~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ubifs/file.c

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * This file is part of UBIFS.
  4  *
  5  * Copyright (C) 2006-2008 Nokia Corporation.
  6  *
  7  * Authors: Artem Bityutskiy (Битюцкий Артём)
  8  *          Adrian Hunter
  9  */
 10 
 11 /*
 12  * This file implements VFS file and inode operations for regular files, device
 13  * nodes and symlinks as well as address space operations.
 14  *
 15  * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
 16  * the page is dirty and is used for optimization purposes - dirty pages are
 17  * not budgeted so the flag shows that 'ubifs_write_end()' should not release
 18  * the budget for this page. The @PG_checked flag is set if full budgeting is
 19  * required for the page e.g., when it corresponds to a file hole or it is
 20  * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because
 21  * it is OK to fail in this function, and the budget is released in
 22  * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry
 23  * information about how the page was budgeted, to make it possible to release
 24  * the budget properly.
 25  *
 26  * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
 27  * implement. However, this is not true for 'ubifs_writepage()', which may be
 28  * called with @i_mutex unlocked. For example, when flusher thread is doing
 29  * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
 30  * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
 31  * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
 32  * 'ubifs_writepage()' we are only guaranteed that the page is locked.
 33  *
 34  * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the
 35  * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
 36  * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
 37  * set as well. However, UBIFS disables readahead.
 38  */
 39 
 40 #include "ubifs.h"
 41 #include <linux/mount.h>
 42 #include <linux/slab.h>
 43 #include <linux/migrate.h>
 44 
 45 static int read_block(struct inode *inode, void *addr, unsigned int block,
 46                       struct ubifs_data_node *dn)
 47 {
 48         struct ubifs_info *c = inode->i_sb->s_fs_info;
 49         int err, len, out_len;
 50         union ubifs_key key;
 51         unsigned int dlen;
 52 
 53         data_key_init(c, &key, inode->i_ino, block);
 54         err = ubifs_tnc_lookup(c, &key, dn);
 55         if (err) {
 56                 if (err == -ENOENT)
 57                         /* Not found, so it must be a hole */
 58                         memset(addr, 0, UBIFS_BLOCK_SIZE);
 59                 return err;
 60         }
 61 
 62         ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
 63                      ubifs_inode(inode)->creat_sqnum);
 64         len = le32_to_cpu(dn->size);
 65         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
 66                 goto dump;
 67 
 68         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
 69 
 70         if (IS_ENCRYPTED(inode)) {
 71                 err = ubifs_decrypt(inode, dn, &dlen, block);
 72                 if (err)
 73                         goto dump;
 74         }
 75 
 76         out_len = UBIFS_BLOCK_SIZE;
 77         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
 78                                le16_to_cpu(dn->compr_type));
 79         if (err || len != out_len)
 80                 goto dump;
 81 
 82         /*
 83          * Data length can be less than a full block, even for blocks that are
 84          * not the last in the file (e.g., as a result of making a hole and
 85          * appending data). Ensure that the remainder is zeroed out.
 86          */
 87         if (len < UBIFS_BLOCK_SIZE)
 88                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
 89 
 90         return 0;
 91 
 92 dump:
 93         ubifs_err(c, "bad data node (block %u, inode %lu)",
 94                   block, inode->i_ino);
 95         ubifs_dump_node(c, dn, UBIFS_MAX_DATA_NODE_SZ);
 96         return -EINVAL;
 97 }
 98 
 99 static int do_readpage(struct page *page)
100 {
101         void *addr;
102         int err = 0, i;
103         unsigned int block, beyond;
104         struct ubifs_data_node *dn;
105         struct inode *inode = page->mapping->host;
106         struct ubifs_info *c = inode->i_sb->s_fs_info;
107         loff_t i_size = i_size_read(inode);
108 
109         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
110                 inode->i_ino, page->index, i_size, page->flags);
111         ubifs_assert(c, !PageChecked(page));
112         ubifs_assert(c, !PagePrivate(page));
113 
114         addr = kmap(page);
115 
116         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
117         beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
118         if (block >= beyond) {
119                 /* Reading beyond inode */
120                 SetPageChecked(page);
121                 memset(addr, 0, PAGE_SIZE);
122                 goto out;
123         }
124 
125         dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
126         if (!dn) {
127                 err = -ENOMEM;
128                 goto error;
129         }
130 
131         i = 0;
132         while (1) {
133                 int ret;
134 
135                 if (block >= beyond) {
136                         /* Reading beyond inode */
137                         err = -ENOENT;
138                         memset(addr, 0, UBIFS_BLOCK_SIZE);
139                 } else {
140                         ret = read_block(inode, addr, block, dn);
141                         if (ret) {
142                                 err = ret;
143                                 if (err != -ENOENT)
144                                         break;
145                         } else if (block + 1 == beyond) {
146                                 int dlen = le32_to_cpu(dn->size);
147                                 int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
148 
149                                 if (ilen && ilen < dlen)
150                                         memset(addr + ilen, 0, dlen - ilen);
151                         }
152                 }
153                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
154                         break;
155                 block += 1;
156                 addr += UBIFS_BLOCK_SIZE;
157         }
158         if (err) {
159                 struct ubifs_info *c = inode->i_sb->s_fs_info;
160                 if (err == -ENOENT) {
161                         /* Not found, so it must be a hole */
162                         SetPageChecked(page);
163                         dbg_gen("hole");
164                         goto out_free;
165                 }
166                 ubifs_err(c, "cannot read page %lu of inode %lu, error %d",
167                           page->index, inode->i_ino, err);
168                 goto error;
169         }
170 
171 out_free:
172         kfree(dn);
173 out:
174         SetPageUptodate(page);
175         ClearPageError(page);
176         flush_dcache_page(page);
177         kunmap(page);
178         return 0;
179 
180 error:
181         kfree(dn);
182         ClearPageUptodate(page);
183         SetPageError(page);
184         flush_dcache_page(page);
185         kunmap(page);
186         return err;
187 }
188 
189 /**
190  * release_new_page_budget - release budget of a new page.
191  * @c: UBIFS file-system description object
192  *
193  * This is a helper function which releases budget corresponding to the budget
194  * of one new page of data.
195  */
196 static void release_new_page_budget(struct ubifs_info *c)
197 {
198         struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
199 
200         ubifs_release_budget(c, &req);
201 }
202 
203 /**
204  * release_existing_page_budget - release budget of an existing page.
205  * @c: UBIFS file-system description object
206  *
207  * This is a helper function which releases budget corresponding to the budget
208  * of changing one page of data which already exists on the flash media.
209  */
210 static void release_existing_page_budget(struct ubifs_info *c)
211 {
212         struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
213 
214         ubifs_release_budget(c, &req);
215 }
216 
217 static int write_begin_slow(struct address_space *mapping,
218                             loff_t pos, unsigned len, struct page **pagep,
219                             unsigned flags)
220 {
221         struct inode *inode = mapping->host;
222         struct ubifs_info *c = inode->i_sb->s_fs_info;
223         pgoff_t index = pos >> PAGE_SHIFT;
224         struct ubifs_budget_req req = { .new_page = 1 };
225         int err, appending = !!(pos + len > inode->i_size);
226         struct page *page;
227 
228         dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
229                 inode->i_ino, pos, len, inode->i_size);
230 
231         /*
232          * At the slow path we have to budget before locking the page, because
233          * budgeting may force write-back, which would wait on locked pages and
234          * deadlock if we had the page locked. At this point we do not know
235          * anything about the page, so assume that this is a new page which is
236          * written to a hole. This corresponds to largest budget. Later the
237          * budget will be amended if this is not true.
238          */
239         if (appending)
240                 /* We are appending data, budget for inode change */
241                 req.dirtied_ino = 1;
242 
243         err = ubifs_budget_space(c, &req);
244         if (unlikely(err))
245                 return err;
246 
247         page = grab_cache_page_write_begin(mapping, index, flags);
248         if (unlikely(!page)) {
249                 ubifs_release_budget(c, &req);
250                 return -ENOMEM;
251         }
252 
253         if (!PageUptodate(page)) {
254                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE)
255                         SetPageChecked(page);
256                 else {
257                         err = do_readpage(page);
258                         if (err) {
259                                 unlock_page(page);
260                                 put_page(page);
261                                 ubifs_release_budget(c, &req);
262                                 return err;
263                         }
264                 }
265 
266                 SetPageUptodate(page);
267                 ClearPageError(page);
268         }
269 
270         if (PagePrivate(page))
271                 /*
272                  * The page is dirty, which means it was budgeted twice:
273                  *   o first time the budget was allocated by the task which
274                  *     made the page dirty and set the PG_private flag;
275                  *   o and then we budgeted for it for the second time at the
276                  *     very beginning of this function.
277                  *
278                  * So what we have to do is to release the page budget we
279                  * allocated.
280                  */
281                 release_new_page_budget(c);
282         else if (!PageChecked(page))
283                 /*
284                  * We are changing a page which already exists on the media.
285                  * This means that changing the page does not make the amount
286                  * of indexing information larger, and this part of the budget
287                  * which we have already acquired may be released.
288                  */
289                 ubifs_convert_page_budget(c);
290 
291         if (appending) {
292                 struct ubifs_inode *ui = ubifs_inode(inode);
293 
294                 /*
295                  * 'ubifs_write_end()' is optimized from the fast-path part of
296                  * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
297                  * if data is appended.
298                  */
299                 mutex_lock(&ui->ui_mutex);
300                 if (ui->dirty)
301                         /*
302                          * The inode is dirty already, so we may free the
303                          * budget we allocated.
304                          */
305                         ubifs_release_dirty_inode_budget(c, ui);
306         }
307 
308         *pagep = page;
309         return 0;
310 }
311 
312 /**
313  * allocate_budget - allocate budget for 'ubifs_write_begin()'.
314  * @c: UBIFS file-system description object
315  * @page: page to allocate budget for
316  * @ui: UBIFS inode object the page belongs to
317  * @appending: non-zero if the page is appended
318  *
319  * This is a helper function for 'ubifs_write_begin()' which allocates budget
320  * for the operation. The budget is allocated differently depending on whether
321  * this is appending, whether the page is dirty or not, and so on. This
322  * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
323  * in case of success and %-ENOSPC in case of failure.
324  */
325 static int allocate_budget(struct ubifs_info *c, struct page *page,
326                            struct ubifs_inode *ui, int appending)
327 {
328         struct ubifs_budget_req req = { .fast = 1 };
329 
330         if (PagePrivate(page)) {
331                 if (!appending)
332                         /*
333                          * The page is dirty and we are not appending, which
334                          * means no budget is needed at all.
335                          */
336                         return 0;
337 
338                 mutex_lock(&ui->ui_mutex);
339                 if (ui->dirty)
340                         /*
341                          * The page is dirty and we are appending, so the inode
342                          * has to be marked as dirty. However, it is already
343                          * dirty, so we do not need any budget. We may return,
344                          * but @ui->ui_mutex hast to be left locked because we
345                          * should prevent write-back from flushing the inode
346                          * and freeing the budget. The lock will be released in
347                          * 'ubifs_write_end()'.
348                          */
349                         return 0;
350 
351                 /*
352                  * The page is dirty, we are appending, the inode is clean, so
353                  * we need to budget the inode change.
354                  */
355                 req.dirtied_ino = 1;
356         } else {
357                 if (PageChecked(page))
358                         /*
359                          * The page corresponds to a hole and does not
360                          * exist on the media. So changing it makes
361                          * make the amount of indexing information
362                          * larger, and we have to budget for a new
363                          * page.
364                          */
365                         req.new_page = 1;
366                 else
367                         /*
368                          * Not a hole, the change will not add any new
369                          * indexing information, budget for page
370                          * change.
371                          */
372                         req.dirtied_page = 1;
373 
374                 if (appending) {
375                         mutex_lock(&ui->ui_mutex);
376                         if (!ui->dirty)
377                                 /*
378                                  * The inode is clean but we will have to mark
379                                  * it as dirty because we are appending. This
380                                  * needs a budget.
381                                  */
382                                 req.dirtied_ino = 1;
383                 }
384         }
385 
386         return ubifs_budget_space(c, &req);
387 }
388 
389 /*
390  * This function is called when a page of data is going to be written. Since
391  * the page of data will not necessarily go to the flash straight away, UBIFS
392  * has to reserve space on the media for it, which is done by means of
393  * budgeting.
394  *
395  * This is the hot-path of the file-system and we are trying to optimize it as
396  * much as possible. For this reasons it is split on 2 parts - slow and fast.
397  *
398  * There many budgeting cases:
399  *     o a new page is appended - we have to budget for a new page and for
400  *       changing the inode; however, if the inode is already dirty, there is
401  *       no need to budget for it;
402  *     o an existing clean page is changed - we have budget for it; if the page
403  *       does not exist on the media (a hole), we have to budget for a new
404  *       page; otherwise, we may budget for changing an existing page; the
405  *       difference between these cases is that changing an existing page does
406  *       not introduce anything new to the FS indexing information, so it does
407  *       not grow, and smaller budget is acquired in this case;
408  *     o an existing dirty page is changed - no need to budget at all, because
409  *       the page budget has been acquired by earlier, when the page has been
410  *       marked dirty.
411  *
412  * UBIFS budgeting sub-system may force write-back if it thinks there is no
413  * space to reserve. This imposes some locking restrictions and makes it
414  * impossible to take into account the above cases, and makes it impossible to
415  * optimize budgeting.
416  *
417  * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
418  * there is a plenty of flash space and the budget will be acquired quickly,
419  * without forcing write-back. The slow path does not make this assumption.
420  */
421 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
422                              loff_t pos, unsigned len, unsigned flags,
423                              struct page **pagep, void **fsdata)
424 {
425         struct inode *inode = mapping->host;
426         struct ubifs_info *c = inode->i_sb->s_fs_info;
427         struct ubifs_inode *ui = ubifs_inode(inode);
428         pgoff_t index = pos >> PAGE_SHIFT;
429         int err, appending = !!(pos + len > inode->i_size);
430         int skipped_read = 0;
431         struct page *page;
432 
433         ubifs_assert(c, ubifs_inode(inode)->ui_size == inode->i_size);
434         ubifs_assert(c, !c->ro_media && !c->ro_mount);
435 
436         if (unlikely(c->ro_error))
437                 return -EROFS;
438 
439         /* Try out the fast-path part first */
440         page = grab_cache_page_write_begin(mapping, index, flags);
441         if (unlikely(!page))
442                 return -ENOMEM;
443 
444         if (!PageUptodate(page)) {
445                 /* The page is not loaded from the flash */
446                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE) {
447                         /*
448                          * We change whole page so no need to load it. But we
449                          * do not know whether this page exists on the media or
450                          * not, so we assume the latter because it requires
451                          * larger budget. The assumption is that it is better
452                          * to budget a bit more than to read the page from the
453                          * media. Thus, we are setting the @PG_checked flag
454                          * here.
455                          */
456                         SetPageChecked(page);
457                         skipped_read = 1;
458                 } else {
459                         err = do_readpage(page);
460                         if (err) {
461                                 unlock_page(page);
462                                 put_page(page);
463                                 return err;
464                         }
465                 }
466 
467                 SetPageUptodate(page);
468                 ClearPageError(page);
469         }
470 
471         err = allocate_budget(c, page, ui, appending);
472         if (unlikely(err)) {
473                 ubifs_assert(c, err == -ENOSPC);
474                 /*
475                  * If we skipped reading the page because we were going to
476                  * write all of it, then it is not up to date.
477                  */
478                 if (skipped_read) {
479                         ClearPageChecked(page);
480                         ClearPageUptodate(page);
481                 }
482                 /*
483                  * Budgeting failed which means it would have to force
484                  * write-back but didn't, because we set the @fast flag in the
485                  * request. Write-back cannot be done now, while we have the
486                  * page locked, because it would deadlock. Unlock and free
487                  * everything and fall-back to slow-path.
488                  */
489                 if (appending) {
490                         ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
491                         mutex_unlock(&ui->ui_mutex);
492                 }
493                 unlock_page(page);
494                 put_page(page);
495 
496                 return write_begin_slow(mapping, pos, len, pagep, flags);
497         }
498 
499         /*
500          * Whee, we acquired budgeting quickly - without involving
501          * garbage-collection, committing or forcing write-back. We return
502          * with @ui->ui_mutex locked if we are appending pages, and unlocked
503          * otherwise. This is an optimization (slightly hacky though).
504          */
505         *pagep = page;
506         return 0;
507 
508 }
509 
510 /**
511  * cancel_budget - cancel budget.
512  * @c: UBIFS file-system description object
513  * @page: page to cancel budget for
514  * @ui: UBIFS inode object the page belongs to
515  * @appending: non-zero if the page is appended
516  *
517  * This is a helper function for a page write operation. It unlocks the
518  * @ui->ui_mutex in case of appending.
519  */
520 static void cancel_budget(struct ubifs_info *c, struct page *page,
521                           struct ubifs_inode *ui, int appending)
522 {
523         if (appending) {
524                 if (!ui->dirty)
525                         ubifs_release_dirty_inode_budget(c, ui);
526                 mutex_unlock(&ui->ui_mutex);
527         }
528         if (!PagePrivate(page)) {
529                 if (PageChecked(page))
530                         release_new_page_budget(c);
531                 else
532                         release_existing_page_budget(c);
533         }
534 }
535 
536 static int ubifs_write_end(struct file *file, struct address_space *mapping,
537                            loff_t pos, unsigned len, unsigned copied,
538                            struct page *page, void *fsdata)
539 {
540         struct inode *inode = mapping->host;
541         struct ubifs_inode *ui = ubifs_inode(inode);
542         struct ubifs_info *c = inode->i_sb->s_fs_info;
543         loff_t end_pos = pos + len;
544         int appending = !!(end_pos > inode->i_size);
545 
546         dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
547                 inode->i_ino, pos, page->index, len, copied, inode->i_size);
548 
549         if (unlikely(copied < len && len == PAGE_SIZE)) {
550                 /*
551                  * VFS copied less data to the page that it intended and
552                  * declared in its '->write_begin()' call via the @len
553                  * argument. If the page was not up-to-date, and @len was
554                  * @PAGE_SIZE, the 'ubifs_write_begin()' function did
555                  * not load it from the media (for optimization reasons). This
556                  * means that part of the page contains garbage. So read the
557                  * page now.
558                  */
559                 dbg_gen("copied %d instead of %d, read page and repeat",
560                         copied, len);
561                 cancel_budget(c, page, ui, appending);
562                 ClearPageChecked(page);
563 
564                 /*
565                  * Return 0 to force VFS to repeat the whole operation, or the
566                  * error code if 'do_readpage()' fails.
567                  */
568                 copied = do_readpage(page);
569                 goto out;
570         }
571 
572         if (!PagePrivate(page)) {
573                 SetPagePrivate(page);
574                 atomic_long_inc(&c->dirty_pg_cnt);
575                 __set_page_dirty_nobuffers(page);
576         }
577 
578         if (appending) {
579                 i_size_write(inode, end_pos);
580                 ui->ui_size = end_pos;
581                 /*
582                  * Note, we do not set @I_DIRTY_PAGES (which means that the
583                  * inode has dirty pages), this has been done in
584                  * '__set_page_dirty_nobuffers()'.
585                  */
586                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
587                 ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
588                 mutex_unlock(&ui->ui_mutex);
589         }
590 
591 out:
592         unlock_page(page);
593         put_page(page);
594         return copied;
595 }
596 
597 /**
598  * populate_page - copy data nodes into a page for bulk-read.
599  * @c: UBIFS file-system description object
600  * @page: page
601  * @bu: bulk-read information
602  * @n: next zbranch slot
603  *
604  * This function returns %0 on success and a negative error code on failure.
605  */
606 static int populate_page(struct ubifs_info *c, struct page *page,
607                          struct bu_info *bu, int *n)
608 {
609         int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
610         struct inode *inode = page->mapping->host;
611         loff_t i_size = i_size_read(inode);
612         unsigned int page_block;
613         void *addr, *zaddr;
614         pgoff_t end_index;
615 
616         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
617                 inode->i_ino, page->index, i_size, page->flags);
618 
619         addr = zaddr = kmap(page);
620 
621         end_index = (i_size - 1) >> PAGE_SHIFT;
622         if (!i_size || page->index > end_index) {
623                 hole = 1;
624                 memset(addr, 0, PAGE_SIZE);
625                 goto out_hole;
626         }
627 
628         page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
629         while (1) {
630                 int err, len, out_len, dlen;
631 
632                 if (nn >= bu->cnt) {
633                         hole = 1;
634                         memset(addr, 0, UBIFS_BLOCK_SIZE);
635                 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
636                         struct ubifs_data_node *dn;
637 
638                         dn = bu->buf + (bu->zbranch[nn].offs - offs);
639 
640                         ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
641                                      ubifs_inode(inode)->creat_sqnum);
642 
643                         len = le32_to_cpu(dn->size);
644                         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
645                                 goto out_err;
646 
647                         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
648                         out_len = UBIFS_BLOCK_SIZE;
649 
650                         if (IS_ENCRYPTED(inode)) {
651                                 err = ubifs_decrypt(inode, dn, &dlen, page_block);
652                                 if (err)
653                                         goto out_err;
654                         }
655 
656                         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
657                                                le16_to_cpu(dn->compr_type));
658                         if (err || len != out_len)
659                                 goto out_err;
660 
661                         if (len < UBIFS_BLOCK_SIZE)
662                                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
663 
664                         nn += 1;
665                         read = (i << UBIFS_BLOCK_SHIFT) + len;
666                 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
667                         nn += 1;
668                         continue;
669                 } else {
670                         hole = 1;
671                         memset(addr, 0, UBIFS_BLOCK_SIZE);
672                 }
673                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
674                         break;
675                 addr += UBIFS_BLOCK_SIZE;
676                 page_block += 1;
677         }
678 
679         if (end_index == page->index) {
680                 int len = i_size & (PAGE_SIZE - 1);
681 
682                 if (len && len < read)
683                         memset(zaddr + len, 0, read - len);
684         }
685 
686 out_hole:
687         if (hole) {
688                 SetPageChecked(page);
689                 dbg_gen("hole");
690         }
691 
692         SetPageUptodate(page);
693         ClearPageError(page);
694         flush_dcache_page(page);
695         kunmap(page);
696         *n = nn;
697         return 0;
698 
699 out_err:
700         ClearPageUptodate(page);
701         SetPageError(page);
702         flush_dcache_page(page);
703         kunmap(page);
704         ubifs_err(c, "bad data node (block %u, inode %lu)",
705                   page_block, inode->i_ino);
706         return -EINVAL;
707 }
708 
709 /**
710  * ubifs_do_bulk_read - do bulk-read.
711  * @c: UBIFS file-system description object
712  * @bu: bulk-read information
713  * @page1: first page to read
714  *
715  * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
716  */
717 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
718                               struct page *page1)
719 {
720         pgoff_t offset = page1->index, end_index;
721         struct address_space *mapping = page1->mapping;
722         struct inode *inode = mapping->host;
723         struct ubifs_inode *ui = ubifs_inode(inode);
724         int err, page_idx, page_cnt, ret = 0, n = 0;
725         int allocate = bu->buf ? 0 : 1;
726         loff_t isize;
727         gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS;
728 
729         err = ubifs_tnc_get_bu_keys(c, bu);
730         if (err)
731                 goto out_warn;
732 
733         if (bu->eof) {
734                 /* Turn off bulk-read at the end of the file */
735                 ui->read_in_a_row = 1;
736                 ui->bulk_read = 0;
737         }
738 
739         page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
740         if (!page_cnt) {
741                 /*
742                  * This happens when there are multiple blocks per page and the
743                  * blocks for the first page we are looking for, are not
744                  * together. If all the pages were like this, bulk-read would
745                  * reduce performance, so we turn it off for a while.
746                  */
747                 goto out_bu_off;
748         }
749 
750         if (bu->cnt) {
751                 if (allocate) {
752                         /*
753                          * Allocate bulk-read buffer depending on how many data
754                          * nodes we are going to read.
755                          */
756                         bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
757                                       bu->zbranch[bu->cnt - 1].len -
758                                       bu->zbranch[0].offs;
759                         ubifs_assert(c, bu->buf_len > 0);
760                         ubifs_assert(c, bu->buf_len <= c->leb_size);
761                         bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
762                         if (!bu->buf)
763                                 goto out_bu_off;
764                 }
765 
766                 err = ubifs_tnc_bulk_read(c, bu);
767                 if (err)
768                         goto out_warn;
769         }
770 
771         err = populate_page(c, page1, bu, &n);
772         if (err)
773                 goto out_warn;
774 
775         unlock_page(page1);
776         ret = 1;
777 
778         isize = i_size_read(inode);
779         if (isize == 0)
780                 goto out_free;
781         end_index = ((isize - 1) >> PAGE_SHIFT);
782 
783         for (page_idx = 1; page_idx < page_cnt; page_idx++) {
784                 pgoff_t page_offset = offset + page_idx;
785                 struct page *page;
786 
787                 if (page_offset > end_index)
788                         break;
789                 page = pagecache_get_page(mapping, page_offset,
790                                  FGP_LOCK|FGP_ACCESSED|FGP_CREAT|FGP_NOWAIT,
791                                  ra_gfp_mask);
792                 if (!page)
793                         break;
794                 if (!PageUptodate(page))
795                         err = populate_page(c, page, bu, &n);
796                 unlock_page(page);
797                 put_page(page);
798                 if (err)
799                         break;
800         }
801 
802         ui->last_page_read = offset + page_idx - 1;
803 
804 out_free:
805         if (allocate)
806                 kfree(bu->buf);
807         return ret;
808 
809 out_warn:
810         ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
811         goto out_free;
812 
813 out_bu_off:
814         ui->read_in_a_row = ui->bulk_read = 0;
815         goto out_free;
816 }
817 
818 /**
819  * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
820  * @page: page from which to start bulk-read.
821  *
822  * Some flash media are capable of reading sequentially at faster rates. UBIFS
823  * bulk-read facility is designed to take advantage of that, by reading in one
824  * go consecutive data nodes that are also located consecutively in the same
825  * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
826  */
827 static int ubifs_bulk_read(struct page *page)
828 {
829         struct inode *inode = page->mapping->host;
830         struct ubifs_info *c = inode->i_sb->s_fs_info;
831         struct ubifs_inode *ui = ubifs_inode(inode);
832         pgoff_t index = page->index, last_page_read = ui->last_page_read;
833         struct bu_info *bu;
834         int err = 0, allocated = 0;
835 
836         ui->last_page_read = index;
837         if (!c->bulk_read)
838                 return 0;
839 
840         /*
841          * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
842          * so don't bother if we cannot lock the mutex.
843          */
844         if (!mutex_trylock(&ui->ui_mutex))
845                 return 0;
846 
847         if (index != last_page_read + 1) {
848                 /* Turn off bulk-read if we stop reading sequentially */
849                 ui->read_in_a_row = 1;
850                 if (ui->bulk_read)
851                         ui->bulk_read = 0;
852                 goto out_unlock;
853         }
854 
855         if (!ui->bulk_read) {
856                 ui->read_in_a_row += 1;
857                 if (ui->read_in_a_row < 3)
858                         goto out_unlock;
859                 /* Three reads in a row, so switch on bulk-read */
860                 ui->bulk_read = 1;
861         }
862 
863         /*
864          * If possible, try to use pre-allocated bulk-read information, which
865          * is protected by @c->bu_mutex.
866          */
867         if (mutex_trylock(&c->bu_mutex))
868                 bu = &c->bu;
869         else {
870                 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
871                 if (!bu)
872                         goto out_unlock;
873 
874                 bu->buf = NULL;
875                 allocated = 1;
876         }
877 
878         bu->buf_len = c->max_bu_buf_len;
879         data_key_init(c, &bu->key, inode->i_ino,
880                       page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
881         err = ubifs_do_bulk_read(c, bu, page);
882 
883         if (!allocated)
884                 mutex_unlock(&c->bu_mutex);
885         else
886                 kfree(bu);
887 
888 out_unlock:
889         mutex_unlock(&ui->ui_mutex);
890         return err;
891 }
892 
893 static int ubifs_readpage(struct file *file, struct page *page)
894 {
895         if (ubifs_bulk_read(page))
896                 return 0;
897         do_readpage(page);
898         unlock_page(page);
899         return 0;
900 }
901 
902 static int do_writepage(struct page *page, int len)
903 {
904         int err = 0, i, blen;
905         unsigned int block;
906         void *addr;
907         union ubifs_key key;
908         struct inode *inode = page->mapping->host;
909         struct ubifs_info *c = inode->i_sb->s_fs_info;
910 
911 #ifdef UBIFS_DEBUG
912         struct ubifs_inode *ui = ubifs_inode(inode);
913         spin_lock(&ui->ui_lock);
914         ubifs_assert(c, page->index <= ui->synced_i_size >> PAGE_SHIFT);
915         spin_unlock(&ui->ui_lock);
916 #endif
917 
918         /* Update radix tree tags */
919         set_page_writeback(page);
920 
921         addr = kmap(page);
922         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
923         i = 0;
924         while (len) {
925                 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
926                 data_key_init(c, &key, inode->i_ino, block);
927                 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
928                 if (err)
929                         break;
930                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
931                         break;
932                 block += 1;
933                 addr += blen;
934                 len -= blen;
935         }
936         if (err) {
937                 SetPageError(page);
938                 ubifs_err(c, "cannot write page %lu of inode %lu, error %d",
939                           page->index, inode->i_ino, err);
940                 ubifs_ro_mode(c, err);
941         }
942 
943         ubifs_assert(c, PagePrivate(page));
944         if (PageChecked(page))
945                 release_new_page_budget(c);
946         else
947                 release_existing_page_budget(c);
948 
949         atomic_long_dec(&c->dirty_pg_cnt);
950         ClearPagePrivate(page);
951         ClearPageChecked(page);
952 
953         kunmap(page);
954         unlock_page(page);
955         end_page_writeback(page);
956         return err;
957 }
958 
959 /*
960  * When writing-back dirty inodes, VFS first writes-back pages belonging to the
961  * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
962  * situation when a we have an inode with size 0, then a megabyte of data is
963  * appended to the inode, then write-back starts and flushes some amount of the
964  * dirty pages, the journal becomes full, commit happens and finishes, and then
965  * an unclean reboot happens. When the file system is mounted next time, the
966  * inode size would still be 0, but there would be many pages which are beyond
967  * the inode size, they would be indexed and consume flash space. Because the
968  * journal has been committed, the replay would not be able to detect this
969  * situation and correct the inode size. This means UBIFS would have to scan
970  * whole index and correct all inode sizes, which is long an unacceptable.
971  *
972  * To prevent situations like this, UBIFS writes pages back only if they are
973  * within the last synchronized inode size, i.e. the size which has been
974  * written to the flash media last time. Otherwise, UBIFS forces inode
975  * write-back, thus making sure the on-flash inode contains current inode size,
976  * and then keeps writing pages back.
977  *
978  * Some locking issues explanation. 'ubifs_writepage()' first is called with
979  * the page locked, and it locks @ui_mutex. However, write-back does take inode
980  * @i_mutex, which means other VFS operations may be run on this inode at the
981  * same time. And the problematic one is truncation to smaller size, from where
982  * we have to call 'truncate_setsize()', which first changes @inode->i_size,
983  * then drops the truncated pages. And while dropping the pages, it takes the
984  * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
985  * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
986  * This means that @inode->i_size is changed while @ui_mutex is unlocked.
987  *
988  * XXX(truncate): with the new truncate sequence this is not true anymore,
989  * and the calls to truncate_setsize can be move around freely.  They should
990  * be moved to the very end of the truncate sequence.
991  *
992  * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
993  * inode size. How do we do this if @inode->i_size may became smaller while we
994  * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
995  * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
996  * internally and updates it under @ui_mutex.
997  *
998  * Q: why we do not worry that if we race with truncation, we may end up with a
999  * situation when the inode is truncated while we are in the middle of
1000  * 'do_writepage()', so we do write beyond inode size?
1001  * A: If we are in the middle of 'do_writepage()', truncation would be locked
1002  * on the page lock and it would not write the truncated inode node to the
1003  * journal before we have finished.
1004  */
1005 static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
1006 {
1007         struct inode *inode = page->mapping->host;
1008         struct ubifs_info *c = inode->i_sb->s_fs_info;
1009         struct ubifs_inode *ui = ubifs_inode(inode);
1010         loff_t i_size =  i_size_read(inode), synced_i_size;
1011         pgoff_t end_index = i_size >> PAGE_SHIFT;
1012         int err, len = i_size & (PAGE_SIZE - 1);
1013         void *kaddr;
1014 
1015         dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1016                 inode->i_ino, page->index, page->flags);
1017         ubifs_assert(c, PagePrivate(page));
1018 
1019         /* Is the page fully outside @i_size? (truncate in progress) */
1020         if (page->index > end_index || (page->index == end_index && !len)) {
1021                 err = 0;
1022                 goto out_unlock;
1023         }
1024 
1025         spin_lock(&ui->ui_lock);
1026         synced_i_size = ui->synced_i_size;
1027         spin_unlock(&ui->ui_lock);
1028 
1029         /* Is the page fully inside @i_size? */
1030         if (page->index < end_index) {
1031                 if (page->index >= synced_i_size >> PAGE_SHIFT) {
1032                         err = inode->i_sb->s_op->write_inode(inode, NULL);
1033                         if (err)
1034                                 goto out_unlock;
1035                         /*
1036                          * The inode has been written, but the write-buffer has
1037                          * not been synchronized, so in case of an unclean
1038                          * reboot we may end up with some pages beyond inode
1039                          * size, but they would be in the journal (because
1040                          * commit flushes write buffers) and recovery would deal
1041                          * with this.
1042                          */
1043                 }
1044                 return do_writepage(page, PAGE_SIZE);
1045         }
1046 
1047         /*
1048          * The page straddles @i_size. It must be zeroed out on each and every
1049          * writepage invocation because it may be mmapped. "A file is mapped
1050          * in multiples of the page size. For a file that is not a multiple of
1051          * the page size, the remaining memory is zeroed when mapped, and
1052          * writes to that region are not written out to the file."
1053          */
1054         kaddr = kmap_atomic(page);
1055         memset(kaddr + len, 0, PAGE_SIZE - len);
1056         flush_dcache_page(page);
1057         kunmap_atomic(kaddr);
1058 
1059         if (i_size > synced_i_size) {
1060                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1061                 if (err)
1062                         goto out_unlock;
1063         }
1064 
1065         return do_writepage(page, len);
1066 
1067 out_unlock:
1068         unlock_page(page);
1069         return err;
1070 }
1071 
1072 /**
1073  * do_attr_changes - change inode attributes.
1074  * @inode: inode to change attributes for
1075  * @attr: describes attributes to change
1076  */
1077 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1078 {
1079         if (attr->ia_valid & ATTR_UID)
1080                 inode->i_uid = attr->ia_uid;
1081         if (attr->ia_valid & ATTR_GID)
1082                 inode->i_gid = attr->ia_gid;
1083         if (attr->ia_valid & ATTR_ATIME)
1084                 inode->i_atime = attr->ia_atime;
1085         if (attr->ia_valid & ATTR_MTIME)
1086                 inode->i_mtime = attr->ia_mtime;
1087         if (attr->ia_valid & ATTR_CTIME)
1088                 inode->i_ctime = attr->ia_ctime;
1089         if (attr->ia_valid & ATTR_MODE) {
1090                 umode_t mode = attr->ia_mode;
1091 
1092                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1093                         mode &= ~S_ISGID;
1094                 inode->i_mode = mode;
1095         }
1096 }
1097 
1098 /**
1099  * do_truncation - truncate an inode.
1100  * @c: UBIFS file-system description object
1101  * @inode: inode to truncate
1102  * @attr: inode attribute changes description
1103  *
1104  * This function implements VFS '->setattr()' call when the inode is truncated
1105  * to a smaller size. Returns zero in case of success and a negative error code
1106  * in case of failure.
1107  */
1108 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1109                          const struct iattr *attr)
1110 {
1111         int err;
1112         struct ubifs_budget_req req;
1113         loff_t old_size = inode->i_size, new_size = attr->ia_size;
1114         int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1115         struct ubifs_inode *ui = ubifs_inode(inode);
1116 
1117         dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1118         memset(&req, 0, sizeof(struct ubifs_budget_req));
1119 
1120         /*
1121          * If this is truncation to a smaller size, and we do not truncate on a
1122          * block boundary, budget for changing one data block, because the last
1123          * block will be re-written.
1124          */
1125         if (new_size & (UBIFS_BLOCK_SIZE - 1))
1126                 req.dirtied_page = 1;
1127 
1128         req.dirtied_ino = 1;
1129         /* A funny way to budget for truncation node */
1130         req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1131         err = ubifs_budget_space(c, &req);
1132         if (err) {
1133                 /*
1134                  * Treat truncations to zero as deletion and always allow them,
1135                  * just like we do for '->unlink()'.
1136                  */
1137                 if (new_size || err != -ENOSPC)
1138                         return err;
1139                 budgeted = 0;
1140         }
1141 
1142         truncate_setsize(inode, new_size);
1143 
1144         if (offset) {
1145                 pgoff_t index = new_size >> PAGE_SHIFT;
1146                 struct page *page;
1147 
1148                 page = find_lock_page(inode->i_mapping, index);
1149                 if (page) {
1150                         if (PageDirty(page)) {
1151                                 /*
1152                                  * 'ubifs_jnl_truncate()' will try to truncate
1153                                  * the last data node, but it contains
1154                                  * out-of-date data because the page is dirty.
1155                                  * Write the page now, so that
1156                                  * 'ubifs_jnl_truncate()' will see an already
1157                                  * truncated (and up to date) data node.
1158                                  */
1159                                 ubifs_assert(c, PagePrivate(page));
1160 
1161                                 clear_page_dirty_for_io(page);
1162                                 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1163                                         offset = new_size &
1164                                                  (PAGE_SIZE - 1);
1165                                 err = do_writepage(page, offset);
1166                                 put_page(page);
1167                                 if (err)
1168                                         goto out_budg;
1169                                 /*
1170                                  * We could now tell 'ubifs_jnl_truncate()' not
1171                                  * to read the last block.
1172                                  */
1173                         } else {
1174                                 /*
1175                                  * We could 'kmap()' the page and pass the data
1176                                  * to 'ubifs_jnl_truncate()' to save it from
1177                                  * having to read it.
1178                                  */
1179                                 unlock_page(page);
1180                                 put_page(page);
1181                         }
1182                 }
1183         }
1184 
1185         mutex_lock(&ui->ui_mutex);
1186         ui->ui_size = inode->i_size;
1187         /* Truncation changes inode [mc]time */
1188         inode->i_mtime = inode->i_ctime = current_time(inode);
1189         /* Other attributes may be changed at the same time as well */
1190         do_attr_changes(inode, attr);
1191         err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1192         mutex_unlock(&ui->ui_mutex);
1193 
1194 out_budg:
1195         if (budgeted)
1196                 ubifs_release_budget(c, &req);
1197         else {
1198                 c->bi.nospace = c->bi.nospace_rp = 0;
1199                 smp_wmb();
1200         }
1201         return err;
1202 }
1203 
1204 /**
1205  * do_setattr - change inode attributes.
1206  * @c: UBIFS file-system description object
1207  * @inode: inode to change attributes for
1208  * @attr: inode attribute changes description
1209  *
1210  * This function implements VFS '->setattr()' call for all cases except
1211  * truncations to smaller size. Returns zero in case of success and a negative
1212  * error code in case of failure.
1213  */
1214 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1215                       const struct iattr *attr)
1216 {
1217         int err, release;
1218         loff_t new_size = attr->ia_size;
1219         struct ubifs_inode *ui = ubifs_inode(inode);
1220         struct ubifs_budget_req req = { .dirtied_ino = 1,
1221                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1222 
1223         err = ubifs_budget_space(c, &req);
1224         if (err)
1225                 return err;
1226 
1227         if (attr->ia_valid & ATTR_SIZE) {
1228                 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1229                 truncate_setsize(inode, new_size);
1230         }
1231 
1232         mutex_lock(&ui->ui_mutex);
1233         if (attr->ia_valid & ATTR_SIZE) {
1234                 /* Truncation changes inode [mc]time */
1235                 inode->i_mtime = inode->i_ctime = current_time(inode);
1236                 /* 'truncate_setsize()' changed @i_size, update @ui_size */
1237                 ui->ui_size = inode->i_size;
1238         }
1239 
1240         do_attr_changes(inode, attr);
1241 
1242         release = ui->dirty;
1243         if (attr->ia_valid & ATTR_SIZE)
1244                 /*
1245                  * Inode length changed, so we have to make sure
1246                  * @I_DIRTY_DATASYNC is set.
1247                  */
1248                  __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1249         else
1250                 mark_inode_dirty_sync(inode);
1251         mutex_unlock(&ui->ui_mutex);
1252 
1253         if (release)
1254                 ubifs_release_budget(c, &req);
1255         if (IS_SYNC(inode))
1256                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1257         return err;
1258 }
1259 
1260 int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1261 {
1262         int err;
1263         struct inode *inode = d_inode(dentry);
1264         struct ubifs_info *c = inode->i_sb->s_fs_info;
1265 
1266         dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1267                 inode->i_ino, inode->i_mode, attr->ia_valid);
1268         err = setattr_prepare(dentry, attr);
1269         if (err)
1270                 return err;
1271 
1272         err = dbg_check_synced_i_size(c, inode);
1273         if (err)
1274                 return err;
1275 
1276         err = fscrypt_prepare_setattr(dentry, attr);
1277         if (err)
1278                 return err;
1279 
1280         if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1281                 /* Truncation to a smaller size */
1282                 err = do_truncation(c, inode, attr);
1283         else
1284                 err = do_setattr(c, inode, attr);
1285 
1286         return err;
1287 }
1288 
1289 static void ubifs_invalidatepage(struct page *page, unsigned int offset,
1290                                  unsigned int length)
1291 {
1292         struct inode *inode = page->mapping->host;
1293         struct ubifs_info *c = inode->i_sb->s_fs_info;
1294 
1295         ubifs_assert(c, PagePrivate(page));
1296         if (offset || length < PAGE_SIZE)
1297                 /* Partial page remains dirty */
1298                 return;
1299 
1300         if (PageChecked(page))
1301                 release_new_page_budget(c);
1302         else
1303                 release_existing_page_budget(c);
1304 
1305         atomic_long_dec(&c->dirty_pg_cnt);
1306         ClearPagePrivate(page);
1307         ClearPageChecked(page);
1308 }
1309 
1310 int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1311 {
1312         struct inode *inode = file->f_mapping->host;
1313         struct ubifs_info *c = inode->i_sb->s_fs_info;
1314         int err;
1315 
1316         dbg_gen("syncing inode %lu", inode->i_ino);
1317 
1318         if (c->ro_mount)
1319                 /*
1320                  * For some really strange reasons VFS does not filter out
1321                  * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1322                  */
1323                 return 0;
1324 
1325         err = file_write_and_wait_range(file, start, end);
1326         if (err)
1327                 return err;
1328         inode_lock(inode);
1329 
1330         /* Synchronize the inode unless this is a 'datasync()' call. */
1331         if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1332                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1333                 if (err)
1334                         goto out;
1335         }
1336 
1337         /*
1338          * Nodes related to this inode may still sit in a write-buffer. Flush
1339          * them.
1340          */
1341         err = ubifs_sync_wbufs_by_inode(c, inode);
1342 out:
1343         inode_unlock(inode);
1344         return err;
1345 }
1346 
1347 /**
1348  * mctime_update_needed - check if mtime or ctime update is needed.
1349  * @inode: the inode to do the check for
1350  * @now: current time
1351  *
1352  * This helper function checks if the inode mtime/ctime should be updated or
1353  * not. If current values of the time-stamps are within the UBIFS inode time
1354  * granularity, they are not updated. This is an optimization.
1355  */
1356 static inline int mctime_update_needed(const struct inode *inode,
1357                                        const struct timespec64 *now)
1358 {
1359         if (!timespec64_equal(&inode->i_mtime, now) ||
1360             !timespec64_equal(&inode->i_ctime, now))
1361                 return 1;
1362         return 0;
1363 }
1364 
1365 /**
1366  * ubifs_update_time - update time of inode.
1367  * @inode: inode to update
1368  *
1369  * This function updates time of the inode.
1370  */
1371 int ubifs_update_time(struct inode *inode, struct timespec64 *time,
1372                              int flags)
1373 {
1374         struct ubifs_inode *ui = ubifs_inode(inode);
1375         struct ubifs_info *c = inode->i_sb->s_fs_info;
1376         struct ubifs_budget_req req = { .dirtied_ino = 1,
1377                         .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1378         int err, release;
1379 
1380         if (!IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1381                 return generic_update_time(inode, time, flags);
1382 
1383         err = ubifs_budget_space(c, &req);
1384         if (err)
1385                 return err;
1386 
1387         mutex_lock(&ui->ui_mutex);
1388         if (flags & S_ATIME)
1389                 inode->i_atime = *time;
1390         if (flags & S_CTIME)
1391                 inode->i_ctime = *time;
1392         if (flags & S_MTIME)
1393                 inode->i_mtime = *time;
1394 
1395         release = ui->dirty;
1396         __mark_inode_dirty(inode, I_DIRTY_SYNC);
1397         mutex_unlock(&ui->ui_mutex);
1398         if (release)
1399                 ubifs_release_budget(c, &req);
1400         return 0;
1401 }
1402 
1403 /**
1404  * update_mctime - update mtime and ctime of an inode.
1405  * @inode: inode to update
1406  *
1407  * This function updates mtime and ctime of the inode if it is not equivalent to
1408  * current time. Returns zero in case of success and a negative error code in
1409  * case of failure.
1410  */
1411 static int update_mctime(struct inode *inode)
1412 {
1413         struct timespec64 now = current_time(inode);
1414         struct ubifs_inode *ui = ubifs_inode(inode);
1415         struct ubifs_info *c = inode->i_sb->s_fs_info;
1416 
1417         if (mctime_update_needed(inode, &now)) {
1418                 int err, release;
1419                 struct ubifs_budget_req req = { .dirtied_ino = 1,
1420                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1421 
1422                 err = ubifs_budget_space(c, &req);
1423                 if (err)
1424                         return err;
1425 
1426                 mutex_lock(&ui->ui_mutex);
1427                 inode->i_mtime = inode->i_ctime = current_time(inode);
1428                 release = ui->dirty;
1429                 mark_inode_dirty_sync(inode);
1430                 mutex_unlock(&ui->ui_mutex);
1431                 if (release)
1432                         ubifs_release_budget(c, &req);
1433         }
1434 
1435         return 0;
1436 }
1437 
1438 static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1439 {
1440         int err = update_mctime(file_inode(iocb->ki_filp));
1441         if (err)
1442                 return err;
1443 
1444         return generic_file_write_iter(iocb, from);
1445 }
1446 
1447 static int ubifs_set_page_dirty(struct page *page)
1448 {
1449         int ret;
1450         struct inode *inode = page->mapping->host;
1451         struct ubifs_info *c = inode->i_sb->s_fs_info;
1452 
1453         ret = __set_page_dirty_nobuffers(page);
1454         /*
1455          * An attempt to dirty a page without budgeting for it - should not
1456          * happen.
1457          */
1458         ubifs_assert(c, ret == 0);
1459         return ret;
1460 }
1461 
1462 #ifdef CONFIG_MIGRATION
1463 static int ubifs_migrate_page(struct address_space *mapping,
1464                 struct page *newpage, struct page *page, enum migrate_mode mode)
1465 {
1466         int rc;
1467 
1468         rc = migrate_page_move_mapping(mapping, newpage, page, 0);
1469         if (rc != MIGRATEPAGE_SUCCESS)
1470                 return rc;
1471 
1472         if (PagePrivate(page)) {
1473                 ClearPagePrivate(page);
1474                 SetPagePrivate(newpage);
1475         }
1476 
1477         if (mode != MIGRATE_SYNC_NO_COPY)
1478                 migrate_page_copy(newpage, page);
1479         else
1480                 migrate_page_states(newpage, page);
1481         return MIGRATEPAGE_SUCCESS;
1482 }
1483 #endif
1484 
1485 static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1486 {
1487         struct inode *inode = page->mapping->host;
1488         struct ubifs_info *c = inode->i_sb->s_fs_info;
1489 
1490         /*
1491          * An attempt to release a dirty page without budgeting for it - should
1492          * not happen.
1493          */
1494         if (PageWriteback(page))
1495                 return 0;
1496         ubifs_assert(c, PagePrivate(page));
1497         ubifs_assert(c, 0);
1498         ClearPagePrivate(page);
1499         ClearPageChecked(page);
1500         return 1;
1501 }
1502 
1503 /*
1504  * mmap()d file has taken write protection fault and is being made writable.
1505  * UBIFS must ensure page is budgeted for.
1506  */
1507 static vm_fault_t ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1508 {
1509         struct page *page = vmf->page;
1510         struct inode *inode = file_inode(vmf->vma->vm_file);
1511         struct ubifs_info *c = inode->i_sb->s_fs_info;
1512         struct timespec64 now = current_time(inode);
1513         struct ubifs_budget_req req = { .new_page = 1 };
1514         int err, update_time;
1515 
1516         dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1517                 i_size_read(inode));
1518         ubifs_assert(c, !c->ro_media && !c->ro_mount);
1519 
1520         if (unlikely(c->ro_error))
1521                 return VM_FAULT_SIGBUS; /* -EROFS */
1522 
1523         /*
1524          * We have not locked @page so far so we may budget for changing the
1525          * page. Note, we cannot do this after we locked the page, because
1526          * budgeting may cause write-back which would cause deadlock.
1527          *
1528          * At the moment we do not know whether the page is dirty or not, so we
1529          * assume that it is not and budget for a new page. We could look at
1530          * the @PG_private flag and figure this out, but we may race with write
1531          * back and the page state may change by the time we lock it, so this
1532          * would need additional care. We do not bother with this at the
1533          * moment, although it might be good idea to do. Instead, we allocate
1534          * budget for a new page and amend it later on if the page was in fact
1535          * dirty.
1536          *
1537          * The budgeting-related logic of this function is similar to what we
1538          * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1539          * for more comments.
1540          */
1541         update_time = mctime_update_needed(inode, &now);
1542         if (update_time)
1543                 /*
1544                  * We have to change inode time stamp which requires extra
1545                  * budgeting.
1546                  */
1547                 req.dirtied_ino = 1;
1548 
1549         err = ubifs_budget_space(c, &req);
1550         if (unlikely(err)) {
1551                 if (err == -ENOSPC)
1552                         ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1553                                    inode->i_ino);
1554                 return VM_FAULT_SIGBUS;
1555         }
1556 
1557         lock_page(page);
1558         if (unlikely(page->mapping != inode->i_mapping ||
1559                      page_offset(page) > i_size_read(inode))) {
1560                 /* Page got truncated out from underneath us */
1561                 goto sigbus;
1562         }
1563 
1564         if (PagePrivate(page))
1565                 release_new_page_budget(c);
1566         else {
1567                 if (!PageChecked(page))
1568                         ubifs_convert_page_budget(c);
1569                 SetPagePrivate(page);
1570                 atomic_long_inc(&c->dirty_pg_cnt);
1571                 __set_page_dirty_nobuffers(page);
1572         }
1573 
1574         if (update_time) {
1575                 int release;
1576                 struct ubifs_inode *ui = ubifs_inode(inode);
1577 
1578                 mutex_lock(&ui->ui_mutex);
1579                 inode->i_mtime = inode->i_ctime = current_time(inode);
1580                 release = ui->dirty;
1581                 mark_inode_dirty_sync(inode);
1582                 mutex_unlock(&ui->ui_mutex);
1583                 if (release)
1584                         ubifs_release_dirty_inode_budget(c, ui);
1585         }
1586 
1587         wait_for_stable_page(page);
1588         return VM_FAULT_LOCKED;
1589 
1590 sigbus:
1591         unlock_page(page);
1592         ubifs_release_budget(c, &req);
1593         return VM_FAULT_SIGBUS;
1594 }
1595 
1596 static const struct vm_operations_struct ubifs_file_vm_ops = {
1597         .fault        = filemap_fault,
1598         .map_pages = filemap_map_pages,
1599         .page_mkwrite = ubifs_vm_page_mkwrite,
1600 };
1601 
1602 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1603 {
1604         int err;
1605 
1606         err = generic_file_mmap(file, vma);
1607         if (err)
1608                 return err;
1609         vma->vm_ops = &ubifs_file_vm_ops;
1610 
1611         if (IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1612                 file_accessed(file);
1613 
1614         return 0;
1615 }
1616 
1617 static const char *ubifs_get_link(struct dentry *dentry,
1618                                             struct inode *inode,
1619                                             struct delayed_call *done)
1620 {
1621         struct ubifs_inode *ui = ubifs_inode(inode);
1622 
1623         if (!IS_ENCRYPTED(inode))
1624                 return ui->data;
1625 
1626         if (!dentry)
1627                 return ERR_PTR(-ECHILD);
1628 
1629         return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1630 }
1631 
1632 const struct address_space_operations ubifs_file_address_operations = {
1633         .readpage       = ubifs_readpage,
1634         .writepage      = ubifs_writepage,
1635         .write_begin    = ubifs_write_begin,
1636         .write_end      = ubifs_write_end,
1637         .invalidatepage = ubifs_invalidatepage,
1638         .set_page_dirty = ubifs_set_page_dirty,
1639 #ifdef CONFIG_MIGRATION
1640         .migratepage    = ubifs_migrate_page,
1641 #endif
1642         .releasepage    = ubifs_releasepage,
1643 };
1644 
1645 const struct inode_operations ubifs_file_inode_operations = {
1646         .setattr     = ubifs_setattr,
1647         .getattr     = ubifs_getattr,
1648         .listxattr   = ubifs_listxattr,
1649         .update_time = ubifs_update_time,
1650 };
1651 
1652 const struct inode_operations ubifs_symlink_inode_operations = {
1653         .get_link    = ubifs_get_link,
1654         .setattr     = ubifs_setattr,
1655         .getattr     = ubifs_getattr,
1656         .listxattr   = ubifs_listxattr,
1657         .update_time = ubifs_update_time,
1658 };
1659 
1660 const struct file_operations ubifs_file_operations = {
1661         .llseek         = generic_file_llseek,
1662         .read_iter      = generic_file_read_iter,
1663         .write_iter     = ubifs_write_iter,
1664         .mmap           = ubifs_file_mmap,
1665         .fsync          = ubifs_fsync,
1666         .unlocked_ioctl = ubifs_ioctl,
1667         .splice_read    = generic_file_splice_read,
1668         .splice_write   = iter_file_splice_write,
1669         .open           = fscrypt_file_open,
1670 #ifdef CONFIG_COMPAT
1671         .compat_ioctl   = ubifs_compat_ioctl,
1672 #endif
1673 };
1674 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp