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
);
75 ssize_t size
= XFS_FSB_TO_B(mp
, count_fsb
);
77 if (IS_DAX(VFS_I(ip
)))
78 return dax_clear_blocks(VFS_I(ip
), block
, size
);
81 * let the block layer decide on the fastest method of
82 * implementing the zeroing.
84 return sb_issue_zeroout(mp
->m_super
, block
, count_fsb
, GFP_NOFS
);
89 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
90 * caller. Frees all the extents that need freeing, which must be done
91 * last due to locking considerations. We never free any extents in
92 * the first transaction.
94 * If an inode *ip is provided, rejoin it to the transaction if
95 * the transaction was committed.
99 struct xfs_trans
**tp
, /* transaction pointer addr */
100 struct xfs_bmap_free
*flist
, /* i/o: list extents to free */
101 struct xfs_inode
*ip
)
103 struct xfs_efd_log_item
*efd
; /* extent free data */
104 struct xfs_efi_log_item
*efi
; /* extent free intention */
105 int error
; /* error return value */
106 int committed
;/* xact committed or not */
107 struct xfs_bmap_free_item
*free
; /* free extent item */
108 struct xfs_bmap_free_item
*next
; /* next item on free list */
110 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
111 if (flist
->xbf_count
== 0)
114 efi
= xfs_trans_get_efi(*tp
, flist
->xbf_count
);
115 for (free
= flist
->xbf_first
; free
; free
= free
->xbfi_next
)
116 xfs_trans_log_efi_extent(*tp
, efi
, free
->xbfi_startblock
,
117 free
->xbfi_blockcount
);
119 error
= __xfs_trans_roll(tp
, ip
, &committed
);
122 * If the transaction was committed, drop the EFD reference
123 * since we're bailing out of here. The other reference is
124 * dropped when the EFI hits the AIL.
126 * If the transaction was not committed, the EFI is freed by the
127 * EFI item unlock handler on abort. Also, we have a new
128 * transaction so we should return committed=1 even though we're
129 * returning an error.
132 xfs_efi_release(efi
);
133 xfs_force_shutdown((*tp
)->t_mountp
,
134 (error
== -EFSCORRUPTED
) ?
135 SHUTDOWN_CORRUPT_INCORE
:
136 SHUTDOWN_META_IO_ERROR
);
142 * Get an EFD and free each extent in the list, logging to the EFD in
143 * the process. The remaining bmap free list is cleaned up by the caller
146 efd
= xfs_trans_get_efd(*tp
, efi
, flist
->xbf_count
);
147 for (free
= flist
->xbf_first
; free
!= NULL
; free
= next
) {
148 next
= free
->xbfi_next
;
150 error
= xfs_trans_free_extent(*tp
, efd
, free
->xbfi_startblock
,
151 free
->xbfi_blockcount
);
155 xfs_bmap_del_free(flist
, NULL
, free
);
163 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
165 xfs_alloctype_t atype
= 0; /* type for allocation routines */
166 int error
; /* error return value */
167 xfs_mount_t
*mp
; /* mount point structure */
168 xfs_extlen_t prod
= 0; /* product factor for allocators */
169 xfs_extlen_t ralen
= 0; /* realtime allocation length */
170 xfs_extlen_t align
; /* minimum allocation alignment */
173 mp
= ap
->ip
->i_mount
;
174 align
= xfs_get_extsz_hint(ap
->ip
);
175 prod
= align
/ mp
->m_sb
.sb_rextsize
;
176 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
177 align
, 1, ap
->eof
, 0,
178 ap
->conv
, &ap
->offset
, &ap
->length
);
182 ASSERT(ap
->length
% mp
->m_sb
.sb_rextsize
== 0);
185 * If the offset & length are not perfectly aligned
186 * then kill prod, it will just get us in trouble.
188 if (do_mod(ap
->offset
, align
) || ap
->length
% align
)
191 * Set ralen to be the actual requested length in rtextents.
193 ralen
= ap
->length
/ mp
->m_sb
.sb_rextsize
;
195 * If the old value was close enough to MAXEXTLEN that
196 * we rounded up to it, cut it back so it's valid again.
197 * Note that if it's a really large request (bigger than
198 * MAXEXTLEN), we don't hear about that number, and can't
199 * adjust the starting point to match it.
201 if (ralen
* mp
->m_sb
.sb_rextsize
>= MAXEXTLEN
)
202 ralen
= MAXEXTLEN
/ mp
->m_sb
.sb_rextsize
;
205 * Lock out other modifications to the RT bitmap inode.
207 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
208 xfs_trans_ijoin(ap
->tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
211 * If it's an allocation to an empty file at offset 0,
212 * pick an extent that will space things out in the rt area.
214 if (ap
->eof
&& ap
->offset
== 0) {
215 xfs_rtblock_t
uninitialized_var(rtx
); /* realtime extent no */
217 error
= xfs_rtpick_extent(mp
, ap
->tp
, ralen
, &rtx
);
220 ap
->blkno
= rtx
* mp
->m_sb
.sb_rextsize
;
225 xfs_bmap_adjacent(ap
);
228 * Realtime allocation, done through xfs_rtallocate_extent.
230 atype
= ap
->blkno
== 0 ? XFS_ALLOCTYPE_ANY_AG
: XFS_ALLOCTYPE_NEAR_BNO
;
231 do_div(ap
->blkno
, mp
->m_sb
.sb_rextsize
);
234 if ((error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1, ap
->length
,
235 &ralen
, atype
, ap
->wasdel
, prod
, &rtb
)))
237 if (rtb
== NULLFSBLOCK
&& prod
> 1 &&
238 (error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1,
239 ap
->length
, &ralen
, atype
,
240 ap
->wasdel
, 1, &rtb
)))
243 if (ap
->blkno
!= NULLFSBLOCK
) {
244 ap
->blkno
*= mp
->m_sb
.sb_rextsize
;
245 ralen
*= mp
->m_sb
.sb_rextsize
;
247 ap
->ip
->i_d
.di_nblocks
+= ralen
;
248 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
250 ap
->ip
->i_delayed_blks
-= ralen
;
252 * Adjust the disk quota also. This was reserved
255 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
256 ap
->wasdel
? XFS_TRANS_DQ_DELRTBCOUNT
:
257 XFS_TRANS_DQ_RTBCOUNT
, (long) ralen
);
259 /* Zero the extent if we were asked to do so */
260 if (ap
->userdata
& XFS_ALLOC_USERDATA_ZERO
) {
261 error
= xfs_zero_extent(ap
->ip
, ap
->blkno
, ap
->length
);
272 * Check if the endoff is outside the last extent. If so the caller will grow
273 * the allocation to a stripe unit boundary. All offsets are considered outside
274 * the end of file for an empty fork, so 1 is returned in *eof in that case.
278 struct xfs_inode
*ip
,
279 xfs_fileoff_t endoff
,
283 struct xfs_bmbt_irec rec
;
286 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, eof
);
290 *eof
= endoff
>= rec
.br_startoff
+ rec
.br_blockcount
;
295 * Extent tree block counting routines.
299 * Count leaf blocks given a range of extent records.
302 xfs_bmap_count_leaves(
310 for (b
= 0; b
< numrecs
; b
++) {
311 xfs_bmbt_rec_host_t
*frp
= xfs_iext_get_ext(ifp
, idx
+ b
);
312 *count
+= xfs_bmbt_get_blockcount(frp
);
317 * Count leaf blocks given a range of extent records originally
321 xfs_bmap_disk_count_leaves(
322 struct xfs_mount
*mp
,
323 struct xfs_btree_block
*block
,
330 for (b
= 1; b
<= numrecs
; b
++) {
331 frp
= XFS_BMBT_REC_ADDR(mp
, block
, b
);
332 *count
+= xfs_bmbt_disk_get_blockcount(frp
);
337 * Recursively walks each level of a btree
338 * to count total fsblocks in use.
340 STATIC
int /* error */
342 xfs_mount_t
*mp
, /* file system mount point */
343 xfs_trans_t
*tp
, /* transaction pointer */
344 xfs_ifork_t
*ifp
, /* inode fork pointer */
345 xfs_fsblock_t blockno
, /* file system block number */
346 int levelin
, /* level in btree */
347 int *count
) /* Count of blocks */
353 xfs_fsblock_t bno
= blockno
;
354 xfs_fsblock_t nextbno
;
355 struct xfs_btree_block
*block
, *nextblock
;
358 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
, XFS_BMAP_BTREE_REF
,
363 block
= XFS_BUF_TO_BLOCK(bp
);
366 /* Not at node above leaves, count this level of nodes */
367 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
368 while (nextbno
!= NULLFSBLOCK
) {
369 error
= xfs_btree_read_bufl(mp
, tp
, nextbno
, 0, &nbp
,
375 nextblock
= XFS_BUF_TO_BLOCK(nbp
);
376 nextbno
= be64_to_cpu(nextblock
->bb_u
.l
.bb_rightsib
);
377 xfs_trans_brelse(tp
, nbp
);
380 /* Dive to the next level */
381 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
382 bno
= be64_to_cpu(*pp
);
383 if (unlikely((error
=
384 xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
)) < 0)) {
385 xfs_trans_brelse(tp
, bp
);
386 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
387 XFS_ERRLEVEL_LOW
, mp
);
388 return -EFSCORRUPTED
;
390 xfs_trans_brelse(tp
, bp
);
392 /* count all level 1 nodes and their leaves */
394 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
395 numrecs
= be16_to_cpu(block
->bb_numrecs
);
396 xfs_bmap_disk_count_leaves(mp
, block
, numrecs
, count
);
397 xfs_trans_brelse(tp
, bp
);
398 if (nextbno
== NULLFSBLOCK
)
401 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
407 block
= XFS_BUF_TO_BLOCK(bp
);
414 * Count fsblocks of the given fork.
417 xfs_bmap_count_blocks(
418 xfs_trans_t
*tp
, /* transaction pointer */
419 xfs_inode_t
*ip
, /* incore inode */
420 int whichfork
, /* data or attr fork */
421 int *count
) /* out: count of blocks */
423 struct xfs_btree_block
*block
; /* current btree block */
424 xfs_fsblock_t bno
; /* block # of "block" */
425 xfs_ifork_t
*ifp
; /* fork structure */
426 int level
; /* btree level, for checking */
427 xfs_mount_t
*mp
; /* file system mount structure */
428 __be64
*pp
; /* pointer to block address */
432 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
433 if ( XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
) {
434 xfs_bmap_count_leaves(ifp
, 0,
435 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
),
441 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
443 block
= ifp
->if_broot
;
444 level
= be16_to_cpu(block
->bb_level
);
446 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
447 bno
= be64_to_cpu(*pp
);
448 ASSERT(bno
!= NULLFSBLOCK
);
449 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
450 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
452 if (unlikely(xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
) < 0)) {
453 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW
,
455 return -EFSCORRUPTED
;
462 * returns 1 for success, 0 if we failed to map the extent.
465 xfs_getbmapx_fix_eof_hole(
466 xfs_inode_t
*ip
, /* xfs incore inode pointer */
467 struct getbmapx
*out
, /* output structure */
468 int prealloced
, /* this is a file with
469 * preallocated data space */
470 __int64_t end
, /* last block requested */
471 xfs_fsblock_t startblock
)
474 xfs_mount_t
*mp
; /* file system mount point */
475 xfs_ifork_t
*ifp
; /* inode fork pointer */
476 xfs_extnum_t lastx
; /* last extent pointer */
477 xfs_fileoff_t fileblock
;
479 if (startblock
== HOLESTARTBLOCK
) {
482 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)));
483 fixlen
-= out
->bmv_offset
;
484 if (prealloced
&& out
->bmv_offset
+ out
->bmv_length
== end
) {
485 /* Came to hole at EOF. Trim it. */
488 out
->bmv_length
= fixlen
;
491 if (startblock
== DELAYSTARTBLOCK
)
494 out
->bmv_block
= xfs_fsb_to_db(ip
, startblock
);
495 fileblock
= XFS_BB_TO_FSB(ip
->i_mount
, out
->bmv_offset
);
496 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
497 if (xfs_iext_bno_to_ext(ifp
, fileblock
, &lastx
) &&
498 (lastx
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))-1))
499 out
->bmv_oflags
|= BMV_OF_LAST
;
506 * Get inode's extents as described in bmv, and format for output.
507 * Calls formatter to fill the user's buffer until all extents
508 * are mapped, until the passed-in bmv->bmv_count slots have
509 * been filled, or until the formatter short-circuits the loop,
510 * if it is tracking filled-in extents on its own.
515 struct getbmapx
*bmv
, /* user bmap structure */
516 xfs_bmap_format_t formatter
, /* format to user */
517 void *arg
) /* formatter arg */
519 __int64_t bmvend
; /* last block requested */
520 int error
= 0; /* return value */
521 __int64_t fixlen
; /* length for -1 case */
522 int i
; /* extent number */
523 int lock
; /* lock state */
524 xfs_bmbt_irec_t
*map
; /* buffer for user's data */
525 xfs_mount_t
*mp
; /* file system mount point */
526 int nex
; /* # of user extents can do */
527 int nexleft
; /* # of user extents left */
528 int subnex
; /* # of bmapi's can do */
529 int nmap
; /* number of map entries */
530 struct getbmapx
*out
; /* output structure */
531 int whichfork
; /* data or attr fork */
532 int prealloced
; /* this is a file with
533 * preallocated data space */
534 int iflags
; /* interface flags */
535 int bmapi_flags
; /* flags for xfs_bmapi */
539 iflags
= bmv
->bmv_iflags
;
540 whichfork
= iflags
& BMV_IF_ATTRFORK
? XFS_ATTR_FORK
: XFS_DATA_FORK
;
542 if (whichfork
== XFS_ATTR_FORK
) {
543 if (XFS_IFORK_Q(ip
)) {
544 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
&&
545 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_BTREE
&&
546 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)
549 ip
->i_d
.di_aformat
!= 0 &&
550 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
)) {
551 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW
,
553 return -EFSCORRUPTED
;
559 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
&&
560 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
&&
561 ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
)
564 if (xfs_get_extsz_hint(ip
) ||
565 ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
|XFS_DIFLAG_APPEND
)){
567 fixlen
= mp
->m_super
->s_maxbytes
;
570 fixlen
= XFS_ISIZE(ip
);
574 if (bmv
->bmv_length
== -1) {
575 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, fixlen
));
577 max_t(__int64_t
, fixlen
- bmv
->bmv_offset
, 0);
578 } else if (bmv
->bmv_length
== 0) {
579 bmv
->bmv_entries
= 0;
581 } else if (bmv
->bmv_length
< 0) {
585 nex
= bmv
->bmv_count
- 1;
588 bmvend
= bmv
->bmv_offset
+ bmv
->bmv_length
;
591 if (bmv
->bmv_count
> ULONG_MAX
/ sizeof(struct getbmapx
))
593 out
= kmem_zalloc_large(bmv
->bmv_count
* sizeof(struct getbmapx
), 0);
597 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
598 if (whichfork
== XFS_DATA_FORK
) {
599 if (!(iflags
& BMV_IF_DELALLOC
) &&
600 (ip
->i_delayed_blks
|| XFS_ISIZE(ip
) > ip
->i_d
.di_size
)) {
601 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
603 goto out_unlock_iolock
;
606 * Even after flushing the inode, there can still be
607 * delalloc blocks on the inode beyond EOF due to
608 * speculative preallocation. These are not removed
609 * until the release function is called or the inode
610 * is inactivated. Hence we cannot assert here that
611 * ip->i_delayed_blks == 0.
615 lock
= xfs_ilock_data_map_shared(ip
);
617 lock
= xfs_ilock_attr_map_shared(ip
);
621 * Don't let nex be bigger than the number of extents
622 * we can have assuming alternating holes and real extents.
624 if (nex
> XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1)
625 nex
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1;
627 bmapi_flags
= xfs_bmapi_aflag(whichfork
);
628 if (!(iflags
& BMV_IF_PREALLOC
))
629 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
632 * Allocate enough space to handle "subnex" maps at a time.
636 map
= kmem_alloc(subnex
* sizeof(*map
), KM_MAYFAIL
| KM_NOFS
);
638 goto out_unlock_ilock
;
640 bmv
->bmv_entries
= 0;
642 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0 &&
643 (whichfork
== XFS_ATTR_FORK
|| !(iflags
& BMV_IF_DELALLOC
))) {
651 nmap
= (nexleft
> subnex
) ? subnex
: nexleft
;
652 error
= xfs_bmapi_read(ip
, XFS_BB_TO_FSBT(mp
, bmv
->bmv_offset
),
653 XFS_BB_TO_FSB(mp
, bmv
->bmv_length
),
654 map
, &nmap
, bmapi_flags
);
657 ASSERT(nmap
<= subnex
);
659 for (i
= 0; i
< nmap
&& nexleft
&& bmv
->bmv_length
; i
++) {
660 out
[cur_ext
].bmv_oflags
= 0;
661 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
)
662 out
[cur_ext
].bmv_oflags
|= BMV_OF_PREALLOC
;
663 else if (map
[i
].br_startblock
== DELAYSTARTBLOCK
)
664 out
[cur_ext
].bmv_oflags
|= BMV_OF_DELALLOC
;
665 out
[cur_ext
].bmv_offset
=
666 XFS_FSB_TO_BB(mp
, map
[i
].br_startoff
);
667 out
[cur_ext
].bmv_length
=
668 XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
669 out
[cur_ext
].bmv_unused1
= 0;
670 out
[cur_ext
].bmv_unused2
= 0;
673 * delayed allocation extents that start beyond EOF can
674 * occur due to speculative EOF allocation when the
675 * delalloc extent is larger than the largest freespace
676 * extent at conversion time. These extents cannot be
677 * converted by data writeback, so can exist here even
678 * if we are not supposed to be finding delalloc
681 if (map
[i
].br_startblock
== DELAYSTARTBLOCK
&&
682 map
[i
].br_startoff
<= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)))
683 ASSERT((iflags
& BMV_IF_DELALLOC
) != 0);
685 if (map
[i
].br_startblock
== HOLESTARTBLOCK
&&
686 whichfork
== XFS_ATTR_FORK
) {
687 /* came to the end of attribute fork */
688 out
[cur_ext
].bmv_oflags
|= BMV_OF_LAST
;
692 if (!xfs_getbmapx_fix_eof_hole(ip
, &out
[cur_ext
],
694 map
[i
].br_startblock
))
698 out
[cur_ext
].bmv_offset
+
699 out
[cur_ext
].bmv_length
;
701 max_t(__int64_t
, 0, bmvend
- bmv
->bmv_offset
);
704 * In case we don't want to return the hole,
705 * don't increase cur_ext so that we can reuse
706 * it in the next loop.
708 if ((iflags
& BMV_IF_NO_HOLES
) &&
709 map
[i
].br_startblock
== HOLESTARTBLOCK
) {
710 memset(&out
[cur_ext
], 0, sizeof(out
[cur_ext
]));
718 } while (nmap
&& nexleft
&& bmv
->bmv_length
);
723 xfs_iunlock(ip
, lock
);
725 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
727 for (i
= 0; i
< cur_ext
; i
++) {
728 int full
= 0; /* user array is full */
730 /* format results & advance arg */
731 error
= formatter(&arg
, &out
[i
], &full
);
741 * dead simple method of punching delalyed allocation blocks from a range in
742 * the inode. Walks a block at a time so will be slow, but is only executed in
743 * rare error cases so the overhead is not critical. This will always punch out
744 * both the start and end blocks, even if the ranges only partially overlap
745 * them, so it is up to the caller to ensure that partial blocks are not
749 xfs_bmap_punch_delalloc_range(
750 struct xfs_inode
*ip
,
751 xfs_fileoff_t start_fsb
,
752 xfs_fileoff_t length
)
754 xfs_fileoff_t remaining
= length
;
757 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
761 xfs_bmbt_irec_t imap
;
763 xfs_fsblock_t firstblock
;
764 xfs_bmap_free_t flist
;
767 * Map the range first and check that it is a delalloc extent
768 * before trying to unmap the range. Otherwise we will be
769 * trying to remove a real extent (which requires a
770 * transaction) or a hole, which is probably a bad idea...
772 error
= xfs_bmapi_read(ip
, start_fsb
, 1, &imap
, &nimaps
,
776 /* something screwed, just bail */
777 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
778 xfs_alert(ip
->i_mount
,
779 "Failed delalloc mapping lookup ino %lld fsb %lld.",
780 ip
->i_ino
, start_fsb
);
788 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
789 /* been converted, ignore */
792 WARN_ON(imap
.br_blockcount
== 0);
795 * Note: while we initialise the firstblock/flist pair, they
796 * should never be used because blocks should never be
797 * allocated or freed for a delalloc extent and hence we need
798 * don't cancel or finish them after the xfs_bunmapi() call.
800 xfs_bmap_init(&flist
, &firstblock
);
801 error
= xfs_bunmapi(NULL
, ip
, start_fsb
, 1, 0, 1, &firstblock
,
806 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
810 } while(remaining
> 0);
816 * Test whether it is appropriate to check an inode for and free post EOF
817 * blocks. The 'force' parameter determines whether we should also consider
818 * regular files that are marked preallocated or append-only.
821 xfs_can_free_eofblocks(struct xfs_inode
*ip
, bool force
)
823 /* prealloc/delalloc exists only on regular files */
824 if (!S_ISREG(ip
->i_d
.di_mode
))
828 * Zero sized files with no cached pages and delalloc blocks will not
829 * have speculative prealloc/delalloc blocks to remove.
831 if (VFS_I(ip
)->i_size
== 0 &&
832 VFS_I(ip
)->i_mapping
->nrpages
== 0 &&
833 ip
->i_delayed_blks
== 0)
836 /* If we haven't read in the extent list, then don't do it now. */
837 if (!(ip
->i_df
.if_flags
& XFS_IFEXTENTS
))
841 * Do not free real preallocated or append-only files unless the file
842 * has delalloc blocks and we are forced to remove them.
844 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
))
845 if (!force
|| ip
->i_delayed_blks
== 0)
852 * This is called by xfs_inactive to free any blocks beyond eof
853 * when the link count isn't zero and by xfs_dm_punch_hole() when
854 * punching a hole to EOF.
864 xfs_fileoff_t end_fsb
;
865 xfs_fileoff_t last_fsb
;
866 xfs_filblks_t map_len
;
868 xfs_bmbt_irec_t imap
;
871 * Figure out if there are any blocks beyond the end
872 * of the file. If not, then there is nothing to do.
874 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_ISIZE(ip
));
875 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
876 if (last_fsb
<= end_fsb
)
878 map_len
= last_fsb
- end_fsb
;
881 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
882 error
= xfs_bmapi_read(ip
, end_fsb
, map_len
, &imap
, &nimaps
, 0);
883 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
885 if (!error
&& (nimaps
!= 0) &&
886 (imap
.br_startblock
!= HOLESTARTBLOCK
||
887 ip
->i_delayed_blks
)) {
889 * Attach the dquots to the inode up front.
891 error
= xfs_qm_dqattach(ip
, 0);
896 * There are blocks after the end of file.
897 * Free them up now by truncating the file to
900 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
903 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
904 xfs_trans_cancel(tp
);
909 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
911 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
912 xfs_trans_cancel(tp
);
914 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
918 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
919 xfs_trans_ijoin(tp
, ip
, 0);
922 * Do not update the on-disk file size. If we update the
923 * on-disk file size and then the system crashes before the
924 * contents of the file are flushed to disk then the files
925 * may be full of holes (ie NULL files bug).
927 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
,
931 * If we get an error at this point we simply don't
932 * bother truncating the file.
934 xfs_trans_cancel(tp
);
936 error
= xfs_trans_commit(tp
);
938 xfs_inode_clear_eofblocks_tag(ip
);
941 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
943 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
949 xfs_alloc_file_space(
950 struct xfs_inode
*ip
,
955 xfs_mount_t
*mp
= ip
->i_mount
;
957 xfs_filblks_t allocated_fsb
;
958 xfs_filblks_t allocatesize_fsb
;
959 xfs_extlen_t extsz
, temp
;
960 xfs_fileoff_t startoffset_fsb
;
961 xfs_fsblock_t firstfsb
;
966 xfs_bmbt_irec_t imaps
[1], *imapp
;
967 xfs_bmap_free_t free_list
;
968 uint qblocks
, resblks
, resrtextents
;
971 trace_xfs_alloc_file_space(ip
);
973 if (XFS_FORCED_SHUTDOWN(mp
))
976 error
= xfs_qm_dqattach(ip
, 0);
983 rt
= XFS_IS_REALTIME_INODE(ip
);
984 extsz
= xfs_get_extsz_hint(ip
);
989 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
990 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
993 * Allocate file space until done or until there is an error
995 while (allocatesize_fsb
&& !error
) {
999 * Determine space reservations for data/realtime.
1001 if (unlikely(extsz
)) {
1002 s
= startoffset_fsb
;
1005 e
= startoffset_fsb
+ allocatesize_fsb
;
1006 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
1008 if ((temp
= do_mod(e
, extsz
)))
1012 e
= allocatesize_fsb
;
1016 * The transaction reservation is limited to a 32-bit block
1017 * count, hence we need to limit the number of blocks we are
1018 * trying to reserve to avoid an overflow. We can't allocate
1019 * more than @nimaps extents, and an extent is limited on disk
1020 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1022 resblks
= min_t(xfs_fileoff_t
, (e
- s
), (MAXEXTLEN
* nimaps
));
1024 resrtextents
= qblocks
= resblks
;
1025 resrtextents
/= mp
->m_sb
.sb_rextsize
;
1026 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1027 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
1030 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resblks
);
1031 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
1035 * Allocate and setup the transaction.
1037 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1038 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
,
1039 resblks
, resrtextents
);
1041 * Check for running out of space
1045 * Free the transaction structure.
1047 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1048 xfs_trans_cancel(tp
);
1051 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1052 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
1057 xfs_trans_ijoin(tp
, ip
, 0);
1059 xfs_bmap_init(&free_list
, &firstfsb
);
1060 error
= xfs_bmapi_write(tp
, ip
, startoffset_fsb
,
1061 allocatesize_fsb
, alloc_type
, &firstfsb
,
1062 resblks
, imapp
, &nimaps
, &free_list
);
1067 * Complete the transaction
1069 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1073 error
= xfs_trans_commit(tp
);
1074 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1078 allocated_fsb
= imapp
->br_blockcount
;
1085 startoffset_fsb
+= allocated_fsb
;
1086 allocatesize_fsb
-= allocated_fsb
;
1091 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1092 xfs_bmap_cancel(&free_list
);
1093 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
1095 error1
: /* Just cancel transaction */
1096 xfs_trans_cancel(tp
);
1097 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1102 * Zero file bytes between startoff and endoff inclusive.
1103 * The iolock is held exclusive and no blocks are buffered.
1105 * This function is used by xfs_free_file_space() to zero
1106 * partial blocks when the range to free is not block aligned.
1107 * When unreserving space with boundaries that are not block
1108 * aligned we round up the start and round down the end
1109 * boundaries and then use this function to zero the parts of
1110 * the blocks that got dropped during the rounding.
1113 xfs_zero_remaining_bytes(
1118 xfs_bmbt_irec_t imap
;
1119 xfs_fileoff_t offset_fsb
;
1120 xfs_off_t lastoffset
;
1123 xfs_mount_t
*mp
= ip
->i_mount
;
1128 * Avoid doing I/O beyond eof - it's not necessary
1129 * since nothing can read beyond eof. The space will
1130 * be zeroed when the file is extended anyway.
1132 if (startoff
>= XFS_ISIZE(ip
))
1135 if (endoff
> XFS_ISIZE(ip
))
1136 endoff
= XFS_ISIZE(ip
);
1138 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
1141 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1144 lock_mode
= xfs_ilock_data_map_shared(ip
);
1145 error
= xfs_bmapi_read(ip
, offset_fsb
, 1, &imap
, &nimap
, 0);
1146 xfs_iunlock(ip
, lock_mode
);
1148 if (error
|| nimap
< 1)
1150 ASSERT(imap
.br_blockcount
>= 1);
1151 ASSERT(imap
.br_startoff
== offset_fsb
);
1152 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1154 if (imap
.br_startblock
== HOLESTARTBLOCK
||
1155 imap
.br_state
== XFS_EXT_UNWRITTEN
) {
1156 /* skip the entire extent */
1157 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+
1158 imap
.br_blockcount
) - 1;
1162 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
1163 if (lastoffset
> endoff
)
1164 lastoffset
= endoff
;
1166 /* DAX can just zero the backing device directly */
1167 if (IS_DAX(VFS_I(ip
))) {
1168 error
= dax_zero_page_range(VFS_I(ip
), offset
,
1169 lastoffset
- offset
+ 1,
1170 xfs_get_blocks_direct
);
1176 error
= xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip
) ?
1177 mp
->m_rtdev_targp
: mp
->m_ddev_targp
,
1178 xfs_fsb_to_db(ip
, imap
.br_startblock
),
1179 BTOBB(mp
->m_sb
.sb_blocksize
),
1185 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
1186 0, lastoffset
- offset
+ 1);
1188 error
= xfs_bwrite(bp
);
1197 xfs_free_file_space(
1198 struct xfs_inode
*ip
,
1203 xfs_fileoff_t endoffset_fsb
;
1205 xfs_fsblock_t firstfsb
;
1206 xfs_bmap_free_t free_list
;
1207 xfs_bmbt_irec_t imap
;
1209 xfs_off_t iendoffset
;
1216 xfs_fileoff_t startoffset_fsb
;
1221 trace_xfs_free_file_space(ip
);
1223 error
= xfs_qm_dqattach(ip
, 0);
1228 if (len
<= 0) /* if nothing being freed */
1230 rt
= XFS_IS_REALTIME_INODE(ip
);
1231 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
1232 endoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
+ len
);
1234 /* wait for the completion of any pending DIOs */
1235 inode_dio_wait(VFS_I(ip
));
1237 rounding
= max_t(xfs_off_t
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
1238 ioffset
= round_down(offset
, rounding
);
1239 iendoffset
= round_up(offset
+ len
, rounding
) - 1;
1240 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
, ioffset
,
1244 truncate_pagecache_range(VFS_I(ip
), ioffset
, iendoffset
);
1247 * Need to zero the stuff we're not freeing, on disk.
1248 * If it's a realtime file & can't use unwritten extents then we
1249 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1250 * will take care of it for us.
1252 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
1254 error
= xfs_bmapi_read(ip
, startoffset_fsb
, 1,
1258 ASSERT(nimap
== 0 || nimap
== 1);
1259 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1262 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1263 block
= imap
.br_startblock
;
1264 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
1266 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
1269 error
= xfs_bmapi_read(ip
, endoffset_fsb
- 1, 1,
1273 ASSERT(nimap
== 0 || nimap
== 1);
1274 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1275 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1277 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
1278 endoffset_fsb
-= mod
;
1281 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
1283 * One contiguous piece to clear
1285 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
1288 * Some full blocks, possibly two pieces to clear
1290 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
1291 error
= xfs_zero_remaining_bytes(ip
, offset
,
1292 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
1294 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
1295 error
= xfs_zero_remaining_bytes(ip
,
1296 XFS_FSB_TO_B(mp
, endoffset_fsb
),
1301 * free file space until done or until there is an error
1303 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1304 while (!error
&& !done
) {
1307 * allocate and setup the transaction. Allow this
1308 * transaction to dip into the reserve blocks to ensure
1309 * the freeing of the space succeeds at ENOSPC.
1311 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1312 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
, resblks
, 0);
1315 * check for running out of space
1319 * Free the transaction structure.
1321 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1322 xfs_trans_cancel(tp
);
1325 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1326 error
= xfs_trans_reserve_quota(tp
, mp
,
1327 ip
->i_udquot
, ip
->i_gdquot
, ip
->i_pdquot
,
1328 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
1332 xfs_trans_ijoin(tp
, ip
, 0);
1335 * issue the bunmapi() call to free the blocks
1337 xfs_bmap_init(&free_list
, &firstfsb
);
1338 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
1339 endoffset_fsb
- startoffset_fsb
,
1340 0, 2, &firstfsb
, &free_list
, &done
);
1345 * complete the transaction
1347 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1351 error
= xfs_trans_commit(tp
);
1352 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1359 xfs_bmap_cancel(&free_list
);
1361 xfs_trans_cancel(tp
);
1362 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1367 * Preallocate and zero a range of a file. This mechanism has the allocation
1368 * semantics of fallocate and in addition converts data in the range to zeroes.
1371 xfs_zero_file_space(
1372 struct xfs_inode
*ip
,
1376 struct xfs_mount
*mp
= ip
->i_mount
;
1380 trace_xfs_zero_file_space(ip
);
1382 blksize
= 1 << mp
->m_sb
.sb_blocklog
;
1385 * Punch a hole and prealloc the range. We use hole punch rather than
1386 * unwritten extent conversion for two reasons:
1388 * 1.) Hole punch handles partial block zeroing for us.
1390 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1391 * by virtue of the hole punch.
1393 error
= xfs_free_file_space(ip
, offset
, len
);
1397 error
= xfs_alloc_file_space(ip
, round_down(offset
, blksize
),
1398 round_up(offset
+ len
, blksize
) -
1399 round_down(offset
, blksize
),
1400 XFS_BMAPI_PREALLOC
);
1407 * @next_fsb will keep track of the extent currently undergoing shift.
1408 * @stop_fsb will keep track of the extent at which we have to stop.
1409 * If we are shifting left, we will start with block (offset + len) and
1410 * shift each extent till last extent.
1411 * If we are shifting right, we will start with last extent inside file space
1412 * and continue until we reach the block corresponding to offset.
1415 xfs_shift_file_space(
1416 struct xfs_inode
*ip
,
1419 enum shift_direction direction
)
1422 struct xfs_mount
*mp
= ip
->i_mount
;
1423 struct xfs_trans
*tp
;
1425 struct xfs_bmap_free free_list
;
1426 xfs_fsblock_t first_block
;
1427 xfs_fileoff_t stop_fsb
;
1428 xfs_fileoff_t next_fsb
;
1429 xfs_fileoff_t shift_fsb
;
1431 ASSERT(direction
== SHIFT_LEFT
|| direction
== SHIFT_RIGHT
);
1433 if (direction
== SHIFT_LEFT
) {
1434 next_fsb
= XFS_B_TO_FSB(mp
, offset
+ len
);
1435 stop_fsb
= XFS_B_TO_FSB(mp
, VFS_I(ip
)->i_size
);
1438 * If right shift, delegate the work of initialization of
1439 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1441 next_fsb
= NULLFSBLOCK
;
1442 stop_fsb
= XFS_B_TO_FSB(mp
, offset
);
1445 shift_fsb
= XFS_B_TO_FSB(mp
, len
);
1448 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1449 * into the accessible region of the file.
1451 if (xfs_can_free_eofblocks(ip
, true)) {
1452 error
= xfs_free_eofblocks(mp
, ip
, false);
1458 * Writeback and invalidate cache for the remainder of the file as we're
1459 * about to shift down every extent from offset to EOF.
1461 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
1465 error
= invalidate_inode_pages2_range(VFS_I(ip
)->i_mapping
,
1466 offset
>> PAGE_CACHE_SHIFT
, -1);
1471 * The extent shiting code works on extent granularity. So, if
1472 * stop_fsb is not the starting block of extent, we need to split
1473 * the extent at stop_fsb.
1475 if (direction
== SHIFT_RIGHT
) {
1476 error
= xfs_bmap_split_extent(ip
, stop_fsb
);
1481 while (!error
&& !done
) {
1482 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1484 * We would need to reserve permanent block for transaction.
1485 * This will come into picture when after shifting extent into
1486 * hole we found that adjacent extents can be merged which
1487 * may lead to freeing of a block during record update.
1489 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
,
1490 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0);
1492 xfs_trans_cancel(tp
);
1496 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1497 error
= xfs_trans_reserve_quota(tp
, mp
, ip
->i_udquot
,
1498 ip
->i_gdquot
, ip
->i_pdquot
,
1499 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0,
1500 XFS_QMOPT_RES_REGBLKS
);
1502 goto out_trans_cancel
;
1504 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1506 xfs_bmap_init(&free_list
, &first_block
);
1509 * We are using the write transaction in which max 2 bmbt
1510 * updates are allowed
1512 error
= xfs_bmap_shift_extents(tp
, ip
, &next_fsb
, shift_fsb
,
1513 &done
, stop_fsb
, &first_block
, &free_list
,
1514 direction
, XFS_BMAP_MAX_SHIFT_EXTENTS
);
1516 goto out_bmap_cancel
;
1518 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1520 goto out_bmap_cancel
;
1522 error
= xfs_trans_commit(tp
);
1528 xfs_bmap_cancel(&free_list
);
1530 xfs_trans_cancel(tp
);
1535 * xfs_collapse_file_space()
1536 * This routine frees disk space and shift extent for the given file.
1537 * The first thing we do is to free data blocks in the specified range
1538 * by calling xfs_free_file_space(). It would also sync dirty data
1539 * and invalidate page cache over the region on which collapse range
1540 * is working. And Shift extent records to the left to cover a hole.
1547 xfs_collapse_file_space(
1548 struct xfs_inode
*ip
,
1554 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1555 trace_xfs_collapse_file_space(ip
);
1557 error
= xfs_free_file_space(ip
, offset
, len
);
1561 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_LEFT
);
1565 * xfs_insert_file_space()
1566 * This routine create hole space by shifting extents for the given file.
1567 * The first thing we do is to sync dirty data and invalidate page cache
1568 * over the region on which insert range is working. And split an extent
1569 * to two extents at given offset by calling xfs_bmap_split_extent.
1570 * And shift all extent records which are laying between [offset,
1571 * last allocated extent] to the right to reserve hole range.
1577 xfs_insert_file_space(
1578 struct xfs_inode
*ip
,
1582 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1583 trace_xfs_insert_file_space(ip
);
1585 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_RIGHT
);
1589 * We need to check that the format of the data fork in the temporary inode is
1590 * valid for the target inode before doing the swap. This is not a problem with
1591 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1592 * data fork depending on the space the attribute fork is taking so we can get
1593 * invalid formats on the target inode.
1595 * E.g. target has space for 7 extents in extent format, temp inode only has
1596 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1597 * btree, but when swapped it needs to be in extent format. Hence we can't just
1598 * blindly swap data forks on attr2 filesystems.
1600 * Note that we check the swap in both directions so that we don't end up with
1601 * a corrupt temporary inode, either.
1603 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1604 * inode will prevent this situation from occurring, so all we do here is
1605 * reject and log the attempt. basically we are putting the responsibility on
1606 * userspace to get this right.
1609 xfs_swap_extents_check_format(
1610 xfs_inode_t
*ip
, /* target inode */
1611 xfs_inode_t
*tip
) /* tmp inode */
1614 /* Should never get a local format */
1615 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
||
1616 tip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
1620 * if the target inode has less extents that then temporary inode then
1621 * why did userspace call us?
1623 if (ip
->i_d
.di_nextents
< tip
->i_d
.di_nextents
)
1627 * if the target inode is in extent form and the temp inode is in btree
1628 * form then we will end up with the target inode in the wrong format
1629 * as we already know there are less extents in the temp inode.
1631 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1632 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
)
1635 /* Check temp in extent form to max in target */
1636 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1637 XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) >
1638 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1641 /* Check target in extent form to max in temp */
1642 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1643 XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) >
1644 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1648 * If we are in a btree format, check that the temp root block will fit
1649 * in the target and that it has enough extents to be in btree format
1652 * Note that we have to be careful to allow btree->extent conversions
1653 * (a common defrag case) which will occur when the temp inode is in
1656 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1657 if (XFS_IFORK_BOFF(ip
) &&
1658 XFS_BMAP_BMDR_SPACE(tip
->i_df
.if_broot
) > XFS_IFORK_BOFF(ip
))
1660 if (XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) <=
1661 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1665 /* Reciprocal target->temp btree format checks */
1666 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1667 if (XFS_IFORK_BOFF(tip
) &&
1668 XFS_BMAP_BMDR_SPACE(ip
->i_df
.if_broot
) > XFS_IFORK_BOFF(tip
))
1670 if (XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) <=
1671 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1679 xfs_swap_extent_flush(
1680 struct xfs_inode
*ip
)
1684 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
1687 truncate_pagecache_range(VFS_I(ip
), 0, -1);
1689 /* Verify O_DIRECT for ftmp */
1690 if (VFS_I(ip
)->i_mapping
->nrpages
)
1697 xfs_inode_t
*ip
, /* target inode */
1698 xfs_inode_t
*tip
, /* tmp inode */
1701 xfs_mount_t
*mp
= ip
->i_mount
;
1703 xfs_bstat_t
*sbp
= &sxp
->sx_stat
;
1704 xfs_ifork_t
*tempifp
, *ifp
, *tifp
;
1705 int src_log_flags
, target_log_flags
;
1712 tempifp
= kmem_alloc(sizeof(xfs_ifork_t
), KM_MAYFAIL
);
1719 * Lock the inodes against other IO, page faults and truncate to
1720 * begin with. Then we can ensure the inodes are flushed and have no
1721 * page cache safely. Once we have done this we can take the ilocks and
1722 * do the rest of the checks.
1724 lock_flags
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1725 xfs_lock_two_inodes(ip
, tip
, XFS_IOLOCK_EXCL
);
1726 xfs_lock_two_inodes(ip
, tip
, XFS_MMAPLOCK_EXCL
);
1728 /* Verify that both files have the same format */
1729 if ((ip
->i_d
.di_mode
& S_IFMT
) != (tip
->i_d
.di_mode
& S_IFMT
)) {
1734 /* Verify both files are either real-time or non-realtime */
1735 if (XFS_IS_REALTIME_INODE(ip
) != XFS_IS_REALTIME_INODE(tip
)) {
1740 error
= xfs_swap_extent_flush(ip
);
1743 error
= xfs_swap_extent_flush(tip
);
1747 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SWAPEXT
);
1748 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
1750 xfs_trans_cancel(tp
);
1755 * Lock and join the inodes to the tansaction so that transaction commit
1756 * or cancel will unlock the inodes from this point onwards.
1758 xfs_lock_two_inodes(ip
, tip
, XFS_ILOCK_EXCL
);
1759 lock_flags
|= XFS_ILOCK_EXCL
;
1760 xfs_trans_ijoin(tp
, ip
, lock_flags
);
1761 xfs_trans_ijoin(tp
, tip
, lock_flags
);
1764 /* Verify all data are being swapped */
1765 if (sxp
->sx_offset
!= 0 ||
1766 sxp
->sx_length
!= ip
->i_d
.di_size
||
1767 sxp
->sx_length
!= tip
->i_d
.di_size
) {
1769 goto out_trans_cancel
;
1772 trace_xfs_swap_extent_before(ip
, 0);
1773 trace_xfs_swap_extent_before(tip
, 1);
1775 /* check inode formats now that data is flushed */
1776 error
= xfs_swap_extents_check_format(ip
, tip
);
1779 "%s: inode 0x%llx format is incompatible for exchanging.",
1780 __func__
, ip
->i_ino
);
1781 goto out_trans_cancel
;
1785 * Compare the current change & modify times with that
1786 * passed in. If they differ, we abort this swap.
1787 * This is the mechanism used to ensure the calling
1788 * process that the file was not changed out from
1791 if ((sbp
->bs_ctime
.tv_sec
!= VFS_I(ip
)->i_ctime
.tv_sec
) ||
1792 (sbp
->bs_ctime
.tv_nsec
!= VFS_I(ip
)->i_ctime
.tv_nsec
) ||
1793 (sbp
->bs_mtime
.tv_sec
!= VFS_I(ip
)->i_mtime
.tv_sec
) ||
1794 (sbp
->bs_mtime
.tv_nsec
!= VFS_I(ip
)->i_mtime
.tv_nsec
)) {
1796 goto out_trans_cancel
;
1799 * Count the number of extended attribute blocks
1801 if ( ((XFS_IFORK_Q(ip
) != 0) && (ip
->i_d
.di_anextents
> 0)) &&
1802 (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1803 error
= xfs_bmap_count_blocks(tp
, ip
, XFS_ATTR_FORK
, &aforkblks
);
1805 goto out_trans_cancel
;
1807 if ( ((XFS_IFORK_Q(tip
) != 0) && (tip
->i_d
.di_anextents
> 0)) &&
1808 (tip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1809 error
= xfs_bmap_count_blocks(tp
, tip
, XFS_ATTR_FORK
,
1812 goto out_trans_cancel
;
1816 * Before we've swapped the forks, lets set the owners of the forks
1817 * appropriately. We have to do this as we are demand paging the btree
1818 * buffers, and so the validation done on read will expect the owner
1819 * field to be correctly set. Once we change the owners, we can swap the
1822 * Note the trickiness in setting the log flags - we set the owner log
1823 * flag on the opposite inode (i.e. the inode we are setting the new
1824 * owner to be) because once we swap the forks and log that, log
1825 * recovery is going to see the fork as owned by the swapped inode,
1826 * not the pre-swapped inodes.
1828 src_log_flags
= XFS_ILOG_CORE
;
1829 target_log_flags
= XFS_ILOG_CORE
;
1830 if (ip
->i_d
.di_version
== 3 &&
1831 ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1832 target_log_flags
|= XFS_ILOG_DOWNER
;
1833 error
= xfs_bmbt_change_owner(tp
, ip
, XFS_DATA_FORK
,
1836 goto out_trans_cancel
;
1839 if (tip
->i_d
.di_version
== 3 &&
1840 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1841 src_log_flags
|= XFS_ILOG_DOWNER
;
1842 error
= xfs_bmbt_change_owner(tp
, tip
, XFS_DATA_FORK
,
1845 goto out_trans_cancel
;
1849 * Swap the data forks of the inodes
1853 *tempifp
= *ifp
; /* struct copy */
1854 *ifp
= *tifp
; /* struct copy */
1855 *tifp
= *tempifp
; /* struct copy */
1858 * Fix the on-disk inode values
1860 tmp
= (__uint64_t
)ip
->i_d
.di_nblocks
;
1861 ip
->i_d
.di_nblocks
= tip
->i_d
.di_nblocks
- taforkblks
+ aforkblks
;
1862 tip
->i_d
.di_nblocks
= tmp
+ taforkblks
- aforkblks
;
1864 tmp
= (__uint64_t
) ip
->i_d
.di_nextents
;
1865 ip
->i_d
.di_nextents
= tip
->i_d
.di_nextents
;
1866 tip
->i_d
.di_nextents
= tmp
;
1868 tmp
= (__uint64_t
) ip
->i_d
.di_format
;
1869 ip
->i_d
.di_format
= tip
->i_d
.di_format
;
1870 tip
->i_d
.di_format
= tmp
;
1873 * The extents in the source inode could still contain speculative
1874 * preallocation beyond EOF (e.g. the file is open but not modified
1875 * while defrag is in progress). In that case, we need to copy over the
1876 * number of delalloc blocks the data fork in the source inode is
1877 * tracking beyond EOF so that when the fork is truncated away when the
1878 * temporary inode is unlinked we don't underrun the i_delayed_blks
1879 * counter on that inode.
1881 ASSERT(tip
->i_delayed_blks
== 0);
1882 tip
->i_delayed_blks
= ip
->i_delayed_blks
;
1883 ip
->i_delayed_blks
= 0;
1885 switch (ip
->i_d
.di_format
) {
1886 case XFS_DINODE_FMT_EXTENTS
:
1887 /* If the extents fit in the inode, fix the
1888 * pointer. Otherwise it's already NULL or
1889 * pointing to the extent.
1891 if (ip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1892 ifp
->if_u1
.if_extents
=
1893 ifp
->if_u2
.if_inline_ext
;
1895 src_log_flags
|= XFS_ILOG_DEXT
;
1897 case XFS_DINODE_FMT_BTREE
:
1898 ASSERT(ip
->i_d
.di_version
< 3 ||
1899 (src_log_flags
& XFS_ILOG_DOWNER
));
1900 src_log_flags
|= XFS_ILOG_DBROOT
;
1904 switch (tip
->i_d
.di_format
) {
1905 case XFS_DINODE_FMT_EXTENTS
:
1906 /* If the extents fit in the inode, fix the
1907 * pointer. Otherwise it's already NULL or
1908 * pointing to the extent.
1910 if (tip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1911 tifp
->if_u1
.if_extents
=
1912 tifp
->if_u2
.if_inline_ext
;
1914 target_log_flags
|= XFS_ILOG_DEXT
;
1916 case XFS_DINODE_FMT_BTREE
:
1917 target_log_flags
|= XFS_ILOG_DBROOT
;
1918 ASSERT(tip
->i_d
.di_version
< 3 ||
1919 (target_log_flags
& XFS_ILOG_DOWNER
));
1923 xfs_trans_log_inode(tp
, ip
, src_log_flags
);
1924 xfs_trans_log_inode(tp
, tip
, target_log_flags
);
1927 * If this is a synchronous mount, make sure that the
1928 * transaction goes to disk before returning to the user.
1930 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1931 xfs_trans_set_sync(tp
);
1933 error
= xfs_trans_commit(tp
);
1935 trace_xfs_swap_extent_after(ip
, 0);
1936 trace_xfs_swap_extent_after(tip
, 1);
1942 xfs_iunlock(ip
, lock_flags
);
1943 xfs_iunlock(tip
, lock_flags
);
1947 xfs_trans_cancel(tp
);