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_writepage - Write page for writeback mappings
87 * @wbc: The writeback control
89 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
91 struct inode *inode = page->mapping->host;
92 struct gfs2_inode *ip = GFS2_I(inode);
93 struct gfs2_sbd *sdp = GFS2_SB(inode);
94 struct iomap_writepage_ctx wpc = { };
96 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
98 if (current->journal_info)
100 return iomap_writepage(page, wbc, &wpc, &gfs2_writeback_ops);
103 redirty_page_for_writepage(wbc, page);
110 * gfs2_write_jdata_page - gfs2 jdata-specific version of block_write_full_page
111 * @page: The page to write
112 * @wbc: The writeback control
114 * This is the same as calling block_write_full_page, but it also
115 * writes pages outside of i_size
117 static int gfs2_write_jdata_page(struct page *page,
118 struct writeback_control *wbc)
120 struct inode * const inode = page->mapping->host;
121 loff_t i_size = i_size_read(inode);
122 const pgoff_t end_index = i_size >> PAGE_SHIFT;
126 * The page straddles i_size. It must be zeroed out on each and every
127 * writepage invocation because it may be mmapped. "A file is mapped
128 * in multiples of the page size. For a file that is not a multiple of
129 * the page size, the remaining memory is zeroed when mapped, and
130 * writes to that region are not written out to the file."
132 offset = i_size & (PAGE_SIZE - 1);
133 if (page->index == end_index && offset)
134 zero_user_segment(page, offset, PAGE_SIZE);
136 return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc,
137 end_buffer_async_write);
141 * __gfs2_jdata_writepage - The core of jdata writepage
142 * @page: The page to write
143 * @wbc: The writeback control
145 * This is shared between writepage and writepages and implements the
146 * core of the writepage operation. If a transaction is required then
147 * PageChecked will have been set and the transaction will have
148 * already been started before this is called.
151 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
153 struct inode *inode = page->mapping->host;
154 struct gfs2_inode *ip = GFS2_I(inode);
155 struct gfs2_sbd *sdp = GFS2_SB(inode);
157 if (PageChecked(page)) {
158 ClearPageChecked(page);
159 if (!page_has_buffers(page)) {
160 create_empty_buffers(page, inode->i_sb->s_blocksize,
161 BIT(BH_Dirty)|BIT(BH_Uptodate));
163 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize);
165 return gfs2_write_jdata_page(page, wbc);
169 * gfs2_jdata_writepage - Write complete page
170 * @page: Page to write
171 * @wbc: The writeback control
177 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
179 struct inode *inode = page->mapping->host;
180 struct gfs2_inode *ip = GFS2_I(inode);
181 struct gfs2_sbd *sdp = GFS2_SB(inode);
183 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
185 if (PageChecked(page) || current->journal_info)
187 return __gfs2_jdata_writepage(page, wbc);
190 redirty_page_for_writepage(wbc, page);
197 * gfs2_writepages - Write a bunch of dirty pages back to disk
198 * @mapping: The mapping to write
199 * @wbc: Write-back control
201 * Used for both ordered and writeback modes.
203 static int gfs2_writepages(struct address_space *mapping,
204 struct writeback_control *wbc)
206 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
207 struct iomap_writepage_ctx wpc = { };
211 * Even if we didn't write any pages here, we might still be holding
212 * dirty pages in the ail. We forcibly flush the ail because we don't
213 * want balance_dirty_pages() to loop indefinitely trying to write out
214 * pages held in the ail that it can't find.
216 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
218 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
223 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
224 * @mapping: The mapping
225 * @wbc: The writeback control
226 * @pvec: The vector of pages
227 * @nr_pages: The number of pages to write
228 * @done_index: Page index
230 * Returns: non-zero if loop should terminate, zero otherwise
233 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
234 struct writeback_control *wbc,
235 struct pagevec *pvec,
239 struct inode *inode = mapping->host;
240 struct gfs2_sbd *sdp = GFS2_SB(inode);
241 unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
245 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
249 for(i = 0; i < nr_pages; i++) {
250 struct page *page = pvec->pages[i];
252 *done_index = page->index;
256 if (unlikely(page->mapping != mapping)) {
262 if (!PageDirty(page)) {
263 /* someone wrote it for us */
264 goto continue_unlock;
267 if (PageWriteback(page)) {
268 if (wbc->sync_mode != WB_SYNC_NONE)
269 wait_on_page_writeback(page);
271 goto continue_unlock;
274 BUG_ON(PageWriteback(page));
275 if (!clear_page_dirty_for_io(page))
276 goto continue_unlock;
278 trace_wbc_writepage(wbc, inode_to_bdi(inode));
280 ret = __gfs2_jdata_writepage(page, wbc);
282 if (ret == AOP_WRITEPAGE_ACTIVATE) {
288 * done_index is set past this page,
289 * so media errors will not choke
290 * background writeout for the entire
291 * file. This has consequences for
292 * range_cyclic semantics (ie. it may
293 * not be suitable for data integrity
296 *done_index = page->index + 1;
303 * We stop writing back only if we are not doing
304 * integrity sync. In case of integrity sync we have to
305 * keep going until we have written all the pages
306 * we tagged for writeback prior to entering this loop.
308 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
319 * gfs2_write_cache_jdata - Like write_cache_pages but different
320 * @mapping: The mapping to write
321 * @wbc: The writeback control
323 * The reason that we use our own function here is that we need to
324 * start transactions before we grab page locks. This allows us
325 * to get the ordering right.
328 static int gfs2_write_cache_jdata(struct address_space *mapping,
329 struct writeback_control *wbc)
335 pgoff_t writeback_index;
344 if (wbc->range_cyclic) {
345 writeback_index = mapping->writeback_index; /* prev offset */
346 index = writeback_index;
353 index = wbc->range_start >> PAGE_SHIFT;
354 end = wbc->range_end >> PAGE_SHIFT;
355 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
357 cycled = 1; /* ignore range_cyclic tests */
359 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
360 tag = PAGECACHE_TAG_TOWRITE;
362 tag = PAGECACHE_TAG_DIRTY;
365 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
366 tag_pages_for_writeback(mapping, index, end);
368 while (!done && (index <= end)) {
369 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
374 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
379 pagevec_release(&pvec);
383 if (!cycled && !done) {
386 * We hit the last page and there is more work to be done: wrap
387 * back to the start of the file
391 end = writeback_index - 1;
395 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
396 mapping->writeback_index = done_index;
403 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
404 * @mapping: The mapping to write
405 * @wbc: The writeback control
409 static int gfs2_jdata_writepages(struct address_space *mapping,
410 struct writeback_control *wbc)
412 struct gfs2_inode *ip = GFS2_I(mapping->host);
413 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
416 ret = gfs2_write_cache_jdata(mapping, wbc);
417 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
418 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
419 GFS2_LFC_JDATA_WPAGES);
420 ret = gfs2_write_cache_jdata(mapping, wbc);
426 * stuffed_readpage - Fill in a Linux page with stuffed file data
432 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
434 struct buffer_head *dibh;
435 u64 dsize = i_size_read(&ip->i_inode);
440 * Due to the order of unstuffing files and ->fault(), we can be
441 * asked for a zero page in the case of a stuffed file being extended,
442 * so we need to supply one here. It doesn't happen often.
444 if (unlikely(page->index)) {
445 zero_user(page, 0, PAGE_SIZE);
446 SetPageUptodate(page);
450 error = gfs2_meta_inode_buffer(ip, &dibh);
454 kaddr = kmap_atomic(page);
455 if (dsize > gfs2_max_stuffed_size(ip))
456 dsize = gfs2_max_stuffed_size(ip);
457 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
458 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
459 kunmap_atomic(kaddr);
460 flush_dcache_page(page);
462 SetPageUptodate(page);
468 * gfs2_read_folio - read a folio from a file
469 * @file: The file to read
470 * @folio: The folio in the file
472 static int gfs2_read_folio(struct file *file, struct folio *folio)
474 struct inode *inode = folio->mapping->host;
475 struct gfs2_inode *ip = GFS2_I(inode);
476 struct gfs2_sbd *sdp = GFS2_SB(inode);
479 if (!gfs2_is_jdata(ip) ||
480 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
481 error = iomap_read_folio(folio, &gfs2_iomap_ops);
482 } else if (gfs2_is_stuffed(ip)) {
483 error = stuffed_readpage(ip, &folio->page);
486 error = mpage_read_folio(folio, gfs2_block_map);
489 if (unlikely(gfs2_withdrawn(sdp)))
496 * gfs2_internal_read - read an internal file
497 * @ip: The gfs2 inode
498 * @buf: The buffer to fill
499 * @pos: The file position
500 * @size: The amount to read
504 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
507 struct address_space *mapping = ip->i_inode.i_mapping;
508 unsigned long index = *pos >> PAGE_SHIFT;
509 unsigned offset = *pos & (PAGE_SIZE - 1);
517 if (offset + size > PAGE_SIZE)
518 amt = PAGE_SIZE - offset;
519 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
521 return PTR_ERR(page);
522 p = kmap_atomic(page);
523 memcpy(buf + copied, p + offset, amt);
529 } while(copied < size);
535 * gfs2_readahead - Read a bunch of pages at once
536 * @rac: Read-ahead control structure
539 * 1. This is only for readahead, so we can simply ignore any things
540 * which are slightly inconvenient (such as locking conflicts between
541 * the page lock and the glock) and return having done no I/O. Its
542 * obviously not something we'd want to do on too regular a basis.
543 * Any I/O we ignore at this time will be done via readpage later.
544 * 2. We don't handle stuffed files here we let readpage do the honours.
545 * 3. mpage_readahead() does most of the heavy lifting in the common case.
546 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
549 static void gfs2_readahead(struct readahead_control *rac)
551 struct inode *inode = rac->mapping->host;
552 struct gfs2_inode *ip = GFS2_I(inode);
554 if (gfs2_is_stuffed(ip))
556 else if (gfs2_is_jdata(ip))
557 mpage_readahead(rac, gfs2_block_map);
559 iomap_readahead(rac, &gfs2_iomap_ops);
563 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
564 * @inode: the rindex inode
566 void adjust_fs_space(struct inode *inode)
568 struct gfs2_sbd *sdp = GFS2_SB(inode);
569 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
570 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
571 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
572 struct buffer_head *m_bh;
573 u64 fs_total, new_free;
575 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
578 /* Total up the file system space, according to the latest rindex. */
579 fs_total = gfs2_ri_total(sdp);
580 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
583 spin_lock(&sdp->sd_statfs_spin);
584 gfs2_statfs_change_in(m_sc, m_bh->b_data +
585 sizeof(struct gfs2_dinode));
586 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
587 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
590 spin_unlock(&sdp->sd_statfs_spin);
591 fs_warn(sdp, "File system extended by %llu blocks.\n",
592 (unsigned long long)new_free);
593 gfs2_statfs_change(sdp, new_free, new_free, 0);
595 update_statfs(sdp, m_bh);
598 sdp->sd_rindex_uptodate = 0;
602 static bool jdata_dirty_folio(struct address_space *mapping,
605 if (current->journal_info)
606 folio_set_checked(folio);
607 return block_dirty_folio(mapping, folio);
611 * gfs2_bmap - Block map function
612 * @mapping: Address space info
613 * @lblock: The block to map
615 * Returns: The disk address for the block or 0 on hole or error
618 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
620 struct gfs2_inode *ip = GFS2_I(mapping->host);
621 struct gfs2_holder i_gh;
625 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
629 if (!gfs2_is_stuffed(ip))
630 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
632 gfs2_glock_dq_uninit(&i_gh);
637 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
639 struct gfs2_bufdata *bd;
643 clear_buffer_dirty(bh);
646 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
647 list_del_init(&bd->bd_list);
649 spin_lock(&sdp->sd_ail_lock);
650 gfs2_remove_from_journal(bh, REMOVE_JDATA);
651 spin_unlock(&sdp->sd_ail_lock);
655 clear_buffer_mapped(bh);
656 clear_buffer_req(bh);
657 clear_buffer_new(bh);
658 gfs2_log_unlock(sdp);
662 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
665 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
666 size_t stop = offset + length;
667 int partial_page = (offset || length < folio_size(folio));
668 struct buffer_head *bh, *head;
669 unsigned long pos = 0;
671 BUG_ON(!folio_test_locked(folio));
673 folio_clear_checked(folio);
674 head = folio_buffers(folio);
680 if (pos + bh->b_size > stop)
684 gfs2_discard(sdp, bh);
686 bh = bh->b_this_page;
687 } while (bh != head);
690 filemap_release_folio(folio, 0);
694 * gfs2_release_folio - free the metadata associated with a folio
695 * @folio: the folio that's being released
696 * @gfp_mask: passed from Linux VFS, ignored by us
698 * Calls try_to_free_buffers() to free the buffers and put the folio if the
699 * buffers can be released.
701 * Returns: true if the folio was put or else false
704 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
706 struct address_space *mapping = folio->mapping;
707 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
708 struct buffer_head *bh, *head;
709 struct gfs2_bufdata *bd;
711 head = folio_buffers(folio);
716 * mm accommodates an old ext3 case where clean folios might
717 * not have had the dirty bit cleared. Thus, it can send actual
718 * dirty folios to ->release_folio() via shrink_active_list().
720 * As a workaround, we skip folios that contain dirty buffers
721 * below. Once ->release_folio isn't called on dirty folios
722 * anymore, we can warn on dirty buffers like we used to here
729 if (atomic_read(&bh->b_count))
734 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
736 bh = bh->b_this_page;
737 } while (bh != head);
743 gfs2_assert_warn(sdp, bd->bd_bh == bh);
745 bh->b_private = NULL;
747 * The bd may still be queued as a revoke, in which
748 * case we must not dequeue nor free it.
750 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
751 list_del_init(&bd->bd_list);
752 if (list_empty(&bd->bd_list))
753 kmem_cache_free(gfs2_bufdata_cachep, bd);
756 bh = bh->b_this_page;
757 } while (bh != head);
758 gfs2_log_unlock(sdp);
760 return try_to_free_buffers(folio);
763 gfs2_log_unlock(sdp);
767 static const struct address_space_operations gfs2_aops = {
768 .writepage = gfs2_writepage,
769 .writepages = gfs2_writepages,
770 .read_folio = gfs2_read_folio,
771 .readahead = gfs2_readahead,
772 .dirty_folio = filemap_dirty_folio,
773 .release_folio = iomap_release_folio,
774 .invalidate_folio = iomap_invalidate_folio,
776 .direct_IO = noop_direct_IO,
777 .migratepage = iomap_migrate_page,
778 .is_partially_uptodate = iomap_is_partially_uptodate,
779 .error_remove_page = generic_error_remove_page,
782 static const struct address_space_operations gfs2_jdata_aops = {
783 .writepage = gfs2_jdata_writepage,
784 .writepages = gfs2_jdata_writepages,
785 .read_folio = gfs2_read_folio,
786 .readahead = gfs2_readahead,
787 .dirty_folio = jdata_dirty_folio,
789 .invalidate_folio = gfs2_invalidate_folio,
790 .release_folio = gfs2_release_folio,
791 .is_partially_uptodate = block_is_partially_uptodate,
792 .error_remove_page = generic_error_remove_page,
795 void gfs2_set_aops(struct inode *inode)
797 if (gfs2_is_jdata(GFS2_I(inode)))
798 inode->i_mapping->a_ops = &gfs2_jdata_aops;
800 inode->i_mapping->a_ops = &gfs2_aops;