1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
6 #include "libxfs_priv.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
17 #include "xfs_inode.h"
18 #include "xfs_btree.h"
19 #include "xfs_trans.h"
20 #include "xfs_alloc.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_errortag.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_trace.h"
26 #include "xfs_attr_leaf.h"
27 #include "xfs_quota_defs.h"
29 #include "xfs_ag_resv.h"
30 #include "xfs_refcount.h"
33 kmem_zone_t
*xfs_bmap_free_item_zone
;
36 * Miscellaneous helper functions
40 * Compute and fill in the value of the maximum depth of a bmap btree
41 * in this filesystem. Done once, during mount.
44 xfs_bmap_compute_maxlevels(
45 xfs_mount_t
*mp
, /* file system mount structure */
46 int whichfork
) /* data or attr fork */
48 int level
; /* btree level */
49 uint maxblocks
; /* max blocks at this level */
50 uint maxleafents
; /* max leaf entries possible */
51 int maxrootrecs
; /* max records in root block */
52 int minleafrecs
; /* min records in leaf block */
53 int minnoderecs
; /* min records in node block */
54 int sz
; /* root block size */
57 * The maximum number of extents in a file, hence the maximum number of
58 * leaf entries, is controlled by the size of the on-disk extent count,
59 * either a signed 32-bit number for the data fork, or a signed 16-bit
60 * number for the attr fork.
62 * Note that we can no longer assume that if we are in ATTR1 that
63 * the fork offset of all the inodes will be
64 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
65 * with ATTR2 and then mounted back with ATTR1, keeping the
66 * di_forkoff's fixed but probably at various positions. Therefore,
67 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
68 * of a minimum size available.
70 if (whichfork
== XFS_DATA_FORK
) {
71 maxleafents
= MAXEXTNUM
;
72 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
74 maxleafents
= MAXAEXTNUM
;
75 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
77 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
78 minleafrecs
= mp
->m_bmap_dmnr
[0];
79 minnoderecs
= mp
->m_bmap_dmnr
[1];
80 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
81 for (level
= 1; maxblocks
> 1; level
++) {
82 if (maxblocks
<= maxrootrecs
)
85 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
87 mp
->m_bm_maxlevels
[whichfork
] = level
;
90 STATIC
int /* error */
92 struct xfs_btree_cur
*cur
,
93 struct xfs_bmbt_irec
*irec
,
94 int *stat
) /* success/failure */
96 cur
->bc_rec
.b
= *irec
;
97 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
100 STATIC
int /* error */
101 xfs_bmbt_lookup_first(
102 struct xfs_btree_cur
*cur
,
103 int *stat
) /* success/failure */
105 cur
->bc_rec
.b
.br_startoff
= 0;
106 cur
->bc_rec
.b
.br_startblock
= 0;
107 cur
->bc_rec
.b
.br_blockcount
= 0;
108 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
112 * Check if the inode needs to be converted to btree format.
114 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
116 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
118 return whichfork
!= XFS_COW_FORK
&&
119 ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
&&
120 ifp
->if_nextents
> XFS_IFORK_MAXEXT(ip
, whichfork
);
124 * Check if the inode should be converted to extent format.
126 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
128 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
130 return whichfork
!= XFS_COW_FORK
&&
131 ifp
->if_format
== XFS_DINODE_FMT_BTREE
&&
132 ifp
->if_nextents
<= XFS_IFORK_MAXEXT(ip
, whichfork
);
136 * Update the record referred to by cur to the value given by irec
137 * This either works (return 0) or gets an EFSCORRUPTED error.
141 struct xfs_btree_cur
*cur
,
142 struct xfs_bmbt_irec
*irec
)
144 union xfs_btree_rec rec
;
146 xfs_bmbt_disk_set_all(&rec
.bmbt
, irec
);
147 return xfs_btree_update(cur
, &rec
);
151 * Compute the worst-case number of indirect blocks that will be used
152 * for ip's delayed extent of length "len".
155 xfs_bmap_worst_indlen(
156 xfs_inode_t
*ip
, /* incore inode pointer */
157 xfs_filblks_t len
) /* delayed extent length */
159 int level
; /* btree level number */
160 int maxrecs
; /* maximum record count at this level */
161 xfs_mount_t
*mp
; /* mount structure */
162 xfs_filblks_t rval
; /* return value */
165 maxrecs
= mp
->m_bmap_dmxr
[0];
166 for (level
= 0, rval
= 0;
167 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
170 do_div(len
, maxrecs
);
173 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
176 maxrecs
= mp
->m_bmap_dmxr
[1];
182 * Calculate the default attribute fork offset for newly created inodes.
185 xfs_default_attroffset(
186 struct xfs_inode
*ip
)
188 struct xfs_mount
*mp
= ip
->i_mount
;
191 if (mp
->m_sb
.sb_inodesize
== 256)
192 offset
= XFS_LITINO(mp
) - XFS_BMDR_SPACE_CALC(MINABTPTRS
);
194 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
196 ASSERT(offset
< XFS_LITINO(mp
));
201 * Helper routine to reset inode di_forkoff field when switching
202 * attribute fork from local to extent format - we reset it where
203 * possible to make space available for inline data fork extents.
206 xfs_bmap_forkoff_reset(
210 if (whichfork
== XFS_ATTR_FORK
&&
211 ip
->i_df
.if_format
!= XFS_DINODE_FMT_DEV
&&
212 ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
) {
213 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
215 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
216 ip
->i_d
.di_forkoff
= dfl_forkoff
;
221 STATIC
struct xfs_buf
*
223 struct xfs_btree_cur
*cur
,
226 struct xfs_log_item
*lip
;
232 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
233 if (!cur
->bc_bufs
[i
])
235 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
236 return cur
->bc_bufs
[i
];
239 /* Chase down all the log items to see if the bp is there */
240 list_for_each_entry(lip
, &cur
->bc_tp
->t_items
, li_trans
) {
241 struct xfs_buf_log_item
*bip
= (struct xfs_buf_log_item
*)lip
;
243 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
244 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
253 struct xfs_btree_block
*block
,
259 __be64
*pp
, *thispa
; /* pointer to block address */
260 xfs_bmbt_key_t
*prevp
, *keyp
;
262 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
265 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
266 dmxr
= mp
->m_bmap_dmxr
[0];
267 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
270 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
271 be64_to_cpu(keyp
->br_startoff
));
276 * Compare the block numbers to see if there are dups.
279 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
281 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
283 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
285 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
287 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
288 if (*thispa
== *pp
) {
289 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
291 (unsigned long long)be64_to_cpu(*thispa
));
292 xfs_err(mp
, "%s: ptrs are equal in node\n",
294 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
301 * Check that the extents for the inode ip are in the right order in all
302 * btree leaves. THis becomes prohibitively expensive for large extent count
303 * files, so don't bother with inodes that have more than 10,000 extents in
304 * them. The btree record ordering checks will still be done, so for such large
305 * bmapbt constructs that is going to catch most corruptions.
308 xfs_bmap_check_leaf_extents(
309 xfs_btree_cur_t
*cur
, /* btree cursor or null */
310 xfs_inode_t
*ip
, /* incore inode pointer */
311 int whichfork
) /* data or attr fork */
313 struct xfs_mount
*mp
= ip
->i_mount
;
314 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
315 struct xfs_btree_block
*block
; /* current btree block */
316 xfs_fsblock_t bno
; /* block # of "block" */
317 struct xfs_buf
*bp
; /* buffer for "block" */
318 int error
; /* error return value */
319 xfs_extnum_t i
=0, j
; /* index into the extents list */
320 int level
; /* btree level, for checking */
321 __be64
*pp
; /* pointer to block address */
322 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
323 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
324 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
327 if (ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
330 /* skip large extent count inodes */
331 if (ip
->i_df
.if_nextents
> 10000)
335 block
= ifp
->if_broot
;
337 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
339 level
= be16_to_cpu(block
->bb_level
);
341 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
342 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
343 bno
= be64_to_cpu(*pp
);
345 ASSERT(bno
!= NULLFSBLOCK
);
346 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
347 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
350 * Go down the tree until leaf level is reached, following the first
351 * pointer (leftmost) at each level.
353 while (level
-- > 0) {
354 /* See if buf is in cur first */
356 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
359 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, &bp
,
365 block
= XFS_BUF_TO_BLOCK(bp
);
370 * Check this block for basic sanity (increasing keys and
371 * no duplicate blocks).
374 xfs_check_block(block
, mp
, 0, 0);
375 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
376 bno
= be64_to_cpu(*pp
);
377 if (XFS_IS_CORRUPT(mp
, !xfs_verify_fsbno(mp
, bno
))) {
378 error
= -EFSCORRUPTED
;
383 xfs_trans_brelse(NULL
, bp
);
388 * Here with bp and block set to the leftmost leaf node in the tree.
393 * Loop over all leaf nodes checking that all extents are in the right order.
396 xfs_fsblock_t nextbno
;
397 xfs_extnum_t num_recs
;
400 num_recs
= xfs_btree_get_numrecs(block
);
403 * Read-ahead the next leaf block, if any.
406 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
409 * Check all the extents to make sure they are OK.
410 * If we had a previous block, the last entry should
411 * conform with the first entry in this one.
414 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
416 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
417 xfs_bmbt_disk_get_blockcount(&last
) <=
418 xfs_bmbt_disk_get_startoff(ep
));
420 for (j
= 1; j
< num_recs
; j
++) {
421 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
422 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
423 xfs_bmbt_disk_get_blockcount(ep
) <=
424 xfs_bmbt_disk_get_startoff(nextp
));
432 xfs_trans_brelse(NULL
, bp
);
436 * If we've reached the end, stop.
438 if (bno
== NULLFSBLOCK
)
442 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
445 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, &bp
,
451 block
= XFS_BUF_TO_BLOCK(bp
);
457 xfs_warn(mp
, "%s: at error0", __func__
);
459 xfs_trans_brelse(NULL
, bp
);
461 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
463 xfs_err(mp
, "%s: CORRUPTED BTREE OR SOMETHING", __func__
);
464 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
469 * Validate that the bmbt_irecs being returned from bmapi are valid
470 * given the caller's original parameters. Specifically check the
471 * ranges of the returned irecs to ensure that they only extend beyond
472 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
475 xfs_bmap_validate_ret(
479 xfs_bmbt_irec_t
*mval
,
483 int i
; /* index to map values */
485 ASSERT(ret_nmap
<= nmap
);
487 for (i
= 0; i
< ret_nmap
; i
++) {
488 ASSERT(mval
[i
].br_blockcount
> 0);
489 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
490 ASSERT(mval
[i
].br_startoff
>= bno
);
491 ASSERT(mval
[i
].br_blockcount
<= len
);
492 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
495 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
496 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
500 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
501 mval
[i
].br_startoff
);
502 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
503 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
504 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
505 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
510 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
511 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
515 * bmap free list manipulation functions
519 * Add the extent to the list of extents to be free at transaction end.
520 * The list is maintained sorted (by block number).
524 struct xfs_trans
*tp
,
527 const struct xfs_owner_info
*oinfo
,
530 struct xfs_extent_free_item
*new; /* new element */
532 struct xfs_mount
*mp
= tp
->t_mountp
;
536 ASSERT(bno
!= NULLFSBLOCK
);
538 ASSERT(len
<= MAXEXTLEN
);
539 ASSERT(!isnullstartblock(bno
));
540 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
541 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
542 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
543 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
544 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
545 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
547 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
549 new = kmem_cache_alloc(xfs_bmap_free_item_zone
,
550 GFP_KERNEL
| __GFP_NOFAIL
);
551 new->xefi_startblock
= bno
;
552 new->xefi_blockcount
= (xfs_extlen_t
)len
;
554 new->xefi_oinfo
= *oinfo
;
556 new->xefi_oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
557 new->xefi_skip_discard
= skip_discard
;
558 trace_xfs_bmap_free_defer(tp
->t_mountp
,
559 XFS_FSB_TO_AGNO(tp
->t_mountp
, bno
), 0,
560 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bno
), len
);
561 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_FREE
, &new->xefi_list
);
565 * Inode fork format manipulation functions
569 * Convert the inode format to extent format if it currently is in btree format,
570 * but the extent list is small enough that it fits into the extent format.
572 * Since the extents are already in-core, all we have to do is give up the space
573 * for the btree root and pitch the leaf block.
575 STATIC
int /* error */
576 xfs_bmap_btree_to_extents(
577 struct xfs_trans
*tp
, /* transaction pointer */
578 struct xfs_inode
*ip
, /* incore inode pointer */
579 struct xfs_btree_cur
*cur
, /* btree cursor */
580 int *logflagsp
, /* inode logging flags */
581 int whichfork
) /* data or attr fork */
583 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
584 struct xfs_mount
*mp
= ip
->i_mount
;
585 struct xfs_btree_block
*rblock
= ifp
->if_broot
;
586 struct xfs_btree_block
*cblock
;/* child btree block */
587 xfs_fsblock_t cbno
; /* child block number */
588 struct xfs_buf
*cbp
; /* child block's buffer */
589 int error
; /* error return value */
590 __be64
*pp
; /* ptr to block address */
591 struct xfs_owner_info oinfo
;
593 /* check if we actually need the extent format first: */
594 if (!xfs_bmap_wants_extents(ip
, whichfork
))
598 ASSERT(whichfork
!= XFS_COW_FORK
);
599 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
600 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_BTREE
);
601 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
602 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
603 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
605 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
606 cbno
= be64_to_cpu(*pp
);
608 if (XFS_IS_CORRUPT(cur
->bc_mp
, !xfs_btree_check_lptr(cur
, cbno
, 1)))
609 return -EFSCORRUPTED
;
611 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, &cbp
, XFS_BMAP_BTREE_REF
,
615 cblock
= XFS_BUF_TO_BLOCK(cbp
);
616 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
618 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, whichfork
);
619 xfs_bmap_add_free(cur
->bc_tp
, cbno
, 1, &oinfo
);
620 ip
->i_d
.di_nblocks
--;
621 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
622 xfs_trans_binval(tp
, cbp
);
623 if (cur
->bc_bufs
[0] == cbp
)
624 cur
->bc_bufs
[0] = NULL
;
625 xfs_iroot_realloc(ip
, -1, whichfork
);
626 ASSERT(ifp
->if_broot
== NULL
);
627 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
628 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
629 *logflagsp
|= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
634 * Convert an extents-format file into a btree-format file.
635 * The new file will have a root block (in the inode) and a single child block.
637 STATIC
int /* error */
638 xfs_bmap_extents_to_btree(
639 struct xfs_trans
*tp
, /* transaction pointer */
640 struct xfs_inode
*ip
, /* incore inode pointer */
641 struct xfs_btree_cur
**curp
, /* cursor returned to caller */
642 int wasdel
, /* converting a delayed alloc */
643 int *logflagsp
, /* inode logging flags */
644 int whichfork
) /* data or attr fork */
646 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
647 struct xfs_buf
*abp
; /* buffer for ablock */
648 struct xfs_alloc_arg args
; /* allocation arguments */
649 struct xfs_bmbt_rec
*arp
; /* child record pointer */
650 struct xfs_btree_block
*block
; /* btree root block */
651 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
652 int error
; /* error return value */
653 struct xfs_ifork
*ifp
; /* inode fork pointer */
654 struct xfs_bmbt_key
*kp
; /* root block key pointer */
655 struct xfs_mount
*mp
; /* mount structure */
656 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
657 struct xfs_iext_cursor icur
;
658 struct xfs_bmbt_irec rec
;
659 xfs_extnum_t cnt
= 0;
662 ASSERT(whichfork
!= XFS_COW_FORK
);
663 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
664 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
);
667 * Make space in the inode incore. This needs to be undone if we fail
668 * to expand the root.
670 xfs_iroot_realloc(ip
, 1, whichfork
);
671 ifp
->if_flags
|= XFS_IFBROOT
;
676 block
= ifp
->if_broot
;
677 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
678 XFS_BTNUM_BMAP
, 1, 1, ip
->i_ino
,
679 XFS_BTREE_LONG_PTRS
);
681 * Need a cursor. Can't allocate until bb_level is filled in.
683 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
684 cur
->bc_ino
.flags
= wasdel
? XFS_BTCUR_BMBT_WASDEL
: 0;
686 * Convert to a btree with two levels, one record in root.
688 ifp
->if_format
= XFS_DINODE_FMT_BTREE
;
689 memset(&args
, 0, sizeof(args
));
692 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, ip
->i_ino
, whichfork
);
693 if (tp
->t_firstblock
== NULLFSBLOCK
) {
694 args
.type
= XFS_ALLOCTYPE_START_BNO
;
695 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
696 } else if (tp
->t_flags
& XFS_TRANS_LOWMODE
) {
697 args
.type
= XFS_ALLOCTYPE_START_BNO
;
698 args
.fsbno
= tp
->t_firstblock
;
700 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
701 args
.fsbno
= tp
->t_firstblock
;
703 args
.minlen
= args
.maxlen
= args
.prod
= 1;
704 args
.wasdel
= wasdel
;
706 error
= xfs_alloc_vextent(&args
);
708 goto out_root_realloc
;
710 if (WARN_ON_ONCE(args
.fsbno
== NULLFSBLOCK
)) {
712 goto out_root_realloc
;
716 * Allocation can't fail, the space was reserved.
718 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
||
719 args
.agno
>= XFS_FSB_TO_AGNO(mp
, tp
->t_firstblock
));
720 tp
->t_firstblock
= args
.fsbno
;
721 cur
->bc_ino
.allocated
++;
722 ip
->i_d
.di_nblocks
++;
723 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
724 error
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
725 XFS_FSB_TO_DADDR(mp
, args
.fsbno
),
726 mp
->m_bsize
, 0, &abp
);
728 goto out_unreserve_dquot
;
731 * Fill in the child block.
733 abp
->b_ops
= &xfs_bmbt_buf_ops
;
734 ablock
= XFS_BUF_TO_BLOCK(abp
);
735 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
736 XFS_BTNUM_BMAP
, 0, 0, ip
->i_ino
,
737 XFS_BTREE_LONG_PTRS
);
739 for_each_xfs_iext(ifp
, &icur
, &rec
) {
740 if (isnullstartblock(rec
.br_startblock
))
742 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1 + cnt
);
743 xfs_bmbt_disk_set_all(arp
, &rec
);
746 ASSERT(cnt
== ifp
->if_nextents
);
747 xfs_btree_set_numrecs(ablock
, cnt
);
750 * Fill in the root key and pointer.
752 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
753 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
754 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
755 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
756 be16_to_cpu(block
->bb_level
)));
757 *pp
= cpu_to_be64(args
.fsbno
);
760 * Do all this logging at the end so that
761 * the root is at the right level.
763 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
764 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
765 ASSERT(*curp
== NULL
);
767 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
771 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
773 xfs_iroot_realloc(ip
, -1, whichfork
);
774 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
775 ASSERT(ifp
->if_broot
== NULL
);
776 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
782 * Convert a local file to an extents file.
783 * This code is out of bounds for data forks of regular files,
784 * since the file data needs to get logged so things will stay consistent.
785 * (The bmap-level manipulations are ok, though).
788 xfs_bmap_local_to_extents_empty(
789 struct xfs_trans
*tp
,
790 struct xfs_inode
*ip
,
793 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
795 ASSERT(whichfork
!= XFS_COW_FORK
);
796 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_LOCAL
);
797 ASSERT(ifp
->if_bytes
== 0);
798 ASSERT(ifp
->if_nextents
== 0);
800 xfs_bmap_forkoff_reset(ip
, whichfork
);
801 ifp
->if_flags
&= ~XFS_IFINLINE
;
802 ifp
->if_flags
|= XFS_IFEXTENTS
;
803 ifp
->if_u1
.if_root
= NULL
;
805 ifp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
806 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
810 STATIC
int /* error */
811 xfs_bmap_local_to_extents(
812 xfs_trans_t
*tp
, /* transaction pointer */
813 xfs_inode_t
*ip
, /* incore inode pointer */
814 xfs_extlen_t total
, /* total blocks needed by transaction */
815 int *logflagsp
, /* inode logging flags */
817 void (*init_fn
)(struct xfs_trans
*tp
,
819 struct xfs_inode
*ip
,
820 struct xfs_ifork
*ifp
))
823 int flags
; /* logging flags returned */
824 struct xfs_ifork
*ifp
; /* inode fork pointer */
825 xfs_alloc_arg_t args
; /* allocation arguments */
826 struct xfs_buf
*bp
; /* buffer for extent block */
827 struct xfs_bmbt_irec rec
;
828 struct xfs_iext_cursor icur
;
831 * We don't want to deal with the case of keeping inode data inline yet.
832 * So sending the data fork of a regular inode is invalid.
834 ASSERT(!(S_ISREG(VFS_I(ip
)->i_mode
) && whichfork
== XFS_DATA_FORK
));
835 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
836 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_LOCAL
);
838 if (!ifp
->if_bytes
) {
839 xfs_bmap_local_to_extents_empty(tp
, ip
, whichfork
);
840 flags
= XFS_ILOG_CORE
;
846 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
)) == XFS_IFINLINE
);
847 memset(&args
, 0, sizeof(args
));
849 args
.mp
= ip
->i_mount
;
850 xfs_rmap_ino_owner(&args
.oinfo
, ip
->i_ino
, whichfork
, 0);
852 * Allocate a block. We know we need only one, since the
853 * file currently fits in an inode.
855 if (tp
->t_firstblock
== NULLFSBLOCK
) {
856 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
857 args
.type
= XFS_ALLOCTYPE_START_BNO
;
859 args
.fsbno
= tp
->t_firstblock
;
860 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
863 args
.minlen
= args
.maxlen
= args
.prod
= 1;
864 error
= xfs_alloc_vextent(&args
);
868 /* Can't fail, the space was reserved. */
869 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
870 ASSERT(args
.len
== 1);
871 tp
->t_firstblock
= args
.fsbno
;
872 error
= xfs_trans_get_buf(tp
, args
.mp
->m_ddev_targp
,
873 XFS_FSB_TO_DADDR(args
.mp
, args
.fsbno
),
874 args
.mp
->m_bsize
, 0, &bp
);
879 * Initialize the block, copy the data and log the remote buffer.
881 * The callout is responsible for logging because the remote format
882 * might differ from the local format and thus we don't know how much to
883 * log here. Note that init_fn must also set the buffer log item type
886 init_fn(tp
, bp
, ip
, ifp
);
888 /* account for the change in fork size */
889 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
890 xfs_bmap_local_to_extents_empty(tp
, ip
, whichfork
);
891 flags
|= XFS_ILOG_CORE
;
893 ifp
->if_u1
.if_root
= NULL
;
897 rec
.br_startblock
= args
.fsbno
;
898 rec
.br_blockcount
= 1;
899 rec
.br_state
= XFS_EXT_NORM
;
900 xfs_iext_first(ifp
, &icur
);
901 xfs_iext_insert(ip
, &icur
, &rec
, 0);
903 ifp
->if_nextents
= 1;
904 ip
->i_d
.di_nblocks
= 1;
905 xfs_trans_mod_dquot_byino(tp
, ip
,
906 XFS_TRANS_DQ_BCOUNT
, 1L);
907 flags
|= xfs_ilog_fext(whichfork
);
915 * Called from xfs_bmap_add_attrfork to handle btree format files.
917 STATIC
int /* error */
918 xfs_bmap_add_attrfork_btree(
919 xfs_trans_t
*tp
, /* transaction pointer */
920 xfs_inode_t
*ip
, /* incore inode pointer */
921 int *flags
) /* inode logging flags */
923 xfs_btree_cur_t
*cur
; /* btree cursor */
924 int error
; /* error return value */
925 xfs_mount_t
*mp
; /* file system mount struct */
926 int stat
; /* newroot status */
929 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
930 *flags
|= XFS_ILOG_DBROOT
;
932 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
933 error
= xfs_bmbt_lookup_first(cur
, &stat
);
936 /* must be at least one entry */
937 if (XFS_IS_CORRUPT(mp
, stat
!= 1)) {
938 error
= -EFSCORRUPTED
;
941 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
944 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
947 cur
->bc_ino
.allocated
= 0;
948 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
952 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
957 * Called from xfs_bmap_add_attrfork to handle extents format files.
959 STATIC
int /* error */
960 xfs_bmap_add_attrfork_extents(
961 struct xfs_trans
*tp
, /* transaction pointer */
962 struct xfs_inode
*ip
, /* incore inode pointer */
963 int *flags
) /* inode logging flags */
965 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
966 int error
; /* error return value */
968 if (ip
->i_df
.if_nextents
* sizeof(struct xfs_bmbt_rec
) <=
972 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0, flags
,
975 cur
->bc_ino
.allocated
= 0;
976 xfs_btree_del_cursor(cur
, error
);
982 * Called from xfs_bmap_add_attrfork to handle local format files. Each
983 * different data fork content type needs a different callout to do the
984 * conversion. Some are basic and only require special block initialisation
985 * callouts for the data formating, others (directories) are so specialised they
986 * handle everything themselves.
988 * XXX (dgc): investigate whether directory conversion can use the generic
989 * formatting callout. It should be possible - it's just a very complex
992 STATIC
int /* error */
993 xfs_bmap_add_attrfork_local(
994 struct xfs_trans
*tp
, /* transaction pointer */
995 struct xfs_inode
*ip
, /* incore inode pointer */
996 int *flags
) /* inode logging flags */
998 struct xfs_da_args dargs
; /* args for dir/attr code */
1000 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1003 if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
1004 memset(&dargs
, 0, sizeof(dargs
));
1005 dargs
.geo
= ip
->i_mount
->m_dir_geo
;
1007 dargs
.total
= dargs
.geo
->fsbcount
;
1008 dargs
.whichfork
= XFS_DATA_FORK
;
1010 return xfs_dir2_sf_to_block(&dargs
);
1013 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1014 return xfs_bmap_local_to_extents(tp
, ip
, 1, flags
,
1016 xfs_symlink_local_to_remote
);
1018 /* should only be called for types that support local format data */
1020 return -EFSCORRUPTED
;
1023 /* Set an inode attr fork off based on the format */
1025 xfs_bmap_set_attrforkoff(
1026 struct xfs_inode
*ip
,
1030 switch (ip
->i_df
.if_format
) {
1031 case XFS_DINODE_FMT_DEV
:
1032 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1034 case XFS_DINODE_FMT_LOCAL
:
1035 case XFS_DINODE_FMT_EXTENTS
:
1036 case XFS_DINODE_FMT_BTREE
:
1037 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1038 if (!ip
->i_d
.di_forkoff
)
1039 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1040 else if ((ip
->i_mount
->m_flags
& XFS_MOUNT_ATTR2
) && version
)
1052 * Convert inode from non-attributed to attributed.
1053 * Must not be in a transaction, ip must not be locked.
1055 int /* error code */
1056 xfs_bmap_add_attrfork(
1057 xfs_inode_t
*ip
, /* incore inode pointer */
1058 int size
, /* space new attribute needs */
1059 int rsvd
) /* xact may use reserved blks */
1061 xfs_mount_t
*mp
; /* mount structure */
1062 xfs_trans_t
*tp
; /* transaction pointer */
1063 int blks
; /* space reservation */
1064 int version
= 1; /* superblock attr version */
1065 int logflags
; /* logging flags */
1066 int error
; /* error return value */
1068 ASSERT(XFS_IFORK_Q(ip
) == 0);
1071 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1073 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1075 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_addafork
, blks
, 0,
1076 rsvd
? XFS_TRANS_RESERVE
: 0, &tp
);
1080 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1081 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1082 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1083 XFS_QMOPT_RES_REGBLKS
);
1086 if (XFS_IFORK_Q(ip
))
1089 xfs_trans_ijoin(tp
, ip
, 0);
1090 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1091 error
= xfs_bmap_set_attrforkoff(ip
, size
, &version
);
1094 ASSERT(ip
->i_afp
== NULL
);
1096 ip
->i_afp
= kmem_cache_zalloc(xfs_ifork_zone
,
1097 GFP_KERNEL
| __GFP_NOFAIL
);
1099 ip
->i_afp
->if_format
= XFS_DINODE_FMT_EXTENTS
;
1100 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1102 switch (ip
->i_df
.if_format
) {
1103 case XFS_DINODE_FMT_LOCAL
:
1104 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &logflags
);
1106 case XFS_DINODE_FMT_EXTENTS
:
1107 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &logflags
);
1109 case XFS_DINODE_FMT_BTREE
:
1110 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &logflags
);
1117 xfs_trans_log_inode(tp
, ip
, logflags
);
1120 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1121 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1122 bool log_sb
= false;
1124 spin_lock(&mp
->m_sb_lock
);
1125 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1126 xfs_sb_version_addattr(&mp
->m_sb
);
1129 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1130 xfs_sb_version_addattr2(&mp
->m_sb
);
1133 spin_unlock(&mp
->m_sb_lock
);
1138 error
= xfs_trans_commit(tp
);
1139 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1143 xfs_trans_cancel(tp
);
1144 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1149 * Internal and external extent tree search functions.
1152 struct xfs_iread_state
{
1153 struct xfs_iext_cursor icur
;
1154 xfs_extnum_t loaded
;
1157 /* Stuff every bmbt record from this block into the incore extent map. */
1159 xfs_iread_bmbt_block(
1160 struct xfs_btree_cur
*cur
,
1164 struct xfs_iread_state
*ir
= priv
;
1165 struct xfs_mount
*mp
= cur
->bc_mp
;
1166 struct xfs_inode
*ip
= cur
->bc_ino
.ip
;
1167 struct xfs_btree_block
*block
;
1169 struct xfs_bmbt_rec
*frp
;
1170 xfs_extnum_t num_recs
;
1172 int whichfork
= cur
->bc_ino
.whichfork
;
1173 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1175 block
= xfs_btree_get_block(cur
, level
, &bp
);
1177 /* Abort if we find more records than nextents. */
1178 num_recs
= xfs_btree_get_numrecs(block
);
1179 if (unlikely(ir
->loaded
+ num_recs
> ifp
->if_nextents
)) {
1180 xfs_warn(ip
->i_mount
, "corrupt dinode %llu, (btree extents).",
1181 (unsigned long long)ip
->i_ino
);
1182 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
, __func__
, block
,
1183 sizeof(*block
), __this_address
);
1184 return -EFSCORRUPTED
;
1187 /* Copy records into the incore cache. */
1188 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1189 for (j
= 0; j
< num_recs
; j
++, frp
++, ir
->loaded
++) {
1190 struct xfs_bmbt_irec
new;
1193 xfs_bmbt_disk_get_all(frp
, &new);
1194 fa
= xfs_bmap_validate_extent(ip
, whichfork
, &new);
1196 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
,
1197 "xfs_iread_extents(2)", frp
,
1199 return -EFSCORRUPTED
;
1201 xfs_iext_insert(ip
, &ir
->icur
, &new,
1202 xfs_bmap_fork_to_state(whichfork
));
1203 trace_xfs_read_extent(ip
, &ir
->icur
,
1204 xfs_bmap_fork_to_state(whichfork
), _THIS_IP_
);
1205 xfs_iext_next(ifp
, &ir
->icur
);
1212 * Read in extents from a btree-format inode.
1216 struct xfs_trans
*tp
,
1217 struct xfs_inode
*ip
,
1220 struct xfs_iread_state ir
;
1221 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1222 struct xfs_mount
*mp
= ip
->i_mount
;
1223 struct xfs_btree_cur
*cur
;
1226 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1228 if (XFS_IS_CORRUPT(mp
, ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)) {
1229 error
= -EFSCORRUPTED
;
1234 xfs_iext_first(ifp
, &ir
.icur
);
1235 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
1236 error
= xfs_btree_visit_blocks(cur
, xfs_iread_bmbt_block
,
1237 XFS_BTREE_VISIT_RECORDS
, &ir
);
1238 xfs_btree_del_cursor(cur
, error
);
1242 if (XFS_IS_CORRUPT(mp
, ir
.loaded
!= ifp
->if_nextents
)) {
1243 error
= -EFSCORRUPTED
;
1246 ASSERT(ir
.loaded
== xfs_iext_count(ifp
));
1248 ifp
->if_flags
|= XFS_IFEXTENTS
;
1251 xfs_iext_destroy(ifp
);
1256 * Returns the relative block number of the first unused block(s) in the given
1257 * fork with at least "len" logically contiguous blocks free. This is the
1258 * lowest-address hole if the fork has holes, else the first block past the end
1259 * of fork. Return 0 if the fork is currently local (in-inode).
1262 xfs_bmap_first_unused(
1263 struct xfs_trans
*tp
, /* transaction pointer */
1264 struct xfs_inode
*ip
, /* incore inode */
1265 xfs_extlen_t len
, /* size of hole to find */
1266 xfs_fileoff_t
*first_unused
, /* unused block */
1267 int whichfork
) /* data or attr fork */
1269 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1270 struct xfs_bmbt_irec got
;
1271 struct xfs_iext_cursor icur
;
1272 xfs_fileoff_t lastaddr
= 0;
1273 xfs_fileoff_t lowest
, max
;
1276 if (ifp
->if_format
== XFS_DINODE_FMT_LOCAL
) {
1281 ASSERT(xfs_ifork_has_extents(ifp
));
1283 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1284 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1289 lowest
= max
= *first_unused
;
1290 for_each_xfs_iext(ifp
, &icur
, &got
) {
1292 * See if the hole before this extent will work.
1294 if (got
.br_startoff
>= lowest
+ len
&&
1295 got
.br_startoff
- max
>= len
)
1297 lastaddr
= got
.br_startoff
+ got
.br_blockcount
;
1298 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1301 *first_unused
= max
;
1306 * Returns the file-relative block number of the last block - 1 before
1307 * last_block (input value) in the file.
1308 * This is not based on i_size, it is based on the extent records.
1309 * Returns 0 for local files, as they do not have extent records.
1312 xfs_bmap_last_before(
1313 struct xfs_trans
*tp
, /* transaction pointer */
1314 struct xfs_inode
*ip
, /* incore inode */
1315 xfs_fileoff_t
*last_block
, /* last block */
1316 int whichfork
) /* data or attr fork */
1318 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1319 struct xfs_bmbt_irec got
;
1320 struct xfs_iext_cursor icur
;
1323 switch (ifp
->if_format
) {
1324 case XFS_DINODE_FMT_LOCAL
:
1327 case XFS_DINODE_FMT_BTREE
:
1328 case XFS_DINODE_FMT_EXTENTS
:
1332 return -EFSCORRUPTED
;
1335 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1336 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1341 if (!xfs_iext_lookup_extent_before(ip
, ifp
, last_block
, &icur
, &got
))
1347 xfs_bmap_last_extent(
1348 struct xfs_trans
*tp
,
1349 struct xfs_inode
*ip
,
1351 struct xfs_bmbt_irec
*rec
,
1354 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1355 struct xfs_iext_cursor icur
;
1358 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1359 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1364 xfs_iext_last(ifp
, &icur
);
1365 if (!xfs_iext_get_extent(ifp
, &icur
, rec
))
1373 * Check the last inode extent to determine whether this allocation will result
1374 * in blocks being allocated at the end of the file. When we allocate new data
1375 * blocks at the end of the file which do not start at the previous data block,
1376 * we will try to align the new blocks at stripe unit boundaries.
1378 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1379 * at, or past the EOF.
1383 struct xfs_bmalloca
*bma
,
1386 struct xfs_bmbt_irec rec
;
1391 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1402 * Check if we are allocation or past the last extent, or at least into
1403 * the last delayed allocated extent.
1405 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1406 (bma
->offset
>= rec
.br_startoff
&&
1407 isnullstartblock(rec
.br_startblock
));
1412 * Returns the file-relative block number of the first block past eof in
1413 * the file. This is not based on i_size, it is based on the extent records.
1414 * Returns 0 for local files, as they do not have extent records.
1417 xfs_bmap_last_offset(
1418 struct xfs_inode
*ip
,
1419 xfs_fileoff_t
*last_block
,
1422 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1423 struct xfs_bmbt_irec rec
;
1429 if (ifp
->if_format
== XFS_DINODE_FMT_LOCAL
)
1432 if (XFS_IS_CORRUPT(ip
->i_mount
, !xfs_ifork_has_extents(ifp
)))
1433 return -EFSCORRUPTED
;
1435 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1436 if (error
|| is_empty
)
1439 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1444 * Returns whether the selected fork of the inode has exactly one
1445 * block or not. For the data fork we check this matches di_size,
1446 * implying the file's range is 0..bsize-1.
1448 int /* 1=>1 block, 0=>otherwise */
1450 struct xfs_inode
*ip
, /* incore inode */
1451 int whichfork
) /* data or attr fork */
1453 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1454 int rval
; /* return value */
1455 struct xfs_bmbt_irec s
; /* internal version of extent */
1456 struct xfs_iext_cursor icur
;
1459 if (whichfork
== XFS_DATA_FORK
)
1460 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1462 if (ifp
->if_nextents
!= 1)
1464 if (ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
1466 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1467 xfs_iext_first(ifp
, &icur
);
1468 xfs_iext_get_extent(ifp
, &icur
, &s
);
1469 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1470 if (rval
&& whichfork
== XFS_DATA_FORK
)
1471 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1476 * Extent tree manipulation functions used during allocation.
1480 * Convert a delayed allocation to a real allocation.
1482 STATIC
int /* error */
1483 xfs_bmap_add_extent_delay_real(
1484 struct xfs_bmalloca
*bma
,
1487 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
1488 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1489 struct xfs_bmbt_irec
*new = &bma
->got
;
1490 int error
; /* error return value */
1491 int i
; /* temp state */
1492 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1493 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1494 /* left is 0, right is 1, prev is 2 */
1495 int rval
=0; /* return value (logging flags) */
1496 int state
= xfs_bmap_fork_to_state(whichfork
);
1497 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1498 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1499 xfs_filblks_t temp
=0; /* value for da_new calculations */
1500 int tmp_rval
; /* partial logging flags */
1501 struct xfs_bmbt_irec old
;
1503 ASSERT(whichfork
!= XFS_ATTR_FORK
);
1504 ASSERT(!isnullstartblock(new->br_startblock
));
1506 (bma
->cur
->bc_ino
.flags
& XFS_BTCUR_BMBT_WASDEL
));
1508 XFS_STATS_INC(mp
, xs_add_exlist
);
1515 * Set up a bunch of variables to make the tests simpler.
1517 xfs_iext_get_extent(ifp
, &bma
->icur
, &PREV
);
1518 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1519 ASSERT(isnullstartblock(PREV
.br_startblock
));
1520 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1521 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1523 da_old
= startblockval(PREV
.br_startblock
);
1527 * Set flags determining what part of the previous delayed allocation
1528 * extent is being replaced by a real allocation.
1530 if (PREV
.br_startoff
== new->br_startoff
)
1531 state
|= BMAP_LEFT_FILLING
;
1532 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1533 state
|= BMAP_RIGHT_FILLING
;
1536 * Check and set flags if this segment has a left neighbor.
1537 * Don't set contiguous if the combined extent would be too large.
1539 if (xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &LEFT
)) {
1540 state
|= BMAP_LEFT_VALID
;
1541 if (isnullstartblock(LEFT
.br_startblock
))
1542 state
|= BMAP_LEFT_DELAY
;
1545 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1546 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1547 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1548 LEFT
.br_state
== new->br_state
&&
1549 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1550 state
|= BMAP_LEFT_CONTIG
;
1553 * Check and set flags if this segment has a right neighbor.
1554 * Don't set contiguous if the combined extent would be too large.
1555 * Also check for all-three-contiguous being too large.
1557 if (xfs_iext_peek_next_extent(ifp
, &bma
->icur
, &RIGHT
)) {
1558 state
|= BMAP_RIGHT_VALID
;
1559 if (isnullstartblock(RIGHT
.br_startblock
))
1560 state
|= BMAP_RIGHT_DELAY
;
1563 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1564 new_endoff
== RIGHT
.br_startoff
&&
1565 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1566 new->br_state
== RIGHT
.br_state
&&
1567 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1568 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1569 BMAP_RIGHT_FILLING
)) !=
1570 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1571 BMAP_RIGHT_FILLING
) ||
1572 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1574 state
|= BMAP_RIGHT_CONTIG
;
1578 * Switch out based on the FILLING and CONTIG state bits.
1580 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1581 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1582 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1583 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1585 * Filling in all of a previously delayed allocation extent.
1586 * The left and right neighbors are both contiguous with new.
1588 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
1590 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1591 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1592 xfs_iext_prev(ifp
, &bma
->icur
);
1593 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1596 if (bma
->cur
== NULL
)
1597 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1599 rval
= XFS_ILOG_CORE
;
1600 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1603 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1604 error
= -EFSCORRUPTED
;
1607 error
= xfs_btree_delete(bma
->cur
, &i
);
1610 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1611 error
= -EFSCORRUPTED
;
1614 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1617 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1618 error
= -EFSCORRUPTED
;
1621 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1627 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1629 * Filling in all of a previously delayed allocation extent.
1630 * The left neighbor is contiguous, the right is not.
1633 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
1635 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1636 xfs_iext_prev(ifp
, &bma
->icur
);
1637 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1639 if (bma
->cur
== NULL
)
1640 rval
= XFS_ILOG_DEXT
;
1643 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1646 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1647 error
= -EFSCORRUPTED
;
1650 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1656 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1658 * Filling in all of a previously delayed allocation extent.
1659 * The right neighbor is contiguous, the left is not. Take care
1660 * with delay -> unwritten extent allocation here because the
1661 * delalloc record we are overwriting is always written.
1663 PREV
.br_startblock
= new->br_startblock
;
1664 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
1665 PREV
.br_state
= new->br_state
;
1667 xfs_iext_next(ifp
, &bma
->icur
);
1668 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1669 xfs_iext_prev(ifp
, &bma
->icur
);
1670 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1672 if (bma
->cur
== NULL
)
1673 rval
= XFS_ILOG_DEXT
;
1676 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1679 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1680 error
= -EFSCORRUPTED
;
1683 error
= xfs_bmbt_update(bma
->cur
, &PREV
);
1689 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1691 * Filling in all of a previously delayed allocation extent.
1692 * Neither the left nor right neighbors are contiguous with
1695 PREV
.br_startblock
= new->br_startblock
;
1696 PREV
.br_state
= new->br_state
;
1697 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1700 if (bma
->cur
== NULL
)
1701 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1703 rval
= XFS_ILOG_CORE
;
1704 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1707 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1708 error
= -EFSCORRUPTED
;
1711 error
= xfs_btree_insert(bma
->cur
, &i
);
1714 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1715 error
= -EFSCORRUPTED
;
1721 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1723 * Filling in the first part of a previous delayed allocation.
1724 * The left neighbor is contiguous.
1727 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1728 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1729 startblockval(PREV
.br_startblock
));
1731 LEFT
.br_blockcount
+= new->br_blockcount
;
1733 PREV
.br_blockcount
= temp
;
1734 PREV
.br_startoff
+= new->br_blockcount
;
1735 PREV
.br_startblock
= nullstartblock(da_new
);
1737 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1738 xfs_iext_prev(ifp
, &bma
->icur
);
1739 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1741 if (bma
->cur
== NULL
)
1742 rval
= XFS_ILOG_DEXT
;
1745 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1748 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1749 error
= -EFSCORRUPTED
;
1752 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1758 case BMAP_LEFT_FILLING
:
1760 * Filling in the first part of a previous delayed allocation.
1761 * The left neighbor is not contiguous.
1763 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1766 if (bma
->cur
== NULL
)
1767 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1769 rval
= XFS_ILOG_CORE
;
1770 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1773 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1774 error
= -EFSCORRUPTED
;
1777 error
= xfs_btree_insert(bma
->cur
, &i
);
1780 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1781 error
= -EFSCORRUPTED
;
1786 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1787 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1788 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1794 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1795 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1796 startblockval(PREV
.br_startblock
) -
1797 (bma
->cur
? bma
->cur
->bc_ino
.allocated
: 0));
1799 PREV
.br_startoff
= new_endoff
;
1800 PREV
.br_blockcount
= temp
;
1801 PREV
.br_startblock
= nullstartblock(da_new
);
1802 xfs_iext_next(ifp
, &bma
->icur
);
1803 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1804 xfs_iext_prev(ifp
, &bma
->icur
);
1807 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1809 * Filling in the last part of a previous delayed allocation.
1810 * The right neighbor is contiguous with the new allocation.
1813 RIGHT
.br_startoff
= new->br_startoff
;
1814 RIGHT
.br_startblock
= new->br_startblock
;
1815 RIGHT
.br_blockcount
+= new->br_blockcount
;
1817 if (bma
->cur
== NULL
)
1818 rval
= XFS_ILOG_DEXT
;
1821 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1824 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1825 error
= -EFSCORRUPTED
;
1828 error
= xfs_bmbt_update(bma
->cur
, &RIGHT
);
1833 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1834 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1835 startblockval(PREV
.br_startblock
));
1837 PREV
.br_blockcount
= temp
;
1838 PREV
.br_startblock
= nullstartblock(da_new
);
1840 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1841 xfs_iext_next(ifp
, &bma
->icur
);
1842 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &RIGHT
);
1845 case BMAP_RIGHT_FILLING
:
1847 * Filling in the last part of a previous delayed allocation.
1848 * The right neighbor is not contiguous.
1850 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1853 if (bma
->cur
== NULL
)
1854 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1856 rval
= XFS_ILOG_CORE
;
1857 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1860 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1861 error
= -EFSCORRUPTED
;
1864 error
= xfs_btree_insert(bma
->cur
, &i
);
1867 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1868 error
= -EFSCORRUPTED
;
1873 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1874 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1875 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1881 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1882 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1883 startblockval(PREV
.br_startblock
) -
1884 (bma
->cur
? bma
->cur
->bc_ino
.allocated
: 0));
1886 PREV
.br_startblock
= nullstartblock(da_new
);
1887 PREV
.br_blockcount
= temp
;
1888 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1889 xfs_iext_next(ifp
, &bma
->icur
);
1894 * Filling in the middle part of a previous delayed allocation.
1895 * Contiguity is impossible here.
1896 * This case is avoided almost all the time.
1898 * We start with a delayed allocation:
1900 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1903 * and we are allocating:
1904 * +rrrrrrrrrrrrrrrrr+
1907 * and we set it up for insertion as:
1908 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1910 * PREV @ idx LEFT RIGHT
1911 * inserted at idx + 1
1915 /* LEFT is the new middle */
1918 /* RIGHT is the new right */
1919 RIGHT
.br_state
= PREV
.br_state
;
1920 RIGHT
.br_startoff
= new_endoff
;
1921 RIGHT
.br_blockcount
=
1922 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
1923 RIGHT
.br_startblock
=
1924 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1925 RIGHT
.br_blockcount
));
1928 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
1929 PREV
.br_startblock
=
1930 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1931 PREV
.br_blockcount
));
1932 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1934 xfs_iext_next(ifp
, &bma
->icur
);
1935 xfs_iext_insert(bma
->ip
, &bma
->icur
, &RIGHT
, state
);
1936 xfs_iext_insert(bma
->ip
, &bma
->icur
, &LEFT
, state
);
1939 if (bma
->cur
== NULL
)
1940 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1942 rval
= XFS_ILOG_CORE
;
1943 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1946 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
1947 error
= -EFSCORRUPTED
;
1950 error
= xfs_btree_insert(bma
->cur
, &i
);
1953 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
1954 error
= -EFSCORRUPTED
;
1959 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1960 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1961 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1967 da_new
= startblockval(PREV
.br_startblock
) +
1968 startblockval(RIGHT
.br_startblock
);
1971 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1972 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1973 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1974 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1975 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1976 case BMAP_LEFT_CONTIG
:
1977 case BMAP_RIGHT_CONTIG
:
1979 * These cases are all impossible.
1984 /* add reverse mapping unless caller opted out */
1985 if (!(bma
->flags
& XFS_BMAPI_NORMAP
))
1986 xfs_rmap_map_extent(bma
->tp
, bma
->ip
, whichfork
, new);
1988 /* convert to a btree if necessary */
1989 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1990 int tmp_logflags
; /* partial log flag return val */
1992 ASSERT(bma
->cur
== NULL
);
1993 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1994 &bma
->cur
, da_old
> 0, &tmp_logflags
,
1996 bma
->logflags
|= tmp_logflags
;
2001 if (da_new
!= da_old
)
2002 xfs_mod_delalloc(mp
, (int64_t)da_new
- da_old
);
2005 da_new
+= bma
->cur
->bc_ino
.allocated
;
2006 bma
->cur
->bc_ino
.allocated
= 0;
2009 /* adjust for changes in reserved delayed indirect blocks */
2010 if (da_new
!= da_old
) {
2011 ASSERT(state
== 0 || da_new
< da_old
);
2012 error
= xfs_mod_fdblocks(mp
, (int64_t)(da_old
- da_new
),
2016 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
2018 if (whichfork
!= XFS_COW_FORK
)
2019 bma
->logflags
|= rval
;
2027 * Convert an unwritten allocation to a real allocation or vice versa.
2030 xfs_bmap_add_extent_unwritten_real(
2031 struct xfs_trans
*tp
,
2032 xfs_inode_t
*ip
, /* incore inode pointer */
2034 struct xfs_iext_cursor
*icur
,
2035 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2036 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2037 int *logflagsp
) /* inode logging flags */
2039 xfs_btree_cur_t
*cur
; /* btree cursor */
2040 int error
; /* error return value */
2041 int i
; /* temp state */
2042 struct xfs_ifork
*ifp
; /* inode fork pointer */
2043 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2044 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2045 /* left is 0, right is 1, prev is 2 */
2046 int rval
=0; /* return value (logging flags) */
2047 int state
= xfs_bmap_fork_to_state(whichfork
);
2048 struct xfs_mount
*mp
= ip
->i_mount
;
2049 struct xfs_bmbt_irec old
;
2054 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2056 ASSERT(!isnullstartblock(new->br_startblock
));
2058 XFS_STATS_INC(mp
, xs_add_exlist
);
2065 * Set up a bunch of variables to make the tests simpler.
2068 xfs_iext_get_extent(ifp
, icur
, &PREV
);
2069 ASSERT(new->br_state
!= PREV
.br_state
);
2070 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2071 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2072 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2075 * Set flags determining what part of the previous oldext allocation
2076 * extent is being replaced by a newext allocation.
2078 if (PREV
.br_startoff
== new->br_startoff
)
2079 state
|= BMAP_LEFT_FILLING
;
2080 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2081 state
|= BMAP_RIGHT_FILLING
;
2084 * Check and set flags if this segment has a left neighbor.
2085 * Don't set contiguous if the combined extent would be too large.
2087 if (xfs_iext_peek_prev_extent(ifp
, icur
, &LEFT
)) {
2088 state
|= BMAP_LEFT_VALID
;
2089 if (isnullstartblock(LEFT
.br_startblock
))
2090 state
|= BMAP_LEFT_DELAY
;
2093 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2094 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2095 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2096 LEFT
.br_state
== new->br_state
&&
2097 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2098 state
|= BMAP_LEFT_CONTIG
;
2101 * Check and set flags if this segment has a right neighbor.
2102 * Don't set contiguous if the combined extent would be too large.
2103 * Also check for all-three-contiguous being too large.
2105 if (xfs_iext_peek_next_extent(ifp
, icur
, &RIGHT
)) {
2106 state
|= BMAP_RIGHT_VALID
;
2107 if (isnullstartblock(RIGHT
.br_startblock
))
2108 state
|= BMAP_RIGHT_DELAY
;
2111 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2112 new_endoff
== RIGHT
.br_startoff
&&
2113 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2114 new->br_state
== RIGHT
.br_state
&&
2115 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2116 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2117 BMAP_RIGHT_FILLING
)) !=
2118 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2119 BMAP_RIGHT_FILLING
) ||
2120 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2122 state
|= BMAP_RIGHT_CONTIG
;
2125 * Switch out based on the FILLING and CONTIG state bits.
2127 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2128 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2129 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2130 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2132 * Setting all of a previous oldext extent to newext.
2133 * The left and right neighbors are both contiguous with new.
2135 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
2137 xfs_iext_remove(ip
, icur
, state
);
2138 xfs_iext_remove(ip
, icur
, state
);
2139 xfs_iext_prev(ifp
, icur
);
2140 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2141 ifp
->if_nextents
-= 2;
2143 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2145 rval
= XFS_ILOG_CORE
;
2146 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2149 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2150 error
= -EFSCORRUPTED
;
2153 if ((error
= xfs_btree_delete(cur
, &i
)))
2155 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2156 error
= -EFSCORRUPTED
;
2159 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2161 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2162 error
= -EFSCORRUPTED
;
2165 if ((error
= xfs_btree_delete(cur
, &i
)))
2167 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2168 error
= -EFSCORRUPTED
;
2171 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2173 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2174 error
= -EFSCORRUPTED
;
2177 error
= xfs_bmbt_update(cur
, &LEFT
);
2183 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2185 * Setting all of a previous oldext extent to newext.
2186 * The left neighbor is contiguous, the right is not.
2188 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
2190 xfs_iext_remove(ip
, icur
, state
);
2191 xfs_iext_prev(ifp
, icur
);
2192 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2195 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2197 rval
= XFS_ILOG_CORE
;
2198 error
= xfs_bmbt_lookup_eq(cur
, &PREV
, &i
);
2201 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2202 error
= -EFSCORRUPTED
;
2205 if ((error
= xfs_btree_delete(cur
, &i
)))
2207 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2208 error
= -EFSCORRUPTED
;
2211 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2213 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2214 error
= -EFSCORRUPTED
;
2217 error
= xfs_bmbt_update(cur
, &LEFT
);
2223 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2225 * Setting all of a previous oldext extent to newext.
2226 * The right neighbor is contiguous, the left is not.
2228 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
2229 PREV
.br_state
= new->br_state
;
2231 xfs_iext_next(ifp
, icur
);
2232 xfs_iext_remove(ip
, icur
, state
);
2233 xfs_iext_prev(ifp
, icur
);
2234 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2238 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2240 rval
= XFS_ILOG_CORE
;
2241 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2244 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2245 error
= -EFSCORRUPTED
;
2248 if ((error
= xfs_btree_delete(cur
, &i
)))
2250 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2251 error
= -EFSCORRUPTED
;
2254 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2256 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2257 error
= -EFSCORRUPTED
;
2260 error
= xfs_bmbt_update(cur
, &PREV
);
2266 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2268 * Setting all of a previous oldext extent to newext.
2269 * Neither the left nor right neighbors are contiguous with
2272 PREV
.br_state
= new->br_state
;
2273 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2276 rval
= XFS_ILOG_DEXT
;
2279 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2282 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2283 error
= -EFSCORRUPTED
;
2286 error
= xfs_bmbt_update(cur
, &PREV
);
2292 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2294 * Setting the first part of a previous oldext extent to newext.
2295 * The left neighbor is contiguous.
2297 LEFT
.br_blockcount
+= new->br_blockcount
;
2300 PREV
.br_startoff
+= new->br_blockcount
;
2301 PREV
.br_startblock
+= new->br_blockcount
;
2302 PREV
.br_blockcount
-= new->br_blockcount
;
2304 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2305 xfs_iext_prev(ifp
, icur
);
2306 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2309 rval
= XFS_ILOG_DEXT
;
2312 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2315 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2316 error
= -EFSCORRUPTED
;
2319 error
= xfs_bmbt_update(cur
, &PREV
);
2322 error
= xfs_btree_decrement(cur
, 0, &i
);
2325 error
= xfs_bmbt_update(cur
, &LEFT
);
2331 case BMAP_LEFT_FILLING
:
2333 * Setting the first part of a previous oldext extent to newext.
2334 * The left neighbor is not contiguous.
2337 PREV
.br_startoff
+= new->br_blockcount
;
2338 PREV
.br_startblock
+= new->br_blockcount
;
2339 PREV
.br_blockcount
-= new->br_blockcount
;
2341 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2342 xfs_iext_insert(ip
, icur
, new, state
);
2346 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2348 rval
= XFS_ILOG_CORE
;
2349 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2352 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2353 error
= -EFSCORRUPTED
;
2356 error
= xfs_bmbt_update(cur
, &PREV
);
2359 cur
->bc_rec
.b
= *new;
2360 if ((error
= xfs_btree_insert(cur
, &i
)))
2362 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2363 error
= -EFSCORRUPTED
;
2369 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2371 * Setting the last part of a previous oldext extent to newext.
2372 * The right neighbor is contiguous with the new allocation.
2375 PREV
.br_blockcount
-= new->br_blockcount
;
2377 RIGHT
.br_startoff
= new->br_startoff
;
2378 RIGHT
.br_startblock
= new->br_startblock
;
2379 RIGHT
.br_blockcount
+= new->br_blockcount
;
2381 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2382 xfs_iext_next(ifp
, icur
);
2383 xfs_iext_update_extent(ip
, state
, icur
, &RIGHT
);
2386 rval
= XFS_ILOG_DEXT
;
2389 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2392 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2393 error
= -EFSCORRUPTED
;
2396 error
= xfs_bmbt_update(cur
, &PREV
);
2399 error
= xfs_btree_increment(cur
, 0, &i
);
2402 error
= xfs_bmbt_update(cur
, &RIGHT
);
2408 case BMAP_RIGHT_FILLING
:
2410 * Setting the last part of a previous oldext extent to newext.
2411 * The right neighbor is not contiguous.
2414 PREV
.br_blockcount
-= new->br_blockcount
;
2416 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2417 xfs_iext_next(ifp
, icur
);
2418 xfs_iext_insert(ip
, icur
, new, state
);
2422 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2424 rval
= XFS_ILOG_CORE
;
2425 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2428 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2429 error
= -EFSCORRUPTED
;
2432 error
= xfs_bmbt_update(cur
, &PREV
);
2435 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2438 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2439 error
= -EFSCORRUPTED
;
2442 if ((error
= xfs_btree_insert(cur
, &i
)))
2444 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2445 error
= -EFSCORRUPTED
;
2453 * Setting the middle part of a previous oldext extent to
2454 * newext. Contiguity is impossible here.
2455 * One extent becomes three extents.
2458 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
2461 r
[1].br_startoff
= new_endoff
;
2462 r
[1].br_blockcount
=
2463 old
.br_startoff
+ old
.br_blockcount
- new_endoff
;
2464 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2465 r
[1].br_state
= PREV
.br_state
;
2467 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2468 xfs_iext_next(ifp
, icur
);
2469 xfs_iext_insert(ip
, icur
, &r
[1], state
);
2470 xfs_iext_insert(ip
, icur
, &r
[0], state
);
2471 ifp
->if_nextents
+= 2;
2474 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2476 rval
= XFS_ILOG_CORE
;
2477 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2480 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2481 error
= -EFSCORRUPTED
;
2484 /* new right extent - oldext */
2485 error
= xfs_bmbt_update(cur
, &r
[1]);
2488 /* new left extent - oldext */
2489 cur
->bc_rec
.b
= PREV
;
2490 if ((error
= xfs_btree_insert(cur
, &i
)))
2492 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2493 error
= -EFSCORRUPTED
;
2497 * Reset the cursor to the position of the new extent
2498 * we are about to insert as we can't trust it after
2499 * the previous insert.
2501 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2504 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2505 error
= -EFSCORRUPTED
;
2508 /* new middle extent - newext */
2509 if ((error
= xfs_btree_insert(cur
, &i
)))
2511 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2512 error
= -EFSCORRUPTED
;
2518 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2519 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2520 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2521 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2522 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2523 case BMAP_LEFT_CONTIG
:
2524 case BMAP_RIGHT_CONTIG
:
2526 * These cases are all impossible.
2531 /* update reverse mappings */
2532 xfs_rmap_convert_extent(mp
, tp
, ip
, whichfork
, new);
2534 /* convert to a btree if necessary */
2535 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2536 int tmp_logflags
; /* partial log flag return val */
2538 ASSERT(cur
== NULL
);
2539 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
2540 &tmp_logflags
, whichfork
);
2541 *logflagsp
|= tmp_logflags
;
2546 /* clear out the allocated field, done with it now in any case. */
2548 cur
->bc_ino
.allocated
= 0;
2552 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
2562 * Convert a hole to a delayed allocation.
2565 xfs_bmap_add_extent_hole_delay(
2566 xfs_inode_t
*ip
, /* incore inode pointer */
2568 struct xfs_iext_cursor
*icur
,
2569 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2571 struct xfs_ifork
*ifp
; /* inode fork pointer */
2572 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2573 xfs_filblks_t newlen
=0; /* new indirect size */
2574 xfs_filblks_t oldlen
=0; /* old indirect size */
2575 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2576 int state
= xfs_bmap_fork_to_state(whichfork
);
2577 xfs_filblks_t temp
; /* temp for indirect calculations */
2579 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2580 ASSERT(isnullstartblock(new->br_startblock
));
2583 * Check and set flags if this segment has a left neighbor
2585 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2586 state
|= BMAP_LEFT_VALID
;
2587 if (isnullstartblock(left
.br_startblock
))
2588 state
|= BMAP_LEFT_DELAY
;
2592 * Check and set flags if the current (right) segment exists.
2593 * If it doesn't exist, we're converting the hole at end-of-file.
2595 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2596 state
|= BMAP_RIGHT_VALID
;
2597 if (isnullstartblock(right
.br_startblock
))
2598 state
|= BMAP_RIGHT_DELAY
;
2602 * Set contiguity flags on the left and right neighbors.
2603 * Don't let extents get too large, even if the pieces are contiguous.
2605 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2606 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2607 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2608 state
|= BMAP_LEFT_CONTIG
;
2610 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2611 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2612 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2613 (!(state
& BMAP_LEFT_CONTIG
) ||
2614 (left
.br_blockcount
+ new->br_blockcount
+
2615 right
.br_blockcount
<= MAXEXTLEN
)))
2616 state
|= BMAP_RIGHT_CONTIG
;
2619 * Switch out based on the contiguity flags.
2621 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2622 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2624 * New allocation is contiguous with delayed allocations
2625 * on the left and on the right.
2626 * Merge all three into a single extent record.
2628 temp
= left
.br_blockcount
+ new->br_blockcount
+
2629 right
.br_blockcount
;
2631 oldlen
= startblockval(left
.br_startblock
) +
2632 startblockval(new->br_startblock
) +
2633 startblockval(right
.br_startblock
);
2634 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2636 left
.br_startblock
= nullstartblock(newlen
);
2637 left
.br_blockcount
= temp
;
2639 xfs_iext_remove(ip
, icur
, state
);
2640 xfs_iext_prev(ifp
, icur
);
2641 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2644 case BMAP_LEFT_CONTIG
:
2646 * New allocation is contiguous with a delayed allocation
2648 * Merge the new allocation with the left neighbor.
2650 temp
= left
.br_blockcount
+ new->br_blockcount
;
2652 oldlen
= startblockval(left
.br_startblock
) +
2653 startblockval(new->br_startblock
);
2654 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2656 left
.br_blockcount
= temp
;
2657 left
.br_startblock
= nullstartblock(newlen
);
2659 xfs_iext_prev(ifp
, icur
);
2660 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2663 case BMAP_RIGHT_CONTIG
:
2665 * New allocation is contiguous with a delayed allocation
2667 * Merge the new allocation with the right neighbor.
2669 temp
= new->br_blockcount
+ right
.br_blockcount
;
2670 oldlen
= startblockval(new->br_startblock
) +
2671 startblockval(right
.br_startblock
);
2672 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2674 right
.br_startoff
= new->br_startoff
;
2675 right
.br_startblock
= nullstartblock(newlen
);
2676 right
.br_blockcount
= temp
;
2677 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2682 * New allocation is not contiguous with another
2683 * delayed allocation.
2684 * Insert a new entry.
2686 oldlen
= newlen
= 0;
2687 xfs_iext_insert(ip
, icur
, new, state
);
2690 if (oldlen
!= newlen
) {
2691 ASSERT(oldlen
> newlen
);
2692 xfs_mod_fdblocks(ip
->i_mount
, (int64_t)(oldlen
- newlen
),
2695 * Nothing to do for disk quota accounting here.
2697 xfs_mod_delalloc(ip
->i_mount
, (int64_t)newlen
- oldlen
);
2702 * Convert a hole to a real allocation.
2704 STATIC
int /* error */
2705 xfs_bmap_add_extent_hole_real(
2706 struct xfs_trans
*tp
,
2707 struct xfs_inode
*ip
,
2709 struct xfs_iext_cursor
*icur
,
2710 struct xfs_btree_cur
**curp
,
2711 struct xfs_bmbt_irec
*new,
2715 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2716 struct xfs_mount
*mp
= ip
->i_mount
;
2717 struct xfs_btree_cur
*cur
= *curp
;
2718 int error
; /* error return value */
2719 int i
; /* temp state */
2720 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2721 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2722 int rval
=0; /* return value (logging flags) */
2723 int state
= xfs_bmap_fork_to_state(whichfork
);
2724 struct xfs_bmbt_irec old
;
2726 ASSERT(!isnullstartblock(new->br_startblock
));
2727 ASSERT(!cur
|| !(cur
->bc_ino
.flags
& XFS_BTCUR_BMBT_WASDEL
));
2729 XFS_STATS_INC(mp
, xs_add_exlist
);
2732 * Check and set flags if this segment has a left neighbor.
2734 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2735 state
|= BMAP_LEFT_VALID
;
2736 if (isnullstartblock(left
.br_startblock
))
2737 state
|= BMAP_LEFT_DELAY
;
2741 * Check and set flags if this segment has a current value.
2742 * Not true if we're inserting into the "hole" at eof.
2744 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2745 state
|= BMAP_RIGHT_VALID
;
2746 if (isnullstartblock(right
.br_startblock
))
2747 state
|= BMAP_RIGHT_DELAY
;
2751 * We're inserting a real allocation between "left" and "right".
2752 * Set the contiguity flags. Don't let extents get too large.
2754 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2755 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2756 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
2757 left
.br_state
== new->br_state
&&
2758 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2759 state
|= BMAP_LEFT_CONTIG
;
2761 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2762 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2763 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
2764 new->br_state
== right
.br_state
&&
2765 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2766 (!(state
& BMAP_LEFT_CONTIG
) ||
2767 left
.br_blockcount
+ new->br_blockcount
+
2768 right
.br_blockcount
<= MAXEXTLEN
))
2769 state
|= BMAP_RIGHT_CONTIG
;
2773 * Select which case we're in here, and implement it.
2775 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2776 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2778 * New allocation is contiguous with real allocations on the
2779 * left and on the right.
2780 * Merge all three into a single extent record.
2782 left
.br_blockcount
+= new->br_blockcount
+ right
.br_blockcount
;
2784 xfs_iext_remove(ip
, icur
, state
);
2785 xfs_iext_prev(ifp
, icur
);
2786 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2790 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2792 rval
= XFS_ILOG_CORE
;
2793 error
= xfs_bmbt_lookup_eq(cur
, &right
, &i
);
2796 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2797 error
= -EFSCORRUPTED
;
2800 error
= xfs_btree_delete(cur
, &i
);
2803 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2804 error
= -EFSCORRUPTED
;
2807 error
= xfs_btree_decrement(cur
, 0, &i
);
2810 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2811 error
= -EFSCORRUPTED
;
2814 error
= xfs_bmbt_update(cur
, &left
);
2820 case BMAP_LEFT_CONTIG
:
2822 * New allocation is contiguous with a real allocation
2824 * Merge the new allocation with the left neighbor.
2827 left
.br_blockcount
+= new->br_blockcount
;
2829 xfs_iext_prev(ifp
, icur
);
2830 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2833 rval
= xfs_ilog_fext(whichfork
);
2836 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2839 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2840 error
= -EFSCORRUPTED
;
2843 error
= xfs_bmbt_update(cur
, &left
);
2849 case BMAP_RIGHT_CONTIG
:
2851 * New allocation is contiguous with a real allocation
2853 * Merge the new allocation with the right neighbor.
2857 right
.br_startoff
= new->br_startoff
;
2858 right
.br_startblock
= new->br_startblock
;
2859 right
.br_blockcount
+= new->br_blockcount
;
2860 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2863 rval
= xfs_ilog_fext(whichfork
);
2866 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2869 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2870 error
= -EFSCORRUPTED
;
2873 error
= xfs_bmbt_update(cur
, &right
);
2881 * New allocation is not contiguous with another
2883 * Insert a new entry.
2885 xfs_iext_insert(ip
, icur
, new, state
);
2889 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2891 rval
= XFS_ILOG_CORE
;
2892 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2895 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
2896 error
= -EFSCORRUPTED
;
2899 error
= xfs_btree_insert(cur
, &i
);
2902 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
2903 error
= -EFSCORRUPTED
;
2910 /* add reverse mapping unless caller opted out */
2911 if (!(flags
& XFS_BMAPI_NORMAP
))
2912 xfs_rmap_map_extent(tp
, ip
, whichfork
, new);
2914 /* convert to a btree if necessary */
2915 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2916 int tmp_logflags
; /* partial log flag return val */
2918 ASSERT(cur
== NULL
);
2919 error
= xfs_bmap_extents_to_btree(tp
, ip
, curp
, 0,
2920 &tmp_logflags
, whichfork
);
2921 *logflagsp
|= tmp_logflags
;
2927 /* clear out the allocated field, done with it now in any case. */
2929 cur
->bc_ino
.allocated
= 0;
2931 xfs_bmap_check_leaf_extents(cur
, ip
, whichfork
);
2938 * Functions used in the extent read, allocate and remove paths
2942 * Adjust the size of the new extent based on di_extsize and rt extsize.
2945 xfs_bmap_extsize_align(
2947 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
2948 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
2949 xfs_extlen_t extsz
, /* align to this extent size */
2950 int rt
, /* is this a realtime inode? */
2951 int eof
, /* is extent at end-of-file? */
2952 int delay
, /* creating delalloc extent? */
2953 int convert
, /* overwriting unwritten extent? */
2954 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
2955 xfs_extlen_t
*lenp
) /* in/out: aligned length */
2957 xfs_fileoff_t orig_off
; /* original offset */
2958 xfs_extlen_t orig_alen
; /* original length */
2959 xfs_fileoff_t orig_end
; /* original off+len */
2960 xfs_fileoff_t nexto
; /* next file offset */
2961 xfs_fileoff_t prevo
; /* previous file offset */
2962 xfs_fileoff_t align_off
; /* temp for offset */
2963 xfs_extlen_t align_alen
; /* temp for length */
2964 xfs_extlen_t temp
; /* temp for calculations */
2969 orig_off
= align_off
= *offp
;
2970 orig_alen
= align_alen
= *lenp
;
2971 orig_end
= orig_off
+ orig_alen
;
2974 * If this request overlaps an existing extent, then don't
2975 * attempt to perform any additional alignment.
2977 if (!delay
&& !eof
&&
2978 (orig_off
>= gotp
->br_startoff
) &&
2979 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
2984 * If the file offset is unaligned vs. the extent size
2985 * we need to align it. This will be possible unless
2986 * the file was previously written with a kernel that didn't
2987 * perform this alignment, or if a truncate shot us in the
2990 div_u64_rem(orig_off
, extsz
, &temp
);
2996 /* Same adjustment for the end of the requested area. */
2997 temp
= (align_alen
% extsz
);
2999 align_alen
+= extsz
- temp
;
3002 * For large extent hint sizes, the aligned extent might be larger than
3003 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3004 * the length back under MAXEXTLEN. The outer allocation loops handle
3005 * short allocation just fine, so it is safe to do this. We only want to
3006 * do it when we are forced to, though, because it means more allocation
3007 * operations are required.
3009 while (align_alen
> MAXEXTLEN
)
3010 align_alen
-= extsz
;
3011 ASSERT(align_alen
<= MAXEXTLEN
);
3014 * If the previous block overlaps with this proposed allocation
3015 * then move the start forward without adjusting the length.
3017 if (prevp
->br_startoff
!= NULLFILEOFF
) {
3018 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
3019 prevo
= prevp
->br_startoff
;
3021 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
3024 if (align_off
!= orig_off
&& align_off
< prevo
)
3027 * If the next block overlaps with this proposed allocation
3028 * then move the start back without adjusting the length,
3029 * but not before offset 0.
3030 * This may of course make the start overlap previous block,
3031 * and if we hit the offset 0 limit then the next block
3032 * can still overlap too.
3034 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
3035 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
3036 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
3037 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
3039 nexto
= gotp
->br_startoff
;
3041 nexto
= NULLFILEOFF
;
3043 align_off
+ align_alen
!= orig_end
&&
3044 align_off
+ align_alen
> nexto
)
3045 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
3047 * If we're now overlapping the next or previous extent that
3048 * means we can't fit an extsz piece in this hole. Just move
3049 * the start forward to the first valid spot and set
3050 * the length so we hit the end.
3052 if (align_off
!= orig_off
&& align_off
< prevo
)
3054 if (align_off
+ align_alen
!= orig_end
&&
3055 align_off
+ align_alen
> nexto
&&
3056 nexto
!= NULLFILEOFF
) {
3057 ASSERT(nexto
> prevo
);
3058 align_alen
= nexto
- align_off
;
3062 * If realtime, and the result isn't a multiple of the realtime
3063 * extent size we need to remove blocks until it is.
3065 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
3067 * We're not covering the original request, or
3068 * we won't be able to once we fix the length.
3070 if (orig_off
< align_off
||
3071 orig_end
> align_off
+ align_alen
||
3072 align_alen
- temp
< orig_alen
)
3075 * Try to fix it by moving the start up.
3077 if (align_off
+ temp
<= orig_off
) {
3082 * Try to fix it by moving the end in.
3084 else if (align_off
+ align_alen
- temp
>= orig_end
)
3087 * Set the start to the minimum then trim the length.
3090 align_alen
-= orig_off
- align_off
;
3091 align_off
= orig_off
;
3092 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3095 * Result doesn't cover the request, fail it.
3097 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3100 ASSERT(orig_off
>= align_off
);
3101 /* see MAXEXTLEN handling above */
3102 ASSERT(orig_end
<= align_off
+ align_alen
||
3103 align_alen
+ extsz
> MAXEXTLEN
);
3107 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3108 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3109 if (prevp
->br_startoff
!= NULLFILEOFF
)
3110 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3118 #define XFS_ALLOC_GAP_UNITS 4
3122 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3124 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3125 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3126 xfs_mount_t
*mp
; /* mount point structure */
3127 int nullfb
; /* true if ap->firstblock isn't set */
3128 int rt
; /* true if inode is realtime */
3130 #define ISVALID(x,y) \
3132 (x) < mp->m_sb.sb_rblocks : \
3133 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3134 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3135 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3137 mp
= ap
->ip
->i_mount
;
3138 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3139 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) &&
3140 (ap
->datatype
& XFS_ALLOC_USERDATA
);
3141 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3142 ap
->tp
->t_firstblock
);
3144 * If allocating at eof, and there's a previous real block,
3145 * try to use its last block as our starting point.
3147 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3148 !isnullstartblock(ap
->prev
.br_startblock
) &&
3149 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3150 ap
->prev
.br_startblock
)) {
3151 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3153 * Adjust for the gap between prevp and us.
3155 adjust
= ap
->offset
-
3156 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3158 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3159 ap
->blkno
+= adjust
;
3162 * If not at eof, then compare the two neighbor blocks.
3163 * Figure out whether either one gives us a good starting point,
3164 * and pick the better one.
3166 else if (!ap
->eof
) {
3167 xfs_fsblock_t gotbno
; /* right side block number */
3168 xfs_fsblock_t gotdiff
=0; /* right side difference */
3169 xfs_fsblock_t prevbno
; /* left side block number */
3170 xfs_fsblock_t prevdiff
=0; /* left side difference */
3173 * If there's a previous (left) block, select a requested
3174 * start block based on it.
3176 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3177 !isnullstartblock(ap
->prev
.br_startblock
) &&
3178 (prevbno
= ap
->prev
.br_startblock
+
3179 ap
->prev
.br_blockcount
) &&
3180 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3182 * Calculate gap to end of previous block.
3184 adjust
= prevdiff
= ap
->offset
-
3185 (ap
->prev
.br_startoff
+
3186 ap
->prev
.br_blockcount
);
3188 * Figure the startblock based on the previous block's
3189 * end and the gap size.
3191 * If the gap is large relative to the piece we're
3192 * allocating, or using it gives us an invalid block
3193 * number, then just use the end of the previous block.
3195 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3196 ISVALID(prevbno
+ prevdiff
,
3197 ap
->prev
.br_startblock
))
3202 * If the firstblock forbids it, can't use it,
3205 if (!rt
&& !nullfb
&&
3206 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3207 prevbno
= NULLFSBLOCK
;
3210 * No previous block or can't follow it, just default.
3213 prevbno
= NULLFSBLOCK
;
3215 * If there's a following (right) block, select a requested
3216 * start block based on it.
3218 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3220 * Calculate gap to start of next block.
3222 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3224 * Figure the startblock based on the next block's
3225 * start and the gap size.
3227 gotbno
= ap
->got
.br_startblock
;
3230 * If the gap is large relative to the piece we're
3231 * allocating, or using it gives us an invalid block
3232 * number, then just use the start of the next block
3233 * offset by our length.
3235 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3236 ISVALID(gotbno
- gotdiff
, gotbno
))
3238 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3239 gotbno
-= ap
->length
;
3240 gotdiff
+= adjust
- ap
->length
;
3244 * If the firstblock forbids it, can't use it,
3247 if (!rt
&& !nullfb
&&
3248 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3249 gotbno
= NULLFSBLOCK
;
3252 * No next block, just default.
3255 gotbno
= NULLFSBLOCK
;
3257 * If both valid, pick the better one, else the only good
3258 * one, else ap->blkno is already set (to 0 or the inode block).
3260 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3261 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3262 else if (prevbno
!= NULLFSBLOCK
)
3263 ap
->blkno
= prevbno
;
3264 else if (gotbno
!= NULLFSBLOCK
)
3271 xfs_bmap_longest_free_extent(
3272 struct xfs_trans
*tp
,
3277 struct xfs_mount
*mp
= tp
->t_mountp
;
3278 struct xfs_perag
*pag
;
3279 xfs_extlen_t longest
;
3282 pag
= xfs_perag_get(mp
, ag
);
3283 if (!pag
->pagf_init
) {
3284 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3286 /* Couldn't lock the AGF, so skip this AG. */
3287 if (error
== -EAGAIN
) {
3295 longest
= xfs_alloc_longest_free_extent(pag
,
3296 xfs_alloc_min_freelist(mp
, pag
),
3297 xfs_ag_resv_needed(pag
, XFS_AG_RESV_NONE
));
3298 if (*blen
< longest
)
3307 xfs_bmap_select_minlen(
3308 struct xfs_bmalloca
*ap
,
3309 struct xfs_alloc_arg
*args
,
3313 if (notinit
|| *blen
< ap
->minlen
) {
3315 * Since we did a BUF_TRYLOCK above, it is possible that
3316 * there is space for this request.
3318 args
->minlen
= ap
->minlen
;
3319 } else if (*blen
< args
->maxlen
) {
3321 * If the best seen length is less than the request length,
3322 * use the best as the minimum.
3324 args
->minlen
= *blen
;
3327 * Otherwise we've seen an extent as big as maxlen, use that
3330 args
->minlen
= args
->maxlen
;
3335 xfs_bmap_btalloc_nullfb(
3336 struct xfs_bmalloca
*ap
,
3337 struct xfs_alloc_arg
*args
,
3340 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3341 xfs_agnumber_t ag
, startag
;
3345 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3346 args
->total
= ap
->total
;
3348 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3349 if (startag
== NULLAGNUMBER
)
3352 while (*blen
< args
->maxlen
) {
3353 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3358 if (++ag
== mp
->m_sb
.sb_agcount
)
3364 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3369 xfs_bmap_btalloc_filestreams(
3370 struct xfs_bmalloca
*ap
,
3371 struct xfs_alloc_arg
*args
,
3374 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3379 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3380 args
->total
= ap
->total
;
3382 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3383 if (ag
== NULLAGNUMBER
)
3386 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3390 if (*blen
< args
->maxlen
) {
3391 error
= xfs_filestream_new_ag(ap
, &ag
);
3395 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3402 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3405 * Set the failure fallback case to look in the selected AG as stream
3408 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3412 /* Update all inode and quota accounting for the allocation we just did. */
3414 xfs_bmap_btalloc_accounting(
3415 struct xfs_bmalloca
*ap
,
3416 struct xfs_alloc_arg
*args
)
3418 if (ap
->flags
& XFS_BMAPI_COWFORK
) {
3420 * COW fork blocks are in-core only and thus are treated as
3421 * in-core quota reservation (like delalloc blocks) even when
3422 * converted to real blocks. The quota reservation is not
3423 * accounted to disk until blocks are remapped to the data
3424 * fork. So if these blocks were previously delalloc, we
3425 * already have quota reservation and there's nothing to do
3429 xfs_mod_delalloc(ap
->ip
->i_mount
, -(int64_t)args
->len
);
3434 * Otherwise, we've allocated blocks in a hole. The transaction
3435 * has acquired in-core quota reservation for this extent.
3436 * Rather than account these as real blocks, however, we reduce
3437 * the transaction quota reservation based on the allocation.
3438 * This essentially transfers the transaction quota reservation
3439 * to that of a delalloc extent.
3441 ap
->ip
->i_delayed_blks
+= args
->len
;
3442 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
, XFS_TRANS_DQ_RES_BLKS
,
3447 /* data/attr fork only */
3448 ap
->ip
->i_d
.di_nblocks
+= args
->len
;
3449 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3451 ap
->ip
->i_delayed_blks
-= args
->len
;
3452 xfs_mod_delalloc(ap
->ip
->i_mount
, -(int64_t)args
->len
);
3454 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3455 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
: XFS_TRANS_DQ_BCOUNT
,
3460 xfs_bmap_compute_alignments(
3461 struct xfs_bmalloca
*ap
,
3462 struct xfs_alloc_arg
*args
)
3464 struct xfs_mount
*mp
= args
->mp
;
3465 xfs_extlen_t align
= 0; /* minimum allocation alignment */
3466 int stripe_align
= 0;
3468 /* stripe alignment for allocation is determined by mount parameters */
3469 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3470 stripe_align
= mp
->m_swidth
;
3471 else if (mp
->m_dalign
)
3472 stripe_align
= mp
->m_dalign
;
3474 if (ap
->flags
& XFS_BMAPI_COWFORK
)
3475 align
= xfs_get_cowextsz_hint(ap
->ip
);
3476 else if (ap
->datatype
& XFS_ALLOC_USERDATA
)
3477 align
= xfs_get_extsz_hint(ap
->ip
);
3479 if (xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
, align
, 0,
3480 ap
->eof
, 0, ap
->conv
, &ap
->offset
,
3486 /* apply extent size hints if obtained earlier */
3489 div_u64_rem(ap
->offset
, args
->prod
, &args
->mod
);
3491 args
->mod
= args
->prod
- args
->mod
;
3492 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_SIZE
) {
3496 args
->prod
= PAGE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3497 div_u64_rem(ap
->offset
, args
->prod
, &args
->mod
);
3499 args
->mod
= args
->prod
- args
->mod
;
3502 return stripe_align
;
3506 xfs_bmap_process_allocated_extent(
3507 struct xfs_bmalloca
*ap
,
3508 struct xfs_alloc_arg
*args
,
3509 xfs_fileoff_t orig_offset
,
3510 xfs_extlen_t orig_length
)
3514 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3517 * check the allocation happened at the same or higher AG than
3518 * the first block that was allocated.
3521 XFS_FSB_TO_AGNO(args
->mp
, ap
->tp
->t_firstblock
) <=
3522 XFS_FSB_TO_AGNO(args
->mp
, args
->fsbno
));
3524 ap
->blkno
= args
->fsbno
;
3526 ap
->tp
->t_firstblock
= args
->fsbno
;
3527 ap
->length
= args
->len
;
3529 * If the extent size hint is active, we tried to round the
3530 * caller's allocation request offset down to extsz and the
3531 * length up to another extsz boundary. If we found a free
3532 * extent we mapped it in starting at this new offset. If the
3533 * newly mapped space isn't long enough to cover any of the
3534 * range of offsets that was originally requested, move the
3535 * mapping up so that we can fill as much of the caller's
3536 * original request as possible. Free space is apparently
3537 * very fragmented so we're unlikely to be able to satisfy the
3540 if (ap
->length
<= orig_length
)
3541 ap
->offset
= orig_offset
;
3542 else if (ap
->offset
+ ap
->length
< orig_offset
+ orig_length
)
3543 ap
->offset
= orig_offset
+ orig_length
- ap
->length
;
3544 xfs_bmap_btalloc_accounting(ap
, args
);
3549 xfs_bmap_exact_minlen_extent_alloc(
3550 struct xfs_bmalloca
*ap
)
3552 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3553 struct xfs_alloc_arg args
= { .tp
= ap
->tp
, .mp
= mp
};
3554 xfs_fileoff_t orig_offset
;
3555 xfs_extlen_t orig_length
;
3560 if (ap
->minlen
!= 1) {
3561 ap
->blkno
= NULLFSBLOCK
;
3566 orig_offset
= ap
->offset
;
3567 orig_length
= ap
->length
;
3569 args
.alloc_minlen_only
= 1;
3571 xfs_bmap_compute_alignments(ap
, &args
);
3573 if (ap
->tp
->t_firstblock
== NULLFSBLOCK
) {
3575 * Unlike the longest extent available in an AG, we don't track
3576 * the length of an AG's shortest extent.
3577 * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3578 * hence we can afford to start traversing from the 0th AG since
3579 * we need not be concerned about a drop in performance in
3580 * "debug only" code paths.
3582 ap
->blkno
= XFS_AGB_TO_FSB(mp
, 0, 0);
3584 ap
->blkno
= ap
->tp
->t_firstblock
;
3587 args
.fsbno
= ap
->blkno
;
3588 args
.oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
3589 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3590 args
.total
= args
.minlen
= args
.maxlen
= ap
->minlen
;
3593 args
.minalignslop
= 0;
3595 args
.minleft
= ap
->minleft
;
3596 args
.wasdel
= ap
->wasdel
;
3597 args
.resv
= XFS_AG_RESV_NONE
;
3598 args
.datatype
= ap
->datatype
;
3600 error
= xfs_alloc_vextent(&args
);
3604 if (args
.fsbno
!= NULLFSBLOCK
) {
3605 xfs_bmap_process_allocated_extent(ap
, &args
, orig_offset
,
3608 ap
->blkno
= NULLFSBLOCK
;
3616 #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3622 struct xfs_bmalloca
*ap
)
3624 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3625 struct xfs_alloc_arg args
= { .tp
= ap
->tp
, .mp
= mp
};
3626 xfs_alloctype_t atype
= 0;
3627 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3629 xfs_fileoff_t orig_offset
;
3630 xfs_extlen_t orig_length
;
3632 xfs_extlen_t nextminlen
= 0;
3633 int nullfb
; /* true if ap->firstblock isn't set */
3640 orig_offset
= ap
->offset
;
3641 orig_length
= ap
->length
;
3643 stripe_align
= xfs_bmap_compute_alignments(ap
, &args
);
3645 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3646 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3647 ap
->tp
->t_firstblock
);
3649 if ((ap
->datatype
& XFS_ALLOC_USERDATA
) &&
3650 xfs_inode_is_filestream(ap
->ip
)) {
3651 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3652 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3653 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3655 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3658 ap
->blkno
= ap
->tp
->t_firstblock
;
3660 xfs_bmap_adjacent(ap
);
3663 * If allowed, use ap->blkno; otherwise must use firstblock since
3664 * it's in the right allocation group.
3666 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3669 ap
->blkno
= ap
->tp
->t_firstblock
;
3671 * Normal allocation, done through xfs_alloc_vextent.
3673 tryagain
= isaligned
= 0;
3674 args
.fsbno
= ap
->blkno
;
3675 args
.oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
3677 /* Trim the allocation back to the maximum an AG can fit. */
3678 args
.maxlen
= min(ap
->length
, mp
->m_ag_max_usable
);
3682 * Search for an allocation group with a single extent large
3683 * enough for the request. If one isn't found, then adjust
3684 * the minimum allocation size to the largest space found.
3686 if ((ap
->datatype
& XFS_ALLOC_USERDATA
) &&
3687 xfs_inode_is_filestream(ap
->ip
))
3688 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3690 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3693 } else if (ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) {
3694 if (xfs_inode_is_filestream(ap
->ip
))
3695 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3697 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3698 args
.total
= args
.minlen
= ap
->minlen
;
3700 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3701 args
.total
= ap
->total
;
3702 args
.minlen
= ap
->minlen
;
3706 * If we are not low on available data blocks, and the underlying
3707 * logical volume manager is a stripe, and the file offset is zero then
3708 * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof
3709 * is only set if the allocation length is >= the stripe unit and the
3710 * allocation offset is at the end of file.
3712 if (!(ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) && ap
->aeof
) {
3714 args
.alignment
= stripe_align
;
3718 * Adjust minlen to try and preserve alignment if we
3719 * can't guarantee an aligned maxlen extent.
3721 if (blen
> args
.alignment
&&
3722 blen
<= args
.maxlen
+ args
.alignment
)
3723 args
.minlen
= blen
- args
.alignment
;
3724 args
.minalignslop
= 0;
3727 * First try an exact bno allocation.
3728 * If it fails then do a near or start bno
3729 * allocation with alignment turned on.
3733 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3736 * Compute the minlen+alignment for the
3737 * next case. Set slop so that the value
3738 * of minlen+alignment+slop doesn't go up
3739 * between the calls.
3741 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3742 nextminlen
= blen
- stripe_align
;
3744 nextminlen
= args
.minlen
;
3745 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3747 nextminlen
+ stripe_align
-
3750 args
.minalignslop
= 0;
3754 args
.minalignslop
= 0;
3756 args
.minleft
= ap
->minleft
;
3757 args
.wasdel
= ap
->wasdel
;
3758 args
.resv
= XFS_AG_RESV_NONE
;
3759 args
.datatype
= ap
->datatype
;
3761 error
= xfs_alloc_vextent(&args
);
3765 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3767 * Exact allocation failed. Now try with alignment
3771 args
.fsbno
= ap
->blkno
;
3772 args
.alignment
= stripe_align
;
3773 args
.minlen
= nextminlen
;
3774 args
.minalignslop
= 0;
3776 if ((error
= xfs_alloc_vextent(&args
)))
3779 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3781 * allocation failed, so turn off alignment and
3785 args
.fsbno
= ap
->blkno
;
3787 if ((error
= xfs_alloc_vextent(&args
)))
3790 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3791 args
.minlen
> ap
->minlen
) {
3792 args
.minlen
= ap
->minlen
;
3793 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3794 args
.fsbno
= ap
->blkno
;
3795 if ((error
= xfs_alloc_vextent(&args
)))
3798 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3800 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3801 args
.total
= ap
->minlen
;
3802 if ((error
= xfs_alloc_vextent(&args
)))
3804 ap
->tp
->t_flags
|= XFS_TRANS_LOWMODE
;
3807 if (args
.fsbno
!= NULLFSBLOCK
) {
3808 xfs_bmap_process_allocated_extent(ap
, &args
, orig_offset
,
3811 ap
->blkno
= NULLFSBLOCK
;
3817 /* Trim extent to fit a logical block range. */
3820 struct xfs_bmbt_irec
*irec
,
3824 xfs_fileoff_t distance
;
3825 xfs_fileoff_t end
= bno
+ len
;
3827 if (irec
->br_startoff
+ irec
->br_blockcount
<= bno
||
3828 irec
->br_startoff
>= end
) {
3829 irec
->br_blockcount
= 0;
3833 if (irec
->br_startoff
< bno
) {
3834 distance
= bno
- irec
->br_startoff
;
3835 if (isnullstartblock(irec
->br_startblock
))
3836 irec
->br_startblock
= DELAYSTARTBLOCK
;
3837 if (irec
->br_startblock
!= DELAYSTARTBLOCK
&&
3838 irec
->br_startblock
!= HOLESTARTBLOCK
)
3839 irec
->br_startblock
+= distance
;
3840 irec
->br_startoff
+= distance
;
3841 irec
->br_blockcount
-= distance
;
3844 if (end
< irec
->br_startoff
+ irec
->br_blockcount
) {
3845 distance
= irec
->br_startoff
+ irec
->br_blockcount
- end
;
3846 irec
->br_blockcount
-= distance
;
3851 * Trim the returned map to the required bounds
3855 struct xfs_bmbt_irec
*mval
,
3856 struct xfs_bmbt_irec
*got
,
3864 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3865 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3867 if (isnullstartblock(got
->br_startblock
))
3868 mval
->br_startblock
= DELAYSTARTBLOCK
;
3874 ASSERT((*bno
>= obno
) || (n
== 0));
3876 mval
->br_startoff
= *bno
;
3877 if (isnullstartblock(got
->br_startblock
))
3878 mval
->br_startblock
= DELAYSTARTBLOCK
;
3880 mval
->br_startblock
= got
->br_startblock
+
3881 (*bno
- got
->br_startoff
);
3883 * Return the minimum of what we got and what we asked for for
3884 * the length. We can use the len variable here because it is
3885 * modified below and we could have been there before coming
3886 * here if the first part of the allocation didn't overlap what
3889 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
3890 got
->br_blockcount
- (*bno
- got
->br_startoff
));
3891 mval
->br_state
= got
->br_state
;
3892 ASSERT(mval
->br_blockcount
<= len
);
3897 * Update and validate the extent map to return
3900 xfs_bmapi_update_map(
3901 struct xfs_bmbt_irec
**map
,
3909 xfs_bmbt_irec_t
*mval
= *map
;
3911 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
3912 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
3913 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
3914 (mval
->br_startoff
< obno
));
3916 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
3918 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
3919 /* update previous map with new information */
3920 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
3921 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
3922 ASSERT(mval
->br_state
== mval
[-1].br_state
);
3923 mval
[-1].br_blockcount
= mval
->br_blockcount
;
3924 mval
[-1].br_state
= mval
->br_state
;
3925 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
3926 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
3927 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
3928 mval
->br_startblock
== mval
[-1].br_startblock
+
3929 mval
[-1].br_blockcount
&&
3930 mval
[-1].br_state
== mval
->br_state
) {
3931 ASSERT(mval
->br_startoff
==
3932 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
3933 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3934 } else if (*n
> 0 &&
3935 mval
->br_startblock
== DELAYSTARTBLOCK
&&
3936 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
3937 mval
->br_startoff
==
3938 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
3939 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3940 mval
[-1].br_state
= mval
->br_state
;
3941 } else if (!((*n
== 0) &&
3942 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
3951 * Map file blocks to filesystem blocks without allocation.
3955 struct xfs_inode
*ip
,
3958 struct xfs_bmbt_irec
*mval
,
3962 struct xfs_mount
*mp
= ip
->i_mount
;
3963 int whichfork
= xfs_bmapi_whichfork(flags
);
3964 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3965 struct xfs_bmbt_irec got
;
3968 struct xfs_iext_cursor icur
;
3974 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
| XFS_BMAPI_ENTIRE
)));
3975 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
3977 if (WARN_ON_ONCE(!ifp
))
3978 return -EFSCORRUPTED
;
3980 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
3981 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
))
3982 return -EFSCORRUPTED
;
3984 if (XFS_FORCED_SHUTDOWN(mp
))
3987 XFS_STATS_INC(mp
, xs_blk_mapr
);
3989 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
3990 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
3995 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
))
4000 while (bno
< end
&& n
< *nmap
) {
4001 /* Reading past eof, act as though there's a hole up to end. */
4003 got
.br_startoff
= end
;
4004 if (got
.br_startoff
> bno
) {
4005 /* Reading in a hole. */
4006 mval
->br_startoff
= bno
;
4007 mval
->br_startblock
= HOLESTARTBLOCK
;
4008 mval
->br_blockcount
=
4009 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4010 mval
->br_state
= XFS_EXT_NORM
;
4011 bno
+= mval
->br_blockcount
;
4012 len
-= mval
->br_blockcount
;
4018 /* set up the extent map to return. */
4019 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4020 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4022 /* If we're done, stop now. */
4023 if (bno
>= end
|| n
>= *nmap
)
4026 /* Else go on to the next record. */
4027 if (!xfs_iext_next_extent(ifp
, &icur
, &got
))
4035 * Add a delayed allocation extent to an inode. Blocks are reserved from the
4036 * global pool and the extent inserted into the inode in-core extent tree.
4038 * On entry, got refers to the first extent beyond the offset of the extent to
4039 * allocate or eof is specified if no such extent exists. On return, got refers
4040 * to the extent record that was inserted to the inode fork.
4042 * Note that the allocated extent may have been merged with contiguous extents
4043 * during insertion into the inode fork. Thus, got does not reflect the current
4044 * state of the inode fork on return. If necessary, the caller can use lastx to
4045 * look up the updated record in the inode fork.
4048 xfs_bmapi_reserve_delalloc(
4049 struct xfs_inode
*ip
,
4053 xfs_filblks_t prealloc
,
4054 struct xfs_bmbt_irec
*got
,
4055 struct xfs_iext_cursor
*icur
,
4058 struct xfs_mount
*mp
= ip
->i_mount
;
4059 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4061 xfs_extlen_t indlen
;
4063 xfs_fileoff_t aoff
= off
;
4066 * Cap the alloc length. Keep track of prealloc so we know whether to
4067 * tag the inode before we return.
4069 alen
= XFS_FILBLKS_MIN(len
+ prealloc
, MAXEXTLEN
);
4071 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4072 if (prealloc
&& alen
>= len
)
4073 prealloc
= alen
- len
;
4075 /* Figure out the extent size, adjust alen */
4076 if (whichfork
== XFS_COW_FORK
) {
4077 struct xfs_bmbt_irec prev
;
4078 xfs_extlen_t extsz
= xfs_get_cowextsz_hint(ip
);
4080 if (!xfs_iext_peek_prev_extent(ifp
, icur
, &prev
))
4081 prev
.br_startoff
= NULLFILEOFF
;
4083 error
= xfs_bmap_extsize_align(mp
, got
, &prev
, extsz
, 0, eof
,
4084 1, 0, &aoff
, &alen
);
4089 * Make a transaction-less quota reservation for delayed allocation
4090 * blocks. This number gets adjusted later. We return if we haven't
4091 * allocated blocks already inside this loop.
4093 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4094 XFS_QMOPT_RES_REGBLKS
);
4099 * Split changing sb for alen and indlen since they could be coming
4100 * from different places.
4102 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4105 error
= xfs_mod_fdblocks(mp
, -((int64_t)alen
), false);
4107 goto out_unreserve_quota
;
4109 error
= xfs_mod_fdblocks(mp
, -((int64_t)indlen
), false);
4111 goto out_unreserve_blocks
;
4114 ip
->i_delayed_blks
+= alen
;
4115 xfs_mod_delalloc(ip
->i_mount
, alen
+ indlen
);
4117 got
->br_startoff
= aoff
;
4118 got
->br_startblock
= nullstartblock(indlen
);
4119 got
->br_blockcount
= alen
;
4120 got
->br_state
= XFS_EXT_NORM
;
4122 xfs_bmap_add_extent_hole_delay(ip
, whichfork
, icur
, got
);
4125 * Tag the inode if blocks were preallocated. Note that COW fork
4126 * preallocation can occur at the start or end of the extent, even when
4127 * prealloc == 0, so we must also check the aligned offset and length.
4129 if (whichfork
== XFS_DATA_FORK
&& prealloc
)
4130 xfs_inode_set_eofblocks_tag(ip
);
4131 if (whichfork
== XFS_COW_FORK
&& (prealloc
|| aoff
< off
|| alen
> len
))
4132 xfs_inode_set_cowblocks_tag(ip
);
4136 out_unreserve_blocks
:
4137 xfs_mod_fdblocks(mp
, alen
, false);
4138 out_unreserve_quota
:
4139 if (XFS_IS_QUOTA_ON(mp
))
4140 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4141 XFS_QMOPT_RES_REGBLKS
);
4146 xfs_bmap_alloc_userdata(
4147 struct xfs_bmalloca
*bma
)
4149 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4150 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4154 * Set the data type being allocated. For the data fork, the first data
4155 * in the file is treated differently to all other allocations. For the
4156 * attribute fork, we only need to ensure the allocated range is not on
4159 bma
->datatype
= XFS_ALLOC_NOBUSY
;
4160 if (whichfork
== XFS_DATA_FORK
) {
4161 bma
->datatype
|= XFS_ALLOC_USERDATA
;
4162 if (bma
->offset
== 0)
4163 bma
->datatype
|= XFS_ALLOC_INITIAL_USER_DATA
;
4165 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
) {
4166 error
= xfs_bmap_isaeof(bma
, whichfork
);
4171 if (XFS_IS_REALTIME_INODE(bma
->ip
))
4172 return xfs_bmap_rtalloc(bma
);
4175 if (unlikely(XFS_TEST_ERROR(false, mp
,
4176 XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT
)))
4177 return xfs_bmap_exact_minlen_extent_alloc(bma
);
4179 return xfs_bmap_btalloc(bma
);
4184 struct xfs_bmalloca
*bma
)
4186 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4187 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4188 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4189 int tmp_logflags
= 0;
4192 ASSERT(bma
->length
> 0);
4195 * For the wasdelay case, we could also just allocate the stuff asked
4196 * for in this bmap call but that wouldn't be as good.
4199 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4200 bma
->offset
= bma
->got
.br_startoff
;
4201 if (!xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &bma
->prev
))
4202 bma
->prev
.br_startoff
= NULLFILEOFF
;
4204 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4206 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4207 bma
->got
.br_startoff
- bma
->offset
);
4210 if (bma
->flags
& XFS_BMAPI_CONTIG
)
4211 bma
->minlen
= bma
->length
;
4215 if (bma
->flags
& XFS_BMAPI_METADATA
) {
4216 if (unlikely(XFS_TEST_ERROR(false, mp
,
4217 XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT
)))
4218 error
= xfs_bmap_exact_minlen_extent_alloc(bma
);
4220 error
= xfs_bmap_btalloc(bma
);
4222 error
= xfs_bmap_alloc_userdata(bma
);
4224 if (error
|| bma
->blkno
== NULLFSBLOCK
)
4227 if (bma
->flags
& XFS_BMAPI_ZERO
) {
4228 error
= xfs_zero_extent(bma
->ip
, bma
->blkno
, bma
->length
);
4233 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
)
4234 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4236 * Bump the number of extents we've allocated
4242 bma
->cur
->bc_ino
.flags
=
4243 bma
->wasdel
? XFS_BTCUR_BMBT_WASDEL
: 0;
4245 bma
->got
.br_startoff
= bma
->offset
;
4246 bma
->got
.br_startblock
= bma
->blkno
;
4247 bma
->got
.br_blockcount
= bma
->length
;
4248 bma
->got
.br_state
= XFS_EXT_NORM
;
4250 if (bma
->flags
& XFS_BMAPI_PREALLOC
)
4251 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4254 error
= xfs_bmap_add_extent_delay_real(bma
, whichfork
);
4256 error
= xfs_bmap_add_extent_hole_real(bma
->tp
, bma
->ip
,
4257 whichfork
, &bma
->icur
, &bma
->cur
, &bma
->got
,
4258 &bma
->logflags
, bma
->flags
);
4260 bma
->logflags
|= tmp_logflags
;
4265 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4266 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4267 * the neighbouring ones.
4269 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4271 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4272 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4273 bma
->offset
+ bma
->length
);
4274 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4275 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4280 xfs_bmapi_convert_unwritten(
4281 struct xfs_bmalloca
*bma
,
4282 struct xfs_bmbt_irec
*mval
,
4286 int whichfork
= xfs_bmapi_whichfork(flags
);
4287 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4288 int tmp_logflags
= 0;
4291 /* check if we need to do unwritten->real conversion */
4292 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4293 (flags
& XFS_BMAPI_PREALLOC
))
4296 /* check if we need to do real->unwritten conversion */
4297 if (mval
->br_state
== XFS_EXT_NORM
&&
4298 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4299 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4303 * Modify (by adding) the state flag, if writing.
4305 ASSERT(mval
->br_blockcount
<= len
);
4306 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4307 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4308 bma
->ip
, whichfork
);
4310 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4311 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4314 * Before insertion into the bmbt, zero the range being converted
4317 if (flags
& XFS_BMAPI_ZERO
) {
4318 error
= xfs_zero_extent(bma
->ip
, mval
->br_startblock
,
4319 mval
->br_blockcount
);
4324 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, whichfork
,
4325 &bma
->icur
, &bma
->cur
, mval
, &tmp_logflags
);
4327 * Log the inode core unconditionally in the unwritten extent conversion
4328 * path because the conversion might not have done so (e.g., if the
4329 * extent count hasn't changed). We need to make sure the inode is dirty
4330 * in the transaction for the sake of fsync(), even if nothing has
4331 * changed, because fsync() will not force the log for this transaction
4332 * unless it sees the inode pinned.
4334 * Note: If we're only converting cow fork extents, there aren't
4335 * any on-disk updates to make, so we don't need to log anything.
4337 if (whichfork
!= XFS_COW_FORK
)
4338 bma
->logflags
|= tmp_logflags
| XFS_ILOG_CORE
;
4343 * Update our extent pointer, given that
4344 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4345 * of the neighbouring ones.
4347 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4350 * We may have combined previously unwritten space with written space,
4351 * so generate another request.
4353 if (mval
->br_blockcount
< len
)
4358 static inline xfs_extlen_t
4360 struct xfs_trans
*tp
,
4361 struct xfs_inode
*ip
,
4364 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, fork
);
4366 if (tp
&& tp
->t_firstblock
!= NULLFSBLOCK
)
4368 if (ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
4370 return be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4374 * Log whatever the flags say, even if error. Otherwise we might miss detecting
4375 * a case where the data is changed, there's an error, and it's not logged so we
4376 * don't shutdown when we should. Don't bother logging extents/btree changes if
4377 * we converted to the other format.
4381 struct xfs_bmalloca
*bma
,
4385 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4387 if ((bma
->logflags
& xfs_ilog_fext(whichfork
)) &&
4388 ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
4389 bma
->logflags
&= ~xfs_ilog_fext(whichfork
);
4390 else if ((bma
->logflags
& xfs_ilog_fbroot(whichfork
)) &&
4391 ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
4392 bma
->logflags
&= ~xfs_ilog_fbroot(whichfork
);
4395 xfs_trans_log_inode(bma
->tp
, bma
->ip
, bma
->logflags
);
4397 xfs_btree_del_cursor(bma
->cur
, error
);
4401 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4402 * extent state if necessary. Details behaviour is controlled by the flags
4403 * parameter. Only allocates blocks from a single allocation group, to avoid
4408 struct xfs_trans
*tp
, /* transaction pointer */
4409 struct xfs_inode
*ip
, /* incore inode */
4410 xfs_fileoff_t bno
, /* starting file offs. mapped */
4411 xfs_filblks_t len
, /* length to map in file */
4412 int flags
, /* XFS_BMAPI_... */
4413 xfs_extlen_t total
, /* total blocks needed */
4414 struct xfs_bmbt_irec
*mval
, /* output: map values */
4415 int *nmap
) /* i/o: mval size/count */
4417 struct xfs_bmalloca bma
= {
4422 struct xfs_mount
*mp
= ip
->i_mount
;
4423 int whichfork
= xfs_bmapi_whichfork(flags
);
4424 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4425 xfs_fileoff_t end
; /* end of mapped file region */
4426 bool eof
= false; /* after the end of extents */
4427 int error
; /* error return */
4428 int n
; /* current extent index */
4429 xfs_fileoff_t obno
; /* old block number (offset) */
4432 xfs_fileoff_t orig_bno
; /* original block number value */
4433 int orig_flags
; /* original flags arg value */
4434 xfs_filblks_t orig_len
; /* original value of len arg */
4435 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4436 int orig_nmap
; /* original value of *nmap */
4446 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4449 ASSERT(ifp
->if_format
!= XFS_DINODE_FMT_LOCAL
);
4450 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4451 ASSERT(!(flags
& XFS_BMAPI_REMAP
));
4453 /* zeroing is for currently only for data extents, not metadata */
4454 ASSERT((flags
& (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
)) !=
4455 (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
));
4457 * we can allocate unwritten extents or pre-zero allocated blocks,
4458 * but it makes no sense to do both at once. This would result in
4459 * zeroing the unwritten extent twice, but it still being an
4460 * unwritten extent....
4462 ASSERT((flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
)) !=
4463 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
));
4465 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
4466 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
4467 return -EFSCORRUPTED
;
4470 if (XFS_FORCED_SHUTDOWN(mp
))
4473 XFS_STATS_INC(mp
, xs_blk_mapw
);
4475 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4476 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4481 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &bma
.icur
, &bma
.got
))
4483 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4484 bma
.prev
.br_startoff
= NULLFILEOFF
;
4485 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4490 while (bno
< end
&& n
< *nmap
) {
4491 bool need_alloc
= false, wasdelay
= false;
4493 /* in hole or beyond EOF? */
4494 if (eof
|| bma
.got
.br_startoff
> bno
) {
4496 * CoW fork conversions should /never/ hit EOF or
4497 * holes. There should always be something for us
4500 ASSERT(!((flags
& XFS_BMAPI_CONVERT
) &&
4501 (flags
& XFS_BMAPI_COWFORK
)));
4504 } else if (isnullstartblock(bma
.got
.br_startblock
)) {
4509 * First, deal with the hole before the allocated space
4510 * that we found, if any.
4512 if (need_alloc
|| wasdelay
) {
4514 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4515 bma
.wasdel
= wasdelay
;
4520 * There's a 32/64 bit type mismatch between the
4521 * allocation length request (which can be 64 bits in
4522 * length) and the bma length request, which is
4523 * xfs_extlen_t and therefore 32 bits. Hence we have to
4524 * check for 32-bit overflows and handle them here.
4526 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4527 bma
.length
= MAXEXTLEN
;
4532 ASSERT(bma
.length
> 0);
4533 error
= xfs_bmapi_allocate(&bma
);
4536 if (bma
.blkno
== NULLFSBLOCK
)
4540 * If this is a CoW allocation, record the data in
4541 * the refcount btree for orphan recovery.
4543 if (whichfork
== XFS_COW_FORK
)
4544 xfs_refcount_alloc_cow_extent(tp
, bma
.blkno
,
4548 /* Deal with the allocated space we found. */
4549 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4552 /* Execute unwritten extent conversion if necessary */
4553 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4554 if (error
== -EAGAIN
)
4559 /* update the extent map to return */
4560 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4563 * If we're done, stop now. Stop when we've allocated
4564 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4565 * the transaction may get too big.
4567 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4570 /* Else go on to the next record. */
4572 if (!xfs_iext_next_extent(ifp
, &bma
.icur
, &bma
.got
))
4577 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4582 ASSERT(ifp
->if_format
!= XFS_DINODE_FMT_BTREE
||
4583 ifp
->if_nextents
> XFS_IFORK_MAXEXT(ip
, whichfork
));
4584 xfs_bmapi_finish(&bma
, whichfork
, 0);
4585 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4589 xfs_bmapi_finish(&bma
, whichfork
, error
);
4594 * Convert an existing delalloc extent to real blocks based on file offset. This
4595 * attempts to allocate the entire delalloc extent and may require multiple
4596 * invocations to allocate the target offset if a large enough physical extent
4600 xfs_bmapi_convert_delalloc(
4601 struct xfs_inode
*ip
,
4604 struct iomap
*iomap
,
4607 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4608 struct xfs_mount
*mp
= ip
->i_mount
;
4609 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4610 struct xfs_bmalloca bma
= { NULL
};
4612 struct xfs_trans
*tp
;
4615 if (whichfork
== XFS_COW_FORK
)
4616 flags
|= IOMAP_F_SHARED
;
4619 * Space for the extent and indirect blocks was reserved when the
4620 * delalloc extent was created so there's no need to do so here.
4622 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0, 0,
4623 XFS_TRANS_RESERVE
, &tp
);
4627 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4629 error
= xfs_iext_count_may_overflow(ip
, whichfork
,
4630 XFS_IEXT_ADD_NOSPLIT_CNT
);
4632 goto out_trans_cancel
;
4634 xfs_trans_ijoin(tp
, ip
, 0);
4636 if (!xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &bma
.icur
, &bma
.got
) ||
4637 bma
.got
.br_startoff
> offset_fsb
) {
4639 * No extent found in the range we are trying to convert. This
4640 * should only happen for the COW fork, where another thread
4641 * might have moved the extent to the data fork in the meantime.
4643 WARN_ON_ONCE(whichfork
!= XFS_COW_FORK
);
4645 goto out_trans_cancel
;
4649 * If we find a real extent here we raced with another thread converting
4650 * the extent. Just return the real extent at this offset.
4652 if (!isnullstartblock(bma
.got
.br_startblock
)) {
4653 xfs_bmbt_to_iomap(ip
, iomap
, &bma
.got
, flags
);
4654 *seq
= READ_ONCE(ifp
->if_seq
);
4655 goto out_trans_cancel
;
4661 bma
.offset
= bma
.got
.br_startoff
;
4662 bma
.length
= max_t(xfs_filblks_t
, bma
.got
.br_blockcount
, MAXEXTLEN
);
4663 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4666 * When we're converting the delalloc reservations backing dirty pages
4667 * in the page cache, we must be careful about how we create the new
4670 * New CoW fork extents are created unwritten, turned into real extents
4671 * when we're about to write the data to disk, and mapped into the data
4672 * fork after the write finishes. End of story.
4674 * New data fork extents must be mapped in as unwritten and converted
4675 * to real extents after the write succeeds to avoid exposing stale
4676 * disk contents if we crash.
4678 bma
.flags
= XFS_BMAPI_PREALLOC
;
4679 if (whichfork
== XFS_COW_FORK
)
4680 bma
.flags
|= XFS_BMAPI_COWFORK
;
4682 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4683 bma
.prev
.br_startoff
= NULLFILEOFF
;
4685 error
= xfs_bmapi_allocate(&bma
);
4690 if (WARN_ON_ONCE(bma
.blkno
== NULLFSBLOCK
))
4692 error
= -EFSCORRUPTED
;
4693 if (WARN_ON_ONCE(!xfs_valid_startblock(ip
, bma
.got
.br_startblock
)))
4696 XFS_STATS_ADD(mp
, xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, bma
.length
));
4697 XFS_STATS_INC(mp
, xs_xstrat_quick
);
4699 ASSERT(!isnullstartblock(bma
.got
.br_startblock
));
4700 xfs_bmbt_to_iomap(ip
, iomap
, &bma
.got
, flags
);
4701 *seq
= READ_ONCE(ifp
->if_seq
);
4703 if (whichfork
== XFS_COW_FORK
)
4704 xfs_refcount_alloc_cow_extent(tp
, bma
.blkno
, bma
.length
);
4706 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4711 xfs_bmapi_finish(&bma
, whichfork
, 0);
4712 error
= xfs_trans_commit(tp
);
4713 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4717 xfs_bmapi_finish(&bma
, whichfork
, error
);
4719 xfs_trans_cancel(tp
);
4720 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4726 struct xfs_trans
*tp
,
4727 struct xfs_inode
*ip
,
4730 xfs_fsblock_t startblock
,
4733 struct xfs_mount
*mp
= ip
->i_mount
;
4734 struct xfs_ifork
*ifp
;
4735 struct xfs_btree_cur
*cur
= NULL
;
4736 struct xfs_bmbt_irec got
;
4737 struct xfs_iext_cursor icur
;
4738 int whichfork
= xfs_bmapi_whichfork(flags
);
4739 int logflags
= 0, error
;
4741 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4743 ASSERT(len
<= (xfs_filblks_t
)MAXEXTLEN
);
4744 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4745 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
|
4746 XFS_BMAPI_NORMAP
)));
4747 ASSERT((flags
& (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
)) !=
4748 (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
));
4750 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
4751 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
4752 return -EFSCORRUPTED
;
4755 if (XFS_FORCED_SHUTDOWN(mp
))
4758 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4759 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4764 if (xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
)) {
4765 /* make sure we only reflink into a hole. */
4766 ASSERT(got
.br_startoff
> bno
);
4767 ASSERT(got
.br_startoff
- bno
>= len
);
4770 ip
->i_d
.di_nblocks
+= len
;
4771 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4773 if (ifp
->if_flags
& XFS_IFBROOT
) {
4774 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
4775 cur
->bc_ino
.flags
= 0;
4778 got
.br_startoff
= bno
;
4779 got
.br_startblock
= startblock
;
4780 got
.br_blockcount
= len
;
4781 if (flags
& XFS_BMAPI_PREALLOC
)
4782 got
.br_state
= XFS_EXT_UNWRITTEN
;
4784 got
.br_state
= XFS_EXT_NORM
;
4786 error
= xfs_bmap_add_extent_hole_real(tp
, ip
, whichfork
, &icur
,
4787 &cur
, &got
, &logflags
, flags
);
4791 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
, whichfork
);
4794 if (ip
->i_df
.if_format
!= XFS_DINODE_FMT_EXTENTS
)
4795 logflags
&= ~XFS_ILOG_DEXT
;
4796 else if (ip
->i_df
.if_format
!= XFS_DINODE_FMT_BTREE
)
4797 logflags
&= ~XFS_ILOG_DBROOT
;
4800 xfs_trans_log_inode(tp
, ip
, logflags
);
4802 xfs_btree_del_cursor(cur
, error
);
4807 * When a delalloc extent is split (e.g., due to a hole punch), the original
4808 * indlen reservation must be shared across the two new extents that are left
4811 * Given the original reservation and the worst case indlen for the two new
4812 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4813 * reservation fairly across the two new extents. If necessary, steal available
4814 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4815 * ores == 1). The number of stolen blocks is returned. The availability and
4816 * subsequent accounting of stolen blocks is the responsibility of the caller.
4818 static xfs_filblks_t
4819 xfs_bmap_split_indlen(
4820 xfs_filblks_t ores
, /* original res. */
4821 xfs_filblks_t
*indlen1
, /* ext1 worst indlen */
4822 xfs_filblks_t
*indlen2
, /* ext2 worst indlen */
4823 xfs_filblks_t avail
) /* stealable blocks */
4825 xfs_filblks_t len1
= *indlen1
;
4826 xfs_filblks_t len2
= *indlen2
;
4827 xfs_filblks_t nres
= len1
+ len2
; /* new total res. */
4828 xfs_filblks_t stolen
= 0;
4829 xfs_filblks_t resfactor
;
4832 * Steal as many blocks as we can to try and satisfy the worst case
4833 * indlen for both new extents.
4835 if (ores
< nres
&& avail
)
4836 stolen
= XFS_FILBLKS_MIN(nres
- ores
, avail
);
4839 /* nothing else to do if we've satisfied the new reservation */
4844 * We can't meet the total required reservation for the two extents.
4845 * Calculate the percent of the overall shortage between both extents
4846 * and apply this percentage to each of the requested indlen values.
4847 * This distributes the shortage fairly and reduces the chances that one
4848 * of the two extents is left with nothing when extents are repeatedly
4851 resfactor
= (ores
* 100);
4852 do_div(resfactor
, nres
);
4857 ASSERT(len1
+ len2
<= ores
);
4858 ASSERT(len1
< *indlen1
&& len2
< *indlen2
);
4861 * Hand out the remainder to each extent. If one of the two reservations
4862 * is zero, we want to make sure that one gets a block first. The loop
4863 * below starts with len1, so hand len2 a block right off the bat if it
4866 ores
-= (len1
+ len2
);
4867 ASSERT((*indlen1
- len1
) + (*indlen2
- len2
) >= ores
);
4868 if (ores
&& !len2
&& *indlen2
) {
4873 if (len1
< *indlen1
) {
4879 if (len2
< *indlen2
) {
4892 xfs_bmap_del_extent_delay(
4893 struct xfs_inode
*ip
,
4895 struct xfs_iext_cursor
*icur
,
4896 struct xfs_bmbt_irec
*got
,
4897 struct xfs_bmbt_irec
*del
)
4899 struct xfs_mount
*mp
= ip
->i_mount
;
4900 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4901 struct xfs_bmbt_irec
new;
4902 int64_t da_old
, da_new
, da_diff
= 0;
4903 xfs_fileoff_t del_endoff
, got_endoff
;
4904 xfs_filblks_t got_indlen
, new_indlen
, stolen
;
4905 int state
= xfs_bmap_fork_to_state(whichfork
);
4909 XFS_STATS_INC(mp
, xs_del_exlist
);
4911 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4912 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4913 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4914 da_old
= startblockval(got
->br_startblock
);
4917 ASSERT(del
->br_blockcount
> 0);
4918 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4919 ASSERT(got_endoff
>= del_endoff
);
4922 uint64_t rtexts
= XFS_FSB_TO_B(mp
, del
->br_blockcount
);
4924 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
4925 xfs_mod_frextents(mp
, rtexts
);
4929 * Update the inode delalloc counter now and wait to update the
4930 * sb counters as we might have to borrow some blocks for the
4931 * indirect block accounting.
4933 error
= xfs_trans_unreserve_quota_nblks(NULL
, ip
, del
->br_blockcount
, 0,
4934 isrt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4937 ip
->i_delayed_blks
-= del
->br_blockcount
;
4939 if (got
->br_startoff
== del
->br_startoff
)
4940 state
|= BMAP_LEFT_FILLING
;
4941 if (got_endoff
== del_endoff
)
4942 state
|= BMAP_RIGHT_FILLING
;
4944 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
4945 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
4947 * Matches the whole extent. Delete the entry.
4949 xfs_iext_remove(ip
, icur
, state
);
4950 xfs_iext_prev(ifp
, icur
);
4952 case BMAP_LEFT_FILLING
:
4954 * Deleting the first part of the extent.
4956 got
->br_startoff
= del_endoff
;
4957 got
->br_blockcount
-= del
->br_blockcount
;
4958 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4959 got
->br_blockcount
), da_old
);
4960 got
->br_startblock
= nullstartblock((int)da_new
);
4961 xfs_iext_update_extent(ip
, state
, icur
, got
);
4963 case BMAP_RIGHT_FILLING
:
4965 * Deleting the last part of the extent.
4967 got
->br_blockcount
= got
->br_blockcount
- del
->br_blockcount
;
4968 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4969 got
->br_blockcount
), da_old
);
4970 got
->br_startblock
= nullstartblock((int)da_new
);
4971 xfs_iext_update_extent(ip
, state
, icur
, got
);
4975 * Deleting the middle of the extent.
4977 * Distribute the original indlen reservation across the two new
4978 * extents. Steal blocks from the deleted extent if necessary.
4979 * Stealing blocks simply fudges the fdblocks accounting below.
4980 * Warn if either of the new indlen reservations is zero as this
4981 * can lead to delalloc problems.
4983 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4984 got_indlen
= xfs_bmap_worst_indlen(ip
, got
->br_blockcount
);
4986 new.br_blockcount
= got_endoff
- del_endoff
;
4987 new_indlen
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
4989 WARN_ON_ONCE(!got_indlen
|| !new_indlen
);
4990 stolen
= xfs_bmap_split_indlen(da_old
, &got_indlen
, &new_indlen
,
4991 del
->br_blockcount
);
4993 got
->br_startblock
= nullstartblock((int)got_indlen
);
4995 new.br_startoff
= del_endoff
;
4996 new.br_state
= got
->br_state
;
4997 new.br_startblock
= nullstartblock((int)new_indlen
);
4999 xfs_iext_update_extent(ip
, state
, icur
, got
);
5000 xfs_iext_next(ifp
, icur
);
5001 xfs_iext_insert(ip
, icur
, &new, state
);
5003 da_new
= got_indlen
+ new_indlen
- stolen
;
5004 del
->br_blockcount
-= stolen
;
5008 ASSERT(da_old
>= da_new
);
5009 da_diff
= da_old
- da_new
;
5011 da_diff
+= del
->br_blockcount
;
5013 xfs_mod_fdblocks(mp
, da_diff
, false);
5014 xfs_mod_delalloc(mp
, -da_diff
);
5020 xfs_bmap_del_extent_cow(
5021 struct xfs_inode
*ip
,
5022 struct xfs_iext_cursor
*icur
,
5023 struct xfs_bmbt_irec
*got
,
5024 struct xfs_bmbt_irec
*del
)
5026 struct xfs_mount
*mp
= ip
->i_mount
;
5027 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
5028 struct xfs_bmbt_irec
new;
5029 xfs_fileoff_t del_endoff
, got_endoff
;
5030 int state
= BMAP_COWFORK
;
5032 XFS_STATS_INC(mp
, xs_del_exlist
);
5034 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5035 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
5037 ASSERT(del
->br_blockcount
> 0);
5038 ASSERT(got
->br_startoff
<= del
->br_startoff
);
5039 ASSERT(got_endoff
>= del_endoff
);
5040 ASSERT(!isnullstartblock(got
->br_startblock
));
5042 if (got
->br_startoff
== del
->br_startoff
)
5043 state
|= BMAP_LEFT_FILLING
;
5044 if (got_endoff
== del_endoff
)
5045 state
|= BMAP_RIGHT_FILLING
;
5047 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
5048 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
5050 * Matches the whole extent. Delete the entry.
5052 xfs_iext_remove(ip
, icur
, state
);
5053 xfs_iext_prev(ifp
, icur
);
5055 case BMAP_LEFT_FILLING
:
5057 * Deleting the first part of the extent.
5059 got
->br_startoff
= del_endoff
;
5060 got
->br_blockcount
-= del
->br_blockcount
;
5061 got
->br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5062 xfs_iext_update_extent(ip
, state
, icur
, got
);
5064 case BMAP_RIGHT_FILLING
:
5066 * Deleting the last part of the extent.
5068 got
->br_blockcount
-= del
->br_blockcount
;
5069 xfs_iext_update_extent(ip
, state
, icur
, got
);
5073 * Deleting the middle of the extent.
5075 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
5077 new.br_startoff
= del_endoff
;
5078 new.br_blockcount
= got_endoff
- del_endoff
;
5079 new.br_state
= got
->br_state
;
5080 new.br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
5082 xfs_iext_update_extent(ip
, state
, icur
, got
);
5083 xfs_iext_next(ifp
, icur
);
5084 xfs_iext_insert(ip
, icur
, &new, state
);
5087 ip
->i_delayed_blks
-= del
->br_blockcount
;
5091 * Called by xfs_bmapi to update file extent records and the btree
5092 * after removing space.
5094 STATIC
int /* error */
5095 xfs_bmap_del_extent_real(
5096 xfs_inode_t
*ip
, /* incore inode pointer */
5097 xfs_trans_t
*tp
, /* current transaction pointer */
5098 struct xfs_iext_cursor
*icur
,
5099 xfs_btree_cur_t
*cur
, /* if null, not a btree */
5100 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
5101 int *logflagsp
, /* inode logging flags */
5102 int whichfork
, /* data or attr fork */
5103 int bflags
) /* bmapi flags */
5105 xfs_fsblock_t del_endblock
=0; /* first block past del */
5106 xfs_fileoff_t del_endoff
; /* first offset past del */
5107 int do_fx
; /* free extent at end of routine */
5108 int error
; /* error return value */
5109 int flags
= 0;/* inode logging flags */
5110 struct xfs_bmbt_irec got
; /* current extent entry */
5111 xfs_fileoff_t got_endoff
; /* first offset past got */
5112 int i
; /* temp state */
5113 struct xfs_ifork
*ifp
; /* inode fork pointer */
5114 xfs_mount_t
*mp
; /* mount structure */
5115 xfs_filblks_t nblks
; /* quota/sb block count */
5116 xfs_bmbt_irec_t
new; /* new record to be inserted */
5118 uint qfield
; /* quota field to update */
5119 int state
= xfs_bmap_fork_to_state(whichfork
);
5120 struct xfs_bmbt_irec old
;
5123 XFS_STATS_INC(mp
, xs_del_exlist
);
5125 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5126 ASSERT(del
->br_blockcount
> 0);
5127 xfs_iext_get_extent(ifp
, icur
, &got
);
5128 ASSERT(got
.br_startoff
<= del
->br_startoff
);
5129 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
5130 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
5131 ASSERT(got_endoff
>= del_endoff
);
5132 ASSERT(!isnullstartblock(got
.br_startblock
));
5137 * If it's the case where the directory code is running with no block
5138 * reservation, and the deleted block is in the middle of its extent,
5139 * and the resulting insert of an extent would cause transformation to
5140 * btree format, then reject it. The calling code will then swap blocks
5141 * around instead. We have to do this now, rather than waiting for the
5142 * conversion to btree format, since the transaction will be dirty then.
5144 if (tp
->t_blk_res
== 0 &&
5145 ifp
->if_format
== XFS_DINODE_FMT_EXTENTS
&&
5146 ifp
->if_nextents
>= XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5147 del
->br_startoff
> got
.br_startoff
&& del_endoff
< got_endoff
)
5150 flags
= XFS_ILOG_CORE
;
5151 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
5155 len
= div_u64_rem(del
->br_blockcount
, mp
->m_sb
.sb_rextsize
,
5159 if (!(bflags
& XFS_BMAPI_REMAP
)) {
5162 bno
= div_u64_rem(del
->br_startblock
,
5163 mp
->m_sb
.sb_rextsize
, &mod
);
5166 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
5172 nblks
= len
* mp
->m_sb
.sb_rextsize
;
5173 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
5176 nblks
= del
->br_blockcount
;
5177 qfield
= XFS_TRANS_DQ_BCOUNT
;
5180 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
5182 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5185 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5186 error
= -EFSCORRUPTED
;
5191 if (got
.br_startoff
== del
->br_startoff
)
5192 state
|= BMAP_LEFT_FILLING
;
5193 if (got_endoff
== del_endoff
)
5194 state
|= BMAP_RIGHT_FILLING
;
5196 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
5197 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
5199 * Matches the whole extent. Delete the entry.
5201 xfs_iext_remove(ip
, icur
, state
);
5202 xfs_iext_prev(ifp
, icur
);
5205 flags
|= XFS_ILOG_CORE
;
5207 flags
|= xfs_ilog_fext(whichfork
);
5210 if ((error
= xfs_btree_delete(cur
, &i
)))
5212 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5213 error
= -EFSCORRUPTED
;
5217 case BMAP_LEFT_FILLING
:
5219 * Deleting the first part of the extent.
5221 got
.br_startoff
= del_endoff
;
5222 got
.br_startblock
= del_endblock
;
5223 got
.br_blockcount
-= del
->br_blockcount
;
5224 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5226 flags
|= xfs_ilog_fext(whichfork
);
5229 error
= xfs_bmbt_update(cur
, &got
);
5233 case BMAP_RIGHT_FILLING
:
5235 * Deleting the last part of the extent.
5237 got
.br_blockcount
-= del
->br_blockcount
;
5238 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5240 flags
|= xfs_ilog_fext(whichfork
);
5243 error
= xfs_bmbt_update(cur
, &got
);
5249 * Deleting the middle of the extent.
5253 * For directories, -ENOSPC is returned since a directory entry
5254 * remove operation must not fail due to low extent count
5255 * availability. -ENOSPC will be handled by higher layers of XFS
5256 * by letting the corresponding empty Data/Free blocks to linger
5257 * until a future remove operation. Dabtree blocks would be
5258 * swapped with the last block in the leaf space and then the
5259 * new last block will be unmapped.
5261 * The above logic also applies to the source directory entry of
5262 * a rename operation.
5264 error
= xfs_iext_count_may_overflow(ip
, whichfork
, 1);
5266 ASSERT(S_ISDIR(VFS_I(ip
)->i_mode
) &&
5267 whichfork
== XFS_DATA_FORK
);
5274 got
.br_blockcount
= del
->br_startoff
- got
.br_startoff
;
5275 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5277 new.br_startoff
= del_endoff
;
5278 new.br_blockcount
= got_endoff
- del_endoff
;
5279 new.br_state
= got
.br_state
;
5280 new.br_startblock
= del_endblock
;
5282 flags
|= XFS_ILOG_CORE
;
5284 error
= xfs_bmbt_update(cur
, &got
);
5287 error
= xfs_btree_increment(cur
, 0, &i
);
5290 cur
->bc_rec
.b
= new;
5291 error
= xfs_btree_insert(cur
, &i
);
5292 if (error
&& error
!= -ENOSPC
)
5295 * If get no-space back from btree insert, it tried a
5296 * split, and we have a zero block reservation. Fix up
5297 * our state and return the error.
5299 if (error
== -ENOSPC
) {
5301 * Reset the cursor, don't trust it after any
5304 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5307 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5308 error
= -EFSCORRUPTED
;
5312 * Update the btree record back
5313 * to the original value.
5315 error
= xfs_bmbt_update(cur
, &old
);
5319 * Reset the extent record back
5320 * to the original value.
5322 xfs_iext_update_extent(ip
, state
, icur
, &old
);
5327 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
5328 error
= -EFSCORRUPTED
;
5332 flags
|= xfs_ilog_fext(whichfork
);
5335 xfs_iext_next(ifp
, icur
);
5336 xfs_iext_insert(ip
, icur
, &new, state
);
5340 /* remove reverse mapping */
5341 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, del
);
5344 * If we need to, add to list of extents to delete.
5346 if (do_fx
&& !(bflags
& XFS_BMAPI_REMAP
)) {
5347 if (xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
) {
5348 xfs_refcount_decrease_extent(tp
, del
);
5350 __xfs_bmap_add_free(tp
, del
->br_startblock
,
5351 del
->br_blockcount
, NULL
,
5352 (bflags
& XFS_BMAPI_NODISCARD
) ||
5353 del
->br_state
== XFS_EXT_UNWRITTEN
);
5358 * Adjust inode # blocks in the file.
5361 ip
->i_d
.di_nblocks
-= nblks
;
5363 * Adjust quota data.
5365 if (qfield
&& !(bflags
& XFS_BMAPI_REMAP
))
5366 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5374 * Unmap (remove) blocks from a file.
5375 * If nexts is nonzero then the number of extents to remove is limited to
5376 * that value. If not all extents in the block range can be removed then
5381 struct xfs_trans
*tp
, /* transaction pointer */
5382 struct xfs_inode
*ip
, /* incore inode */
5383 xfs_fileoff_t start
, /* first file offset deleted */
5384 xfs_filblks_t
*rlen
, /* i/o: amount remaining */
5385 int flags
, /* misc flags */
5386 xfs_extnum_t nexts
) /* number of extents max */
5388 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
5389 struct xfs_bmbt_irec del
; /* extent being deleted */
5390 int error
; /* error return value */
5391 xfs_extnum_t extno
; /* extent number in list */
5392 struct xfs_bmbt_irec got
; /* current extent record */
5393 struct xfs_ifork
*ifp
; /* inode fork pointer */
5394 int isrt
; /* freeing in rt area */
5395 int logflags
; /* transaction logging flags */
5396 xfs_extlen_t mod
; /* rt extent offset */
5397 struct xfs_mount
*mp
= ip
->i_mount
;
5398 int tmp_logflags
; /* partial logging flags */
5399 int wasdel
; /* was a delayed alloc extent */
5400 int whichfork
; /* data or attribute fork */
5402 xfs_filblks_t len
= *rlen
; /* length to unmap in file */
5403 xfs_fileoff_t max_len
;
5404 xfs_agnumber_t prev_agno
= NULLAGNUMBER
, agno
;
5406 struct xfs_iext_cursor icur
;
5409 trace_xfs_bunmap(ip
, start
, len
, flags
, _RET_IP_
);
5411 whichfork
= xfs_bmapi_whichfork(flags
);
5412 ASSERT(whichfork
!= XFS_COW_FORK
);
5413 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5414 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)))
5415 return -EFSCORRUPTED
;
5416 if (XFS_FORCED_SHUTDOWN(mp
))
5419 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5424 * Guesstimate how many blocks we can unmap without running the risk of
5425 * blowing out the transaction with a mix of EFIs and reflink
5428 if (tp
&& xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
)
5429 max_len
= min(len
, xfs_refcount_max_unmap(tp
->t_log_res
));
5433 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5434 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5436 if (xfs_iext_count(ifp
) == 0) {
5440 XFS_STATS_INC(mp
, xs_blk_unmap
);
5441 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5444 if (!xfs_iext_lookup_extent_before(ip
, ifp
, &end
, &icur
, &got
)) {
5451 if (ifp
->if_flags
& XFS_IFBROOT
) {
5452 ASSERT(ifp
->if_format
== XFS_DINODE_FMT_BTREE
);
5453 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5454 cur
->bc_ino
.flags
= 0;
5460 * Synchronize by locking the bitmap inode.
5462 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTBITMAP
);
5463 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5464 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTSUM
);
5465 xfs_trans_ijoin(tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
5469 while (end
!= (xfs_fileoff_t
)-1 && end
>= start
&&
5470 (nexts
== 0 || extno
< nexts
) && max_len
> 0) {
5472 * Is the found extent after a hole in which end lives?
5473 * Just back up to the previous extent, if so.
5475 if (got
.br_startoff
> end
&&
5476 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5481 * Is the last block of this extent before the range
5482 * we're supposed to delete? If so, we're done.
5484 end
= XFS_FILEOFF_MIN(end
,
5485 got
.br_startoff
+ got
.br_blockcount
- 1);
5489 * Then deal with the (possibly delayed) allocated space
5493 wasdel
= isnullstartblock(del
.br_startblock
);
5496 * Make sure we don't touch multiple AGF headers out of order
5497 * in a single transaction, as that could cause AB-BA deadlocks.
5499 if (!wasdel
&& !isrt
) {
5500 agno
= XFS_FSB_TO_AGNO(mp
, del
.br_startblock
);
5501 if (prev_agno
!= NULLAGNUMBER
&& prev_agno
> agno
)
5505 if (got
.br_startoff
< start
) {
5506 del
.br_startoff
= start
;
5507 del
.br_blockcount
-= start
- got
.br_startoff
;
5509 del
.br_startblock
+= start
- got
.br_startoff
;
5511 if (del
.br_startoff
+ del
.br_blockcount
> end
+ 1)
5512 del
.br_blockcount
= end
+ 1 - del
.br_startoff
;
5514 /* How much can we safely unmap? */
5515 if (max_len
< del
.br_blockcount
) {
5516 del
.br_startoff
+= del
.br_blockcount
- max_len
;
5518 del
.br_startblock
+= del
.br_blockcount
- max_len
;
5519 del
.br_blockcount
= max_len
;
5525 sum
= del
.br_startblock
+ del
.br_blockcount
;
5526 div_u64_rem(sum
, mp
->m_sb
.sb_rextsize
, &mod
);
5529 * Realtime extent not lined up at the end.
5530 * The extent could have been split into written
5531 * and unwritten pieces, or we could just be
5532 * unmapping part of it. But we can't really
5533 * get rid of part of a realtime extent.
5535 if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5537 * This piece is unwritten, or we're not
5538 * using unwritten extents. Skip over it.
5541 end
-= mod
> del
.br_blockcount
?
5542 del
.br_blockcount
: mod
;
5543 if (end
< got
.br_startoff
&&
5544 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5551 * It's written, turn it unwritten.
5552 * This is better than zeroing it.
5554 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5555 ASSERT(tp
->t_blk_res
> 0);
5557 * If this spans a realtime extent boundary,
5558 * chop it back to the start of the one we end at.
5560 if (del
.br_blockcount
> mod
) {
5561 del
.br_startoff
+= del
.br_blockcount
- mod
;
5562 del
.br_startblock
+= del
.br_blockcount
- mod
;
5563 del
.br_blockcount
= mod
;
5565 del
.br_state
= XFS_EXT_UNWRITTEN
;
5566 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5567 whichfork
, &icur
, &cur
, &del
,
5573 div_u64_rem(del
.br_startblock
, mp
->m_sb
.sb_rextsize
, &mod
);
5575 xfs_extlen_t off
= mp
->m_sb
.sb_rextsize
- mod
;
5578 * Realtime extent is lined up at the end but not
5579 * at the front. We'll get rid of full extents if
5582 if (del
.br_blockcount
> off
) {
5583 del
.br_blockcount
-= off
;
5584 del
.br_startoff
+= off
;
5585 del
.br_startblock
+= off
;
5586 } else if (del
.br_startoff
== start
&&
5587 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5588 tp
->t_blk_res
== 0)) {
5590 * Can't make it unwritten. There isn't
5591 * a full extent here so just skip it.
5593 ASSERT(end
>= del
.br_blockcount
);
5594 end
-= del
.br_blockcount
;
5595 if (got
.br_startoff
> end
&&
5596 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5601 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5602 struct xfs_bmbt_irec prev
;
5603 xfs_fileoff_t unwrite_start
;
5606 * This one is already unwritten.
5607 * It must have a written left neighbor.
5608 * Unwrite the killed part of that one and
5611 if (!xfs_iext_prev_extent(ifp
, &icur
, &prev
))
5613 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5614 ASSERT(!isnullstartblock(prev
.br_startblock
));
5615 ASSERT(del
.br_startblock
==
5616 prev
.br_startblock
+ prev
.br_blockcount
);
5617 unwrite_start
= max3(start
,
5618 del
.br_startoff
- mod
,
5620 mod
= unwrite_start
- prev
.br_startoff
;
5621 prev
.br_startoff
= unwrite_start
;
5622 prev
.br_startblock
+= mod
;
5623 prev
.br_blockcount
-= mod
;
5624 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5625 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5626 ip
, whichfork
, &icur
, &cur
,
5632 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5633 del
.br_state
= XFS_EXT_UNWRITTEN
;
5634 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5635 ip
, whichfork
, &icur
, &cur
,
5645 error
= xfs_bmap_del_extent_delay(ip
, whichfork
, &icur
,
5648 error
= xfs_bmap_del_extent_real(ip
, tp
, &icur
, cur
,
5649 &del
, &tmp_logflags
, whichfork
,
5651 logflags
|= tmp_logflags
;
5657 max_len
-= del
.br_blockcount
;
5658 end
= del
.br_startoff
- 1;
5661 * If not done go on to the next (previous) record.
5663 if (end
!= (xfs_fileoff_t
)-1 && end
>= start
) {
5664 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
5665 (got
.br_startoff
> end
&&
5666 !xfs_iext_prev_extent(ifp
, &icur
, &got
))) {
5673 if (done
|| end
== (xfs_fileoff_t
)-1 || end
< start
)
5676 *rlen
= end
- start
+ 1;
5679 * Convert to a btree if necessary.
5681 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5682 ASSERT(cur
== NULL
);
5683 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
5684 &tmp_logflags
, whichfork
);
5685 logflags
|= tmp_logflags
;
5687 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
,
5693 * Log everything. Do this after conversion, there's no point in
5694 * logging the extent records if we've converted to btree format.
5696 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5697 ifp
->if_format
!= XFS_DINODE_FMT_EXTENTS
)
5698 logflags
&= ~xfs_ilog_fext(whichfork
);
5699 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5700 ifp
->if_format
!= XFS_DINODE_FMT_BTREE
)
5701 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5703 * Log inode even in the error case, if the transaction
5704 * is dirty we'll need to shut down the filesystem.
5707 xfs_trans_log_inode(tp
, ip
, logflags
);
5710 cur
->bc_ino
.allocated
= 0;
5711 xfs_btree_del_cursor(cur
, error
);
5716 /* Unmap a range of a file. */
5720 struct xfs_inode
*ip
,
5729 error
= __xfs_bunmapi(tp
, ip
, bno
, &len
, flags
, nexts
);
5735 * Determine whether an extent shift can be accomplished by a merge with the
5736 * extent that precedes the target hole of the shift.
5740 struct xfs_bmbt_irec
*left
, /* preceding extent */
5741 struct xfs_bmbt_irec
*got
, /* current extent to shift */
5742 xfs_fileoff_t shift
) /* shift fsb */
5744 xfs_fileoff_t startoff
;
5746 startoff
= got
->br_startoff
- shift
;
5749 * The extent, once shifted, must be adjacent in-file and on-disk with
5750 * the preceding extent.
5752 if ((left
->br_startoff
+ left
->br_blockcount
!= startoff
) ||
5753 (left
->br_startblock
+ left
->br_blockcount
!= got
->br_startblock
) ||
5754 (left
->br_state
!= got
->br_state
) ||
5755 (left
->br_blockcount
+ got
->br_blockcount
> MAXEXTLEN
))
5762 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5763 * hole in the file. If an extent shift would result in the extent being fully
5764 * adjacent to the extent that currently precedes the hole, we can merge with
5765 * the preceding extent rather than do the shift.
5767 * This function assumes the caller has verified a shift-by-merge is possible
5768 * with the provided extents via xfs_bmse_can_merge().
5772 struct xfs_trans
*tp
,
5773 struct xfs_inode
*ip
,
5775 xfs_fileoff_t shift
, /* shift fsb */
5776 struct xfs_iext_cursor
*icur
,
5777 struct xfs_bmbt_irec
*got
, /* extent to shift */
5778 struct xfs_bmbt_irec
*left
, /* preceding extent */
5779 struct xfs_btree_cur
*cur
,
5780 int *logflags
) /* output */
5782 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5783 struct xfs_bmbt_irec
new;
5784 xfs_filblks_t blockcount
;
5786 struct xfs_mount
*mp
= ip
->i_mount
;
5788 blockcount
= left
->br_blockcount
+ got
->br_blockcount
;
5790 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5791 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5792 ASSERT(xfs_bmse_can_merge(left
, got
, shift
));
5795 new.br_blockcount
= blockcount
;
5798 * Update the on-disk extent count, the btree if necessary and log the
5802 *logflags
|= XFS_ILOG_CORE
;
5804 *logflags
|= XFS_ILOG_DEXT
;
5808 /* lookup and remove the extent to merge */
5809 error
= xfs_bmbt_lookup_eq(cur
, got
, &i
);
5812 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5813 return -EFSCORRUPTED
;
5815 error
= xfs_btree_delete(cur
, &i
);
5818 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5819 return -EFSCORRUPTED
;
5821 /* lookup and update size of the previous extent */
5822 error
= xfs_bmbt_lookup_eq(cur
, left
, &i
);
5825 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5826 return -EFSCORRUPTED
;
5828 error
= xfs_bmbt_update(cur
, &new);
5832 /* change to extent format if required after extent removal */
5833 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, logflags
, whichfork
);
5838 xfs_iext_remove(ip
, icur
, 0);
5839 xfs_iext_prev(ifp
, icur
);
5840 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5843 /* update reverse mapping. rmap functions merge the rmaps for us */
5844 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, got
);
5845 memcpy(&new, got
, sizeof(new));
5846 new.br_startoff
= left
->br_startoff
+ left
->br_blockcount
;
5847 xfs_rmap_map_extent(tp
, ip
, whichfork
, &new);
5852 xfs_bmap_shift_update_extent(
5853 struct xfs_trans
*tp
,
5854 struct xfs_inode
*ip
,
5856 struct xfs_iext_cursor
*icur
,
5857 struct xfs_bmbt_irec
*got
,
5858 struct xfs_btree_cur
*cur
,
5860 xfs_fileoff_t startoff
)
5862 struct xfs_mount
*mp
= ip
->i_mount
;
5863 struct xfs_bmbt_irec prev
= *got
;
5866 *logflags
|= XFS_ILOG_CORE
;
5868 got
->br_startoff
= startoff
;
5871 error
= xfs_bmbt_lookup_eq(cur
, &prev
, &i
);
5874 if (XFS_IS_CORRUPT(mp
, i
!= 1))
5875 return -EFSCORRUPTED
;
5877 error
= xfs_bmbt_update(cur
, got
);
5881 *logflags
|= XFS_ILOG_DEXT
;
5884 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5887 /* update reverse mapping */
5888 xfs_rmap_unmap_extent(tp
, ip
, whichfork
, &prev
);
5889 xfs_rmap_map_extent(tp
, ip
, whichfork
, got
);
5894 xfs_bmap_collapse_extents(
5895 struct xfs_trans
*tp
,
5896 struct xfs_inode
*ip
,
5897 xfs_fileoff_t
*next_fsb
,
5898 xfs_fileoff_t offset_shift_fsb
,
5901 int whichfork
= XFS_DATA_FORK
;
5902 struct xfs_mount
*mp
= ip
->i_mount
;
5903 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5904 struct xfs_btree_cur
*cur
= NULL
;
5905 struct xfs_bmbt_irec got
, prev
;
5906 struct xfs_iext_cursor icur
;
5907 xfs_fileoff_t new_startoff
;
5911 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
5912 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
5913 return -EFSCORRUPTED
;
5916 if (XFS_FORCED_SHUTDOWN(mp
))
5919 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
5921 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5922 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5927 if (ifp
->if_flags
& XFS_IFBROOT
) {
5928 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5929 cur
->bc_ino
.flags
= 0;
5932 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
5936 if (XFS_IS_CORRUPT(mp
, isnullstartblock(got
.br_startblock
))) {
5937 error
= -EFSCORRUPTED
;
5941 new_startoff
= got
.br_startoff
- offset_shift_fsb
;
5942 if (xfs_iext_peek_prev_extent(ifp
, &icur
, &prev
)) {
5943 if (new_startoff
< prev
.br_startoff
+ prev
.br_blockcount
) {
5948 if (xfs_bmse_can_merge(&prev
, &got
, offset_shift_fsb
)) {
5949 error
= xfs_bmse_merge(tp
, ip
, whichfork
,
5950 offset_shift_fsb
, &icur
, &got
, &prev
,
5957 if (got
.br_startoff
< offset_shift_fsb
) {
5963 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
5964 cur
, &logflags
, new_startoff
);
5969 if (!xfs_iext_next_extent(ifp
, &icur
, &got
)) {
5974 *next_fsb
= got
.br_startoff
;
5977 xfs_btree_del_cursor(cur
, error
);
5979 xfs_trans_log_inode(tp
, ip
, logflags
);
5983 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5985 xfs_bmap_can_insert_extents(
5986 struct xfs_inode
*ip
,
5988 xfs_fileoff_t shift
)
5990 struct xfs_bmbt_irec got
;
5994 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5996 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
5999 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
6000 error
= xfs_bmap_last_extent(NULL
, ip
, XFS_DATA_FORK
, &got
, &is_empty
);
6001 if (!error
&& !is_empty
&& got
.br_startoff
>= off
&&
6002 ((got
.br_startoff
+ shift
) & BMBT_STARTOFF_MASK
) < got
.br_startoff
)
6004 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
6010 xfs_bmap_insert_extents(
6011 struct xfs_trans
*tp
,
6012 struct xfs_inode
*ip
,
6013 xfs_fileoff_t
*next_fsb
,
6014 xfs_fileoff_t offset_shift_fsb
,
6016 xfs_fileoff_t stop_fsb
)
6018 int whichfork
= XFS_DATA_FORK
;
6019 struct xfs_mount
*mp
= ip
->i_mount
;
6020 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6021 struct xfs_btree_cur
*cur
= NULL
;
6022 struct xfs_bmbt_irec got
, next
;
6023 struct xfs_iext_cursor icur
;
6024 xfs_fileoff_t new_startoff
;
6028 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
6029 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
6030 return -EFSCORRUPTED
;
6033 if (XFS_FORCED_SHUTDOWN(mp
))
6036 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
6038 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6039 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6044 if (ifp
->if_flags
& XFS_IFBROOT
) {
6045 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6046 cur
->bc_ino
.flags
= 0;
6049 if (*next_fsb
== NULLFSBLOCK
) {
6050 xfs_iext_last(ifp
, &icur
);
6051 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
6052 stop_fsb
> got
.br_startoff
) {
6057 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
6062 if (XFS_IS_CORRUPT(mp
, isnullstartblock(got
.br_startblock
))) {
6063 error
= -EFSCORRUPTED
;
6067 if (XFS_IS_CORRUPT(mp
, stop_fsb
> got
.br_startoff
)) {
6068 error
= -EFSCORRUPTED
;
6072 new_startoff
= got
.br_startoff
+ offset_shift_fsb
;
6073 if (xfs_iext_peek_next_extent(ifp
, &icur
, &next
)) {
6074 if (new_startoff
+ got
.br_blockcount
> next
.br_startoff
) {
6080 * Unlike a left shift (which involves a hole punch), a right
6081 * shift does not modify extent neighbors in any way. We should
6082 * never find mergeable extents in this scenario. Check anyways
6083 * and warn if we encounter two extents that could be one.
6085 if (xfs_bmse_can_merge(&got
, &next
, offset_shift_fsb
))
6089 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
6090 cur
, &logflags
, new_startoff
);
6094 if (!xfs_iext_prev_extent(ifp
, &icur
, &got
) ||
6095 stop_fsb
>= got
.br_startoff
+ got
.br_blockcount
) {
6100 *next_fsb
= got
.br_startoff
;
6103 xfs_btree_del_cursor(cur
, error
);
6105 xfs_trans_log_inode(tp
, ip
, logflags
);
6110 * Splits an extent into two extents at split_fsb block such that it is the
6111 * first block of the current_ext. @ext is a target extent to be split.
6112 * @split_fsb is a block where the extents is split. If split_fsb lies in a
6113 * hole or the first block of extents, just return 0.
6116 xfs_bmap_split_extent(
6117 struct xfs_trans
*tp
,
6118 struct xfs_inode
*ip
,
6119 xfs_fileoff_t split_fsb
)
6121 int whichfork
= XFS_DATA_FORK
;
6122 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6123 struct xfs_btree_cur
*cur
= NULL
;
6124 struct xfs_bmbt_irec got
;
6125 struct xfs_bmbt_irec
new; /* split extent */
6126 struct xfs_mount
*mp
= ip
->i_mount
;
6127 xfs_fsblock_t gotblkcnt
; /* new block count for got */
6128 struct xfs_iext_cursor icur
;
6133 if (XFS_IS_CORRUPT(mp
, !xfs_ifork_has_extents(ifp
)) ||
6134 XFS_TEST_ERROR(false, mp
, XFS_ERRTAG_BMAPIFORMAT
)) {
6135 return -EFSCORRUPTED
;
6138 if (XFS_FORCED_SHUTDOWN(mp
))
6141 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
6142 /* Read in all the extents */
6143 error
= xfs_iread_extents(tp
, ip
, whichfork
);
6149 * If there are not extents, or split_fsb lies in a hole we are done.
6151 if (!xfs_iext_lookup_extent(ip
, ifp
, split_fsb
, &icur
, &got
) ||
6152 got
.br_startoff
>= split_fsb
)
6155 gotblkcnt
= split_fsb
- got
.br_startoff
;
6156 new.br_startoff
= split_fsb
;
6157 new.br_startblock
= got
.br_startblock
+ gotblkcnt
;
6158 new.br_blockcount
= got
.br_blockcount
- gotblkcnt
;
6159 new.br_state
= got
.br_state
;
6161 if (ifp
->if_flags
& XFS_IFBROOT
) {
6162 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
6163 cur
->bc_ino
.flags
= 0;
6164 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
6167 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
6168 error
= -EFSCORRUPTED
;
6173 got
.br_blockcount
= gotblkcnt
;
6174 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), &icur
,
6177 logflags
= XFS_ILOG_CORE
;
6179 error
= xfs_bmbt_update(cur
, &got
);
6183 logflags
|= XFS_ILOG_DEXT
;
6185 /* Add new extent */
6186 xfs_iext_next(ifp
, &icur
);
6187 xfs_iext_insert(ip
, &icur
, &new, 0);
6191 error
= xfs_bmbt_lookup_eq(cur
, &new, &i
);
6194 if (XFS_IS_CORRUPT(mp
, i
!= 0)) {
6195 error
= -EFSCORRUPTED
;
6198 error
= xfs_btree_insert(cur
, &i
);
6201 if (XFS_IS_CORRUPT(mp
, i
!= 1)) {
6202 error
= -EFSCORRUPTED
;
6208 * Convert to a btree if necessary.
6210 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
6211 int tmp_logflags
; /* partial log flag return val */
6213 ASSERT(cur
== NULL
);
6214 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
6215 &tmp_logflags
, whichfork
);
6216 logflags
|= tmp_logflags
;
6221 cur
->bc_ino
.allocated
= 0;
6222 xfs_btree_del_cursor(cur
, error
);
6226 xfs_trans_log_inode(tp
, ip
, logflags
);
6230 /* Deferred mapping is only for real extents in the data fork. */
6232 xfs_bmap_is_update_needed(
6233 struct xfs_bmbt_irec
*bmap
)
6235 return bmap
->br_startblock
!= HOLESTARTBLOCK
&&
6236 bmap
->br_startblock
!= DELAYSTARTBLOCK
;
6239 /* Record a bmap intent. */
6242 struct xfs_trans
*tp
,
6243 enum xfs_bmap_intent_type type
,
6244 struct xfs_inode
*ip
,
6246 struct xfs_bmbt_irec
*bmap
)
6248 struct xfs_bmap_intent
*bi
;
6250 trace_xfs_bmap_defer(tp
->t_mountp
,
6251 XFS_FSB_TO_AGNO(tp
->t_mountp
, bmap
->br_startblock
),
6253 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bmap
->br_startblock
),
6254 ip
->i_ino
, whichfork
,
6256 bmap
->br_blockcount
,
6259 bi
= kmem_alloc(sizeof(struct xfs_bmap_intent
), KM_NOFS
);
6260 INIT_LIST_HEAD(&bi
->bi_list
);
6263 bi
->bi_whichfork
= whichfork
;
6264 bi
->bi_bmap
= *bmap
;
6266 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_BMAP
, &bi
->bi_list
);
6270 /* Map an extent into a file. */
6272 xfs_bmap_map_extent(
6273 struct xfs_trans
*tp
,
6274 struct xfs_inode
*ip
,
6275 struct xfs_bmbt_irec
*PREV
)
6277 if (!xfs_bmap_is_update_needed(PREV
))
6280 __xfs_bmap_add(tp
, XFS_BMAP_MAP
, ip
, XFS_DATA_FORK
, PREV
);
6283 /* Unmap an extent out of a file. */
6285 xfs_bmap_unmap_extent(
6286 struct xfs_trans
*tp
,
6287 struct xfs_inode
*ip
,
6288 struct xfs_bmbt_irec
*PREV
)
6290 if (!xfs_bmap_is_update_needed(PREV
))
6293 __xfs_bmap_add(tp
, XFS_BMAP_UNMAP
, ip
, XFS_DATA_FORK
, PREV
);
6297 * Process one of the deferred bmap operations. We pass back the
6298 * btree cursor to maintain our lock on the bmapbt between calls.
6301 xfs_bmap_finish_one(
6302 struct xfs_trans
*tp
,
6303 struct xfs_inode
*ip
,
6304 enum xfs_bmap_intent_type type
,
6306 xfs_fileoff_t startoff
,
6307 xfs_fsblock_t startblock
,
6308 xfs_filblks_t
*blockcount
,
6313 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
);
6315 trace_xfs_bmap_deferred(tp
->t_mountp
,
6316 XFS_FSB_TO_AGNO(tp
->t_mountp
, startblock
), type
,
6317 XFS_FSB_TO_AGBNO(tp
->t_mountp
, startblock
),
6318 ip
->i_ino
, whichfork
, startoff
, *blockcount
, state
);
6320 if (WARN_ON_ONCE(whichfork
!= XFS_DATA_FORK
))
6321 return -EFSCORRUPTED
;
6323 if (XFS_TEST_ERROR(false, tp
->t_mountp
,
6324 XFS_ERRTAG_BMAP_FINISH_ONE
))
6329 error
= xfs_bmapi_remap(tp
, ip
, startoff
, *blockcount
,
6333 case XFS_BMAP_UNMAP
:
6334 error
= __xfs_bunmapi(tp
, ip
, startoff
, blockcount
,
6335 XFS_BMAPI_REMAP
, 1);
6339 error
= -EFSCORRUPTED
;
6345 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6347 xfs_bmap_validate_extent(
6348 struct xfs_inode
*ip
,
6350 struct xfs_bmbt_irec
*irec
)
6352 struct xfs_mount
*mp
= ip
->i_mount
;
6354 if (!xfs_verify_fileext(mp
, irec
->br_startoff
, irec
->br_blockcount
))
6355 return __this_address
;
6357 if (XFS_IS_REALTIME_INODE(ip
) && whichfork
== XFS_DATA_FORK
) {
6358 if (!xfs_verify_rtext(mp
, irec
->br_startblock
,
6359 irec
->br_blockcount
))
6360 return __this_address
;
6362 if (!xfs_verify_fsbext(mp
, irec
->br_startblock
,
6363 irec
->br_blockcount
))
6364 return __this_address
;
6366 if (irec
->br_state
!= XFS_EXT_NORM
&& whichfork
!= XFS_DATA_FORK
)
6367 return __this_address
;