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
gfs2_trans_add_databufs(struct gfs2_inode * ip,struct folio * folio,size_t from,size_t len)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
gfs2_get_block_noalloc(struct inode * inode,sector_t lblock,struct buffer_head * bh_result,int create)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_page
86 * @folio: The folio to write
87 * @wbc: The writeback control
88 *
89 * This is the same as calling block_write_full_page, but it also
90 * writes pages outside of i_size
91 */
gfs2_write_jdata_folio(struct folio * folio,struct writeback_control * wbc)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, end_buffer_async_write);
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 * This is shared between writepage and writepages and implements the
120 * core of the writepage operation. If a transaction is required then
121 * the checked flag will have been set and the transaction will have
122 * already been started before this is called.
123 */
__gfs2_jdata_write_folio(struct folio * folio,struct writeback_control * wbc)124 static int __gfs2_jdata_write_folio(struct folio *folio,
125 struct writeback_control *wbc)
126 {
127 struct inode *inode = folio->mapping->host;
128 struct gfs2_inode *ip = GFS2_I(inode);
129
130 if (folio_test_checked(folio)) {
131 folio_clear_checked(folio);
132 if (!folio_buffers(folio)) {
133 folio_create_empty_buffers(folio,
134 inode->i_sb->s_blocksize,
135 BIT(BH_Dirty)|BIT(BH_Uptodate));
136 }
137 gfs2_trans_add_databufs(ip, folio, 0, folio_size(folio));
138 }
139 return gfs2_write_jdata_folio(folio, wbc);
140 }
141
142 /**
143 * gfs2_jdata_writepage - Write complete page
144 * @page: Page to write
145 * @wbc: The writeback control
146 *
147 * Returns: errno
148 *
149 */
150
gfs2_jdata_writepage(struct page * page,struct writeback_control * wbc)151 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
152 {
153 struct folio *folio = page_folio(page);
154 struct inode *inode = page->mapping->host;
155 struct gfs2_inode *ip = GFS2_I(inode);
156 struct gfs2_sbd *sdp = GFS2_SB(inode);
157
158 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
159 goto out;
160 if (folio_test_checked(folio) || current->journal_info)
161 goto out_ignore;
162 return __gfs2_jdata_write_folio(folio, wbc);
163
164 out_ignore:
165 folio_redirty_for_writepage(wbc, folio);
166 out:
167 folio_unlock(folio);
168 return 0;
169 }
170
171 /**
172 * gfs2_writepages - Write a bunch of dirty pages back to disk
173 * @mapping: The mapping to write
174 * @wbc: Write-back control
175 *
176 * Used for both ordered and writeback modes.
177 */
gfs2_writepages(struct address_space * mapping,struct writeback_control * wbc)178 static int gfs2_writepages(struct address_space *mapping,
179 struct writeback_control *wbc)
180 {
181 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
182 struct iomap_writepage_ctx wpc = { };
183 int ret;
184
185 /*
186 * Even if we didn't write enough pages here, we might still be holding
187 * dirty pages in the ail. We forcibly flush the ail because we don't
188 * want balance_dirty_pages() to loop indefinitely trying to write out
189 * pages held in the ail that it can't find.
190 */
191 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
192 if (ret == 0 && wbc->nr_to_write > 0)
193 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
194 return ret;
195 }
196
197 /**
198 * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
199 * @mapping: The mapping
200 * @wbc: The writeback control
201 * @fbatch: The batch of folios
202 * @done_index: Page index
203 *
204 * Returns: non-zero if loop should terminate, zero otherwise
205 */
206
gfs2_write_jdata_batch(struct address_space * mapping,struct writeback_control * wbc,struct folio_batch * fbatch,pgoff_t * done_index)207 static int gfs2_write_jdata_batch(struct address_space *mapping,
208 struct writeback_control *wbc,
209 struct folio_batch *fbatch,
210 pgoff_t *done_index)
211 {
212 struct inode *inode = mapping->host;
213 struct gfs2_sbd *sdp = GFS2_SB(inode);
214 unsigned nrblocks;
215 int i;
216 int ret;
217 int nr_pages = 0;
218 int nr_folios = folio_batch_count(fbatch);
219
220 for (i = 0; i < nr_folios; i++)
221 nr_pages += folio_nr_pages(fbatch->folios[i]);
222 nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
223
224 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
225 if (ret < 0)
226 return ret;
227
228 for (i = 0; i < nr_folios; i++) {
229 struct folio *folio = fbatch->folios[i];
230
231 *done_index = folio->index;
232
233 folio_lock(folio);
234
235 if (unlikely(folio->mapping != mapping)) {
236 continue_unlock:
237 folio_unlock(folio);
238 continue;
239 }
240
241 if (!folio_test_dirty(folio)) {
242 /* someone wrote it for us */
243 goto continue_unlock;
244 }
245
246 if (folio_test_writeback(folio)) {
247 if (wbc->sync_mode != WB_SYNC_NONE)
248 folio_wait_writeback(folio);
249 else
250 goto continue_unlock;
251 }
252
253 BUG_ON(folio_test_writeback(folio));
254 if (!folio_clear_dirty_for_io(folio))
255 goto continue_unlock;
256
257 trace_wbc_writepage(wbc, inode_to_bdi(inode));
258
259 ret = __gfs2_jdata_write_folio(folio, wbc);
260 if (unlikely(ret)) {
261 if (ret == AOP_WRITEPAGE_ACTIVATE) {
262 folio_unlock(folio);
263 ret = 0;
264 } else {
265
266 /*
267 * done_index is set past this page,
268 * so media errors will not choke
269 * background writeout for the entire
270 * file. This has consequences for
271 * range_cyclic semantics (ie. it may
272 * not be suitable for data integrity
273 * writeout).
274 */
275 *done_index = folio_next_index(folio);
276 ret = 1;
277 break;
278 }
279 }
280
281 /*
282 * We stop writing back only if we are not doing
283 * integrity sync. In case of integrity sync we have to
284 * keep going until we have written all the pages
285 * we tagged for writeback prior to entering this loop.
286 */
287 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
288 ret = 1;
289 break;
290 }
291
292 }
293 gfs2_trans_end(sdp);
294 return ret;
295 }
296
297 /**
298 * gfs2_write_cache_jdata - Like write_cache_pages but different
299 * @mapping: The mapping to write
300 * @wbc: The writeback control
301 *
302 * The reason that we use our own function here is that we need to
303 * start transactions before we grab page locks. This allows us
304 * to get the ordering right.
305 */
306
gfs2_write_cache_jdata(struct address_space * mapping,struct writeback_control * wbc)307 static int gfs2_write_cache_jdata(struct address_space *mapping,
308 struct writeback_control *wbc)
309 {
310 int ret = 0;
311 int done = 0;
312 struct folio_batch fbatch;
313 int nr_folios;
314 pgoff_t writeback_index;
315 pgoff_t index;
316 pgoff_t end;
317 pgoff_t done_index;
318 int cycled;
319 int range_whole = 0;
320 xa_mark_t tag;
321
322 folio_batch_init(&fbatch);
323 if (wbc->range_cyclic) {
324 writeback_index = mapping->writeback_index; /* prev offset */
325 index = writeback_index;
326 if (index == 0)
327 cycled = 1;
328 else
329 cycled = 0;
330 end = -1;
331 } else {
332 index = wbc->range_start >> PAGE_SHIFT;
333 end = wbc->range_end >> PAGE_SHIFT;
334 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
335 range_whole = 1;
336 cycled = 1; /* ignore range_cyclic tests */
337 }
338 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
339 tag = PAGECACHE_TAG_TOWRITE;
340 else
341 tag = PAGECACHE_TAG_DIRTY;
342
343 retry:
344 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
345 tag_pages_for_writeback(mapping, index, end);
346 done_index = index;
347 while (!done && (index <= end)) {
348 nr_folios = filemap_get_folios_tag(mapping, &index, end,
349 tag, &fbatch);
350 if (nr_folios == 0)
351 break;
352
353 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
354 &done_index);
355 if (ret)
356 done = 1;
357 if (ret > 0)
358 ret = 0;
359 folio_batch_release(&fbatch);
360 cond_resched();
361 }
362
363 if (!cycled && !done) {
364 /*
365 * range_cyclic:
366 * We hit the last page and there is more work to be done: wrap
367 * back to the start of the file
368 */
369 cycled = 1;
370 index = 0;
371 end = writeback_index - 1;
372 goto retry;
373 }
374
375 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
376 mapping->writeback_index = done_index;
377
378 return ret;
379 }
380
381
382 /**
383 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
384 * @mapping: The mapping to write
385 * @wbc: The writeback control
386 *
387 */
388
gfs2_jdata_writepages(struct address_space * mapping,struct writeback_control * wbc)389 static int gfs2_jdata_writepages(struct address_space *mapping,
390 struct writeback_control *wbc)
391 {
392 struct gfs2_inode *ip = GFS2_I(mapping->host);
393 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
394 int ret;
395
396 ret = gfs2_write_cache_jdata(mapping, wbc);
397 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
398 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
399 GFS2_LFC_JDATA_WPAGES);
400 ret = gfs2_write_cache_jdata(mapping, wbc);
401 }
402 return ret;
403 }
404
405 /**
406 * stuffed_readpage - Fill in a Linux page with stuffed file data
407 * @ip: the inode
408 * @page: the page
409 *
410 * Returns: errno
411 */
stuffed_readpage(struct gfs2_inode * ip,struct page * page)412 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
413 {
414 struct buffer_head *dibh;
415 u64 dsize = i_size_read(&ip->i_inode);
416 void *kaddr;
417 int error;
418
419 /*
420 * Due to the order of unstuffing files and ->fault(), we can be
421 * asked for a zero page in the case of a stuffed file being extended,
422 * so we need to supply one here. It doesn't happen often.
423 */
424 if (unlikely(page->index)) {
425 zero_user(page, 0, PAGE_SIZE);
426 SetPageUptodate(page);
427 return 0;
428 }
429
430 error = gfs2_meta_inode_buffer(ip, &dibh);
431 if (error)
432 return error;
433
434 kaddr = kmap_local_page(page);
435 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
436 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
437 kunmap_local(kaddr);
438 flush_dcache_page(page);
439 brelse(dibh);
440 SetPageUptodate(page);
441
442 return 0;
443 }
444
445 /**
446 * gfs2_read_folio - read a folio from a file
447 * @file: The file to read
448 * @folio: The folio in the file
449 */
gfs2_read_folio(struct file * file,struct folio * folio)450 static int gfs2_read_folio(struct file *file, struct folio *folio)
451 {
452 struct inode *inode = folio->mapping->host;
453 struct gfs2_inode *ip = GFS2_I(inode);
454 struct gfs2_sbd *sdp = GFS2_SB(inode);
455 int error;
456
457 if (!gfs2_is_jdata(ip) ||
458 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
459 error = iomap_read_folio(folio, &gfs2_iomap_ops);
460 } else if (gfs2_is_stuffed(ip)) {
461 error = stuffed_readpage(ip, &folio->page);
462 folio_unlock(folio);
463 } else {
464 error = mpage_read_folio(folio, gfs2_block_map);
465 }
466
467 if (gfs2_withdrawing_or_withdrawn(sdp))
468 return -EIO;
469
470 return error;
471 }
472
473 /**
474 * gfs2_internal_read - read an internal file
475 * @ip: The gfs2 inode
476 * @buf: The buffer to fill
477 * @pos: The file position
478 * @size: The amount to read
479 *
480 */
481
gfs2_internal_read(struct gfs2_inode * ip,char * buf,loff_t * pos,size_t size)482 ssize_t gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
483 size_t size)
484 {
485 struct address_space *mapping = ip->i_inode.i_mapping;
486 unsigned long index = *pos >> PAGE_SHIFT;
487 size_t copied = 0;
488
489 do {
490 size_t offset, chunk;
491 struct folio *folio;
492
493 folio = read_cache_folio(mapping, index, gfs2_read_folio, NULL);
494 if (IS_ERR(folio)) {
495 if (PTR_ERR(folio) == -EINTR)
496 continue;
497 return PTR_ERR(folio);
498 }
499 offset = *pos + copied - folio_pos(folio);
500 chunk = min(size - copied, folio_size(folio) - offset);
501 memcpy_from_folio(buf + copied, folio, offset, chunk);
502 index = folio_next_index(folio);
503 folio_put(folio);
504 copied += chunk;
505 } while(copied < size);
506 (*pos) += size;
507 return size;
508 }
509
510 /**
511 * gfs2_readahead - Read a bunch of pages at once
512 * @rac: Read-ahead control structure
513 *
514 * Some notes:
515 * 1. This is only for readahead, so we can simply ignore any things
516 * which are slightly inconvenient (such as locking conflicts between
517 * the page lock and the glock) and return having done no I/O. Its
518 * obviously not something we'd want to do on too regular a basis.
519 * Any I/O we ignore at this time will be done via readpage later.
520 * 2. We don't handle stuffed files here we let readpage do the honours.
521 * 3. mpage_readahead() does most of the heavy lifting in the common case.
522 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
523 */
524
gfs2_readahead(struct readahead_control * rac)525 static void gfs2_readahead(struct readahead_control *rac)
526 {
527 struct inode *inode = rac->mapping->host;
528 struct gfs2_inode *ip = GFS2_I(inode);
529
530 if (gfs2_is_stuffed(ip))
531 ;
532 else if (gfs2_is_jdata(ip))
533 mpage_readahead(rac, gfs2_block_map);
534 else
535 iomap_readahead(rac, &gfs2_iomap_ops);
536 }
537
538 /**
539 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
540 * @inode: the rindex inode
541 */
adjust_fs_space(struct inode * inode)542 void adjust_fs_space(struct inode *inode)
543 {
544 struct gfs2_sbd *sdp = GFS2_SB(inode);
545 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
546 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
547 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
548 struct buffer_head *m_bh;
549 u64 fs_total, new_free;
550
551 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
552 return;
553
554 /* Total up the file system space, according to the latest rindex. */
555 fs_total = gfs2_ri_total(sdp);
556 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
557 goto out;
558
559 spin_lock(&sdp->sd_statfs_spin);
560 gfs2_statfs_change_in(m_sc, m_bh->b_data +
561 sizeof(struct gfs2_dinode));
562 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
563 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
564 else
565 new_free = 0;
566 spin_unlock(&sdp->sd_statfs_spin);
567 fs_warn(sdp, "File system extended by %llu blocks.\n",
568 (unsigned long long)new_free);
569 gfs2_statfs_change(sdp, new_free, new_free, 0);
570
571 update_statfs(sdp, m_bh);
572 brelse(m_bh);
573 out:
574 sdp->sd_rindex_uptodate = 0;
575 gfs2_trans_end(sdp);
576 }
577
jdata_dirty_folio(struct address_space * mapping,struct folio * folio)578 static bool jdata_dirty_folio(struct address_space *mapping,
579 struct folio *folio)
580 {
581 if (current->journal_info)
582 folio_set_checked(folio);
583 return block_dirty_folio(mapping, folio);
584 }
585
586 /**
587 * gfs2_bmap - Block map function
588 * @mapping: Address space info
589 * @lblock: The block to map
590 *
591 * Returns: The disk address for the block or 0 on hole or error
592 */
593
gfs2_bmap(struct address_space * mapping,sector_t lblock)594 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
595 {
596 struct gfs2_inode *ip = GFS2_I(mapping->host);
597 struct gfs2_holder i_gh;
598 sector_t dblock = 0;
599 int error;
600
601 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
602 if (error)
603 return 0;
604
605 if (!gfs2_is_stuffed(ip))
606 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
607
608 gfs2_glock_dq_uninit(&i_gh);
609
610 return dblock;
611 }
612
gfs2_discard(struct gfs2_sbd * sdp,struct buffer_head * bh)613 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
614 {
615 struct gfs2_bufdata *bd;
616
617 lock_buffer(bh);
618 gfs2_log_lock(sdp);
619 clear_buffer_dirty(bh);
620 bd = bh->b_private;
621 if (bd) {
622 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
623 list_del_init(&bd->bd_list);
624 else {
625 spin_lock(&sdp->sd_ail_lock);
626 gfs2_remove_from_journal(bh, REMOVE_JDATA);
627 spin_unlock(&sdp->sd_ail_lock);
628 }
629 }
630 bh->b_bdev = NULL;
631 clear_buffer_mapped(bh);
632 clear_buffer_req(bh);
633 clear_buffer_new(bh);
634 gfs2_log_unlock(sdp);
635 unlock_buffer(bh);
636 }
637
gfs2_invalidate_folio(struct folio * folio,size_t offset,size_t length)638 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
639 size_t length)
640 {
641 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
642 size_t stop = offset + length;
643 int partial_page = (offset || length < folio_size(folio));
644 struct buffer_head *bh, *head;
645 unsigned long pos = 0;
646
647 BUG_ON(!folio_test_locked(folio));
648 if (!partial_page)
649 folio_clear_checked(folio);
650 head = folio_buffers(folio);
651 if (!head)
652 goto out;
653
654 bh = head;
655 do {
656 if (pos + bh->b_size > stop)
657 return;
658
659 if (offset <= pos)
660 gfs2_discard(sdp, bh);
661 pos += bh->b_size;
662 bh = bh->b_this_page;
663 } while (bh != head);
664 out:
665 if (!partial_page)
666 filemap_release_folio(folio, 0);
667 }
668
669 /**
670 * gfs2_release_folio - free the metadata associated with a folio
671 * @folio: the folio that's being released
672 * @gfp_mask: passed from Linux VFS, ignored by us
673 *
674 * Calls try_to_free_buffers() to free the buffers and put the folio if the
675 * buffers can be released.
676 *
677 * Returns: true if the folio was put or else false
678 */
679
gfs2_release_folio(struct folio * folio,gfp_t gfp_mask)680 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
681 {
682 struct address_space *mapping = folio->mapping;
683 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
684 struct buffer_head *bh, *head;
685 struct gfs2_bufdata *bd;
686
687 head = folio_buffers(folio);
688 if (!head)
689 return false;
690
691 /*
692 * mm accommodates an old ext3 case where clean folios might
693 * not have had the dirty bit cleared. Thus, it can send actual
694 * dirty folios to ->release_folio() via shrink_active_list().
695 *
696 * As a workaround, we skip folios that contain dirty buffers
697 * below. Once ->release_folio isn't called on dirty folios
698 * anymore, we can warn on dirty buffers like we used to here
699 * again.
700 */
701
702 gfs2_log_lock(sdp);
703 bh = head;
704 do {
705 if (atomic_read(&bh->b_count))
706 goto cannot_release;
707 bd = bh->b_private;
708 if (bd && bd->bd_tr)
709 goto cannot_release;
710 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
711 goto cannot_release;
712 bh = bh->b_this_page;
713 } while (bh != head);
714
715 bh = head;
716 do {
717 bd = bh->b_private;
718 if (bd) {
719 gfs2_assert_warn(sdp, bd->bd_bh == bh);
720 bd->bd_bh = NULL;
721 bh->b_private = NULL;
722 /*
723 * The bd may still be queued as a revoke, in which
724 * case we must not dequeue nor free it.
725 */
726 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
727 list_del_init(&bd->bd_list);
728 if (list_empty(&bd->bd_list))
729 kmem_cache_free(gfs2_bufdata_cachep, bd);
730 }
731
732 bh = bh->b_this_page;
733 } while (bh != head);
734 gfs2_log_unlock(sdp);
735
736 return try_to_free_buffers(folio);
737
738 cannot_release:
739 gfs2_log_unlock(sdp);
740 return false;
741 }
742
743 static const struct address_space_operations gfs2_aops = {
744 .writepages = gfs2_writepages,
745 .read_folio = gfs2_read_folio,
746 .readahead = gfs2_readahead,
747 .dirty_folio = iomap_dirty_folio,
748 .release_folio = iomap_release_folio,
749 .invalidate_folio = iomap_invalidate_folio,
750 .bmap = gfs2_bmap,
751 .migrate_folio = filemap_migrate_folio,
752 .is_partially_uptodate = iomap_is_partially_uptodate,
753 .error_remove_page = generic_error_remove_page,
754 };
755
756 static const struct address_space_operations gfs2_jdata_aops = {
757 .writepage = gfs2_jdata_writepage,
758 .writepages = gfs2_jdata_writepages,
759 .read_folio = gfs2_read_folio,
760 .readahead = gfs2_readahead,
761 .dirty_folio = jdata_dirty_folio,
762 .bmap = gfs2_bmap,
763 .invalidate_folio = gfs2_invalidate_folio,
764 .release_folio = gfs2_release_folio,
765 .is_partially_uptodate = block_is_partially_uptodate,
766 .error_remove_page = generic_error_remove_page,
767 };
768
gfs2_set_aops(struct inode * inode)769 void gfs2_set_aops(struct inode *inode)
770 {
771 if (gfs2_is_jdata(GFS2_I(inode)))
772 inode->i_mapping->a_ops = &gfs2_jdata_aops;
773 else
774 inode->i_mapping->a_ops = &gfs2_aops;
775 }
776