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>
43 * Prime number of hash buckets since address is used as the key.
46 #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
47 static wait_queue_head_t xfs_ioend_wq[NVSYNC];
54 for (i = 0; i < NVSYNC; i++)
55 init_waitqueue_head(&xfs_ioend_wq[i]);
62 wait_queue_head_t *wq = to_ioend_wq(ip);
64 wait_event(*wq, (atomic_read(&ip->i_iocount) == 0));
71 if (atomic_dec_and_test(&ip->i_iocount))
72 wake_up(to_ioend_wq(ip));
81 struct buffer_head *bh, *head;
83 *delalloc = *unwritten = 0;
85 bh = head = page_buffers(page);
87 if (buffer_unwritten(bh))
89 else if (buffer_delay(bh))
91 } while ((bh = bh->b_this_page) != head);
94 STATIC struct block_device *
95 xfs_find_bdev_for_inode(
98 struct xfs_inode *ip = XFS_I(inode);
99 struct xfs_mount *mp = ip->i_mount;
101 if (XFS_IS_REALTIME_INODE(ip))
102 return mp->m_rtdev_targp->bt_bdev;
104 return mp->m_ddev_targp->bt_bdev;
108 * We're now finished for good with this ioend structure.
109 * Update the page state via the associated buffer_heads,
110 * release holds on the inode and bio, and finally free
111 * up memory. Do not use the ioend after this.
117 struct buffer_head *bh, *next;
118 struct xfs_inode *ip = XFS_I(ioend->io_inode);
120 for (bh = ioend->io_buffer_head; bh; bh = next) {
121 next = bh->b_private;
122 bh->b_end_io(bh, !ioend->io_error);
126 * Volume managers supporting multiple paths can send back ENODEV
127 * when the final path disappears. In this case continuing to fill
128 * the page cache with dirty data which cannot be written out is
129 * evil, so prevent that.
131 if (unlikely(ioend->io_error == -ENODEV)) {
132 xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ,
137 mempool_free(ioend, xfs_ioend_pool);
141 * If the end of the current ioend is beyond the current EOF,
142 * return the new EOF value, otherwise zero.
148 xfs_inode_t *ip = XFS_I(ioend->io_inode);
152 bsize = ioend->io_offset + ioend->io_size;
153 isize = MAX(ip->i_size, ip->i_new_size);
154 isize = MIN(isize, bsize);
155 return isize > ip->i_d.di_size ? isize : 0;
159 * Update on-disk file size now that data has been written to disk. The
160 * current in-memory file size is i_size. If a write is beyond eof i_new_size
161 * will be the intended file size until i_size is updated. If this write does
162 * not extend all the way to the valid file size then restrict this update to
163 * the end of the write.
165 * This function does not block as blocking on the inode lock in IO completion
166 * can lead to IO completion order dependency deadlocks.. If it can't get the
167 * inode ilock it will return EAGAIN. Callers must handle this.
173 xfs_inode_t *ip = XFS_I(ioend->io_inode);
176 if (unlikely(ioend->io_error))
179 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
182 isize = xfs_ioend_new_eof(ioend);
184 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
185 ip->i_d.di_size = isize;
186 xfs_mark_inode_dirty(ip);
189 xfs_iunlock(ip, XFS_ILOCK_EXCL);
194 * Schedule IO completion handling on the final put of an ioend.
198 struct xfs_ioend *ioend)
200 if (atomic_dec_and_test(&ioend->io_remaining)) {
201 if (ioend->io_type == IO_UNWRITTEN)
202 queue_work(xfsconvertd_workqueue, &ioend->io_work);
204 queue_work(xfsdatad_workqueue, &ioend->io_work);
209 * IO write completion.
213 struct work_struct *work)
215 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
216 struct xfs_inode *ip = XFS_I(ioend->io_inode);
220 * For unwritten extents we need to issue transactions to convert a
221 * range to normal written extens after the data I/O has finished.
223 if (ioend->io_type == IO_UNWRITTEN &&
224 likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) {
226 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
229 ioend->io_error = error;
233 * We might have to update the on-disk file size after extending
236 error = xfs_setfilesize(ioend);
237 ASSERT(!error || error == EAGAIN);
240 * If we didn't complete processing of the ioend, requeue it to the
241 * tail of the workqueue for another attempt later. Otherwise destroy
244 if (error == EAGAIN) {
245 atomic_inc(&ioend->io_remaining);
246 xfs_finish_ioend(ioend);
247 /* ensure we don't spin on blocked ioends */
251 aio_complete(ioend->io_iocb, ioend->io_result, 0);
252 xfs_destroy_ioend(ioend);
257 * Call IO completion handling in caller context on the final put of an ioend.
260 xfs_finish_ioend_sync(
261 struct xfs_ioend *ioend)
263 if (atomic_dec_and_test(&ioend->io_remaining))
264 xfs_end_io(&ioend->io_work);
268 * Allocate and initialise an IO completion structure.
269 * We need to track unwritten extent write completion here initially.
270 * We'll need to extend this for updating the ondisk inode size later
280 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
283 * Set the count to 1 initially, which will prevent an I/O
284 * completion callback from happening before we have started
285 * all the I/O from calling the completion routine too early.
287 atomic_set(&ioend->io_remaining, 1);
289 ioend->io_list = NULL;
290 ioend->io_type = type;
291 ioend->io_inode = inode;
292 ioend->io_buffer_head = NULL;
293 ioend->io_buffer_tail = NULL;
294 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
295 ioend->io_offset = 0;
297 ioend->io_iocb = NULL;
298 ioend->io_result = 0;
300 INIT_WORK(&ioend->io_work, xfs_end_io);
308 struct xfs_bmbt_irec *imap,
312 struct xfs_inode *ip = XFS_I(inode);
313 struct xfs_mount *mp = ip->i_mount;
314 ssize_t count = 1 << inode->i_blkbits;
315 xfs_fileoff_t offset_fsb, end_fsb;
317 int bmapi_flags = XFS_BMAPI_ENTIRE;
320 if (XFS_FORCED_SHUTDOWN(mp))
321 return -XFS_ERROR(EIO);
323 if (type == IO_UNWRITTEN)
324 bmapi_flags |= XFS_BMAPI_IGSTATE;
326 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
328 return -XFS_ERROR(EAGAIN);
329 xfs_ilock(ip, XFS_ILOCK_SHARED);
332 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
333 (ip->i_df.if_flags & XFS_IFEXTENTS));
334 ASSERT(offset <= mp->m_maxioffset);
336 if (offset + count > mp->m_maxioffset)
337 count = mp->m_maxioffset - offset;
338 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
339 offset_fsb = XFS_B_TO_FSBT(mp, offset);
340 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
341 bmapi_flags, NULL, 0, imap, &nimaps, NULL);
342 xfs_iunlock(ip, XFS_ILOCK_SHARED);
345 return -XFS_ERROR(error);
347 if (type == IO_DELALLOC &&
348 (!nimaps || isnullstartblock(imap->br_startblock))) {
349 error = xfs_iomap_write_allocate(ip, offset, count, imap);
351 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
352 return -XFS_ERROR(error);
356 if (type == IO_UNWRITTEN) {
358 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
359 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
363 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
370 struct xfs_bmbt_irec *imap,
373 offset >>= inode->i_blkbits;
375 return offset >= imap->br_startoff &&
376 offset < imap->br_startoff + imap->br_blockcount;
380 * BIO completion handler for buffered IO.
387 xfs_ioend_t *ioend = bio->bi_private;
389 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
390 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
392 /* Toss bio and pass work off to an xfsdatad thread */
393 bio->bi_private = NULL;
394 bio->bi_end_io = NULL;
397 xfs_finish_ioend(ioend);
401 xfs_submit_ioend_bio(
402 struct writeback_control *wbc,
406 atomic_inc(&ioend->io_remaining);
407 bio->bi_private = ioend;
408 bio->bi_end_io = xfs_end_bio;
411 * If the I/O is beyond EOF we mark the inode dirty immediately
412 * but don't update the inode size until I/O completion.
414 if (xfs_ioend_new_eof(ioend))
415 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
417 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
422 struct buffer_head *bh)
424 int nvecs = bio_get_nr_vecs(bh->b_bdev);
425 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
427 ASSERT(bio->bi_private == NULL);
428 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
429 bio->bi_bdev = bh->b_bdev;
434 xfs_start_buffer_writeback(
435 struct buffer_head *bh)
437 ASSERT(buffer_mapped(bh));
438 ASSERT(buffer_locked(bh));
439 ASSERT(!buffer_delay(bh));
440 ASSERT(!buffer_unwritten(bh));
442 mark_buffer_async_write(bh);
443 set_buffer_uptodate(bh);
444 clear_buffer_dirty(bh);
448 xfs_start_page_writeback(
453 ASSERT(PageLocked(page));
454 ASSERT(!PageWriteback(page));
456 clear_page_dirty_for_io(page);
457 set_page_writeback(page);
459 /* If no buffers on the page are to be written, finish it here */
461 end_page_writeback(page);
464 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
466 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
470 * Submit all of the bios for all of the ioends we have saved up, covering the
471 * initial writepage page and also any probed pages.
473 * Because we may have multiple ioends spanning a page, we need to start
474 * writeback on all the buffers before we submit them for I/O. If we mark the
475 * buffers as we got, then we can end up with a page that only has buffers
476 * marked async write and I/O complete on can occur before we mark the other
477 * buffers async write.
479 * The end result of this is that we trip a bug in end_page_writeback() because
480 * we call it twice for the one page as the code in end_buffer_async_write()
481 * assumes that all buffers on the page are started at the same time.
483 * The fix is two passes across the ioend list - one to start writeback on the
484 * buffer_heads, and then submit them for I/O on the second pass.
488 struct writeback_control *wbc,
491 xfs_ioend_t *head = ioend;
493 struct buffer_head *bh;
495 sector_t lastblock = 0;
497 /* Pass 1 - start writeback */
499 next = ioend->io_list;
500 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
501 xfs_start_buffer_writeback(bh);
502 } while ((ioend = next) != NULL);
504 /* Pass 2 - submit I/O */
507 next = ioend->io_list;
510 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
514 bio = xfs_alloc_ioend_bio(bh);
515 } else if (bh->b_blocknr != lastblock + 1) {
516 xfs_submit_ioend_bio(wbc, ioend, bio);
520 if (bio_add_buffer(bio, bh) != bh->b_size) {
521 xfs_submit_ioend_bio(wbc, ioend, bio);
525 lastblock = bh->b_blocknr;
528 xfs_submit_ioend_bio(wbc, ioend, bio);
529 xfs_finish_ioend(ioend);
530 } while ((ioend = next) != NULL);
534 * Cancel submission of all buffer_heads so far in this endio.
535 * Toss the endio too. Only ever called for the initial page
536 * in a writepage request, so only ever one page.
543 struct buffer_head *bh, *next_bh;
546 next = ioend->io_list;
547 bh = ioend->io_buffer_head;
549 next_bh = bh->b_private;
550 clear_buffer_async_write(bh);
552 } while ((bh = next_bh) != NULL);
554 xfs_ioend_wake(XFS_I(ioend->io_inode));
555 mempool_free(ioend, xfs_ioend_pool);
556 } while ((ioend = next) != NULL);
560 * Test to see if we've been building up a completion structure for
561 * earlier buffers -- if so, we try to append to this ioend if we
562 * can, otherwise we finish off any current ioend and start another.
563 * Return true if we've finished the given ioend.
568 struct buffer_head *bh,
571 xfs_ioend_t **result,
574 xfs_ioend_t *ioend = *result;
576 if (!ioend || need_ioend || type != ioend->io_type) {
577 xfs_ioend_t *previous = *result;
579 ioend = xfs_alloc_ioend(inode, type);
580 ioend->io_offset = offset;
581 ioend->io_buffer_head = bh;
582 ioend->io_buffer_tail = bh;
584 previous->io_list = ioend;
587 ioend->io_buffer_tail->b_private = bh;
588 ioend->io_buffer_tail = bh;
591 bh->b_private = NULL;
592 ioend->io_size += bh->b_size;
598 struct buffer_head *bh,
599 struct xfs_bmbt_irec *imap,
603 struct xfs_mount *m = XFS_I(inode)->i_mount;
604 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
605 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
607 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
608 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
610 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
611 ((offset - iomap_offset) >> inode->i_blkbits);
613 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
616 set_buffer_mapped(bh);
622 struct buffer_head *bh,
623 struct xfs_bmbt_irec *imap,
626 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
627 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
629 xfs_map_buffer(inode, bh, imap, offset);
630 set_buffer_mapped(bh);
631 clear_buffer_delay(bh);
632 clear_buffer_unwritten(bh);
636 * Test if a given page is suitable for writing as part of an unwritten
637 * or delayed allocate extent.
644 if (PageWriteback(page))
647 if (page->mapping && page_has_buffers(page)) {
648 struct buffer_head *bh, *head;
651 bh = head = page_buffers(page);
653 if (buffer_unwritten(bh))
654 acceptable = (type == IO_UNWRITTEN);
655 else if (buffer_delay(bh))
656 acceptable = (type == IO_DELALLOC);
657 else if (buffer_dirty(bh) && buffer_mapped(bh))
658 acceptable = (type == IO_OVERWRITE);
661 } while ((bh = bh->b_this_page) != head);
671 * Allocate & map buffers for page given the extent map. Write it out.
672 * except for the original page of a writepage, this is called on
673 * delalloc/unwritten pages only, for the original page it is possible
674 * that the page has no mapping at all.
681 struct xfs_bmbt_irec *imap,
682 xfs_ioend_t **ioendp,
683 struct writeback_control *wbc)
685 struct buffer_head *bh, *head;
686 xfs_off_t end_offset;
687 unsigned long p_offset;
690 int count = 0, done = 0, uptodate = 1;
691 xfs_off_t offset = page_offset(page);
693 if (page->index != tindex)
695 if (!trylock_page(page))
697 if (PageWriteback(page))
698 goto fail_unlock_page;
699 if (page->mapping != inode->i_mapping)
700 goto fail_unlock_page;
701 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
702 goto fail_unlock_page;
705 * page_dirty is initially a count of buffers on the page before
706 * EOF and is decremented as we move each into a cleanable state.
710 * End offset is the highest offset that this page should represent.
711 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
712 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
713 * hence give us the correct page_dirty count. On any other page,
714 * it will be zero and in that case we need page_dirty to be the
715 * count of buffers on the page.
717 end_offset = min_t(unsigned long long,
718 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
721 len = 1 << inode->i_blkbits;
722 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
724 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
725 page_dirty = p_offset / len;
727 bh = head = page_buffers(page);
729 if (offset >= end_offset)
731 if (!buffer_uptodate(bh))
733 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
738 if (buffer_unwritten(bh) || buffer_delay(bh) ||
740 if (buffer_unwritten(bh))
742 else if (buffer_delay(bh))
747 if (!xfs_imap_valid(inode, imap, offset)) {
753 if (type != IO_OVERWRITE)
754 xfs_map_at_offset(inode, bh, imap, offset);
755 xfs_add_to_ioend(inode, bh, offset, type,
763 } while (offset += len, (bh = bh->b_this_page) != head);
765 if (uptodate && bh == head)
766 SetPageUptodate(page);
769 if (--wbc->nr_to_write <= 0 &&
770 wbc->sync_mode == WB_SYNC_NONE)
773 xfs_start_page_writeback(page, !page_dirty, count);
783 * Convert & write out a cluster of pages in the same extent as defined
784 * by mp and following the start page.
790 struct xfs_bmbt_irec *imap,
791 xfs_ioend_t **ioendp,
792 struct writeback_control *wbc,
798 pagevec_init(&pvec, 0);
799 while (!done && tindex <= tlast) {
800 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
802 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
805 for (i = 0; i < pagevec_count(&pvec); i++) {
806 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
812 pagevec_release(&pvec);
818 xfs_vm_invalidatepage(
820 unsigned long offset)
822 trace_xfs_invalidatepage(page->mapping->host, page, offset);
823 block_invalidatepage(page, offset);
827 * If the page has delalloc buffers on it, we need to punch them out before we
828 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
829 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
830 * is done on that same region - the delalloc extent is returned when none is
831 * supposed to be there.
833 * We prevent this by truncating away the delalloc regions on the page before
834 * invalidating it. Because they are delalloc, we can do this without needing a
835 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
836 * truncation without a transaction as there is no space left for block
837 * reservation (typically why we see a ENOSPC in writeback).
839 * This is not a performance critical path, so for now just do the punching a
840 * buffer head at a time.
843 xfs_aops_discard_page(
846 struct inode *inode = page->mapping->host;
847 struct xfs_inode *ip = XFS_I(inode);
848 struct buffer_head *bh, *head;
849 loff_t offset = page_offset(page);
851 if (!xfs_is_delayed_page(page, IO_DELALLOC))
854 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
857 xfs_alert(ip->i_mount,
858 "page discard on page %p, inode 0x%llx, offset %llu.",
859 page, ip->i_ino, offset);
861 xfs_ilock(ip, XFS_ILOCK_EXCL);
862 bh = head = page_buffers(page);
865 xfs_fileoff_t start_fsb;
867 if (!buffer_delay(bh))
870 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
871 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
873 /* something screwed, just bail */
874 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
875 xfs_alert(ip->i_mount,
876 "page discard unable to remove delalloc mapping.");
881 offset += 1 << inode->i_blkbits;
883 } while ((bh = bh->b_this_page) != head);
885 xfs_iunlock(ip, XFS_ILOCK_EXCL);
887 xfs_vm_invalidatepage(page, 0);
892 * Write out a dirty page.
894 * For delalloc space on the page we need to allocate space and flush it.
895 * For unwritten space on the page we need to start the conversion to
896 * regular allocated space.
897 * For any other dirty buffer heads on the page we should flush them.
902 struct writeback_control *wbc)
904 struct inode *inode = page->mapping->host;
905 struct buffer_head *bh, *head;
906 struct xfs_bmbt_irec imap;
907 xfs_ioend_t *ioend = NULL, *iohead = NULL;
910 __uint64_t end_offset;
911 pgoff_t end_index, last_index;
913 int err, imap_valid = 0, uptodate = 1;
917 trace_xfs_writepage(inode, page, 0);
919 ASSERT(page_has_buffers(page));
922 * Refuse to write the page out if we are called from reclaim context.
924 * This avoids stack overflows when called from deeply used stacks in
925 * random callers for direct reclaim or memcg reclaim. We explicitly
926 * allow reclaim from kswapd as the stack usage there is relatively low.
928 * This should really be done by the core VM, but until that happens
929 * filesystems like XFS, btrfs and ext4 have to take care of this
932 if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC)
936 * Given that we do not allow direct reclaim to call us, we should
937 * never be called while in a filesystem transaction.
939 if (WARN_ON(current->flags & PF_FSTRANS))
942 /* Is this page beyond the end of the file? */
943 offset = i_size_read(inode);
944 end_index = offset >> PAGE_CACHE_SHIFT;
945 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
946 if (page->index >= end_index) {
947 if ((page->index >= end_index + 1) ||
948 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
954 end_offset = min_t(unsigned long long,
955 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
957 len = 1 << inode->i_blkbits;
959 bh = head = page_buffers(page);
960 offset = page_offset(page);
963 if (wbc->sync_mode == WB_SYNC_NONE)
969 if (offset >= end_offset)
971 if (!buffer_uptodate(bh))
975 * set_page_dirty dirties all buffers in a page, independent
976 * of their state. The dirty state however is entirely
977 * meaningless for holes (!mapped && uptodate), so skip
978 * buffers covering holes here.
980 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
985 if (buffer_unwritten(bh)) {
986 if (type != IO_UNWRITTEN) {
990 } else if (buffer_delay(bh)) {
991 if (type != IO_DELALLOC) {
995 } else if (buffer_uptodate(bh)) {
996 if (type != IO_OVERWRITE) {
1001 if (PageUptodate(page)) {
1002 ASSERT(buffer_mapped(bh));
1009 imap_valid = xfs_imap_valid(inode, &imap, offset);
1012 * If we didn't have a valid mapping then we need to
1013 * put the new mapping into a separate ioend structure.
1014 * This ensures non-contiguous extents always have
1015 * separate ioends, which is particularly important
1016 * for unwritten extent conversion at I/O completion
1020 err = xfs_map_blocks(inode, offset, &imap, type,
1024 imap_valid = xfs_imap_valid(inode, &imap, offset);
1028 if (type != IO_OVERWRITE)
1029 xfs_map_at_offset(inode, bh, &imap, offset);
1030 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1038 } while (offset += len, ((bh = bh->b_this_page) != head));
1040 if (uptodate && bh == head)
1041 SetPageUptodate(page);
1043 xfs_start_page_writeback(page, 1, count);
1045 if (ioend && imap_valid) {
1046 xfs_off_t end_index;
1048 end_index = imap.br_startoff + imap.br_blockcount;
1051 end_index <<= inode->i_blkbits;
1054 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1056 /* check against file size */
1057 if (end_index > last_index)
1058 end_index = last_index;
1060 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1065 xfs_submit_ioend(wbc, iohead);
1071 xfs_cancel_ioend(iohead);
1076 xfs_aops_discard_page(page);
1077 ClearPageUptodate(page);
1082 redirty_page_for_writepage(wbc, page);
1089 struct address_space *mapping,
1090 struct writeback_control *wbc)
1092 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1093 return generic_writepages(mapping, wbc);
1097 * Called to move a page into cleanable state - and from there
1098 * to be released. The page should already be clean. We always
1099 * have buffer heads in this call.
1101 * Returns 1 if the page is ok to release, 0 otherwise.
1108 int delalloc, unwritten;
1110 trace_xfs_releasepage(page->mapping->host, page, 0);
1112 xfs_count_page_state(page, &delalloc, &unwritten);
1114 if (WARN_ON(delalloc))
1116 if (WARN_ON(unwritten))
1119 return try_to_free_buffers(page);
1124 struct inode *inode,
1126 struct buffer_head *bh_result,
1130 struct xfs_inode *ip = XFS_I(inode);
1131 struct xfs_mount *mp = ip->i_mount;
1132 xfs_fileoff_t offset_fsb, end_fsb;
1135 struct xfs_bmbt_irec imap;
1141 if (XFS_FORCED_SHUTDOWN(mp))
1142 return -XFS_ERROR(EIO);
1144 offset = (xfs_off_t)iblock << inode->i_blkbits;
1145 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1146 size = bh_result->b_size;
1148 if (!create && direct && offset >= i_size_read(inode))
1152 lockmode = XFS_ILOCK_EXCL;
1153 xfs_ilock(ip, lockmode);
1155 lockmode = xfs_ilock_map_shared(ip);
1158 ASSERT(offset <= mp->m_maxioffset);
1159 if (offset + size > mp->m_maxioffset)
1160 size = mp->m_maxioffset - offset;
1161 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1162 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1164 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
1165 XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL);
1171 (imap.br_startblock == HOLESTARTBLOCK ||
1172 imap.br_startblock == DELAYSTARTBLOCK))) {
1174 error = xfs_iomap_write_direct(ip, offset, size,
1177 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1182 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1183 } else if (nimaps) {
1184 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1186 trace_xfs_get_blocks_notfound(ip, offset, size);
1189 xfs_iunlock(ip, lockmode);
1191 if (imap.br_startblock != HOLESTARTBLOCK &&
1192 imap.br_startblock != DELAYSTARTBLOCK) {
1194 * For unwritten extents do not report a disk address on
1195 * the read case (treat as if we're reading into a hole).
1197 if (create || !ISUNWRITTEN(&imap))
1198 xfs_map_buffer(inode, bh_result, &imap, offset);
1199 if (create && ISUNWRITTEN(&imap)) {
1201 bh_result->b_private = inode;
1202 set_buffer_unwritten(bh_result);
1207 * If this is a realtime file, data may be on a different device.
1208 * to that pointed to from the buffer_head b_bdev currently.
1210 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1213 * If we previously allocated a block out beyond eof and we are now
1214 * coming back to use it then we will need to flag it as new even if it
1215 * has a disk address.
1217 * With sub-block writes into unwritten extents we also need to mark
1218 * the buffer as new so that the unwritten parts of the buffer gets
1222 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1223 (offset >= i_size_read(inode)) ||
1224 (new || ISUNWRITTEN(&imap))))
1225 set_buffer_new(bh_result);
1227 if (imap.br_startblock == DELAYSTARTBLOCK) {
1230 set_buffer_uptodate(bh_result);
1231 set_buffer_mapped(bh_result);
1232 set_buffer_delay(bh_result);
1237 * If this is O_DIRECT or the mpage code calling tell them how large
1238 * the mapping is, so that we can avoid repeated get_blocks calls.
1240 if (direct || size > (1 << inode->i_blkbits)) {
1241 xfs_off_t mapping_size;
1243 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1244 mapping_size <<= inode->i_blkbits;
1246 ASSERT(mapping_size > 0);
1247 if (mapping_size > size)
1248 mapping_size = size;
1249 if (mapping_size > LONG_MAX)
1250 mapping_size = LONG_MAX;
1252 bh_result->b_size = mapping_size;
1258 xfs_iunlock(ip, lockmode);
1264 struct inode *inode,
1266 struct buffer_head *bh_result,
1269 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1273 xfs_get_blocks_direct(
1274 struct inode *inode,
1276 struct buffer_head *bh_result,
1279 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1283 * Complete a direct I/O write request.
1285 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1286 * need to issue a transaction to convert the range from unwritten to written
1287 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1288 * to do this and we are done. But in case this was a successful AIO
1289 * request this handler is called from interrupt context, from which we
1290 * can't start transactions. In that case offload the I/O completion to
1291 * the workqueues we also use for buffered I/O completion.
1294 xfs_end_io_direct_write(
1302 struct xfs_ioend *ioend = iocb->private;
1303 struct inode *inode = ioend->io_inode;
1306 * blockdev_direct_IO can return an error even after the I/O
1307 * completion handler was called. Thus we need to protect
1308 * against double-freeing.
1310 iocb->private = NULL;
1312 ioend->io_offset = offset;
1313 ioend->io_size = size;
1314 if (private && size > 0)
1315 ioend->io_type = IO_UNWRITTEN;
1319 * If we are converting an unwritten extent we need to delay
1320 * the AIO completion until after the unwrittent extent
1321 * conversion has completed, otherwise do it ASAP.
1323 if (ioend->io_type == IO_UNWRITTEN) {
1324 ioend->io_iocb = iocb;
1325 ioend->io_result = ret;
1327 aio_complete(iocb, ret, 0);
1329 xfs_finish_ioend(ioend);
1331 xfs_finish_ioend_sync(ioend);
1334 /* XXX: probably should move into the real I/O completion handler */
1335 inode_dio_done(inode);
1342 const struct iovec *iov,
1344 unsigned long nr_segs)
1346 struct inode *inode = iocb->ki_filp->f_mapping->host;
1347 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1351 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1353 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1355 xfs_get_blocks_direct,
1356 xfs_end_io_direct_write, NULL, 0);
1357 if (ret != -EIOCBQUEUED && iocb->private)
1358 xfs_destroy_ioend(iocb->private);
1360 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1362 xfs_get_blocks_direct,
1370 xfs_vm_write_failed(
1371 struct address_space *mapping,
1374 struct inode *inode = mapping->host;
1376 if (to > inode->i_size) {
1378 * punch out the delalloc blocks we have already allocated. We
1379 * don't call xfs_setattr() to do this as we may be in the
1380 * middle of a multi-iovec write and so the vfs inode->i_size
1381 * will not match the xfs ip->i_size and so it will zero too
1382 * much. Hence we jus truncate the page cache to zero what is
1383 * necessary and punch the delalloc blocks directly.
1385 struct xfs_inode *ip = XFS_I(inode);
1386 xfs_fileoff_t start_fsb;
1387 xfs_fileoff_t end_fsb;
1390 truncate_pagecache(inode, to, inode->i_size);
1393 * Check if there are any blocks that are outside of i_size
1394 * that need to be trimmed back.
1396 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1397 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1398 if (end_fsb <= start_fsb)
1401 xfs_ilock(ip, XFS_ILOCK_EXCL);
1402 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1403 end_fsb - start_fsb);
1405 /* something screwed, just bail */
1406 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1407 xfs_alert(ip->i_mount,
1408 "xfs_vm_write_failed: unable to clean up ino %lld",
1412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1419 struct address_space *mapping,
1423 struct page **pagep,
1428 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1429 pagep, xfs_get_blocks);
1431 xfs_vm_write_failed(mapping, pos + len);
1438 struct address_space *mapping,
1447 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1448 if (unlikely(ret < len))
1449 xfs_vm_write_failed(mapping, pos + len);
1455 struct address_space *mapping,
1458 struct inode *inode = (struct inode *)mapping->host;
1459 struct xfs_inode *ip = XFS_I(inode);
1461 trace_xfs_vm_bmap(XFS_I(inode));
1462 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1463 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1464 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1465 return generic_block_bmap(mapping, block, xfs_get_blocks);
1470 struct file *unused,
1473 return mpage_readpage(page, xfs_get_blocks);
1478 struct file *unused,
1479 struct address_space *mapping,
1480 struct list_head *pages,
1483 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1486 const struct address_space_operations xfs_address_space_operations = {
1487 .readpage = xfs_vm_readpage,
1488 .readpages = xfs_vm_readpages,
1489 .writepage = xfs_vm_writepage,
1490 .writepages = xfs_vm_writepages,
1491 .releasepage = xfs_vm_releasepage,
1492 .invalidatepage = xfs_vm_invalidatepage,
1493 .write_begin = xfs_vm_write_begin,
1494 .write_end = xfs_vm_write_end,
1495 .bmap = xfs_vm_bmap,
1496 .direct_IO = xfs_vm_direct_IO,
1497 .migratepage = buffer_migrate_page,
1498 .is_partially_uptodate = block_is_partially_uptodate,
1499 .error_remove_page = generic_error_remove_page,