path); | |
1301 | return; | |
1302 | } | |
1303 | ||
1304 | /* reiserfs_read_locked_inode is called to read the inode off disk, and it | |
1305 | ** does a make_bad_inode when things go wrong. But, we need to make sure | |
1306 | ** and clear the key in the private portion of the inode, otherwise a | |
1307 | ** corresponding iput might try to delete whatever object the inode last | |
1308 | ** represented. | |
1309 | */ | |
1310 | static void reiserfs_make_bad_inode(struct inode *inode) { | |
1311 | memset(INODE_PKEY(inode), 0, KEY_SIZE); | |
1312 | make_bad_inode(inode); | |
1313 | } | |
1314 | ||
1315 | // | |
1316 | // initially this function was derived from minix or ext2's analog and | |
1317 | // evolved as the prototype did | |
1318 | // | |
1319 | ||
1320 | int reiserfs_init_locked_inode (struct inode * inode, void *p) | |
1321 | { | |
1322 | struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p ; | |
1323 | inode->i_ino = args->objectid; | |
1324 | INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid); | |
1325 | return 0; | |
1326 | } | |
1327 | ||
1328 | /* looks for stat data in the tree, and fills up the fields of in-core | |
1329 | inode stat data fields */ | |
1330 | void reiserfs_read_locked_inode (struct inode * inode, struct reiserfs_iget_args *args) | |
1331 | { | |
1332 | INITIALIZE_PATH (path_to_sd); | |
1333 | struct cpu_key key; | |
1334 | unsigned long dirino; | |
1335 | int retval; | |
1336 | ||
1337 | dirino = args->dirid ; | |
1338 | ||
1339 | /* set version 1, version 2 could be used too, because stat data | |
1340 | key is the same in both versions */ | |
1341 | key.version = KEY_FORMAT_3_5; | |
1342 | key.on_disk_key.k_dir_id = dirino; | |
1343 | key.on_disk_key.k_objectid = inode->i_ino; | |
1344 | key.on_disk_key.u.k_offset_v1.k_offset = SD_OFFSET; | |
1345 | key.on_disk_key.u.k_offset_v1.k_uniqueness = SD_UNIQUENESS; | |
1346 | ||
1347 | /* look for the object's stat data */ | |
1348 | retval = search_item (inode->i_sb, &key, &path_to_sd); | |
1349 | if (retval == IO_ERROR) { | |
1350 | reiserfs_warning (inode->i_sb, "vs-13070: reiserfs_read_locked_inode: " | |
1351 | "i/o failure occurred trying to find stat data of %K", | |
1352 | &key); | |
1353 | reiserfs_make_bad_inode(inode) ; | |
1354 | return; | |
1355 | } | |
1356 | if (retval != ITEM_FOUND) { | |
1357 | /* a stale NFS handle can trigger this without it being an error */ | |
1358 | pathrelse (&path_to_sd); | |
1359 | reiserfs_make_bad_inode(inode) ; | |
1360 | inode->i_nlink = 0; | |
1361 | return; | |
1362 | } | |
1363 | ||
1364 | init_inode (inode, &path_to_sd); | |
1365 | ||
1366 | /* It is possible that knfsd is trying to access inode of a file | |
1367 | that is being removed from the disk by some other thread. As we | |
1368 | update sd on unlink all that is required is to check for nlink | |
1369 | here. This bug was first found by Sizif when debugging | |
1370 | SquidNG/Butterfly, forgotten, and found again after Philippe | |
1371 | Gramoulle <philippe.gramoulle@mmania.com> reproduced it. | |
1372 | ||
1373 | More logical fix would require changes in fs/inode.c:iput() to | |
1374 | remove inode from hash-table _after_ fs cleaned disk stuff up and | |
1375 | in iget() to return NULL if I_FREEING inode is found in | |
1376 | hash-table. */ | |
1377 | /* Currently there is one place where it's ok to meet inode with | |
1378 | nlink==0: processing of open-unlinked and half-truncated files | |
1379 | during mount (fs/reiserfs/super.c:finish_unfinished()). */ | |
1380 | if( ( inode -> i_nlink == 0 ) && | |
1381 | ! REISERFS_SB(inode -> i_sb) -> s_is_unlinked_ok ) { | |
1382 | reiserfs_warning (inode->i_sb, | |
1383 | "vs-13075: reiserfs_read_locked_inode: " | |
1384 | "dead inode read from disk %K. " | |
1385 | "This is likely to be race with knfsd. Ignore", | |
1386 | &key ); | |
1387 | reiserfs_make_bad_inode( inode ); | |
1388 | } | |
1389 | ||
1390 | reiserfs_check_path(&path_to_sd) ; /* init inode should be relsing */ | |
1391 | ||
1392 | } | |
1393 | ||
1394 | /** | |
1395 | * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked(). | |
1396 | * | |
1397 | * @inode: inode from hash table to check | |
1398 | * @opaque: "cookie" passed to iget5_locked(). This is &reiserfs_iget_args. | |
1399 | * | |
1400 | * This function is called by iget5_locked() to distinguish reiserfs inodes | |
1401 | * having the same inode numbers. Such inodes can only exist due to some | |
1402 | * error condition. One of them should be bad. Inodes with identical | |
1403 | * inode numbers (objectids) are distinguished by parent directory ids. | |
1404 | * | |
1405 | */ | |
1406 | int reiserfs_find_actor( struct inode *inode, void *opaque ) | |
1407 | { | |
1408 | struct reiserfs_iget_args *args; | |
1409 | ||
1410 | args = opaque; | |
1411 | /* args is already in CPU order */ | |
1412 | return (inode->i_ino == args->objectid) && | |
1413 | (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid); | |
1414 | } | |
1415 | ||
1416 | struct inode * reiserfs_iget (struct super_block * s, const struct cpu_key * key) | |
1417 | { | |
1418 | struct inode * inode; | |
1419 | struct reiserfs_iget_args args ; | |
1420 | ||
1421 | args.objectid = key->on_disk_key.k_objectid ; | |
1422 | args.dirid = key->on_disk_key.k_dir_id ; | |
1423 | inode = iget5_locked (s, key->on_disk_key.k_objectid, | |
1424 | reiserfs_find_actor, reiserfs_init_locked_inode, (void *)(&args)); | |
1425 | if (!inode) | |
1426 | return ERR_PTR(-ENOMEM) ; | |
1427 | ||
1428 | if (inode->i_state & I_NEW) { | |
1429 | reiserfs_read_locked_inode(inode, &args); | |
1430 | unlock_new_inode(inode); | |
1431 | } | |
1432 | ||
1433 | if (comp_short_keys (INODE_PKEY (inode), key) || is_bad_inode (inode)) { | |
1434 | /* either due to i/o error or a stale NFS handle */ | |
1435 | iput (inode); | |
1436 | inode = NULL; | |
1437 | } | |
1438 | return inode; | |
1439 | } | |
1440 | ||
1441 | struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp) | |
1442 | { | |
1443 | __u32 *data = vobjp; | |
1444 | struct cpu_key key ; | |
1445 | struct dentry *result; | |
1446 | struct inode *inode; | |
1447 | ||
1448 | key.on_disk_key.k_objectid = data[0] ; | |
1449 | key.on_disk_key.k_dir_id = data[1] ; | |
1450 | reiserfs_write_lock(sb); | |
1451 | inode = reiserfs_iget(sb, &key) ; | |
1452 | if (inode && !IS_ERR(inode) && data[2] != 0 && | |
1453 | data[2] != inode->i_generation) { | |
1454 | iput(inode) ; | |
1455 | inode = NULL ; | |
1456 | } | |
1457 | reiserfs_write_unlock(sb); | |
1458 | if (!inode) | |
1459 | inode = ERR_PTR(-ESTALE); | |
1460 | if (IS_ERR(inode)) | |
1461 | return ERR_PTR(PTR_ERR(inode)); | |
1462 | result = d_alloc_anon(inode); | |
1463 | if (!result) { | |
1464 | iput(inode); | |
1465 | return ERR_PTR(-ENOMEM); | |
1466 | } | |
1467 | return result; | |
1468 | } | |
1469 | ||
1470 | struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 *data, | |
1471 | int len, int fhtype, | |
1472 | int (*acceptable)(void *contect, struct dentry *de), | |
1473 | void *context) { | |
1474 | __u32 obj[3], parent[3]; | |
1475 | ||
1476 | /* fhtype happens to reflect the number of u32s encoded. | |
1477 | * due to a bug in earlier code, fhtype might indicate there | |
1478 | * are more u32s then actually fitted. | |
1479 | * so if fhtype seems to be more than len, reduce fhtype. | |
1480 | * Valid types are: | |
1481 | * 2 - objectid + dir_id - legacy support | |
1482 | * 3 - objectid + dir_id + generation | |
1483 | * 4 - objectid + dir_id + objectid and dirid of parent - legacy | |
1484 | * 5 - objectid + dir_id + generation + objectid and dirid of parent | |
1485 | * 6 - as above plus generation of directory | |
1486 | * 6 does not fit in NFSv2 handles | |
1487 | */ | |
1488 | if (fhtype > len) { | |
1489 | if (fhtype != 6 || len != 5) | |
1490 | reiserfs_warning (sb, "nfsd/reiserfs, fhtype=%d, len=%d - odd", | |
1491 | fhtype, len); | |
1492 | fhtype = 5; | |
1493 | } | |
1494 | ||
1495 | obj[0] = data[0]; | |
1496 | obj[1] = data[1]; | |
1497 | if (fhtype == 3 || fhtype >= 5) | |
1498 | obj[2] = data[2]; | |
1499 | else obj[2] = 0; /* generation number */ | |
1500 | ||
1501 | if (fhtype >= 4) { | |
1502 | parent[0] = data[fhtype>=5?3:2] ; | |
1503 | parent[1] = data[fhtype>=5?4:3] ; | |
1504 | if (fhtype == 6) | |
1505 | parent[2] = data[5]; | |
1506 | else parent[2] = 0; | |
1507 | } | |
1508 | return sb->s_export_op->find_exported_dentry(sb, obj, fhtype < 4 ? NULL : parent, | |
1509 | acceptable, context); | |
1510 | } | |
1511 | ||
1512 | int reiserfs_encode_fh(struct dentry *dentry, __u32 *data, int *lenp, int need_parent) { | |
1513 | struct inode *inode = dentry->d_inode ; | |
1514 | int maxlen = *lenp; | |
1515 | ||
1516 | if (maxlen < 3) | |
1517 | return 255 ; | |
1518 | ||
1519 | data[0] = inode->i_ino ; | |
1520 | data[1] = le32_to_cpu(INODE_PKEY (inode)->k_dir_id) ; | |
1521 | data[2] = inode->i_generation ; | |
1522 | *lenp = 3 ; | |
1523 | /* no room for directory info? return what we've stored so far */ | |
1524 | if (maxlen < 5 || ! need_parent) | |
1525 | return 3 ; | |
1526 | ||
1527 | spin_lock(&dentry->d_lock); | |
1528 | inode = dentry->d_parent->d_inode ; | |
1529 | data[3] = inode->i_ino ; | |
1530 | data[4] = le32_to_cpu(INODE_PKEY (inode)->k_dir_id) ; | |
1531 | *lenp = 5 ; | |
1532 | if (maxlen >= 6) { | |
1533 | data[5] = inode->i_generation ; | |
1534 | *lenp = 6 ; | |
1535 | } | |
1536 | spin_unlock(&dentry->d_lock); | |
1537 | return *lenp ; | |
1538 | } | |
1539 | ||
1540 | ||
1541 | /* looks for stat data, then copies fields to it, marks the buffer | |
1542 | containing stat data as dirty */ | |
1543 | /* reiserfs inodes are never really dirty, since the dirty inode call | |
1544 | ** always logs them. This call allows the VFS inode marking routines | |
1545 | ** to properly mark inodes for datasync and such, but only actually | |
1546 | ** does something when called for a synchronous update. | |
1547 | */ | |
1548 | int reiserfs_write_inode (struct inode * inode, int do_sync) { | |
1549 | struct reiserfs_transaction_handle th ; | |
1550 | int jbegin_count = 1 ; | |
1551 | ||
1552 | if (inode->i_sb->s_flags & MS_RDONLY) | |
1553 | return -EROFS; | |
1554 | /* memory pressure can sometimes initiate write_inode calls with sync == 1, | |
1555 | ** these cases are just when the system needs ram, not when the | |
1556 | ** inode needs to reach disk for safety, and they can safely be | |
1557 | ** ignored because the altered inode has already been logged. | |
1558 | */ | |
1559 | if (do_sync && !(current->flags & PF_MEMALLOC)) { | |
1560 | reiserfs_write_lock(inode->i_sb); | |
1561 | if (!journal_begin(&th, inode->i_sb, jbegin_count)) { | |
1562 | reiserfs_update_sd (&th, inode); | |
1563 | journal_end_sync(&th, inode->i_sb, jbegin_count) ; | |
1564 | } | |
1565 | reiserfs_write_unlock(inode->i_sb); | |
1566 | } | |
1567 | return 0; | |
1568 | } | |
1569 | ||
1570 | /* stat data of new object is inserted already, this inserts the item | |
1571 | containing "." and ".." entries */ | |
1572 | static int reiserfs_new_directory (struct reiserfs_transaction_handle *th, | |
1573 | struct inode *inode, | |
1574 | struct item_head * ih, struct path * path, | |
1575 | struct inode * dir) | |
1576 | { | |
1577 | struct super_block * sb = th->t_super; | |
1578 | char empty_dir [EMPTY_DIR_SIZE]; | |
1579 | char * body = empty_dir; | |
1580 | struct cpu_key key; | |
1581 | int retval; | |
1582 | ||
1583 | BUG_ON (!th->t_trans_id); | |
1584 | ||
1585 | _make_cpu_key (&key, KEY_FORMAT_3_5, le32_to_cpu (ih->ih_key.k_dir_id), | |
1586 | le32_to_cpu (ih->ih_key.k_objectid), DOT_OFFSET, TYPE_DIRENTRY, 3/*key length*/); | |
1587 | ||
1588 | /* compose item head for new item. Directories consist of items of | |
1589 | old type (ITEM_VERSION_1). Do not set key (second arg is 0), it | |
1590 | is done by reiserfs_new_inode */ | |
1591 | if (old_format_only (sb)) { | |
1592 | make_le_item_head (ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET, TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2); | |
1593 | ||
1594 | make_empty_dir_item_v1 (body, ih->ih_key.k_dir_id, ih->ih_key.k_objectid, | |
1595 | INODE_PKEY (dir)->k_dir_id, | |
1596 | INODE_PKEY (dir)->k_objectid ); | |
1597 | } else { | |
1598 | make_le_item_head (ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET, TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2); | |
1599 | ||
1600 | make_empty_dir_item (body, ih->ih_key.k_dir_id, ih->ih_key.k_objectid, | |
1601 | INODE_PKEY (dir)->k_dir_id, | |
1602 | INODE_PKEY (dir)->k_objectid ); | |
1603 | } | |
1604 | ||
1605 | /* look for place in the tree for new item */ | |
1606 | retval = search_item (sb, &key, path); | |
1607 | if (retval == IO_ERROR) { | |
1608 | reiserfs_warning (sb, "vs-13080: reiserfs_new_directory: " | |
1609 | "i/o failure occurred creating new directory"); | |
1610 | return -EIO; | |
1611 | } | |
1612 | if (retval == ITEM_FOUND) { | |
1613 | pathrelse (path); | |
1614 | reiserfs_warning (sb, "vs-13070: reiserfs_new_directory: " | |
1615 | "object with this key exists (%k)", &(ih->ih_key)); | |
1616 | return -EEXIST; | |
1617 | } | |
1618 | ||
1619 | /* insert item, that is empty directory item */ | |
1620 | return reiserfs_insert_item (th, path, &key, ih, inode, body); | |
1621 | } | |
1622 | ||
1623 | ||
1624 | /* stat data of object has been inserted, this inserts the item | |
1625 | containing the body of symlink */ | |
1626 | static int reiserfs_new_symlink (struct reiserfs_transaction_handle *th, | |
1627 | struct inode *inode, /* Inode of symlink */ | |
1628 | struct item_head * ih, | |
1629 | struct path * path, const char * symname, int item_len) | |
1630 | { | |
1631 | struct super_block * sb = th->t_super; | |
1632 | struct cpu_key key; | |
1633 | int retval; | |
1634 | ||
1635 | BUG_ON (!th->t_trans_id); | |
1636 | ||
1637 | _make_cpu_key (&key, KEY_FORMAT_3_5, | |
1638 | le32_to_cpu (ih->ih_key.k_dir_id), | |
1639 | le32_to_cpu (ih->ih_key.k_objectid), | |
1640 | 1, TYPE_DIRECT, 3/*key length*/); | |
1641 | ||
1642 | make_le_item_head (ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len, 0/*free_space*/); | |
1643 | ||
1644 | /* look for place in the tree for new item */ | |
1645 | retval = search_item (sb, &key, path); | |
1646 | if (retval == IO_ERROR) { | |
1647 | reiserfs_warning (sb, "vs-13080: reiserfs_new_symlinik: " | |
1648 | "i/o failure occurred creating new symlink"); | |
1649 | return -EIO; | |
1650 | } | |
1651 | if (retval == ITEM_FOUND) { | |
1652 | pathrelse (path); | |
1653 | reiserfs_warning (sb, "vs-13080: reiserfs_new_symlink: " | |
1654 | "object with this key exists (%k)", &(ih->ih_key)); | |
1655 | return -EEXIST; | |
1656 | } | |
1657 | ||
1658 | /* insert item, that is body of symlink */ | |
1659 | return reiserfs_insert_item (th, path, &key, ih, inode, symname); | |
1660 | } | |
1661 | ||
1662 | ||
1663 | /* inserts the stat data into the tree, and then calls | |
1664 | reiserfs_new_directory (to insert ".", ".." item if new object is | |
1665 | directory) or reiserfs_new_symlink (to insert symlink body if new | |
1666 | object is symlink) or nothing (if new object is regular file) | |
1667 | ||
1668 | NOTE! uid and gid must already be set in the inode. If we return | |
1669 | non-zero due to an error, we have to drop the quota previously allocated | |
1670 | for the fresh inode. This can only be done outside a transaction, so | |
1671 | if we return non-zero, we also end the transaction. */ | |
1672 | int reiserfs_new_inode (struct reiserfs_transaction_handle *th, | |
1673 | struct inode * dir, int mode, | |
1674 | const char * symname, | |
1675 | /* 0 for regular, EMTRY_DIR_SIZE for dirs, | |
1676 | strlen (symname) for symlinks)*/ | |
1677 | loff_t i_size, struct dentry *dentry, | |
1678 | struct inode *inode) | |
1679 | { | |
1680 | struct super_block * sb; | |
1681 | INITIALIZE_PATH (path_to_key); | |
1682 | struct cpu_key key; | |
1683 | struct item_head ih; | |
1684 | struct stat_data sd; | |
1685 | int retval; | |
1686 | int err; | |
1687 | ||
1688 | BUG_ON (!th->t_trans_id); | |
1689 | ||
1690 | if (DQUOT_ALLOC_INODE(inode)) { | |
1691 | err = -EDQUOT; | |
1692 | goto out_end_trans; | |
1693 | } | |
1694 | if (!dir || !dir->i_nlink) { | |
1695 | err = -EPERM; | |
1696 | goto out_bad_inode; | |
1697 | } | |
1698 | ||
1699 | sb = dir->i_sb; | |
1700 | ||
1701 | /* item head of new item */ | |
1702 | ih.ih_key.k_dir_id = reiserfs_choose_packing(dir); | |
1703 | ih.ih_key.k_objectid = cpu_to_le32 (reiserfs_get_unused_objectid (th)); | |
1704 | if (!ih.ih_key.k_objectid) { | |
1705 | err = -ENOMEM; | |
1706 | goto out_bad_inode ; | |
1707 | } | |
1708 | if (old_format_only (sb)) | |
1709 | /* not a perfect generation count, as object ids can be reused, but | |
1710 | ** this is as good as reiserfs can do right now. | |
1711 | ** note that the private part of inode isn't filled in yet, we have | |
1712 | ** to use the directory. | |
1713 | */ | |
1714 | inode->i_generation = le32_to_cpu (INODE_PKEY (dir)->k_objectid); | |
1715 | else | |
1716 | #if defined( USE_INODE_GENERATION_COUNTER ) | |
1717 | inode->i_generation = le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation); | |
1718 | #else | |
1719 | inode->i_generation = ++event; | |
1720 | #endif | |
1721 | ||
1722 | /* fill stat data */ | |
1723 | inode->i_nlink = (S_ISDIR (mode) ? 2 : 1); | |
1724 | ||
1725 | /* uid and gid must already be set by the caller for quota init */ | |
1726 | ||
1727 | /* symlink cannot be immutable or append only, right? */ | |
1728 | if( S_ISLNK( inode -> i_mode ) ) | |
1729 | inode -> i_flags &= ~ ( S_IMMUTABLE | S_APPEND ); | |
1730 | ||
1731 | inode->i_mtime = inode->i_atime = inode->i_ctime = | |
1732 | CURRENT_TIME_SEC; | |
1733 | inode->i_size = i_size; | |
1734 | inode->i_blocks = 0; | |
1735 | inode->i_bytes = 0; | |
1736 | REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 : | |
1737 | U32_MAX/*NO_BYTES_IN_DIRECT_ITEM*/; | |
1738 | ||
1739 | INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list )); | |
1740 | REISERFS_I(inode)->i_flags = 0; | |
1741 | REISERFS_I(inode)->i_prealloc_block = 0; | |
1742 | REISERFS_I(inode)->i_prealloc_count = 0; | |
1743 | REISERFS_I(inode)->i_trans_id = 0; | |
1744 | REISERFS_I(inode)->i_jl = NULL; | |
1745 | REISERFS_I(inode)->i_attrs = | |
1746 | REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK; | |
1747 | sd_attrs_to_i_attrs( REISERFS_I(inode) -> i_attrs, inode ); | |
1748 | REISERFS_I(inode)->i_acl_access = NULL; | |
1749 | REISERFS_I(inode)->i_acl_default = NULL; | |
1750 | init_rwsem (&REISERFS_I(inode)->xattr_sem); | |
1751 | ||
1752 | if (old_format_only (sb)) | |
1753 | make_le_item_head (&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET, TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT); | |
1754 | else | |
1755 | make_le_item_head (&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET, TYPE_STAT_DATA, SD_SIZE, MAX_US_INT); | |
1756 | ||
1757 | /* key to search for correct place for new stat data */ | |
1758 | _make_cpu_key (&key, KEY_FORMAT_3_6, le32_to_cpu (ih.ih_key.k_dir_id), | |
1759 | le32_to_cpu (ih.ih_key.k_objectid), SD_OFFSET, TYPE_STAT_DATA, 3/*key length*/); | |
1760 | ||
1761 | /* find proper place for inserting of stat data */ | |
1762 | retval = search_item (sb, &key, &path_to_key); | |
1763 | if (retval == IO_ERROR) { | |
1764 | err = -EIO; | |
1765 | goto out_bad_inode; | |
1766 | } | |
1767 | if (retval == ITEM_FOUND) { | |
1768 | pathrelse (&path_to_key); | |
1769 | err = -EEXIST; | |
1770 | goto out_bad_inode; | |
1771 | } | |
1772 | if (old_format_only (sb)) { | |
1773 | if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) { | |
1774 | pathrelse (&path_to_key); | |
1775 | /* i_uid or i_gid is too big to be stored in stat data v3.5 */ | |
1776 | err = -EINVAL; | |
1777 | goto out_bad_inode; | |
1778 | } | |
1779 | inode2sd_v1 (&sd, inode, inode->i_size); | |
1780 | } else { | |
1781 | inode2sd (&sd, inode, inode->i_size); | |
1782 | } | |
1783 | // these do not go to on-disk stat data | |
1784 | inode->i_ino = le32_to_cpu (ih.ih_key.k_objectid); | |
1785 | inode->i_blksize = reiserfs_default_io_size; | |
1786 | ||
1787 | // store in in-core inode the key of stat data and version all | |
1788 | // object items will have (directory items will have old offset | |
1789 | // format, other new objects will consist of new items) | |
1790 | memcpy (INODE_PKEY (inode), &(ih.ih_key), KEY_SIZE); | |
1791 | if (old_format_only (sb) || S_ISDIR(mode) || S_ISLNK(mode)) | |
1792 | set_inode_item_key_version (inode, KEY_FORMAT_3_5); | |
1793 | else | |
1794 | set_inode_item_key_version (inode, KEY_FORMAT_3_6); | |
1795 | if (old_format_only (sb)) | |
1796 | set_inode_sd_version (inode, STAT_DATA_V1); | |
1797 | else | |
1798 | set_inode_sd_version (inode, STAT_DATA_V2); | |
1799 | ||
1800 | /* insert the stat data into the tree */ | |
1801 | #ifdef DISPLACE_NEW_PACKING_LOCALITIES | |
1802 | if (REISERFS_I(dir)->new_packing_locality) | |
1803 | th->displace_new_blocks = 1; | |
1804 | #endif | |
1805 | retval = reiserfs_insert_item (th, &path_to_key, &key, &ih, inode, (char *)(&sd)); | |
1806 | if (retval) { | |
1807 | err = retval; | |
1808 | reiserfs_check_path(&path_to_key) ; | |
1809 | goto out_bad_inode; | |
1810 | } | |
1811 | ||
1812 | #ifdef DISPLACE_NEW_PACKING_LOCALITIES | |
1813 | if (!th->displace_new_blocks) | |
1814 | REISERFS_I(dir)->new_packing_locality = 0; | |
1815 | #endif | |
1816 | if (S_ISDIR(mode)) { | |
1817 | /* insert item with "." and ".." */ | |
1818 | retval = reiserfs_new_directory (th, inode, &ih, &path_to_key, dir); | |
1819 | } | |
1820 | ||
1821 | if (S_ISLNK(mode)) { | |
1822 | /* insert body of symlink */ | |
1823 | if (!old_format_only (sb)) | |
1824 | i_size = ROUND_UP(i_size); | |
1825 | retval = reiserfs_new_symlink (th, inode, &ih, &path_to_key, symname, i_size); | |
1826 | } | |
1827 | if (retval) { | |
1828 | err = retval; | |
1829 | reiserfs_check_path(&path_to_key) ; | |
1830 | journal_end(th, th->t_super, th->t_blocks_allocated); | |
1831 | goto out_inserted_sd; | |
1832 | } | |
1833 | ||
1834 | /* XXX CHECK THIS */ | |
1835 | if (reiserfs_posixacl (inode->i_sb)) { | |
1836 | retval = reiserfs_inherit_default_acl (dir, dentry, inode); | |
1837 | if (retval) { | |
1838 | err = retval; | |
1839 | reiserfs_check_path(&path_to_key) ; | |
1840 | journal_end(th, th->t_super, th->t_blocks_allocated); | |
1841 | goto out_inserted_sd; | |
1842 | } | |
1843 | } else if (inode->i_sb->s_flags & MS_POSIXACL) { | |
1844 | reiserfs_warning (inode->i_sb, "ACLs aren't enabled in the fs, " | |
1845 | "but vfs thinks they are!"); | |
1846 | } else if (is_reiserfs_priv_object (dir)) { | |
1847 | reiserfs_mark_inode_private (inode); | |
1848 | } | |
1849 | ||
1850 | insert_inode_hash (inode); | |
1851 | reiserfs_update_sd(th, inode); | |
1852 | reiserfs_check_path(&path_to_key) ; | |
1853 | ||
1854 | return 0; | |
1855 | ||
1856 | /* it looks like you can easily compress these two goto targets into | |
1857 | * one. Keeping it like this doesn't actually hurt anything, and they | |
1858 | * are place holders for what the quota code actually needs. | |
1859 | */ | |
1860 | out_bad_inode: | |
1861 | /* Invalidate the object, nothing was inserted yet */ | |
1862 | INODE_PKEY(inode)->k_objectid = 0; | |
1863 | ||
1864 | /* Quota change must be inside a transaction for journaling */ | |
1865 | DQUOT_FREE_INODE(inode); | |
1866 | ||
1867 | out_end_trans: | |
1868 | journal_end(th, th->t_super, th->t_blocks_allocated) ; | |
1869 | /* Drop can be outside and it needs more credits so it's better to have it outside */ | |
1870 | DQUOT_DROP(inode); | |
1871 | inode->i_flags |= S_NOQUOTA; | |
1872 | make_bad_inode(inode); | |
1873 | ||
1874 | out_inserted_sd: | |
1875 | inode->i_nlink = 0; | |
1876 | th->t_trans_id = 0; /* so the caller can't use this handle later */ | |
1877 | iput(inode); | |
1878 | return err; | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | ** finds the tail page in the page cache, | |
1883 | ** reads the last block in. | |
1884 | ** | |
1885 | ** On success, page_result is set to a locked, pinned page, and bh_result | |
1886 | ** is set to an up to date buffer for the last block in the file. returns 0. | |
1887 | ** | |
1888 | ** tail conversion is not done, so bh_result might not be valid for writing | |
1889 | ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before | |
1890 | ** trying to write the block. | |
1891 | ** | |
1892 | ** on failure, nonzero is returned, page_result and bh_result are untouched. | |
1893 | */ | |
1894 | static int grab_tail_page(struct inode *p_s_inode, | |
1895 | struct page **page_result, | |
1896 | struct buffer_head **bh_result) { | |
1897 | ||
1898 | /* we want the page with the last byte in the file, | |
1899 | ** not the page that will hold the next byte for appending | |
1900 | */ | |
1901 | unsigned long index = (p_s_inode->i_size-1) >> PAGE_CACHE_SHIFT ; | |
1902 | unsigned long pos = 0 ; | |
1903 | unsigned long start = 0 ; | |
1904 | unsigned long blocksize = p_s_inode->i_sb->s_blocksize ; | |
1905 | unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1) ; | |
1906 | struct buffer_head *bh ; | |
1907 | struct buffer_head *head ; | |
1908 | struct page * page ; | |
1909 | int error ; | |
1910 | ||
1911 | /* we know that we are only called with inode->i_size > 0. | |
1912 | ** we also know that a file tail can never be as big as a block | |
1913 | ** If i_size % blocksize == 0, our file is currently block aligned | |
1914 | ** and it won't need converting or zeroing after a truncate. | |
1915 | */ | |
1916 | if ((offset & (blocksize - 1)) == 0) { | |
1917 | return -ENOENT ; | |
1918 | } | |
1919 | page = grab_cache_page(p_s_inode->i_mapping, index) ; | |
1920 | error = -ENOMEM ; | |
1921 | if (!page) { | |
1922 | goto out ; | |
1923 | } | |
1924 | /* start within the page of the last block in the file */ | |
1925 | start = (offset / blocksize) * blocksize ; | |
1926 | ||
1927 | error = block_prepare_write(page, start, offset, | |
1928 | reiserfs_get_block_create_0) ; | |
1929 | if (error) | |
1930 | goto unlock ; | |
1931 | ||
1932 | head = page_buffers(page) ; | |
1933 | bh = head; | |
1934 | do { | |
1935 | if (pos >= start) { | |
1936 | break ; | |
1937 | } | |
1938 | bh = bh->b_this_page ; | |
1939 | pos += blocksize ; | |
1940 | } while(bh != head) ; | |
1941 | ||
1942 | if (!buffer_uptodate(bh)) { | |
1943 | /* note, this should never happen, prepare_write should | |
1944 | ** be taking care of this for us. If the buffer isn't up to date, | |
1945 | ** I've screwed up the code to find the buffer, or the code to | |
1946 | ** call prepare_write | |
1947 | */ | |
1948 | reiserfs_warning (p_s_inode->i_sb, | |
1949 | "clm-6000: error reading block %lu on dev %s", | |
1950 | bh->b_blocknr, | |
1951 | reiserfs_bdevname (p_s_inode->i_sb)) ; | |
1952 | error = -EIO ; | |
1953 | goto unlock ; | |
1954 | } | |
1955 | *bh_result = bh ; | |
1956 | *page_result = page ; | |
1957 | ||
1958 | out: | |
1959 | return error ; | |
1960 | ||
1961 | unlock: | |
1962 | unlock_page(page) ; | |
1963 | page_cache_release(page) ; | |
1964 | return error ; | |
1965 | } | |
1966 | ||
1967 | /* | |
1968 | ** vfs version of truncate file. Must NOT be called with | |
1969 | ** a transaction already started. | |
1970 | ** | |
1971 | ** some code taken from block_truncate_page | |
1972 | */ | |
1973 | int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps) { | |
1974 | struct reiserfs_transaction_handle th ; | |
1975 | /* we want the offset for the first byte after the end of the file */ | |
1976 | unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1) ; | |
1977 | unsigned blocksize = p_s_inode->i_sb->s_blocksize ; | |
1978 | unsigned length ; | |
1979 | struct page *page = NULL ; | |
1980 | int error ; | |
1981 | struct buffer_head *bh = NULL ; | |
1982 | ||
1983 | reiserfs_write_lock(p_s_inode->i_sb); | |
1984 | ||
1985 | if (p_s_inode->i_size > 0) { | |
1986 | if ((error = grab_tail_page(p_s_inode, &page, &bh))) { | |
1987 | // -ENOENT means we truncated past the end of the file, | |
1988 | // and get_block_create_0 could not find a block to read in, | |
1989 | // which is ok. | |
1990 | if (error != -ENOENT) | |
1991 | reiserfs_warning (p_s_inode->i_sb, | |
1992 | "clm-6001: grab_tail_page failed %d", | |
1993 | error); | |
1994 | page = NULL ; | |
1995 | bh = NULL ; | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | /* so, if page != NULL, we have a buffer head for the offset at | |
2000 | ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0, | |
2001 | ** then we have an unformatted node. Otherwise, we have a direct item, | |
2002 | ** and no zeroing is required on disk. We zero after the truncate, | |
2003 | ** because the truncate might pack the item anyway | |
2004 | ** (it will unmap bh if it packs). | |
2005 | */ | |
2006 | /* it is enough to reserve space in transaction for 2 balancings: | |
2007 | one for "save" link adding and another for the first | |
2008 | cut_from_item. 1 is for update_sd */ | |
2009 | error = journal_begin (&th, p_s_inode->i_sb, | |
2010 | JOURNAL_PER_BALANCE_CNT * 2 + 1); | |
2011 | if (error) | |
2012 | goto out; | |
2013 | reiserfs_update_inode_transaction(p_s_inode) ; | |
2014 | if (update_timestamps) | |
2015 | /* we are doing real truncate: if the system crashes before the last | |
2016 | transaction of truncating gets committed - on reboot the file | |
2017 | either appears truncated properly or not truncated at all */ | |
2018 | add_save_link (&th, p_s_inode, 1); | |
2019 | error = reiserfs_do_truncate (&th, p_s_inode, page, update_timestamps) ; | |
2020 | if (error) | |
2021 | goto out; | |
2022 | error = journal_end (&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1); | |
2023 | if (error) | |
2024 | goto out; | |
2025 | ||
2026 | if (update_timestamps) { | |
2027 | error = remove_save_link (p_s_inode, 1/* truncate */); | |
2028 | if (error) | |
2029 | goto out; | |
2030 | } | |
2031 | ||
2032 | if (page) { | |
2033 | length = offset & (blocksize - 1) ; | |
2034 | /* if we are not on a block boundary */ | |
2035 | if (length) { | |
2036 | char *kaddr; | |
2037 | ||
2038 | length = blocksize - length ; | |
2039 | kaddr = kmap_atomic(page, KM_USER0) ; | |
2040 | memset(kaddr + offset, 0, length) ; | |
2041 | flush_dcache_page(page) ; | |
2042 | kunmap_atomic(kaddr, KM_USER0) ; | |
2043 | if (buffer_mapped(bh) && bh->b_blocknr != 0) { | |
2044 | mark_buffer_dirty(bh) ; | |
2045 | } | |
2046 | } | |
2047 | unlock_page(page) ; | |
2048 | page_cache_release(page) ; | |
2049 | } | |
2050 | ||
2051 | reiserfs_write_unlock(p_s_inode->i_sb); | |
2052 | return 0; | |
2053 | out: | |
2054 | if (page) { | |
2055 | unlock_page (page); | |
2056 | page_cache_release (page); | |
2057 | } | |
2058 | reiserfs_write_unlock(p_s_inode->i_sb); | |
2059 | return error; | |
2060 | } | |
2061 | ||
2062 | static int map_block_for_writepage(struct inode *inode, | |
2063 | struct buffer_head *bh_result, | |
2064 | unsigned long block) { | |
2065 | struct reiserfs_transaction_handle th ; | |
2066 | int fs_gen ; | |
2067 | struct item_head tmp_ih ; | |
2068 | struct item_head *ih ; | |
2069 | struct buffer_head *bh ; | |
2070 | __u32 *item ; | |
2071 | struct cpu_key key ; | |
2072 | INITIALIZE_PATH(path) ; | |
2073 | int pos_in_item ; | |
2074 | int jbegin_count = JOURNAL_PER_BALANCE_CNT ; | |
2075 | loff_t byte_offset = (block << inode->i_sb->s_blocksize_bits) + 1 ; | |
2076 | int retval ; | |
2077 | int use_get_block = 0 ; | |
2078 | int bytes_copied = 0 ; | |
2079 | int copy_size ; | |
2080 | int trans_running = 0; | |
2081 | ||
2082 | /* catch places below that try to log something without starting a trans */ | |
2083 | th.t_trans_id = 0; | |
2084 | ||
2085 | if (!buffer_uptodate(bh_result)) { | |
2086 | return -EIO; | |
2087 | } | |
2088 | ||
2089 | kmap(bh_result->b_page) ; | |
2090 | start_over: | |
2091 | reiserfs_write_lock(inode->i_sb); | |
2092 | make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3) ; | |
2093 | ||
2094 | research: | |
2095 | retval = search_for_position_by_key(inode->i_sb, &key, &path) ; | |
2096 | if (retval != POSITION_FOUND) { | |
2097 | use_get_block = 1; | |
2098 | goto out ; | |
2099 | } | |
2100 | ||
2101 | bh = get_last_bh(&path) ; | |
2102 | ih = get_ih(&path) ; | |
2103 | item = get_item(&path) ; | |
2104 | pos_in_item = path.pos_in_item ; | |
2105 | ||
2106 | /* we've found an unformatted node */ | |
2107 | if (indirect_item_found(retval, ih)) { | |
2108 | if (bytes_copied > 0) { | |
2109 | reiserfs_warning (inode->i_sb, "clm-6002: bytes_copied %d", | |
2110 | bytes_copied) ; | |
2111 | } | |
2112 | if (!get_block_num(item, pos_in_item)) { | |
2113 | /* crap, we are writing to a hole */ | |
2114 | use_get_block = 1; | |
2115 | goto out ; | |
2116 | } | |
2117 | set_block_dev_mapped(bh_result, get_block_num(item,pos_in_item),inode); | |
2118 | } else if (is_direct_le_ih(ih)) { | |
2119 | char *p ; | |
2120 | p = page_address(bh_result->b_page) ; | |
2121 | p += (byte_offset -1) & (PAGE_CACHE_SIZE - 1) ; | |
2122 | copy_size = ih_item_len(ih) - pos_in_item; | |
2123 | ||
2124 | fs_gen = get_generation(inode->i_sb) ; | |
2125 | copy_item_head(&tmp_ih, ih) ; | |
2126 | ||
2127 | if (!trans_running) { | |
2128 | /* vs-3050 is gone, no need to drop the path */ | |
2129 | retval = journal_begin(&th, inode->i_sb, jbegin_count) ; | |
2130 | if (retval) | |
2131 | goto out; | |
2132 | reiserfs_update_inode_transaction(inode) ; | |
2133 | trans_running = 1; | |
2134 | if (fs_changed(fs_gen, inode->i_sb) && item_moved(&tmp_ih, &path)) { | |
2135 | reiserfs_restore_prepared_buffer(inode->i_sb, bh) ; | |
2136 | goto research; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | reiserfs_prepare_for_journal(inode->i_sb, bh, 1) ; | |
2141 | ||
2142 | if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) { | |
2143 | reiserfs_restore_prepared_buffer(inode->i_sb, bh) ; | |
2144 | goto research; | |
2145 | } | |
2146 | ||
2147 | memcpy( B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied, copy_size) ; | |
2148 | ||
2149 | journal_mark_dirty(&th, inode->i_sb, bh) ; | |
2150 | bytes_copied += copy_size ; | |
2151 | set_block_dev_mapped(bh_result, 0, inode); | |
2152 | ||
2153 | /* are there still bytes left? */ | |
2154 | if (bytes_copied < bh_result->b_size && | |
2155 | (byte_offset + bytes_copied) < inode->i_size) { | |
2156 | set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + copy_size) ; | |
2157 | goto research ; | |
2158 | } | |
2159 | } else { | |
2160 | reiserfs_warning (inode->i_sb, | |
2161 | "clm-6003: bad item inode %lu, device %s", | |
2162 | inode->i_ino, reiserfs_bdevname (inode->i_sb)) ; | |
2163 | retval = -EIO ; | |
2164 | goto out ; | |
2165 | } | |
2166 | retval = 0 ; | |
2167 | ||
2168 | out: | |
2169 | pathrelse(&path) ; | |
2170 | if (trans_running) { | |
2171 | int err = journal_end(&th, inode->i_sb, jbegin_count) ; | |
2172 | if (err) | |
2173 | retval = err; | |
2174 | trans_running = 0; | |
2175 | } | |
2176 | reiserfs_write_unlock(inode->i_sb); | |
2177 | ||
2178 | /* this is where we fill in holes in the file. */ | |
2179 | if (use_get_block) { | |
2180 | retval = reiserfs_get_block(inode, block, bh_result, | |
2181 | GET_BLOCK_CREATE | GET_BLOCK_NO_ISEM | | |
2182 | GET_BLOCK_NO_DANGLE); | |
2183 | if (!retval) { | |
2184 | if (!buffer_mapped(bh_result) || bh_result->b_blocknr == 0) { | |
2185 | /* get_block failed to find a mapped unformatted node. */ | |
2186 | use_get_block = 0 ; | |
2187 | goto start_over ; | |
2188 | } | |
2189 | } | |
2190 | } | |
2191 | kunmap(bh_result->b_page) ; | |
2192 | ||
2193 | if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) { | |
2194 | /* we've copied data from the page into the direct item, so the | |
2195 | * buffer in the page is now clean, mark it to reflect that. | |
2196 | */ | |
2197 | lock_buffer(bh_result); | |
2198 | clear_buffer_dirty(bh_result); | |
2199 | unlock_buffer(bh_result); | |
2200 | } | |
2201 | return retval ; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * mason@suse.com: updated in 2.5.54 to follow the same general io | |
2206 | * start/recovery path as __block_write_full_page, along with special | |
2207 | * code to handle reiserfs tails. | |
2208 | */ | |
2209 | static int reiserfs_write_full_page(struct page *page, struct writeback_control *wbc) { | |
2210 | struct inode *inode = page->mapping->host ; | |
2211 | unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT ; | |
2212 | int error = 0; | |
2213 | unsigned long block ; | |
2214 | struct buffer_head *head, *bh; | |
2215 | int partial = 0 ; | |
2216 | int nr = 0; | |
2217 | int checked = PageChecked(page); | |
2218 | struct reiserfs_transaction_handle th; | |
2219 | struct super_block *s = inode->i_sb; | |
2220 | int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize; | |
2221 | th.t_trans_id = 0; | |
2222 | ||
2223 | /* The page dirty bit is cleared before writepage is called, which | |
2224 | * means we have to tell create_empty_buffers to make dirty buffers | |
2225 | * The page really should be up to date at this point, so tossing | |
2226 | * in the BH_Uptodate is just a sanity check. | |
2227 | */ | |
2228 | if (!page_has_buffers(page)) { | |
2229 | create_empty_buffers(page, s->s_blocksize, | |
2230 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
2231 | } | |
2232 | head = page_buffers(page) ; | |
2233 | ||
2234 | /* last page in the file, zero out any contents past the | |
2235 | ** last byte in the file | |
2236 | */ | |
2237 | if (page->index >= end_index) { | |
2238 | char *kaddr; | |
2239 | unsigned last_offset; | |
2240 | ||
2241 | last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1) ; | |
2242 | /* no file contents in this page */ | |
2243 | if (page->index >= end_index + 1 || !last_offset) { | |
2244 | unlock_page(page); | |
2245 | return 0; | |
2246 | } | |
2247 | kaddr = kmap_atomic(page, KM_USER0); | |
2248 | memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE-last_offset) ; | |
2249 | flush_dcache_page(page) ; | |
2250 | kunmap_atomic(kaddr, KM_USER0) ; | |
2251 | } | |
2252 | bh = head ; | |
2253 | block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits) ; | |
2254 | /* first map all the buffers, logging any direct items we find */ | |
2255 | do { | |
2256 | if ((checked || buffer_dirty(bh)) && (!buffer_mapped(bh) || | |
2257 | (buffer_mapped(bh) && bh->b_blocknr == 0))) { | |
2258 | /* not mapped yet, or it points to a direct item, search | |
2259 | * the btree for the mapping info, and log any direct | |
2260 | * items found | |
2261 | */ | |
2262 | if ((error = map_block_for_writepage(inode, bh, block))) { | |
2263 | goto fail ; | |
2264 | } | |
2265 | } | |
2266 | bh = bh->b_this_page; | |
2267 | block++; | |
2268 | } while(bh != head) ; | |
2269 | ||
2270 | /* | |
2271 | * we start the transaction after map_block_for_writepage, | |
2272 | * because it can create holes in the file (an unbounded operation). | |
2273 | * starting it here, we can make a reliable estimate for how many | |
2274 | * blocks we're going to log | |
2275 | */ | |
2276 | if (checked) { | |
2277 | ClearPageChecked(page); | |
2278 | reiserfs_write_lock(s); | |
2279 | error = journal_begin(&th, s, bh_per_page + 1); | |
2280 | if (error) { | |
2281 | reiserfs_write_unlock(s); | |
2282 | goto fail; | |
2283 | } | |
2284 | reiserfs_update_inode_transaction(inode); | |
2285 | } | |
2286 | /* now go through and lock any dirty buffers on the page */ | |
2287 | do { | |
2288 | get_bh(bh); | |
2289 | if (!buffer_mapped(bh)) | |
2290 | continue; | |
2291 | if (buffer_mapped(bh) && bh->b_blocknr == 0) | |
2292 | continue; | |
2293 | ||
2294 | if (checked) { | |
2295 | reiserfs_prepare_for_journal(s, bh, 1); | |
2296 | journal_mark_dirty(&th, s, bh); | |
2297 | continue; | |
2298 | } | |
2299 | /* from this point on, we know the buffer is mapped to a | |
2300 | * real block and not a direct item | |
2301 | */ | |
2302 | if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) { | |
2303 | lock_buffer(bh); | |
2304 | } else { | |
2305 | if (test_set_buffer_locked(bh)) { | |
2306 | redirty_page_for_writepage(wbc, page); | |
2307 | continue; | |
2308 | } | |
2309 | } | |
2310 | if (test_clear_buffer_dirty(bh)) { | |
2311 | mark_buffer_async_write(bh); | |
2312 | } else { | |
2313 | unlock_buffer(bh); | |
2314 | } | |
2315 | } while((bh = bh->b_this_page) != head); | |
2316 | ||
2317 | if (checked) { | |
2318 | error = journal_end(&th, s, bh_per_page + 1); | |
2319 | reiserfs_write_unlock(s); | |
2320 | if (error) | |
2321 | goto fail; | |
2322 | } | |
2323 | BUG_ON(PageWriteback(page)); | |
2324 | set_page_writeback(page); | |
2325 | unlock_page(page); | |
2326 | ||
2327 | /* | |
2328 | * since any buffer might be the only dirty buffer on the page, | |
2329 | * the first submit_bh can bring the page out of writeback. | |
2330 | * be careful with the buffers. | |
2331 | */ | |
2332 | do { | |
2333 | struct buffer_head *next = bh->b_this_page; | |
2334 | if (buffer_async_write(bh)) { | |
2335 | submit_bh(WRITE, bh); | |
2336 | nr++; | |
2337 | } | |
2338 | put_bh(bh); | |
2339 | bh = next; | |
2340 | } while(bh != head); | |
2341 | ||
2342 | error = 0; | |
2343 | done: | |
2344 | if (nr == 0) { | |
2345 | /* | |
2346 | * if this page only had a direct item, it is very possible for | |
2347 | * no io to be required without there being an error. Or, | |
2348 | * someone else could have locked them and sent them down the | |
2349 | * pipe without locking the page | |
2350 | */ | |
2351 | bh = head ; | |
2352 | do { | |
2353 | if (!buffer_uptodate(bh)) { | |
2354 | partial = 1; | |
2355 | break; | |
2356 | } | |
2357 | bh = bh->b_this_page; | |
2358 | } while(bh != head); | |
2359 | if (!partial) | |
2360 | SetPageUptodate(page); | |
2361 | end_page_writeback(page); | |
2362 | } | |
2363 | return error; | |
2364 | ||
2365 | fail: | |
2366 | /* catches various errors, we need to make sure any valid dirty blocks | |
2367 | * get to the media. The page is currently locked and not marked for | |
2368 | * writeback | |
2369 | */ | |
2370 | ClearPageUptodate(page); | |
2371 | bh = head; | |
2372 | do { | |
2373 | get_bh(bh); | |
2374 | if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) { | |
2375 | lock_buffer(bh); | |
2376 | mark_buffer_async_write(bh); | |
2377 | } else { | |
2378 | /* | |
2379 | * clear any dirty bits that might have come from getting | |
2380 | * attached to a dirty page | |
2381 | */ | |
2382 | clear_buffer_dirty(bh); | |
2383 | } | |
2384 | bh = bh->b_this_page; | |
2385 | } while(bh != head); | |
2386 | SetPageError(page); | |
2387 | BUG_ON(PageWriteback(page)); | |
2388 | set_page_writeback(page); | |
2389 | unlock_page(page); | |
2390 | do { | |
2391 | struct buffer_head *next = bh->b_this_page; | |
2392 | if (buffer_async_write(bh)) { | |
2393 | clear_buffer_dirty(bh); | |
2394 | submit_bh(WRITE, bh); | |
2395 | nr++; | |
2396 | } | |
2397 | put_bh(bh); | |
2398 | bh = next; | |
2399 | } while(bh != head); | |
2400 | goto done; | |
2401 | } | |
2402 | ||
2403 | ||
2404 | static int reiserfs_readpage (struct file *f, struct page * page) | |
2405 | { | |
2406 | return block_read_full_page (page, reiserfs_get_block); | |
2407 | } | |
2408 | ||
2409 | ||
2410 | static int reiserfs_writepage (struct page * page, struct writeback_control *wbc) | |
2411 | { | |
2412 | struct inode *inode = page->mapping->host ; | |
2413 | reiserfs_wait_on_write_block(inode->i_sb) ; | |
2414 | return reiserfs_write_full_page(page, wbc) ; | |
2415 | } | |
2416 | ||
2417 | static int reiserfs_prepare_write(struct file *f, struct page *page, | |
2418 | unsigned from, unsigned to) { | |
2419 | struct inode *inode = page->mapping->host ; | |
2420 | int ret; | |
2421 | int old_ref = 0; | |
2422 | ||
2423 | reiserfs_wait_on_write_block(inode->i_sb) ; | |
2424 | fix_tail_page_for_writing(page) ; | |
2425 | if (reiserfs_transaction_running(inode->i_sb)) { | |
2426 | struct reiserfs_transaction_handle *th; | |
2427 | th = (struct reiserfs_transaction_handle *)current->journal_info; | |
2428 | BUG_ON (!th->t_refcount); | |
2429 | BUG_ON (!th->t_trans_id); | |
2430 | old_ref = th->t_refcount; | |
2431 | th->t_refcount++; | |
2432 | } | |
2433 | ||
2434 | ret = block_prepare_write(page, from, to, reiserfs_get_block) ; | |
2435 | if (ret && reiserfs_transaction_running(inode->i_sb)) { | |
2436 | struct reiserfs_transaction_handle *th = current->journal_info; | |
2437 | /* this gets a little ugly. If reiserfs_get_block returned an | |
2438 | * error and left a transacstion running, we've got to close it, | |
2439 | * and we've got to free handle if it was a persistent transaction. | |
2440 | * | |
2441 | * But, if we had nested into an existing transaction, we need | |
2442 | * to just drop the ref count on the handle. | |
2443 | * | |
2444 | * If old_ref == 0, the transaction is from reiserfs_get_block, | |
2445 | * and it was a persistent trans. Otherwise, it was nested above. | |
2446 | */ | |
2447 | if (th->t_refcount > old_ref) { | |
2448 | if (old_ref) | |
2449 | th->t_refcount--; | |
2450 | else { | |
2451 | int err; | |
2452 | reiserfs_write_lock(inode->i_sb); | |
2453 | err = reiserfs_end_persistent_transaction(th); | |
2454 | reiserfs_write_unlock(inode->i_sb); | |
2455 | if (err) | |
2456 | ret = err; | |
2457 | } | |
2458 | } | |
2459 | } | |
2460 | return ret; | |
2461 | ||
2462 | } | |
2463 | ||
2464 | ||
2465 | static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block) { | |
2466 | return generic_block_bmap(as, block, reiserfs_bmap) ; | |
2467 | } | |
2468 | ||
2469 | static int reiserfs_commit_write(struct file *f, struct page *page, | |
2470 | unsigned from, unsigned to) { | |
2471 | struct inode *inode = page->mapping->host ; | |
2472 | loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
2473 | int ret = 0; | |
2474 | int update_sd = 0; | |
2475 | struct reiserfs_transaction_handle *th = NULL; | |
2476 | ||
2477 | reiserfs_wait_on_write_block(inode->i_sb) ; | |
2478 | if (reiserfs_transaction_running(inode->i_sb)) { | |
2479 | th = current->journal_info; | |
2480 | } | |
2481 | reiserfs_commit_page(inode, page, from, to); | |
2482 | ||
2483 | /* generic_commit_write does this for us, but does not update the | |
2484 | ** transaction tracking stuff when the size changes. So, we have | |
2485 | ** to do the i_size updates here. | |
2486 | */ | |
2487 | if (pos > inode->i_size) { | |
2488 | struct reiserfs_transaction_handle myth ; | |
2489 | reiserfs_write_lock(inode->i_sb); | |
2490 | /* If the file have grown beyond the border where it | |
2491 | can have a tail, unmark it as needing a tail | |
2492 | packing */ | |
2493 | if ( (have_large_tails (inode->i_sb) && inode->i_size > i_block_size (inode)*4) || | |
2494 | (have_small_tails (inode->i_sb) && inode->i_size > i_block_size(inode)) ) | |
2495 | REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask ; | |
2496 | ||
2497 | ret = journal_begin(&myth, inode->i_sb, 1) ; | |
2498 | if (ret) { | |
2499 | reiserfs_write_unlock(inode->i_sb); | |
2500 | goto journal_error; | |
2501 | } | |
2502 | reiserfs_update_inode_transaction(inode) ; | |
2503 | inode->i_size = pos ; | |
2504 | reiserfs_update_sd(&myth, inode) ; | |
2505 | update_sd = 1; | |
2506 | ret = journal_end(&myth, inode->i_sb, 1) ; | |
2507 | reiserfs_write_unlock(inode->i_sb); | |
2508 | if (ret) | |
2509 | goto journal_error; | |
2510 | } | |
2511 | if (th) { | |
2512 | reiserfs_write_lock(inode->i_sb); | |
2513 | if (!update_sd) | |
2514 | reiserfs_update_sd(th, inode) ; | |
2515 | ret = reiserfs_end_persistent_transaction(th); | |
2516 | reiserfs_write_unlock(inode->i_sb); | |
2517 | if (ret) | |
2518 | goto out; | |
2519 | } | |
2520 | ||
2521 | /* we test for O_SYNC here so we can commit the transaction | |
2522 | ** for any packed tails the file might have had | |
2523 | */ | |
2524 | if (f && (f->f_flags & O_SYNC)) { | |
2525 | reiserfs_write_lock(inode->i_sb); | |
2526 | ret = reiserfs_commit_for_inode(inode) ; | |
2527 | reiserfs_write_unlock(inode->i_sb); | |
2528 | } | |
2529 | out: | |
2530 | return ret ; | |
2531 | ||
2532 | journal_error: | |
2533 | if (th) { | |
2534 | reiserfs_write_lock(inode->i_sb); | |
2535 | if (!update_sd) | |
2536 | reiserfs_update_sd(th, inode) ; | |
2537 | ret = reiserfs_end_persistent_transaction(th); | |
2538 | reiserfs_write_unlock(inode->i_sb); | |
2539 | } | |
2540 | ||
2541 | return ret; | |
2542 | } | |
2543 | ||
2544 | void sd_attrs_to_i_attrs( __u16 sd_attrs, struct inode *inode ) | |
2545 | { | |
2546 | if( reiserfs_attrs( inode -> i_sb ) ) { | |
2547 | if( sd_attrs & REISERFS_SYNC_FL ) | |
2548 | inode -> i_flags |= S_SYNC; | |
2549 | else | |
2550 | inode -> i_flags &= ~S_SYNC; | |
2551 | if( sd_attrs & REISERFS_IMMUTABLE_FL ) | |
2552 | inode -> i_flags |= S_IMMUTABLE; | |
2553 | else | |
2554 | inode -> i_flags &= ~S_IMMUTABLE; | |
2555 | if( sd_attrs & REISERFS_APPEND_FL ) | |
2556 | inode -> i_flags |= S_APPEND; | |
2557 | else | |
2558 | inode -> i_flags &= ~S_APPEND; | |
2559 | if( sd_attrs & REISERFS_NOATIME_FL ) | |
2560 | inode -> i_flags |= S_NOATIME; | |
2561 | else | |
2562 | inode -> i_flags &= ~S_NOATIME; | |
2563 | if( sd_attrs & REISERFS_NOTAIL_FL ) | |
2564 | REISERFS_I(inode)->i_flags |= i_nopack_mask; | |
2565 | else | |
2566 | REISERFS_I(inode)->i_flags &= ~i_nopack_mask; | |
2567 | } | |
2568 | } | |
2569 | ||
2570 | void i_attrs_to_sd_attrs( struct inode *inode, __u16 *sd_attrs ) | |
2571 | { | |
2572 | if( reiserfs_attrs( inode -> i_sb ) ) { | |
2573 | if( inode -> i_flags & S_IMMUTABLE ) | |
2574 | *sd_attrs |= REISERFS_IMMUTABLE_FL; | |
2575 | else | |
2576 | *sd_attrs &= ~REISERFS_IMMUTABLE_FL; | |
2577 | if( inode -> i_flags & S_SYNC ) | |
2578 | *sd_attrs |= REISERFS_SYNC_FL; | |
2579 | else | |
2580 | *sd_attrs &= ~REISERFS_SYNC_FL; | |
2581 | if( inode -> i_flags & S_NOATIME ) | |
2582 | *sd_attrs |= REISERFS_NOATIME_FL; | |
2583 | else | |
2584 | *sd_attrs &= ~REISERFS_NOATIME_FL; | |
2585 | if( REISERFS_I(inode)->i_flags & i_nopack_mask ) | |
2586 | *sd_attrs |= REISERFS_NOTAIL_FL; | |
2587 | else | |
2588 | *sd_attrs &= ~REISERFS_NOTAIL_FL; | |
2589 | } | |
2590 | } | |
2591 | ||
2592 | /* decide if this buffer needs to stay around for data logging or ordered | |
2593 | ** write purposes | |
2594 | */ | |
2595 | static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh) | |
2596 | { | |
2597 | int ret = 1 ; | |
2598 | struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb) ; | |
2599 | ||
2600 | spin_lock(&j->j_dirty_buffers_lock) ; | |
2601 | if (!buffer_mapped(bh)) { | |
2602 | goto free_jh; | |
2603 | } | |
2604 | /* the page is locked, and the only places that log a data buffer | |
2605 | * also lock the page. | |
2606 | */ | |
2607 | if (reiserfs_file_data_log(inode)) { | |
2608 | /* | |
2609 | * very conservative, leave the buffer pinned if | |
2610 | * anyone might need it. | |
2611 | */ | |
2612 | if (buffer_journaled(bh) || buffer_journal_dirty(bh)) { | |
2613 | ret = 0 ; | |
2614 | } | |
2615 | } else | |
2616 | if (buffer_dirty(bh) || buffer_locked(bh)) { | |
2617 | struct reiserfs_journal_list *jl; | |
2618 | struct reiserfs_jh *jh = bh->b_private; | |
2619 | ||
2620 | /* why is this safe? | |
2621 | * reiserfs_setattr updates i_size in the on disk | |
2622 | * stat data before allowing vmtruncate to be called. | |
2623 | * | |
2624 | * If buffer was put onto the ordered list for this | |
2625 | * transaction, we know for sure either this transaction | |
2626 | * or an older one already has updated i_size on disk, | |
2627 | * and this ordered data won't be referenced in the file | |
2628 | * if we crash. | |
2629 | * | |
2630 | * if the buffer was put onto the ordered list for an older | |
2631 | * transaction, we need to leave it around | |
2632 | */ | |
2633 | if (jh && (jl = jh->jl) && jl != SB_JOURNAL(inode->i_sb)->j_current_jl) | |
2634 | ret = 0; | |
2635 | } | |
2636 | free_jh: | |
2637 | if (ret && bh->b_private) { | |
2638 | reiserfs_free_jh(bh); | |
2639 | } | |
2640 | spin_unlock(&j->j_dirty_buffers_lock) ; | |
2641 | return ret ; | |
2642 | } | |
2643 | ||
2644 | /* clm -- taken from fs/buffer.c:block_invalidate_page */ | |
2645 | static int reiserfs_invalidatepage(struct page *page, unsigned long offset) | |
2646 | { | |
2647 | struct buffer_head *head, *bh, *next; | |
2648 | struct inode *inode = page->mapping->host; | |
2649 | unsigned int curr_off = 0; | |
2650 | int ret = 1; | |
2651 | ||
2652 | BUG_ON(!PageLocked(page)); | |
2653 | ||
2654 | if (offset == 0) | |
2655 | ClearPageChecked(page); | |
2656 | ||
2657 | if (!page_has_buffers(page)) | |
2658 | goto out; | |
2659 | ||
2660 | head = page_buffers(page); | |
2661 | bh = head; | |
2662 | do { | |
2663 | unsigned int next_off = curr_off + bh->b_size; | |
2664 | next = bh->b_this_page; | |
2665 | ||
2666 | /* | |
2667 | * is this block fully invalidated? | |
2668 | */ | |
2669 | if (offset <= curr_off) { | |
2670 | if (invalidatepage_can_drop(inode, bh)) | |
2671 | reiserfs_unmap_buffer(bh); | |
2672 | else | |
2673 | ret = 0; | |
2674 | } | |
2675 | curr_off = next_off; | |
2676 | bh = next; | |
2677 | } while (bh != head); | |
2678 | ||
2679 | /* | |
2680 | * We release buffers only if the entire page is being invalidated. | |
2681 | * The get_block cached value has been unconditionally invalidated, | |
2682 | * so real IO is not possible anymore. | |
2683 | */ | |
2684 | if (!offset && ret) | |
2685 | ret = try_to_release_page(page, 0); | |
2686 | out: | |
2687 | return ret; | |
2688 | } | |
2689 | ||
2690 | static int reiserfs_set_page_dirty(struct page *page) { | |
2691 | struct inode *inode = page->mapping->host; | |
2692 | if (reiserfs_file_data_log(inode)) { | |
2693 | SetPageChecked(page); | |
2694 | return __set_page_dirty_nobuffers(page); | |
2695 | } | |
2696 | return __set_page_dirty_buffers(page); | |
2697 | } | |
2698 | ||
2699 | /* | |
2700 | * Returns 1 if the page's buffers were dropped. The page is locked. | |
2701 | * | |
2702 | * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads | |
2703 | * in the buffers at page_buffers(page). | |
2704 | * | |
2705 | * even in -o notail mode, we can't be sure an old mount without -o notail | |
2706 | * didn't create files with tails. | |
2707 | */ | |
2708 | static int reiserfs_releasepage(struct page *page, int unused_gfp_flags) | |
2709 | { | |
2710 | struct inode *inode = page->mapping->host ; | |
2711 | struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb) ; | |
2712 | struct buffer_head *head ; | |
2713 | struct buffer_head *bh ; | |
2714 | int ret = 1 ; | |
2715 | ||
2716 | WARN_ON(PageChecked(page)); | |
2717 | spin_lock(&j->j_dirty_buffers_lock) ; | |
2718 | head = page_buffers(page) ; | |
2719 | bh = head ; | |
2720 | do { | |
2721 | if (bh->b_private) { | |
2722 | if (!buffer_dirty(bh) && !buffer_locked(bh)) { | |
2723 | reiserfs_free_jh(bh); | |
2724 | } else { | |
2725 | ret = 0 ; | |
2726 | break ; | |
2727 | } | |
2728 | } | |
2729 | bh = bh->b_this_page ; | |
2730 | } while (bh != head) ; | |
2731 | if (ret) | |
2732 | ret = try_to_free_buffers(page) ; | |
2733 | spin_unlock(&j->j_dirty_buffers_lock) ; | |
2734 | return ret ; | |
2735 | } | |
2736 | ||
2737 | /* We thank Mingming Cao for helping us understand in great detail what | |
2738 | to do in this section of the code. */ | |
2739 | static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb, | |
2740 | const struct iovec *iov, loff_t offset, unsigned long nr_segs) | |
2741 | { | |
2742 | struct file *file = iocb->ki_filp; | |
2743 | struct inode *inode = file->f_mapping->host; | |
2744 | ||
2745 | return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
2746 | offset, nr_segs, reiserfs_get_blocks_direct_io, NULL); | |
2747 | } | |
2748 | ||
2749 | int reiserfs_setattr(struct dentry *dentry, struct iattr *attr) { | |
2750 | struct inode *inode = dentry->d_inode ; | |
2751 | int error ; | |
2752 | unsigned int ia_valid = attr->ia_valid; | |
2753 | reiserfs_write_lock(inode->i_sb); | |
2754 | if (attr->ia_valid & ATTR_SIZE) { | |
2755 | /* version 2 items will be caught by the s_maxbytes check | |
2756 | ** done for us in vmtruncate | |
2757 | */ | |
2758 | if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 && | |
2759 | attr->ia_size > MAX_NON_LFS) { | |
2760 | error = -EFBIG ; | |
2761 | goto out; | |
2762 | } | |
2763 | /* fill in hole pointers in the expanding truncate case. */ | |
2764 | if (attr->ia_size > inode->i_size) { | |
2765 | error = generic_cont_expand(inode, attr->ia_size) ; | |
2766 | if (REISERFS_I(inode)->i_prealloc_count > 0) { | |
2767 | int err; | |
2768 | struct reiserfs_transaction_handle th ; | |
2769 | /* we're changing at most 2 bitmaps, inode + super */ | |
2770 | err = journal_begin(&th, inode->i_sb, 4) ; | |
2771 | if (!err) { | |
2772 | reiserfs_discard_prealloc (&th, inode); | |
2773 | err = journal_end(&th, inode->i_sb, 4) ; | |
2774 | } | |
2775 | if (err) | |
2776 | error = err; | |
2777 | } | |
2778 | if (error) | |
2779 | goto out; | |
2780 | } | |
2781 | } | |
2782 | ||
2783 | if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) || | |
2784 | ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) && | |
2785 | (get_inode_sd_version (inode) == STAT_DATA_V1)) { | |
2786 | /* stat data of format v3.5 has 16 bit uid and gid */ | |
2787 | error = -EINVAL; | |
2788 | goto out; | |
2789 | } | |
2790 | ||
2791 | error = inode_change_ok(inode, attr) ; | |
2792 | if (!error) { | |
2793 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
2794 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
2795 | error = reiserfs_chown_xattrs (inode, attr); | |
2796 | ||
2797 | if (!error) { | |
2798 | struct reiserfs_transaction_handle th; | |
2799 | ||
2800 | /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */ | |
2801 | journal_begin(&th, inode->i_sb, 4*REISERFS_QUOTA_INIT_BLOCKS+2); | |
2802 | error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0; | |
2803 | if (error) { | |
2804 | journal_end(&th, inode->i_sb, 4*REISERFS_QUOTA_INIT_BLOCKS+2); | |
2805 | goto out; | |
2806 | } | |
2807 | /* Update corresponding info in inode so that everything is in | |
2808 | * one transaction */ | |
2809 | if (attr->ia_valid & ATTR_UID) | |
2810 | inode->i_uid = attr->ia_uid; | |
2811 | if (attr->ia_valid & ATTR_GID) | |
2812 | inode->i_gid = attr->ia_gid; | |
2813 | mark_inode_dirty(inode); | |
2814 | journal_end(&th, inode->i_sb, 4*REISERFS_QUOTA_INIT_BLOCKS+2); | |
2815 | } | |
2816 | } | |
2817 | if (!error) | |
2818 | error = inode_setattr(inode, attr) ; | |
2819 | } | |
2820 | ||
2821 | ||
2822 | if (!error && reiserfs_posixacl (inode->i_sb)) { | |
2823 | if (attr->ia_valid & ATTR_MODE) | |
2824 | error = reiserfs_acl_chmod (inode); | |
2825 | } | |
2826 | ||
2827 | out: | |
2828 | reiserfs_write_unlock(inode->i_sb); | |
2829 | return error ; | |
2830 | } | |
2831 | ||
2832 | ||
2833 | ||
2834 | struct address_space_operations reiserfs_address_space_operations = { | |
2835 | .writepage = reiserfs_writepage, | |
2836 | .readpage = reiserfs_readpage, | |
2837 | .readpages = reiserfs_readpages, | |
2838 | .releasepage = reiserfs_releasepage, | |
2839 | .invalidatepage = reiserfs_invalidatepage, | |
2840 | .sync_page = block_sync_page, | |
2841 | .prepare_write = reiserfs_prepare_write, | |
2842 | .commit_write = reiserfs_commit_write, | |
2843 | .bmap = reiserfs_aop_bmap, | |
2844 | .direct_IO = reiserfs_direct_IO, | |
2845 | .set_page_dirty = reiserfs_set_page_dirty, | |
2846 | } ; |