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
26 #include "xfs_trans.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dir_sf.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_alloc.h"
38 #include "xfs_btree.h"
39 #include "xfs_error.h"
41 #include "xfs_iomap.h"
42 #include <linux/mpage.h>
43 #include <linux/pagevec.h>
44 #include <linux/writeback.h>
53 struct buffer_head *bh, *head;
55 *delalloc = *unmapped = *unwritten = 0;
57 bh = head = page_buffers(page);
59 if (buffer_uptodate(bh) && !buffer_mapped(bh))
61 else if (buffer_unwritten(bh) && !buffer_delay(bh))
62 clear_buffer_unwritten(bh);
63 else if (buffer_unwritten(bh))
65 else if (buffer_delay(bh))
67 } while ((bh = bh->b_this_page) != head);
70 #if defined(XFS_RW_TRACE)
79 bhv_vnode_t *vp = vn_from_inode(inode);
80 loff_t isize = i_size_read(inode);
81 loff_t offset = page_offset(page);
82 int delalloc = -1, unmapped = -1, unwritten = -1;
84 if (page_has_buffers(page))
85 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
91 ktrace_enter(ip->i_rwtrace,
92 (void *)((unsigned long)tag),
96 (void *)((unsigned long)mask),
97 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
99 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
100 (void *)((unsigned long)(isize & 0xffffffff)),
101 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
102 (void *)((unsigned long)(offset & 0xffffffff)),
103 (void *)((unsigned long)delalloc),
104 (void *)((unsigned long)unmapped),
105 (void *)((unsigned long)unwritten),
106 (void *)((unsigned long)current_pid()),
110 #define xfs_page_trace(tag, inode, page, mask)
114 * Schedule IO completion handling on a xfsdatad if this was
115 * the final hold on this ioend.
121 if (atomic_dec_and_test(&ioend->io_remaining))
122 queue_work(xfsdatad_workqueue, &ioend->io_work);
126 * We're now finished for good with this ioend structure.
127 * Update the page state via the associated buffer_heads,
128 * release holds on the inode and bio, and finally free
129 * up memory. Do not use the ioend after this.
135 struct buffer_head *bh, *next;
137 for (bh = ioend->io_buffer_head; bh; bh = next) {
138 next = bh->b_private;
139 bh->b_end_io(bh, !ioend->io_error);
141 if (unlikely(ioend->io_error))
142 vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__);
143 vn_iowake(ioend->io_vnode);
144 mempool_free(ioend, xfs_ioend_pool);
148 * Buffered IO write completion for delayed allocate extents.
149 * TODO: Update ondisk isize now that we know the file data
150 * has been flushed (i.e. the notorious "NULL file" problem).
153 xfs_end_bio_delalloc(
156 xfs_ioend_t *ioend = data;
158 xfs_destroy_ioend(ioend);
162 * Buffered IO write completion for regular, written extents.
168 xfs_ioend_t *ioend = data;
170 xfs_destroy_ioend(ioend);
174 * IO write completion for unwritten extents.
176 * Issue transactions to convert a buffer range from unwritten
177 * to written extents.
180 xfs_end_bio_unwritten(
183 xfs_ioend_t *ioend = data;
184 bhv_vnode_t *vp = ioend->io_vnode;
185 xfs_off_t offset = ioend->io_offset;
186 size_t size = ioend->io_size;
188 if (likely(!ioend->io_error))
189 bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL);
190 xfs_destroy_ioend(ioend);
194 * Allocate and initialise an IO completion structure.
195 * We need to track unwritten extent write completion here initially.
196 * We'll need to extend this for updating the ondisk inode size later
206 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
209 * Set the count to 1 initially, which will prevent an I/O
210 * completion callback from happening before we have started
211 * all the I/O from calling the completion routine too early.
213 atomic_set(&ioend->io_remaining, 1);
215 ioend->io_list = NULL;
216 ioend->io_type = type;
217 ioend->io_vnode = vn_from_inode(inode);
218 ioend->io_buffer_head = NULL;
219 ioend->io_buffer_tail = NULL;
220 atomic_inc(&ioend->io_vnode->v_iocount);
221 ioend->io_offset = 0;
224 if (type == IOMAP_UNWRITTEN)
225 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
226 else if (type == IOMAP_DELAY)
227 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend);
229 INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend);
242 bhv_vnode_t *vp = vn_from_inode(inode);
243 int error, nmaps = 1;
245 error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps);
246 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
256 return offset >= iomapp->iomap_offset &&
257 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
261 * BIO completion handler for buffered IO.
266 unsigned int bytes_done,
269 xfs_ioend_t *ioend = bio->bi_private;
274 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
275 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
277 /* Toss bio and pass work off to an xfsdatad thread */
278 bio->bi_private = NULL;
279 bio->bi_end_io = NULL;
282 xfs_finish_ioend(ioend);
287 xfs_submit_ioend_bio(
291 atomic_inc(&ioend->io_remaining);
293 bio->bi_private = ioend;
294 bio->bi_end_io = xfs_end_bio;
296 submit_bio(WRITE, bio);
297 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
303 struct buffer_head *bh)
306 int nvecs = bio_get_nr_vecs(bh->b_bdev);
309 bio = bio_alloc(GFP_NOIO, nvecs);
313 ASSERT(bio->bi_private == NULL);
314 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
315 bio->bi_bdev = bh->b_bdev;
321 xfs_start_buffer_writeback(
322 struct buffer_head *bh)
324 ASSERT(buffer_mapped(bh));
325 ASSERT(buffer_locked(bh));
326 ASSERT(!buffer_delay(bh));
327 ASSERT(!buffer_unwritten(bh));
329 mark_buffer_async_write(bh);
330 set_buffer_uptodate(bh);
331 clear_buffer_dirty(bh);
335 xfs_start_page_writeback(
337 struct writeback_control *wbc,
341 ASSERT(PageLocked(page));
342 ASSERT(!PageWriteback(page));
343 set_page_writeback(page);
345 clear_page_dirty(page);
348 end_page_writeback(page);
349 wbc->pages_skipped++; /* We didn't write this page */
353 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
355 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
359 * Submit all of the bios for all of the ioends we have saved up, covering the
360 * initial writepage page and also any probed pages.
362 * Because we may have multiple ioends spanning a page, we need to start
363 * writeback on all the buffers before we submit them for I/O. If we mark the
364 * buffers as we got, then we can end up with a page that only has buffers
365 * marked async write and I/O complete on can occur before we mark the other
366 * buffers async write.
368 * The end result of this is that we trip a bug in end_page_writeback() because
369 * we call it twice for the one page as the code in end_buffer_async_write()
370 * assumes that all buffers on the page are started at the same time.
372 * The fix is two passes across the ioend list - one to start writeback on the
373 * buffer_heads, and then submit them for I/O on the second pass.
379 xfs_ioend_t *head = ioend;
381 struct buffer_head *bh;
383 sector_t lastblock = 0;
385 /* Pass 1 - start writeback */
387 next = ioend->io_list;
388 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
389 xfs_start_buffer_writeback(bh);
391 } while ((ioend = next) != NULL);
393 /* Pass 2 - submit I/O */
396 next = ioend->io_list;
399 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
403 bio = xfs_alloc_ioend_bio(bh);
404 } else if (bh->b_blocknr != lastblock + 1) {
405 xfs_submit_ioend_bio(ioend, bio);
409 if (bio_add_buffer(bio, bh) != bh->b_size) {
410 xfs_submit_ioend_bio(ioend, bio);
414 lastblock = bh->b_blocknr;
417 xfs_submit_ioend_bio(ioend, bio);
418 xfs_finish_ioend(ioend);
419 } while ((ioend = next) != NULL);
423 * Cancel submission of all buffer_heads so far in this endio.
424 * Toss the endio too. Only ever called for the initial page
425 * in a writepage request, so only ever one page.
432 struct buffer_head *bh, *next_bh;
435 next = ioend->io_list;
436 bh = ioend->io_buffer_head;
438 next_bh = bh->b_private;
439 clear_buffer_async_write(bh);
441 } while ((bh = next_bh) != NULL);
443 vn_iowake(ioend->io_vnode);
444 mempool_free(ioend, xfs_ioend_pool);
445 } while ((ioend = next) != NULL);
449 * Test to see if we've been building up a completion structure for
450 * earlier buffers -- if so, we try to append to this ioend if we
451 * can, otherwise we finish off any current ioend and start another.
452 * Return true if we've finished the given ioend.
457 struct buffer_head *bh,
460 xfs_ioend_t **result,
463 xfs_ioend_t *ioend = *result;
465 if (!ioend || need_ioend || type != ioend->io_type) {
466 xfs_ioend_t *previous = *result;
468 ioend = xfs_alloc_ioend(inode, type);
469 ioend->io_offset = offset;
470 ioend->io_buffer_head = bh;
471 ioend->io_buffer_tail = bh;
473 previous->io_list = ioend;
476 ioend->io_buffer_tail->b_private = bh;
477 ioend->io_buffer_tail = bh;
480 bh->b_private = NULL;
481 ioend->io_size += bh->b_size;
486 struct buffer_head *bh,
493 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
495 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
496 ((offset - mp->iomap_offset) >> block_bits);
498 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
501 set_buffer_mapped(bh);
506 struct buffer_head *bh,
511 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
512 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
515 xfs_map_buffer(bh, iomapp, offset, block_bits);
516 bh->b_bdev = iomapp->iomap_target->bt_bdev;
517 set_buffer_mapped(bh);
518 clear_buffer_delay(bh);
519 clear_buffer_unwritten(bh);
523 * Look for a page at index that is suitable for clustering.
528 unsigned int pg_offset,
533 if (PageWriteback(page))
536 if (page->mapping && PageDirty(page)) {
537 if (page_has_buffers(page)) {
538 struct buffer_head *bh, *head;
540 bh = head = page_buffers(page);
542 if (!buffer_uptodate(bh))
544 if (mapped != buffer_mapped(bh))
547 if (ret >= pg_offset)
549 } while ((bh = bh->b_this_page) != head);
551 ret = mapped ? 0 : PAGE_CACHE_SIZE;
560 struct page *startpage,
561 struct buffer_head *bh,
562 struct buffer_head *head,
566 pgoff_t tindex, tlast, tloff;
570 /* First sum forwards in this page */
572 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
575 } while ((bh = bh->b_this_page) != head);
577 /* if we reached the end of the page, sum forwards in following pages */
578 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
579 tindex = startpage->index + 1;
581 /* Prune this back to avoid pathological behavior */
582 tloff = min(tlast, startpage->index + 64);
584 pagevec_init(&pvec, 0);
585 while (!done && tindex <= tloff) {
586 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
588 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
591 for (i = 0; i < pagevec_count(&pvec); i++) {
592 struct page *page = pvec.pages[i];
593 size_t pg_offset, len = 0;
595 if (tindex == tlast) {
597 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
603 pg_offset = PAGE_CACHE_SIZE;
605 if (page->index == tindex && !TestSetPageLocked(page)) {
606 len = xfs_probe_page(page, pg_offset, mapped);
619 pagevec_release(&pvec);
627 * Test if a given page is suitable for writing as part of an unwritten
628 * or delayed allocate extent.
635 if (PageWriteback(page))
638 if (page->mapping && page_has_buffers(page)) {
639 struct buffer_head *bh, *head;
642 bh = head = page_buffers(page);
644 if (buffer_unwritten(bh))
645 acceptable = (type == IOMAP_UNWRITTEN);
646 else if (buffer_delay(bh))
647 acceptable = (type == IOMAP_DELAY);
648 else if (buffer_dirty(bh) && buffer_mapped(bh))
649 acceptable = (type == 0);
652 } while ((bh = bh->b_this_page) != head);
662 * Allocate & map buffers for page given the extent map. Write it out.
663 * except for the original page of a writepage, this is called on
664 * delalloc/unwritten pages only, for the original page it is possible
665 * that the page has no mapping at all.
673 xfs_ioend_t **ioendp,
674 struct writeback_control *wbc,
678 struct buffer_head *bh, *head;
679 xfs_off_t end_offset;
680 unsigned long p_offset;
682 int bbits = inode->i_blkbits;
684 int count = 0, done = 0, uptodate = 1;
685 xfs_off_t offset = page_offset(page);
687 if (page->index != tindex)
689 if (TestSetPageLocked(page))
691 if (PageWriteback(page))
692 goto fail_unlock_page;
693 if (page->mapping != inode->i_mapping)
694 goto fail_unlock_page;
695 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
696 goto fail_unlock_page;
699 * page_dirty is initially a count of buffers on the page before
700 * EOF and is decremented as we move each into a cleanable state.
704 * End offset is the highest offset that this page should represent.
705 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
706 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
707 * hence give us the correct page_dirty count. On any other page,
708 * it will be zero and in that case we need page_dirty to be the
709 * count of buffers on the page.
711 end_offset = min_t(unsigned long long,
712 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
715 len = 1 << inode->i_blkbits;
716 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
718 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
719 page_dirty = p_offset / len;
721 bh = head = page_buffers(page);
723 if (offset >= end_offset)
725 if (!buffer_uptodate(bh))
727 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
732 if (buffer_unwritten(bh) || buffer_delay(bh)) {
733 if (buffer_unwritten(bh))
734 type = IOMAP_UNWRITTEN;
738 if (!xfs_iomap_valid(mp, offset)) {
743 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
744 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
746 xfs_map_at_offset(bh, offset, bbits, mp);
748 xfs_add_to_ioend(inode, bh, offset,
751 set_buffer_dirty(bh);
753 mark_buffer_dirty(bh);
759 if (buffer_mapped(bh) && all_bh && startio) {
761 xfs_add_to_ioend(inode, bh, offset,
769 } while (offset += len, (bh = bh->b_this_page) != head);
771 if (uptodate && bh == head)
772 SetPageUptodate(page);
776 struct backing_dev_info *bdi;
778 bdi = inode->i_mapping->backing_dev_info;
780 if (bdi_write_congested(bdi)) {
781 wbc->encountered_congestion = 1;
783 } else if (wbc->nr_to_write <= 0) {
787 xfs_start_page_writeback(page, wbc, !page_dirty, count);
798 * Convert & write out a cluster of pages in the same extent as defined
799 * by mp and following the start page.
806 xfs_ioend_t **ioendp,
807 struct writeback_control *wbc,
815 pagevec_init(&pvec, 0);
816 while (!done && tindex <= tlast) {
817 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
819 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
822 for (i = 0; i < pagevec_count(&pvec); i++) {
823 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
824 iomapp, ioendp, wbc, startio, all_bh);
829 pagevec_release(&pvec);
835 * Calling this without startio set means we are being asked to make a dirty
836 * page ready for freeing it's buffers. When called with startio set then
837 * we are coming from writepage.
839 * When called with startio set it is important that we write the WHOLE
841 * The bh->b_state's cannot know if any of the blocks or which block for
842 * that matter are dirty due to mmap writes, and therefore bh uptodate is
843 * only valid if the page itself isn't completely uptodate. Some layers
844 * may clear the page dirty flag prior to calling write page, under the
845 * assumption the entire page will be written out; by not writing out the
846 * whole page the page can be reused before all valid dirty data is
847 * written out. Note: in the case of a page that has been dirty'd by
848 * mapwrite and but partially setup by block_prepare_write the
849 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
850 * valid state, thus the whole page must be written out thing.
854 xfs_page_state_convert(
857 struct writeback_control *wbc,
859 int unmapped) /* also implies page uptodate */
861 struct buffer_head *bh, *head;
863 xfs_ioend_t *ioend = NULL, *iohead = NULL;
865 unsigned long p_offset = 0;
867 __uint64_t end_offset;
868 pgoff_t end_index, last_index, tlast;
870 int flags, err, iomap_valid = 0, uptodate = 1;
871 int page_dirty, count = 0;
873 int all_bh = unmapped;
876 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
877 trylock |= BMAPI_TRYLOCK;
880 /* Is this page beyond the end of the file? */
881 offset = i_size_read(inode);
882 end_index = offset >> PAGE_CACHE_SHIFT;
883 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
884 if (page->index >= end_index) {
885 if ((page->index >= end_index + 1) ||
886 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
894 * page_dirty is initially a count of buffers on the page before
895 * EOF and is decremented as we move each into a cleanable state.
899 * End offset is the highest offset that this page should represent.
900 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
901 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
902 * hence give us the correct page_dirty count. On any other page,
903 * it will be zero and in that case we need page_dirty to be the
904 * count of buffers on the page.
906 end_offset = min_t(unsigned long long,
907 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
908 len = 1 << inode->i_blkbits;
909 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
911 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
912 page_dirty = p_offset / len;
914 bh = head = page_buffers(page);
915 offset = page_offset(page);
919 /* TODO: cleanup count and page_dirty */
922 if (offset >= end_offset)
924 if (!buffer_uptodate(bh))
926 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
928 * the iomap is actually still valid, but the ioend
929 * isn't. shouldn't happen too often.
936 iomap_valid = xfs_iomap_valid(&iomap, offset);
939 * First case, map an unwritten extent and prepare for
940 * extent state conversion transaction on completion.
942 * Second case, allocate space for a delalloc buffer.
943 * We can return EAGAIN here in the release page case.
945 * Third case, an unmapped buffer was found, and we are
946 * in a path where we need to write the whole page out.
948 if (buffer_unwritten(bh) || buffer_delay(bh) ||
949 ((buffer_uptodate(bh) || PageUptodate(page)) &&
950 !buffer_mapped(bh) && (unmapped || startio))) {
952 * Make sure we don't use a read-only iomap
954 if (flags == BMAPI_READ)
957 if (buffer_unwritten(bh)) {
958 type = IOMAP_UNWRITTEN;
959 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
960 } else if (buffer_delay(bh)) {
962 flags = BMAPI_ALLOCATE | trylock;
965 flags = BMAPI_WRITE | BMAPI_MMAP;
969 if (type == IOMAP_NEW) {
970 size = xfs_probe_cluster(inode,
976 err = xfs_map_blocks(inode, offset, size,
980 iomap_valid = xfs_iomap_valid(&iomap, offset);
983 xfs_map_at_offset(bh, offset,
984 inode->i_blkbits, &iomap);
986 xfs_add_to_ioend(inode, bh, offset,
990 set_buffer_dirty(bh);
992 mark_buffer_dirty(bh);
997 } else if (buffer_uptodate(bh) && startio) {
999 * we got here because the buffer is already mapped.
1000 * That means it must already have extents allocated
1001 * underneath it. Map the extent by reading it.
1003 if (!iomap_valid || type != 0) {
1005 size = xfs_probe_cluster(inode, page, bh,
1007 err = xfs_map_blocks(inode, offset, size,
1011 iomap_valid = xfs_iomap_valid(&iomap, offset);
1015 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1016 ASSERT(buffer_mapped(bh));
1019 xfs_add_to_ioend(inode, bh, offset, type,
1020 &ioend, !iomap_valid);
1026 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1027 (unmapped || startio)) {
1034 } while (offset += len, ((bh = bh->b_this_page) != head));
1036 if (uptodate && bh == head)
1037 SetPageUptodate(page);
1040 xfs_start_page_writeback(page, wbc, 1, count);
1042 if (ioend && iomap_valid) {
1043 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1045 tlast = min_t(pgoff_t, offset, last_index);
1046 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1047 wbc, startio, all_bh, tlast);
1051 xfs_submit_ioend(iohead);
1057 xfs_cancel_ioend(iohead);
1060 * If it's delalloc and we have nowhere to put it,
1061 * throw it away, unless the lower layers told
1064 if (err != -EAGAIN) {
1066 block_invalidatepage(page, 0);
1067 ClearPageUptodate(page);
1073 * writepage: Called from one of two places:
1075 * 1. we are flushing a delalloc buffer head.
1077 * 2. we are writing out a dirty page. Typically the page dirty
1078 * state is cleared before we get here. In this case is it
1079 * conceivable we have no buffer heads.
1081 * For delalloc space on the page we need to allocate space and
1082 * flush it. For unmapped buffer heads on the page we should
1083 * allocate space if the page is uptodate. For any other dirty
1084 * buffer heads on the page we should flush them.
1086 * If we detect that a transaction would be required to flush
1087 * the page, we have to check the process flags first, if we
1088 * are already in a transaction or disk I/O during allocations
1089 * is off, we need to fail the writepage and redirty the page.
1095 struct writeback_control *wbc)
1099 int delalloc, unmapped, unwritten;
1100 struct inode *inode = page->mapping->host;
1102 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1105 * We need a transaction if:
1106 * 1. There are delalloc buffers on the page
1107 * 2. The page is uptodate and we have unmapped buffers
1108 * 3. The page is uptodate and we have no buffers
1109 * 4. There are unwritten buffers on the page
1112 if (!page_has_buffers(page)) {
1116 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1117 if (!PageUptodate(page))
1119 need_trans = delalloc + unmapped + unwritten;
1123 * If we need a transaction and the process flags say
1124 * we are already in a transaction, or no IO is allowed
1125 * then mark the page dirty again and leave the page
1128 if (current_test_flags(PF_FSTRANS) && need_trans)
1132 * Delay hooking up buffer heads until we have
1133 * made our go/no-go decision.
1135 if (!page_has_buffers(page))
1136 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1139 * Convert delayed allocate, unwritten or unmapped space
1140 * to real space and flush out to disk.
1142 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1143 if (error == -EAGAIN)
1145 if (unlikely(error < 0))
1151 redirty_page_for_writepage(wbc, page);
1161 struct address_space *mapping,
1162 struct writeback_control *wbc)
1164 struct bhv_vnode *vp = vn_from_inode(mapping->host);
1168 return generic_writepages(mapping, wbc);
1172 * Called to move a page into cleanable state - and from there
1173 * to be released. Possibly the page is already clean. We always
1174 * have buffer heads in this call.
1176 * Returns 0 if the page is ok to release, 1 otherwise.
1178 * Possible scenarios are:
1180 * 1. We are being called to release a page which has been written
1181 * to via regular I/O. buffer heads will be dirty and possibly
1182 * delalloc. If no delalloc buffer heads in this case then we
1183 * can just return zero.
1185 * 2. We are called to release a page which has been written via
1186 * mmap, all we need to do is ensure there is no delalloc
1187 * state in the buffer heads, if not we can let the caller
1188 * free them and we should come back later via writepage.
1195 struct inode *inode = page->mapping->host;
1196 int dirty, delalloc, unmapped, unwritten;
1197 struct writeback_control wbc = {
1198 .sync_mode = WB_SYNC_ALL,
1202 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1204 if (!page_has_buffers(page))
1207 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1208 if (!delalloc && !unwritten)
1211 if (!(gfp_mask & __GFP_FS))
1214 /* If we are already inside a transaction or the thread cannot
1215 * do I/O, we cannot release this page.
1217 if (current_test_flags(PF_FSTRANS))
1221 * Convert delalloc space to real space, do not flush the
1222 * data out to disk, that will be done by the caller.
1223 * Never need to allocate space here - we will always
1224 * come back to writepage in that case.
1226 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1227 if (dirty == 0 && !unwritten)
1232 return try_to_free_buffers(page);
1237 struct inode *inode,
1239 struct buffer_head *bh_result,
1242 bmapi_flags_t flags)
1244 bhv_vnode_t *vp = vn_from_inode(inode);
1251 offset = (xfs_off_t)iblock << inode->i_blkbits;
1252 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1253 size = bh_result->b_size;
1254 error = bhv_vop_bmap(vp, offset, size,
1255 create ? flags : BMAPI_READ, &iomap, &niomap);
1261 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1263 * For unwritten extents do not report a disk address on
1264 * the read case (treat as if we're reading into a hole).
1266 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1267 xfs_map_buffer(bh_result, &iomap, offset,
1270 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1272 bh_result->b_private = inode;
1273 set_buffer_unwritten(bh_result);
1274 set_buffer_delay(bh_result);
1279 * If this is a realtime file, data may be on a different device.
1280 * to that pointed to from the buffer_head b_bdev currently.
1282 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1285 * If we previously allocated a block out beyond eof and we are
1286 * now coming back to use it then we will need to flag it as new
1287 * even if it has a disk address.
1290 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1291 (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
1292 set_buffer_new(bh_result);
1294 if (iomap.iomap_flags & IOMAP_DELAY) {
1297 set_buffer_uptodate(bh_result);
1298 set_buffer_mapped(bh_result);
1299 set_buffer_delay(bh_result);
1303 if (direct || size > (1 << inode->i_blkbits)) {
1304 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1305 offset = min_t(xfs_off_t,
1306 iomap.iomap_bsize - iomap.iomap_delta, size);
1307 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1315 struct inode *inode,
1317 struct buffer_head *bh_result,
1320 return __xfs_get_blocks(inode, iblock,
1321 bh_result, create, 0, BMAPI_WRITE);
1325 xfs_get_blocks_direct(
1326 struct inode *inode,
1328 struct buffer_head *bh_result,
1331 return __xfs_get_blocks(inode, iblock,
1332 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1342 xfs_ioend_t *ioend = iocb->private;
1345 * Non-NULL private data means we need to issue a transaction to
1346 * convert a range from unwritten to written extents. This needs
1347 * to happen from process context but aio+dio I/O completion
1348 * happens from irq context so we need to defer it to a workqueue.
1349 * This is not necessary for synchronous direct I/O, but we do
1350 * it anyway to keep the code uniform and simpler.
1352 * The core direct I/O code might be changed to always call the
1353 * completion handler in the future, in which case all this can
1356 if (private && size > 0) {
1357 ioend->io_offset = offset;
1358 ioend->io_size = size;
1359 xfs_finish_ioend(ioend);
1362 xfs_destroy_ioend(ioend);
1366 * blockdev_direct_IO can return an error even after the I/O
1367 * completion handler was called. Thus we need to protect
1368 * against double-freeing.
1370 iocb->private = NULL;
1377 const struct iovec *iov,
1379 unsigned long nr_segs)
1381 struct file *file = iocb->ki_filp;
1382 struct inode *inode = file->f_mapping->host;
1383 bhv_vnode_t *vp = vn_from_inode(inode);
1389 error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps);
1393 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1395 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1396 iomap.iomap_target->bt_bdev,
1397 iov, offset, nr_segs,
1398 xfs_get_blocks_direct,
1401 if (unlikely(ret <= 0 && iocb->private))
1402 xfs_destroy_ioend(iocb->private);
1407 xfs_vm_prepare_write(
1413 return block_prepare_write(page, from, to, xfs_get_blocks);
1418 struct address_space *mapping,
1421 struct inode *inode = (struct inode *)mapping->host;
1422 bhv_vnode_t *vp = vn_from_inode(inode);
1424 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1425 bhv_vop_rwlock(vp, VRWLOCK_READ);
1426 bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
1427 bhv_vop_rwunlock(vp, VRWLOCK_READ);
1428 return generic_block_bmap(mapping, block, xfs_get_blocks);
1433 struct file *unused,
1436 return mpage_readpage(page, xfs_get_blocks);
1441 struct file *unused,
1442 struct address_space *mapping,
1443 struct list_head *pages,
1446 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1450 xfs_vm_invalidatepage(
1452 unsigned long offset)
1454 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1455 page->mapping->host, page, offset);
1456 block_invalidatepage(page, offset);
1459 struct address_space_operations xfs_address_space_operations = {
1460 .readpage = xfs_vm_readpage,
1461 .readpages = xfs_vm_readpages,
1462 .writepage = xfs_vm_writepage,
1463 .writepages = xfs_vm_writepages,
1464 .sync_page = block_sync_page,
1465 .releasepage = xfs_vm_releasepage,
1466 .invalidatepage = xfs_vm_invalidatepage,
1467 .prepare_write = xfs_vm_prepare_write,
1468 .commit_write = generic_commit_write,
1469 .bmap = xfs_vm_bmap,
1470 .direct_IO = xfs_vm_direct_IO,
1471 .migratepage = buffer_migrate_page,