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>
46 STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
48 #if defined(XFS_RW_TRACE)
58 vnode_t *vp = LINVFS_GET_VP(inode);
59 loff_t isize = i_size_read(inode);
60 loff_t offset = page_offset(page);
61 int delalloc = -1, unmapped = -1, unwritten = -1;
63 if (page_has_buffers(page))
64 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
66 bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
71 ktrace_enter(ip->i_rwtrace,
72 (void *)((unsigned long)tag),
76 (void *)((unsigned long)mask),
77 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
78 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
79 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
80 (void *)((unsigned long)(isize & 0xffffffff)),
81 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
82 (void *)((unsigned long)(offset & 0xffffffff)),
83 (void *)((unsigned long)delalloc),
84 (void *)((unsigned long)unmapped),
85 (void *)((unsigned long)unwritten),
90 #define xfs_page_trace(tag, inode, page, mask)
94 * Schedule IO completion handling on a xfsdatad if this was
95 * the final hold on this ioend.
101 if (atomic_dec_and_test(&ioend->io_remaining))
102 queue_work(xfsdatad_workqueue, &ioend->io_work);
106 * We're now finished for good with this ioend structure.
107 * Update the page state via the associated buffer_heads,
108 * release holds on the inode and bio, and finally free
109 * up memory. Do not use the ioend after this.
115 struct buffer_head *bh, *next;
117 for (bh = ioend->io_buffer_head; bh; bh = next) {
118 next = bh->b_private;
119 bh->b_end_io(bh, ioend->io_uptodate);
122 vn_iowake(ioend->io_vnode);
123 mempool_free(ioend, xfs_ioend_pool);
127 * Buffered IO write completion for delayed allocate extents.
128 * TODO: Update ondisk isize now that we know the file data
129 * has been flushed (i.e. the notorious "NULL file" problem).
132 xfs_end_bio_delalloc(
135 xfs_ioend_t *ioend = data;
137 xfs_destroy_ioend(ioend);
141 * Buffered IO write completion for regular, written extents.
147 xfs_ioend_t *ioend = data;
149 xfs_destroy_ioend(ioend);
153 * IO write completion for unwritten extents.
155 * Issue transactions to convert a buffer range from unwritten
156 * to written extents.
159 xfs_end_bio_unwritten(
162 xfs_ioend_t *ioend = data;
163 vnode_t *vp = ioend->io_vnode;
164 xfs_off_t offset = ioend->io_offset;
165 size_t size = ioend->io_size;
168 if (ioend->io_uptodate)
169 VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
170 xfs_destroy_ioend(ioend);
174 * Allocate and initialise an IO completion structure.
175 * We need to track unwritten extent write completion here initially.
176 * We'll need to extend this for updating the ondisk inode size later
186 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
189 * Set the count to 1 initially, which will prevent an I/O
190 * completion callback from happening before we have started
191 * all the I/O from calling the completion routine too early.
193 atomic_set(&ioend->io_remaining, 1);
194 ioend->io_uptodate = 1; /* cleared if any I/O fails */
195 ioend->io_list = NULL;
196 ioend->io_type = type;
197 ioend->io_vnode = LINVFS_GET_VP(inode);
198 ioend->io_buffer_head = NULL;
199 ioend->io_buffer_tail = NULL;
200 atomic_inc(&ioend->io_vnode->v_iocount);
201 ioend->io_offset = 0;
204 if (type == IOMAP_UNWRITTEN)
205 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
206 else if (type == IOMAP_DELAY)
207 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend);
209 INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend);
222 vnode_t *vp = LINVFS_GET_VP(inode);
223 int error, nmaps = 1;
225 VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
226 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
236 return offset >= iomapp->iomap_offset &&
237 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
241 * BIO completion handler for buffered IO.
246 unsigned int bytes_done,
249 xfs_ioend_t *ioend = bio->bi_private;
255 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
257 /* Toss bio and pass work off to an xfsdatad thread */
258 if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
259 ioend->io_uptodate = 0;
260 bio->bi_private = NULL;
261 bio->bi_end_io = NULL;
264 xfs_finish_ioend(ioend);
269 xfs_submit_ioend_bio(
273 atomic_inc(&ioend->io_remaining);
275 bio->bi_private = ioend;
276 bio->bi_end_io = xfs_end_bio;
278 submit_bio(WRITE, bio);
279 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
285 struct buffer_head *bh)
288 int nvecs = bio_get_nr_vecs(bh->b_bdev);
291 bio = bio_alloc(GFP_NOIO, nvecs);
295 ASSERT(bio->bi_private == NULL);
296 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
297 bio->bi_bdev = bh->b_bdev;
303 xfs_start_buffer_writeback(
304 struct buffer_head *bh)
306 ASSERT(buffer_mapped(bh));
307 ASSERT(buffer_locked(bh));
308 ASSERT(!buffer_delay(bh));
309 ASSERT(!buffer_unwritten(bh));
311 mark_buffer_async_write(bh);
312 set_buffer_uptodate(bh);
313 clear_buffer_dirty(bh);
317 xfs_start_page_writeback(
319 struct writeback_control *wbc,
323 ASSERT(PageLocked(page));
324 ASSERT(!PageWriteback(page));
325 set_page_writeback(page);
327 clear_page_dirty(page);
330 end_page_writeback(page);
331 wbc->pages_skipped++; /* We didn't write this page */
335 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
337 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
341 * Submit all of the bios for all of the ioends we have saved up,
342 * covering the initial writepage page and also any probed pages.
349 struct buffer_head *bh;
351 sector_t lastblock = 0;
354 next = ioend->io_list;
357 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
358 xfs_start_buffer_writeback(bh);
362 bio = xfs_alloc_ioend_bio(bh);
363 } else if (bh->b_blocknr != lastblock + 1) {
364 xfs_submit_ioend_bio(ioend, bio);
368 if (bio_add_buffer(bio, bh) != bh->b_size) {
369 xfs_submit_ioend_bio(ioend, bio);
373 lastblock = bh->b_blocknr;
376 xfs_submit_ioend_bio(ioend, bio);
377 xfs_finish_ioend(ioend);
378 } while ((ioend = next) != NULL);
382 * Cancel submission of all buffer_heads so far in this endio.
383 * Toss the endio too. Only ever called for the initial page
384 * in a writepage request, so only ever one page.
391 struct buffer_head *bh, *next_bh;
394 next = ioend->io_list;
395 bh = ioend->io_buffer_head;
397 next_bh = bh->b_private;
398 clear_buffer_async_write(bh);
400 } while ((bh = next_bh) != NULL);
402 vn_iowake(ioend->io_vnode);
403 mempool_free(ioend, xfs_ioend_pool);
404 } while ((ioend = next) != NULL);
408 * Test to see if we've been building up a completion structure for
409 * earlier buffers -- if so, we try to append to this ioend if we
410 * can, otherwise we finish off any current ioend and start another.
411 * Return true if we've finished the given ioend.
416 struct buffer_head *bh,
419 xfs_ioend_t **result,
422 xfs_ioend_t *ioend = *result;
424 if (!ioend || need_ioend || type != ioend->io_type) {
425 xfs_ioend_t *previous = *result;
427 ioend = xfs_alloc_ioend(inode, type);
428 ioend->io_offset = offset;
429 ioend->io_buffer_head = bh;
430 ioend->io_buffer_tail = bh;
432 previous->io_list = ioend;
435 ioend->io_buffer_tail->b_private = bh;
436 ioend->io_buffer_tail = bh;
439 bh->b_private = NULL;
440 ioend->io_size += bh->b_size;
445 struct buffer_head *bh,
453 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
454 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
455 ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);
457 sector_shift = block_bits - BBSHIFT;
458 bn = (iomapp->iomap_bn >> sector_shift) +
459 ((offset - iomapp->iomap_offset) >> block_bits);
461 ASSERT(bn || (iomapp->iomap_flags & IOMAP_REALTIME));
462 ASSERT((bn << sector_shift) >= iomapp->iomap_bn);
466 bh->b_bdev = iomapp->iomap_target->bt_bdev;
467 set_buffer_mapped(bh);
468 clear_buffer_delay(bh);
469 clear_buffer_unwritten(bh);
473 * Look for a page at index that is suitable for clustering.
478 unsigned int pg_offset,
483 if (PageWriteback(page))
486 if (page->mapping && PageDirty(page)) {
487 if (page_has_buffers(page)) {
488 struct buffer_head *bh, *head;
490 bh = head = page_buffers(page);
492 if (!buffer_uptodate(bh))
494 if (mapped != buffer_mapped(bh))
497 if (ret >= pg_offset)
499 } while ((bh = bh->b_this_page) != head);
501 ret = mapped ? 0 : PAGE_CACHE_SIZE;
510 struct page *startpage,
511 struct buffer_head *bh,
512 struct buffer_head *head,
516 pgoff_t tindex, tlast, tloff;
520 /* First sum forwards in this page */
522 if (mapped != buffer_mapped(bh))
525 } while ((bh = bh->b_this_page) != head);
527 /* if we reached the end of the page, sum forwards in following pages */
528 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
529 tindex = startpage->index + 1;
531 /* Prune this back to avoid pathological behavior */
532 tloff = min(tlast, startpage->index + 64);
534 pagevec_init(&pvec, 0);
535 while (!done && tindex <= tloff) {
536 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
538 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
541 for (i = 0; i < pagevec_count(&pvec); i++) {
542 struct page *page = pvec.pages[i];
543 size_t pg_offset, len = 0;
545 if (tindex == tlast) {
547 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
553 pg_offset = PAGE_CACHE_SIZE;
555 if (page->index == tindex && !TestSetPageLocked(page)) {
556 len = xfs_probe_page(page, pg_offset, mapped);
569 pagevec_release(&pvec);
577 * Test if a given page is suitable for writing as part of an unwritten
578 * or delayed allocate extent.
585 if (PageWriteback(page))
588 if (page->mapping && page_has_buffers(page)) {
589 struct buffer_head *bh, *head;
592 bh = head = page_buffers(page);
594 if (buffer_unwritten(bh))
595 acceptable = (type == IOMAP_UNWRITTEN);
596 else if (buffer_delay(bh))
597 acceptable = (type == IOMAP_DELAY);
598 else if (buffer_mapped(bh))
599 acceptable = (type == 0);
602 } while ((bh = bh->b_this_page) != head);
612 * Allocate & map buffers for page given the extent map. Write it out.
613 * except for the original page of a writepage, this is called on
614 * delalloc/unwritten pages only, for the original page it is possible
615 * that the page has no mapping at all.
623 xfs_ioend_t **ioendp,
624 struct writeback_control *wbc,
628 struct buffer_head *bh, *head;
629 xfs_off_t end_offset;
630 unsigned long p_offset;
632 int bbits = inode->i_blkbits;
634 int count = 0, done = 0, uptodate = 1;
635 xfs_off_t offset = page_offset(page);
637 if (page->index != tindex)
639 if (TestSetPageLocked(page))
641 if (PageWriteback(page))
642 goto fail_unlock_page;
643 if (page->mapping != inode->i_mapping)
644 goto fail_unlock_page;
645 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
646 goto fail_unlock_page;
649 * page_dirty is initially a count of buffers on the page before
650 * EOF and is decrememted as we move each into a cleanable state.
654 * End offset is the highest offset that this page should represent.
655 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
656 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
657 * hence give us the correct page_dirty count. On any other page,
658 * it will be zero and in that case we need page_dirty to be the
659 * count of buffers on the page.
661 end_offset = min_t(unsigned long long,
662 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
665 len = 1 << inode->i_blkbits;
666 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
668 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
669 page_dirty = p_offset / len;
671 bh = head = page_buffers(page);
673 if (offset >= end_offset)
675 if (!buffer_uptodate(bh))
677 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
682 if (buffer_unwritten(bh) || buffer_delay(bh)) {
683 if (buffer_unwritten(bh))
684 type = IOMAP_UNWRITTEN;
688 if (!xfs_iomap_valid(mp, offset)) {
693 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
694 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
696 xfs_map_at_offset(bh, offset, bbits, mp);
698 xfs_add_to_ioend(inode, bh, offset,
701 set_buffer_dirty(bh);
703 mark_buffer_dirty(bh);
709 if (buffer_mapped(bh) && all_bh && startio) {
711 xfs_add_to_ioend(inode, bh, offset,
719 } while (offset += len, (bh = bh->b_this_page) != head);
721 if (uptodate && bh == head)
722 SetPageUptodate(page);
726 struct backing_dev_info *bdi;
728 bdi = inode->i_mapping->backing_dev_info;
729 if (bdi_write_congested(bdi)) {
730 wbc->encountered_congestion = 1;
732 } else if (--wbc->nr_to_write <= 0) {
736 xfs_start_page_writeback(page, wbc, !page_dirty, count);
747 * Convert & write out a cluster of pages in the same extent as defined
748 * by mp and following the start page.
755 xfs_ioend_t **ioendp,
756 struct writeback_control *wbc,
764 pagevec_init(&pvec, 0);
765 while (!done && tindex <= tlast) {
766 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
768 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
771 for (i = 0; i < pagevec_count(&pvec); i++) {
772 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
773 iomapp, ioendp, wbc, startio, all_bh);
778 pagevec_release(&pvec);
784 * Calling this without startio set means we are being asked to make a dirty
785 * page ready for freeing it's buffers. When called with startio set then
786 * we are coming from writepage.
788 * When called with startio set it is important that we write the WHOLE
790 * The bh->b_state's cannot know if any of the blocks or which block for
791 * that matter are dirty due to mmap writes, and therefore bh uptodate is
792 * only vaild if the page itself isn't completely uptodate. Some layers
793 * may clear the page dirty flag prior to calling write page, under the
794 * assumption the entire page will be written out; by not writing out the
795 * whole page the page can be reused before all valid dirty data is
796 * written out. Note: in the case of a page that has been dirty'd by
797 * mapwrite and but partially setup by block_prepare_write the
798 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
799 * valid state, thus the whole page must be written out thing.
803 xfs_page_state_convert(
806 struct writeback_control *wbc,
808 int unmapped) /* also implies page uptodate */
810 struct buffer_head *bh, *head;
812 xfs_ioend_t *ioend = NULL, *iohead = NULL;
814 unsigned long p_offset = 0;
816 __uint64_t end_offset;
817 pgoff_t end_index, last_index, tlast;
819 int flags, err, iomap_valid = 0, uptodate = 1;
820 int page_dirty, count = 0, trylock_flag = 0;
821 int all_bh = unmapped;
823 /* wait for other IO threads? */
824 if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking))
825 trylock_flag |= BMAPI_TRYLOCK;
827 /* Is this page beyond the end of the file? */
828 offset = i_size_read(inode);
829 end_index = offset >> PAGE_CACHE_SHIFT;
830 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
831 if (page->index >= end_index) {
832 if ((page->index >= end_index + 1) ||
833 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
841 * page_dirty is initially a count of buffers on the page before
842 * EOF and is decrememted as we move each into a cleanable state.
846 * End offset is the highest offset that this page should represent.
847 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
848 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
849 * hence give us the correct page_dirty count. On any other page,
850 * it will be zero and in that case we need page_dirty to be the
851 * count of buffers on the page.
853 end_offset = min_t(unsigned long long,
854 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
855 len = 1 << inode->i_blkbits;
856 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
858 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
859 page_dirty = p_offset / len;
861 bh = head = page_buffers(page);
862 offset = page_offset(page);
866 /* TODO: cleanup count and page_dirty */
869 if (offset >= end_offset)
871 if (!buffer_uptodate(bh))
873 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
875 * the iomap is actually still valid, but the ioend
876 * isn't. shouldn't happen too often.
883 iomap_valid = xfs_iomap_valid(&iomap, offset);
886 * First case, map an unwritten extent and prepare for
887 * extent state conversion transaction on completion.
889 * Second case, allocate space for a delalloc buffer.
890 * We can return EAGAIN here in the release page case.
892 * Third case, an unmapped buffer was found, and we are
893 * in a path where we need to write the whole page out.
895 if (buffer_unwritten(bh) || buffer_delay(bh) ||
896 ((buffer_uptodate(bh) || PageUptodate(page)) &&
897 !buffer_mapped(bh) && (unmapped || startio))) {
899 * Make sure we don't use a read-only iomap
901 if (flags == BMAPI_READ)
904 if (buffer_unwritten(bh)) {
905 type = IOMAP_UNWRITTEN;
906 flags = BMAPI_WRITE|BMAPI_IGNSTATE;
907 } else if (buffer_delay(bh)) {
909 flags = BMAPI_ALLOCATE;
911 flags |= trylock_flag;
914 flags = BMAPI_WRITE|BMAPI_MMAP;
918 if (type == IOMAP_NEW) {
919 size = xfs_probe_cluster(inode,
925 err = xfs_map_blocks(inode, offset, size,
929 iomap_valid = xfs_iomap_valid(&iomap, offset);
932 xfs_map_at_offset(bh, offset,
933 inode->i_blkbits, &iomap);
935 xfs_add_to_ioend(inode, bh, offset,
939 set_buffer_dirty(bh);
941 mark_buffer_dirty(bh);
946 } else if (buffer_uptodate(bh) && startio) {
948 * we got here because the buffer is already mapped.
949 * That means it must already have extents allocated
950 * underneath it. Map the extent by reading it.
952 if (!iomap_valid || type != 0) {
954 size = xfs_probe_cluster(inode, page, bh,
956 err = xfs_map_blocks(inode, offset, size,
960 iomap_valid = xfs_iomap_valid(&iomap, offset);
964 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
965 ASSERT(buffer_mapped(bh));
968 xfs_add_to_ioend(inode, bh, offset, type,
969 &ioend, !iomap_valid);
975 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
976 (unmapped || startio)) {
983 } while (offset += len, ((bh = bh->b_this_page) != head));
985 if (uptodate && bh == head)
986 SetPageUptodate(page);
989 xfs_start_page_writeback(page, wbc, 1, count);
991 if (ioend && iomap_valid) {
992 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
994 tlast = min_t(pgoff_t, offset, last_index);
995 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
996 wbc, startio, all_bh, tlast);
1000 xfs_submit_ioend(iohead);
1006 xfs_cancel_ioend(iohead);
1009 * If it's delalloc and we have nowhere to put it,
1010 * throw it away, unless the lower layers told
1013 if (err != -EAGAIN) {
1015 block_invalidatepage(page, 0);
1016 ClearPageUptodate(page);
1023 struct inode *inode,
1025 unsigned long blocks,
1026 struct buffer_head *bh_result,
1029 bmapi_flags_t flags)
1031 vnode_t *vp = LINVFS_GET_VP(inode);
1038 offset = (xfs_off_t)iblock << inode->i_blkbits;
1040 size = (ssize_t) min_t(xfs_off_t, LONG_MAX,
1041 (xfs_off_t)blocks << inode->i_blkbits);
1043 size = 1 << inode->i_blkbits;
1045 VOP_BMAP(vp, offset, size,
1046 create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
1053 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1057 /* For unwritten extents do not report a disk address on
1058 * the read case (treat as if we're reading into a hole).
1060 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1061 delta = offset - iomap.iomap_offset;
1062 delta >>= inode->i_blkbits;
1064 bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
1066 BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
1067 bh_result->b_blocknr = bn;
1068 set_buffer_mapped(bh_result);
1070 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1072 bh_result->b_private = inode;
1073 set_buffer_unwritten(bh_result);
1074 set_buffer_delay(bh_result);
1078 /* If this is a realtime file, data might be on a new device */
1079 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1081 /* If we previously allocated a block out beyond eof and
1082 * we are now coming back to use it then we will need to
1083 * flag it as new even if it has a disk address.
1086 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1087 (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
1088 set_buffer_new(bh_result);
1090 if (iomap.iomap_flags & IOMAP_DELAY) {
1093 set_buffer_uptodate(bh_result);
1094 set_buffer_mapped(bh_result);
1095 set_buffer_delay(bh_result);
1100 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1101 offset = min_t(xfs_off_t,
1102 iomap.iomap_bsize - iomap.iomap_delta,
1103 (xfs_off_t)blocks << inode->i_blkbits);
1104 bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset);
1112 struct inode *inode,
1114 struct buffer_head *bh_result,
1117 return __linvfs_get_block(inode, iblock, 0, bh_result,
1118 create, 0, BMAPI_WRITE);
1122 linvfs_get_blocks_direct(
1123 struct inode *inode,
1125 unsigned long max_blocks,
1126 struct buffer_head *bh_result,
1129 return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
1130 create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1134 linvfs_end_io_direct(
1140 xfs_ioend_t *ioend = iocb->private;
1143 * Non-NULL private data means we need to issue a transaction to
1144 * convert a range from unwritten to written extents. This needs
1145 * to happen from process contect but aio+dio I/O completion
1146 * happens from irq context so we need to defer it to a workqueue.
1147 * This is not nessecary for synchronous direct I/O, but we do
1148 * it anyway to keep the code uniform and simpler.
1150 * The core direct I/O code might be changed to always call the
1151 * completion handler in the future, in which case all this can
1154 if (private && size > 0) {
1155 ioend->io_offset = offset;
1156 ioend->io_size = size;
1157 xfs_finish_ioend(ioend);
1160 xfs_destroy_ioend(ioend);
1164 * blockdev_direct_IO can return an error even afer the I/O
1165 * completion handler was called. Thus we need to protect
1166 * against double-freeing.
1168 iocb->private = NULL;
1175 const struct iovec *iov,
1177 unsigned long nr_segs)
1179 struct file *file = iocb->ki_filp;
1180 struct inode *inode = file->f_mapping->host;
1181 vnode_t *vp = LINVFS_GET_VP(inode);
1187 VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
1191 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1193 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1194 iomap.iomap_target->bt_bdev,
1195 iov, offset, nr_segs,
1196 linvfs_get_blocks_direct,
1197 linvfs_end_io_direct);
1199 if (unlikely(ret <= 0 && iocb->private))
1200 xfs_destroy_ioend(iocb->private);
1207 struct address_space *mapping,
1210 struct inode *inode = (struct inode *)mapping->host;
1211 vnode_t *vp = LINVFS_GET_VP(inode);
1214 vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);
1216 VOP_RWLOCK(vp, VRWLOCK_READ);
1217 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
1218 VOP_RWUNLOCK(vp, VRWLOCK_READ);
1219 return generic_block_bmap(mapping, block, linvfs_get_block);
1224 struct file *unused,
1227 return mpage_readpage(page, linvfs_get_block);
1232 struct file *unused,
1233 struct address_space *mapping,
1234 struct list_head *pages,
1237 return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
1241 xfs_count_page_state(
1247 struct buffer_head *bh, *head;
1249 *delalloc = *unmapped = *unwritten = 0;
1251 bh = head = page_buffers(page);
1253 if (buffer_uptodate(bh) && !buffer_mapped(bh))
1255 else if (buffer_unwritten(bh) && !buffer_delay(bh))
1256 clear_buffer_unwritten(bh);
1257 else if (buffer_unwritten(bh))
1259 else if (buffer_delay(bh))
1261 } while ((bh = bh->b_this_page) != head);
1266 * writepage: Called from one of two places:
1268 * 1. we are flushing a delalloc buffer head.
1270 * 2. we are writing out a dirty page. Typically the page dirty
1271 * state is cleared before we get here. In this case is it
1272 * conceivable we have no buffer heads.
1274 * For delalloc space on the page we need to allocate space and
1275 * flush it. For unmapped buffer heads on the page we should
1276 * allocate space if the page is uptodate. For any other dirty
1277 * buffer heads on the page we should flush them.
1279 * If we detect that a transaction would be required to flush
1280 * the page, we have to check the process flags first, if we
1281 * are already in a transaction or disk I/O during allocations
1282 * is off, we need to fail the writepage and redirty the page.
1288 struct writeback_control *wbc)
1292 int delalloc, unmapped, unwritten;
1293 struct inode *inode = page->mapping->host;
1295 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1298 * We need a transaction if:
1299 * 1. There are delalloc buffers on the page
1300 * 2. The page is uptodate and we have unmapped buffers
1301 * 3. The page is uptodate and we have no buffers
1302 * 4. There are unwritten buffers on the page
1305 if (!page_has_buffers(page)) {
1309 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1310 if (!PageUptodate(page))
1312 need_trans = delalloc + unmapped + unwritten;
1316 * If we need a transaction and the process flags say
1317 * we are already in a transaction, or no IO is allowed
1318 * then mark the page dirty again and leave the page
1321 if (PFLAGS_TEST_FSTRANS() && need_trans)
1325 * Delay hooking up buffer heads until we have
1326 * made our go/no-go decision.
1328 if (!page_has_buffers(page))
1329 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1332 * Convert delayed allocate, unwritten or unmapped space
1333 * to real space and flush out to disk.
1335 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1336 if (error == -EAGAIN)
1338 if (unlikely(error < 0))
1344 redirty_page_for_writepage(wbc, page);
1353 linvfs_invalidate_page(
1355 unsigned long offset)
1357 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1358 page->mapping->host, page, offset);
1359 return block_invalidatepage(page, offset);
1363 * Called to move a page into cleanable state - and from there
1364 * to be released. Possibly the page is already clean. We always
1365 * have buffer heads in this call.
1367 * Returns 0 if the page is ok to release, 1 otherwise.
1369 * Possible scenarios are:
1371 * 1. We are being called to release a page which has been written
1372 * to via regular I/O. buffer heads will be dirty and possibly
1373 * delalloc. If no delalloc buffer heads in this case then we
1374 * can just return zero.
1376 * 2. We are called to release a page which has been written via
1377 * mmap, all we need to do is ensure there is no delalloc
1378 * state in the buffer heads, if not we can let the caller
1379 * free them and we should come back later via writepage.
1382 linvfs_release_page(
1386 struct inode *inode = page->mapping->host;
1387 int dirty, delalloc, unmapped, unwritten;
1388 struct writeback_control wbc = {
1389 .sync_mode = WB_SYNC_ALL,
1393 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1395 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1396 if (!delalloc && !unwritten)
1399 if (!(gfp_mask & __GFP_FS))
1402 /* If we are already inside a transaction or the thread cannot
1403 * do I/O, we cannot release this page.
1405 if (PFLAGS_TEST_FSTRANS())
1409 * Convert delalloc space to real space, do not flush the
1410 * data out to disk, that will be done by the caller.
1411 * Never need to allocate space here - we will always
1412 * come back to writepage in that case.
1414 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1415 if (dirty == 0 && !unwritten)
1420 return try_to_free_buffers(page);
1424 linvfs_prepare_write(
1430 return block_prepare_write(page, from, to, linvfs_get_block);
1433 struct address_space_operations linvfs_aops = {
1434 .readpage = linvfs_readpage,
1435 .readpages = linvfs_readpages,
1436 .writepage = linvfs_writepage,
1437 .sync_page = block_sync_page,
1438 .releasepage = linvfs_release_page,
1439 .invalidatepage = linvfs_invalidate_page,
1440 .prepare_write = linvfs_prepare_write,
1441 .commit_write = generic_commit_write,
1442 .bmap = linvfs_bmap,
1443 .direct_IO = linvfs_direct_IO,