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
25 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_error.h"
39 #include "xfs_iomap.h"
40 #include "xfs_vnodeops.h"
41 #include <linux/mpage.h>
42 #include <linux/pagevec.h>
43 #include <linux/writeback.h>
52 struct buffer_head *bh, *head;
54 *delalloc = *unmapped = *unwritten = 0;
56 bh = head = page_buffers(page);
58 if (buffer_uptodate(bh) && !buffer_mapped(bh))
60 else if (buffer_unwritten(bh))
62 else if (buffer_delay(bh))
64 } while ((bh = bh->b_this_page) != head);
67 #if defined(XFS_RW_TRACE)
76 bhv_vnode_t *vp = vn_from_inode(inode);
77 loff_t isize = i_size_read(inode);
78 loff_t offset = page_offset(page);
79 int delalloc = -1, unmapped = -1, unwritten = -1;
81 if (page_has_buffers(page))
82 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
88 ktrace_enter(ip->i_rwtrace,
89 (void *)((unsigned long)tag),
94 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
95 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
96 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
97 (void *)((unsigned long)(isize & 0xffffffff)),
98 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
99 (void *)((unsigned long)(offset & 0xffffffff)),
100 (void *)((unsigned long)delalloc),
101 (void *)((unsigned long)unmapped),
102 (void *)((unsigned long)unwritten),
103 (void *)((unsigned long)current_pid()),
107 #define xfs_page_trace(tag, inode, page, pgoff)
111 * Schedule IO completion handling on a xfsdatad if this was
112 * the final hold on this ioend. If we are asked to wait,
113 * flush the workqueue.
120 if (atomic_dec_and_test(&ioend->io_remaining)) {
121 queue_work(xfsdatad_workqueue, &ioend->io_work);
123 flush_workqueue(xfsdatad_workqueue);
128 * We're now finished for good with this ioend structure.
129 * Update the page state via the associated buffer_heads,
130 * release holds on the inode and bio, and finally free
131 * up memory. Do not use the ioend after this.
137 struct buffer_head *bh, *next;
139 for (bh = ioend->io_buffer_head; bh; bh = next) {
140 next = bh->b_private;
141 bh->b_end_io(bh, !ioend->io_error);
143 if (unlikely(ioend->io_error)) {
144 vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
147 vn_iowake(XFS_I(ioend->io_inode));
148 mempool_free(ioend, xfs_ioend_pool);
152 * Update on-disk file size now that data has been written to disk.
153 * The current in-memory file size is i_size. If a write is beyond
154 * eof io_new_size will be the intended file size until i_size is
155 * updated. If this write does not extend all the way to the valid
156 * file size then restrict this update to the end of the write.
162 xfs_inode_t *ip = XFS_I(ioend->io_inode);
166 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
167 ASSERT(ioend->io_type != IOMAP_READ);
169 if (unlikely(ioend->io_error))
172 bsize = ioend->io_offset + ioend->io_size;
174 xfs_ilock(ip, XFS_ILOCK_EXCL);
176 isize = MAX(ip->i_size, ip->i_iocore.io_new_size);
177 isize = MIN(isize, bsize);
179 if (ip->i_d.di_size < isize) {
180 ip->i_d.di_size = isize;
181 ip->i_update_core = 1;
182 ip->i_update_size = 1;
183 mark_inode_dirty_sync(ioend->io_inode);
186 xfs_iunlock(ip, XFS_ILOCK_EXCL);
190 * Buffered IO write completion for delayed allocate extents.
193 xfs_end_bio_delalloc(
194 struct work_struct *work)
197 container_of(work, xfs_ioend_t, io_work);
199 xfs_setfilesize(ioend);
200 xfs_destroy_ioend(ioend);
204 * Buffered IO write completion for regular, written extents.
208 struct work_struct *work)
211 container_of(work, xfs_ioend_t, io_work);
213 xfs_setfilesize(ioend);
214 xfs_destroy_ioend(ioend);
218 * IO write completion for unwritten extents.
220 * Issue transactions to convert a buffer range from unwritten
221 * to written extents.
224 xfs_end_bio_unwritten(
225 struct work_struct *work)
228 container_of(work, xfs_ioend_t, io_work);
229 xfs_off_t offset = ioend->io_offset;
230 size_t size = ioend->io_size;
232 if (likely(!ioend->io_error)) {
233 xfs_bmap(XFS_I(ioend->io_inode), offset, size,
234 BMAPI_UNWRITTEN, NULL, NULL);
235 xfs_setfilesize(ioend);
237 xfs_destroy_ioend(ioend);
241 * IO read completion for regular, written extents.
245 struct work_struct *work)
248 container_of(work, xfs_ioend_t, io_work);
250 xfs_destroy_ioend(ioend);
254 * Allocate and initialise an IO completion structure.
255 * We need to track unwritten extent write completion here initially.
256 * We'll need to extend this for updating the ondisk inode size later
266 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
269 * Set the count to 1 initially, which will prevent an I/O
270 * completion callback from happening before we have started
271 * all the I/O from calling the completion routine too early.
273 atomic_set(&ioend->io_remaining, 1);
275 ioend->io_list = NULL;
276 ioend->io_type = type;
277 ioend->io_inode = inode;
278 ioend->io_buffer_head = NULL;
279 ioend->io_buffer_tail = NULL;
280 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
281 ioend->io_offset = 0;
284 if (type == IOMAP_UNWRITTEN)
285 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
286 else if (type == IOMAP_DELAY)
287 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
288 else if (type == IOMAP_READ)
289 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
291 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
304 xfs_inode_t *ip = XFS_I(inode);
305 int error, nmaps = 1;
307 error = xfs_bmap(ip, offset, count,
308 flags, mapp, &nmaps);
309 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
310 xfs_iflags_set(ip, XFS_IMODIFIED);
319 return offset >= iomapp->iomap_offset &&
320 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
324 * BIO completion handler for buffered IO.
331 xfs_ioend_t *ioend = bio->bi_private;
333 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
334 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
336 /* Toss bio and pass work off to an xfsdatad thread */
337 bio->bi_private = NULL;
338 bio->bi_end_io = NULL;
341 xfs_finish_ioend(ioend, 0);
345 xfs_submit_ioend_bio(
349 atomic_inc(&ioend->io_remaining);
351 bio->bi_private = ioend;
352 bio->bi_end_io = xfs_end_bio;
354 submit_bio(WRITE, bio);
355 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
361 struct buffer_head *bh)
364 int nvecs = bio_get_nr_vecs(bh->b_bdev);
367 bio = bio_alloc(GFP_NOIO, nvecs);
371 ASSERT(bio->bi_private == NULL);
372 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
373 bio->bi_bdev = bh->b_bdev;
379 xfs_start_buffer_writeback(
380 struct buffer_head *bh)
382 ASSERT(buffer_mapped(bh));
383 ASSERT(buffer_locked(bh));
384 ASSERT(!buffer_delay(bh));
385 ASSERT(!buffer_unwritten(bh));
387 mark_buffer_async_write(bh);
388 set_buffer_uptodate(bh);
389 clear_buffer_dirty(bh);
393 xfs_start_page_writeback(
395 struct writeback_control *wbc,
399 ASSERT(PageLocked(page));
400 ASSERT(!PageWriteback(page));
402 clear_page_dirty_for_io(page);
403 set_page_writeback(page);
405 /* If no buffers on the page are to be written, finish it here */
407 end_page_writeback(page);
410 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
412 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
416 * Submit all of the bios for all of the ioends we have saved up, covering the
417 * initial writepage page and also any probed pages.
419 * Because we may have multiple ioends spanning a page, we need to start
420 * writeback on all the buffers before we submit them for I/O. If we mark the
421 * buffers as we got, then we can end up with a page that only has buffers
422 * marked async write and I/O complete on can occur before we mark the other
423 * buffers async write.
425 * The end result of this is that we trip a bug in end_page_writeback() because
426 * we call it twice for the one page as the code in end_buffer_async_write()
427 * assumes that all buffers on the page are started at the same time.
429 * The fix is two passes across the ioend list - one to start writeback on the
430 * buffer_heads, and then submit them for I/O on the second pass.
436 xfs_ioend_t *head = ioend;
438 struct buffer_head *bh;
440 sector_t lastblock = 0;
442 /* Pass 1 - start writeback */
444 next = ioend->io_list;
445 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
446 xfs_start_buffer_writeback(bh);
448 } while ((ioend = next) != NULL);
450 /* Pass 2 - submit I/O */
453 next = ioend->io_list;
456 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
460 bio = xfs_alloc_ioend_bio(bh);
461 } else if (bh->b_blocknr != lastblock + 1) {
462 xfs_submit_ioend_bio(ioend, bio);
466 if (bio_add_buffer(bio, bh) != bh->b_size) {
467 xfs_submit_ioend_bio(ioend, bio);
471 lastblock = bh->b_blocknr;
474 xfs_submit_ioend_bio(ioend, bio);
475 xfs_finish_ioend(ioend, 0);
476 } while ((ioend = next) != NULL);
480 * Cancel submission of all buffer_heads so far in this endio.
481 * Toss the endio too. Only ever called for the initial page
482 * in a writepage request, so only ever one page.
489 struct buffer_head *bh, *next_bh;
492 next = ioend->io_list;
493 bh = ioend->io_buffer_head;
495 next_bh = bh->b_private;
496 clear_buffer_async_write(bh);
498 } while ((bh = next_bh) != NULL);
500 vn_iowake(XFS_I(ioend->io_inode));
501 mempool_free(ioend, xfs_ioend_pool);
502 } while ((ioend = next) != NULL);
506 * Test to see if we've been building up a completion structure for
507 * earlier buffers -- if so, we try to append to this ioend if we
508 * can, otherwise we finish off any current ioend and start another.
509 * Return true if we've finished the given ioend.
514 struct buffer_head *bh,
517 xfs_ioend_t **result,
520 xfs_ioend_t *ioend = *result;
522 if (!ioend || need_ioend || type != ioend->io_type) {
523 xfs_ioend_t *previous = *result;
525 ioend = xfs_alloc_ioend(inode, type);
526 ioend->io_offset = offset;
527 ioend->io_buffer_head = bh;
528 ioend->io_buffer_tail = bh;
530 previous->io_list = ioend;
533 ioend->io_buffer_tail->b_private = bh;
534 ioend->io_buffer_tail = bh;
537 bh->b_private = NULL;
538 ioend->io_size += bh->b_size;
543 struct buffer_head *bh,
550 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
552 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
553 ((offset - mp->iomap_offset) >> block_bits);
555 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
558 set_buffer_mapped(bh);
563 struct buffer_head *bh,
568 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
569 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
572 xfs_map_buffer(bh, iomapp, offset, block_bits);
573 bh->b_bdev = iomapp->iomap_target->bt_bdev;
574 set_buffer_mapped(bh);
575 clear_buffer_delay(bh);
576 clear_buffer_unwritten(bh);
580 * Look for a page at index that is suitable for clustering.
585 unsigned int pg_offset,
590 if (PageWriteback(page))
593 if (page->mapping && PageDirty(page)) {
594 if (page_has_buffers(page)) {
595 struct buffer_head *bh, *head;
597 bh = head = page_buffers(page);
599 if (!buffer_uptodate(bh))
601 if (mapped != buffer_mapped(bh))
604 if (ret >= pg_offset)
606 } while ((bh = bh->b_this_page) != head);
608 ret = mapped ? 0 : PAGE_CACHE_SIZE;
617 struct page *startpage,
618 struct buffer_head *bh,
619 struct buffer_head *head,
623 pgoff_t tindex, tlast, tloff;
627 /* First sum forwards in this page */
629 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
632 } while ((bh = bh->b_this_page) != head);
634 /* if we reached the end of the page, sum forwards in following pages */
635 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
636 tindex = startpage->index + 1;
638 /* Prune this back to avoid pathological behavior */
639 tloff = min(tlast, startpage->index + 64);
641 pagevec_init(&pvec, 0);
642 while (!done && tindex <= tloff) {
643 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
645 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
648 for (i = 0; i < pagevec_count(&pvec); i++) {
649 struct page *page = pvec.pages[i];
650 size_t pg_offset, pg_len = 0;
652 if (tindex == tlast) {
654 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
660 pg_offset = PAGE_CACHE_SIZE;
662 if (page->index == tindex && !TestSetPageLocked(page)) {
663 pg_len = xfs_probe_page(page, pg_offset, mapped);
676 pagevec_release(&pvec);
684 * Test if a given page is suitable for writing as part of an unwritten
685 * or delayed allocate extent.
692 if (PageWriteback(page))
695 if (page->mapping && page_has_buffers(page)) {
696 struct buffer_head *bh, *head;
699 bh = head = page_buffers(page);
701 if (buffer_unwritten(bh))
702 acceptable = (type == IOMAP_UNWRITTEN);
703 else if (buffer_delay(bh))
704 acceptable = (type == IOMAP_DELAY);
705 else if (buffer_dirty(bh) && buffer_mapped(bh))
706 acceptable = (type == IOMAP_NEW);
709 } while ((bh = bh->b_this_page) != head);
719 * Allocate & map buffers for page given the extent map. Write it out.
720 * except for the original page of a writepage, this is called on
721 * delalloc/unwritten pages only, for the original page it is possible
722 * that the page has no mapping at all.
730 xfs_ioend_t **ioendp,
731 struct writeback_control *wbc,
735 struct buffer_head *bh, *head;
736 xfs_off_t end_offset;
737 unsigned long p_offset;
739 int bbits = inode->i_blkbits;
741 int count = 0, done = 0, uptodate = 1;
742 xfs_off_t offset = page_offset(page);
744 if (page->index != tindex)
746 if (TestSetPageLocked(page))
748 if (PageWriteback(page))
749 goto fail_unlock_page;
750 if (page->mapping != inode->i_mapping)
751 goto fail_unlock_page;
752 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
753 goto fail_unlock_page;
756 * page_dirty is initially a count of buffers on the page before
757 * EOF and is decremented as we move each into a cleanable state.
761 * End offset is the highest offset that this page should represent.
762 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
763 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
764 * hence give us the correct page_dirty count. On any other page,
765 * it will be zero and in that case we need page_dirty to be the
766 * count of buffers on the page.
768 end_offset = min_t(unsigned long long,
769 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
772 len = 1 << inode->i_blkbits;
773 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
775 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
776 page_dirty = p_offset / len;
778 bh = head = page_buffers(page);
780 if (offset >= end_offset)
782 if (!buffer_uptodate(bh))
784 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
789 if (buffer_unwritten(bh) || buffer_delay(bh)) {
790 if (buffer_unwritten(bh))
791 type = IOMAP_UNWRITTEN;
795 if (!xfs_iomap_valid(mp, offset)) {
800 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
801 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
803 xfs_map_at_offset(bh, offset, bbits, mp);
805 xfs_add_to_ioend(inode, bh, offset,
808 set_buffer_dirty(bh);
810 mark_buffer_dirty(bh);
816 if (buffer_mapped(bh) && all_bh && startio) {
818 xfs_add_to_ioend(inode, bh, offset,
826 } while (offset += len, (bh = bh->b_this_page) != head);
828 if (uptodate && bh == head)
829 SetPageUptodate(page);
833 struct backing_dev_info *bdi;
835 bdi = inode->i_mapping->backing_dev_info;
837 if (bdi_write_congested(bdi)) {
838 wbc->encountered_congestion = 1;
840 } else if (wbc->nr_to_write <= 0) {
844 xfs_start_page_writeback(page, wbc, !page_dirty, count);
855 * Convert & write out a cluster of pages in the same extent as defined
856 * by mp and following the start page.
863 xfs_ioend_t **ioendp,
864 struct writeback_control *wbc,
872 pagevec_init(&pvec, 0);
873 while (!done && tindex <= tlast) {
874 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
876 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
879 for (i = 0; i < pagevec_count(&pvec); i++) {
880 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
881 iomapp, ioendp, wbc, startio, all_bh);
886 pagevec_release(&pvec);
892 * Calling this without startio set means we are being asked to make a dirty
893 * page ready for freeing it's buffers. When called with startio set then
894 * we are coming from writepage.
896 * When called with startio set it is important that we write the WHOLE
898 * The bh->b_state's cannot know if any of the blocks or which block for
899 * that matter are dirty due to mmap writes, and therefore bh uptodate is
900 * only valid if the page itself isn't completely uptodate. Some layers
901 * may clear the page dirty flag prior to calling write page, under the
902 * assumption the entire page will be written out; by not writing out the
903 * whole page the page can be reused before all valid dirty data is
904 * written out. Note: in the case of a page that has been dirty'd by
905 * mapwrite and but partially setup by block_prepare_write the
906 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
907 * valid state, thus the whole page must be written out thing.
911 xfs_page_state_convert(
914 struct writeback_control *wbc,
916 int unmapped) /* also implies page uptodate */
918 struct buffer_head *bh, *head;
920 xfs_ioend_t *ioend = NULL, *iohead = NULL;
922 unsigned long p_offset = 0;
924 __uint64_t end_offset;
925 pgoff_t end_index, last_index, tlast;
927 int flags, err, iomap_valid = 0, uptodate = 1;
928 int page_dirty, count = 0;
930 int all_bh = unmapped;
933 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
934 trylock |= BMAPI_TRYLOCK;
937 /* Is this page beyond the end of the file? */
938 offset = i_size_read(inode);
939 end_index = offset >> PAGE_CACHE_SHIFT;
940 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
941 if (page->index >= end_index) {
942 if ((page->index >= end_index + 1) ||
943 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
951 * page_dirty is initially a count of buffers on the page before
952 * EOF and is decremented as we move each into a cleanable state.
956 * End offset is the highest offset that this page should represent.
957 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
958 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
959 * hence give us the correct page_dirty count. On any other page,
960 * it will be zero and in that case we need page_dirty to be the
961 * count of buffers on the page.
963 end_offset = min_t(unsigned long long,
964 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
965 len = 1 << inode->i_blkbits;
966 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
968 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
969 page_dirty = p_offset / len;
971 bh = head = page_buffers(page);
972 offset = page_offset(page);
976 /* TODO: cleanup count and page_dirty */
979 if (offset >= end_offset)
981 if (!buffer_uptodate(bh))
983 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
985 * the iomap is actually still valid, but the ioend
986 * isn't. shouldn't happen too often.
993 iomap_valid = xfs_iomap_valid(&iomap, offset);
996 * First case, map an unwritten extent and prepare for
997 * extent state conversion transaction on completion.
999 * Second case, allocate space for a delalloc buffer.
1000 * We can return EAGAIN here in the release page case.
1002 * Third case, an unmapped buffer was found, and we are
1003 * in a path where we need to write the whole page out.
1005 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1006 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1007 !buffer_mapped(bh) && (unmapped || startio))) {
1011 * Make sure we don't use a read-only iomap
1013 if (flags == BMAPI_READ)
1016 if (buffer_unwritten(bh)) {
1017 type = IOMAP_UNWRITTEN;
1018 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1019 } else if (buffer_delay(bh)) {
1021 flags = BMAPI_ALLOCATE | trylock;
1024 flags = BMAPI_WRITE | BMAPI_MMAP;
1029 * if we didn't have a valid mapping then we
1030 * need to ensure that we put the new mapping
1031 * in a new ioend structure. This needs to be
1032 * done to ensure that the ioends correctly
1033 * reflect the block mappings at io completion
1034 * for unwritten extent conversion.
1037 if (type == IOMAP_NEW) {
1038 size = xfs_probe_cluster(inode,
1044 err = xfs_map_blocks(inode, offset, size,
1048 iomap_valid = xfs_iomap_valid(&iomap, offset);
1051 xfs_map_at_offset(bh, offset,
1052 inode->i_blkbits, &iomap);
1054 xfs_add_to_ioend(inode, bh, offset,
1058 set_buffer_dirty(bh);
1060 mark_buffer_dirty(bh);
1065 } else if (buffer_uptodate(bh) && startio) {
1067 * we got here because the buffer is already mapped.
1068 * That means it must already have extents allocated
1069 * underneath it. Map the extent by reading it.
1071 if (!iomap_valid || flags != BMAPI_READ) {
1073 size = xfs_probe_cluster(inode, page, bh,
1075 err = xfs_map_blocks(inode, offset, size,
1079 iomap_valid = xfs_iomap_valid(&iomap, offset);
1083 * We set the type to IOMAP_NEW in case we are doing a
1084 * small write at EOF that is extending the file but
1085 * without needing an allocation. We need to update the
1086 * file size on I/O completion in this case so it is
1087 * the same case as having just allocated a new extent
1088 * that we are writing into for the first time.
1091 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1092 ASSERT(buffer_mapped(bh));
1095 xfs_add_to_ioend(inode, bh, offset, type,
1096 &ioend, !iomap_valid);
1102 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1103 (unmapped || startio)) {
1110 } while (offset += len, ((bh = bh->b_this_page) != head));
1112 if (uptodate && bh == head)
1113 SetPageUptodate(page);
1116 xfs_start_page_writeback(page, wbc, 1, count);
1118 if (ioend && iomap_valid) {
1119 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1121 tlast = min_t(pgoff_t, offset, last_index);
1122 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1123 wbc, startio, all_bh, tlast);
1127 xfs_submit_ioend(iohead);
1133 xfs_cancel_ioend(iohead);
1136 * If it's delalloc and we have nowhere to put it,
1137 * throw it away, unless the lower layers told
1140 if (err != -EAGAIN) {
1142 block_invalidatepage(page, 0);
1143 ClearPageUptodate(page);
1149 * writepage: Called from one of two places:
1151 * 1. we are flushing a delalloc buffer head.
1153 * 2. we are writing out a dirty page. Typically the page dirty
1154 * state is cleared before we get here. In this case is it
1155 * conceivable we have no buffer heads.
1157 * For delalloc space on the page we need to allocate space and
1158 * flush it. For unmapped buffer heads on the page we should
1159 * allocate space if the page is uptodate. For any other dirty
1160 * buffer heads on the page we should flush them.
1162 * If we detect that a transaction would be required to flush
1163 * the page, we have to check the process flags first, if we
1164 * are already in a transaction or disk I/O during allocations
1165 * is off, we need to fail the writepage and redirty the page.
1171 struct writeback_control *wbc)
1175 int delalloc, unmapped, unwritten;
1176 struct inode *inode = page->mapping->host;
1178 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1181 * We need a transaction if:
1182 * 1. There are delalloc buffers on the page
1183 * 2. The page is uptodate and we have unmapped buffers
1184 * 3. The page is uptodate and we have no buffers
1185 * 4. There are unwritten buffers on the page
1188 if (!page_has_buffers(page)) {
1192 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1193 if (!PageUptodate(page))
1195 need_trans = delalloc + unmapped + unwritten;
1199 * If we need a transaction and the process flags say
1200 * we are already in a transaction, or no IO is allowed
1201 * then mark the page dirty again and leave the page
1204 if (current_test_flags(PF_FSTRANS) && need_trans)
1208 * Delay hooking up buffer heads until we have
1209 * made our go/no-go decision.
1211 if (!page_has_buffers(page))
1212 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1215 * Convert delayed allocate, unwritten or unmapped space
1216 * to real space and flush out to disk.
1218 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1219 if (error == -EAGAIN)
1221 if (unlikely(error < 0))
1227 redirty_page_for_writepage(wbc, page);
1237 struct address_space *mapping,
1238 struct writeback_control *wbc)
1240 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1241 return generic_writepages(mapping, wbc);
1245 * Called to move a page into cleanable state - and from there
1246 * to be released. Possibly the page is already clean. We always
1247 * have buffer heads in this call.
1249 * Returns 0 if the page is ok to release, 1 otherwise.
1251 * Possible scenarios are:
1253 * 1. We are being called to release a page which has been written
1254 * to via regular I/O. buffer heads will be dirty and possibly
1255 * delalloc. If no delalloc buffer heads in this case then we
1256 * can just return zero.
1258 * 2. We are called to release a page which has been written via
1259 * mmap, all we need to do is ensure there is no delalloc
1260 * state in the buffer heads, if not we can let the caller
1261 * free them and we should come back later via writepage.
1268 struct inode *inode = page->mapping->host;
1269 int dirty, delalloc, unmapped, unwritten;
1270 struct writeback_control wbc = {
1271 .sync_mode = WB_SYNC_ALL,
1275 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1277 if (!page_has_buffers(page))
1280 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1281 if (!delalloc && !unwritten)
1284 if (!(gfp_mask & __GFP_FS))
1287 /* If we are already inside a transaction or the thread cannot
1288 * do I/O, we cannot release this page.
1290 if (current_test_flags(PF_FSTRANS))
1294 * Convert delalloc space to real space, do not flush the
1295 * data out to disk, that will be done by the caller.
1296 * Never need to allocate space here - we will always
1297 * come back to writepage in that case.
1299 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1300 if (dirty == 0 && !unwritten)
1305 return try_to_free_buffers(page);
1310 struct inode *inode,
1312 struct buffer_head *bh_result,
1315 bmapi_flags_t flags)
1323 offset = (xfs_off_t)iblock << inode->i_blkbits;
1324 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1325 size = bh_result->b_size;
1326 error = xfs_bmap(XFS_I(inode), offset, size,
1327 create ? flags : BMAPI_READ, &iomap, &niomap);
1333 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1335 * For unwritten extents do not report a disk address on
1336 * the read case (treat as if we're reading into a hole).
1338 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1339 xfs_map_buffer(bh_result, &iomap, offset,
1342 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1344 bh_result->b_private = inode;
1345 set_buffer_unwritten(bh_result);
1350 * If this is a realtime file, data may be on a different device.
1351 * to that pointed to from the buffer_head b_bdev currently.
1353 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1356 * If we previously allocated a block out beyond eof and we are now
1357 * coming back to use it then we will need to flag it as new even if it
1358 * has a disk address.
1360 * With sub-block writes into unwritten extents we also need to mark
1361 * the buffer as new so that the unwritten parts of the buffer gets
1365 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1366 (offset >= i_size_read(inode)) ||
1367 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1368 set_buffer_new(bh_result);
1370 if (iomap.iomap_flags & IOMAP_DELAY) {
1373 set_buffer_uptodate(bh_result);
1374 set_buffer_mapped(bh_result);
1375 set_buffer_delay(bh_result);
1379 if (direct || size > (1 << inode->i_blkbits)) {
1380 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1381 offset = min_t(xfs_off_t,
1382 iomap.iomap_bsize - iomap.iomap_delta, size);
1383 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1391 struct inode *inode,
1393 struct buffer_head *bh_result,
1396 return __xfs_get_blocks(inode, iblock,
1397 bh_result, create, 0, BMAPI_WRITE);
1401 xfs_get_blocks_direct(
1402 struct inode *inode,
1404 struct buffer_head *bh_result,
1407 return __xfs_get_blocks(inode, iblock,
1408 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1418 xfs_ioend_t *ioend = iocb->private;
1421 * Non-NULL private data means we need to issue a transaction to
1422 * convert a range from unwritten to written extents. This needs
1423 * to happen from process context but aio+dio I/O completion
1424 * happens from irq context so we need to defer it to a workqueue.
1425 * This is not necessary for synchronous direct I/O, but we do
1426 * it anyway to keep the code uniform and simpler.
1428 * Well, if only it were that simple. Because synchronous direct I/O
1429 * requires extent conversion to occur *before* we return to userspace,
1430 * we have to wait for extent conversion to complete. Look at the
1431 * iocb that has been passed to us to determine if this is AIO or
1432 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1433 * workqueue and wait for it to complete.
1435 * The core direct I/O code might be changed to always call the
1436 * completion handler in the future, in which case all this can
1439 ioend->io_offset = offset;
1440 ioend->io_size = size;
1441 if (ioend->io_type == IOMAP_READ) {
1442 xfs_finish_ioend(ioend, 0);
1443 } else if (private && size > 0) {
1444 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1447 * A direct I/O write ioend starts it's life in unwritten
1448 * state in case they map an unwritten extent. This write
1449 * didn't map an unwritten extent so switch it's completion
1452 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1453 xfs_finish_ioend(ioend, 0);
1457 * blockdev_direct_IO can return an error even after the I/O
1458 * completion handler was called. Thus we need to protect
1459 * against double-freeing.
1461 iocb->private = NULL;
1468 const struct iovec *iov,
1470 unsigned long nr_segs)
1472 struct file *file = iocb->ki_filp;
1473 struct inode *inode = file->f_mapping->host;
1479 error = xfs_bmap(XFS_I(inode), offset, 0,
1480 BMAPI_DEVICE, &iomap, &maps);
1485 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1486 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1487 iomap.iomap_target->bt_bdev,
1488 iov, offset, nr_segs,
1489 xfs_get_blocks_direct,
1492 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1493 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1494 iomap.iomap_target->bt_bdev,
1495 iov, offset, nr_segs,
1496 xfs_get_blocks_direct,
1500 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1501 xfs_destroy_ioend(iocb->private);
1508 struct address_space *mapping,
1512 struct page **pagep,
1516 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1522 struct address_space *mapping,
1525 struct inode *inode = (struct inode *)mapping->host;
1526 struct xfs_inode *ip = XFS_I(inode);
1528 vn_trace_entry(XFS_I(inode), __FUNCTION__,
1529 (inst_t *)__return_address);
1530 xfs_rwlock(ip, VRWLOCK_READ);
1531 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1532 xfs_rwunlock(ip, VRWLOCK_READ);
1533 return generic_block_bmap(mapping, block, xfs_get_blocks);
1538 struct file *unused,
1541 return mpage_readpage(page, xfs_get_blocks);
1546 struct file *unused,
1547 struct address_space *mapping,
1548 struct list_head *pages,
1551 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1555 xfs_vm_invalidatepage(
1557 unsigned long offset)
1559 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1560 page->mapping->host, page, offset);
1561 block_invalidatepage(page, offset);
1564 const struct address_space_operations xfs_address_space_operations = {
1565 .readpage = xfs_vm_readpage,
1566 .readpages = xfs_vm_readpages,
1567 .writepage = xfs_vm_writepage,
1568 .writepages = xfs_vm_writepages,
1569 .sync_page = block_sync_page,
1570 .releasepage = xfs_vm_releasepage,
1571 .invalidatepage = xfs_vm_invalidatepage,
1572 .write_begin = xfs_vm_write_begin,
1573 .write_end = generic_write_end,
1574 .bmap = xfs_vm_bmap,
1575 .direct_IO = xfs_vm_direct_IO,
1576 .migratepage = buffer_migrate_page,