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 unsigned int from, unsigned int len)
43 struct buffer_head *head = folio_buffers(folio);
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);
131 if (PageChecked(page)) {
132 ClearPageChecked(page);
133 if (!page_has_buffers(page)) {
134 create_empty_buffers(page, inode->i_sb->s_blocksize,
135 BIT(BH_Dirty)|BIT(BH_Uptodate));
137 gfs2_trans_add_databufs(ip, page_folio(page), 0, PAGE_SIZE);
139 return gfs2_write_jdata_page(page, wbc);
143 * gfs2_jdata_writepage - Write complete page
144 * @page: Page to write
145 * @wbc: The writeback control
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 (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
159 if (PageChecked(page) || current->journal_info)
161 return __gfs2_jdata_writepage(page, wbc);
164 redirty_page_for_writepage(wbc, page);
171 * gfs2_writepages - Write a bunch of dirty pages back to disk
172 * @mapping: The mapping to write
173 * @wbc: Write-back control
175 * Used for both ordered and writeback modes.
177 static int gfs2_writepages(struct address_space *mapping,
178 struct writeback_control *wbc)
180 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
181 struct iomap_writepage_ctx wpc = { };
185 * Even if we didn't write any 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.
190 ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
192 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
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
203 * Returns: non-zero if loop should terminate, zero otherwise
206 static int gfs2_write_jdata_batch(struct address_space *mapping,
207 struct writeback_control *wbc,
208 struct folio_batch *fbatch,
211 struct inode *inode = mapping->host;
212 struct gfs2_sbd *sdp = GFS2_SB(inode);
217 int nr_folios = folio_batch_count(fbatch);
219 for (i = 0; i < nr_folios; i++)
220 nr_pages += folio_nr_pages(fbatch->folios[i]);
221 nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
223 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
227 for (i = 0; i < nr_folios; i++) {
228 struct folio *folio = fbatch->folios[i];
230 *done_index = folio->index;
234 if (unlikely(folio->mapping != mapping)) {
240 if (!folio_test_dirty(folio)) {
241 /* someone wrote it for us */
242 goto continue_unlock;
245 if (folio_test_writeback(folio)) {
246 if (wbc->sync_mode != WB_SYNC_NONE)
247 folio_wait_writeback(folio);
249 goto continue_unlock;
252 BUG_ON(folio_test_writeback(folio));
253 if (!folio_clear_dirty_for_io(folio))
254 goto continue_unlock;
256 trace_wbc_writepage(wbc, inode_to_bdi(inode));
258 ret = __gfs2_jdata_writepage(&folio->page, wbc);
260 if (ret == AOP_WRITEPAGE_ACTIVATE) {
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
274 *done_index = folio->index +
275 folio_nr_pages(folio);
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.
287 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
298 * gfs2_write_cache_jdata - Like write_cache_pages but different
299 * @mapping: The mapping to write
300 * @wbc: The writeback control
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.
307 static int gfs2_write_cache_jdata(struct address_space *mapping,
308 struct writeback_control *wbc)
312 struct folio_batch fbatch;
314 pgoff_t writeback_index;
322 folio_batch_init(&fbatch);
323 if (wbc->range_cyclic) {
324 writeback_index = mapping->writeback_index; /* prev offset */
325 index = writeback_index;
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)
336 cycled = 1; /* ignore range_cyclic tests */
338 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
339 tag = PAGECACHE_TAG_TOWRITE;
341 tag = PAGECACHE_TAG_DIRTY;
344 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
345 tag_pages_for_writeback(mapping, index, end);
347 while (!done && (index <= end)) {
348 nr_folios = filemap_get_folios_tag(mapping, &index, end,
353 ret = gfs2_write_jdata_batch(mapping, wbc, &fbatch,
359 folio_batch_release(&fbatch);
363 if (!cycled && !done) {
366 * We hit the last page and there is more work to be done: wrap
367 * back to the start of the file
371 end = writeback_index - 1;
375 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
376 mapping->writeback_index = done_index;
383 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
384 * @mapping: The mapping to write
385 * @wbc: The writeback control
389 static int gfs2_jdata_writepages(struct address_space *mapping,
390 struct writeback_control *wbc)
392 struct gfs2_inode *ip = GFS2_I(mapping->host);
393 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
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);
406 * stuffed_readpage - Fill in a Linux page with stuffed file data
412 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
414 struct buffer_head *dibh;
415 u64 dsize = i_size_read(&ip->i_inode);
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.
424 if (unlikely(page->index)) {
425 zero_user(page, 0, PAGE_SIZE);
426 SetPageUptodate(page);
430 error = gfs2_meta_inode_buffer(ip, &dibh);
434 kaddr = kmap_atomic(page);
435 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
436 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
437 kunmap_atomic(kaddr);
438 flush_dcache_page(page);
440 SetPageUptodate(page);
446 * gfs2_read_folio - read a folio from a file
447 * @file: The file to read
448 * @folio: The folio in the file
450 static int gfs2_read_folio(struct file *file, struct folio *folio)
452 struct inode *inode = folio->mapping->host;
453 struct gfs2_inode *ip = GFS2_I(inode);
454 struct gfs2_sbd *sdp = GFS2_SB(inode);
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);
464 error = mpage_read_folio(folio, gfs2_block_map);
467 if (unlikely(gfs2_withdrawn(sdp)))
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
482 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
485 struct address_space *mapping = ip->i_inode.i_mapping;
486 unsigned long index = *pos >> PAGE_SHIFT;
487 unsigned offset = *pos & (PAGE_SIZE - 1);
495 if (offset + size > PAGE_SIZE)
496 amt = PAGE_SIZE - offset;
497 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
499 return PTR_ERR(page);
500 p = kmap_atomic(page);
501 memcpy(buf + copied, p + offset, amt);
507 } while(copied < size);
513 * gfs2_readahead - Read a bunch of pages at once
514 * @rac: Read-ahead control structure
517 * 1. This is only for readahead, so we can simply ignore any things
518 * which are slightly inconvenient (such as locking conflicts between
519 * the page lock and the glock) and return having done no I/O. Its
520 * obviously not something we'd want to do on too regular a basis.
521 * Any I/O we ignore at this time will be done via readpage later.
522 * 2. We don't handle stuffed files here we let readpage do the honours.
523 * 3. mpage_readahead() does most of the heavy lifting in the common case.
524 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
527 static void gfs2_readahead(struct readahead_control *rac)
529 struct inode *inode = rac->mapping->host;
530 struct gfs2_inode *ip = GFS2_I(inode);
532 if (gfs2_is_stuffed(ip))
534 else if (gfs2_is_jdata(ip))
535 mpage_readahead(rac, gfs2_block_map);
537 iomap_readahead(rac, &gfs2_iomap_ops);
541 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
542 * @inode: the rindex inode
544 void adjust_fs_space(struct inode *inode)
546 struct gfs2_sbd *sdp = GFS2_SB(inode);
547 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
548 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
549 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
550 struct buffer_head *m_bh;
551 u64 fs_total, new_free;
553 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
556 /* Total up the file system space, according to the latest rindex. */
557 fs_total = gfs2_ri_total(sdp);
558 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
561 spin_lock(&sdp->sd_statfs_spin);
562 gfs2_statfs_change_in(m_sc, m_bh->b_data +
563 sizeof(struct gfs2_dinode));
564 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
565 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
568 spin_unlock(&sdp->sd_statfs_spin);
569 fs_warn(sdp, "File system extended by %llu blocks.\n",
570 (unsigned long long)new_free);
571 gfs2_statfs_change(sdp, new_free, new_free, 0);
573 update_statfs(sdp, m_bh);
576 sdp->sd_rindex_uptodate = 0;
580 static bool jdata_dirty_folio(struct address_space *mapping,
583 if (current->journal_info)
584 folio_set_checked(folio);
585 return block_dirty_folio(mapping, folio);
589 * gfs2_bmap - Block map function
590 * @mapping: Address space info
591 * @lblock: The block to map
593 * Returns: The disk address for the block or 0 on hole or error
596 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
598 struct gfs2_inode *ip = GFS2_I(mapping->host);
599 struct gfs2_holder i_gh;
603 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
607 if (!gfs2_is_stuffed(ip))
608 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
610 gfs2_glock_dq_uninit(&i_gh);
615 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
617 struct gfs2_bufdata *bd;
621 clear_buffer_dirty(bh);
624 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
625 list_del_init(&bd->bd_list);
627 spin_lock(&sdp->sd_ail_lock);
628 gfs2_remove_from_journal(bh, REMOVE_JDATA);
629 spin_unlock(&sdp->sd_ail_lock);
633 clear_buffer_mapped(bh);
634 clear_buffer_req(bh);
635 clear_buffer_new(bh);
636 gfs2_log_unlock(sdp);
640 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
643 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
644 size_t stop = offset + length;
645 int partial_page = (offset || length < folio_size(folio));
646 struct buffer_head *bh, *head;
647 unsigned long pos = 0;
649 BUG_ON(!folio_test_locked(folio));
651 folio_clear_checked(folio);
652 head = folio_buffers(folio);
658 if (pos + bh->b_size > stop)
662 gfs2_discard(sdp, bh);
664 bh = bh->b_this_page;
665 } while (bh != head);
668 filemap_release_folio(folio, 0);
672 * gfs2_release_folio - free the metadata associated with a folio
673 * @folio: the folio that's being released
674 * @gfp_mask: passed from Linux VFS, ignored by us
676 * Calls try_to_free_buffers() to free the buffers and put the folio if the
677 * buffers can be released.
679 * Returns: true if the folio was put or else false
682 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
684 struct address_space *mapping = folio->mapping;
685 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
686 struct buffer_head *bh, *head;
687 struct gfs2_bufdata *bd;
689 head = folio_buffers(folio);
694 * mm accommodates an old ext3 case where clean folios might
695 * not have had the dirty bit cleared. Thus, it can send actual
696 * dirty folios to ->release_folio() via shrink_active_list().
698 * As a workaround, we skip folios that contain dirty buffers
699 * below. Once ->release_folio isn't called on dirty folios
700 * anymore, we can warn on dirty buffers like we used to here
707 if (atomic_read(&bh->b_count))
712 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
714 bh = bh->b_this_page;
715 } while (bh != head);
721 gfs2_assert_warn(sdp, bd->bd_bh == bh);
723 bh->b_private = NULL;
725 * The bd may still be queued as a revoke, in which
726 * case we must not dequeue nor free it.
728 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
729 list_del_init(&bd->bd_list);
730 if (list_empty(&bd->bd_list))
731 kmem_cache_free(gfs2_bufdata_cachep, bd);
734 bh = bh->b_this_page;
735 } while (bh != head);
736 gfs2_log_unlock(sdp);
738 return try_to_free_buffers(folio);
741 gfs2_log_unlock(sdp);
745 static const struct address_space_operations gfs2_aops = {
746 .writepages = gfs2_writepages,
747 .read_folio = gfs2_read_folio,
748 .readahead = gfs2_readahead,
749 .dirty_folio = filemap_dirty_folio,
750 .release_folio = iomap_release_folio,
751 .invalidate_folio = iomap_invalidate_folio,
753 .direct_IO = noop_direct_IO,
754 .migrate_folio = filemap_migrate_folio,
755 .is_partially_uptodate = iomap_is_partially_uptodate,
756 .error_remove_page = generic_error_remove_page,
759 static const struct address_space_operations gfs2_jdata_aops = {
760 .writepage = gfs2_jdata_writepage,
761 .writepages = gfs2_jdata_writepages,
762 .read_folio = gfs2_read_folio,
763 .readahead = gfs2_readahead,
764 .dirty_folio = jdata_dirty_folio,
766 .invalidate_folio = gfs2_invalidate_folio,
767 .release_folio = gfs2_release_folio,
768 .is_partially_uptodate = block_is_partially_uptodate,
769 .error_remove_page = generic_error_remove_page,
772 void gfs2_set_aops(struct inode *inode)
774 if (gfs2_is_jdata(GFS2_I(inode)))
775 inode->i_mapping->a_ops = &gfs2_jdata_aops;
777 inode->i_mapping->a_ops = &gfs2_aops;