1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
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>
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>
40 void gfs2_trans_add_databufs(struct gfs2_inode *ip, struct folio *folio,
41 size_t from, size_t len)
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;
49 for (bh = head, start = 0; bh != head || !start;
50 bh = bh->b_this_page, start = end) {
56 set_buffer_uptodate(bh);
57 gfs2_trans_add_data(ip->i_gl, bh);
62 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
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
71 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
72 struct buffer_head *bh_result, int create)
76 error = gfs2_block_map(inode, lblock, bh_result, 0);
79 if (!buffer_mapped(bh_result))
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
89 * This is the same as calling block_write_full_folio, but it also
90 * writes pages outside of i_size
92 static int gfs2_write_jdata_folio(struct folio *folio,
93 struct writeback_control *wbc)
95 struct inode * const inode = folio->mapping->host;
96 loff_t i_size = i_size_read(inode);
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."
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),
110 return __block_write_full_folio(inode, folio, gfs2_get_block_noalloc,
115 * __gfs2_jdata_write_folio - The core of jdata writepage
116 * @folio: The folio to write
117 * @wbc: The writeback control
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.
123 static int __gfs2_jdata_write_folio(struct folio *folio,
124 struct writeback_control *wbc)
126 struct inode *inode = folio->mapping->host;
127 struct gfs2_inode *ip = GFS2_I(inode);
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));
136 gfs2_trans_add_databufs(ip, folio, 0, folio_size(folio));
138 return gfs2_write_jdata_folio(folio, wbc);
142 * gfs2_writepages - Write a bunch of dirty pages back to disk
143 * @mapping: The mapping to write
144 * @wbc: Write-back control
146 * Used for both ordered and writeback modes.
148 static int gfs2_writepages(struct address_space *mapping,
149 struct writeback_control *wbc)
151 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
152 struct iomap_writepage_ctx wpc = { };
156 * Even if we didn't write enough pages here, we might still be holding
157 * dirty pages in the ail. We forcibly flush the ail because we don't
158 * want balance_dirty_pages() to loop indefinitely trying to write out
159 * pages held in the ail that it can't find.
161 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
162 if (ret == 0 && wbc->nr_to_write > 0)
163 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
168 * gfs2_write_jdata_batch - Write back a folio batch's worth of folios
169 * @mapping: The mapping
170 * @wbc: The writeback control
171 * @fbatch: The batch of folios
172 * @done_index: Page index
174 * Returns: non-zero if loop should terminate, zero otherwise
177 static int gfs2_write_jdata_batch(struct address_space *mapping,
178 struct writeback_control *wbc,
179 struct folio_batch *fbatch,
182 struct inode *inode = mapping->host;
183 struct gfs2_sbd *sdp = GFS2_SB(inode);
188 int nr_folios = folio_batch_count(fbatch);
190 for (i = 0; i < nr_folios; i++)
191 size += folio_size(fbatch->folios[i]);
192 nrblocks = size >> inode->i_blkbits;
194 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
198 for (i = 0; i < nr_folios; i++) {
199 struct folio *folio = fbatch->folios[i];
201 *done_index = folio->index;
205 if (unlikely(folio->mapping != mapping)) {
211 if (!folio_test_dirty(folio)) {
212 /* someone wrote it for us */
213 goto continue_unlock;
216 if (folio_test_writeback(folio)) {
217 if (wbc->sync_mode != WB_SYNC_NONE)
218 folio_wait_writeback(folio);
220 goto continue_unlock;
223 BUG_ON(folio_test_writeback(folio));
224 if (!folio_clear_dirty_for_io(folio))
225 goto continue_unlock;
227 trace_wbc_writepage(wbc, inode_to_bdi(inode));
229 ret = __gfs2_jdata_write_folio(folio, wbc);
231 if (ret == AOP_WRITEPAGE_ACTIVATE) {
237 * done_index is set past this page,
238 * so media errors will not choke
239 * background writeout for the entire
240 * file. This has consequences for
241 * range_cyclic semantics (ie. it may
242 * not be suitable for data integrity
245 *done_index = folio_next_index(folio);
252 * We stop writing back only if we are not doing
253 * integrity sync. In case of integrity sync we have to
254 * keep going until we have written all the pages
255 * we tagged for writeback prior to entering this loop.
257 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
268 * gfs2_write_cache_jdata - Like write_cache_pages but different
269 * @mapping: The mapping to write
270 * @wbc: The writeback control
272 * The reason that we use our own function here is that we need to
273 * start transactions before we grab page locks. This allows us
274 * to get the ordering right.
277 static int gfs2_write_cache_jdata(struct address_space *mapping,
278 struct writeback_control *wbc)
282 struct folio_batch fbatch;
284 pgoff_t writeback_index;
292 folio_batch_init(&fbatch);
293 if (wbc->range_cyclic) {
294 writeback_index = mapping->writeback_index; /* prev offset */
295 index = writeback_index;
302 index = wbc->range_start >> PAGE_SHIFT;
303 end = wbc->range_end >> PAGE_SHIFT;
304 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
306 cycled = 1; /* ignore range_cyclic tests */
308 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
309 tag = PAGECACHE_TAG_TOWRITE;
311 tag = PAGECACHE_TAG_DIRTY;
314 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
315 tag_pages_for_writeback(mapping, index, end);
317 while (!done && (index <= end)) {
318 nr_folios = filemap_get_folios_tag(mapping, &index, end,
323 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
329 folio_batch_release(&fbatch);
333 if (!cycled && !done) {
336 * We hit the last page and there is more work to be done: wrap
337 * back to the start of the file
341 end = writeback_index - 1;
345 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
346 mapping->writeback_index = done_index;
353 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
354 * @mapping: The mapping to write
355 * @wbc: The writeback control
359 static int gfs2_jdata_writepages(struct address_space *mapping,
360 struct writeback_control *wbc)
362 struct gfs2_inode *ip = GFS2_I(mapping->host);
363 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
366 ret = gfs2_write_cache_jdata(mapping, wbc);
367 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
368 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
369 GFS2_LFC_JDATA_WPAGES);
370 ret = gfs2_write_cache_jdata(mapping, wbc);
376 * stuffed_read_folio - Fill in a Linux folio with stuffed file data
382 static int stuffed_read_folio(struct gfs2_inode *ip, struct folio *folio)
384 struct buffer_head *dibh = NULL;
385 size_t dsize = i_size_read(&ip->i_inode);
390 * Due to the order of unstuffing files and ->fault(), we can be
391 * asked for a zero folio in the case of a stuffed file being extended,
392 * so we need to supply one here. It doesn't happen often.
394 if (unlikely(folio->index)) {
397 error = gfs2_meta_inode_buffer(ip, &dibh);
400 from = dibh->b_data + sizeof(struct gfs2_dinode);
403 folio_fill_tail(folio, 0, from, dsize);
406 folio_end_read(folio, error == 0);
412 * gfs2_read_folio - read a folio from a file
413 * @file: The file to read
414 * @folio: The folio in the file
416 static int gfs2_read_folio(struct file *file, struct folio *folio)
418 struct inode *inode = folio->mapping->host;
419 struct gfs2_inode *ip = GFS2_I(inode);
420 struct gfs2_sbd *sdp = GFS2_SB(inode);
423 if (!gfs2_is_jdata(ip) ||
424 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
425 error = iomap_read_folio(folio, &gfs2_iomap_ops);
426 } else if (gfs2_is_stuffed(ip)) {
427 error = stuffed_read_folio(ip, folio);
429 error = mpage_read_folio(folio, gfs2_block_map);
432 if (gfs2_withdrawing_or_withdrawn(sdp))
439 * gfs2_internal_read - read an internal file
440 * @ip: The gfs2 inode
441 * @buf: The buffer to fill
442 * @pos: The file position
443 * @size: The amount to read
447 ssize_t gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
450 struct address_space *mapping = ip->i_inode.i_mapping;
451 unsigned long index = *pos >> PAGE_SHIFT;
455 size_t offset, chunk;
458 folio = read_cache_folio(mapping, index, gfs2_read_folio, NULL);
460 if (PTR_ERR(folio) == -EINTR)
462 return PTR_ERR(folio);
464 offset = *pos + copied - folio_pos(folio);
465 chunk = min(size - copied, folio_size(folio) - offset);
466 memcpy_from_folio(buf + copied, folio, offset, chunk);
467 index = folio_next_index(folio);
470 } while(copied < size);
476 * gfs2_readahead - Read a bunch of pages at once
477 * @rac: Read-ahead control structure
480 * 1. This is only for readahead, so we can simply ignore any things
481 * which are slightly inconvenient (such as locking conflicts between
482 * the page lock and the glock) and return having done no I/O. Its
483 * obviously not something we'd want to do on too regular a basis.
484 * Any I/O we ignore at this time will be done via readpage later.
485 * 2. We don't handle stuffed files here we let readpage do the honours.
486 * 3. mpage_readahead() does most of the heavy lifting in the common case.
487 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
490 static void gfs2_readahead(struct readahead_control *rac)
492 struct inode *inode = rac->mapping->host;
493 struct gfs2_inode *ip = GFS2_I(inode);
495 if (gfs2_is_stuffed(ip))
497 else if (gfs2_is_jdata(ip))
498 mpage_readahead(rac, gfs2_block_map);
500 iomap_readahead(rac, &gfs2_iomap_ops);
504 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
505 * @inode: the rindex inode
507 void adjust_fs_space(struct inode *inode)
509 struct gfs2_sbd *sdp = GFS2_SB(inode);
510 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
511 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
512 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
513 struct buffer_head *m_bh;
514 u64 fs_total, new_free;
516 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
519 /* Total up the file system space, according to the latest rindex. */
520 fs_total = gfs2_ri_total(sdp);
521 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
524 spin_lock(&sdp->sd_statfs_spin);
525 gfs2_statfs_change_in(m_sc, m_bh->b_data +
526 sizeof(struct gfs2_dinode));
527 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
528 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
531 spin_unlock(&sdp->sd_statfs_spin);
532 fs_warn(sdp, "File system extended by %llu blocks.\n",
533 (unsigned long long)new_free);
534 gfs2_statfs_change(sdp, new_free, new_free, 0);
536 update_statfs(sdp, m_bh);
539 sdp->sd_rindex_uptodate = 0;
543 static bool jdata_dirty_folio(struct address_space *mapping,
546 if (current->journal_info)
547 folio_set_checked(folio);
548 return block_dirty_folio(mapping, folio);
552 * gfs2_bmap - Block map function
553 * @mapping: Address space info
554 * @lblock: The block to map
556 * Returns: The disk address for the block or 0 on hole or error
559 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
561 struct gfs2_inode *ip = GFS2_I(mapping->host);
562 struct gfs2_holder i_gh;
566 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
570 if (!gfs2_is_stuffed(ip))
571 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
573 gfs2_glock_dq_uninit(&i_gh);
578 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
580 struct gfs2_bufdata *bd;
584 clear_buffer_dirty(bh);
587 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
588 list_del_init(&bd->bd_list);
590 spin_lock(&sdp->sd_ail_lock);
591 gfs2_remove_from_journal(bh, REMOVE_JDATA);
592 spin_unlock(&sdp->sd_ail_lock);
596 clear_buffer_mapped(bh);
597 clear_buffer_req(bh);
598 clear_buffer_new(bh);
599 gfs2_log_unlock(sdp);
603 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
606 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
607 size_t stop = offset + length;
608 int partial_page = (offset || length < folio_size(folio));
609 struct buffer_head *bh, *head;
610 unsigned long pos = 0;
612 BUG_ON(!folio_test_locked(folio));
614 folio_clear_checked(folio);
615 head = folio_buffers(folio);
621 if (pos + bh->b_size > stop)
625 gfs2_discard(sdp, bh);
627 bh = bh->b_this_page;
628 } while (bh != head);
631 filemap_release_folio(folio, 0);
635 * gfs2_release_folio - free the metadata associated with a folio
636 * @folio: the folio that's being released
637 * @gfp_mask: passed from Linux VFS, ignored by us
639 * Calls try_to_free_buffers() to free the buffers and put the folio if the
640 * buffers can be released.
642 * Returns: true if the folio was put or else false
645 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
647 struct address_space *mapping = folio->mapping;
648 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
649 struct buffer_head *bh, *head;
650 struct gfs2_bufdata *bd;
652 head = folio_buffers(folio);
657 * mm accommodates an old ext3 case where clean folios might
658 * not have had the dirty bit cleared. Thus, it can send actual
659 * dirty folios to ->release_folio() via shrink_active_list().
661 * As a workaround, we skip folios that contain dirty buffers
662 * below. Once ->release_folio isn't called on dirty folios
663 * anymore, we can warn on dirty buffers like we used to here
670 if (atomic_read(&bh->b_count))
675 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
677 bh = bh->b_this_page;
678 } while (bh != head);
684 gfs2_assert_warn(sdp, bd->bd_bh == bh);
686 bh->b_private = NULL;
688 * The bd may still be queued as a revoke, in which
689 * case we must not dequeue nor free it.
691 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
692 list_del_init(&bd->bd_list);
693 if (list_empty(&bd->bd_list))
694 kmem_cache_free(gfs2_bufdata_cachep, bd);
697 bh = bh->b_this_page;
698 } while (bh != head);
699 gfs2_log_unlock(sdp);
701 return try_to_free_buffers(folio);
704 gfs2_log_unlock(sdp);
708 static const struct address_space_operations gfs2_aops = {
709 .writepages = gfs2_writepages,
710 .read_folio = gfs2_read_folio,
711 .readahead = gfs2_readahead,
712 .dirty_folio = iomap_dirty_folio,
713 .release_folio = iomap_release_folio,
714 .invalidate_folio = iomap_invalidate_folio,
716 .migrate_folio = filemap_migrate_folio,
717 .is_partially_uptodate = iomap_is_partially_uptodate,
718 .error_remove_folio = generic_error_remove_folio,
721 static const struct address_space_operations gfs2_jdata_aops = {
722 .writepages = gfs2_jdata_writepages,
723 .read_folio = gfs2_read_folio,
724 .readahead = gfs2_readahead,
725 .dirty_folio = jdata_dirty_folio,
727 .migrate_folio = buffer_migrate_folio,
728 .invalidate_folio = gfs2_invalidate_folio,
729 .release_folio = gfs2_release_folio,
730 .is_partially_uptodate = block_is_partially_uptodate,
731 .error_remove_folio = generic_error_remove_folio,
734 void gfs2_set_aops(struct inode *inode)
736 if (gfs2_is_jdata(GFS2_I(inode)))
737 inode->i_mapping->a_ops = &gfs2_jdata_aops;
739 inode->i_mapping->a_ops = &gfs2_aops;