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 error = xfs_iomap_write_unwritten(ip, offset, size);
349 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
354 if (ioend->io_append_trans)
355 error = xfs_setfilesize_ioend(ioend, error);
356 xfs_destroy_ioend(ioend, error);
363 struct xfs_ioend *ioend = bio->bi_private;
364 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
366 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
367 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
368 else if (ioend->io_append_trans)
369 queue_work(mp->m_data_workqueue, &ioend->io_work);
371 xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status));
378 struct xfs_bmbt_irec *imap,
381 struct xfs_inode *ip = XFS_I(inode);
382 struct xfs_mount *mp = ip->i_mount;
383 ssize_t count = i_blocksize(inode);
384 xfs_fileoff_t offset_fsb, end_fsb;
386 int bmapi_flags = XFS_BMAPI_ENTIRE;
389 if (XFS_FORCED_SHUTDOWN(mp))
392 ASSERT(type != XFS_IO_COW);
393 if (type == XFS_IO_UNWRITTEN)
394 bmapi_flags |= XFS_BMAPI_IGSTATE;
396 xfs_ilock(ip, XFS_ILOCK_SHARED);
397 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
398 (ip->i_df.if_flags & XFS_IFEXTENTS));
399 ASSERT(offset <= mp->m_super->s_maxbytes);
401 if (offset + count > mp->m_super->s_maxbytes)
402 count = mp->m_super->s_maxbytes - offset;
403 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
404 offset_fsb = XFS_B_TO_FSBT(mp, offset);
405 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
406 imap, &nimaps, bmapi_flags);
408 * Truncate an overwrite extent if there's a pending CoW
409 * reservation before the end of this extent. This forces us
410 * to come back to writepage to take care of the CoW.
412 if (nimaps && type == XFS_IO_OVERWRITE)
413 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
414 xfs_iunlock(ip, XFS_ILOCK_SHARED);
419 if (type == XFS_IO_DELALLOC &&
420 (!nimaps || isnullstartblock(imap->br_startblock))) {
421 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
424 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
429 if (type == XFS_IO_UNWRITTEN) {
431 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
432 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
436 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
443 struct xfs_bmbt_irec *imap,
446 offset >>= inode->i_blkbits;
448 return offset >= imap->br_startoff &&
449 offset < imap->br_startoff + imap->br_blockcount;
453 xfs_start_buffer_writeback(
454 struct buffer_head *bh)
456 ASSERT(buffer_mapped(bh));
457 ASSERT(buffer_locked(bh));
458 ASSERT(!buffer_delay(bh));
459 ASSERT(!buffer_unwritten(bh));
462 set_buffer_async_write(bh);
463 set_buffer_uptodate(bh);
464 clear_buffer_dirty(bh);
468 xfs_start_page_writeback(
472 ASSERT(PageLocked(page));
473 ASSERT(!PageWriteback(page));
476 * if the page was not fully cleaned, we need to ensure that the higher
477 * layers come back to it correctly. That means we need to keep the page
478 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
479 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
480 * write this page in this writeback sweep will be made.
483 clear_page_dirty_for_io(page);
484 set_page_writeback(page);
486 set_page_writeback_keepwrite(page);
491 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
493 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
497 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
498 * it, and we submit that bio. The ioend may be used for multiple bio
499 * submissions, so we only want to allocate an append transaction for the ioend
500 * once. In the case of multiple bio submission, each bio will take an IO
501 * reference to the ioend to ensure that the ioend completion is only done once
502 * all bios have been submitted and the ioend is really done.
504 * If @fail is non-zero, it means that we have a situation where some part of
505 * the submission process has failed after we have marked paged for writeback
506 * and unlocked them. In this situation, we need to fail the bio and ioend
507 * rather than submit it to IO. This typically only happens on a filesystem
512 struct writeback_control *wbc,
513 struct xfs_ioend *ioend,
516 /* Convert CoW extents to regular */
517 if (!status && ioend->io_type == XFS_IO_COW) {
518 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
519 ioend->io_offset, ioend->io_size);
522 /* Reserve log space if we might write beyond the on-disk inode size. */
524 ioend->io_type != XFS_IO_UNWRITTEN &&
525 xfs_ioend_is_append(ioend) &&
526 !ioend->io_append_trans)
527 status = xfs_setfilesize_trans_alloc(ioend);
529 ioend->io_bio->bi_private = ioend;
530 ioend->io_bio->bi_end_io = xfs_end_bio;
531 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
534 * If we are failing the IO now, just mark the ioend with an
535 * error and finish it. This will run IO completion immediately
536 * as there is only one reference to the ioend at this point in
540 ioend->io_bio->bi_status = errno_to_blk_status(status);
541 bio_endio(ioend->io_bio);
545 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
546 submit_bio(ioend->io_bio);
551 xfs_init_bio_from_bh(
553 struct buffer_head *bh)
555 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
556 bio_set_dev(bio, bh->b_bdev);
559 static struct xfs_ioend *
564 struct buffer_head *bh)
566 struct xfs_ioend *ioend;
569 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
570 xfs_init_bio_from_bh(bio, bh);
572 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
573 INIT_LIST_HEAD(&ioend->io_list);
574 ioend->io_type = type;
575 ioend->io_inode = inode;
577 ioend->io_offset = offset;
578 INIT_WORK(&ioend->io_work, xfs_end_io);
579 ioend->io_append_trans = NULL;
585 * Allocate a new bio, and chain the old bio to the new one.
587 * Note that we have to do perform the chaining in this unintuitive order
588 * so that the bi_private linkage is set up in the right direction for the
589 * traversal in xfs_destroy_ioend().
593 struct xfs_ioend *ioend,
594 struct writeback_control *wbc,
595 struct buffer_head *bh)
599 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
600 xfs_init_bio_from_bh(new, bh);
602 bio_chain(ioend->io_bio, new);
603 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
604 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
605 ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint;
606 submit_bio(ioend->io_bio);
611 * Test to see if we've been building up a completion structure for
612 * earlier buffers -- if so, we try to append to this ioend if we
613 * can, otherwise we finish off any current ioend and start another.
614 * Return the ioend we finished off so that the caller can submit it
615 * once it has finished processing the dirty page.
620 struct buffer_head *bh,
622 struct xfs_writepage_ctx *wpc,
623 struct writeback_control *wbc,
624 struct list_head *iolist)
626 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
627 bh->b_blocknr != wpc->last_block + 1 ||
628 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
630 list_add(&wpc->ioend->io_list, iolist);
631 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
635 * If the buffer doesn't fit into the bio we need to allocate a new
636 * one. This shouldn't happen more than once for a given buffer.
638 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
639 xfs_chain_bio(wpc->ioend, wbc, bh);
641 wpc->ioend->io_size += bh->b_size;
642 wpc->last_block = bh->b_blocknr;
643 xfs_start_buffer_writeback(bh);
649 struct buffer_head *bh,
650 struct xfs_bmbt_irec *imap,
654 struct xfs_mount *m = XFS_I(inode)->i_mount;
655 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
656 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
658 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
659 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
661 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
662 ((offset - iomap_offset) >> inode->i_blkbits);
664 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
667 set_buffer_mapped(bh);
673 struct buffer_head *bh,
674 struct xfs_bmbt_irec *imap,
677 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
678 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
680 xfs_map_buffer(inode, bh, imap, offset);
681 set_buffer_mapped(bh);
682 clear_buffer_delay(bh);
683 clear_buffer_unwritten(bh);
687 * Test if a given page contains at least one buffer of a given @type.
688 * If @check_all_buffers is true, then we walk all the buffers in the page to
689 * try to find one of the type passed in. If it is not set, then the caller only
690 * needs to check the first buffer on the page for a match.
696 bool check_all_buffers)
698 struct buffer_head *bh;
699 struct buffer_head *head;
701 if (PageWriteback(page))
705 if (!page_has_buffers(page))
708 bh = head = page_buffers(page);
710 if (buffer_unwritten(bh)) {
711 if (type == XFS_IO_UNWRITTEN)
713 } else if (buffer_delay(bh)) {
714 if (type == XFS_IO_DELALLOC)
716 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
717 if (type == XFS_IO_OVERWRITE)
721 /* If we are only checking the first buffer, we are done now. */
722 if (!check_all_buffers)
724 } while ((bh = bh->b_this_page) != head);
730 xfs_vm_invalidatepage(
735 trace_xfs_invalidatepage(page->mapping->host, page, offset,
737 block_invalidatepage(page, offset, length);
741 * If the page has delalloc buffers on it, we need to punch them out before we
742 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
743 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
744 * is done on that same region - the delalloc extent is returned when none is
745 * supposed to be there.
747 * We prevent this by truncating away the delalloc regions on the page before
748 * invalidating it. Because they are delalloc, we can do this without needing a
749 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
750 * truncation without a transaction as there is no space left for block
751 * reservation (typically why we see a ENOSPC in writeback).
753 * This is not a performance critical path, so for now just do the punching a
754 * buffer head at a time.
757 xfs_aops_discard_page(
760 struct inode *inode = page->mapping->host;
761 struct xfs_inode *ip = XFS_I(inode);
762 struct buffer_head *bh, *head;
763 loff_t offset = page_offset(page);
765 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
768 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
771 xfs_alert(ip->i_mount,
772 "page discard on page %p, inode 0x%llx, offset %llu.",
773 page, ip->i_ino, offset);
775 xfs_ilock(ip, XFS_ILOCK_EXCL);
776 bh = head = page_buffers(page);
779 xfs_fileoff_t start_fsb;
781 if (!buffer_delay(bh))
784 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
785 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
787 /* something screwed, just bail */
788 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
789 xfs_alert(ip->i_mount,
790 "page discard unable to remove delalloc mapping.");
795 offset += i_blocksize(inode);
797 } while ((bh = bh->b_this_page) != head);
799 xfs_iunlock(ip, XFS_ILOCK_EXCL);
801 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
807 struct xfs_writepage_ctx *wpc,
810 unsigned int *new_type)
812 struct xfs_inode *ip = XFS_I(inode);
813 struct xfs_bmbt_irec imap;
818 * If we already have a valid COW mapping keep using it.
820 if (wpc->io_type == XFS_IO_COW) {
821 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
822 if (wpc->imap_valid) {
823 *new_type = XFS_IO_COW;
829 * Else we need to check if there is a COW mapping at this offset.
831 xfs_ilock(ip, XFS_ILOCK_SHARED);
832 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
833 xfs_iunlock(ip, XFS_ILOCK_SHARED);
839 * And if the COW mapping has a delayed extent here we need to
840 * allocate real space for it now.
842 if (isnullstartblock(imap.br_startblock)) {
843 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
849 wpc->io_type = *new_type = XFS_IO_COW;
850 wpc->imap_valid = true;
856 * We implement an immediate ioend submission policy here to avoid needing to
857 * chain multiple ioends and hence nest mempool allocations which can violate
858 * forward progress guarantees we need to provide. The current ioend we are
859 * adding buffers to is cached on the writepage context, and if the new buffer
860 * does not append to the cached ioend it will create a new ioend and cache that
863 * If a new ioend is created and cached, the old ioend is returned and queued
864 * locally for submission once the entire page is processed or an error has been
865 * detected. While ioends are submitted immediately after they are completed,
866 * batching optimisations are provided by higher level block plugging.
868 * At the end of a writeback pass, there will be a cached ioend remaining on the
869 * writepage context that the caller will need to submit.
873 struct xfs_writepage_ctx *wpc,
874 struct writeback_control *wbc,
880 LIST_HEAD(submit_list);
881 struct xfs_ioend *ioend, *next;
882 struct buffer_head *bh, *head;
883 ssize_t len = i_blocksize(inode);
887 unsigned int new_type;
889 bh = head = page_buffers(page);
890 offset = page_offset(page);
892 if (offset >= end_offset)
894 if (!buffer_uptodate(bh))
898 * set_page_dirty dirties all buffers in a page, independent
899 * of their state. The dirty state however is entirely
900 * meaningless for holes (!mapped && uptodate), so skip
901 * buffers covering holes here.
903 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
904 wpc->imap_valid = false;
908 if (buffer_unwritten(bh))
909 new_type = XFS_IO_UNWRITTEN;
910 else if (buffer_delay(bh))
911 new_type = XFS_IO_DELALLOC;
912 else if (buffer_uptodate(bh))
913 new_type = XFS_IO_OVERWRITE;
915 if (PageUptodate(page))
916 ASSERT(buffer_mapped(bh));
918 * This buffer is not uptodate and will not be
919 * written to disk. Ensure that we will put any
920 * subsequent writeable buffers into a new
923 wpc->imap_valid = false;
927 if (xfs_is_reflink_inode(XFS_I(inode))) {
928 error = xfs_map_cow(wpc, inode, offset, &new_type);
933 if (wpc->io_type != new_type) {
934 wpc->io_type = new_type;
935 wpc->imap_valid = false;
939 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
941 if (!wpc->imap_valid) {
942 error = xfs_map_blocks(inode, offset, &wpc->imap,
946 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
949 if (wpc->imap_valid) {
951 if (wpc->io_type != XFS_IO_OVERWRITE)
952 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
953 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
957 } while (offset += len, ((bh = bh->b_this_page) != head));
959 if (uptodate && bh == head)
960 SetPageUptodate(page);
962 ASSERT(wpc->ioend || list_empty(&submit_list));
966 * On error, we have to fail the ioend here because we have locked
967 * buffers in the ioend. If we don't do this, we'll deadlock
968 * invalidating the page as that tries to lock the buffers on the page.
969 * Also, because we may have set pages under writeback, we have to make
970 * sure we run IO completion to mark the error state of the IO
971 * appropriately, so we can't cancel the ioend directly here. That means
972 * we have to mark this page as under writeback if we included any
973 * buffers from it in the ioend chain so that completion treats it
976 * If we didn't include the page in the ioend, the on error we can
977 * simply discard and unlock it as there are no other users of the page
978 * or it's buffers right now. The caller will still need to trigger
979 * submission of outstanding ioends on the writepage context so they are
980 * treated correctly on error.
983 xfs_start_page_writeback(page, !error);
986 * Preserve the original error if there was one, otherwise catch
987 * submission errors here and propagate into subsequent ioend
990 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
993 list_del_init(&ioend->io_list);
994 error2 = xfs_submit_ioend(wbc, ioend, error);
995 if (error2 && !error)
999 xfs_aops_discard_page(page);
1000 ClearPageUptodate(page);
1004 * We can end up here with no error and nothing to write if we
1005 * race with a partial page truncate on a sub-page block sized
1006 * filesystem. In that case we need to mark the page clean.
1008 xfs_start_page_writeback(page, 1);
1009 end_page_writeback(page);
1012 mapping_set_error(page->mapping, error);
1017 * Write out a dirty page.
1019 * For delalloc space on the page we need to allocate space and flush it.
1020 * For unwritten space on the page we need to start the conversion to
1021 * regular allocated space.
1022 * For any other dirty buffer heads on the page we should flush them.
1027 struct writeback_control *wbc,
1030 struct xfs_writepage_ctx *wpc = data;
1031 struct inode *inode = page->mapping->host;
1033 uint64_t end_offset;
1036 trace_xfs_writepage(inode, page, 0, 0);
1038 ASSERT(page_has_buffers(page));
1041 * Refuse to write the page out if we are called from reclaim context.
1043 * This avoids stack overflows when called from deeply used stacks in
1044 * random callers for direct reclaim or memcg reclaim. We explicitly
1045 * allow reclaim from kswapd as the stack usage there is relatively low.
1047 * This should never happen except in the case of a VM regression so
1050 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1055 * Given that we do not allow direct reclaim to call us, we should
1056 * never be called while in a filesystem transaction.
1058 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
1062 * Is this page beyond the end of the file?
1064 * The page index is less than the end_index, adjust the end_offset
1065 * to the highest offset that this page should represent.
1066 * -----------------------------------------------------
1067 * | file mapping | <EOF> |
1068 * -----------------------------------------------------
1069 * | Page ... | Page N-2 | Page N-1 | Page N | |
1070 * ^--------------------------------^----------|--------
1071 * | desired writeback range | see else |
1072 * ---------------------------------^------------------|
1074 offset = i_size_read(inode);
1075 end_index = offset >> PAGE_SHIFT;
1076 if (page->index < end_index)
1077 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1080 * Check whether the page to write out is beyond or straddles
1082 * -------------------------------------------------------
1083 * | file mapping | <EOF> |
1084 * -------------------------------------------------------
1085 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1086 * ^--------------------------------^-----------|---------
1088 * ---------------------------------^-----------|--------|
1090 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1093 * Skip the page if it is fully outside i_size, e.g. due to a
1094 * truncate operation that is in progress. We must redirty the
1095 * page so that reclaim stops reclaiming it. Otherwise
1096 * xfs_vm_releasepage() is called on it and gets confused.
1098 * Note that the end_index is unsigned long, it would overflow
1099 * if the given offset is greater than 16TB on 32-bit system
1100 * and if we do check the page is fully outside i_size or not
1101 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1102 * will be evaluated to 0. Hence this page will be redirtied
1103 * and be written out repeatedly which would result in an
1104 * infinite loop, the user program that perform this operation
1105 * will hang. Instead, we can verify this situation by checking
1106 * if the page to write is totally beyond the i_size or if it's
1107 * offset is just equal to the EOF.
1109 if (page->index > end_index ||
1110 (page->index == end_index && offset_into_page == 0))
1114 * The page straddles i_size. It must be zeroed out on each
1115 * and every writepage invocation because it may be mmapped.
1116 * "A file is mapped in multiples of the page size. For a file
1117 * that is not a multiple of the page size, the remaining
1118 * memory is zeroed when mapped, and writes to that region are
1119 * not written out to the file."
1121 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1123 /* Adjust the end_offset to the end of file */
1124 end_offset = offset;
1127 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1130 redirty_page_for_writepage(wbc, page);
1138 struct writeback_control *wbc)
1140 struct xfs_writepage_ctx wpc = {
1141 .io_type = XFS_IO_INVALID,
1145 ret = xfs_do_writepage(page, wbc, &wpc);
1147 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1153 struct address_space *mapping,
1154 struct writeback_control *wbc)
1156 struct xfs_writepage_ctx wpc = {
1157 .io_type = XFS_IO_INVALID,
1161 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1162 if (dax_mapping(mapping))
1163 return dax_writeback_mapping_range(mapping,
1164 xfs_find_bdev_for_inode(mapping->host), wbc);
1166 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1168 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1173 * Called to move a page into cleanable state - and from there
1174 * to be released. The page should already be clean. We always
1175 * have buffer heads in this call.
1177 * Returns 1 if the page is ok to release, 0 otherwise.
1184 int delalloc, unwritten;
1186 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1189 * mm accommodates an old ext3 case where clean pages might not have had
1190 * the dirty bit cleared. Thus, it can send actual dirty pages to
1191 * ->releasepage() via shrink_active_list(). Conversely,
1192 * block_invalidatepage() can send pages that are still marked dirty
1193 * but otherwise have invalidated buffers.
1195 * We want to release the latter to avoid unnecessary buildup of the
1196 * LRU, skip the former and warn if we've left any lingering
1197 * delalloc/unwritten buffers on clean pages. Skip pages with delalloc
1198 * or unwritten buffers and warn if the page is not dirty. Otherwise
1199 * try to release the buffers.
1201 xfs_count_page_state(page, &delalloc, &unwritten);
1204 WARN_ON_ONCE(!PageDirty(page));
1208 WARN_ON_ONCE(!PageDirty(page));
1212 return try_to_free_buffers(page);
1216 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1217 * is, so that we can avoid repeated get_blocks calls.
1219 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1220 * for blocks beyond EOF must be marked new so that sub block regions can be
1221 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1222 * was just allocated or is unwritten, otherwise the callers would overwrite
1223 * existing data with zeros. Hence we have to split the mapping into a range up
1224 * to and including EOF, and a second mapping for beyond EOF.
1228 struct inode *inode,
1230 struct buffer_head *bh_result,
1231 struct xfs_bmbt_irec *imap,
1235 xfs_off_t mapping_size;
1237 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1238 mapping_size <<= inode->i_blkbits;
1240 ASSERT(mapping_size > 0);
1241 if (mapping_size > size)
1242 mapping_size = size;
1243 if (offset < i_size_read(inode) &&
1244 offset + mapping_size >= i_size_read(inode)) {
1245 /* limit mapping to block that spans EOF */
1246 mapping_size = roundup_64(i_size_read(inode) - offset,
1247 i_blocksize(inode));
1249 if (mapping_size > LONG_MAX)
1250 mapping_size = LONG_MAX;
1252 bh_result->b_size = mapping_size;
1257 struct inode *inode,
1259 struct buffer_head *bh_result,
1262 struct xfs_inode *ip = XFS_I(inode);
1263 struct xfs_mount *mp = ip->i_mount;
1264 xfs_fileoff_t offset_fsb, end_fsb;
1267 struct xfs_bmbt_irec imap;
1274 if (XFS_FORCED_SHUTDOWN(mp))
1277 offset = (xfs_off_t)iblock << inode->i_blkbits;
1278 ASSERT(bh_result->b_size >= i_blocksize(inode));
1279 size = bh_result->b_size;
1281 if (offset >= i_size_read(inode))
1285 * Direct I/O is usually done on preallocated files, so try getting
1286 * a block mapping without an exclusive lock first.
1288 lockmode = xfs_ilock_data_map_shared(ip);
1290 ASSERT(offset <= mp->m_super->s_maxbytes);
1291 if (offset + size > mp->m_super->s_maxbytes)
1292 size = mp->m_super->s_maxbytes - offset;
1293 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1294 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1296 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1297 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1302 trace_xfs_get_blocks_found(ip, offset, size,
1303 imap.br_state == XFS_EXT_UNWRITTEN ?
1304 XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap);
1305 xfs_iunlock(ip, lockmode);
1307 trace_xfs_get_blocks_notfound(ip, offset, size);
1311 /* trim mapping down to size requested */
1312 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1315 * For unwritten extents do not report a disk address in the buffered
1316 * read case (treat as if we're reading into a hole).
1318 if (xfs_bmap_is_real_extent(&imap))
1319 xfs_map_buffer(inode, bh_result, &imap, offset);
1322 * If this is a realtime file, data may be on a different device.
1323 * to that pointed to from the buffer_head b_bdev currently.
1325 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1329 xfs_iunlock(ip, lockmode);
1336 struct iov_iter *iter)
1339 * We just need the method present so that open/fcntl allow direct I/O.
1346 struct address_space *mapping,
1349 struct inode *inode = (struct inode *)mapping->host;
1350 struct xfs_inode *ip = XFS_I(inode);
1352 trace_xfs_vm_bmap(XFS_I(inode));
1355 * The swap code (ab-)uses ->bmap to get a block mapping and then
1356 * bypasseѕ the file system for actual I/O. We really can't allow
1357 * that on reflinks inodes, so we have to skip out here. And yes,
1358 * 0 is the magic code for a bmap error.
1360 * Since we don't pass back blockdev info, we can't return bmap
1361 * information for rt files either.
1363 if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip))
1366 filemap_write_and_wait(mapping);
1367 return generic_block_bmap(mapping, block, xfs_get_blocks);
1372 struct file *unused,
1375 trace_xfs_vm_readpage(page->mapping->host, 1);
1376 return mpage_readpage(page, xfs_get_blocks);
1381 struct file *unused,
1382 struct address_space *mapping,
1383 struct list_head *pages,
1386 trace_xfs_vm_readpages(mapping->host, nr_pages);
1387 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1391 * This is basically a copy of __set_page_dirty_buffers() with one
1392 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1393 * dirty, we'll never be able to clean them because we don't write buffers
1394 * beyond EOF, and that means we can't invalidate pages that span EOF
1395 * that have been marked dirty. Further, the dirty state can leak into
1396 * the file interior if the file is extended, resulting in all sorts of
1397 * bad things happening as the state does not match the underlying data.
1399 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1400 * this only exist because of bufferheads and how the generic code manages them.
1403 xfs_vm_set_page_dirty(
1406 struct address_space *mapping = page->mapping;
1407 struct inode *inode = mapping->host;
1412 if (unlikely(!mapping))
1413 return !TestSetPageDirty(page);
1415 end_offset = i_size_read(inode);
1416 offset = page_offset(page);
1418 spin_lock(&mapping->private_lock);
1419 if (page_has_buffers(page)) {
1420 struct buffer_head *head = page_buffers(page);
1421 struct buffer_head *bh = head;
1424 if (offset < end_offset)
1425 set_buffer_dirty(bh);
1426 bh = bh->b_this_page;
1427 offset += i_blocksize(inode);
1428 } while (bh != head);
1431 * Lock out page->mem_cgroup migration to keep PageDirty
1432 * synchronized with per-memcg dirty page counters.
1434 lock_page_memcg(page);
1435 newly_dirty = !TestSetPageDirty(page);
1436 spin_unlock(&mapping->private_lock);
1439 /* sigh - __set_page_dirty() is static, so copy it here, too */
1440 unsigned long flags;
1442 spin_lock_irqsave(&mapping->tree_lock, flags);
1443 if (page->mapping) { /* Race with truncate? */
1444 WARN_ON_ONCE(!PageUptodate(page));
1445 account_page_dirtied(page, mapping);
1446 radix_tree_tag_set(&mapping->page_tree,
1447 page_index(page), PAGECACHE_TAG_DIRTY);
1449 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1451 unlock_page_memcg(page);
1453 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1457 const struct address_space_operations xfs_address_space_operations = {
1458 .readpage = xfs_vm_readpage,
1459 .readpages = xfs_vm_readpages,
1460 .writepage = xfs_vm_writepage,
1461 .writepages = xfs_vm_writepages,
1462 .set_page_dirty = xfs_vm_set_page_dirty,
1463 .releasepage = xfs_vm_releasepage,
1464 .invalidatepage = xfs_vm_invalidatepage,
1465 .bmap = xfs_vm_bmap,
1466 .direct_IO = xfs_vm_direct_IO,
1467 .migratepage = buffer_migrate_page,
1468 .is_partially_uptodate = block_is_partially_uptodate,
1469 .error_remove_page = generic_error_remove_page,