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_page_add_databufs(struct gfs2_inode *ip, struct page *page,
41 unsigned int from, unsigned int len)
43 struct buffer_head *head = page_buffers(page);
44 unsigned int bsize = head->b_size;
45 struct buffer_head *bh;
46 unsigned int to = from + len;
47 unsigned int start, end;
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_page - gfs2 jdata-specific version of block_write_full_page
86 * @page: The page to write
87 * @wbc: The writeback control
89 * This is the same as calling block_write_full_page, but it also
90 * writes pages outside of i_size
92 static int gfs2_write_jdata_page(struct page *page,
93 struct writeback_control *wbc)
95 struct inode * const inode = page->mapping->host;
96 loff_t i_size = i_size_read(inode);
97 const pgoff_t end_index = i_size >> PAGE_SHIFT;
101 * The page straddles i_size. It must be zeroed out on each and every
102 * writepage invocation because it may be mmapped. "A file is mapped
103 * in multiples of the page size. For a file that is not a multiple of
104 * the page size, the remaining memory is zeroed when mapped, and
105 * writes to that region are not written out to the file."
107 offset = i_size & (PAGE_SIZE - 1);
108 if (page->index == end_index && offset)
109 zero_user_segment(page, offset, PAGE_SIZE);
111 return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc,
112 end_buffer_async_write);
116 * __gfs2_jdata_writepage - The core of jdata writepage
117 * @page: The page to write
118 * @wbc: The writeback control
120 * This is shared between writepage and writepages and implements the
121 * core of the writepage operation. If a transaction is required then
122 * PageChecked will have been set and the transaction will have
123 * already been started before this is called.
126 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
128 struct inode *inode = page->mapping->host;
129 struct gfs2_inode *ip = GFS2_I(inode);
130 struct gfs2_sbd *sdp = GFS2_SB(inode);
132 if (PageChecked(page)) {
133 ClearPageChecked(page);
134 if (!page_has_buffers(page)) {
135 create_empty_buffers(page, inode->i_sb->s_blocksize,
136 BIT(BH_Dirty)|BIT(BH_Uptodate));
138 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize);
140 return gfs2_write_jdata_page(page, wbc);
144 * gfs2_jdata_writepage - Write complete page
145 * @page: Page to write
146 * @wbc: The writeback control
152 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
154 struct inode *inode = page->mapping->host;
155 struct gfs2_inode *ip = GFS2_I(inode);
156 struct gfs2_sbd *sdp = GFS2_SB(inode);
158 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
160 if (PageChecked(page) || current->journal_info)
162 return __gfs2_jdata_writepage(page, wbc);
165 redirty_page_for_writepage(wbc, page);
172 * gfs2_writepages - Write a bunch of dirty pages back to disk
173 * @mapping: The mapping to write
174 * @wbc: Write-back control
176 * Used for both ordered and writeback modes.
178 static int gfs2_writepages(struct address_space *mapping,
179 struct writeback_control *wbc)
181 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
182 struct iomap_writepage_ctx wpc = { };
186 * Even if we didn't write any 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.
191 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
193 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
198 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
199 * @mapping: The mapping
200 * @wbc: The writeback control
201 * @pvec: The vector of pages
202 * @nr_pages: The number of pages to write
203 * @done_index: Page index
205 * Returns: non-zero if loop should terminate, zero otherwise
208 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
209 struct writeback_control *wbc,
210 struct pagevec *pvec,
214 struct inode *inode = mapping->host;
215 struct gfs2_sbd *sdp = GFS2_SB(inode);
216 unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
220 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
224 for(i = 0; i < nr_pages; i++) {
225 struct page *page = pvec->pages[i];
227 *done_index = page->index;
231 if (unlikely(page->mapping != mapping)) {
237 if (!PageDirty(page)) {
238 /* someone wrote it for us */
239 goto continue_unlock;
242 if (PageWriteback(page)) {
243 if (wbc->sync_mode != WB_SYNC_NONE)
244 wait_on_page_writeback(page);
246 goto continue_unlock;
249 BUG_ON(PageWriteback(page));
250 if (!clear_page_dirty_for_io(page))
251 goto continue_unlock;
253 trace_wbc_writepage(wbc, inode_to_bdi(inode));
255 ret = __gfs2_jdata_writepage(page, wbc);
257 if (ret == AOP_WRITEPAGE_ACTIVATE) {
263 * done_index is set past this page,
264 * so media errors will not choke
265 * background writeout for the entire
266 * file. This has consequences for
267 * range_cyclic semantics (ie. it may
268 * not be suitable for data integrity
271 *done_index = page->index + 1;
278 * We stop writing back only if we are not doing
279 * integrity sync. In case of integrity sync we have to
280 * keep going until we have written all the pages
281 * we tagged for writeback prior to entering this loop.
283 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
294 * gfs2_write_cache_jdata - Like write_cache_pages but different
295 * @mapping: The mapping to write
296 * @wbc: The writeback control
298 * The reason that we use our own function here is that we need to
299 * start transactions before we grab page locks. This allows us
300 * to get the ordering right.
303 static int gfs2_write_cache_jdata(struct address_space *mapping,
304 struct writeback_control *wbc)
310 pgoff_t writeback_index;
319 if (wbc->range_cyclic) {
320 writeback_index = mapping->writeback_index; /* prev offset */
321 index = writeback_index;
328 index = wbc->range_start >> PAGE_SHIFT;
329 end = wbc->range_end >> PAGE_SHIFT;
330 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
332 cycled = 1; /* ignore range_cyclic tests */
334 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
335 tag = PAGECACHE_TAG_TOWRITE;
337 tag = PAGECACHE_TAG_DIRTY;
340 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
341 tag_pages_for_writeback(mapping, index, end);
343 while (!done && (index <= end)) {
344 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
349 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
354 pagevec_release(&pvec);
358 if (!cycled && !done) {
361 * We hit the last page and there is more work to be done: wrap
362 * back to the start of the file
366 end = writeback_index - 1;
370 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
371 mapping->writeback_index = done_index;
378 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
379 * @mapping: The mapping to write
380 * @wbc: The writeback control
384 static int gfs2_jdata_writepages(struct address_space *mapping,
385 struct writeback_control *wbc)
387 struct gfs2_inode *ip = GFS2_I(mapping->host);
388 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
391 ret = gfs2_write_cache_jdata(mapping, wbc);
392 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
393 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
394 GFS2_LFC_JDATA_WPAGES);
395 ret = gfs2_write_cache_jdata(mapping, wbc);
401 * stuffed_readpage - Fill in a Linux page with stuffed file data
407 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
409 struct buffer_head *dibh;
410 u64 dsize = i_size_read(&ip->i_inode);
415 * Due to the order of unstuffing files and ->fault(), we can be
416 * asked for a zero page in the case of a stuffed file being extended,
417 * so we need to supply one here. It doesn't happen often.
419 if (unlikely(page->index)) {
420 zero_user(page, 0, PAGE_SIZE);
421 SetPageUptodate(page);
425 error = gfs2_meta_inode_buffer(ip, &dibh);
429 kaddr = kmap_atomic(page);
430 if (dsize > gfs2_max_stuffed_size(ip))
431 dsize = gfs2_max_stuffed_size(ip);
432 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
433 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
434 kunmap_atomic(kaddr);
435 flush_dcache_page(page);
437 SetPageUptodate(page);
443 * gfs2_read_folio - read a folio from a file
444 * @file: The file to read
445 * @folio: The folio in the file
447 static int gfs2_read_folio(struct file *file, struct folio *folio)
449 struct inode *inode = folio->mapping->host;
450 struct gfs2_inode *ip = GFS2_I(inode);
451 struct gfs2_sbd *sdp = GFS2_SB(inode);
454 if (!gfs2_is_jdata(ip) ||
455 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
456 error = iomap_read_folio(folio, &gfs2_iomap_ops);
457 } else if (gfs2_is_stuffed(ip)) {
458 error = stuffed_readpage(ip, &folio->page);
461 error = mpage_read_folio(folio, gfs2_block_map);
464 if (unlikely(gfs2_withdrawn(sdp)))
471 * gfs2_internal_read - read an internal file
472 * @ip: The gfs2 inode
473 * @buf: The buffer to fill
474 * @pos: The file position
475 * @size: The amount to read
479 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
482 struct address_space *mapping = ip->i_inode.i_mapping;
483 unsigned long index = *pos >> PAGE_SHIFT;
484 unsigned offset = *pos & (PAGE_SIZE - 1);
492 if (offset + size > PAGE_SIZE)
493 amt = PAGE_SIZE - offset;
494 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
496 return PTR_ERR(page);
497 p = kmap_atomic(page);
498 memcpy(buf + copied, p + offset, amt);
504 } while(copied < size);
510 * gfs2_readahead - Read a bunch of pages at once
511 * @rac: Read-ahead control structure
514 * 1. This is only for readahead, so we can simply ignore any things
515 * which are slightly inconvenient (such as locking conflicts between
516 * the page lock and the glock) and return having done no I/O. Its
517 * obviously not something we'd want to do on too regular a basis.
518 * Any I/O we ignore at this time will be done via readpage later.
519 * 2. We don't handle stuffed files here we let readpage do the honours.
520 * 3. mpage_readahead() does most of the heavy lifting in the common case.
521 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
524 static void gfs2_readahead(struct readahead_control *rac)
526 struct inode *inode = rac->mapping->host;
527 struct gfs2_inode *ip = GFS2_I(inode);
529 if (gfs2_is_stuffed(ip))
531 else if (gfs2_is_jdata(ip))
532 mpage_readahead(rac, gfs2_block_map);
534 iomap_readahead(rac, &gfs2_iomap_ops);
538 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
539 * @inode: the rindex inode
541 void adjust_fs_space(struct inode *inode)
543 struct gfs2_sbd *sdp = GFS2_SB(inode);
544 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
545 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
546 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
547 struct buffer_head *m_bh;
548 u64 fs_total, new_free;
550 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
553 /* Total up the file system space, according to the latest rindex. */
554 fs_total = gfs2_ri_total(sdp);
555 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
558 spin_lock(&sdp->sd_statfs_spin);
559 gfs2_statfs_change_in(m_sc, m_bh->b_data +
560 sizeof(struct gfs2_dinode));
561 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
562 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
565 spin_unlock(&sdp->sd_statfs_spin);
566 fs_warn(sdp, "File system extended by %llu blocks.\n",
567 (unsigned long long)new_free);
568 gfs2_statfs_change(sdp, new_free, new_free, 0);
570 update_statfs(sdp, m_bh);
573 sdp->sd_rindex_uptodate = 0;
577 static bool jdata_dirty_folio(struct address_space *mapping,
580 if (current->journal_info)
581 folio_set_checked(folio);
582 return block_dirty_folio(mapping, folio);
586 * gfs2_bmap - Block map function
587 * @mapping: Address space info
588 * @lblock: The block to map
590 * Returns: The disk address for the block or 0 on hole or error
593 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
595 struct gfs2_inode *ip = GFS2_I(mapping->host);
596 struct gfs2_holder i_gh;
600 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
604 if (!gfs2_is_stuffed(ip))
605 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
607 gfs2_glock_dq_uninit(&i_gh);
612 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
614 struct gfs2_bufdata *bd;
618 clear_buffer_dirty(bh);
621 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
622 list_del_init(&bd->bd_list);
624 spin_lock(&sdp->sd_ail_lock);
625 gfs2_remove_from_journal(bh, REMOVE_JDATA);
626 spin_unlock(&sdp->sd_ail_lock);
630 clear_buffer_mapped(bh);
631 clear_buffer_req(bh);
632 clear_buffer_new(bh);
633 gfs2_log_unlock(sdp);
637 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
640 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
641 size_t stop = offset + length;
642 int partial_page = (offset || length < folio_size(folio));
643 struct buffer_head *bh, *head;
644 unsigned long pos = 0;
646 BUG_ON(!folio_test_locked(folio));
648 folio_clear_checked(folio);
649 head = folio_buffers(folio);
655 if (pos + bh->b_size > stop)
659 gfs2_discard(sdp, bh);
661 bh = bh->b_this_page;
662 } while (bh != head);
665 filemap_release_folio(folio, 0);
669 * gfs2_release_folio - free the metadata associated with a folio
670 * @folio: the folio that's being released
671 * @gfp_mask: passed from Linux VFS, ignored by us
673 * Calls try_to_free_buffers() to free the buffers and put the folio if the
674 * buffers can be released.
676 * Returns: true if the folio was put or else false
679 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
681 struct address_space *mapping = folio->mapping;
682 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
683 struct buffer_head *bh, *head;
684 struct gfs2_bufdata *bd;
686 head = folio_buffers(folio);
691 * mm accommodates an old ext3 case where clean folios might
692 * not have had the dirty bit cleared. Thus, it can send actual
693 * dirty folios to ->release_folio() via shrink_active_list().
695 * As a workaround, we skip folios that contain dirty buffers
696 * below. Once ->release_folio isn't called on dirty folios
697 * anymore, we can warn on dirty buffers like we used to here
704 if (atomic_read(&bh->b_count))
709 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
711 bh = bh->b_this_page;
712 } while (bh != head);
718 gfs2_assert_warn(sdp, bd->bd_bh == bh);
720 bh->b_private = NULL;
722 * The bd may still be queued as a revoke, in which
723 * case we must not dequeue nor free it.
725 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
726 list_del_init(&bd->bd_list);
727 if (list_empty(&bd->bd_list))
728 kmem_cache_free(gfs2_bufdata_cachep, bd);
731 bh = bh->b_this_page;
732 } while (bh != head);
733 gfs2_log_unlock(sdp);
735 return try_to_free_buffers(folio);
738 gfs2_log_unlock(sdp);
742 static const struct address_space_operations gfs2_aops = {
743 .writepages = gfs2_writepages,
744 .read_folio = gfs2_read_folio,
745 .readahead = gfs2_readahead,
746 .dirty_folio = filemap_dirty_folio,
747 .release_folio = iomap_release_folio,
748 .invalidate_folio = iomap_invalidate_folio,
750 .direct_IO = noop_direct_IO,
751 .migrate_folio = filemap_migrate_folio,
752 .is_partially_uptodate = iomap_is_partially_uptodate,
753 .error_remove_page = generic_error_remove_page,
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,
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,
769 void gfs2_set_aops(struct inode *inode)
771 if (gfs2_is_jdata(GFS2_I(inode)))
772 inode->i_mapping->a_ops = &gfs2_jdata_aops;
774 inode->i_mapping->a_ops = &gfs2_aops;