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

TOMOYO Linux Cross Reference
Linux/fs/gfs2/aops.c

Version: ~ [ linux-6.11-rc3 ] ~ [ linux-6.10.4 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.45 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.104 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.164 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.223 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.281 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.319 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ 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  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
  4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
  5  */
  6 
  7 #include <linux/sched.h>
  8 #include <linux/slab.h>
  9 #include <linux/spinlock.h>
 10 #include <linux/completion.h>
 11 #include <linux/buffer_head.h>
 12 #include <linux/pagemap.h>
 13 #include <linux/pagevec.h>
 14 #include <linux/mpage.h>
 15 #include <linux/fs.h>
 16 #include <linux/writeback.h>
 17 #include <linux/swap.h>
 18 #include <linux/gfs2_ondisk.h>
 19 #include <linux/backing-dev.h>
 20 #include <linux/uio.h>
 21 #include <trace/events/writeback.h>
 22 #include <linux/sched/signal.h>
 23 
 24 #include "gfs2.h"
 25 #include "incore.h"
 26 #include "bmap.h"
 27 #include "glock.h"
 28 #include "inode.h"
 29 #include "log.h"
 30 #include "meta_io.h"
 31 #include "quota.h"
 32 #include "trans.h"
 33 #include "rgrp.h"
 34 #include "super.h"
 35 #include "util.h"
 36 #include "glops.h"
 37 #include "aops.h"
 38 
 39 
 40 void gfs2_trans_add_databufs(struct gfs2_inode *ip, struct folio *folio,
 41                              size_t from, size_t len)
 42 {
 43         struct buffer_head *head = folio_buffers(folio);
 44         unsigned int bsize = head->b_size;
 45         struct buffer_head *bh;
 46         size_t to = from + len;
 47         size_t start, end;
 48 
 49         for (bh = head, start = 0; bh != head || !start;
 50              bh = bh->b_this_page, start = end) {
 51                 end = start + bsize;
 52                 if (end <= from)
 53                         continue;
 54                 if (start >= to)
 55                         break;
 56                 set_buffer_uptodate(bh);
 57                 gfs2_trans_add_data(ip->i_gl, bh);
 58         }
 59 }
 60 
 61 /**
 62  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
 63  * @inode: The inode
 64  * @lblock: The block number to look up
 65  * @bh_result: The buffer head to return the result in
 66  * @create: Non-zero if we may add block to the file
 67  *
 68  * Returns: errno
 69  */
 70 
 71 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
 72                                   struct buffer_head *bh_result, int create)
 73 {
 74         int error;
 75 
 76         error = gfs2_block_map(inode, lblock, bh_result, 0);
 77         if (error)
 78                 return error;
 79         if (!buffer_mapped(bh_result))
 80                 return -ENODATA;
 81         return 0;
 82 }
 83 
 84 /**
 85  * gfs2_write_jdata_folio - gfs2 jdata-specific version of block_write_full_folio
 86  * @folio: The folio to write
 87  * @wbc: The writeback control
 88  *
 89  * This is the same as calling block_write_full_folio, but it also
 90  * writes pages outside of i_size
 91  */
 92 static int gfs2_write_jdata_folio(struct folio *folio,
 93                                  struct writeback_control *wbc)
 94 {
 95         struct inode * const inode = folio->mapping->host;
 96         loff_t i_size = i_size_read(inode);
 97 
 98         /*
 99          * The folio straddles i_size.  It must be zeroed out on each and every
100          * writepage invocation because it may be mmapped.  "A file is mapped
101          * in multiples of the page size.  For a file that is not a multiple of
102          * the page size, the remaining memory is zeroed when mapped, and
103          * writes to that region are not written out to the file."
104          */
105         if (folio_pos(folio) < i_size &&
106             i_size < folio_pos(folio) + folio_size(folio))
107                 folio_zero_segment(folio, offset_in_folio(folio, i_size),
108                                 folio_size(folio));
109 
110         return __block_write_full_folio(inode, folio, gfs2_get_block_noalloc,
111                         wbc);
112 }
113 
114 /**
115  * __gfs2_jdata_write_folio - The core of jdata writepage
116  * @folio: The folio to write
117  * @wbc: The writeback control
118  *
119  * Implements the core of write back. If a transaction is required then
120  * the checked flag will have been set and the transaction will have
121  * already been started before this is called.
122  */
123 static int __gfs2_jdata_write_folio(struct folio *folio,
124                 struct writeback_control *wbc)
125 {
126         struct inode *inode = folio->mapping->host;
127         struct gfs2_inode *ip = GFS2_I(inode);
128 
129         if (folio_test_checked(folio)) {
130                 folio_clear_checked(folio);
131                 if (!folio_buffers(folio)) {
132                         create_empty_buffers(folio,
133                                         inode->i_sb->s_blocksize,
134                                         BIT(BH_Dirty)|BIT(BH_Uptodate));
135                 }
136                 gfs2_trans_add_databufs(ip, folio, 0, folio_size(folio));
137         }
138         return gfs2_write_jdata_folio(folio, wbc);
139 }
140 
141 /**
142  * gfs2_jdata_writepage - Write complete page
143  * @page: Page to write
144  * @wbc: The writeback control
145  *
146  * Returns: errno
147  *
148  */
149 
150 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
151 {
152         struct folio *folio = page_folio(page);
153         struct inode *inode = page->mapping->host;
154         struct gfs2_inode *ip = GFS2_I(inode);
155         struct gfs2_sbd *sdp = GFS2_SB(inode);
156 
157         if (gfs2_assert_withdraw(sdp, ip->i_gl->gl_state == LM_ST_EXCLUSIVE))
158                 goto out;
159         if (folio_test_checked(folio) || current->journal_info)
160                 goto out_ignore;
161         return __gfs2_jdata_write_folio(folio, wbc);
162 
163 out_ignore:
164         folio_redirty_for_writepage(wbc, folio);
165 out:
166         folio_unlock(folio);
167         return 0;
168 }
169 
170 /**
171  * gfs2_writepages - Write a bunch of dirty pages back to disk
172  * @mapping: The mapping to write
173  * @wbc: Write-back control
174  *
175  * Used for both ordered and writeback modes.
176  */
177 static int gfs2_writepages(struct address_space *mapping,
178                            struct writeback_control *wbc)
179 {
180         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
181         struct iomap_writepage_ctx wpc = { };
182         int ret;
183 
184         /*
185          * Even if we didn't write enough pages here, we might still be holding
186          * dirty pages in the ail. We forcibly flush the ail because we don't
187          * want balance_dirty_pages() to loop indefinitely trying to write out
188          * pages held in the ail that it can't find.
189          */
190         ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
191         if (ret == 0 && wbc->nr_to_write > 0)
192                 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
193         return ret;
194 }
195 
196 /**
197  * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
198  * @mapping: The mapping
199  * @wbc: The writeback control
200  * @fbatch: The batch of folios
201  * @done_index: Page index
202  *
203  * Returns: non-zero if loop should terminate, zero otherwise
204  */
205 
206 static int gfs2_write_jdata_batch(struct address_space *mapping,
207                                     struct writeback_control *wbc,
208                                     struct folio_batch *fbatch,
209                                     pgoff_t *done_index)
210 {
211         struct inode *inode = mapping->host;
212         struct gfs2_sbd *sdp = GFS2_SB(inode);
213         unsigned nrblocks;
214         int i;
215         int ret;
216         size_t size = 0;
217         int nr_folios = folio_batch_count(fbatch);
218 
219         for (i = 0; i < nr_folios; i++)
220                 size += folio_size(fbatch->folios[i]);
221         nrblocks = size >> inode->i_blkbits;
222 
223         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
224         if (ret < 0)
225                 return ret;
226 
227         for (i = 0; i < nr_folios; i++) {
228                 struct folio *folio = fbatch->folios[i];
229 
230                 *done_index = folio->index;
231 
232                 folio_lock(folio);
233 
234                 if (unlikely(folio->mapping != mapping)) {
235 continue_unlock:
236                         folio_unlock(folio);
237                         continue;
238                 }
239 
240                 if (!folio_test_dirty(folio)) {
241                         /* someone wrote it for us */
242                         goto continue_unlock;
243                 }
244 
245                 if (folio_test_writeback(folio)) {
246                         if (wbc->sync_mode != WB_SYNC_NONE)
247                                 folio_wait_writeback(folio);
248                         else
249                                 goto continue_unlock;
250                 }
251 
252                 BUG_ON(folio_test_writeback(folio));
253                 if (!folio_clear_dirty_for_io(folio))
254                         goto continue_unlock;
255 
256                 trace_wbc_writepage(wbc, inode_to_bdi(inode));
257 
258                 ret = __gfs2_jdata_write_folio(folio, wbc);
259                 if (unlikely(ret)) {
260                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
261                                 folio_unlock(folio);
262                                 ret = 0;
263                         } else {
264 
265                                 /*
266                                  * done_index is set past this page,
267                                  * so media errors will not choke
268                                  * background writeout for the entire
269                                  * file. This has consequences for
270                                  * range_cyclic semantics (ie. it may
271                                  * not be suitable for data integrity
272                                  * writeout).
273                                  */
274                                 *done_index = folio_next_index(folio);
275                                 ret = 1;
276                                 break;
277                         }
278                 }
279 
280                 /*
281                  * We stop writing back only if we are not doing
282                  * integrity sync. In case of integrity sync we have to
283                  * keep going until we have written all the pages
284                  * we tagged for writeback prior to entering this loop.
285                  */
286                 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
287                         ret = 1;
288                         break;
289                 }
290 
291         }
292         gfs2_trans_end(sdp);
293         return ret;
294 }
295 
296 /**
297  * gfs2_write_cache_jdata - Like write_cache_pages but different
298  * @mapping: The mapping to write
299  * @wbc: The writeback control
300  *
301  * The reason that we use our own function here is that we need to
302  * start transactions before we grab page locks. This allows us
303  * to get the ordering right.
304  */
305 
306 static int gfs2_write_cache_jdata(struct address_space *mapping,
307                                   struct writeback_control *wbc)
308 {
309         int ret = 0;
310         int done = 0;
311         struct folio_batch fbatch;
312         int nr_folios;
313         pgoff_t writeback_index;
314         pgoff_t index;
315         pgoff_t end;
316         pgoff_t done_index;
317         int cycled;
318         int range_whole = 0;
319         xa_mark_t tag;
320 
321         folio_batch_init(&fbatch);
322         if (wbc->range_cyclic) {
323                 writeback_index = mapping->writeback_index; /* prev offset */
324                 index = writeback_index;
325                 if (index == 0)
326                         cycled = 1;
327                 else
328                         cycled = 0;
329                 end = -1;
330         } else {
331                 index = wbc->range_start >> PAGE_SHIFT;
332                 end = wbc->range_end >> PAGE_SHIFT;
333                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
334                         range_whole = 1;
335                 cycled = 1; /* ignore range_cyclic tests */
336         }
337         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
338                 tag = PAGECACHE_TAG_TOWRITE;
339         else
340                 tag = PAGECACHE_TAG_DIRTY;
341 
342 retry:
343         if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
344                 tag_pages_for_writeback(mapping, index, end);
345         done_index = index;
346         while (!done && (index <= end)) {
347                 nr_folios = filemap_get_folios_tag(mapping, &index, end,
348                                 tag, &fbatch);
349                 if (nr_folios == 0)
350                         break;
351 
352                 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
353                                 &done_index);
354                 if (ret)
355                         done = 1;
356                 if (ret > 0)
357                         ret = 0;
358                 folio_batch_release(&fbatch);
359                 cond_resched();
360         }
361 
362         if (!cycled && !done) {
363                 /*
364                  * range_cyclic:
365                  * We hit the last page and there is more work to be done: wrap
366                  * back to the start of the file
367                  */
368                 cycled = 1;
369                 index = 0;
370                 end = writeback_index - 1;
371                 goto retry;
372         }
373 
374         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
375                 mapping->writeback_index = done_index;
376 
377         return ret;
378 }
379 
380 
381 /**
382  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
383  * @mapping: The mapping to write
384  * @wbc: The writeback control
385  * 
386  */
387 
388 static int gfs2_jdata_writepages(struct address_space *mapping,
389                                  struct writeback_control *wbc)
390 {
391         struct gfs2_inode *ip = GFS2_I(mapping->host);
392         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
393         int ret;
394 
395         ret = gfs2_write_cache_jdata(mapping, wbc);
396         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
397                 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
398                                GFS2_LFC_JDATA_WPAGES);
399                 ret = gfs2_write_cache_jdata(mapping, wbc);
400         }
401         return ret;
402 }
403 
404 /**
405  * stuffed_read_folio - Fill in a Linux folio with stuffed file data
406  * @ip: the inode
407  * @folio: the folio
408  *
409  * Returns: errno
410  */
411 static int stuffed_read_folio(struct gfs2_inode *ip, struct folio *folio)
412 {
413         struct buffer_head *dibh = NULL;
414         size_t dsize = i_size_read(&ip->i_inode);
415         void *from = NULL;
416         int error = 0;
417 
418         /*
419          * Due to the order of unstuffing files and ->fault(), we can be
420          * asked for a zero folio in the case of a stuffed file being extended,
421          * so we need to supply one here. It doesn't happen often.
422          */
423         if (unlikely(folio->index)) {
424                 dsize = 0;
425         } else {
426                 error = gfs2_meta_inode_buffer(ip, &dibh);
427                 if (error)
428                         goto out;
429                 from = dibh->b_data + sizeof(struct gfs2_dinode);
430         }
431 
432         folio_fill_tail(folio, 0, from, dsize);
433         brelse(dibh);
434 out:
435         folio_end_read(folio, error == 0);
436 
437         return error;
438 }
439 
440 /**
441  * gfs2_read_folio - read a folio from a file
442  * @file: The file to read
443  * @folio: The folio in the file
444  */
445 static int gfs2_read_folio(struct file *file, struct folio *folio)
446 {
447         struct inode *inode = folio->mapping->host;
448         struct gfs2_inode *ip = GFS2_I(inode);
449         struct gfs2_sbd *sdp = GFS2_SB(inode);
450         int error;
451 
452         if (!gfs2_is_jdata(ip) ||
453             (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
454                 error = iomap_read_folio(folio, &gfs2_iomap_ops);
455         } else if (gfs2_is_stuffed(ip)) {
456                 error = stuffed_read_folio(ip, folio);
457         } else {
458                 error = mpage_read_folio(folio, gfs2_block_map);
459         }
460 
461         if (gfs2_withdrawing_or_withdrawn(sdp))
462                 return -EIO;
463 
464         return error;
465 }
466 
467 /**
468  * gfs2_internal_read - read an internal file
469  * @ip: The gfs2 inode
470  * @buf: The buffer to fill
471  * @pos: The file position
472  * @size: The amount to read
473  *
474  */
475 
476 ssize_t gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
477                            size_t size)
478 {
479         struct address_space *mapping = ip->i_inode.i_mapping;
480         unsigned long index = *pos >> PAGE_SHIFT;
481         size_t copied = 0;
482 
483         do {
484                 size_t offset, chunk;
485                 struct folio *folio;
486 
487                 folio = read_cache_folio(mapping, index, gfs2_read_folio, NULL);
488                 if (IS_ERR(folio)) {
489                         if (PTR_ERR(folio) == -EINTR)
490                                 continue;
491                         return PTR_ERR(folio);
492                 }
493                 offset = *pos + copied - folio_pos(folio);
494                 chunk = min(size - copied, folio_size(folio) - offset);
495                 memcpy_from_folio(buf + copied, folio, offset, chunk);
496                 index = folio_next_index(folio);
497                 folio_put(folio);
498                 copied += chunk;
499         } while(copied < size);
500         (*pos) += size;
501         return size;
502 }
503 
504 /**
505  * gfs2_readahead - Read a bunch of pages at once
506  * @rac: Read-ahead control structure
507  *
508  * Some notes:
509  * 1. This is only for readahead, so we can simply ignore any things
510  *    which are slightly inconvenient (such as locking conflicts between
511  *    the page lock and the glock) and return having done no I/O. Its
512  *    obviously not something we'd want to do on too regular a basis.
513  *    Any I/O we ignore at this time will be done via readpage later.
514  * 2. We don't handle stuffed files here we let readpage do the honours.
515  * 3. mpage_readahead() does most of the heavy lifting in the common case.
516  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
517  */
518 
519 static void gfs2_readahead(struct readahead_control *rac)
520 {
521         struct inode *inode = rac->mapping->host;
522         struct gfs2_inode *ip = GFS2_I(inode);
523 
524         if (gfs2_is_stuffed(ip))
525                 ;
526         else if (gfs2_is_jdata(ip))
527                 mpage_readahead(rac, gfs2_block_map);
528         else
529                 iomap_readahead(rac, &gfs2_iomap_ops);
530 }
531 
532 /**
533  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
534  * @inode: the rindex inode
535  */
536 void adjust_fs_space(struct inode *inode)
537 {
538         struct gfs2_sbd *sdp = GFS2_SB(inode);
539         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
540         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
541         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
542         struct buffer_head *m_bh;
543         u64 fs_total, new_free;
544 
545         if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
546                 return;
547 
548         /* Total up the file system space, according to the latest rindex. */
549         fs_total = gfs2_ri_total(sdp);
550         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
551                 goto out;
552 
553         spin_lock(&sdp->sd_statfs_spin);
554         gfs2_statfs_change_in(m_sc, m_bh->b_data +
555                               sizeof(struct gfs2_dinode));
556         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
557                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
558         else
559                 new_free = 0;
560         spin_unlock(&sdp->sd_statfs_spin);
561         fs_warn(sdp, "File system extended by %llu blocks.\n",
562                 (unsigned long long)new_free);
563         gfs2_statfs_change(sdp, new_free, new_free, 0);
564 
565         update_statfs(sdp, m_bh);
566         brelse(m_bh);
567 out:
568         sdp->sd_rindex_uptodate = 0;
569         gfs2_trans_end(sdp);
570 }
571 
572 static bool jdata_dirty_folio(struct address_space *mapping,
573                 struct folio *folio)
574 {
575         if (current->journal_info)
576                 folio_set_checked(folio);
577         return block_dirty_folio(mapping, folio);
578 }
579 
580 /**
581  * gfs2_bmap - Block map function
582  * @mapping: Address space info
583  * @lblock: The block to map
584  *
585  * Returns: The disk address for the block or 0 on hole or error
586  */
587 
588 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
589 {
590         struct gfs2_inode *ip = GFS2_I(mapping->host);
591         struct gfs2_holder i_gh;
592         sector_t dblock = 0;
593         int error;
594 
595         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
596         if (error)
597                 return 0;
598 
599         if (!gfs2_is_stuffed(ip))
600                 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
601 
602         gfs2_glock_dq_uninit(&i_gh);
603 
604         return dblock;
605 }
606 
607 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
608 {
609         struct gfs2_bufdata *bd;
610 
611         lock_buffer(bh);
612         gfs2_log_lock(sdp);
613         clear_buffer_dirty(bh);
614         bd = bh->b_private;
615         if (bd) {
616                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
617                         list_del_init(&bd->bd_list);
618                 else {
619                         spin_lock(&sdp->sd_ail_lock);
620                         gfs2_remove_from_journal(bh, REMOVE_JDATA);
621                         spin_unlock(&sdp->sd_ail_lock);
622                 }
623         }
624         bh->b_bdev = NULL;
625         clear_buffer_mapped(bh);
626         clear_buffer_req(bh);
627         clear_buffer_new(bh);
628         gfs2_log_unlock(sdp);
629         unlock_buffer(bh);
630 }
631 
632 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
633                                 size_t length)
634 {
635         struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
636         size_t stop = offset + length;
637         int partial_page = (offset || length < folio_size(folio));
638         struct buffer_head *bh, *head;
639         unsigned long pos = 0;
640 
641         BUG_ON(!folio_test_locked(folio));
642         if (!partial_page)
643                 folio_clear_checked(folio);
644         head = folio_buffers(folio);
645         if (!head)
646                 goto out;
647 
648         bh = head;
649         do {
650                 if (pos + bh->b_size > stop)
651                         return;
652 
653                 if (offset <= pos)
654                         gfs2_discard(sdp, bh);
655                 pos += bh->b_size;
656                 bh = bh->b_this_page;
657         } while (bh != head);
658 out:
659         if (!partial_page)
660                 filemap_release_folio(folio, 0);
661 }
662 
663 /**
664  * gfs2_release_folio - free the metadata associated with a folio
665  * @folio: the folio that's being released
666  * @gfp_mask: passed from Linux VFS, ignored by us
667  *
668  * Calls try_to_free_buffers() to free the buffers and put the folio if the
669  * buffers can be released.
670  *
671  * Returns: true if the folio was put or else false
672  */
673 
674 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
675 {
676         struct address_space *mapping = folio->mapping;
677         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
678         struct buffer_head *bh, *head;
679         struct gfs2_bufdata *bd;
680 
681         head = folio_buffers(folio);
682         if (!head)
683                 return false;
684 
685         /*
686          * mm accommodates an old ext3 case where clean folios might
687          * not have had the dirty bit cleared.  Thus, it can send actual
688          * dirty folios to ->release_folio() via shrink_active_list().
689          *
690          * As a workaround, we skip folios that contain dirty buffers
691          * below.  Once ->release_folio isn't called on dirty folios
692          * anymore, we can warn on dirty buffers like we used to here
693          * again.
694          */
695 
696         gfs2_log_lock(sdp);
697         bh = head;
698         do {
699                 if (atomic_read(&bh->b_count))
700                         goto cannot_release;
701                 bd = bh->b_private;
702                 if (bd && bd->bd_tr)
703                         goto cannot_release;
704                 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
705                         goto cannot_release;
706                 bh = bh->b_this_page;
707         } while (bh != head);
708 
709         bh = head;
710         do {
711                 bd = bh->b_private;
712                 if (bd) {
713                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
714                         bd->bd_bh = NULL;
715                         bh->b_private = NULL;
716                         /*
717                          * The bd may still be queued as a revoke, in which
718                          * case we must not dequeue nor free it.
719                          */
720                         if (!bd->bd_blkno && !list_empty(&bd->bd_list))
721                                 list_del_init(&bd->bd_list);
722                         if (list_empty(&bd->bd_list))
723                                 kmem_cache_free(gfs2_bufdata_cachep, bd);
724                 }
725 
726                 bh = bh->b_this_page;
727         } while (bh != head);
728         gfs2_log_unlock(sdp);
729 
730         return try_to_free_buffers(folio);
731 
732 cannot_release:
733         gfs2_log_unlock(sdp);
734         return false;
735 }
736 
737 static const struct address_space_operations gfs2_aops = {
738         .writepages = gfs2_writepages,
739         .read_folio = gfs2_read_folio,
740         .readahead = gfs2_readahead,
741         .dirty_folio = iomap_dirty_folio,
742         .release_folio = iomap_release_folio,
743         .invalidate_folio = iomap_invalidate_folio,
744         .bmap = gfs2_bmap,
745         .migrate_folio = filemap_migrate_folio,
746         .is_partially_uptodate = iomap_is_partially_uptodate,
747         .error_remove_folio = generic_error_remove_folio,
748 };
749 
750 static const struct address_space_operations gfs2_jdata_aops = {
751         .writepage = gfs2_jdata_writepage,
752         .writepages = gfs2_jdata_writepages,
753         .read_folio = gfs2_read_folio,
754         .readahead = gfs2_readahead,
755         .dirty_folio = jdata_dirty_folio,
756         .bmap = gfs2_bmap,
757         .migrate_folio = buffer_migrate_folio,
758         .invalidate_folio = gfs2_invalidate_folio,
759         .release_folio = gfs2_release_folio,
760         .is_partially_uptodate = block_is_partially_uptodate,
761         .error_remove_folio = generic_error_remove_folio,
762 };
763 
764 void gfs2_set_aops(struct inode *inode)
765 {
766         if (gfs2_is_jdata(GFS2_I(inode)))
767                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
768         else
769                 inode->i_mapping->a_ops = &gfs2_aops;
770 }
771 

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

kernel.org | git.kernel.org | LWN.net | Project Home | 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.

sflogo.php