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 (error
== -EFSCORRUPTED
) ?
129 SHUTDOWN_CORRUPT_INCORE
:
130 SHUTDOWN_META_IO_ERROR
);
136 * Get an EFD and free each extent in the list, logging to the EFD in
137 * the process. The remaining bmap free list is cleaned up by the caller
140 efd
= xfs_trans_get_efd(*tp
, efi
, flist
->xbf_count
);
141 for (free
= flist
->xbf_first
; free
!= NULL
; free
= next
) {
142 next
= free
->xbfi_next
;
144 error
= xfs_trans_free_extent(*tp
, efd
, free
->xbfi_startblock
,
145 free
->xbfi_blockcount
);
149 xfs_bmap_del_free(flist
, NULL
, free
);
157 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
159 xfs_alloctype_t atype
= 0; /* type for allocation routines */
160 int error
; /* error return value */
161 xfs_mount_t
*mp
; /* mount point structure */
162 xfs_extlen_t prod
= 0; /* product factor for allocators */
163 xfs_extlen_t ralen
= 0; /* realtime allocation length */
164 xfs_extlen_t align
; /* minimum allocation alignment */
167 mp
= ap
->ip
->i_mount
;
168 align
= xfs_get_extsz_hint(ap
->ip
);
169 prod
= align
/ mp
->m_sb
.sb_rextsize
;
170 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
171 align
, 1, ap
->eof
, 0,
172 ap
->conv
, &ap
->offset
, &ap
->length
);
176 ASSERT(ap
->length
% mp
->m_sb
.sb_rextsize
== 0);
179 * If the offset & length are not perfectly aligned
180 * then kill prod, it will just get us in trouble.
182 if (do_mod(ap
->offset
, align
) || ap
->length
% align
)
185 * Set ralen to be the actual requested length in rtextents.
187 ralen
= ap
->length
/ mp
->m_sb
.sb_rextsize
;
189 * If the old value was close enough to MAXEXTLEN that
190 * we rounded up to it, cut it back so it's valid again.
191 * Note that if it's a really large request (bigger than
192 * MAXEXTLEN), we don't hear about that number, and can't
193 * adjust the starting point to match it.
195 if (ralen
* mp
->m_sb
.sb_rextsize
>= MAXEXTLEN
)
196 ralen
= MAXEXTLEN
/ mp
->m_sb
.sb_rextsize
;
199 * Lock out modifications to both the RT bitmap and summary inodes
201 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
202 xfs_trans_ijoin(ap
->tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
203 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
);
204 xfs_trans_ijoin(ap
->tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
207 * If it's an allocation to an empty file at offset 0,
208 * pick an extent that will space things out in the rt area.
210 if (ap
->eof
&& ap
->offset
== 0) {
211 xfs_rtblock_t
uninitialized_var(rtx
); /* realtime extent no */
213 error
= xfs_rtpick_extent(mp
, ap
->tp
, ralen
, &rtx
);
216 ap
->blkno
= rtx
* mp
->m_sb
.sb_rextsize
;
221 xfs_bmap_adjacent(ap
);
224 * Realtime allocation, done through xfs_rtallocate_extent.
226 atype
= ap
->blkno
== 0 ? XFS_ALLOCTYPE_ANY_AG
: XFS_ALLOCTYPE_NEAR_BNO
;
227 do_div(ap
->blkno
, mp
->m_sb
.sb_rextsize
);
230 if ((error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1, ap
->length
,
231 &ralen
, atype
, ap
->wasdel
, prod
, &rtb
)))
233 if (rtb
== NULLFSBLOCK
&& prod
> 1 &&
234 (error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1,
235 ap
->length
, &ralen
, atype
,
236 ap
->wasdel
, 1, &rtb
)))
239 if (ap
->blkno
!= NULLFSBLOCK
) {
240 ap
->blkno
*= mp
->m_sb
.sb_rextsize
;
241 ralen
*= mp
->m_sb
.sb_rextsize
;
243 ap
->ip
->i_d
.di_nblocks
+= ralen
;
244 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
246 ap
->ip
->i_delayed_blks
-= ralen
;
248 * Adjust the disk quota also. This was reserved
251 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
252 ap
->wasdel
? XFS_TRANS_DQ_DELRTBCOUNT
:
253 XFS_TRANS_DQ_RTBCOUNT
, (long) ralen
);
255 /* Zero the extent if we were asked to do so */
256 if (ap
->userdata
& XFS_ALLOC_USERDATA_ZERO
) {
257 error
= xfs_zero_extent(ap
->ip
, ap
->blkno
, ap
->length
);
268 * Check if the endoff is outside the last extent. If so the caller will grow
269 * the allocation to a stripe unit boundary. All offsets are considered outside
270 * the end of file for an empty fork, so 1 is returned in *eof in that case.
274 struct xfs_inode
*ip
,
275 xfs_fileoff_t endoff
,
279 struct xfs_bmbt_irec rec
;
282 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, eof
);
286 *eof
= endoff
>= rec
.br_startoff
+ rec
.br_blockcount
;
291 * Extent tree block counting routines.
295 * Count leaf blocks given a range of extent records.
298 xfs_bmap_count_leaves(
306 for (b
= 0; b
< numrecs
; b
++) {
307 xfs_bmbt_rec_host_t
*frp
= xfs_iext_get_ext(ifp
, idx
+ b
);
308 *count
+= xfs_bmbt_get_blockcount(frp
);
313 * Count leaf blocks given a range of extent records originally
317 xfs_bmap_disk_count_leaves(
318 struct xfs_mount
*mp
,
319 struct xfs_btree_block
*block
,
326 for (b
= 1; b
<= numrecs
; b
++) {
327 frp
= XFS_BMBT_REC_ADDR(mp
, block
, b
);
328 *count
+= xfs_bmbt_disk_get_blockcount(frp
);
333 * Recursively walks each level of a btree
334 * to count total fsblocks in use.
336 STATIC
int /* error */
338 xfs_mount_t
*mp
, /* file system mount point */
339 xfs_trans_t
*tp
, /* transaction pointer */
340 xfs_ifork_t
*ifp
, /* inode fork pointer */
341 xfs_fsblock_t blockno
, /* file system block number */
342 int levelin
, /* level in btree */
343 int *count
) /* Count of blocks */
349 xfs_fsblock_t bno
= blockno
;
350 xfs_fsblock_t nextbno
;
351 struct xfs_btree_block
*block
, *nextblock
;
354 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
, XFS_BMAP_BTREE_REF
,
359 block
= XFS_BUF_TO_BLOCK(bp
);
362 /* Not at node above leaves, count this level of nodes */
363 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
364 while (nextbno
!= NULLFSBLOCK
) {
365 error
= xfs_btree_read_bufl(mp
, tp
, nextbno
, 0, &nbp
,
371 nextblock
= XFS_BUF_TO_BLOCK(nbp
);
372 nextbno
= be64_to_cpu(nextblock
->bb_u
.l
.bb_rightsib
);
373 xfs_trans_brelse(tp
, nbp
);
376 /* Dive to the next level */
377 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
378 bno
= be64_to_cpu(*pp
);
379 if (unlikely((error
=
380 xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
)) < 0)) {
381 xfs_trans_brelse(tp
, bp
);
382 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
383 XFS_ERRLEVEL_LOW
, mp
);
384 return -EFSCORRUPTED
;
386 xfs_trans_brelse(tp
, bp
);
388 /* count all level 1 nodes and their leaves */
390 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
391 numrecs
= be16_to_cpu(block
->bb_numrecs
);
392 xfs_bmap_disk_count_leaves(mp
, block
, numrecs
, count
);
393 xfs_trans_brelse(tp
, bp
);
394 if (nextbno
== NULLFSBLOCK
)
397 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
403 block
= XFS_BUF_TO_BLOCK(bp
);
410 * Count fsblocks of the given fork.
413 xfs_bmap_count_blocks(
414 xfs_trans_t
*tp
, /* transaction pointer */
415 xfs_inode_t
*ip
, /* incore inode */
416 int whichfork
, /* data or attr fork */
417 int *count
) /* out: count of blocks */
419 struct xfs_btree_block
*block
; /* current btree block */
420 xfs_fsblock_t bno
; /* block # of "block" */
421 xfs_ifork_t
*ifp
; /* fork structure */
422 int level
; /* btree level, for checking */
423 xfs_mount_t
*mp
; /* file system mount structure */
424 __be64
*pp
; /* pointer to block address */
428 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
429 if ( XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
) {
430 xfs_bmap_count_leaves(ifp
, 0,
431 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
),
437 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
439 block
= ifp
->if_broot
;
440 level
= be16_to_cpu(block
->bb_level
);
442 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
443 bno
= be64_to_cpu(*pp
);
444 ASSERT(bno
!= NULLFSBLOCK
);
445 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
446 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
448 if (unlikely(xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
) < 0)) {
449 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW
,
451 return -EFSCORRUPTED
;
458 * returns 1 for success, 0 if we failed to map the extent.
461 xfs_getbmapx_fix_eof_hole(
462 xfs_inode_t
*ip
, /* xfs incore inode pointer */
463 struct getbmapx
*out
, /* output structure */
464 int prealloced
, /* this is a file with
465 * preallocated data space */
466 __int64_t end
, /* last block requested */
467 xfs_fsblock_t startblock
)
470 xfs_mount_t
*mp
; /* file system mount point */
471 xfs_ifork_t
*ifp
; /* inode fork pointer */
472 xfs_extnum_t lastx
; /* last extent pointer */
473 xfs_fileoff_t fileblock
;
475 if (startblock
== HOLESTARTBLOCK
) {
478 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)));
479 fixlen
-= out
->bmv_offset
;
480 if (prealloced
&& out
->bmv_offset
+ out
->bmv_length
== end
) {
481 /* Came to hole at EOF. Trim it. */
484 out
->bmv_length
= fixlen
;
487 if (startblock
== DELAYSTARTBLOCK
)
490 out
->bmv_block
= xfs_fsb_to_db(ip
, startblock
);
491 fileblock
= XFS_BB_TO_FSB(ip
->i_mount
, out
->bmv_offset
);
492 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
493 if (xfs_iext_bno_to_ext(ifp
, fileblock
, &lastx
) &&
494 (lastx
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))-1))
495 out
->bmv_oflags
|= BMV_OF_LAST
;
502 * Get inode's extents as described in bmv, and format for output.
503 * Calls formatter to fill the user's buffer until all extents
504 * are mapped, until the passed-in bmv->bmv_count slots have
505 * been filled, or until the formatter short-circuits the loop,
506 * if it is tracking filled-in extents on its own.
511 struct getbmapx
*bmv
, /* user bmap structure */
512 xfs_bmap_format_t formatter
, /* format to user */
513 void *arg
) /* formatter arg */
515 __int64_t bmvend
; /* last block requested */
516 int error
= 0; /* return value */
517 __int64_t fixlen
; /* length for -1 case */
518 int i
; /* extent number */
519 int lock
; /* lock state */
520 xfs_bmbt_irec_t
*map
; /* buffer for user's data */
521 xfs_mount_t
*mp
; /* file system mount point */
522 int nex
; /* # of user extents can do */
523 int nexleft
; /* # of user extents left */
524 int subnex
; /* # of bmapi's can do */
525 int nmap
; /* number of map entries */
526 struct getbmapx
*out
; /* output structure */
527 int whichfork
; /* data or attr fork */
528 int prealloced
; /* this is a file with
529 * preallocated data space */
530 int iflags
; /* interface flags */
531 int bmapi_flags
; /* flags for xfs_bmapi */
535 iflags
= bmv
->bmv_iflags
;
536 whichfork
= iflags
& BMV_IF_ATTRFORK
? XFS_ATTR_FORK
: XFS_DATA_FORK
;
538 if (whichfork
== XFS_ATTR_FORK
) {
539 if (XFS_IFORK_Q(ip
)) {
540 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
&&
541 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_BTREE
&&
542 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)
545 ip
->i_d
.di_aformat
!= 0 &&
546 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
)) {
547 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW
,
549 return -EFSCORRUPTED
;
555 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
&&
556 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
&&
557 ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
)
560 if (xfs_get_extsz_hint(ip
) ||
561 ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
|XFS_DIFLAG_APPEND
)){
563 fixlen
= mp
->m_super
->s_maxbytes
;
566 fixlen
= XFS_ISIZE(ip
);
570 if (bmv
->bmv_length
== -1) {
571 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, fixlen
));
573 max_t(__int64_t
, fixlen
- bmv
->bmv_offset
, 0);
574 } else if (bmv
->bmv_length
== 0) {
575 bmv
->bmv_entries
= 0;
577 } else if (bmv
->bmv_length
< 0) {
581 nex
= bmv
->bmv_count
- 1;
584 bmvend
= bmv
->bmv_offset
+ bmv
->bmv_length
;
587 if (bmv
->bmv_count
> ULONG_MAX
/ sizeof(struct getbmapx
))
589 out
= kmem_zalloc_large(bmv
->bmv_count
* sizeof(struct getbmapx
), 0);
593 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
594 if (whichfork
== XFS_DATA_FORK
) {
595 if (!(iflags
& BMV_IF_DELALLOC
) &&
596 (ip
->i_delayed_blks
|| XFS_ISIZE(ip
) > ip
->i_d
.di_size
)) {
597 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
599 goto out_unlock_iolock
;
602 * Even after flushing the inode, there can still be
603 * delalloc blocks on the inode beyond EOF due to
604 * speculative preallocation. These are not removed
605 * until the release function is called or the inode
606 * is inactivated. Hence we cannot assert here that
607 * ip->i_delayed_blks == 0.
611 lock
= xfs_ilock_data_map_shared(ip
);
613 lock
= xfs_ilock_attr_map_shared(ip
);
617 * Don't let nex be bigger than the number of extents
618 * we can have assuming alternating holes and real extents.
620 if (nex
> XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1)
621 nex
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1;
623 bmapi_flags
= xfs_bmapi_aflag(whichfork
);
624 if (!(iflags
& BMV_IF_PREALLOC
))
625 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
628 * Allocate enough space to handle "subnex" maps at a time.
632 map
= kmem_alloc(subnex
* sizeof(*map
), KM_MAYFAIL
| KM_NOFS
);
634 goto out_unlock_ilock
;
636 bmv
->bmv_entries
= 0;
638 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0 &&
639 (whichfork
== XFS_ATTR_FORK
|| !(iflags
& BMV_IF_DELALLOC
))) {
647 nmap
= (nexleft
> subnex
) ? subnex
: nexleft
;
648 error
= xfs_bmapi_read(ip
, XFS_BB_TO_FSBT(mp
, bmv
->bmv_offset
),
649 XFS_BB_TO_FSB(mp
, bmv
->bmv_length
),
650 map
, &nmap
, bmapi_flags
);
653 ASSERT(nmap
<= subnex
);
655 for (i
= 0; i
< nmap
&& nexleft
&& bmv
->bmv_length
; i
++) {
656 out
[cur_ext
].bmv_oflags
= 0;
657 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
)
658 out
[cur_ext
].bmv_oflags
|= BMV_OF_PREALLOC
;
659 else if (map
[i
].br_startblock
== DELAYSTARTBLOCK
)
660 out
[cur_ext
].bmv_oflags
|= BMV_OF_DELALLOC
;
661 out
[cur_ext
].bmv_offset
=
662 XFS_FSB_TO_BB(mp
, map
[i
].br_startoff
);
663 out
[cur_ext
].bmv_length
=
664 XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
665 out
[cur_ext
].bmv_unused1
= 0;
666 out
[cur_ext
].bmv_unused2
= 0;
669 * delayed allocation extents that start beyond EOF can
670 * occur due to speculative EOF allocation when the
671 * delalloc extent is larger than the largest freespace
672 * extent at conversion time. These extents cannot be
673 * converted by data writeback, so can exist here even
674 * if we are not supposed to be finding delalloc
677 if (map
[i
].br_startblock
== DELAYSTARTBLOCK
&&
678 map
[i
].br_startoff
<= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)))
679 ASSERT((iflags
& BMV_IF_DELALLOC
) != 0);
681 if (map
[i
].br_startblock
== HOLESTARTBLOCK
&&
682 whichfork
== XFS_ATTR_FORK
) {
683 /* came to the end of attribute fork */
684 out
[cur_ext
].bmv_oflags
|= BMV_OF_LAST
;
688 if (!xfs_getbmapx_fix_eof_hole(ip
, &out
[cur_ext
],
690 map
[i
].br_startblock
))
694 out
[cur_ext
].bmv_offset
+
695 out
[cur_ext
].bmv_length
;
697 max_t(__int64_t
, 0, bmvend
- bmv
->bmv_offset
);
700 * In case we don't want to return the hole,
701 * don't increase cur_ext so that we can reuse
702 * it in the next loop.
704 if ((iflags
& BMV_IF_NO_HOLES
) &&
705 map
[i
].br_startblock
== HOLESTARTBLOCK
) {
706 memset(&out
[cur_ext
], 0, sizeof(out
[cur_ext
]));
714 } while (nmap
&& nexleft
&& bmv
->bmv_length
);
719 xfs_iunlock(ip
, lock
);
721 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
723 for (i
= 0; i
< cur_ext
; i
++) {
724 int full
= 0; /* user array is full */
726 /* format results & advance arg */
727 error
= formatter(&arg
, &out
[i
], &full
);
737 * dead simple method of punching delalyed allocation blocks from a range in
738 * the inode. Walks a block at a time so will be slow, but is only executed in
739 * rare error cases so the overhead is not critical. This will always punch out
740 * both the start and end blocks, even if the ranges only partially overlap
741 * them, so it is up to the caller to ensure that partial blocks are not
745 xfs_bmap_punch_delalloc_range(
746 struct xfs_inode
*ip
,
747 xfs_fileoff_t start_fsb
,
748 xfs_fileoff_t length
)
750 xfs_fileoff_t remaining
= length
;
753 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
757 xfs_bmbt_irec_t imap
;
759 xfs_fsblock_t firstblock
;
760 xfs_bmap_free_t flist
;
763 * Map the range first and check that it is a delalloc extent
764 * before trying to unmap the range. Otherwise we will be
765 * trying to remove a real extent (which requires a
766 * transaction) or a hole, which is probably a bad idea...
768 error
= xfs_bmapi_read(ip
, start_fsb
, 1, &imap
, &nimaps
,
772 /* something screwed, just bail */
773 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
774 xfs_alert(ip
->i_mount
,
775 "Failed delalloc mapping lookup ino %lld fsb %lld.",
776 ip
->i_ino
, start_fsb
);
784 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
785 /* been converted, ignore */
788 WARN_ON(imap
.br_blockcount
== 0);
791 * Note: while we initialise the firstblock/flist pair, they
792 * should never be used because blocks should never be
793 * allocated or freed for a delalloc extent and hence we need
794 * don't cancel or finish them after the xfs_bunmapi() call.
796 xfs_bmap_init(&flist
, &firstblock
);
797 error
= xfs_bunmapi(NULL
, ip
, start_fsb
, 1, 0, 1, &firstblock
,
802 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
806 } while(remaining
> 0);
812 * Test whether it is appropriate to check an inode for and free post EOF
813 * blocks. The 'force' parameter determines whether we should also consider
814 * regular files that are marked preallocated or append-only.
817 xfs_can_free_eofblocks(struct xfs_inode
*ip
, bool force
)
819 /* prealloc/delalloc exists only on regular files */
820 if (!S_ISREG(VFS_I(ip
)->i_mode
))
824 * Zero sized files with no cached pages and delalloc blocks will not
825 * have speculative prealloc/delalloc blocks to remove.
827 if (VFS_I(ip
)->i_size
== 0 &&
828 VFS_I(ip
)->i_mapping
->nrpages
== 0 &&
829 ip
->i_delayed_blks
== 0)
832 /* If we haven't read in the extent list, then don't do it now. */
833 if (!(ip
->i_df
.if_flags
& XFS_IFEXTENTS
))
837 * Do not free real preallocated or append-only files unless the file
838 * has delalloc blocks and we are forced to remove them.
840 if (ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
))
841 if (!force
|| ip
->i_delayed_blks
== 0)
848 * This is called by xfs_inactive to free any blocks beyond eof
849 * when the link count isn't zero and by xfs_dm_punch_hole() when
850 * punching a hole to EOF.
860 xfs_fileoff_t end_fsb
;
861 xfs_fileoff_t last_fsb
;
862 xfs_filblks_t map_len
;
864 xfs_bmbt_irec_t imap
;
867 * Figure out if there are any blocks beyond the end
868 * of the file. If not, then there is nothing to do.
870 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_ISIZE(ip
));
871 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
872 if (last_fsb
<= end_fsb
)
874 map_len
= last_fsb
- end_fsb
;
877 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
878 error
= xfs_bmapi_read(ip
, end_fsb
, map_len
, &imap
, &nimaps
, 0);
879 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
881 if (!error
&& (nimaps
!= 0) &&
882 (imap
.br_startblock
!= HOLESTARTBLOCK
||
883 ip
->i_delayed_blks
)) {
885 * Attach the dquots to the inode up front.
887 error
= xfs_qm_dqattach(ip
, 0);
892 * There are blocks after the end of file.
893 * Free them up now by truncating the file to
896 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
899 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
900 xfs_trans_cancel(tp
);
905 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
907 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
908 xfs_trans_cancel(tp
);
910 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
914 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
915 xfs_trans_ijoin(tp
, ip
, 0);
918 * Do not update the on-disk file size. If we update the
919 * on-disk file size and then the system crashes before the
920 * contents of the file are flushed to disk then the files
921 * may be full of holes (ie NULL files bug).
923 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
,
927 * If we get an error at this point we simply don't
928 * bother truncating the file.
930 xfs_trans_cancel(tp
);
932 error
= xfs_trans_commit(tp
);
934 xfs_inode_clear_eofblocks_tag(ip
);
937 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
939 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
945 xfs_alloc_file_space(
946 struct xfs_inode
*ip
,
951 xfs_mount_t
*mp
= ip
->i_mount
;
953 xfs_filblks_t allocated_fsb
;
954 xfs_filblks_t allocatesize_fsb
;
955 xfs_extlen_t extsz
, temp
;
956 xfs_fileoff_t startoffset_fsb
;
957 xfs_fsblock_t firstfsb
;
962 xfs_bmbt_irec_t imaps
[1], *imapp
;
963 xfs_bmap_free_t free_list
;
964 uint qblocks
, resblks
, resrtextents
;
967 trace_xfs_alloc_file_space(ip
);
969 if (XFS_FORCED_SHUTDOWN(mp
))
972 error
= xfs_qm_dqattach(ip
, 0);
979 rt
= XFS_IS_REALTIME_INODE(ip
);
980 extsz
= xfs_get_extsz_hint(ip
);
985 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
986 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
989 * Allocate file space until done or until there is an error
991 while (allocatesize_fsb
&& !error
) {
995 * Determine space reservations for data/realtime.
997 if (unlikely(extsz
)) {
1001 e
= startoffset_fsb
+ allocatesize_fsb
;
1002 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
1004 if ((temp
= do_mod(e
, extsz
)))
1008 e
= allocatesize_fsb
;
1012 * The transaction reservation is limited to a 32-bit block
1013 * count, hence we need to limit the number of blocks we are
1014 * trying to reserve to avoid an overflow. We can't allocate
1015 * more than @nimaps extents, and an extent is limited on disk
1016 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1018 resblks
= min_t(xfs_fileoff_t
, (e
- s
), (MAXEXTLEN
* nimaps
));
1020 resrtextents
= qblocks
= resblks
;
1021 resrtextents
/= mp
->m_sb
.sb_rextsize
;
1022 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1023 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
1026 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resblks
);
1027 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
1031 * Allocate and setup the transaction.
1033 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1034 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
,
1035 resblks
, resrtextents
);
1037 * Check for running out of space
1041 * Free the transaction structure.
1043 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1044 xfs_trans_cancel(tp
);
1047 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1048 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
1053 xfs_trans_ijoin(tp
, ip
, 0);
1055 xfs_bmap_init(&free_list
, &firstfsb
);
1056 error
= xfs_bmapi_write(tp
, ip
, startoffset_fsb
,
1057 allocatesize_fsb
, alloc_type
, &firstfsb
,
1058 resblks
, imapp
, &nimaps
, &free_list
);
1063 * Complete the transaction
1065 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1069 error
= xfs_trans_commit(tp
);
1070 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1074 allocated_fsb
= imapp
->br_blockcount
;
1081 startoffset_fsb
+= allocated_fsb
;
1082 allocatesize_fsb
-= allocated_fsb
;
1087 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1088 xfs_bmap_cancel(&free_list
);
1089 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
1091 error1
: /* Just cancel transaction */
1092 xfs_trans_cancel(tp
);
1093 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1098 * Zero file bytes between startoff and endoff inclusive.
1099 * The iolock is held exclusive and no blocks are buffered.
1101 * This function is used by xfs_free_file_space() to zero
1102 * partial blocks when the range to free is not block aligned.
1103 * When unreserving space with boundaries that are not block
1104 * aligned we round up the start and round down the end
1105 * boundaries and then use this function to zero the parts of
1106 * the blocks that got dropped during the rounding.
1109 xfs_zero_remaining_bytes(
1114 xfs_bmbt_irec_t imap
;
1115 xfs_fileoff_t offset_fsb
;
1116 xfs_off_t lastoffset
;
1119 xfs_mount_t
*mp
= ip
->i_mount
;
1124 * Avoid doing I/O beyond eof - it's not necessary
1125 * since nothing can read beyond eof. The space will
1126 * be zeroed when the file is extended anyway.
1128 if (startoff
>= XFS_ISIZE(ip
))
1131 if (endoff
> XFS_ISIZE(ip
))
1132 endoff
= XFS_ISIZE(ip
);
1134 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
1137 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1140 lock_mode
= xfs_ilock_data_map_shared(ip
);
1141 error
= xfs_bmapi_read(ip
, offset_fsb
, 1, &imap
, &nimap
, 0);
1142 xfs_iunlock(ip
, lock_mode
);
1144 if (error
|| nimap
< 1)
1146 ASSERT(imap
.br_blockcount
>= 1);
1147 ASSERT(imap
.br_startoff
== offset_fsb
);
1148 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1150 if (imap
.br_startblock
== HOLESTARTBLOCK
||
1151 imap
.br_state
== XFS_EXT_UNWRITTEN
) {
1152 /* skip the entire extent */
1153 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+
1154 imap
.br_blockcount
) - 1;
1158 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
1159 if (lastoffset
> endoff
)
1160 lastoffset
= endoff
;
1162 /* DAX can just zero the backing device directly */
1163 if (IS_DAX(VFS_I(ip
))) {
1164 error
= dax_zero_page_range(VFS_I(ip
), offset
,
1165 lastoffset
- offset
+ 1,
1166 xfs_get_blocks_direct
);
1172 error
= xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip
) ?
1173 mp
->m_rtdev_targp
: mp
->m_ddev_targp
,
1174 xfs_fsb_to_db(ip
, imap
.br_startblock
),
1175 BTOBB(mp
->m_sb
.sb_blocksize
),
1181 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
1182 0, lastoffset
- offset
+ 1);
1184 error
= xfs_bwrite(bp
);
1193 xfs_free_file_space(
1194 struct xfs_inode
*ip
,
1199 xfs_fileoff_t endoffset_fsb
;
1201 xfs_fsblock_t firstfsb
;
1202 xfs_bmap_free_t free_list
;
1203 xfs_bmbt_irec_t imap
;
1205 xfs_off_t iendoffset
;
1212 xfs_fileoff_t startoffset_fsb
;
1217 trace_xfs_free_file_space(ip
);
1219 error
= xfs_qm_dqattach(ip
, 0);
1224 if (len
<= 0) /* if nothing being freed */
1226 rt
= XFS_IS_REALTIME_INODE(ip
);
1227 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
1228 endoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
+ len
);
1230 /* wait for the completion of any pending DIOs */
1231 inode_dio_wait(VFS_I(ip
));
1233 rounding
= max_t(xfs_off_t
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_SIZE
);
1234 ioffset
= round_down(offset
, rounding
);
1235 iendoffset
= round_up(offset
+ len
, rounding
) - 1;
1236 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
, ioffset
,
1240 truncate_pagecache_range(VFS_I(ip
), ioffset
, iendoffset
);
1243 * Need to zero the stuff we're not freeing, on disk.
1244 * If it's a realtime file & can't use unwritten extents then we
1245 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1246 * will take care of it for us.
1248 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
1250 error
= xfs_bmapi_read(ip
, startoffset_fsb
, 1,
1254 ASSERT(nimap
== 0 || nimap
== 1);
1255 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1258 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1259 block
= imap
.br_startblock
;
1260 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
1262 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
1265 error
= xfs_bmapi_read(ip
, endoffset_fsb
- 1, 1,
1269 ASSERT(nimap
== 0 || nimap
== 1);
1270 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1271 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1273 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
1274 endoffset_fsb
-= mod
;
1277 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
1279 * One contiguous piece to clear
1281 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
1284 * Some full blocks, possibly two pieces to clear
1286 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
1287 error
= xfs_zero_remaining_bytes(ip
, offset
,
1288 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
1290 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
1291 error
= xfs_zero_remaining_bytes(ip
,
1292 XFS_FSB_TO_B(mp
, endoffset_fsb
),
1297 * free file space until done or until there is an error
1299 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1300 while (!error
&& !done
) {
1303 * allocate and setup the transaction. Allow this
1304 * transaction to dip into the reserve blocks to ensure
1305 * the freeing of the space succeeds at ENOSPC.
1307 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1308 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
, resblks
, 0);
1311 * check for running out of space
1315 * Free the transaction structure.
1317 ASSERT(error
== -ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1318 xfs_trans_cancel(tp
);
1321 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1322 error
= xfs_trans_reserve_quota(tp
, mp
,
1323 ip
->i_udquot
, ip
->i_gdquot
, ip
->i_pdquot
,
1324 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
1328 xfs_trans_ijoin(tp
, ip
, 0);
1331 * issue the bunmapi() call to free the blocks
1333 xfs_bmap_init(&free_list
, &firstfsb
);
1334 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
1335 endoffset_fsb
- startoffset_fsb
,
1336 0, 2, &firstfsb
, &free_list
, &done
);
1341 * complete the transaction
1343 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1347 error
= xfs_trans_commit(tp
);
1348 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1355 xfs_bmap_cancel(&free_list
);
1357 xfs_trans_cancel(tp
);
1358 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1363 * Preallocate and zero a range of a file. This mechanism has the allocation
1364 * semantics of fallocate and in addition converts data in the range to zeroes.
1367 xfs_zero_file_space(
1368 struct xfs_inode
*ip
,
1372 struct xfs_mount
*mp
= ip
->i_mount
;
1376 trace_xfs_zero_file_space(ip
);
1378 blksize
= 1 << mp
->m_sb
.sb_blocklog
;
1381 * Punch a hole and prealloc the range. We use hole punch rather than
1382 * unwritten extent conversion for two reasons:
1384 * 1.) Hole punch handles partial block zeroing for us.
1386 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1387 * by virtue of the hole punch.
1389 error
= xfs_free_file_space(ip
, offset
, len
);
1393 error
= xfs_alloc_file_space(ip
, round_down(offset
, blksize
),
1394 round_up(offset
+ len
, blksize
) -
1395 round_down(offset
, blksize
),
1396 XFS_BMAPI_PREALLOC
);
1403 * @next_fsb will keep track of the extent currently undergoing shift.
1404 * @stop_fsb will keep track of the extent at which we have to stop.
1405 * If we are shifting left, we will start with block (offset + len) and
1406 * shift each extent till last extent.
1407 * If we are shifting right, we will start with last extent inside file space
1408 * and continue until we reach the block corresponding to offset.
1411 xfs_shift_file_space(
1412 struct xfs_inode
*ip
,
1415 enum shift_direction direction
)
1418 struct xfs_mount
*mp
= ip
->i_mount
;
1419 struct xfs_trans
*tp
;
1421 struct xfs_bmap_free free_list
;
1422 xfs_fsblock_t first_block
;
1423 xfs_fileoff_t stop_fsb
;
1424 xfs_fileoff_t next_fsb
;
1425 xfs_fileoff_t shift_fsb
;
1427 ASSERT(direction
== SHIFT_LEFT
|| direction
== SHIFT_RIGHT
);
1429 if (direction
== SHIFT_LEFT
) {
1430 next_fsb
= XFS_B_TO_FSB(mp
, offset
+ len
);
1431 stop_fsb
= XFS_B_TO_FSB(mp
, VFS_I(ip
)->i_size
);
1434 * If right shift, delegate the work of initialization of
1435 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1437 next_fsb
= NULLFSBLOCK
;
1438 stop_fsb
= XFS_B_TO_FSB(mp
, offset
);
1441 shift_fsb
= XFS_B_TO_FSB(mp
, len
);
1444 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1445 * into the accessible region of the file.
1447 if (xfs_can_free_eofblocks(ip
, true)) {
1448 error
= xfs_free_eofblocks(mp
, ip
, false);
1454 * Writeback and invalidate cache for the remainder of the file as we're
1455 * about to shift down every extent from offset to EOF.
1457 error
= filemap_write_and_wait_range(VFS_I(ip
)->i_mapping
,
1461 error
= invalidate_inode_pages2_range(VFS_I(ip
)->i_mapping
,
1462 offset
>> PAGE_SHIFT
, -1);
1467 * The extent shiting code works on extent granularity. So, if
1468 * stop_fsb is not the starting block of extent, we need to split
1469 * the extent at stop_fsb.
1471 if (direction
== SHIFT_RIGHT
) {
1472 error
= xfs_bmap_split_extent(ip
, stop_fsb
);
1477 while (!error
&& !done
) {
1478 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1480 * We would need to reserve permanent block for transaction.
1481 * This will come into picture when after shifting extent into
1482 * hole we found that adjacent extents can be merged which
1483 * may lead to freeing of a block during record update.
1485 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_write
,
1486 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0);
1488 xfs_trans_cancel(tp
);
1492 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1493 error
= xfs_trans_reserve_quota(tp
, mp
, ip
->i_udquot
,
1494 ip
->i_gdquot
, ip
->i_pdquot
,
1495 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0,
1496 XFS_QMOPT_RES_REGBLKS
);
1498 goto out_trans_cancel
;
1500 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1502 xfs_bmap_init(&free_list
, &first_block
);
1505 * We are using the write transaction in which max 2 bmbt
1506 * updates are allowed
1508 error
= xfs_bmap_shift_extents(tp
, ip
, &next_fsb
, shift_fsb
,
1509 &done
, stop_fsb
, &first_block
, &free_list
,
1510 direction
, XFS_BMAP_MAX_SHIFT_EXTENTS
);
1512 goto out_bmap_cancel
;
1514 error
= xfs_bmap_finish(&tp
, &free_list
, NULL
);
1516 goto out_bmap_cancel
;
1518 error
= xfs_trans_commit(tp
);
1524 xfs_bmap_cancel(&free_list
);
1526 xfs_trans_cancel(tp
);
1531 * xfs_collapse_file_space()
1532 * This routine frees disk space and shift extent for the given file.
1533 * The first thing we do is to free data blocks in the specified range
1534 * by calling xfs_free_file_space(). It would also sync dirty data
1535 * and invalidate page cache over the region on which collapse range
1536 * is working. And Shift extent records to the left to cover a hole.
1543 xfs_collapse_file_space(
1544 struct xfs_inode
*ip
,
1550 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1551 trace_xfs_collapse_file_space(ip
);
1553 error
= xfs_free_file_space(ip
, offset
, len
);
1557 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_LEFT
);
1561 * xfs_insert_file_space()
1562 * This routine create hole space by shifting extents for the given file.
1563 * The first thing we do is to sync dirty data and invalidate page cache
1564 * over the region on which insert range is working. And split an extent
1565 * to two extents at given offset by calling xfs_bmap_split_extent.
1566 * And shift all extent records which are laying between [offset,
1567 * last allocated extent] to the right to reserve hole range.
1573 xfs_insert_file_space(
1574 struct xfs_inode
*ip
,
1578 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1579 trace_xfs_insert_file_space(ip
);
1581 return xfs_shift_file_space(ip
, offset
, len
, SHIFT_RIGHT
);
1585 * We need to check that the format of the data fork in the temporary inode is
1586 * valid for the target inode before doing the swap. This is not a problem with
1587 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1588 * data fork depending on the space the attribute fork is taking so we can get
1589 * invalid formats on the target inode.
1591 * E.g. target has space for 7 extents in extent format, temp inode only has
1592 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1593 * btree, but when swapped it needs to be in extent format. Hence we can't just
1594 * blindly swap data forks on attr2 filesystems.
1596 * Note that we check the swap in both directions so that we don't end up with
1597 * a corrupt temporary inode, either.
1599 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1600 * inode will prevent this situation from occurring, so all we do here is
1601 * reject and log the attempt. basically we are putting the responsibility on
1602 * userspace to get this right.
1605 xfs_swap_extents_check_format(
1606 xfs_inode_t
*ip
, /* target inode */
1607 xfs_inode_t
*tip
) /* tmp inode */
1610 /* Should never get a local format */
1611 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
||
1612 tip
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
1616 * if the target inode has less extents that then temporary inode then
1617 * why did userspace call us?
1619 if (ip
->i_d
.di_nextents
< tip
->i_d
.di_nextents
)
1623 * if the target inode is in extent form and the temp inode is in btree
1624 * form then we will end up with the target inode in the wrong format
1625 * as we already know there are less extents in the temp inode.
1627 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1628 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
)
1631 /* Check temp in extent form to max in target */
1632 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1633 XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) >
1634 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1637 /* Check target in extent form to max in temp */
1638 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1639 XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) >
1640 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1644 * If we are in a btree format, check that the temp root block will fit
1645 * in the target and that it has enough extents to be in btree format
1648 * Note that we have to be careful to allow btree->extent conversions
1649 * (a common defrag case) which will occur when the temp inode is in
1652 if (tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1653 if (XFS_IFORK_BOFF(ip
) &&
1654 XFS_BMAP_BMDR_SPACE(tip
->i_df
.if_broot
) > XFS_IFORK_BOFF(ip
))
1656 if (XFS_IFORK_NEXTENTS(tip
, XFS_DATA_FORK
) <=
1657 XFS_IFORK_MAXEXT(ip
, XFS_DATA_FORK
))
1661 /* Reciprocal target->temp btree format checks */
1662 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1663 if (XFS_IFORK_BOFF(tip
) &&
1664 XFS_BMAP_BMDR_SPACE(ip
->i_df
.if_broot
) > XFS_IFORK_BOFF(tip
))
1666 if (XFS_IFORK_NEXTENTS(ip
, XFS_DATA_FORK
) <=
1667 XFS_IFORK_MAXEXT(tip
, XFS_DATA_FORK
))
1675 xfs_swap_extent_flush(
1676 struct xfs_inode
*ip
)
1680 error
= filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
1683 truncate_pagecache_range(VFS_I(ip
), 0, -1);
1685 /* Verify O_DIRECT for ftmp */
1686 if (VFS_I(ip
)->i_mapping
->nrpages
)
1693 xfs_inode_t
*ip
, /* target inode */
1694 xfs_inode_t
*tip
, /* tmp inode */
1697 xfs_mount_t
*mp
= ip
->i_mount
;
1699 xfs_bstat_t
*sbp
= &sxp
->sx_stat
;
1700 xfs_ifork_t
*tempifp
, *ifp
, *tifp
;
1701 int src_log_flags
, target_log_flags
;
1708 tempifp
= kmem_alloc(sizeof(xfs_ifork_t
), KM_MAYFAIL
);
1715 * Lock the inodes against other IO, page faults and truncate to
1716 * begin with. Then we can ensure the inodes are flushed and have no
1717 * page cache safely. Once we have done this we can take the ilocks and
1718 * do the rest of the checks.
1720 lock_flags
= XFS_IOLOCK_EXCL
| XFS_MMAPLOCK_EXCL
;
1721 xfs_lock_two_inodes(ip
, tip
, XFS_IOLOCK_EXCL
);
1722 xfs_lock_two_inodes(ip
, tip
, XFS_MMAPLOCK_EXCL
);
1724 /* Verify that both files have the same format */
1725 if ((VFS_I(ip
)->i_mode
& S_IFMT
) != (VFS_I(tip
)->i_mode
& S_IFMT
)) {
1730 /* Verify both files are either real-time or non-realtime */
1731 if (XFS_IS_REALTIME_INODE(ip
) != XFS_IS_REALTIME_INODE(tip
)) {
1736 error
= xfs_swap_extent_flush(ip
);
1739 error
= xfs_swap_extent_flush(tip
);
1743 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SWAPEXT
);
1744 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 0, 0);
1746 xfs_trans_cancel(tp
);
1751 * Lock and join the inodes to the tansaction so that transaction commit
1752 * or cancel will unlock the inodes from this point onwards.
1754 xfs_lock_two_inodes(ip
, tip
, XFS_ILOCK_EXCL
);
1755 lock_flags
|= XFS_ILOCK_EXCL
;
1756 xfs_trans_ijoin(tp
, ip
, lock_flags
);
1757 xfs_trans_ijoin(tp
, tip
, lock_flags
);
1760 /* Verify all data are being swapped */
1761 if (sxp
->sx_offset
!= 0 ||
1762 sxp
->sx_length
!= ip
->i_d
.di_size
||
1763 sxp
->sx_length
!= tip
->i_d
.di_size
) {
1765 goto out_trans_cancel
;
1768 trace_xfs_swap_extent_before(ip
, 0);
1769 trace_xfs_swap_extent_before(tip
, 1);
1771 /* check inode formats now that data is flushed */
1772 error
= xfs_swap_extents_check_format(ip
, tip
);
1775 "%s: inode 0x%llx format is incompatible for exchanging.",
1776 __func__
, ip
->i_ino
);
1777 goto out_trans_cancel
;
1781 * Compare the current change & modify times with that
1782 * passed in. If they differ, we abort this swap.
1783 * This is the mechanism used to ensure the calling
1784 * process that the file was not changed out from
1787 if ((sbp
->bs_ctime
.tv_sec
!= VFS_I(ip
)->i_ctime
.tv_sec
) ||
1788 (sbp
->bs_ctime
.tv_nsec
!= VFS_I(ip
)->i_ctime
.tv_nsec
) ||
1789 (sbp
->bs_mtime
.tv_sec
!= VFS_I(ip
)->i_mtime
.tv_sec
) ||
1790 (sbp
->bs_mtime
.tv_nsec
!= VFS_I(ip
)->i_mtime
.tv_nsec
)) {
1792 goto out_trans_cancel
;
1795 * Count the number of extended attribute blocks
1797 if ( ((XFS_IFORK_Q(ip
) != 0) && (ip
->i_d
.di_anextents
> 0)) &&
1798 (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1799 error
= xfs_bmap_count_blocks(tp
, ip
, XFS_ATTR_FORK
, &aforkblks
);
1801 goto out_trans_cancel
;
1803 if ( ((XFS_IFORK_Q(tip
) != 0) && (tip
->i_d
.di_anextents
> 0)) &&
1804 (tip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)) {
1805 error
= xfs_bmap_count_blocks(tp
, tip
, XFS_ATTR_FORK
,
1808 goto out_trans_cancel
;
1812 * Before we've swapped the forks, lets set the owners of the forks
1813 * appropriately. We have to do this as we are demand paging the btree
1814 * buffers, and so the validation done on read will expect the owner
1815 * field to be correctly set. Once we change the owners, we can swap the
1818 * Note the trickiness in setting the log flags - we set the owner log
1819 * flag on the opposite inode (i.e. the inode we are setting the new
1820 * owner to be) because once we swap the forks and log that, log
1821 * recovery is going to see the fork as owned by the swapped inode,
1822 * not the pre-swapped inodes.
1824 src_log_flags
= XFS_ILOG_CORE
;
1825 target_log_flags
= XFS_ILOG_CORE
;
1826 if (ip
->i_d
.di_version
== 3 &&
1827 ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1828 target_log_flags
|= XFS_ILOG_DOWNER
;
1829 error
= xfs_bmbt_change_owner(tp
, ip
, XFS_DATA_FORK
,
1832 goto out_trans_cancel
;
1835 if (tip
->i_d
.di_version
== 3 &&
1836 tip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) {
1837 src_log_flags
|= XFS_ILOG_DOWNER
;
1838 error
= xfs_bmbt_change_owner(tp
, tip
, XFS_DATA_FORK
,
1841 goto out_trans_cancel
;
1845 * Swap the data forks of the inodes
1849 *tempifp
= *ifp
; /* struct copy */
1850 *ifp
= *tifp
; /* struct copy */
1851 *tifp
= *tempifp
; /* struct copy */
1854 * Fix the on-disk inode values
1856 tmp
= (__uint64_t
)ip
->i_d
.di_nblocks
;
1857 ip
->i_d
.di_nblocks
= tip
->i_d
.di_nblocks
- taforkblks
+ aforkblks
;
1858 tip
->i_d
.di_nblocks
= tmp
+ taforkblks
- aforkblks
;
1860 tmp
= (__uint64_t
) ip
->i_d
.di_nextents
;
1861 ip
->i_d
.di_nextents
= tip
->i_d
.di_nextents
;
1862 tip
->i_d
.di_nextents
= tmp
;
1864 tmp
= (__uint64_t
) ip
->i_d
.di_format
;
1865 ip
->i_d
.di_format
= tip
->i_d
.di_format
;
1866 tip
->i_d
.di_format
= tmp
;
1869 * The extents in the source inode could still contain speculative
1870 * preallocation beyond EOF (e.g. the file is open but not modified
1871 * while defrag is in progress). In that case, we need to copy over the
1872 * number of delalloc blocks the data fork in the source inode is
1873 * tracking beyond EOF so that when the fork is truncated away when the
1874 * temporary inode is unlinked we don't underrun the i_delayed_blks
1875 * counter on that inode.
1877 ASSERT(tip
->i_delayed_blks
== 0);
1878 tip
->i_delayed_blks
= ip
->i_delayed_blks
;
1879 ip
->i_delayed_blks
= 0;
1881 switch (ip
->i_d
.di_format
) {
1882 case XFS_DINODE_FMT_EXTENTS
:
1883 /* If the extents fit in the inode, fix the
1884 * pointer. Otherwise it's already NULL or
1885 * pointing to the extent.
1887 if (ip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1888 ifp
->if_u1
.if_extents
=
1889 ifp
->if_u2
.if_inline_ext
;
1891 src_log_flags
|= XFS_ILOG_DEXT
;
1893 case XFS_DINODE_FMT_BTREE
:
1894 ASSERT(ip
->i_d
.di_version
< 3 ||
1895 (src_log_flags
& XFS_ILOG_DOWNER
));
1896 src_log_flags
|= XFS_ILOG_DBROOT
;
1900 switch (tip
->i_d
.di_format
) {
1901 case XFS_DINODE_FMT_EXTENTS
:
1902 /* If the extents fit in the inode, fix the
1903 * pointer. Otherwise it's already NULL or
1904 * pointing to the extent.
1906 if (tip
->i_d
.di_nextents
<= XFS_INLINE_EXTS
) {
1907 tifp
->if_u1
.if_extents
=
1908 tifp
->if_u2
.if_inline_ext
;
1910 target_log_flags
|= XFS_ILOG_DEXT
;
1912 case XFS_DINODE_FMT_BTREE
:
1913 target_log_flags
|= XFS_ILOG_DBROOT
;
1914 ASSERT(tip
->i_d
.di_version
< 3 ||
1915 (target_log_flags
& XFS_ILOG_DOWNER
));
1919 xfs_trans_log_inode(tp
, ip
, src_log_flags
);
1920 xfs_trans_log_inode(tp
, tip
, target_log_flags
);
1923 * If this is a synchronous mount, make sure that the
1924 * transaction goes to disk before returning to the user.
1926 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
1927 xfs_trans_set_sync(tp
);
1929 error
= xfs_trans_commit(tp
);
1931 trace_xfs_swap_extent_after(ip
, 0);
1932 trace_xfs_swap_extent_after(tip
, 1);
1938 xfs_iunlock(ip
, lock_flags
);
1939 xfs_iunlock(tip
, lock_flags
);
1943 xfs_trans_cancel(tp
);