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 "xfs_reflink.h"
35 #include <linux/gfp.h>
36 #include <linux/mpage.h>
37 #include <linux/pagevec.h>
38 #include <linux/writeback.h>
41 * structure owned by writepages passed to individual writepage calls
43 struct xfs_writepage_ctx {
44 struct xfs_bmbt_irec imap;
47 struct xfs_ioend *ioend;
57 struct buffer_head *bh, *head;
59 *delalloc = *unwritten = 0;
61 bh = head = page_buffers(page);
63 if (buffer_unwritten(bh))
65 else if (buffer_delay(bh))
67 } while ((bh = bh->b_this_page) != head);
71 xfs_find_bdev_for_inode(
74 struct xfs_inode *ip = XFS_I(inode);
75 struct xfs_mount *mp = ip->i_mount;
77 if (XFS_IS_REALTIME_INODE(ip))
78 return mp->m_rtdev_targp->bt_bdev;
80 return mp->m_ddev_targp->bt_bdev;
84 * We're now finished for good with this page. Update the page state via the
85 * associated buffer_heads, paying attention to the start and end offsets that
86 * we need to process on the page.
88 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
89 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
90 * the page at all, as we may be racing with memory reclaim and it can free both
91 * the bufferhead chain and the page as it will see the page as clean and
95 xfs_finish_page_writeback(
100 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
101 struct buffer_head *head, *bh, *next;
102 unsigned int off = 0;
105 ASSERT(bvec->bv_offset < PAGE_SIZE);
106 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
107 ASSERT(end < PAGE_SIZE);
108 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
110 bh = head = page_buffers(bvec->bv_page);
114 next = bh->b_this_page;
115 if (off < bvec->bv_offset)
119 bh->b_end_io(bh, !error);
122 } while ((bh = next) != head);
126 * We're now finished for good with this ioend structure. Update the page
127 * state, release holds on bios, and finally free up memory. Do not use the
132 struct xfs_ioend *ioend,
135 struct inode *inode = ioend->io_inode;
136 struct bio *last = ioend->io_bio;
137 struct bio *bio, *next;
139 for (bio = &ioend->io_inline_bio; bio; bio = next) {
140 struct bio_vec *bvec;
144 * For the last bio, bi_private points to the ioend, so we
145 * need to explicitly end the iteration here.
150 next = bio->bi_private;
152 /* walk each page on bio, ending page IO on them */
153 bio_for_each_segment_all(bvec, bio, i)
154 xfs_finish_page_writeback(inode, bvec, error);
161 * Fast and loose check if this write could update the on-disk inode size.
163 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
165 return ioend->io_offset + ioend->io_size >
166 XFS_I(ioend->io_inode)->i_d.di_size;
170 xfs_setfilesize_trans_alloc(
171 struct xfs_ioend *ioend)
173 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
174 struct xfs_trans *tp;
177 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
181 ioend->io_append_trans = tp;
184 * We may pass freeze protection with a transaction. So tell lockdep
187 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
189 * We hand off the transaction to the completion thread now, so
190 * clear the flag here.
192 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
197 * Update on-disk file size now that data has been written to disk.
201 struct xfs_inode *ip,
202 struct xfs_trans *tp,
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209 isize = xfs_new_eof(ip, offset + size);
211 xfs_iunlock(ip, XFS_ILOCK_EXCL);
212 xfs_trans_cancel(tp);
216 trace_xfs_setfilesize(ip, offset, size);
218 ip->i_d.di_size = isize;
219 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
220 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
222 return xfs_trans_commit(tp);
227 struct xfs_inode *ip,
231 struct xfs_mount *mp = ip->i_mount;
232 struct xfs_trans *tp;
235 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
239 return __xfs_setfilesize(ip, tp, offset, size);
243 xfs_setfilesize_ioend(
244 struct xfs_ioend *ioend,
247 struct xfs_inode *ip = XFS_I(ioend->io_inode);
248 struct xfs_trans *tp = ioend->io_append_trans;
251 * The transaction may have been allocated in the I/O submission thread,
252 * thus we need to mark ourselves as being in a transaction manually.
253 * Similarly for freeze protection.
255 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
256 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
258 /* we abort the update if there was an IO error */
260 xfs_trans_cancel(tp);
264 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
268 * IO write completion.
272 struct work_struct *work)
274 struct xfs_ioend *ioend =
275 container_of(work, struct xfs_ioend, io_work);
276 struct xfs_inode *ip = XFS_I(ioend->io_inode);
277 int error = ioend->io_bio->bi_error;
280 * Set an error if the mount has shut down and proceed with end I/O
281 * processing so it can perform whatever cleanups are necessary.
283 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
287 * For a CoW extent, we need to move the mapping from the CoW fork
288 * to the data fork. If instead an error happened, just dump the
291 if (ioend->io_type == XFS_IO_COW) {
294 if (ioend->io_bio->bi_error) {
295 error = xfs_reflink_cancel_cow_range(ip,
296 ioend->io_offset, ioend->io_size);
299 error = xfs_reflink_end_cow(ip, ioend->io_offset,
306 * For unwritten extents we need to issue transactions to convert a
307 * range to normal written extens after the data I/O has finished.
308 * Detecting and handling completion IO errors is done individually
309 * for each case as different cleanup operations need to be performed
312 if (ioend->io_type == XFS_IO_UNWRITTEN) {
315 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
317 } else if (ioend->io_append_trans) {
318 error = xfs_setfilesize_ioend(ioend, error);
320 ASSERT(!xfs_ioend_is_append(ioend) ||
321 ioend->io_type == XFS_IO_COW);
325 xfs_destroy_ioend(ioend, error);
332 struct xfs_ioend *ioend = bio->bi_private;
333 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
335 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
336 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
337 else if (ioend->io_append_trans)
338 queue_work(mp->m_data_workqueue, &ioend->io_work);
340 xfs_destroy_ioend(ioend, bio->bi_error);
347 struct xfs_bmbt_irec *imap,
350 struct xfs_inode *ip = XFS_I(inode);
351 struct xfs_mount *mp = ip->i_mount;
352 ssize_t count = 1 << inode->i_blkbits;
353 xfs_fileoff_t offset_fsb, end_fsb;
355 int bmapi_flags = XFS_BMAPI_ENTIRE;
358 if (XFS_FORCED_SHUTDOWN(mp))
361 ASSERT(type != XFS_IO_COW);
362 if (type == XFS_IO_UNWRITTEN)
363 bmapi_flags |= XFS_BMAPI_IGSTATE;
365 xfs_ilock(ip, XFS_ILOCK_SHARED);
366 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
367 (ip->i_df.if_flags & XFS_IFEXTENTS));
368 ASSERT(offset <= mp->m_super->s_maxbytes);
370 if (offset + count > mp->m_super->s_maxbytes)
371 count = mp->m_super->s_maxbytes - offset;
372 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
373 offset_fsb = XFS_B_TO_FSBT(mp, offset);
374 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
375 imap, &nimaps, bmapi_flags);
377 * Truncate an overwrite extent if there's a pending CoW
378 * reservation before the end of this extent. This forces us
379 * to come back to writepage to take care of the CoW.
381 if (nimaps && type == XFS_IO_OVERWRITE)
382 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
383 xfs_iunlock(ip, XFS_ILOCK_SHARED);
388 if (type == XFS_IO_DELALLOC &&
389 (!nimaps || isnullstartblock(imap->br_startblock))) {
390 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
393 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
398 if (type == XFS_IO_UNWRITTEN) {
400 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
401 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
405 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
412 struct xfs_bmbt_irec *imap,
415 offset >>= inode->i_blkbits;
417 return offset >= imap->br_startoff &&
418 offset < imap->br_startoff + imap->br_blockcount;
422 xfs_start_buffer_writeback(
423 struct buffer_head *bh)
425 ASSERT(buffer_mapped(bh));
426 ASSERT(buffer_locked(bh));
427 ASSERT(!buffer_delay(bh));
428 ASSERT(!buffer_unwritten(bh));
430 mark_buffer_async_write(bh);
431 set_buffer_uptodate(bh);
432 clear_buffer_dirty(bh);
436 xfs_start_page_writeback(
440 ASSERT(PageLocked(page));
441 ASSERT(!PageWriteback(page));
444 * if the page was not fully cleaned, we need to ensure that the higher
445 * layers come back to it correctly. That means we need to keep the page
446 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
447 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
448 * write this page in this writeback sweep will be made.
451 clear_page_dirty_for_io(page);
452 set_page_writeback(page);
454 set_page_writeback_keepwrite(page);
459 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
461 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
465 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
466 * it, and we submit that bio. The ioend may be used for multiple bio
467 * submissions, so we only want to allocate an append transaction for the ioend
468 * once. In the case of multiple bio submission, each bio will take an IO
469 * reference to the ioend to ensure that the ioend completion is only done once
470 * all bios have been submitted and the ioend is really done.
472 * If @fail is non-zero, it means that we have a situation where some part of
473 * the submission process has failed after we have marked paged for writeback
474 * and unlocked them. In this situation, we need to fail the bio and ioend
475 * rather than submit it to IO. This typically only happens on a filesystem
480 struct writeback_control *wbc,
481 struct xfs_ioend *ioend,
484 /* Convert CoW extents to regular */
485 if (!status && ioend->io_type == XFS_IO_COW) {
486 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
487 ioend->io_offset, ioend->io_size);
490 /* Reserve log space if we might write beyond the on-disk inode size. */
492 ioend->io_type != XFS_IO_UNWRITTEN &&
493 xfs_ioend_is_append(ioend) &&
494 !ioend->io_append_trans)
495 status = xfs_setfilesize_trans_alloc(ioend);
497 ioend->io_bio->bi_private = ioend;
498 ioend->io_bio->bi_end_io = xfs_end_bio;
499 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
502 * If we are failing the IO now, just mark the ioend with an
503 * error and finish it. This will run IO completion immediately
504 * as there is only one reference to the ioend at this point in
508 ioend->io_bio->bi_error = status;
509 bio_endio(ioend->io_bio);
513 submit_bio(ioend->io_bio);
518 xfs_init_bio_from_bh(
520 struct buffer_head *bh)
522 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
523 bio->bi_bdev = bh->b_bdev;
526 static struct xfs_ioend *
531 struct buffer_head *bh)
533 struct xfs_ioend *ioend;
536 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
537 xfs_init_bio_from_bh(bio, bh);
539 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
540 INIT_LIST_HEAD(&ioend->io_list);
541 ioend->io_type = type;
542 ioend->io_inode = inode;
544 ioend->io_offset = offset;
545 INIT_WORK(&ioend->io_work, xfs_end_io);
546 ioend->io_append_trans = NULL;
552 * Allocate a new bio, and chain the old bio to the new one.
554 * Note that we have to do perform the chaining in this unintuitive order
555 * so that the bi_private linkage is set up in the right direction for the
556 * traversal in xfs_destroy_ioend().
560 struct xfs_ioend *ioend,
561 struct writeback_control *wbc,
562 struct buffer_head *bh)
566 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
567 xfs_init_bio_from_bh(new, bh);
569 bio_chain(ioend->io_bio, new);
570 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
571 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
572 submit_bio(ioend->io_bio);
577 * Test to see if we've been building up a completion structure for
578 * earlier buffers -- if so, we try to append to this ioend if we
579 * can, otherwise we finish off any current ioend and start another.
580 * Return the ioend we finished off so that the caller can submit it
581 * once it has finished processing the dirty page.
586 struct buffer_head *bh,
588 struct xfs_writepage_ctx *wpc,
589 struct writeback_control *wbc,
590 struct list_head *iolist)
592 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
593 bh->b_blocknr != wpc->last_block + 1 ||
594 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
596 list_add(&wpc->ioend->io_list, iolist);
597 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
601 * If the buffer doesn't fit into the bio we need to allocate a new
602 * one. This shouldn't happen more than once for a given buffer.
604 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
605 xfs_chain_bio(wpc->ioend, wbc, bh);
607 wpc->ioend->io_size += bh->b_size;
608 wpc->last_block = bh->b_blocknr;
609 xfs_start_buffer_writeback(bh);
615 struct buffer_head *bh,
616 struct xfs_bmbt_irec *imap,
620 struct xfs_mount *m = XFS_I(inode)->i_mount;
621 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
622 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
624 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
625 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
627 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
628 ((offset - iomap_offset) >> inode->i_blkbits);
630 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
633 set_buffer_mapped(bh);
639 struct buffer_head *bh,
640 struct xfs_bmbt_irec *imap,
643 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
644 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
646 xfs_map_buffer(inode, bh, imap, offset);
647 set_buffer_mapped(bh);
648 clear_buffer_delay(bh);
649 clear_buffer_unwritten(bh);
653 * Test if a given page contains at least one buffer of a given @type.
654 * If @check_all_buffers is true, then we walk all the buffers in the page to
655 * try to find one of the type passed in. If it is not set, then the caller only
656 * needs to check the first buffer on the page for a match.
662 bool check_all_buffers)
664 struct buffer_head *bh;
665 struct buffer_head *head;
667 if (PageWriteback(page))
671 if (!page_has_buffers(page))
674 bh = head = page_buffers(page);
676 if (buffer_unwritten(bh)) {
677 if (type == XFS_IO_UNWRITTEN)
679 } else if (buffer_delay(bh)) {
680 if (type == XFS_IO_DELALLOC)
682 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
683 if (type == XFS_IO_OVERWRITE)
687 /* If we are only checking the first buffer, we are done now. */
688 if (!check_all_buffers)
690 } while ((bh = bh->b_this_page) != head);
696 xfs_vm_invalidatepage(
701 trace_xfs_invalidatepage(page->mapping->host, page, offset,
703 block_invalidatepage(page, offset, length);
707 * If the page has delalloc buffers on it, we need to punch them out before we
708 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
709 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
710 * is done on that same region - the delalloc extent is returned when none is
711 * supposed to be there.
713 * We prevent this by truncating away the delalloc regions on the page before
714 * invalidating it. Because they are delalloc, we can do this without needing a
715 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
716 * truncation without a transaction as there is no space left for block
717 * reservation (typically why we see a ENOSPC in writeback).
719 * This is not a performance critical path, so for now just do the punching a
720 * buffer head at a time.
723 xfs_aops_discard_page(
726 struct inode *inode = page->mapping->host;
727 struct xfs_inode *ip = XFS_I(inode);
728 struct buffer_head *bh, *head;
729 loff_t offset = page_offset(page);
731 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
734 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
737 xfs_alert(ip->i_mount,
738 "page discard on page %p, inode 0x%llx, offset %llu.",
739 page, ip->i_ino, offset);
741 xfs_ilock(ip, XFS_ILOCK_EXCL);
742 bh = head = page_buffers(page);
745 xfs_fileoff_t start_fsb;
747 if (!buffer_delay(bh))
750 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
751 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
753 /* something screwed, just bail */
754 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
755 xfs_alert(ip->i_mount,
756 "page discard unable to remove delalloc mapping.");
761 offset += 1 << inode->i_blkbits;
763 } while ((bh = bh->b_this_page) != head);
765 xfs_iunlock(ip, XFS_ILOCK_EXCL);
767 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
773 struct xfs_writepage_ctx *wpc,
776 unsigned int *new_type)
778 struct xfs_inode *ip = XFS_I(inode);
779 struct xfs_bmbt_irec imap;
784 * If we already have a valid COW mapping keep using it.
786 if (wpc->io_type == XFS_IO_COW) {
787 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
788 if (wpc->imap_valid) {
789 *new_type = XFS_IO_COW;
795 * Else we need to check if there is a COW mapping at this offset.
797 xfs_ilock(ip, XFS_ILOCK_SHARED);
798 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
799 xfs_iunlock(ip, XFS_ILOCK_SHARED);
805 * And if the COW mapping has a delayed extent here we need to
806 * allocate real space for it now.
808 if (isnullstartblock(imap.br_startblock)) {
809 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
815 wpc->io_type = *new_type = XFS_IO_COW;
816 wpc->imap_valid = true;
822 * We implement an immediate ioend submission policy here to avoid needing to
823 * chain multiple ioends and hence nest mempool allocations which can violate
824 * forward progress guarantees we need to provide. The current ioend we are
825 * adding buffers to is cached on the writepage context, and if the new buffer
826 * does not append to the cached ioend it will create a new ioend and cache that
829 * If a new ioend is created and cached, the old ioend is returned and queued
830 * locally for submission once the entire page is processed or an error has been
831 * detected. While ioends are submitted immediately after they are completed,
832 * batching optimisations are provided by higher level block plugging.
834 * At the end of a writeback pass, there will be a cached ioend remaining on the
835 * writepage context that the caller will need to submit.
839 struct xfs_writepage_ctx *wpc,
840 struct writeback_control *wbc,
844 __uint64_t end_offset)
846 LIST_HEAD(submit_list);
847 struct xfs_ioend *ioend, *next;
848 struct buffer_head *bh, *head;
849 ssize_t len = 1 << inode->i_blkbits;
853 unsigned int new_type;
855 bh = head = page_buffers(page);
856 offset = page_offset(page);
858 if (offset >= end_offset)
860 if (!buffer_uptodate(bh))
864 * set_page_dirty dirties all buffers in a page, independent
865 * of their state. The dirty state however is entirely
866 * meaningless for holes (!mapped && uptodate), so skip
867 * buffers covering holes here.
869 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
870 wpc->imap_valid = false;
874 if (buffer_unwritten(bh))
875 new_type = XFS_IO_UNWRITTEN;
876 else if (buffer_delay(bh))
877 new_type = XFS_IO_DELALLOC;
878 else if (buffer_uptodate(bh))
879 new_type = XFS_IO_OVERWRITE;
881 if (PageUptodate(page))
882 ASSERT(buffer_mapped(bh));
884 * This buffer is not uptodate and will not be
885 * written to disk. Ensure that we will put any
886 * subsequent writeable buffers into a new
889 wpc->imap_valid = false;
893 if (xfs_is_reflink_inode(XFS_I(inode))) {
894 error = xfs_map_cow(wpc, inode, offset, &new_type);
899 if (wpc->io_type != new_type) {
900 wpc->io_type = new_type;
901 wpc->imap_valid = false;
905 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
907 if (!wpc->imap_valid) {
908 error = xfs_map_blocks(inode, offset, &wpc->imap,
912 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
915 if (wpc->imap_valid) {
917 if (wpc->io_type != XFS_IO_OVERWRITE)
918 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
919 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
923 } while (offset += len, ((bh = bh->b_this_page) != head));
925 if (uptodate && bh == head)
926 SetPageUptodate(page);
928 ASSERT(wpc->ioend || list_empty(&submit_list));
932 * On error, we have to fail the ioend here because we have locked
933 * buffers in the ioend. If we don't do this, we'll deadlock
934 * invalidating the page as that tries to lock the buffers on the page.
935 * Also, because we may have set pages under writeback, we have to make
936 * sure we run IO completion to mark the error state of the IO
937 * appropriately, so we can't cancel the ioend directly here. That means
938 * we have to mark this page as under writeback if we included any
939 * buffers from it in the ioend chain so that completion treats it
942 * If we didn't include the page in the ioend, the on error we can
943 * simply discard and unlock it as there are no other users of the page
944 * or it's buffers right now. The caller will still need to trigger
945 * submission of outstanding ioends on the writepage context so they are
946 * treated correctly on error.
949 xfs_start_page_writeback(page, !error);
952 * Preserve the original error if there was one, otherwise catch
953 * submission errors here and propagate into subsequent ioend
956 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
959 list_del_init(&ioend->io_list);
960 error2 = xfs_submit_ioend(wbc, ioend, error);
961 if (error2 && !error)
965 xfs_aops_discard_page(page);
966 ClearPageUptodate(page);
970 * We can end up here with no error and nothing to write if we
971 * race with a partial page truncate on a sub-page block sized
972 * filesystem. In that case we need to mark the page clean.
974 xfs_start_page_writeback(page, 1);
975 end_page_writeback(page);
978 mapping_set_error(page->mapping, error);
983 * Write out a dirty page.
985 * For delalloc space on the page we need to allocate space and flush it.
986 * For unwritten space on the page we need to start the conversion to
987 * regular allocated space.
988 * For any other dirty buffer heads on the page we should flush them.
993 struct writeback_control *wbc,
996 struct xfs_writepage_ctx *wpc = data;
997 struct inode *inode = page->mapping->host;
999 __uint64_t end_offset;
1002 trace_xfs_writepage(inode, page, 0, 0);
1004 ASSERT(page_has_buffers(page));
1007 * Refuse to write the page out if we are called from reclaim context.
1009 * This avoids stack overflows when called from deeply used stacks in
1010 * random callers for direct reclaim or memcg reclaim. We explicitly
1011 * allow reclaim from kswapd as the stack usage there is relatively low.
1013 * This should never happen except in the case of a VM regression so
1016 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1021 * Given that we do not allow direct reclaim to call us, we should
1022 * never be called while in a filesystem transaction.
1024 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
1028 * Is this page beyond the end of the file?
1030 * The page index is less than the end_index, adjust the end_offset
1031 * to the highest offset that this page should represent.
1032 * -----------------------------------------------------
1033 * | file mapping | <EOF> |
1034 * -----------------------------------------------------
1035 * | Page ... | Page N-2 | Page N-1 | Page N | |
1036 * ^--------------------------------^----------|--------
1037 * | desired writeback range | see else |
1038 * ---------------------------------^------------------|
1040 offset = i_size_read(inode);
1041 end_index = offset >> PAGE_SHIFT;
1042 if (page->index < end_index)
1043 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1046 * Check whether the page to write out is beyond or straddles
1048 * -------------------------------------------------------
1049 * | file mapping | <EOF> |
1050 * -------------------------------------------------------
1051 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1052 * ^--------------------------------^-----------|---------
1054 * ---------------------------------^-----------|--------|
1056 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1059 * Skip the page if it is fully outside i_size, e.g. due to a
1060 * truncate operation that is in progress. We must redirty the
1061 * page so that reclaim stops reclaiming it. Otherwise
1062 * xfs_vm_releasepage() is called on it and gets confused.
1064 * Note that the end_index is unsigned long, it would overflow
1065 * if the given offset is greater than 16TB on 32-bit system
1066 * and if we do check the page is fully outside i_size or not
1067 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1068 * will be evaluated to 0. Hence this page will be redirtied
1069 * and be written out repeatedly which would result in an
1070 * infinite loop, the user program that perform this operation
1071 * will hang. Instead, we can verify this situation by checking
1072 * if the page to write is totally beyond the i_size or if it's
1073 * offset is just equal to the EOF.
1075 if (page->index > end_index ||
1076 (page->index == end_index && offset_into_page == 0))
1080 * The page straddles i_size. It must be zeroed out on each
1081 * and every writepage invocation because it may be mmapped.
1082 * "A file is mapped in multiples of the page size. For a file
1083 * that is not a multiple of the page size, the remaining
1084 * memory is zeroed when mapped, and writes to that region are
1085 * not written out to the file."
1087 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1089 /* Adjust the end_offset to the end of file */
1090 end_offset = offset;
1093 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1096 redirty_page_for_writepage(wbc, page);
1104 struct writeback_control *wbc)
1106 struct xfs_writepage_ctx wpc = {
1107 .io_type = XFS_IO_INVALID,
1111 ret = xfs_do_writepage(page, wbc, &wpc);
1113 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1119 struct address_space *mapping,
1120 struct writeback_control *wbc)
1122 struct xfs_writepage_ctx wpc = {
1123 .io_type = XFS_IO_INVALID,
1127 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1128 if (dax_mapping(mapping))
1129 return dax_writeback_mapping_range(mapping,
1130 xfs_find_bdev_for_inode(mapping->host), wbc);
1132 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1134 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1139 * Called to move a page into cleanable state - and from there
1140 * to be released. The page should already be clean. We always
1141 * have buffer heads in this call.
1143 * Returns 1 if the page is ok to release, 0 otherwise.
1150 int delalloc, unwritten;
1152 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1155 * mm accommodates an old ext3 case where clean pages might not have had
1156 * the dirty bit cleared. Thus, it can send actual dirty pages to
1157 * ->releasepage() via shrink_active_list(). Conversely,
1158 * block_invalidatepage() can send pages that are still marked dirty
1159 * but otherwise have invalidated buffers.
1161 * We want to release the latter to avoid unnecessary buildup of the
1162 * LRU, skip the former and warn if we've left any lingering
1163 * delalloc/unwritten buffers on clean pages. Skip pages with delalloc
1164 * or unwritten buffers and warn if the page is not dirty. Otherwise
1165 * try to release the buffers.
1167 xfs_count_page_state(page, &delalloc, &unwritten);
1170 WARN_ON_ONCE(!PageDirty(page));
1174 WARN_ON_ONCE(!PageDirty(page));
1178 return try_to_free_buffers(page);
1182 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1183 * is, so that we can avoid repeated get_blocks calls.
1185 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1186 * for blocks beyond EOF must be marked new so that sub block regions can be
1187 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1188 * was just allocated or is unwritten, otherwise the callers would overwrite
1189 * existing data with zeros. Hence we have to split the mapping into a range up
1190 * to and including EOF, and a second mapping for beyond EOF.
1194 struct inode *inode,
1196 struct buffer_head *bh_result,
1197 struct xfs_bmbt_irec *imap,
1201 xfs_off_t mapping_size;
1203 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1204 mapping_size <<= inode->i_blkbits;
1206 ASSERT(mapping_size > 0);
1207 if (mapping_size > size)
1208 mapping_size = size;
1209 if (offset < i_size_read(inode) &&
1210 offset + mapping_size >= i_size_read(inode)) {
1211 /* limit mapping to block that spans EOF */
1212 mapping_size = roundup_64(i_size_read(inode) - offset,
1213 1 << inode->i_blkbits);
1215 if (mapping_size > LONG_MAX)
1216 mapping_size = LONG_MAX;
1218 bh_result->b_size = mapping_size;
1223 struct inode *inode,
1225 struct buffer_head *bh_result,
1228 struct xfs_inode *ip = XFS_I(inode);
1229 struct xfs_mount *mp = ip->i_mount;
1230 xfs_fileoff_t offset_fsb, end_fsb;
1233 struct xfs_bmbt_irec imap;
1240 if (XFS_FORCED_SHUTDOWN(mp))
1243 offset = (xfs_off_t)iblock << inode->i_blkbits;
1244 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1245 size = bh_result->b_size;
1247 if (offset >= i_size_read(inode))
1251 * Direct I/O is usually done on preallocated files, so try getting
1252 * a block mapping without an exclusive lock first.
1254 lockmode = xfs_ilock_data_map_shared(ip);
1256 ASSERT(offset <= mp->m_super->s_maxbytes);
1257 if (offset + size > mp->m_super->s_maxbytes)
1258 size = mp->m_super->s_maxbytes - offset;
1259 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1260 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1262 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1263 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1268 trace_xfs_get_blocks_found(ip, offset, size,
1269 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1270 : XFS_IO_OVERWRITE, &imap);
1271 xfs_iunlock(ip, lockmode);
1273 trace_xfs_get_blocks_notfound(ip, offset, size);
1277 /* trim mapping down to size requested */
1278 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1281 * For unwritten extents do not report a disk address in the buffered
1282 * read case (treat as if we're reading into a hole).
1284 if (imap.br_startblock != HOLESTARTBLOCK &&
1285 imap.br_startblock != DELAYSTARTBLOCK &&
1286 !ISUNWRITTEN(&imap))
1287 xfs_map_buffer(inode, bh_result, &imap, offset);
1290 * If this is a realtime file, data may be on a different device.
1291 * to that pointed to from the buffer_head b_bdev currently.
1293 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1297 xfs_iunlock(ip, lockmode);
1304 struct iov_iter *iter)
1307 * We just need the method present so that open/fcntl allow direct I/O.
1314 struct address_space *mapping,
1317 struct inode *inode = (struct inode *)mapping->host;
1318 struct xfs_inode *ip = XFS_I(inode);
1320 trace_xfs_vm_bmap(XFS_I(inode));
1323 * The swap code (ab-)uses ->bmap to get a block mapping and then
1324 * bypasseѕ the file system for actual I/O. We really can't allow
1325 * that on reflinks inodes, so we have to skip out here. And yes,
1326 * 0 is the magic code for a bmap error..
1328 if (xfs_is_reflink_inode(ip))
1331 filemap_write_and_wait(mapping);
1332 return generic_block_bmap(mapping, block, xfs_get_blocks);
1337 struct file *unused,
1340 trace_xfs_vm_readpage(page->mapping->host, 1);
1341 return mpage_readpage(page, xfs_get_blocks);
1346 struct file *unused,
1347 struct address_space *mapping,
1348 struct list_head *pages,
1351 trace_xfs_vm_readpages(mapping->host, nr_pages);
1352 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1356 * This is basically a copy of __set_page_dirty_buffers() with one
1357 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1358 * dirty, we'll never be able to clean them because we don't write buffers
1359 * beyond EOF, and that means we can't invalidate pages that span EOF
1360 * that have been marked dirty. Further, the dirty state can leak into
1361 * the file interior if the file is extended, resulting in all sorts of
1362 * bad things happening as the state does not match the underlying data.
1364 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1365 * this only exist because of bufferheads and how the generic code manages them.
1368 xfs_vm_set_page_dirty(
1371 struct address_space *mapping = page->mapping;
1372 struct inode *inode = mapping->host;
1377 if (unlikely(!mapping))
1378 return !TestSetPageDirty(page);
1380 end_offset = i_size_read(inode);
1381 offset = page_offset(page);
1383 spin_lock(&mapping->private_lock);
1384 if (page_has_buffers(page)) {
1385 struct buffer_head *head = page_buffers(page);
1386 struct buffer_head *bh = head;
1389 if (offset < end_offset)
1390 set_buffer_dirty(bh);
1391 bh = bh->b_this_page;
1392 offset += 1 << inode->i_blkbits;
1393 } while (bh != head);
1396 * Lock out page->mem_cgroup migration to keep PageDirty
1397 * synchronized with per-memcg dirty page counters.
1399 lock_page_memcg(page);
1400 newly_dirty = !TestSetPageDirty(page);
1401 spin_unlock(&mapping->private_lock);
1404 /* sigh - __set_page_dirty() is static, so copy it here, too */
1405 unsigned long flags;
1407 spin_lock_irqsave(&mapping->tree_lock, flags);
1408 if (page->mapping) { /* Race with truncate? */
1409 WARN_ON_ONCE(!PageUptodate(page));
1410 account_page_dirtied(page, mapping);
1411 radix_tree_tag_set(&mapping->page_tree,
1412 page_index(page), PAGECACHE_TAG_DIRTY);
1414 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1416 unlock_page_memcg(page);
1418 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1422 const struct address_space_operations xfs_address_space_operations = {
1423 .readpage = xfs_vm_readpage,
1424 .readpages = xfs_vm_readpages,
1425 .writepage = xfs_vm_writepage,
1426 .writepages = xfs_vm_writepages,
1427 .set_page_dirty = xfs_vm_set_page_dirty,
1428 .releasepage = xfs_vm_releasepage,
1429 .invalidatepage = xfs_vm_invalidatepage,
1430 .bmap = xfs_vm_bmap,
1431 .direct_IO = xfs_vm_direct_IO,
1432 .migratepage = buffer_migrate_page,
1433 .is_partially_uptodate = block_is_partially_uptodate,
1434 .error_remove_page = generic_error_remove_page,