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 struct inode
*inode
= iocb
->ki_filp
->f_dentry
->d_inode
;
158 ASSERT(!private || inode
== (struct inode
*)private);
160 /* private indicates an unwritten extent lay beneath this IO */
161 if (private && size
> 0) {
162 vnode_t
*vp
= LINVFS_GET_VP(inode
);
165 VOP_BMAP(vp
, offset
, size
, BMAPI_UNWRITTEN
, NULL
, NULL
, error
);
177 vnode_t
*vp
= LINVFS_GET_VP(inode
);
178 int error
, nmaps
= 1;
180 VOP_BMAP(vp
, offset
, count
, flags
, mapp
, &nmaps
, error
);
181 if (!error
&& (flags
& (BMAPI_WRITE
|BMAPI_ALLOCATE
)))
187 * Finds the corresponding mapping in block @map array of the
188 * given @offset within a @page.
194 unsigned long offset
)
196 loff_t full_offset
; /* offset from start of file */
198 ASSERT(offset
< PAGE_CACHE_SIZE
);
200 full_offset
= page
->index
; /* NB: using 64bit number */
201 full_offset
<<= PAGE_CACHE_SHIFT
; /* offset from file start */
202 full_offset
+= offset
; /* offset from page start */
204 if (full_offset
< iomapp
->iomap_offset
)
206 if (iomapp
->iomap_offset
+ (iomapp
->iomap_bsize
-1) >= full_offset
)
214 struct buffer_head
*bh
,
215 unsigned long offset
,
223 ASSERT(!(iomapp
->iomap_flags
& IOMAP_HOLE
));
224 ASSERT(!(iomapp
->iomap_flags
& IOMAP_DELAY
));
225 ASSERT(iomapp
->iomap_bn
!= IOMAP_DADDR_NULL
);
228 delta
<<= PAGE_CACHE_SHIFT
;
230 delta
-= iomapp
->iomap_offset
;
231 delta
>>= block_bits
;
233 sector_shift
= block_bits
- BBSHIFT
;
234 bn
= iomapp
->iomap_bn
>> sector_shift
;
236 BUG_ON(!bn
&& !(iomapp
->iomap_flags
& IOMAP_REALTIME
));
237 ASSERT((bn
<< sector_shift
) >= iomapp
->iomap_bn
);
241 bh
->b_bdev
= iomapp
->iomap_target
->pbr_bdev
;
242 set_buffer_mapped(bh
);
243 clear_buffer_delay(bh
);
247 * Look for a page at index which is unlocked and contains our
248 * unwritten extent flagged buffers at its head. Returns page
249 * locked and with an extra reference count, and length of the
250 * unwritten extent component on this page that we can write,
251 * in units of filesystem blocks.
254 xfs_probe_unwritten_page(
255 struct address_space
*mapping
,
259 unsigned long max_offset
,
265 page
= find_trylock_page(mapping
, index
);
268 if (PageWriteback(page
))
271 if (page
->mapping
&& page_has_buffers(page
)) {
272 struct buffer_head
*bh
, *head
;
273 unsigned long p_offset
= 0;
276 bh
= head
= page_buffers(page
);
278 if (!buffer_unwritten(bh
) || !buffer_uptodate(bh
))
280 if (!xfs_offset_to_map(page
, iomapp
, p_offset
))
282 if (p_offset
>= max_offset
)
284 xfs_map_at_offset(page
, bh
, p_offset
, bbits
, iomapp
);
285 set_buffer_unwritten_io(bh
);
287 p_offset
+= bh
->b_size
;
289 } while ((bh
= bh
->b_this_page
) != head
);
301 * Look for a page at index which is unlocked and not mapped
302 * yet - clustering for mmap write case.
305 xfs_probe_unmapped_page(
306 struct address_space
*mapping
,
308 unsigned int pg_offset
)
313 page
= find_trylock_page(mapping
, index
);
316 if (PageWriteback(page
))
319 if (page
->mapping
&& PageDirty(page
)) {
320 if (page_has_buffers(page
)) {
321 struct buffer_head
*bh
, *head
;
323 bh
= head
= page_buffers(page
);
325 if (buffer_mapped(bh
) || !buffer_uptodate(bh
))
328 if (ret
>= pg_offset
)
330 } while ((bh
= bh
->b_this_page
) != head
);
332 ret
= PAGE_CACHE_SIZE
;
341 xfs_probe_unmapped_cluster(
343 struct page
*startpage
,
344 struct buffer_head
*bh
,
345 struct buffer_head
*head
)
347 pgoff_t tindex
, tlast
, tloff
;
348 unsigned int pg_offset
, len
, total
= 0;
349 struct address_space
*mapping
= inode
->i_mapping
;
351 /* First sum forwards in this page */
353 if (buffer_mapped(bh
))
356 } while ((bh
= bh
->b_this_page
) != head
);
358 /* If we reached the end of the page, sum forwards in
362 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
363 /* Prune this back to avoid pathological behavior */
364 tloff
= min(tlast
, startpage
->index
+ 64);
365 for (tindex
= startpage
->index
+ 1; tindex
< tloff
; tindex
++) {
366 len
= xfs_probe_unmapped_page(mapping
, tindex
,
372 if (tindex
== tlast
&&
373 (pg_offset
= i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
374 total
+= xfs_probe_unmapped_page(mapping
,
382 * Probe for a given page (index) in the inode and test if it is delayed
383 * and without unwritten buffers. Returns page locked and with an extra
387 xfs_probe_delalloc_page(
393 page
= find_trylock_page(inode
->i_mapping
, index
);
396 if (PageWriteback(page
))
399 if (page
->mapping
&& page_has_buffers(page
)) {
400 struct buffer_head
*bh
, *head
;
403 bh
= head
= page_buffers(page
);
405 if (buffer_unwritten(bh
)) {
408 } else if (buffer_delay(bh
)) {
411 } while ((bh
= bh
->b_this_page
) != head
);
425 struct page
*start_page
,
426 struct buffer_head
*head
,
427 struct buffer_head
*curr
,
428 unsigned long p_offset
,
431 struct writeback_control
*wbc
,
435 struct buffer_head
*bh
= curr
;
439 unsigned long nblocks
= 0;
441 offset
= start_page
->index
;
442 offset
<<= PAGE_CACHE_SHIFT
;
445 /* get an "empty" pagebuf to manage IO completion
446 * Proper values will be set before returning */
447 pb
= pagebuf_lookup(iomapp
->iomap_target
, 0, 0, 0);
451 /* Take a reference to the inode to prevent it from
452 * being reclaimed while we have outstanding unwritten
455 if ((igrab(inode
)) != inode
) {
460 /* Set the count to 1 initially, this will stop an I/O
461 * completion callout which happens before we have started
462 * all the I/O from calling pagebuf_iodone too early.
464 atomic_set(&pb
->pb_io_remaining
, 1);
466 /* First map forwards in the page consecutive buffers
467 * covering this unwritten extent
470 if (!buffer_unwritten(bh
))
472 tmp
= xfs_offset_to_map(start_page
, iomapp
, p_offset
);
475 xfs_map_at_offset(start_page
, bh
, p_offset
, block_bits
, iomapp
);
476 set_buffer_unwritten_io(bh
);
478 p_offset
+= bh
->b_size
;
480 } while ((bh
= bh
->b_this_page
) != head
);
482 atomic_add(nblocks
, &pb
->pb_io_remaining
);
484 /* If we reached the end of the page, map forwards in any
485 * following pages which are also covered by this extent.
488 struct address_space
*mapping
= inode
->i_mapping
;
489 pgoff_t tindex
, tloff
, tlast
;
491 unsigned int pg_offset
, bbits
= inode
->i_blkbits
;
494 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
495 tloff
= (iomapp
->iomap_offset
+ iomapp
->iomap_bsize
) >> PAGE_CACHE_SHIFT
;
496 tloff
= min(tlast
, tloff
);
497 for (tindex
= start_page
->index
+ 1; tindex
< tloff
; tindex
++) {
498 page
= xfs_probe_unwritten_page(mapping
,
500 PAGE_CACHE_SIZE
, &bs
, bbits
);
504 atomic_add(bs
, &pb
->pb_io_remaining
);
505 xfs_convert_page(inode
, page
, iomapp
, wbc
, pb
,
507 /* stop if converting the next page might add
508 * enough blocks that the corresponding byte
509 * count won't fit in our ulong page buf length */
510 if (nblocks
>= ((ULONG_MAX
- PAGE_SIZE
) >> block_bits
))
514 if (tindex
== tlast
&&
515 (pg_offset
= (i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1)))) {
516 page
= xfs_probe_unwritten_page(mapping
,
518 pg_offset
, &bs
, bbits
);
521 atomic_add(bs
, &pb
->pb_io_remaining
);
522 xfs_convert_page(inode
, page
, iomapp
, wbc
, pb
,
524 if (nblocks
>= ((ULONG_MAX
- PAGE_SIZE
) >> block_bits
))
531 size
= nblocks
; /* NB: using 64bit number here */
532 size
<<= block_bits
; /* convert fsb's to byte range */
536 XFS_BUF_SET_SIZE(pb
, size
);
537 XFS_BUF_SET_COUNT(pb
, size
);
538 XFS_BUF_SET_OFFSET(pb
, offset
);
539 XFS_BUF_SET_FSPRIVATE(pb
, LINVFS_GET_VP(inode
));
540 XFS_BUF_SET_IODONE_FUNC(pb
, linvfs_unwritten_convert
);
542 if (atomic_dec_and_test(&pb
->pb_io_remaining
) == 1) {
543 pagebuf_iodone(pb
, 1, 1);
552 struct writeback_control
*wbc
,
553 struct buffer_head
*bh_arr
[],
558 struct buffer_head
*bh
;
561 BUG_ON(PageWriteback(page
));
563 set_page_writeback(page
);
565 clear_page_dirty(page
);
569 for (i
= 0; i
< bh_count
; i
++) {
571 mark_buffer_async_write(bh
);
572 if (buffer_unwritten(bh
))
573 set_buffer_unwritten_io(bh
);
574 set_buffer_uptodate(bh
);
575 clear_buffer_dirty(bh
);
578 for (i
= 0; i
< bh_count
; i
++)
579 submit_bh(WRITE
, bh_arr
[i
]);
581 if (probed_page
&& clear_dirty
)
582 wbc
->nr_to_write
--; /* Wrote an "extra" page */
587 * Allocate & map buffers for page given the extent map. Write it out.
588 * except for the original page of a writepage, this is called on
589 * delalloc/unwritten pages only, for the original page it is possible
590 * that the page has no mapping at all.
597 struct writeback_control
*wbc
,
602 struct buffer_head
*bh_arr
[MAX_BUF_PER_PAGE
], *bh
, *head
;
603 xfs_iomap_t
*mp
= iomapp
, *tmp
;
604 unsigned long offset
, end_offset
;
606 int bbits
= inode
->i_blkbits
;
609 end_offset
= (i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1));
612 * page_dirty is initially a count of buffers on the page before
613 * EOF and is decrememted as we move each into a cleanable state.
615 len
= 1 << inode
->i_blkbits
;
616 end_offset
= max(end_offset
, PAGE_CACHE_SIZE
);
617 end_offset
= roundup(end_offset
, len
);
618 page_dirty
= end_offset
/ len
;
621 bh
= head
= page_buffers(page
);
623 if (offset
>= end_offset
)
625 if (!(PageUptodate(page
) || buffer_uptodate(bh
)))
627 if (buffer_mapped(bh
) && all_bh
&&
628 !(buffer_unwritten(bh
) || buffer_delay(bh
))) {
631 bh_arr
[index
++] = bh
;
636 tmp
= xfs_offset_to_map(page
, mp
, offset
);
639 ASSERT(!(tmp
->iomap_flags
& IOMAP_HOLE
));
640 ASSERT(!(tmp
->iomap_flags
& IOMAP_DELAY
));
642 /* If this is a new unwritten extent buffer (i.e. one
643 * that we haven't passed in private data for, we must
644 * now map this buffer too.
646 if (buffer_unwritten(bh
) && !bh
->b_end_io
) {
647 ASSERT(tmp
->iomap_flags
& IOMAP_UNWRITTEN
);
648 xfs_map_unwritten(inode
, page
, head
, bh
, offset
,
649 bbits
, tmp
, wbc
, startio
, all_bh
);
650 } else if (! (buffer_unwritten(bh
) && buffer_locked(bh
))) {
651 xfs_map_at_offset(page
, bh
, offset
, bbits
, tmp
);
652 if (buffer_unwritten(bh
)) {
653 set_buffer_unwritten_io(bh
);
654 bh
->b_private
= private;
659 bh_arr
[index
++] = bh
;
661 set_buffer_dirty(bh
);
663 mark_buffer_dirty(bh
);
666 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
668 if (startio
&& index
) {
669 xfs_submit_page(page
, wbc
, bh_arr
, index
, 1, !page_dirty
);
676 * Convert & write out a cluster of pages in the same extent as defined
677 * by mp and following the start page.
684 struct writeback_control
*wbc
,
691 for (; tindex
<= tlast
; tindex
++) {
692 page
= xfs_probe_delalloc_page(inode
, tindex
);
695 xfs_convert_page(inode
, page
, iomapp
, wbc
, NULL
,
701 * Calling this without startio set means we are being asked to make a dirty
702 * page ready for freeing it's buffers. When called with startio set then
703 * we are coming from writepage.
705 * When called with startio set it is important that we write the WHOLE
707 * The bh->b_state's cannot know if any of the blocks or which block for
708 * that matter are dirty due to mmap writes, and therefore bh uptodate is
709 * only vaild if the page itself isn't completely uptodate. Some layers
710 * may clear the page dirty flag prior to calling write page, under the
711 * assumption the entire page will be written out; by not writing out the
712 * whole page the page can be reused before all valid dirty data is
713 * written out. Note: in the case of a page that has been dirty'd by
714 * mapwrite and but partially setup by block_prepare_write the
715 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
716 * valid state, thus the whole page must be written out thing.
720 xfs_page_state_convert(
723 struct writeback_control
*wbc
,
725 int unmapped
) /* also implies page uptodate */
727 struct buffer_head
*bh_arr
[MAX_BUF_PER_PAGE
], *bh
, *head
;
728 xfs_iomap_t
*iomp
, iomap
;
730 unsigned long p_offset
= 0;
731 __uint64_t end_offset
;
732 pgoff_t end_index
, last_index
, tlast
;
733 int len
, err
, i
, cnt
= 0, uptodate
= 1;
737 /* wait for other IO threads? */
738 flags
= (startio
&& wbc
->sync_mode
!= WB_SYNC_NONE
) ? 0 : BMAPI_TRYLOCK
;
740 /* Is this page beyond the end of the file? */
741 offset
= i_size_read(inode
);
742 end_index
= offset
>> PAGE_CACHE_SHIFT
;
743 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
744 if (page
->index
>= end_index
) {
745 if ((page
->index
>= end_index
+ 1) ||
746 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
752 end_offset
= min_t(unsigned long long,
753 (loff_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
, offset
);
754 offset
= (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
;
757 * page_dirty is initially a count of buffers on the page before
758 * EOF and is decrememted as we move each into a cleanable state.
760 len
= 1 << inode
->i_blkbits
;
761 p_offset
= max(p_offset
, PAGE_CACHE_SIZE
);
762 p_offset
= roundup(p_offset
, len
);
763 page_dirty
= p_offset
/ len
;
767 bh
= head
= page_buffers(page
);
770 if (offset
>= end_offset
)
772 if (!buffer_uptodate(bh
))
774 if (!(PageUptodate(page
) || buffer_uptodate(bh
)) && !startio
)
778 iomp
= xfs_offset_to_map(page
, &iomap
, p_offset
);
782 * First case, map an unwritten extent and prepare for
783 * extent state conversion transaction on completion.
785 if (buffer_unwritten(bh
)) {
789 err
= xfs_map_blocks(inode
, offset
, len
, &iomap
,
790 BMAPI_READ
|BMAPI_IGNSTATE
);
794 iomp
= xfs_offset_to_map(page
, &iomap
,
799 err
= xfs_map_unwritten(inode
, page
,
801 inode
->i_blkbits
, iomp
,
802 wbc
, startio
, unmapped
);
807 set_bit(BH_Lock
, &bh
->b_state
);
809 BUG_ON(!buffer_locked(bh
));
814 * Second case, allocate space for a delalloc buffer.
815 * We can return EAGAIN here in the release page case.
817 } else if (buffer_delay(bh
)) {
819 err
= xfs_map_blocks(inode
, offset
, len
, &iomap
,
820 BMAPI_ALLOCATE
| flags
);
824 iomp
= xfs_offset_to_map(page
, &iomap
,
828 xfs_map_at_offset(page
, bh
, p_offset
,
829 inode
->i_blkbits
, iomp
);
833 set_buffer_dirty(bh
);
835 mark_buffer_dirty(bh
);
839 } else if ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
840 (unmapped
|| startio
)) {
842 if (!buffer_mapped(bh
)) {
846 * Getting here implies an unmapped buffer
847 * was found, and we are in a path where we
848 * need to write the whole page out.
851 size
= xfs_probe_unmapped_cluster(
852 inode
, page
, bh
, head
);
853 err
= xfs_map_blocks(inode
, offset
,
855 BMAPI_WRITE
|BMAPI_MMAP
);
859 iomp
= xfs_offset_to_map(page
, &iomap
,
863 xfs_map_at_offset(page
,
865 inode
->i_blkbits
, iomp
);
869 set_buffer_dirty(bh
);
871 mark_buffer_dirty(bh
);
875 } else if (startio
) {
876 if (buffer_uptodate(bh
) &&
877 !test_and_set_bit(BH_Lock
, &bh
->b_state
)) {
883 } while (offset
+= len
, p_offset
+= len
,
884 ((bh
= bh
->b_this_page
) != head
));
886 if (uptodate
&& bh
== head
)
887 SetPageUptodate(page
);
890 xfs_submit_page(page
, wbc
, bh_arr
, cnt
, 0, !page_dirty
);
894 offset
= (iomp
->iomap_offset
+ iomp
->iomap_bsize
- 1) >>
896 tlast
= min_t(pgoff_t
, offset
, last_index
);
897 xfs_cluster_write(inode
, page
->index
+ 1, iomp
, wbc
,
898 startio
, unmapped
, tlast
);
904 for (i
= 0; i
< cnt
; i
++) {
905 unlock_buffer(bh_arr
[i
]);
909 * If it's delalloc and we have nowhere to put it,
910 * throw it away, unless the lower layers told
913 if (err
!= -EAGAIN
) {
915 block_invalidatepage(page
, 0);
917 ClearPageUptodate(page
);
926 unsigned long blocks
,
927 struct buffer_head
*bh_result
,
932 vnode_t
*vp
= LINVFS_GET_VP(inode
);
937 loff_t offset
= (loff_t
)iblock
<< inode
->i_blkbits
;
940 size
= blocks
<< inode
->i_blkbits
;
942 size
= 1 << inode
->i_blkbits
;
944 VOP_BMAP(vp
, offset
, size
,
945 create
? flags
: BMAPI_READ
, &iomap
, &retpbbm
, error
);
952 if (iomap
.iomap_bn
!= IOMAP_DADDR_NULL
) {
956 /* For unwritten extents do not report a disk address on
957 * the read case (treat as if we're reading into a hole).
959 if (create
|| !(iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
960 delta
= offset
- iomap
.iomap_offset
;
961 delta
>>= inode
->i_blkbits
;
963 bn
= iomap
.iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
);
965 BUG_ON(!bn
&& !(iomap
.iomap_flags
& IOMAP_REALTIME
));
966 bh_result
->b_blocknr
= bn
;
967 set_buffer_mapped(bh_result
);
969 if (create
&& (iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
971 bh_result
->b_private
= inode
;
972 set_buffer_unwritten(bh_result
);
973 set_buffer_delay(bh_result
);
977 /* If this is a realtime file, data might be on a new device */
978 bh_result
->b_bdev
= iomap
.iomap_target
->pbr_bdev
;
980 /* If we previously allocated a block out beyond eof and
981 * we are now coming back to use it then we will need to
982 * flag it as new even if it has a disk address.
985 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
986 (offset
>= i_size_read(inode
)) || (iomap
.iomap_flags
& IOMAP_NEW
))) {
987 set_buffer_new(bh_result
);
990 if (iomap
.iomap_flags
& IOMAP_DELAY
) {
993 set_buffer_uptodate(bh_result
);
994 set_buffer_mapped(bh_result
);
995 set_buffer_delay(bh_result
);
1000 bh_result
->b_size
= (ssize_t
)min(
1001 (loff_t
)(iomap
.iomap_bsize
- iomap
.iomap_delta
),
1002 (loff_t
)(blocks
<< inode
->i_blkbits
));
1010 struct inode
*inode
,
1012 struct buffer_head
*bh_result
,
1015 return __linvfs_get_block(inode
, iblock
, 0, bh_result
,
1016 create
, 0, BMAPI_WRITE
);
1020 linvfs_get_blocks_direct(
1021 struct inode
*inode
,
1023 unsigned long max_blocks
,
1024 struct buffer_head
*bh_result
,
1027 return __linvfs_get_block(inode
, iblock
, max_blocks
, bh_result
,
1028 create
, 1, BMAPI_WRITE
|BMAPI_DIRECT
);
1035 const struct iovec
*iov
,
1037 unsigned long nr_segs
)
1039 struct file
*file
= iocb
->ki_filp
;
1040 struct inode
*inode
= file
->f_mapping
->host
;
1041 vnode_t
*vp
= LINVFS_GET_VP(inode
);
1046 VOP_BMAP(vp
, offset
, 0, BMAPI_DEVICE
, &iomap
, &maps
, error
);
1050 return blockdev_direct_IO_own_locking(rw
, iocb
, inode
,
1051 iomap
.iomap_target
->pbr_bdev
,
1052 iov
, offset
, nr_segs
,
1053 linvfs_get_blocks_direct
,
1054 linvfs_unwritten_convert_direct
);
1060 struct address_space
*mapping
,
1063 struct inode
*inode
= (struct inode
*)mapping
->host
;
1064 vnode_t
*vp
= LINVFS_GET_VP(inode
);
1067 vn_trace_entry(vp
, "linvfs_bmap", (inst_t
*)__return_address
);
1069 VOP_RWLOCK(vp
, VRWLOCK_READ
);
1070 VOP_FLUSH_PAGES(vp
, (xfs_off_t
)0, -1, 0, FI_REMAPF
, error
);
1071 VOP_RWUNLOCK(vp
, VRWLOCK_READ
);
1072 return generic_block_bmap(mapping
, block
, linvfs_get_block
);
1077 struct file
*unused
,
1080 return mpage_readpage(page
, linvfs_get_block
);
1085 struct file
*unused
,
1086 struct address_space
*mapping
,
1087 struct list_head
*pages
,
1090 return mpage_readpages(mapping
, pages
, nr_pages
, linvfs_get_block
);
1094 xfs_count_page_state(
1100 struct buffer_head
*bh
, *head
;
1102 *delalloc
= *unmapped
= *unwritten
= 0;
1104 bh
= head
= page_buffers(page
);
1106 if (buffer_uptodate(bh
) && !buffer_mapped(bh
))
1108 else if (buffer_unwritten(bh
) && !buffer_delay(bh
))
1109 clear_buffer_unwritten(bh
);
1110 else if (buffer_unwritten(bh
))
1112 else if (buffer_delay(bh
))
1114 } while ((bh
= bh
->b_this_page
) != head
);
1119 * writepage: Called from one of two places:
1121 * 1. we are flushing a delalloc buffer head.
1123 * 2. we are writing out a dirty page. Typically the page dirty
1124 * state is cleared before we get here. In this case is it
1125 * conceivable we have no buffer heads.
1127 * For delalloc space on the page we need to allocate space and
1128 * flush it. For unmapped buffer heads on the page we should
1129 * allocate space if the page is uptodate. For any other dirty
1130 * buffer heads on the page we should flush them.
1132 * If we detect that a transaction would be required to flush
1133 * the page, we have to check the process flags first, if we
1134 * are already in a transaction or disk I/O during allocations
1135 * is off, we need to fail the writepage and redirty the page.
1141 struct writeback_control
*wbc
)
1145 int delalloc
, unmapped
, unwritten
;
1146 struct inode
*inode
= page
->mapping
->host
;
1148 xfs_page_trace(XFS_WRITEPAGE_ENTER
, inode
, page
, 0);
1151 * We need a transaction if:
1152 * 1. There are delalloc buffers on the page
1153 * 2. The page is uptodate and we have unmapped buffers
1154 * 3. The page is uptodate and we have no buffers
1155 * 4. There are unwritten buffers on the page
1158 if (!page_has_buffers(page
)) {
1162 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1163 if (!PageUptodate(page
))
1165 need_trans
= delalloc
+ unmapped
+ unwritten
;
1169 * If we need a transaction and the process flags say
1170 * we are already in a transaction, or no IO is allowed
1171 * then mark the page dirty again and leave the page
1174 if (PFLAGS_TEST_FSTRANS() && need_trans
)
1178 * Delay hooking up buffer heads until we have
1179 * made our go/no-go decision.
1181 if (!page_has_buffers(page
))
1182 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
1185 * Convert delayed allocate, unwritten or unmapped space
1186 * to real space and flush out to disk.
1188 error
= xfs_page_state_convert(inode
, page
, wbc
, 1, unmapped
);
1189 if (error
== -EAGAIN
)
1191 if (unlikely(error
< 0))
1197 redirty_page_for_writepage(wbc
, page
);
1206 * Called to move a page into cleanable state - and from there
1207 * to be released. Possibly the page is already clean. We always
1208 * have buffer heads in this call.
1210 * Returns 0 if the page is ok to release, 1 otherwise.
1212 * Possible scenarios are:
1214 * 1. We are being called to release a page which has been written
1215 * to via regular I/O. buffer heads will be dirty and possibly
1216 * delalloc. If no delalloc buffer heads in this case then we
1217 * can just return zero.
1219 * 2. We are called to release a page which has been written via
1220 * mmap, all we need to do is ensure there is no delalloc
1221 * state in the buffer heads, if not we can let the caller
1222 * free them and we should come back later via writepage.
1225 linvfs_release_page(
1229 struct inode
*inode
= page
->mapping
->host
;
1230 int dirty
, delalloc
, unmapped
, unwritten
;
1231 struct writeback_control wbc
= {
1232 .sync_mode
= WB_SYNC_ALL
,
1236 xfs_page_trace(XFS_RELEASEPAGE_ENTER
, inode
, page
, gfp_mask
);
1238 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1239 if (!delalloc
&& !unwritten
)
1242 if (!(gfp_mask
& __GFP_FS
))
1245 /* If we are already inside a transaction or the thread cannot
1246 * do I/O, we cannot release this page.
1248 if (PFLAGS_TEST_FSTRANS())
1252 * Convert delalloc space to real space, do not flush the
1253 * data out to disk, that will be done by the caller.
1254 * Never need to allocate space here - we will always
1255 * come back to writepage in that case.
1257 dirty
= xfs_page_state_convert(inode
, page
, &wbc
, 0, 0);
1258 if (dirty
== 0 && !unwritten
)
1263 return try_to_free_buffers(page
);
1267 linvfs_prepare_write(
1273 return block_prepare_write(page
, from
, to
, linvfs_get_block
);
1276 struct address_space_operations linvfs_aops
= {
1277 .readpage
= linvfs_readpage
,
1278 .readpages
= linvfs_readpages
,
1279 .writepage
= linvfs_writepage
,
1280 .sync_page
= block_sync_page
,
1281 .releasepage
= linvfs_release_page
,
1282 .prepare_write
= linvfs_prepare_write
,
1283 .commit_write
= generic_commit_write
,
1284 .bmap
= linvfs_bmap
,
1285 .direct_IO
= linvfs_direct_IO
,