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
24 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
32 #include "xfs_iomap.h"
33 #include "xfs_vnodeops.h"
34 #include "xfs_trace.h"
36 #include <linux/gfp.h>
37 #include <linux/mpage.h>
38 #include <linux/pagevec.h>
39 #include <linux/writeback.h>
47 struct buffer_head *bh, *head;
49 *delalloc = *unwritten = 0;
51 bh = head = page_buffers(page);
53 if (buffer_unwritten(bh))
55 else if (buffer_delay(bh))
57 } while ((bh = bh->b_this_page) != head);
60 STATIC struct block_device *
61 xfs_find_bdev_for_inode(
64 struct xfs_inode *ip = XFS_I(inode);
65 struct xfs_mount *mp = ip->i_mount;
67 if (XFS_IS_REALTIME_INODE(ip))
68 return mp->m_rtdev_targp->bt_bdev;
70 return mp->m_ddev_targp->bt_bdev;
74 * We're now finished for good with this ioend structure.
75 * Update the page state via the associated buffer_heads,
76 * release holds on the inode and bio, and finally free
77 * up memory. Do not use the ioend after this.
83 struct buffer_head *bh, *next;
85 for (bh = ioend->io_buffer_head; bh; bh = next) {
87 bh->b_end_io(bh, !ioend->io_error);
91 inode_dio_done(ioend->io_inode);
92 if (ioend->io_isasync) {
93 aio_complete(ioend->io_iocb, ioend->io_error ?
94 ioend->io_error : ioend->io_result, 0);
98 mempool_free(ioend, xfs_ioend_pool);
102 * If the end of the current ioend is beyond the current EOF,
103 * return the new EOF value, otherwise zero.
109 xfs_inode_t *ip = XFS_I(ioend->io_inode);
113 bsize = ioend->io_offset + ioend->io_size;
114 isize = MAX(ip->i_size, ip->i_new_size);
115 isize = MIN(isize, bsize);
116 return isize > ip->i_d.di_size ? isize : 0;
120 * Fast and loose check if this write could update the on-disk inode size.
122 static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
124 return ioend->io_offset + ioend->io_size >
125 XFS_I(ioend->io_inode)->i_d.di_size;
129 * Update on-disk file size now that data has been written to disk. The
130 * current in-memory file size is i_size. If a write is beyond eof i_new_size
131 * will be the intended file size until i_size is updated. If this write does
132 * not extend all the way to the valid file size then restrict this update to
133 * the end of the write.
135 * This function does not block as blocking on the inode lock in IO completion
136 * can lead to IO completion order dependency deadlocks.. If it can't get the
137 * inode ilock it will return EAGAIN. Callers must handle this.
143 xfs_inode_t *ip = XFS_I(ioend->io_inode);
146 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
149 isize = xfs_ioend_new_eof(ioend);
151 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
152 ip->i_d.di_size = isize;
153 xfs_mark_inode_dirty(ip);
156 xfs_iunlock(ip, XFS_ILOCK_EXCL);
161 * Schedule IO completion handling on the final put of an ioend.
163 * If there is no work to do we might as well call it a day and free the
168 struct xfs_ioend *ioend)
170 if (atomic_dec_and_test(&ioend->io_remaining)) {
171 if (ioend->io_type == IO_UNWRITTEN)
172 queue_work(xfsconvertd_workqueue, &ioend->io_work);
173 else if (xfs_ioend_is_append(ioend))
174 queue_work(xfsdatad_workqueue, &ioend->io_work);
176 xfs_destroy_ioend(ioend);
181 * IO write completion.
185 struct work_struct *work)
187 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
188 struct xfs_inode *ip = XFS_I(ioend->io_inode);
191 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
192 ioend->io_error = -EIO;
199 * For unwritten extents we need to issue transactions to convert a
200 * range to normal written extens after the data I/O has finished.
202 if (ioend->io_type == IO_UNWRITTEN) {
203 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
206 ioend->io_error = -error;
212 * We might have to update the on-disk file size after extending
215 error = xfs_setfilesize(ioend);
216 ASSERT(!error || error == EAGAIN);
220 * If we didn't complete processing of the ioend, requeue it to the
221 * tail of the workqueue for another attempt later. Otherwise destroy
224 if (error == EAGAIN) {
225 atomic_inc(&ioend->io_remaining);
226 xfs_finish_ioend(ioend);
227 /* ensure we don't spin on blocked ioends */
230 xfs_destroy_ioend(ioend);
235 * Call IO completion handling in caller context on the final put of an ioend.
238 xfs_finish_ioend_sync(
239 struct xfs_ioend *ioend)
241 if (atomic_dec_and_test(&ioend->io_remaining))
242 xfs_end_io(&ioend->io_work);
246 * Allocate and initialise an IO completion structure.
247 * We need to track unwritten extent write completion here initially.
248 * We'll need to extend this for updating the ondisk inode size later
258 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
261 * Set the count to 1 initially, which will prevent an I/O
262 * completion callback from happening before we have started
263 * all the I/O from calling the completion routine too early.
265 atomic_set(&ioend->io_remaining, 1);
266 ioend->io_isasync = 0;
268 ioend->io_list = NULL;
269 ioend->io_type = type;
270 ioend->io_inode = inode;
271 ioend->io_buffer_head = NULL;
272 ioend->io_buffer_tail = NULL;
273 ioend->io_offset = 0;
275 ioend->io_iocb = NULL;
276 ioend->io_result = 0;
278 INIT_WORK(&ioend->io_work, xfs_end_io);
286 struct xfs_bmbt_irec *imap,
290 struct xfs_inode *ip = XFS_I(inode);
291 struct xfs_mount *mp = ip->i_mount;
292 ssize_t count = 1 << inode->i_blkbits;
293 xfs_fileoff_t offset_fsb, end_fsb;
295 int bmapi_flags = XFS_BMAPI_ENTIRE;
298 if (XFS_FORCED_SHUTDOWN(mp))
299 return -XFS_ERROR(EIO);
301 if (type == IO_UNWRITTEN)
302 bmapi_flags |= XFS_BMAPI_IGSTATE;
304 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
306 return -XFS_ERROR(EAGAIN);
307 xfs_ilock(ip, XFS_ILOCK_SHARED);
310 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
311 (ip->i_df.if_flags & XFS_IFEXTENTS));
312 ASSERT(offset <= mp->m_maxioffset);
314 if (offset + count > mp->m_maxioffset)
315 count = mp->m_maxioffset - offset;
316 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
317 offset_fsb = XFS_B_TO_FSBT(mp, offset);
318 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
319 imap, &nimaps, bmapi_flags);
320 xfs_iunlock(ip, XFS_ILOCK_SHARED);
323 return -XFS_ERROR(error);
325 if (type == IO_DELALLOC &&
326 (!nimaps || isnullstartblock(imap->br_startblock))) {
327 error = xfs_iomap_write_allocate(ip, offset, count, imap);
329 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
330 return -XFS_ERROR(error);
334 if (type == IO_UNWRITTEN) {
336 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
337 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
341 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
348 struct xfs_bmbt_irec *imap,
351 offset >>= inode->i_blkbits;
353 return offset >= imap->br_startoff &&
354 offset < imap->br_startoff + imap->br_blockcount;
358 * BIO completion handler for buffered IO.
365 xfs_ioend_t *ioend = bio->bi_private;
367 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
368 if (!ioend->io_error && !test_bit(BIO_UPTODATE, &bio->bi_flags))
369 ioend->io_error = error;
371 /* Toss bio and pass work off to an xfsdatad thread */
372 bio->bi_private = NULL;
373 bio->bi_end_io = NULL;
376 xfs_finish_ioend(ioend);
380 xfs_submit_ioend_bio(
381 struct writeback_control *wbc,
385 atomic_inc(&ioend->io_remaining);
386 bio->bi_private = ioend;
387 bio->bi_end_io = xfs_end_bio;
390 * If the I/O is beyond EOF we mark the inode dirty immediately
391 * but don't update the inode size until I/O completion.
393 if (xfs_ioend_new_eof(ioend))
394 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
396 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
401 struct buffer_head *bh)
403 int nvecs = bio_get_nr_vecs(bh->b_bdev);
404 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
406 ASSERT(bio->bi_private == NULL);
407 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
408 bio->bi_bdev = bh->b_bdev;
413 xfs_start_buffer_writeback(
414 struct buffer_head *bh)
416 ASSERT(buffer_mapped(bh));
417 ASSERT(buffer_locked(bh));
418 ASSERT(!buffer_delay(bh));
419 ASSERT(!buffer_unwritten(bh));
421 mark_buffer_async_write(bh);
422 set_buffer_uptodate(bh);
423 clear_buffer_dirty(bh);
427 xfs_start_page_writeback(
432 ASSERT(PageLocked(page));
433 ASSERT(!PageWriteback(page));
435 clear_page_dirty_for_io(page);
436 set_page_writeback(page);
438 /* If no buffers on the page are to be written, finish it here */
440 end_page_writeback(page);
443 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
445 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
449 * Submit all of the bios for all of the ioends we have saved up, covering the
450 * initial writepage page and also any probed pages.
452 * Because we may have multiple ioends spanning a page, we need to start
453 * writeback on all the buffers before we submit them for I/O. If we mark the
454 * buffers as we got, then we can end up with a page that only has buffers
455 * marked async write and I/O complete on can occur before we mark the other
456 * buffers async write.
458 * The end result of this is that we trip a bug in end_page_writeback() because
459 * we call it twice for the one page as the code in end_buffer_async_write()
460 * assumes that all buffers on the page are started at the same time.
462 * The fix is two passes across the ioend list - one to start writeback on the
463 * buffer_heads, and then submit them for I/O on the second pass.
467 struct writeback_control *wbc,
470 xfs_ioend_t *head = ioend;
472 struct buffer_head *bh;
474 sector_t lastblock = 0;
476 /* Pass 1 - start writeback */
478 next = ioend->io_list;
479 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
480 xfs_start_buffer_writeback(bh);
481 } while ((ioend = next) != NULL);
483 /* Pass 2 - submit I/O */
486 next = ioend->io_list;
489 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
493 bio = xfs_alloc_ioend_bio(bh);
494 } else if (bh->b_blocknr != lastblock + 1) {
495 xfs_submit_ioend_bio(wbc, ioend, bio);
499 if (bio_add_buffer(bio, bh) != bh->b_size) {
500 xfs_submit_ioend_bio(wbc, ioend, bio);
504 lastblock = bh->b_blocknr;
507 xfs_submit_ioend_bio(wbc, ioend, bio);
508 xfs_finish_ioend(ioend);
509 } while ((ioend = next) != NULL);
513 * Cancel submission of all buffer_heads so far in this endio.
514 * Toss the endio too. Only ever called for the initial page
515 * in a writepage request, so only ever one page.
522 struct buffer_head *bh, *next_bh;
525 next = ioend->io_list;
526 bh = ioend->io_buffer_head;
528 next_bh = bh->b_private;
529 clear_buffer_async_write(bh);
531 } while ((bh = next_bh) != NULL);
533 mempool_free(ioend, xfs_ioend_pool);
534 } while ((ioend = next) != NULL);
538 * Test to see if we've been building up a completion structure for
539 * earlier buffers -- if so, we try to append to this ioend if we
540 * can, otherwise we finish off any current ioend and start another.
541 * Return true if we've finished the given ioend.
546 struct buffer_head *bh,
549 xfs_ioend_t **result,
552 xfs_ioend_t *ioend = *result;
554 if (!ioend || need_ioend || type != ioend->io_type) {
555 xfs_ioend_t *previous = *result;
557 ioend = xfs_alloc_ioend(inode, type);
558 ioend->io_offset = offset;
559 ioend->io_buffer_head = bh;
560 ioend->io_buffer_tail = bh;
562 previous->io_list = ioend;
565 ioend->io_buffer_tail->b_private = bh;
566 ioend->io_buffer_tail = bh;
569 bh->b_private = NULL;
570 ioend->io_size += bh->b_size;
576 struct buffer_head *bh,
577 struct xfs_bmbt_irec *imap,
581 struct xfs_mount *m = XFS_I(inode)->i_mount;
582 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
583 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
585 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
586 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
588 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
589 ((offset - iomap_offset) >> inode->i_blkbits);
591 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
594 set_buffer_mapped(bh);
600 struct buffer_head *bh,
601 struct xfs_bmbt_irec *imap,
604 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
605 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
607 xfs_map_buffer(inode, bh, imap, offset);
608 set_buffer_mapped(bh);
609 clear_buffer_delay(bh);
610 clear_buffer_unwritten(bh);
614 * Test if a given page is suitable for writing as part of an unwritten
615 * or delayed allocate extent.
622 if (PageWriteback(page))
625 if (page->mapping && page_has_buffers(page)) {
626 struct buffer_head *bh, *head;
629 bh = head = page_buffers(page);
631 if (buffer_unwritten(bh))
632 acceptable = (type == IO_UNWRITTEN);
633 else if (buffer_delay(bh))
634 acceptable = (type == IO_DELALLOC);
635 else if (buffer_dirty(bh) && buffer_mapped(bh))
636 acceptable = (type == IO_OVERWRITE);
639 } while ((bh = bh->b_this_page) != head);
649 * Allocate & map buffers for page given the extent map. Write it out.
650 * except for the original page of a writepage, this is called on
651 * delalloc/unwritten pages only, for the original page it is possible
652 * that the page has no mapping at all.
659 struct xfs_bmbt_irec *imap,
660 xfs_ioend_t **ioendp,
661 struct writeback_control *wbc)
663 struct buffer_head *bh, *head;
664 xfs_off_t end_offset;
665 unsigned long p_offset;
668 int count = 0, done = 0, uptodate = 1;
669 xfs_off_t offset = page_offset(page);
671 if (page->index != tindex)
673 if (!trylock_page(page))
675 if (PageWriteback(page))
676 goto fail_unlock_page;
677 if (page->mapping != inode->i_mapping)
678 goto fail_unlock_page;
679 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
680 goto fail_unlock_page;
683 * page_dirty is initially a count of buffers on the page before
684 * EOF and is decremented as we move each into a cleanable state.
688 * End offset is the highest offset that this page should represent.
689 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
690 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
691 * hence give us the correct page_dirty count. On any other page,
692 * it will be zero and in that case we need page_dirty to be the
693 * count of buffers on the page.
695 end_offset = min_t(unsigned long long,
696 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
699 len = 1 << inode->i_blkbits;
700 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
702 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
703 page_dirty = p_offset / len;
705 bh = head = page_buffers(page);
707 if (offset >= end_offset)
709 if (!buffer_uptodate(bh))
711 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
716 if (buffer_unwritten(bh) || buffer_delay(bh) ||
718 if (buffer_unwritten(bh))
720 else if (buffer_delay(bh))
725 if (!xfs_imap_valid(inode, imap, offset)) {
731 if (type != IO_OVERWRITE)
732 xfs_map_at_offset(inode, bh, imap, offset);
733 xfs_add_to_ioend(inode, bh, offset, type,
741 } while (offset += len, (bh = bh->b_this_page) != head);
743 if (uptodate && bh == head)
744 SetPageUptodate(page);
747 if (--wbc->nr_to_write <= 0 &&
748 wbc->sync_mode == WB_SYNC_NONE)
751 xfs_start_page_writeback(page, !page_dirty, count);
761 * Convert & write out a cluster of pages in the same extent as defined
762 * by mp and following the start page.
768 struct xfs_bmbt_irec *imap,
769 xfs_ioend_t **ioendp,
770 struct writeback_control *wbc,
776 pagevec_init(&pvec, 0);
777 while (!done && tindex <= tlast) {
778 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
780 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
783 for (i = 0; i < pagevec_count(&pvec); i++) {
784 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
790 pagevec_release(&pvec);
796 xfs_vm_invalidatepage(
798 unsigned long offset)
800 trace_xfs_invalidatepage(page->mapping->host, page, offset);
801 block_invalidatepage(page, offset);
805 * If the page has delalloc buffers on it, we need to punch them out before we
806 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
807 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
808 * is done on that same region - the delalloc extent is returned when none is
809 * supposed to be there.
811 * We prevent this by truncating away the delalloc regions on the page before
812 * invalidating it. Because they are delalloc, we can do this without needing a
813 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
814 * truncation without a transaction as there is no space left for block
815 * reservation (typically why we see a ENOSPC in writeback).
817 * This is not a performance critical path, so for now just do the punching a
818 * buffer head at a time.
821 xfs_aops_discard_page(
824 struct inode *inode = page->mapping->host;
825 struct xfs_inode *ip = XFS_I(inode);
826 struct buffer_head *bh, *head;
827 loff_t offset = page_offset(page);
829 if (!xfs_is_delayed_page(page, IO_DELALLOC))
832 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
835 xfs_alert(ip->i_mount,
836 "page discard on page %p, inode 0x%llx, offset %llu.",
837 page, ip->i_ino, offset);
839 xfs_ilock(ip, XFS_ILOCK_EXCL);
840 bh = head = page_buffers(page);
843 xfs_fileoff_t start_fsb;
845 if (!buffer_delay(bh))
848 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
849 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
851 /* something screwed, just bail */
852 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
853 xfs_alert(ip->i_mount,
854 "page discard unable to remove delalloc mapping.");
859 offset += 1 << inode->i_blkbits;
861 } while ((bh = bh->b_this_page) != head);
863 xfs_iunlock(ip, XFS_ILOCK_EXCL);
865 xfs_vm_invalidatepage(page, 0);
870 * Write out a dirty page.
872 * For delalloc space on the page we need to allocate space and flush it.
873 * For unwritten space on the page we need to start the conversion to
874 * regular allocated space.
875 * For any other dirty buffer heads on the page we should flush them.
880 struct writeback_control *wbc)
882 struct inode *inode = page->mapping->host;
883 struct buffer_head *bh, *head;
884 struct xfs_bmbt_irec imap;
885 xfs_ioend_t *ioend = NULL, *iohead = NULL;
888 __uint64_t end_offset;
889 pgoff_t end_index, last_index;
891 int err, imap_valid = 0, uptodate = 1;
895 trace_xfs_writepage(inode, page, 0);
897 ASSERT(page_has_buffers(page));
900 * Refuse to write the page out if we are called from reclaim context.
902 * This avoids stack overflows when called from deeply used stacks in
903 * random callers for direct reclaim or memcg reclaim. We explicitly
904 * allow reclaim from kswapd as the stack usage there is relatively low.
906 * This should never happen except in the case of a VM regression so
909 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
914 * Given that we do not allow direct reclaim to call us, we should
915 * never be called while in a filesystem transaction.
917 if (WARN_ON(current->flags & PF_FSTRANS))
920 /* Is this page beyond the end of the file? */
921 offset = i_size_read(inode);
922 end_index = offset >> PAGE_CACHE_SHIFT;
923 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
924 if (page->index >= end_index) {
925 if ((page->index >= end_index + 1) ||
926 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
932 end_offset = min_t(unsigned long long,
933 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
935 len = 1 << inode->i_blkbits;
937 bh = head = page_buffers(page);
938 offset = page_offset(page);
941 if (wbc->sync_mode == WB_SYNC_NONE)
947 if (offset >= end_offset)
949 if (!buffer_uptodate(bh))
953 * set_page_dirty dirties all buffers in a page, independent
954 * of their state. The dirty state however is entirely
955 * meaningless for holes (!mapped && uptodate), so skip
956 * buffers covering holes here.
958 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
963 if (buffer_unwritten(bh)) {
964 if (type != IO_UNWRITTEN) {
968 } else if (buffer_delay(bh)) {
969 if (type != IO_DELALLOC) {
973 } else if (buffer_uptodate(bh)) {
974 if (type != IO_OVERWRITE) {
979 if (PageUptodate(page)) {
980 ASSERT(buffer_mapped(bh));
987 imap_valid = xfs_imap_valid(inode, &imap, offset);
990 * If we didn't have a valid mapping then we need to
991 * put the new mapping into a separate ioend structure.
992 * This ensures non-contiguous extents always have
993 * separate ioends, which is particularly important
994 * for unwritten extent conversion at I/O completion
998 err = xfs_map_blocks(inode, offset, &imap, type,
1002 imap_valid = xfs_imap_valid(inode, &imap, offset);
1006 if (type != IO_OVERWRITE)
1007 xfs_map_at_offset(inode, bh, &imap, offset);
1008 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1016 } while (offset += len, ((bh = bh->b_this_page) != head));
1018 if (uptodate && bh == head)
1019 SetPageUptodate(page);
1021 xfs_start_page_writeback(page, 1, count);
1023 if (ioend && imap_valid) {
1024 xfs_off_t end_index;
1026 end_index = imap.br_startoff + imap.br_blockcount;
1029 end_index <<= inode->i_blkbits;
1032 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1034 /* check against file size */
1035 if (end_index > last_index)
1036 end_index = last_index;
1038 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1043 xfs_submit_ioend(wbc, iohead);
1049 xfs_cancel_ioend(iohead);
1054 xfs_aops_discard_page(page);
1055 ClearPageUptodate(page);
1060 redirty_page_for_writepage(wbc, page);
1067 struct address_space *mapping,
1068 struct writeback_control *wbc)
1070 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1071 return generic_writepages(mapping, wbc);
1075 * Called to move a page into cleanable state - and from there
1076 * to be released. The page should already be clean. We always
1077 * have buffer heads in this call.
1079 * Returns 1 if the page is ok to release, 0 otherwise.
1086 int delalloc, unwritten;
1088 trace_xfs_releasepage(page->mapping->host, page, 0);
1090 xfs_count_page_state(page, &delalloc, &unwritten);
1092 if (WARN_ON(delalloc))
1094 if (WARN_ON(unwritten))
1097 return try_to_free_buffers(page);
1102 struct inode *inode,
1104 struct buffer_head *bh_result,
1108 struct xfs_inode *ip = XFS_I(inode);
1109 struct xfs_mount *mp = ip->i_mount;
1110 xfs_fileoff_t offset_fsb, end_fsb;
1113 struct xfs_bmbt_irec imap;
1119 if (XFS_FORCED_SHUTDOWN(mp))
1120 return -XFS_ERROR(EIO);
1122 offset = (xfs_off_t)iblock << inode->i_blkbits;
1123 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1124 size = bh_result->b_size;
1126 if (!create && direct && offset >= i_size_read(inode))
1130 lockmode = XFS_ILOCK_EXCL;
1131 xfs_ilock(ip, lockmode);
1133 lockmode = xfs_ilock_map_shared(ip);
1136 ASSERT(offset <= mp->m_maxioffset);
1137 if (offset + size > mp->m_maxioffset)
1138 size = mp->m_maxioffset - offset;
1139 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1140 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1142 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1143 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1149 (imap.br_startblock == HOLESTARTBLOCK ||
1150 imap.br_startblock == DELAYSTARTBLOCK))) {
1152 error = xfs_iomap_write_direct(ip, offset, size,
1155 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1160 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1161 } else if (nimaps) {
1162 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1164 trace_xfs_get_blocks_notfound(ip, offset, size);
1167 xfs_iunlock(ip, lockmode);
1169 if (imap.br_startblock != HOLESTARTBLOCK &&
1170 imap.br_startblock != DELAYSTARTBLOCK) {
1172 * For unwritten extents do not report a disk address on
1173 * the read case (treat as if we're reading into a hole).
1175 if (create || !ISUNWRITTEN(&imap))
1176 xfs_map_buffer(inode, bh_result, &imap, offset);
1177 if (create && ISUNWRITTEN(&imap)) {
1179 bh_result->b_private = inode;
1180 set_buffer_unwritten(bh_result);
1185 * If this is a realtime file, data may be on a different device.
1186 * to that pointed to from the buffer_head b_bdev currently.
1188 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1191 * If we previously allocated a block out beyond eof and we are now
1192 * coming back to use it then we will need to flag it as new even if it
1193 * has a disk address.
1195 * With sub-block writes into unwritten extents we also need to mark
1196 * the buffer as new so that the unwritten parts of the buffer gets
1200 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1201 (offset >= i_size_read(inode)) ||
1202 (new || ISUNWRITTEN(&imap))))
1203 set_buffer_new(bh_result);
1205 if (imap.br_startblock == DELAYSTARTBLOCK) {
1208 set_buffer_uptodate(bh_result);
1209 set_buffer_mapped(bh_result);
1210 set_buffer_delay(bh_result);
1215 * If this is O_DIRECT or the mpage code calling tell them how large
1216 * the mapping is, so that we can avoid repeated get_blocks calls.
1218 if (direct || size > (1 << inode->i_blkbits)) {
1219 xfs_off_t mapping_size;
1221 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1222 mapping_size <<= inode->i_blkbits;
1224 ASSERT(mapping_size > 0);
1225 if (mapping_size > size)
1226 mapping_size = size;
1227 if (mapping_size > LONG_MAX)
1228 mapping_size = LONG_MAX;
1230 bh_result->b_size = mapping_size;
1236 xfs_iunlock(ip, lockmode);
1242 struct inode *inode,
1244 struct buffer_head *bh_result,
1247 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1251 xfs_get_blocks_direct(
1252 struct inode *inode,
1254 struct buffer_head *bh_result,
1257 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1261 * Complete a direct I/O write request.
1263 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1264 * need to issue a transaction to convert the range from unwritten to written
1265 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1266 * to do this and we are done. But in case this was a successful AIO
1267 * request this handler is called from interrupt context, from which we
1268 * can't start transactions. In that case offload the I/O completion to
1269 * the workqueues we also use for buffered I/O completion.
1272 xfs_end_io_direct_write(
1280 struct xfs_ioend *ioend = iocb->private;
1283 * blockdev_direct_IO can return an error even after the I/O
1284 * completion handler was called. Thus we need to protect
1285 * against double-freeing.
1287 iocb->private = NULL;
1289 ioend->io_offset = offset;
1290 ioend->io_size = size;
1291 ioend->io_iocb = iocb;
1292 ioend->io_result = ret;
1293 if (private && size > 0)
1294 ioend->io_type = IO_UNWRITTEN;
1297 ioend->io_isasync = 1;
1298 xfs_finish_ioend(ioend);
1300 xfs_finish_ioend_sync(ioend);
1308 const struct iovec *iov,
1310 unsigned long nr_segs)
1312 struct inode *inode = iocb->ki_filp->f_mapping->host;
1313 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1317 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1319 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1321 xfs_get_blocks_direct,
1322 xfs_end_io_direct_write, NULL, 0);
1323 if (ret != -EIOCBQUEUED && iocb->private)
1324 xfs_destroy_ioend(iocb->private);
1326 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1328 xfs_get_blocks_direct,
1336 xfs_vm_write_failed(
1337 struct address_space *mapping,
1340 struct inode *inode = mapping->host;
1342 if (to > inode->i_size) {
1344 * punch out the delalloc blocks we have already allocated. We
1345 * don't call xfs_setattr() to do this as we may be in the
1346 * middle of a multi-iovec write and so the vfs inode->i_size
1347 * will not match the xfs ip->i_size and so it will zero too
1348 * much. Hence we jus truncate the page cache to zero what is
1349 * necessary and punch the delalloc blocks directly.
1351 struct xfs_inode *ip = XFS_I(inode);
1352 xfs_fileoff_t start_fsb;
1353 xfs_fileoff_t end_fsb;
1356 truncate_pagecache(inode, to, inode->i_size);
1359 * Check if there are any blocks that are outside of i_size
1360 * that need to be trimmed back.
1362 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1363 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1364 if (end_fsb <= start_fsb)
1367 xfs_ilock(ip, XFS_ILOCK_EXCL);
1368 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1369 end_fsb - start_fsb);
1371 /* something screwed, just bail */
1372 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1373 xfs_alert(ip->i_mount,
1374 "xfs_vm_write_failed: unable to clean up ino %lld",
1378 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1385 struct address_space *mapping,
1389 struct page **pagep,
1394 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1395 pagep, xfs_get_blocks);
1397 xfs_vm_write_failed(mapping, pos + len);
1404 struct address_space *mapping,
1413 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1414 if (unlikely(ret < len))
1415 xfs_vm_write_failed(mapping, pos + len);
1421 struct address_space *mapping,
1424 struct inode *inode = (struct inode *)mapping->host;
1425 struct xfs_inode *ip = XFS_I(inode);
1427 trace_xfs_vm_bmap(XFS_I(inode));
1428 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1429 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1430 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1431 return generic_block_bmap(mapping, block, xfs_get_blocks);
1436 struct file *unused,
1439 return mpage_readpage(page, xfs_get_blocks);
1444 struct file *unused,
1445 struct address_space *mapping,
1446 struct list_head *pages,
1449 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1452 const struct address_space_operations xfs_address_space_operations = {
1453 .readpage = xfs_vm_readpage,
1454 .readpages = xfs_vm_readpages,
1455 .writepage = xfs_vm_writepage,
1456 .writepages = xfs_vm_writepages,
1457 .releasepage = xfs_vm_releasepage,
1458 .invalidatepage = xfs_vm_invalidatepage,
1459 .write_begin = xfs_vm_write_begin,
1460 .write_end = xfs_vm_write_end,
1461 .bmap = xfs_vm_bmap,
1462 .direct_IO = xfs_vm_direct_IO,
1463 .migratepage = buffer_migrate_page,
1464 .is_partially_uptodate = block_is_partially_uptodate,
1465 .error_remove_page = generic_error_remove_page,