2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs_shared.h"
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_trans_resv.h"
23 #include "xfs_mount.h"
24 #include "xfs_inode.h"
25 #include "xfs_trans.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_alloc.h"
28 #include "xfs_error.h"
29 #include "xfs_iomap.h"
30 #include "xfs_trace.h"
32 #include "xfs_bmap_util.h"
33 #include "xfs_bmap_btree.h"
34 #include <linux/aio.h>
35 #include <linux/gfp.h>
36 #include <linux/mpage.h>
37 #include <linux/pagevec.h>
38 #include <linux/writeback.h>
46 struct buffer_head *bh, *head;
48 *delalloc = *unwritten = 0;
50 bh = head = page_buffers(page);
52 if (buffer_unwritten(bh))
54 else if (buffer_delay(bh))
56 } while ((bh = bh->b_this_page) != head);
59 STATIC struct block_device *
60 xfs_find_bdev_for_inode(
63 struct xfs_inode *ip = XFS_I(inode);
64 struct xfs_mount *mp = ip->i_mount;
66 if (XFS_IS_REALTIME_INODE(ip))
67 return mp->m_rtdev_targp->bt_bdev;
69 return mp->m_ddev_targp->bt_bdev;
73 * We're now finished for good with this ioend structure.
74 * Update the page state via the associated buffer_heads,
75 * release holds on the inode and bio, and finally free
76 * up memory. Do not use the ioend after this.
82 struct buffer_head *bh, *next;
84 for (bh = ioend->io_buffer_head; bh; bh = next) {
86 bh->b_end_io(bh, !ioend->io_error);
89 mempool_free(ioend, xfs_ioend_pool);
93 * Fast and loose check if this write could update the on-disk inode size.
95 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
97 return ioend->io_offset + ioend->io_size >
98 XFS_I(ioend->io_inode)->i_d.di_size;
102 xfs_setfilesize_trans_alloc(
103 struct xfs_ioend *ioend)
105 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
106 struct xfs_trans *tp;
109 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
111 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
113 xfs_trans_cancel(tp, 0);
117 ioend->io_append_trans = tp;
120 * We may pass freeze protection with a transaction. So tell lockdep
123 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
126 * We hand off the transaction to the completion thread now, so
127 * clear the flag here.
129 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
134 * Update on-disk file size now that data has been written to disk.
138 struct xfs_ioend *ioend)
140 struct xfs_inode *ip = XFS_I(ioend->io_inode);
141 struct xfs_trans *tp = ioend->io_append_trans;
145 * The transaction may have been allocated in the I/O submission thread,
146 * thus we need to mark ourselves as beeing in a transaction manually.
147 * Similarly for freeze protection.
149 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
150 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
153 xfs_ilock(ip, XFS_ILOCK_EXCL);
154 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
156 xfs_iunlock(ip, XFS_ILOCK_EXCL);
157 xfs_trans_cancel(tp, 0);
161 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
163 ip->i_d.di_size = isize;
164 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
165 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
167 return xfs_trans_commit(tp, 0);
171 * Schedule IO completion handling on the final put of an ioend.
173 * If there is no work to do we might as well call it a day and free the
178 struct xfs_ioend *ioend)
180 if (atomic_dec_and_test(&ioend->io_remaining)) {
181 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
183 if (ioend->io_type == XFS_IO_UNWRITTEN)
184 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
185 else if (ioend->io_append_trans ||
186 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
187 queue_work(mp->m_data_workqueue, &ioend->io_work);
189 xfs_destroy_ioend(ioend);
194 * IO write completion.
198 struct work_struct *work)
200 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
201 struct xfs_inode *ip = XFS_I(ioend->io_inode);
204 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
205 ioend->io_error = -EIO;
212 * For unwritten extents we need to issue transactions to convert a
213 * range to normal written extens after the data I/O has finished.
215 if (ioend->io_type == XFS_IO_UNWRITTEN) {
216 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
218 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
220 * For direct I/O we do not know if we need to allocate blocks
221 * or not so we can't preallocate an append transaction as that
222 * results in nested reservations and log space deadlocks. Hence
223 * allocate the transaction here. While this is sub-optimal and
224 * can block IO completion for some time, we're stuck with doing
225 * it this way until we can pass the ioend to the direct IO
226 * allocation callbacks and avoid nesting that way.
228 error = xfs_setfilesize_trans_alloc(ioend);
231 error = xfs_setfilesize(ioend);
232 } else if (ioend->io_append_trans) {
233 error = xfs_setfilesize(ioend);
235 ASSERT(!xfs_ioend_is_append(ioend));
240 ioend->io_error = error;
241 xfs_destroy_ioend(ioend);
245 * Call IO completion handling in caller context on the final put of an ioend.
248 xfs_finish_ioend_sync(
249 struct xfs_ioend *ioend)
251 if (atomic_dec_and_test(&ioend->io_remaining))
252 xfs_end_io(&ioend->io_work);
256 * Allocate and initialise an IO completion structure.
257 * We need to track unwritten extent write completion here initially.
258 * We'll need to extend this for updating the ondisk inode size later
268 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
271 * Set the count to 1 initially, which will prevent an I/O
272 * completion callback from happening before we have started
273 * all the I/O from calling the completion routine too early.
275 atomic_set(&ioend->io_remaining, 1);
276 ioend->io_isdirect = 0;
278 ioend->io_list = NULL;
279 ioend->io_type = type;
280 ioend->io_inode = inode;
281 ioend->io_buffer_head = NULL;
282 ioend->io_buffer_tail = NULL;
283 ioend->io_offset = 0;
285 ioend->io_append_trans = NULL;
287 INIT_WORK(&ioend->io_work, xfs_end_io);
295 struct xfs_bmbt_irec *imap,
299 struct xfs_inode *ip = XFS_I(inode);
300 struct xfs_mount *mp = ip->i_mount;
301 ssize_t count = 1 << inode->i_blkbits;
302 xfs_fileoff_t offset_fsb, end_fsb;
304 int bmapi_flags = XFS_BMAPI_ENTIRE;
307 if (XFS_FORCED_SHUTDOWN(mp))
310 if (type == XFS_IO_UNWRITTEN)
311 bmapi_flags |= XFS_BMAPI_IGSTATE;
313 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
316 xfs_ilock(ip, XFS_ILOCK_SHARED);
319 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
320 (ip->i_df.if_flags & XFS_IFEXTENTS));
321 ASSERT(offset <= mp->m_super->s_maxbytes);
323 if (offset + count > mp->m_super->s_maxbytes)
324 count = mp->m_super->s_maxbytes - offset;
325 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
326 offset_fsb = XFS_B_TO_FSBT(mp, offset);
327 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
328 imap, &nimaps, bmapi_flags);
329 xfs_iunlock(ip, XFS_ILOCK_SHARED);
334 if (type == XFS_IO_DELALLOC &&
335 (!nimaps || isnullstartblock(imap->br_startblock))) {
336 error = xfs_iomap_write_allocate(ip, offset, imap);
338 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
343 if (type == XFS_IO_UNWRITTEN) {
345 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
346 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
350 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
357 struct xfs_bmbt_irec *imap,
360 offset >>= inode->i_blkbits;
362 return offset >= imap->br_startoff &&
363 offset < imap->br_startoff + imap->br_blockcount;
367 * BIO completion handler for buffered IO.
374 xfs_ioend_t *ioend = bio->bi_private;
376 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
377 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
379 /* Toss bio and pass work off to an xfsdatad thread */
380 bio->bi_private = NULL;
381 bio->bi_end_io = NULL;
384 xfs_finish_ioend(ioend);
388 xfs_submit_ioend_bio(
389 struct writeback_control *wbc,
393 atomic_inc(&ioend->io_remaining);
394 bio->bi_private = ioend;
395 bio->bi_end_io = xfs_end_bio;
396 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
401 struct buffer_head *bh)
403 int nvecs = bio_get_nr_vecs(bh->b_bdev);
404 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
406 ASSERT(bio->bi_private == NULL);
407 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
408 bio->bi_bdev = bh->b_bdev;
413 xfs_start_buffer_writeback(
414 struct buffer_head *bh)
416 ASSERT(buffer_mapped(bh));
417 ASSERT(buffer_locked(bh));
418 ASSERT(!buffer_delay(bh));
419 ASSERT(!buffer_unwritten(bh));
421 mark_buffer_async_write(bh);
422 set_buffer_uptodate(bh);
423 clear_buffer_dirty(bh);
427 xfs_start_page_writeback(
432 ASSERT(PageLocked(page));
433 ASSERT(!PageWriteback(page));
436 * if the page was not fully cleaned, we need to ensure that the higher
437 * layers come back to it correctly. That means we need to keep the page
438 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
439 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
440 * write this page in this writeback sweep will be made.
443 clear_page_dirty_for_io(page);
444 set_page_writeback(page);
446 set_page_writeback_keepwrite(page);
450 /* If no buffers on the page are to be written, finish it here */
452 end_page_writeback(page);
455 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
457 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
461 * Submit all of the bios for all of the ioends we have saved up, covering the
462 * initial writepage page and also any probed pages.
464 * Because we may have multiple ioends spanning a page, we need to start
465 * writeback on all the buffers before we submit them for I/O. If we mark the
466 * buffers as we got, then we can end up with a page that only has buffers
467 * marked async write and I/O complete on can occur before we mark the other
468 * buffers async write.
470 * The end result of this is that we trip a bug in end_page_writeback() because
471 * we call it twice for the one page as the code in end_buffer_async_write()
472 * assumes that all buffers on the page are started at the same time.
474 * The fix is two passes across the ioend list - one to start writeback on the
475 * buffer_heads, and then submit them for I/O on the second pass.
477 * If @fail is non-zero, it means that we have a situation where some part of
478 * the submission process has failed after we have marked paged for writeback
479 * and unlocked them. In this situation, we need to fail the ioend chain rather
480 * than submit it to IO. This typically only happens on a filesystem shutdown.
484 struct writeback_control *wbc,
488 xfs_ioend_t *head = ioend;
490 struct buffer_head *bh;
492 sector_t lastblock = 0;
494 /* Pass 1 - start writeback */
496 next = ioend->io_list;
497 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
498 xfs_start_buffer_writeback(bh);
499 } while ((ioend = next) != NULL);
501 /* Pass 2 - submit I/O */
504 next = ioend->io_list;
508 * If we are failing the IO now, just mark the ioend with an
509 * error and finish it. This will run IO completion immediately
510 * as there is only one reference to the ioend at this point in
514 ioend->io_error = fail;
515 xfs_finish_ioend(ioend);
519 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
523 bio = xfs_alloc_ioend_bio(bh);
524 } else if (bh->b_blocknr != lastblock + 1) {
525 xfs_submit_ioend_bio(wbc, ioend, bio);
529 if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
530 xfs_submit_ioend_bio(wbc, ioend, bio);
534 lastblock = bh->b_blocknr;
537 xfs_submit_ioend_bio(wbc, ioend, bio);
538 xfs_finish_ioend(ioend);
539 } while ((ioend = next) != NULL);
543 * Cancel submission of all buffer_heads so far in this endio.
544 * Toss the endio too. Only ever called for the initial page
545 * in a writepage request, so only ever one page.
552 struct buffer_head *bh, *next_bh;
555 next = ioend->io_list;
556 bh = ioend->io_buffer_head;
558 next_bh = bh->b_private;
559 clear_buffer_async_write(bh);
561 * The unwritten flag is cleared when added to the
562 * ioend. We're not submitting for I/O so mark the
563 * buffer unwritten again for next time around.
565 if (ioend->io_type == XFS_IO_UNWRITTEN)
566 set_buffer_unwritten(bh);
568 } while ((bh = next_bh) != NULL);
570 mempool_free(ioend, xfs_ioend_pool);
571 } while ((ioend = next) != NULL);
575 * Test to see if we've been building up a completion structure for
576 * earlier buffers -- if so, we try to append to this ioend if we
577 * can, otherwise we finish off any current ioend and start another.
578 * Return true if we've finished the given ioend.
583 struct buffer_head *bh,
586 xfs_ioend_t **result,
589 xfs_ioend_t *ioend = *result;
591 if (!ioend || need_ioend || type != ioend->io_type) {
592 xfs_ioend_t *previous = *result;
594 ioend = xfs_alloc_ioend(inode, type);
595 ioend->io_offset = offset;
596 ioend->io_buffer_head = bh;
597 ioend->io_buffer_tail = bh;
599 previous->io_list = ioend;
602 ioend->io_buffer_tail->b_private = bh;
603 ioend->io_buffer_tail = bh;
606 bh->b_private = NULL;
607 ioend->io_size += bh->b_size;
613 struct buffer_head *bh,
614 struct xfs_bmbt_irec *imap,
618 struct xfs_mount *m = XFS_I(inode)->i_mount;
619 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
620 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
622 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
623 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
625 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
626 ((offset - iomap_offset) >> inode->i_blkbits);
628 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
631 set_buffer_mapped(bh);
637 struct buffer_head *bh,
638 struct xfs_bmbt_irec *imap,
641 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
642 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
644 xfs_map_buffer(inode, bh, imap, offset);
645 set_buffer_mapped(bh);
646 clear_buffer_delay(bh);
647 clear_buffer_unwritten(bh);
651 * Test if a given page contains at least one buffer of a given @type.
652 * If @check_all_buffers is true, then we walk all the buffers in the page to
653 * try to find one of the type passed in. If it is not set, then the caller only
654 * needs to check the first buffer on the page for a match.
660 bool check_all_buffers)
662 struct buffer_head *bh;
663 struct buffer_head *head;
665 if (PageWriteback(page))
669 if (!page_has_buffers(page))
672 bh = head = page_buffers(page);
674 if (buffer_unwritten(bh)) {
675 if (type == XFS_IO_UNWRITTEN)
677 } else if (buffer_delay(bh)) {
678 if (type == XFS_IO_DELALLOC)
680 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
681 if (type == XFS_IO_OVERWRITE)
685 /* If we are only checking the first buffer, we are done now. */
686 if (!check_all_buffers)
688 } while ((bh = bh->b_this_page) != head);
694 * Allocate & map buffers for page given the extent map. Write it out.
695 * except for the original page of a writepage, this is called on
696 * delalloc/unwritten pages only, for the original page it is possible
697 * that the page has no mapping at all.
704 struct xfs_bmbt_irec *imap,
705 xfs_ioend_t **ioendp,
706 struct writeback_control *wbc)
708 struct buffer_head *bh, *head;
709 xfs_off_t end_offset;
710 unsigned long p_offset;
713 int count = 0, done = 0, uptodate = 1;
714 xfs_off_t offset = page_offset(page);
716 if (page->index != tindex)
718 if (!trylock_page(page))
720 if (PageWriteback(page))
721 goto fail_unlock_page;
722 if (page->mapping != inode->i_mapping)
723 goto fail_unlock_page;
724 if (!xfs_check_page_type(page, (*ioendp)->io_type, false))
725 goto fail_unlock_page;
728 * page_dirty is initially a count of buffers on the page before
729 * EOF and is decremented as we move each into a cleanable state.
733 * End offset is the highest offset that this page should represent.
734 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
735 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
736 * hence give us the correct page_dirty count. On any other page,
737 * it will be zero and in that case we need page_dirty to be the
738 * count of buffers on the page.
740 end_offset = min_t(unsigned long long,
741 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
745 * If the current map does not span the entire page we are about to try
746 * to write, then give up. The only way we can write a page that spans
747 * multiple mappings in a single writeback iteration is via the
748 * xfs_vm_writepage() function. Data integrity writeback requires the
749 * entire page to be written in a single attempt, otherwise the part of
750 * the page we don't write here doesn't get written as part of the data
753 * For normal writeback, we also don't attempt to write partial pages
754 * here as it simply means that write_cache_pages() will see it under
755 * writeback and ignore the page until some point in the future, at
756 * which time this will be the only page in the file that needs
757 * writeback. Hence for more optimal IO patterns, we should always
758 * avoid partial page writeback due to multiple mappings on a page here.
760 if (!xfs_imap_valid(inode, imap, end_offset))
761 goto fail_unlock_page;
763 len = 1 << inode->i_blkbits;
764 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
766 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
767 page_dirty = p_offset / len;
770 * The moment we find a buffer that doesn't match our current type
771 * specification or can't be written, abort the loop and start
772 * writeback. As per the above xfs_imap_valid() check, only
773 * xfs_vm_writepage() can handle partial page writeback fully - we are
774 * limited here to the buffers that are contiguous with the current
775 * ioend, and hence a buffer we can't write breaks that contiguity and
776 * we have to defer the rest of the IO to xfs_vm_writepage().
778 bh = head = page_buffers(page);
780 if (offset >= end_offset)
782 if (!buffer_uptodate(bh))
784 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
789 if (buffer_unwritten(bh) || buffer_delay(bh) ||
791 if (buffer_unwritten(bh))
792 type = XFS_IO_UNWRITTEN;
793 else if (buffer_delay(bh))
794 type = XFS_IO_DELALLOC;
796 type = XFS_IO_OVERWRITE;
799 * imap should always be valid because of the above
800 * partial page end_offset check on the imap.
802 ASSERT(xfs_imap_valid(inode, imap, offset));
805 if (type != XFS_IO_OVERWRITE)
806 xfs_map_at_offset(inode, bh, imap, offset);
807 xfs_add_to_ioend(inode, bh, offset, type,
816 } while (offset += len, (bh = bh->b_this_page) != head);
818 if (uptodate && bh == head)
819 SetPageUptodate(page);
822 if (--wbc->nr_to_write <= 0 &&
823 wbc->sync_mode == WB_SYNC_NONE)
826 xfs_start_page_writeback(page, !page_dirty, count);
836 * Convert & write out a cluster of pages in the same extent as defined
837 * by mp and following the start page.
843 struct xfs_bmbt_irec *imap,
844 xfs_ioend_t **ioendp,
845 struct writeback_control *wbc,
851 pagevec_init(&pvec, 0);
852 while (!done && tindex <= tlast) {
853 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
855 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
858 for (i = 0; i < pagevec_count(&pvec); i++) {
859 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
865 pagevec_release(&pvec);
871 xfs_vm_invalidatepage(
876 trace_xfs_invalidatepage(page->mapping->host, page, offset,
878 block_invalidatepage(page, offset, length);
882 * If the page has delalloc buffers on it, we need to punch them out before we
883 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
884 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
885 * is done on that same region - the delalloc extent is returned when none is
886 * supposed to be there.
888 * We prevent this by truncating away the delalloc regions on the page before
889 * invalidating it. Because they are delalloc, we can do this without needing a
890 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
891 * truncation without a transaction as there is no space left for block
892 * reservation (typically why we see a ENOSPC in writeback).
894 * This is not a performance critical path, so for now just do the punching a
895 * buffer head at a time.
898 xfs_aops_discard_page(
901 struct inode *inode = page->mapping->host;
902 struct xfs_inode *ip = XFS_I(inode);
903 struct buffer_head *bh, *head;
904 loff_t offset = page_offset(page);
906 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
909 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
912 xfs_alert(ip->i_mount,
913 "page discard on page %p, inode 0x%llx, offset %llu.",
914 page, ip->i_ino, offset);
916 xfs_ilock(ip, XFS_ILOCK_EXCL);
917 bh = head = page_buffers(page);
920 xfs_fileoff_t start_fsb;
922 if (!buffer_delay(bh))
925 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
926 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
928 /* something screwed, just bail */
929 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
930 xfs_alert(ip->i_mount,
931 "page discard unable to remove delalloc mapping.");
936 offset += 1 << inode->i_blkbits;
938 } while ((bh = bh->b_this_page) != head);
940 xfs_iunlock(ip, XFS_ILOCK_EXCL);
942 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
947 * Write out a dirty page.
949 * For delalloc space on the page we need to allocate space and flush it.
950 * For unwritten space on the page we need to start the conversion to
951 * regular allocated space.
952 * For any other dirty buffer heads on the page we should flush them.
957 struct writeback_control *wbc)
959 struct inode *inode = page->mapping->host;
960 struct buffer_head *bh, *head;
961 struct xfs_bmbt_irec imap;
962 xfs_ioend_t *ioend = NULL, *iohead = NULL;
965 __uint64_t end_offset;
966 pgoff_t end_index, last_index;
968 int err, imap_valid = 0, uptodate = 1;
972 trace_xfs_writepage(inode, page, 0, 0);
974 ASSERT(page_has_buffers(page));
977 * Refuse to write the page out if we are called from reclaim context.
979 * This avoids stack overflows when called from deeply used stacks in
980 * random callers for direct reclaim or memcg reclaim. We explicitly
981 * allow reclaim from kswapd as the stack usage there is relatively low.
983 * This should never happen except in the case of a VM regression so
986 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
991 * Given that we do not allow direct reclaim to call us, we should
992 * never be called while in a filesystem transaction.
994 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
997 /* Is this page beyond the end of the file? */
998 offset = i_size_read(inode);
999 end_index = offset >> PAGE_CACHE_SHIFT;
1000 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
1003 * The page index is less than the end_index, adjust the end_offset
1004 * to the highest offset that this page should represent.
1005 * -----------------------------------------------------
1006 * | file mapping | <EOF> |
1007 * -----------------------------------------------------
1008 * | Page ... | Page N-2 | Page N-1 | Page N | |
1009 * ^--------------------------------^----------|--------
1010 * | desired writeback range | see else |
1011 * ---------------------------------^------------------|
1013 if (page->index < end_index)
1014 end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT;
1017 * Check whether the page to write out is beyond or straddles
1019 * -------------------------------------------------------
1020 * | file mapping | <EOF> |
1021 * -------------------------------------------------------
1022 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1023 * ^--------------------------------^-----------|---------
1025 * ---------------------------------^-----------|--------|
1027 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
1030 * Skip the page if it is fully outside i_size, e.g. due to a
1031 * truncate operation that is in progress. We must redirty the
1032 * page so that reclaim stops reclaiming it. Otherwise
1033 * xfs_vm_releasepage() is called on it and gets confused.
1035 * Note that the end_index is unsigned long, it would overflow
1036 * if the given offset is greater than 16TB on 32-bit system
1037 * and if we do check the page is fully outside i_size or not
1038 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1039 * will be evaluated to 0. Hence this page will be redirtied
1040 * and be written out repeatedly which would result in an
1041 * infinite loop, the user program that perform this operation
1042 * will hang. Instead, we can verify this situation by checking
1043 * if the page to write is totally beyond the i_size or if it's
1044 * offset is just equal to the EOF.
1046 if (page->index > end_index ||
1047 (page->index == end_index && offset_into_page == 0))
1051 * The page straddles i_size. It must be zeroed out on each
1052 * and every writepage invocation because it may be mmapped.
1053 * "A file is mapped in multiples of the page size. For a file
1054 * that is not a multiple of the page size, the remaining
1055 * memory is zeroed when mapped, and writes to that region are
1056 * not written out to the file."
1058 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
1060 /* Adjust the end_offset to the end of file */
1061 end_offset = offset;
1064 len = 1 << inode->i_blkbits;
1066 bh = head = page_buffers(page);
1067 offset = page_offset(page);
1068 type = XFS_IO_OVERWRITE;
1070 if (wbc->sync_mode == WB_SYNC_NONE)
1076 if (offset >= end_offset)
1078 if (!buffer_uptodate(bh))
1082 * set_page_dirty dirties all buffers in a page, independent
1083 * of their state. The dirty state however is entirely
1084 * meaningless for holes (!mapped && uptodate), so skip
1085 * buffers covering holes here.
1087 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
1092 if (buffer_unwritten(bh)) {
1093 if (type != XFS_IO_UNWRITTEN) {
1094 type = XFS_IO_UNWRITTEN;
1097 } else if (buffer_delay(bh)) {
1098 if (type != XFS_IO_DELALLOC) {
1099 type = XFS_IO_DELALLOC;
1102 } else if (buffer_uptodate(bh)) {
1103 if (type != XFS_IO_OVERWRITE) {
1104 type = XFS_IO_OVERWRITE;
1108 if (PageUptodate(page))
1109 ASSERT(buffer_mapped(bh));
1111 * This buffer is not uptodate and will not be
1112 * written to disk. Ensure that we will put any
1113 * subsequent writeable buffers into a new
1121 imap_valid = xfs_imap_valid(inode, &imap, offset);
1124 * If we didn't have a valid mapping then we need to
1125 * put the new mapping into a separate ioend structure.
1126 * This ensures non-contiguous extents always have
1127 * separate ioends, which is particularly important
1128 * for unwritten extent conversion at I/O completion
1132 err = xfs_map_blocks(inode, offset, &imap, type,
1136 imap_valid = xfs_imap_valid(inode, &imap, offset);
1140 if (type != XFS_IO_OVERWRITE)
1141 xfs_map_at_offset(inode, bh, &imap, offset);
1142 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1150 } while (offset += len, ((bh = bh->b_this_page) != head));
1152 if (uptodate && bh == head)
1153 SetPageUptodate(page);
1155 xfs_start_page_writeback(page, 1, count);
1157 /* if there is no IO to be submitted for this page, we are done */
1164 * Any errors from this point onwards need tobe reported through the IO
1165 * completion path as we have marked the initial page as under writeback
1169 xfs_off_t end_index;
1171 end_index = imap.br_startoff + imap.br_blockcount;
1174 end_index <<= inode->i_blkbits;
1177 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1179 /* check against file size */
1180 if (end_index > last_index)
1181 end_index = last_index;
1183 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1189 * Reserve log space if we might write beyond the on-disk inode size.
1192 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1193 err = xfs_setfilesize_trans_alloc(ioend);
1195 xfs_submit_ioend(wbc, iohead, err);
1201 xfs_cancel_ioend(iohead);
1206 xfs_aops_discard_page(page);
1207 ClearPageUptodate(page);
1212 redirty_page_for_writepage(wbc, page);
1219 struct address_space *mapping,
1220 struct writeback_control *wbc)
1222 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1223 return generic_writepages(mapping, wbc);
1227 * Called to move a page into cleanable state - and from there
1228 * to be released. The page should already be clean. We always
1229 * have buffer heads in this call.
1231 * Returns 1 if the page is ok to release, 0 otherwise.
1238 int delalloc, unwritten;
1240 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1242 xfs_count_page_state(page, &delalloc, &unwritten);
1244 if (WARN_ON_ONCE(delalloc))
1246 if (WARN_ON_ONCE(unwritten))
1249 return try_to_free_buffers(page);
1254 struct inode *inode,
1256 struct buffer_head *bh_result,
1260 struct xfs_inode *ip = XFS_I(inode);
1261 struct xfs_mount *mp = ip->i_mount;
1262 xfs_fileoff_t offset_fsb, end_fsb;
1265 struct xfs_bmbt_irec imap;
1271 if (XFS_FORCED_SHUTDOWN(mp))
1274 offset = (xfs_off_t)iblock << inode->i_blkbits;
1275 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1276 size = bh_result->b_size;
1278 if (!create && direct && offset >= i_size_read(inode))
1282 * Direct I/O is usually done on preallocated files, so try getting
1283 * a block mapping without an exclusive lock first. For buffered
1284 * writes we already have the exclusive iolock anyway, so avoiding
1285 * a lock roundtrip here by taking the ilock exclusive from the
1286 * beginning is a useful micro optimization.
1288 if (create && !direct) {
1289 lockmode = XFS_ILOCK_EXCL;
1290 xfs_ilock(ip, lockmode);
1292 lockmode = xfs_ilock_data_map_shared(ip);
1295 ASSERT(offset <= mp->m_super->s_maxbytes);
1296 if (offset + size > mp->m_super->s_maxbytes)
1297 size = mp->m_super->s_maxbytes - offset;
1298 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1299 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1301 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1302 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1308 (imap.br_startblock == HOLESTARTBLOCK ||
1309 imap.br_startblock == DELAYSTARTBLOCK))) {
1310 if (direct || xfs_get_extsz_hint(ip)) {
1312 * Drop the ilock in preparation for starting the block
1313 * allocation transaction. It will be retaken
1314 * exclusively inside xfs_iomap_write_direct for the
1315 * actual allocation.
1317 xfs_iunlock(ip, lockmode);
1318 error = xfs_iomap_write_direct(ip, offset, size,
1325 * Delalloc reservations do not require a transaction,
1326 * we can go on without dropping the lock here. If we
1327 * are allocating a new delalloc block, make sure that
1328 * we set the new flag so that we mark the buffer new so
1329 * that we know that it is newly allocated if the write
1332 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1334 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1338 xfs_iunlock(ip, lockmode);
1341 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1342 } else if (nimaps) {
1343 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1344 xfs_iunlock(ip, lockmode);
1346 trace_xfs_get_blocks_notfound(ip, offset, size);
1350 if (imap.br_startblock != HOLESTARTBLOCK &&
1351 imap.br_startblock != DELAYSTARTBLOCK) {
1353 * For unwritten extents do not report a disk address on
1354 * the read case (treat as if we're reading into a hole).
1356 if (create || !ISUNWRITTEN(&imap))
1357 xfs_map_buffer(inode, bh_result, &imap, offset);
1358 if (create && ISUNWRITTEN(&imap)) {
1360 bh_result->b_private = inode;
1361 set_buffer_defer_completion(bh_result);
1363 set_buffer_unwritten(bh_result);
1368 * If this is a realtime file, data may be on a different device.
1369 * to that pointed to from the buffer_head b_bdev currently.
1371 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1374 * If we previously allocated a block out beyond eof and we are now
1375 * coming back to use it then we will need to flag it as new even if it
1376 * has a disk address.
1378 * With sub-block writes into unwritten extents we also need to mark
1379 * the buffer as new so that the unwritten parts of the buffer gets
1383 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1384 (offset >= i_size_read(inode)) ||
1385 (new || ISUNWRITTEN(&imap))))
1386 set_buffer_new(bh_result);
1388 if (imap.br_startblock == DELAYSTARTBLOCK) {
1391 set_buffer_uptodate(bh_result);
1392 set_buffer_mapped(bh_result);
1393 set_buffer_delay(bh_result);
1398 * If this is O_DIRECT or the mpage code calling tell them how large
1399 * the mapping is, so that we can avoid repeated get_blocks calls.
1401 * If the mapping spans EOF, then we have to break the mapping up as the
1402 * mapping for blocks beyond EOF must be marked new so that sub block
1403 * regions can be correctly zeroed. We can't do this for mappings within
1404 * EOF unless the mapping was just allocated or is unwritten, otherwise
1405 * the callers would overwrite existing data with zeros. Hence we have
1406 * to split the mapping into a range up to and including EOF, and a
1407 * second mapping for beyond EOF.
1409 if (direct || size > (1 << inode->i_blkbits)) {
1410 xfs_off_t mapping_size;
1412 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1413 mapping_size <<= inode->i_blkbits;
1415 ASSERT(mapping_size > 0);
1416 if (mapping_size > size)
1417 mapping_size = size;
1418 if (offset < i_size_read(inode) &&
1419 offset + mapping_size >= i_size_read(inode)) {
1420 /* limit mapping to block that spans EOF */
1421 mapping_size = roundup_64(i_size_read(inode) - offset,
1422 1 << inode->i_blkbits);
1424 if (mapping_size > LONG_MAX)
1425 mapping_size = LONG_MAX;
1427 bh_result->b_size = mapping_size;
1433 xfs_iunlock(ip, lockmode);
1439 struct inode *inode,
1441 struct buffer_head *bh_result,
1444 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1448 xfs_get_blocks_direct(
1449 struct inode *inode,
1451 struct buffer_head *bh_result,
1454 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1458 * Complete a direct I/O write request.
1460 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1461 * need to issue a transaction to convert the range from unwritten to written
1462 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1463 * to do this and we are done. But in case this was a successful AIO
1464 * request this handler is called from interrupt context, from which we
1465 * can't start transactions. In that case offload the I/O completion to
1466 * the workqueues we also use for buffered I/O completion.
1469 xfs_end_io_direct_write(
1475 struct xfs_ioend *ioend = iocb->private;
1478 * While the generic direct I/O code updates the inode size, it does
1479 * so only after the end_io handler is called, which means our
1480 * end_io handler thinks the on-disk size is outside the in-core
1481 * size. To prevent this just update it a little bit earlier here.
1483 if (offset + size > i_size_read(ioend->io_inode))
1484 i_size_write(ioend->io_inode, offset + size);
1487 * blockdev_direct_IO can return an error even after the I/O
1488 * completion handler was called. Thus we need to protect
1489 * against double-freeing.
1491 iocb->private = NULL;
1493 ioend->io_offset = offset;
1494 ioend->io_size = size;
1495 if (private && size > 0)
1496 ioend->io_type = XFS_IO_UNWRITTEN;
1498 xfs_finish_ioend_sync(ioend);
1505 struct iov_iter *iter,
1508 struct inode *inode = iocb->ki_filp->f_mapping->host;
1509 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1510 struct xfs_ioend *ioend = NULL;
1514 size_t size = iov_iter_count(iter);
1517 * We cannot preallocate a size update transaction here as we
1518 * don't know whether allocation is necessary or not. Hence we
1519 * can only tell IO completion that one is necessary if we are
1520 * not doing unwritten extent conversion.
1522 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
1523 if (offset + size > XFS_I(inode)->i_d.di_size)
1524 ioend->io_isdirect = 1;
1526 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1527 offset, xfs_get_blocks_direct,
1528 xfs_end_io_direct_write, NULL,
1530 if (ret != -EIOCBQUEUED && iocb->private)
1531 goto out_destroy_ioend;
1533 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1534 offset, xfs_get_blocks_direct,
1541 xfs_destroy_ioend(ioend);
1546 * Punch out the delalloc blocks we have already allocated.
1548 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1549 * as the page is still locked at this point.
1552 xfs_vm_kill_delalloc_range(
1553 struct inode *inode,
1557 struct xfs_inode *ip = XFS_I(inode);
1558 xfs_fileoff_t start_fsb;
1559 xfs_fileoff_t end_fsb;
1562 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1563 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1564 if (end_fsb <= start_fsb)
1567 xfs_ilock(ip, XFS_ILOCK_EXCL);
1568 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1569 end_fsb - start_fsb);
1571 /* something screwed, just bail */
1572 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1573 xfs_alert(ip->i_mount,
1574 "xfs_vm_write_failed: unable to clean up ino %lld",
1578 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1582 xfs_vm_write_failed(
1583 struct inode *inode,
1588 loff_t block_offset;
1591 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1592 loff_t to = from + len;
1593 struct buffer_head *bh, *head;
1596 * The request pos offset might be 32 or 64 bit, this is all fine
1597 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1598 * platform, the high 32-bit will be masked off if we evaluate the
1599 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1600 * 0xfffff000 as an unsigned long, hence the result is incorrect
1601 * which could cause the following ASSERT failed in most cases.
1602 * In order to avoid this, we can evaluate the block_offset of the
1603 * start of the page by using shifts rather than masks the mismatch
1606 block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
1608 ASSERT(block_offset + from == pos);
1610 head = page_buffers(page);
1612 for (bh = head; bh != head || !block_start;
1613 bh = bh->b_this_page, block_start = block_end,
1614 block_offset += bh->b_size) {
1615 block_end = block_start + bh->b_size;
1617 /* skip buffers before the write */
1618 if (block_end <= from)
1621 /* if the buffer is after the write, we're done */
1622 if (block_start >= to)
1625 if (!buffer_delay(bh))
1628 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1631 xfs_vm_kill_delalloc_range(inode, block_offset,
1632 block_offset + bh->b_size);
1635 * This buffer does not contain data anymore. make sure anyone
1636 * who finds it knows that for certain.
1638 clear_buffer_delay(bh);
1639 clear_buffer_uptodate(bh);
1640 clear_buffer_mapped(bh);
1641 clear_buffer_new(bh);
1642 clear_buffer_dirty(bh);
1648 * This used to call block_write_begin(), but it unlocks and releases the page
1649 * on error, and we need that page to be able to punch stale delalloc blocks out
1650 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1651 * the appropriate point.
1656 struct address_space *mapping,
1660 struct page **pagep,
1663 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1667 ASSERT(len <= PAGE_CACHE_SIZE);
1669 page = grab_cache_page_write_begin(mapping, index, flags);
1673 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1674 if (unlikely(status)) {
1675 struct inode *inode = mapping->host;
1676 size_t isize = i_size_read(inode);
1678 xfs_vm_write_failed(inode, page, pos, len);
1682 * If the write is beyond EOF, we only want to kill blocks
1683 * allocated in this write, not blocks that were previously
1684 * written successfully.
1686 if (pos + len > isize) {
1687 ssize_t start = max_t(ssize_t, pos, isize);
1689 truncate_pagecache_range(inode, start, pos + len);
1692 page_cache_release(page);
1701 * On failure, we only need to kill delalloc blocks beyond EOF in the range of
1702 * this specific write because they will never be written. Previous writes
1703 * beyond EOF where block allocation succeeded do not need to be trashed, so
1704 * only new blocks from this write should be trashed. For blocks within
1705 * EOF, generic_write_end() zeros them so they are safe to leave alone and be
1706 * written with all the other valid data.
1711 struct address_space *mapping,
1720 ASSERT(len <= PAGE_CACHE_SIZE);
1722 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1723 if (unlikely(ret < len)) {
1724 struct inode *inode = mapping->host;
1725 size_t isize = i_size_read(inode);
1726 loff_t to = pos + len;
1729 /* only kill blocks in this write beyond EOF */
1732 xfs_vm_kill_delalloc_range(inode, isize, to);
1733 truncate_pagecache_range(inode, isize, to);
1741 struct address_space *mapping,
1744 struct inode *inode = (struct inode *)mapping->host;
1745 struct xfs_inode *ip = XFS_I(inode);
1747 trace_xfs_vm_bmap(XFS_I(inode));
1748 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1749 filemap_write_and_wait(mapping);
1750 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1751 return generic_block_bmap(mapping, block, xfs_get_blocks);
1756 struct file *unused,
1759 return mpage_readpage(page, xfs_get_blocks);
1764 struct file *unused,
1765 struct address_space *mapping,
1766 struct list_head *pages,
1769 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1773 * This is basically a copy of __set_page_dirty_buffers() with one
1774 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1775 * dirty, we'll never be able to clean them because we don't write buffers
1776 * beyond EOF, and that means we can't invalidate pages that span EOF
1777 * that have been marked dirty. Further, the dirty state can leak into
1778 * the file interior if the file is extended, resulting in all sorts of
1779 * bad things happening as the state does not match the underlying data.
1781 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1782 * this only exist because of bufferheads and how the generic code manages them.
1785 xfs_vm_set_page_dirty(
1788 struct address_space *mapping = page->mapping;
1789 struct inode *inode = mapping->host;
1794 if (unlikely(!mapping))
1795 return !TestSetPageDirty(page);
1797 end_offset = i_size_read(inode);
1798 offset = page_offset(page);
1800 spin_lock(&mapping->private_lock);
1801 if (page_has_buffers(page)) {
1802 struct buffer_head *head = page_buffers(page);
1803 struct buffer_head *bh = head;
1806 if (offset < end_offset)
1807 set_buffer_dirty(bh);
1808 bh = bh->b_this_page;
1809 offset += 1 << inode->i_blkbits;
1810 } while (bh != head);
1812 newly_dirty = !TestSetPageDirty(page);
1813 spin_unlock(&mapping->private_lock);
1816 /* sigh - __set_page_dirty() is static, so copy it here, too */
1817 unsigned long flags;
1819 spin_lock_irqsave(&mapping->tree_lock, flags);
1820 if (page->mapping) { /* Race with truncate? */
1821 WARN_ON_ONCE(!PageUptodate(page));
1822 account_page_dirtied(page, mapping);
1823 radix_tree_tag_set(&mapping->page_tree,
1824 page_index(page), PAGECACHE_TAG_DIRTY);
1826 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1827 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1832 const struct address_space_operations xfs_address_space_operations = {
1833 .readpage = xfs_vm_readpage,
1834 .readpages = xfs_vm_readpages,
1835 .writepage = xfs_vm_writepage,
1836 .writepages = xfs_vm_writepages,
1837 .set_page_dirty = xfs_vm_set_page_dirty,
1838 .releasepage = xfs_vm_releasepage,
1839 .invalidatepage = xfs_vm_invalidatepage,
1840 .write_begin = xfs_vm_write_begin,
1841 .write_end = xfs_vm_write_end,
1842 .bmap = xfs_vm_bmap,
1843 .direct_IO = xfs_vm_direct_IO,
1844 .migratepage = buffer_migrate_page,
1845 .is_partially_uptodate = block_is_partially_uptodate,
1846 .error_remove_page = generic_error_remove_page,