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.
91 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
92 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
93 * the page at all, as we may be racing with memory reclaim and it can free both
94 * the bufferhead chain and the page as it will see the page as clean and
98 xfs_finish_page_writeback(
100 struct bio_vec *bvec,
103 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
104 struct buffer_head *head, *bh, *next;
105 unsigned int off = 0;
108 ASSERT(bvec->bv_offset < PAGE_SIZE);
109 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
110 ASSERT(end < PAGE_SIZE);
111 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
113 bh = head = page_buffers(bvec->bv_page);
117 next = bh->b_this_page;
118 if (off < bvec->bv_offset)
122 bh->b_end_io(bh, !error);
125 } while ((bh = next) != head);
129 * We're now finished for good with this ioend structure. Update the page
130 * state, release holds on bios, and finally free up memory. Do not use the
135 struct xfs_ioend *ioend,
138 struct inode *inode = ioend->io_inode;
139 struct bio *last = ioend->io_bio;
140 struct bio *bio, *next;
142 for (bio = &ioend->io_inline_bio; bio; bio = next) {
143 struct bio_vec *bvec;
147 * For the last bio, bi_private points to the ioend, so we
148 * need to explicitly end the iteration here.
153 next = bio->bi_private;
155 /* walk each page on bio, ending page IO on them */
156 bio_for_each_segment_all(bvec, bio, i)
157 xfs_finish_page_writeback(inode, bvec, error);
164 * Fast and loose check if this write could update the on-disk inode size.
166 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
168 return ioend->io_offset + ioend->io_size >
169 XFS_I(ioend->io_inode)->i_d.di_size;
173 xfs_setfilesize_trans_alloc(
174 struct xfs_ioend *ioend)
176 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
177 struct xfs_trans *tp;
180 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
184 ioend->io_append_trans = tp;
187 * We may pass freeze protection with a transaction. So tell lockdep
190 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
192 * We hand off the transaction to the completion thread now, so
193 * clear the flag here.
195 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
200 * Update on-disk file size now that data has been written to disk.
204 struct xfs_inode *ip,
205 struct xfs_trans *tp,
211 xfs_ilock(ip, XFS_ILOCK_EXCL);
212 isize = xfs_new_eof(ip, offset + size);
214 xfs_iunlock(ip, XFS_ILOCK_EXCL);
215 xfs_trans_cancel(tp);
219 trace_xfs_setfilesize(ip, offset, size);
221 ip->i_d.di_size = isize;
222 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
223 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
225 return xfs_trans_commit(tp);
230 struct xfs_inode *ip,
234 struct xfs_mount *mp = ip->i_mount;
235 struct xfs_trans *tp;
238 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
242 return __xfs_setfilesize(ip, tp, offset, size);
246 xfs_setfilesize_ioend(
247 struct xfs_ioend *ioend,
250 struct xfs_inode *ip = XFS_I(ioend->io_inode);
251 struct xfs_trans *tp = ioend->io_append_trans;
254 * The transaction may have been allocated in the I/O submission thread,
255 * thus we need to mark ourselves as being in a transaction manually.
256 * Similarly for freeze protection.
258 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
259 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
261 /* we abort the update if there was an IO error */
263 xfs_trans_cancel(tp);
267 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
271 * IO write completion.
275 struct work_struct *work)
277 struct xfs_ioend *ioend =
278 container_of(work, struct xfs_ioend, io_work);
279 struct xfs_inode *ip = XFS_I(ioend->io_inode);
280 int error = ioend->io_bio->bi_error;
283 * Set an error if the mount has shut down and proceed with end I/O
284 * processing so it can perform whatever cleanups are necessary.
286 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
290 * For unwritten extents we need to issue transactions to convert a
291 * range to normal written extens after the data I/O has finished.
292 * Detecting and handling completion IO errors is done individually
293 * for each case as different cleanup operations need to be performed
296 if (ioend->io_type == XFS_IO_UNWRITTEN) {
299 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
301 } else if (ioend->io_append_trans) {
302 error = xfs_setfilesize_ioend(ioend, error);
304 ASSERT(!xfs_ioend_is_append(ioend));
308 xfs_destroy_ioend(ioend, error);
315 struct xfs_ioend *ioend = bio->bi_private;
316 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
318 if (ioend->io_type == XFS_IO_UNWRITTEN)
319 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
320 else if (ioend->io_append_trans)
321 queue_work(mp->m_data_workqueue, &ioend->io_work);
323 xfs_destroy_ioend(ioend, bio->bi_error);
330 struct xfs_bmbt_irec *imap,
333 struct xfs_inode *ip = XFS_I(inode);
334 struct xfs_mount *mp = ip->i_mount;
335 ssize_t count = 1 << inode->i_blkbits;
336 xfs_fileoff_t offset_fsb, end_fsb;
338 int bmapi_flags = XFS_BMAPI_ENTIRE;
341 if (XFS_FORCED_SHUTDOWN(mp))
344 if (type == XFS_IO_UNWRITTEN)
345 bmapi_flags |= XFS_BMAPI_IGSTATE;
347 xfs_ilock(ip, XFS_ILOCK_SHARED);
348 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
349 (ip->i_df.if_flags & XFS_IFEXTENTS));
350 ASSERT(offset <= mp->m_super->s_maxbytes);
352 if (offset + count > mp->m_super->s_maxbytes)
353 count = mp->m_super->s_maxbytes - offset;
354 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
355 offset_fsb = XFS_B_TO_FSBT(mp, offset);
356 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
357 imap, &nimaps, bmapi_flags);
358 xfs_iunlock(ip, XFS_ILOCK_SHARED);
363 if (type == XFS_IO_DELALLOC &&
364 (!nimaps || isnullstartblock(imap->br_startblock))) {
365 error = xfs_iomap_write_allocate(ip, offset, imap);
367 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
372 if (type == XFS_IO_UNWRITTEN) {
374 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
375 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
379 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
386 struct xfs_bmbt_irec *imap,
389 offset >>= inode->i_blkbits;
391 return offset >= imap->br_startoff &&
392 offset < imap->br_startoff + imap->br_blockcount;
396 xfs_start_buffer_writeback(
397 struct buffer_head *bh)
399 ASSERT(buffer_mapped(bh));
400 ASSERT(buffer_locked(bh));
401 ASSERT(!buffer_delay(bh));
402 ASSERT(!buffer_unwritten(bh));
404 mark_buffer_async_write(bh);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
410 xfs_start_page_writeback(
414 ASSERT(PageLocked(page));
415 ASSERT(!PageWriteback(page));
418 * if the page was not fully cleaned, we need to ensure that the higher
419 * layers come back to it correctly. That means we need to keep the page
420 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
421 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
422 * write this page in this writeback sweep will be made.
425 clear_page_dirty_for_io(page);
426 set_page_writeback(page);
428 set_page_writeback_keepwrite(page);
433 static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
435 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
439 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
440 * it, and we submit that bio. The ioend may be used for multiple bio
441 * submissions, so we only want to allocate an append transaction for the ioend
442 * once. In the case of multiple bio submission, each bio will take an IO
443 * reference to the ioend to ensure that the ioend completion is only done once
444 * all bios have been submitted and the ioend is really done.
446 * If @fail is non-zero, it means that we have a situation where some part of
447 * the submission process has failed after we have marked paged for writeback
448 * and unlocked them. In this situation, we need to fail the bio and ioend
449 * rather than submit it to IO. This typically only happens on a filesystem
454 struct writeback_control *wbc,
455 struct xfs_ioend *ioend,
458 /* Reserve log space if we might write beyond the on-disk inode size. */
460 ioend->io_type != XFS_IO_UNWRITTEN &&
461 xfs_ioend_is_append(ioend) &&
462 !ioend->io_append_trans)
463 status = xfs_setfilesize_trans_alloc(ioend);
465 ioend->io_bio->bi_private = ioend;
466 ioend->io_bio->bi_end_io = xfs_end_bio;
467 bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE,
468 (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0);
470 * If we are failing the IO now, just mark the ioend with an
471 * error and finish it. This will run IO completion immediately
472 * as there is only one reference to the ioend at this point in
476 ioend->io_bio->bi_error = status;
477 bio_endio(ioend->io_bio);
481 submit_bio(ioend->io_bio);
486 xfs_init_bio_from_bh(
488 struct buffer_head *bh)
490 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
491 bio->bi_bdev = bh->b_bdev;
494 static struct xfs_ioend *
499 struct buffer_head *bh)
501 struct xfs_ioend *ioend;
504 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
505 xfs_init_bio_from_bh(bio, bh);
507 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
508 INIT_LIST_HEAD(&ioend->io_list);
509 ioend->io_type = type;
510 ioend->io_inode = inode;
512 ioend->io_offset = offset;
513 INIT_WORK(&ioend->io_work, xfs_end_io);
514 ioend->io_append_trans = NULL;
520 * Allocate a new bio, and chain the old bio to the new one.
522 * Note that we have to do perform the chaining in this unintuitive order
523 * so that the bi_private linkage is set up in the right direction for the
524 * traversal in xfs_destroy_ioend().
528 struct xfs_ioend *ioend,
529 struct writeback_control *wbc,
530 struct buffer_head *bh)
534 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
535 xfs_init_bio_from_bh(new, bh);
537 bio_chain(ioend->io_bio, new);
538 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
539 bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE,
540 (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0);
541 submit_bio(ioend->io_bio);
546 * Test to see if we've been building up a completion structure for
547 * earlier buffers -- if so, we try to append to this ioend if we
548 * can, otherwise we finish off any current ioend and start another.
549 * Return the ioend we finished off so that the caller can submit it
550 * once it has finished processing the dirty page.
555 struct buffer_head *bh,
557 struct xfs_writepage_ctx *wpc,
558 struct writeback_control *wbc,
559 struct list_head *iolist)
561 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
562 bh->b_blocknr != wpc->last_block + 1 ||
563 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
565 list_add(&wpc->ioend->io_list, iolist);
566 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
570 * If the buffer doesn't fit into the bio we need to allocate a new
571 * one. This shouldn't happen more than once for a given buffer.
573 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
574 xfs_chain_bio(wpc->ioend, wbc, bh);
576 wpc->ioend->io_size += bh->b_size;
577 wpc->last_block = bh->b_blocknr;
578 xfs_start_buffer_writeback(bh);
584 struct buffer_head *bh,
585 struct xfs_bmbt_irec *imap,
589 struct xfs_mount *m = XFS_I(inode)->i_mount;
590 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
591 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
593 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
594 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
596 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
597 ((offset - iomap_offset) >> inode->i_blkbits);
599 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
602 set_buffer_mapped(bh);
608 struct buffer_head *bh,
609 struct xfs_bmbt_irec *imap,
612 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
613 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
615 xfs_map_buffer(inode, bh, imap, offset);
616 set_buffer_mapped(bh);
617 clear_buffer_delay(bh);
618 clear_buffer_unwritten(bh);
622 * Test if a given page contains at least one buffer of a given @type.
623 * If @check_all_buffers is true, then we walk all the buffers in the page to
624 * try to find one of the type passed in. If it is not set, then the caller only
625 * needs to check the first buffer on the page for a match.
631 bool check_all_buffers)
633 struct buffer_head *bh;
634 struct buffer_head *head;
636 if (PageWriteback(page))
640 if (!page_has_buffers(page))
643 bh = head = page_buffers(page);
645 if (buffer_unwritten(bh)) {
646 if (type == XFS_IO_UNWRITTEN)
648 } else if (buffer_delay(bh)) {
649 if (type == XFS_IO_DELALLOC)
651 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
652 if (type == XFS_IO_OVERWRITE)
656 /* If we are only checking the first buffer, we are done now. */
657 if (!check_all_buffers)
659 } while ((bh = bh->b_this_page) != head);
665 xfs_vm_invalidatepage(
670 trace_xfs_invalidatepage(page->mapping->host, page, offset,
672 block_invalidatepage(page, offset, length);
676 * If the page has delalloc buffers on it, we need to punch them out before we
677 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
678 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
679 * is done on that same region - the delalloc extent is returned when none is
680 * supposed to be there.
682 * We prevent this by truncating away the delalloc regions on the page before
683 * invalidating it. Because they are delalloc, we can do this without needing a
684 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
685 * truncation without a transaction as there is no space left for block
686 * reservation (typically why we see a ENOSPC in writeback).
688 * This is not a performance critical path, so for now just do the punching a
689 * buffer head at a time.
692 xfs_aops_discard_page(
695 struct inode *inode = page->mapping->host;
696 struct xfs_inode *ip = XFS_I(inode);
697 struct buffer_head *bh, *head;
698 loff_t offset = page_offset(page);
700 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
703 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
706 xfs_alert(ip->i_mount,
707 "page discard on page %p, inode 0x%llx, offset %llu.",
708 page, ip->i_ino, offset);
710 xfs_ilock(ip, XFS_ILOCK_EXCL);
711 bh = head = page_buffers(page);
714 xfs_fileoff_t start_fsb;
716 if (!buffer_delay(bh))
719 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
720 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
722 /* something screwed, just bail */
723 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
724 xfs_alert(ip->i_mount,
725 "page discard unable to remove delalloc mapping.");
730 offset += 1 << inode->i_blkbits;
732 } while ((bh = bh->b_this_page) != head);
734 xfs_iunlock(ip, XFS_ILOCK_EXCL);
736 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
741 * We implement an immediate ioend submission policy here to avoid needing to
742 * chain multiple ioends and hence nest mempool allocations which can violate
743 * forward progress guarantees we need to provide. The current ioend we are
744 * adding buffers to is cached on the writepage context, and if the new buffer
745 * does not append to the cached ioend it will create a new ioend and cache that
748 * If a new ioend is created and cached, the old ioend is returned and queued
749 * locally for submission once the entire page is processed or an error has been
750 * detected. While ioends are submitted immediately after they are completed,
751 * batching optimisations are provided by higher level block plugging.
753 * At the end of a writeback pass, there will be a cached ioend remaining on the
754 * writepage context that the caller will need to submit.
758 struct xfs_writepage_ctx *wpc,
759 struct writeback_control *wbc,
763 __uint64_t end_offset)
765 LIST_HEAD(submit_list);
766 struct xfs_ioend *ioend, *next;
767 struct buffer_head *bh, *head;
768 ssize_t len = 1 << inode->i_blkbits;
773 bh = head = page_buffers(page);
774 offset = page_offset(page);
776 if (offset >= end_offset)
778 if (!buffer_uptodate(bh))
782 * set_page_dirty dirties all buffers in a page, independent
783 * of their state. The dirty state however is entirely
784 * meaningless for holes (!mapped && uptodate), so skip
785 * buffers covering holes here.
787 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
788 wpc->imap_valid = false;
792 if (buffer_unwritten(bh)) {
793 if (wpc->io_type != XFS_IO_UNWRITTEN) {
794 wpc->io_type = XFS_IO_UNWRITTEN;
795 wpc->imap_valid = false;
797 } else if (buffer_delay(bh)) {
798 if (wpc->io_type != XFS_IO_DELALLOC) {
799 wpc->io_type = XFS_IO_DELALLOC;
800 wpc->imap_valid = false;
802 } else if (buffer_uptodate(bh)) {
803 if (wpc->io_type != XFS_IO_OVERWRITE) {
804 wpc->io_type = XFS_IO_OVERWRITE;
805 wpc->imap_valid = false;
808 if (PageUptodate(page))
809 ASSERT(buffer_mapped(bh));
811 * This buffer is not uptodate and will not be
812 * written to disk. Ensure that we will put any
813 * subsequent writeable buffers into a new
816 wpc->imap_valid = false;
821 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
823 if (!wpc->imap_valid) {
824 error = xfs_map_blocks(inode, offset, &wpc->imap,
828 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
831 if (wpc->imap_valid) {
833 if (wpc->io_type != XFS_IO_OVERWRITE)
834 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
835 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
839 } while (offset += len, ((bh = bh->b_this_page) != head));
841 if (uptodate && bh == head)
842 SetPageUptodate(page);
844 ASSERT(wpc->ioend || list_empty(&submit_list));
848 * On error, we have to fail the ioend here because we have locked
849 * buffers in the ioend. If we don't do this, we'll deadlock
850 * invalidating the page as that tries to lock the buffers on the page.
851 * Also, because we may have set pages under writeback, we have to make
852 * sure we run IO completion to mark the error state of the IO
853 * appropriately, so we can't cancel the ioend directly here. That means
854 * we have to mark this page as under writeback if we included any
855 * buffers from it in the ioend chain so that completion treats it
858 * If we didn't include the page in the ioend, the on error we can
859 * simply discard and unlock it as there are no other users of the page
860 * or it's buffers right now. The caller will still need to trigger
861 * submission of outstanding ioends on the writepage context so they are
862 * treated correctly on error.
865 xfs_start_page_writeback(page, !error);
868 * Preserve the original error if there was one, otherwise catch
869 * submission errors here and propagate into subsequent ioend
872 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
875 list_del_init(&ioend->io_list);
876 error2 = xfs_submit_ioend(wbc, ioend, error);
877 if (error2 && !error)
881 xfs_aops_discard_page(page);
882 ClearPageUptodate(page);
886 * We can end up here with no error and nothing to write if we
887 * race with a partial page truncate on a sub-page block sized
888 * filesystem. In that case we need to mark the page clean.
890 xfs_start_page_writeback(page, 1);
891 end_page_writeback(page);
894 mapping_set_error(page->mapping, error);
899 * Write out a dirty page.
901 * For delalloc space on the page we need to allocate space and flush it.
902 * For unwritten space on the page we need to start the conversion to
903 * regular allocated space.
904 * For any other dirty buffer heads on the page we should flush them.
909 struct writeback_control *wbc,
912 struct xfs_writepage_ctx *wpc = data;
913 struct inode *inode = page->mapping->host;
915 __uint64_t end_offset;
918 trace_xfs_writepage(inode, page, 0, 0);
920 ASSERT(page_has_buffers(page));
923 * Refuse to write the page out if we are called from reclaim context.
925 * This avoids stack overflows when called from deeply used stacks in
926 * random callers for direct reclaim or memcg reclaim. We explicitly
927 * allow reclaim from kswapd as the stack usage there is relatively low.
929 * This should never happen except in the case of a VM regression so
932 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
937 * Given that we do not allow direct reclaim to call us, we should
938 * never be called while in a filesystem transaction.
940 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
944 * Is this page beyond the end of the file?
946 * The page index is less than the end_index, adjust the end_offset
947 * to the highest offset that this page should represent.
948 * -----------------------------------------------------
949 * | file mapping | <EOF> |
950 * -----------------------------------------------------
951 * | Page ... | Page N-2 | Page N-1 | Page N | |
952 * ^--------------------------------^----------|--------
953 * | desired writeback range | see else |
954 * ---------------------------------^------------------|
956 offset = i_size_read(inode);
957 end_index = offset >> PAGE_SHIFT;
958 if (page->index < end_index)
959 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
962 * Check whether the page to write out is beyond or straddles
964 * -------------------------------------------------------
965 * | file mapping | <EOF> |
966 * -------------------------------------------------------
967 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
968 * ^--------------------------------^-----------|---------
970 * ---------------------------------^-----------|--------|
972 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
975 * Skip the page if it is fully outside i_size, e.g. due to a
976 * truncate operation that is in progress. We must redirty the
977 * page so that reclaim stops reclaiming it. Otherwise
978 * xfs_vm_releasepage() is called on it and gets confused.
980 * Note that the end_index is unsigned long, it would overflow
981 * if the given offset is greater than 16TB on 32-bit system
982 * and if we do check the page is fully outside i_size or not
983 * via "if (page->index >= end_index + 1)" as "end_index + 1"
984 * will be evaluated to 0. Hence this page will be redirtied
985 * and be written out repeatedly which would result in an
986 * infinite loop, the user program that perform this operation
987 * will hang. Instead, we can verify this situation by checking
988 * if the page to write is totally beyond the i_size or if it's
989 * offset is just equal to the EOF.
991 if (page->index > end_index ||
992 (page->index == end_index && offset_into_page == 0))
996 * The page straddles i_size. It must be zeroed out on each
997 * and every writepage invocation because it may be mmapped.
998 * "A file is mapped in multiples of the page size. For a file
999 * that is not a multiple of the page size, the remaining
1000 * memory is zeroed when mapped, and writes to that region are
1001 * not written out to the file."
1003 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1005 /* Adjust the end_offset to the end of file */
1006 end_offset = offset;
1009 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1012 redirty_page_for_writepage(wbc, page);
1020 struct writeback_control *wbc)
1022 struct xfs_writepage_ctx wpc = {
1023 .io_type = XFS_IO_INVALID,
1027 ret = xfs_do_writepage(page, wbc, &wpc);
1029 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1035 struct address_space *mapping,
1036 struct writeback_control *wbc)
1038 struct xfs_writepage_ctx wpc = {
1039 .io_type = XFS_IO_INVALID,
1043 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1044 if (dax_mapping(mapping))
1045 return dax_writeback_mapping_range(mapping,
1046 xfs_find_bdev_for_inode(mapping->host), wbc);
1048 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1050 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1055 * Called to move a page into cleanable state - and from there
1056 * to be released. The page should already be clean. We always
1057 * have buffer heads in this call.
1059 * Returns 1 if the page is ok to release, 0 otherwise.
1066 int delalloc, unwritten;
1068 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1071 * mm accommodates an old ext3 case where clean pages might not have had
1072 * the dirty bit cleared. Thus, it can send actual dirty pages to
1073 * ->releasepage() via shrink_active_list(). Conversely,
1074 * block_invalidatepage() can send pages that are still marked dirty
1075 * but otherwise have invalidated buffers.
1077 * We've historically freed buffers on the latter. Instead, quietly
1078 * filter out all dirty pages to avoid spurious buffer state warnings.
1079 * This can likely be removed once shrink_active_list() is fixed.
1081 if (PageDirty(page))
1084 xfs_count_page_state(page, &delalloc, &unwritten);
1086 if (WARN_ON_ONCE(delalloc))
1088 if (WARN_ON_ONCE(unwritten))
1091 return try_to_free_buffers(page);
1095 * When we map a DIO buffer, we may need to pass flags to
1096 * xfs_end_io_direct_write to tell it what kind of write IO we are doing.
1098 * Note that for DIO, an IO to the highest supported file block offset (i.e.
1099 * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64
1100 * bit variable. Hence if we see this overflow, we have to assume that the IO is
1101 * extending the file size. We won't know for sure until IO completion is run
1102 * and the actual max write offset is communicated to the IO completion
1107 struct inode *inode,
1108 struct buffer_head *bh_result,
1109 struct xfs_bmbt_irec *imap,
1112 uintptr_t *flags = (uintptr_t *)&bh_result->b_private;
1113 xfs_off_t size = bh_result->b_size;
1115 trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
1116 ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap);
1118 if (ISUNWRITTEN(imap)) {
1119 *flags |= XFS_DIO_FLAG_UNWRITTEN;
1120 set_buffer_defer_completion(bh_result);
1121 } else if (offset + size > i_size_read(inode) || offset + size < 0) {
1122 *flags |= XFS_DIO_FLAG_APPEND;
1123 set_buffer_defer_completion(bh_result);
1128 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1129 * is, so that we can avoid repeated get_blocks calls.
1131 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1132 * for blocks beyond EOF must be marked new so that sub block regions can be
1133 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1134 * was just allocated or is unwritten, otherwise the callers would overwrite
1135 * existing data with zeros. Hence we have to split the mapping into a range up
1136 * to and including EOF, and a second mapping for beyond EOF.
1140 struct inode *inode,
1142 struct buffer_head *bh_result,
1143 struct xfs_bmbt_irec *imap,
1147 xfs_off_t mapping_size;
1149 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1150 mapping_size <<= inode->i_blkbits;
1152 ASSERT(mapping_size > 0);
1153 if (mapping_size > size)
1154 mapping_size = size;
1155 if (offset < i_size_read(inode) &&
1156 offset + mapping_size >= i_size_read(inode)) {
1157 /* limit mapping to block that spans EOF */
1158 mapping_size = roundup_64(i_size_read(inode) - offset,
1159 1 << inode->i_blkbits);
1161 if (mapping_size > LONG_MAX)
1162 mapping_size = LONG_MAX;
1164 bh_result->b_size = mapping_size;
1169 struct inode *inode,
1171 struct buffer_head *bh_result,
1176 struct xfs_inode *ip = XFS_I(inode);
1177 struct xfs_mount *mp = ip->i_mount;
1178 xfs_fileoff_t offset_fsb, end_fsb;
1181 struct xfs_bmbt_irec imap;
1187 BUG_ON(create && !direct);
1189 if (XFS_FORCED_SHUTDOWN(mp))
1192 offset = (xfs_off_t)iblock << inode->i_blkbits;
1193 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1194 size = bh_result->b_size;
1196 if (!create && offset >= i_size_read(inode))
1200 * Direct I/O is usually done on preallocated files, so try getting
1201 * a block mapping without an exclusive lock first.
1203 lockmode = xfs_ilock_data_map_shared(ip);
1205 ASSERT(offset <= mp->m_super->s_maxbytes);
1206 if (offset + size > mp->m_super->s_maxbytes)
1207 size = mp->m_super->s_maxbytes - offset;
1208 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1209 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1211 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1212 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1216 /* for DAX, we convert unwritten extents directly */
1219 (imap.br_startblock == HOLESTARTBLOCK ||
1220 imap.br_startblock == DELAYSTARTBLOCK) ||
1221 (IS_DAX(inode) && ISUNWRITTEN(&imap)))) {
1223 * xfs_iomap_write_direct() expects the shared lock. It
1224 * is unlocked on return.
1226 if (lockmode == XFS_ILOCK_EXCL)
1227 xfs_ilock_demote(ip, lockmode);
1229 error = xfs_iomap_write_direct(ip, offset, size,
1235 trace_xfs_get_blocks_alloc(ip, offset, size,
1236 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1237 : XFS_IO_DELALLOC, &imap);
1238 } else if (nimaps) {
1239 trace_xfs_get_blocks_found(ip, offset, size,
1240 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1241 : XFS_IO_OVERWRITE, &imap);
1242 xfs_iunlock(ip, lockmode);
1244 trace_xfs_get_blocks_notfound(ip, offset, size);
1248 if (IS_DAX(inode) && create) {
1249 ASSERT(!ISUNWRITTEN(&imap));
1250 /* zeroing is not needed at a higher layer */
1254 /* trim mapping down to size requested */
1255 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1258 * For unwritten extents do not report a disk address in the buffered
1259 * read case (treat as if we're reading into a hole).
1261 if (imap.br_startblock != HOLESTARTBLOCK &&
1262 imap.br_startblock != DELAYSTARTBLOCK &&
1263 (create || !ISUNWRITTEN(&imap))) {
1264 xfs_map_buffer(inode, bh_result, &imap, offset);
1265 if (ISUNWRITTEN(&imap))
1266 set_buffer_unwritten(bh_result);
1267 /* direct IO needs special help */
1270 ASSERT(!ISUNWRITTEN(&imap));
1272 xfs_map_direct(inode, bh_result, &imap, offset);
1277 * If this is a realtime file, data may be on a different device.
1278 * to that pointed to from the buffer_head b_bdev currently.
1280 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1283 * If we previously allocated a block out beyond eof and we are now
1284 * coming back to use it then we will need to flag it as new even if it
1285 * has a disk address.
1287 * With sub-block writes into unwritten extents we also need to mark
1288 * the buffer as new so that the unwritten parts of the buffer gets
1292 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1293 (offset >= i_size_read(inode)) ||
1294 (new || ISUNWRITTEN(&imap))))
1295 set_buffer_new(bh_result);
1297 BUG_ON(direct && imap.br_startblock == DELAYSTARTBLOCK);
1302 xfs_iunlock(ip, lockmode);
1308 struct inode *inode,
1310 struct buffer_head *bh_result,
1313 return __xfs_get_blocks(inode, iblock, bh_result, create, false, false);
1317 xfs_get_blocks_direct(
1318 struct inode *inode,
1320 struct buffer_head *bh_result,
1323 return __xfs_get_blocks(inode, iblock, bh_result, create, true, false);
1327 xfs_get_blocks_dax_fault(
1328 struct inode *inode,
1330 struct buffer_head *bh_result,
1333 return __xfs_get_blocks(inode, iblock, bh_result, create, true, true);
1337 * Complete a direct I/O write request.
1339 * xfs_map_direct passes us some flags in the private data to tell us what to
1340 * do. If no flags are set, then the write IO is an overwrite wholly within
1341 * the existing allocated file size and so there is nothing for us to do.
1343 * Note that in this case the completion can be called in interrupt context,
1344 * whereas if we have flags set we will always be called in task context
1345 * (i.e. from a workqueue).
1348 xfs_end_io_direct_write(
1354 struct inode *inode = file_inode(iocb->ki_filp);
1355 struct xfs_inode *ip = XFS_I(inode);
1356 uintptr_t flags = (uintptr_t)private;
1359 trace_xfs_end_io_direct_write(ip, offset, size);
1361 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1368 * The flags tell us whether we are doing unwritten extent conversions
1369 * or an append transaction that updates the on-disk file size. These
1370 * cases are the only cases where we should *potentially* be needing
1371 * to update the VFS inode size.
1374 ASSERT(offset + size <= i_size_read(inode));
1379 * We need to update the in-core inode size here so that we don't end up
1380 * with the on-disk inode size being outside the in-core inode size. We
1381 * have no other method of updating EOF for AIO, so always do it here
1384 * We need to lock the test/set EOF update as we can be racing with
1385 * other IO completions here to update the EOF. Failing to serialise
1386 * here can result in EOF moving backwards and Bad Things Happen when
1389 spin_lock(&ip->i_flags_lock);
1390 if (offset + size > i_size_read(inode))
1391 i_size_write(inode, offset + size);
1392 spin_unlock(&ip->i_flags_lock);
1394 if (flags & XFS_DIO_FLAG_UNWRITTEN) {
1395 trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
1397 error = xfs_iomap_write_unwritten(ip, offset, size);
1398 } else if (flags & XFS_DIO_FLAG_APPEND) {
1399 trace_xfs_end_io_direct_write_append(ip, offset, size);
1401 error = xfs_setfilesize(ip, offset, size);
1410 struct iov_iter *iter)
1413 * We just need the method present so that open/fcntl allow direct I/O.
1420 struct address_space *mapping,
1423 struct inode *inode = (struct inode *)mapping->host;
1424 struct xfs_inode *ip = XFS_I(inode);
1426 trace_xfs_vm_bmap(XFS_I(inode));
1427 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1428 filemap_write_and_wait(mapping);
1429 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1430 return generic_block_bmap(mapping, block, xfs_get_blocks);
1435 struct file *unused,
1438 trace_xfs_vm_readpage(page->mapping->host, 1);
1439 return mpage_readpage(page, xfs_get_blocks);
1444 struct file *unused,
1445 struct address_space *mapping,
1446 struct list_head *pages,
1449 trace_xfs_vm_readpages(mapping->host, nr_pages);
1450 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1454 * This is basically a copy of __set_page_dirty_buffers() with one
1455 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1456 * dirty, we'll never be able to clean them because we don't write buffers
1457 * beyond EOF, and that means we can't invalidate pages that span EOF
1458 * that have been marked dirty. Further, the dirty state can leak into
1459 * the file interior if the file is extended, resulting in all sorts of
1460 * bad things happening as the state does not match the underlying data.
1462 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1463 * this only exist because of bufferheads and how the generic code manages them.
1466 xfs_vm_set_page_dirty(
1469 struct address_space *mapping = page->mapping;
1470 struct inode *inode = mapping->host;
1475 if (unlikely(!mapping))
1476 return !TestSetPageDirty(page);
1478 end_offset = i_size_read(inode);
1479 offset = page_offset(page);
1481 spin_lock(&mapping->private_lock);
1482 if (page_has_buffers(page)) {
1483 struct buffer_head *head = page_buffers(page);
1484 struct buffer_head *bh = head;
1487 if (offset < end_offset)
1488 set_buffer_dirty(bh);
1489 bh = bh->b_this_page;
1490 offset += 1 << inode->i_blkbits;
1491 } while (bh != head);
1494 * Lock out page->mem_cgroup migration to keep PageDirty
1495 * synchronized with per-memcg dirty page counters.
1497 lock_page_memcg(page);
1498 newly_dirty = !TestSetPageDirty(page);
1499 spin_unlock(&mapping->private_lock);
1502 /* sigh - __set_page_dirty() is static, so copy it here, too */
1503 unsigned long flags;
1505 spin_lock_irqsave(&mapping->tree_lock, flags);
1506 if (page->mapping) { /* Race with truncate? */
1507 WARN_ON_ONCE(!PageUptodate(page));
1508 account_page_dirtied(page, mapping);
1509 radix_tree_tag_set(&mapping->page_tree,
1510 page_index(page), PAGECACHE_TAG_DIRTY);
1512 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1514 unlock_page_memcg(page);
1516 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1520 const struct address_space_operations xfs_address_space_operations = {
1521 .readpage = xfs_vm_readpage,
1522 .readpages = xfs_vm_readpages,
1523 .writepage = xfs_vm_writepage,
1524 .writepages = xfs_vm_writepages,
1525 .set_page_dirty = xfs_vm_set_page_dirty,
1526 .releasepage = xfs_vm_releasepage,
1527 .invalidatepage = xfs_vm_invalidatepage,
1528 .bmap = xfs_vm_bmap,
1529 .direct_IO = xfs_vm_direct_IO,
1530 .migratepage = buffer_migrate_page,
1531 .is_partially_uptodate = block_is_partially_uptodate,
1532 .error_remove_page = generic_error_remove_page,