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 xfs_find_daxdev_for_inode(
87 struct xfs_inode *ip = XFS_I(inode);
88 struct xfs_mount *mp = ip->i_mount;
90 if (XFS_IS_REALTIME_INODE(ip))
91 return mp->m_rtdev_targp->bt_daxdev;
93 return mp->m_ddev_targp->bt_daxdev;
97 * We're now finished for good with this page. Update the page state via the
98 * associated buffer_heads, paying attention to the start and end offsets that
99 * we need to process on the page.
101 * Note that we open code the action in end_buffer_async_write here so that we
102 * only have to iterate over the buffers attached to the page once. This is not
103 * only more efficient, but also ensures that we only calls end_page_writeback
104 * at the end of the iteration, and thus avoids the pitfall of having the page
105 * and buffers potentially freed after every call to end_buffer_async_write.
108 xfs_finish_page_writeback(
110 struct bio_vec *bvec,
113 struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head;
115 unsigned int off = 0;
118 ASSERT(bvec->bv_offset < PAGE_SIZE);
119 ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0);
120 ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE);
121 ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0);
123 local_irq_save(flags);
124 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
126 if (off >= bvec->bv_offset &&
127 off < bvec->bv_offset + bvec->bv_len) {
128 ASSERT(buffer_async_write(bh));
129 ASSERT(bh->b_end_io == NULL);
132 mark_buffer_write_io_error(bh);
133 clear_buffer_uptodate(bh);
134 SetPageError(bvec->bv_page);
136 set_buffer_uptodate(bh);
138 clear_buffer_async_write(bh);
140 } else if (buffer_async_write(bh)) {
141 ASSERT(buffer_locked(bh));
145 } while ((bh = bh->b_this_page) != head);
146 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
147 local_irq_restore(flags);
150 end_page_writeback(bvec->bv_page);
154 * We're now finished for good with this ioend structure. Update the page
155 * state, release holds on bios, and finally free up memory. Do not use the
160 struct xfs_ioend *ioend,
163 struct inode *inode = ioend->io_inode;
164 struct bio *bio = &ioend->io_inline_bio;
165 struct bio *last = ioend->io_bio, *next;
166 u64 start = bio->bi_iter.bi_sector;
167 bool quiet = bio_flagged(bio, BIO_QUIET);
169 for (bio = &ioend->io_inline_bio; bio; bio = next) {
170 struct bio_vec *bvec;
174 * For the last bio, bi_private points to the ioend, so we
175 * need to explicitly end the iteration here.
180 next = bio->bi_private;
182 /* walk each page on bio, ending page IO on them */
183 bio_for_each_segment_all(bvec, bio, i)
184 xfs_finish_page_writeback(inode, bvec, error);
189 if (unlikely(error && !quiet)) {
190 xfs_err_ratelimited(XFS_I(inode)->i_mount,
191 "writeback error on sector %llu", start);
196 * Fast and loose check if this write could update the on-disk inode size.
198 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
200 return ioend->io_offset + ioend->io_size >
201 XFS_I(ioend->io_inode)->i_d.di_size;
205 xfs_setfilesize_trans_alloc(
206 struct xfs_ioend *ioend)
208 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
209 struct xfs_trans *tp;
212 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
216 ioend->io_append_trans = tp;
219 * We may pass freeze protection with a transaction. So tell lockdep
222 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
224 * We hand off the transaction to the completion thread now, so
225 * clear the flag here.
227 current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
232 * Update on-disk file size now that data has been written to disk.
236 struct xfs_inode *ip,
237 struct xfs_trans *tp,
243 xfs_ilock(ip, XFS_ILOCK_EXCL);
244 isize = xfs_new_eof(ip, offset + size);
246 xfs_iunlock(ip, XFS_ILOCK_EXCL);
247 xfs_trans_cancel(tp);
251 trace_xfs_setfilesize(ip, offset, size);
253 ip->i_d.di_size = isize;
254 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
255 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
257 return xfs_trans_commit(tp);
262 struct xfs_inode *ip,
266 struct xfs_mount *mp = ip->i_mount;
267 struct xfs_trans *tp;
270 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
274 return __xfs_setfilesize(ip, tp, offset, size);
278 xfs_setfilesize_ioend(
279 struct xfs_ioend *ioend,
282 struct xfs_inode *ip = XFS_I(ioend->io_inode);
283 struct xfs_trans *tp = ioend->io_append_trans;
286 * The transaction may have been allocated in the I/O submission thread,
287 * thus we need to mark ourselves as being in a transaction manually.
288 * Similarly for freeze protection.
290 current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
291 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
293 /* we abort the update if there was an IO error */
295 xfs_trans_cancel(tp);
299 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
303 * IO write completion.
307 struct work_struct *work)
309 struct xfs_ioend *ioend =
310 container_of(work, struct xfs_ioend, io_work);
311 struct xfs_inode *ip = XFS_I(ioend->io_inode);
312 xfs_off_t offset = ioend->io_offset;
313 size_t size = ioend->io_size;
317 * Just clean up the in-memory strutures if the fs has been shut down.
319 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
325 * Clean up any COW blocks on an I/O error.
327 error = blk_status_to_errno(ioend->io_bio->bi_status);
328 if (unlikely(error)) {
329 switch (ioend->io_type) {
331 xfs_reflink_cancel_cow_range(ip, offset, size, true);
339 * Success: commit the COW or unwritten blocks if needed.
341 switch (ioend->io_type) {
343 error = xfs_reflink_end_cow(ip, offset, size);
345 case XFS_IO_UNWRITTEN:
346 /* writeback should never update isize */
347 error = xfs_iomap_write_unwritten(ip, offset, size, false);
350 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
355 if (ioend->io_append_trans)
356 error = xfs_setfilesize_ioend(ioend, error);
357 xfs_destroy_ioend(ioend, error);
364 struct xfs_ioend *ioend = bio->bi_private;
365 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
367 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
368 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
369 else if (ioend->io_append_trans)
370 queue_work(mp->m_data_workqueue, &ioend->io_work);
372 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
379 struct xfs_bmbt_irec *imap,
382 struct xfs_inode *ip = XFS_I(inode);
383 struct xfs_mount *mp = ip->i_mount;
384 ssize_t count = i_blocksize(inode);
385 xfs_fileoff_t offset_fsb, end_fsb;
387 int bmapi_flags = XFS_BMAPI_ENTIRE;
390 if (XFS_FORCED_SHUTDOWN(mp))
393 ASSERT(type != XFS_IO_COW);
394 if (type == XFS_IO_UNWRITTEN)
395 bmapi_flags |= XFS_BMAPI_IGSTATE;
397 xfs_ilock(ip, XFS_ILOCK_SHARED);
398 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
399 (ip->i_df.if_flags & XFS_IFEXTENTS));
400 ASSERT(offset <= mp->m_super->s_maxbytes);
402 if ((xfs_ufsize_t)offset + count > mp->m_super->s_maxbytes)
403 count = mp->m_super->s_maxbytes - offset;
404 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
405 offset_fsb = XFS_B_TO_FSBT(mp, offset);
406 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
407 imap, &nimaps, bmapi_flags);
409 * Truncate an overwrite extent if there's a pending CoW
410 * reservation before the end of this extent. This forces us
411 * to come back to writepage to take care of the CoW.
413 if (nimaps && type == XFS_IO_OVERWRITE)
414 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
415 xfs_iunlock(ip, XFS_ILOCK_SHARED);
420 if (type == XFS_IO_DELALLOC &&
421 (!nimaps || isnullstartblock(imap->br_startblock))) {
422 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
425 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
430 if (type == XFS_IO_UNWRITTEN) {
432 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
433 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
437 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
444 struct xfs_bmbt_irec *imap,
447 offset >>= inode->i_blkbits;
450 * We have to make sure the cached mapping is within EOF to protect
451 * against eofblocks trimming on file release leaving us with a stale
452 * mapping. Otherwise, a page for a subsequent file extending buffered
453 * write could get picked up by this writeback cycle and written to the
456 * Note that what we really want here is a generic mapping invalidation
457 * mechanism to protect us from arbitrary extent modifying contexts, not
460 xfs_trim_extent_eof(imap, XFS_I(inode));
462 return offset >= imap->br_startoff &&
463 offset < imap->br_startoff + imap->br_blockcount;
467 xfs_start_buffer_writeback(
468 struct buffer_head *bh)
470 ASSERT(buffer_mapped(bh));
471 ASSERT(buffer_locked(bh));
472 ASSERT(!buffer_delay(bh));
473 ASSERT(!buffer_unwritten(bh));
476 set_buffer_async_write(bh);
477 set_buffer_uptodate(bh);
478 clear_buffer_dirty(bh);
482 xfs_start_page_writeback(
486 ASSERT(PageLocked(page));
487 ASSERT(!PageWriteback(page));
490 * if the page was not fully cleaned, we need to ensure that the higher
491 * layers come back to it correctly. That means we need to keep the page
492 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
493 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
494 * write this page in this writeback sweep will be made.
497 clear_page_dirty_for_io(page);
498 set_page_writeback(page);
500 set_page_writeback_keepwrite(page);
505 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
507 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
511 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
512 * it, and we submit that bio. The ioend may be used for multiple bio
513 * submissions, so we only want to allocate an append transaction for the ioend
514 * once. In the case of multiple bio submission, each bio will take an IO
515 * reference to the ioend to ensure that the ioend completion is only done once
516 * all bios have been submitted and the ioend is really done.
518 * If @fail is non-zero, it means that we have a situation where some part of
519 * the submission process has failed after we have marked paged for writeback
520 * and unlocked them. In this situation, we need to fail the bio and ioend
521 * rather than submit it to IO. This typically only happens on a filesystem
526 struct writeback_control *wbc,
527 struct xfs_ioend *ioend,
530 /* Convert CoW extents to regular */
531 if (!status && ioend->io_type == XFS_IO_COW) {
532 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
533 ioend->io_offset, ioend->io_size);
536 /* Reserve log space if we might write beyond the on-disk inode size. */
538 ioend->io_type != XFS_IO_UNWRITTEN &&
539 xfs_ioend_is_append(ioend) &&
540 !ioend->io_append_trans)
541 status = xfs_setfilesize_trans_alloc(ioend);
543 ioend->io_bio->bi_private = ioend;
544 ioend->io_bio->bi_end_io = xfs_end_bio;
545 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
548 * If we are failing the IO now, just mark the ioend with an
549 * error and finish it. This will run IO completion immediately
550 * as there is only one reference to the ioend at this point in
554 ioend->io_bio->bi_status = errno_to_blk_status(status);
555 bio_endio(ioend->io_bio);
559 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
560 submit_bio(ioend->io_bio);
565 xfs_init_bio_from_bh(
567 struct buffer_head *bh)
569 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
570 bio_set_dev(bio, bh->b_bdev);
573 static struct xfs_ioend *
578 struct buffer_head *bh)
580 struct xfs_ioend *ioend;
583 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
584 xfs_init_bio_from_bh(bio, bh);
586 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
587 INIT_LIST_HEAD(&ioend->io_list);
588 ioend->io_type = type;
589 ioend->io_inode = inode;
591 ioend->io_offset = offset;
592 INIT_WORK(&ioend->io_work, xfs_end_io);
593 ioend->io_append_trans = NULL;
599 * Allocate a new bio, and chain the old bio to the new one.
601 * Note that we have to do perform the chaining in this unintuitive order
602 * so that the bi_private linkage is set up in the right direction for the
603 * traversal in xfs_destroy_ioend().
607 struct xfs_ioend *ioend,
608 struct writeback_control *wbc,
609 struct buffer_head *bh)
613 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
614 xfs_init_bio_from_bh(new, bh);
616 bio_chain(ioend->io_bio, new);
617 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
618 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
619 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
620 submit_bio(ioend->io_bio);
625 * Test to see if we've been building up a completion structure for
626 * earlier buffers -- if so, we try to append to this ioend if we
627 * can, otherwise we finish off any current ioend and start another.
628 * Return the ioend we finished off so that the caller can submit it
629 * once it has finished processing the dirty page.
634 struct buffer_head *bh,
636 struct xfs_writepage_ctx *wpc,
637 struct writeback_control *wbc,
638 struct list_head *iolist)
640 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
641 bh->b_blocknr != wpc->last_block + 1 ||
642 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
644 list_add(&wpc->ioend->io_list, iolist);
645 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
649 * If the buffer doesn't fit into the bio we need to allocate a new
650 * one. This shouldn't happen more than once for a given buffer.
652 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
653 xfs_chain_bio(wpc->ioend, wbc, bh);
655 wpc->ioend->io_size += bh->b_size;
656 wpc->last_block = bh->b_blocknr;
657 xfs_start_buffer_writeback(bh);
663 struct buffer_head *bh,
664 struct xfs_bmbt_irec *imap,
668 struct xfs_mount *m = XFS_I(inode)->i_mount;
669 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
670 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
672 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
673 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
675 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
676 ((offset - iomap_offset) >> inode->i_blkbits);
678 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
681 set_buffer_mapped(bh);
687 struct buffer_head *bh,
688 struct xfs_bmbt_irec *imap,
691 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
692 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
694 xfs_map_buffer(inode, bh, imap, offset);
695 set_buffer_mapped(bh);
696 clear_buffer_delay(bh);
697 clear_buffer_unwritten(bh);
701 * Test if a given page contains at least one buffer of a given @type.
702 * If @check_all_buffers is true, then we walk all the buffers in the page to
703 * try to find one of the type passed in. If it is not set, then the caller only
704 * needs to check the first buffer on the page for a match.
710 bool check_all_buffers)
712 struct buffer_head *bh;
713 struct buffer_head *head;
715 if (PageWriteback(page))
719 if (!page_has_buffers(page))
722 bh = head = page_buffers(page);
724 if (buffer_unwritten(bh)) {
725 if (type == XFS_IO_UNWRITTEN)
727 } else if (buffer_delay(bh)) {
728 if (type == XFS_IO_DELALLOC)
730 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
731 if (type == XFS_IO_OVERWRITE)
735 /* If we are only checking the first buffer, we are done now. */
736 if (!check_all_buffers)
738 } while ((bh = bh->b_this_page) != head);
744 xfs_vm_invalidatepage(
749 trace_xfs_invalidatepage(page->mapping->host, page, offset,
753 * If we are invalidating the entire page, clear the dirty state from it
754 * so that we can check for attempts to release dirty cached pages in
755 * xfs_vm_releasepage().
757 if (offset == 0 && length >= PAGE_SIZE)
758 cancel_dirty_page(page);
759 block_invalidatepage(page, offset, length);
763 * If the page has delalloc buffers on it, we need to punch them out before we
764 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
765 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
766 * is done on that same region - the delalloc extent is returned when none is
767 * supposed to be there.
769 * We prevent this by truncating away the delalloc regions on the page before
770 * invalidating it. Because they are delalloc, we can do this without needing a
771 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
772 * truncation without a transaction as there is no space left for block
773 * reservation (typically why we see a ENOSPC in writeback).
775 * This is not a performance critical path, so for now just do the punching a
776 * buffer head at a time.
779 xfs_aops_discard_page(
782 struct inode *inode = page->mapping->host;
783 struct xfs_inode *ip = XFS_I(inode);
784 struct buffer_head *bh, *head;
785 loff_t offset = page_offset(page);
787 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
790 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
793 xfs_alert(ip->i_mount,
794 "page discard on page %p, inode 0x%llx, offset %llu.",
795 page, ip->i_ino, offset);
797 xfs_ilock(ip, XFS_ILOCK_EXCL);
798 bh = head = page_buffers(page);
801 xfs_fileoff_t start_fsb;
803 if (!buffer_delay(bh))
806 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
807 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
809 /* something screwed, just bail */
810 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
811 xfs_alert(ip->i_mount,
812 "page discard unable to remove delalloc mapping.");
817 offset += i_blocksize(inode);
819 } while ((bh = bh->b_this_page) != head);
821 xfs_iunlock(ip, XFS_ILOCK_EXCL);
823 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
829 struct xfs_writepage_ctx *wpc,
832 unsigned int *new_type)
834 struct xfs_inode *ip = XFS_I(inode);
835 struct xfs_bmbt_irec imap;
840 * If we already have a valid COW mapping keep using it.
842 if (wpc->io_type == XFS_IO_COW) {
843 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
844 if (wpc->imap_valid) {
845 *new_type = XFS_IO_COW;
851 * Else we need to check if there is a COW mapping at this offset.
853 xfs_ilock(ip, XFS_ILOCK_SHARED);
854 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
855 xfs_iunlock(ip, XFS_ILOCK_SHARED);
861 * And if the COW mapping has a delayed extent here we need to
862 * allocate real space for it now.
864 if (isnullstartblock(imap.br_startblock)) {
865 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
871 wpc->io_type = *new_type = XFS_IO_COW;
872 wpc->imap_valid = true;
878 * We implement an immediate ioend submission policy here to avoid needing to
879 * chain multiple ioends and hence nest mempool allocations which can violate
880 * forward progress guarantees we need to provide. The current ioend we are
881 * adding buffers to is cached on the writepage context, and if the new buffer
882 * does not append to the cached ioend it will create a new ioend and cache that
885 * If a new ioend is created and cached, the old ioend is returned and queued
886 * locally for submission once the entire page is processed or an error has been
887 * detected. While ioends are submitted immediately after they are completed,
888 * batching optimisations are provided by higher level block plugging.
890 * At the end of a writeback pass, there will be a cached ioend remaining on the
891 * writepage context that the caller will need to submit.
895 struct xfs_writepage_ctx *wpc,
896 struct writeback_control *wbc,
902 LIST_HEAD(submit_list);
903 struct xfs_ioend *ioend, *next;
904 struct buffer_head *bh, *head;
905 ssize_t len = i_blocksize(inode);
909 unsigned int new_type;
911 bh = head = page_buffers(page);
912 offset = page_offset(page);
914 if (offset >= end_offset)
916 if (!buffer_uptodate(bh))
920 * set_page_dirty dirties all buffers in a page, independent
921 * of their state. The dirty state however is entirely
922 * meaningless for holes (!mapped && uptodate), so skip
923 * buffers covering holes here.
925 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
926 wpc->imap_valid = false;
930 if (buffer_unwritten(bh))
931 new_type = XFS_IO_UNWRITTEN;
932 else if (buffer_delay(bh))
933 new_type = XFS_IO_DELALLOC;
934 else if (buffer_uptodate(bh))
935 new_type = XFS_IO_OVERWRITE;
937 if (PageUptodate(page))
938 ASSERT(buffer_mapped(bh));
940 * This buffer is not uptodate and will not be
941 * written to disk. Ensure that we will put any
942 * subsequent writeable buffers into a new
945 wpc->imap_valid = false;
949 if (xfs_is_reflink_inode(XFS_I(inode))) {
950 error = xfs_map_cow(wpc, inode, offset, &new_type);
955 if (wpc->io_type != new_type) {
956 wpc->io_type = new_type;
957 wpc->imap_valid = false;
961 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
963 if (!wpc->imap_valid) {
964 error = xfs_map_blocks(inode, offset, &wpc->imap,
968 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
971 if (wpc->imap_valid) {
973 if (wpc->io_type != XFS_IO_OVERWRITE)
974 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
975 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
979 } while (offset += len, ((bh = bh->b_this_page) != head));
981 if (uptodate && bh == head)
982 SetPageUptodate(page);
984 ASSERT(wpc->ioend || list_empty(&submit_list));
988 * On error, we have to fail the ioend here because we have locked
989 * buffers in the ioend. If we don't do this, we'll deadlock
990 * invalidating the page as that tries to lock the buffers on the page.
991 * Also, because we may have set pages under writeback, we have to make
992 * sure we run IO completion to mark the error state of the IO
993 * appropriately, so we can't cancel the ioend directly here. That means
994 * we have to mark this page as under writeback if we included any
995 * buffers from it in the ioend chain so that completion treats it
998 * If we didn't include the page in the ioend, the on error we can
999 * simply discard and unlock it as there are no other users of the page
1000 * or it's buffers right now. The caller will still need to trigger
1001 * submission of outstanding ioends on the writepage context so they are
1002 * treated correctly on error.
1005 xfs_start_page_writeback(page, !error);
1008 * Preserve the original error if there was one, otherwise catch
1009 * submission errors here and propagate into subsequent ioend
1012 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
1015 list_del_init(&ioend->io_list);
1016 error2 = xfs_submit_ioend(wbc, ioend, error);
1017 if (error2 && !error)
1021 xfs_aops_discard_page(page);
1022 ClearPageUptodate(page);
1026 * We can end up here with no error and nothing to write if we
1027 * race with a partial page truncate on a sub-page block sized
1028 * filesystem. In that case we need to mark the page clean.
1030 xfs_start_page_writeback(page, 1);
1031 end_page_writeback(page);
1034 mapping_set_error(page->mapping, error);
1039 * Write out a dirty page.
1041 * For delalloc space on the page we need to allocate space and flush it.
1042 * For unwritten space on the page we need to start the conversion to
1043 * regular allocated space.
1044 * For any other dirty buffer heads on the page we should flush them.
1049 struct writeback_control *wbc,
1052 struct xfs_writepage_ctx *wpc = data;
1053 struct inode *inode = page->mapping->host;
1055 uint64_t end_offset;
1058 trace_xfs_writepage(inode, page, 0, 0);
1060 ASSERT(page_has_buffers(page));
1063 * Refuse to write the page out if we are called from reclaim context.
1065 * This avoids stack overflows when called from deeply used stacks in
1066 * random callers for direct reclaim or memcg reclaim. We explicitly
1067 * allow reclaim from kswapd as the stack usage there is relatively low.
1069 * This should never happen except in the case of a VM regression so
1072 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1077 * Given that we do not allow direct reclaim to call us, we should
1078 * never be called while in a filesystem transaction.
1080 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
1084 * Is this page beyond the end of the file?
1086 * The page index is less than the end_index, adjust the end_offset
1087 * to the highest offset that this page should represent.
1088 * -----------------------------------------------------
1089 * | file mapping | <EOF> |
1090 * -----------------------------------------------------
1091 * | Page ... | Page N-2 | Page N-1 | Page N | |
1092 * ^--------------------------------^----------|--------
1093 * | desired writeback range | see else |
1094 * ---------------------------------^------------------|
1096 offset = i_size_read(inode);
1097 end_index = offset >> PAGE_SHIFT;
1098 if (page->index < end_index)
1099 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1102 * Check whether the page to write out is beyond or straddles
1104 * -------------------------------------------------------
1105 * | file mapping | <EOF> |
1106 * -------------------------------------------------------
1107 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1108 * ^--------------------------------^-----------|---------
1110 * ---------------------------------^-----------|--------|
1112 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1115 * Skip the page if it is fully outside i_size, e.g. due to a
1116 * truncate operation that is in progress. We must redirty the
1117 * page so that reclaim stops reclaiming it. Otherwise
1118 * xfs_vm_releasepage() is called on it and gets confused.
1120 * Note that the end_index is unsigned long, it would overflow
1121 * if the given offset is greater than 16TB on 32-bit system
1122 * and if we do check the page is fully outside i_size or not
1123 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1124 * will be evaluated to 0. Hence this page will be redirtied
1125 * and be written out repeatedly which would result in an
1126 * infinite loop, the user program that perform this operation
1127 * will hang. Instead, we can verify this situation by checking
1128 * if the page to write is totally beyond the i_size or if it's
1129 * offset is just equal to the EOF.
1131 if (page->index > end_index ||
1132 (page->index == end_index && offset_into_page == 0))
1136 * The page straddles i_size. It must be zeroed out on each
1137 * and every writepage invocation because it may be mmapped.
1138 * "A file is mapped in multiples of the page size. For a file
1139 * that is not a multiple of the page size, the remaining
1140 * memory is zeroed when mapped, and writes to that region are
1141 * not written out to the file."
1143 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1145 /* Adjust the end_offset to the end of file */
1146 end_offset = offset;
1149 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1152 redirty_page_for_writepage(wbc, page);
1160 struct writeback_control *wbc)
1162 struct xfs_writepage_ctx wpc = {
1163 .io_type = XFS_IO_INVALID,
1167 ret = xfs_do_writepage(page, wbc, &wpc);
1169 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1175 struct address_space *mapping,
1176 struct writeback_control *wbc)
1178 struct xfs_writepage_ctx wpc = {
1179 .io_type = XFS_IO_INVALID,
1183 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1184 if (dax_mapping(mapping))
1185 return dax_writeback_mapping_range(mapping,
1186 xfs_find_bdev_for_inode(mapping->host), wbc);
1188 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1190 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1195 * Called to move a page into cleanable state - and from there
1196 * to be released. The page should already be clean. We always
1197 * have buffer heads in this call.
1199 * Returns 1 if the page is ok to release, 0 otherwise.
1206 int delalloc, unwritten;
1208 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1211 * mm accommodates an old ext3 case where clean pages might not have had
1212 * the dirty bit cleared. Thus, it can send actual dirty pages to
1213 * ->releasepage() via shrink_active_list(). Conversely,
1214 * block_invalidatepage() can send pages that are still marked dirty but
1215 * otherwise have invalidated buffers.
1217 * We want to release the latter to avoid unnecessary buildup of the
1218 * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages
1219 * that are entirely invalidated and need to be released. Hence the
1220 * only time we should get dirty pages here is through
1221 * shrink_active_list() and so we can simply skip those now.
1223 * warn if we've left any lingering delalloc/unwritten buffers on clean
1224 * or invalidated pages we are about to release.
1226 if (PageDirty(page))
1229 xfs_count_page_state(page, &delalloc, &unwritten);
1231 if (WARN_ON_ONCE(delalloc))
1233 if (WARN_ON_ONCE(unwritten))
1236 return try_to_free_buffers(page);
1240 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1241 * is, so that we can avoid repeated get_blocks calls.
1243 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1244 * for blocks beyond EOF must be marked new so that sub block regions can be
1245 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1246 * was just allocated or is unwritten, otherwise the callers would overwrite
1247 * existing data with zeros. Hence we have to split the mapping into a range up
1248 * to and including EOF, and a second mapping for beyond EOF.
1252 struct inode *inode,
1254 struct buffer_head *bh_result,
1255 struct xfs_bmbt_irec *imap,
1259 xfs_off_t mapping_size;
1261 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1262 mapping_size <<= inode->i_blkbits;
1264 ASSERT(mapping_size > 0);
1265 if (mapping_size > size)
1266 mapping_size = size;
1267 if (offset < i_size_read(inode) &&
1268 (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) {
1269 /* limit mapping to block that spans EOF */
1270 mapping_size = roundup_64(i_size_read(inode) - offset,
1271 i_blocksize(inode));
1273 if (mapping_size > LONG_MAX)
1274 mapping_size = LONG_MAX;
1276 bh_result->b_size = mapping_size;
1281 struct inode *inode,
1283 struct buffer_head *bh_result,
1286 struct xfs_inode *ip = XFS_I(inode);
1287 struct xfs_mount *mp = ip->i_mount;
1288 xfs_fileoff_t offset_fsb, end_fsb;
1291 struct xfs_bmbt_irec imap;
1298 if (XFS_FORCED_SHUTDOWN(mp))
1301 offset = (xfs_off_t)iblock << inode->i_blkbits;
1302 ASSERT(bh_result->b_size >= i_blocksize(inode));
1303 size = bh_result->b_size;
1305 if (offset >= i_size_read(inode))
1309 * Direct I/O is usually done on preallocated files, so try getting
1310 * a block mapping without an exclusive lock first.
1312 lockmode = xfs_ilock_data_map_shared(ip);
1314 ASSERT(offset <= mp->m_super->s_maxbytes);
1315 if ((xfs_ufsize_t)offset + size > mp->m_super->s_maxbytes)
1316 size = mp->m_super->s_maxbytes - offset;
1317 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1318 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1320 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1321 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1326 trace_xfs_get_blocks_found(ip, offset, size,
1327 imap.br_state == XFS_EXT_UNWRITTEN ?
1328 XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap);
1329 xfs_iunlock(ip, lockmode);
1331 trace_xfs_get_blocks_notfound(ip, offset, size);
1335 /* trim mapping down to size requested */
1336 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1339 * For unwritten extents do not report a disk address in the buffered
1340 * read case (treat as if we're reading into a hole).
1342 if (xfs_bmap_is_real_extent(&imap))
1343 xfs_map_buffer(inode, bh_result, &imap, offset);
1346 * If this is a realtime file, data may be on a different device.
1347 * to that pointed to from the buffer_head b_bdev currently.
1349 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1353 xfs_iunlock(ip, lockmode);
1360 struct iov_iter *iter)
1363 * We just need the method present so that open/fcntl allow direct I/O.
1370 struct address_space *mapping,
1373 struct inode *inode = (struct inode *)mapping->host;
1374 struct xfs_inode *ip = XFS_I(inode);
1376 trace_xfs_vm_bmap(XFS_I(inode));
1379 * The swap code (ab-)uses ->bmap to get a block mapping and then
1380 * bypasseѕ the file system for actual I/O. We really can't allow
1381 * that on reflinks inodes, so we have to skip out here. And yes,
1382 * 0 is the magic code for a bmap error.
1384 * Since we don't pass back blockdev info, we can't return bmap
1385 * information for rt files either.
1387 if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip))
1390 filemap_write_and_wait(mapping);
1391 return generic_block_bmap(mapping, block, xfs_get_blocks);
1396 struct file *unused,
1399 trace_xfs_vm_readpage(page->mapping->host, 1);
1400 return mpage_readpage(page, xfs_get_blocks);
1405 struct file *unused,
1406 struct address_space *mapping,
1407 struct list_head *pages,
1410 trace_xfs_vm_readpages(mapping->host, nr_pages);
1411 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1415 * This is basically a copy of __set_page_dirty_buffers() with one
1416 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1417 * dirty, we'll never be able to clean them because we don't write buffers
1418 * beyond EOF, and that means we can't invalidate pages that span EOF
1419 * that have been marked dirty. Further, the dirty state can leak into
1420 * the file interior if the file is extended, resulting in all sorts of
1421 * bad things happening as the state does not match the underlying data.
1423 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1424 * this only exist because of bufferheads and how the generic code manages them.
1427 xfs_vm_set_page_dirty(
1430 struct address_space *mapping = page->mapping;
1431 struct inode *inode = mapping->host;
1436 if (unlikely(!mapping))
1437 return !TestSetPageDirty(page);
1439 end_offset = i_size_read(inode);
1440 offset = page_offset(page);
1442 spin_lock(&mapping->private_lock);
1443 if (page_has_buffers(page)) {
1444 struct buffer_head *head = page_buffers(page);
1445 struct buffer_head *bh = head;
1448 if (offset < end_offset)
1449 set_buffer_dirty(bh);
1450 bh = bh->b_this_page;
1451 offset += i_blocksize(inode);
1452 } while (bh != head);
1455 * Lock out page->mem_cgroup migration to keep PageDirty
1456 * synchronized with per-memcg dirty page counters.
1458 lock_page_memcg(page);
1459 newly_dirty = !TestSetPageDirty(page);
1460 spin_unlock(&mapping->private_lock);
1463 /* sigh - __set_page_dirty() is static, so copy it here, too */
1464 unsigned long flags;
1466 spin_lock_irqsave(&mapping->tree_lock, flags);
1467 if (page->mapping) { /* Race with truncate? */
1468 WARN_ON_ONCE(!PageUptodate(page));
1469 account_page_dirtied(page, mapping);
1470 radix_tree_tag_set(&mapping->page_tree,
1471 page_index(page), PAGECACHE_TAG_DIRTY);
1473 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1475 unlock_page_memcg(page);
1477 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1481 const struct address_space_operations xfs_address_space_operations = {
1482 .readpage = xfs_vm_readpage,
1483 .readpages = xfs_vm_readpages,
1484 .writepage = xfs_vm_writepage,
1485 .writepages = xfs_vm_writepages,
1486 .set_page_dirty = xfs_vm_set_page_dirty,
1487 .releasepage = xfs_vm_releasepage,
1488 .invalidatepage = xfs_vm_invalidatepage,
1489 .bmap = xfs_vm_bmap,
1490 .direct_IO = xfs_vm_direct_IO,
1491 .migratepage = buffer_migrate_page,
1492 .is_partially_uptodate = block_is_partially_uptodate,
1493 .error_remove_page = generic_error_remove_page,