1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
19 #include "xfs_inode.h"
20 #include "xfs_btree.h"
21 #include "xfs_trans.h"
22 #include "xfs_inode_item.h"
23 #include "xfs_extfree_item.h"
24 #include "xfs_alloc.h"
26 #include "xfs_bmap_util.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_rtalloc.h"
29 #include "xfs_errortag.h"
30 #include "xfs_error.h"
31 #include "xfs_quota.h"
32 #include "xfs_trans_space.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_trace.h"
35 #include "xfs_symlink.h"
36 #include "xfs_attr_leaf.h"
37 #include "xfs_filestream.h"
39 #include "xfs_ag_resv.h"
40 #include "xfs_refcount.h"
41 #include "xfs_icache.h"
44 kmem_zone_t
*xfs_bmap_free_item_zone
;
47 * Miscellaneous helper functions
51 * Compute and fill in the value of the maximum depth of a bmap btree
52 * in this filesystem. Done once, during mount.
55 xfs_bmap_compute_maxlevels(
56 xfs_mount_t
*mp
, /* file system mount structure */
57 int whichfork
) /* data or attr fork */
59 int level
; /* btree level */
60 uint maxblocks
; /* max blocks at this level */
61 uint maxleafents
; /* max leaf entries possible */
62 int maxrootrecs
; /* max records in root block */
63 int minleafrecs
; /* min records in leaf block */
64 int minnoderecs
; /* min records in node block */
65 int sz
; /* root block size */
68 * The maximum number of extents in a file, hence the maximum
69 * number of leaf entries, is controlled by the type of di_nextents
70 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
71 * (a signed 16-bit number, xfs_aextnum_t).
73 * Note that we can no longer assume that if we are in ATTR1 that
74 * the fork offset of all the inodes will be
75 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
76 * with ATTR2 and then mounted back with ATTR1, keeping the
77 * di_forkoff's fixed but probably at various positions. Therefore,
78 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
79 * of a minimum size available.
81 if (whichfork
== XFS_DATA_FORK
) {
82 maxleafents
= MAXEXTNUM
;
83 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
85 maxleafents
= MAXAEXTNUM
;
86 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
88 maxrootrecs
= xfs_bmdr_maxrecs(sz
, 0);
89 minleafrecs
= mp
->m_bmap_dmnr
[0];
90 minnoderecs
= mp
->m_bmap_dmnr
[1];
91 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
92 for (level
= 1; maxblocks
> 1; level
++) {
93 if (maxblocks
<= maxrootrecs
)
96 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
98 mp
->m_bm_maxlevels
[whichfork
] = level
;
101 STATIC
int /* error */
103 struct xfs_btree_cur
*cur
,
104 struct xfs_bmbt_irec
*irec
,
105 int *stat
) /* success/failure */
107 cur
->bc_rec
.b
= *irec
;
108 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
111 STATIC
int /* error */
112 xfs_bmbt_lookup_first(
113 struct xfs_btree_cur
*cur
,
114 int *stat
) /* success/failure */
116 cur
->bc_rec
.b
.br_startoff
= 0;
117 cur
->bc_rec
.b
.br_startblock
= 0;
118 cur
->bc_rec
.b
.br_blockcount
= 0;
119 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
123 * Check if the inode needs to be converted to btree format.
125 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
127 return whichfork
!= XFS_COW_FORK
&&
128 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
129 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
130 XFS_IFORK_MAXEXT(ip
, whichfork
);
134 * Check if the inode should be converted to extent format.
136 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
138 return whichfork
!= XFS_COW_FORK
&&
139 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
140 XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
141 XFS_IFORK_MAXEXT(ip
, whichfork
);
145 * Update the record referred to by cur to the value given by irec
146 * This either works (return 0) or gets an EFSCORRUPTED error.
150 struct xfs_btree_cur
*cur
,
151 struct xfs_bmbt_irec
*irec
)
153 union xfs_btree_rec rec
;
155 xfs_bmbt_disk_set_all(&rec
.bmbt
, irec
);
156 return xfs_btree_update(cur
, &rec
);
160 * Compute the worst-case number of indirect blocks that will be used
161 * for ip's delayed extent of length "len".
164 xfs_bmap_worst_indlen(
165 xfs_inode_t
*ip
, /* incore inode pointer */
166 xfs_filblks_t len
) /* delayed extent length */
168 int level
; /* btree level number */
169 int maxrecs
; /* maximum record count at this level */
170 xfs_mount_t
*mp
; /* mount structure */
171 xfs_filblks_t rval
; /* return value */
174 maxrecs
= mp
->m_bmap_dmxr
[0];
175 for (level
= 0, rval
= 0;
176 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
179 do_div(len
, maxrecs
);
182 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
185 maxrecs
= mp
->m_bmap_dmxr
[1];
191 * Calculate the default attribute fork offset for newly created inodes.
194 xfs_default_attroffset(
195 struct xfs_inode
*ip
)
197 struct xfs_mount
*mp
= ip
->i_mount
;
200 if (mp
->m_sb
.sb_inodesize
== 256) {
201 offset
= XFS_LITINO(mp
, ip
->i_d
.di_version
) -
202 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
204 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
207 ASSERT(offset
< XFS_LITINO(mp
, ip
->i_d
.di_version
));
212 * Helper routine to reset inode di_forkoff field when switching
213 * attribute fork from local to extent format - we reset it where
214 * possible to make space available for inline data fork extents.
217 xfs_bmap_forkoff_reset(
221 if (whichfork
== XFS_ATTR_FORK
&&
222 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
223 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
224 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
226 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
227 ip
->i_d
.di_forkoff
= dfl_forkoff
;
232 STATIC
struct xfs_buf
*
234 struct xfs_btree_cur
*cur
,
237 struct xfs_log_item
*lip
;
243 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
244 if (!cur
->bc_bufs
[i
])
246 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
247 return cur
->bc_bufs
[i
];
250 /* Chase down all the log items to see if the bp is there */
251 list_for_each_entry(lip
, &cur
->bc_tp
->t_items
, li_trans
) {
252 struct xfs_buf_log_item
*bip
= (struct xfs_buf_log_item
*)lip
;
254 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
255 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
264 struct xfs_btree_block
*block
,
270 __be64
*pp
, *thispa
; /* pointer to block address */
271 xfs_bmbt_key_t
*prevp
, *keyp
;
273 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
276 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
277 dmxr
= mp
->m_bmap_dmxr
[0];
278 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
281 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
282 be64_to_cpu(keyp
->br_startoff
));
287 * Compare the block numbers to see if there are dups.
290 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
292 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
294 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
296 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
298 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
299 if (*thispa
== *pp
) {
300 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
302 (unsigned long long)be64_to_cpu(*thispa
));
303 xfs_err(mp
, "%s: ptrs are equal in node\n",
305 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
312 * Check that the extents for the inode ip are in the right order in all
313 * btree leaves. THis becomes prohibitively expensive for large extent count
314 * files, so don't bother with inodes that have more than 10,000 extents in
315 * them. The btree record ordering checks will still be done, so for such large
316 * bmapbt constructs that is going to catch most corruptions.
319 xfs_bmap_check_leaf_extents(
320 xfs_btree_cur_t
*cur
, /* btree cursor or null */
321 xfs_inode_t
*ip
, /* incore inode pointer */
322 int whichfork
) /* data or attr fork */
324 struct xfs_btree_block
*block
; /* current btree block */
325 xfs_fsblock_t bno
; /* block # of "block" */
326 xfs_buf_t
*bp
; /* buffer for "block" */
327 int error
; /* error return value */
328 xfs_extnum_t i
=0, j
; /* index into the extents list */
329 struct xfs_ifork
*ifp
; /* fork structure */
330 int level
; /* btree level, for checking */
331 xfs_mount_t
*mp
; /* file system mount structure */
332 __be64
*pp
; /* pointer to block address */
333 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
334 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
335 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
338 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
342 /* skip large extent count inodes */
343 if (ip
->i_d
.di_nextents
> 10000)
348 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
349 block
= ifp
->if_broot
;
351 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
353 level
= be16_to_cpu(block
->bb_level
);
355 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
356 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
357 bno
= be64_to_cpu(*pp
);
359 ASSERT(bno
!= NULLFSBLOCK
);
360 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
361 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
364 * Go down the tree until leaf level is reached, following the first
365 * pointer (leftmost) at each level.
367 while (level
-- > 0) {
368 /* See if buf is in cur first */
370 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
373 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
379 block
= XFS_BUF_TO_BLOCK(bp
);
384 * Check this block for basic sanity (increasing keys and
385 * no duplicate blocks).
388 xfs_check_block(block
, mp
, 0, 0);
389 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
390 bno
= be64_to_cpu(*pp
);
391 XFS_WANT_CORRUPTED_GOTO(mp
,
392 xfs_verify_fsbno(mp
, bno
), error0
);
395 xfs_trans_brelse(NULL
, bp
);
400 * Here with bp and block set to the leftmost leaf node in the tree.
405 * Loop over all leaf nodes checking that all extents are in the right order.
408 xfs_fsblock_t nextbno
;
409 xfs_extnum_t num_recs
;
412 num_recs
= xfs_btree_get_numrecs(block
);
415 * Read-ahead the next leaf block, if any.
418 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
421 * Check all the extents to make sure they are OK.
422 * If we had a previous block, the last entry should
423 * conform with the first entry in this one.
426 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
428 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
429 xfs_bmbt_disk_get_blockcount(&last
) <=
430 xfs_bmbt_disk_get_startoff(ep
));
432 for (j
= 1; j
< num_recs
; j
++) {
433 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
434 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
435 xfs_bmbt_disk_get_blockcount(ep
) <=
436 xfs_bmbt_disk_get_startoff(nextp
));
444 xfs_trans_brelse(NULL
, bp
);
448 * If we've reached the end, stop.
450 if (bno
== NULLFSBLOCK
)
454 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
457 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
463 block
= XFS_BUF_TO_BLOCK(bp
);
469 xfs_warn(mp
, "%s: at error0", __func__
);
471 xfs_trans_brelse(NULL
, bp
);
473 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
475 xfs_err(mp
, "%s: CORRUPTED BTREE OR SOMETHING", __func__
);
476 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
481 * Validate that the bmbt_irecs being returned from bmapi are valid
482 * given the caller's original parameters. Specifically check the
483 * ranges of the returned irecs to ensure that they only extend beyond
484 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
487 xfs_bmap_validate_ret(
491 xfs_bmbt_irec_t
*mval
,
495 int i
; /* index to map values */
497 ASSERT(ret_nmap
<= nmap
);
499 for (i
= 0; i
< ret_nmap
; i
++) {
500 ASSERT(mval
[i
].br_blockcount
> 0);
501 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
502 ASSERT(mval
[i
].br_startoff
>= bno
);
503 ASSERT(mval
[i
].br_blockcount
<= len
);
504 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
507 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
508 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
512 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
513 mval
[i
].br_startoff
);
514 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
515 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
516 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
517 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
522 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
523 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
527 * bmap free list manipulation functions
531 * Add the extent to the list of extents to be free at transaction end.
532 * The list is maintained sorted (by block number).
536 struct xfs_trans
*tp
,
539 const struct xfs_owner_info
*oinfo
,
542 struct xfs_extent_free_item
*new; /* new element */
544 struct xfs_mount
*mp
= tp
->t_mountp
;
548 ASSERT(bno
!= NULLFSBLOCK
);
550 ASSERT(len
<= MAXEXTLEN
);
551 ASSERT(!isnullstartblock(bno
));
552 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
553 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
554 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
555 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
556 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
557 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
559 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
561 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
562 new->xefi_startblock
= bno
;
563 new->xefi_blockcount
= (xfs_extlen_t
)len
;
565 new->xefi_oinfo
= *oinfo
;
567 new->xefi_oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
568 new->xefi_skip_discard
= skip_discard
;
569 trace_xfs_bmap_free_defer(tp
->t_mountp
,
570 XFS_FSB_TO_AGNO(tp
->t_mountp
, bno
), 0,
571 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bno
), len
);
572 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_FREE
, &new->xefi_list
);
576 * Inode fork format manipulation functions
580 * Convert the inode format to extent format if it currently is in btree format,
581 * but the extent list is small enough that it fits into the extent format.
583 * Since the extents are already in-core, all we have to do is give up the space
584 * for the btree root and pitch the leaf block.
586 STATIC
int /* error */
587 xfs_bmap_btree_to_extents(
588 struct xfs_trans
*tp
, /* transaction pointer */
589 struct xfs_inode
*ip
, /* incore inode pointer */
590 struct xfs_btree_cur
*cur
, /* btree cursor */
591 int *logflagsp
, /* inode logging flags */
592 int whichfork
) /* data or attr fork */
594 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
595 struct xfs_mount
*mp
= ip
->i_mount
;
596 struct xfs_btree_block
*rblock
= ifp
->if_broot
;
597 struct xfs_btree_block
*cblock
;/* child btree block */
598 xfs_fsblock_t cbno
; /* child block number */
599 xfs_buf_t
*cbp
; /* child block's buffer */
600 int error
; /* error return value */
601 __be64
*pp
; /* ptr to block address */
602 struct xfs_owner_info oinfo
;
604 /* check if we actually need the extent format first: */
605 if (!xfs_bmap_wants_extents(ip
, whichfork
))
609 ASSERT(whichfork
!= XFS_COW_FORK
);
610 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
611 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
612 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
613 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
614 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
616 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
617 cbno
= be64_to_cpu(*pp
);
619 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
,
620 xfs_btree_check_lptr(cur
, cbno
, 1));
622 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
, XFS_BMAP_BTREE_REF
,
626 cblock
= XFS_BUF_TO_BLOCK(cbp
);
627 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
629 xfs_rmap_ino_bmbt_owner(&oinfo
, ip
->i_ino
, whichfork
);
630 xfs_bmap_add_free(cur
->bc_tp
, cbno
, 1, &oinfo
);
631 ip
->i_d
.di_nblocks
--;
632 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
633 xfs_trans_binval(tp
, cbp
);
634 if (cur
->bc_bufs
[0] == cbp
)
635 cur
->bc_bufs
[0] = NULL
;
636 xfs_iroot_realloc(ip
, -1, whichfork
);
637 ASSERT(ifp
->if_broot
== NULL
);
638 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
639 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
640 *logflagsp
|= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
645 * Convert an extents-format file into a btree-format file.
646 * The new file will have a root block (in the inode) and a single child block.
648 STATIC
int /* error */
649 xfs_bmap_extents_to_btree(
650 struct xfs_trans
*tp
, /* transaction pointer */
651 struct xfs_inode
*ip
, /* incore inode pointer */
652 struct xfs_btree_cur
**curp
, /* cursor returned to caller */
653 int wasdel
, /* converting a delayed alloc */
654 int *logflagsp
, /* inode logging flags */
655 int whichfork
) /* data or attr fork */
657 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
658 struct xfs_buf
*abp
; /* buffer for ablock */
659 struct xfs_alloc_arg args
; /* allocation arguments */
660 struct xfs_bmbt_rec
*arp
; /* child record pointer */
661 struct xfs_btree_block
*block
; /* btree root block */
662 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
663 int error
; /* error return value */
664 struct xfs_ifork
*ifp
; /* inode fork pointer */
665 struct xfs_bmbt_key
*kp
; /* root block key pointer */
666 struct xfs_mount
*mp
; /* mount structure */
667 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
668 struct xfs_iext_cursor icur
;
669 struct xfs_bmbt_irec rec
;
670 xfs_extnum_t cnt
= 0;
673 ASSERT(whichfork
!= XFS_COW_FORK
);
674 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
675 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
678 * Make space in the inode incore. This needs to be undone if we fail
679 * to expand the root.
681 xfs_iroot_realloc(ip
, 1, whichfork
);
682 ifp
->if_flags
|= XFS_IFBROOT
;
687 block
= ifp
->if_broot
;
688 xfs_btree_init_block_int(mp
, block
, XFS_BUF_DADDR_NULL
,
689 XFS_BTNUM_BMAP
, 1, 1, ip
->i_ino
,
690 XFS_BTREE_LONG_PTRS
);
692 * Need a cursor. Can't allocate until bb_level is filled in.
694 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
695 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
697 * Convert to a btree with two levels, one record in root.
699 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
700 memset(&args
, 0, sizeof(args
));
703 xfs_rmap_ino_bmbt_owner(&args
.oinfo
, ip
->i_ino
, whichfork
);
704 if (tp
->t_firstblock
== NULLFSBLOCK
) {
705 args
.type
= XFS_ALLOCTYPE_START_BNO
;
706 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
707 } else if (tp
->t_flags
& XFS_TRANS_LOWMODE
) {
708 args
.type
= XFS_ALLOCTYPE_START_BNO
;
709 args
.fsbno
= tp
->t_firstblock
;
711 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
712 args
.fsbno
= tp
->t_firstblock
;
714 args
.minlen
= args
.maxlen
= args
.prod
= 1;
715 args
.wasdel
= wasdel
;
717 error
= xfs_alloc_vextent(&args
);
719 goto out_root_realloc
;
721 if (WARN_ON_ONCE(args
.fsbno
== NULLFSBLOCK
)) {
723 goto out_root_realloc
;
727 * Allocation can't fail, the space was reserved.
729 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
||
730 args
.agno
>= XFS_FSB_TO_AGNO(mp
, tp
->t_firstblock
));
731 tp
->t_firstblock
= args
.fsbno
;
732 cur
->bc_private
.b
.allocated
++;
733 ip
->i_d
.di_nblocks
++;
734 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
735 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
737 error
= -EFSCORRUPTED
;
738 goto out_unreserve_dquot
;
742 * Fill in the child block.
744 abp
->b_ops
= &xfs_bmbt_buf_ops
;
745 ablock
= XFS_BUF_TO_BLOCK(abp
);
746 xfs_btree_init_block_int(mp
, ablock
, abp
->b_bn
,
747 XFS_BTNUM_BMAP
, 0, 0, ip
->i_ino
,
748 XFS_BTREE_LONG_PTRS
);
750 for_each_xfs_iext(ifp
, &icur
, &rec
) {
751 if (isnullstartblock(rec
.br_startblock
))
753 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1 + cnt
);
754 xfs_bmbt_disk_set_all(arp
, &rec
);
757 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
758 xfs_btree_set_numrecs(ablock
, cnt
);
761 * Fill in the root key and pointer.
763 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
764 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
765 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
766 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
767 be16_to_cpu(block
->bb_level
)));
768 *pp
= cpu_to_be64(args
.fsbno
);
771 * Do all this logging at the end so that
772 * the root is at the right level.
774 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
775 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
776 ASSERT(*curp
== NULL
);
778 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
782 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
784 xfs_iroot_realloc(ip
, -1, whichfork
);
785 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
786 ASSERT(ifp
->if_broot
== NULL
);
787 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
793 * Convert a local file to an extents file.
794 * This code is out of bounds for data forks of regular files,
795 * since the file data needs to get logged so things will stay consistent.
796 * (The bmap-level manipulations are ok, though).
799 xfs_bmap_local_to_extents_empty(
800 struct xfs_inode
*ip
,
803 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
805 ASSERT(whichfork
!= XFS_COW_FORK
);
806 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
807 ASSERT(ifp
->if_bytes
== 0);
808 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
810 xfs_bmap_forkoff_reset(ip
, whichfork
);
811 ifp
->if_flags
&= ~XFS_IFINLINE
;
812 ifp
->if_flags
|= XFS_IFEXTENTS
;
813 ifp
->if_u1
.if_root
= NULL
;
815 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
819 STATIC
int /* error */
820 xfs_bmap_local_to_extents(
821 xfs_trans_t
*tp
, /* transaction pointer */
822 xfs_inode_t
*ip
, /* incore inode pointer */
823 xfs_extlen_t total
, /* total blocks needed by transaction */
824 int *logflagsp
, /* inode logging flags */
826 void (*init_fn
)(struct xfs_trans
*tp
,
828 struct xfs_inode
*ip
,
829 struct xfs_ifork
*ifp
))
832 int flags
; /* logging flags returned */
833 struct xfs_ifork
*ifp
; /* inode fork pointer */
834 xfs_alloc_arg_t args
; /* allocation arguments */
835 xfs_buf_t
*bp
; /* buffer for extent block */
836 struct xfs_bmbt_irec rec
;
837 struct xfs_iext_cursor icur
;
840 * We don't want to deal with the case of keeping inode data inline yet.
841 * So sending the data fork of a regular inode is invalid.
843 ASSERT(!(S_ISREG(VFS_I(ip
)->i_mode
) && whichfork
== XFS_DATA_FORK
));
844 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
845 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
847 if (!ifp
->if_bytes
) {
848 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
849 flags
= XFS_ILOG_CORE
;
855 ASSERT((ifp
->if_flags
& (XFS_IFINLINE
|XFS_IFEXTENTS
)) == XFS_IFINLINE
);
856 memset(&args
, 0, sizeof(args
));
858 args
.mp
= ip
->i_mount
;
859 xfs_rmap_ino_owner(&args
.oinfo
, ip
->i_ino
, whichfork
, 0);
861 * Allocate a block. We know we need only one, since the
862 * file currently fits in an inode.
864 if (tp
->t_firstblock
== NULLFSBLOCK
) {
865 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
866 args
.type
= XFS_ALLOCTYPE_START_BNO
;
868 args
.fsbno
= tp
->t_firstblock
;
869 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
872 args
.minlen
= args
.maxlen
= args
.prod
= 1;
873 error
= xfs_alloc_vextent(&args
);
877 /* Can't fail, the space was reserved. */
878 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
879 ASSERT(args
.len
== 1);
880 tp
->t_firstblock
= args
.fsbno
;
881 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
884 * Initialize the block, copy the data and log the remote buffer.
886 * The callout is responsible for logging because the remote format
887 * might differ from the local format and thus we don't know how much to
888 * log here. Note that init_fn must also set the buffer log item type
891 init_fn(tp
, bp
, ip
, ifp
);
893 /* account for the change in fork size */
894 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
895 xfs_bmap_local_to_extents_empty(ip
, whichfork
);
896 flags
|= XFS_ILOG_CORE
;
898 ifp
->if_u1
.if_root
= NULL
;
902 rec
.br_startblock
= args
.fsbno
;
903 rec
.br_blockcount
= 1;
904 rec
.br_state
= XFS_EXT_NORM
;
905 xfs_iext_first(ifp
, &icur
);
906 xfs_iext_insert(ip
, &icur
, &rec
, 0);
908 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
909 ip
->i_d
.di_nblocks
= 1;
910 xfs_trans_mod_dquot_byino(tp
, ip
,
911 XFS_TRANS_DQ_BCOUNT
, 1L);
912 flags
|= xfs_ilog_fext(whichfork
);
920 * Called from xfs_bmap_add_attrfork to handle btree format files.
922 STATIC
int /* error */
923 xfs_bmap_add_attrfork_btree(
924 xfs_trans_t
*tp
, /* transaction pointer */
925 xfs_inode_t
*ip
, /* incore inode pointer */
926 int *flags
) /* inode logging flags */
928 xfs_btree_cur_t
*cur
; /* btree cursor */
929 int error
; /* error return value */
930 xfs_mount_t
*mp
; /* file system mount struct */
931 int stat
; /* newroot status */
934 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
935 *flags
|= XFS_ILOG_DBROOT
;
937 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
938 error
= xfs_bmbt_lookup_first(cur
, &stat
);
941 /* must be at least one entry */
942 XFS_WANT_CORRUPTED_GOTO(mp
, stat
== 1, error0
);
943 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
946 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
949 cur
->bc_private
.b
.allocated
= 0;
950 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
954 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
959 * Called from xfs_bmap_add_attrfork to handle extents format files.
961 STATIC
int /* error */
962 xfs_bmap_add_attrfork_extents(
963 struct xfs_trans
*tp
, /* transaction pointer */
964 struct xfs_inode
*ip
, /* incore inode pointer */
965 int *flags
) /* inode logging flags */
967 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
968 int error
; /* error return value */
970 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
973 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0, flags
,
976 cur
->bc_private
.b
.allocated
= 0;
977 xfs_btree_del_cursor(cur
, error
);
983 * Called from xfs_bmap_add_attrfork to handle local format files. Each
984 * different data fork content type needs a different callout to do the
985 * conversion. Some are basic and only require special block initialisation
986 * callouts for the data formating, others (directories) are so specialised they
987 * handle everything themselves.
989 * XXX (dgc): investigate whether directory conversion can use the generic
990 * formatting callout. It should be possible - it's just a very complex
993 STATIC
int /* error */
994 xfs_bmap_add_attrfork_local(
995 struct xfs_trans
*tp
, /* transaction pointer */
996 struct xfs_inode
*ip
, /* incore inode pointer */
997 int *flags
) /* inode logging flags */
999 struct xfs_da_args dargs
; /* args for dir/attr code */
1001 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
1004 if (S_ISDIR(VFS_I(ip
)->i_mode
)) {
1005 memset(&dargs
, 0, sizeof(dargs
));
1006 dargs
.geo
= ip
->i_mount
->m_dir_geo
;
1008 dargs
.total
= dargs
.geo
->fsbcount
;
1009 dargs
.whichfork
= XFS_DATA_FORK
;
1011 return xfs_dir2_sf_to_block(&dargs
);
1014 if (S_ISLNK(VFS_I(ip
)->i_mode
))
1015 return xfs_bmap_local_to_extents(tp
, ip
, 1, flags
,
1017 xfs_symlink_local_to_remote
);
1019 /* should only be called for types that support local format data */
1021 return -EFSCORRUPTED
;
1024 /* Set an inode attr fork off based on the format */
1026 xfs_bmap_set_attrforkoff(
1027 struct xfs_inode
*ip
,
1031 switch (ip
->i_d
.di_format
) {
1032 case XFS_DINODE_FMT_DEV
:
1033 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
1035 case XFS_DINODE_FMT_LOCAL
:
1036 case XFS_DINODE_FMT_EXTENTS
:
1037 case XFS_DINODE_FMT_BTREE
:
1038 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
1039 if (!ip
->i_d
.di_forkoff
)
1040 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
1041 else if ((ip
->i_mount
->m_flags
& XFS_MOUNT_ATTR2
) && version
)
1053 * Convert inode from non-attributed to attributed.
1054 * Must not be in a transaction, ip must not be locked.
1056 int /* error code */
1057 xfs_bmap_add_attrfork(
1058 xfs_inode_t
*ip
, /* incore inode pointer */
1059 int size
, /* space new attribute needs */
1060 int rsvd
) /* xact may use reserved blks */
1062 xfs_mount_t
*mp
; /* mount structure */
1063 xfs_trans_t
*tp
; /* transaction pointer */
1064 int blks
; /* space reservation */
1065 int version
= 1; /* superblock attr version */
1066 int logflags
; /* logging flags */
1067 int error
; /* error return value */
1069 ASSERT(XFS_IFORK_Q(ip
) == 0);
1072 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
1074 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
1076 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_addafork
, blks
, 0,
1077 rsvd
? XFS_TRANS_RESERVE
: 0, &tp
);
1081 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1082 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
1083 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
1084 XFS_QMOPT_RES_REGBLKS
);
1087 if (XFS_IFORK_Q(ip
))
1089 if (ip
->i_d
.di_anextents
!= 0) {
1090 error
= -EFSCORRUPTED
;
1093 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
1095 * For inodes coming from pre-6.2 filesystems.
1097 ASSERT(ip
->i_d
.di_aformat
== 0);
1098 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
1101 xfs_trans_ijoin(tp
, ip
, 0);
1102 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1103 error
= xfs_bmap_set_attrforkoff(ip
, size
, &version
);
1106 ASSERT(ip
->i_afp
== NULL
);
1107 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
1108 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
1110 switch (ip
->i_d
.di_format
) {
1111 case XFS_DINODE_FMT_LOCAL
:
1112 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &logflags
);
1114 case XFS_DINODE_FMT_EXTENTS
:
1115 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &logflags
);
1117 case XFS_DINODE_FMT_BTREE
:
1118 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &logflags
);
1125 xfs_trans_log_inode(tp
, ip
, logflags
);
1128 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
1129 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
1130 bool log_sb
= false;
1132 spin_lock(&mp
->m_sb_lock
);
1133 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
1134 xfs_sb_version_addattr(&mp
->m_sb
);
1137 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
1138 xfs_sb_version_addattr2(&mp
->m_sb
);
1141 spin_unlock(&mp
->m_sb_lock
);
1146 error
= xfs_trans_commit(tp
);
1147 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1151 xfs_trans_cancel(tp
);
1152 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1157 * Internal and external extent tree search functions.
1161 * Read in extents from a btree-format inode.
1165 struct xfs_trans
*tp
,
1166 struct xfs_inode
*ip
,
1169 struct xfs_mount
*mp
= ip
->i_mount
;
1170 int state
= xfs_bmap_fork_to_state(whichfork
);
1171 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1172 xfs_extnum_t nextents
= XFS_IFORK_NEXTENTS(ip
, whichfork
);
1173 struct xfs_btree_block
*block
= ifp
->if_broot
;
1174 struct xfs_iext_cursor icur
;
1175 struct xfs_bmbt_irec
new;
1183 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1185 if (unlikely(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
1186 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
1187 return -EFSCORRUPTED
;
1191 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1193 level
= be16_to_cpu(block
->bb_level
);
1194 if (unlikely(level
== 0)) {
1195 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
1196 return -EFSCORRUPTED
;
1198 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
1199 bno
= be64_to_cpu(*pp
);
1202 * Go down the tree until leaf level is reached, following the first
1203 * pointer (leftmost) at each level.
1205 while (level
-- > 0) {
1206 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1207 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1210 block
= XFS_BUF_TO_BLOCK(bp
);
1213 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
1214 bno
= be64_to_cpu(*pp
);
1215 XFS_WANT_CORRUPTED_GOTO(mp
,
1216 xfs_verify_fsbno(mp
, bno
), out_brelse
);
1217 xfs_trans_brelse(tp
, bp
);
1221 * Here with bp and block set to the leftmost leaf node in the tree.
1224 xfs_iext_first(ifp
, &icur
);
1227 * Loop over all leaf nodes. Copy information to the extent records.
1230 xfs_bmbt_rec_t
*frp
;
1231 xfs_fsblock_t nextbno
;
1232 xfs_extnum_t num_recs
;
1234 num_recs
= xfs_btree_get_numrecs(block
);
1235 if (unlikely(i
+ num_recs
> nextents
)) {
1236 xfs_warn(ip
->i_mount
,
1237 "corrupt dinode %Lu, (btree extents).",
1238 (unsigned long long) ip
->i_ino
);
1239 xfs_inode_verifier_error(ip
, -EFSCORRUPTED
,
1240 __func__
, block
, sizeof(*block
),
1242 error
= -EFSCORRUPTED
;
1246 * Read-ahead the next leaf block, if any.
1248 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
1249 if (nextbno
!= NULLFSBLOCK
)
1250 xfs_btree_reada_bufl(mp
, nextbno
, 1,
1253 * Copy records into the extent records.
1255 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
1256 for (j
= 0; j
< num_recs
; j
++, frp
++, i
++) {
1259 xfs_bmbt_disk_get_all(frp
, &new);
1260 fa
= xfs_bmap_validate_extent(ip
, whichfork
, &new);
1262 error
= -EFSCORRUPTED
;
1263 xfs_inode_verifier_error(ip
, error
,
1264 "xfs_iread_extents(2)",
1265 frp
, sizeof(*frp
), fa
);
1268 xfs_iext_insert(ip
, &icur
, &new, state
);
1269 trace_xfs_read_extent(ip
, &icur
, state
, _THIS_IP_
);
1270 xfs_iext_next(ifp
, &icur
);
1272 xfs_trans_brelse(tp
, bp
);
1275 * If we've reached the end, stop.
1277 if (bno
== NULLFSBLOCK
)
1279 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
1280 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
1283 block
= XFS_BUF_TO_BLOCK(bp
);
1286 if (i
!= XFS_IFORK_NEXTENTS(ip
, whichfork
)) {
1287 error
= -EFSCORRUPTED
;
1290 ASSERT(i
== xfs_iext_count(ifp
));
1292 ifp
->if_flags
|= XFS_IFEXTENTS
;
1296 xfs_trans_brelse(tp
, bp
);
1298 xfs_iext_destroy(ifp
);
1303 * Returns the relative block number of the first unused block(s) in the given
1304 * fork with at least "len" logically contiguous blocks free. This is the
1305 * lowest-address hole if the fork has holes, else the first block past the end
1306 * of fork. Return 0 if the fork is currently local (in-inode).
1309 xfs_bmap_first_unused(
1310 struct xfs_trans
*tp
, /* transaction pointer */
1311 struct xfs_inode
*ip
, /* incore inode */
1312 xfs_extlen_t len
, /* size of hole to find */
1313 xfs_fileoff_t
*first_unused
, /* unused block */
1314 int whichfork
) /* data or attr fork */
1316 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1317 struct xfs_bmbt_irec got
;
1318 struct xfs_iext_cursor icur
;
1319 xfs_fileoff_t lastaddr
= 0;
1320 xfs_fileoff_t lowest
, max
;
1323 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
1324 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
1325 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
1327 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
1332 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1333 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1338 lowest
= max
= *first_unused
;
1339 for_each_xfs_iext(ifp
, &icur
, &got
) {
1341 * See if the hole before this extent will work.
1343 if (got
.br_startoff
>= lowest
+ len
&&
1344 got
.br_startoff
- max
>= len
)
1346 lastaddr
= got
.br_startoff
+ got
.br_blockcount
;
1347 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
1350 *first_unused
= max
;
1355 * Returns the file-relative block number of the last block - 1 before
1356 * last_block (input value) in the file.
1357 * This is not based on i_size, it is based on the extent records.
1358 * Returns 0 for local files, as they do not have extent records.
1361 xfs_bmap_last_before(
1362 struct xfs_trans
*tp
, /* transaction pointer */
1363 struct xfs_inode
*ip
, /* incore inode */
1364 xfs_fileoff_t
*last_block
, /* last block */
1365 int whichfork
) /* data or attr fork */
1367 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1368 struct xfs_bmbt_irec got
;
1369 struct xfs_iext_cursor icur
;
1372 switch (XFS_IFORK_FORMAT(ip
, whichfork
)) {
1373 case XFS_DINODE_FMT_LOCAL
:
1376 case XFS_DINODE_FMT_BTREE
:
1377 case XFS_DINODE_FMT_EXTENTS
:
1383 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1384 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1389 if (!xfs_iext_lookup_extent_before(ip
, ifp
, last_block
, &icur
, &got
))
1395 xfs_bmap_last_extent(
1396 struct xfs_trans
*tp
,
1397 struct xfs_inode
*ip
,
1399 struct xfs_bmbt_irec
*rec
,
1402 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1403 struct xfs_iext_cursor icur
;
1406 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
1407 error
= xfs_iread_extents(tp
, ip
, whichfork
);
1412 xfs_iext_last(ifp
, &icur
);
1413 if (!xfs_iext_get_extent(ifp
, &icur
, rec
))
1421 * Check the last inode extent to determine whether this allocation will result
1422 * in blocks being allocated at the end of the file. When we allocate new data
1423 * blocks at the end of the file which do not start at the previous data block,
1424 * we will try to align the new blocks at stripe unit boundaries.
1426 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1427 * at, or past the EOF.
1431 struct xfs_bmalloca
*bma
,
1434 struct xfs_bmbt_irec rec
;
1439 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
1450 * Check if we are allocation or past the last extent, or at least into
1451 * the last delayed allocated extent.
1453 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
1454 (bma
->offset
>= rec
.br_startoff
&&
1455 isnullstartblock(rec
.br_startblock
));
1460 * Returns the file-relative block number of the first block past eof in
1461 * the file. This is not based on i_size, it is based on the extent records.
1462 * Returns 0 for local files, as they do not have extent records.
1465 xfs_bmap_last_offset(
1466 struct xfs_inode
*ip
,
1467 xfs_fileoff_t
*last_block
,
1470 struct xfs_bmbt_irec rec
;
1476 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
)
1479 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
1480 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1483 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
1484 if (error
|| is_empty
)
1487 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
1492 * Returns whether the selected fork of the inode has exactly one
1493 * block or not. For the data fork we check this matches di_size,
1494 * implying the file's range is 0..bsize-1.
1496 int /* 1=>1 block, 0=>otherwise */
1498 xfs_inode_t
*ip
, /* incore inode */
1499 int whichfork
) /* data or attr fork */
1501 struct xfs_ifork
*ifp
; /* inode fork pointer */
1502 int rval
; /* return value */
1503 xfs_bmbt_irec_t s
; /* internal version of extent */
1504 struct xfs_iext_cursor icur
;
1507 if (whichfork
== XFS_DATA_FORK
)
1508 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
1510 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
1512 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
1514 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1515 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
1516 xfs_iext_first(ifp
, &icur
);
1517 xfs_iext_get_extent(ifp
, &icur
, &s
);
1518 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
1519 if (rval
&& whichfork
== XFS_DATA_FORK
)
1520 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
1525 * Extent tree manipulation functions used during allocation.
1529 * Convert a delayed allocation to a real allocation.
1531 STATIC
int /* error */
1532 xfs_bmap_add_extent_delay_real(
1533 struct xfs_bmalloca
*bma
,
1536 struct xfs_bmbt_irec
*new = &bma
->got
;
1537 int error
; /* error return value */
1538 int i
; /* temp state */
1539 struct xfs_ifork
*ifp
; /* inode fork pointer */
1540 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1541 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1542 /* left is 0, right is 1, prev is 2 */
1543 int rval
=0; /* return value (logging flags) */
1544 int state
= xfs_bmap_fork_to_state(whichfork
);
1545 xfs_filblks_t da_new
; /* new count del alloc blocks used */
1546 xfs_filblks_t da_old
; /* old count del alloc blocks used */
1547 xfs_filblks_t temp
=0; /* value for da_new calculations */
1548 int tmp_rval
; /* partial logging flags */
1549 struct xfs_mount
*mp
;
1550 xfs_extnum_t
*nextents
;
1551 struct xfs_bmbt_irec old
;
1553 mp
= bma
->ip
->i_mount
;
1554 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1555 ASSERT(whichfork
!= XFS_ATTR_FORK
);
1556 nextents
= (whichfork
== XFS_COW_FORK
? &bma
->ip
->i_cnextents
:
1557 &bma
->ip
->i_d
.di_nextents
);
1559 ASSERT(!isnullstartblock(new->br_startblock
));
1561 (bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
1563 XFS_STATS_INC(mp
, xs_add_exlist
);
1570 * Set up a bunch of variables to make the tests simpler.
1572 xfs_iext_get_extent(ifp
, &bma
->icur
, &PREV
);
1573 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1574 ASSERT(isnullstartblock(PREV
.br_startblock
));
1575 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1576 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1578 da_old
= startblockval(PREV
.br_startblock
);
1582 * Set flags determining what part of the previous delayed allocation
1583 * extent is being replaced by a real allocation.
1585 if (PREV
.br_startoff
== new->br_startoff
)
1586 state
|= BMAP_LEFT_FILLING
;
1587 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1588 state
|= BMAP_RIGHT_FILLING
;
1591 * Check and set flags if this segment has a left neighbor.
1592 * Don't set contiguous if the combined extent would be too large.
1594 if (xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &LEFT
)) {
1595 state
|= BMAP_LEFT_VALID
;
1596 if (isnullstartblock(LEFT
.br_startblock
))
1597 state
|= BMAP_LEFT_DELAY
;
1600 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1601 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1602 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1603 LEFT
.br_state
== new->br_state
&&
1604 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1605 state
|= BMAP_LEFT_CONTIG
;
1608 * Check and set flags if this segment has a right neighbor.
1609 * Don't set contiguous if the combined extent would be too large.
1610 * Also check for all-three-contiguous being too large.
1612 if (xfs_iext_peek_next_extent(ifp
, &bma
->icur
, &RIGHT
)) {
1613 state
|= BMAP_RIGHT_VALID
;
1614 if (isnullstartblock(RIGHT
.br_startblock
))
1615 state
|= BMAP_RIGHT_DELAY
;
1618 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1619 new_endoff
== RIGHT
.br_startoff
&&
1620 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1621 new->br_state
== RIGHT
.br_state
&&
1622 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1623 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1624 BMAP_RIGHT_FILLING
)) !=
1625 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1626 BMAP_RIGHT_FILLING
) ||
1627 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1629 state
|= BMAP_RIGHT_CONTIG
;
1633 * Switch out based on the FILLING and CONTIG state bits.
1635 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1636 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1637 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1638 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1640 * Filling in all of a previously delayed allocation extent.
1641 * The left and right neighbors are both contiguous with new.
1643 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
1645 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1646 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1647 xfs_iext_prev(ifp
, &bma
->icur
);
1648 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1651 if (bma
->cur
== NULL
)
1652 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1654 rval
= XFS_ILOG_CORE
;
1655 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1658 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1659 error
= xfs_btree_delete(bma
->cur
, &i
);
1662 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1663 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1666 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1667 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1673 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1675 * Filling in all of a previously delayed allocation extent.
1676 * The left neighbor is contiguous, the right is not.
1679 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
1681 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1682 xfs_iext_prev(ifp
, &bma
->icur
);
1683 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1685 if (bma
->cur
== NULL
)
1686 rval
= XFS_ILOG_DEXT
;
1689 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1692 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1693 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1699 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1701 * Filling in all of a previously delayed allocation extent.
1702 * The right neighbor is contiguous, the left is not. Take care
1703 * with delay -> unwritten extent allocation here because the
1704 * delalloc record we are overwriting is always written.
1706 PREV
.br_startblock
= new->br_startblock
;
1707 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
1708 PREV
.br_state
= new->br_state
;
1710 xfs_iext_next(ifp
, &bma
->icur
);
1711 xfs_iext_remove(bma
->ip
, &bma
->icur
, state
);
1712 xfs_iext_prev(ifp
, &bma
->icur
);
1713 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1715 if (bma
->cur
== NULL
)
1716 rval
= XFS_ILOG_DEXT
;
1719 error
= xfs_bmbt_lookup_eq(bma
->cur
, &RIGHT
, &i
);
1722 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1723 error
= xfs_bmbt_update(bma
->cur
, &PREV
);
1729 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1731 * Filling in all of a previously delayed allocation extent.
1732 * Neither the left nor right neighbors are contiguous with
1735 PREV
.br_startblock
= new->br_startblock
;
1736 PREV
.br_state
= new->br_state
;
1737 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1740 if (bma
->cur
== NULL
)
1741 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1743 rval
= XFS_ILOG_CORE
;
1744 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1747 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1748 error
= xfs_btree_insert(bma
->cur
, &i
);
1751 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1755 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1757 * Filling in the first part of a previous delayed allocation.
1758 * The left neighbor is contiguous.
1761 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1762 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1763 startblockval(PREV
.br_startblock
));
1765 LEFT
.br_blockcount
+= new->br_blockcount
;
1767 PREV
.br_blockcount
= temp
;
1768 PREV
.br_startoff
+= new->br_blockcount
;
1769 PREV
.br_startblock
= nullstartblock(da_new
);
1771 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1772 xfs_iext_prev(ifp
, &bma
->icur
);
1773 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &LEFT
);
1775 if (bma
->cur
== NULL
)
1776 rval
= XFS_ILOG_DEXT
;
1779 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1782 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1783 error
= xfs_bmbt_update(bma
->cur
, &LEFT
);
1789 case BMAP_LEFT_FILLING
:
1791 * Filling in the first part of a previous delayed allocation.
1792 * The left neighbor is not contiguous.
1794 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1796 if (bma
->cur
== NULL
)
1797 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1799 rval
= XFS_ILOG_CORE
;
1800 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1803 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1804 error
= xfs_btree_insert(bma
->cur
, &i
);
1807 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1810 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1811 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1812 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1818 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1819 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1820 startblockval(PREV
.br_startblock
) -
1821 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
1823 PREV
.br_startoff
= new_endoff
;
1824 PREV
.br_blockcount
= temp
;
1825 PREV
.br_startblock
= nullstartblock(da_new
);
1826 xfs_iext_next(ifp
, &bma
->icur
);
1827 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1828 xfs_iext_prev(ifp
, &bma
->icur
);
1831 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1833 * Filling in the last part of a previous delayed allocation.
1834 * The right neighbor is contiguous with the new allocation.
1837 RIGHT
.br_startoff
= new->br_startoff
;
1838 RIGHT
.br_startblock
= new->br_startblock
;
1839 RIGHT
.br_blockcount
+= new->br_blockcount
;
1841 if (bma
->cur
== NULL
)
1842 rval
= XFS_ILOG_DEXT
;
1845 error
= xfs_bmbt_lookup_eq(bma
->cur
, &old
, &i
);
1848 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1849 error
= xfs_bmbt_update(bma
->cur
, &RIGHT
);
1854 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1855 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1856 startblockval(PREV
.br_startblock
));
1858 PREV
.br_blockcount
= temp
;
1859 PREV
.br_startblock
= nullstartblock(da_new
);
1861 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1862 xfs_iext_next(ifp
, &bma
->icur
);
1863 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &RIGHT
);
1866 case BMAP_RIGHT_FILLING
:
1868 * Filling in the last part of a previous delayed allocation.
1869 * The right neighbor is not contiguous.
1871 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, new);
1873 if (bma
->cur
== NULL
)
1874 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1876 rval
= XFS_ILOG_CORE
;
1877 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1880 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1881 error
= xfs_btree_insert(bma
->cur
, &i
);
1884 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1887 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1888 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1889 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1895 temp
= PREV
.br_blockcount
- new->br_blockcount
;
1896 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
1897 startblockval(PREV
.br_startblock
) -
1898 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
1900 PREV
.br_startblock
= nullstartblock(da_new
);
1901 PREV
.br_blockcount
= temp
;
1902 xfs_iext_insert(bma
->ip
, &bma
->icur
, &PREV
, state
);
1903 xfs_iext_next(ifp
, &bma
->icur
);
1908 * Filling in the middle part of a previous delayed allocation.
1909 * Contiguity is impossible here.
1910 * This case is avoided almost all the time.
1912 * We start with a delayed allocation:
1914 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1917 * and we are allocating:
1918 * +rrrrrrrrrrrrrrrrr+
1921 * and we set it up for insertion as:
1922 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1924 * PREV @ idx LEFT RIGHT
1925 * inserted at idx + 1
1929 /* LEFT is the new middle */
1932 /* RIGHT is the new right */
1933 RIGHT
.br_state
= PREV
.br_state
;
1934 RIGHT
.br_startoff
= new_endoff
;
1935 RIGHT
.br_blockcount
=
1936 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
1937 RIGHT
.br_startblock
=
1938 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1939 RIGHT
.br_blockcount
));
1942 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
1943 PREV
.br_startblock
=
1944 nullstartblock(xfs_bmap_worst_indlen(bma
->ip
,
1945 PREV
.br_blockcount
));
1946 xfs_iext_update_extent(bma
->ip
, state
, &bma
->icur
, &PREV
);
1948 xfs_iext_next(ifp
, &bma
->icur
);
1949 xfs_iext_insert(bma
->ip
, &bma
->icur
, &RIGHT
, state
);
1950 xfs_iext_insert(bma
->ip
, &bma
->icur
, &LEFT
, state
);
1953 if (bma
->cur
== NULL
)
1954 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1956 rval
= XFS_ILOG_CORE
;
1957 error
= xfs_bmbt_lookup_eq(bma
->cur
, new, &i
);
1960 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
1961 error
= xfs_btree_insert(bma
->cur
, &i
);
1964 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
1967 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1968 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1969 &bma
->cur
, 1, &tmp_rval
, whichfork
);
1975 da_new
= startblockval(PREV
.br_startblock
) +
1976 startblockval(RIGHT
.br_startblock
);
1979 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1980 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1981 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1982 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1983 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1984 case BMAP_LEFT_CONTIG
:
1985 case BMAP_RIGHT_CONTIG
:
1987 * These cases are all impossible.
1992 /* add reverse mapping unless caller opted out */
1993 if (!(bma
->flags
& XFS_BMAPI_NORMAP
)) {
1994 error
= xfs_rmap_map_extent(bma
->tp
, bma
->ip
, whichfork
, new);
1999 /* convert to a btree if necessary */
2000 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
2001 int tmp_logflags
; /* partial log flag return val */
2003 ASSERT(bma
->cur
== NULL
);
2004 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
2005 &bma
->cur
, da_old
> 0, &tmp_logflags
,
2007 bma
->logflags
|= tmp_logflags
;
2013 da_new
+= bma
->cur
->bc_private
.b
.allocated
;
2014 bma
->cur
->bc_private
.b
.allocated
= 0;
2017 /* adjust for changes in reserved delayed indirect blocks */
2018 if (da_new
!= da_old
) {
2019 ASSERT(state
== 0 || da_new
< da_old
);
2020 error
= xfs_mod_fdblocks(mp
, (int64_t)(da_old
- da_new
),
2024 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
2026 if (whichfork
!= XFS_COW_FORK
)
2027 bma
->logflags
|= rval
;
2035 * Convert an unwritten allocation to a real allocation or vice versa.
2038 xfs_bmap_add_extent_unwritten_real(
2039 struct xfs_trans
*tp
,
2040 xfs_inode_t
*ip
, /* incore inode pointer */
2042 struct xfs_iext_cursor
*icur
,
2043 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
2044 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
2045 int *logflagsp
) /* inode logging flags */
2047 xfs_btree_cur_t
*cur
; /* btree cursor */
2048 int error
; /* error return value */
2049 int i
; /* temp state */
2050 struct xfs_ifork
*ifp
; /* inode fork pointer */
2051 xfs_fileoff_t new_endoff
; /* end offset of new entry */
2052 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
2053 /* left is 0, right is 1, prev is 2 */
2054 int rval
=0; /* return value (logging flags) */
2055 int state
= xfs_bmap_fork_to_state(whichfork
);
2056 struct xfs_mount
*mp
= ip
->i_mount
;
2057 struct xfs_bmbt_irec old
;
2062 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2064 ASSERT(!isnullstartblock(new->br_startblock
));
2066 XFS_STATS_INC(mp
, xs_add_exlist
);
2073 * Set up a bunch of variables to make the tests simpler.
2076 xfs_iext_get_extent(ifp
, icur
, &PREV
);
2077 ASSERT(new->br_state
!= PREV
.br_state
);
2078 new_endoff
= new->br_startoff
+ new->br_blockcount
;
2079 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
2080 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
2083 * Set flags determining what part of the previous oldext allocation
2084 * extent is being replaced by a newext allocation.
2086 if (PREV
.br_startoff
== new->br_startoff
)
2087 state
|= BMAP_LEFT_FILLING
;
2088 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
2089 state
|= BMAP_RIGHT_FILLING
;
2092 * Check and set flags if this segment has a left neighbor.
2093 * Don't set contiguous if the combined extent would be too large.
2095 if (xfs_iext_peek_prev_extent(ifp
, icur
, &LEFT
)) {
2096 state
|= BMAP_LEFT_VALID
;
2097 if (isnullstartblock(LEFT
.br_startblock
))
2098 state
|= BMAP_LEFT_DELAY
;
2101 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2102 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
2103 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
2104 LEFT
.br_state
== new->br_state
&&
2105 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2106 state
|= BMAP_LEFT_CONTIG
;
2109 * Check and set flags if this segment has a right neighbor.
2110 * Don't set contiguous if the combined extent would be too large.
2111 * Also check for all-three-contiguous being too large.
2113 if (xfs_iext_peek_next_extent(ifp
, icur
, &RIGHT
)) {
2114 state
|= BMAP_RIGHT_VALID
;
2115 if (isnullstartblock(RIGHT
.br_startblock
))
2116 state
|= BMAP_RIGHT_DELAY
;
2119 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2120 new_endoff
== RIGHT
.br_startoff
&&
2121 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
2122 new->br_state
== RIGHT
.br_state
&&
2123 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
2124 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2125 BMAP_RIGHT_FILLING
)) !=
2126 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
2127 BMAP_RIGHT_FILLING
) ||
2128 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
2130 state
|= BMAP_RIGHT_CONTIG
;
2133 * Switch out based on the FILLING and CONTIG state bits.
2135 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2136 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
2137 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
2138 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2140 * Setting all of a previous oldext extent to newext.
2141 * The left and right neighbors are both contiguous with new.
2143 LEFT
.br_blockcount
+= PREV
.br_blockcount
+ RIGHT
.br_blockcount
;
2145 xfs_iext_remove(ip
, icur
, state
);
2146 xfs_iext_remove(ip
, icur
, state
);
2147 xfs_iext_prev(ifp
, icur
);
2148 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2149 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2150 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 2);
2152 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2154 rval
= XFS_ILOG_CORE
;
2155 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2158 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2159 if ((error
= xfs_btree_delete(cur
, &i
)))
2161 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2162 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2164 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2165 if ((error
= xfs_btree_delete(cur
, &i
)))
2167 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2168 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2170 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2171 error
= xfs_bmbt_update(cur
, &LEFT
);
2177 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2179 * Setting all of a previous oldext extent to newext.
2180 * The left neighbor is contiguous, the right is not.
2182 LEFT
.br_blockcount
+= PREV
.br_blockcount
;
2184 xfs_iext_remove(ip
, icur
, state
);
2185 xfs_iext_prev(ifp
, icur
);
2186 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2187 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2188 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2190 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2192 rval
= XFS_ILOG_CORE
;
2193 error
= xfs_bmbt_lookup_eq(cur
, &PREV
, &i
);
2196 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2197 if ((error
= xfs_btree_delete(cur
, &i
)))
2199 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2200 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2202 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2203 error
= xfs_bmbt_update(cur
, &LEFT
);
2209 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2211 * Setting all of a previous oldext extent to newext.
2212 * The right neighbor is contiguous, the left is not.
2214 PREV
.br_blockcount
+= RIGHT
.br_blockcount
;
2215 PREV
.br_state
= new->br_state
;
2217 xfs_iext_next(ifp
, icur
);
2218 xfs_iext_remove(ip
, icur
, state
);
2219 xfs_iext_prev(ifp
, icur
);
2220 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2222 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2223 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2225 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2227 rval
= XFS_ILOG_CORE
;
2228 error
= xfs_bmbt_lookup_eq(cur
, &RIGHT
, &i
);
2231 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2232 if ((error
= xfs_btree_delete(cur
, &i
)))
2234 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2235 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
2237 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2238 error
= xfs_bmbt_update(cur
, &PREV
);
2244 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
2246 * Setting all of a previous oldext extent to newext.
2247 * Neither the left nor right neighbors are contiguous with
2250 PREV
.br_state
= new->br_state
;
2251 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2254 rval
= XFS_ILOG_DEXT
;
2257 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2260 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2261 error
= xfs_bmbt_update(cur
, &PREV
);
2267 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
2269 * Setting the first part of a previous oldext extent to newext.
2270 * The left neighbor is contiguous.
2272 LEFT
.br_blockcount
+= new->br_blockcount
;
2275 PREV
.br_startoff
+= new->br_blockcount
;
2276 PREV
.br_startblock
+= new->br_blockcount
;
2277 PREV
.br_blockcount
-= new->br_blockcount
;
2279 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2280 xfs_iext_prev(ifp
, icur
);
2281 xfs_iext_update_extent(ip
, state
, icur
, &LEFT
);
2284 rval
= XFS_ILOG_DEXT
;
2287 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2290 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2291 error
= xfs_bmbt_update(cur
, &PREV
);
2294 error
= xfs_btree_decrement(cur
, 0, &i
);
2297 error
= xfs_bmbt_update(cur
, &LEFT
);
2303 case BMAP_LEFT_FILLING
:
2305 * Setting the first part of a previous oldext extent to newext.
2306 * The left neighbor is not contiguous.
2309 PREV
.br_startoff
+= new->br_blockcount
;
2310 PREV
.br_startblock
+= new->br_blockcount
;
2311 PREV
.br_blockcount
-= new->br_blockcount
;
2313 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2314 xfs_iext_insert(ip
, icur
, new, state
);
2315 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2316 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2318 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2320 rval
= XFS_ILOG_CORE
;
2321 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2324 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2325 error
= xfs_bmbt_update(cur
, &PREV
);
2328 cur
->bc_rec
.b
= *new;
2329 if ((error
= xfs_btree_insert(cur
, &i
)))
2331 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2335 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
2337 * Setting the last part of a previous oldext extent to newext.
2338 * The right neighbor is contiguous with the new allocation.
2341 PREV
.br_blockcount
-= new->br_blockcount
;
2343 RIGHT
.br_startoff
= new->br_startoff
;
2344 RIGHT
.br_startblock
= new->br_startblock
;
2345 RIGHT
.br_blockcount
+= new->br_blockcount
;
2347 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2348 xfs_iext_next(ifp
, icur
);
2349 xfs_iext_update_extent(ip
, state
, icur
, &RIGHT
);
2352 rval
= XFS_ILOG_DEXT
;
2355 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2358 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2359 error
= xfs_bmbt_update(cur
, &PREV
);
2362 error
= xfs_btree_increment(cur
, 0, &i
);
2365 error
= xfs_bmbt_update(cur
, &RIGHT
);
2371 case BMAP_RIGHT_FILLING
:
2373 * Setting the last part of a previous oldext extent to newext.
2374 * The right neighbor is not contiguous.
2377 PREV
.br_blockcount
-= new->br_blockcount
;
2379 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2380 xfs_iext_next(ifp
, icur
);
2381 xfs_iext_insert(ip
, icur
, new, state
);
2383 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2384 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2386 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2388 rval
= XFS_ILOG_CORE
;
2389 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2392 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2393 error
= xfs_bmbt_update(cur
, &PREV
);
2396 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2399 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2400 if ((error
= xfs_btree_insert(cur
, &i
)))
2402 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2408 * Setting the middle part of a previous oldext extent to
2409 * newext. Contiguity is impossible here.
2410 * One extent becomes three extents.
2413 PREV
.br_blockcount
= new->br_startoff
- PREV
.br_startoff
;
2416 r
[1].br_startoff
= new_endoff
;
2417 r
[1].br_blockcount
=
2418 old
.br_startoff
+ old
.br_blockcount
- new_endoff
;
2419 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
2420 r
[1].br_state
= PREV
.br_state
;
2422 xfs_iext_update_extent(ip
, state
, icur
, &PREV
);
2423 xfs_iext_next(ifp
, icur
);
2424 xfs_iext_insert(ip
, icur
, &r
[1], state
);
2425 xfs_iext_insert(ip
, icur
, &r
[0], state
);
2427 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2428 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 2);
2430 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
2432 rval
= XFS_ILOG_CORE
;
2433 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2436 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2437 /* new right extent - oldext */
2438 error
= xfs_bmbt_update(cur
, &r
[1]);
2441 /* new left extent - oldext */
2442 cur
->bc_rec
.b
= PREV
;
2443 if ((error
= xfs_btree_insert(cur
, &i
)))
2445 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2447 * Reset the cursor to the position of the new extent
2448 * we are about to insert as we can't trust it after
2449 * the previous insert.
2451 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2454 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2455 /* new middle extent - newext */
2456 if ((error
= xfs_btree_insert(cur
, &i
)))
2458 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2462 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2463 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2464 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
2465 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
2466 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2467 case BMAP_LEFT_CONTIG
:
2468 case BMAP_RIGHT_CONTIG
:
2470 * These cases are all impossible.
2475 /* update reverse mappings */
2476 error
= xfs_rmap_convert_extent(mp
, tp
, ip
, whichfork
, new);
2480 /* convert to a btree if necessary */
2481 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2482 int tmp_logflags
; /* partial log flag return val */
2484 ASSERT(cur
== NULL
);
2485 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
2486 &tmp_logflags
, whichfork
);
2487 *logflagsp
|= tmp_logflags
;
2492 /* clear out the allocated field, done with it now in any case. */
2494 cur
->bc_private
.b
.allocated
= 0;
2498 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
2508 * Convert a hole to a delayed allocation.
2511 xfs_bmap_add_extent_hole_delay(
2512 xfs_inode_t
*ip
, /* incore inode pointer */
2514 struct xfs_iext_cursor
*icur
,
2515 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
2517 struct xfs_ifork
*ifp
; /* inode fork pointer */
2518 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2519 xfs_filblks_t newlen
=0; /* new indirect size */
2520 xfs_filblks_t oldlen
=0; /* old indirect size */
2521 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2522 int state
= xfs_bmap_fork_to_state(whichfork
);
2523 xfs_filblks_t temp
; /* temp for indirect calculations */
2525 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2526 ASSERT(isnullstartblock(new->br_startblock
));
2529 * Check and set flags if this segment has a left neighbor
2531 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2532 state
|= BMAP_LEFT_VALID
;
2533 if (isnullstartblock(left
.br_startblock
))
2534 state
|= BMAP_LEFT_DELAY
;
2538 * Check and set flags if the current (right) segment exists.
2539 * If it doesn't exist, we're converting the hole at end-of-file.
2541 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2542 state
|= BMAP_RIGHT_VALID
;
2543 if (isnullstartblock(right
.br_startblock
))
2544 state
|= BMAP_RIGHT_DELAY
;
2548 * Set contiguity flags on the left and right neighbors.
2549 * Don't let extents get too large, even if the pieces are contiguous.
2551 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
2552 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2553 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2554 state
|= BMAP_LEFT_CONTIG
;
2556 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
2557 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2558 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2559 (!(state
& BMAP_LEFT_CONTIG
) ||
2560 (left
.br_blockcount
+ new->br_blockcount
+
2561 right
.br_blockcount
<= MAXEXTLEN
)))
2562 state
|= BMAP_RIGHT_CONTIG
;
2565 * Switch out based on the contiguity flags.
2567 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2568 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2570 * New allocation is contiguous with delayed allocations
2571 * on the left and on the right.
2572 * Merge all three into a single extent record.
2574 temp
= left
.br_blockcount
+ new->br_blockcount
+
2575 right
.br_blockcount
;
2577 oldlen
= startblockval(left
.br_startblock
) +
2578 startblockval(new->br_startblock
) +
2579 startblockval(right
.br_startblock
);
2580 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2582 left
.br_startblock
= nullstartblock(newlen
);
2583 left
.br_blockcount
= temp
;
2585 xfs_iext_remove(ip
, icur
, state
);
2586 xfs_iext_prev(ifp
, icur
);
2587 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2590 case BMAP_LEFT_CONTIG
:
2592 * New allocation is contiguous with a delayed allocation
2594 * Merge the new allocation with the left neighbor.
2596 temp
= left
.br_blockcount
+ new->br_blockcount
;
2598 oldlen
= startblockval(left
.br_startblock
) +
2599 startblockval(new->br_startblock
);
2600 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2602 left
.br_blockcount
= temp
;
2603 left
.br_startblock
= nullstartblock(newlen
);
2605 xfs_iext_prev(ifp
, icur
);
2606 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2609 case BMAP_RIGHT_CONTIG
:
2611 * New allocation is contiguous with a delayed allocation
2613 * Merge the new allocation with the right neighbor.
2615 temp
= new->br_blockcount
+ right
.br_blockcount
;
2616 oldlen
= startblockval(new->br_startblock
) +
2617 startblockval(right
.br_startblock
);
2618 newlen
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2620 right
.br_startoff
= new->br_startoff
;
2621 right
.br_startblock
= nullstartblock(newlen
);
2622 right
.br_blockcount
= temp
;
2623 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2628 * New allocation is not contiguous with another
2629 * delayed allocation.
2630 * Insert a new entry.
2632 oldlen
= newlen
= 0;
2633 xfs_iext_insert(ip
, icur
, new, state
);
2636 if (oldlen
!= newlen
) {
2637 ASSERT(oldlen
> newlen
);
2638 xfs_mod_fdblocks(ip
->i_mount
, (int64_t)(oldlen
- newlen
),
2641 * Nothing to do for disk quota accounting here.
2647 * Convert a hole to a real allocation.
2649 STATIC
int /* error */
2650 xfs_bmap_add_extent_hole_real(
2651 struct xfs_trans
*tp
,
2652 struct xfs_inode
*ip
,
2654 struct xfs_iext_cursor
*icur
,
2655 struct xfs_btree_cur
**curp
,
2656 struct xfs_bmbt_irec
*new,
2660 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2661 struct xfs_mount
*mp
= ip
->i_mount
;
2662 struct xfs_btree_cur
*cur
= *curp
;
2663 int error
; /* error return value */
2664 int i
; /* temp state */
2665 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
2666 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
2667 int rval
=0; /* return value (logging flags) */
2668 int state
= xfs_bmap_fork_to_state(whichfork
);
2669 struct xfs_bmbt_irec old
;
2671 ASSERT(!isnullstartblock(new->br_startblock
));
2672 ASSERT(!cur
|| !(cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
2674 XFS_STATS_INC(mp
, xs_add_exlist
);
2677 * Check and set flags if this segment has a left neighbor.
2679 if (xfs_iext_peek_prev_extent(ifp
, icur
, &left
)) {
2680 state
|= BMAP_LEFT_VALID
;
2681 if (isnullstartblock(left
.br_startblock
))
2682 state
|= BMAP_LEFT_DELAY
;
2686 * Check and set flags if this segment has a current value.
2687 * Not true if we're inserting into the "hole" at eof.
2689 if (xfs_iext_get_extent(ifp
, icur
, &right
)) {
2690 state
|= BMAP_RIGHT_VALID
;
2691 if (isnullstartblock(right
.br_startblock
))
2692 state
|= BMAP_RIGHT_DELAY
;
2696 * We're inserting a real allocation between "left" and "right".
2697 * Set the contiguity flags. Don't let extents get too large.
2699 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
2700 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
2701 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
2702 left
.br_state
== new->br_state
&&
2703 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
2704 state
|= BMAP_LEFT_CONTIG
;
2706 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
2707 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
2708 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
2709 new->br_state
== right
.br_state
&&
2710 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
2711 (!(state
& BMAP_LEFT_CONTIG
) ||
2712 left
.br_blockcount
+ new->br_blockcount
+
2713 right
.br_blockcount
<= MAXEXTLEN
))
2714 state
|= BMAP_RIGHT_CONTIG
;
2718 * Select which case we're in here, and implement it.
2720 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
2721 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
2723 * New allocation is contiguous with real allocations on the
2724 * left and on the right.
2725 * Merge all three into a single extent record.
2727 left
.br_blockcount
+= new->br_blockcount
+ right
.br_blockcount
;
2729 xfs_iext_remove(ip
, icur
, state
);
2730 xfs_iext_prev(ifp
, icur
);
2731 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2733 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2734 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2736 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2738 rval
= XFS_ILOG_CORE
;
2739 error
= xfs_bmbt_lookup_eq(cur
, &right
, &i
);
2742 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2743 error
= xfs_btree_delete(cur
, &i
);
2746 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2747 error
= xfs_btree_decrement(cur
, 0, &i
);
2750 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2751 error
= xfs_bmbt_update(cur
, &left
);
2757 case BMAP_LEFT_CONTIG
:
2759 * New allocation is contiguous with a real allocation
2761 * Merge the new allocation with the left neighbor.
2764 left
.br_blockcount
+= new->br_blockcount
;
2766 xfs_iext_prev(ifp
, icur
);
2767 xfs_iext_update_extent(ip
, state
, icur
, &left
);
2770 rval
= xfs_ilog_fext(whichfork
);
2773 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2776 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2777 error
= xfs_bmbt_update(cur
, &left
);
2783 case BMAP_RIGHT_CONTIG
:
2785 * New allocation is contiguous with a real allocation
2787 * Merge the new allocation with the right neighbor.
2791 right
.br_startoff
= new->br_startoff
;
2792 right
.br_startblock
= new->br_startblock
;
2793 right
.br_blockcount
+= new->br_blockcount
;
2794 xfs_iext_update_extent(ip
, state
, icur
, &right
);
2797 rval
= xfs_ilog_fext(whichfork
);
2800 error
= xfs_bmbt_lookup_eq(cur
, &old
, &i
);
2803 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2804 error
= xfs_bmbt_update(cur
, &right
);
2812 * New allocation is not contiguous with another
2814 * Insert a new entry.
2816 xfs_iext_insert(ip
, icur
, new, state
);
2817 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2818 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2820 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2822 rval
= XFS_ILOG_CORE
;
2823 error
= xfs_bmbt_lookup_eq(cur
, new, &i
);
2826 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, done
);
2827 error
= xfs_btree_insert(cur
, &i
);
2830 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
2835 /* add reverse mapping unless caller opted out */
2836 if (!(flags
& XFS_BMAPI_NORMAP
)) {
2837 error
= xfs_rmap_map_extent(tp
, ip
, whichfork
, new);
2842 /* convert to a btree if necessary */
2843 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
2844 int tmp_logflags
; /* partial log flag return val */
2846 ASSERT(cur
== NULL
);
2847 error
= xfs_bmap_extents_to_btree(tp
, ip
, curp
, 0,
2848 &tmp_logflags
, whichfork
);
2849 *logflagsp
|= tmp_logflags
;
2855 /* clear out the allocated field, done with it now in any case. */
2857 cur
->bc_private
.b
.allocated
= 0;
2859 xfs_bmap_check_leaf_extents(cur
, ip
, whichfork
);
2866 * Functions used in the extent read, allocate and remove paths
2870 * Adjust the size of the new extent based on di_extsize and rt extsize.
2873 xfs_bmap_extsize_align(
2875 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
2876 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
2877 xfs_extlen_t extsz
, /* align to this extent size */
2878 int rt
, /* is this a realtime inode? */
2879 int eof
, /* is extent at end-of-file? */
2880 int delay
, /* creating delalloc extent? */
2881 int convert
, /* overwriting unwritten extent? */
2882 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
2883 xfs_extlen_t
*lenp
) /* in/out: aligned length */
2885 xfs_fileoff_t orig_off
; /* original offset */
2886 xfs_extlen_t orig_alen
; /* original length */
2887 xfs_fileoff_t orig_end
; /* original off+len */
2888 xfs_fileoff_t nexto
; /* next file offset */
2889 xfs_fileoff_t prevo
; /* previous file offset */
2890 xfs_fileoff_t align_off
; /* temp for offset */
2891 xfs_extlen_t align_alen
; /* temp for length */
2892 xfs_extlen_t temp
; /* temp for calculations */
2897 orig_off
= align_off
= *offp
;
2898 orig_alen
= align_alen
= *lenp
;
2899 orig_end
= orig_off
+ orig_alen
;
2902 * If this request overlaps an existing extent, then don't
2903 * attempt to perform any additional alignment.
2905 if (!delay
&& !eof
&&
2906 (orig_off
>= gotp
->br_startoff
) &&
2907 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
2912 * If the file offset is unaligned vs. the extent size
2913 * we need to align it. This will be possible unless
2914 * the file was previously written with a kernel that didn't
2915 * perform this alignment, or if a truncate shot us in the
2918 div_u64_rem(orig_off
, extsz
, &temp
);
2924 /* Same adjustment for the end of the requested area. */
2925 temp
= (align_alen
% extsz
);
2927 align_alen
+= extsz
- temp
;
2930 * For large extent hint sizes, the aligned extent might be larger than
2931 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
2932 * the length back under MAXEXTLEN. The outer allocation loops handle
2933 * short allocation just fine, so it is safe to do this. We only want to
2934 * do it when we are forced to, though, because it means more allocation
2935 * operations are required.
2937 while (align_alen
> MAXEXTLEN
)
2938 align_alen
-= extsz
;
2939 ASSERT(align_alen
<= MAXEXTLEN
);
2942 * If the previous block overlaps with this proposed allocation
2943 * then move the start forward without adjusting the length.
2945 if (prevp
->br_startoff
!= NULLFILEOFF
) {
2946 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
2947 prevo
= prevp
->br_startoff
;
2949 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
2952 if (align_off
!= orig_off
&& align_off
< prevo
)
2955 * If the next block overlaps with this proposed allocation
2956 * then move the start back without adjusting the length,
2957 * but not before offset 0.
2958 * This may of course make the start overlap previous block,
2959 * and if we hit the offset 0 limit then the next block
2960 * can still overlap too.
2962 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
2963 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
2964 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
2965 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
2967 nexto
= gotp
->br_startoff
;
2969 nexto
= NULLFILEOFF
;
2971 align_off
+ align_alen
!= orig_end
&&
2972 align_off
+ align_alen
> nexto
)
2973 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
2975 * If we're now overlapping the next or previous extent that
2976 * means we can't fit an extsz piece in this hole. Just move
2977 * the start forward to the first valid spot and set
2978 * the length so we hit the end.
2980 if (align_off
!= orig_off
&& align_off
< prevo
)
2982 if (align_off
+ align_alen
!= orig_end
&&
2983 align_off
+ align_alen
> nexto
&&
2984 nexto
!= NULLFILEOFF
) {
2985 ASSERT(nexto
> prevo
);
2986 align_alen
= nexto
- align_off
;
2990 * If realtime, and the result isn't a multiple of the realtime
2991 * extent size we need to remove blocks until it is.
2993 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
2995 * We're not covering the original request, or
2996 * we won't be able to once we fix the length.
2998 if (orig_off
< align_off
||
2999 orig_end
> align_off
+ align_alen
||
3000 align_alen
- temp
< orig_alen
)
3003 * Try to fix it by moving the start up.
3005 if (align_off
+ temp
<= orig_off
) {
3010 * Try to fix it by moving the end in.
3012 else if (align_off
+ align_alen
- temp
>= orig_end
)
3015 * Set the start to the minimum then trim the length.
3018 align_alen
-= orig_off
- align_off
;
3019 align_off
= orig_off
;
3020 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
3023 * Result doesn't cover the request, fail it.
3025 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
3028 ASSERT(orig_off
>= align_off
);
3029 /* see MAXEXTLEN handling above */
3030 ASSERT(orig_end
<= align_off
+ align_alen
||
3031 align_alen
+ extsz
> MAXEXTLEN
);
3035 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
3036 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
3037 if (prevp
->br_startoff
!= NULLFILEOFF
)
3038 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
3046 #define XFS_ALLOC_GAP_UNITS 4
3050 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3052 xfs_fsblock_t adjust
; /* adjustment to block numbers */
3053 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3054 xfs_mount_t
*mp
; /* mount point structure */
3055 int nullfb
; /* true if ap->firstblock isn't set */
3056 int rt
; /* true if inode is realtime */
3058 #define ISVALID(x,y) \
3060 (x) < mp->m_sb.sb_rblocks : \
3061 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3062 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3063 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3065 mp
= ap
->ip
->i_mount
;
3066 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3067 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) &&
3068 xfs_alloc_is_userdata(ap
->datatype
);
3069 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3070 ap
->tp
->t_firstblock
);
3072 * If allocating at eof, and there's a previous real block,
3073 * try to use its last block as our starting point.
3075 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3076 !isnullstartblock(ap
->prev
.br_startblock
) &&
3077 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
3078 ap
->prev
.br_startblock
)) {
3079 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
3081 * Adjust for the gap between prevp and us.
3083 adjust
= ap
->offset
-
3084 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
3086 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
3087 ap
->blkno
+= adjust
;
3090 * If not at eof, then compare the two neighbor blocks.
3091 * Figure out whether either one gives us a good starting point,
3092 * and pick the better one.
3094 else if (!ap
->eof
) {
3095 xfs_fsblock_t gotbno
; /* right side block number */
3096 xfs_fsblock_t gotdiff
=0; /* right side difference */
3097 xfs_fsblock_t prevbno
; /* left side block number */
3098 xfs_fsblock_t prevdiff
=0; /* left side difference */
3101 * If there's a previous (left) block, select a requested
3102 * start block based on it.
3104 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
3105 !isnullstartblock(ap
->prev
.br_startblock
) &&
3106 (prevbno
= ap
->prev
.br_startblock
+
3107 ap
->prev
.br_blockcount
) &&
3108 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
3110 * Calculate gap to end of previous block.
3112 adjust
= prevdiff
= ap
->offset
-
3113 (ap
->prev
.br_startoff
+
3114 ap
->prev
.br_blockcount
);
3116 * Figure the startblock based on the previous block's
3117 * end and the gap size.
3119 * If the gap is large relative to the piece we're
3120 * allocating, or using it gives us an invalid block
3121 * number, then just use the end of the previous block.
3123 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3124 ISVALID(prevbno
+ prevdiff
,
3125 ap
->prev
.br_startblock
))
3130 * If the firstblock forbids it, can't use it,
3133 if (!rt
&& !nullfb
&&
3134 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
3135 prevbno
= NULLFSBLOCK
;
3138 * No previous block or can't follow it, just default.
3141 prevbno
= NULLFSBLOCK
;
3143 * If there's a following (right) block, select a requested
3144 * start block based on it.
3146 if (!isnullstartblock(ap
->got
.br_startblock
)) {
3148 * Calculate gap to start of next block.
3150 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
3152 * Figure the startblock based on the next block's
3153 * start and the gap size.
3155 gotbno
= ap
->got
.br_startblock
;
3158 * If the gap is large relative to the piece we're
3159 * allocating, or using it gives us an invalid block
3160 * number, then just use the start of the next block
3161 * offset by our length.
3163 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
3164 ISVALID(gotbno
- gotdiff
, gotbno
))
3166 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
3167 gotbno
-= ap
->length
;
3168 gotdiff
+= adjust
- ap
->length
;
3172 * If the firstblock forbids it, can't use it,
3175 if (!rt
&& !nullfb
&&
3176 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
3177 gotbno
= NULLFSBLOCK
;
3180 * No next block, just default.
3183 gotbno
= NULLFSBLOCK
;
3185 * If both valid, pick the better one, else the only good
3186 * one, else ap->blkno is already set (to 0 or the inode block).
3188 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
3189 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
3190 else if (prevbno
!= NULLFSBLOCK
)
3191 ap
->blkno
= prevbno
;
3192 else if (gotbno
!= NULLFSBLOCK
)
3199 xfs_bmap_longest_free_extent(
3200 struct xfs_trans
*tp
,
3205 struct xfs_mount
*mp
= tp
->t_mountp
;
3206 struct xfs_perag
*pag
;
3207 xfs_extlen_t longest
;
3210 pag
= xfs_perag_get(mp
, ag
);
3211 if (!pag
->pagf_init
) {
3212 error
= xfs_alloc_pagf_init(mp
, tp
, ag
, XFS_ALLOC_FLAG_TRYLOCK
);
3216 if (!pag
->pagf_init
) {
3222 longest
= xfs_alloc_longest_free_extent(pag
,
3223 xfs_alloc_min_freelist(mp
, pag
),
3224 xfs_ag_resv_needed(pag
, XFS_AG_RESV_NONE
));
3225 if (*blen
< longest
)
3234 xfs_bmap_select_minlen(
3235 struct xfs_bmalloca
*ap
,
3236 struct xfs_alloc_arg
*args
,
3240 if (notinit
|| *blen
< ap
->minlen
) {
3242 * Since we did a BUF_TRYLOCK above, it is possible that
3243 * there is space for this request.
3245 args
->minlen
= ap
->minlen
;
3246 } else if (*blen
< args
->maxlen
) {
3248 * If the best seen length is less than the request length,
3249 * use the best as the minimum.
3251 args
->minlen
= *blen
;
3254 * Otherwise we've seen an extent as big as maxlen, use that
3257 args
->minlen
= args
->maxlen
;
3262 xfs_bmap_btalloc_nullfb(
3263 struct xfs_bmalloca
*ap
,
3264 struct xfs_alloc_arg
*args
,
3267 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3268 xfs_agnumber_t ag
, startag
;
3272 args
->type
= XFS_ALLOCTYPE_START_BNO
;
3273 args
->total
= ap
->total
;
3275 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3276 if (startag
== NULLAGNUMBER
)
3279 while (*blen
< args
->maxlen
) {
3280 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3285 if (++ag
== mp
->m_sb
.sb_agcount
)
3291 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3296 xfs_bmap_btalloc_filestreams(
3297 struct xfs_bmalloca
*ap
,
3298 struct xfs_alloc_arg
*args
,
3301 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
3306 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
3307 args
->total
= ap
->total
;
3309 ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
3310 if (ag
== NULLAGNUMBER
)
3313 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
, ¬init
);
3317 if (*blen
< args
->maxlen
) {
3318 error
= xfs_filestream_new_ag(ap
, &ag
);
3322 error
= xfs_bmap_longest_free_extent(args
->tp
, ag
, blen
,
3329 xfs_bmap_select_minlen(ap
, args
, blen
, notinit
);
3332 * Set the failure fallback case to look in the selected AG as stream
3335 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3339 /* Update all inode and quota accounting for the allocation we just did. */
3341 xfs_bmap_btalloc_accounting(
3342 struct xfs_bmalloca
*ap
,
3343 struct xfs_alloc_arg
*args
)
3345 if (ap
->flags
& XFS_BMAPI_COWFORK
) {
3347 * COW fork blocks are in-core only and thus are treated as
3348 * in-core quota reservation (like delalloc blocks) even when
3349 * converted to real blocks. The quota reservation is not
3350 * accounted to disk until blocks are remapped to the data
3351 * fork. So if these blocks were previously delalloc, we
3352 * already have quota reservation and there's nothing to do
3359 * Otherwise, we've allocated blocks in a hole. The transaction
3360 * has acquired in-core quota reservation for this extent.
3361 * Rather than account these as real blocks, however, we reduce
3362 * the transaction quota reservation based on the allocation.
3363 * This essentially transfers the transaction quota reservation
3364 * to that of a delalloc extent.
3366 ap
->ip
->i_delayed_blks
+= args
->len
;
3367 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
, XFS_TRANS_DQ_RES_BLKS
,
3372 /* data/attr fork only */
3373 ap
->ip
->i_d
.di_nblocks
+= args
->len
;
3374 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
3376 ap
->ip
->i_delayed_blks
-= args
->len
;
3377 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
3378 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
: XFS_TRANS_DQ_BCOUNT
,
3384 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3386 xfs_mount_t
*mp
; /* mount point structure */
3387 xfs_alloctype_t atype
= 0; /* type for allocation routines */
3388 xfs_extlen_t align
= 0; /* minimum allocation alignment */
3389 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
3391 xfs_alloc_arg_t args
;
3392 xfs_fileoff_t orig_offset
;
3393 xfs_extlen_t orig_length
;
3395 xfs_extlen_t nextminlen
= 0;
3396 int nullfb
; /* true if ap->firstblock isn't set */
3403 orig_offset
= ap
->offset
;
3404 orig_length
= ap
->length
;
3406 mp
= ap
->ip
->i_mount
;
3408 /* stripe alignment for allocation is determined by mount parameters */
3410 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
3411 stripe_align
= mp
->m_swidth
;
3412 else if (mp
->m_dalign
)
3413 stripe_align
= mp
->m_dalign
;
3415 if (ap
->flags
& XFS_BMAPI_COWFORK
)
3416 align
= xfs_get_cowextsz_hint(ap
->ip
);
3417 else if (xfs_alloc_is_userdata(ap
->datatype
))
3418 align
= xfs_get_extsz_hint(ap
->ip
);
3420 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
3421 align
, 0, ap
->eof
, 0, ap
->conv
,
3422 &ap
->offset
, &ap
->length
);
3428 nullfb
= ap
->tp
->t_firstblock
== NULLFSBLOCK
;
3429 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
,
3430 ap
->tp
->t_firstblock
);
3432 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3433 xfs_inode_is_filestream(ap
->ip
)) {
3434 ag
= xfs_filestream_lookup_ag(ap
->ip
);
3435 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
3436 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
3438 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
3441 ap
->blkno
= ap
->tp
->t_firstblock
;
3443 xfs_bmap_adjacent(ap
);
3446 * If allowed, use ap->blkno; otherwise must use firstblock since
3447 * it's in the right allocation group.
3449 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
3452 ap
->blkno
= ap
->tp
->t_firstblock
;
3454 * Normal allocation, done through xfs_alloc_vextent.
3456 tryagain
= isaligned
= 0;
3457 memset(&args
, 0, sizeof(args
));
3460 args
.fsbno
= ap
->blkno
;
3461 args
.oinfo
= XFS_RMAP_OINFO_SKIP_UPDATE
;
3463 /* Trim the allocation back to the maximum an AG can fit. */
3464 args
.maxlen
= min(ap
->length
, mp
->m_ag_max_usable
);
3468 * Search for an allocation group with a single extent large
3469 * enough for the request. If one isn't found, then adjust
3470 * the minimum allocation size to the largest space found.
3472 if (xfs_alloc_is_userdata(ap
->datatype
) &&
3473 xfs_inode_is_filestream(ap
->ip
))
3474 error
= xfs_bmap_btalloc_filestreams(ap
, &args
, &blen
);
3476 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
3479 } else if (ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) {
3480 if (xfs_inode_is_filestream(ap
->ip
))
3481 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3483 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3484 args
.total
= args
.minlen
= ap
->minlen
;
3486 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3487 args
.total
= ap
->total
;
3488 args
.minlen
= ap
->minlen
;
3490 /* apply extent size hints if obtained earlier */
3493 div_u64_rem(ap
->offset
, args
.prod
, &args
.mod
);
3495 args
.mod
= args
.prod
- args
.mod
;
3496 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_SIZE
) {
3500 args
.prod
= PAGE_SIZE
>> mp
->m_sb
.sb_blocklog
;
3501 div_u64_rem(ap
->offset
, args
.prod
, &args
.mod
);
3503 args
.mod
= args
.prod
- args
.mod
;
3506 * If we are not low on available data blocks, and the
3507 * underlying logical volume manager is a stripe, and
3508 * the file offset is zero then try to allocate data
3509 * blocks on stripe unit boundary.
3510 * NOTE: ap->aeof is only set if the allocation length
3511 * is >= the stripe unit and the allocation offset is
3512 * at the end of file.
3514 if (!(ap
->tp
->t_flags
& XFS_TRANS_LOWMODE
) && ap
->aeof
) {
3516 args
.alignment
= stripe_align
;
3520 * Adjust for alignment
3522 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
3523 args
.minlen
= blen
- args
.alignment
;
3524 args
.minalignslop
= 0;
3527 * First try an exact bno allocation.
3528 * If it fails then do a near or start bno
3529 * allocation with alignment turned on.
3533 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
3536 * Compute the minlen+alignment for the
3537 * next case. Set slop so that the value
3538 * of minlen+alignment+slop doesn't go up
3539 * between the calls.
3541 if (blen
> stripe_align
&& blen
<= args
.maxlen
)
3542 nextminlen
= blen
- stripe_align
;
3544 nextminlen
= args
.minlen
;
3545 if (nextminlen
+ stripe_align
> args
.minlen
+ 1)
3547 nextminlen
+ stripe_align
-
3550 args
.minalignslop
= 0;
3554 args
.minalignslop
= 0;
3556 args
.minleft
= ap
->minleft
;
3557 args
.wasdel
= ap
->wasdel
;
3558 args
.resv
= XFS_AG_RESV_NONE
;
3559 args
.datatype
= ap
->datatype
;
3560 if (ap
->datatype
& XFS_ALLOC_USERDATA_ZERO
)
3563 error
= xfs_alloc_vextent(&args
);
3567 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
3569 * Exact allocation failed. Now try with alignment
3573 args
.fsbno
= ap
->blkno
;
3574 args
.alignment
= stripe_align
;
3575 args
.minlen
= nextminlen
;
3576 args
.minalignslop
= 0;
3578 if ((error
= xfs_alloc_vextent(&args
)))
3581 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
3583 * allocation failed, so turn off alignment and
3587 args
.fsbno
= ap
->blkno
;
3589 if ((error
= xfs_alloc_vextent(&args
)))
3592 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
3593 args
.minlen
> ap
->minlen
) {
3594 args
.minlen
= ap
->minlen
;
3595 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3596 args
.fsbno
= ap
->blkno
;
3597 if ((error
= xfs_alloc_vextent(&args
)))
3600 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
3602 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
3603 args
.total
= ap
->minlen
;
3604 if ((error
= xfs_alloc_vextent(&args
)))
3606 ap
->tp
->t_flags
|= XFS_TRANS_LOWMODE
;
3608 if (args
.fsbno
!= NULLFSBLOCK
) {
3610 * check the allocation happened at the same or higher AG than
3611 * the first block that was allocated.
3613 ASSERT(ap
->tp
->t_firstblock
== NULLFSBLOCK
||
3614 XFS_FSB_TO_AGNO(mp
, ap
->tp
->t_firstblock
) <=
3615 XFS_FSB_TO_AGNO(mp
, args
.fsbno
));
3617 ap
->blkno
= args
.fsbno
;
3618 if (ap
->tp
->t_firstblock
== NULLFSBLOCK
)
3619 ap
->tp
->t_firstblock
= args
.fsbno
;
3620 ASSERT(nullfb
|| fb_agno
<= args
.agno
);
3621 ap
->length
= args
.len
;
3623 * If the extent size hint is active, we tried to round the
3624 * caller's allocation request offset down to extsz and the
3625 * length up to another extsz boundary. If we found a free
3626 * extent we mapped it in starting at this new offset. If the
3627 * newly mapped space isn't long enough to cover any of the
3628 * range of offsets that was originally requested, move the
3629 * mapping up so that we can fill as much of the caller's
3630 * original request as possible. Free space is apparently
3631 * very fragmented so we're unlikely to be able to satisfy the
3634 if (ap
->length
<= orig_length
)
3635 ap
->offset
= orig_offset
;
3636 else if (ap
->offset
+ ap
->length
< orig_offset
+ orig_length
)
3637 ap
->offset
= orig_offset
+ orig_length
- ap
->length
;
3638 xfs_bmap_btalloc_accounting(ap
, &args
);
3640 ap
->blkno
= NULLFSBLOCK
;
3647 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3648 * It figures out where to ask the underlying allocator to put the new extent.
3652 struct xfs_bmalloca
*ap
) /* bmap alloc argument struct */
3654 if (XFS_IS_REALTIME_INODE(ap
->ip
) &&
3655 xfs_alloc_is_userdata(ap
->datatype
))
3656 return xfs_bmap_rtalloc(ap
);
3657 return xfs_bmap_btalloc(ap
);
3660 /* Trim extent to fit a logical block range. */
3663 struct xfs_bmbt_irec
*irec
,
3667 xfs_fileoff_t distance
;
3668 xfs_fileoff_t end
= bno
+ len
;
3670 if (irec
->br_startoff
+ irec
->br_blockcount
<= bno
||
3671 irec
->br_startoff
>= end
) {
3672 irec
->br_blockcount
= 0;
3676 if (irec
->br_startoff
< bno
) {
3677 distance
= bno
- irec
->br_startoff
;
3678 if (isnullstartblock(irec
->br_startblock
))
3679 irec
->br_startblock
= DELAYSTARTBLOCK
;
3680 if (irec
->br_startblock
!= DELAYSTARTBLOCK
&&
3681 irec
->br_startblock
!= HOLESTARTBLOCK
)
3682 irec
->br_startblock
+= distance
;
3683 irec
->br_startoff
+= distance
;
3684 irec
->br_blockcount
-= distance
;
3687 if (end
< irec
->br_startoff
+ irec
->br_blockcount
) {
3688 distance
= irec
->br_startoff
+ irec
->br_blockcount
- end
;
3689 irec
->br_blockcount
-= distance
;
3694 * Trim the returned map to the required bounds
3698 struct xfs_bmbt_irec
*mval
,
3699 struct xfs_bmbt_irec
*got
,
3707 if ((flags
& XFS_BMAPI_ENTIRE
) ||
3708 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
3710 if (isnullstartblock(got
->br_startblock
))
3711 mval
->br_startblock
= DELAYSTARTBLOCK
;
3717 ASSERT((*bno
>= obno
) || (n
== 0));
3719 mval
->br_startoff
= *bno
;
3720 if (isnullstartblock(got
->br_startblock
))
3721 mval
->br_startblock
= DELAYSTARTBLOCK
;
3723 mval
->br_startblock
= got
->br_startblock
+
3724 (*bno
- got
->br_startoff
);
3726 * Return the minimum of what we got and what we asked for for
3727 * the length. We can use the len variable here because it is
3728 * modified below and we could have been there before coming
3729 * here if the first part of the allocation didn't overlap what
3732 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
3733 got
->br_blockcount
- (*bno
- got
->br_startoff
));
3734 mval
->br_state
= got
->br_state
;
3735 ASSERT(mval
->br_blockcount
<= len
);
3740 * Update and validate the extent map to return
3743 xfs_bmapi_update_map(
3744 struct xfs_bmbt_irec
**map
,
3752 xfs_bmbt_irec_t
*mval
= *map
;
3754 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
3755 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
3756 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
3757 (mval
->br_startoff
< obno
));
3759 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
3761 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
3762 /* update previous map with new information */
3763 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
3764 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
3765 ASSERT(mval
->br_state
== mval
[-1].br_state
);
3766 mval
[-1].br_blockcount
= mval
->br_blockcount
;
3767 mval
[-1].br_state
= mval
->br_state
;
3768 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
3769 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
3770 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
3771 mval
->br_startblock
== mval
[-1].br_startblock
+
3772 mval
[-1].br_blockcount
&&
3773 mval
[-1].br_state
== mval
->br_state
) {
3774 ASSERT(mval
->br_startoff
==
3775 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
3776 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3777 } else if (*n
> 0 &&
3778 mval
->br_startblock
== DELAYSTARTBLOCK
&&
3779 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
3780 mval
->br_startoff
==
3781 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
3782 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
3783 mval
[-1].br_state
= mval
->br_state
;
3784 } else if (!((*n
== 0) &&
3785 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
3794 * Map file blocks to filesystem blocks without allocation.
3798 struct xfs_inode
*ip
,
3801 struct xfs_bmbt_irec
*mval
,
3805 struct xfs_mount
*mp
= ip
->i_mount
;
3806 struct xfs_ifork
*ifp
;
3807 struct xfs_bmbt_irec got
;
3810 struct xfs_iext_cursor icur
;
3814 int whichfork
= xfs_bmapi_whichfork(flags
);
3817 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
3818 XFS_BMAPI_COWFORK
)));
3819 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
));
3821 if (unlikely(XFS_TEST_ERROR(
3822 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
3823 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
3824 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
3825 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
3826 return -EFSCORRUPTED
;
3829 if (XFS_FORCED_SHUTDOWN(mp
))
3832 XFS_STATS_INC(mp
, xs_blk_mapr
);
3834 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3836 /* No CoW fork? Return a hole. */
3837 if (whichfork
== XFS_COW_FORK
&& !ifp
) {
3838 mval
->br_startoff
= bno
;
3839 mval
->br_startblock
= HOLESTARTBLOCK
;
3840 mval
->br_blockcount
= len
;
3841 mval
->br_state
= XFS_EXT_NORM
;
3846 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
3847 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
3852 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
))
3857 while (bno
< end
&& n
< *nmap
) {
3858 /* Reading past eof, act as though there's a hole up to end. */
3860 got
.br_startoff
= end
;
3861 if (got
.br_startoff
> bno
) {
3862 /* Reading in a hole. */
3863 mval
->br_startoff
= bno
;
3864 mval
->br_startblock
= HOLESTARTBLOCK
;
3865 mval
->br_blockcount
=
3866 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
3867 mval
->br_state
= XFS_EXT_NORM
;
3868 bno
+= mval
->br_blockcount
;
3869 len
-= mval
->br_blockcount
;
3875 /* set up the extent map to return. */
3876 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
3877 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
3879 /* If we're done, stop now. */
3880 if (bno
>= end
|| n
>= *nmap
)
3883 /* Else go on to the next record. */
3884 if (!xfs_iext_next_extent(ifp
, &icur
, &got
))
3892 * Add a delayed allocation extent to an inode. Blocks are reserved from the
3893 * global pool and the extent inserted into the inode in-core extent tree.
3895 * On entry, got refers to the first extent beyond the offset of the extent to
3896 * allocate or eof is specified if no such extent exists. On return, got refers
3897 * to the extent record that was inserted to the inode fork.
3899 * Note that the allocated extent may have been merged with contiguous extents
3900 * during insertion into the inode fork. Thus, got does not reflect the current
3901 * state of the inode fork on return. If necessary, the caller can use lastx to
3902 * look up the updated record in the inode fork.
3905 xfs_bmapi_reserve_delalloc(
3906 struct xfs_inode
*ip
,
3910 xfs_filblks_t prealloc
,
3911 struct xfs_bmbt_irec
*got
,
3912 struct xfs_iext_cursor
*icur
,
3915 struct xfs_mount
*mp
= ip
->i_mount
;
3916 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3918 xfs_extlen_t indlen
;
3920 xfs_fileoff_t aoff
= off
;
3923 * Cap the alloc length. Keep track of prealloc so we know whether to
3924 * tag the inode before we return.
3926 alen
= XFS_FILBLKS_MIN(len
+ prealloc
, MAXEXTLEN
);
3928 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
3929 if (prealloc
&& alen
>= len
)
3930 prealloc
= alen
- len
;
3932 /* Figure out the extent size, adjust alen */
3933 if (whichfork
== XFS_COW_FORK
) {
3934 struct xfs_bmbt_irec prev
;
3935 xfs_extlen_t extsz
= xfs_get_cowextsz_hint(ip
);
3937 if (!xfs_iext_peek_prev_extent(ifp
, icur
, &prev
))
3938 prev
.br_startoff
= NULLFILEOFF
;
3940 error
= xfs_bmap_extsize_align(mp
, got
, &prev
, extsz
, 0, eof
,
3941 1, 0, &aoff
, &alen
);
3946 * Make a transaction-less quota reservation for delayed allocation
3947 * blocks. This number gets adjusted later. We return if we haven't
3948 * allocated blocks already inside this loop.
3950 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
3951 XFS_QMOPT_RES_REGBLKS
);
3956 * Split changing sb for alen and indlen since they could be coming
3957 * from different places.
3959 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
3962 error
= xfs_mod_fdblocks(mp
, -((int64_t)alen
), false);
3964 goto out_unreserve_quota
;
3966 error
= xfs_mod_fdblocks(mp
, -((int64_t)indlen
), false);
3968 goto out_unreserve_blocks
;
3971 ip
->i_delayed_blks
+= alen
;
3973 got
->br_startoff
= aoff
;
3974 got
->br_startblock
= nullstartblock(indlen
);
3975 got
->br_blockcount
= alen
;
3976 got
->br_state
= XFS_EXT_NORM
;
3978 xfs_bmap_add_extent_hole_delay(ip
, whichfork
, icur
, got
);
3981 * Tag the inode if blocks were preallocated. Note that COW fork
3982 * preallocation can occur at the start or end of the extent, even when
3983 * prealloc == 0, so we must also check the aligned offset and length.
3985 if (whichfork
== XFS_DATA_FORK
&& prealloc
)
3986 xfs_inode_set_eofblocks_tag(ip
);
3987 if (whichfork
== XFS_COW_FORK
&& (prealloc
|| aoff
< off
|| alen
> len
))
3988 xfs_inode_set_cowblocks_tag(ip
);
3992 out_unreserve_blocks
:
3993 xfs_mod_fdblocks(mp
, alen
, false);
3994 out_unreserve_quota
:
3995 if (XFS_IS_QUOTA_ON(mp
))
3996 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
3997 XFS_QMOPT_RES_REGBLKS
);
4003 struct xfs_bmalloca
*bma
)
4005 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4006 int whichfork
= xfs_bmapi_whichfork(bma
->flags
);
4007 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4008 int tmp_logflags
= 0;
4011 ASSERT(bma
->length
> 0);
4014 * For the wasdelay case, we could also just allocate the stuff asked
4015 * for in this bmap call but that wouldn't be as good.
4018 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4019 bma
->offset
= bma
->got
.br_startoff
;
4020 xfs_iext_peek_prev_extent(ifp
, &bma
->icur
, &bma
->prev
);
4022 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4024 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4025 bma
->got
.br_startoff
- bma
->offset
);
4029 * Set the data type being allocated. For the data fork, the first data
4030 * in the file is treated differently to all other allocations. For the
4031 * attribute fork, we only need to ensure the allocated range is not on
4034 if (!(bma
->flags
& XFS_BMAPI_METADATA
)) {
4035 bma
->datatype
= XFS_ALLOC_NOBUSY
;
4036 if (whichfork
== XFS_DATA_FORK
) {
4037 if (bma
->offset
== 0)
4038 bma
->datatype
|= XFS_ALLOC_INITIAL_USER_DATA
;
4040 bma
->datatype
|= XFS_ALLOC_USERDATA
;
4042 if (bma
->flags
& XFS_BMAPI_ZERO
)
4043 bma
->datatype
|= XFS_ALLOC_USERDATA_ZERO
;
4046 bma
->minlen
= (bma
->flags
& XFS_BMAPI_CONTIG
) ? bma
->length
: 1;
4049 * Only want to do the alignment at the eof if it is userdata and
4050 * allocation length is larger than a stripe unit.
4052 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
&&
4053 !(bma
->flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4054 error
= xfs_bmap_isaeof(bma
, whichfork
);
4059 error
= xfs_bmap_alloc(bma
);
4063 if (bma
->blkno
== NULLFSBLOCK
)
4065 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
)
4066 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4068 * Bump the number of extents we've allocated
4074 bma
->cur
->bc_private
.b
.flags
=
4075 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4077 bma
->got
.br_startoff
= bma
->offset
;
4078 bma
->got
.br_startblock
= bma
->blkno
;
4079 bma
->got
.br_blockcount
= bma
->length
;
4080 bma
->got
.br_state
= XFS_EXT_NORM
;
4083 * In the data fork, a wasdelay extent has been initialized, so
4084 * shouldn't be flagged as unwritten.
4086 * For the cow fork, however, we convert delalloc reservations
4087 * (extents allocated for speculative preallocation) to
4088 * allocated unwritten extents, and only convert the unwritten
4089 * extents to real extents when we're about to write the data.
4091 if ((!bma
->wasdel
|| (bma
->flags
& XFS_BMAPI_COWFORK
)) &&
4092 (bma
->flags
& XFS_BMAPI_PREALLOC
))
4093 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4096 error
= xfs_bmap_add_extent_delay_real(bma
, whichfork
);
4098 error
= xfs_bmap_add_extent_hole_real(bma
->tp
, bma
->ip
,
4099 whichfork
, &bma
->icur
, &bma
->cur
, &bma
->got
,
4100 &bma
->logflags
, bma
->flags
);
4102 bma
->logflags
|= tmp_logflags
;
4107 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4108 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4109 * the neighbouring ones.
4111 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4113 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4114 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4115 bma
->offset
+ bma
->length
);
4116 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4117 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4122 xfs_bmapi_convert_unwritten(
4123 struct xfs_bmalloca
*bma
,
4124 struct xfs_bmbt_irec
*mval
,
4128 int whichfork
= xfs_bmapi_whichfork(flags
);
4129 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4130 int tmp_logflags
= 0;
4133 /* check if we need to do unwritten->real conversion */
4134 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4135 (flags
& XFS_BMAPI_PREALLOC
))
4138 /* check if we need to do real->unwritten conversion */
4139 if (mval
->br_state
== XFS_EXT_NORM
&&
4140 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4141 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4145 * Modify (by adding) the state flag, if writing.
4147 ASSERT(mval
->br_blockcount
<= len
);
4148 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4149 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4150 bma
->ip
, whichfork
);
4152 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4153 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4156 * Before insertion into the bmbt, zero the range being converted
4159 if (flags
& XFS_BMAPI_ZERO
) {
4160 error
= xfs_zero_extent(bma
->ip
, mval
->br_startblock
,
4161 mval
->br_blockcount
);
4166 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, whichfork
,
4167 &bma
->icur
, &bma
->cur
, mval
, &tmp_logflags
);
4169 * Log the inode core unconditionally in the unwritten extent conversion
4170 * path because the conversion might not have done so (e.g., if the
4171 * extent count hasn't changed). We need to make sure the inode is dirty
4172 * in the transaction for the sake of fsync(), even if nothing has
4173 * changed, because fsync() will not force the log for this transaction
4174 * unless it sees the inode pinned.
4176 * Note: If we're only converting cow fork extents, there aren't
4177 * any on-disk updates to make, so we don't need to log anything.
4179 if (whichfork
!= XFS_COW_FORK
)
4180 bma
->logflags
|= tmp_logflags
| XFS_ILOG_CORE
;
4185 * Update our extent pointer, given that
4186 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4187 * of the neighbouring ones.
4189 xfs_iext_get_extent(ifp
, &bma
->icur
, &bma
->got
);
4192 * We may have combined previously unwritten space with written space,
4193 * so generate another request.
4195 if (mval
->br_blockcount
< len
)
4200 static inline xfs_extlen_t
4202 struct xfs_trans
*tp
,
4203 struct xfs_inode
*ip
,
4206 if (tp
&& tp
->t_firstblock
!= NULLFSBLOCK
)
4208 if (XFS_IFORK_FORMAT(ip
, fork
) != XFS_DINODE_FMT_BTREE
)
4210 return be16_to_cpu(XFS_IFORK_PTR(ip
, fork
)->if_broot
->bb_level
) + 1;
4214 * Log whatever the flags say, even if error. Otherwise we might miss detecting
4215 * a case where the data is changed, there's an error, and it's not logged so we
4216 * don't shutdown when we should. Don't bother logging extents/btree changes if
4217 * we converted to the other format.
4221 struct xfs_bmalloca
*bma
,
4225 if ((bma
->logflags
& xfs_ilog_fext(whichfork
)) &&
4226 XFS_IFORK_FORMAT(bma
->ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4227 bma
->logflags
&= ~xfs_ilog_fext(whichfork
);
4228 else if ((bma
->logflags
& xfs_ilog_fbroot(whichfork
)) &&
4229 XFS_IFORK_FORMAT(bma
->ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
4230 bma
->logflags
&= ~xfs_ilog_fbroot(whichfork
);
4233 xfs_trans_log_inode(bma
->tp
, bma
->ip
, bma
->logflags
);
4235 xfs_btree_del_cursor(bma
->cur
, error
);
4239 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4240 * extent state if necessary. Details behaviour is controlled by the flags
4241 * parameter. Only allocates blocks from a single allocation group, to avoid
4246 struct xfs_trans
*tp
, /* transaction pointer */
4247 struct xfs_inode
*ip
, /* incore inode */
4248 xfs_fileoff_t bno
, /* starting file offs. mapped */
4249 xfs_filblks_t len
, /* length to map in file */
4250 int flags
, /* XFS_BMAPI_... */
4251 xfs_extlen_t total
, /* total blocks needed */
4252 struct xfs_bmbt_irec
*mval
, /* output: map values */
4253 int *nmap
) /* i/o: mval size/count */
4255 struct xfs_bmalloca bma
= {
4260 struct xfs_mount
*mp
= ip
->i_mount
;
4261 struct xfs_ifork
*ifp
;
4262 xfs_fileoff_t end
; /* end of mapped file region */
4263 bool eof
= false; /* after the end of extents */
4264 int error
; /* error return */
4265 int n
; /* current extent index */
4266 xfs_fileoff_t obno
; /* old block number (offset) */
4267 int whichfork
; /* data or attr fork */
4270 xfs_fileoff_t orig_bno
; /* original block number value */
4271 int orig_flags
; /* original flags arg value */
4272 xfs_filblks_t orig_len
; /* original value of len arg */
4273 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4274 int orig_nmap
; /* original value of *nmap */
4282 whichfork
= xfs_bmapi_whichfork(flags
);
4285 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4288 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
);
4289 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4290 ASSERT(!(flags
& XFS_BMAPI_REMAP
));
4292 /* zeroing is for currently only for data extents, not metadata */
4293 ASSERT((flags
& (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
)) !=
4294 (XFS_BMAPI_METADATA
| XFS_BMAPI_ZERO
));
4296 * we can allocate unwritten extents or pre-zero allocated blocks,
4297 * but it makes no sense to do both at once. This would result in
4298 * zeroing the unwritten extent twice, but it still being an
4299 * unwritten extent....
4301 ASSERT((flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
)) !=
4302 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_ZERO
));
4304 if (unlikely(XFS_TEST_ERROR(
4305 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4306 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4307 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
4308 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW
, mp
);
4309 return -EFSCORRUPTED
;
4312 if (XFS_FORCED_SHUTDOWN(mp
))
4315 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4317 XFS_STATS_INC(mp
, xs_blk_mapw
);
4319 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4320 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4325 if (!xfs_iext_lookup_extent(ip
, ifp
, bno
, &bma
.icur
, &bma
.got
))
4327 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4328 bma
.prev
.br_startoff
= NULLFILEOFF
;
4329 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4334 while (bno
< end
&& n
< *nmap
) {
4335 bool need_alloc
= false, wasdelay
= false;
4337 /* in hole or beyond EOF? */
4338 if (eof
|| bma
.got
.br_startoff
> bno
) {
4340 * CoW fork conversions should /never/ hit EOF or
4341 * holes. There should always be something for us
4344 ASSERT(!((flags
& XFS_BMAPI_CONVERT
) &&
4345 (flags
& XFS_BMAPI_COWFORK
)));
4348 } else if (isnullstartblock(bma
.got
.br_startblock
)) {
4353 * First, deal with the hole before the allocated space
4354 * that we found, if any.
4356 if (need_alloc
|| wasdelay
) {
4358 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4359 bma
.wasdel
= wasdelay
;
4364 * There's a 32/64 bit type mismatch between the
4365 * allocation length request (which can be 64 bits in
4366 * length) and the bma length request, which is
4367 * xfs_extlen_t and therefore 32 bits. Hence we have to
4368 * check for 32-bit overflows and handle them here.
4370 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
4371 bma
.length
= MAXEXTLEN
;
4376 ASSERT(bma
.length
> 0);
4377 error
= xfs_bmapi_allocate(&bma
);
4380 if (bma
.blkno
== NULLFSBLOCK
)
4384 * If this is a CoW allocation, record the data in
4385 * the refcount btree for orphan recovery.
4387 if (whichfork
== XFS_COW_FORK
) {
4388 error
= xfs_refcount_alloc_cow_extent(tp
,
4389 bma
.blkno
, bma
.length
);
4395 /* Deal with the allocated space we found. */
4396 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
4399 /* Execute unwritten extent conversion if necessary */
4400 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
4401 if (error
== -EAGAIN
)
4406 /* update the extent map to return */
4407 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4410 * If we're done, stop now. Stop when we've allocated
4411 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4412 * the transaction may get too big.
4414 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
4417 /* Else go on to the next record. */
4419 if (!xfs_iext_next_extent(ifp
, &bma
.icur
, &bma
.got
))
4424 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4429 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
4430 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
4431 XFS_IFORK_MAXEXT(ip
, whichfork
));
4432 xfs_bmapi_finish(&bma
, whichfork
, 0);
4433 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
4437 xfs_bmapi_finish(&bma
, whichfork
, error
);
4442 * Convert an existing delalloc extent to real blocks based on file offset. This
4443 * attempts to allocate the entire delalloc extent and may require multiple
4444 * invocations to allocate the target offset if a large enough physical extent
4448 xfs_bmapi_convert_delalloc(
4449 struct xfs_inode
*ip
,
4451 xfs_fileoff_t offset_fsb
,
4452 struct xfs_bmbt_irec
*imap
,
4455 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4456 struct xfs_mount
*mp
= ip
->i_mount
;
4457 struct xfs_bmalloca bma
= { NULL
};
4458 struct xfs_trans
*tp
;
4462 * Space for the extent and indirect blocks was reserved when the
4463 * delalloc extent was created so there's no need to do so here.
4465 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0, 0,
4466 XFS_TRANS_RESERVE
, &tp
);
4470 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4471 xfs_trans_ijoin(tp
, ip
, 0);
4473 if (!xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &bma
.icur
, &bma
.got
) ||
4474 bma
.got
.br_startoff
> offset_fsb
) {
4476 * No extent found in the range we are trying to convert. This
4477 * should only happen for the COW fork, where another thread
4478 * might have moved the extent to the data fork in the meantime.
4480 WARN_ON_ONCE(whichfork
!= XFS_COW_FORK
);
4482 goto out_trans_cancel
;
4486 * If we find a real extent here we raced with another thread converting
4487 * the extent. Just return the real extent at this offset.
4489 if (!isnullstartblock(bma
.got
.br_startblock
)) {
4491 *seq
= READ_ONCE(ifp
->if_seq
);
4492 goto out_trans_cancel
;
4498 bma
.offset
= bma
.got
.br_startoff
;
4499 bma
.length
= max_t(xfs_filblks_t
, bma
.got
.br_blockcount
, MAXEXTLEN
);
4500 bma
.total
= XFS_EXTENTADD_SPACE_RES(ip
->i_mount
, XFS_DATA_FORK
);
4501 bma
.minleft
= xfs_bmapi_minleft(tp
, ip
, whichfork
);
4502 if (whichfork
== XFS_COW_FORK
)
4503 bma
.flags
= XFS_BMAPI_COWFORK
| XFS_BMAPI_PREALLOC
;
4505 if (!xfs_iext_peek_prev_extent(ifp
, &bma
.icur
, &bma
.prev
))
4506 bma
.prev
.br_startoff
= NULLFILEOFF
;
4508 error
= xfs_bmapi_allocate(&bma
);
4513 if (WARN_ON_ONCE(bma
.blkno
== NULLFSBLOCK
))
4515 error
= -EFSCORRUPTED
;
4516 if (WARN_ON_ONCE(!bma
.got
.br_startblock
&& !XFS_IS_REALTIME_INODE(ip
)))
4519 XFS_STATS_ADD(mp
, xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, bma
.length
));
4520 XFS_STATS_INC(mp
, xs_xstrat_quick
);
4522 ASSERT(!isnullstartblock(bma
.got
.br_startblock
));
4524 *seq
= READ_ONCE(ifp
->if_seq
);
4526 if (whichfork
== XFS_COW_FORK
) {
4527 error
= xfs_refcount_alloc_cow_extent(tp
, bma
.blkno
,
4533 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
, &bma
.logflags
,
4538 xfs_bmapi_finish(&bma
, whichfork
, 0);
4539 error
= xfs_trans_commit(tp
);
4540 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4544 xfs_bmapi_finish(&bma
, whichfork
, error
);
4546 xfs_trans_cancel(tp
);
4547 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4553 struct xfs_trans
*tp
,
4554 struct xfs_inode
*ip
,
4557 xfs_fsblock_t startblock
,
4560 struct xfs_mount
*mp
= ip
->i_mount
;
4561 struct xfs_ifork
*ifp
;
4562 struct xfs_btree_cur
*cur
= NULL
;
4563 struct xfs_bmbt_irec got
;
4564 struct xfs_iext_cursor icur
;
4565 int whichfork
= xfs_bmapi_whichfork(flags
);
4566 int logflags
= 0, error
;
4568 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4570 ASSERT(len
<= (xfs_filblks_t
)MAXEXTLEN
);
4571 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
4572 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
|
4573 XFS_BMAPI_NORMAP
)));
4574 ASSERT((flags
& (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
)) !=
4575 (XFS_BMAPI_ATTRFORK
| XFS_BMAPI_PREALLOC
));
4577 if (unlikely(XFS_TEST_ERROR(
4578 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4579 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4580 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
4581 XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW
, mp
);
4582 return -EFSCORRUPTED
;
4585 if (XFS_FORCED_SHUTDOWN(mp
))
4588 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4589 error
= xfs_iread_extents(tp
, ip
, whichfork
);
4594 if (xfs_iext_lookup_extent(ip
, ifp
, bno
, &icur
, &got
)) {
4595 /* make sure we only reflink into a hole. */
4596 ASSERT(got
.br_startoff
> bno
);
4597 ASSERT(got
.br_startoff
- bno
>= len
);
4600 ip
->i_d
.di_nblocks
+= len
;
4601 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4603 if (ifp
->if_flags
& XFS_IFBROOT
) {
4604 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
4605 cur
->bc_private
.b
.flags
= 0;
4608 got
.br_startoff
= bno
;
4609 got
.br_startblock
= startblock
;
4610 got
.br_blockcount
= len
;
4611 if (flags
& XFS_BMAPI_PREALLOC
)
4612 got
.br_state
= XFS_EXT_UNWRITTEN
;
4614 got
.br_state
= XFS_EXT_NORM
;
4616 error
= xfs_bmap_add_extent_hole_real(tp
, ip
, whichfork
, &icur
,
4617 &cur
, &got
, &logflags
, flags
);
4621 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
, whichfork
);
4624 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
)
4625 logflags
&= ~XFS_ILOG_DEXT
;
4626 else if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
)
4627 logflags
&= ~XFS_ILOG_DBROOT
;
4630 xfs_trans_log_inode(tp
, ip
, logflags
);
4632 xfs_btree_del_cursor(cur
, error
);
4637 * When a delalloc extent is split (e.g., due to a hole punch), the original
4638 * indlen reservation must be shared across the two new extents that are left
4641 * Given the original reservation and the worst case indlen for the two new
4642 * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4643 * reservation fairly across the two new extents. If necessary, steal available
4644 * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4645 * ores == 1). The number of stolen blocks is returned. The availability and
4646 * subsequent accounting of stolen blocks is the responsibility of the caller.
4648 static xfs_filblks_t
4649 xfs_bmap_split_indlen(
4650 xfs_filblks_t ores
, /* original res. */
4651 xfs_filblks_t
*indlen1
, /* ext1 worst indlen */
4652 xfs_filblks_t
*indlen2
, /* ext2 worst indlen */
4653 xfs_filblks_t avail
) /* stealable blocks */
4655 xfs_filblks_t len1
= *indlen1
;
4656 xfs_filblks_t len2
= *indlen2
;
4657 xfs_filblks_t nres
= len1
+ len2
; /* new total res. */
4658 xfs_filblks_t stolen
= 0;
4659 xfs_filblks_t resfactor
;
4662 * Steal as many blocks as we can to try and satisfy the worst case
4663 * indlen for both new extents.
4665 if (ores
< nres
&& avail
)
4666 stolen
= XFS_FILBLKS_MIN(nres
- ores
, avail
);
4669 /* nothing else to do if we've satisfied the new reservation */
4674 * We can't meet the total required reservation for the two extents.
4675 * Calculate the percent of the overall shortage between both extents
4676 * and apply this percentage to each of the requested indlen values.
4677 * This distributes the shortage fairly and reduces the chances that one
4678 * of the two extents is left with nothing when extents are repeatedly
4681 resfactor
= (ores
* 100);
4682 do_div(resfactor
, nres
);
4687 ASSERT(len1
+ len2
<= ores
);
4688 ASSERT(len1
< *indlen1
&& len2
< *indlen2
);
4691 * Hand out the remainder to each extent. If one of the two reservations
4692 * is zero, we want to make sure that one gets a block first. The loop
4693 * below starts with len1, so hand len2 a block right off the bat if it
4696 ores
-= (len1
+ len2
);
4697 ASSERT((*indlen1
- len1
) + (*indlen2
- len2
) >= ores
);
4698 if (ores
&& !len2
&& *indlen2
) {
4703 if (len1
< *indlen1
) {
4709 if (len2
< *indlen2
) {
4722 xfs_bmap_del_extent_delay(
4723 struct xfs_inode
*ip
,
4725 struct xfs_iext_cursor
*icur
,
4726 struct xfs_bmbt_irec
*got
,
4727 struct xfs_bmbt_irec
*del
)
4729 struct xfs_mount
*mp
= ip
->i_mount
;
4730 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4731 struct xfs_bmbt_irec
new;
4732 int64_t da_old
, da_new
, da_diff
= 0;
4733 xfs_fileoff_t del_endoff
, got_endoff
;
4734 xfs_filblks_t got_indlen
, new_indlen
, stolen
;
4735 int state
= xfs_bmap_fork_to_state(whichfork
);
4739 XFS_STATS_INC(mp
, xs_del_exlist
);
4741 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4742 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4743 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4744 da_old
= startblockval(got
->br_startblock
);
4747 ASSERT(del
->br_blockcount
> 0);
4748 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4749 ASSERT(got_endoff
>= del_endoff
);
4752 uint64_t rtexts
= XFS_FSB_TO_B(mp
, del
->br_blockcount
);
4754 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
4755 xfs_mod_frextents(mp
, rtexts
);
4759 * Update the inode delalloc counter now and wait to update the
4760 * sb counters as we might have to borrow some blocks for the
4761 * indirect block accounting.
4763 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
,
4764 -((long)del
->br_blockcount
), 0,
4765 isrt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4768 ip
->i_delayed_blks
-= del
->br_blockcount
;
4770 if (got
->br_startoff
== del
->br_startoff
)
4771 state
|= BMAP_LEFT_FILLING
;
4772 if (got_endoff
== del_endoff
)
4773 state
|= BMAP_RIGHT_FILLING
;
4775 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
4776 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
4778 * Matches the whole extent. Delete the entry.
4780 xfs_iext_remove(ip
, icur
, state
);
4781 xfs_iext_prev(ifp
, icur
);
4783 case BMAP_LEFT_FILLING
:
4785 * Deleting the first part of the extent.
4787 got
->br_startoff
= del_endoff
;
4788 got
->br_blockcount
-= del
->br_blockcount
;
4789 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4790 got
->br_blockcount
), da_old
);
4791 got
->br_startblock
= nullstartblock((int)da_new
);
4792 xfs_iext_update_extent(ip
, state
, icur
, got
);
4794 case BMAP_RIGHT_FILLING
:
4796 * Deleting the last part of the extent.
4798 got
->br_blockcount
= got
->br_blockcount
- del
->br_blockcount
;
4799 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
,
4800 got
->br_blockcount
), da_old
);
4801 got
->br_startblock
= nullstartblock((int)da_new
);
4802 xfs_iext_update_extent(ip
, state
, icur
, got
);
4806 * Deleting the middle of the extent.
4808 * Distribute the original indlen reservation across the two new
4809 * extents. Steal blocks from the deleted extent if necessary.
4810 * Stealing blocks simply fudges the fdblocks accounting below.
4811 * Warn if either of the new indlen reservations is zero as this
4812 * can lead to delalloc problems.
4814 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4815 got_indlen
= xfs_bmap_worst_indlen(ip
, got
->br_blockcount
);
4817 new.br_blockcount
= got_endoff
- del_endoff
;
4818 new_indlen
= xfs_bmap_worst_indlen(ip
, new.br_blockcount
);
4820 WARN_ON_ONCE(!got_indlen
|| !new_indlen
);
4821 stolen
= xfs_bmap_split_indlen(da_old
, &got_indlen
, &new_indlen
,
4822 del
->br_blockcount
);
4824 got
->br_startblock
= nullstartblock((int)got_indlen
);
4826 new.br_startoff
= del_endoff
;
4827 new.br_state
= got
->br_state
;
4828 new.br_startblock
= nullstartblock((int)new_indlen
);
4830 xfs_iext_update_extent(ip
, state
, icur
, got
);
4831 xfs_iext_next(ifp
, icur
);
4832 xfs_iext_insert(ip
, icur
, &new, state
);
4834 da_new
= got_indlen
+ new_indlen
- stolen
;
4835 del
->br_blockcount
-= stolen
;
4839 ASSERT(da_old
>= da_new
);
4840 da_diff
= da_old
- da_new
;
4842 da_diff
+= del
->br_blockcount
;
4844 xfs_mod_fdblocks(mp
, da_diff
, false);
4849 xfs_bmap_del_extent_cow(
4850 struct xfs_inode
*ip
,
4851 struct xfs_iext_cursor
*icur
,
4852 struct xfs_bmbt_irec
*got
,
4853 struct xfs_bmbt_irec
*del
)
4855 struct xfs_mount
*mp
= ip
->i_mount
;
4856 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_COW_FORK
);
4857 struct xfs_bmbt_irec
new;
4858 xfs_fileoff_t del_endoff
, got_endoff
;
4859 int state
= BMAP_COWFORK
;
4861 XFS_STATS_INC(mp
, xs_del_exlist
);
4863 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4864 got_endoff
= got
->br_startoff
+ got
->br_blockcount
;
4866 ASSERT(del
->br_blockcount
> 0);
4867 ASSERT(got
->br_startoff
<= del
->br_startoff
);
4868 ASSERT(got_endoff
>= del_endoff
);
4869 ASSERT(!isnullstartblock(got
->br_startblock
));
4871 if (got
->br_startoff
== del
->br_startoff
)
4872 state
|= BMAP_LEFT_FILLING
;
4873 if (got_endoff
== del_endoff
)
4874 state
|= BMAP_RIGHT_FILLING
;
4876 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
4877 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
4879 * Matches the whole extent. Delete the entry.
4881 xfs_iext_remove(ip
, icur
, state
);
4882 xfs_iext_prev(ifp
, icur
);
4884 case BMAP_LEFT_FILLING
:
4886 * Deleting the first part of the extent.
4888 got
->br_startoff
= del_endoff
;
4889 got
->br_blockcount
-= del
->br_blockcount
;
4890 got
->br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
4891 xfs_iext_update_extent(ip
, state
, icur
, got
);
4893 case BMAP_RIGHT_FILLING
:
4895 * Deleting the last part of the extent.
4897 got
->br_blockcount
-= del
->br_blockcount
;
4898 xfs_iext_update_extent(ip
, state
, icur
, got
);
4902 * Deleting the middle of the extent.
4904 got
->br_blockcount
= del
->br_startoff
- got
->br_startoff
;
4906 new.br_startoff
= del_endoff
;
4907 new.br_blockcount
= got_endoff
- del_endoff
;
4908 new.br_state
= got
->br_state
;
4909 new.br_startblock
= del
->br_startblock
+ del
->br_blockcount
;
4911 xfs_iext_update_extent(ip
, state
, icur
, got
);
4912 xfs_iext_next(ifp
, icur
);
4913 xfs_iext_insert(ip
, icur
, &new, state
);
4916 ip
->i_delayed_blks
-= del
->br_blockcount
;
4920 * Called by xfs_bmapi to update file extent records and the btree
4921 * after removing space.
4923 STATIC
int /* error */
4924 xfs_bmap_del_extent_real(
4925 xfs_inode_t
*ip
, /* incore inode pointer */
4926 xfs_trans_t
*tp
, /* current transaction pointer */
4927 struct xfs_iext_cursor
*icur
,
4928 xfs_btree_cur_t
*cur
, /* if null, not a btree */
4929 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
4930 int *logflagsp
, /* inode logging flags */
4931 int whichfork
, /* data or attr fork */
4932 int bflags
) /* bmapi flags */
4934 xfs_fsblock_t del_endblock
=0; /* first block past del */
4935 xfs_fileoff_t del_endoff
; /* first offset past del */
4936 int do_fx
; /* free extent at end of routine */
4937 int error
; /* error return value */
4938 int flags
= 0;/* inode logging flags */
4939 struct xfs_bmbt_irec got
; /* current extent entry */
4940 xfs_fileoff_t got_endoff
; /* first offset past got */
4941 int i
; /* temp state */
4942 struct xfs_ifork
*ifp
; /* inode fork pointer */
4943 xfs_mount_t
*mp
; /* mount structure */
4944 xfs_filblks_t nblks
; /* quota/sb block count */
4945 xfs_bmbt_irec_t
new; /* new record to be inserted */
4947 uint qfield
; /* quota field to update */
4948 int state
= xfs_bmap_fork_to_state(whichfork
);
4949 struct xfs_bmbt_irec old
;
4952 XFS_STATS_INC(mp
, xs_del_exlist
);
4954 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4955 ASSERT(del
->br_blockcount
> 0);
4956 xfs_iext_get_extent(ifp
, icur
, &got
);
4957 ASSERT(got
.br_startoff
<= del
->br_startoff
);
4958 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
4959 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
4960 ASSERT(got_endoff
>= del_endoff
);
4961 ASSERT(!isnullstartblock(got
.br_startblock
));
4966 * If it's the case where the directory code is running with no block
4967 * reservation, and the deleted block is in the middle of its extent,
4968 * and the resulting insert of an extent would cause transformation to
4969 * btree format, then reject it. The calling code will then swap blocks
4970 * around instead. We have to do this now, rather than waiting for the
4971 * conversion to btree format, since the transaction will be dirty then.
4973 if (tp
->t_blk_res
== 0 &&
4974 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
4975 XFS_IFORK_NEXTENTS(ip
, whichfork
) >=
4976 XFS_IFORK_MAXEXT(ip
, whichfork
) &&
4977 del
->br_startoff
> got
.br_startoff
&& del_endoff
< got_endoff
)
4980 flags
= XFS_ILOG_CORE
;
4981 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
4986 bno
= div_u64_rem(del
->br_startblock
, mp
->m_sb
.sb_rextsize
,
4989 len
= div_u64_rem(del
->br_blockcount
, mp
->m_sb
.sb_rextsize
,
4993 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
4997 nblks
= len
* mp
->m_sb
.sb_rextsize
;
4998 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
5001 nblks
= del
->br_blockcount
;
5002 qfield
= XFS_TRANS_DQ_BCOUNT
;
5005 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
5007 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5010 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5013 if (got
.br_startoff
== del
->br_startoff
)
5014 state
|= BMAP_LEFT_FILLING
;
5015 if (got_endoff
== del_endoff
)
5016 state
|= BMAP_RIGHT_FILLING
;
5018 switch (state
& (BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
)) {
5019 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
5021 * Matches the whole extent. Delete the entry.
5023 xfs_iext_remove(ip
, icur
, state
);
5024 xfs_iext_prev(ifp
, icur
);
5025 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5026 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5027 flags
|= XFS_ILOG_CORE
;
5029 flags
|= xfs_ilog_fext(whichfork
);
5032 if ((error
= xfs_btree_delete(cur
, &i
)))
5034 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5036 case BMAP_LEFT_FILLING
:
5038 * Deleting the first part of the extent.
5040 got
.br_startoff
= del_endoff
;
5041 got
.br_startblock
= del_endblock
;
5042 got
.br_blockcount
-= del
->br_blockcount
;
5043 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5045 flags
|= xfs_ilog_fext(whichfork
);
5048 error
= xfs_bmbt_update(cur
, &got
);
5052 case BMAP_RIGHT_FILLING
:
5054 * Deleting the last part of the extent.
5056 got
.br_blockcount
-= del
->br_blockcount
;
5057 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5059 flags
|= xfs_ilog_fext(whichfork
);
5062 error
= xfs_bmbt_update(cur
, &got
);
5068 * Deleting the middle of the extent.
5072 got
.br_blockcount
= del
->br_startoff
- got
.br_startoff
;
5073 xfs_iext_update_extent(ip
, state
, icur
, &got
);
5075 new.br_startoff
= del_endoff
;
5076 new.br_blockcount
= got_endoff
- del_endoff
;
5077 new.br_state
= got
.br_state
;
5078 new.br_startblock
= del_endblock
;
5080 flags
|= XFS_ILOG_CORE
;
5082 error
= xfs_bmbt_update(cur
, &got
);
5085 error
= xfs_btree_increment(cur
, 0, &i
);
5088 cur
->bc_rec
.b
= new;
5089 error
= xfs_btree_insert(cur
, &i
);
5090 if (error
&& error
!= -ENOSPC
)
5093 * If get no-space back from btree insert, it tried a
5094 * split, and we have a zero block reservation. Fix up
5095 * our state and return the error.
5097 if (error
== -ENOSPC
) {
5099 * Reset the cursor, don't trust it after any
5102 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5105 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5107 * Update the btree record back
5108 * to the original value.
5110 error
= xfs_bmbt_update(cur
, &old
);
5114 * Reset the extent record back
5115 * to the original value.
5117 xfs_iext_update_extent(ip
, state
, icur
, &old
);
5122 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, done
);
5124 flags
|= xfs_ilog_fext(whichfork
);
5125 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5126 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
5127 xfs_iext_next(ifp
, icur
);
5128 xfs_iext_insert(ip
, icur
, &new, state
);
5132 /* remove reverse mapping */
5133 error
= xfs_rmap_unmap_extent(tp
, ip
, whichfork
, del
);
5138 * If we need to, add to list of extents to delete.
5140 if (do_fx
&& !(bflags
& XFS_BMAPI_REMAP
)) {
5141 if (xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
) {
5142 error
= xfs_refcount_decrease_extent(tp
, del
);
5146 __xfs_bmap_add_free(tp
, del
->br_startblock
,
5147 del
->br_blockcount
, NULL
,
5148 (bflags
& XFS_BMAPI_NODISCARD
) ||
5149 del
->br_state
== XFS_EXT_UNWRITTEN
);
5154 * Adjust inode # blocks in the file.
5157 ip
->i_d
.di_nblocks
-= nblks
;
5159 * Adjust quota data.
5161 if (qfield
&& !(bflags
& XFS_BMAPI_REMAP
))
5162 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
5170 * Unmap (remove) blocks from a file.
5171 * If nexts is nonzero then the number of extents to remove is limited to
5172 * that value. If not all extents in the block range can be removed then
5177 struct xfs_trans
*tp
, /* transaction pointer */
5178 struct xfs_inode
*ip
, /* incore inode */
5179 xfs_fileoff_t start
, /* first file offset deleted */
5180 xfs_filblks_t
*rlen
, /* i/o: amount remaining */
5181 int flags
, /* misc flags */
5182 xfs_extnum_t nexts
) /* number of extents max */
5184 struct xfs_btree_cur
*cur
; /* bmap btree cursor */
5185 struct xfs_bmbt_irec del
; /* extent being deleted */
5186 int error
; /* error return value */
5187 xfs_extnum_t extno
; /* extent number in list */
5188 struct xfs_bmbt_irec got
; /* current extent record */
5189 struct xfs_ifork
*ifp
; /* inode fork pointer */
5190 int isrt
; /* freeing in rt area */
5191 int logflags
; /* transaction logging flags */
5192 xfs_extlen_t mod
; /* rt extent offset */
5193 struct xfs_mount
*mp
; /* mount structure */
5194 int tmp_logflags
; /* partial logging flags */
5195 int wasdel
; /* was a delayed alloc extent */
5196 int whichfork
; /* data or attribute fork */
5198 xfs_filblks_t len
= *rlen
; /* length to unmap in file */
5199 xfs_fileoff_t max_len
;
5200 xfs_agnumber_t prev_agno
= NULLAGNUMBER
, agno
;
5202 struct xfs_iext_cursor icur
;
5205 trace_xfs_bunmap(ip
, start
, len
, flags
, _RET_IP_
);
5207 whichfork
= xfs_bmapi_whichfork(flags
);
5208 ASSERT(whichfork
!= XFS_COW_FORK
);
5209 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5211 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5212 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5213 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5215 return -EFSCORRUPTED
;
5218 if (XFS_FORCED_SHUTDOWN(mp
))
5221 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5226 * Guesstimate how many blocks we can unmap without running the risk of
5227 * blowing out the transaction with a mix of EFIs and reflink
5230 if (tp
&& xfs_is_reflink_inode(ip
) && whichfork
== XFS_DATA_FORK
)
5231 max_len
= min(len
, xfs_refcount_max_unmap(tp
->t_log_res
));
5235 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5236 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5238 if (xfs_iext_count(ifp
) == 0) {
5242 XFS_STATS_INC(mp
, xs_blk_unmap
);
5243 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5246 if (!xfs_iext_lookup_extent_before(ip
, ifp
, &end
, &icur
, &got
)) {
5253 if (ifp
->if_flags
& XFS_IFBROOT
) {
5254 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5255 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5256 cur
->bc_private
.b
.flags
= 0;
5262 * Synchronize by locking the bitmap inode.
5264 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTBITMAP
);
5265 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5266 xfs_ilock(mp
->m_rsumip
, XFS_ILOCK_EXCL
|XFS_ILOCK_RTSUM
);
5267 xfs_trans_ijoin(tp
, mp
->m_rsumip
, XFS_ILOCK_EXCL
);
5271 while (end
!= (xfs_fileoff_t
)-1 && end
>= start
&&
5272 (nexts
== 0 || extno
< nexts
) && max_len
> 0) {
5274 * Is the found extent after a hole in which end lives?
5275 * Just back up to the previous extent, if so.
5277 if (got
.br_startoff
> end
&&
5278 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5283 * Is the last block of this extent before the range
5284 * we're supposed to delete? If so, we're done.
5286 end
= XFS_FILEOFF_MIN(end
,
5287 got
.br_startoff
+ got
.br_blockcount
- 1);
5291 * Then deal with the (possibly delayed) allocated space
5295 wasdel
= isnullstartblock(del
.br_startblock
);
5298 * Make sure we don't touch multiple AGF headers out of order
5299 * in a single transaction, as that could cause AB-BA deadlocks.
5302 agno
= XFS_FSB_TO_AGNO(mp
, del
.br_startblock
);
5303 if (prev_agno
!= NULLAGNUMBER
&& prev_agno
> agno
)
5307 if (got
.br_startoff
< start
) {
5308 del
.br_startoff
= start
;
5309 del
.br_blockcount
-= start
- got
.br_startoff
;
5311 del
.br_startblock
+= start
- got
.br_startoff
;
5313 if (del
.br_startoff
+ del
.br_blockcount
> end
+ 1)
5314 del
.br_blockcount
= end
+ 1 - del
.br_startoff
;
5316 /* How much can we safely unmap? */
5317 if (max_len
< del
.br_blockcount
) {
5318 del
.br_startoff
+= del
.br_blockcount
- max_len
;
5320 del
.br_startblock
+= del
.br_blockcount
- max_len
;
5321 del
.br_blockcount
= max_len
;
5327 sum
= del
.br_startblock
+ del
.br_blockcount
;
5328 div_u64_rem(sum
, mp
->m_sb
.sb_rextsize
, &mod
);
5331 * Realtime extent not lined up at the end.
5332 * The extent could have been split into written
5333 * and unwritten pieces, or we could just be
5334 * unmapping part of it. But we can't really
5335 * get rid of part of a realtime extent.
5337 if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5339 * This piece is unwritten, or we're not
5340 * using unwritten extents. Skip over it.
5343 end
-= mod
> del
.br_blockcount
?
5344 del
.br_blockcount
: mod
;
5345 if (end
< got
.br_startoff
&&
5346 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5353 * It's written, turn it unwritten.
5354 * This is better than zeroing it.
5356 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5357 ASSERT(tp
->t_blk_res
> 0);
5359 * If this spans a realtime extent boundary,
5360 * chop it back to the start of the one we end at.
5362 if (del
.br_blockcount
> mod
) {
5363 del
.br_startoff
+= del
.br_blockcount
- mod
;
5364 del
.br_startblock
+= del
.br_blockcount
- mod
;
5365 del
.br_blockcount
= mod
;
5367 del
.br_state
= XFS_EXT_UNWRITTEN
;
5368 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5369 whichfork
, &icur
, &cur
, &del
,
5375 div_u64_rem(del
.br_startblock
, mp
->m_sb
.sb_rextsize
, &mod
);
5378 * Realtime extent is lined up at the end but not
5379 * at the front. We'll get rid of full extents if
5382 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5383 if (del
.br_blockcount
> mod
) {
5384 del
.br_blockcount
-= mod
;
5385 del
.br_startoff
+= mod
;
5386 del
.br_startblock
+= mod
;
5387 } else if (del
.br_startoff
== start
&&
5388 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5389 tp
->t_blk_res
== 0)) {
5391 * Can't make it unwritten. There isn't
5392 * a full extent here so just skip it.
5394 ASSERT(end
>= del
.br_blockcount
);
5395 end
-= del
.br_blockcount
;
5396 if (got
.br_startoff
> end
&&
5397 !xfs_iext_prev_extent(ifp
, &icur
, &got
)) {
5402 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5403 struct xfs_bmbt_irec prev
;
5406 * This one is already unwritten.
5407 * It must have a written left neighbor.
5408 * Unwrite the killed part of that one and
5411 if (!xfs_iext_prev_extent(ifp
, &icur
, &prev
))
5413 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5414 ASSERT(!isnullstartblock(prev
.br_startblock
));
5415 ASSERT(del
.br_startblock
==
5416 prev
.br_startblock
+ prev
.br_blockcount
);
5417 if (prev
.br_startoff
< start
) {
5418 mod
= start
- prev
.br_startoff
;
5419 prev
.br_blockcount
-= mod
;
5420 prev
.br_startblock
+= mod
;
5421 prev
.br_startoff
= start
;
5423 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5424 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5425 ip
, whichfork
, &icur
, &cur
,
5431 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5432 del
.br_state
= XFS_EXT_UNWRITTEN
;
5433 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5434 ip
, whichfork
, &icur
, &cur
,
5444 error
= xfs_bmap_del_extent_delay(ip
, whichfork
, &icur
,
5447 error
= xfs_bmap_del_extent_real(ip
, tp
, &icur
, cur
,
5448 &del
, &tmp_logflags
, whichfork
,
5450 logflags
|= tmp_logflags
;
5456 max_len
-= del
.br_blockcount
;
5457 end
= del
.br_startoff
- 1;
5460 * If not done go on to the next (previous) record.
5462 if (end
!= (xfs_fileoff_t
)-1 && end
>= start
) {
5463 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
5464 (got
.br_startoff
> end
&&
5465 !xfs_iext_prev_extent(ifp
, &icur
, &got
))) {
5472 if (done
|| end
== (xfs_fileoff_t
)-1 || end
< start
)
5475 *rlen
= end
- start
+ 1;
5478 * Convert to a btree if necessary.
5480 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5481 ASSERT(cur
== NULL
);
5482 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
5483 &tmp_logflags
, whichfork
);
5484 logflags
|= tmp_logflags
;
5486 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &logflags
,
5492 * Log everything. Do this after conversion, there's no point in
5493 * logging the extent records if we've converted to btree format.
5495 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5496 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5497 logflags
&= ~xfs_ilog_fext(whichfork
);
5498 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5499 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5500 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5502 * Log inode even in the error case, if the transaction
5503 * is dirty we'll need to shut down the filesystem.
5506 xfs_trans_log_inode(tp
, ip
, logflags
);
5509 cur
->bc_private
.b
.allocated
= 0;
5510 xfs_btree_del_cursor(cur
, error
);
5515 /* Unmap a range of a file. */
5519 struct xfs_inode
*ip
,
5528 error
= __xfs_bunmapi(tp
, ip
, bno
, &len
, flags
, nexts
);
5534 * Determine whether an extent shift can be accomplished by a merge with the
5535 * extent that precedes the target hole of the shift.
5539 struct xfs_bmbt_irec
*left
, /* preceding extent */
5540 struct xfs_bmbt_irec
*got
, /* current extent to shift */
5541 xfs_fileoff_t shift
) /* shift fsb */
5543 xfs_fileoff_t startoff
;
5545 startoff
= got
->br_startoff
- shift
;
5548 * The extent, once shifted, must be adjacent in-file and on-disk with
5549 * the preceding extent.
5551 if ((left
->br_startoff
+ left
->br_blockcount
!= startoff
) ||
5552 (left
->br_startblock
+ left
->br_blockcount
!= got
->br_startblock
) ||
5553 (left
->br_state
!= got
->br_state
) ||
5554 (left
->br_blockcount
+ got
->br_blockcount
> MAXEXTLEN
))
5561 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5562 * hole in the file. If an extent shift would result in the extent being fully
5563 * adjacent to the extent that currently precedes the hole, we can merge with
5564 * the preceding extent rather than do the shift.
5566 * This function assumes the caller has verified a shift-by-merge is possible
5567 * with the provided extents via xfs_bmse_can_merge().
5571 struct xfs_trans
*tp
,
5572 struct xfs_inode
*ip
,
5574 xfs_fileoff_t shift
, /* shift fsb */
5575 struct xfs_iext_cursor
*icur
,
5576 struct xfs_bmbt_irec
*got
, /* extent to shift */
5577 struct xfs_bmbt_irec
*left
, /* preceding extent */
5578 struct xfs_btree_cur
*cur
,
5579 int *logflags
) /* output */
5581 struct xfs_bmbt_irec
new;
5582 xfs_filblks_t blockcount
;
5584 struct xfs_mount
*mp
= ip
->i_mount
;
5586 blockcount
= left
->br_blockcount
+ got
->br_blockcount
;
5588 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5589 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
5590 ASSERT(xfs_bmse_can_merge(left
, got
, shift
));
5593 new.br_blockcount
= blockcount
;
5596 * Update the on-disk extent count, the btree if necessary and log the
5599 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5600 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
5601 *logflags
|= XFS_ILOG_CORE
;
5603 *logflags
|= XFS_ILOG_DEXT
;
5607 /* lookup and remove the extent to merge */
5608 error
= xfs_bmbt_lookup_eq(cur
, got
, &i
);
5611 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5613 error
= xfs_btree_delete(cur
, &i
);
5616 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5618 /* lookup and update size of the previous extent */
5619 error
= xfs_bmbt_lookup_eq(cur
, left
, &i
);
5622 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5624 error
= xfs_bmbt_update(cur
, &new);
5629 xfs_iext_remove(ip
, icur
, 0);
5630 xfs_iext_prev(XFS_IFORK_PTR(ip
, whichfork
), icur
);
5631 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5634 /* update reverse mapping. rmap functions merge the rmaps for us */
5635 error
= xfs_rmap_unmap_extent(tp
, ip
, whichfork
, got
);
5638 memcpy(&new, got
, sizeof(new));
5639 new.br_startoff
= left
->br_startoff
+ left
->br_blockcount
;
5640 return xfs_rmap_map_extent(tp
, ip
, whichfork
, &new);
5644 xfs_bmap_shift_update_extent(
5645 struct xfs_trans
*tp
,
5646 struct xfs_inode
*ip
,
5648 struct xfs_iext_cursor
*icur
,
5649 struct xfs_bmbt_irec
*got
,
5650 struct xfs_btree_cur
*cur
,
5652 xfs_fileoff_t startoff
)
5654 struct xfs_mount
*mp
= ip
->i_mount
;
5655 struct xfs_bmbt_irec prev
= *got
;
5658 *logflags
|= XFS_ILOG_CORE
;
5660 got
->br_startoff
= startoff
;
5663 error
= xfs_bmbt_lookup_eq(cur
, &prev
, &i
);
5666 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
5668 error
= xfs_bmbt_update(cur
, got
);
5672 *logflags
|= XFS_ILOG_DEXT
;
5675 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), icur
,
5678 /* update reverse mapping */
5679 error
= xfs_rmap_unmap_extent(tp
, ip
, whichfork
, &prev
);
5682 return xfs_rmap_map_extent(tp
, ip
, whichfork
, got
);
5686 xfs_bmap_collapse_extents(
5687 struct xfs_trans
*tp
,
5688 struct xfs_inode
*ip
,
5689 xfs_fileoff_t
*next_fsb
,
5690 xfs_fileoff_t offset_shift_fsb
,
5693 int whichfork
= XFS_DATA_FORK
;
5694 struct xfs_mount
*mp
= ip
->i_mount
;
5695 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5696 struct xfs_btree_cur
*cur
= NULL
;
5697 struct xfs_bmbt_irec got
, prev
;
5698 struct xfs_iext_cursor icur
;
5699 xfs_fileoff_t new_startoff
;
5703 if (unlikely(XFS_TEST_ERROR(
5704 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5705 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
5706 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
5707 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
5708 return -EFSCORRUPTED
;
5711 if (XFS_FORCED_SHUTDOWN(mp
))
5714 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
5716 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5717 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5722 if (ifp
->if_flags
& XFS_IFBROOT
) {
5723 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5724 cur
->bc_private
.b
.flags
= 0;
5727 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
5731 XFS_WANT_CORRUPTED_GOTO(mp
, !isnullstartblock(got
.br_startblock
),
5734 new_startoff
= got
.br_startoff
- offset_shift_fsb
;
5735 if (xfs_iext_peek_prev_extent(ifp
, &icur
, &prev
)) {
5736 if (new_startoff
< prev
.br_startoff
+ prev
.br_blockcount
) {
5741 if (xfs_bmse_can_merge(&prev
, &got
, offset_shift_fsb
)) {
5742 error
= xfs_bmse_merge(tp
, ip
, whichfork
,
5743 offset_shift_fsb
, &icur
, &got
, &prev
,
5750 if (got
.br_startoff
< offset_shift_fsb
) {
5756 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
5757 cur
, &logflags
, new_startoff
);
5762 if (!xfs_iext_next_extent(ifp
, &icur
, &got
)) {
5767 *next_fsb
= got
.br_startoff
;
5770 xfs_btree_del_cursor(cur
, error
);
5772 xfs_trans_log_inode(tp
, ip
, logflags
);
5776 /* Make sure we won't be right-shifting an extent past the maximum bound. */
5778 xfs_bmap_can_insert_extents(
5779 struct xfs_inode
*ip
,
5781 xfs_fileoff_t shift
)
5783 struct xfs_bmbt_irec got
;
5787 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
5789 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
5792 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
5793 error
= xfs_bmap_last_extent(NULL
, ip
, XFS_DATA_FORK
, &got
, &is_empty
);
5794 if (!error
&& !is_empty
&& got
.br_startoff
>= off
&&
5795 ((got
.br_startoff
+ shift
) & BMBT_STARTOFF_MASK
) < got
.br_startoff
)
5797 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
5803 xfs_bmap_insert_extents(
5804 struct xfs_trans
*tp
,
5805 struct xfs_inode
*ip
,
5806 xfs_fileoff_t
*next_fsb
,
5807 xfs_fileoff_t offset_shift_fsb
,
5809 xfs_fileoff_t stop_fsb
)
5811 int whichfork
= XFS_DATA_FORK
;
5812 struct xfs_mount
*mp
= ip
->i_mount
;
5813 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5814 struct xfs_btree_cur
*cur
= NULL
;
5815 struct xfs_bmbt_irec got
, next
;
5816 struct xfs_iext_cursor icur
;
5817 xfs_fileoff_t new_startoff
;
5821 if (unlikely(XFS_TEST_ERROR(
5822 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5823 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
5824 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
5825 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
5826 return -EFSCORRUPTED
;
5829 if (XFS_FORCED_SHUTDOWN(mp
))
5832 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
));
5834 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5835 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5840 if (ifp
->if_flags
& XFS_IFBROOT
) {
5841 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5842 cur
->bc_private
.b
.flags
= 0;
5845 if (*next_fsb
== NULLFSBLOCK
) {
5846 xfs_iext_last(ifp
, &icur
);
5847 if (!xfs_iext_get_extent(ifp
, &icur
, &got
) ||
5848 stop_fsb
> got
.br_startoff
) {
5853 if (!xfs_iext_lookup_extent(ip
, ifp
, *next_fsb
, &icur
, &got
)) {
5858 XFS_WANT_CORRUPTED_GOTO(mp
, !isnullstartblock(got
.br_startblock
),
5861 if (stop_fsb
>= got
.br_startoff
+ got
.br_blockcount
) {
5866 new_startoff
= got
.br_startoff
+ offset_shift_fsb
;
5867 if (xfs_iext_peek_next_extent(ifp
, &icur
, &next
)) {
5868 if (new_startoff
+ got
.br_blockcount
> next
.br_startoff
) {
5874 * Unlike a left shift (which involves a hole punch), a right
5875 * shift does not modify extent neighbors in any way. We should
5876 * never find mergeable extents in this scenario. Check anyways
5877 * and warn if we encounter two extents that could be one.
5879 if (xfs_bmse_can_merge(&got
, &next
, offset_shift_fsb
))
5883 error
= xfs_bmap_shift_update_extent(tp
, ip
, whichfork
, &icur
, &got
,
5884 cur
, &logflags
, new_startoff
);
5888 if (!xfs_iext_prev_extent(ifp
, &icur
, &got
) ||
5889 stop_fsb
>= got
.br_startoff
+ got
.br_blockcount
) {
5894 *next_fsb
= got
.br_startoff
;
5897 xfs_btree_del_cursor(cur
, error
);
5899 xfs_trans_log_inode(tp
, ip
, logflags
);
5904 * Splits an extent into two extents at split_fsb block such that it is the
5905 * first block of the current_ext. @ext is a target extent to be split.
5906 * @split_fsb is a block where the extents is split. If split_fsb lies in a
5907 * hole or the first block of extents, just return 0.
5910 xfs_bmap_split_extent_at(
5911 struct xfs_trans
*tp
,
5912 struct xfs_inode
*ip
,
5913 xfs_fileoff_t split_fsb
)
5915 int whichfork
= XFS_DATA_FORK
;
5916 struct xfs_btree_cur
*cur
= NULL
;
5917 struct xfs_bmbt_irec got
;
5918 struct xfs_bmbt_irec
new; /* split extent */
5919 struct xfs_mount
*mp
= ip
->i_mount
;
5920 struct xfs_ifork
*ifp
;
5921 xfs_fsblock_t gotblkcnt
; /* new block count for got */
5922 struct xfs_iext_cursor icur
;
5927 if (unlikely(XFS_TEST_ERROR(
5928 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5929 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
5930 mp
, XFS_ERRTAG_BMAPIFORMAT
))) {
5931 XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
5932 XFS_ERRLEVEL_LOW
, mp
);
5933 return -EFSCORRUPTED
;
5936 if (XFS_FORCED_SHUTDOWN(mp
))
5939 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5940 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5941 /* Read in all the extents */
5942 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5948 * If there are not extents, or split_fsb lies in a hole we are done.
5950 if (!xfs_iext_lookup_extent(ip
, ifp
, split_fsb
, &icur
, &got
) ||
5951 got
.br_startoff
>= split_fsb
)
5954 gotblkcnt
= split_fsb
- got
.br_startoff
;
5955 new.br_startoff
= split_fsb
;
5956 new.br_startblock
= got
.br_startblock
+ gotblkcnt
;
5957 new.br_blockcount
= got
.br_blockcount
- gotblkcnt
;
5958 new.br_state
= got
.br_state
;
5960 if (ifp
->if_flags
& XFS_IFBROOT
) {
5961 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5962 cur
->bc_private
.b
.flags
= 0;
5963 error
= xfs_bmbt_lookup_eq(cur
, &got
, &i
);
5966 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
5969 got
.br_blockcount
= gotblkcnt
;
5970 xfs_iext_update_extent(ip
, xfs_bmap_fork_to_state(whichfork
), &icur
,
5973 logflags
= XFS_ILOG_CORE
;
5975 error
= xfs_bmbt_update(cur
, &got
);
5979 logflags
|= XFS_ILOG_DEXT
;
5981 /* Add new extent */
5982 xfs_iext_next(ifp
, &icur
);
5983 xfs_iext_insert(ip
, &icur
, &new, 0);
5984 XFS_IFORK_NEXT_SET(ip
, whichfork
,
5985 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
5988 error
= xfs_bmbt_lookup_eq(cur
, &new, &i
);
5991 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, del_cursor
);
5992 error
= xfs_btree_insert(cur
, &i
);
5995 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, del_cursor
);
5999 * Convert to a btree if necessary.
6001 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
6002 int tmp_logflags
; /* partial log flag return val */
6004 ASSERT(cur
== NULL
);
6005 error
= xfs_bmap_extents_to_btree(tp
, ip
, &cur
, 0,
6006 &tmp_logflags
, whichfork
);
6007 logflags
|= tmp_logflags
;
6012 cur
->bc_private
.b
.allocated
= 0;
6013 xfs_btree_del_cursor(cur
, error
);
6017 xfs_trans_log_inode(tp
, ip
, logflags
);
6022 xfs_bmap_split_extent(
6023 struct xfs_inode
*ip
,
6024 xfs_fileoff_t split_fsb
)
6026 struct xfs_mount
*mp
= ip
->i_mount
;
6027 struct xfs_trans
*tp
;
6030 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
,
6031 XFS_DIOSTRAT_SPACE_RES(mp
, 0), 0, 0, &tp
);
6035 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
6036 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
6038 error
= xfs_bmap_split_extent_at(tp
, ip
, split_fsb
);
6042 return xfs_trans_commit(tp
);
6045 xfs_trans_cancel(tp
);
6049 /* Deferred mapping is only for real extents in the data fork. */
6051 xfs_bmap_is_update_needed(
6052 struct xfs_bmbt_irec
*bmap
)
6054 return bmap
->br_startblock
!= HOLESTARTBLOCK
&&
6055 bmap
->br_startblock
!= DELAYSTARTBLOCK
;
6058 /* Record a bmap intent. */
6061 struct xfs_trans
*tp
,
6062 enum xfs_bmap_intent_type type
,
6063 struct xfs_inode
*ip
,
6065 struct xfs_bmbt_irec
*bmap
)
6067 struct xfs_bmap_intent
*bi
;
6069 trace_xfs_bmap_defer(tp
->t_mountp
,
6070 XFS_FSB_TO_AGNO(tp
->t_mountp
, bmap
->br_startblock
),
6072 XFS_FSB_TO_AGBNO(tp
->t_mountp
, bmap
->br_startblock
),
6073 ip
->i_ino
, whichfork
,
6075 bmap
->br_blockcount
,
6078 bi
= kmem_alloc(sizeof(struct xfs_bmap_intent
), KM_SLEEP
| KM_NOFS
);
6079 INIT_LIST_HEAD(&bi
->bi_list
);
6082 bi
->bi_whichfork
= whichfork
;
6083 bi
->bi_bmap
= *bmap
;
6085 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_BMAP
, &bi
->bi_list
);
6089 /* Map an extent into a file. */
6091 xfs_bmap_map_extent(
6092 struct xfs_trans
*tp
,
6093 struct xfs_inode
*ip
,
6094 struct xfs_bmbt_irec
*PREV
)
6096 if (!xfs_bmap_is_update_needed(PREV
))
6099 return __xfs_bmap_add(tp
, XFS_BMAP_MAP
, ip
, XFS_DATA_FORK
, PREV
);
6102 /* Unmap an extent out of a file. */
6104 xfs_bmap_unmap_extent(
6105 struct xfs_trans
*tp
,
6106 struct xfs_inode
*ip
,
6107 struct xfs_bmbt_irec
*PREV
)
6109 if (!xfs_bmap_is_update_needed(PREV
))
6112 return __xfs_bmap_add(tp
, XFS_BMAP_UNMAP
, ip
, XFS_DATA_FORK
, PREV
);
6116 * Process one of the deferred bmap operations. We pass back the
6117 * btree cursor to maintain our lock on the bmapbt between calls.
6120 xfs_bmap_finish_one(
6121 struct xfs_trans
*tp
,
6122 struct xfs_inode
*ip
,
6123 enum xfs_bmap_intent_type type
,
6125 xfs_fileoff_t startoff
,
6126 xfs_fsblock_t startblock
,
6127 xfs_filblks_t
*blockcount
,
6132 ASSERT(tp
->t_firstblock
== NULLFSBLOCK
);
6134 trace_xfs_bmap_deferred(tp
->t_mountp
,
6135 XFS_FSB_TO_AGNO(tp
->t_mountp
, startblock
), type
,
6136 XFS_FSB_TO_AGBNO(tp
->t_mountp
, startblock
),
6137 ip
->i_ino
, whichfork
, startoff
, *blockcount
, state
);
6139 if (WARN_ON_ONCE(whichfork
!= XFS_DATA_FORK
))
6140 return -EFSCORRUPTED
;
6142 if (XFS_TEST_ERROR(false, tp
->t_mountp
,
6143 XFS_ERRTAG_BMAP_FINISH_ONE
))
6148 error
= xfs_bmapi_remap(tp
, ip
, startoff
, *blockcount
,
6152 case XFS_BMAP_UNMAP
:
6153 error
= __xfs_bunmapi(tp
, ip
, startoff
, blockcount
,
6154 XFS_BMAPI_REMAP
, 1);
6158 error
= -EFSCORRUPTED
;
6164 /* Check that an inode's extent does not have invalid flags or bad ranges. */
6166 xfs_bmap_validate_extent(
6167 struct xfs_inode
*ip
,
6169 struct xfs_bmbt_irec
*irec
)
6171 struct xfs_mount
*mp
= ip
->i_mount
;
6172 xfs_fsblock_t endfsb
;
6175 isrt
= XFS_IS_REALTIME_INODE(ip
);
6176 endfsb
= irec
->br_startblock
+ irec
->br_blockcount
- 1;
6178 if (!xfs_verify_rtbno(mp
, irec
->br_startblock
))
6179 return __this_address
;
6180 if (!xfs_verify_rtbno(mp
, endfsb
))
6181 return __this_address
;
6183 if (!xfs_verify_fsbno(mp
, irec
->br_startblock
))
6184 return __this_address
;
6185 if (!xfs_verify_fsbno(mp
, endfsb
))
6186 return __this_address
;
6187 if (XFS_FSB_TO_AGNO(mp
, irec
->br_startblock
) !=
6188 XFS_FSB_TO_AGNO(mp
, endfsb
))
6189 return __this_address
;
6191 if (irec
->br_state
!= XFS_EXT_NORM
&& whichfork
!= XFS_DATA_FORK
)
6192 return __this_address
;