2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_trans.h"
31 #include "xfs_extfree_item.h"
32 #include "xfs_alloc.h"
34 #include "xfs_bmap_util.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_quota.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trace.h"
41 #include "xfs_icache.h"
44 /* Kernel only BMAP related definitions and functions */
47 * Convert the given file system block to a disk block. We have to treat it
48 * differently based on whether the file is a real time file or not, because the
52 xfs_fsb_to_db(struct xfs_inode
*ip
, xfs_fsblock_t fsb
)
54 return (XFS_IS_REALTIME_INODE(ip
) ? \
55 (xfs_daddr_t
)XFS_FSB_TO_BB((ip
)->i_mount
, (fsb
)) : \
56 XFS_FSB_TO_DADDR((ip
)->i_mount
, (fsb
)));
60 * Routine to zero an extent on disk allocated to the specific inode.
62 * The VFS functions take a linearised filesystem block offset, so we have to
63 * convert the sparse xfs fsb to the right format first.
64 * VFS types are real funky, too.
69 xfs_fsblock_t start_fsb
,
72 struct xfs_mount
*mp
= ip
->i_mount
;
73 xfs_daddr_t sector
= xfs_fsb_to_db(ip
, start_fsb
);
74 sector_t block
= XFS_BB_TO_FSBT(mp
, sector
);
76 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip
)),
77 block
<< (mp
->m_super
->s_blocksize_bits
- 9),
78 count_fsb
<< (mp
->m_super
->s_blocksize_bits
- 9),
83 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
84 * caller. Frees all the extents that need freeing, which must be done
85 * last due to locking considerations. We never free any extents in
86 * the first transaction.
88 * If an inode *ip is provided, rejoin it to the transaction if
89 * the transaction was committed.
93 struct xfs_trans
**tp
, /* transaction pointer addr */
94 struct xfs_bmap_free
*flist
, /* i/o: list extents to free */
97 struct xfs_efd_log_item
*efd
; /* extent free data */
98 struct xfs_efi_log_item
*efi
; /* extent free intention */
99 int error
; /* error return value */
100 int committed
;/* xact committed or not */
101 struct xfs_bmap_free_item
*free
; /* free extent item */
102 struct xfs_bmap_free_item
*next
; /* next item on free list */
104 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
105 if (flist
->xbf_count
== 0)
108 efi
= xfs_trans_get_efi(*tp
, flist
->xbf_count
);
109 for (free
= flist
->xbf_first
; free
; free
= free
->xbfi_next
)
110 xfs_trans_log_efi_extent(*tp
, efi
, free
->xbfi_startblock
,
111 free
->xbfi_blockcount
);
113 error
= __xfs_trans_roll(tp
, ip
, &committed
);
116 * If the transaction was committed, drop the EFD reference
117 * since we're bailing out of here. The other reference is
118 * dropped when the EFI hits the AIL.
120 * If the transaction was not committed, the EFI is freed by the
121 * EFI item unlock handler on abort. Also, we have a new
122 * transaction so we should return committed=1 even though we're
123 * returning an error.
126 xfs_efi_release(efi
);
127 xfs_force_shutdown((*tp
)->t_mountp
,
128 SHUTDOWN_META_IO_ERROR
);
134 * Get an EFD and free each extent in the list, logging to the EFD in
135 * the process. The remaining bmap free list is cleaned up by the caller
138 efd
= xfs_trans_get_efd(*tp
, efi
, flist
->xbf_count
);
139 for (free
= flist
->xbf_first
; free
!= NULL
; free
= next
) {
140 next
= free
->xbfi_next
;
142 error
= xfs_trans_free_extent(*tp
, efd
, free
->xbfi_startblock
,
143 free
->xbfi_blockcount
);
147 xfs_bmap_del_free(flist
, NULL
, free
);
155 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
157 xfs_alloctype_t atype
= 0; /* type for allocation routines */
158 int error
; /* error return value */
159 xfs_mount_t
*mp
; /* mount point structure */
160 xfs_extlen_t prod
= 0; /* product factor for allocators */
161 xfs_extlen_t ralen
= 0; /* realtime allocation length */
162 xfs_extlen_t align
; /* minimum allocation alignment */
165 mp
= ap
->ip
->i_mount
;
166 align
= xfs_get_extsz_hint(ap
->ip
);
167 prod
= align
/ mp
->m_sb
.sb_rextsize
;
168 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
169 align
, 1, ap
->eof
, 0,
170 ap
->conv
, &ap
->offset
, &ap
->length
);
174 ASSERT(ap
->length
% mp
->m_sb
.sb_rextsize
== 0);
177 * If the offset & length are not perfectly aligned
178 * then kill prod, it will just get us in trouble.
180 if (do_mod(ap
->offset
, align
) || ap
->length
% align
)
183 * Set ralen to be the actual requested length in rtextents.
185 ralen
= ap
->length
/ mp
->m_sb
.sb_rextsize
;
187 * If the old value was close enough to MAXEXTLEN that
188 * we rounded up to it, cut it back so it's valid again.
189 * Note that if it's a really large request (bigger than
190 * MAXEXTLEN), we don't hear about that number, and can't
191 * adjust the starting point to match it.
193 if (ralen
* mp
->m_sb
.sb_rextsize
>= MAXEXTLEN
)
194 ralen
= MAXEXTLEN
/ mp
->m_sb
.sb_rextsize
;
197 * Lock out modifications to both the RT bitmap and summary inodes
199 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
200 xfs_trans_ijoin(ap
->tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
201 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
);
202 xfs_trans_ijoin(ap
->tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
205 * If it's an allocation to an empty file at offset 0,
206 * pick an extent that will space things out in the rt area.
208 if (ap
->eof
&& ap
->offset
== 0) {
209 xfs_rtblock_t
uninitialized_var(rtx
); /* realtime extent no */
211 error
= xfs_rtpick_extent(mp
, ap
->tp
, ralen
, &rtx
);
214 ap
->blkno
= rtx
* mp
->m_sb
.sb_rextsize
;
219 xfs_bmap_adjacent(ap
);
222 * Realtime allocation, done through xfs_rtallocate_extent.
224 atype
= ap
->blkno
== 0 ? XFS_ALLOCTYPE_ANY_AG
: XFS_ALLOCTYPE_NEAR_BNO
;
225 do_div(ap
->blkno
, mp
->m_sb
.sb_rextsize
);
228 if ((error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1, ap
->length
,
229 &ralen
, atype
, ap
->wasdel
, prod
, &rtb
)))
231 if (rtb
== NULLFSBLOCK
&& prod
> 1 &&
232 (error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1,
233 ap
->length
, &ralen
, atype
,
234 ap
->wasdel
, 1, &rtb
)))
237 if (ap
->blkno
!= NULLFSBLOCK
) {
238 ap
->blkno
*= mp
->m_sb
.sb_rextsize
;
239 ralen
*= mp
->m_sb
.sb_rextsize
;
241 ap
->ip
->i_d
.di_nblocks
+= ralen
;
242 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
244 ap
->ip
->i_delayed_blks
-= ralen
;
246 * Adjust the disk quota also. This was reserved
249 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
250 ap
->wasdel
? XFS_TRANS_DQ_DELRTBCOUNT
:
251 XFS_TRANS_DQ_RTBCOUNT
, (long) ralen
);
253 /* Zero the extent if we were asked to do so */
254 if (ap
->userdata
& XFS_ALLOC_USERDATA_ZERO
) {
255 error
= xfs_zero_extent(ap
->ip
, ap
->blkno
, ap
->length
);
266 * Check if the endoff is outside the last extent. If so the caller will grow
267 * the allocation to a stripe unit boundary. All offsets are considered outside
268 * the end of file for an empty fork, so 1 is returned in *eof in that case.
272 struct xfs_inode
*ip
,
273 xfs_fileoff_t endoff
,
277 struct xfs_bmbt_irec rec
;
280 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, eof
);
284 *eof
= endoff
>= rec
.br_startoff
+ rec
.br_blockcount
;
289 * Extent tree block counting routines.
293 * Count leaf blocks given a range of extent records.
296 xfs_bmap_count_leaves(
304 for (b
= 0; b
< numrecs
; b
++) {
305 xfs_bmbt_rec_host_t
*frp
= xfs_iext_get_ext(ifp
, idx
+ b
);
306 *count
+= xfs_bmbt_get_blockcount(frp
);
311 * Count leaf blocks given a range of extent records originally
315 xfs_bmap_disk_count_leaves(
316 struct xfs_mount
*mp
,
317 struct xfs_btree_block
*block
,
324 for (b
= 1; b
<= numrecs
; b
++) {
325 frp
= XFS_BMBT_REC_ADDR(mp
, block
, b
);
326 *count
+= xfs_bmbt_disk_get_blockcount(frp
);
331 * Recursively walks each level of a btree
332 * to count total fsblocks in use.
334 STATIC
int /* error */
336 xfs_mount_t
*mp
, /* file system mount point */
337 xfs_trans_t
*tp
, /* transaction pointer */
338 xfs_ifork_t
*ifp
, /* inode fork pointer */
339 xfs_fsblock_t blockno
, /* file system block number */
340 int levelin
, /* level in btree */
341 int *count
) /* Count of blocks */
347 xfs_fsblock_t bno
= blockno
;
348 xfs_fsblock_t nextbno
;
349 struct xfs_btree_block
*block
, *nextblock
;
352 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
, XFS_BMAP_BTREE_REF
,
357 block
= XFS_BUF_TO_BLOCK(bp
);
360 /* Not at node above leaves, count this level of nodes */
361 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
362 while (nextbno
!= NULLFSBLOCK
) {
363 error
= xfs_btree_read_bufl(mp
, tp
, nextbno
, 0, &nbp
,
369 nextblock
= XFS_BUF_TO_BLOCK(nbp
);
370 nextbno
= be64_to_cpu(nextblock
->bb_u
.l
.bb_rightsib
);
371 xfs_trans_brelse(tp
, nbp
);
374 /* Dive to the next level */
375 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
376 bno
= be64_to_cpu(*pp
);
377 if (unlikely((error
=
378 xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
)) < 0)) {
379 xfs_trans_brelse(tp
, bp
);
380 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
381 XFS_ERRLEVEL_LOW
, mp
);
382 return -EFSCORRUPTED
;
384 xfs_trans_brelse(tp
, bp
);
386 /* count all level 1 nodes and their leaves */
388 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
389 numrecs
= be16_to_cpu(block
->bb_numrecs
);
390 xfs_bmap_disk_count_leaves(mp
, block
, numrecs
, count
);
391 xfs_trans_brelse(tp
, bp
);
392 if (nextbno
== NULLFSBLOCK
)
395 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
401 block
= XFS_BUF_TO_BLOCK(bp
);
408 * Count fsblocks of the given fork.
410 static int /* error */
411 xfs_bmap_count_blocks(
412 xfs_trans_t
*tp
, /* transaction pointer */
413 xfs_inode_t
*ip
, /* incore inode */
414 int whichfork
, /* data or attr fork */
415 int *count
) /* out: count of blocks */
417 struct xfs_btree_block
*block
; /* current btree block */
418 xfs_fsblock_t bno
; /* block # of "block" */
419 xfs_ifork_t
*ifp
; /* fork structure */
420 int level
; /* btree level, for checking */
421 xfs_mount_t
*mp
; /* file system mount structure */
422 __be64
*pp
; /* pointer to block address */
426 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
427 if ( XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
) {
428 xfs_bmap_count_leaves(ifp
, 0,
429 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
),
435 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
437 block
= ifp
->if_broot
;
438 level
= be16_to_cpu(block
->bb_level
);
440 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
441 bno
= be64_to_cpu(*pp
);
442 ASSERT(bno
!= NULLFSBLOCK
);
443 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
444 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
446 if (unlikely(xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
) < 0)) {
447 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW
,
449 return -EFSCORRUPTED
;
456 * returns 1 for success, 0 if we failed to map the extent.
459 xfs_getbmapx_fix_eof_hole(
460 xfs_inode_t
*ip
, /* xfs incore inode pointer */
461 struct getbmapx
*out
, /* output structure */
462 int prealloced
, /* this is a file with
463 * preallocated data space */
464 __int64_t end
, /* last block requested */
465 xfs_fsblock_t startblock
)
468 xfs_mount_t
*mp
; /* file system mount point */
469 xfs_ifork_t
*ifp
; /* inode fork pointer */
470 xfs_extnum_t lastx
; /* last extent pointer */
471 xfs_fileoff_t fileblock
;
473 if (startblock
== HOLESTARTBLOCK
) {
476 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)));
477 fixlen
-= out
->bmv_offset
;
478 if (prealloced
&& out
->bmv_offset
+ out
->bmv_length
== end
) {
479 /* Came to hole at EOF. Trim it. */
482 out
->bmv_length
= fixlen
;
485 if (startblock
== DELAYSTARTBLOCK
)
488 out
->bmv_block
= xfs_fsb_to_db(ip
, startblock
);
489 fileblock
= XFS_BB_TO_FSB(ip
->i_mount
, out
->bmv_offset
);
490 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
491 if (xfs_iext_bno_to_ext(ifp
, fileblock
, &lastx
) &&
492 (lastx
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))-1))
493 out
->bmv_oflags
|= BMV_OF_LAST
;
500 * Get inode's extents as described in bmv, and format for output.
501 * Calls formatter to fill the user's buffer until all extents
502 * are mapped, until the passed-in bmv->bmv_count slots have
503 * been filled, or until the formatter short-circuits the loop,
504 * if it is tracking filled-in extents on its own.
509 struct getbmapx
*bmv
, /* user bmap structure */
510 xfs_bmap_format_t formatter
, /* format to user */
511 void *arg
) /* formatter arg */
513 __int64_t bmvend
; /* last block requested */
514 int error
= 0; /* return value */
515 __int64_t fixlen
; /* length for -1 case */
516 int i
; /* extent number */
517 int lock
; /* lock state */
518 xfs_bmbt_irec_t
*map
; /* buffer for user's data */
519 xfs_mount_t
*mp
; /* file system mount point */
520 int nex
; /* # of user extents can do */
521 int nexleft
; /* # of user extents left */
522 int subnex
; /* # of bmapi's can do */
523 int nmap
; /* number of map entries */
524 struct getbmapx
*out
; /* output structure */
525 int whichfork
; /* data or attr fork */
526 int prealloced
; /* this is a file with
527 * preallocated data space */
528 int iflags
; /* interface flags */
529 int bmapi_flags
; /* flags for xfs_bmapi */
533 iflags
= bmv
->bmv_iflags
;
534 whichfork
= iflags
& BMV_IF_ATTRFORK
? XFS_ATTR_FORK
: XFS_DATA_FORK
;
536 if (whichfork
== XFS_ATTR_FORK
) {
537 if (XFS_IFORK_Q(ip
)) {
538 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
&&
539 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_BTREE
&&
540 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)
543 ip
->i_d
.di_aformat
!= 0 &&
544 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
)) {
545 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW
,
547 return -EFSCORRUPTED
;
553 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
&&
554 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
&&
555 ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
)
558 if (xfs_get_extsz_hint(ip
) ||
559 ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
|XFS_DIFLAG_APPEND
)){
561 fixlen
= mp
->m_super
->s_maxbytes
;
564 fixlen
= XFS_ISIZE(ip
);
568 if (bmv
->bmv_length
== -1) {
569 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, fixlen
));
571 max_t(__int64_t
, fixlen
- bmv
->bmv_offset
, 0);
572 } else if (bmv
->bmv_length
== 0) {
573 bmv
->bmv_entries
= 0;
575 } else if (bmv
->bmv_length
< 0) {
579 nex
= bmv
->bmv_count
- 1;
582 bmvend
= bmv
->bmv_offset
+ bmv
->bmv_length
;
585 if (bmv
->bmv_count
> ULONG_MAX
/ sizeof(struct getbmapx
))
587 out
= kmem_zalloc_large(bmv
->bmv_count
* sizeof(struct getbmapx
), 0);
591 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
592 if (whichfork
== XFS_DATA_FORK
) {
593 if (!(iflags
& BMV_IF_DELALLOC
) &&
594 (ip
->i_delayed_blks
|| XFS_ISIZE(ip
) > ip
->i_d
.di_size
)) {
595 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
597 goto out_unlock_iolock
;
600 * Even after flushing the inode, there can still be
601 * delalloc blocks on the inode beyond EOF due to
602 * speculative preallocation. These are not removed
603 * until the release function is called or the inode
604 * is inactivated. Hence we cannot assert here that
605 * ip->i_delayed_blks == 0.
609 lock
= xfs_ilock_data_map_shared(ip
);
611 lock
= xfs_ilock_attr_map_shared(ip
);
615 * Don't let nex be bigger than the number of extents
616 * we can have assuming alternating holes and real extents.
618 if (nex
> XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1)
619 nex
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1;
621 bmapi_flags
= xfs_bmapi_aflag(whichfork
);
622 if (!(iflags
& BMV_IF_PREALLOC
))
623 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
626 * Allocate enough space to handle "subnex" maps at a time.
630 map
= kmem_alloc(subnex
* sizeof(*map
), KM_MAYFAIL
| KM_NOFS
);
632 goto out_unlock_ilock
;
634 bmv
->bmv_entries
= 0;
636 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0 &&
637 (whichfork
== XFS_ATTR_FORK
|| !(iflags
& BMV_IF_DELALLOC
))) {
645 nmap
= (nexleft
> subnex
) ? subnex
: nexleft
;
646 error
= xfs_bmapi_read(ip
, XFS_BB_TO_FSBT(mp
, bmv
->bmv_offset
),
647 XFS_BB_TO_FSB(mp
, bmv
->bmv_length
),
648 map
, &nmap
, bmapi_flags
);
651 ASSERT(nmap
<= subnex
);
653 for (i
= 0; i
< nmap
&& nexleft
&& bmv
->bmv_length
; i
++) {
654 out
[cur_ext
].bmv_oflags
= 0;
655 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
)
656 out
[cur_ext
].bmv_oflags
|= BMV_OF_PREALLOC
;
657 else if (map
[i
].br_startblock
== DELAYSTARTBLOCK
)
658 out
[cur_ext
].bmv_oflags
|= BMV_OF_DELALLOC
;
659 out
[cur_ext
].bmv_offset
=
660 XFS_FSB_TO_BB(mp
, map
[i
].br_startoff
);
661 out
[cur_ext
].bmv_length
=
662 XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
663 out
[cur_ext
].bmv_unused1
= 0;
664 out
[cur_ext
].bmv_unused2
= 0;
667 * delayed allocation extents that start beyond EOF can
668 * occur due to speculative EOF allocation when the
669 * delalloc extent is larger than the largest freespace
670 * extent at conversion time. These extents cannot be
671 * converted by data writeback, so can exist here even
672 * if we are not supposed to be finding delalloc
675 if (map
[i
].br_startblock
== DELAYSTARTBLOCK
&&
676 map
[i
].br_startoff
<= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)))
677 ASSERT((iflags
& BMV_IF_DELALLOC
) != 0);
679 if (map
[i
].br_startblock
== HOLESTARTBLOCK
&&
680 whichfork
== XFS_ATTR_FORK
) {
681 /* came to the end of attribute fork */
682 out
[cur_ext
].bmv_oflags
|= BMV_OF_LAST
;
686 if (!xfs_getbmapx_fix_eof_hole(ip
, &out
[cur_ext
],
688 map
[i
].br_startblock
))
692 out
[cur_ext
].bmv_offset
+
693 out
[cur_ext
].bmv_length
;
695 max_t(__int64_t
, 0, bmvend
- bmv
->bmv_offset
);
698 * In case we don't want to return the hole,
699 * don't increase cur_ext so that we can reuse
700 * it in the next loop.
702 if ((iflags
& BMV_IF_NO_HOLES
) &&
703 map
[i
].br_startblock
== HOLESTARTBLOCK
) {
704 memset(&out
[cur_ext
], 0, sizeof(out
[cur_ext
]));
712 } while (nmap
&& nexleft
&& bmv
->bmv_length
);
717 xfs_iunlock(ip
, lock
);
719 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
721 for (i
= 0; i
< cur_ext
; i
++) {
722 int full
= 0; /* user array is full */
724 /* format results & advance arg */
725 error
= formatter(&arg
, &out
[i
], &full
);
735 * dead simple method of punching delalyed allocation blocks from a range in
736 * the inode. Walks a block at a time so will be slow, but is only executed in
737 * rare error cases so the overhead is not critical. This will always punch out
738 * both the start and end blocks, even if the ranges only partially overlap
739 * them, so it is up to the caller to ensure that partial blocks are not
743 xfs_bmap_punch_delalloc_range(
744 struct xfs_inode
*ip
,
745 xfs_fileoff_t start_fsb
,
746 xfs_fileoff_t length
)
748 xfs_fileoff_t remaining
= length
;
751 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
755 xfs_bmbt_irec_t imap
;
757 xfs_fsblock_t firstblock
;
758 xfs_bmap_free_t flist
;
761 * Map the range first and check that it is a delalloc extent
762 * before trying to unmap the range. Otherwise we will be
763 * trying to remove a real extent (which requires a
764 * transaction) or a hole, which is probably a bad idea...
766 error
= xfs_bmapi_read(ip
, start_fsb
, 1, &imap
, &nimaps
,
770 /* something screwed, just bail */
771 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
772 xfs_alert(ip
->i_mount
,
773 "Failed delalloc mapping lookup ino %lld fsb %lld.",
774 ip
->i_ino
, start_fsb
);
782 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
783 /* been converted, ignore */
786 WARN_ON(imap
.br_blockcount
== 0);
789 * Note: while we initialise the firstblock/flist pair, they
790 * should never be used because blocks should never be
791 * allocated or freed for a delalloc extent and hence we need
792 * don't cancel or finish them after the xfs_bunmapi() call.
794 xfs_bmap_init(&flist
, &firstblock
);
795 error
= xfs_bunmapi(NULL
, ip
, start_fsb
, 1, 0, 1, &firstblock
,
800 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
804 } while(remaining
> 0);
810 * Test whether it is appropriate to check an inode for and free post EOF
811 * blocks. The 'force' parameter determines whether we should also consider
812 * regular files that are marked preallocated or append-only.
815 xfs_can_free_eofblocks(struct xfs_inode
*ip
, bool force
)
817 /* prealloc/delalloc exists only on regular files */
818 if (!S_ISREG(VFS_I(ip
)->i_mode
))
822 * Zero sized files with no cached pages and delalloc blocks will not
823 * have speculative prealloc/delalloc blocks to remove.
825 if (VFS_I(ip
)->i_size
== 0 &&
826 VFS_I(ip
)->i_mapping
->nrpages
== 0 &&
827 ip
->i_delayed_blks
== 0)
830 /* If we haven't read in the extent list, then don't do it now. */
831 if (!(ip
->i_df
.if_flags
& XFS_IFEXTENTS
))
835 * Do not free real preallocated or append-only files unless the file
836 * has delalloc blocks and we are forced to remove them.
838 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
))
839 if (!force
|| ip
->i_delayed_blks
== 0)
846 * This is called by xfs_inactive to free any blocks beyond eof
847 * when the link count isn't zero and by xfs_dm_punch_hole() when
848 * punching a hole to EOF.
858 xfs_fileoff_t end_fsb
;
859 xfs_fileoff_t last_fsb
;
860 xfs_filblks_t map_len
;
862 xfs_bmbt_irec_t imap
;
865 * Figure out if there are any blocks beyond the end
866 * of the file. If not, then there is nothing to do.
868 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_ISIZE(ip
));
869 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
870 if (last_fsb
<= end_fsb
)
872 map_len
= last_fsb
- end_fsb
;
875 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
876 error
= xfs_bmapi_read(ip
, end_fsb
, map_len
, &imap
, &nimaps
, 0);
877 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
879 if (!error
&& (nimaps
!= 0) &&
880 (imap
.br_startblock
!= HOLESTARTBLOCK
||
881 ip
->i_delayed_blks
)) {
883 * Attach the dquots to the inode up front.
885 error
= xfs_qm_dqattach(ip
, 0);
890 * There are blocks after the end of file.
891 * Free them up now by truncating the file to
895 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
))
899 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0,
902 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
904 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
908 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
909 xfs_trans_ijoin(tp
, ip
, 0);
912 * Do not update the on-disk file size. If we update the
913 * on-disk file size and then the system crashes before the
914 * contents of the file are flushed to disk then the files
915 * may be full of holes (ie NULL files bug).
917 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
,
921 * If we get an error at this point we simply don't
922 * bother truncating the file.
924 xfs_trans_cancel(tp
);
926 error
= xfs_trans_commit(tp
);
928 xfs_inode_clear_eofblocks_tag(ip
);
931 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
933 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
939 xfs_alloc_file_space(
940 struct xfs_inode
*ip
,
945 xfs_mount_t
*mp
= ip
->i_mount
;
947 xfs_filblks_t allocated_fsb
;
948 xfs_filblks_t allocatesize_fsb
;
949 xfs_extlen_t extsz
, temp
;
950 xfs_fileoff_t startoffset_fsb
;
951 xfs_fsblock_t firstfsb
;
956 xfs_bmbt_irec_t imaps
[1], *imapp
;
957 xfs_bmap_free_t free_list
;
958 uint qblocks
, resblks
, resrtextents
;
961 trace_xfs_alloc_file_space(ip
);
963 if (XFS_FORCED_SHUTDOWN(mp
))
966 error
= xfs_qm_dqattach(ip
, 0);
973 rt
= XFS_IS_REALTIME_INODE(ip
);
974 extsz
= xfs_get_extsz_hint(ip
);
979 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
980 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
983 * Allocate file space until done or until there is an error
985 while (allocatesize_fsb
&& !error
) {
989 * Determine space reservations for data/realtime.
991 if (unlikely(extsz
)) {
995 e
= startoffset_fsb
+ allocatesize_fsb
;
996 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
998 if ((temp
= do_mod(e
, extsz
)))
1002 e
= allocatesize_fsb
;
1006 * The transaction reservation is limited to a 32-bit block
1007 * count, hence we need to limit the number of blocks we are
1008 * trying to reserve to avoid an overflow. We can't allocate
1009 * more than @nimaps extents, and an extent is limited on disk
1010 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1012 resblks
= min_t(xfs_fileoff_t
, (e
- s
), (MAXEXTLEN
* nimaps
));
1014 resrtextents
= qblocks
= resblks
;
1015 resrtextents
/= mp
->m_sb
.sb_rextsize
;
1016 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1017 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
1020 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resblks
);
1021 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
1025 * Allocate and setup the transaction.
1027 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, resblks
,
1028 resrtextents
, 0, &tp
);
1031 * Check for running out of space
1035 * Free the transaction structure.
1037 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1040 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1041 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
1046 xfs_trans_ijoin(tp
, ip
, 0);
1048 xfs_bmap_init(&free_list
, &firstfsb
);
1049 error
= xfs_bmapi_write(tp
, ip
, startoffset_fsb
,
1050 allocatesize_fsb
, alloc_type
, &firstfsb
,
1051 resblks
, imapp
, &nimaps
, &free_list
);
1056 * Complete the transaction
1058 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1062 error
= xfs_trans_commit(tp
);
1063 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1067 allocated_fsb
= imapp
->br_blockcount
;
1074 startoffset_fsb
+= allocated_fsb
;
1075 allocatesize_fsb
-= allocated_fsb
;
1080 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1081 xfs_bmap_cancel(&free_list
);
1082 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
1084 error1
: /* Just cancel transaction */
1085 xfs_trans_cancel(tp
);
1086 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1092 struct xfs_inode
*ip
,
1093 xfs_fileoff_t startoffset_fsb
,
1094 xfs_filblks_t len_fsb
,
1097 struct xfs_mount
*mp
= ip
->i_mount
;
1098 struct xfs_trans
*tp
;
1099 struct xfs_bmap_free free_list
;
1100 xfs_fsblock_t firstfsb
;
1101 uint resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1104 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, resblks
, 0, 0, &tp
);
1106 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1110 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1111 error
= xfs_trans_reserve_quota(tp
, mp
, ip
->i_udquot
, ip
->i_gdquot
,
1112 ip
->i_pdquot
, resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
1114 goto out_trans_cancel
;
1116 xfs_trans_ijoin(tp
, ip
, 0);
1118 xfs_bmap_init(&free_list
, &firstfsb
);
1119 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
, len_fsb
, 0, 2, &firstfsb
,
1122 goto out_bmap_cancel
;
1124 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1126 goto out_bmap_cancel
;
1128 error
= xfs_trans_commit(tp
);
1130 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1134 xfs_bmap_cancel(&free_list
);
1136 xfs_trans_cancel(tp
);
1141 xfs_adjust_extent_unmap_boundaries(
1142 struct xfs_inode
*ip
,
1143 xfs_fileoff_t
*startoffset_fsb
,
1144 xfs_fileoff_t
*endoffset_fsb
)
1146 struct xfs_mount
*mp
= ip
->i_mount
;
1147 struct xfs_bmbt_irec imap
;
1149 xfs_extlen_t mod
= 0;
1152 error
= xfs_bmapi_read(ip
, *startoffset_fsb
, 1, &imap
, &nimap
, 0);
1156 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1159 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1160 block
= imap
.br_startblock
;
1161 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
1163 *startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
1167 error
= xfs_bmapi_read(ip
, *endoffset_fsb
- 1, 1, &imap
, &nimap
, 0);
1171 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1172 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1174 if (mod
&& mod
!= mp
->m_sb
.sb_rextsize
)
1175 *endoffset_fsb
-= mod
;
1182 xfs_flush_unmap_range(
1183 struct xfs_inode
*ip
,
1187 struct xfs_mount
*mp
= ip
->i_mount
;
1188 struct inode
*inode
= VFS_I(ip
);
1189 xfs_off_t rounding
, start
, end
;
1192 /* wait for the completion of any pending DIOs */
1193 inode_dio_wait(inode
);
1195 rounding
= max_t(xfs_off_t
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_SIZE
);
1196 start
= round_down(offset
, rounding
);
1197 end
= round_up(offset
+ len
, rounding
) - 1;
1199 error
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1202 truncate_pagecache_range(inode
, start
, end
);
1207 xfs_free_file_space(
1208 struct xfs_inode
*ip
,
1212 struct xfs_mount
*mp
= ip
->i_mount
;
1213 xfs_fileoff_t startoffset_fsb
;
1214 xfs_fileoff_t endoffset_fsb
;
1215 int done
= 0, error
;
1217 trace_xfs_free_file_space(ip
);
1219 error
= xfs_qm_dqattach(ip
, 0);
1223 if (len
<= 0) /* if nothing being freed */
1226 error
= xfs_flush_unmap_range(ip
, offset
, len
);
1230 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
1231 endoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
+ len
);
1234 * Need to zero the stuff we're not freeing, on disk. If it's a RT file
1235 * and we can't use unwritten extents then we actually need to ensure
1236 * to zero the whole extent, otherwise we just need to take of block
1237 * boundaries, and xfs_bunmapi will handle the rest.
1239 if (XFS_IS_REALTIME_INODE(ip
) &&
1240 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
1241 error
= xfs_adjust_extent_unmap_boundaries(ip
, &startoffset_fsb
,
1247 if (endoffset_fsb
> startoffset_fsb
) {
1249 error
= xfs_unmap_extent(ip
, startoffset_fsb
,
1250 endoffset_fsb
- startoffset_fsb
, &done
);
1257 * Now that we've unmap all full blocks we'll have to zero out any
1258 * partial block at the beginning and/or end. xfs_zero_range is
1259 * smart enough to skip any holes, including those we just created.
1261 return xfs_zero_range(ip
, offset
, len
, NULL
);
1265 * Preallocate and zero a range of a file. This mechanism has the allocation
1266 * semantics of fallocate and in addition converts data in the range to zeroes.
1269 xfs_zero_file_space(
1270 struct xfs_inode
*ip
,
1274 struct xfs_mount
*mp
= ip
->i_mount
;
1278 trace_xfs_zero_file_space(ip
);
1280 blksize
= 1 << mp
->m_sb
.sb_blocklog
;
1283 * Punch a hole and prealloc the range. We use hole punch rather than
1284 * unwritten extent conversion for two reasons:
1286 * 1.) Hole punch handles partial block zeroing for us.
1288 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1289 * by virtue of the hole punch.
1291 error
= xfs_free_file_space(ip
, offset
, len
);
1295 error
= xfs_alloc_file_space(ip
, round_down(offset
, blksize
),
1296 round_up(offset
+ len
, blksize
) -
1297 round_down(offset
, blksize
),
1298 XFS_BMAPI_PREALLOC
);
1305 * @next_fsb will keep track of the extent currently undergoing shift.
1306 * @stop_fsb will keep track of the extent at which we have to stop.
1307 * If we are shifting left, we will start with block (offset + len) and
1308 * shift each extent till last extent.
1309 * If we are shifting right, we will start with last extent inside file space
1310 * and continue until we reach the block corresponding to offset.
1313 xfs_shift_file_space(
1314 struct xfs_inode
*ip
,
1317 enum shift_direction direction
)
1320 struct xfs_mount
*mp
= ip
->i_mount
;
1321 struct xfs_trans
*tp
;
1323 struct xfs_bmap_free free_list
;
1324 xfs_fsblock_t first_block
;
1325 xfs_fileoff_t stop_fsb
;
1326 xfs_fileoff_t next_fsb
;
1327 xfs_fileoff_t shift_fsb
;
1329 ASSERT(direction
== SHIFT_LEFT
|| direction
== SHIFT_RIGHT
);
1331 if (direction
== SHIFT_LEFT
) {
1332 next_fsb
= XFS_B_TO_FSB(mp
, offset
+ len
);
1333 stop_fsb
= XFS_B_TO_FSB(mp
, VFS_I(ip
)->i_size
);
1336 * If right shift, delegate the work of initialization of
1337 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1339 next_fsb
= NULLFSBLOCK
;
1340 stop_fsb
= XFS_B_TO_FSB(mp
, offset
);
1343 shift_fsb
= XFS_B_TO_FSB(mp
, len
);
1346 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1347 * into the accessible region of the file.
1349 if (xfs_can_free_eofblocks(ip
, true)) {
1350 error
= xfs_free_eofblocks(mp
, ip
, false);
1356 * Writeback and invalidate cache for the remainder of the file as we're
1357 * about to shift down every extent from offset to EOF.
1359 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
1363 error
= invalidate_inode_pages2_range(VFS_I(ip
)->i_mapping
,
1364 offset
>> PAGE_SHIFT
, -1);
1369 * The extent shiting code works on extent granularity. So, if
1370 * stop_fsb is not the starting block of extent, we need to split
1371 * the extent at stop_fsb.
1373 if (direction
== SHIFT_RIGHT
) {
1374 error
= xfs_bmap_split_extent(ip
, stop_fsb
);
1379 while (!error
&& !done
) {
1381 * We would need to reserve permanent block for transaction.
1382 * This will come into picture when after shifting extent into
1383 * hole we found that adjacent extents can be merged which
1384 * may lead to freeing of a block during record update.
1386 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
,
1387 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0, 0, &tp
);
1391 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1392 error
= xfs_trans_reserve_quota(tp
, mp
, ip
->i_udquot
,
1393 ip
->i_gdquot
, ip
->i_pdquot
,
1394 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0,
1395 XFS_QMOPT_RES_REGBLKS
);
1397 goto out_trans_cancel
;
1399 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1401 xfs_bmap_init(&free_list
, &first_block
);
1404 * We are using the write transaction in which max 2 bmbt
1405 * updates are allowed
1407 error
= xfs_bmap_shift_extents(tp
, ip
, &next_fsb
, shift_fsb
,
1408 &done
, stop_fsb
, &first_block
, &free_list
,
1409 direction
, XFS_BMAP_MAX_SHIFT_EXTENTS
);
1411 goto out_bmap_cancel
;
1413 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1415 goto out_bmap_cancel
;
1417 error
= xfs_trans_commit(tp
);
1423 xfs_bmap_cancel(&free_list
);
1425 xfs_trans_cancel(tp
);
1430 * xfs_collapse_file_space()
1431 * This routine frees disk space and shift extent for the given file.
1432 * The first thing we do is to free data blocks in the specified range
1433 * by calling xfs_free_file_space(). It would also sync dirty data
1434 * and invalidate page cache over the region on which collapse range
1435 * is working. And Shift extent records to the left to cover a hole.
1442 xfs_collapse_file_space(
1443 struct xfs_inode
*ip
,
1449 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1450 trace_xfs_collapse_file_space(ip
);
1452 error
= xfs_free_file_space(ip
, offset
, len
);
1456 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_LEFT
);
1460 * xfs_insert_file_space()
1461 * This routine create hole space by shifting extents for the given file.
1462 * The first thing we do is to sync dirty data and invalidate page cache
1463 * over the region on which insert range is working. And split an extent
1464 * to two extents at given offset by calling xfs_bmap_split_extent.
1465 * And shift all extent records which are laying between [offset,
1466 * last allocated extent] to the right to reserve hole range.
1472 xfs_insert_file_space(
1473 struct xfs_inode
*ip
,
1477 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1478 trace_xfs_insert_file_space(ip
);
1480 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_RIGHT
);
1484 * We need to check that the format of the data fork in the temporary inode is
1485 * valid for the target inode before doing the swap. This is not a problem with
1486 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1487 * data fork depending on the space the attribute fork is taking so we can get
1488 * invalid formats on the target inode.
1490 * E.g. target has space for 7 extents in extent format, temp inode only has
1491 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1492 * btree, but when swapped it needs to be in extent format. Hence we can't just
1493 * blindly swap data forks on attr2 filesystems.
1495 * Note that we check the swap in both directions so that we don't end up with
1496 * a corrupt temporary inode, either.
1498 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1499 * inode will prevent this situation from occurring, so all we do here is
1500 * reject and log the attempt. basically we are putting the responsibility on
1501 * userspace to get this right.
1504 xfs_swap_extents_check_format(
1505 xfs_inode_t
*ip
, /* target inode */
1506 xfs_inode_t
*tip
) /* tmp inode */
1509 /* Should never get a local format */
1510 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
||
1511 tip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
1515 * if the target inode has less extents that then temporary inode then
1516 * why did userspace call us?
1518 if (ip
->i_d
.di_nextents
< tip
->i_d
.di_nextents
)
1522 * if the target inode is in extent form and the temp inode is in btree
1523 * form then we will end up with the target inode in the wrong format
1524 * as we already know there are less extents in the temp inode.
1526 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1527 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
)
1530 /* Check temp in extent form to max in target */
1531 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1532 XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) >
1533 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1536 /* Check target in extent form to max in temp */
1537 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1538 XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) >
1539 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1543 * If we are in a btree format, check that the temp root block will fit
1544 * in the target and that it has enough extents to be in btree format
1547 * Note that we have to be careful to allow btree->extent conversions
1548 * (a common defrag case) which will occur when the temp inode is in
1551 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1552 if (XFS_IFORK_BOFF(ip
) &&
1553 XFS_BMAP_BMDR_SPACE(tip
->i_df
.if_broot
) > XFS_IFORK_BOFF(ip
))
1555 if (XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) <=
1556 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1560 /* Reciprocal target->temp btree format checks */
1561 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1562 if (XFS_IFORK_BOFF(tip
) &&
1563 XFS_BMAP_BMDR_SPACE(ip
->i_df
.if_broot
) > XFS_IFORK_BOFF(tip
))
1565 if (XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) <=
1566 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1574 xfs_swap_extent_flush(
1575 struct xfs_inode
*ip
)
1579 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
1582 truncate_pagecache_range(VFS_I(ip
), 0, -1);
1584 /* Verify O_DIRECT for ftmp */
1585 if (VFS_I(ip
)->i_mapping
->nrpages
)
1592 xfs_inode_t
*ip
, /* target inode */
1593 xfs_inode_t
*tip
, /* tmp inode */
1596 xfs_mount_t
*mp
= ip
->i_mount
;
1598 xfs_bstat_t
*sbp
= &sxp
->sx_stat
;
1599 xfs_ifork_t
*tempifp
, *ifp
, *tifp
;
1600 int src_log_flags
, target_log_flags
;
1607 tempifp
= kmem_alloc(sizeof(xfs_ifork_t
), KM_MAYFAIL
);
1614 * Lock the inodes against other IO, page faults and truncate to
1615 * begin with. Then we can ensure the inodes are flushed and have no
1616 * page cache safely. Once we have done this we can take the ilocks and
1617 * do the rest of the checks.
1619 lock_flags
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1620 xfs_lock_two_inodes(ip
, tip
, XFS_IOLOCK_EXCL
);
1621 xfs_lock_two_inodes(ip
, tip
, XFS_MMAPLOCK_EXCL
);
1623 /* Verify that both files have the same format */
1624 if ((VFS_I(ip
)->i_mode
& S_IFMT
) != (VFS_I(tip
)->i_mode
& S_IFMT
)) {
1629 /* Verify both files are either real-time or non-realtime */
1630 if (XFS_IS_REALTIME_INODE(ip
) != XFS_IS_REALTIME_INODE(tip
)) {
1635 error
= xfs_swap_extent_flush(ip
);
1638 error
= xfs_swap_extent_flush(tip
);
1642 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_ichange
, 0, 0, 0, &tp
);
1647 * Lock and join the inodes to the tansaction so that transaction commit
1648 * or cancel will unlock the inodes from this point onwards.
1650 xfs_lock_two_inodes(ip
, tip
, XFS_ILOCK_EXCL
);
1651 lock_flags
|= XFS_ILOCK_EXCL
;
1652 xfs_trans_ijoin(tp
, ip
, lock_flags
);
1653 xfs_trans_ijoin(tp
, tip
, lock_flags
);
1656 /* Verify all data are being swapped */
1657 if (sxp
->sx_offset
!= 0 ||
1658 sxp
->sx_length
!= ip
->i_d
.di_size
||
1659 sxp
->sx_length
!= tip
->i_d
.di_size
) {
1661 goto out_trans_cancel
;
1664 trace_xfs_swap_extent_before(ip
, 0);
1665 trace_xfs_swap_extent_before(tip
, 1);
1667 /* check inode formats now that data is flushed */
1668 error
= xfs_swap_extents_check_format(ip
, tip
);
1671 "%s: inode 0x%llx format is incompatible for exchanging.",
1672 __func__
, ip
->i_ino
);
1673 goto out_trans_cancel
;
1677 * Compare the current change & modify times with that
1678 * passed in. If they differ, we abort this swap.
1679 * This is the mechanism used to ensure the calling
1680 * process that the file was not changed out from
1683 if ((sbp
->bs_ctime
.tv_sec
!= VFS_I(ip
)->i_ctime
.tv_sec
) ||
1684 (sbp
->bs_ctime
.tv_nsec
!= VFS_I(ip
)->i_ctime
.tv_nsec
) ||
1685 (sbp
->bs_mtime
.tv_sec
!= VFS_I(ip
)->i_mtime
.tv_sec
) ||
1686 (sbp
->bs_mtime
.tv_nsec
!= VFS_I(ip
)->i_mtime
.tv_nsec
)) {
1688 goto out_trans_cancel
;
1691 * Count the number of extended attribute blocks
1693 if ( ((XFS_IFORK_Q(ip
) != 0) && (ip
->i_d
.di_anextents
> 0)) &&
1694 (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1695 error
= xfs_bmap_count_blocks(tp
, ip
, XFS_ATTR_FORK
, &aforkblks
);
1697 goto out_trans_cancel
;
1699 if ( ((XFS_IFORK_Q(tip
) != 0) && (tip
->i_d
.di_anextents
> 0)) &&
1700 (tip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1701 error
= xfs_bmap_count_blocks(tp
, tip
, XFS_ATTR_FORK
,
1704 goto out_trans_cancel
;
1708 * Before we've swapped the forks, lets set the owners of the forks
1709 * appropriately. We have to do this as we are demand paging the btree
1710 * buffers, and so the validation done on read will expect the owner
1711 * field to be correctly set. Once we change the owners, we can swap the
1714 * Note the trickiness in setting the log flags - we set the owner log
1715 * flag on the opposite inode (i.e. the inode we are setting the new
1716 * owner to be) because once we swap the forks and log that, log
1717 * recovery is going to see the fork as owned by the swapped inode,
1718 * not the pre-swapped inodes.
1720 src_log_flags
= XFS_ILOG_CORE
;
1721 target_log_flags
= XFS_ILOG_CORE
;
1722 if (ip
->i_d
.di_version
== 3 &&
1723 ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1724 target_log_flags
|= XFS_ILOG_DOWNER
;
1725 error
= xfs_bmbt_change_owner(tp
, ip
, XFS_DATA_FORK
,
1728 goto out_trans_cancel
;
1731 if (tip
->i_d
.di_version
== 3 &&
1732 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1733 src_log_flags
|= XFS_ILOG_DOWNER
;
1734 error
= xfs_bmbt_change_owner(tp
, tip
, XFS_DATA_FORK
,
1737 goto out_trans_cancel
;
1741 * Swap the data forks of the inodes
1745 *tempifp
= *ifp
; /* struct copy */
1746 *ifp
= *tifp
; /* struct copy */
1747 *tifp
= *tempifp
; /* struct copy */
1750 * Fix the on-disk inode values
1752 tmp
= (__uint64_t
)ip
->i_d
.di_nblocks
;
1753 ip
->i_d
.di_nblocks
= tip
->i_d
.di_nblocks
- taforkblks
+ aforkblks
;
1754 tip
->i_d
.di_nblocks
= tmp
+ taforkblks
- aforkblks
;
1756 tmp
= (__uint64_t
) ip
->i_d
.di_nextents
;
1757 ip
->i_d
.di_nextents
= tip
->i_d
.di_nextents
;
1758 tip
->i_d
.di_nextents
= tmp
;
1760 tmp
= (__uint64_t
) ip
->i_d
.di_format
;
1761 ip
->i_d
.di_format
= tip
->i_d
.di_format
;
1762 tip
->i_d
.di_format
= tmp
;
1765 * The extents in the source inode could still contain speculative
1766 * preallocation beyond EOF (e.g. the file is open but not modified
1767 * while defrag is in progress). In that case, we need to copy over the
1768 * number of delalloc blocks the data fork in the source inode is
1769 * tracking beyond EOF so that when the fork is truncated away when the
1770 * temporary inode is unlinked we don't underrun the i_delayed_blks
1771 * counter on that inode.
1773 ASSERT(tip
->i_delayed_blks
== 0);
1774 tip
->i_delayed_blks
= ip
->i_delayed_blks
;
1775 ip
->i_delayed_blks
= 0;
1777 switch (ip
->i_d
.di_format
) {
1778 case XFS_DINODE_FMT_EXTENTS
:
1779 /* If the extents fit in the inode, fix the
1780 * pointer. Otherwise it's already NULL or
1781 * pointing to the extent.
1783 if (ip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1784 ifp
->if_u1
.if_extents
=
1785 ifp
->if_u2
.if_inline_ext
;
1787 src_log_flags
|= XFS_ILOG_DEXT
;
1789 case XFS_DINODE_FMT_BTREE
:
1790 ASSERT(ip
->i_d
.di_version
< 3 ||
1791 (src_log_flags
& XFS_ILOG_DOWNER
));
1792 src_log_flags
|= XFS_ILOG_DBROOT
;
1796 switch (tip
->i_d
.di_format
) {
1797 case XFS_DINODE_FMT_EXTENTS
:
1798 /* If the extents fit in the inode, fix the
1799 * pointer. Otherwise it's already NULL or
1800 * pointing to the extent.
1802 if (tip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1803 tifp
->if_u1
.if_extents
=
1804 tifp
->if_u2
.if_inline_ext
;
1806 target_log_flags
|= XFS_ILOG_DEXT
;
1808 case XFS_DINODE_FMT_BTREE
:
1809 target_log_flags
|= XFS_ILOG_DBROOT
;
1810 ASSERT(tip
->i_d
.di_version
< 3 ||
1811 (target_log_flags
& XFS_ILOG_DOWNER
));
1815 xfs_trans_log_inode(tp
, ip
, src_log_flags
);
1816 xfs_trans_log_inode(tp
, tip
, target_log_flags
);
1819 * If this is a synchronous mount, make sure that the
1820 * transaction goes to disk before returning to the user.
1822 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1823 xfs_trans_set_sync(tp
);
1825 error
= xfs_trans_commit(tp
);
1827 trace_xfs_swap_extent_after(ip
, 0);
1828 trace_xfs_swap_extent_after(tip
, 1);
1834 xfs_iunlock(ip
, lock_flags
);
1835 xfs_iunlock(tip
, lock_flags
);
1839 xfs_trans_cancel(tp
);