2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs_shared.h"
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_trans_resv.h"
23 #include "xfs_mount.h"
24 #include "xfs_inode.h"
25 #include "xfs_trans.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_alloc.h"
28 #include "xfs_error.h"
29 #include "xfs_iomap.h"
30 #include "xfs_trace.h"
32 #include "xfs_bmap_util.h"
33 #include "xfs_bmap_btree.h"
34 #include <linux/gfp.h>
35 #include <linux/mpage.h>
36 #include <linux/pagevec.h>
37 #include <linux/writeback.h>
39 /* flags for direct write completions */
40 #define XFS_DIO_FLAG_UNWRITTEN (1 << 0)
41 #define XFS_DIO_FLAG_APPEND (1 << 1)
44 * structure owned by writepages passed to individual writepage calls
46 struct xfs_writepage_ctx {
47 struct xfs_bmbt_irec imap;
50 struct xfs_ioend *ioend;
60 struct buffer_head *bh, *head;
62 *delalloc = *unwritten = 0;
64 bh = head = page_buffers(page);
66 if (buffer_unwritten(bh))
68 else if (buffer_delay(bh))
70 } while ((bh = bh->b_this_page) != head);
74 xfs_find_bdev_for_inode(
77 struct xfs_inode *ip = XFS_I(inode);
78 struct xfs_mount *mp = ip->i_mount;
80 if (XFS_IS_REALTIME_INODE(ip))
81 return mp->m_rtdev_targp->bt_bdev;
83 return mp->m_ddev_targp->bt_bdev;
87 * We're now finished for good with this page. Update the page state via the
88 * associated buffer_heads, paying attention to the start and end offsets that
89 * we need to process on the page.
92 xfs_finish_page_writeback(
97 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
98 struct buffer_head *head, *bh;
101 ASSERT(bvec->bv_offset < PAGE_SIZE);
102 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
103 ASSERT(end < PAGE_SIZE);
104 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
106 bh = head = page_buffers(bvec->bv_page);
109 if (off < bvec->bv_offset)
113 bh->b_end_io(bh, !error);
116 } while ((bh = bh->b_this_page) != head);
120 * We're now finished for good with this ioend structure. Update the page
121 * state, release holds on bios, and finally free up memory. Do not use the
126 struct xfs_ioend *ioend,
129 struct inode *inode = ioend->io_inode;
130 struct bio *last = ioend->io_bio;
131 struct bio *bio, *next;
133 for (bio = &ioend->io_inline_bio; bio; bio = next) {
134 struct bio_vec *bvec;
138 * For the last bio, bi_private points to the ioend, so we
139 * need to explicitly end the iteration here.
144 next = bio->bi_private;
146 /* walk each page on bio, ending page IO on them */
147 bio_for_each_segment_all(bvec, bio, i)
148 xfs_finish_page_writeback(inode, bvec, error);
155 * Fast and loose check if this write could update the on-disk inode size.
157 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
159 return ioend->io_offset + ioend->io_size >
160 XFS_I(ioend->io_inode)->i_d.di_size;
164 xfs_setfilesize_trans_alloc(
165 struct xfs_ioend *ioend)
167 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
168 struct xfs_trans *tp;
171 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
175 ioend->io_append_trans = tp;
178 * We may pass freeze protection with a transaction. So tell lockdep
181 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
183 * We hand off the transaction to the completion thread now, so
184 * clear the flag here.
186 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
191 * Update on-disk file size now that data has been written to disk.
195 struct xfs_inode *ip,
196 struct xfs_trans *tp,
202 xfs_ilock(ip, XFS_ILOCK_EXCL);
203 isize = xfs_new_eof(ip, offset + size);
205 xfs_iunlock(ip, XFS_ILOCK_EXCL);
206 xfs_trans_cancel(tp);
210 trace_xfs_setfilesize(ip, offset, size);
212 ip->i_d.di_size = isize;
213 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
214 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
216 return xfs_trans_commit(tp);
220 xfs_setfilesize_ioend(
221 struct xfs_ioend *ioend,
224 struct xfs_inode *ip = XFS_I(ioend->io_inode);
225 struct xfs_trans *tp = ioend->io_append_trans;
228 * The transaction may have been allocated in the I/O submission thread,
229 * thus we need to mark ourselves as being in a transaction manually.
230 * Similarly for freeze protection.
232 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
233 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
235 /* we abort the update if there was an IO error */
237 xfs_trans_cancel(tp);
241 return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
245 * IO write completion.
249 struct work_struct *work)
251 struct xfs_ioend *ioend =
252 container_of(work, struct xfs_ioend, io_work);
253 struct xfs_inode *ip = XFS_I(ioend->io_inode);
254 int error = ioend->io_bio->bi_error;
257 * Set an error if the mount has shut down and proceed with end I/O
258 * processing so it can perform whatever cleanups are necessary.
260 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
264 * For unwritten extents we need to issue transactions to convert a
265 * range to normal written extens after the data I/O has finished.
266 * Detecting and handling completion IO errors is done individually
267 * for each case as different cleanup operations need to be performed
270 if (ioend->io_type == XFS_IO_UNWRITTEN) {
273 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
275 } else if (ioend->io_append_trans) {
276 error = xfs_setfilesize_ioend(ioend, error);
278 ASSERT(!xfs_ioend_is_append(ioend));
282 xfs_destroy_ioend(ioend, error);
289 struct xfs_ioend *ioend = bio->bi_private;
290 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
292 if (ioend->io_type == XFS_IO_UNWRITTEN)
293 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
294 else if (ioend->io_append_trans)
295 queue_work(mp->m_data_workqueue, &ioend->io_work);
297 xfs_destroy_ioend(ioend, bio->bi_error);
304 struct xfs_bmbt_irec *imap,
307 struct xfs_inode *ip = XFS_I(inode);
308 struct xfs_mount *mp = ip->i_mount;
309 ssize_t count = 1 << inode->i_blkbits;
310 xfs_fileoff_t offset_fsb, end_fsb;
312 int bmapi_flags = XFS_BMAPI_ENTIRE;
315 if (XFS_FORCED_SHUTDOWN(mp))
318 if (type == XFS_IO_UNWRITTEN)
319 bmapi_flags |= XFS_BMAPI_IGSTATE;
321 xfs_ilock(ip, XFS_ILOCK_SHARED);
322 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
323 (ip->i_df.if_flags & XFS_IFEXTENTS));
324 ASSERT(offset <= mp->m_super->s_maxbytes);
326 if (offset + count > mp->m_super->s_maxbytes)
327 count = mp->m_super->s_maxbytes - offset;
328 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
329 offset_fsb = XFS_B_TO_FSBT(mp, offset);
330 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
331 imap, &nimaps, bmapi_flags);
332 xfs_iunlock(ip, XFS_ILOCK_SHARED);
337 if (type == XFS_IO_DELALLOC &&
338 (!nimaps || isnullstartblock(imap->br_startblock))) {
339 error = xfs_iomap_write_allocate(ip, offset, imap);
341 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
346 if (type == XFS_IO_UNWRITTEN) {
348 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
349 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
353 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
360 struct xfs_bmbt_irec *imap,
363 offset >>= inode->i_blkbits;
365 return offset >= imap->br_startoff &&
366 offset < imap->br_startoff + imap->br_blockcount;
370 xfs_start_buffer_writeback(
371 struct buffer_head *bh)
373 ASSERT(buffer_mapped(bh));
374 ASSERT(buffer_locked(bh));
375 ASSERT(!buffer_delay(bh));
376 ASSERT(!buffer_unwritten(bh));
378 mark_buffer_async_write(bh);
379 set_buffer_uptodate(bh);
380 clear_buffer_dirty(bh);
384 xfs_start_page_writeback(
388 ASSERT(PageLocked(page));
389 ASSERT(!PageWriteback(page));
392 * if the page was not fully cleaned, we need to ensure that the higher
393 * layers come back to it correctly. That means we need to keep the page
394 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
395 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
396 * write this page in this writeback sweep will be made.
399 clear_page_dirty_for_io(page);
400 set_page_writeback(page);
402 set_page_writeback_keepwrite(page);
407 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
409 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
413 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
414 * it, and we submit that bio. The ioend may be used for multiple bio
415 * submissions, so we only want to allocate an append transaction for the ioend
416 * once. In the case of multiple bio submission, each bio will take an IO
417 * reference to the ioend to ensure that the ioend completion is only done once
418 * all bios have been submitted and the ioend is really done.
420 * If @fail is non-zero, it means that we have a situation where some part of
421 * the submission process has failed after we have marked paged for writeback
422 * and unlocked them. In this situation, we need to fail the bio and ioend
423 * rather than submit it to IO. This typically only happens on a filesystem
428 struct writeback_control *wbc,
429 struct xfs_ioend *ioend,
432 /* Reserve log space if we might write beyond the on-disk inode size. */
434 ioend->io_type != XFS_IO_UNWRITTEN &&
435 xfs_ioend_is_append(ioend) &&
436 !ioend->io_append_trans)
437 status = xfs_setfilesize_trans_alloc(ioend);
439 ioend->io_bio->bi_private = ioend;
440 ioend->io_bio->bi_end_io = xfs_end_bio;
443 * If we are failing the IO now, just mark the ioend with an
444 * error and finish it. This will run IO completion immediately
445 * as there is only one reference to the ioend at this point in
449 ioend->io_bio->bi_error = status;
450 bio_endio(ioend->io_bio);
454 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE,
460 xfs_init_bio_from_bh(
462 struct buffer_head *bh)
464 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
465 bio->bi_bdev = bh->b_bdev;
468 static struct xfs_ioend *
473 struct buffer_head *bh)
475 struct xfs_ioend *ioend;
478 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
479 xfs_init_bio_from_bh(bio, bh);
481 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
482 INIT_LIST_HEAD(&ioend->io_list);
483 ioend->io_type = type;
484 ioend->io_inode = inode;
486 ioend->io_offset = offset;
487 INIT_WORK(&ioend->io_work, xfs_end_io);
488 ioend->io_append_trans = NULL;
494 * Allocate a new bio, and chain the old bio to the new one.
496 * Note that we have to do perform the chaining in this unintuitive order
497 * so that the bi_private linkage is set up in the right direction for the
498 * traversal in xfs_destroy_ioend().
502 struct xfs_ioend *ioend,
503 struct writeback_control *wbc,
504 struct buffer_head *bh)
508 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
509 xfs_init_bio_from_bh(new, bh);
511 bio_chain(ioend->io_bio, new);
512 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
513 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE,
519 * Test to see if we've been building up a completion structure for
520 * earlier buffers -- if so, we try to append to this ioend if we
521 * can, otherwise we finish off any current ioend and start another.
522 * Return the ioend we finished off so that the caller can submit it
523 * once it has finished processing the dirty page.
528 struct buffer_head *bh,
530 struct xfs_writepage_ctx *wpc,
531 struct writeback_control *wbc,
532 struct list_head *iolist)
534 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
535 bh->b_blocknr != wpc->last_block + 1 ||
536 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
538 list_add(&wpc->ioend->io_list, iolist);
539 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
543 * If the buffer doesn't fit into the bio we need to allocate a new
544 * one. This shouldn't happen more than once for a given buffer.
546 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
547 xfs_chain_bio(wpc->ioend, wbc, bh);
549 wpc->ioend->io_size += bh->b_size;
550 wpc->last_block = bh->b_blocknr;
551 xfs_start_buffer_writeback(bh);
557 struct buffer_head *bh,
558 struct xfs_bmbt_irec *imap,
562 struct xfs_mount *m = XFS_I(inode)->i_mount;
563 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
564 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
566 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
567 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
569 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
570 ((offset - iomap_offset) >> inode->i_blkbits);
572 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
575 set_buffer_mapped(bh);
581 struct buffer_head *bh,
582 struct xfs_bmbt_irec *imap,
585 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
586 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
588 xfs_map_buffer(inode, bh, imap, offset);
589 set_buffer_mapped(bh);
590 clear_buffer_delay(bh);
591 clear_buffer_unwritten(bh);
595 * Test if a given page contains at least one buffer of a given @type.
596 * If @check_all_buffers is true, then we walk all the buffers in the page to
597 * try to find one of the type passed in. If it is not set, then the caller only
598 * needs to check the first buffer on the page for a match.
604 bool check_all_buffers)
606 struct buffer_head *bh;
607 struct buffer_head *head;
609 if (PageWriteback(page))
613 if (!page_has_buffers(page))
616 bh = head = page_buffers(page);
618 if (buffer_unwritten(bh)) {
619 if (type == XFS_IO_UNWRITTEN)
621 } else if (buffer_delay(bh)) {
622 if (type == XFS_IO_DELALLOC)
624 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
625 if (type == XFS_IO_OVERWRITE)
629 /* If we are only checking the first buffer, we are done now. */
630 if (!check_all_buffers)
632 } while ((bh = bh->b_this_page) != head);
638 xfs_vm_invalidatepage(
643 trace_xfs_invalidatepage(page->mapping->host, page, offset,
645 block_invalidatepage(page, offset, length);
649 * If the page has delalloc buffers on it, we need to punch them out before we
650 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
651 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
652 * is done on that same region - the delalloc extent is returned when none is
653 * supposed to be there.
655 * We prevent this by truncating away the delalloc regions on the page before
656 * invalidating it. Because they are delalloc, we can do this without needing a
657 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
658 * truncation without a transaction as there is no space left for block
659 * reservation (typically why we see a ENOSPC in writeback).
661 * This is not a performance critical path, so for now just do the punching a
662 * buffer head at a time.
665 xfs_aops_discard_page(
668 struct inode *inode = page->mapping->host;
669 struct xfs_inode *ip = XFS_I(inode);
670 struct buffer_head *bh, *head;
671 loff_t offset = page_offset(page);
673 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
676 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
679 xfs_alert(ip->i_mount,
680 "page discard on page %p, inode 0x%llx, offset %llu.",
681 page, ip->i_ino, offset);
683 xfs_ilock(ip, XFS_ILOCK_EXCL);
684 bh = head = page_buffers(page);
687 xfs_fileoff_t start_fsb;
689 if (!buffer_delay(bh))
692 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
693 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
695 /* something screwed, just bail */
696 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
697 xfs_alert(ip->i_mount,
698 "page discard unable to remove delalloc mapping.");
703 offset += 1 << inode->i_blkbits;
705 } while ((bh = bh->b_this_page) != head);
707 xfs_iunlock(ip, XFS_ILOCK_EXCL);
709 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
714 * We implement an immediate ioend submission policy here to avoid needing to
715 * chain multiple ioends and hence nest mempool allocations which can violate
716 * forward progress guarantees we need to provide. The current ioend we are
717 * adding buffers to is cached on the writepage context, and if the new buffer
718 * does not append to the cached ioend it will create a new ioend and cache that
721 * If a new ioend is created and cached, the old ioend is returned and queued
722 * locally for submission once the entire page is processed or an error has been
723 * detected. While ioends are submitted immediately after they are completed,
724 * batching optimisations are provided by higher level block plugging.
726 * At the end of a writeback pass, there will be a cached ioend remaining on the
727 * writepage context that the caller will need to submit.
731 struct xfs_writepage_ctx *wpc,
732 struct writeback_control *wbc,
736 __uint64_t end_offset)
738 LIST_HEAD(submit_list);
739 struct xfs_ioend *ioend, *next;
740 struct buffer_head *bh, *head;
741 ssize_t len = 1 << inode->i_blkbits;
746 bh = head = page_buffers(page);
747 offset = page_offset(page);
749 if (offset >= end_offset)
751 if (!buffer_uptodate(bh))
755 * set_page_dirty dirties all buffers in a page, independent
756 * of their state. The dirty state however is entirely
757 * meaningless for holes (!mapped && uptodate), so skip
758 * buffers covering holes here.
760 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
761 wpc->imap_valid = false;
765 if (buffer_unwritten(bh)) {
766 if (wpc->io_type != XFS_IO_UNWRITTEN) {
767 wpc->io_type = XFS_IO_UNWRITTEN;
768 wpc->imap_valid = false;
770 } else if (buffer_delay(bh)) {
771 if (wpc->io_type != XFS_IO_DELALLOC) {
772 wpc->io_type = XFS_IO_DELALLOC;
773 wpc->imap_valid = false;
775 } else if (buffer_uptodate(bh)) {
776 if (wpc->io_type != XFS_IO_OVERWRITE) {
777 wpc->io_type = XFS_IO_OVERWRITE;
778 wpc->imap_valid = false;
781 if (PageUptodate(page))
782 ASSERT(buffer_mapped(bh));
784 * This buffer is not uptodate and will not be
785 * written to disk. Ensure that we will put any
786 * subsequent writeable buffers into a new
789 wpc->imap_valid = false;
794 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
796 if (!wpc->imap_valid) {
797 error = xfs_map_blocks(inode, offset, &wpc->imap,
801 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
804 if (wpc->imap_valid) {
806 if (wpc->io_type != XFS_IO_OVERWRITE)
807 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
808 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
812 } while (offset += len, ((bh = bh->b_this_page) != head));
814 if (uptodate && bh == head)
815 SetPageUptodate(page);
817 ASSERT(wpc->ioend || list_empty(&submit_list));
821 * On error, we have to fail the ioend here because we have locked
822 * buffers in the ioend. If we don't do this, we'll deadlock
823 * invalidating the page as that tries to lock the buffers on the page.
824 * Also, because we may have set pages under writeback, we have to make
825 * sure we run IO completion to mark the error state of the IO
826 * appropriately, so we can't cancel the ioend directly here. That means
827 * we have to mark this page as under writeback if we included any
828 * buffers from it in the ioend chain so that completion treats it
831 * If we didn't include the page in the ioend, the on error we can
832 * simply discard and unlock it as there are no other users of the page
833 * or it's buffers right now. The caller will still need to trigger
834 * submission of outstanding ioends on the writepage context so they are
835 * treated correctly on error.
838 xfs_start_page_writeback(page, !error);
841 * Preserve the original error if there was one, otherwise catch
842 * submission errors here and propagate into subsequent ioend
845 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
848 list_del_init(&ioend->io_list);
849 error2 = xfs_submit_ioend(wbc, ioend, error);
850 if (error2 && !error)
854 xfs_aops_discard_page(page);
855 ClearPageUptodate(page);
859 * We can end up here with no error and nothing to write if we
860 * race with a partial page truncate on a sub-page block sized
861 * filesystem. In that case we need to mark the page clean.
863 xfs_start_page_writeback(page, 1);
864 end_page_writeback(page);
867 mapping_set_error(page->mapping, error);
872 * Write out a dirty page.
874 * For delalloc space on the page we need to allocate space and flush it.
875 * For unwritten space on the page we need to start the conversion to
876 * regular allocated space.
877 * For any other dirty buffer heads on the page we should flush them.
882 struct writeback_control *wbc,
885 struct xfs_writepage_ctx *wpc = data;
886 struct inode *inode = page->mapping->host;
888 __uint64_t end_offset;
891 trace_xfs_writepage(inode, page, 0, 0);
893 ASSERT(page_has_buffers(page));
896 * Refuse to write the page out if we are called from reclaim context.
898 * This avoids stack overflows when called from deeply used stacks in
899 * random callers for direct reclaim or memcg reclaim. We explicitly
900 * allow reclaim from kswapd as the stack usage there is relatively low.
902 * This should never happen except in the case of a VM regression so
905 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
910 * Given that we do not allow direct reclaim to call us, we should
911 * never be called while in a filesystem transaction.
913 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
917 * Is this page beyond the end of the file?
919 * The page index is less than the end_index, adjust the end_offset
920 * to the highest offset that this page should represent.
921 * -----------------------------------------------------
922 * | file mapping | <EOF> |
923 * -----------------------------------------------------
924 * | Page ... | Page N-2 | Page N-1 | Page N | |
925 * ^--------------------------------^----------|--------
926 * | desired writeback range | see else |
927 * ---------------------------------^------------------|
929 offset = i_size_read(inode);
930 end_index = offset >> PAGE_SHIFT;
931 if (page->index < end_index)
932 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
935 * Check whether the page to write out is beyond or straddles
937 * -------------------------------------------------------
938 * | file mapping | <EOF> |
939 * -------------------------------------------------------
940 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
941 * ^--------------------------------^-----------|---------
943 * ---------------------------------^-----------|--------|
945 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
948 * Skip the page if it is fully outside i_size, e.g. due to a
949 * truncate operation that is in progress. We must redirty the
950 * page so that reclaim stops reclaiming it. Otherwise
951 * xfs_vm_releasepage() is called on it and gets confused.
953 * Note that the end_index is unsigned long, it would overflow
954 * if the given offset is greater than 16TB on 32-bit system
955 * and if we do check the page is fully outside i_size or not
956 * via "if (page->index >= end_index + 1)" as "end_index + 1"
957 * will be evaluated to 0. Hence this page will be redirtied
958 * and be written out repeatedly which would result in an
959 * infinite loop, the user program that perform this operation
960 * will hang. Instead, we can verify this situation by checking
961 * if the page to write is totally beyond the i_size or if it's
962 * offset is just equal to the EOF.
964 if (page->index > end_index ||
965 (page->index == end_index && offset_into_page == 0))
969 * The page straddles i_size. It must be zeroed out on each
970 * and every writepage invocation because it may be mmapped.
971 * "A file is mapped in multiples of the page size. For a file
972 * that is not a multiple of the page size, the remaining
973 * memory is zeroed when mapped, and writes to that region are
974 * not written out to the file."
976 zero_user_segment(page, offset_into_page, PAGE_SIZE);
978 /* Adjust the end_offset to the end of file */
982 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
985 redirty_page_for_writepage(wbc, page);
993 struct writeback_control *wbc)
995 struct xfs_writepage_ctx wpc = {
996 .io_type = XFS_IO_INVALID,
1000 ret = xfs_do_writepage(page, wbc, &wpc);
1002 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1008 struct address_space *mapping,
1009 struct writeback_control *wbc)
1011 struct xfs_writepage_ctx wpc = {
1012 .io_type = XFS_IO_INVALID,
1016 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1017 if (dax_mapping(mapping))
1018 return dax_writeback_mapping_range(mapping,
1019 xfs_find_bdev_for_inode(mapping->host), wbc);
1021 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1023 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1028 * Called to move a page into cleanable state - and from there
1029 * to be released. The page should already be clean. We always
1030 * have buffer heads in this call.
1032 * Returns 1 if the page is ok to release, 0 otherwise.
1039 int delalloc, unwritten;
1041 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1043 xfs_count_page_state(page, &delalloc, &unwritten);
1045 if (WARN_ON_ONCE(delalloc))
1047 if (WARN_ON_ONCE(unwritten))
1050 return try_to_free_buffers(page);
1054 * When we map a DIO buffer, we may need to pass flags to
1055 * xfs_end_io_direct_write to tell it what kind of write IO we are doing.
1057 * Note that for DIO, an IO to the highest supported file block offset (i.e.
1058 * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64
1059 * bit variable. Hence if we see this overflow, we have to assume that the IO is
1060 * extending the file size. We won't know for sure until IO completion is run
1061 * and the actual max write offset is communicated to the IO completion
1066 struct inode *inode,
1067 struct buffer_head *bh_result,
1068 struct xfs_bmbt_irec *imap,
1071 uintptr_t *flags = (uintptr_t *)&bh_result->b_private;
1072 xfs_off_t size = bh_result->b_size;
1074 trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
1075 ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap);
1077 if (ISUNWRITTEN(imap)) {
1078 *flags |= XFS_DIO_FLAG_UNWRITTEN;
1079 set_buffer_defer_completion(bh_result);
1080 } else if (offset + size > i_size_read(inode) || offset + size < 0) {
1081 *flags |= XFS_DIO_FLAG_APPEND;
1082 set_buffer_defer_completion(bh_result);
1087 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1088 * is, so that we can avoid repeated get_blocks calls.
1090 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1091 * for blocks beyond EOF must be marked new so that sub block regions can be
1092 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1093 * was just allocated or is unwritten, otherwise the callers would overwrite
1094 * existing data with zeros. Hence we have to split the mapping into a range up
1095 * to and including EOF, and a second mapping for beyond EOF.
1099 struct inode *inode,
1101 struct buffer_head *bh_result,
1102 struct xfs_bmbt_irec *imap,
1106 xfs_off_t mapping_size;
1108 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1109 mapping_size <<= inode->i_blkbits;
1111 ASSERT(mapping_size > 0);
1112 if (mapping_size > size)
1113 mapping_size = size;
1114 if (offset < i_size_read(inode) &&
1115 offset + mapping_size >= i_size_read(inode)) {
1116 /* limit mapping to block that spans EOF */
1117 mapping_size = roundup_64(i_size_read(inode) - offset,
1118 1 << inode->i_blkbits);
1120 if (mapping_size > LONG_MAX)
1121 mapping_size = LONG_MAX;
1123 bh_result->b_size = mapping_size;
1128 struct inode *inode,
1130 struct buffer_head *bh_result,
1135 struct xfs_inode *ip = XFS_I(inode);
1136 struct xfs_mount *mp = ip->i_mount;
1137 xfs_fileoff_t offset_fsb, end_fsb;
1140 struct xfs_bmbt_irec imap;
1146 BUG_ON(create && !direct);
1148 if (XFS_FORCED_SHUTDOWN(mp))
1151 offset = (xfs_off_t)iblock << inode->i_blkbits;
1152 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1153 size = bh_result->b_size;
1155 if (!create && offset >= i_size_read(inode))
1159 * Direct I/O is usually done on preallocated files, so try getting
1160 * a block mapping without an exclusive lock first.
1162 lockmode = xfs_ilock_data_map_shared(ip);
1164 ASSERT(offset <= mp->m_super->s_maxbytes);
1165 if (offset + size > mp->m_super->s_maxbytes)
1166 size = mp->m_super->s_maxbytes - offset;
1167 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1168 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1170 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1171 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1175 /* for DAX, we convert unwritten extents directly */
1178 (imap.br_startblock == HOLESTARTBLOCK ||
1179 imap.br_startblock == DELAYSTARTBLOCK) ||
1180 (IS_DAX(inode) && ISUNWRITTEN(&imap)))) {
1182 * xfs_iomap_write_direct() expects the shared lock. It
1183 * is unlocked on return.
1185 if (lockmode == XFS_ILOCK_EXCL)
1186 xfs_ilock_demote(ip, lockmode);
1188 error = xfs_iomap_write_direct(ip, offset, size,
1194 trace_xfs_get_blocks_alloc(ip, offset, size,
1195 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1196 : XFS_IO_DELALLOC, &imap);
1197 } else if (nimaps) {
1198 trace_xfs_get_blocks_found(ip, offset, size,
1199 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1200 : XFS_IO_OVERWRITE, &imap);
1201 xfs_iunlock(ip, lockmode);
1203 trace_xfs_get_blocks_notfound(ip, offset, size);
1207 if (IS_DAX(inode) && create) {
1208 ASSERT(!ISUNWRITTEN(&imap));
1209 /* zeroing is not needed at a higher layer */
1213 /* trim mapping down to size requested */
1214 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1217 * For unwritten extents do not report a disk address in the buffered
1218 * read case (treat as if we're reading into a hole).
1220 if (imap.br_startblock != HOLESTARTBLOCK &&
1221 imap.br_startblock != DELAYSTARTBLOCK &&
1222 (create || !ISUNWRITTEN(&imap))) {
1223 xfs_map_buffer(inode, bh_result, &imap, offset);
1224 if (ISUNWRITTEN(&imap))
1225 set_buffer_unwritten(bh_result);
1226 /* direct IO needs special help */
1229 ASSERT(!ISUNWRITTEN(&imap));
1231 xfs_map_direct(inode, bh_result, &imap, offset);
1236 * If this is a realtime file, data may be on a different device.
1237 * to that pointed to from the buffer_head b_bdev currently.
1239 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1242 * If we previously allocated a block out beyond eof and we are now
1243 * coming back to use it then we will need to flag it as new even if it
1244 * has a disk address.
1246 * With sub-block writes into unwritten extents we also need to mark
1247 * the buffer as new so that the unwritten parts of the buffer gets
1251 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1252 (offset >= i_size_read(inode)) ||
1253 (new || ISUNWRITTEN(&imap))))
1254 set_buffer_new(bh_result);
1256 BUG_ON(direct && imap.br_startblock == DELAYSTARTBLOCK);
1261 xfs_iunlock(ip, lockmode);
1267 struct inode *inode,
1269 struct buffer_head *bh_result,
1272 return __xfs_get_blocks(inode, iblock, bh_result, create, false, false);
1276 xfs_get_blocks_direct(
1277 struct inode *inode,
1279 struct buffer_head *bh_result,
1282 return __xfs_get_blocks(inode, iblock, bh_result, create, true, false);
1286 xfs_get_blocks_dax_fault(
1287 struct inode *inode,
1289 struct buffer_head *bh_result,
1292 return __xfs_get_blocks(inode, iblock, bh_result, create, true, true);
1296 * Complete a direct I/O write request.
1298 * xfs_map_direct passes us some flags in the private data to tell us what to
1299 * do. If no flags are set, then the write IO is an overwrite wholly within
1300 * the existing allocated file size and so there is nothing for us to do.
1302 * Note that in this case the completion can be called in interrupt context,
1303 * whereas if we have flags set we will always be called in task context
1304 * (i.e. from a workqueue).
1307 xfs_end_io_direct_write(
1313 struct inode *inode = file_inode(iocb->ki_filp);
1314 struct xfs_inode *ip = XFS_I(inode);
1315 struct xfs_mount *mp = ip->i_mount;
1316 uintptr_t flags = (uintptr_t)private;
1319 trace_xfs_end_io_direct_write(ip, offset, size);
1321 if (XFS_FORCED_SHUTDOWN(mp))
1328 * The flags tell us whether we are doing unwritten extent conversions
1329 * or an append transaction that updates the on-disk file size. These
1330 * cases are the only cases where we should *potentially* be needing
1331 * to update the VFS inode size.
1334 ASSERT(offset + size <= i_size_read(inode));
1339 * We need to update the in-core inode size here so that we don't end up
1340 * with the on-disk inode size being outside the in-core inode size. We
1341 * have no other method of updating EOF for AIO, so always do it here
1344 * We need to lock the test/set EOF update as we can be racing with
1345 * other IO completions here to update the EOF. Failing to serialise
1346 * here can result in EOF moving backwards and Bad Things Happen when
1349 spin_lock(&ip->i_flags_lock);
1350 if (offset + size > i_size_read(inode))
1351 i_size_write(inode, offset + size);
1352 spin_unlock(&ip->i_flags_lock);
1354 if (flags & XFS_DIO_FLAG_UNWRITTEN) {
1355 trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
1357 error = xfs_iomap_write_unwritten(ip, offset, size);
1358 } else if (flags & XFS_DIO_FLAG_APPEND) {
1359 struct xfs_trans *tp;
1361 trace_xfs_end_io_direct_write_append(ip, offset, size);
1363 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0,
1366 error = xfs_setfilesize(ip, tp, offset, size);
1375 struct iov_iter *iter)
1378 * We just need the method present so that open/fcntl allow direct I/O.
1385 struct address_space *mapping,
1388 struct inode *inode = (struct inode *)mapping->host;
1389 struct xfs_inode *ip = XFS_I(inode);
1391 trace_xfs_vm_bmap(XFS_I(inode));
1392 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1393 filemap_write_and_wait(mapping);
1394 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1395 return generic_block_bmap(mapping, block, xfs_get_blocks);
1400 struct file *unused,
1403 trace_xfs_vm_readpage(page->mapping->host, 1);
1404 return mpage_readpage(page, xfs_get_blocks);
1409 struct file *unused,
1410 struct address_space *mapping,
1411 struct list_head *pages,
1414 trace_xfs_vm_readpages(mapping->host, nr_pages);
1415 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1419 * This is basically a copy of __set_page_dirty_buffers() with one
1420 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1421 * dirty, we'll never be able to clean them because we don't write buffers
1422 * beyond EOF, and that means we can't invalidate pages that span EOF
1423 * that have been marked dirty. Further, the dirty state can leak into
1424 * the file interior if the file is extended, resulting in all sorts of
1425 * bad things happening as the state does not match the underlying data.
1427 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1428 * this only exist because of bufferheads and how the generic code manages them.
1431 xfs_vm_set_page_dirty(
1434 struct address_space *mapping = page->mapping;
1435 struct inode *inode = mapping->host;
1440 if (unlikely(!mapping))
1441 return !TestSetPageDirty(page);
1443 end_offset = i_size_read(inode);
1444 offset = page_offset(page);
1446 spin_lock(&mapping->private_lock);
1447 if (page_has_buffers(page)) {
1448 struct buffer_head *head = page_buffers(page);
1449 struct buffer_head *bh = head;
1452 if (offset < end_offset)
1453 set_buffer_dirty(bh);
1454 bh = bh->b_this_page;
1455 offset += 1 << inode->i_blkbits;
1456 } while (bh != head);
1459 * Lock out page->mem_cgroup migration to keep PageDirty
1460 * synchronized with per-memcg dirty page counters.
1462 lock_page_memcg(page);
1463 newly_dirty = !TestSetPageDirty(page);
1464 spin_unlock(&mapping->private_lock);
1467 /* sigh - __set_page_dirty() is static, so copy it here, too */
1468 unsigned long flags;
1470 spin_lock_irqsave(&mapping->tree_lock, flags);
1471 if (page->mapping) { /* Race with truncate? */
1472 WARN_ON_ONCE(!PageUptodate(page));
1473 account_page_dirtied(page, mapping);
1474 radix_tree_tag_set(&mapping->page_tree,
1475 page_index(page), PAGECACHE_TAG_DIRTY);
1477 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1479 unlock_page_memcg(page);
1481 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1485 const struct address_space_operations xfs_address_space_operations = {
1486 .readpage = xfs_vm_readpage,
1487 .readpages = xfs_vm_readpages,
1488 .writepage = xfs_vm_writepage,
1489 .writepages = xfs_vm_writepages,
1490 .set_page_dirty = xfs_vm_set_page_dirty,
1491 .releasepage = xfs_vm_releasepage,
1492 .invalidatepage = xfs_vm_invalidatepage,
1493 .bmap = xfs_vm_bmap,
1494 .direct_IO = xfs_vm_direct_IO,
1495 .migratepage = buffer_migrate_page,
1496 .is_partially_uptodate = block_is_partially_uptodate,
1497 .error_remove_page = generic_error_remove_page,