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

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

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * This file is part of UBIFS.
  3  *
  4  * Copyright (C) 2006-2008 Nokia Corporation.
  5  *
  6  * This program is free software; you can redistribute it and/or modify it
  7  * under the terms of the GNU General Public License version 2 as published by
  8  * the Free Software Foundation.
  9  *
 10  * This program is distributed in the hope that it will be useful, but WITHOUT
 11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 13  * more details.
 14  *
 15  * You should have received a copy of the GNU General Public License along with
 16  * this program; if not, write to the Free Software Foundation, Inc., 51
 17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 18  *
 19  * Authors: Artem Bityutskiy (Битюцкий Артём)
 20  *          Adrian Hunter
 21  */
 22 
 23 /*
 24  * This file implements VFS file and inode operations for regular files, device
 25  * nodes and symlinks as well as address space operations.
 26  *
 27  * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
 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.
 37  *
 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 flusher thread is doing
 41  * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
 42  * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
 43  * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
 44  * 'ubifs_writepage()' we are only guaranteed that the page is locked.
 45  *
 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.
 50  */
 51 
 52 #include "ubifs.h"
 53 #include <linux/mount.h>
 54 #include <linux/slab.h>
 55 #include <linux/migrate.h>
 56 
 57 static int read_block(struct inode *inode, void *addr, unsigned int block,
 58                       struct ubifs_data_node *dn)
 59 {
 60         struct ubifs_info *c = inode->i_sb->s_fs_info;
 61         int err, len, out_len;
 62         union ubifs_key key;
 63         unsigned int dlen;
 64 
 65         data_key_init(c, &key, inode->i_ino, block);
 66         err = ubifs_tnc_lookup(c, &key, dn);
 67         if (err) {
 68                 if (err == -ENOENT)
 69                         /* Not found, so it must be a hole */
 70                         memset(addr, 0, UBIFS_BLOCK_SIZE);
 71                 return err;
 72         }
 73 
 74         ubifs_assert(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;
 79 
 80         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
 81 
 82         if (ubifs_crypt_is_encrypted(inode)) {
 83                 err = ubifs_decrypt(inode, dn, &dlen, block);
 84                 if (err)
 85                         goto dump;
 86         }
 87 
 88         out_len = UBIFS_BLOCK_SIZE;
 89         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
 90                                le16_to_cpu(dn->compr_type));
 91         if (err || len != out_len)
 92                 goto dump;
 93 
 94         /*
 95          * Data length can be less than a full block, even for blocks that are
 96          * not the last in the file (e.g., as a result of making a hole and
 97          * appending data). Ensure that the remainder is zeroed out.
 98          */
 99         if (len < UBIFS_BLOCK_SIZE)
100                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
101 
102         return 0;
103 
104 dump:
105         ubifs_err(c, "bad data node (block %u, inode %lu)",
106                   block, inode->i_ino);
107         ubifs_dump_node(c, dn);
108         return -EINVAL;
109 }
110 
111 static int do_readpage(struct page *page)
112 {
113         void *addr;
114         int err = 0, i;
115         unsigned int block, beyond;
116         struct ubifs_data_node *dn;
117         struct inode *inode = page->mapping->host;
118         loff_t i_size = i_size_read(inode);
119 
120         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
121                 inode->i_ino, page->index, i_size, page->flags);
122         ubifs_assert(!PageChecked(page));
123         ubifs_assert(!PagePrivate(page));
124 
125         addr = kmap(page);
126 
127         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
128         beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
129         if (block >= beyond) {
130                 /* Reading beyond inode */
131                 SetPageChecked(page);
132                 memset(addr, 0, PAGE_SIZE);
133                 goto out;
134         }
135 
136         dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
137         if (!dn) {
138                 err = -ENOMEM;
139                 goto error;
140         }
141 
142         i = 0;
143         while (1) {
144                 int ret;
145 
146                 if (block >= beyond) {
147                         /* Reading beyond inode */
148                         err = -ENOENT;
149                         memset(addr, 0, UBIFS_BLOCK_SIZE);
150                 } else {
151                         ret = read_block(inode, addr, block, dn);
152                         if (ret) {
153                                 err = ret;
154                                 if (err != -ENOENT)
155                                         break;
156                         } else if (block + 1 == beyond) {
157                                 int dlen = le32_to_cpu(dn->size);
158                                 int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
159 
160                                 if (ilen && ilen < dlen)
161                                         memset(addr + ilen, 0, dlen - ilen);
162                         }
163                 }
164                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
165                         break;
166                 block += 1;
167                 addr += UBIFS_BLOCK_SIZE;
168         }
169         if (err) {
170                 struct ubifs_info *c = inode->i_sb->s_fs_info;
171                 if (err == -ENOENT) {
172                         /* Not found, so it must be a hole */
173                         SetPageChecked(page);
174                         dbg_gen("hole");
175                         goto out_free;
176                 }
177                 ubifs_err(c, "cannot read page %lu of inode %lu, error %d",
178                           page->index, inode->i_ino, err);
179                 goto error;
180         }
181 
182 out_free:
183         kfree(dn);
184 out:
185         SetPageUptodate(page);
186         ClearPageError(page);
187         flush_dcache_page(page);
188         kunmap(page);
189         return 0;
190 
191 error:
192         kfree(dn);
193         ClearPageUptodate(page);
194         SetPageError(page);
195         flush_dcache_page(page);
196         kunmap(page);
197         return err;
198 }
199 
200 /**
201  * release_new_page_budget - release budget of a new page.
202  * @c: UBIFS file-system description object
203  *
204  * This is a helper function which releases budget corresponding to the budget
205  * of one new page of data.
206  */
207 static void release_new_page_budget(struct ubifs_info *c)
208 {
209         struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
210 
211         ubifs_release_budget(c, &req);
212 }
213 
214 /**
215  * release_existing_page_budget - release budget of an existing page.
216  * @c: UBIFS file-system description object
217  *
218  * This is a helper function which releases budget corresponding to the budget
219  * of changing one one page of data which already exists on the flash media.
220  */
221 static void release_existing_page_budget(struct ubifs_info *c)
222 {
223         struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
224 
225         ubifs_release_budget(c, &req);
226 }
227 
228 static int write_begin_slow(struct address_space *mapping,
229                             loff_t pos, unsigned len, struct page **pagep,
230                             unsigned flags)
231 {
232         struct inode *inode = mapping->host;
233         struct ubifs_info *c = inode->i_sb->s_fs_info;
234         pgoff_t index = pos >> PAGE_SHIFT;
235         struct ubifs_budget_req req = { .new_page = 1 };
236         int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
237         struct page *page;
238 
239         dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
240                 inode->i_ino, pos, len, inode->i_size);
241 
242         /*
243          * At the slow path we have to budget before locking the page, because
244          * budgeting may force write-back, which would wait on locked pages and
245          * deadlock if we had the page locked. At this point we do not know
246          * anything about the page, so assume that this is a new page which is
247          * written to a hole. This corresponds to largest budget. Later the
248          * budget will be amended if this is not true.
249          */
250         if (appending)
251                 /* We are appending data, budget for inode change */
252                 req.dirtied_ino = 1;
253 
254         err = ubifs_budget_space(c, &req);
255         if (unlikely(err))
256                 return err;
257 
258         page = grab_cache_page_write_begin(mapping, index, flags);
259         if (unlikely(!page)) {
260                 ubifs_release_budget(c, &req);
261                 return -ENOMEM;
262         }
263 
264         if (!PageUptodate(page)) {
265                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE)
266                         SetPageChecked(page);
267                 else {
268                         err = do_readpage(page);
269                         if (err) {
270                                 unlock_page(page);
271                                 put_page(page);
272                                 ubifs_release_budget(c, &req);
273                                 return err;
274                         }
275                 }
276 
277                 SetPageUptodate(page);
278                 ClearPageError(page);
279         }
280 
281         if (PagePrivate(page))
282                 /*
283                  * The page is dirty, which means it was budgeted twice:
284                  *   o first time the budget was allocated by the task which
285                  *     made the page dirty and set the PG_private flag;
286                  *   o and then we budgeted for it for the second time at the
287                  *     very beginning of this function.
288                  *
289                  * So what we have to do is to release the page budget we
290                  * allocated.
291                  */
292                 release_new_page_budget(c);
293         else if (!PageChecked(page))
294                 /*
295                  * We are changing a page which already exists on the media.
296                  * This means that changing the page does not make the amount
297                  * of indexing information larger, and this part of the budget
298                  * which we have already acquired may be released.
299                  */
300                 ubifs_convert_page_budget(c);
301 
302         if (appending) {
303                 struct ubifs_inode *ui = ubifs_inode(inode);
304 
305                 /*
306                  * 'ubifs_write_end()' is optimized from the fast-path part of
307                  * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
308                  * if data is appended.
309                  */
310                 mutex_lock(&ui->ui_mutex);
311                 if (ui->dirty)
312                         /*
313                          * The inode is dirty already, so we may free the
314                          * budget we allocated.
315                          */
316                         ubifs_release_dirty_inode_budget(c, ui);
317         }
318 
319         *pagep = page;
320         return 0;
321 }
322 
323 /**
324  * allocate_budget - allocate budget for 'ubifs_write_begin()'.
325  * @c: UBIFS file-system description object
326  * @page: page to allocate budget for
327  * @ui: UBIFS inode object the page belongs to
328  * @appending: non-zero if the page is appended
329  *
330  * This is a helper function for 'ubifs_write_begin()' which allocates budget
331  * for the operation. The budget is allocated differently depending on whether
332  * this is appending, whether the page is dirty or not, and so on. This
333  * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
334  * in case of success and %-ENOSPC in case of failure.
335  */
336 static int allocate_budget(struct ubifs_info *c, struct page *page,
337                            struct ubifs_inode *ui, int appending)
338 {
339         struct ubifs_budget_req req = { .fast = 1 };
340 
341         if (PagePrivate(page)) {
342                 if (!appending)
343                         /*
344                          * The page is dirty and we are not appending, which
345                          * means no budget is needed at all.
346                          */
347                         return 0;
348 
349                 mutex_lock(&ui->ui_mutex);
350                 if (ui->dirty)
351                         /*
352                          * The page is dirty and we are appending, so the inode
353                          * has to be marked as dirty. However, it is already
354                          * dirty, so we do not need any budget. We may return,
355                          * but @ui->ui_mutex hast to be left locked because we
356                          * should prevent write-back from flushing the inode
357                          * and freeing the budget. The lock will be released in
358                          * 'ubifs_write_end()'.
359                          */
360                         return 0;
361 
362                 /*
363                  * The page is dirty, we are appending, the inode is clean, so
364                  * we need to budget the inode change.
365                  */
366                 req.dirtied_ino = 1;
367         } else {
368                 if (PageChecked(page))
369                         /*
370                          * The page corresponds to a hole and does not
371                          * exist on the media. So changing it makes
372                          * make the amount of indexing information
373                          * larger, and we have to budget for a new
374                          * page.
375                          */
376                         req.new_page = 1;
377                 else
378                         /*
379                          * Not a hole, the change will not add any new
380                          * indexing information, budget for page
381                          * change.
382                          */
383                         req.dirtied_page = 1;
384 
385                 if (appending) {
386                         mutex_lock(&ui->ui_mutex);
387                         if (!ui->dirty)
388                                 /*
389                                  * The inode is clean but we will have to mark
390                                  * it as dirty because we are appending. This
391                                  * needs a budget.
392                                  */
393                                 req.dirtied_ino = 1;
394                 }
395         }
396 
397         return ubifs_budget_space(c, &req);
398 }
399 
400 /*
401  * This function is called when a page of data is going to be written. Since
402  * the page of data will not necessarily go to the flash straight away, UBIFS
403  * has to reserve space on the media for it, which is done by means of
404  * budgeting.
405  *
406  * This is the hot-path of the file-system and we are trying to optimize it as
407  * much as possible. For this reasons it is split on 2 parts - slow and fast.
408  *
409  * There many budgeting cases:
410  *     o a new page is appended - we have to budget for a new page and for
411  *       changing the inode; however, if the inode is already dirty, there is
412  *       no need to budget for it;
413  *     o an existing clean page is changed - we have budget for it; if the page
414  *       does not exist on the media (a hole), we have to budget for a new
415  *       page; otherwise, we may budget for changing an existing page; the
416  *       difference between these cases is that changing an existing page does
417  *       not introduce anything new to the FS indexing information, so it does
418  *       not grow, and smaller budget is acquired in this case;
419  *     o an existing dirty page is changed - no need to budget at all, because
420  *       the page budget has been acquired by earlier, when the page has been
421  *       marked dirty.
422  *
423  * UBIFS budgeting sub-system may force write-back if it thinks there is no
424  * space to reserve. This imposes some locking restrictions and makes it
425  * impossible to take into account the above cases, and makes it impossible to
426  * optimize budgeting.
427  *
428  * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
429  * there is a plenty of flash space and the budget will be acquired quickly,
430  * without forcing write-back. The slow path does not make this assumption.
431  */
432 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
433                              loff_t pos, unsigned len, unsigned flags,
434                              struct page **pagep, void **fsdata)
435 {
436         struct inode *inode = mapping->host;
437         struct ubifs_info *c = inode->i_sb->s_fs_info;
438         struct ubifs_inode *ui = ubifs_inode(inode);
439         pgoff_t index = pos >> PAGE_SHIFT;
440         int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
441         int skipped_read = 0;
442         struct page *page;
443 
444         ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
445         ubifs_assert(!c->ro_media && !c->ro_mount);
446 
447         if (unlikely(c->ro_error))
448                 return -EROFS;
449 
450         /* Try out the fast-path part first */
451         page = grab_cache_page_write_begin(mapping, index, flags);
452         if (unlikely(!page))
453                 return -ENOMEM;
454 
455         if (!PageUptodate(page)) {
456                 /* The page is not loaded from the flash */
457                 if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE) {
458                         /*
459                          * We change whole page so no need to load it. But we
460                          * do not know whether this page exists on the media or
461                          * not, so we assume the latter because it requires
462                          * larger budget. The assumption is that it is better
463                          * to budget a bit more than to read the page from the
464                          * media. Thus, we are setting the @PG_checked flag
465                          * here.
466                          */
467                         SetPageChecked(page);
468                         skipped_read = 1;
469                 } else {
470                         err = do_readpage(page);
471                         if (err) {
472                                 unlock_page(page);
473                                 put_page(page);
474                                 return err;
475                         }
476                 }
477 
478                 SetPageUptodate(page);
479                 ClearPageError(page);
480         }
481 
482         err = allocate_budget(c, page, ui, appending);
483         if (unlikely(err)) {
484                 ubifs_assert(err == -ENOSPC);
485                 /*
486                  * If we skipped reading the page because we were going to
487                  * write all of it, then it is not up to date.
488                  */
489                 if (skipped_read) {
490                         ClearPageChecked(page);
491                         ClearPageUptodate(page);
492                 }
493                 /*
494                  * Budgeting failed which means it would have to force
495                  * write-back but didn't, because we set the @fast flag in the
496                  * request. Write-back cannot be done now, while we have the
497                  * page locked, because it would deadlock. Unlock and free
498                  * everything and fall-back to slow-path.
499                  */
500                 if (appending) {
501                         ubifs_assert(mutex_is_locked(&ui->ui_mutex));
502                         mutex_unlock(&ui->ui_mutex);
503                 }
504                 unlock_page(page);
505                 put_page(page);
506 
507                 return write_begin_slow(mapping, pos, len, pagep, flags);
508         }
509 
510         /*
511          * Whee, we acquired budgeting quickly - without involving
512          * garbage-collection, committing or forcing write-back. We return
513          * with @ui->ui_mutex locked if we are appending pages, and unlocked
514          * otherwise. This is an optimization (slightly hacky though).
515          */
516         *pagep = page;
517         return 0;
518 
519 }
520 
521 /**
522  * cancel_budget - cancel budget.
523  * @c: UBIFS file-system description object
524  * @page: page to cancel budget for
525  * @ui: UBIFS inode object the page belongs to
526  * @appending: non-zero if the page is appended
527  *
528  * This is a helper function for a page write operation. It unlocks the
529  * @ui->ui_mutex in case of appending.
530  */
531 static void cancel_budget(struct ubifs_info *c, struct page *page,
532                           struct ubifs_inode *ui, int appending)
533 {
534         if (appending) {
535                 if (!ui->dirty)
536                         ubifs_release_dirty_inode_budget(c, ui);
537                 mutex_unlock(&ui->ui_mutex);
538         }
539         if (!PagePrivate(page)) {
540                 if (PageChecked(page))
541                         release_new_page_budget(c);
542                 else
543                         release_existing_page_budget(c);
544         }
545 }
546 
547 static int ubifs_write_end(struct file *file, struct address_space *mapping,
548                            loff_t pos, unsigned len, unsigned copied,
549                            struct page *page, void *fsdata)
550 {
551         struct inode *inode = mapping->host;
552         struct ubifs_inode *ui = ubifs_inode(inode);
553         struct ubifs_info *c = inode->i_sb->s_fs_info;
554         loff_t end_pos = pos + len;
555         int appending = !!(end_pos > inode->i_size);
556 
557         dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
558                 inode->i_ino, pos, page->index, len, copied, inode->i_size);
559 
560         if (unlikely(copied < len && len == PAGE_SIZE)) {
561                 /*
562                  * VFS copied less data to the page that it intended and
563                  * declared in its '->write_begin()' call via the @len
564                  * argument. If the page was not up-to-date, and @len was
565                  * @PAGE_SIZE, the 'ubifs_write_begin()' function did
566                  * not load it from the media (for optimization reasons). This
567                  * means that part of the page contains garbage. So read the
568                  * page now.
569                  */
570                 dbg_gen("copied %d instead of %d, read page and repeat",
571                         copied, len);
572                 cancel_budget(c, page, ui, appending);
573                 ClearPageChecked(page);
574 
575                 /*
576                  * Return 0 to force VFS to repeat the whole operation, or the
577                  * error code if 'do_readpage()' fails.
578                  */
579                 copied = do_readpage(page);
580                 goto out;
581         }
582 
583         if (!PagePrivate(page)) {
584                 SetPagePrivate(page);
585                 atomic_long_inc(&c->dirty_pg_cnt);
586                 __set_page_dirty_nobuffers(page);
587         }
588 
589         if (appending) {
590                 i_size_write(inode, end_pos);
591                 ui->ui_size = end_pos;
592                 /*
593                  * Note, we do not set @I_DIRTY_PAGES (which means that the
594                  * inode has dirty pages), this has been done in
595                  * '__set_page_dirty_nobuffers()'.
596                  */
597                 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
598                 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
599                 mutex_unlock(&ui->ui_mutex);
600         }
601 
602 out:
603         unlock_page(page);
604         put_page(page);
605         return copied;
606 }
607 
608 /**
609  * populate_page - copy data nodes into a page for bulk-read.
610  * @c: UBIFS file-system description object
611  * @page: page
612  * @bu: bulk-read information
613  * @n: next zbranch slot
614  *
615  * This function returns %0 on success and a negative error code on failure.
616  */
617 static int populate_page(struct ubifs_info *c, struct page *page,
618                          struct bu_info *bu, int *n)
619 {
620         int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
621         struct inode *inode = page->mapping->host;
622         loff_t i_size = i_size_read(inode);
623         unsigned int page_block;
624         void *addr, *zaddr;
625         pgoff_t end_index;
626 
627         dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
628                 inode->i_ino, page->index, i_size, page->flags);
629 
630         addr = zaddr = kmap(page);
631 
632         end_index = (i_size - 1) >> PAGE_SHIFT;
633         if (!i_size || page->index > end_index) {
634                 hole = 1;
635                 memset(addr, 0, PAGE_SIZE);
636                 goto out_hole;
637         }
638 
639         page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
640         while (1) {
641                 int err, len, out_len, dlen;
642 
643                 if (nn >= bu->cnt) {
644                         hole = 1;
645                         memset(addr, 0, UBIFS_BLOCK_SIZE);
646                 } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
647                         struct ubifs_data_node *dn;
648 
649                         dn = bu->buf + (bu->zbranch[nn].offs - offs);
650 
651                         ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
652                                      ubifs_inode(inode)->creat_sqnum);
653 
654                         len = le32_to_cpu(dn->size);
655                         if (len <= 0 || len > UBIFS_BLOCK_SIZE)
656                                 goto out_err;
657 
658                         dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
659                         out_len = UBIFS_BLOCK_SIZE;
660 
661                         if (ubifs_crypt_is_encrypted(inode)) {
662                                 err = ubifs_decrypt(inode, dn, &dlen, page_block);
663                                 if (err)
664                                         goto out_err;
665                         }
666 
667                         err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
668                                                le16_to_cpu(dn->compr_type));
669                         if (err || len != out_len)
670                                 goto out_err;
671 
672                         if (len < UBIFS_BLOCK_SIZE)
673                                 memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
674 
675                         nn += 1;
676                         read = (i << UBIFS_BLOCK_SHIFT) + len;
677                 } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
678                         nn += 1;
679                         continue;
680                 } else {
681                         hole = 1;
682                         memset(addr, 0, UBIFS_BLOCK_SIZE);
683                 }
684                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
685                         break;
686                 addr += UBIFS_BLOCK_SIZE;
687                 page_block += 1;
688         }
689 
690         if (end_index == page->index) {
691                 int len = i_size & (PAGE_SIZE - 1);
692 
693                 if (len && len < read)
694                         memset(zaddr + len, 0, read - len);
695         }
696 
697 out_hole:
698         if (hole) {
699                 SetPageChecked(page);
700                 dbg_gen("hole");
701         }
702 
703         SetPageUptodate(page);
704         ClearPageError(page);
705         flush_dcache_page(page);
706         kunmap(page);
707         *n = nn;
708         return 0;
709 
710 out_err:
711         ClearPageUptodate(page);
712         SetPageError(page);
713         flush_dcache_page(page);
714         kunmap(page);
715         ubifs_err(c, "bad data node (block %u, inode %lu)",
716                   page_block, inode->i_ino);
717         return -EINVAL;
718 }
719 
720 /**
721  * ubifs_do_bulk_read - do bulk-read.
722  * @c: UBIFS file-system description object
723  * @bu: bulk-read information
724  * @page1: first page to read
725  *
726  * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
727  */
728 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
729                               struct page *page1)
730 {
731         pgoff_t offset = page1->index, end_index;
732         struct address_space *mapping = page1->mapping;
733         struct inode *inode = mapping->host;
734         struct ubifs_inode *ui = ubifs_inode(inode);
735         int err, page_idx, page_cnt, ret = 0, n = 0;
736         int allocate = bu->buf ? 0 : 1;
737         loff_t isize;
738         gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS;
739 
740         err = ubifs_tnc_get_bu_keys(c, bu);
741         if (err)
742                 goto out_warn;
743 
744         if (bu->eof) {
745                 /* Turn off bulk-read at the end of the file */
746                 ui->read_in_a_row = 1;
747                 ui->bulk_read = 0;
748         }
749 
750         page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
751         if (!page_cnt) {
752                 /*
753                  * This happens when there are multiple blocks per page and the
754                  * blocks for the first page we are looking for, are not
755                  * together. If all the pages were like this, bulk-read would
756                  * reduce performance, so we turn it off for a while.
757                  */
758                 goto out_bu_off;
759         }
760 
761         if (bu->cnt) {
762                 if (allocate) {
763                         /*
764                          * Allocate bulk-read buffer depending on how many data
765                          * nodes we are going to read.
766                          */
767                         bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
768                                       bu->zbranch[bu->cnt - 1].len -
769                                       bu->zbranch[0].offs;
770                         ubifs_assert(bu->buf_len > 0);
771                         ubifs_assert(bu->buf_len <= c->leb_size);
772                         bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
773                         if (!bu->buf)
774                                 goto out_bu_off;
775                 }
776 
777                 err = ubifs_tnc_bulk_read(c, bu);
778                 if (err)
779                         goto out_warn;
780         }
781 
782         err = populate_page(c, page1, bu, &n);
783         if (err)
784                 goto out_warn;
785 
786         unlock_page(page1);
787         ret = 1;
788 
789         isize = i_size_read(inode);
790         if (isize == 0)
791                 goto out_free;
792         end_index = ((isize - 1) >> PAGE_SHIFT);
793 
794         for (page_idx = 1; page_idx < page_cnt; page_idx++) {
795                 pgoff_t page_offset = offset + page_idx;
796                 struct page *page;
797 
798                 if (page_offset > end_index)
799                         break;
800                 page = find_or_create_page(mapping, page_offset, ra_gfp_mask);
801                 if (!page)
802                         break;
803                 if (!PageUptodate(page))
804                         err = populate_page(c, page, bu, &n);
805                 unlock_page(page);
806                 put_page(page);
807                 if (err)
808                         break;
809         }
810 
811         ui->last_page_read = offset + page_idx - 1;
812 
813 out_free:
814         if (allocate)
815                 kfree(bu->buf);
816         return ret;
817 
818 out_warn:
819         ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
820         goto out_free;
821 
822 out_bu_off:
823         ui->read_in_a_row = ui->bulk_read = 0;
824         goto out_free;
825 }
826 
827 /**
828  * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
829  * @page: page from which to start bulk-read.
830  *
831  * Some flash media are capable of reading sequentially at faster rates. UBIFS
832  * bulk-read facility is designed to take advantage of that, by reading in one
833  * go consecutive data nodes that are also located consecutively in the same
834  * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
835  */
836 static int ubifs_bulk_read(struct page *page)
837 {
838         struct inode *inode = page->mapping->host;
839         struct ubifs_info *c = inode->i_sb->s_fs_info;
840         struct ubifs_inode *ui = ubifs_inode(inode);
841         pgoff_t index = page->index, last_page_read = ui->last_page_read;
842         struct bu_info *bu;
843         int err = 0, allocated = 0;
844 
845         ui->last_page_read = index;
846         if (!c->bulk_read)
847                 return 0;
848 
849         /*
850          * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
851          * so don't bother if we cannot lock the mutex.
852          */
853         if (!mutex_trylock(&ui->ui_mutex))
854                 return 0;
855 
856         if (index != last_page_read + 1) {
857                 /* Turn off bulk-read if we stop reading sequentially */
858                 ui->read_in_a_row = 1;
859                 if (ui->bulk_read)
860                         ui->bulk_read = 0;
861                 goto out_unlock;
862         }
863 
864         if (!ui->bulk_read) {
865                 ui->read_in_a_row += 1;
866                 if (ui->read_in_a_row < 3)
867                         goto out_unlock;
868                 /* Three reads in a row, so switch on bulk-read */
869                 ui->bulk_read = 1;
870         }
871 
872         /*
873          * If possible, try to use pre-allocated bulk-read information, which
874          * is protected by @c->bu_mutex.
875          */
876         if (mutex_trylock(&c->bu_mutex))
877                 bu = &c->bu;
878         else {
879                 bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
880                 if (!bu)
881                         goto out_unlock;
882 
883                 bu->buf = NULL;
884                 allocated = 1;
885         }
886 
887         bu->buf_len = c->max_bu_buf_len;
888         data_key_init(c, &bu->key, inode->i_ino,
889                       page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
890         err = ubifs_do_bulk_read(c, bu, page);
891 
892         if (!allocated)
893                 mutex_unlock(&c->bu_mutex);
894         else
895                 kfree(bu);
896 
897 out_unlock:
898         mutex_unlock(&ui->ui_mutex);
899         return err;
900 }
901 
902 static int ubifs_readpage(struct file *file, struct page *page)
903 {
904         if (ubifs_bulk_read(page))
905                 return 0;
906         do_readpage(page);
907         unlock_page(page);
908         return 0;
909 }
910 
911 static int do_writepage(struct page *page, int len)
912 {
913         int err = 0, i, blen;
914         unsigned int block;
915         void *addr;
916         union ubifs_key key;
917         struct inode *inode = page->mapping->host;
918         struct ubifs_info *c = inode->i_sb->s_fs_info;
919 
920 #ifdef UBIFS_DEBUG
921         struct ubifs_inode *ui = ubifs_inode(inode);
922         spin_lock(&ui->ui_lock);
923         ubifs_assert(page->index <= ui->synced_i_size >> PAGE_SHIFT);
924         spin_unlock(&ui->ui_lock);
925 #endif
926 
927         /* Update radix tree tags */
928         set_page_writeback(page);
929 
930         addr = kmap(page);
931         block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
932         i = 0;
933         while (len) {
934                 blen = min_t(int, len, UBIFS_BLOCK_SIZE);
935                 data_key_init(c, &key, inode->i_ino, block);
936                 err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
937                 if (err)
938                         break;
939                 if (++i >= UBIFS_BLOCKS_PER_PAGE)
940                         break;
941                 block += 1;
942                 addr += blen;
943                 len -= blen;
944         }
945         if (err) {
946                 SetPageError(page);
947                 ubifs_err(c, "cannot write page %lu of inode %lu, error %d",
948                           page->index, inode->i_ino, err);
949                 ubifs_ro_mode(c, err);
950         }
951 
952         ubifs_assert(PagePrivate(page));
953         if (PageChecked(page))
954                 release_new_page_budget(c);
955         else
956                 release_existing_page_budget(c);
957 
958         atomic_long_dec(&c->dirty_pg_cnt);
959         ClearPagePrivate(page);
960         ClearPageChecked(page);
961 
962         kunmap(page);
963         unlock_page(page);
964         end_page_writeback(page);
965         return err;
966 }
967 
968 /*
969  * When writing-back dirty inodes, VFS first writes-back pages belonging to the
970  * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
971  * situation when a we have an inode with size 0, then a megabyte of data is
972  * appended to the inode, then write-back starts and flushes some amount of the
973  * dirty pages, the journal becomes full, commit happens and finishes, and then
974  * an unclean reboot happens. When the file system is mounted next time, the
975  * inode size would still be 0, but there would be many pages which are beyond
976  * the inode size, they would be indexed and consume flash space. Because the
977  * journal has been committed, the replay would not be able to detect this
978  * situation and correct the inode size. This means UBIFS would have to scan
979  * whole index and correct all inode sizes, which is long an unacceptable.
980  *
981  * To prevent situations like this, UBIFS writes pages back only if they are
982  * within the last synchronized inode size, i.e. the size which has been
983  * written to the flash media last time. Otherwise, UBIFS forces inode
984  * write-back, thus making sure the on-flash inode contains current inode size,
985  * and then keeps writing pages back.
986  *
987  * Some locking issues explanation. 'ubifs_writepage()' first is called with
988  * the page locked, and it locks @ui_mutex. However, write-back does take inode
989  * @i_mutex, which means other VFS operations may be run on this inode at the
990  * same time. And the problematic one is truncation to smaller size, from where
991  * we have to call 'truncate_setsize()', which first changes @inode->i_size,
992  * then drops the truncated pages. And while dropping the pages, it takes the
993  * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
994  * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
995  * This means that @inode->i_size is changed while @ui_mutex is unlocked.
996  *
997  * XXX(truncate): with the new truncate sequence this is not true anymore,
998  * and the calls to truncate_setsize can be move around freely.  They should
999  * be moved to the very end of the truncate sequence.
1000  *
1001  * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
1002  * inode size. How do we do this if @inode->i_size may became smaller while we
1003  * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
1004  * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
1005  * internally and updates it under @ui_mutex.
1006  *
1007  * Q: why we do not worry that if we race with truncation, we may end up with a
1008  * situation when the inode is truncated while we are in the middle of
1009  * 'do_writepage()', so we do write beyond inode size?
1010  * A: If we are in the middle of 'do_writepage()', truncation would be locked
1011  * on the page lock and it would not write the truncated inode node to the
1012  * journal before we have finished.
1013  */
1014 static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
1015 {
1016         struct inode *inode = page->mapping->host;
1017         struct ubifs_inode *ui = ubifs_inode(inode);
1018         loff_t i_size =  i_size_read(inode), synced_i_size;
1019         pgoff_t end_index = i_size >> PAGE_SHIFT;
1020         int err, len = i_size & (PAGE_SIZE - 1);
1021         void *kaddr;
1022 
1023         dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1024                 inode->i_ino, page->index, page->flags);
1025         ubifs_assert(PagePrivate(page));
1026 
1027         /* Is the page fully outside @i_size? (truncate in progress) */
1028         if (page->index > end_index || (page->index == end_index && !len)) {
1029                 err = 0;
1030                 goto out_unlock;
1031         }
1032 
1033         spin_lock(&ui->ui_lock);
1034         synced_i_size = ui->synced_i_size;
1035         spin_unlock(&ui->ui_lock);
1036 
1037         /* Is the page fully inside @i_size? */
1038         if (page->index < end_index) {
1039                 if (page->index >= synced_i_size >> PAGE_SHIFT) {
1040                         err = inode->i_sb->s_op->write_inode(inode, NULL);
1041                         if (err)
1042                                 goto out_unlock;
1043                         /*
1044                          * The inode has been written, but the write-buffer has
1045                          * not been synchronized, so in case of an unclean
1046                          * reboot we may end up with some pages beyond inode
1047                          * size, but they would be in the journal (because
1048                          * commit flushes write buffers) and recovery would deal
1049                          * with this.
1050                          */
1051                 }
1052                 return do_writepage(page, PAGE_SIZE);
1053         }
1054 
1055         /*
1056          * The page straddles @i_size. It must be zeroed out on each and every
1057          * writepage invocation because it may be mmapped. "A file is mapped
1058          * in multiples of the page size. For a file that is not a multiple of
1059          * the page size, the remaining memory is zeroed when mapped, and
1060          * writes to that region are not written out to the file."
1061          */
1062         kaddr = kmap_atomic(page);
1063         memset(kaddr + len, 0, PAGE_SIZE - len);
1064         flush_dcache_page(page);
1065         kunmap_atomic(kaddr);
1066 
1067         if (i_size > synced_i_size) {
1068                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1069                 if (err)
1070                         goto out_unlock;
1071         }
1072 
1073         return do_writepage(page, len);
1074 
1075 out_unlock:
1076         unlock_page(page);
1077         return err;
1078 }
1079 
1080 /**
1081  * do_attr_changes - change inode attributes.
1082  * @inode: inode to change attributes for
1083  * @attr: describes attributes to change
1084  */
1085 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1086 {
1087         if (attr->ia_valid & ATTR_UID)
1088                 inode->i_uid = attr->ia_uid;
1089         if (attr->ia_valid & ATTR_GID)
1090                 inode->i_gid = attr->ia_gid;
1091         if (attr->ia_valid & ATTR_ATIME)
1092                 inode->i_atime = timespec_trunc(attr->ia_atime,
1093                                                 inode->i_sb->s_time_gran);
1094         if (attr->ia_valid & ATTR_MTIME)
1095                 inode->i_mtime = timespec_trunc(attr->ia_mtime,
1096                                                 inode->i_sb->s_time_gran);
1097         if (attr->ia_valid & ATTR_CTIME)
1098                 inode->i_ctime = timespec_trunc(attr->ia_ctime,
1099                                                 inode->i_sb->s_time_gran);
1100         if (attr->ia_valid & ATTR_MODE) {
1101                 umode_t mode = attr->ia_mode;
1102 
1103                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1104                         mode &= ~S_ISGID;
1105                 inode->i_mode = mode;
1106         }
1107 }
1108 
1109 /**
1110  * do_truncation - truncate an inode.
1111  * @c: UBIFS file-system description object
1112  * @inode: inode to truncate
1113  * @attr: inode attribute changes description
1114  *
1115  * This function implements VFS '->setattr()' call when the inode is truncated
1116  * to a smaller size. Returns zero in case of success and a negative error code
1117  * in case of failure.
1118  */
1119 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1120                          const struct iattr *attr)
1121 {
1122         int err;
1123         struct ubifs_budget_req req;
1124         loff_t old_size = inode->i_size, new_size = attr->ia_size;
1125         int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1126         struct ubifs_inode *ui = ubifs_inode(inode);
1127 
1128         dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1129         memset(&req, 0, sizeof(struct ubifs_budget_req));
1130 
1131         /*
1132          * If this is truncation to a smaller size, and we do not truncate on a
1133          * block boundary, budget for changing one data block, because the last
1134          * block will be re-written.
1135          */
1136         if (new_size & (UBIFS_BLOCK_SIZE - 1))
1137                 req.dirtied_page = 1;
1138 
1139         req.dirtied_ino = 1;
1140         /* A funny way to budget for truncation node */
1141         req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1142         err = ubifs_budget_space(c, &req);
1143         if (err) {
1144                 /*
1145                  * Treat truncations to zero as deletion and always allow them,
1146                  * just like we do for '->unlink()'.
1147                  */
1148                 if (new_size || err != -ENOSPC)
1149                         return err;
1150                 budgeted = 0;
1151         }
1152 
1153         truncate_setsize(inode, new_size);
1154 
1155         if (offset) {
1156                 pgoff_t index = new_size >> PAGE_SHIFT;
1157                 struct page *page;
1158 
1159                 page = find_lock_page(inode->i_mapping, index);
1160                 if (page) {
1161                         if (PageDirty(page)) {
1162                                 /*
1163                                  * 'ubifs_jnl_truncate()' will try to truncate
1164                                  * the last data node, but it contains
1165                                  * out-of-date data because the page is dirty.
1166                                  * Write the page now, so that
1167                                  * 'ubifs_jnl_truncate()' will see an already
1168                                  * truncated (and up to date) data node.
1169                                  */
1170                                 ubifs_assert(PagePrivate(page));
1171 
1172                                 clear_page_dirty_for_io(page);
1173                                 if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1174                                         offset = new_size &
1175                                                  (PAGE_SIZE - 1);
1176                                 err = do_writepage(page, offset);
1177                                 put_page(page);
1178                                 if (err)
1179                                         goto out_budg;
1180                                 /*
1181                                  * We could now tell 'ubifs_jnl_truncate()' not
1182                                  * to read the last block.
1183                                  */
1184                         } else {
1185                                 /*
1186                                  * We could 'kmap()' the page and pass the data
1187                                  * to 'ubifs_jnl_truncate()' to save it from
1188                                  * having to read it.
1189                                  */
1190                                 unlock_page(page);
1191                                 put_page(page);
1192                         }
1193                 }
1194         }
1195 
1196         mutex_lock(&ui->ui_mutex);
1197         ui->ui_size = inode->i_size;
1198         /* Truncation changes inode [mc]time */
1199         inode->i_mtime = inode->i_ctime = current_time(inode);
1200         /* Other attributes may be changed at the same time as well */
1201         do_attr_changes(inode, attr);
1202         err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1203         mutex_unlock(&ui->ui_mutex);
1204 
1205 out_budg:
1206         if (budgeted)
1207                 ubifs_release_budget(c, &req);
1208         else {
1209                 c->bi.nospace = c->bi.nospace_rp = 0;
1210                 smp_wmb();
1211         }
1212         return err;
1213 }
1214 
1215 /**
1216  * do_setattr - change inode attributes.
1217  * @c: UBIFS file-system description object
1218  * @inode: inode to change attributes for
1219  * @attr: inode attribute changes description
1220  *
1221  * This function implements VFS '->setattr()' call for all cases except
1222  * truncations to smaller size. Returns zero in case of success and a negative
1223  * error code in case of failure.
1224  */
1225 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1226                       const struct iattr *attr)
1227 {
1228         int err, release;
1229         loff_t new_size = attr->ia_size;
1230         struct ubifs_inode *ui = ubifs_inode(inode);
1231         struct ubifs_budget_req req = { .dirtied_ino = 1,
1232                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1233 
1234         err = ubifs_budget_space(c, &req);
1235         if (err)
1236                 return err;
1237 
1238         if (attr->ia_valid & ATTR_SIZE) {
1239                 dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1240                 truncate_setsize(inode, new_size);
1241         }
1242 
1243         mutex_lock(&ui->ui_mutex);
1244         if (attr->ia_valid & ATTR_SIZE) {
1245                 /* Truncation changes inode [mc]time */
1246                 inode->i_mtime = inode->i_ctime = current_time(inode);
1247                 /* 'truncate_setsize()' changed @i_size, update @ui_size */
1248                 ui->ui_size = inode->i_size;
1249         }
1250 
1251         do_attr_changes(inode, attr);
1252 
1253         release = ui->dirty;
1254         if (attr->ia_valid & ATTR_SIZE)
1255                 /*
1256                  * Inode length changed, so we have to make sure
1257                  * @I_DIRTY_DATASYNC is set.
1258                  */
1259                  __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1260         else
1261                 mark_inode_dirty_sync(inode);
1262         mutex_unlock(&ui->ui_mutex);
1263 
1264         if (release)
1265                 ubifs_release_budget(c, &req);
1266         if (IS_SYNC(inode))
1267                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1268         return err;
1269 }
1270 
1271 int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1272 {
1273         int err;
1274         struct inode *inode = d_inode(dentry);
1275         struct ubifs_info *c = inode->i_sb->s_fs_info;
1276 
1277         dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1278                 inode->i_ino, inode->i_mode, attr->ia_valid);
1279         err = setattr_prepare(dentry, attr);
1280         if (err)
1281                 return err;
1282 
1283         err = dbg_check_synced_i_size(c, inode);
1284         if (err)
1285                 return err;
1286 
1287         err = fscrypt_prepare_setattr(dentry, attr);
1288         if (err)
1289                 return err;
1290 
1291         if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1292                 /* Truncation to a smaller size */
1293                 err = do_truncation(c, inode, attr);
1294         else
1295                 err = do_setattr(c, inode, attr);
1296 
1297         return err;
1298 }
1299 
1300 static void ubifs_invalidatepage(struct page *page, unsigned int offset,
1301                                  unsigned int length)
1302 {
1303         struct inode *inode = page->mapping->host;
1304         struct ubifs_info *c = inode->i_sb->s_fs_info;
1305 
1306         ubifs_assert(PagePrivate(page));
1307         if (offset || length < PAGE_SIZE)
1308                 /* Partial page remains dirty */
1309                 return;
1310 
1311         if (PageChecked(page))
1312                 release_new_page_budget(c);
1313         else
1314                 release_existing_page_budget(c);
1315 
1316         atomic_long_dec(&c->dirty_pg_cnt);
1317         ClearPagePrivate(page);
1318         ClearPageChecked(page);
1319 }
1320 
1321 int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1322 {
1323         struct inode *inode = file->f_mapping->host;
1324         struct ubifs_info *c = inode->i_sb->s_fs_info;
1325         int err;
1326 
1327         dbg_gen("syncing inode %lu", inode->i_ino);
1328 
1329         if (c->ro_mount)
1330                 /*
1331                  * For some really strange reasons VFS does not filter out
1332                  * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1333                  */
1334                 return 0;
1335 
1336         err = file_write_and_wait_range(file, start, end);
1337         if (err)
1338                 return err;
1339         inode_lock(inode);
1340 
1341         /* Synchronize the inode unless this is a 'datasync()' call. */
1342         if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1343                 err = inode->i_sb->s_op->write_inode(inode, NULL);
1344                 if (err)
1345                         goto out;
1346         }
1347 
1348         /*
1349          * Nodes related to this inode may still sit in a write-buffer. Flush
1350          * them.
1351          */
1352         err = ubifs_sync_wbufs_by_inode(c, inode);
1353 out:
1354         inode_unlock(inode);
1355         return err;
1356 }
1357 
1358 /**
1359  * mctime_update_needed - check if mtime or ctime update is needed.
1360  * @inode: the inode to do the check for
1361  * @now: current time
1362  *
1363  * This helper function checks if the inode mtime/ctime should be updated or
1364  * not. If current values of the time-stamps are within the UBIFS inode time
1365  * granularity, they are not updated. This is an optimization.
1366  */
1367 static inline int mctime_update_needed(const struct inode *inode,
1368                                        const struct timespec *now)
1369 {
1370         if (!timespec_equal(&inode->i_mtime, now) ||
1371             !timespec_equal(&inode->i_ctime, now))
1372                 return 1;
1373         return 0;
1374 }
1375 
1376 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1377 /**
1378  * ubifs_update_time - update time of inode.
1379  * @inode: inode to update
1380  *
1381  * This function updates time of the inode.
1382  */
1383 int ubifs_update_time(struct inode *inode, struct timespec *time,
1384                              int flags)
1385 {
1386         struct ubifs_inode *ui = ubifs_inode(inode);
1387         struct ubifs_info *c = inode->i_sb->s_fs_info;
1388         struct ubifs_budget_req req = { .dirtied_ino = 1,
1389                         .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1390         int iflags = I_DIRTY_TIME;
1391         int err, release;
1392 
1393         err = ubifs_budget_space(c, &req);
1394         if (err)
1395                 return err;
1396 
1397         mutex_lock(&ui->ui_mutex);
1398         if (flags & S_ATIME)
1399                 inode->i_atime = *time;
1400         if (flags & S_CTIME)
1401                 inode->i_ctime = *time;
1402         if (flags & S_MTIME)
1403                 inode->i_mtime = *time;
1404 
1405         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
1406                 iflags |= I_DIRTY_SYNC;
1407 
1408         release = ui->dirty;
1409         __mark_inode_dirty(inode, iflags);
1410         mutex_unlock(&ui->ui_mutex);
1411         if (release)
1412                 ubifs_release_budget(c, &req);
1413         return 0;
1414 }
1415 #endif
1416 
1417 /**
1418  * update_mctime - update mtime and ctime of an inode.
1419  * @inode: inode to update
1420  *
1421  * This function updates mtime and ctime of the inode if it is not equivalent to
1422  * current time. Returns zero in case of success and a negative error code in
1423  * case of failure.
1424  */
1425 static int update_mctime(struct inode *inode)
1426 {
1427         struct timespec now = current_time(inode);
1428         struct ubifs_inode *ui = ubifs_inode(inode);
1429         struct ubifs_info *c = inode->i_sb->s_fs_info;
1430 
1431         if (mctime_update_needed(inode, &now)) {
1432                 int err, release;
1433                 struct ubifs_budget_req req = { .dirtied_ino = 1,
1434                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1435 
1436                 err = ubifs_budget_space(c, &req);
1437                 if (err)
1438                         return err;
1439 
1440                 mutex_lock(&ui->ui_mutex);
1441                 inode->i_mtime = inode->i_ctime = current_time(inode);
1442                 release = ui->dirty;
1443                 mark_inode_dirty_sync(inode);
1444                 mutex_unlock(&ui->ui_mutex);
1445                 if (release)
1446                         ubifs_release_budget(c, &req);
1447         }
1448 
1449         return 0;
1450 }
1451 
1452 static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1453 {
1454         int err = update_mctime(file_inode(iocb->ki_filp));
1455         if (err)
1456                 return err;
1457 
1458         return generic_file_write_iter(iocb, from);
1459 }
1460 
1461 static int ubifs_set_page_dirty(struct page *page)
1462 {
1463         int ret;
1464 
1465         ret = __set_page_dirty_nobuffers(page);
1466         /*
1467          * An attempt to dirty a page without budgeting for it - should not
1468          * happen.
1469          */
1470         ubifs_assert(ret == 0);
1471         return ret;
1472 }
1473 
1474 #ifdef CONFIG_MIGRATION
1475 static int ubifs_migrate_page(struct address_space *mapping,
1476                 struct page *newpage, struct page *page, enum migrate_mode mode)
1477 {
1478         int rc;
1479 
1480         rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
1481         if (rc != MIGRATEPAGE_SUCCESS)
1482                 return rc;
1483 
1484         if (PagePrivate(page)) {
1485                 ClearPagePrivate(page);
1486                 SetPagePrivate(newpage);
1487         }
1488 
1489         if (mode != MIGRATE_SYNC_NO_COPY)
1490                 migrate_page_copy(newpage, page);
1491         else
1492                 migrate_page_states(newpage, page);
1493         return MIGRATEPAGE_SUCCESS;
1494 }
1495 #endif
1496 
1497 static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1498 {
1499         /*
1500          * An attempt to release a dirty page without budgeting for it - should
1501          * not happen.
1502          */
1503         if (PageWriteback(page))
1504                 return 0;
1505         ubifs_assert(PagePrivate(page));
1506         ubifs_assert(0);
1507         ClearPagePrivate(page);
1508         ClearPageChecked(page);
1509         return 1;
1510 }
1511 
1512 /*
1513  * mmap()d file has taken write protection fault and is being made writable.
1514  * UBIFS must ensure page is budgeted for.
1515  */
1516 static int ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1517 {
1518         struct page *page = vmf->page;
1519         struct inode *inode = file_inode(vmf->vma->vm_file);
1520         struct ubifs_info *c = inode->i_sb->s_fs_info;
1521         struct timespec now = current_time(inode);
1522         struct ubifs_budget_req req = { .new_page = 1 };
1523         int err, update_time;
1524 
1525         dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1526                 i_size_read(inode));
1527         ubifs_assert(!c->ro_media && !c->ro_mount);
1528 
1529         if (unlikely(c->ro_error))
1530                 return VM_FAULT_SIGBUS; /* -EROFS */
1531 
1532         /*
1533          * We have not locked @page so far so we may budget for changing the
1534          * page. Note, we cannot do this after we locked the page, because
1535          * budgeting may cause write-back which would cause deadlock.
1536          *
1537          * At the moment we do not know whether the page is dirty or not, so we
1538          * assume that it is not and budget for a new page. We could look at
1539          * the @PG_private flag and figure this out, but we may race with write
1540          * back and the page state may change by the time we lock it, so this
1541          * would need additional care. We do not bother with this at the
1542          * moment, although it might be good idea to do. Instead, we allocate
1543          * budget for a new page and amend it later on if the page was in fact
1544          * dirty.
1545          *
1546          * The budgeting-related logic of this function is similar to what we
1547          * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1548          * for more comments.
1549          */
1550         update_time = mctime_update_needed(inode, &now);
1551         if (update_time)
1552                 /*
1553                  * We have to change inode time stamp which requires extra
1554                  * budgeting.
1555                  */
1556                 req.dirtied_ino = 1;
1557 
1558         err = ubifs_budget_space(c, &req);
1559         if (unlikely(err)) {
1560                 if (err == -ENOSPC)
1561                         ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1562                                    inode->i_ino);
1563                 return VM_FAULT_SIGBUS;
1564         }
1565 
1566         lock_page(page);
1567         if (unlikely(page->mapping != inode->i_mapping ||
1568                      page_offset(page) > i_size_read(inode))) {
1569                 /* Page got truncated out from underneath us */
1570                 err = -EINVAL;
1571                 goto out_unlock;
1572         }
1573 
1574         if (PagePrivate(page))
1575                 release_new_page_budget(c);
1576         else {
1577                 if (!PageChecked(page))
1578                         ubifs_convert_page_budget(c);
1579                 SetPagePrivate(page);
1580                 atomic_long_inc(&c->dirty_pg_cnt);
1581                 __set_page_dirty_nobuffers(page);
1582         }
1583 
1584         if (update_time) {
1585                 int release;
1586                 struct ubifs_inode *ui = ubifs_inode(inode);
1587 
1588                 mutex_lock(&ui->ui_mutex);
1589                 inode->i_mtime = inode->i_ctime = current_time(inode);
1590                 release = ui->dirty;
1591                 mark_inode_dirty_sync(inode);
1592                 mutex_unlock(&ui->ui_mutex);
1593                 if (release)
1594                         ubifs_release_dirty_inode_budget(c, ui);
1595         }
1596 
1597         wait_for_stable_page(page);
1598         return VM_FAULT_LOCKED;
1599 
1600 out_unlock:
1601         unlock_page(page);
1602         ubifs_release_budget(c, &req);
1603         if (err)
1604                 err = VM_FAULT_SIGBUS;
1605         return err;
1606 }
1607 
1608 static const struct vm_operations_struct ubifs_file_vm_ops = {
1609         .fault        = filemap_fault,
1610         .map_pages = filemap_map_pages,
1611         .page_mkwrite = ubifs_vm_page_mkwrite,
1612 };
1613 
1614 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1615 {
1616         int err;
1617 
1618         err = generic_file_mmap(file, vma);
1619         if (err)
1620                 return err;
1621         vma->vm_ops = &ubifs_file_vm_ops;
1622 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1623         file_accessed(file);
1624 #endif
1625         return 0;
1626 }
1627 
1628 static const char *ubifs_get_link(struct dentry *dentry,
1629                                             struct inode *inode,
1630                                             struct delayed_call *done)
1631 {
1632         struct ubifs_inode *ui = ubifs_inode(inode);
1633 
1634         if (!IS_ENCRYPTED(inode))
1635                 return ui->data;
1636 
1637         if (!dentry)
1638                 return ERR_PTR(-ECHILD);
1639 
1640         return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1641 }
1642 
1643 const struct address_space_operations ubifs_file_address_operations = {
1644         .readpage       = ubifs_readpage,
1645         .writepage      = ubifs_writepage,
1646         .write_begin    = ubifs_write_begin,
1647         .write_end      = ubifs_write_end,
1648         .invalidatepage = ubifs_invalidatepage,
1649         .set_page_dirty = ubifs_set_page_dirty,
1650 #ifdef CONFIG_MIGRATION
1651         .migratepage    = ubifs_migrate_page,
1652 #endif
1653         .releasepage    = ubifs_releasepage,
1654 };
1655 
1656 const struct inode_operations ubifs_file_inode_operations = {
1657         .setattr     = ubifs_setattr,
1658         .getattr     = ubifs_getattr,
1659         .listxattr   = ubifs_listxattr,
1660 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1661         .update_time = ubifs_update_time,
1662 #endif
1663 };
1664 
1665 const struct inode_operations ubifs_symlink_inode_operations = {
1666         .get_link    = ubifs_get_link,
1667         .setattr     = ubifs_setattr,
1668         .getattr     = ubifs_getattr,
1669         .listxattr   = ubifs_listxattr,
1670 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
1671         .update_time = ubifs_update_time,
1672 #endif
1673 };
1674 
1675 const struct file_operations ubifs_file_operations = {
1676         .llseek         = generic_file_llseek,
1677         .read_iter      = generic_file_read_iter,
1678         .write_iter     = ubifs_write_iter,
1679         .mmap           = ubifs_file_mmap,
1680         .fsync          = ubifs_fsync,
1681         .unlocked_ioctl = ubifs_ioctl,
1682         .splice_read    = generic_file_splice_read,
1683         .splice_write   = iter_file_splice_write,
1684         .open           = fscrypt_file_open,
1685 #ifdef CONFIG_COMPAT
1686         .compat_ioctl   = ubifs_compat_ioctl,
1687 #endif
1688 };
1689 

~ [ 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