2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
39 #include "xfs_trans.h"
40 #include "xfs_dmapi.h"
41 #include "xfs_mount.h"
42 #include "xfs_bmap_btree.h"
43 #include "xfs_alloc_btree.h"
44 #include "xfs_ialloc_btree.h"
45 #include "xfs_alloc.h"
46 #include "xfs_btree.h"
47 #include "xfs_attr_sf.h"
48 #include "xfs_dir_sf.h"
49 #include "xfs_dir2_sf.h"
50 #include "xfs_dinode.h"
51 #include "xfs_inode.h"
52 #include "xfs_error.h"
54 #include "xfs_iomap.h"
55 #include <linux/mpage.h>
56 #include <linux/writeback.h>
58 STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
59 STATIC void xfs_convert_page(struct inode *, struct page *, xfs_iomap_t *,
60 struct writeback_control *wbc, void *, int, int);
62 #if defined(XFS_RW_TRACE)
72 vnode_t *vp = LINVFS_GET_VP(inode);
73 loff_t isize = i_size_read(inode);
74 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
75 int delalloc = -1, unmapped = -1, unwritten = -1;
77 if (page_has_buffers(page))
78 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
80 bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
85 ktrace_enter(ip->i_rwtrace,
86 (void *)((unsigned long)tag),
90 (void *)((unsigned long)mask),
91 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
92 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
93 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
94 (void *)((unsigned long)(isize & 0xffffffff)),
95 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
96 (void *)((unsigned long)(offset & 0xffffffff)),
97 (void *)((unsigned long)delalloc),
98 (void *)((unsigned long)unmapped),
99 (void *)((unsigned long)unwritten),
104 #define xfs_page_trace(tag, inode, page, mask)
108 linvfs_unwritten_done(
109 struct buffer_head *bh,
112 xfs_buf_t *pb = (xfs_buf_t *)bh->b_private;
114 ASSERT(buffer_unwritten(bh));
116 clear_buffer_unwritten(bh);
118 pagebuf_ioerror(pb, EIO);
119 if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
120 pagebuf_iodone(pb, 1, 1);
122 end_buffer_async_write(bh, uptodate);
126 * Issue transactions to convert a buffer range from unwritten
127 * to written extents (buffered IO).
130 linvfs_unwritten_convert(
133 vnode_t *vp = XFS_BUF_FSPRIVATE(bp, vnode_t *);
136 BUG_ON(atomic_read(&bp->pb_hold) < 1);
137 VOP_BMAP(vp, XFS_BUF_OFFSET(bp), XFS_BUF_SIZE(bp),
138 BMAPI_UNWRITTEN, NULL, NULL, error);
139 XFS_BUF_SET_FSPRIVATE(bp, NULL);
140 XFS_BUF_CLR_IODONE_FUNC(bp);
141 XFS_BUF_UNDATAIO(bp);
142 iput(LINVFS_GET_IP(vp));
143 pagebuf_iodone(bp, 0, 0);
147 * Issue transactions to convert a buffer range from unwritten
148 * to written extents (direct IO).
151 linvfs_unwritten_convert_direct(
157 ASSERT(!private || inode == (struct inode *)private);
159 /* private indicates an unwritten extent lay beneath this IO */
160 if (private && size > 0) {
161 vnode_t *vp = LINVFS_GET_VP(inode);
164 VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
176 vnode_t *vp = LINVFS_GET_VP(inode);
177 int error, nmaps = 1;
179 VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
180 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
186 * Finds the corresponding mapping in block @map array of the
187 * given @offset within a @page.
193 unsigned long offset)
195 loff_t full_offset; /* offset from start of file */
197 ASSERT(offset < PAGE_CACHE_SIZE);
199 full_offset = page->index; /* NB: using 64bit number */
200 full_offset <<= PAGE_CACHE_SHIFT; /* offset from file start */
201 full_offset += offset; /* offset from page start */
203 if (full_offset < iomapp->iomap_offset)
205 if (iomapp->iomap_offset + (iomapp->iomap_bsize -1) >= full_offset)
213 struct buffer_head *bh,
214 unsigned long offset,
222 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
223 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
224 ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);
227 delta <<= PAGE_CACHE_SHIFT;
229 delta -= iomapp->iomap_offset;
230 delta >>= block_bits;
232 sector_shift = block_bits - BBSHIFT;
233 bn = iomapp->iomap_bn >> sector_shift;
235 BUG_ON(!bn && !(iomapp->iomap_flags & IOMAP_REALTIME));
236 ASSERT((bn << sector_shift) >= iomapp->iomap_bn);
240 bh->b_bdev = iomapp->iomap_target->pbr_bdev;
241 set_buffer_mapped(bh);
242 clear_buffer_delay(bh);
246 * Look for a page at index which is unlocked and contains our
247 * unwritten extent flagged buffers at its head. Returns page
248 * locked and with an extra reference count, and length of the
249 * unwritten extent component on this page that we can write,
250 * in units of filesystem blocks.
253 xfs_probe_unwritten_page(
254 struct address_space *mapping,
258 unsigned long max_offset,
264 page = find_trylock_page(mapping, index);
267 if (PageWriteback(page))
270 if (page->mapping && page_has_buffers(page)) {
271 struct buffer_head *bh, *head;
272 unsigned long p_offset = 0;
275 bh = head = page_buffers(page);
277 if (!buffer_unwritten(bh) || !buffer_uptodate(bh))
279 if (!xfs_offset_to_map(page, iomapp, p_offset))
281 if (p_offset >= max_offset)
283 xfs_map_at_offset(page, bh, p_offset, bbits, iomapp);
284 set_buffer_unwritten_io(bh);
286 p_offset += bh->b_size;
288 } while ((bh = bh->b_this_page) != head);
300 * Look for a page at index which is unlocked and not mapped
301 * yet - clustering for mmap write case.
304 xfs_probe_unmapped_page(
305 struct address_space *mapping,
307 unsigned int pg_offset)
312 page = find_trylock_page(mapping, index);
315 if (PageWriteback(page))
318 if (page->mapping && PageDirty(page)) {
319 if (page_has_buffers(page)) {
320 struct buffer_head *bh, *head;
322 bh = head = page_buffers(page);
324 if (buffer_mapped(bh) || !buffer_uptodate(bh))
327 if (ret >= pg_offset)
329 } while ((bh = bh->b_this_page) != head);
331 ret = PAGE_CACHE_SIZE;
340 xfs_probe_unmapped_cluster(
342 struct page *startpage,
343 struct buffer_head *bh,
344 struct buffer_head *head)
346 pgoff_t tindex, tlast, tloff;
347 unsigned int pg_offset, len, total = 0;
348 struct address_space *mapping = inode->i_mapping;
350 /* First sum forwards in this page */
352 if (buffer_mapped(bh))
355 } while ((bh = bh->b_this_page) != head);
357 /* If we reached the end of the page, sum forwards in
361 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
362 /* Prune this back to avoid pathological behavior */
363 tloff = min(tlast, startpage->index + 64);
364 for (tindex = startpage->index + 1; tindex < tloff; tindex++) {
365 len = xfs_probe_unmapped_page(mapping, tindex,
371 if (tindex == tlast &&
372 (pg_offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
373 total += xfs_probe_unmapped_page(mapping,
381 * Probe for a given page (index) in the inode and test if it is delayed
382 * and without unwritten buffers. Returns page locked and with an extra
386 xfs_probe_delalloc_page(
392 page = find_trylock_page(inode->i_mapping, index);
395 if (PageWriteback(page))
398 if (page->mapping && page_has_buffers(page)) {
399 struct buffer_head *bh, *head;
402 bh = head = page_buffers(page);
404 if (buffer_unwritten(bh)) {
407 } else if (buffer_delay(bh)) {
410 } while ((bh = bh->b_this_page) != head);
424 struct page *start_page,
425 struct buffer_head *head,
426 struct buffer_head *curr,
427 unsigned long p_offset,
430 struct writeback_control *wbc,
434 struct buffer_head *bh = curr;
438 unsigned long nblocks = 0;
440 offset = start_page->index;
441 offset <<= PAGE_CACHE_SHIFT;
444 /* get an "empty" pagebuf to manage IO completion
445 * Proper values will be set before returning */
446 pb = pagebuf_lookup(iomapp->iomap_target, 0, 0, 0);
450 /* Take a reference to the inode to prevent it from
451 * being reclaimed while we have outstanding unwritten
454 if ((igrab(inode)) != inode) {
459 /* Set the count to 1 initially, this will stop an I/O
460 * completion callout which happens before we have started
461 * all the I/O from calling pagebuf_iodone too early.
463 atomic_set(&pb->pb_io_remaining, 1);
465 /* First map forwards in the page consecutive buffers
466 * covering this unwritten extent
469 if (!buffer_unwritten(bh))
471 tmp = xfs_offset_to_map(start_page, iomapp, p_offset);
474 xfs_map_at_offset(start_page, bh, p_offset, block_bits, iomapp);
475 set_buffer_unwritten_io(bh);
477 p_offset += bh->b_size;
479 } while ((bh = bh->b_this_page) != head);
481 atomic_add(nblocks, &pb->pb_io_remaining);
483 /* If we reached the end of the page, map forwards in any
484 * following pages which are also covered by this extent.
487 struct address_space *mapping = inode->i_mapping;
488 pgoff_t tindex, tloff, tlast;
490 unsigned int pg_offset, bbits = inode->i_blkbits;
493 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
494 tloff = (iomapp->iomap_offset + iomapp->iomap_bsize) >> PAGE_CACHE_SHIFT;
495 tloff = min(tlast, tloff);
496 for (tindex = start_page->index + 1; tindex < tloff; tindex++) {
497 page = xfs_probe_unwritten_page(mapping,
499 PAGE_CACHE_SIZE, &bs, bbits);
503 atomic_add(bs, &pb->pb_io_remaining);
504 xfs_convert_page(inode, page, iomapp, wbc, pb,
506 /* stop if converting the next page might add
507 * enough blocks that the corresponding byte
508 * count won't fit in our ulong page buf length */
509 if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
513 if (tindex == tlast &&
514 (pg_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1)))) {
515 page = xfs_probe_unwritten_page(mapping,
517 pg_offset, &bs, bbits);
520 atomic_add(bs, &pb->pb_io_remaining);
521 xfs_convert_page(inode, page, iomapp, wbc, pb,
523 if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
530 size = nblocks; /* NB: using 64bit number here */
531 size <<= block_bits; /* convert fsb's to byte range */
535 XFS_BUF_SET_SIZE(pb, size);
536 XFS_BUF_SET_COUNT(pb, size);
537 XFS_BUF_SET_OFFSET(pb, offset);
538 XFS_BUF_SET_FSPRIVATE(pb, LINVFS_GET_VP(inode));
539 XFS_BUF_SET_IODONE_FUNC(pb, linvfs_unwritten_convert);
541 if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
542 pagebuf_iodone(pb, 1, 1);
551 struct writeback_control *wbc,
552 struct buffer_head *bh_arr[],
557 struct buffer_head *bh;
560 BUG_ON(PageWriteback(page));
561 set_page_writeback(page);
563 clear_page_dirty(page);
567 for (i = 0; i < bh_count; i++) {
569 mark_buffer_async_write(bh);
570 if (buffer_unwritten(bh))
571 set_buffer_unwritten_io(bh);
572 set_buffer_uptodate(bh);
573 clear_buffer_dirty(bh);
576 for (i = 0; i < bh_count; i++)
577 submit_bh(WRITE, bh_arr[i]);
579 if (probed_page && clear_dirty)
580 wbc->nr_to_write--; /* Wrote an "extra" page */
582 end_page_writeback(page);
583 wbc->pages_skipped++; /* We didn't write this page */
588 * Allocate & map buffers for page given the extent map. Write it out.
589 * except for the original page of a writepage, this is called on
590 * delalloc/unwritten pages only, for the original page it is possible
591 * that the page has no mapping at all.
598 struct writeback_control *wbc,
603 struct buffer_head *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
604 xfs_iomap_t *mp = iomapp, *tmp;
605 unsigned long end, offset;
607 int i = 0, index = 0;
608 int bbits = inode->i_blkbits;
610 end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
611 if (page->index < end_index) {
612 end = PAGE_CACHE_SIZE;
614 end = i_size_read(inode) & (PAGE_CACHE_SIZE-1);
616 bh = head = page_buffers(page);
621 if (!(PageUptodate(page) || buffer_uptodate(bh)))
623 if (buffer_mapped(bh) && all_bh &&
624 !(buffer_unwritten(bh) || buffer_delay(bh))) {
627 bh_arr[index++] = bh;
631 tmp = xfs_offset_to_map(page, mp, offset);
634 ASSERT(!(tmp->iomap_flags & IOMAP_HOLE));
635 ASSERT(!(tmp->iomap_flags & IOMAP_DELAY));
637 /* If this is a new unwritten extent buffer (i.e. one
638 * that we haven't passed in private data for, we must
639 * now map this buffer too.
641 if (buffer_unwritten(bh) && !bh->b_end_io) {
642 ASSERT(tmp->iomap_flags & IOMAP_UNWRITTEN);
643 xfs_map_unwritten(inode, page, head, bh, offset,
644 bbits, tmp, wbc, startio, all_bh);
645 } else if (! (buffer_unwritten(bh) && buffer_locked(bh))) {
646 xfs_map_at_offset(page, bh, offset, bbits, tmp);
647 if (buffer_unwritten(bh)) {
648 set_buffer_unwritten_io(bh);
649 bh->b_private = private;
654 bh_arr[index++] = bh;
656 set_buffer_dirty(bh);
658 mark_buffer_dirty(bh);
660 } while (i++, (bh = bh->b_this_page) != head);
663 xfs_submit_page(page, wbc, bh_arr, index, 1, index == i);
670 * Convert & write out a cluster of pages in the same extent as defined
671 * by mp and following the start page.
678 struct writeback_control *wbc,
685 for (; tindex <= tlast; tindex++) {
686 page = xfs_probe_delalloc_page(inode, tindex);
689 xfs_convert_page(inode, page, iomapp, wbc, NULL,
695 * Calling this without startio set means we are being asked to make a dirty
696 * page ready for freeing it's buffers. When called with startio set then
697 * we are coming from writepage.
699 * When called with startio set it is important that we write the WHOLE
701 * The bh->b_state's cannot know if any of the blocks or which block for
702 * that matter are dirty due to mmap writes, and therefore bh uptodate is
703 * only vaild if the page itself isn't completely uptodate. Some layers
704 * may clear the page dirty flag prior to calling write page, under the
705 * assumption the entire page will be written out; by not writing out the
706 * whole page the page can be reused before all valid dirty data is
707 * written out. Note: in the case of a page that has been dirty'd by
708 * mapwrite and but partially setup by block_prepare_write the
709 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
710 * valid state, thus the whole page must be written out thing.
714 xfs_page_state_convert(
717 struct writeback_control *wbc,
719 int unmapped) /* also implies page uptodate */
721 struct buffer_head *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
722 xfs_iomap_t *iomp, iomap;
724 unsigned long p_offset = 0;
725 __uint64_t end_offset;
726 pgoff_t end_index, last_index, tlast;
727 int len, err, i, cnt = 0, uptodate = 1;
728 int flags = startio ? 0 : BMAPI_TRYLOCK;
729 int page_dirty, delalloc = 0;
731 /* Is this page beyond the end of the file? */
732 offset = i_size_read(inode);
733 end_index = offset >> PAGE_CACHE_SHIFT;
734 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
735 if (page->index >= end_index) {
736 if ((page->index >= end_index + 1) ||
737 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
743 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
744 end_offset = min_t(unsigned long long,
745 offset + PAGE_CACHE_SIZE, i_size_read(inode));
747 bh = head = page_buffers(page);
751 * page_dirty is initially a count of buffers on the page and
752 * is decrememted as we move each into a cleanable state.
755 page_dirty = PAGE_CACHE_SIZE / len;
758 if (offset >= end_offset)
760 if (!buffer_uptodate(bh))
762 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio)
766 iomp = xfs_offset_to_map(page, &iomap, p_offset);
770 * First case, map an unwritten extent and prepare for
771 * extent state conversion transaction on completion.
773 if (buffer_unwritten(bh)) {
777 err = xfs_map_blocks(inode, offset, len, &iomap,
778 BMAPI_READ|BMAPI_IGNSTATE);
782 iomp = xfs_offset_to_map(page, &iomap,
787 err = xfs_map_unwritten(inode, page,
789 inode->i_blkbits, iomp,
790 wbc, startio, unmapped);
795 set_bit(BH_Lock, &bh->b_state);
797 BUG_ON(!buffer_locked(bh));
802 * Second case, allocate space for a delalloc buffer.
803 * We can return EAGAIN here in the release page case.
805 } else if (buffer_delay(bh)) {
808 err = xfs_map_blocks(inode, offset, len, &iomap,
809 BMAPI_ALLOCATE | flags);
813 iomp = xfs_offset_to_map(page, &iomap,
817 xfs_map_at_offset(page, bh, p_offset,
818 inode->i_blkbits, iomp);
822 set_buffer_dirty(bh);
824 mark_buffer_dirty(bh);
828 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
829 (unmapped || startio)) {
831 if (!buffer_mapped(bh)) {
835 * Getting here implies an unmapped buffer
836 * was found, and we are in a path where we
837 * need to write the whole page out.
840 size = xfs_probe_unmapped_cluster(
841 inode, page, bh, head);
842 err = xfs_map_blocks(inode, offset,
844 BMAPI_WRITE|BMAPI_MMAP);
848 iomp = xfs_offset_to_map(page, &iomap,
852 xfs_map_at_offset(page,
854 inode->i_blkbits, iomp);
858 set_buffer_dirty(bh);
860 mark_buffer_dirty(bh);
864 } else if (startio) {
865 if (buffer_uptodate(bh) &&
866 !test_and_set_bit(BH_Lock, &bh->b_state)) {
872 } while (offset += len, p_offset += len,
873 ((bh = bh->b_this_page) != head));
875 if (uptodate && bh == head)
876 SetPageUptodate(page);
879 xfs_submit_page(page, wbc, bh_arr, cnt, 0, 1);
882 tlast = (iomp->iomap_offset + iomp->iomap_bsize - 1) >>
884 if (delalloc && (tlast > last_index))
886 xfs_cluster_write(inode, page->index + 1, iomp, wbc,
887 startio, unmapped, tlast);
893 for (i = 0; i < cnt; i++) {
894 unlock_buffer(bh_arr[i]);
898 * If it's delalloc and we have nowhere to put it,
899 * throw it away, unless the lower layers told
902 if (err != -EAGAIN) {
904 block_invalidatepage(page, 0);
906 ClearPageUptodate(page);
915 unsigned long blocks,
916 struct buffer_head *bh_result,
921 vnode_t *vp = LINVFS_GET_VP(inode);
926 loff_t offset = (loff_t)iblock << inode->i_blkbits;
929 size = blocks << inode->i_blkbits;
931 size = 1 << inode->i_blkbits;
933 VOP_BMAP(vp, offset, size,
934 create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
941 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
945 /* For unwritten extents do not report a disk address on
946 * the read case (treat as if we're reading into a hole).
948 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
949 delta = offset - iomap.iomap_offset;
950 delta >>= inode->i_blkbits;
952 bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
954 BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
955 bh_result->b_blocknr = bn;
956 set_buffer_mapped(bh_result);
958 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
960 bh_result->b_private = inode;
961 set_buffer_unwritten(bh_result);
962 set_buffer_delay(bh_result);
966 /* If this is a realtime file, data might be on a new device */
967 bh_result->b_bdev = iomap.iomap_target->pbr_bdev;
969 /* If we previously allocated a block out beyond eof and
970 * we are now coming back to use it then we will need to
971 * flag it as new even if it has a disk address.
974 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
975 (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) {
976 set_buffer_new(bh_result);
979 if (iomap.iomap_flags & IOMAP_DELAY) {
982 set_buffer_uptodate(bh_result);
983 set_buffer_mapped(bh_result);
984 set_buffer_delay(bh_result);
989 bh_result->b_size = (ssize_t)min(
990 (loff_t)(iomap.iomap_bsize - iomap.iomap_delta),
991 (loff_t)(blocks << inode->i_blkbits));
1001 struct buffer_head *bh_result,
1004 return __linvfs_get_block(inode, iblock, 0, bh_result,
1005 create, 0, BMAPI_WRITE);
1009 linvfs_get_blocks_direct(
1010 struct inode *inode,
1012 unsigned long max_blocks,
1013 struct buffer_head *bh_result,
1016 return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
1017 create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1024 const struct iovec *iov,
1026 unsigned long nr_segs)
1028 struct file *file = iocb->ki_filp;
1029 struct inode *inode = file->f_mapping->host;
1030 vnode_t *vp = LINVFS_GET_VP(inode);
1035 VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
1039 return blockdev_direct_IO_own_locking(rw, iocb, inode,
1040 iomap.iomap_target->pbr_bdev,
1041 iov, offset, nr_segs,
1042 linvfs_get_blocks_direct,
1043 linvfs_unwritten_convert_direct);
1049 struct address_space *mapping,
1052 struct inode *inode = (struct inode *)mapping->host;
1053 vnode_t *vp = LINVFS_GET_VP(inode);
1056 vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);
1058 VOP_RWLOCK(vp, VRWLOCK_READ);
1059 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
1060 VOP_RWUNLOCK(vp, VRWLOCK_READ);
1061 return generic_block_bmap(mapping, block, linvfs_get_block);
1066 struct file *unused,
1069 return mpage_readpage(page, linvfs_get_block);
1074 struct file *unused,
1075 struct address_space *mapping,
1076 struct list_head *pages,
1079 return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
1083 xfs_count_page_state(
1089 struct buffer_head *bh, *head;
1091 *delalloc = *unmapped = *unwritten = 0;
1093 bh = head = page_buffers(page);
1095 if (buffer_uptodate(bh) && !buffer_mapped(bh))
1097 else if (buffer_unwritten(bh) && !buffer_delay(bh))
1098 clear_buffer_unwritten(bh);
1099 else if (buffer_unwritten(bh))
1101 else if (buffer_delay(bh))
1103 } while ((bh = bh->b_this_page) != head);
1108 * writepage: Called from one of two places:
1110 * 1. we are flushing a delalloc buffer head.
1112 * 2. we are writing out a dirty page. Typically the page dirty
1113 * state is cleared before we get here. In this case is it
1114 * conceivable we have no buffer heads.
1116 * For delalloc space on the page we need to allocate space and
1117 * flush it. For unmapped buffer heads on the page we should
1118 * allocate space if the page is uptodate. For any other dirty
1119 * buffer heads on the page we should flush them.
1121 * If we detect that a transaction would be required to flush
1122 * the page, we have to check the process flags first, if we
1123 * are already in a transaction or disk I/O during allocations
1124 * is off, we need to fail the writepage and redirty the page.
1130 struct writeback_control *wbc)
1134 int delalloc, unmapped, unwritten;
1135 struct inode *inode = page->mapping->host;
1137 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1140 * We need a transaction if:
1141 * 1. There are delalloc buffers on the page
1142 * 2. The page is uptodate and we have unmapped buffers
1143 * 3. The page is uptodate and we have no buffers
1144 * 4. There are unwritten buffers on the page
1147 if (!page_has_buffers(page)) {
1151 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1152 if (!PageUptodate(page))
1154 need_trans = delalloc + unmapped + unwritten;
1158 * If we need a transaction and the process flags say
1159 * we are already in a transaction, or no IO is allowed
1160 * then mark the page dirty again and leave the page
1163 if (PFLAGS_TEST_FSTRANS() && need_trans)
1167 * Delay hooking up buffer heads until we have
1168 * made our go/no-go decision.
1170 if (!page_has_buffers(page))
1171 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1174 * Convert delayed allocate, unwritten or unmapped space
1175 * to real space and flush out to disk.
1177 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1178 if (error == -EAGAIN)
1180 if (unlikely(error < 0))
1186 redirty_page_for_writepage(wbc, page);
1195 * Called to move a page into cleanable state - and from there
1196 * to be released. Possibly the page is already clean. We always
1197 * have buffer heads in this call.
1199 * Returns 0 if the page is ok to release, 1 otherwise.
1201 * Possible scenarios are:
1203 * 1. We are being called to release a page which has been written
1204 * to via regular I/O. buffer heads will be dirty and possibly
1205 * delalloc. If no delalloc buffer heads in this case then we
1206 * can just return zero.
1208 * 2. We are called to release a page which has been written via
1209 * mmap, all we need to do is ensure there is no delalloc
1210 * state in the buffer heads, if not we can let the caller
1211 * free them and we should come back later via writepage.
1214 linvfs_release_page(
1218 struct inode *inode = page->mapping->host;
1219 int dirty, delalloc, unmapped, unwritten;
1220 struct writeback_control wbc = {
1221 .sync_mode = WB_SYNC_ALL,
1225 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1227 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1228 if (!delalloc && !unwritten)
1231 if (!(gfp_mask & __GFP_FS))
1234 /* If we are already inside a transaction or the thread cannot
1235 * do I/O, we cannot release this page.
1237 if (PFLAGS_TEST_FSTRANS())
1241 * Convert delalloc space to real space, do not flush the
1242 * data out to disk, that will be done by the caller.
1243 * Never need to allocate space here - we will always
1244 * come back to writepage in that case.
1246 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1247 if (dirty == 0 && !unwritten)
1252 return try_to_free_buffers(page);
1256 linvfs_prepare_write(
1262 return block_prepare_write(page, from, to, linvfs_get_block);
1265 struct address_space_operations linvfs_aops = {
1266 .readpage = linvfs_readpage,
1267 .readpages = linvfs_readpages,
1268 .writepage = linvfs_writepage,
1269 .sync_page = block_sync_page,
1270 .releasepage = linvfs_release_page,
1271 .prepare_write = linvfs_prepare_write,
1272 .commit_write = generic_commit_write,
1273 .bmap = linvfs_bmap,
1274 .direct_IO = linvfs_direct_IO,