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);
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_page_add_databufs(ip, 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 enough 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);
191 if (ret == 0 && wbc->nr_to_write > 0)
192 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
197 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
198 * @mapping: The mapping
199 * @wbc: The writeback control
200 * @pvec: The vector of pages
201 * @nr_pages: The number of pages to write
202 * @done_index: Page index
204 * Returns: non-zero if loop should terminate, zero otherwise
207 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
208 struct writeback_control *wbc,
209 struct pagevec *pvec,
213 struct inode *inode = mapping->host;
214 struct gfs2_sbd *sdp = GFS2_SB(inode);
215 unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
219 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
223 for(i = 0; i < nr_pages; i++) {
224 struct page *page = pvec->pages[i];
226 *done_index = page->index;
230 if (unlikely(page->mapping != mapping)) {
236 if (!PageDirty(page)) {
237 /* someone wrote it for us */
238 goto continue_unlock;
241 if (PageWriteback(page)) {
242 if (wbc->sync_mode != WB_SYNC_NONE)
243 wait_on_page_writeback(page);
245 goto continue_unlock;
248 BUG_ON(PageWriteback(page));
249 if (!clear_page_dirty_for_io(page))
250 goto continue_unlock;
252 trace_wbc_writepage(wbc, inode_to_bdi(inode));
254 ret = __gfs2_jdata_writepage(page, wbc);
256 if (ret == AOP_WRITEPAGE_ACTIVATE) {
262 * done_index is set past this page,
263 * so media errors will not choke
264 * background writeout for the entire
265 * file. This has consequences for
266 * range_cyclic semantics (ie. it may
267 * not be suitable for data integrity
270 *done_index = page->index + 1;
277 * We stop writing back only if we are not doing
278 * integrity sync. In case of integrity sync we have to
279 * keep going until we have written all the pages
280 * we tagged for writeback prior to entering this loop.
282 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
293 * gfs2_write_cache_jdata - Like write_cache_pages but different
294 * @mapping: The mapping to write
295 * @wbc: The writeback control
297 * The reason that we use our own function here is that we need to
298 * start transactions before we grab page locks. This allows us
299 * to get the ordering right.
302 static int gfs2_write_cache_jdata(struct address_space *mapping,
303 struct writeback_control *wbc)
309 pgoff_t writeback_index;
318 if (wbc->range_cyclic) {
319 writeback_index = mapping->writeback_index; /* prev offset */
320 index = writeback_index;
327 index = wbc->range_start >> PAGE_SHIFT;
328 end = wbc->range_end >> PAGE_SHIFT;
329 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
331 cycled = 1; /* ignore range_cyclic tests */
333 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
334 tag = PAGECACHE_TAG_TOWRITE;
336 tag = PAGECACHE_TAG_DIRTY;
339 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
340 tag_pages_for_writeback(mapping, index, end);
342 while (!done && (index <= end)) {
343 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
348 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
353 pagevec_release(&pvec);
357 if (!cycled && !done) {
360 * We hit the last page and there is more work to be done: wrap
361 * back to the start of the file
365 end = writeback_index - 1;
369 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
370 mapping->writeback_index = done_index;
377 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
378 * @mapping: The mapping to write
379 * @wbc: The writeback control
383 static int gfs2_jdata_writepages(struct address_space *mapping,
384 struct writeback_control *wbc)
386 struct gfs2_inode *ip = GFS2_I(mapping->host);
387 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
390 ret = gfs2_write_cache_jdata(mapping, wbc);
391 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
392 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
393 GFS2_LFC_JDATA_WPAGES);
394 ret = gfs2_write_cache_jdata(mapping, wbc);
400 * stuffed_readpage - Fill in a Linux page with stuffed file data
406 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
408 struct buffer_head *dibh;
409 u64 dsize = i_size_read(&ip->i_inode);
414 * Due to the order of unstuffing files and ->fault(), we can be
415 * asked for a zero page in the case of a stuffed file being extended,
416 * so we need to supply one here. It doesn't happen often.
418 if (unlikely(page->index)) {
419 zero_user(page, 0, PAGE_SIZE);
420 SetPageUptodate(page);
424 error = gfs2_meta_inode_buffer(ip, &dibh);
428 kaddr = kmap_atomic(page);
429 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
430 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
431 kunmap_atomic(kaddr);
432 flush_dcache_page(page);
434 SetPageUptodate(page);
440 * gfs2_read_folio - read a folio from a file
441 * @file: The file to read
442 * @folio: The folio in the file
444 static int gfs2_read_folio(struct file *file, struct folio *folio)
446 struct inode *inode = folio->mapping->host;
447 struct gfs2_inode *ip = GFS2_I(inode);
448 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 if (!gfs2_is_jdata(ip) ||
452 (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
453 error = iomap_read_folio(folio, &gfs2_iomap_ops);
454 } else if (gfs2_is_stuffed(ip)) {
455 error = stuffed_readpage(ip, &folio->page);
458 error = mpage_read_folio(folio, gfs2_block_map);
461 if (unlikely(gfs2_withdrawn(sdp)))
468 * gfs2_internal_read - read an internal file
469 * @ip: The gfs2 inode
470 * @buf: The buffer to fill
471 * @pos: The file position
472 * @size: The amount to read
476 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
479 struct address_space *mapping = ip->i_inode.i_mapping;
480 unsigned long index = *pos >> PAGE_SHIFT;
481 unsigned offset = *pos & (PAGE_SIZE - 1);
489 if (offset + size > PAGE_SIZE)
490 amt = PAGE_SIZE - offset;
491 page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
493 return PTR_ERR(page);
494 p = kmap_atomic(page);
495 memcpy(buf + copied, p + offset, amt);
501 } while(copied < size);
507 * gfs2_readahead - Read a bunch of pages at once
508 * @rac: Read-ahead control structure
511 * 1. This is only for readahead, so we can simply ignore any things
512 * which are slightly inconvenient (such as locking conflicts between
513 * the page lock and the glock) and return having done no I/O. Its
514 * obviously not something we'd want to do on too regular a basis.
515 * Any I/O we ignore at this time will be done via readpage later.
516 * 2. We don't handle stuffed files here we let readpage do the honours.
517 * 3. mpage_readahead() does most of the heavy lifting in the common case.
518 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
521 static void gfs2_readahead(struct readahead_control *rac)
523 struct inode *inode = rac->mapping->host;
524 struct gfs2_inode *ip = GFS2_I(inode);
526 if (gfs2_is_stuffed(ip))
528 else if (gfs2_is_jdata(ip))
529 mpage_readahead(rac, gfs2_block_map);
531 iomap_readahead(rac, &gfs2_iomap_ops);
535 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
536 * @inode: the rindex inode
538 void adjust_fs_space(struct inode *inode)
540 struct gfs2_sbd *sdp = GFS2_SB(inode);
541 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
542 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
543 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
544 struct buffer_head *m_bh;
545 u64 fs_total, new_free;
547 if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
550 /* Total up the file system space, according to the latest rindex. */
551 fs_total = gfs2_ri_total(sdp);
552 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
555 spin_lock(&sdp->sd_statfs_spin);
556 gfs2_statfs_change_in(m_sc, m_bh->b_data +
557 sizeof(struct gfs2_dinode));
558 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
559 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
562 spin_unlock(&sdp->sd_statfs_spin);
563 fs_warn(sdp, "File system extended by %llu blocks.\n",
564 (unsigned long long)new_free);
565 gfs2_statfs_change(sdp, new_free, new_free, 0);
567 update_statfs(sdp, m_bh);
570 sdp->sd_rindex_uptodate = 0;
574 static bool jdata_dirty_folio(struct address_space *mapping,
577 if (current->journal_info)
578 folio_set_checked(folio);
579 return block_dirty_folio(mapping, folio);
583 * gfs2_bmap - Block map function
584 * @mapping: Address space info
585 * @lblock: The block to map
587 * Returns: The disk address for the block or 0 on hole or error
590 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
592 struct gfs2_inode *ip = GFS2_I(mapping->host);
593 struct gfs2_holder i_gh;
597 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
601 if (!gfs2_is_stuffed(ip))
602 dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
604 gfs2_glock_dq_uninit(&i_gh);
609 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
611 struct gfs2_bufdata *bd;
615 clear_buffer_dirty(bh);
618 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
619 list_del_init(&bd->bd_list);
621 spin_lock(&sdp->sd_ail_lock);
622 gfs2_remove_from_journal(bh, REMOVE_JDATA);
623 spin_unlock(&sdp->sd_ail_lock);
627 clear_buffer_mapped(bh);
628 clear_buffer_req(bh);
629 clear_buffer_new(bh);
630 gfs2_log_unlock(sdp);
634 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
637 struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
638 size_t stop = offset + length;
639 int partial_page = (offset || length < folio_size(folio));
640 struct buffer_head *bh, *head;
641 unsigned long pos = 0;
643 BUG_ON(!folio_test_locked(folio));
645 folio_clear_checked(folio);
646 head = folio_buffers(folio);
652 if (pos + bh->b_size > stop)
656 gfs2_discard(sdp, bh);
658 bh = bh->b_this_page;
659 } while (bh != head);
662 filemap_release_folio(folio, 0);
666 * gfs2_release_folio - free the metadata associated with a folio
667 * @folio: the folio that's being released
668 * @gfp_mask: passed from Linux VFS, ignored by us
670 * Calls try_to_free_buffers() to free the buffers and put the folio if the
671 * buffers can be released.
673 * Returns: true if the folio was put or else false
676 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
678 struct address_space *mapping = folio->mapping;
679 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
680 struct buffer_head *bh, *head;
681 struct gfs2_bufdata *bd;
683 head = folio_buffers(folio);
688 * mm accommodates an old ext3 case where clean folios might
689 * not have had the dirty bit cleared. Thus, it can send actual
690 * dirty folios to ->release_folio() via shrink_active_list().
692 * As a workaround, we skip folios that contain dirty buffers
693 * below. Once ->release_folio isn't called on dirty folios
694 * anymore, we can warn on dirty buffers like we used to here
701 if (atomic_read(&bh->b_count))
706 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
708 bh = bh->b_this_page;
709 } while (bh != head);
715 gfs2_assert_warn(sdp, bd->bd_bh == bh);
717 bh->b_private = NULL;
719 * The bd may still be queued as a revoke, in which
720 * case we must not dequeue nor free it.
722 if (!bd->bd_blkno && !list_empty(&bd->bd_list))
723 list_del_init(&bd->bd_list);
724 if (list_empty(&bd->bd_list))
725 kmem_cache_free(gfs2_bufdata_cachep, bd);
728 bh = bh->b_this_page;
729 } while (bh != head);
730 gfs2_log_unlock(sdp);
732 return try_to_free_buffers(folio);
735 gfs2_log_unlock(sdp);
739 static const struct address_space_operations gfs2_aops = {
740 .writepages = gfs2_writepages,
741 .read_folio = gfs2_read_folio,
742 .readahead = gfs2_readahead,
743 .dirty_folio = filemap_dirty_folio,
744 .release_folio = iomap_release_folio,
745 .invalidate_folio = iomap_invalidate_folio,
747 .direct_IO = noop_direct_IO,
748 .migrate_folio = filemap_migrate_folio,
749 .is_partially_uptodate = iomap_is_partially_uptodate,
750 .error_remove_page = generic_error_remove_page,
753 static const struct address_space_operations gfs2_jdata_aops = {
754 .writepage = gfs2_jdata_writepage,
755 .writepages = gfs2_jdata_writepages,
756 .read_folio = gfs2_read_folio,
757 .readahead = gfs2_readahead,
758 .dirty_folio = jdata_dirty_folio,
760 .invalidate_folio = gfs2_invalidate_folio,
761 .release_folio = gfs2_release_folio,
762 .is_partially_uptodate = block_is_partially_uptodate,
763 .error_remove_page = generic_error_remove_page,
766 void gfs2_set_aops(struct inode *inode)
768 if (gfs2_is_jdata(GFS2_I(inode)))
769 inode->i_mapping->a_ops = &gfs2_jdata_aops;
771 inode->i_mapping->a_ops = &gfs2_aops;