2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "libxfs_priv.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_alloc.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_trans_space.h"
38 #include "xfs_trace.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_quota_defs.h"
42 #include "xfs_ag_resv.h"
43 #include "xfs_refcount.h"
44 #include "xfs_rmap_btree.h"
47 kmem_zone_t
*xfs_bmap_free_item_zone
;
50 * Miscellaneous helper functions
54 * Compute and fill in the value of the maximum depth of a bmap btree
55 * in this filesystem. Done once, during mount.
58 xfs_bmap_compute_maxlevels(
59 xfs_mount_t
*mp
, /* file system mount structure */
60 int whichfork
) /* data or attr fork */
62 int level
; /* btree level */
63 uint maxblocks
; /* max blocks at this level */
64 uint maxleafents
; /* max leaf entries possible */
65 int maxrootrecs
; /* max records in root block */
66 int minleafrecs
; /* min records in leaf block */
67 int minnoderecs
; /* min records in node block */
68 int sz
; /* root block size */
71 * The maximum number of extents in a file, hence the maximum
72 * number of leaf entries, is controlled by the type of di_nextents
73 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
74 * (a signed 16-bit number, xfs_aextnum_t).
76 * Note that we can no longer assume that if we are in ATTR1 that
77 * the fork offset of all the inodes will be
78 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
79 * with ATTR2 and then mounted back with ATTR1, keeping the
80 * di_forkoff's fixed but probably at various positions. Therefore,
81 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
82 * of a minimum size available.
84 if (whichfork
== XFS_DATA_FORK
) {
85 maxleafents
= MAXEXTNUM
;
86 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
88 maxleafents
= MAXAEXTNUM
;
89 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
91 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
92 minleafrecs
= mp
->m_bmap_dmnr
[0];
93 minnoderecs
= mp
->m_bmap_dmnr
[1];
94 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
95 for (level
= 1; maxblocks
> 1; level
++) {
96 if (maxblocks
<= maxrootrecs
)
99 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
101 mp
->m_bm_maxlevels
[whichfork
] = level
;
104 STATIC
int /* error */
106 struct xfs_btree_cur
*cur
,
110 int *stat
) /* success/failure */
112 cur
->bc_rec
.b
.br_startoff
= off
;
113 cur
->bc_rec
.b
.br_startblock
= bno
;
114 cur
->bc_rec
.b
.br_blockcount
= len
;
115 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
118 STATIC
int /* error */
120 struct xfs_btree_cur
*cur
,
124 int *stat
) /* success/failure */
126 cur
->bc_rec
.b
.br_startoff
= off
;
127 cur
->bc_rec
.b
.br_startblock
= bno
;
128 cur
->bc_rec
.b
.br_blockcount
= len
;
129 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
133 * Check if the inode needs to be converted to btree format.
135 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
137 return whichfork
!= XFS_COW_FORK
&&
138 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
139 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
140 XFS_IFORK_MAXEXT(ip
, whichfork
);
144 * Check if the inode should be converted to extent format.
146 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
148 return whichfork
!= XFS_COW_FORK
&&
149 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
150 XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
151 XFS_IFORK_MAXEXT(ip
, whichfork
);
155 * Update the record referred to by cur to the value given
156 * by [off, bno, len, state].
157 * This either works (return 0) or gets an EFSCORRUPTED error.
161 struct xfs_btree_cur
*cur
,
167 union xfs_btree_rec rec
;
169 xfs_bmbt_disk_set_allf(&rec
.bmbt
, off
, bno
, len
, state
);
170 return xfs_btree_update(cur
, &rec
);
174 * Compute the worst-case number of indirect blocks that will be used
175 * for ip's delayed extent of length "len".
178 xfs_bmap_worst_indlen(
179 xfs_inode_t
*ip
, /* incore inode pointer */
180 xfs_filblks_t len
) /* delayed extent length */
182 int level
; /* btree level number */
183 int maxrecs
; /* maximum record count at this level */
184 xfs_mount_t
*mp
; /* mount structure */
185 xfs_filblks_t rval
; /* return value */
186 xfs_filblks_t orig_len
;
190 /* Calculate the worst-case size of the bmbt. */
192 maxrecs
= mp
->m_bmap_dmxr
[0];
193 for (level
= 0, rval
= 0;
194 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
197 do_div(len
, maxrecs
);
200 rval
+= XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
205 maxrecs
= mp
->m_bmap_dmxr
[1];
208 /* Calculate the worst-case size of the rmapbt. */
209 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
210 rval
+= 1 + xfs_rmapbt_calc_size(mp
, orig_len
) +
211 mp
->m_rmap_maxlevels
;
217 * Calculate the default attribute fork offset for newly created inodes.
220 xfs_default_attroffset(
221 struct xfs_inode
*ip
)
223 struct xfs_mount
*mp
= ip
->i_mount
;
226 if (mp
->m_sb
.sb_inodesize
== 256) {
227 offset
= XFS_LITINO(mp
, ip
->i_d
.di_version
) -
228 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
230 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
233 ASSERT(offset
< XFS_LITINO(mp
, ip
->i_d
.di_version
));
238 * Helper routine to reset inode di_forkoff field when switching
239 * attribute fork from local to extent format - we reset it where
240 * possible to make space available for inline data fork extents.
243 xfs_bmap_forkoff_reset(
247 if (whichfork
== XFS_ATTR_FORK
&&
248 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
249 ip
->i_d
.di_format
!= XFS_DINODE_FMT_UUID
&&
250 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
251 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
253 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
254 ip
->i_d
.di_forkoff
= dfl_forkoff
;
259 STATIC
struct xfs_buf
*
261 struct xfs_btree_cur
*cur
,
264 struct xfs_log_item_desc
*lidp
;
270 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
271 if (!cur
->bc_bufs
[i
])
273 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
274 return cur
->bc_bufs
[i
];
277 /* Chase down all the log items to see if the bp is there */
278 list_for_each_entry(lidp
, &cur
->bc_tp
->t_items
, lid_trans
) {
279 struct xfs_buf_log_item
*bip
;
280 bip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
281 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
282 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
291 struct xfs_btree_block
*block
,
297 __be64
*pp
, *thispa
; /* pointer to block address */
298 xfs_bmbt_key_t
*prevp
, *keyp
;
300 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
303 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
304 dmxr
= mp
->m_bmap_dmxr
[0];
305 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
308 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
309 be64_to_cpu(keyp
->br_startoff
));
314 * Compare the block numbers to see if there are dups.
317 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
319 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
321 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
323 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
325 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
326 if (*thispa
== *pp
) {
327 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
329 (unsigned long long)be64_to_cpu(*thispa
));
330 panic("%s: ptrs are equal in node\n",
338 * Check that the extents for the inode ip are in the right order in all
339 * btree leaves. THis becomes prohibitively expensive for large extent count
340 * files, so don't bother with inodes that have more than 10,000 extents in
341 * them. The btree record ordering checks will still be done, so for such large
342 * bmapbt constructs that is going to catch most corruptions.
345 xfs_bmap_check_leaf_extents(
346 xfs_btree_cur_t
*cur
, /* btree cursor or null */
347 xfs_inode_t
*ip
, /* incore inode pointer */
348 int whichfork
) /* data or attr fork */
350 struct xfs_btree_block
*block
; /* current btree block */
351 xfs_fsblock_t bno
; /* block # of "block" */
352 xfs_buf_t
*bp
; /* buffer for "block" */
353 int error
; /* error return value */
354 xfs_extnum_t i
=0, j
; /* index into the extents list */
355 xfs_ifork_t
*ifp
; /* fork structure */
356 int level
; /* btree level, for checking */
357 xfs_mount_t
*mp
; /* file system mount structure */
358 __be64
*pp
; /* pointer to block address */
359 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
360 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
361 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
364 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
368 /* skip large extent count inodes */
369 if (ip
->i_d
.di_nextents
> 10000)
374 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
375 block
= ifp
->if_broot
;
377 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
379 level
= be16_to_cpu(block
->bb_level
);
381 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
382 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
383 bno
= be64_to_cpu(*pp
);
385 ASSERT(bno
!= NULLFSBLOCK
);
386 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
387 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
390 * Go down the tree until leaf level is reached, following the first
391 * pointer (leftmost) at each level.
393 while (level
-- > 0) {
394 /* See if buf is in cur first */
396 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
399 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
405 block
= XFS_BUF_TO_BLOCK(bp
);
410 * Check this block for basic sanity (increasing keys and
411 * no duplicate blocks).
414 xfs_check_block(block
, mp
, 0, 0);
415 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
416 bno
= be64_to_cpu(*pp
);
417 XFS_WANT_CORRUPTED_GOTO(mp
,
418 XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
421 xfs_trans_brelse(NULL
, bp
);
426 * Here with bp and block set to the leftmost leaf node in the tree.
431 * Loop over all leaf nodes checking that all extents are in the right order.
434 xfs_fsblock_t nextbno
;
435 xfs_extnum_t num_recs
;
438 num_recs
= xfs_btree_get_numrecs(block
);
441 * Read-ahead the next leaf block, if any.
444 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
447 * Check all the extents to make sure they are OK.
448 * If we had a previous block, the last entry should
449 * conform with the first entry in this one.
452 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
454 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
455 xfs_bmbt_disk_get_blockcount(&last
) <=
456 xfs_bmbt_disk_get_startoff(ep
));
458 for (j
= 1; j
< num_recs
; j
++) {
459 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
460 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
461 xfs_bmbt_disk_get_blockcount(ep
) <=
462 xfs_bmbt_disk_get_startoff(nextp
));
470 xfs_trans_brelse(NULL
, bp
);
474 * If we've reached the end, stop.
476 if (bno
== NULLFSBLOCK
)
480 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
483 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
489 block
= XFS_BUF_TO_BLOCK(bp
);
495 xfs_warn(mp
, "%s: at error0", __func__
);
497 xfs_trans_brelse(NULL
, bp
);
499 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
501 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__
);
506 * Add bmap trace insert entries for all the contents of the extent records.
509 xfs_bmap_trace_exlist(
510 xfs_inode_t
*ip
, /* incore inode pointer */
511 xfs_extnum_t cnt
, /* count of entries in the list */
512 int whichfork
, /* data or attr or cow fork */
513 unsigned long caller_ip
)
515 xfs_extnum_t idx
; /* extent record index */
516 xfs_ifork_t
*ifp
; /* inode fork pointer */
519 if (whichfork
== XFS_ATTR_FORK
)
520 state
|= BMAP_ATTRFORK
;
521 else if (whichfork
== XFS_COW_FORK
)
522 state
|= BMAP_COWFORK
;
524 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
525 ASSERT(cnt
== xfs_iext_count(ifp
));
526 for (idx
= 0; idx
< cnt
; idx
++)
527 trace_xfs_extlist(ip
, idx
, state
, caller_ip
);
531 * Validate that the bmbt_irecs being returned from bmapi are valid
532 * given the caller's original parameters. Specifically check the
533 * ranges of the returned irecs to ensure that they only extend beyond
534 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
537 xfs_bmap_validate_ret(
541 xfs_bmbt_irec_t
*mval
,
545 int i
; /* index to map values */
547 ASSERT(ret_nmap
<= nmap
);
549 for (i
= 0; i
< ret_nmap
; i
++) {
550 ASSERT(mval
[i
].br_blockcount
> 0);
551 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
552 ASSERT(mval
[i
].br_startoff
>= bno
);
553 ASSERT(mval
[i
].br_blockcount
<= len
);
554 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
557 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
558 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
562 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
563 mval
[i
].br_startoff
);
564 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
565 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
566 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
567 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
572 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
573 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
577 * bmap free list manipulation functions
581 * Add the extent to the list of extents to be free at transaction end.
582 * The list is maintained sorted (by block number).
586 struct xfs_mount
*mp
,
587 struct xfs_defer_ops
*dfops
,
590 struct xfs_owner_info
*oinfo
)
592 struct xfs_extent_free_item
*new; /* new element */
597 ASSERT(bno
!= NULLFSBLOCK
);
599 ASSERT(len
<= MAXEXTLEN
);
600 ASSERT(!isnullstartblock(bno
));
601 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
602 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
603 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
604 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
605 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
606 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
608 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
610 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
611 new->xefi_startblock
= bno
;
612 new->xefi_blockcount
= (xfs_extlen_t
)len
;
614 new->xefi_oinfo
= *oinfo
;
616 xfs_rmap_skip_owner_update(&new->xefi_oinfo
);
617 trace_xfs_bmap_free_defer(mp
, XFS_FSB_TO_AGNO(mp
, bno
), 0,
618 XFS_FSB_TO_AGBNO(mp
, bno
), len
);
619 xfs_defer_add(dfops
, XFS_DEFER_OPS_TYPE_FREE
, &new->xefi_list
);
623 * Inode fork format manipulation functions
627 * Transform a btree format file with only one leaf node, where the
628 * extents list will fit in the inode, into an extents format file.
629 * Since the file extents are already in-core, all we have to do is
630 * give up the space for the btree root and pitch the leaf block.
632 STATIC
int /* error */
633 xfs_bmap_btree_to_extents(
634 xfs_trans_t
*tp
, /* transaction pointer */
635 xfs_inode_t
*ip
, /* incore inode pointer */
636 xfs_btree_cur_t
*cur
, /* btree cursor */
637 int *logflagsp
, /* inode logging flags */
638 int whichfork
) /* data or attr fork */
641 struct xfs_btree_block
*cblock
;/* child btree block */
642 xfs_fsblock_t cbno
; /* child block number */
643 xfs_buf_t
*cbp
; /* child block's buffer */
644 int error
; /* error return value */
645 xfs_ifork_t
*ifp
; /* inode fork data */
646 xfs_mount_t
*mp
; /* mount point structure */
647 __be64
*pp
; /* ptr to block address */
648 struct xfs_btree_block
*rblock
;/* root btree block */
649 struct xfs_owner_info oinfo
;
652 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
653 ASSERT(whichfork
!= XFS_COW_FORK
);
654 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
655 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
656 rblock
= ifp
->if_broot
;
657 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
658 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
659 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
660 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
661 cbno
= be64_to_cpu(*pp
);
664 if ((error
= xfs_btree_check_lptr(cur
, cbno
, 1)))
667 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
, XFS_BMAP_BTREE_REF
,
671 cblock
= XFS_BUF_TO_BLOCK(cbp
);
672 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
674 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, whichfork
);
675 xfs_bmap_add_free(mp
, cur
->bc_private
.b
.dfops
, cbno
, 1, &oinfo
);
676 ip
->i_d
.di_nblocks
--;
677 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
678 xfs_trans_binval(tp
, cbp
);
679 if (cur
->bc_bufs
[0] == cbp
)
680 cur
->bc_bufs
[0] = NULL
;
681 xfs_iroot_realloc(ip
, -1, whichfork
);
682 ASSERT(ifp
->if_broot
== NULL
);
683 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
684 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
685 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
690 * Convert an extents-format file into a btree-format file.
691 * The new file will have a root block (in the inode) and a single child block.
693 STATIC
int /* error */
694 xfs_bmap_extents_to_btree(
695 xfs_trans_t
*tp
, /* transaction pointer */
696 xfs_inode_t
*ip
, /* incore inode pointer */
697 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
698 struct xfs_defer_ops
*dfops
, /* blocks freed in xaction */
699 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
700 int wasdel
, /* converting a delayed alloc */
701 int *logflagsp
, /* inode logging flags */
702 int whichfork
) /* data or attr fork */
704 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
705 xfs_buf_t
*abp
; /* buffer for ablock */
706 xfs_alloc_arg_t args
; /* allocation arguments */
707 xfs_bmbt_rec_t
*arp
; /* child record pointer */
708 struct xfs_btree_block
*block
; /* btree root block */
709 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
710 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
711 int error
; /* error return value */
712 xfs_extnum_t i
, cnt
; /* extent record index */
713 xfs_ifork_t
*ifp
; /* inode fork pointer */
714 xfs_bmbt_key_t
*kp
; /* root block key pointer */
715 xfs_mount_t
*mp
; /* mount structure */
716 xfs_extnum_t nextents
; /* number of file extents */
717 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
720 ASSERT(whichfork
!= XFS_COW_FORK
);
721 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
722 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
725 * Make space in the inode incore.
727 xfs_iroot_realloc(ip
, 1, whichfork
);
728 ifp
->if_flags
|= XFS_IFBROOT
;
733 block
= ifp
->if_broot
;
734 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
735 XFS_BTNUM_BMAP
, 1, 1, ip
->i_ino
,
736 XFS_BTREE_LONG_PTRS
);
738 * Need a cursor. Can't allocate until bb_level is filled in.
740 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
741 cur
->bc_private
.b
.firstblock
= *firstblock
;
742 cur
->bc_private
.b
.dfops
= dfops
;
743 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
745 * Convert to a btree with two levels, one record in root.
747 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
748 memset(&args
, 0, sizeof(args
));
751 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, ip
->i_ino
, whichfork
);
752 args
.firstblock
= *firstblock
;
753 if (*firstblock
== NULLFSBLOCK
) {
754 args
.type
= XFS_ALLOCTYPE_START_BNO
;
755 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
756 } else if (dfops
->dop_low
) {
757 args
.type
= XFS_ALLOCTYPE_START_BNO
;
759 args
.fsbno
= *firstblock
;
761 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
762 args
.fsbno
= *firstblock
;
764 args
.minlen
= args
.maxlen
= args
.prod
= 1;
765 args
.wasdel
= wasdel
;
767 if ((error
= xfs_alloc_vextent(&args
))) {
768 xfs_iroot_realloc(ip
, -1, whichfork
);
769 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
774 * During a CoW operation, the allocation and bmbt updates occur in
775 * different transactions. The mapping code tries to put new bmbt
776 * blocks near extents being mapped, but the only way to guarantee this
777 * is if the alloc and the mapping happen in a single transaction that
778 * has a block reservation. That isn't the case here, so if we run out
779 * of space we'll try again with another AG.
781 if (xfs_sb_version_hasreflink(&cur
->bc_mp
->m_sb
) &&
782 args
.fsbno
== NULLFSBLOCK
&&
783 args
.type
== XFS_ALLOCTYPE_NEAR_BNO
) {
784 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
787 if (WARN_ON_ONCE(args
.fsbno
== NULLFSBLOCK
)) {
788 xfs_iroot_realloc(ip
, -1, whichfork
);
789 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
793 * Allocation can't fail, the space was reserved.
795 ASSERT(*firstblock
== NULLFSBLOCK
||
796 args
.agno
>= XFS_FSB_TO_AGNO(mp
, *firstblock
));
797 *firstblock
= cur
->bc_private
.b
.firstblock
= args
.fsbno
;
798 cur
->bc_private
.b
.allocated
++;
799 ip
->i_d
.di_nblocks
++;
800 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
801 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
803 * Fill in the child block.
805 abp
->b_ops
= &xfs_bmbt_buf_ops
;
806 ablock
= XFS_BUF_TO_BLOCK(abp
);
807 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
808 XFS_BTNUM_BMAP
, 0, 0, ip
->i_ino
,
809 XFS_BTREE_LONG_PTRS
);
811 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
812 nextents
= xfs_iext_count(ifp
);
813 for (cnt
= i
= 0; i
< nextents
; i
++) {
814 ep
= xfs_iext_get_ext(ifp
, i
);
815 if (!isnullstartblock(xfs_bmbt_get_startblock(ep
))) {
816 arp
->l0
= cpu_to_be64(ep
->l0
);
817 arp
->l1
= cpu_to_be64(ep
->l1
);
821 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
822 xfs_btree_set_numrecs(ablock
, cnt
);
825 * Fill in the root key and pointer.
827 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
828 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
829 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
830 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
831 be16_to_cpu(block
->bb_level
)));
832 *pp
= cpu_to_be64(args
.fsbno
);
835 * Do all this logging at the end so that
836 * the root is at the right level.
838 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
839 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
840 ASSERT(*curp
== NULL
);
842 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
847 * Convert a local file to an extents file.
848 * This code is out of bounds for data forks of regular files,
849 * since the file data needs to get logged so things will stay consistent.
850 * (The bmap-level manipulations are ok, though).
853 xfs_bmap_local_to_extents_empty(
854 struct xfs_inode
*ip
,
857 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
859 ASSERT(whichfork
!= XFS_COW_FORK
);
860 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
861 ASSERT(ifp
->if_bytes
== 0);
862 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
864 xfs_bmap_forkoff_reset(ip
, whichfork
);
865 ifp
->if_flags
&= ~XFS_IFINLINE
;
866 ifp
->if_flags
|= XFS_IFEXTENTS
;
867 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
871 STATIC
int /* error */
872 xfs_bmap_local_to_extents(
873 xfs_trans_t
*tp
, /* transaction pointer */
874 xfs_inode_t
*ip
, /* incore inode pointer */
875 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
876 xfs_extlen_t total
, /* total blocks needed by transaction */
877 int *logflagsp
, /* inode logging flags */
879 void (*init_fn
)(struct xfs_trans
*tp
,
881 struct xfs_inode
*ip
,
882 struct xfs_ifork
*ifp
))
885 int flags
; /* logging flags returned */
886 xfs_ifork_t
*ifp
; /* inode fork pointer */
887 xfs_alloc_arg_t args
; /* allocation arguments */
888 xfs_buf_t
*bp
; /* buffer for extent block */
889 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
892 * We don't want to deal with the case of keeping inode data inline yet.
893 * So sending the data fork of a regular inode is invalid.
895 ASSERT(!(S_ISREG(VFS_I(ip
)->i_mode
) && whichfork
== XFS_DATA_FORK
));
896 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
897 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
899 if (!ifp
->if_bytes
) {
900 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
901 flags
= XFS_ILOG_CORE
;
907 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
|XFS_IFEXTIREC
)) ==
909 memset(&args
, 0, sizeof(args
));
911 args
.mp
= ip
->i_mount
;
912 xfs_rmap_ino_owner(&args
.oinfo
, ip
->i_ino
, whichfork
, 0);
913 args
.firstblock
= *firstblock
;
915 * Allocate a block. We know we need only one, since the
916 * file currently fits in an inode.
918 if (*firstblock
== NULLFSBLOCK
) {
920 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
921 args
.type
= XFS_ALLOCTYPE_START_BNO
;
923 args
.fsbno
= *firstblock
;
924 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
927 args
.minlen
= args
.maxlen
= args
.prod
= 1;
928 error
= xfs_alloc_vextent(&args
);
933 * During a CoW operation, the allocation and bmbt updates occur in
934 * different transactions. The mapping code tries to put new bmbt
935 * blocks near extents being mapped, but the only way to guarantee this
936 * is if the alloc and the mapping happen in a single transaction that
937 * has a block reservation. That isn't the case here, so if we run out
938 * of space we'll try again with another AG.
940 if (xfs_sb_version_hasreflink(&ip
->i_mount
->m_sb
) &&
941 args
.fsbno
== NULLFSBLOCK
&&
942 args
.type
== XFS_ALLOCTYPE_NEAR_BNO
) {
945 /* Can't fail, the space was reserved. */
946 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
947 ASSERT(args
.len
== 1);
948 *firstblock
= args
.fsbno
;
949 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
952 * Initialize the block, copy the data and log the remote buffer.
954 * The callout is responsible for logging because the remote format
955 * might differ from the local format and thus we don't know how much to
956 * log here. Note that init_fn must also set the buffer log item type
959 init_fn(tp
, bp
, ip
, ifp
);
961 /* account for the change in fork size */
962 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
963 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
964 flags
|= XFS_ILOG_CORE
;
966 xfs_iext_add(ifp
, 0, 1);
967 ep
= xfs_iext_get_ext(ifp
, 0);
968 xfs_bmbt_set_allf(ep
, 0, args
.fsbno
, 1, XFS_EXT_NORM
);
969 trace_xfs_bmap_post_update(ip
, 0,
970 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0,
972 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
973 ip
->i_d
.di_nblocks
= 1;
974 xfs_trans_mod_dquot_byino(tp
, ip
,
975 XFS_TRANS_DQ_BCOUNT
, 1L);
976 flags
|= xfs_ilog_fext(whichfork
);
984 * Called from xfs_bmap_add_attrfork to handle btree format files.
986 STATIC
int /* error */
987 xfs_bmap_add_attrfork_btree(
988 xfs_trans_t
*tp
, /* transaction pointer */
989 xfs_inode_t
*ip
, /* incore inode pointer */
990 xfs_fsblock_t
*firstblock
, /* first block allocated */
991 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
992 int *flags
) /* inode logging flags */
994 xfs_btree_cur_t
*cur
; /* btree cursor */
995 int error
; /* error return value */
996 xfs_mount_t
*mp
; /* file system mount struct */
997 int stat
; /* newroot status */
1000 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
1001 *flags
|= XFS_ILOG_DBROOT
;
1003 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
1004 cur
->bc_private
.b
.dfops
= dfops
;
1005 cur
->bc_private
.b
.firstblock
= *firstblock
;
1006 if ((error
= xfs_bmbt_lookup_ge(cur
, 0, 0, 0, &stat
)))
1008 /* must be at least one entry */
1009 XFS_WANT_CORRUPTED_GOTO(mp
, stat
== 1, error0
);
1010 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
1013 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1016 *firstblock
= cur
->bc_private
.b
.firstblock
;
1017 cur
->bc_private
.b
.allocated
= 0;
1018 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1022 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1027 * Called from xfs_bmap_add_attrfork to handle extents format files.
1029 STATIC
int /* error */
1030 xfs_bmap_add_attrfork_extents(
1031 xfs_trans_t
*tp
, /* transaction pointer */
1032 xfs_inode_t
*ip
, /* incore inode pointer */
1033 xfs_fsblock_t
*firstblock
, /* first block allocated */
1034 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
1035 int *flags
) /* inode logging flags */
1037 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
1038 int error
; /* error return value */
1040 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
1043 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, dfops
, &cur
, 0,
1044 flags
, XFS_DATA_FORK
);
1046 cur
->bc_private
.b
.allocated
= 0;
1047 xfs_btree_del_cursor(cur
,
1048 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
1054 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1055 * different data fork content type needs a different callout to do the
1056 * conversion. Some are basic and only require special block initialisation
1057 * callouts for the data formating, others (directories) are so specialised they
1058 * handle everything themselves.
1060 * XXX (dgc): investigate whether directory conversion can use the generic
1061 * formatting callout. It should be possible - it's just a very complex
1064 STATIC
int /* error */
1065 xfs_bmap_add_attrfork_local(
1066 xfs_trans_t
*tp
, /* transaction pointer */
1067 xfs_inode_t
*ip
, /* incore inode pointer */
1068 xfs_fsblock_t
*firstblock
, /* first block allocated */
1069 struct xfs_defer_ops
*dfops
, /* blocks to free at commit */
1070 int *flags
) /* inode logging flags */
1072 xfs_da_args_t dargs
; /* args for dir/attr code */
1074 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1077 if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
1078 memset(&dargs
, 0, sizeof(dargs
));
1079 dargs
.geo
= ip
->i_mount
->m_dir_geo
;
1081 dargs
.firstblock
= firstblock
;
1082 dargs
.dfops
= dfops
;
1083 dargs
.total
= dargs
.geo
->fsbcount
;
1084 dargs
.whichfork
= XFS_DATA_FORK
;
1086 return xfs_dir2_sf_to_block(&dargs
);
1089 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1090 return xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1,
1091 flags
, XFS_DATA_FORK
,
1092 xfs_symlink_local_to_remote
);
1094 /* should only be called for types that support local format data */
1096 return -EFSCORRUPTED
;
1100 * Convert inode from non-attributed to attributed.
1101 * Must not be in a transaction, ip must not be locked.
1103 int /* error code */
1104 xfs_bmap_add_attrfork(
1105 xfs_inode_t
*ip
, /* incore inode pointer */
1106 int size
, /* space new attribute needs */
1107 int rsvd
) /* xact may use reserved blks */
1109 xfs_fsblock_t firstblock
; /* 1st block/ag allocated */
1110 struct xfs_defer_ops dfops
; /* freed extent records */
1111 xfs_mount_t
*mp
; /* mount structure */
1112 xfs_trans_t
*tp
; /* transaction pointer */
1113 int blks
; /* space reservation */
1114 int version
= 1; /* superblock attr version */
1115 int logflags
; /* logging flags */
1116 int error
; /* error return value */
1118 ASSERT(XFS_IFORK_Q(ip
) == 0);
1121 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1123 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1125 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_addafork
, blks
, 0,
1126 rsvd
? XFS_TRANS_RESERVE
: 0, &tp
);
1130 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1131 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1132 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1133 XFS_QMOPT_RES_REGBLKS
);
1136 if (XFS_IFORK_Q(ip
))
1138 if (ip
->i_d
.di_anextents
!= 0) {
1139 error
= -EFSCORRUPTED
;
1142 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
1144 * For inodes coming from pre-6.2 filesystems.
1146 ASSERT(ip
->i_d
.di_aformat
== 0);
1147 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1150 xfs_trans_ijoin(tp
, ip
, 0);
1151 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1153 switch (ip
->i_d
.di_format
) {
1154 case XFS_DINODE_FMT_DEV
:
1155 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1157 case XFS_DINODE_FMT_UUID
:
1158 ip
->i_d
.di_forkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
1160 case XFS_DINODE_FMT_LOCAL
:
1161 case XFS_DINODE_FMT_EXTENTS
:
1162 case XFS_DINODE_FMT_BTREE
:
1163 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1164 if (!ip
->i_d
.di_forkoff
)
1165 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1166 else if (mp
->m_flags
& XFS_MOUNT_ATTR2
)
1175 ASSERT(ip
->i_afp
== NULL
);
1176 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
1177 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1179 xfs_defer_init(&dfops
, &firstblock
);
1180 switch (ip
->i_d
.di_format
) {
1181 case XFS_DINODE_FMT_LOCAL
:
1182 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &firstblock
, &dfops
,
1185 case XFS_DINODE_FMT_EXTENTS
:
1186 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &firstblock
,
1189 case XFS_DINODE_FMT_BTREE
:
1190 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &firstblock
, &dfops
,
1198 xfs_trans_log_inode(tp
, ip
, logflags
);
1201 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1202 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1203 bool log_sb
= false;
1205 spin_lock(&mp
->m_sb_lock
);
1206 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1207 xfs_sb_version_addattr(&mp
->m_sb
);
1210 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1211 xfs_sb_version_addattr2(&mp
->m_sb
);
1214 spin_unlock(&mp
->m_sb_lock
);
1219 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
1222 error
= xfs_trans_commit(tp
);
1223 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1227 xfs_defer_cancel(&dfops
);
1229 xfs_trans_cancel(tp
);
1230 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1235 * Internal and external extent tree search functions.
1239 * Read in the extents to if_extents.
1240 * All inode fields are set up by caller, we just traverse the btree
1241 * and copy the records in. If the file system cannot contain unwritten
1242 * extents, the records are checked for no "state" flags.
1245 xfs_bmap_read_extents(
1246 xfs_trans_t
*tp
, /* transaction pointer */
1247 xfs_inode_t
*ip
, /* incore inode */
1248 int whichfork
) /* data or attr fork */
1250 struct xfs_btree_block
*block
; /* current btree block */
1251 xfs_fsblock_t bno
; /* block # of "block" */
1252 xfs_buf_t
*bp
; /* buffer for "block" */
1253 int error
; /* error return value */
1254 xfs_exntfmt_t exntf
; /* XFS_EXTFMT_NOSTATE, if checking */
1255 xfs_extnum_t i
, j
; /* index into the extents list */
1256 xfs_ifork_t
*ifp
; /* fork structure */
1257 int level
; /* btree level, for checking */
1258 xfs_mount_t
*mp
; /* file system mount structure */
1259 __be64
*pp
; /* pointer to block address */
1261 xfs_extnum_t room
; /* number of entries there's room for */
1264 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1265 exntf
= (whichfork
!= XFS_DATA_FORK
) ? XFS_EXTFMT_NOSTATE
:
1266 XFS_EXTFMT_INODE(ip
);
1267 block
= ifp
->if_broot
;
1269 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1271 level
= be16_to_cpu(block
->bb_level
);
1273 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
1274 bno
= be64_to_cpu(*pp
);
1277 * Go down the tree until leaf level is reached, following the first
1278 * pointer (leftmost) at each level.
1280 while (level
-- > 0) {
1281 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1282 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1285 block
= XFS_BUF_TO_BLOCK(bp
);
1288 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
1289 bno
= be64_to_cpu(*pp
);
1290 XFS_WANT_CORRUPTED_GOTO(mp
,
1291 XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
1292 xfs_trans_brelse(tp
, bp
);
1295 * Here with bp and block set to the leftmost leaf node in the tree.
1297 room
= xfs_iext_count(ifp
);
1300 * Loop over all leaf nodes. Copy information to the extent records.
1303 xfs_bmbt_rec_t
*frp
;
1304 xfs_fsblock_t nextbno
;
1305 xfs_extnum_t num_recs
;
1308 num_recs
= xfs_btree_get_numrecs(block
);
1309 if (unlikely(i
+ num_recs
> room
)) {
1310 ASSERT(i
+ num_recs
<= room
);
1311 xfs_warn(ip
->i_mount
,
1312 "corrupt dinode %Lu, (btree extents).",
1313 (unsigned long long) ip
->i_ino
);
1314 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1315 XFS_ERRLEVEL_LOW
, ip
->i_mount
, block
);
1319 * Read-ahead the next leaf block, if any.
1321 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
1322 if (nextbno
!= NULLFSBLOCK
)
1323 xfs_btree_reada_bufl(mp
, nextbno
, 1,
1326 * Copy records into the extent records.
1328 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1330 for (j
= 0; j
< num_recs
; j
++, i
++, frp
++) {
1331 xfs_bmbt_rec_host_t
*trp
= xfs_iext_get_ext(ifp
, i
);
1332 trp
->l0
= be64_to_cpu(frp
->l0
);
1333 trp
->l1
= be64_to_cpu(frp
->l1
);
1335 if (exntf
== XFS_EXTFMT_NOSTATE
) {
1337 * Check all attribute bmap btree records and
1338 * any "older" data bmap btree records for a
1339 * set bit in the "extent flag" position.
1341 if (unlikely(xfs_check_nostate_extents(ifp
,
1342 start
, num_recs
))) {
1343 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1349 xfs_trans_brelse(tp
, bp
);
1352 * If we've reached the end, stop.
1354 if (bno
== NULLFSBLOCK
)
1356 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1357 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1360 block
= XFS_BUF_TO_BLOCK(bp
);
1362 if (i
!= XFS_IFORK_NEXTENTS(ip
, whichfork
))
1363 return -EFSCORRUPTED
;
1364 ASSERT(i
== xfs_iext_count(ifp
));
1365 XFS_BMAP_TRACE_EXLIST(ip
, i
, whichfork
);
1368 xfs_trans_brelse(tp
, bp
);
1369 return -EFSCORRUPTED
;
1373 * Returns the file-relative block number of the first unused block(s)
1374 * in the file with at least "len" logically contiguous blocks free.
1375 * This is the lowest-address hole if the file has holes, else the first block
1376 * past the end of file.
1377 * Return 0 if the file is currently local (in-inode).
1380 xfs_bmap_first_unused(
1381 xfs_trans_t
*tp
, /* transaction pointer */
1382 xfs_inode_t
*ip
, /* incore inode */
1383 xfs_extlen_t len
, /* size of hole to find */
1384 xfs_fileoff_t
*first_unused
, /* unused block */
1385 int whichfork
) /* data or attr fork */
1387 int error
; /* error return value */
1388 int idx
; /* extent record index */
1389 xfs_ifork_t
*ifp
; /* inode fork pointer */
1390 xfs_fileoff_t lastaddr
; /* last block number seen */
1391 xfs_fileoff_t lowest
; /* lowest useful block */
1392 xfs_fileoff_t max
; /* starting useful block */
1393 xfs_fileoff_t off
; /* offset for this block */
1394 xfs_extnum_t nextents
; /* number of extent entries */
1396 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
1397 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
1398 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
1399 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1403 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1404 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
1405 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
1407 lowest
= *first_unused
;
1408 nextents
= xfs_iext_count(ifp
);
1409 for (idx
= 0, lastaddr
= 0, max
= lowest
; idx
< nextents
; idx
++) {
1410 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, idx
);
1411 off
= xfs_bmbt_get_startoff(ep
);
1413 * See if the hole before this extent will work.
1415 if (off
>= lowest
+ len
&& off
- max
>= len
) {
1416 *first_unused
= max
;
1419 lastaddr
= off
+ xfs_bmbt_get_blockcount(ep
);
1420 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1422 *first_unused
= max
;
1427 * Returns the file-relative block number of the last block - 1 before
1428 * last_block (input value) in the file.
1429 * This is not based on i_size, it is based on the extent records.
1430 * Returns 0 for local files, as they do not have extent records.
1433 xfs_bmap_last_before(
1434 struct xfs_trans
*tp
, /* transaction pointer */
1435 struct xfs_inode
*ip
, /* incore inode */
1436 xfs_fileoff_t
*last_block
, /* last block */
1437 int whichfork
) /* data or attr fork */
1439 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1440 struct xfs_bmbt_irec got
;
1444 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
1445 case XFS_DINODE_FMT_LOCAL
:
1448 case XFS_DINODE_FMT_BTREE
:
1449 case XFS_DINODE_FMT_EXTENTS
:
1455 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1456 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1461 if (xfs_iext_lookup_extent(ip
, ifp
, *last_block
- 1, &idx
, &got
)) {
1462 if (got
.br_startoff
<= *last_block
- 1)
1466 if (xfs_iext_get_extent(ifp
, idx
- 1, &got
)) {
1467 *last_block
= got
.br_startoff
+ got
.br_blockcount
;
1476 xfs_bmap_last_extent(
1477 struct xfs_trans
*tp
,
1478 struct xfs_inode
*ip
,
1480 struct xfs_bmbt_irec
*rec
,
1483 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1487 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1488 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1493 nextents
= xfs_iext_count(ifp
);
1494 if (nextents
== 0) {
1499 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, nextents
- 1), rec
);
1505 * Check the last inode extent to determine whether this allocation will result
1506 * in blocks being allocated at the end of the file. When we allocate new data
1507 * blocks at the end of the file which do not start at the previous data block,
1508 * we will try to align the new blocks at stripe unit boundaries.
1510 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1511 * at, or past the EOF.
1515 struct xfs_bmalloca
*bma
,
1518 struct xfs_bmbt_irec rec
;
1523 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1534 * Check if we are allocation or past the last extent, or at least into
1535 * the last delayed allocated extent.
1537 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1538 (bma
->offset
>= rec
.br_startoff
&&
1539 isnullstartblock(rec
.br_startblock
));
1544 * Returns the file-relative block number of the first block past eof in
1545 * the file. This is not based on i_size, it is based on the extent records.
1546 * Returns 0 for local files, as they do not have extent records.
1549 xfs_bmap_last_offset(
1550 struct xfs_inode
*ip
,
1551 xfs_fileoff_t
*last_block
,
1554 struct xfs_bmbt_irec rec
;
1560 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
)
1563 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
1564 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1567 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1568 if (error
|| is_empty
)
1571 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1576 * Returns whether the selected fork of the inode has exactly one
1577 * block or not. For the data fork we check this matches di_size,
1578 * implying the file's range is 0..bsize-1.
1580 int /* 1=>1 block, 0=>otherwise */
1582 xfs_inode_t
*ip
, /* incore inode */
1583 int whichfork
) /* data or attr fork */
1585 xfs_bmbt_rec_host_t
*ep
; /* ptr to fork's extent */
1586 xfs_ifork_t
*ifp
; /* inode fork pointer */
1587 int rval
; /* return value */
1588 xfs_bmbt_irec_t s
; /* internal version of extent */
1591 if (whichfork
== XFS_DATA_FORK
)
1592 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1594 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
1596 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1598 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1599 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1600 ep
= xfs_iext_get_ext(ifp
, 0);
1601 xfs_bmbt_get_all(ep
, &s
);
1602 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1603 if (rval
&& whichfork
== XFS_DATA_FORK
)
1604 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1609 * Extent tree manipulation functions used during allocation.
1613 * Convert a delayed allocation to a real allocation.
1615 STATIC
int /* error */
1616 xfs_bmap_add_extent_delay_real(
1617 struct xfs_bmalloca
*bma
,
1620 struct xfs_bmbt_irec
*new = &bma
->got
;
1621 int diff
; /* temp value */
1622 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
1623 int error
; /* error return value */
1624 int i
; /* temp state */
1625 xfs_ifork_t
*ifp
; /* inode fork pointer */
1626 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1627 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1628 /* left is 0, right is 1, prev is 2 */
1629 int rval
=0; /* return value (logging flags) */
1630 int state
= 0;/* state bits, accessed thru macros */
1631 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1632 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1633 xfs_filblks_t temp
=0; /* value for da_new calculations */
1634 xfs_filblks_t temp2
=0;/* value for da_new calculations */
1635 int tmp_rval
; /* partial logging flags */
1636 struct xfs_mount
*mp
;
1637 xfs_extnum_t
*nextents
;
1639 mp
= bma
->ip
->i_mount
;
1640 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1641 ASSERT(whichfork
!= XFS_ATTR_FORK
);
1642 nextents
= (whichfork
== XFS_COW_FORK
? &bma
->ip
->i_cnextents
:
1643 &bma
->ip
->i_d
.di_nextents
);
1645 ASSERT(bma
->idx
>= 0);
1646 ASSERT(bma
->idx
<= xfs_iext_count(ifp
));
1647 ASSERT(!isnullstartblock(new->br_startblock
));
1649 (bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
1651 XFS_STATS_INC(mp
, xs_add_exlist
);
1657 if (whichfork
== XFS_COW_FORK
)
1658 state
|= BMAP_COWFORK
;
1661 * Set up a bunch of variables to make the tests simpler.
1663 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
1664 xfs_bmbt_get_all(ep
, &PREV
);
1665 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1666 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1667 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1669 da_old
= startblockval(PREV
.br_startblock
);
1673 * Set flags determining what part of the previous delayed allocation
1674 * extent is being replaced by a real allocation.
1676 if (PREV
.br_startoff
== new->br_startoff
)
1677 state
|= BMAP_LEFT_FILLING
;
1678 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1679 state
|= BMAP_RIGHT_FILLING
;
1682 * Check and set flags if this segment has a left neighbor.
1683 * Don't set contiguous if the combined extent would be too large.
1686 state
|= BMAP_LEFT_VALID
;
1687 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &LEFT
);
1689 if (isnullstartblock(LEFT
.br_startblock
))
1690 state
|= BMAP_LEFT_DELAY
;
1693 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1694 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1695 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1696 LEFT
.br_state
== new->br_state
&&
1697 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1698 state
|= BMAP_LEFT_CONTIG
;
1701 * Check and set flags if this segment has a right neighbor.
1702 * Don't set contiguous if the combined extent would be too large.
1703 * Also check for all-three-contiguous being too large.
1705 if (bma
->idx
< xfs_iext_count(ifp
) - 1) {
1706 state
|= BMAP_RIGHT_VALID
;
1707 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
+ 1), &RIGHT
);
1709 if (isnullstartblock(RIGHT
.br_startblock
))
1710 state
|= BMAP_RIGHT_DELAY
;
1713 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1714 new_endoff
== RIGHT
.br_startoff
&&
1715 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1716 new->br_state
== RIGHT
.br_state
&&
1717 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1718 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1719 BMAP_RIGHT_FILLING
)) !=
1720 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1721 BMAP_RIGHT_FILLING
) ||
1722 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1724 state
|= BMAP_RIGHT_CONTIG
;
1728 * Switch out based on the FILLING and CONTIG state bits.
1730 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1731 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1732 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1733 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1735 * Filling in all of a previously delayed allocation extent.
1736 * The left and right neighbors are both contiguous with new.
1739 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1740 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1741 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1742 RIGHT
.br_blockcount
);
1743 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1745 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 2, state
);
1747 if (bma
->cur
== NULL
)
1748 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1750 rval
= XFS_ILOG_CORE
;
1751 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1752 RIGHT
.br_startblock
,
1753 RIGHT
.br_blockcount
, &i
);
1756 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1757 error
= xfs_btree_delete(bma
->cur
, &i
);
1760 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1761 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1764 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1765 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1767 LEFT
.br_blockcount
+
1768 PREV
.br_blockcount
+
1769 RIGHT
.br_blockcount
, LEFT
.br_state
);
1775 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1777 * Filling in all of a previously delayed allocation extent.
1778 * The left neighbor is contiguous, the right is not.
1782 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1783 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1784 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
1785 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1787 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1788 if (bma
->cur
== NULL
)
1789 rval
= XFS_ILOG_DEXT
;
1792 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
1793 LEFT
.br_startblock
, LEFT
.br_blockcount
,
1797 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1798 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1800 LEFT
.br_blockcount
+
1801 PREV
.br_blockcount
, LEFT
.br_state
);
1807 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1809 * Filling in all of a previously delayed allocation extent.
1810 * The right neighbor is contiguous, the left is not.
1812 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1813 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1814 xfs_bmbt_set_blockcount(ep
,
1815 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
1816 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1818 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1819 if (bma
->cur
== NULL
)
1820 rval
= XFS_ILOG_DEXT
;
1823 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1824 RIGHT
.br_startblock
,
1825 RIGHT
.br_blockcount
, &i
);
1828 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1829 error
= xfs_bmbt_update(bma
->cur
, PREV
.br_startoff
,
1831 PREV
.br_blockcount
+
1832 RIGHT
.br_blockcount
, PREV
.br_state
);
1838 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1840 * Filling in all of a previously delayed allocation extent.
1841 * Neither the left nor right neighbors are contiguous with
1844 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1845 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
1846 xfs_bmbt_set_state(ep
, new->br_state
);
1847 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1850 if (bma
->cur
== NULL
)
1851 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1853 rval
= XFS_ILOG_CORE
;
1854 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
1855 new->br_startblock
, new->br_blockcount
,
1859 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1860 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1861 error
= xfs_btree_insert(bma
->cur
, &i
);
1864 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1868 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1870 * Filling in the first part of a previous delayed allocation.
1871 * The left neighbor is contiguous.
1873 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1874 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
1875 LEFT
.br_blockcount
+ new->br_blockcount
);
1876 xfs_bmbt_set_startoff(ep
,
1877 PREV
.br_startoff
+ new->br_blockcount
);
1878 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
1880 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1881 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1882 xfs_bmbt_set_blockcount(ep
, temp
);
1883 if (bma
->cur
== NULL
)
1884 rval
= XFS_ILOG_DEXT
;
1887 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
1888 LEFT
.br_startblock
, LEFT
.br_blockcount
,
1892 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1893 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
1895 LEFT
.br_blockcount
+
1901 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1902 startblockval(PREV
.br_startblock
));
1903 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
1904 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1909 case BMAP_LEFT_FILLING
:
1911 * Filling in the first part of a previous delayed allocation.
1912 * The left neighbor is not contiguous.
1914 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1915 xfs_bmbt_set_startoff(ep
, new_endoff
);
1916 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1917 xfs_bmbt_set_blockcount(ep
, temp
);
1918 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
1920 if (bma
->cur
== NULL
)
1921 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1923 rval
= XFS_ILOG_CORE
;
1924 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
1925 new->br_startblock
, new->br_blockcount
,
1929 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1930 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1931 error
= xfs_btree_insert(bma
->cur
, &i
);
1934 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1937 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1938 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1939 bma
->firstblock
, bma
->dfops
,
1940 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1945 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1946 startblockval(PREV
.br_startblock
) -
1947 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
1948 ep
= xfs_iext_get_ext(ifp
, bma
->idx
+ 1);
1949 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
1950 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1953 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1955 * Filling in the last part of a previous delayed allocation.
1956 * The right neighbor is contiguous with the new allocation.
1958 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1959 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1960 xfs_bmbt_set_blockcount(ep
, temp
);
1961 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
+ 1),
1962 new->br_startoff
, new->br_startblock
,
1963 new->br_blockcount
+ RIGHT
.br_blockcount
,
1965 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
1966 if (bma
->cur
== NULL
)
1967 rval
= XFS_ILOG_DEXT
;
1970 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
1971 RIGHT
.br_startblock
,
1972 RIGHT
.br_blockcount
, &i
);
1975 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1976 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
1978 new->br_blockcount
+
1979 RIGHT
.br_blockcount
,
1985 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1986 startblockval(PREV
.br_startblock
));
1987 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1988 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
1989 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1994 case BMAP_RIGHT_FILLING
:
1996 * Filling in the last part of a previous delayed allocation.
1997 * The right neighbor is not contiguous.
1999 temp
= PREV
.br_blockcount
- new->br_blockcount
;
2000 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2001 xfs_bmbt_set_blockcount(ep
, temp
);
2002 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 1, new, state
);
2004 if (bma
->cur
== NULL
)
2005 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2007 rval
= XFS_ILOG_CORE
;
2008 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2009 new->br_startblock
, new->br_blockcount
,
2013 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2014 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2015 error
= xfs_btree_insert(bma
->cur
, &i
);
2018 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2021 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2022 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2023 bma
->firstblock
, bma
->dfops
, &bma
->cur
, 1,
2024 &tmp_rval
, whichfork
);
2029 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
2030 startblockval(PREV
.br_startblock
) -
2031 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2032 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2033 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
2034 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2041 * Filling in the middle part of a previous delayed allocation.
2042 * Contiguity is impossible here.
2043 * This case is avoided almost all the time.
2045 * We start with a delayed allocation:
2047 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2050 * and we are allocating:
2051 * +rrrrrrrrrrrrrrrrr+
2054 * and we set it up for insertion as:
2055 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2057 * PREV @ idx LEFT RIGHT
2058 * inserted at idx + 1
2060 temp
= new->br_startoff
- PREV
.br_startoff
;
2061 temp2
= PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2062 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, 0, _THIS_IP_
);
2063 xfs_bmbt_set_blockcount(ep
, temp
); /* truncate PREV */
2065 RIGHT
.br_state
= PREV
.br_state
;
2066 RIGHT
.br_startblock
= nullstartblock(
2067 (int)xfs_bmap_worst_indlen(bma
->ip
, temp2
));
2068 RIGHT
.br_startoff
= new_endoff
;
2069 RIGHT
.br_blockcount
= temp2
;
2070 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2071 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 2, &LEFT
, state
);
2073 if (bma
->cur
== NULL
)
2074 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2076 rval
= XFS_ILOG_CORE
;
2077 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
2078 new->br_startblock
, new->br_blockcount
,
2082 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2083 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2084 error
= xfs_btree_insert(bma
->cur
, &i
);
2087 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2090 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2091 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2092 bma
->firstblock
, bma
->dfops
, &bma
->cur
,
2093 1, &tmp_rval
, whichfork
);
2098 temp
= xfs_bmap_worst_indlen(bma
->ip
, temp
);
2099 temp2
= xfs_bmap_worst_indlen(bma
->ip
, temp2
);
2100 diff
= (int)(temp
+ temp2
- startblockval(PREV
.br_startblock
) -
2101 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
2103 error
= xfs_mod_fdblocks(bma
->ip
->i_mount
,
2104 -((int64_t)diff
), false);
2110 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
2111 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2112 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
2113 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2114 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, bma
->idx
+ 2),
2115 nullstartblock((int)temp2
));
2116 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
2119 da_new
= temp
+ temp2
;
2122 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2123 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2124 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2125 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2126 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2127 case BMAP_LEFT_CONTIG
:
2128 case BMAP_RIGHT_CONTIG
:
2130 * These cases are all impossible.
2135 /* add reverse mapping */
2136 error
= xfs_rmap_map_extent(mp
, bma
->dfops
, bma
->ip
, whichfork
, new);
2140 /* convert to a btree if necessary */
2141 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2142 int tmp_logflags
; /* partial log flag return val */
2144 ASSERT(bma
->cur
== NULL
);
2145 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2146 bma
->firstblock
, bma
->dfops
, &bma
->cur
,
2147 da_old
> 0, &tmp_logflags
, whichfork
);
2148 bma
->logflags
|= tmp_logflags
;
2153 /* adjust for changes in reserved delayed indirect blocks */
2154 if (da_old
|| da_new
) {
2157 temp
+= bma
->cur
->bc_private
.b
.allocated
;
2158 ASSERT(temp
<= da_old
);
2160 xfs_mod_fdblocks(bma
->ip
->i_mount
,
2161 (int64_t)(da_old
- temp
), false);
2164 /* clear out the allocated field, done with it now in any case. */
2166 bma
->cur
->bc_private
.b
.allocated
= 0;
2168 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
2170 if (whichfork
!= XFS_COW_FORK
)
2171 bma
->logflags
|= rval
;
2179 * Convert an unwritten allocation to a real allocation or vice versa.
2181 STATIC
int /* error */
2182 xfs_bmap_add_extent_unwritten_real(
2183 struct xfs_trans
*tp
,
2184 xfs_inode_t
*ip
, /* incore inode pointer */
2186 xfs_extnum_t
*idx
, /* extent number to update/insert */
2187 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2188 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2189 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
2190 struct xfs_defer_ops
*dfops
, /* list of extents to be freed */
2191 int *logflagsp
) /* inode logging flags */
2193 xfs_btree_cur_t
*cur
; /* btree cursor */
2194 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
2195 int error
; /* error return value */
2196 int i
; /* temp state */
2197 xfs_ifork_t
*ifp
; /* inode fork pointer */
2198 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2199 xfs_exntst_t newext
; /* new extent state */
2200 xfs_exntst_t oldext
; /* old extent state */
2201 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2202 /* left is 0, right is 1, prev is 2 */
2203 int rval
=0; /* return value (logging flags) */
2204 int state
= 0;/* state bits, accessed thru macros */
2205 struct xfs_mount
*mp
= ip
->i_mount
;
2210 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2211 if (whichfork
== XFS_COW_FORK
)
2212 state
|= BMAP_COWFORK
;
2215 ASSERT(*idx
<= xfs_iext_count(ifp
));
2216 ASSERT(!isnullstartblock(new->br_startblock
));
2218 XFS_STATS_INC(mp
, xs_add_exlist
);
2225 * Set up a bunch of variables to make the tests simpler.
2228 ep
= xfs_iext_get_ext(ifp
, *idx
);
2229 xfs_bmbt_get_all(ep
, &PREV
);
2230 newext
= new->br_state
;
2231 oldext
= (newext
== XFS_EXT_UNWRITTEN
) ?
2232 XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
2233 ASSERT(PREV
.br_state
== oldext
);
2234 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2235 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2236 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2239 * Set flags determining what part of the previous oldext allocation
2240 * extent is being replaced by a newext allocation.
2242 if (PREV
.br_startoff
== new->br_startoff
)
2243 state
|= BMAP_LEFT_FILLING
;
2244 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2245 state
|= BMAP_RIGHT_FILLING
;
2248 * Check and set flags if this segment has a left neighbor.
2249 * Don't set contiguous if the combined extent would be too large.
2252 state
|= BMAP_LEFT_VALID
;
2253 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
2255 if (isnullstartblock(LEFT
.br_startblock
))
2256 state
|= BMAP_LEFT_DELAY
;
2259 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2260 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2261 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2262 LEFT
.br_state
== newext
&&
2263 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2264 state
|= BMAP_LEFT_CONTIG
;
2267 * Check and set flags if this segment has a right neighbor.
2268 * Don't set contiguous if the combined extent would be too large.
2269 * Also check for all-three-contiguous being too large.
2271 if (*idx
< xfs_iext_count(ifp
) - 1) {
2272 state
|= BMAP_RIGHT_VALID
;
2273 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
2274 if (isnullstartblock(RIGHT
.br_startblock
))
2275 state
|= BMAP_RIGHT_DELAY
;
2278 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2279 new_endoff
== RIGHT
.br_startoff
&&
2280 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2281 newext
== RIGHT
.br_state
&&
2282 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2283 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2284 BMAP_RIGHT_FILLING
)) !=
2285 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2286 BMAP_RIGHT_FILLING
) ||
2287 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2289 state
|= BMAP_RIGHT_CONTIG
;
2292 * Switch out based on the FILLING and CONTIG state bits.
2294 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2295 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2296 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2297 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2299 * Setting all of a previous oldext extent to newext.
2300 * The left and right neighbors are both contiguous with new.
2304 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2305 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2306 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2307 RIGHT
.br_blockcount
);
2308 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2310 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
2311 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2312 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 2);
2314 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2316 rval
= XFS_ILOG_CORE
;
2317 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2318 RIGHT
.br_startblock
,
2319 RIGHT
.br_blockcount
, &i
)))
2321 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2322 if ((error
= xfs_btree_delete(cur
, &i
)))
2324 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2325 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2327 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2328 if ((error
= xfs_btree_delete(cur
, &i
)))
2330 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2331 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2333 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2334 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2336 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
2337 RIGHT
.br_blockcount
, LEFT
.br_state
)))
2342 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2344 * Setting all of a previous oldext extent to newext.
2345 * The left neighbor is contiguous, the right is not.
2349 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2350 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2351 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
2352 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2354 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2355 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2356 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2358 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2360 rval
= XFS_ILOG_CORE
;
2361 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2362 PREV
.br_startblock
, PREV
.br_blockcount
,
2365 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2366 if ((error
= xfs_btree_delete(cur
, &i
)))
2368 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2369 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2371 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2372 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2374 LEFT
.br_blockcount
+ PREV
.br_blockcount
,
2380 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2382 * Setting all of a previous oldext extent to newext.
2383 * The right neighbor is contiguous, the left is not.
2385 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2386 xfs_bmbt_set_blockcount(ep
,
2387 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
2388 xfs_bmbt_set_state(ep
, newext
);
2389 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2390 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2391 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2392 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2394 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2396 rval
= XFS_ILOG_CORE
;
2397 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
2398 RIGHT
.br_startblock
,
2399 RIGHT
.br_blockcount
, &i
)))
2401 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2402 if ((error
= xfs_btree_delete(cur
, &i
)))
2404 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2405 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2407 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2408 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2410 new->br_blockcount
+ RIGHT
.br_blockcount
,
2416 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2418 * Setting all of a previous oldext extent to newext.
2419 * Neither the left nor right neighbors are contiguous with
2422 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2423 xfs_bmbt_set_state(ep
, newext
);
2424 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2427 rval
= XFS_ILOG_DEXT
;
2430 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2431 new->br_startblock
, new->br_blockcount
,
2434 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2435 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2436 new->br_startblock
, new->br_blockcount
,
2442 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2444 * Setting the first part of a previous oldext extent to newext.
2445 * The left neighbor is contiguous.
2447 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2448 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
2449 LEFT
.br_blockcount
+ new->br_blockcount
);
2450 xfs_bmbt_set_startoff(ep
,
2451 PREV
.br_startoff
+ new->br_blockcount
);
2452 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
2454 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2455 xfs_bmbt_set_startblock(ep
,
2456 new->br_startblock
+ new->br_blockcount
);
2457 xfs_bmbt_set_blockcount(ep
,
2458 PREV
.br_blockcount
- new->br_blockcount
);
2459 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2464 rval
= XFS_ILOG_DEXT
;
2467 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2468 PREV
.br_startblock
, PREV
.br_blockcount
,
2471 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2472 if ((error
= xfs_bmbt_update(cur
,
2473 PREV
.br_startoff
+ new->br_blockcount
,
2474 PREV
.br_startblock
+ new->br_blockcount
,
2475 PREV
.br_blockcount
- new->br_blockcount
,
2478 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2480 error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
2482 LEFT
.br_blockcount
+ new->br_blockcount
,
2489 case BMAP_LEFT_FILLING
:
2491 * Setting the first part of a previous oldext extent to newext.
2492 * The left neighbor is not contiguous.
2494 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2495 ASSERT(ep
&& xfs_bmbt_get_state(ep
) == oldext
);
2496 xfs_bmbt_set_startoff(ep
, new_endoff
);
2497 xfs_bmbt_set_blockcount(ep
,
2498 PREV
.br_blockcount
- new->br_blockcount
);
2499 xfs_bmbt_set_startblock(ep
,
2500 new->br_startblock
+ new->br_blockcount
);
2501 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2503 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2504 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2505 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2507 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2509 rval
= XFS_ILOG_CORE
;
2510 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2511 PREV
.br_startblock
, PREV
.br_blockcount
,
2514 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2515 if ((error
= xfs_bmbt_update(cur
,
2516 PREV
.br_startoff
+ new->br_blockcount
,
2517 PREV
.br_startblock
+ new->br_blockcount
,
2518 PREV
.br_blockcount
- new->br_blockcount
,
2521 cur
->bc_rec
.b
= *new;
2522 if ((error
= xfs_btree_insert(cur
, &i
)))
2524 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2528 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2530 * Setting the last part of a previous oldext extent to newext.
2531 * The right neighbor is contiguous with the new allocation.
2533 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2534 xfs_bmbt_set_blockcount(ep
,
2535 PREV
.br_blockcount
- new->br_blockcount
);
2536 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2540 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2541 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2542 new->br_startoff
, new->br_startblock
,
2543 new->br_blockcount
+ RIGHT
.br_blockcount
, newext
);
2544 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2547 rval
= XFS_ILOG_DEXT
;
2550 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2552 PREV
.br_blockcount
, &i
)))
2554 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2555 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2557 PREV
.br_blockcount
- new->br_blockcount
,
2560 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
2562 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
2564 new->br_blockcount
+ RIGHT
.br_blockcount
,
2570 case BMAP_RIGHT_FILLING
:
2572 * Setting the last part of a previous oldext extent to newext.
2573 * The right neighbor is not contiguous.
2575 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2576 xfs_bmbt_set_blockcount(ep
,
2577 PREV
.br_blockcount
- new->br_blockcount
);
2578 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2581 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2583 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2584 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2586 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2588 rval
= XFS_ILOG_CORE
;
2589 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2590 PREV
.br_startblock
, PREV
.br_blockcount
,
2593 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2594 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
2596 PREV
.br_blockcount
- new->br_blockcount
,
2599 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2600 new->br_startblock
, new->br_blockcount
,
2603 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2604 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
2605 if ((error
= xfs_btree_insert(cur
, &i
)))
2607 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2613 * Setting the middle part of a previous oldext extent to
2614 * newext. Contiguity is impossible here.
2615 * One extent becomes three extents.
2617 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2618 xfs_bmbt_set_blockcount(ep
,
2619 new->br_startoff
- PREV
.br_startoff
);
2620 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2623 r
[1].br_startoff
= new_endoff
;
2624 r
[1].br_blockcount
=
2625 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
2626 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2627 r
[1].br_state
= oldext
;
2630 xfs_iext_insert(ip
, *idx
, 2, &r
[0], state
);
2632 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2633 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 2);
2635 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2637 rval
= XFS_ILOG_CORE
;
2638 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
2639 PREV
.br_startblock
, PREV
.br_blockcount
,
2642 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2643 /* new right extent - oldext */
2644 if ((error
= xfs_bmbt_update(cur
, r
[1].br_startoff
,
2645 r
[1].br_startblock
, r
[1].br_blockcount
,
2648 /* new left extent - oldext */
2649 cur
->bc_rec
.b
= PREV
;
2650 cur
->bc_rec
.b
.br_blockcount
=
2651 new->br_startoff
- PREV
.br_startoff
;
2652 if ((error
= xfs_btree_insert(cur
, &i
)))
2654 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2656 * Reset the cursor to the position of the new extent
2657 * we are about to insert as we can't trust it after
2658 * the previous insert.
2660 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
2661 new->br_startblock
, new->br_blockcount
,
2664 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2665 /* new middle extent - newext */
2666 cur
->bc_rec
.b
.br_state
= new->br_state
;
2667 if ((error
= xfs_btree_insert(cur
, &i
)))
2669 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2673 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2674 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2675 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2676 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2677 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2678 case BMAP_LEFT_CONTIG
:
2679 case BMAP_RIGHT_CONTIG
:
2681 * These cases are all impossible.
2686 /* update reverse mappings */
2687 error
= xfs_rmap_convert_extent(mp
, dfops
, ip
, whichfork
, new);
2691 /* convert to a btree if necessary */
2692 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2693 int tmp_logflags
; /* partial log flag return val */
2695 ASSERT(cur
== NULL
);
2696 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
, dfops
, &cur
,
2697 0, &tmp_logflags
, whichfork
);
2698 *logflagsp
|= tmp_logflags
;
2703 /* clear out the allocated field, done with it now in any case. */
2705 cur
->bc_private
.b
.allocated
= 0;
2709 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
2719 * Convert a hole to a delayed allocation.
2722 xfs_bmap_add_extent_hole_delay(
2723 xfs_inode_t
*ip
, /* incore inode pointer */
2725 xfs_extnum_t
*idx
, /* extent number to update/insert */
2726 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2728 xfs_ifork_t
*ifp
; /* inode fork pointer */
2729 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2730 xfs_filblks_t newlen
=0; /* new indirect size */
2731 xfs_filblks_t oldlen
=0; /* old indirect size */
2732 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2733 int state
; /* state bits, accessed thru macros */
2734 xfs_filblks_t temp
=0; /* temp for indirect calculations */
2736 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2738 if (whichfork
== XFS_COW_FORK
)
2739 state
|= BMAP_COWFORK
;
2740 ASSERT(isnullstartblock(new->br_startblock
));
2743 * Check and set flags if this segment has a left neighbor
2746 state
|= BMAP_LEFT_VALID
;
2747 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
2749 if (isnullstartblock(left
.br_startblock
))
2750 state
|= BMAP_LEFT_DELAY
;
2754 * Check and set flags if the current (right) segment exists.
2755 * If it doesn't exist, we're converting the hole at end-of-file.
2757 if (*idx
< xfs_iext_count(ifp
)) {
2758 state
|= BMAP_RIGHT_VALID
;
2759 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
2761 if (isnullstartblock(right
.br_startblock
))
2762 state
|= BMAP_RIGHT_DELAY
;
2766 * Set contiguity flags on the left and right neighbors.
2767 * Don't let extents get too large, even if the pieces are contiguous.
2769 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2770 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2771 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2772 state
|= BMAP_LEFT_CONTIG
;
2774 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2775 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2776 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2777 (!(state
& BMAP_LEFT_CONTIG
) ||
2778 (left
.br_blockcount
+ new->br_blockcount
+
2779 right
.br_blockcount
<= MAXEXTLEN
)))
2780 state
|= BMAP_RIGHT_CONTIG
;
2783 * Switch out based on the contiguity flags.
2785 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2786 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2788 * New allocation is contiguous with delayed allocations
2789 * on the left and on the right.
2790 * Merge all three into a single extent record.
2793 temp
= left
.br_blockcount
+ new->br_blockcount
+
2794 right
.br_blockcount
;
2796 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2797 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2798 oldlen
= startblockval(left
.br_startblock
) +
2799 startblockval(new->br_startblock
) +
2800 startblockval(right
.br_startblock
);
2801 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2803 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2804 nullstartblock((int)newlen
));
2805 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2807 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2810 case BMAP_LEFT_CONTIG
:
2812 * New allocation is contiguous with a delayed allocation
2814 * Merge the new allocation with the left neighbor.
2817 temp
= left
.br_blockcount
+ new->br_blockcount
;
2819 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2820 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
2821 oldlen
= startblockval(left
.br_startblock
) +
2822 startblockval(new->br_startblock
);
2823 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2825 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
2826 nullstartblock((int)newlen
));
2827 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2830 case BMAP_RIGHT_CONTIG
:
2832 * New allocation is contiguous with a delayed allocation
2834 * Merge the new allocation with the right neighbor.
2836 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2837 temp
= new->br_blockcount
+ right
.br_blockcount
;
2838 oldlen
= startblockval(new->br_startblock
) +
2839 startblockval(right
.br_startblock
);
2840 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2842 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
2844 nullstartblock((int)newlen
), temp
, right
.br_state
);
2845 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2850 * New allocation is not contiguous with another
2851 * delayed allocation.
2852 * Insert a new entry.
2854 oldlen
= newlen
= 0;
2855 xfs_iext_insert(ip
, *idx
, 1, new, state
);
2858 if (oldlen
!= newlen
) {
2859 ASSERT(oldlen
> newlen
);
2860 xfs_mod_fdblocks(ip
->i_mount
, (int64_t)(oldlen
- newlen
),
2863 * Nothing to do for disk quota accounting here.
2869 * Convert a hole to a real allocation.
2871 STATIC
int /* error */
2872 xfs_bmap_add_extent_hole_real(
2873 struct xfs_trans
*tp
,
2874 struct xfs_inode
*ip
,
2877 struct xfs_btree_cur
**curp
,
2878 struct xfs_bmbt_irec
*new,
2879 xfs_fsblock_t
*first
,
2880 struct xfs_defer_ops
*dfops
,
2883 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2884 struct xfs_mount
*mp
= ip
->i_mount
;
2885 struct xfs_btree_cur
*cur
= *curp
;
2886 int error
; /* error return value */
2887 int i
; /* temp state */
2888 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2889 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2890 int rval
=0; /* return value (logging flags) */
2891 int state
; /* state bits, accessed thru macros */
2894 ASSERT(*idx
<= xfs_iext_count(ifp
));
2895 ASSERT(!isnullstartblock(new->br_startblock
));
2896 ASSERT(!cur
|| !(cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
2898 XFS_STATS_INC(mp
, xs_add_exlist
);
2901 if (whichfork
== XFS_ATTR_FORK
)
2902 state
|= BMAP_ATTRFORK
;
2903 if (whichfork
== XFS_COW_FORK
)
2904 state
|= BMAP_COWFORK
;
2907 * Check and set flags if this segment has a left neighbor.
2910 state
|= BMAP_LEFT_VALID
;
2911 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
2912 if (isnullstartblock(left
.br_startblock
))
2913 state
|= BMAP_LEFT_DELAY
;
2917 * Check and set flags if this segment has a current value.
2918 * Not true if we're inserting into the "hole" at eof.
2920 if (*idx
< xfs_iext_count(ifp
)) {
2921 state
|= BMAP_RIGHT_VALID
;
2922 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
2923 if (isnullstartblock(right
.br_startblock
))
2924 state
|= BMAP_RIGHT_DELAY
;
2928 * We're inserting a real allocation between "left" and "right".
2929 * Set the contiguity flags. Don't let extents get too large.
2931 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2932 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2933 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
2934 left
.br_state
== new->br_state
&&
2935 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2936 state
|= BMAP_LEFT_CONTIG
;
2938 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2939 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2940 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
2941 new->br_state
== right
.br_state
&&
2942 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2943 (!(state
& BMAP_LEFT_CONTIG
) ||
2944 left
.br_blockcount
+ new->br_blockcount
+
2945 right
.br_blockcount
<= MAXEXTLEN
))
2946 state
|= BMAP_RIGHT_CONTIG
;
2950 * Select which case we're in here, and implement it.
2952 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2953 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2955 * New allocation is contiguous with real allocations on the
2956 * left and on the right.
2957 * Merge all three into a single extent record.
2960 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2961 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
2962 left
.br_blockcount
+ new->br_blockcount
+
2963 right
.br_blockcount
);
2964 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2966 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
2968 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2969 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2971 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2973 rval
= XFS_ILOG_CORE
;
2974 error
= xfs_bmbt_lookup_eq(cur
, right
.br_startoff
,
2975 right
.br_startblock
, right
.br_blockcount
,
2979 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2980 error
= xfs_btree_delete(cur
, &i
);
2983 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2984 error
= xfs_btree_decrement(cur
, 0, &i
);
2987 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2988 error
= xfs_bmbt_update(cur
, left
.br_startoff
,
2990 left
.br_blockcount
+
2991 new->br_blockcount
+
2992 right
.br_blockcount
,
2999 case BMAP_LEFT_CONTIG
:
3001 * New allocation is contiguous with a real allocation
3003 * Merge the new allocation with the left neighbor.
3006 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
3007 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
3008 left
.br_blockcount
+ new->br_blockcount
);
3009 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
3012 rval
= xfs_ilog_fext(whichfork
);
3015 error
= xfs_bmbt_lookup_eq(cur
, left
.br_startoff
,
3016 left
.br_startblock
, left
.br_blockcount
,
3020 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3021 error
= xfs_bmbt_update(cur
, left
.br_startoff
,
3023 left
.br_blockcount
+
3031 case BMAP_RIGHT_CONTIG
:
3033 * New allocation is contiguous with a real allocation
3035 * Merge the new allocation with the right neighbor.
3037 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
3038 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
3039 new->br_startoff
, new->br_startblock
,
3040 new->br_blockcount
+ right
.br_blockcount
,
3042 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
3045 rval
= xfs_ilog_fext(whichfork
);
3048 error
= xfs_bmbt_lookup_eq(cur
,
3050 right
.br_startblock
,
3051 right
.br_blockcount
, &i
);
3054 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3055 error
= xfs_bmbt_update(cur
, new->br_startoff
,
3057 new->br_blockcount
+
3058 right
.br_blockcount
,
3067 * New allocation is not contiguous with another
3069 * Insert a new entry.
3071 xfs_iext_insert(ip
, *idx
, 1, new, state
);
3072 XFS_IFORK_NEXT_SET(ip
, whichfork
,
3073 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
3075 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
3077 rval
= XFS_ILOG_CORE
;
3078 error
= xfs_bmbt_lookup_eq(cur
,
3081 new->br_blockcount
, &i
);
3084 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
3085 cur
->bc_rec
.b
.br_state
= new->br_state
;
3086 error
= xfs_btree_insert(cur
, &i
);
3089 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
3094 /* add reverse mapping */
3095 error
= xfs_rmap_map_extent(mp
, dfops
, ip
, whichfork
, new);
3099 /* convert to a btree if necessary */
3100 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
3101 int tmp_logflags
; /* partial log flag return val */
3103 ASSERT(cur
== NULL
);
3104 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
, dfops
, curp
,
3105 0, &tmp_logflags
, whichfork
);
3106 *logflagsp
|= tmp_logflags
;
3112 /* clear out the allocated field, done with it now in any case. */
3114 cur
->bc_private
.b
.allocated
= 0;
3116 xfs_bmap_check_leaf_extents(cur
, ip
, whichfork
);
3123 * Functions used in the extent read, allocate and remove paths
3127 * Adjust the size of the new extent based on di_extsize and rt extsize.
3130 xfs_bmap_extsize_align(
3132 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
3133 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
3134 xfs_extlen_t extsz
, /* align to this extent size */
3135 int rt
, /* is this a realtime inode? */
3136 int eof
, /* is extent at end-of-file? */
3137 int delay
, /* creating delalloc extent? */
3138 int convert
, /* overwriting unwritten extent? */
3139 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
3140 xfs_extlen_t
*lenp
) /* in/out: aligned length */
3142 xfs_fileoff_t orig_off
; /* original offset */
3143 xfs_extlen_t orig_alen
; /* original length */
3144 xfs_fileoff_t orig_end
; /* original off+len */
3145 xfs_fileoff_t nexto
; /* next file offset */
3146 xfs_fileoff_t prevo
; /* previous file offset */
3147 xfs_fileoff_t align_off
; /* temp for offset */
3148 xfs_extlen_t align_alen
; /* temp for length */
3149 xfs_extlen_t temp
; /* temp for calculations */
3154 orig_off
= align_off
= *offp
;
3155 orig_alen
= align_alen
= *lenp
;
3156 orig_end
= orig_off
+ orig_alen
;
3159 * If this request overlaps an existing extent, then don't
3160 * attempt to perform any additional alignment.
3162 if (!delay
&& !eof
&&
3163 (orig_off
>= gotp
->br_startoff
) &&
3164 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
3169 * If the file offset is unaligned vs. the extent size
3170 * we need to align it. This will be possible unless
3171 * the file was previously written with a kernel that didn't
3172 * perform this alignment, or if a truncate shot us in the
3175 temp
= do_mod(orig_off
, extsz
);
3181 /* Same adjustment for the end of the requested area. */
3182 temp
= (align_alen
% extsz
);
3184 align_alen
+= extsz
- temp
;
3187 * For large extent hint sizes, the aligned extent might be larger than
3188 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3189 * the length back under MAXEXTLEN. The outer allocation loops handle
3190 * short allocation just fine, so it is safe to do this. We only want to
3191 * do it when we are forced to, though, because it means more allocation
3192 * operations are required.
3194 while (align_alen
> MAXEXTLEN
)
3195 align_alen
-= extsz
;
3196 ASSERT(align_alen
<= MAXEXTLEN
);
3199 * If the previous block overlaps with this proposed allocation
3200 * then move the start forward without adjusting the length.
3202 if (prevp
->br_startoff
!= NULLFILEOFF
) {
3203 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
3204 prevo
= prevp
->br_startoff
;
3206 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
3209 if (align_off
!= orig_off
&& align_off
< prevo
)
3212 * If the next block overlaps with this proposed allocation
3213 * then move the start back without adjusting the length,
3214 * but not before offset 0.
3215 * This may of course make the start overlap previous block,
3216 * and if we hit the offset 0 limit then the next block
3217 * can still overlap too.
3219 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
3220 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
3221 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
3222 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
3224 nexto
= gotp
->br_startoff
;
3226 nexto
= NULLFILEOFF
;
3228 align_off
+ align_alen
!= orig_end
&&
3229 align_off
+ align_alen
> nexto
)
3230 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
3232 * If we're now overlapping the next or previous extent that
3233 * means we can't fit an extsz piece in this hole. Just move
3234 * the start forward to the first valid spot and set
3235 * the length so we hit the end.
3237 if (align_off
!= orig_off
&& align_off
< prevo
)
3239 if (align_off
+ align_alen
!= orig_end
&&
3240 align_off
+ align_alen
> nexto
&&
3241 nexto
!= NULLFILEOFF
) {
3242 ASSERT(nexto
> prevo
);
3243 align_alen
= nexto
- align_off
;
3247 * If realtime, and the result isn't a multiple of the realtime
3248 * extent size we need to remove blocks until it is.
3250 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
3252 * We're not covering the original request, or
3253 * we won't be able to once we fix the length.
3255 if (orig_off
< align_off
||
3256 orig_end
> align_off
+ align_alen
||
3257 align_alen
- temp
< orig_alen
)
3260 * Try to fix it by moving the start up.
3262 if (align_off
+ temp
<= orig_off
) {
3267 * Try to fix it by moving the end in.
3269 else if (align_off
+ align_alen
- temp
>= orig_end
)
3272 * Set the start to the minimum then trim the length.
3275 align_alen
-= orig_off
- align_off
;
3276 align_off
= orig_off
;
3277 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3280 * Result doesn't cover the request, fail it.
3282 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3285 ASSERT(orig_off
>= align_off
);
3286 /* see MAXEXTLEN handling above */
3287 ASSERT(orig_end
<= align_off
+ align_alen
||
3288 align_alen
+ extsz
> MAXEXTLEN
);
3292 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3293 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3294 if (prevp
->br_startoff
!= NULLFILEOFF
)
3295 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3303 #define XFS_ALLOC_GAP_UNITS 4
3307 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3309 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3310 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3311 xfs_mount_t
*mp
; /* mount point structure */
3312 int nullfb
; /* true if ap->firstblock isn't set */
3313 int rt
; /* true if inode is realtime */
3315 #define ISVALID(x,y) \
3317 (x) < mp->m_sb.sb_rblocks : \
3318 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3319 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3320 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3322 mp
= ap
->ip
->i_mount
;
3323 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3324 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) &&
3325 xfs_alloc_is_userdata(ap
->datatype
);
3326 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3328 * If allocating at eof, and there's a previous real block,
3329 * try to use its last block as our starting point.
3331 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3332 !isnullstartblock(ap
->prev
.br_startblock
) &&
3333 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3334 ap
->prev
.br_startblock
)) {
3335 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3337 * Adjust for the gap between prevp and us.
3339 adjust
= ap
->offset
-
3340 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3342 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3343 ap
->blkno
+= adjust
;
3346 * If not at eof, then compare the two neighbor blocks.
3347 * Figure out whether either one gives us a good starting point,
3348 * and pick the better one.
3350 else if (!ap
->eof
) {
3351 xfs_fsblock_t gotbno
; /* right side block number */
3352 xfs_fsblock_t gotdiff
=0; /* right side difference */
3353 xfs_fsblock_t prevbno
; /* left side block number */
3354 xfs_fsblock_t prevdiff
=0; /* left side difference */
3357 * If there's a previous (left) block, select a requested
3358 * start block based on it.
3360 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3361 !isnullstartblock(ap
->prev
.br_startblock
) &&
3362 (prevbno
= ap
->prev
.br_startblock
+
3363 ap
->prev
.br_blockcount
) &&
3364 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3366 * Calculate gap to end of previous block.
3368 adjust
= prevdiff
= ap
->offset
-
3369 (ap
->prev
.br_startoff
+
3370 ap
->prev
.br_blockcount
);
3372 * Figure the startblock based on the previous block's
3373 * end and the gap size.
3375 * If the gap is large relative to the piece we're
3376 * allocating, or using it gives us an invalid block
3377 * number, then just use the end of the previous block.
3379 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3380 ISVALID(prevbno
+ prevdiff
,
3381 ap
->prev
.br_startblock
))
3386 * If the firstblock forbids it, can't use it,
3389 if (!rt
&& !nullfb
&&
3390 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3391 prevbno
= NULLFSBLOCK
;
3394 * No previous block or can't follow it, just default.
3397 prevbno
= NULLFSBLOCK
;
3399 * If there's a following (right) block, select a requested
3400 * start block based on it.
3402 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3404 * Calculate gap to start of next block.
3406 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3408 * Figure the startblock based on the next block's
3409 * start and the gap size.
3411 gotbno
= ap
->got
.br_startblock
;
3414 * If the gap is large relative to the piece we're
3415 * allocating, or using it gives us an invalid block
3416 * number, then just use the start of the next block
3417 * offset by our length.
3419 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3420 ISVALID(gotbno
- gotdiff
, gotbno
))
3422 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3423 gotbno
-= ap
->length
;
3424 gotdiff
+= adjust
- ap
->length
;
3428 * If the firstblock forbids it, can't use it,
3431 if (!rt
&& !nullfb
&&
3432 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3433 gotbno
= NULLFSBLOCK
;
3436 * No next block, just default.
3439 gotbno
= NULLFSBLOCK
;
3441 * If both valid, pick the better one, else the only good
3442 * one, else ap->blkno is already set (to 0 or the inode block).
3444 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3445 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3446 else if (prevbno
!= NULLFSBLOCK
)
3447 ap
->blkno
= prevbno
;
3448 else if (gotbno
!= NULLFSBLOCK
)
3455 xfs_bmap_longest_free_extent(
3456 struct xfs_trans
*tp
,
3461 struct xfs_mount
*mp
= tp
->t_mountp
;
3462 struct xfs_perag
*pag
;
3463 xfs_extlen_t longest
;
3466 pag
= xfs_perag_get(mp
, ag
);
3467 if (!pag
->pagf_init
) {
3468 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3472 if (!pag
->pagf_init
) {
3478 longest
= xfs_alloc_longest_free_extent(mp
, pag
,
3479 xfs_alloc_min_freelist(mp
, pag
),
3480 xfs_ag_resv_needed(pag
, XFS_AG_RESV_NONE
));
3481 if (*blen
< longest
)
3490 xfs_bmap_select_minlen(
3491 struct xfs_bmalloca
*ap
,
3492 struct xfs_alloc_arg
*args
,
3496 if (notinit
|| *blen
< ap
->minlen
) {
3498 * Since we did a BUF_TRYLOCK above, it is possible that
3499 * there is space for this request.
3501 args
->minlen
= ap
->minlen
;
3502 } else if (*blen
< args
->maxlen
) {
3504 * If the best seen length is less than the request length,
3505 * use the best as the minimum.
3507 args
->minlen
= *blen
;
3510 * Otherwise we've seen an extent as big as maxlen, use that
3513 args
->minlen
= args
->maxlen
;
3518 xfs_bmap_btalloc_nullfb(
3519 struct xfs_bmalloca
*ap
,
3520 struct xfs_alloc_arg
*args
,
3523 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3524 xfs_agnumber_t ag
, startag
;
3528 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3529 args
->total
= ap
->total
;
3531 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3532 if (startag
== NULLAGNUMBER
)
3535 while (*blen
< args
->maxlen
) {
3536 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3541 if (++ag
== mp
->m_sb
.sb_agcount
)
3547 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3552 xfs_bmap_btalloc_filestreams(
3553 struct xfs_bmalloca
*ap
,
3554 struct xfs_alloc_arg
*args
,
3557 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3562 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3563 args
->total
= ap
->total
;
3565 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3566 if (ag
== NULLAGNUMBER
)
3569 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3573 if (*blen
< args
->maxlen
) {
3574 error
= xfs_filestream_new_ag(ap
, &ag
);
3578 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3585 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3588 * Set the failure fallback case to look in the selected AG as stream
3591 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3597 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3599 xfs_mount_t
*mp
; /* mount point structure */
3600 xfs_alloctype_t atype
= 0; /* type for allocation routines */
3601 xfs_extlen_t align
= 0; /* minimum allocation alignment */
3602 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3604 xfs_alloc_arg_t args
;
3606 xfs_extlen_t nextminlen
= 0;
3607 int nullfb
; /* true if ap->firstblock isn't set */
3615 mp
= ap
->ip
->i_mount
;
3617 /* stripe alignment for allocation is determined by mount parameters */
3619 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3620 stripe_align
= mp
->m_swidth
;
3621 else if (mp
->m_dalign
)
3622 stripe_align
= mp
->m_dalign
;
3624 if (ap
->flags
& XFS_BMAPI_COWFORK
)
3625 align
= xfs_get_cowextsz_hint(ap
->ip
);
3626 else if (xfs_alloc_is_userdata(ap
->datatype
))
3627 align
= xfs_get_extsz_hint(ap
->ip
);
3629 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
3630 align
, 0, ap
->eof
, 0, ap
->conv
,
3631 &ap
->offset
, &ap
->length
);
3637 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
3638 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
3640 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3641 xfs_inode_is_filestream(ap
->ip
)) {
3642 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3643 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3644 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3646 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3649 ap
->blkno
= *ap
->firstblock
;
3651 xfs_bmap_adjacent(ap
);
3654 * If allowed, use ap->blkno; otherwise must use firstblock since
3655 * it's in the right allocation group.
3657 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3660 ap
->blkno
= *ap
->firstblock
;
3662 * Normal allocation, done through xfs_alloc_vextent.
3664 tryagain
= isaligned
= 0;
3665 memset(&args
, 0, sizeof(args
));
3668 args
.fsbno
= ap
->blkno
;
3669 xfs_rmap_skip_owner_update(&args
.oinfo
);
3671 /* Trim the allocation back to the maximum an AG can fit. */
3672 args
.maxlen
= MIN(ap
->length
, mp
->m_ag_max_usable
);
3673 args
.firstblock
= *ap
->firstblock
;
3677 * Search for an allocation group with a single extent large
3678 * enough for the request. If one isn't found, then adjust
3679 * the minimum allocation size to the largest space found.
3681 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3682 xfs_inode_is_filestream(ap
->ip
))
3683 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3685 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3688 } else if (ap
->dfops
->dop_low
) {
3689 if (xfs_inode_is_filestream(ap
->ip
))
3690 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3692 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3693 args
.total
= args
.minlen
= ap
->minlen
;
3695 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3696 args
.total
= ap
->total
;
3697 args
.minlen
= ap
->minlen
;
3699 /* apply extent size hints if obtained earlier */
3702 if ((args
.mod
= (xfs_extlen_t
)do_mod(ap
->offset
, args
.prod
)))
3703 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3704 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_SIZE
) {
3708 args
.prod
= PAGE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3709 if ((args
.mod
= (xfs_extlen_t
)(do_mod(ap
->offset
, args
.prod
))))
3710 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
3713 * If we are not low on available data blocks, and the
3714 * underlying logical volume manager is a stripe, and
3715 * the file offset is zero then try to allocate data
3716 * blocks on stripe unit boundary.
3717 * NOTE: ap->aeof is only set if the allocation length
3718 * is >= the stripe unit and the allocation offset is
3719 * at the end of file.
3721 if (!ap
->dfops
->dop_low
&& ap
->aeof
) {
3723 args
.alignment
= stripe_align
;
3727 * Adjust for alignment
3729 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
3730 args
.minlen
= blen
- args
.alignment
;
3731 args
.minalignslop
= 0;
3734 * First try an exact bno allocation.
3735 * If it fails then do a near or start bno
3736 * allocation with alignment turned on.
3740 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3743 * Compute the minlen+alignment for the
3744 * next case. Set slop so that the value
3745 * of minlen+alignment+slop doesn't go up
3746 * between the calls.
3748 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3749 nextminlen
= blen
- stripe_align
;
3751 nextminlen
= args
.minlen
;
3752 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3754 nextminlen
+ stripe_align
-
3757 args
.minalignslop
= 0;
3761 args
.minalignslop
= 0;
3763 args
.minleft
= ap
->minleft
;
3764 args
.wasdel
= ap
->wasdel
;
3765 args
.resv
= XFS_AG_RESV_NONE
;
3766 args
.datatype
= ap
->datatype
;
3767 if (ap
->datatype
& XFS_ALLOC_USERDATA_ZERO
)
3770 error
= xfs_alloc_vextent(&args
);
3774 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3776 * Exact allocation failed. Now try with alignment
3780 args
.fsbno
= ap
->blkno
;
3781 args
.alignment
= stripe_align
;
3782 args
.minlen
= nextminlen
;
3783 args
.minalignslop
= 0;
3785 if ((error
= xfs_alloc_vextent(&args
)))
3788 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3790 * allocation failed, so turn off alignment and
3794 args
.fsbno
= ap
->blkno
;
3796 if ((error
= xfs_alloc_vextent(&args
)))
3799 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3800 args
.minlen
> ap
->minlen
) {
3801 args
.minlen
= ap
->minlen
;
3802 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3803 args
.fsbno
= ap
->blkno
;
3804 if ((error
= xfs_alloc_vextent(&args
)))
3807 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3809 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3810 args
.total
= ap
->minlen
;
3811 if ((error
= xfs_alloc_vextent(&args
)))
3813 ap
->dfops
->dop_low
= true;
3815 if (args
.fsbno
!= NULLFSBLOCK
) {
3817 * check the allocation happened at the same or higher AG than
3818 * the first block that was allocated.
3820 ASSERT(*ap
->firstblock
== NULLFSBLOCK
||
3821 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) <=
3822 XFS_FSB_TO_AGNO(mp
, args
.fsbno
));
3824 ap
->blkno
= args
.fsbno
;
3825 if (*ap
->firstblock
== NULLFSBLOCK
)
3826 *ap
->firstblock
= args
.fsbno
;
3827 ASSERT(nullfb
|| fb_agno
<= args
.agno
);
3828 ap
->length
= args
.len
;
3829 if (!(ap
->flags
& XFS_BMAPI_COWFORK
))
3830 ap
->ip
->i_d
.di_nblocks
+= args
.len
;
3831 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3833 ap
->ip
->i_delayed_blks
-= args
.len
;
3835 * Adjust the disk quota also. This was reserved
3838 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3839 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
:
3840 XFS_TRANS_DQ_BCOUNT
,
3843 ap
->blkno
= NULLFSBLOCK
;
3850 * For a remap operation, just "allocate" an extent at the address that the
3851 * caller passed in, and ensure that the AGFL is the right size. The caller
3852 * will then map the "allocated" extent into the file somewhere.
3855 xfs_bmap_remap_alloc(
3856 struct xfs_trans
*tp
,
3857 struct xfs_inode
*ip
,
3858 xfs_fsblock_t startblock
,
3859 xfs_extlen_t length
)
3861 struct xfs_mount
*mp
= tp
->t_mountp
;
3862 struct xfs_alloc_arg args
;
3866 * validate that the block number is legal - the enables us to detect
3867 * and handle a silent filesystem corruption rather than crashing.
3869 memset(&args
, 0, sizeof(struct xfs_alloc_arg
));
3872 args
.agno
= XFS_FSB_TO_AGNO(mp
, startblock
);
3873 args
.agbno
= XFS_FSB_TO_AGBNO(mp
, startblock
);
3875 if (args
.agno
>= mp
->m_sb
.sb_agcount
||
3876 args
.agbno
>= mp
->m_sb
.sb_agblocks
)
3877 return -EFSCORRUPTED
;
3879 /* "Allocate" the extent from the range we passed in. */
3880 trace_xfs_bmap_remap_alloc(ip
, startblock
, length
);
3882 ip
->i_d
.di_nblocks
+= length
;
3883 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3885 /* Fix the freelist, like a real allocator does. */
3886 args
.datatype
= XFS_ALLOC_USERDATA
| XFS_ALLOC_NOBUSY
;
3887 args
.pag
= xfs_perag_get(mp
, args
.agno
);
3891 * The freelist fixing code will decline the allocation if
3892 * the size and shape of the free space doesn't allow for
3893 * allocating the extent and updating all the metadata that
3894 * happens during an allocation. We're remapping, not
3895 * allocating, so skip that check by pretending to be freeing.
3897 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
3898 xfs_perag_put(args
.pag
);
3900 trace_xfs_bmap_remap_alloc_error(ip
, error
, _RET_IP_
);
3905 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3906 * It figures out where to ask the underlying allocator to put the new extent.
3910 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3912 if (XFS_IS_REALTIME_INODE(ap
->ip
) &&
3913 xfs_alloc_is_userdata(ap
->datatype
))
3914 return xfs_bmap_rtalloc(ap
);
3915 return xfs_bmap_btalloc(ap
);
3918 /* Trim extent to fit a logical block range. */
3921 struct xfs_bmbt_irec
*irec
,
3925 xfs_fileoff_t distance
;
3926 xfs_fileoff_t end
= bno
+ len
;
3928 if (irec
->br_startoff
+ irec
->br_blockcount
<= bno
||
3929 irec
->br_startoff
>= end
) {
3930 irec
->br_blockcount
= 0;
3934 if (irec
->br_startoff
< bno
) {
3935 distance
= bno
- irec
->br_startoff
;
3936 if (isnullstartblock(irec
->br_startblock
))
3937 irec
->br_startblock
= DELAYSTARTBLOCK
;
3938 if (irec
->br_startblock
!= DELAYSTARTBLOCK
&&
3939 irec
->br_startblock
!= HOLESTARTBLOCK
)
3940 irec
->br_startblock
+= distance
;
3941 irec
->br_startoff
+= distance
;
3942 irec
->br_blockcount
-= distance
;
3945 if (end
< irec
->br_startoff
+ irec
->br_blockcount
) {
3946 distance
= irec
->br_startoff
+ irec
->br_blockcount
- end
;
3947 irec
->br_blockcount
-= distance
;
3952 * Trim the returned map to the required bounds
3956 struct xfs_bmbt_irec
*mval
,
3957 struct xfs_bmbt_irec
*got
,
3965 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3966 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3968 if (isnullstartblock(got
->br_startblock
))
3969 mval
->br_startblock
= DELAYSTARTBLOCK
;
3975 ASSERT((*bno
>= obno
) || (n
== 0));
3977 mval
->br_startoff
= *bno
;
3978 if (isnullstartblock(got
->br_startblock
))
3979 mval
->br_startblock
= DELAYSTARTBLOCK
;
3981 mval
->br_startblock
= got
->br_startblock
+
3982 (*bno
- got
->br_startoff
);
3984 * Return the minimum of what we got and what we asked for for
3985 * the length. We can use the len variable here because it is
3986 * modified below and we could have been there before coming
3987 * here if the first part of the allocation didn't overlap what
3990 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
3991 got
->br_blockcount
- (*bno
- got
->br_startoff
));
3992 mval
->br_state
= got
->br_state
;
3993 ASSERT(mval
->br_blockcount
<= len
);
3998 * Update and validate the extent map to return
4001 xfs_bmapi_update_map(
4002 struct xfs_bmbt_irec
**map
,
4010 xfs_bmbt_irec_t
*mval
= *map
;
4012 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
4013 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
4014 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
4015 (mval
->br_startoff
< obno
));
4017 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
4019 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
4020 /* update previous map with new information */
4021 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
4022 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
4023 ASSERT(mval
->br_state
== mval
[-1].br_state
);
4024 mval
[-1].br_blockcount
= mval
->br_blockcount
;
4025 mval
[-1].br_state
= mval
->br_state
;
4026 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
4027 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
4028 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
4029 mval
->br_startblock
== mval
[-1].br_startblock
+
4030 mval
[-1].br_blockcount
&&
4031 ((flags
& XFS_BMAPI_IGSTATE
) ||
4032 mval
[-1].br_state
== mval
->br_state
)) {
4033 ASSERT(mval
->br_startoff
==
4034 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
4035 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4036 } else if (*n
> 0 &&
4037 mval
->br_startblock
== DELAYSTARTBLOCK
&&
4038 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
4039 mval
->br_startoff
==
4040 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
4041 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4042 mval
[-1].br_state
= mval
->br_state
;
4043 } else if (!((*n
== 0) &&
4044 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
4053 * Map file blocks to filesystem blocks without allocation.
4057 struct xfs_inode
*ip
,
4060 struct xfs_bmbt_irec
*mval
,
4064 struct xfs_mount
*mp
= ip
->i_mount
;
4065 struct xfs_ifork
*ifp
;
4066 struct xfs_bmbt_irec got
;
4073 int whichfork
= xfs_bmapi_whichfork(flags
);
4076 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
4077 XFS_BMAPI_IGSTATE
|XFS_BMAPI_COWFORK
)));
4078 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
4080 if (unlikely(XFS_TEST_ERROR(
4081 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4082 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4083 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4084 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
4085 return -EFSCORRUPTED
;
4088 if (XFS_FORCED_SHUTDOWN(mp
))
4091 XFS_STATS_INC(mp
, xs_blk_mapr
);
4093 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4095 /* No CoW fork? Return a hole. */
4096 if (whichfork
== XFS_COW_FORK
&& !ifp
) {
4097 mval
->br_startoff
= bno
;
4098 mval
->br_startblock
= HOLESTARTBLOCK
;
4099 mval
->br_blockcount
= len
;
4100 mval
->br_state
= XFS_EXT_NORM
;
4105 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4106 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
4111 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &idx
, &got
))
4116 while (bno
< end
&& n
< *nmap
) {
4117 /* Reading past eof, act as though there's a hole up to end. */
4119 got
.br_startoff
= end
;
4120 if (got
.br_startoff
> bno
) {
4121 /* Reading in a hole. */
4122 mval
->br_startoff
= bno
;
4123 mval
->br_startblock
= HOLESTARTBLOCK
;
4124 mval
->br_blockcount
=
4125 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4126 mval
->br_state
= XFS_EXT_NORM
;
4127 bno
+= mval
->br_blockcount
;
4128 len
-= mval
->br_blockcount
;
4134 /* set up the extent map to return. */
4135 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4136 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4138 /* If we're done, stop now. */
4139 if (bno
>= end
|| n
>= *nmap
)
4142 /* Else go on to the next record. */
4143 if (!xfs_iext_get_extent(ifp
, ++idx
, &got
))
4151 * Add a delayed allocation extent to an inode. Blocks are reserved from the
4152 * global pool and the extent inserted into the inode in-core extent tree.
4154 * On entry, got refers to the first extent beyond the offset of the extent to
4155 * allocate or eof is specified if no such extent exists. On return, got refers
4156 * to the extent record that was inserted to the inode fork.
4158 * Note that the allocated extent may have been merged with contiguous extents
4159 * during insertion into the inode fork. Thus, got does not reflect the current
4160 * state of the inode fork on return. If necessary, the caller can use lastx to
4161 * look up the updated record in the inode fork.
4164 xfs_bmapi_reserve_delalloc(
4165 struct xfs_inode
*ip
,
4169 xfs_filblks_t prealloc
,
4170 struct xfs_bmbt_irec
*got
,
4171 xfs_extnum_t
*lastx
,
4174 struct xfs_mount
*mp
= ip
->i_mount
;
4175 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4177 xfs_extlen_t indlen
;
4178 char rt
= XFS_IS_REALTIME_INODE(ip
);
4181 xfs_fileoff_t aoff
= off
;
4184 * Cap the alloc length. Keep track of prealloc so we know whether to
4185 * tag the inode before we return.
4187 alen
= XFS_FILBLKS_MIN(len
+ prealloc
, MAXEXTLEN
);
4189 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4190 if (prealloc
&& alen
>= len
)
4191 prealloc
= alen
- len
;
4193 /* Figure out the extent size, adjust alen */
4194 if (whichfork
== XFS_COW_FORK
)
4195 extsz
= xfs_get_cowextsz_hint(ip
);
4197 extsz
= xfs_get_extsz_hint(ip
);
4199 struct xfs_bmbt_irec prev
;
4201 if (!xfs_iext_get_extent(ifp
, *lastx
- 1, &prev
))
4202 prev
.br_startoff
= NULLFILEOFF
;
4204 error
= xfs_bmap_extsize_align(mp
, got
, &prev
, extsz
, rt
, eof
,
4205 1, 0, &aoff
, &alen
);
4210 extsz
= alen
/ mp
->m_sb
.sb_rextsize
;
4213 * Make a transaction-less quota reservation for delayed allocation
4214 * blocks. This number gets adjusted later. We return if we haven't
4215 * allocated blocks already inside this loop.
4217 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4218 rt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4223 * Split changing sb for alen and indlen since they could be coming
4224 * from different places.
4226 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4230 error
= xfs_mod_frextents(mp
, -((int64_t)extsz
));
4232 error
= xfs_mod_fdblocks(mp
, -((int64_t)alen
), false);
4236 goto out_unreserve_quota
;
4238 error
= xfs_mod_fdblocks(mp
, -((int64_t)indlen
), false);
4240 goto out_unreserve_blocks
;
4243 ip
->i_delayed_blks
+= alen
;
4245 got
->br_startoff
= aoff
;
4246 got
->br_startblock
= nullstartblock(indlen
);
4247 got
->br_blockcount
= alen
;
4248 got
->br_state
= XFS_EXT_NORM
;
4250 xfs_bmap_add_extent_hole_delay(ip
, whichfork
, lastx
, got
);
4253 * Tag the inode if blocks were preallocated. Note that COW fork
4254 * preallocation can occur at the start or end of the extent, even when
4255 * prealloc == 0, so we must also check the aligned offset and length.
4257 if (whichfork
== XFS_DATA_FORK
&& prealloc
)
4258 xfs_inode_set_eofblocks_tag(ip
);
4259 if (whichfork
== XFS_COW_FORK
&& (prealloc
|| aoff
< off
|| alen
> len
))
4260 xfs_inode_set_cowblocks_tag(ip
);
4264 out_unreserve_blocks
:
4266 xfs_mod_frextents(mp
, extsz
);
4268 xfs_mod_fdblocks(mp
, alen
, false);
4269 out_unreserve_quota
:
4270 if (XFS_IS_QUOTA_ON(mp
))
4271 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0, rt
?
4272 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4278 struct xfs_bmalloca
*bma
)
4280 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4281 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4282 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4283 int tmp_logflags
= 0;
4286 ASSERT(bma
->length
> 0);
4289 * For the wasdelay case, we could also just allocate the stuff asked
4290 * for in this bmap call but that wouldn't be as good.
4293 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4294 bma
->offset
= bma
->got
.br_startoff
;
4296 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
4300 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4302 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4303 bma
->got
.br_startoff
- bma
->offset
);
4307 * Set the data type being allocated. For the data fork, the first data
4308 * in the file is treated differently to all other allocations. For the
4309 * attribute fork, we only need to ensure the allocated range is not on
4312 if (!(bma
->flags
& XFS_BMAPI_METADATA
)) {
4313 bma
->datatype
= XFS_ALLOC_NOBUSY
;
4314 if (whichfork
== XFS_DATA_FORK
) {
4315 if (bma
->offset
== 0)
4316 bma
->datatype
|= XFS_ALLOC_INITIAL_USER_DATA
;
4318 bma
->datatype
|= XFS_ALLOC_USERDATA
;
4320 if (bma
->flags
& XFS_BMAPI_ZERO
)
4321 bma
->datatype
|= XFS_ALLOC_USERDATA_ZERO
;
4324 bma
->minlen
= (bma
->flags
& XFS_BMAPI_CONTIG
) ? bma
->length
: 1;
4327 * Only want to do the alignment at the eof if it is userdata and
4328 * allocation length is larger than a stripe unit.
4330 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
&&
4331 !(bma
->flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4332 error
= xfs_bmap_isaeof(bma
, whichfork
);
4337 error
= xfs_bmap_alloc(bma
);
4342 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4343 if (bma
->blkno
== NULLFSBLOCK
)
4345 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4346 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4347 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4348 bma
->cur
->bc_private
.b
.dfops
= bma
->dfops
;
4351 * Bump the number of extents we've allocated
4357 bma
->cur
->bc_private
.b
.flags
=
4358 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4360 bma
->got
.br_startoff
= bma
->offset
;
4361 bma
->got
.br_startblock
= bma
->blkno
;
4362 bma
->got
.br_blockcount
= bma
->length
;
4363 bma
->got
.br_state
= XFS_EXT_NORM
;
4366 * In the data fork, a wasdelay extent has been initialized, so
4367 * shouldn't be flagged as unwritten.
4369 * For the cow fork, however, we convert delalloc reservations
4370 * (extents allocated for speculative preallocation) to
4371 * allocated unwritten extents, and only convert the unwritten
4372 * extents to real extents when we're about to write the data.
4374 if ((!bma
->wasdel
|| (bma
->flags
& XFS_BMAPI_COWFORK
)) &&
4375 (bma
->flags
& XFS_BMAPI_PREALLOC
) &&
4376 xfs_sb_version_hasextflgbit(&mp
->m_sb
))
4377 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4380 error
= xfs_bmap_add_extent_delay_real(bma
, whichfork
);
4382 error
= xfs_bmap_add_extent_hole_real(bma
->tp
, bma
->ip
,
4383 whichfork
, &bma
->idx
, &bma
->cur
, &bma
->got
,
4384 bma
->firstblock
, bma
->dfops
, &bma
->logflags
);
4386 bma
->logflags
|= tmp_logflags
;
4391 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4392 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4393 * the neighbouring ones.
4395 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4397 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4398 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4399 bma
->offset
+ bma
->length
);
4400 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4401 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4406 xfs_bmapi_convert_unwritten(
4407 struct xfs_bmalloca
*bma
,
4408 struct xfs_bmbt_irec
*mval
,
4412 int whichfork
= xfs_bmapi_whichfork(flags
);
4413 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4414 int tmp_logflags
= 0;
4417 /* check if we need to do unwritten->real conversion */
4418 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4419 (flags
& XFS_BMAPI_PREALLOC
))
4422 /* check if we need to do real->unwritten conversion */
4423 if (mval
->br_state
== XFS_EXT_NORM
&&
4424 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4425 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4429 * Modify (by adding) the state flag, if writing.
4431 ASSERT(mval
->br_blockcount
<= len
);
4432 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4433 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4434 bma
->ip
, whichfork
);
4435 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4436 bma
->cur
->bc_private
.b
.dfops
= bma
->dfops
;
4438 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4439 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4442 * Before insertion into the bmbt, zero the range being converted
4445 if (flags
& XFS_BMAPI_ZERO
) {
4446 error
= xfs_zero_extent(bma
->ip
, mval
->br_startblock
,
4447 mval
->br_blockcount
);
4452 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, whichfork
,
4453 &bma
->idx
, &bma
->cur
, mval
, bma
->firstblock
, bma
->dfops
,
4456 * Log the inode core unconditionally in the unwritten extent conversion
4457 * path because the conversion might not have done so (e.g., if the
4458 * extent count hasn't changed). We need to make sure the inode is dirty
4459 * in the transaction for the sake of fsync(), even if nothing has
4460 * changed, because fsync() will not force the log for this transaction
4461 * unless it sees the inode pinned.
4463 * Note: If we're only converting cow fork extents, there aren't
4464 * any on-disk updates to make, so we don't need to log anything.
4466 if (whichfork
!= XFS_COW_FORK
)
4467 bma
->logflags
|= tmp_logflags
| XFS_ILOG_CORE
;
4472 * Update our extent pointer, given that
4473 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4474 * of the neighbouring ones.
4476 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4479 * We may have combined previously unwritten space with written space,
4480 * so generate another request.
4482 if (mval
->br_blockcount
< len
)
4488 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4489 * extent state if necessary. Details behaviour is controlled by the flags
4490 * parameter. Only allocates blocks from a single allocation group, to avoid
4493 * The returned value in "firstblock" from the first call in a transaction
4494 * must be remembered and presented to subsequent calls in "firstblock".
4495 * An upper bound for the number of blocks to be allocated is supplied to
4496 * the first call in "total"; if no allocation group has that many free
4497 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4501 struct xfs_trans
*tp
, /* transaction pointer */
4502 struct xfs_inode
*ip
, /* incore inode */
4503 xfs_fileoff_t bno
, /* starting file offs. mapped */
4504 xfs_filblks_t len
, /* length to map in file */
4505 int flags
, /* XFS_BMAPI_... */
4506 xfs_fsblock_t
*firstblock
, /* first allocated block
4507 controls a.g. for allocs */
4508 xfs_extlen_t total
, /* total blocks needed */
4509 struct xfs_bmbt_irec
*mval
, /* output: map values */
4510 int *nmap
, /* i/o: mval size/count */
4511 struct xfs_defer_ops
*dfops
) /* i/o: list extents to free */
4513 struct xfs_mount
*mp
= ip
->i_mount
;
4514 struct xfs_ifork
*ifp
;
4515 struct xfs_bmalloca bma
= { NULL
}; /* args for xfs_bmap_alloc */
4516 xfs_fileoff_t end
; /* end of mapped file region */
4517 bool eof
= false; /* after the end of extents */
4518 int error
; /* error return */
4519 int n
; /* current extent index */
4520 xfs_fileoff_t obno
; /* old block number (offset) */
4521 int whichfork
; /* data or attr fork */
4524 xfs_fileoff_t orig_bno
; /* original block number value */
4525 int orig_flags
; /* original flags arg value */
4526 xfs_filblks_t orig_len
; /* original value of len arg */
4527 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4528 int orig_nmap
; /* original value of *nmap */
4536 whichfork
= xfs_bmapi_whichfork(flags
);
4539 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4540 ASSERT(!(flags
& XFS_BMAPI_IGSTATE
));
4541 ASSERT(tp
!= NULL
||
4542 (flags
& (XFS_BMAPI_CONVERT
| XFS_BMAPI_COWFORK
)) ==
4543 (XFS_BMAPI_CONVERT
| XFS_BMAPI_COWFORK
));
4545 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
);
4546 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4547 ASSERT(!(flags
& XFS_BMAPI_REMAP
));
4549 /* zeroing is for currently only for data extents, not metadata */
4550 ASSERT((flags
& (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
)) !=
4551 (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
));
4553 * we can allocate unwritten extents or pre-zero allocated blocks,
4554 * but it makes no sense to do both at once. This would result in
4555 * zeroing the unwritten extent twice, but it still being an
4556 * unwritten extent....
4558 ASSERT((flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
)) !=
4559 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
));
4561 if (unlikely(XFS_TEST_ERROR(
4562 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4563 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4564 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4565 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW
, mp
);
4566 return -EFSCORRUPTED
;
4569 if (XFS_FORCED_SHUTDOWN(mp
))
4572 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4574 XFS_STATS_INC(mp
, xs_blk_mapw
);
4576 if (*firstblock
== NULLFSBLOCK
) {
4577 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
)
4578 bma
.minleft
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4585 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4586 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4595 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &bma
.idx
, &bma
.got
))
4597 if (!xfs_iext_get_extent(ifp
, bma
.idx
- 1, &bma
.prev
))
4598 bma
.prev
.br_startoff
= NULLFILEOFF
;
4604 bma
.firstblock
= firstblock
;
4606 while (bno
< end
&& n
< *nmap
) {
4607 bool need_alloc
= false, wasdelay
= false;
4609 /* in hole or beyoned EOF? */
4610 if (eof
|| bma
.got
.br_startoff
> bno
) {
4611 if (flags
& XFS_BMAPI_DELALLOC
) {
4613 * For the COW fork we can reasonably get a
4614 * request for converting an extent that races
4615 * with other threads already having converted
4616 * part of it, as there converting COW to
4617 * regular blocks is not protected using the
4620 ASSERT(flags
& XFS_BMAPI_COWFORK
);
4621 if (!(flags
& XFS_BMAPI_COWFORK
)) {
4626 if (eof
|| bno
>= end
)
4631 } else if (isnullstartblock(bma
.got
.br_startblock
)) {
4636 * First, deal with the hole before the allocated space
4637 * that we found, if any.
4639 if (need_alloc
|| wasdelay
) {
4641 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4642 bma
.wasdel
= wasdelay
;
4647 * There's a 32/64 bit type mismatch between the
4648 * allocation length request (which can be 64 bits in
4649 * length) and the bma length request, which is
4650 * xfs_extlen_t and therefore 32 bits. Hence we have to
4651 * check for 32-bit overflows and handle them here.
4653 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4654 bma
.length
= MAXEXTLEN
;
4659 ASSERT(bma
.length
> 0);
4660 error
= xfs_bmapi_allocate(&bma
);
4663 if (bma
.blkno
== NULLFSBLOCK
)
4667 * If this is a CoW allocation, record the data in
4668 * the refcount btree for orphan recovery.
4670 if (whichfork
== XFS_COW_FORK
) {
4671 error
= xfs_refcount_alloc_cow_extent(mp
, dfops
,
4672 bma
.blkno
, bma
.length
);
4678 /* Deal with the allocated space we found. */
4679 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4682 /* Execute unwritten extent conversion if necessary */
4683 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4684 if (error
== -EAGAIN
)
4689 /* update the extent map to return */
4690 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4693 * If we're done, stop now. Stop when we've allocated
4694 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4695 * the transaction may get too big.
4697 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4700 /* Else go on to the next record. */
4702 if (!xfs_iext_get_extent(ifp
, ++bma
.idx
, &bma
.got
))
4708 * Transform from btree to extents, give it cur.
4710 if (xfs_bmap_wants_extents(ip
, whichfork
)) {
4711 int tmp_logflags
= 0;
4714 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
,
4715 &tmp_logflags
, whichfork
);
4716 bma
.logflags
|= tmp_logflags
;
4721 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
4722 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
4723 XFS_IFORK_MAXEXT(ip
, whichfork
));
4727 * Log everything. Do this after conversion, there's no point in
4728 * logging the extent records if we've converted to btree format.
4730 if ((bma
.logflags
& xfs_ilog_fext(whichfork
)) &&
4731 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4732 bma
.logflags
&= ~xfs_ilog_fext(whichfork
);
4733 else if ((bma
.logflags
& xfs_ilog_fbroot(whichfork
)) &&
4734 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
4735 bma
.logflags
&= ~xfs_ilog_fbroot(whichfork
);
4737 * Log whatever the flags say, even if error. Otherwise we might miss
4738 * detecting a case where the data is changed, there's an error,
4739 * and it's not logged so we don't shutdown when we should.
4742 xfs_trans_log_inode(tp
, ip
, bma
.logflags
);
4746 ASSERT(*firstblock
== NULLFSBLOCK
||
4747 XFS_FSB_TO_AGNO(mp
, *firstblock
) <=
4749 bma
.cur
->bc_private
.b
.firstblock
));
4750 *firstblock
= bma
.cur
->bc_private
.b
.firstblock
;
4752 xfs_btree_del_cursor(bma
.cur
,
4753 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
4756 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4763 struct xfs_trans
*tp
,
4764 struct xfs_inode
*ip
,
4767 xfs_fsblock_t startblock
,
4768 struct xfs_defer_ops
*dfops
)
4770 struct xfs_mount
*mp
= ip
->i_mount
;
4771 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4772 struct xfs_btree_cur
*cur
= NULL
;
4773 xfs_fsblock_t firstblock
= NULLFSBLOCK
;
4774 struct xfs_bmbt_irec got
;
4776 int logflags
= 0, error
;
4779 ASSERT(len
<= (xfs_filblks_t
)MAXEXTLEN
);
4780 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4782 if (unlikely(XFS_TEST_ERROR(
4783 (XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_EXTENTS
&&
4784 XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_BTREE
),
4785 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4786 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW
, mp
);
4787 return -EFSCORRUPTED
;
4790 if (XFS_FORCED_SHUTDOWN(mp
))
4793 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4794 error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
);
4799 if (xfs_iext_lookup_extent(ip
, ifp
, bno
, &idx
, &got
)) {
4800 /* make sure we only reflink into a hole. */
4801 ASSERT(got
.br_startoff
> bno
);
4802 ASSERT(got
.br_startoff
- bno
>= len
);
4805 error
= xfs_bmap_remap_alloc(tp
, ip
, startblock
, len
);
4809 if (ifp
->if_flags
& XFS_IFBROOT
) {
4810 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
4811 cur
->bc_private
.b
.firstblock
= firstblock
;
4812 cur
->bc_private
.b
.dfops
= dfops
;
4813 cur
->bc_private
.b
.flags
= 0;
4816 got
.br_startoff
= bno
;
4817 got
.br_startblock
= startblock
;
4818 got
.br_blockcount
= len
;
4819 got
.br_state
= XFS_EXT_NORM
;
4821 error
= xfs_bmap_add_extent_hole_real(tp
, ip
, XFS_DATA_FORK
, &idx
, &cur
,
4822 &got
, &firstblock
, dfops
, &logflags
);
4826 if (xfs_bmap_wants_extents(ip
, XFS_DATA_FORK
)) {
4827 int tmp_logflags
= 0;
4829 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
,
4830 &tmp_logflags
, XFS_DATA_FORK
);
4831 logflags
|= tmp_logflags
;
4835 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
)
4836 logflags
&= ~XFS_ILOG_DEXT
;
4837 else if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
)
4838 logflags
&= ~XFS_ILOG_DBROOT
;
4841 xfs_trans_log_inode(tp
, ip
, logflags
);
4843 xfs_btree_del_cursor(cur
,
4844 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
4850 * When a delalloc extent is split (e.g., due to a hole punch), the original
4851 * indlen reservation must be shared across the two new extents that are left
4854 * Given the original reservation and the worst case indlen for the two new
4855 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4856 * reservation fairly across the two new extents. If necessary, steal available
4857 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4858 * ores == 1). The number of stolen blocks is returned. The availability and
4859 * subsequent accounting of stolen blocks is the responsibility of the caller.
4861 static xfs_filblks_t
4862 xfs_bmap_split_indlen(
4863 xfs_filblks_t ores
, /* original res. */
4864 xfs_filblks_t
*indlen1
, /* ext1 worst indlen */
4865 xfs_filblks_t
*indlen2
, /* ext2 worst indlen */
4866 xfs_filblks_t avail
) /* stealable blocks */
4868 xfs_filblks_t len1
= *indlen1
;
4869 xfs_filblks_t len2
= *indlen2
;
4870 xfs_filblks_t nres
= len1
+ len2
; /* new total res. */
4871 xfs_filblks_t stolen
= 0;
4872 xfs_filblks_t resfactor
;
4875 * Steal as many blocks as we can to try and satisfy the worst case
4876 * indlen for both new extents.
4878 if (ores
< nres
&& avail
)
4879 stolen
= XFS_FILBLKS_MIN(nres
- ores
, avail
);
4882 /* nothing else to do if we've satisfied the new reservation */
4887 * We can't meet the total required reservation for the two extents.
4888 * Calculate the percent of the overall shortage between both extents
4889 * and apply this percentage to each of the requested indlen values.
4890 * This distributes the shortage fairly and reduces the chances that one
4891 * of the two extents is left with nothing when extents are repeatedly
4894 resfactor
= (ores
* 100);
4895 do_div(resfactor
, nres
);
4900 ASSERT(len1
+ len2
<= ores
);
4901 ASSERT(len1
< *indlen1
&& len2
< *indlen2
);
4904 * Hand out the remainder to each extent. If one of the two reservations
4905 * is zero, we want to make sure that one gets a block first. The loop
4906 * below starts with len1, so hand len2 a block right off the bat if it
4909 ores
-= (len1
+ len2
);
4910 ASSERT((*indlen1
- len1
) + (*indlen2
- len2
) >= ores
);
4911 if (ores
&& !len2
&& *indlen2
) {
4916 if (len1
< *indlen1
) {
4922 if (len2
< *indlen2
) {
4935 xfs_bmap_del_extent_delay(
4936 struct xfs_inode
*ip
,
4939 struct xfs_bmbt_irec
*got
,
4940 struct xfs_bmbt_irec
*del
)
4942 struct xfs_mount
*mp
= ip
->i_mount
;
4943 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4944 struct xfs_bmbt_irec
new;
4945 int64_t da_old
, da_new
, da_diff
= 0;
4946 xfs_fileoff_t del_endoff
, got_endoff
;
4947 xfs_filblks_t got_indlen
, new_indlen
, stolen
;
4948 int error
= 0, state
= 0;
4951 XFS_STATS_INC(mp
, xs_del_exlist
);
4953 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4954 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4955 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4956 da_old
= startblockval(got
->br_startblock
);
4960 ASSERT(*idx
<= xfs_iext_count(ifp
));
4961 ASSERT(del
->br_blockcount
> 0);
4962 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4963 ASSERT(got_endoff
>= del_endoff
);
4966 int64_t rtexts
= XFS_FSB_TO_B(mp
, del
->br_blockcount
);
4968 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
4969 xfs_mod_frextents(mp
, rtexts
);
4973 * Update the inode delalloc counter now and wait to update the
4974 * sb counters as we might have to borrow some blocks for the
4975 * indirect block accounting.
4977 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
,
4978 -((long)del
->br_blockcount
), 0,
4979 isrt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4982 ip
->i_delayed_blks
-= del
->br_blockcount
;
4984 if (whichfork
== XFS_COW_FORK
)
4985 state
|= BMAP_COWFORK
;
4987 if (got
->br_startoff
== del
->br_startoff
)
4988 state
|= BMAP_LEFT_CONTIG
;
4989 if (got_endoff
== del_endoff
)
4990 state
|= BMAP_RIGHT_CONTIG
;
4992 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
4993 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
4995 * Matches the whole extent. Delete the entry.
4997 xfs_iext_remove(ip
, *idx
, 1, state
);
5000 case BMAP_LEFT_CONTIG
:
5002 * Deleting the first part of the extent.
5004 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5005 got
->br_startoff
= del_endoff
;
5006 got
->br_blockcount
-= del
->br_blockcount
;
5007 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
5008 got
->br_blockcount
), da_old
);
5009 got
->br_startblock
= nullstartblock((int)da_new
);
5010 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5011 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5013 case BMAP_RIGHT_CONTIG
:
5015 * Deleting the last part of the extent.
5017 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5018 got
->br_blockcount
= got
->br_blockcount
- del
->br_blockcount
;
5019 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
5020 got
->br_blockcount
), da_old
);
5021 got
->br_startblock
= nullstartblock((int)da_new
);
5022 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5023 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5027 * Deleting the middle of the extent.
5029 * Distribute the original indlen reservation across the two new
5030 * extents. Steal blocks from the deleted extent if necessary.
5031 * Stealing blocks simply fudges the fdblocks accounting below.
5032 * Warn if either of the new indlen reservations is zero as this
5033 * can lead to delalloc problems.
5035 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5037 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
5038 got_indlen
= xfs_bmap_worst_indlen(ip
, got
->br_blockcount
);
5040 new.br_blockcount
= got_endoff
- del_endoff
;
5041 new_indlen
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
5043 WARN_ON_ONCE(!got_indlen
|| !new_indlen
);
5044 stolen
= xfs_bmap_split_indlen(da_old
, &got_indlen
, &new_indlen
,
5045 del
->br_blockcount
);
5047 got
->br_startblock
= nullstartblock((int)got_indlen
);
5048 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5049 trace_xfs_bmap_post_update(ip
, *idx
, 0, _THIS_IP_
);
5051 new.br_startoff
= del_endoff
;
5052 new.br_state
= got
->br_state
;
5053 new.br_startblock
= nullstartblock((int)new_indlen
);
5056 xfs_iext_insert(ip
, *idx
, 1, &new, state
);
5058 da_new
= got_indlen
+ new_indlen
- stolen
;
5059 del
->br_blockcount
-= stolen
;
5063 ASSERT(da_old
>= da_new
);
5064 da_diff
= da_old
- da_new
;
5066 da_diff
+= del
->br_blockcount
;
5068 xfs_mod_fdblocks(mp
, da_diff
, false);
5073 xfs_bmap_del_extent_cow(
5074 struct xfs_inode
*ip
,
5076 struct xfs_bmbt_irec
*got
,
5077 struct xfs_bmbt_irec
*del
)
5079 struct xfs_mount
*mp
= ip
->i_mount
;
5080 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
5081 struct xfs_bmbt_irec
new;
5082 xfs_fileoff_t del_endoff
, got_endoff
;
5083 int state
= BMAP_COWFORK
;
5085 XFS_STATS_INC(mp
, xs_del_exlist
);
5087 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5088 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
5091 ASSERT(*idx
<= xfs_iext_count(ifp
));
5092 ASSERT(del
->br_blockcount
> 0);
5093 ASSERT(got
->br_startoff
<= del
->br_startoff
);
5094 ASSERT(got_endoff
>= del_endoff
);
5095 ASSERT(!isnullstartblock(got
->br_startblock
));
5097 if (got
->br_startoff
== del
->br_startoff
)
5098 state
|= BMAP_LEFT_CONTIG
;
5099 if (got_endoff
== del_endoff
)
5100 state
|= BMAP_RIGHT_CONTIG
;
5102 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
5103 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
5105 * Matches the whole extent. Delete the entry.
5107 xfs_iext_remove(ip
, *idx
, 1, state
);
5110 case BMAP_LEFT_CONTIG
:
5112 * Deleting the first part of the extent.
5114 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5115 got
->br_startoff
= del_endoff
;
5116 got
->br_blockcount
-= del
->br_blockcount
;
5117 got
->br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5118 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5119 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5121 case BMAP_RIGHT_CONTIG
:
5123 * Deleting the last part of the extent.
5125 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5126 got
->br_blockcount
-= del
->br_blockcount
;
5127 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5128 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5132 * Deleting the middle of the extent.
5134 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5135 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
5136 xfs_bmbt_set_all(xfs_iext_get_ext(ifp
, *idx
), got
);
5137 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5139 new.br_startoff
= del_endoff
;
5140 new.br_blockcount
= got_endoff
- del_endoff
;
5141 new.br_state
= got
->br_state
;
5142 new.br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5145 xfs_iext_insert(ip
, *idx
, 1, &new, state
);
5151 * Called by xfs_bmapi to update file extent records and the btree
5152 * after removing space (or undoing a delayed allocation).
5154 STATIC
int /* error */
5155 xfs_bmap_del_extent(
5156 xfs_inode_t
*ip
, /* incore inode pointer */
5157 xfs_trans_t
*tp
, /* current transaction pointer */
5158 xfs_extnum_t
*idx
, /* extent number to update/delete */
5159 struct xfs_defer_ops
*dfops
, /* list of extents to be freed */
5160 xfs_btree_cur_t
*cur
, /* if null, not a btree */
5161 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
5162 int *logflagsp
, /* inode logging flags */
5163 int whichfork
, /* data or attr fork */
5164 int bflags
) /* bmapi flags */
5166 xfs_filblks_t da_new
; /* new delay-alloc indirect blocks */
5167 xfs_filblks_t da_old
; /* old delay-alloc indirect blocks */
5168 xfs_fsblock_t del_endblock
=0; /* first block past del */
5169 xfs_fileoff_t del_endoff
; /* first offset past del */
5170 int delay
; /* current block is delayed allocated */
5171 int do_fx
; /* free extent at end of routine */
5172 xfs_bmbt_rec_host_t
*ep
; /* current extent entry pointer */
5173 int error
; /* error return value */
5174 int flags
; /* inode logging flags */
5175 xfs_bmbt_irec_t got
; /* current extent entry */
5176 xfs_fileoff_t got_endoff
; /* first offset past got */
5177 int i
; /* temp state */
5178 xfs_ifork_t
*ifp
; /* inode fork pointer */
5179 xfs_mount_t
*mp
; /* mount structure */
5180 xfs_filblks_t nblks
; /* quota/sb block count */
5181 xfs_bmbt_irec_t
new; /* new record to be inserted */
5183 uint qfield
; /* quota field to update */
5184 xfs_filblks_t temp
; /* for indirect length calculations */
5185 xfs_filblks_t temp2
; /* for indirect length calculations */
5189 XFS_STATS_INC(mp
, xs_del_exlist
);
5191 if (whichfork
== XFS_ATTR_FORK
)
5192 state
|= BMAP_ATTRFORK
;
5193 else if (whichfork
== XFS_COW_FORK
)
5194 state
|= BMAP_COWFORK
;
5196 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5197 ASSERT((*idx
>= 0) && (*idx
< xfs_iext_count(ifp
)));
5198 ASSERT(del
->br_blockcount
> 0);
5199 ep
= xfs_iext_get_ext(ifp
, *idx
);
5200 xfs_bmbt_get_all(ep
, &got
);
5201 ASSERT(got
.br_startoff
<= del
->br_startoff
);
5202 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5203 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
5204 ASSERT(got_endoff
>= del_endoff
);
5205 delay
= isnullstartblock(got
.br_startblock
);
5206 ASSERT(isnullstartblock(del
->br_startblock
) == delay
);
5211 * If deleting a real allocation, must free up the disk space.
5214 flags
= XFS_ILOG_CORE
;
5216 * Realtime allocation. Free it and record di_nblocks update.
5218 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
5222 ASSERT(do_mod(del
->br_blockcount
,
5223 mp
->m_sb
.sb_rextsize
) == 0);
5224 ASSERT(do_mod(del
->br_startblock
,
5225 mp
->m_sb
.sb_rextsize
) == 0);
5226 bno
= del
->br_startblock
;
5227 len
= del
->br_blockcount
;
5228 do_div(bno
, mp
->m_sb
.sb_rextsize
);
5229 do_div(len
, mp
->m_sb
.sb_rextsize
);
5230 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
5234 nblks
= len
* mp
->m_sb
.sb_rextsize
;
5235 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
5238 * Ordinary allocation.
5242 nblks
= del
->br_blockcount
;
5243 qfield
= XFS_TRANS_DQ_BCOUNT
;
5246 * Set up del_endblock and cur for later.
5248 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
5250 if ((error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
5251 got
.br_startblock
, got
.br_blockcount
,
5254 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5256 da_old
= da_new
= 0;
5258 da_old
= startblockval(got
.br_startblock
);
5265 * Set flag value to use in switch statement.
5266 * Left-contig is 2, right-contig is 1.
5268 switch (((got
.br_startoff
== del
->br_startoff
) << 1) |
5269 (got_endoff
== del_endoff
)) {
5272 * Matches the whole extent. Delete the entry.
5274 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5275 xfs_iext_remove(ip
, *idx
, 1,
5276 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0);
5281 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5282 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5283 flags
|= XFS_ILOG_CORE
;
5285 flags
|= xfs_ilog_fext(whichfork
);
5288 if ((error
= xfs_btree_delete(cur
, &i
)))
5290 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5295 * Deleting the first part of the extent.
5297 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5298 xfs_bmbt_set_startoff(ep
, del_endoff
);
5299 temp
= got
.br_blockcount
- del
->br_blockcount
;
5300 xfs_bmbt_set_blockcount(ep
, temp
);
5302 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
5304 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5305 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5309 xfs_bmbt_set_startblock(ep
, del_endblock
);
5310 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5312 flags
|= xfs_ilog_fext(whichfork
);
5315 if ((error
= xfs_bmbt_update(cur
, del_endoff
, del_endblock
,
5316 got
.br_blockcount
- del
->br_blockcount
,
5323 * Deleting the last part of the extent.
5325 temp
= got
.br_blockcount
- del
->br_blockcount
;
5326 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5327 xfs_bmbt_set_blockcount(ep
, temp
);
5329 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
5331 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5332 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5336 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5338 flags
|= xfs_ilog_fext(whichfork
);
5341 if ((error
= xfs_bmbt_update(cur
, got
.br_startoff
,
5343 got
.br_blockcount
- del
->br_blockcount
,
5350 * Deleting the middle of the extent.
5352 temp
= del
->br_startoff
- got
.br_startoff
;
5353 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
5354 xfs_bmbt_set_blockcount(ep
, temp
);
5355 new.br_startoff
= del_endoff
;
5356 temp2
= got_endoff
- del_endoff
;
5357 new.br_blockcount
= temp2
;
5358 new.br_state
= got
.br_state
;
5360 new.br_startblock
= del_endblock
;
5361 flags
|= XFS_ILOG_CORE
;
5363 if ((error
= xfs_bmbt_update(cur
,
5365 got
.br_startblock
, temp
,
5368 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
5370 cur
->bc_rec
.b
= new;
5371 error
= xfs_btree_insert(cur
, &i
);
5372 if (error
&& error
!= -ENOSPC
)
5375 * If get no-space back from btree insert,
5376 * it tried a split, and we have a zero
5377 * block reservation.
5378 * Fix up our state and return the error.
5380 if (error
== -ENOSPC
) {
5382 * Reset the cursor, don't trust
5383 * it after any insert operation.
5385 if ((error
= xfs_bmbt_lookup_eq(cur
,
5390 XFS_WANT_CORRUPTED_GOTO(mp
,
5393 * Update the btree record back
5394 * to the original value.
5396 if ((error
= xfs_bmbt_update(cur
,
5403 * Reset the extent record back
5404 * to the original value.
5406 xfs_bmbt_set_blockcount(ep
,
5412 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5414 flags
|= xfs_ilog_fext(whichfork
);
5415 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5416 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
5418 xfs_filblks_t stolen
;
5419 ASSERT(whichfork
== XFS_DATA_FORK
);
5422 * Distribute the original indlen reservation across the
5423 * two new extents. Steal blocks from the deleted extent
5424 * if necessary. Stealing blocks simply fudges the
5425 * fdblocks accounting in xfs_bunmapi().
5427 temp
= xfs_bmap_worst_indlen(ip
, got
.br_blockcount
);
5428 temp2
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
5429 stolen
= xfs_bmap_split_indlen(da_old
, &temp
, &temp2
,
5430 del
->br_blockcount
);
5431 da_new
= temp
+ temp2
- stolen
;
5432 del
->br_blockcount
-= stolen
;
5435 * Set the reservation for each extent. Warn if either
5436 * is zero as this can lead to delalloc problems.
5438 WARN_ON_ONCE(!temp
|| !temp2
);
5439 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
5440 new.br_startblock
= nullstartblock((int)temp2
);
5442 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
5443 xfs_iext_insert(ip
, *idx
+ 1, 1, &new, state
);
5448 /* remove reverse mapping */
5450 error
= xfs_rmap_unmap_extent(mp
, dfops
, ip
, whichfork
, del
);
5456 * If we need to, add to list of extents to delete.
5458 if (do_fx
&& !(bflags
& XFS_BMAPI_REMAP
)) {
5459 if (xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
) {
5460 error
= xfs_refcount_decrease_extent(mp
, dfops
, del
);
5464 xfs_bmap_add_free(mp
, dfops
, del
->br_startblock
,
5465 del
->br_blockcount
, NULL
);
5469 * Adjust inode # blocks in the file.
5472 ip
->i_d
.di_nblocks
-= nblks
;
5474 * Adjust quota data.
5476 if (qfield
&& !(bflags
& XFS_BMAPI_REMAP
))
5477 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5480 * Account for change in delayed indirect blocks.
5481 * Nothing to do for disk quota accounting here.
5483 ASSERT(da_old
>= da_new
);
5484 if (da_old
> da_new
)
5485 xfs_mod_fdblocks(mp
, (int64_t)(da_old
- da_new
), false);
5492 * Unmap (remove) blocks from a file.
5493 * If nexts is nonzero then the number of extents to remove is limited to
5494 * that value. If not all extents in the block range can be removed then
5499 xfs_trans_t
*tp
, /* transaction pointer */
5500 struct xfs_inode
*ip
, /* incore inode */
5501 xfs_fileoff_t bno
, /* starting offset to unmap */
5502 xfs_filblks_t
*rlen
, /* i/o: amount remaining */
5503 int flags
, /* misc flags */
5504 xfs_extnum_t nexts
, /* number of extents max */
5505 xfs_fsblock_t
*firstblock
, /* first allocated block
5506 controls a.g. for allocs */
5507 struct xfs_defer_ops
*dfops
) /* i/o: deferred updates */
5509 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
5510 xfs_bmbt_irec_t del
; /* extent being deleted */
5511 int error
; /* error return value */
5512 xfs_extnum_t extno
; /* extent number in list */
5513 xfs_bmbt_irec_t got
; /* current extent record */
5514 xfs_ifork_t
*ifp
; /* inode fork pointer */
5515 int isrt
; /* freeing in rt area */
5516 xfs_extnum_t lastx
; /* last extent index used */
5517 int logflags
; /* transaction logging flags */
5518 xfs_extlen_t mod
; /* rt extent offset */
5519 xfs_mount_t
*mp
; /* mount structure */
5520 xfs_fileoff_t start
; /* first file offset deleted */
5521 int tmp_logflags
; /* partial logging flags */
5522 int wasdel
; /* was a delayed alloc extent */
5523 int whichfork
; /* data or attribute fork */
5525 xfs_filblks_t len
= *rlen
; /* length to unmap in file */
5527 trace_xfs_bunmap(ip
, bno
, len
, flags
, _RET_IP_
);
5529 whichfork
= xfs_bmapi_whichfork(flags
);
5530 ASSERT(whichfork
!= XFS_COW_FORK
);
5531 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5533 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5534 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5535 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5537 return -EFSCORRUPTED
;
5540 if (XFS_FORCED_SHUTDOWN(mp
))
5543 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5547 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5548 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5550 if (xfs_iext_count(ifp
) == 0) {
5554 XFS_STATS_INC(mp
, xs_blk_unmap
);
5555 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5557 bno
= start
+ len
- 1;
5560 * Check to see if the given block number is past the end of the
5561 * file, back up to the last block if so...
5563 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &lastx
, &got
)) {
5565 xfs_iext_get_extent(ifp
, --lastx
, &got
);
5566 bno
= got
.br_startoff
+ got
.br_blockcount
- 1;
5570 if (ifp
->if_flags
& XFS_IFBROOT
) {
5571 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5572 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5573 cur
->bc_private
.b
.firstblock
= *firstblock
;
5574 cur
->bc_private
.b
.dfops
= dfops
;
5575 cur
->bc_private
.b
.flags
= 0;
5581 * Synchronize by locking the bitmap inode.
5583 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTBITMAP
);
5584 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5585 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTSUM
);
5586 xfs_trans_ijoin(tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
5590 while (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
&& lastx
>= 0 &&
5591 (nexts
== 0 || extno
< nexts
)) {
5593 * Is the found extent after a hole in which bno lives?
5594 * Just back up to the previous extent, if so.
5596 if (got
.br_startoff
> bno
) {
5599 xfs_iext_get_extent(ifp
, lastx
, &got
);
5602 * Is the last block of this extent before the range
5603 * we're supposed to delete? If so, we're done.
5605 bno
= XFS_FILEOFF_MIN(bno
,
5606 got
.br_startoff
+ got
.br_blockcount
- 1);
5610 * Then deal with the (possibly delayed) allocated space
5614 wasdel
= isnullstartblock(del
.br_startblock
);
5615 if (got
.br_startoff
< start
) {
5616 del
.br_startoff
= start
;
5617 del
.br_blockcount
-= start
- got
.br_startoff
;
5619 del
.br_startblock
+= start
- got
.br_startoff
;
5621 if (del
.br_startoff
+ del
.br_blockcount
> bno
+ 1)
5622 del
.br_blockcount
= bno
+ 1 - del
.br_startoff
;
5623 sum
= del
.br_startblock
+ del
.br_blockcount
;
5625 (mod
= do_mod(sum
, mp
->m_sb
.sb_rextsize
))) {
5627 * Realtime extent not lined up at the end.
5628 * The extent could have been split into written
5629 * and unwritten pieces, or we could just be
5630 * unmapping part of it. But we can't really
5631 * get rid of part of a realtime extent.
5633 if (del
.br_state
== XFS_EXT_UNWRITTEN
||
5634 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5636 * This piece is unwritten, or we're not
5637 * using unwritten extents. Skip over it.
5640 bno
-= mod
> del
.br_blockcount
?
5641 del
.br_blockcount
: mod
;
5642 if (bno
< got
.br_startoff
) {
5644 xfs_bmbt_get_all(xfs_iext_get_ext(
5650 * It's written, turn it unwritten.
5651 * This is better than zeroing it.
5653 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5654 ASSERT(tp
->t_blk_res
> 0);
5656 * If this spans a realtime extent boundary,
5657 * chop it back to the start of the one we end at.
5659 if (del
.br_blockcount
> mod
) {
5660 del
.br_startoff
+= del
.br_blockcount
- mod
;
5661 del
.br_startblock
+= del
.br_blockcount
- mod
;
5662 del
.br_blockcount
= mod
;
5664 del
.br_state
= XFS_EXT_UNWRITTEN
;
5665 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5666 whichfork
, &lastx
, &cur
, &del
,
5667 firstblock
, dfops
, &logflags
);
5672 if (isrt
&& (mod
= do_mod(del
.br_startblock
, mp
->m_sb
.sb_rextsize
))) {
5674 * Realtime extent is lined up at the end but not
5675 * at the front. We'll get rid of full extents if
5678 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5679 if (del
.br_blockcount
> mod
) {
5680 del
.br_blockcount
-= mod
;
5681 del
.br_startoff
+= mod
;
5682 del
.br_startblock
+= mod
;
5683 } else if ((del
.br_startoff
== start
&&
5684 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5685 tp
->t_blk_res
== 0)) ||
5686 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5688 * Can't make it unwritten. There isn't
5689 * a full extent here so just skip it.
5691 ASSERT(bno
>= del
.br_blockcount
);
5692 bno
-= del
.br_blockcount
;
5693 if (got
.br_startoff
> bno
&& --lastx
>= 0)
5694 xfs_iext_get_extent(ifp
, lastx
, &got
);
5696 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5697 struct xfs_bmbt_irec prev
;
5700 * This one is already unwritten.
5701 * It must have a written left neighbor.
5702 * Unwrite the killed part of that one and
5706 xfs_iext_get_extent(ifp
, lastx
- 1, &prev
);
5707 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5708 ASSERT(!isnullstartblock(prev
.br_startblock
));
5709 ASSERT(del
.br_startblock
==
5710 prev
.br_startblock
+ prev
.br_blockcount
);
5711 if (prev
.br_startoff
< start
) {
5712 mod
= start
- prev
.br_startoff
;
5713 prev
.br_blockcount
-= mod
;
5714 prev
.br_startblock
+= mod
;
5715 prev
.br_startoff
= start
;
5717 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5719 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5720 ip
, whichfork
, &lastx
, &cur
,
5721 &prev
, firstblock
, dfops
,
5727 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5728 del
.br_state
= XFS_EXT_UNWRITTEN
;
5729 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5730 ip
, whichfork
, &lastx
, &cur
,
5731 &del
, firstblock
, dfops
,
5740 * If it's the case where the directory code is running
5741 * with no block reservation, and the deleted block is in
5742 * the middle of its extent, and the resulting insert
5743 * of an extent would cause transformation to btree format,
5744 * then reject it. The calling code will then swap
5745 * blocks around instead.
5746 * We have to do this now, rather than waiting for the
5747 * conversion to btree format, since the transaction
5750 if (!wasdel
&& tp
->t_blk_res
== 0 &&
5751 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5752 XFS_IFORK_NEXTENTS(ip
, whichfork
) >= /* Note the >= */
5753 XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5754 del
.br_startoff
> got
.br_startoff
&&
5755 del
.br_startoff
+ del
.br_blockcount
<
5756 got
.br_startoff
+ got
.br_blockcount
) {
5762 * Unreserve quota and update realtime free space, if
5763 * appropriate. If delayed allocation, update the inode delalloc
5764 * counter now and wait to update the sb counters as
5765 * xfs_bmap_del_extent() might need to borrow some blocks.
5768 ASSERT(startblockval(del
.br_startblock
) > 0);
5770 xfs_filblks_t rtexts
;
5772 rtexts
= XFS_FSB_TO_B(mp
, del
.br_blockcount
);
5773 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
5774 xfs_mod_frextents(mp
, (int64_t)rtexts
);
5775 (void)xfs_trans_reserve_quota_nblks(NULL
,
5776 ip
, -((long)del
.br_blockcount
), 0,
5777 XFS_QMOPT_RES_RTBLKS
);
5779 (void)xfs_trans_reserve_quota_nblks(NULL
,
5780 ip
, -((long)del
.br_blockcount
), 0,
5781 XFS_QMOPT_RES_REGBLKS
);
5783 ip
->i_delayed_blks
-= del
.br_blockcount
;
5785 cur
->bc_private
.b
.flags
|=
5786 XFS_BTCUR_BPRV_WASDEL
;
5788 cur
->bc_private
.b
.flags
&= ~XFS_BTCUR_BPRV_WASDEL
;
5790 error
= xfs_bmap_del_extent(ip
, tp
, &lastx
, dfops
, cur
, &del
,
5791 &tmp_logflags
, whichfork
, flags
);
5792 logflags
|= tmp_logflags
;
5796 if (!isrt
&& wasdel
)
5797 xfs_mod_fdblocks(mp
, (int64_t)del
.br_blockcount
, false);
5799 bno
= del
.br_startoff
- 1;
5802 * If not done go on to the next (previous) record.
5804 if (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
) {
5806 xfs_iext_get_extent(ifp
, lastx
, &got
);
5807 if (got
.br_startoff
> bno
&& --lastx
>= 0)
5808 xfs_iext_get_extent(ifp
, lastx
, &got
);
5813 if (bno
== (xfs_fileoff_t
)-1 || bno
< start
|| lastx
< 0)
5816 *rlen
= bno
- start
+ 1;
5819 * Convert to a btree if necessary.
5821 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5822 ASSERT(cur
== NULL
);
5823 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, dfops
,
5824 &cur
, 0, &tmp_logflags
, whichfork
);
5825 logflags
|= tmp_logflags
;
5830 * transform from btree to extents, give it cur
5832 else if (xfs_bmap_wants_extents(ip
, whichfork
)) {
5833 ASSERT(cur
!= NULL
);
5834 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &tmp_logflags
,
5836 logflags
|= tmp_logflags
;
5841 * transform from extents to local?
5846 * Log everything. Do this after conversion, there's no point in
5847 * logging the extent records if we've converted to btree format.
5849 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5850 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5851 logflags
&= ~xfs_ilog_fext(whichfork
);
5852 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5853 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5854 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5856 * Log inode even in the error case, if the transaction
5857 * is dirty we'll need to shut down the filesystem.
5860 xfs_trans_log_inode(tp
, ip
, logflags
);
5863 *firstblock
= cur
->bc_private
.b
.firstblock
;
5864 cur
->bc_private
.b
.allocated
= 0;
5866 xfs_btree_del_cursor(cur
,
5867 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5872 /* Unmap a range of a file. */
5876 struct xfs_inode
*ip
,
5881 xfs_fsblock_t
*firstblock
,
5882 struct xfs_defer_ops
*dfops
,
5887 error
= __xfs_bunmapi(tp
, ip
, bno
, &len
, flags
, nexts
, firstblock
,
5894 * Determine whether an extent shift can be accomplished by a merge with the
5895 * extent that precedes the target hole of the shift.
5899 struct xfs_bmbt_irec
*left
, /* preceding extent */
5900 struct xfs_bmbt_irec
*got
, /* current extent to shift */
5901 xfs_fileoff_t shift
) /* shift fsb */
5903 xfs_fileoff_t startoff
;
5905 startoff
= got
->br_startoff
- shift
;
5908 * The extent, once shifted, must be adjacent in-file and on-disk with
5909 * the preceding extent.
5911 if ((left
->br_startoff
+ left
->br_blockcount
!= startoff
) ||
5912 (left
->br_startblock
+ left
->br_blockcount
!= got
->br_startblock
) ||
5913 (left
->br_state
!= got
->br_state
) ||
5914 (left
->br_blockcount
+ got
->br_blockcount
> MAXEXTLEN
))
5921 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5922 * hole in the file. If an extent shift would result in the extent being fully
5923 * adjacent to the extent that currently precedes the hole, we can merge with
5924 * the preceding extent rather than do the shift.
5926 * This function assumes the caller has verified a shift-by-merge is possible
5927 * with the provided extents via xfs_bmse_can_merge().
5931 struct xfs_inode
*ip
,
5933 xfs_fileoff_t shift
, /* shift fsb */
5934 int current_ext
, /* idx of gotp */
5935 struct xfs_bmbt_rec_host
*gotp
, /* extent to shift */
5936 struct xfs_bmbt_rec_host
*leftp
, /* preceding extent */
5937 struct xfs_btree_cur
*cur
,
5938 int *logflags
) /* output */
5940 struct xfs_bmbt_irec got
;
5941 struct xfs_bmbt_irec left
;
5942 xfs_filblks_t blockcount
;
5944 struct xfs_mount
*mp
= ip
->i_mount
;
5946 xfs_bmbt_get_all(gotp
, &got
);
5947 xfs_bmbt_get_all(leftp
, &left
);
5948 blockcount
= left
.br_blockcount
+ got
.br_blockcount
;
5950 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5951 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5952 ASSERT(xfs_bmse_can_merge(&left
, &got
, shift
));
5955 * Merge the in-core extents. Note that the host record pointers and
5956 * current_ext index are invalid once the extent has been removed via
5957 * xfs_iext_remove().
5959 xfs_bmbt_set_blockcount(leftp
, blockcount
);
5960 xfs_iext_remove(ip
, current_ext
, 1, 0);
5963 * Update the on-disk extent count, the btree if necessary and log the
5966 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5967 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5968 *logflags
|= XFS_ILOG_CORE
;
5970 *logflags
|= XFS_ILOG_DEXT
;
5974 /* lookup and remove the extent to merge */
5975 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
, got
.br_startblock
,
5976 got
.br_blockcount
, &i
);
5979 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5981 error
= xfs_btree_delete(cur
, &i
);
5984 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5986 /* lookup and update size of the previous extent */
5987 error
= xfs_bmbt_lookup_eq(cur
, left
.br_startoff
, left
.br_startblock
,
5988 left
.br_blockcount
, &i
);
5991 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5993 left
.br_blockcount
= blockcount
;
5995 return xfs_bmbt_update(cur
, left
.br_startoff
, left
.br_startblock
,
5996 left
.br_blockcount
, left
.br_state
);
6000 * Shift a single extent.
6004 struct xfs_inode
*ip
,
6006 xfs_fileoff_t offset_shift_fsb
,
6008 struct xfs_bmbt_rec_host
*gotp
,
6009 struct xfs_btree_cur
*cur
,
6011 enum shift_direction direction
,
6012 struct xfs_defer_ops
*dfops
)
6014 struct xfs_ifork
*ifp
;
6015 struct xfs_mount
*mp
;
6016 xfs_fileoff_t startoff
;
6017 struct xfs_bmbt_rec_host
*adj_irecp
;
6018 struct xfs_bmbt_irec got
;
6019 struct xfs_bmbt_irec adj_irec
;
6025 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6026 total_extents
= xfs_iext_count(ifp
);
6028 xfs_bmbt_get_all(gotp
, &got
);
6030 /* delalloc extents should be prevented by caller */
6031 XFS_WANT_CORRUPTED_RETURN(mp
, !isnullstartblock(got
.br_startblock
));
6033 if (direction
== SHIFT_LEFT
) {
6034 startoff
= got
.br_startoff
- offset_shift_fsb
;
6037 * Check for merge if we've got an extent to the left,
6038 * otherwise make sure there's enough room at the start
6039 * of the file for the shift.
6041 if (!*current_ext
) {
6042 if (got
.br_startoff
< offset_shift_fsb
)
6044 goto update_current_ext
;
6047 * grab the left extent and check for a large
6050 adj_irecp
= xfs_iext_get_ext(ifp
, *current_ext
- 1);
6051 xfs_bmbt_get_all(adj_irecp
, &adj_irec
);
6054 adj_irec
.br_startoff
+ adj_irec
.br_blockcount
)
6057 /* check whether to merge the extent or shift it down */
6058 if (xfs_bmse_can_merge(&adj_irec
, &got
,
6059 offset_shift_fsb
)) {
6060 error
= xfs_bmse_merge(ip
, whichfork
, offset_shift_fsb
,
6061 *current_ext
, gotp
, adj_irecp
,
6069 startoff
= got
.br_startoff
+ offset_shift_fsb
;
6070 /* nothing to move if this is the last extent */
6071 if (*current_ext
>= (total_extents
- 1))
6072 goto update_current_ext
;
6074 * If this is not the last extent in the file, make sure there
6075 * is enough room between current extent and next extent for
6076 * accommodating the shift.
6078 adj_irecp
= xfs_iext_get_ext(ifp
, *current_ext
+ 1);
6079 xfs_bmbt_get_all(adj_irecp
, &adj_irec
);
6080 if (startoff
+ got
.br_blockcount
> adj_irec
.br_startoff
)
6083 * Unlike a left shift (which involves a hole punch),
6084 * a right shift does not modify extent neighbors
6085 * in any way. We should never find mergeable extents
6086 * in this scenario. Check anyways and warn if we
6087 * encounter two extents that could be one.
6089 if (xfs_bmse_can_merge(&got
, &adj_irec
, offset_shift_fsb
))
6093 * Increment the extent index for the next iteration, update the start
6094 * offset of the in-core extent and update the btree if applicable.
6097 if (direction
== SHIFT_LEFT
)
6101 xfs_bmbt_set_startoff(gotp
, startoff
);
6102 *logflags
|= XFS_ILOG_CORE
;
6105 *logflags
|= XFS_ILOG_DEXT
;
6109 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
, got
.br_startblock
,
6110 got
.br_blockcount
, &i
);
6113 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
6115 got
.br_startoff
= startoff
;
6116 error
= xfs_bmbt_update(cur
, got
.br_startoff
, got
.br_startblock
,
6117 got
.br_blockcount
, got
.br_state
);
6122 /* update reverse mapping */
6123 error
= xfs_rmap_unmap_extent(mp
, dfops
, ip
, whichfork
, &adj_irec
);
6126 adj_irec
.br_startoff
= startoff
;
6127 return xfs_rmap_map_extent(mp
, dfops
, ip
, whichfork
, &adj_irec
);
6131 * Shift extent records to the left/right to cover/create a hole.
6133 * The maximum number of extents to be shifted in a single operation is
6134 * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6135 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6136 * is the length by which each extent is shifted. If there is no hole to shift
6137 * the extents into, this will be considered invalid operation and we abort
6141 xfs_bmap_shift_extents(
6142 struct xfs_trans
*tp
,
6143 struct xfs_inode
*ip
,
6144 xfs_fileoff_t
*next_fsb
,
6145 xfs_fileoff_t offset_shift_fsb
,
6147 xfs_fileoff_t stop_fsb
,
6148 xfs_fsblock_t
*firstblock
,
6149 struct xfs_defer_ops
*dfops
,
6150 enum shift_direction direction
,
6153 struct xfs_btree_cur
*cur
= NULL
;
6154 struct xfs_bmbt_rec_host
*gotp
;
6155 struct xfs_bmbt_irec got
;
6156 struct xfs_mount
*mp
= ip
->i_mount
;
6157 struct xfs_ifork
*ifp
;
6158 xfs_extnum_t nexts
= 0;
6159 xfs_extnum_t current_ext
;
6160 xfs_extnum_t total_extents
;
6161 xfs_extnum_t stop_extent
;
6163 int whichfork
= XFS_DATA_FORK
;
6166 if (unlikely(XFS_TEST_ERROR(
6167 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
6168 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
6169 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
6170 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6171 XFS_ERRLEVEL_LOW
, mp
);
6172 return -EFSCORRUPTED
;
6175 if (XFS_FORCED_SHUTDOWN(mp
))
6178 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
6179 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
6180 ASSERT(direction
== SHIFT_LEFT
|| direction
== SHIFT_RIGHT
);
6181 ASSERT(*next_fsb
!= NULLFSBLOCK
|| direction
== SHIFT_RIGHT
);
6183 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6184 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6185 /* Read in all the extents */
6186 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6191 if (ifp
->if_flags
& XFS_IFBROOT
) {
6192 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6193 cur
->bc_private
.b
.firstblock
= *firstblock
;
6194 cur
->bc_private
.b
.dfops
= dfops
;
6195 cur
->bc_private
.b
.flags
= 0;
6199 * There may be delalloc extents in the data fork before the range we
6200 * are collapsing out, so we cannot use the count of real extents here.
6201 * Instead we have to calculate it from the incore fork.
6203 total_extents
= xfs_iext_count(ifp
);
6204 if (total_extents
== 0) {
6210 * In case of first right shift, we need to initialize next_fsb
6212 if (*next_fsb
== NULLFSBLOCK
) {
6213 gotp
= xfs_iext_get_ext(ifp
, total_extents
- 1);
6214 xfs_bmbt_get_all(gotp
, &got
);
6215 *next_fsb
= got
.br_startoff
;
6216 if (stop_fsb
> *next_fsb
) {
6222 /* Lookup the extent index at which we have to stop */
6223 if (direction
== SHIFT_RIGHT
) {
6224 gotp
= xfs_iext_bno_to_ext(ifp
, stop_fsb
, &stop_extent
);
6225 /* Make stop_extent exclusive of shift range */
6228 stop_extent
= total_extents
;
6231 * Look up the extent index for the fsb where we start shifting. We can
6232 * henceforth iterate with current_ext as extent list changes are locked
6235 * gotp can be null in 2 cases: 1) if there are no extents or 2)
6236 * *next_fsb lies in a hole beyond which there are no extents. Either
6239 gotp
= xfs_iext_bno_to_ext(ifp
, *next_fsb
, ¤t_ext
);
6245 /* some sanity checking before we finally start shifting extents */
6246 if ((direction
== SHIFT_LEFT
&& current_ext
>= stop_extent
) ||
6247 (direction
== SHIFT_RIGHT
&& current_ext
<= stop_extent
)) {
6252 while (nexts
++ < num_exts
) {
6253 error
= xfs_bmse_shift_one(ip
, whichfork
, offset_shift_fsb
,
6254 ¤t_ext
, gotp
, cur
, &logflags
,
6259 * If there was an extent merge during the shift, the extent
6260 * count can change. Update the total and grade the next record.
6262 if (direction
== SHIFT_LEFT
) {
6263 total_extents
= xfs_iext_count(ifp
);
6264 stop_extent
= total_extents
;
6267 if (current_ext
== stop_extent
) {
6269 *next_fsb
= NULLFSBLOCK
;
6272 gotp
= xfs_iext_get_ext(ifp
, current_ext
);
6276 xfs_bmbt_get_all(gotp
, &got
);
6277 *next_fsb
= got
.br_startoff
;
6282 xfs_btree_del_cursor(cur
,
6283 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
6286 xfs_trans_log_inode(tp
, ip
, logflags
);
6292 * Splits an extent into two extents at split_fsb block such that it is
6293 * the first block of the current_ext. @current_ext is a target extent
6294 * to be split. @split_fsb is a block where the extents is split.
6295 * If split_fsb lies in a hole or the first block of extents, just return 0.
6298 xfs_bmap_split_extent_at(
6299 struct xfs_trans
*tp
,
6300 struct xfs_inode
*ip
,
6301 xfs_fileoff_t split_fsb
,
6302 xfs_fsblock_t
*firstfsb
,
6303 struct xfs_defer_ops
*dfops
)
6305 int whichfork
= XFS_DATA_FORK
;
6306 struct xfs_btree_cur
*cur
= NULL
;
6307 struct xfs_bmbt_rec_host
*gotp
;
6308 struct xfs_bmbt_irec got
;
6309 struct xfs_bmbt_irec
new; /* split extent */
6310 struct xfs_mount
*mp
= ip
->i_mount
;
6311 struct xfs_ifork
*ifp
;
6312 xfs_fsblock_t gotblkcnt
; /* new block count for got */
6313 xfs_extnum_t current_ext
;
6318 if (unlikely(XFS_TEST_ERROR(
6319 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
6320 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
6321 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
6322 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6323 XFS_ERRLEVEL_LOW
, mp
);
6324 return -EFSCORRUPTED
;
6327 if (XFS_FORCED_SHUTDOWN(mp
))
6330 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6331 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6332 /* Read in all the extents */
6333 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6339 * gotp can be null in 2 cases: 1) if there are no extents
6340 * or 2) split_fsb lies in a hole beyond which there are
6341 * no extents. Either way, we are done.
6343 gotp
= xfs_iext_bno_to_ext(ifp
, split_fsb
, ¤t_ext
);
6347 xfs_bmbt_get_all(gotp
, &got
);
6350 * Check split_fsb lies in a hole or the start boundary offset
6353 if (got
.br_startoff
>= split_fsb
)
6356 gotblkcnt
= split_fsb
- got
.br_startoff
;
6357 new.br_startoff
= split_fsb
;
6358 new.br_startblock
= got
.br_startblock
+ gotblkcnt
;
6359 new.br_blockcount
= got
.br_blockcount
- gotblkcnt
;
6360 new.br_state
= got
.br_state
;
6362 if (ifp
->if_flags
& XFS_IFBROOT
) {
6363 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6364 cur
->bc_private
.b
.firstblock
= *firstfsb
;
6365 cur
->bc_private
.b
.dfops
= dfops
;
6366 cur
->bc_private
.b
.flags
= 0;
6367 error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
6373 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
6376 xfs_bmbt_set_blockcount(gotp
, gotblkcnt
);
6377 got
.br_blockcount
= gotblkcnt
;
6379 logflags
= XFS_ILOG_CORE
;
6381 error
= xfs_bmbt_update(cur
, got
.br_startoff
,
6388 logflags
|= XFS_ILOG_DEXT
;
6390 /* Add new extent */
6392 xfs_iext_insert(ip
, current_ext
, 1, &new, 0);
6393 XFS_IFORK_NEXT_SET(ip
, whichfork
,
6394 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
6397 error
= xfs_bmbt_lookup_eq(cur
, new.br_startoff
,
6398 new.br_startblock
, new.br_blockcount
,
6402 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, del_cursor
);
6403 cur
->bc_rec
.b
.br_state
= new.br_state
;
6405 error
= xfs_btree_insert(cur
, &i
);
6408 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
6412 * Convert to a btree if necessary.
6414 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
6415 int tmp_logflags
; /* partial log flag return val */
6417 ASSERT(cur
== NULL
);
6418 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstfsb
, dfops
,
6419 &cur
, 0, &tmp_logflags
, whichfork
);
6420 logflags
|= tmp_logflags
;
6425 cur
->bc_private
.b
.allocated
= 0;
6426 xfs_btree_del_cursor(cur
,
6427 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
6431 xfs_trans_log_inode(tp
, ip
, logflags
);
6436 xfs_bmap_split_extent(
6437 struct xfs_inode
*ip
,
6438 xfs_fileoff_t split_fsb
)
6440 struct xfs_mount
*mp
= ip
->i_mount
;
6441 struct xfs_trans
*tp
;
6442 struct xfs_defer_ops dfops
;
6443 xfs_fsblock_t firstfsb
;
6446 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
,
6447 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0, 0, &tp
);
6451 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
6452 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
6454 xfs_defer_init(&dfops
, &firstfsb
);
6456 error
= xfs_bmap_split_extent_at(tp
, ip
, split_fsb
,
6461 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
6465 return xfs_trans_commit(tp
);
6468 xfs_defer_cancel(&dfops
);
6469 xfs_trans_cancel(tp
);
6473 /* Deferred mapping is only for real extents in the data fork. */
6475 xfs_bmap_is_update_needed(
6476 struct xfs_bmbt_irec
*bmap
)
6478 return bmap
->br_startblock
!= HOLESTARTBLOCK
&&
6479 bmap
->br_startblock
!= DELAYSTARTBLOCK
;
6482 /* Record a bmap intent. */
6485 struct xfs_mount
*mp
,
6486 struct xfs_defer_ops
*dfops
,
6487 enum xfs_bmap_intent_type type
,
6488 struct xfs_inode
*ip
,
6490 struct xfs_bmbt_irec
*bmap
)
6493 struct xfs_bmap_intent
*bi
;
6495 trace_xfs_bmap_defer(mp
,
6496 XFS_FSB_TO_AGNO(mp
, bmap
->br_startblock
),
6498 XFS_FSB_TO_AGBNO(mp
, bmap
->br_startblock
),
6499 ip
->i_ino
, whichfork
,
6501 bmap
->br_blockcount
,
6504 bi
= kmem_alloc(sizeof(struct xfs_bmap_intent
), KM_SLEEP
| KM_NOFS
);
6505 INIT_LIST_HEAD(&bi
->bi_list
);
6508 bi
->bi_whichfork
= whichfork
;
6509 bi
->bi_bmap
= *bmap
;
6511 error
= xfs_defer_join(dfops
, bi
->bi_owner
);
6517 xfs_defer_add(dfops
, XFS_DEFER_OPS_TYPE_BMAP
, &bi
->bi_list
);
6521 /* Map an extent into a file. */
6523 xfs_bmap_map_extent(
6524 struct xfs_mount
*mp
,
6525 struct xfs_defer_ops
*dfops
,
6526 struct xfs_inode
*ip
,
6527 struct xfs_bmbt_irec
*PREV
)
6529 if (!xfs_bmap_is_update_needed(PREV
))
6532 return __xfs_bmap_add(mp
, dfops
, XFS_BMAP_MAP
, ip
,
6533 XFS_DATA_FORK
, PREV
);
6536 /* Unmap an extent out of a file. */
6538 xfs_bmap_unmap_extent(
6539 struct xfs_mount
*mp
,
6540 struct xfs_defer_ops
*dfops
,
6541 struct xfs_inode
*ip
,
6542 struct xfs_bmbt_irec
*PREV
)
6544 if (!xfs_bmap_is_update_needed(PREV
))
6547 return __xfs_bmap_add(mp
, dfops
, XFS_BMAP_UNMAP
, ip
,
6548 XFS_DATA_FORK
, PREV
);
6552 * Process one of the deferred bmap operations. We pass back the
6553 * btree cursor to maintain our lock on the bmapbt between calls.
6556 xfs_bmap_finish_one(
6557 struct xfs_trans
*tp
,
6558 struct xfs_defer_ops
*dfops
,
6559 struct xfs_inode
*ip
,
6560 enum xfs_bmap_intent_type type
,
6562 xfs_fileoff_t startoff
,
6563 xfs_fsblock_t startblock
,
6564 xfs_filblks_t blockcount
,
6567 int error
= 0, done
;
6569 trace_xfs_bmap_deferred(tp
->t_mountp
,
6570 XFS_FSB_TO_AGNO(tp
->t_mountp
, startblock
), type
,
6571 XFS_FSB_TO_AGBNO(tp
->t_mountp
, startblock
),
6572 ip
->i_ino
, whichfork
, startoff
, blockcount
, state
);
6574 if (WARN_ON_ONCE(whichfork
!= XFS_DATA_FORK
))
6575 return -EFSCORRUPTED
;
6577 if (XFS_TEST_ERROR(false, tp
->t_mountp
,
6578 XFS_ERRTAG_BMAP_FINISH_ONE
,
6579 XFS_RANDOM_BMAP_FINISH_ONE
))
6584 error
= xfs_bmapi_remap(tp
, ip
, startoff
, blockcount
,
6587 case XFS_BMAP_UNMAP
:
6588 error
= xfs_bunmapi(tp
, ip
, startoff
, blockcount
,
6589 XFS_BMAPI_REMAP
, 1, &startblock
, dfops
, &done
);
6594 error
= -EFSCORRUPTED
;