Linux 2.6.35-rc2
[linux/fpc-iii.git] / fs / ubifs / file.c
blob12f445cee9f7ba90a6b4a33eeb1dc18be5df3a0d
1 /*
2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
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.
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.
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
19 * Authors: Artem Bityutskiy (Битюцкий Артём)
20 * Adrian Hunter
24 * This file implements VFS file and inode operations for regular files, device
25 * nodes and symlinks as well as address space operations.
27 * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
28 * the page is dirty and is used for optimization purposes - dirty pages are
29 * not budgeted so the flag shows that 'ubifs_write_end()' should not release
30 * the budget for this page. The @PG_checked flag is set if full budgeting is
31 * required for the page e.g., when it corresponds to a file hole or it is
32 * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because
33 * it is OK to fail in this function, and the budget is released in
34 * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry
35 * information about how the page was budgeted, to make it possible to release
36 * the budget properly.
38 * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
39 * implement. However, this is not true for 'ubifs_writepage()', which may be
40 * called with @i_mutex unlocked. For example, when pdflush is doing background
41 * write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex. At "normal"
42 * work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g. in the
43 * "sys_write -> alloc_pages -> direct reclaim path". So, in 'ubifs_writepage()'
44 * we are only guaranteed that the page is locked.
46 * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the
47 * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
48 * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
49 * set as well. However, UBIFS disables readahead.
52 #include "ubifs.h"
53 #include <linux/mount.h>
54 #include <linux/namei.h>
55 #include <linux/slab.h>
57 static int read_block(struct inode *inode, void *addr, unsigned int block,
58 struct ubifs_data_node *dn)
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;
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;
74 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
75 ubifs_inode(inode)->creat_sqnum);
76 len = le32_to_cpu(dn->size);
77 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
78 goto dump;
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;
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.
92 if (len < UBIFS_BLOCK_SIZE)
93 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
95 return 0;
97 dump:
98 ubifs_err("bad data node (block %u, inode %lu)",
99 block, inode->i_ino);
100 dbg_dump_node(c, dn);
101 return -EINVAL;
104 static int do_readpage(struct page *page)
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);
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));
118 addr = kmap(page);
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;
129 dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
130 if (!dn) {
131 err = -ENOMEM;
132 goto error;
135 i = 0;
136 while (1) {
137 int ret;
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 } else if (block + 1 == beyond) {
150 int dlen = le32_to_cpu(dn->size);
151 int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
153 if (ilen && ilen < dlen)
154 memset(addr + ilen, 0, dlen - ilen);
157 if (++i >= UBIFS_BLOCKS_PER_PAGE)
158 break;
159 block += 1;
160 addr += UBIFS_BLOCK_SIZE;
162 if (err) {
163 if (err == -ENOENT) {
164 /* Not found, so it must be a hole */
165 SetPageChecked(page);
166 dbg_gen("hole");
167 goto out_free;
169 ubifs_err("cannot read page %lu of inode %lu, error %d",
170 page->index, inode->i_ino, err);
171 goto error;
174 out_free:
175 kfree(dn);
176 out:
177 SetPageUptodate(page);
178 ClearPageError(page);
179 flush_dcache_page(page);
180 kunmap(page);
181 return 0;
183 error:
184 kfree(dn);
185 ClearPageUptodate(page);
186 SetPageError(page);
187 flush_dcache_page(page);
188 kunmap(page);
189 return err;
193 * release_new_page_budget - release budget of a new page.
194 * @c: UBIFS file-system description object
196 * This is a helper function which releases budget corresponding to the budget
197 * of one new page of data.
199 static void release_new_page_budget(struct ubifs_info *c)
201 struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
203 ubifs_release_budget(c, &req);
207 * release_existing_page_budget - release budget of an existing page.
208 * @c: UBIFS file-system description object
210 * This is a helper function which releases budget corresponding to the budget
211 * of changing one one page of data which already exists on the flash media.
213 static void release_existing_page_budget(struct ubifs_info *c)
215 struct ubifs_budget_req req = { .dd_growth = c->page_budget};
217 ubifs_release_budget(c, &req);
220 static int write_begin_slow(struct address_space *mapping,
221 loff_t pos, unsigned len, struct page **pagep,
222 unsigned flags)
224 struct inode *inode = mapping->host;
225 struct ubifs_info *c = inode->i_sb->s_fs_info;
226 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
227 struct ubifs_budget_req req = { .new_page = 1 };
228 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
229 struct page *page;
231 dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
232 inode->i_ino, pos, len, inode->i_size);
235 * At the slow path we have to budget before locking the page, because
236 * budgeting may force write-back, which would wait on locked pages and
237 * deadlock if we had the page locked. At this point we do not know
238 * anything about the page, so assume that this is a new page which is
239 * written to a hole. This corresponds to largest budget. Later the
240 * budget will be amended if this is not true.
242 if (appending)
243 /* We are appending data, budget for inode change */
244 req.dirtied_ino = 1;
246 err = ubifs_budget_space(c, &req);
247 if (unlikely(err))
248 return err;
250 page = grab_cache_page_write_begin(mapping, index, flags);
251 if (unlikely(!page)) {
252 ubifs_release_budget(c, &req);
253 return -ENOMEM;
256 if (!PageUptodate(page)) {
257 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
258 SetPageChecked(page);
259 else {
260 err = do_readpage(page);
261 if (err) {
262 unlock_page(page);
263 page_cache_release(page);
264 return err;
268 SetPageUptodate(page);
269 ClearPageError(page);
272 if (PagePrivate(page))
274 * The page is dirty, which means it was budgeted twice:
275 * o first time the budget was allocated by the task which
276 * made the page dirty and set the PG_private flag;
277 * o and then we budgeted for it for the second time at the
278 * very beginning of this function.
280 * So what we have to do is to release the page budget we
281 * allocated.
283 release_new_page_budget(c);
284 else if (!PageChecked(page))
286 * We are changing a page which already exists on the media.
287 * This means that changing the page does not make the amount
288 * of indexing information larger, and this part of the budget
289 * which we have already acquired may be released.
291 ubifs_convert_page_budget(c);
293 if (appending) {
294 struct ubifs_inode *ui = ubifs_inode(inode);
297 * 'ubifs_write_end()' is optimized from the fast-path part of
298 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
299 * if data is appended.
301 mutex_lock(&ui->ui_mutex);
302 if (ui->dirty)
304 * The inode is dirty already, so we may free the
305 * budget we allocated.
307 ubifs_release_dirty_inode_budget(c, ui);
310 *pagep = page;
311 return 0;
315 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
316 * @c: UBIFS file-system description object
317 * @page: page to allocate budget for
318 * @ui: UBIFS inode object the page belongs to
319 * @appending: non-zero if the page is appended
321 * This is a helper function for 'ubifs_write_begin()' which allocates budget
322 * for the operation. The budget is allocated differently depending on whether
323 * this is appending, whether the page is dirty or not, and so on. This
324 * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
325 * in case of success and %-ENOSPC in case of failure.
327 static int allocate_budget(struct ubifs_info *c, struct page *page,
328 struct ubifs_inode *ui, int appending)
330 struct ubifs_budget_req req = { .fast = 1 };
332 if (PagePrivate(page)) {
333 if (!appending)
335 * The page is dirty and we are not appending, which
336 * means no budget is needed at all.
338 return 0;
340 mutex_lock(&ui->ui_mutex);
341 if (ui->dirty)
343 * The page is dirty and we are appending, so the inode
344 * has to be marked as dirty. However, it is already
345 * dirty, so we do not need any budget. We may return,
346 * but @ui->ui_mutex hast to be left locked because we
347 * should prevent write-back from flushing the inode
348 * and freeing the budget. The lock will be released in
349 * 'ubifs_write_end()'.
351 return 0;
354 * The page is dirty, we are appending, the inode is clean, so
355 * we need to budget the inode change.
357 req.dirtied_ino = 1;
358 } else {
359 if (PageChecked(page))
361 * The page corresponds to a hole and does not
362 * exist on the media. So changing it makes
363 * make the amount of indexing information
364 * larger, and we have to budget for a new
365 * page.
367 req.new_page = 1;
368 else
370 * Not a hole, the change will not add any new
371 * indexing information, budget for page
372 * change.
374 req.dirtied_page = 1;
376 if (appending) {
377 mutex_lock(&ui->ui_mutex);
378 if (!ui->dirty)
380 * The inode is clean but we will have to mark
381 * it as dirty because we are appending. This
382 * needs a budget.
384 req.dirtied_ino = 1;
388 return ubifs_budget_space(c, &req);
392 * This function is called when a page of data is going to be written. Since
393 * the page of data will not necessarily go to the flash straight away, UBIFS
394 * has to reserve space on the media for it, which is done by means of
395 * budgeting.
397 * This is the hot-path of the file-system and we are trying to optimize it as
398 * much as possible. For this reasons it is split on 2 parts - slow and fast.
400 * There many budgeting cases:
401 * o a new page is appended - we have to budget for a new page and for
402 * changing the inode; however, if the inode is already dirty, there is
403 * no need to budget for it;
404 * o an existing clean page is changed - we have budget for it; if the page
405 * does not exist on the media (a hole), we have to budget for a new
406 * page; otherwise, we may budget for changing an existing page; the
407 * difference between these cases is that changing an existing page does
408 * not introduce anything new to the FS indexing information, so it does
409 * not grow, and smaller budget is acquired in this case;
410 * o an existing dirty page is changed - no need to budget at all, because
411 * the page budget has been acquired by earlier, when the page has been
412 * marked dirty.
414 * UBIFS budgeting sub-system may force write-back if it thinks there is no
415 * space to reserve. This imposes some locking restrictions and makes it
416 * impossible to take into account the above cases, and makes it impossible to
417 * optimize budgeting.
419 * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
420 * there is a plenty of flash space and the budget will be acquired quickly,
421 * without forcing write-back. The slow path does not make this assumption.
423 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
424 loff_t pos, unsigned len, unsigned flags,
425 struct page **pagep, void **fsdata)
427 struct inode *inode = mapping->host;
428 struct ubifs_info *c = inode->i_sb->s_fs_info;
429 struct ubifs_inode *ui = ubifs_inode(inode);
430 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
431 int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
432 int skipped_read = 0;
433 struct page *page;
435 ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
437 if (unlikely(c->ro_media))
438 return -EROFS;
440 /* Try out the fast-path part first */
441 page = grab_cache_page_write_begin(mapping, index, flags);
442 if (unlikely(!page))
443 return -ENOMEM;
445 if (!PageUptodate(page)) {
446 /* The page is not loaded from the flash */
447 if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
449 * We change whole page so no need to load it. But we
450 * have to set the @PG_checked flag to make the further
451 * code know that the page is new. This might be not
452 * true, but it is better to budget more than to read
453 * the page from the media.
455 SetPageChecked(page);
456 skipped_read = 1;
457 } else {
458 err = do_readpage(page);
459 if (err) {
460 unlock_page(page);
461 page_cache_release(page);
462 return err;
466 SetPageUptodate(page);
467 ClearPageError(page);
470 err = allocate_budget(c, page, ui, appending);
471 if (unlikely(err)) {
472 ubifs_assert(err == -ENOSPC);
474 * If we skipped reading the page because we were going to
475 * write all of it, then it is not up to date.
477 if (skipped_read) {
478 ClearPageChecked(page);
479 ClearPageUptodate(page);
482 * Budgeting failed which means it would have to force
483 * write-back but didn't, because we set the @fast flag in the
484 * request. Write-back cannot be done now, while we have the
485 * page locked, because it would deadlock. Unlock and free
486 * everything and fall-back to slow-path.
488 if (appending) {
489 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
490 mutex_unlock(&ui->ui_mutex);
492 unlock_page(page);
493 page_cache_release(page);
495 return write_begin_slow(mapping, pos, len, pagep, flags);
499 * Whee, we acquired budgeting quickly - without involving
500 * garbage-collection, committing or forcing write-back. We return
501 * with @ui->ui_mutex locked if we are appending pages, and unlocked
502 * otherwise. This is an optimization (slightly hacky though).
504 *pagep = page;
505 return 0;
510 * cancel_budget - cancel budget.
511 * @c: UBIFS file-system description object
512 * @page: page to cancel budget for
513 * @ui: UBIFS inode object the page belongs to
514 * @appending: non-zero if the page is appended
516 * This is a helper function for a page write operation. It unlocks the
517 * @ui->ui_mutex in case of appending.
519 static void cancel_budget(struct ubifs_info *c, struct page *page,
520 struct ubifs_inode *ui, int appending)
522 if (appending) {
523 if (!ui->dirty)
524 ubifs_release_dirty_inode_budget(c, ui);
525 mutex_unlock(&ui->ui_mutex);
527 if (!PagePrivate(page)) {
528 if (PageChecked(page))
529 release_new_page_budget(c);
530 else
531 release_existing_page_budget(c);
535 static int ubifs_write_end(struct file *file, struct address_space *mapping,
536 loff_t pos, unsigned len, unsigned copied,
537 struct page *page, void *fsdata)
539 struct inode *inode = mapping->host;
540 struct ubifs_inode *ui = ubifs_inode(inode);
541 struct ubifs_info *c = inode->i_sb->s_fs_info;
542 loff_t end_pos = pos + len;
543 int appending = !!(end_pos > inode->i_size);
545 dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
546 inode->i_ino, pos, page->index, len, copied, inode->i_size);
548 if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
550 * VFS copied less data to the page that it intended and
551 * declared in its '->write_begin()' call via the @len
552 * argument. If the page was not up-to-date, and @len was
553 * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
554 * not load it from the media (for optimization reasons). This
555 * means that part of the page contains garbage. So read the
556 * page now.
558 dbg_gen("copied %d instead of %d, read page and repeat",
559 copied, len);
560 cancel_budget(c, page, ui, appending);
563 * Return 0 to force VFS to repeat the whole operation, or the
564 * error code if 'do_readpage()' fails.
566 copied = do_readpage(page);
567 goto out;
570 if (!PagePrivate(page)) {
571 SetPagePrivate(page);
572 atomic_long_inc(&c->dirty_pg_cnt);
573 __set_page_dirty_nobuffers(page);
576 if (appending) {
577 i_size_write(inode, end_pos);
578 ui->ui_size = end_pos;
580 * Note, we do not set @I_DIRTY_PAGES (which means that the
581 * inode has dirty pages), this has been done in
582 * '__set_page_dirty_nobuffers()'.
584 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
585 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
586 mutex_unlock(&ui->ui_mutex);
589 out:
590 unlock_page(page);
591 page_cache_release(page);
592 return copied;
596 * populate_page - copy data nodes into a page for bulk-read.
597 * @c: UBIFS file-system description object
598 * @page: page
599 * @bu: bulk-read information
600 * @n: next zbranch slot
602 * This function returns %0 on success and a negative error code on failure.
604 static int populate_page(struct ubifs_info *c, struct page *page,
605 struct bu_info *bu, int *n)
607 int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
608 struct inode *inode = page->mapping->host;
609 loff_t i_size = i_size_read(inode);
610 unsigned int page_block;
611 void *addr, *zaddr;
612 pgoff_t end_index;
614 dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
615 inode->i_ino, page->index, i_size, page->flags);
617 addr = zaddr = kmap(page);
619 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
620 if (!i_size || page->index > end_index) {
621 hole = 1;
622 memset(addr, 0, PAGE_CACHE_SIZE);
623 goto out_hole;
626 page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
627 while (1) {
628 int err, len, out_len, dlen;
630 if (nn >= bu->cnt) {
631 hole = 1;
632 memset(addr, 0, UBIFS_BLOCK_SIZE);
633 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
634 struct ubifs_data_node *dn;
636 dn = bu->buf + (bu->zbranch[nn].offs - offs);
638 ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
639 ubifs_inode(inode)->creat_sqnum);
641 len = le32_to_cpu(dn->size);
642 if (len <= 0 || len > UBIFS_BLOCK_SIZE)
643 goto out_err;
645 dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
646 out_len = UBIFS_BLOCK_SIZE;
647 err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
648 le16_to_cpu(dn->compr_type));
649 if (err || len != out_len)
650 goto out_err;
652 if (len < UBIFS_BLOCK_SIZE)
653 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
655 nn += 1;
656 read = (i << UBIFS_BLOCK_SHIFT) + len;
657 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
658 nn += 1;
659 continue;
660 } else {
661 hole = 1;
662 memset(addr, 0, UBIFS_BLOCK_SIZE);
664 if (++i >= UBIFS_BLOCKS_PER_PAGE)
665 break;
666 addr += UBIFS_BLOCK_SIZE;
667 page_block += 1;
670 if (end_index == page->index) {
671 int len = i_size & (PAGE_CACHE_SIZE - 1);
673 if (len && len < read)
674 memset(zaddr + len, 0, read - len);
677 out_hole:
678 if (hole) {
679 SetPageChecked(page);
680 dbg_gen("hole");
683 SetPageUptodate(page);
684 ClearPageError(page);
685 flush_dcache_page(page);
686 kunmap(page);
687 *n = nn;
688 return 0;
690 out_err:
691 ClearPageUptodate(page);
692 SetPageError(page);
693 flush_dcache_page(page);
694 kunmap(page);
695 ubifs_err("bad data node (block %u, inode %lu)",
696 page_block, inode->i_ino);
697 return -EINVAL;
701 * ubifs_do_bulk_read - do bulk-read.
702 * @c: UBIFS file-system description object
703 * @bu: bulk-read information
704 * @page1: first page to read
706 * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
708 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
709 struct page *page1)
711 pgoff_t offset = page1->index, end_index;
712 struct address_space *mapping = page1->mapping;
713 struct inode *inode = mapping->host;
714 struct ubifs_inode *ui = ubifs_inode(inode);
715 int err, page_idx, page_cnt, ret = 0, n = 0;
716 int allocate = bu->buf ? 0 : 1;
717 loff_t isize;
719 err = ubifs_tnc_get_bu_keys(c, bu);
720 if (err)
721 goto out_warn;
723 if (bu->eof) {
724 /* Turn off bulk-read at the end of the file */
725 ui->read_in_a_row = 1;
726 ui->bulk_read = 0;
729 page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
730 if (!page_cnt) {
732 * This happens when there are multiple blocks per page and the
733 * blocks for the first page we are looking for, are not
734 * together. If all the pages were like this, bulk-read would
735 * reduce performance, so we turn it off for a while.
737 goto out_bu_off;
740 if (bu->cnt) {
741 if (allocate) {
743 * Allocate bulk-read buffer depending on how many data
744 * nodes we are going to read.
746 bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
747 bu->zbranch[bu->cnt - 1].len -
748 bu->zbranch[0].offs;
749 ubifs_assert(bu->buf_len > 0);
750 ubifs_assert(bu->buf_len <= c->leb_size);
751 bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
752 if (!bu->buf)
753 goto out_bu_off;
756 err = ubifs_tnc_bulk_read(c, bu);
757 if (err)
758 goto out_warn;
761 err = populate_page(c, page1, bu, &n);
762 if (err)
763 goto out_warn;
765 unlock_page(page1);
766 ret = 1;
768 isize = i_size_read(inode);
769 if (isize == 0)
770 goto out_free;
771 end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
773 for (page_idx = 1; page_idx < page_cnt; page_idx++) {
774 pgoff_t page_offset = offset + page_idx;
775 struct page *page;
777 if (page_offset > end_index)
778 break;
779 page = find_or_create_page(mapping, page_offset,
780 GFP_NOFS | __GFP_COLD);
781 if (!page)
782 break;
783 if (!PageUptodate(page))
784 err = populate_page(c, page, bu, &n);
785 unlock_page(page);
786 page_cache_release(page);
787 if (err)
788 break;
791 ui->last_page_read = offset + page_idx - 1;
793 out_free:
794 if (allocate)
795 kfree(bu->buf);
796 return ret;
798 out_warn:
799 ubifs_warn("ignoring error %d and skipping bulk-read", err);
800 goto out_free;
802 out_bu_off:
803 ui->read_in_a_row = ui->bulk_read = 0;
804 goto out_free;
808 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
809 * @page: page from which to start bulk-read.
811 * Some flash media are capable of reading sequentially at faster rates. UBIFS
812 * bulk-read facility is designed to take advantage of that, by reading in one
813 * go consecutive data nodes that are also located consecutively in the same
814 * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
816 static int ubifs_bulk_read(struct page *page)
818 struct inode *inode = page->mapping->host;
819 struct ubifs_info *c = inode->i_sb->s_fs_info;
820 struct ubifs_inode *ui = ubifs_inode(inode);
821 pgoff_t index = page->index, last_page_read = ui->last_page_read;
822 struct bu_info *bu;
823 int err = 0, allocated = 0;
825 ui->last_page_read = index;
826 if (!c->bulk_read)
827 return 0;
830 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
831 * so don't bother if we cannot lock the mutex.
833 if (!mutex_trylock(&ui->ui_mutex))
834 return 0;
836 if (index != last_page_read + 1) {
837 /* Turn off bulk-read if we stop reading sequentially */
838 ui->read_in_a_row = 1;
839 if (ui->bulk_read)
840 ui->bulk_read = 0;
841 goto out_unlock;
844 if (!ui->bulk_read) {
845 ui->read_in_a_row += 1;
846 if (ui->read_in_a_row < 3)
847 goto out_unlock;
848 /* Three reads in a row, so switch on bulk-read */
849 ui->bulk_read = 1;
853 * If possible, try to use pre-allocated bulk-read information, which
854 * is protected by @c->bu_mutex.
856 if (mutex_trylock(&c->bu_mutex))
857 bu = &c->bu;
858 else {
859 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
860 if (!bu)
861 goto out_unlock;
863 bu->buf = NULL;
864 allocated = 1;
867 bu->buf_len = c->max_bu_buf_len;
868 data_key_init(c, &bu->key, inode->i_ino,
869 page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
870 err = ubifs_do_bulk_read(c, bu, page);
872 if (!allocated)
873 mutex_unlock(&c->bu_mutex);
874 else
875 kfree(bu);
877 out_unlock:
878 mutex_unlock(&ui->ui_mutex);
879 return err;
882 static int ubifs_readpage(struct file *file, struct page *page)
884 if (ubifs_bulk_read(page))
885 return 0;
886 do_readpage(page);
887 unlock_page(page);
888 return 0;
891 static int do_writepage(struct page *page, int len)
893 int err = 0, i, blen;
894 unsigned int block;
895 void *addr;
896 union ubifs_key key;
897 struct inode *inode = page->mapping->host;
898 struct ubifs_info *c = inode->i_sb->s_fs_info;
900 #ifdef UBIFS_DEBUG
901 spin_lock(&ui->ui_lock);
902 ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
903 spin_unlock(&ui->ui_lock);
904 #endif
906 /* Update radix tree tags */
907 set_page_writeback(page);
909 addr = kmap(page);
910 block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
911 i = 0;
912 while (len) {
913 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
914 data_key_init(c, &key, inode->i_ino, block);
915 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
916 if (err)
917 break;
918 if (++i >= UBIFS_BLOCKS_PER_PAGE)
919 break;
920 block += 1;
921 addr += blen;
922 len -= blen;
924 if (err) {
925 SetPageError(page);
926 ubifs_err("cannot write page %lu of inode %lu, error %d",
927 page->index, inode->i_ino, err);
928 ubifs_ro_mode(c, err);
931 ubifs_assert(PagePrivate(page));
932 if (PageChecked(page))
933 release_new_page_budget(c);
934 else
935 release_existing_page_budget(c);
937 atomic_long_dec(&c->dirty_pg_cnt);
938 ClearPagePrivate(page);
939 ClearPageChecked(page);
941 kunmap(page);
942 unlock_page(page);
943 end_page_writeback(page);
944 return err;
948 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
949 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
950 * situation when a we have an inode with size 0, then a megabyte of data is
951 * appended to the inode, then write-back starts and flushes some amount of the
952 * dirty pages, the journal becomes full, commit happens and finishes, and then
953 * an unclean reboot happens. When the file system is mounted next time, the
954 * inode size would still be 0, but there would be many pages which are beyond
955 * the inode size, they would be indexed and consume flash space. Because the
956 * journal has been committed, the replay would not be able to detect this
957 * situation and correct the inode size. This means UBIFS would have to scan
958 * whole index and correct all inode sizes, which is long an unacceptable.
960 * To prevent situations like this, UBIFS writes pages back only if they are
961 * within the last synchronized inode size, i.e. the size which has been
962 * written to the flash media last time. Otherwise, UBIFS forces inode
963 * write-back, thus making sure the on-flash inode contains current inode size,
964 * and then keeps writing pages back.
966 * Some locking issues explanation. 'ubifs_writepage()' first is called with
967 * the page locked, and it locks @ui_mutex. However, write-back does take inode
968 * @i_mutex, which means other VFS operations may be run on this inode at the
969 * same time. And the problematic one is truncation to smaller size, from where
970 * we have to call 'simple_setsize()', which first changes @inode->i_size, then
971 * drops the truncated pages. And while dropping the pages, it takes the page
972 * lock. This means that 'do_truncation()' cannot call 'simple_setsize()' with
973 * @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'. This
974 * means that @inode->i_size is changed while @ui_mutex is unlocked.
976 * XXX: with the new truncate the above is not true anymore, the simple_setsize
977 * calls can be replaced with the individual components.
979 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
980 * inode size. How do we do this if @inode->i_size may became smaller while we
981 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
982 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
983 * internally and updates it under @ui_mutex.
985 * Q: why we do not worry that if we race with truncation, we may end up with a
986 * situation when the inode is truncated while we are in the middle of
987 * 'do_writepage()', so we do write beyond inode size?
988 * A: If we are in the middle of 'do_writepage()', truncation would be locked
989 * on the page lock and it would not write the truncated inode node to the
990 * journal before we have finished.
992 static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
994 struct inode *inode = page->mapping->host;
995 struct ubifs_inode *ui = ubifs_inode(inode);
996 loff_t i_size = i_size_read(inode), synced_i_size;
997 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
998 int err, len = i_size & (PAGE_CACHE_SIZE - 1);
999 void *kaddr;
1001 dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1002 inode->i_ino, page->index, page->flags);
1003 ubifs_assert(PagePrivate(page));
1005 /* Is the page fully outside @i_size? (truncate in progress) */
1006 if (page->index > end_index || (page->index == end_index && !len)) {
1007 err = 0;
1008 goto out_unlock;
1011 spin_lock(&ui->ui_lock);
1012 synced_i_size = ui->synced_i_size;
1013 spin_unlock(&ui->ui_lock);
1015 /* Is the page fully inside @i_size? */
1016 if (page->index < end_index) {
1017 if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
1018 err = inode->i_sb->s_op->write_inode(inode, NULL);
1019 if (err)
1020 goto out_unlock;
1022 * The inode has been written, but the write-buffer has
1023 * not been synchronized, so in case of an unclean
1024 * reboot we may end up with some pages beyond inode
1025 * size, but they would be in the journal (because
1026 * commit flushes write buffers) and recovery would deal
1027 * with this.
1030 return do_writepage(page, PAGE_CACHE_SIZE);
1034 * The page straddles @i_size. It must be zeroed out on each and every
1035 * writepage invocation because it may be mmapped. "A file is mapped
1036 * in multiples of the page size. For a file that is not a multiple of
1037 * the page size, the remaining memory is zeroed when mapped, and
1038 * writes to that region are not written out to the file."
1040 kaddr = kmap_atomic(page, KM_USER0);
1041 memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1042 flush_dcache_page(page);
1043 kunmap_atomic(kaddr, KM_USER0);
1045 if (i_size > synced_i_size) {
1046 err = inode->i_sb->s_op->write_inode(inode, NULL);
1047 if (err)
1048 goto out_unlock;
1051 return do_writepage(page, len);
1053 out_unlock:
1054 unlock_page(page);
1055 return err;
1059 * do_attr_changes - change inode attributes.
1060 * @inode: inode to change attributes for
1061 * @attr: describes attributes to change
1063 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1065 if (attr->ia_valid & ATTR_UID)
1066 inode->i_uid = attr->ia_uid;
1067 if (attr->ia_valid & ATTR_GID)
1068 inode->i_gid = attr->ia_gid;
1069 if (attr->ia_valid & ATTR_ATIME)
1070 inode->i_atime = timespec_trunc(attr->ia_atime,
1071 inode->i_sb->s_time_gran);
1072 if (attr->ia_valid & ATTR_MTIME)
1073 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1074 inode->i_sb->s_time_gran);
1075 if (attr->ia_valid & ATTR_CTIME)
1076 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1077 inode->i_sb->s_time_gran);
1078 if (attr->ia_valid & ATTR_MODE) {
1079 umode_t mode = attr->ia_mode;
1081 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1082 mode &= ~S_ISGID;
1083 inode->i_mode = mode;
1088 * do_truncation - truncate an inode.
1089 * @c: UBIFS file-system description object
1090 * @inode: inode to truncate
1091 * @attr: inode attribute changes description
1093 * This function implements VFS '->setattr()' call when the inode is truncated
1094 * to a smaller size. Returns zero in case of success and a negative error code
1095 * in case of failure.
1097 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1098 const struct iattr *attr)
1100 int err;
1101 struct ubifs_budget_req req;
1102 loff_t old_size = inode->i_size, new_size = attr->ia_size;
1103 int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1104 struct ubifs_inode *ui = ubifs_inode(inode);
1106 dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1107 memset(&req, 0, sizeof(struct ubifs_budget_req));
1110 * If this is truncation to a smaller size, and we do not truncate on a
1111 * block boundary, budget for changing one data block, because the last
1112 * block will be re-written.
1114 if (new_size & (UBIFS_BLOCK_SIZE - 1))
1115 req.dirtied_page = 1;
1117 req.dirtied_ino = 1;
1118 /* A funny way to budget for truncation node */
1119 req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1120 err = ubifs_budget_space(c, &req);
1121 if (err) {
1123 * Treat truncations to zero as deletion and always allow them,
1124 * just like we do for '->unlink()'.
1126 if (new_size || err != -ENOSPC)
1127 return err;
1128 budgeted = 0;
1131 err = simple_setsize(inode, new_size);
1132 if (err)
1133 goto out_budg;
1135 if (offset) {
1136 pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1137 struct page *page;
1139 page = find_lock_page(inode->i_mapping, index);
1140 if (page) {
1141 if (PageDirty(page)) {
1143 * 'ubifs_jnl_truncate()' will try to truncate
1144 * the last data node, but it contains
1145 * out-of-date data because the page is dirty.
1146 * Write the page now, so that
1147 * 'ubifs_jnl_truncate()' will see an already
1148 * truncated (and up to date) data node.
1150 ubifs_assert(PagePrivate(page));
1152 clear_page_dirty_for_io(page);
1153 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1154 offset = new_size &
1155 (PAGE_CACHE_SIZE - 1);
1156 err = do_writepage(page, offset);
1157 page_cache_release(page);
1158 if (err)
1159 goto out_budg;
1161 * We could now tell 'ubifs_jnl_truncate()' not
1162 * to read the last block.
1164 } else {
1166 * We could 'kmap()' the page and pass the data
1167 * to 'ubifs_jnl_truncate()' to save it from
1168 * having to read it.
1170 unlock_page(page);
1171 page_cache_release(page);
1176 mutex_lock(&ui->ui_mutex);
1177 ui->ui_size = inode->i_size;
1178 /* Truncation changes inode [mc]time */
1179 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1180 /* Other attributes may be changed at the same time as well */
1181 do_attr_changes(inode, attr);
1182 err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1183 mutex_unlock(&ui->ui_mutex);
1185 out_budg:
1186 if (budgeted)
1187 ubifs_release_budget(c, &req);
1188 else {
1189 c->nospace = c->nospace_rp = 0;
1190 smp_wmb();
1192 return err;
1196 * do_setattr - change inode attributes.
1197 * @c: UBIFS file-system description object
1198 * @inode: inode to change attributes for
1199 * @attr: inode attribute changes description
1201 * This function implements VFS '->setattr()' call for all cases except
1202 * truncations to smaller size. Returns zero in case of success and a negative
1203 * error code in case of failure.
1205 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1206 const struct iattr *attr)
1208 int err, release;
1209 loff_t new_size = attr->ia_size;
1210 struct ubifs_inode *ui = ubifs_inode(inode);
1211 struct ubifs_budget_req req = { .dirtied_ino = 1,
1212 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1214 err = ubifs_budget_space(c, &req);
1215 if (err)
1216 return err;
1218 if (attr->ia_valid & ATTR_SIZE) {
1219 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1220 err = simple_setsize(inode, new_size);
1221 if (err)
1222 goto out;
1225 mutex_lock(&ui->ui_mutex);
1226 if (attr->ia_valid & ATTR_SIZE) {
1227 /* Truncation changes inode [mc]time */
1228 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1229 /* 'simple_setsize()' changed @i_size, update @ui_size */
1230 ui->ui_size = inode->i_size;
1233 do_attr_changes(inode, attr);
1235 release = ui->dirty;
1236 if (attr->ia_valid & ATTR_SIZE)
1238 * Inode length changed, so we have to make sure
1239 * @I_DIRTY_DATASYNC is set.
1241 __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1242 else
1243 mark_inode_dirty_sync(inode);
1244 mutex_unlock(&ui->ui_mutex);
1246 if (release)
1247 ubifs_release_budget(c, &req);
1248 if (IS_SYNC(inode))
1249 err = inode->i_sb->s_op->write_inode(inode, NULL);
1250 return err;
1252 out:
1253 ubifs_release_budget(c, &req);
1254 return err;
1257 int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1259 int err;
1260 struct inode *inode = dentry->d_inode;
1261 struct ubifs_info *c = inode->i_sb->s_fs_info;
1263 dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1264 inode->i_ino, inode->i_mode, attr->ia_valid);
1265 err = inode_change_ok(inode, attr);
1266 if (err)
1267 return err;
1269 err = dbg_check_synced_i_size(inode);
1270 if (err)
1271 return err;
1273 if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1274 /* Truncation to a smaller size */
1275 err = do_truncation(c, inode, attr);
1276 else
1277 err = do_setattr(c, inode, attr);
1279 return err;
1282 static void ubifs_invalidatepage(struct page *page, unsigned long offset)
1284 struct inode *inode = page->mapping->host;
1285 struct ubifs_info *c = inode->i_sb->s_fs_info;
1287 ubifs_assert(PagePrivate(page));
1288 if (offset)
1289 /* Partial page remains dirty */
1290 return;
1292 if (PageChecked(page))
1293 release_new_page_budget(c);
1294 else
1295 release_existing_page_budget(c);
1297 atomic_long_dec(&c->dirty_pg_cnt);
1298 ClearPagePrivate(page);
1299 ClearPageChecked(page);
1302 static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
1304 struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
1306 nd_set_link(nd, ui->data);
1307 return NULL;
1310 int ubifs_fsync(struct file *file, int datasync)
1312 struct inode *inode = file->f_mapping->host;
1313 struct ubifs_info *c = inode->i_sb->s_fs_info;
1314 int err;
1316 dbg_gen("syncing inode %lu", inode->i_ino);
1319 * VFS has already synchronized dirty pages for this inode. Synchronize
1320 * the inode unless this is a 'datasync()' call.
1322 if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1323 err = inode->i_sb->s_op->write_inode(inode, NULL);
1324 if (err)
1325 return err;
1329 * Nodes related to this inode may still sit in a write-buffer. Flush
1330 * them.
1332 err = ubifs_sync_wbufs_by_inode(c, inode);
1333 if (err)
1334 return err;
1336 return 0;
1340 * mctime_update_needed - check if mtime or ctime update is needed.
1341 * @inode: the inode to do the check for
1342 * @now: current time
1344 * This helper function checks if the inode mtime/ctime should be updated or
1345 * not. If current values of the time-stamps are within the UBIFS inode time
1346 * granularity, they are not updated. This is an optimization.
1348 static inline int mctime_update_needed(const struct inode *inode,
1349 const struct timespec *now)
1351 if (!timespec_equal(&inode->i_mtime, now) ||
1352 !timespec_equal(&inode->i_ctime, now))
1353 return 1;
1354 return 0;
1358 * update_ctime - update mtime and ctime of an inode.
1359 * @c: UBIFS file-system description object
1360 * @inode: inode to update
1362 * This function updates mtime and ctime of the inode if it is not equivalent to
1363 * current time. Returns zero in case of success and a negative error code in
1364 * case of failure.
1366 static int update_mctime(struct ubifs_info *c, struct inode *inode)
1368 struct timespec now = ubifs_current_time(inode);
1369 struct ubifs_inode *ui = ubifs_inode(inode);
1371 if (mctime_update_needed(inode, &now)) {
1372 int err, release;
1373 struct ubifs_budget_req req = { .dirtied_ino = 1,
1374 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1376 err = ubifs_budget_space(c, &req);
1377 if (err)
1378 return err;
1380 mutex_lock(&ui->ui_mutex);
1381 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1382 release = ui->dirty;
1383 mark_inode_dirty_sync(inode);
1384 mutex_unlock(&ui->ui_mutex);
1385 if (release)
1386 ubifs_release_budget(c, &req);
1389 return 0;
1392 static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1393 unsigned long nr_segs, loff_t pos)
1395 int err;
1396 struct inode *inode = iocb->ki_filp->f_mapping->host;
1397 struct ubifs_info *c = inode->i_sb->s_fs_info;
1399 err = update_mctime(c, inode);
1400 if (err)
1401 return err;
1403 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1406 static int ubifs_set_page_dirty(struct page *page)
1408 int ret;
1410 ret = __set_page_dirty_nobuffers(page);
1412 * An attempt to dirty a page without budgeting for it - should not
1413 * happen.
1415 ubifs_assert(ret == 0);
1416 return ret;
1419 static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1422 * An attempt to release a dirty page without budgeting for it - should
1423 * not happen.
1425 if (PageWriteback(page))
1426 return 0;
1427 ubifs_assert(PagePrivate(page));
1428 ubifs_assert(0);
1429 ClearPagePrivate(page);
1430 ClearPageChecked(page);
1431 return 1;
1435 * mmap()d file has taken write protection fault and is being made
1436 * writable. UBIFS must ensure page is budgeted for.
1438 static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1440 struct page *page = vmf->page;
1441 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1442 struct ubifs_info *c = inode->i_sb->s_fs_info;
1443 struct timespec now = ubifs_current_time(inode);
1444 struct ubifs_budget_req req = { .new_page = 1 };
1445 int err, update_time;
1447 dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1448 i_size_read(inode));
1449 ubifs_assert(!(inode->i_sb->s_flags & MS_RDONLY));
1451 if (unlikely(c->ro_media))
1452 return VM_FAULT_SIGBUS; /* -EROFS */
1455 * We have not locked @page so far so we may budget for changing the
1456 * page. Note, we cannot do this after we locked the page, because
1457 * budgeting may cause write-back which would cause deadlock.
1459 * At the moment we do not know whether the page is dirty or not, so we
1460 * assume that it is not and budget for a new page. We could look at
1461 * the @PG_private flag and figure this out, but we may race with write
1462 * back and the page state may change by the time we lock it, so this
1463 * would need additional care. We do not bother with this at the
1464 * moment, although it might be good idea to do. Instead, we allocate
1465 * budget for a new page and amend it later on if the page was in fact
1466 * dirty.
1468 * The budgeting-related logic of this function is similar to what we
1469 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1470 * for more comments.
1472 update_time = mctime_update_needed(inode, &now);
1473 if (update_time)
1475 * We have to change inode time stamp which requires extra
1476 * budgeting.
1478 req.dirtied_ino = 1;
1480 err = ubifs_budget_space(c, &req);
1481 if (unlikely(err)) {
1482 if (err == -ENOSPC)
1483 ubifs_warn("out of space for mmapped file "
1484 "(inode number %lu)", inode->i_ino);
1485 return VM_FAULT_SIGBUS;
1488 lock_page(page);
1489 if (unlikely(page->mapping != inode->i_mapping ||
1490 page_offset(page) > i_size_read(inode))) {
1491 /* Page got truncated out from underneath us */
1492 err = -EINVAL;
1493 goto out_unlock;
1496 if (PagePrivate(page))
1497 release_new_page_budget(c);
1498 else {
1499 if (!PageChecked(page))
1500 ubifs_convert_page_budget(c);
1501 SetPagePrivate(page);
1502 atomic_long_inc(&c->dirty_pg_cnt);
1503 __set_page_dirty_nobuffers(page);
1506 if (update_time) {
1507 int release;
1508 struct ubifs_inode *ui = ubifs_inode(inode);
1510 mutex_lock(&ui->ui_mutex);
1511 inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1512 release = ui->dirty;
1513 mark_inode_dirty_sync(inode);
1514 mutex_unlock(&ui->ui_mutex);
1515 if (release)
1516 ubifs_release_dirty_inode_budget(c, ui);
1519 unlock_page(page);
1520 return 0;
1522 out_unlock:
1523 unlock_page(page);
1524 ubifs_release_budget(c, &req);
1525 if (err)
1526 err = VM_FAULT_SIGBUS;
1527 return err;
1530 static const struct vm_operations_struct ubifs_file_vm_ops = {
1531 .fault = filemap_fault,
1532 .page_mkwrite = ubifs_vm_page_mkwrite,
1535 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1537 int err;
1539 /* 'generic_file_mmap()' takes care of NOMMU case */
1540 err = generic_file_mmap(file, vma);
1541 if (err)
1542 return err;
1543 vma->vm_ops = &ubifs_file_vm_ops;
1544 return 0;
1547 const struct address_space_operations ubifs_file_address_operations = {
1548 .readpage = ubifs_readpage,
1549 .writepage = ubifs_writepage,
1550 .write_begin = ubifs_write_begin,
1551 .write_end = ubifs_write_end,
1552 .invalidatepage = ubifs_invalidatepage,
1553 .set_page_dirty = ubifs_set_page_dirty,
1554 .releasepage = ubifs_releasepage,
1557 const struct inode_operations ubifs_file_inode_operations = {
1558 .setattr = ubifs_setattr,
1559 .getattr = ubifs_getattr,
1560 #ifdef CONFIG_UBIFS_FS_XATTR
1561 .setxattr = ubifs_setxattr,
1562 .getxattr = ubifs_getxattr,
1563 .listxattr = ubifs_listxattr,
1564 .removexattr = ubifs_removexattr,
1565 #endif
1568 const struct inode_operations ubifs_symlink_inode_operations = {
1569 .readlink = generic_readlink,
1570 .follow_link = ubifs_follow_link,
1571 .setattr = ubifs_setattr,
1572 .getattr = ubifs_getattr,
1575 const struct file_operations ubifs_file_operations = {
1576 .llseek = generic_file_llseek,
1577 .read = do_sync_read,
1578 .write = do_sync_write,
1579 .aio_read = generic_file_aio_read,
1580 .aio_write = ubifs_aio_write,
1581 .mmap = ubifs_file_mmap,
1582 .fsync = ubifs_fsync,
1583 .unlocked_ioctl = ubifs_ioctl,
1584 .splice_read = generic_file_splice_read,
1585 .splice_write = generic_file_splice_write,
1586 #ifdef CONFIG_COMPAT
1587 .compat_ioctl = ubifs_compat_ioctl,
1588 #endif