aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ubifs/file.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ubifs/file.c')
-rw-r--r--fs/ubifs/file.c1275
1 files changed, 1275 insertions, 0 deletions
diff --git a/fs/ubifs/file.c b/fs/ubifs/file.c
new file mode 100644
index 000000000000..005a3b854d96
--- /dev/null
+++ b/fs/ubifs/file.c
@@ -0,0 +1,1275 @@
1/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors: Artem Bityutskiy (Битюцкий Артём)
20 * Adrian Hunter
21 */
22
23/*
24 * This file implements VFS file and inode operations of regular files, device
25 * nodes and symlinks as well as address space operations.
26 *
27 * UBIFS uses 2 page flags: PG_private and PG_checked. PG_private is set if the
28 * page is dirty and is used for budgeting purposes - dirty pages should not be
29 * budgeted. The PG_checked flag is set if full budgeting is required for the
30 * page e.g., when it corresponds to a file hole or it is just beyond the file
31 * size. The budgeting is done in 'ubifs_write_begin()', because it is OK to
32 * fail in this function, and the budget is released in 'ubifs_write_end()'. So
33 * the PG_private and PG_checked flags carry the information about how the page
34 * was budgeted, to make it possible to release the budget properly.
35 *
36 * A thing to keep in mind: inode's 'i_mutex' is locked in most VFS operations
37 * we implement. However, this is not true for '->writepage()', which might be
38 * called with 'i_mutex' unlocked. For example, when pdflush is performing
39 * write-back, it calls 'writepage()' with unlocked 'i_mutex', although the
40 * inode has 'I_LOCK' flag in this case. At "normal" work-paths 'i_mutex' is
41 * locked in '->writepage', e.g. in "sys_write -> alloc_pages -> direct reclaim
42 * path'. So, in '->writepage()' we are only guaranteed that the page is
43 * locked.
44 *
45 * Similarly, 'i_mutex' does not have to be locked in readpage(), e.g.,
46 * readahead path does not have it locked ("sys_read -> generic_file_aio_read
47 * -> ondemand_readahead -> readpage"). In case of readahead, 'I_LOCK' flag is
48 * not set as well. However, UBIFS disables readahead.
49 *
50 * This, for example means that there might be 2 concurrent '->writepage()'
51 * calls for the same inode, but different inode dirty pages.
52 */
53
54#include "ubifs.h"
55#include <linux/mount.h>
56
57static int read_block(struct inode *inode, void *addr, unsigned int block,
58 struct ubifs_data_node *dn)
59{
60 struct ubifs_info *c = inode->i_sb->s_fs_info;
61 int err, len, out_len;
62 union ubifs_key key;
63 unsigned int dlen;
64
65 data_key_init(c, &key, inode->i_ino, block);
66 err = ubifs_tnc_lookup(c, &key, dn);
67 if (err) {
68 if (err == -ENOENT)
69 /* Not found, so it must be a hole */
70 memset(addr, 0, UBIFS_BLOCK_SIZE);
71 return err;
72 }
73
74 ubifs_assert(dn->ch.sqnum > ubifs_inode(inode)->creat_sqnum);
75
76 len = le32_to_cpu(dn->size);
77 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
78 goto dump;
79
80 dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
81 out_len = UBIFS_BLOCK_SIZE;
82 err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
83 le16_to_cpu(dn->compr_type));
84 if (err || len != out_len)
85 goto dump;
86
87 /*
88 * Data length can be less than a full block, even for blocks that are
89 * not the last in the file (e.g., as a result of making a hole and
90 * appending data). Ensure that the remainder is zeroed out.
91 */
92 if (len < UBIFS_BLOCK_SIZE)
93 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
94
95 return 0;
96
97dump:
98 ubifs_err("bad data node (block %u, inode %lu)",
99 block, inode->i_ino);
100 dbg_dump_node(c, dn);
101 return -EINVAL;
102}
103
104static int do_readpage(struct page *page)
105{
106 void *addr;
107 int err = 0, i;
108 unsigned int block, beyond;
109 struct ubifs_data_node *dn;
110 struct inode *inode = page->mapping->host;
111 loff_t i_size = i_size_read(inode);
112
113 dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
114 inode->i_ino, page->index, i_size, page->flags);
115 ubifs_assert(!PageChecked(page));
116 ubifs_assert(!PagePrivate(page));
117
118 addr = kmap(page);
119
120 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
121 beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
122 if (block >= beyond) {
123 /* Reading beyond inode */
124 SetPageChecked(page);
125 memset(addr, 0, PAGE_CACHE_SIZE);
126 goto out;
127 }
128
129 dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
130 if (!dn) {
131 err = -ENOMEM;
132 goto error;
133 }
134
135 i = 0;
136 while (1) {
137 int ret;
138
139 if (block >= beyond) {
140 /* Reading beyond inode */
141 err = -ENOENT;
142 memset(addr, 0, UBIFS_BLOCK_SIZE);
143 } else {
144 ret = read_block(inode, addr, block, dn);
145 if (ret) {
146 err = ret;
147 if (err != -ENOENT)
148 break;
149 }
150 }
151 if (++i >= UBIFS_BLOCKS_PER_PAGE)
152 break;
153 block += 1;
154 addr += UBIFS_BLOCK_SIZE;
155 }
156 if (err) {
157 if (err == -ENOENT) {
158 /* Not found, so it must be a hole */
159 SetPageChecked(page);
160 dbg_gen("hole");
161 goto out_free;
162 }
163 ubifs_err("cannot read page %lu of inode %lu, error %d",
164 page->index, inode->i_ino, err);
165 goto error;
166 }
167
168out_free:
169 kfree(dn);
170out:
171 SetPageUptodate(page);
172 ClearPageError(page);
173 flush_dcache_page(page);
174 kunmap(page);
175 return 0;
176
177error:
178 kfree(dn);
179 ClearPageUptodate(page);
180 SetPageError(page);
181 flush_dcache_page(page);
182 kunmap(page);
183 return err;
184}
185
186/**
187 * release_new_page_budget - release budget of a new page.
188 * @c: UBIFS file-system description object
189 *
190 * This is a helper function which releases budget corresponding to the budget
191 * of one new page of data.
192 */
193static void release_new_page_budget(struct ubifs_info *c)
194{
195 struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
196
197 ubifs_release_budget(c, &req);
198}
199
200/**
201 * release_existing_page_budget - release budget of an existing page.
202 * @c: UBIFS file-system description object
203 *
204 * This is a helper function which releases budget corresponding to the budget
205 * of changing one one page of data which already exists on the flash media.
206 */
207static void release_existing_page_budget(struct ubifs_info *c)
208{
209 struct ubifs_budget_req req = { .dd_growth = c->page_budget};
210
211 ubifs_release_budget(c, &req);
212}
213
214static int write_begin_slow(struct address_space *mapping,
215 loff_t pos, unsigned len, struct page **pagep)
216{
217 struct inode *inode = mapping->host;
218 struct ubifs_info *c = inode->i_sb->s_fs_info;
219 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
220 struct ubifs_budget_req req = { .new_page = 1 };
221 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
222 struct page *page;
223
224 dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
225 inode->i_ino, pos, len, inode->i_size);
226
227 /*
228 * At the slow path we have to budget before locking the page, because
229 * budgeting may force write-back, which would wait on locked pages and
230 * deadlock if we had the page locked. At this point we do not know
231 * anything about the page, so assume that this is a new page which is
232 * written to a hole. This corresponds to largest budget. Later the
233 * budget will be amended if this is not true.
234 */
235 if (appending)
236 /* We are appending data, budget for inode change */
237 req.dirtied_ino = 1;
238
239 err = ubifs_budget_space(c, &req);
240 if (unlikely(err))
241 return err;
242
243 page = __grab_cache_page(mapping, index);
244 if (unlikely(!page)) {
245 ubifs_release_budget(c, &req);
246 return -ENOMEM;
247 }
248
249 if (!PageUptodate(page)) {
250 if (!(pos & PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
251 SetPageChecked(page);
252 else {
253 err = do_readpage(page);
254 if (err) {
255 unlock_page(page);
256 page_cache_release(page);
257 return err;
258 }
259 }
260
261 SetPageUptodate(page);
262 ClearPageError(page);
263 }
264
265 if (PagePrivate(page))
266 /*
267 * The page is dirty, which means it was budgeted twice:
268 * o first time the budget was allocated by the task which
269 * made the page dirty and set the PG_private flag;
270 * o and then we budgeted for it for the second time at the
271 * very beginning of this function.
272 *
273 * So what we have to do is to release the page budget we
274 * allocated.
275 */
276 release_new_page_budget(c);
277 else if (!PageChecked(page))
278 /*
279 * We are changing a page which already exists on the media.
280 * This means that changing the page does not make the amount
281 * of indexing information larger, and this part of the budget
282 * which we have already acquired may be released.
283 */
284 ubifs_convert_page_budget(c);
285
286 if (appending) {
287 struct ubifs_inode *ui = ubifs_inode(inode);
288
289 /*
290 * 'ubifs_write_end()' is optimized from the fast-path part of
291 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
292 * if data is appended.
293 */
294 mutex_lock(&ui->ui_mutex);
295 if (ui->dirty)
296 /*
297 * The inode is dirty already, so we may free the
298 * budget we allocated.
299 */
300 ubifs_release_dirty_inode_budget(c, ui);
301 }
302
303 *pagep = page;
304 return 0;
305}
306
307/**
308 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
309 * @c: UBIFS file-system description object
310 * @page: page to allocate budget for
311 * @ui: UBIFS inode object the page belongs to
312 * @appending: non-zero if the page is appended
313 *
314 * This is a helper function for 'ubifs_write_begin()' which allocates budget
315 * for the operation. The budget is allocated differently depending on whether
316 * this is appending, whether the page is dirty or not, and so on. This
317 * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
318 * in case of success and %-ENOSPC in case of failure.
319 */
320static int allocate_budget(struct ubifs_info *c, struct page *page,
321 struct ubifs_inode *ui, int appending)
322{
323 struct ubifs_budget_req req = { .fast = 1 };
324
325 if (PagePrivate(page)) {
326 if (!appending)
327 /*
328 * The page is dirty and we are not appending, which
329 * means no budget is needed at all.
330 */
331 return 0;
332
333 mutex_lock(&ui->ui_mutex);
334 if (ui->dirty)
335 /*
336 * The page is dirty and we are appending, so the inode
337 * has to be marked as dirty. However, it is already
338 * dirty, so we do not need any budget. We may return,
339 * but @ui->ui_mutex hast to be left locked because we
340 * should prevent write-back from flushing the inode
341 * and freeing the budget. The lock will be released in
342 * 'ubifs_write_end()'.
343 */
344 return 0;
345
346 /*
347 * The page is dirty, we are appending, the inode is clean, so
348 * we need to budget the inode change.
349 */
350 req.dirtied_ino = 1;
351 } else {
352 if (PageChecked(page))
353 /*
354 * The page corresponds to a hole and does not
355 * exist on the media. So changing it makes
356 * make the amount of indexing information
357 * larger, and we have to budget for a new
358 * page.
359 */
360 req.new_page = 1;
361 else
362 /*
363 * Not a hole, the change will not add any new
364 * indexing information, budget for page
365 * change.
366 */
367 req.dirtied_page = 1;
368
369 if (appending) {
370 mutex_lock(&ui->ui_mutex);
371 if (!ui->dirty)
372 /*
373 * The inode is clean but we will have to mark
374 * it as dirty because we are appending. This
375 * needs a budget.
376 */
377 req.dirtied_ino = 1;
378 }
379 }
380
381 return ubifs_budget_space(c, &req);
382}
383
384/*
385 * This function is called when a page of data is going to be written. Since
386 * the page of data will not necessarily go to the flash straight away, UBIFS
387 * has to reserve space on the media for it, which is done by means of
388 * budgeting.
389 *
390 * This is the hot-path of the file-system and we are trying to optimize it as
391 * much as possible. For this reasons it is split on 2 parts - slow and fast.
392 *
393 * There many budgeting cases:
394 * o a new page is appended - we have to budget for a new page and for
395 * changing the inode; however, if the inode is already dirty, there is
396 * no need to budget for it;
397 * o an existing clean page is changed - we have budget for it; if the page
398 * does not exist on the media (a hole), we have to budget for a new
399 * page; otherwise, we may budget for changing an existing page; the
400 * difference between these cases is that changing an existing page does
401 * not introduce anything new to the FS indexing information, so it does
402 * not grow, and smaller budget is acquired in this case;
403 * o an existing dirty page is changed - no need to budget at all, because
404 * the page budget has been acquired by earlier, when the page has been
405 * marked dirty.
406 *
407 * UBIFS budgeting sub-system may force write-back if it thinks there is no
408 * space to reserve. This imposes some locking restrictions and makes it
409 * impossible to take into account the above cases, and makes it impossible to
410 * optimize budgeting.
411 *
412 * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
413 * there is a plenty of flash space and the budget will be acquired quickly,
414 * without forcing write-back. The slow path does not make this assumption.
415 */
416static int ubifs_write_begin(struct file *file, struct address_space *mapping,
417 loff_t pos, unsigned len, unsigned flags,
418 struct page **pagep, void **fsdata)
419{
420 struct inode *inode = mapping->host;
421 struct ubifs_info *c = inode->i_sb->s_fs_info;
422 struct ubifs_inode *ui = ubifs_inode(inode);
423 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
424 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
425 struct page *page;
426
427
428 ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
429
430 if (unlikely(c->ro_media))
431 return -EROFS;
432
433 /* Try out the fast-path part first */
434 page = __grab_cache_page(mapping, index);
435 if (unlikely(!page))
436 return -ENOMEM;
437
438 if (!PageUptodate(page)) {
439 /* The page is not loaded from the flash */
440 if (!(pos & PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
441 /*
442 * We change whole page so no need to load it. But we
443 * have to set the @PG_checked flag to make the further
444 * code the page is new. This might be not true, but it
445 * is better to budget more that to read the page from
446 * the media.
447 */
448 SetPageChecked(page);
449 else {
450 err = do_readpage(page);
451 if (err) {
452 unlock_page(page);
453 page_cache_release(page);
454 return err;
455 }
456 }
457
458 SetPageUptodate(page);
459 ClearPageError(page);
460 }
461
462 err = allocate_budget(c, page, ui, appending);
463 if (unlikely(err)) {
464 ubifs_assert(err == -ENOSPC);
465 /*
466 * Budgeting failed which means it would have to force
467 * write-back but didn't, because we set the @fast flag in the
468 * request. Write-back cannot be done now, while we have the
469 * page locked, because it would deadlock. Unlock and free
470 * everything and fall-back to slow-path.
471 */
472 if (appending) {
473 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
474 mutex_unlock(&ui->ui_mutex);
475 }
476 unlock_page(page);
477 page_cache_release(page);
478
479 return write_begin_slow(mapping, pos, len, pagep);
480 }
481
482 /*
483 * Whee, we aquired budgeting quickly - without involving
484 * garbage-collection, committing or forceing write-back. We return
485 * with @ui->ui_mutex locked if we are appending pages, and unlocked
486 * otherwise. This is an optimization (slightly hacky though).
487 */
488 *pagep = page;
489 return 0;
490
491}
492
493/**
494 * cancel_budget - cancel budget.
495 * @c: UBIFS file-system description object
496 * @page: page to cancel budget for
497 * @ui: UBIFS inode object the page belongs to
498 * @appending: non-zero if the page is appended
499 *
500 * This is a helper function for a page write operation. It unlocks the
501 * @ui->ui_mutex in case of appending.
502 */
503static void cancel_budget(struct ubifs_info *c, struct page *page,
504 struct ubifs_inode *ui, int appending)
505{
506 if (appending) {
507 if (!ui->dirty)
508 ubifs_release_dirty_inode_budget(c, ui);
509 mutex_unlock(&ui->ui_mutex);
510 }
511 if (!PagePrivate(page)) {
512 if (PageChecked(page))
513 release_new_page_budget(c);
514 else
515 release_existing_page_budget(c);
516 }
517}
518
519static int ubifs_write_end(struct file *file, struct address_space *mapping,
520 loff_t pos, unsigned len, unsigned copied,
521 struct page *page, void *fsdata)
522{
523 struct inode *inode = mapping->host;
524 struct ubifs_inode *ui = ubifs_inode(inode);
525 struct ubifs_info *c = inode->i_sb->s_fs_info;
526 loff_t end_pos = pos + len;
527 int appending = !!(end_pos > inode->i_size);
528
529 dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
530 inode->i_ino, pos, page->index, len, copied, inode->i_size);
531
532 if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
533 /*
534 * VFS copied less data to the page that it intended and
535 * declared in its '->write_begin()' call via the @len
536 * argument. If the page was not up-to-date, and @len was
537 * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
538 * not load it from the media (for optimization reasons). This
539 * means that part of the page contains garbage. So read the
540 * page now.
541 */
542 dbg_gen("copied %d instead of %d, read page and repeat",
543 copied, len);
544 cancel_budget(c, page, ui, appending);
545
546 /*
547 * Return 0 to force VFS to repeat the whole operation, or the
548 * error code if 'do_readpage()' failes.
549 */
550 copied = do_readpage(page);
551 goto out;
552 }
553
554 if (!PagePrivate(page)) {
555 SetPagePrivate(page);
556 atomic_long_inc(&c->dirty_pg_cnt);
557 __set_page_dirty_nobuffers(page);
558 }
559
560 if (appending) {
561 i_size_write(inode, end_pos);
562 ui->ui_size = end_pos;
563 /*
564 * Note, we do not set @I_DIRTY_PAGES (which means that the
565 * inode has dirty pages), this has been done in
566 * '__set_page_dirty_nobuffers()'.
567 */
568 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
569 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
570 mutex_unlock(&ui->ui_mutex);
571 }
572
573out:
574 unlock_page(page);
575 page_cache_release(page);
576 return copied;
577}
578
579static int ubifs_readpage(struct file *file, struct page *page)
580{
581 do_readpage(page);
582 unlock_page(page);
583 return 0;
584}
585
586static int do_writepage(struct page *page, int len)
587{
588 int err = 0, i, blen;
589 unsigned int block;
590 void *addr;
591 union ubifs_key key;
592 struct inode *inode = page->mapping->host;
593 struct ubifs_info *c = inode->i_sb->s_fs_info;
594
595#ifdef UBIFS_DEBUG
596 spin_lock(&ui->ui_lock);
597 ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
598 spin_unlock(&ui->ui_lock);
599#endif
600
601 /* Update radix tree tags */
602 set_page_writeback(page);
603
604 addr = kmap(page);
605 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
606 i = 0;
607 while (len) {
608 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
609 data_key_init(c, &key, inode->i_ino, block);
610 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
611 if (err)
612 break;
613 if (++i >= UBIFS_BLOCKS_PER_PAGE)
614 break;
615 block += 1;
616 addr += blen;
617 len -= blen;
618 }
619 if (err) {
620 SetPageError(page);
621 ubifs_err("cannot write page %lu of inode %lu, error %d",
622 page->index, inode->i_ino, err);
623 ubifs_ro_mode(c, err);
624 }
625
626 ubifs_assert(PagePrivate(page));
627 if (PageChecked(page))
628 release_new_page_budget(c);
629 else
630 release_existing_page_budget(c);
631
632 atomic_long_dec(&c->dirty_pg_cnt);
633 ClearPagePrivate(page);
634 ClearPageChecked(page);
635
636 kunmap(page);
637 unlock_page(page);
638 end_page_writeback(page);
639 return err;
640}
641
642/*
643 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
644 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
645 * situation when a we have an inode with size 0, then a megabyte of data is
646 * appended to the inode, then write-back starts and flushes some amount of the
647 * dirty pages, the journal becomes full, commit happens and finishes, and then
648 * an unclean reboot happens. When the file system is mounted next time, the
649 * inode size would still be 0, but there would be many pages which are beyond
650 * the inode size, they would be indexed and consume flash space. Because the
651 * journal has been committed, the replay would not be able to detect this
652 * situation and correct the inode size. This means UBIFS would have to scan
653 * whole index and correct all inode sizes, which is long an unacceptable.
654 *
655 * To prevent situations like this, UBIFS writes pages back only if they are
656 * within last synchronized inode size, i.e. the the size which has been
657 * written to the flash media last time. Otherwise, UBIFS forces inode
658 * write-back, thus making sure the on-flash inode contains current inode size,
659 * and then keeps writing pages back.
660 *
661 * Some locking issues explanation. 'ubifs_writepage()' first is called with
662 * the page locked, and it locks @ui_mutex. However, write-back does take inode
663 * @i_mutex, which means other VFS operations may be run on this inode at the
664 * same time. And the problematic one is truncation to smaller size, from where
665 * we have to call 'vmtruncate()', which first changes @inode->i_size, then
666 * drops the truncated pages. And while dropping the pages, it takes the page
667 * lock. This means that 'do_truncation()' cannot call 'vmtruncate()' with
668 * @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'. This
669 * means that @inode->i_size is changed while @ui_mutex is unlocked.
670 *
671 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
672 * inode size. How do we do this if @inode->i_size may became smaller while we
673 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
674 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
675 * internally and updates it under @ui_mutex.
676 *
677 * Q: why we do not worry that if we race with truncation, we may end up with a
678 * situation when the inode is truncated while we are in the middle of
679 * 'do_writepage()', so we do write beyond inode size?
680 * A: If we are in the middle of 'do_writepage()', truncation would be locked
681 * on the page lock and it would not write the truncated inode node to the
682 * journal before we have finished.
683 */
684static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
685{
686 struct inode *inode = page->mapping->host;
687 struct ubifs_inode *ui = ubifs_inode(inode);
688 loff_t i_size = i_size_read(inode), synced_i_size;
689 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
690 int err, len = i_size & (PAGE_CACHE_SIZE - 1);
691 void *kaddr;
692
693 dbg_gen("ino %lu, pg %lu, pg flags %#lx",
694 inode->i_ino, page->index, page->flags);
695 ubifs_assert(PagePrivate(page));
696
697 /* Is the page fully outside @i_size? (truncate in progress) */
698 if (page->index > end_index || (page->index == end_index && !len)) {
699 err = 0;
700 goto out_unlock;
701 }
702
703 spin_lock(&ui->ui_lock);
704 synced_i_size = ui->synced_i_size;
705 spin_unlock(&ui->ui_lock);
706
707 /* Is the page fully inside @i_size? */
708 if (page->index < end_index) {
709 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
710 err = inode->i_sb->s_op->write_inode(inode, 1);
711 if (err)
712 goto out_unlock;
713 /*
714 * The inode has been written, but the write-buffer has
715 * not been synchronized, so in case of an unclean
716 * reboot we may end up with some pages beyond inode
717 * size, but they would be in the journal (because
718 * commit flushes write buffers) and recovery would deal
719 * with this.
720 */
721 }
722 return do_writepage(page, PAGE_CACHE_SIZE);
723 }
724
725 /*
726 * The page straddles @i_size. It must be zeroed out on each and every
727 * writepage invocation because it may be mmapped. "A file is mapped
728 * in multiples of the page size. For a file that is not a multiple of
729 * the page size, the remaining memory is zeroed when mapped, and
730 * writes to that region are not written out to the file."
731 */
732 kaddr = kmap_atomic(page, KM_USER0);
733 memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
734 flush_dcache_page(page);
735 kunmap_atomic(kaddr, KM_USER0);
736
737 if (i_size > synced_i_size) {
738 err = inode->i_sb->s_op->write_inode(inode, 1);
739 if (err)
740 goto out_unlock;
741 }
742
743 return do_writepage(page, len);
744
745out_unlock:
746 unlock_page(page);
747 return err;
748}
749
750/**
751 * do_attr_changes - change inode attributes.
752 * @inode: inode to change attributes for
753 * @attr: describes attributes to change
754 */
755static void do_attr_changes(struct inode *inode, const struct iattr *attr)
756{
757 if (attr->ia_valid & ATTR_UID)
758 inode->i_uid = attr->ia_uid;
759 if (attr->ia_valid & ATTR_GID)
760 inode->i_gid = attr->ia_gid;
761 if (attr->ia_valid & ATTR_ATIME)
762 inode->i_atime = timespec_trunc(attr->ia_atime,
763 inode->i_sb->s_time_gran);
764 if (attr->ia_valid & ATTR_MTIME)
765 inode->i_mtime = timespec_trunc(attr->ia_mtime,
766 inode->i_sb->s_time_gran);
767 if (attr->ia_valid & ATTR_CTIME)
768 inode->i_ctime = timespec_trunc(attr->ia_ctime,
769 inode->i_sb->s_time_gran);
770 if (attr->ia_valid & ATTR_MODE) {
771 umode_t mode = attr->ia_mode;
772
773 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
774 mode &= ~S_ISGID;
775 inode->i_mode = mode;
776 }
777}
778
779/**
780 * do_truncation - truncate an inode.
781 * @c: UBIFS file-system description object
782 * @inode: inode to truncate
783 * @attr: inode attribute changes description
784 *
785 * This function implements VFS '->setattr()' call when the inode is truncated
786 * to a smaller size. Returns zero in case of success and a negative error code
787 * in case of failure.
788 */
789static int do_truncation(struct ubifs_info *c, struct inode *inode,
790 const struct iattr *attr)
791{
792 int err;
793 struct ubifs_budget_req req;
794 loff_t old_size = inode->i_size, new_size = attr->ia_size;
795 int offset = new_size & (UBIFS_BLOCK_SIZE - 1);
796 struct ubifs_inode *ui = ubifs_inode(inode);
797
798 dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
799 memset(&req, 0, sizeof(struct ubifs_budget_req));
800
801 /*
802 * If this is truncation to a smaller size, and we do not truncate on a
803 * block boundary, budget for changing one data block, because the last
804 * block will be re-written.
805 */
806 if (new_size & (UBIFS_BLOCK_SIZE - 1))
807 req.dirtied_page = 1;
808
809 req.dirtied_ino = 1;
810 /* A funny way to budget for truncation node */
811 req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
812 err = ubifs_budget_space(c, &req);
813 if (err)
814 return err;
815
816 err = vmtruncate(inode, new_size);
817 if (err)
818 goto out_budg;
819
820 if (offset) {
821 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
822 struct page *page;
823
824 page = find_lock_page(inode->i_mapping, index);
825 if (page) {
826 if (PageDirty(page)) {
827 /*
828 * 'ubifs_jnl_truncate()' will try to truncate
829 * the last data node, but it contains
830 * out-of-date data because the page is dirty.
831 * Write the page now, so that
832 * 'ubifs_jnl_truncate()' will see an already
833 * truncated (and up to date) data node.
834 */
835 ubifs_assert(PagePrivate(page));
836
837 clear_page_dirty_for_io(page);
838 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
839 offset = new_size &
840 (PAGE_CACHE_SIZE - 1);
841 err = do_writepage(page, offset);
842 page_cache_release(page);
843 if (err)
844 goto out_budg;
845 /*
846 * We could now tell 'ubifs_jnl_truncate()' not
847 * to read the last block.
848 */
849 } else {
850 /*
851 * We could 'kmap()' the page and pass the data
852 * to 'ubifs_jnl_truncate()' to save it from
853 * having to read it.
854 */
855 unlock_page(page);
856 page_cache_release(page);
857 }
858 }
859 }
860
861 mutex_lock(&ui->ui_mutex);
862 ui->ui_size = inode->i_size;
863 /* Truncation changes inode [mc]time */
864 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
865 /* The other attributes may be changed at the same time as well */
866 do_attr_changes(inode, attr);
867
868 err = ubifs_jnl_truncate(c, inode, old_size, new_size);
869 mutex_unlock(&ui->ui_mutex);
870out_budg:
871 ubifs_release_budget(c, &req);
872 return err;
873}
874
875/**
876 * do_setattr - change inode attributes.
877 * @c: UBIFS file-system description object
878 * @inode: inode to change attributes for
879 * @attr: inode attribute changes description
880 *
881 * This function implements VFS '->setattr()' call for all cases except
882 * truncations to smaller size. Returns zero in case of success and a negative
883 * error code in case of failure.
884 */
885static int do_setattr(struct ubifs_info *c, struct inode *inode,
886 const struct iattr *attr)
887{
888 int err, release;
889 loff_t new_size = attr->ia_size;
890 struct ubifs_inode *ui = ubifs_inode(inode);
891 struct ubifs_budget_req req = { .dirtied_ino = 1,
892 .dirtied_ino_d = ui->data_len };
893
894 err = ubifs_budget_space(c, &req);
895 if (err)
896 return err;
897
898 if (attr->ia_valid & ATTR_SIZE) {
899 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
900 err = vmtruncate(inode, new_size);
901 if (err)
902 goto out;
903 }
904
905 mutex_lock(&ui->ui_mutex);
906 if (attr->ia_valid & ATTR_SIZE) {
907 /* Truncation changes inode [mc]time */
908 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
909 /* 'vmtruncate()' changed @i_size, update @ui_size */
910 ui->ui_size = inode->i_size;
911 }
912
913 do_attr_changes(inode, attr);
914
915 release = ui->dirty;
916 if (attr->ia_valid & ATTR_SIZE)
917 /*
918 * Inode length changed, so we have to make sure
919 * @I_DIRTY_DATASYNC is set.
920 */
921 __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
922 else
923 mark_inode_dirty_sync(inode);
924 mutex_unlock(&ui->ui_mutex);
925
926 if (release)
927 ubifs_release_budget(c, &req);
928 if (IS_SYNC(inode))
929 err = inode->i_sb->s_op->write_inode(inode, 1);
930 return err;
931
932out:
933 ubifs_release_budget(c, &req);
934 return err;
935}
936
937int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
938{
939 int err;
940 struct inode *inode = dentry->d_inode;
941 struct ubifs_info *c = inode->i_sb->s_fs_info;
942
943 dbg_gen("ino %lu, ia_valid %#x", inode->i_ino, attr->ia_valid);
944 err = inode_change_ok(inode, attr);
945 if (err)
946 return err;
947
948 err = dbg_check_synced_i_size(inode);
949 if (err)
950 return err;
951
952 if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
953 /* Truncation to a smaller size */
954 err = do_truncation(c, inode, attr);
955 else
956 err = do_setattr(c, inode, attr);
957
958 return err;
959}
960
961static void ubifs_invalidatepage(struct page *page, unsigned long offset)
962{
963 struct inode *inode = page->mapping->host;
964 struct ubifs_info *c = inode->i_sb->s_fs_info;
965
966 ubifs_assert(PagePrivate(page));
967 if (offset)
968 /* Partial page remains dirty */
969 return;
970
971 if (PageChecked(page))
972 release_new_page_budget(c);
973 else
974 release_existing_page_budget(c);
975
976 atomic_long_dec(&c->dirty_pg_cnt);
977 ClearPagePrivate(page);
978 ClearPageChecked(page);
979}
980
981static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
982{
983 struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
984
985 nd_set_link(nd, ui->data);
986 return NULL;
987}
988
989int ubifs_fsync(struct file *file, struct dentry *dentry, int datasync)
990{
991 struct inode *inode = dentry->d_inode;
992 struct ubifs_info *c = inode->i_sb->s_fs_info;
993 int err;
994
995 dbg_gen("syncing inode %lu", inode->i_ino);
996
997 /*
998 * VFS has already synchronized dirty pages for this inode. Synchronize
999 * the inode unless this is a 'datasync()' call.
1000 */
1001 if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1002 err = inode->i_sb->s_op->write_inode(inode, 1);
1003 if (err)
1004 return err;
1005 }
1006
1007 /*
1008 * Nodes related to this inode may still sit in a write-buffer. Flush
1009 * them.
1010 */
1011 err = ubifs_sync_wbufs_by_inode(c, inode);
1012 if (err)
1013 return err;
1014
1015 return 0;
1016}
1017
1018/**
1019 * mctime_update_needed - check if mtime or ctime update is needed.
1020 * @inode: the inode to do the check for
1021 * @now: current time
1022 *
1023 * This helper function checks if the inode mtime/ctime should be updated or
1024 * not. If current values of the time-stamps are within the UBIFS inode time
1025 * granularity, they are not updated. This is an optimization.
1026 */
1027static inline int mctime_update_needed(const struct inode *inode,
1028 const struct timespec *now)
1029{
1030 if (!timespec_equal(&inode->i_mtime, now) ||
1031 !timespec_equal(&inode->i_ctime, now))
1032 return 1;
1033 return 0;
1034}
1035
1036/**
1037 * update_ctime - update mtime and ctime of an inode.
1038 * @c: UBIFS file-system description object
1039 * @inode: inode to update
1040 *
1041 * This function updates mtime and ctime of the inode if it is not equivalent to
1042 * current time. Returns zero in case of success and a negative error code in
1043 * case of failure.
1044 */
1045static int update_mctime(struct ubifs_info *c, struct inode *inode)
1046{
1047 struct timespec now = ubifs_current_time(inode);
1048 struct ubifs_inode *ui = ubifs_inode(inode);
1049
1050 if (mctime_update_needed(inode, &now)) {
1051 int err, release;
1052 struct ubifs_budget_req req = { .dirtied_ino = 1,
1053 .dirtied_ino_d = ui->data_len };
1054
1055 err = ubifs_budget_space(c, &req);
1056 if (err)
1057 return err;
1058
1059 mutex_lock(&ui->ui_mutex);
1060 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1061 release = ui->dirty;
1062 mark_inode_dirty_sync(inode);
1063 mutex_unlock(&ui->ui_mutex);
1064 if (release)
1065 ubifs_release_budget(c, &req);
1066 }
1067
1068 return 0;
1069}
1070
1071static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1072 unsigned long nr_segs, loff_t pos)
1073{
1074 int err;
1075 ssize_t ret;
1076 struct inode *inode = iocb->ki_filp->f_mapping->host;
1077 struct ubifs_info *c = inode->i_sb->s_fs_info;
1078
1079 err = update_mctime(c, inode);
1080 if (err)
1081 return err;
1082
1083 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
1084 if (ret < 0)
1085 return ret;
1086
1087 if (ret > 0 && (IS_SYNC(inode) || iocb->ki_filp->f_flags & O_SYNC)) {
1088 err = ubifs_sync_wbufs_by_inode(c, inode);
1089 if (err)
1090 return err;
1091 }
1092
1093 return ret;
1094}
1095
1096static int ubifs_set_page_dirty(struct page *page)
1097{
1098 int ret;
1099
1100 ret = __set_page_dirty_nobuffers(page);
1101 /*
1102 * An attempt to dirty a page without budgeting for it - should not
1103 * happen.
1104 */
1105 ubifs_assert(ret == 0);
1106 return ret;
1107}
1108
1109static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1110{
1111 /*
1112 * An attempt to release a dirty page without budgeting for it - should
1113 * not happen.
1114 */
1115 if (PageWriteback(page))
1116 return 0;
1117 ubifs_assert(PagePrivate(page));
1118 ubifs_assert(0);
1119 ClearPagePrivate(page);
1120 ClearPageChecked(page);
1121 return 1;
1122}
1123
1124/*
1125 * mmap()d file has taken write protection fault and is being made
1126 * writable. UBIFS must ensure page is budgeted for.
1127 */
1128static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1129{
1130 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1131 struct ubifs_info *c = inode->i_sb->s_fs_info;
1132 struct timespec now = ubifs_current_time(inode);
1133 struct ubifs_budget_req req = { .new_page = 1 };
1134 int err, update_time;
1135
1136 dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1137 i_size_read(inode));
1138 ubifs_assert(!(inode->i_sb->s_flags & MS_RDONLY));
1139
1140 if (unlikely(c->ro_media))
1141 return -EROFS;
1142
1143 /*
1144 * We have not locked @page so far so we may budget for changing the
1145 * page. Note, we cannot do this after we locked the page, because
1146 * budgeting may cause write-back which would cause deadlock.
1147 *
1148 * At the moment we do not know whether the page is dirty or not, so we
1149 * assume that it is not and budget for a new page. We could look at
1150 * the @PG_private flag and figure this out, but we may race with write
1151 * back and the page state may change by the time we lock it, so this
1152 * would need additional care. We do not bother with this at the
1153 * moment, although it might be good idea to do. Instead, we allocate
1154 * budget for a new page and amend it later on if the page was in fact
1155 * dirty.
1156 *
1157 * The budgeting-related logic of this function is similar to what we
1158 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1159 * for more comments.
1160 */
1161 update_time = mctime_update_needed(inode, &now);
1162 if (update_time)
1163 /*
1164 * We have to change inode time stamp which requires extra
1165 * budgeting.
1166 */
1167 req.dirtied_ino = 1;
1168
1169 err = ubifs_budget_space(c, &req);
1170 if (unlikely(err)) {
1171 if (err == -ENOSPC)
1172 ubifs_warn("out of space for mmapped file "
1173 "(inode number %lu)", inode->i_ino);
1174 return err;
1175 }
1176
1177 lock_page(page);
1178 if (unlikely(page->mapping != inode->i_mapping ||
1179 page_offset(page) > i_size_read(inode))) {
1180 /* Page got truncated out from underneath us */
1181 err = -EINVAL;
1182 goto out_unlock;
1183 }
1184
1185 if (PagePrivate(page))
1186 release_new_page_budget(c);
1187 else {
1188 if (!PageChecked(page))
1189 ubifs_convert_page_budget(c);
1190 SetPagePrivate(page);
1191 atomic_long_inc(&c->dirty_pg_cnt);
1192 __set_page_dirty_nobuffers(page);
1193 }
1194
1195 if (update_time) {
1196 int release;
1197 struct ubifs_inode *ui = ubifs_inode(inode);
1198
1199 mutex_lock(&ui->ui_mutex);
1200 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1201 release = ui->dirty;
1202 mark_inode_dirty_sync(inode);
1203 mutex_unlock(&ui->ui_mutex);
1204 if (release)
1205 ubifs_release_dirty_inode_budget(c, ui);
1206 }
1207
1208 unlock_page(page);
1209 return 0;
1210
1211out_unlock:
1212 unlock_page(page);
1213 ubifs_release_budget(c, &req);
1214 return err;
1215}
1216
1217static struct vm_operations_struct ubifs_file_vm_ops = {
1218 .fault = filemap_fault,
1219 .page_mkwrite = ubifs_vm_page_mkwrite,
1220};
1221
1222static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1223{
1224 int err;
1225
1226 /* 'generic_file_mmap()' takes care of NOMMU case */
1227 err = generic_file_mmap(file, vma);
1228 if (err)
1229 return err;
1230 vma->vm_ops = &ubifs_file_vm_ops;
1231 return 0;
1232}
1233
1234struct address_space_operations ubifs_file_address_operations = {
1235 .readpage = ubifs_readpage,
1236 .writepage = ubifs_writepage,
1237 .write_begin = ubifs_write_begin,
1238 .write_end = ubifs_write_end,
1239 .invalidatepage = ubifs_invalidatepage,
1240 .set_page_dirty = ubifs_set_page_dirty,
1241 .releasepage = ubifs_releasepage,
1242};
1243
1244struct inode_operations ubifs_file_inode_operations = {
1245 .setattr = ubifs_setattr,
1246 .getattr = ubifs_getattr,
1247#ifdef CONFIG_UBIFS_FS_XATTR
1248 .setxattr = ubifs_setxattr,
1249 .getxattr = ubifs_getxattr,
1250 .listxattr = ubifs_listxattr,
1251 .removexattr = ubifs_removexattr,
1252#endif
1253};
1254
1255struct inode_operations ubifs_symlink_inode_operations = {
1256 .readlink = generic_readlink,
1257 .follow_link = ubifs_follow_link,
1258 .setattr = ubifs_setattr,
1259 .getattr = ubifs_getattr,
1260};
1261
1262struct file_operations ubifs_file_operations = {
1263 .llseek = generic_file_llseek,
1264 .read = do_sync_read,
1265 .write = do_sync_write,
1266 .aio_read = generic_file_aio_read,
1267 .aio_write = ubifs_aio_write,
1268 .mmap = ubifs_file_mmap,
1269 .fsync = ubifs_fsync,
1270 .unlocked_ioctl = ubifs_ioctl,
1271 .splice_read = generic_file_splice_read,
1272#ifdef CONFIG_COMPAT
1273 .compat_ioctl = ubifs_compat_ioctl,
1274#endif
1275};