2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_mount.h"
36 #include "xfs_itable.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_extfree_item.h"
39 #include "xfs_alloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_quota.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
52 kmem_zone_t
*xfs_bmap_free_item_zone
;
55 * Prototypes for internal bmap routines.
60 xfs_bmap_check_leaf_extents(
61 struct xfs_btree_cur
*cur
,
65 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
70 * Called from xfs_bmap_add_attrfork to handle extents format files.
72 STATIC
int /* error */
73 xfs_bmap_add_attrfork_extents(
74 xfs_trans_t
*tp
, /* transaction pointer */
75 xfs_inode_t
*ip
, /* incore inode pointer */
76 xfs_fsblock_t
*firstblock
, /* first block allocated */
77 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
78 int *flags
); /* inode logging flags */
81 * Called from xfs_bmap_add_attrfork to handle local format files.
83 STATIC
int /* error */
84 xfs_bmap_add_attrfork_local(
85 xfs_trans_t
*tp
, /* transaction pointer */
86 xfs_inode_t
*ip
, /* incore inode pointer */
87 xfs_fsblock_t
*firstblock
, /* first block allocated */
88 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
89 int *flags
); /* inode logging flags */
92 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
93 * It figures out where to ask the underlying allocator to put the new extent.
95 STATIC
int /* error */
97 xfs_bmalloca_t
*ap
); /* bmap alloc argument struct */
100 * Transform a btree format file with only one leaf node, where the
101 * extents list will fit in the inode, into an extents format file.
102 * Since the file extents are already in-core, all we have to do is
103 * give up the space for the btree root and pitch the leaf block.
105 STATIC
int /* error */
106 xfs_bmap_btree_to_extents(
107 xfs_trans_t
*tp
, /* transaction pointer */
108 xfs_inode_t
*ip
, /* incore inode pointer */
109 xfs_btree_cur_t
*cur
, /* btree cursor */
110 int *logflagsp
, /* inode logging flags */
111 int whichfork
); /* data or attr fork */
114 * Remove the entry "free" from the free item list. Prev points to the
115 * previous entry, unless "free" is the head of the list.
119 xfs_bmap_free_t
*flist
, /* free item list header */
120 xfs_bmap_free_item_t
*prev
, /* previous item on list, if any */
121 xfs_bmap_free_item_t
*free
); /* list item to be freed */
124 * Convert an extents-format file into a btree-format file.
125 * The new file will have a root block (in the inode) and a single child block.
127 STATIC
int /* error */
128 xfs_bmap_extents_to_btree(
129 xfs_trans_t
*tp
, /* transaction pointer */
130 xfs_inode_t
*ip
, /* incore inode pointer */
131 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
132 xfs_bmap_free_t
*flist
, /* blocks freed in xaction */
133 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
134 int wasdel
, /* converting a delayed alloc */
135 int *logflagsp
, /* inode logging flags */
136 int whichfork
); /* data or attr fork */
139 * Convert a local file to an extents file.
140 * This code is sort of bogus, since the file data needs to get
141 * logged so it won't be lost. The bmap-level manipulations are ok, though.
143 STATIC
int /* error */
144 xfs_bmap_local_to_extents(
145 xfs_trans_t
*tp
, /* transaction pointer */
146 xfs_inode_t
*ip
, /* incore inode pointer */
147 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
148 xfs_extlen_t total
, /* total blocks needed by transaction */
149 int *logflagsp
, /* inode logging flags */
150 int whichfork
, /* data or attr fork */
151 void (*init_fn
)(struct xfs_buf
*bp
,
152 struct xfs_inode
*ip
,
153 struct xfs_ifork
*ifp
));
156 * Search the extents list for the inode, for the extent containing bno.
157 * If bno lies in a hole, point to the next entry. If bno lies past eof,
158 * *eofp will be set, and *prevp will contain the last entry (null if none).
159 * Else, *lastxp will be set to the index of the found
160 * entry; *gotp will contain the entry.
162 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
163 xfs_bmap_search_extents(
164 xfs_inode_t
*ip
, /* incore inode pointer */
165 xfs_fileoff_t bno
, /* block number searched for */
166 int whichfork
, /* data or attr fork */
167 int *eofp
, /* out: end of file found */
168 xfs_extnum_t
*lastxp
, /* out: last extent index */
169 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
170 xfs_bmbt_irec_t
*prevp
); /* out: previous extent entry found */
173 * Compute the worst-case number of indirect blocks that will be used
174 * for ip's delayed extent of length "len".
177 xfs_bmap_worst_indlen(
178 xfs_inode_t
*ip
, /* incore inode pointer */
179 xfs_filblks_t len
); /* delayed extent length */
183 * Perform various validation checks on the values being returned
187 xfs_bmap_validate_ret(
191 xfs_bmbt_irec_t
*mval
,
195 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
203 xfs_fsblock_t blockno
,
208 xfs_bmap_count_leaves(
215 xfs_bmap_disk_count_leaves(
216 struct xfs_mount
*mp
,
217 struct xfs_btree_block
*block
,
222 * Bmap internal routines.
225 STATIC
int /* error */
227 struct xfs_btree_cur
*cur
,
231 int *stat
) /* success/failure */
233 cur
->bc_rec
.b
.br_startoff
= off
;
234 cur
->bc_rec
.b
.br_startblock
= bno
;
235 cur
->bc_rec
.b
.br_blockcount
= len
;
236 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
239 STATIC
int /* error */
241 struct xfs_btree_cur
*cur
,
245 int *stat
) /* success/failure */
247 cur
->bc_rec
.b
.br_startoff
= off
;
248 cur
->bc_rec
.b
.br_startblock
= bno
;
249 cur
->bc_rec
.b
.br_blockcount
= len
;
250 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
254 * Check if the inode needs to be converted to btree format.
256 static inline bool xfs_bmap_needs_btree(struct xfs_inode
*ip
, int whichfork
)
258 return XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
259 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
260 XFS_IFORK_MAXEXT(ip
, whichfork
);
264 * Check if the inode should be converted to extent format.
266 static inline bool xfs_bmap_wants_extents(struct xfs_inode
*ip
, int whichfork
)
268 return XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
269 XFS_IFORK_NEXTENTS(ip
, whichfork
) <=
270 XFS_IFORK_MAXEXT(ip
, whichfork
);
274 * Update the record referred to by cur to the value given
275 * by [off, bno, len, state].
276 * This either works (return 0) or gets an EFSCORRUPTED error.
280 struct xfs_btree_cur
*cur
,
286 union xfs_btree_rec rec
;
288 xfs_bmbt_disk_set_allf(&rec
.bmbt
, off
, bno
, len
, state
);
289 return xfs_btree_update(cur
, &rec
);
293 * Called from xfs_bmap_add_attrfork to handle btree format files.
295 STATIC
int /* error */
296 xfs_bmap_add_attrfork_btree(
297 xfs_trans_t
*tp
, /* transaction pointer */
298 xfs_inode_t
*ip
, /* incore inode pointer */
299 xfs_fsblock_t
*firstblock
, /* first block allocated */
300 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
301 int *flags
) /* inode logging flags */
303 xfs_btree_cur_t
*cur
; /* btree cursor */
304 int error
; /* error return value */
305 xfs_mount_t
*mp
; /* file system mount struct */
306 int stat
; /* newroot status */
309 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
310 *flags
|= XFS_ILOG_DBROOT
;
312 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
313 cur
->bc_private
.b
.flist
= flist
;
314 cur
->bc_private
.b
.firstblock
= *firstblock
;
315 if ((error
= xfs_bmbt_lookup_ge(cur
, 0, 0, 0, &stat
)))
317 /* must be at least one entry */
318 XFS_WANT_CORRUPTED_GOTO(stat
== 1, error0
);
319 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
322 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
323 return XFS_ERROR(ENOSPC
);
325 *firstblock
= cur
->bc_private
.b
.firstblock
;
326 cur
->bc_private
.b
.allocated
= 0;
327 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
331 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
336 * Called from xfs_bmap_add_attrfork to handle extents format files.
338 STATIC
int /* error */
339 xfs_bmap_add_attrfork_extents(
340 xfs_trans_t
*tp
, /* transaction pointer */
341 xfs_inode_t
*ip
, /* incore inode pointer */
342 xfs_fsblock_t
*firstblock
, /* first block allocated */
343 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
344 int *flags
) /* inode logging flags */
346 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
347 int error
; /* error return value */
349 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
352 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
, &cur
, 0,
353 flags
, XFS_DATA_FORK
);
355 cur
->bc_private
.b
.allocated
= 0;
356 xfs_btree_del_cursor(cur
,
357 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
363 * Block initialisation functions for local to extent format conversion.
364 * As these get more complex, they will be moved to the relevant files,
365 * but for now they are too simple to worry about.
368 xfs_bmap_local_to_extents_init_fn(
370 struct xfs_inode
*ip
,
371 struct xfs_ifork
*ifp
)
373 bp
->b_ops
= &xfs_bmbt_buf_ops
;
374 memcpy(bp
->b_addr
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
378 xfs_symlink_local_to_remote(
380 struct xfs_inode
*ip
,
381 struct xfs_ifork
*ifp
)
383 /* remote symlink blocks are not verifiable until CRCs come along */
385 memcpy(bp
->b_addr
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
389 * Called from xfs_bmap_add_attrfork to handle local format files. Each
390 * different data fork content type needs a different callout to do the
391 * conversion. Some are basic and only require special block initialisation
392 * callouts for the data formating, others (directories) are so specialised they
393 * handle everything themselves.
395 * XXX (dgc): investigate whether directory conversion can use the generic
396 * formatting callout. It should be possible - it's just a very complex
397 * formatter. it would also require passing the transaction through to the init
400 STATIC
int /* error */
401 xfs_bmap_add_attrfork_local(
402 xfs_trans_t
*tp
, /* transaction pointer */
403 xfs_inode_t
*ip
, /* incore inode pointer */
404 xfs_fsblock_t
*firstblock
, /* first block allocated */
405 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
406 int *flags
) /* inode logging flags */
408 xfs_da_args_t dargs
; /* args for dir/attr code */
410 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
413 if (S_ISDIR(ip
->i_d
.di_mode
)) {
414 memset(&dargs
, 0, sizeof(dargs
));
416 dargs
.firstblock
= firstblock
;
418 dargs
.total
= ip
->i_mount
->m_dirblkfsbs
;
419 dargs
.whichfork
= XFS_DATA_FORK
;
421 return xfs_dir2_sf_to_block(&dargs
);
424 if (S_ISLNK(ip
->i_d
.di_mode
))
425 return xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1,
426 flags
, XFS_DATA_FORK
,
427 xfs_symlink_local_to_remote
);
429 return xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1, flags
,
431 xfs_bmap_local_to_extents_init_fn
);
435 * Convert a delayed allocation to a real allocation.
437 STATIC
int /* error */
438 xfs_bmap_add_extent_delay_real(
439 struct xfs_bmalloca
*bma
)
441 struct xfs_bmbt_irec
*new = &bma
->got
;
442 int diff
; /* temp value */
443 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
444 int error
; /* error return value */
445 int i
; /* temp state */
446 xfs_ifork_t
*ifp
; /* inode fork pointer */
447 xfs_fileoff_t new_endoff
; /* end offset of new entry */
448 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
449 /* left is 0, right is 1, prev is 2 */
450 int rval
=0; /* return value (logging flags) */
451 int state
= 0;/* state bits, accessed thru macros */
452 xfs_filblks_t da_new
; /* new count del alloc blocks used */
453 xfs_filblks_t da_old
; /* old count del alloc blocks used */
454 xfs_filblks_t temp
=0; /* value for da_new calculations */
455 xfs_filblks_t temp2
=0;/* value for da_new calculations */
456 int tmp_rval
; /* partial logging flags */
458 ifp
= XFS_IFORK_PTR(bma
->ip
, XFS_DATA_FORK
);
460 ASSERT(bma
->idx
>= 0);
461 ASSERT(bma
->idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
462 ASSERT(!isnullstartblock(new->br_startblock
));
464 (bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
466 XFS_STATS_INC(xs_add_exlist
);
473 * Set up a bunch of variables to make the tests simpler.
475 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
476 xfs_bmbt_get_all(ep
, &PREV
);
477 new_endoff
= new->br_startoff
+ new->br_blockcount
;
478 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
479 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
481 da_old
= startblockval(PREV
.br_startblock
);
485 * Set flags determining what part of the previous delayed allocation
486 * extent is being replaced by a real allocation.
488 if (PREV
.br_startoff
== new->br_startoff
)
489 state
|= BMAP_LEFT_FILLING
;
490 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
491 state
|= BMAP_RIGHT_FILLING
;
494 * Check and set flags if this segment has a left neighbor.
495 * Don't set contiguous if the combined extent would be too large.
498 state
|= BMAP_LEFT_VALID
;
499 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &LEFT
);
501 if (isnullstartblock(LEFT
.br_startblock
))
502 state
|= BMAP_LEFT_DELAY
;
505 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
506 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
507 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
508 LEFT
.br_state
== new->br_state
&&
509 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
510 state
|= BMAP_LEFT_CONTIG
;
513 * Check and set flags if this segment has a right neighbor.
514 * Don't set contiguous if the combined extent would be too large.
515 * Also check for all-three-contiguous being too large.
517 if (bma
->idx
< bma
->ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
518 state
|= BMAP_RIGHT_VALID
;
519 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
+ 1), &RIGHT
);
521 if (isnullstartblock(RIGHT
.br_startblock
))
522 state
|= BMAP_RIGHT_DELAY
;
525 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
526 new_endoff
== RIGHT
.br_startoff
&&
527 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
528 new->br_state
== RIGHT
.br_state
&&
529 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
530 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
531 BMAP_RIGHT_FILLING
)) !=
532 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
533 BMAP_RIGHT_FILLING
) ||
534 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
536 state
|= BMAP_RIGHT_CONTIG
;
540 * Switch out based on the FILLING and CONTIG state bits.
542 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
543 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
544 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
545 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
547 * Filling in all of a previously delayed allocation extent.
548 * The left and right neighbors are both contiguous with new.
551 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
552 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
553 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
554 RIGHT
.br_blockcount
);
555 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
557 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 2, state
);
558 bma
->ip
->i_d
.di_nextents
--;
559 if (bma
->cur
== NULL
)
560 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
562 rval
= XFS_ILOG_CORE
;
563 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
565 RIGHT
.br_blockcount
, &i
);
568 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
569 error
= xfs_btree_delete(bma
->cur
, &i
);
572 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
573 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
576 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
577 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
581 RIGHT
.br_blockcount
, LEFT
.br_state
);
587 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
589 * Filling in all of a previously delayed allocation extent.
590 * The left neighbor is contiguous, the right is not.
594 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
595 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
596 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
597 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
599 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
600 if (bma
->cur
== NULL
)
601 rval
= XFS_ILOG_DEXT
;
604 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
605 LEFT
.br_startblock
, LEFT
.br_blockcount
,
609 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
610 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
613 PREV
.br_blockcount
, LEFT
.br_state
);
619 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
621 * Filling in all of a previously delayed allocation extent.
622 * The right neighbor is contiguous, the left is not.
624 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
625 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
626 xfs_bmbt_set_blockcount(ep
,
627 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
628 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
630 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
631 if (bma
->cur
== NULL
)
632 rval
= XFS_ILOG_DEXT
;
635 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
637 RIGHT
.br_blockcount
, &i
);
640 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
641 error
= xfs_bmbt_update(bma
->cur
, PREV
.br_startoff
,
644 RIGHT
.br_blockcount
, PREV
.br_state
);
650 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
652 * Filling in all of a previously delayed allocation extent.
653 * Neither the left nor right neighbors are contiguous with
656 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
657 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
658 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
660 bma
->ip
->i_d
.di_nextents
++;
661 if (bma
->cur
== NULL
)
662 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
664 rval
= XFS_ILOG_CORE
;
665 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
666 new->br_startblock
, new->br_blockcount
,
670 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
671 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
672 error
= xfs_btree_insert(bma
->cur
, &i
);
675 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
679 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
681 * Filling in the first part of a previous delayed allocation.
682 * The left neighbor is contiguous.
684 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
685 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
686 LEFT
.br_blockcount
+ new->br_blockcount
);
687 xfs_bmbt_set_startoff(ep
,
688 PREV
.br_startoff
+ new->br_blockcount
);
689 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
- 1, state
, _THIS_IP_
);
691 temp
= PREV
.br_blockcount
- new->br_blockcount
;
692 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
693 xfs_bmbt_set_blockcount(ep
, temp
);
694 if (bma
->cur
== NULL
)
695 rval
= XFS_ILOG_DEXT
;
698 error
= xfs_bmbt_lookup_eq(bma
->cur
, LEFT
.br_startoff
,
699 LEFT
.br_startblock
, LEFT
.br_blockcount
,
703 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
704 error
= xfs_bmbt_update(bma
->cur
, LEFT
.br_startoff
,
712 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
713 startblockval(PREV
.br_startblock
));
714 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
715 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
720 case BMAP_LEFT_FILLING
:
722 * Filling in the first part of a previous delayed allocation.
723 * The left neighbor is not contiguous.
725 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
726 xfs_bmbt_set_startoff(ep
, new_endoff
);
727 temp
= PREV
.br_blockcount
- new->br_blockcount
;
728 xfs_bmbt_set_blockcount(ep
, temp
);
729 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
730 bma
->ip
->i_d
.di_nextents
++;
731 if (bma
->cur
== NULL
)
732 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
734 rval
= XFS_ILOG_CORE
;
735 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
736 new->br_startblock
, new->br_blockcount
,
740 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
741 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
742 error
= xfs_btree_insert(bma
->cur
, &i
);
745 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
748 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
749 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
750 bma
->firstblock
, bma
->flist
,
751 &bma
->cur
, 1, &tmp_rval
, XFS_DATA_FORK
);
756 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
757 startblockval(PREV
.br_startblock
) -
758 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
759 ep
= xfs_iext_get_ext(ifp
, bma
->idx
+ 1);
760 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
761 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
764 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
766 * Filling in the last part of a previous delayed allocation.
767 * The right neighbor is contiguous with the new allocation.
769 temp
= PREV
.br_blockcount
- new->br_blockcount
;
770 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
771 xfs_bmbt_set_blockcount(ep
, temp
);
772 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
+ 1),
773 new->br_startoff
, new->br_startblock
,
774 new->br_blockcount
+ RIGHT
.br_blockcount
,
776 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 1, state
, _THIS_IP_
);
777 if (bma
->cur
== NULL
)
778 rval
= XFS_ILOG_DEXT
;
781 error
= xfs_bmbt_lookup_eq(bma
->cur
, RIGHT
.br_startoff
,
783 RIGHT
.br_blockcount
, &i
);
786 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
787 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
796 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
797 startblockval(PREV
.br_startblock
));
798 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
799 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
800 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
805 case BMAP_RIGHT_FILLING
:
807 * Filling in the last part of a previous delayed allocation.
808 * The right neighbor is not contiguous.
810 temp
= PREV
.br_blockcount
- new->br_blockcount
;
811 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
812 xfs_bmbt_set_blockcount(ep
, temp
);
813 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 1, new, state
);
814 bma
->ip
->i_d
.di_nextents
++;
815 if (bma
->cur
== NULL
)
816 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
818 rval
= XFS_ILOG_CORE
;
819 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
820 new->br_startblock
, new->br_blockcount
,
824 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
825 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
826 error
= xfs_btree_insert(bma
->cur
, &i
);
829 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
832 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
833 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
834 bma
->firstblock
, bma
->flist
, &bma
->cur
, 1,
835 &tmp_rval
, XFS_DATA_FORK
);
840 da_new
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma
->ip
, temp
),
841 startblockval(PREV
.br_startblock
) -
842 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
843 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
844 xfs_bmbt_set_startblock(ep
, nullstartblock(da_new
));
845 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
852 * Filling in the middle part of a previous delayed allocation.
853 * Contiguity is impossible here.
854 * This case is avoided almost all the time.
856 * We start with a delayed allocation:
858 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
861 * and we are allocating:
862 * +rrrrrrrrrrrrrrrrr+
865 * and we set it up for insertion as:
866 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
868 * PREV @ idx LEFT RIGHT
869 * inserted at idx + 1
871 temp
= new->br_startoff
- PREV
.br_startoff
;
872 temp2
= PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
873 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, 0, _THIS_IP_
);
874 xfs_bmbt_set_blockcount(ep
, temp
); /* truncate PREV */
876 RIGHT
.br_state
= PREV
.br_state
;
877 RIGHT
.br_startblock
= nullstartblock(
878 (int)xfs_bmap_worst_indlen(bma
->ip
, temp2
));
879 RIGHT
.br_startoff
= new_endoff
;
880 RIGHT
.br_blockcount
= temp2
;
881 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
882 xfs_iext_insert(bma
->ip
, bma
->idx
+ 1, 2, &LEFT
, state
);
883 bma
->ip
->i_d
.di_nextents
++;
884 if (bma
->cur
== NULL
)
885 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
887 rval
= XFS_ILOG_CORE
;
888 error
= xfs_bmbt_lookup_eq(bma
->cur
, new->br_startoff
,
889 new->br_startblock
, new->br_blockcount
,
893 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
894 bma
->cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
895 error
= xfs_btree_insert(bma
->cur
, &i
);
898 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
901 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
902 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
903 bma
->firstblock
, bma
->flist
, &bma
->cur
,
904 1, &tmp_rval
, XFS_DATA_FORK
);
909 temp
= xfs_bmap_worst_indlen(bma
->ip
, temp
);
910 temp2
= xfs_bmap_worst_indlen(bma
->ip
, temp2
);
911 diff
= (int)(temp
+ temp2
- startblockval(PREV
.br_startblock
) -
912 (bma
->cur
? bma
->cur
->bc_private
.b
.allocated
: 0));
914 error
= xfs_icsb_modify_counters(bma
->ip
->i_mount
,
916 -((int64_t)diff
), 0);
922 ep
= xfs_iext_get_ext(ifp
, bma
->idx
);
923 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
924 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
925 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
926 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, bma
->idx
+ 2),
927 nullstartblock((int)temp2
));
928 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
+ 2, state
, _THIS_IP_
);
931 da_new
= temp
+ temp2
;
934 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
935 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
936 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
937 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
938 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
939 case BMAP_LEFT_CONTIG
:
940 case BMAP_RIGHT_CONTIG
:
942 * These cases are all impossible.
947 /* convert to a btree if necessary */
948 if (xfs_bmap_needs_btree(bma
->ip
, XFS_DATA_FORK
)) {
949 int tmp_logflags
; /* partial log flag return val */
951 ASSERT(bma
->cur
== NULL
);
952 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
953 bma
->firstblock
, bma
->flist
, &bma
->cur
,
954 da_old
> 0, &tmp_logflags
, XFS_DATA_FORK
);
955 bma
->logflags
|= tmp_logflags
;
960 /* adjust for changes in reserved delayed indirect blocks */
961 if (da_old
|| da_new
) {
964 temp
+= bma
->cur
->bc_private
.b
.allocated
;
965 ASSERT(temp
<= da_old
);
967 xfs_icsb_modify_counters(bma
->ip
->i_mount
,
969 (int64_t)(da_old
- temp
), 0);
972 /* clear out the allocated field, done with it now in any case. */
974 bma
->cur
->bc_private
.b
.allocated
= 0;
976 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, XFS_DATA_FORK
);
978 bma
->logflags
|= rval
;
986 * Convert an unwritten allocation to a real allocation or vice versa.
988 STATIC
int /* error */
989 xfs_bmap_add_extent_unwritten_real(
990 struct xfs_trans
*tp
,
991 xfs_inode_t
*ip
, /* incore inode pointer */
992 xfs_extnum_t
*idx
, /* extent number to update/insert */
993 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
994 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
995 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
996 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
997 int *logflagsp
) /* inode logging flags */
999 xfs_btree_cur_t
*cur
; /* btree cursor */
1000 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
1001 int error
; /* error return value */
1002 int i
; /* temp state */
1003 xfs_ifork_t
*ifp
; /* inode fork pointer */
1004 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1005 xfs_exntst_t newext
; /* new extent state */
1006 xfs_exntst_t oldext
; /* old extent state */
1007 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1008 /* left is 0, right is 1, prev is 2 */
1009 int rval
=0; /* return value (logging flags) */
1010 int state
= 0;/* state bits, accessed thru macros */
1015 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1018 ASSERT(*idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
1019 ASSERT(!isnullstartblock(new->br_startblock
));
1021 XFS_STATS_INC(xs_add_exlist
);
1028 * Set up a bunch of variables to make the tests simpler.
1031 ep
= xfs_iext_get_ext(ifp
, *idx
);
1032 xfs_bmbt_get_all(ep
, &PREV
);
1033 newext
= new->br_state
;
1034 oldext
= (newext
== XFS_EXT_UNWRITTEN
) ?
1035 XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
1036 ASSERT(PREV
.br_state
== oldext
);
1037 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1038 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1039 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1042 * Set flags determining what part of the previous oldext allocation
1043 * extent is being replaced by a newext allocation.
1045 if (PREV
.br_startoff
== new->br_startoff
)
1046 state
|= BMAP_LEFT_FILLING
;
1047 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1048 state
|= BMAP_RIGHT_FILLING
;
1051 * Check and set flags if this segment has a left neighbor.
1052 * Don't set contiguous if the combined extent would be too large.
1055 state
|= BMAP_LEFT_VALID
;
1056 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
1058 if (isnullstartblock(LEFT
.br_startblock
))
1059 state
|= BMAP_LEFT_DELAY
;
1062 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1063 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1064 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1065 LEFT
.br_state
== newext
&&
1066 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1067 state
|= BMAP_LEFT_CONTIG
;
1070 * Check and set flags if this segment has a right neighbor.
1071 * Don't set contiguous if the combined extent would be too large.
1072 * Also check for all-three-contiguous being too large.
1074 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
1075 state
|= BMAP_RIGHT_VALID
;
1076 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
1077 if (isnullstartblock(RIGHT
.br_startblock
))
1078 state
|= BMAP_RIGHT_DELAY
;
1081 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1082 new_endoff
== RIGHT
.br_startoff
&&
1083 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1084 newext
== RIGHT
.br_state
&&
1085 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1086 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1087 BMAP_RIGHT_FILLING
)) !=
1088 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1089 BMAP_RIGHT_FILLING
) ||
1090 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1092 state
|= BMAP_RIGHT_CONTIG
;
1095 * Switch out based on the FILLING and CONTIG state bits.
1097 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1098 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1099 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1100 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1102 * Setting all of a previous oldext extent to newext.
1103 * The left and right neighbors are both contiguous with new.
1107 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1108 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1109 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1110 RIGHT
.br_blockcount
);
1111 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1113 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
1114 ip
->i_d
.di_nextents
-= 2;
1116 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1118 rval
= XFS_ILOG_CORE
;
1119 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
1120 RIGHT
.br_startblock
,
1121 RIGHT
.br_blockcount
, &i
)))
1123 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1124 if ((error
= xfs_btree_delete(cur
, &i
)))
1126 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1127 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1129 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1130 if ((error
= xfs_btree_delete(cur
, &i
)))
1132 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1133 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1135 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1136 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
1138 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1139 RIGHT
.br_blockcount
, LEFT
.br_state
)))
1144 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1146 * Setting all of a previous oldext extent to newext.
1147 * The left neighbor is contiguous, the right is not.
1151 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1152 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1153 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
1154 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1156 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1157 ip
->i_d
.di_nextents
--;
1159 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1161 rval
= XFS_ILOG_CORE
;
1162 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1163 PREV
.br_startblock
, PREV
.br_blockcount
,
1166 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1167 if ((error
= xfs_btree_delete(cur
, &i
)))
1169 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1170 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1172 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1173 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
1175 LEFT
.br_blockcount
+ PREV
.br_blockcount
,
1181 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1183 * Setting all of a previous oldext extent to newext.
1184 * The right neighbor is contiguous, the left is not.
1186 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1187 xfs_bmbt_set_blockcount(ep
,
1188 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
1189 xfs_bmbt_set_state(ep
, newext
);
1190 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1191 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1192 ip
->i_d
.di_nextents
--;
1194 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1196 rval
= XFS_ILOG_CORE
;
1197 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
1198 RIGHT
.br_startblock
,
1199 RIGHT
.br_blockcount
, &i
)))
1201 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1202 if ((error
= xfs_btree_delete(cur
, &i
)))
1204 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1205 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1207 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1208 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1210 new->br_blockcount
+ RIGHT
.br_blockcount
,
1216 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1218 * Setting all of a previous oldext extent to newext.
1219 * Neither the left nor right neighbors are contiguous with
1222 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1223 xfs_bmbt_set_state(ep
, newext
);
1224 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1227 rval
= XFS_ILOG_DEXT
;
1230 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1231 new->br_startblock
, new->br_blockcount
,
1234 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1235 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1236 new->br_startblock
, new->br_blockcount
,
1242 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1244 * Setting the first part of a previous oldext extent to newext.
1245 * The left neighbor is contiguous.
1247 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
1248 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
1249 LEFT
.br_blockcount
+ new->br_blockcount
);
1250 xfs_bmbt_set_startoff(ep
,
1251 PREV
.br_startoff
+ new->br_blockcount
);
1252 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
1254 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1255 xfs_bmbt_set_startblock(ep
,
1256 new->br_startblock
+ new->br_blockcount
);
1257 xfs_bmbt_set_blockcount(ep
,
1258 PREV
.br_blockcount
- new->br_blockcount
);
1259 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1264 rval
= XFS_ILOG_DEXT
;
1267 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1268 PREV
.br_startblock
, PREV
.br_blockcount
,
1271 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1272 if ((error
= xfs_bmbt_update(cur
,
1273 PREV
.br_startoff
+ new->br_blockcount
,
1274 PREV
.br_startblock
+ new->br_blockcount
,
1275 PREV
.br_blockcount
- new->br_blockcount
,
1278 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1280 error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
1282 LEFT
.br_blockcount
+ new->br_blockcount
,
1289 case BMAP_LEFT_FILLING
:
1291 * Setting the first part of a previous oldext extent to newext.
1292 * The left neighbor is not contiguous.
1294 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1295 ASSERT(ep
&& xfs_bmbt_get_state(ep
) == oldext
);
1296 xfs_bmbt_set_startoff(ep
, new_endoff
);
1297 xfs_bmbt_set_blockcount(ep
,
1298 PREV
.br_blockcount
- new->br_blockcount
);
1299 xfs_bmbt_set_startblock(ep
,
1300 new->br_startblock
+ new->br_blockcount
);
1301 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1303 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1304 ip
->i_d
.di_nextents
++;
1306 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1308 rval
= XFS_ILOG_CORE
;
1309 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1310 PREV
.br_startblock
, PREV
.br_blockcount
,
1313 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1314 if ((error
= xfs_bmbt_update(cur
,
1315 PREV
.br_startoff
+ new->br_blockcount
,
1316 PREV
.br_startblock
+ new->br_blockcount
,
1317 PREV
.br_blockcount
- new->br_blockcount
,
1320 cur
->bc_rec
.b
= *new;
1321 if ((error
= xfs_btree_insert(cur
, &i
)))
1323 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1327 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1329 * Setting the last part of a previous oldext extent to newext.
1330 * The right neighbor is contiguous with the new allocation.
1332 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1333 xfs_bmbt_set_blockcount(ep
,
1334 PREV
.br_blockcount
- new->br_blockcount
);
1335 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1339 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1340 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
1341 new->br_startoff
, new->br_startblock
,
1342 new->br_blockcount
+ RIGHT
.br_blockcount
, newext
);
1343 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1346 rval
= XFS_ILOG_DEXT
;
1349 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1351 PREV
.br_blockcount
, &i
)))
1353 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1354 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
1356 PREV
.br_blockcount
- new->br_blockcount
,
1359 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
1361 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1363 new->br_blockcount
+ RIGHT
.br_blockcount
,
1369 case BMAP_RIGHT_FILLING
:
1371 * Setting the last part of a previous oldext extent to newext.
1372 * The right neighbor is not contiguous.
1374 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1375 xfs_bmbt_set_blockcount(ep
,
1376 PREV
.br_blockcount
- new->br_blockcount
);
1377 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1380 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1382 ip
->i_d
.di_nextents
++;
1384 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1386 rval
= XFS_ILOG_CORE
;
1387 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1388 PREV
.br_startblock
, PREV
.br_blockcount
,
1391 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1392 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
1394 PREV
.br_blockcount
- new->br_blockcount
,
1397 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1398 new->br_startblock
, new->br_blockcount
,
1401 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1402 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1403 if ((error
= xfs_btree_insert(cur
, &i
)))
1405 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1411 * Setting the middle part of a previous oldext extent to
1412 * newext. Contiguity is impossible here.
1413 * One extent becomes three extents.
1415 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1416 xfs_bmbt_set_blockcount(ep
,
1417 new->br_startoff
- PREV
.br_startoff
);
1418 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1421 r
[1].br_startoff
= new_endoff
;
1422 r
[1].br_blockcount
=
1423 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
1424 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
1425 r
[1].br_state
= oldext
;
1428 xfs_iext_insert(ip
, *idx
, 2, &r
[0], state
);
1430 ip
->i_d
.di_nextents
+= 2;
1432 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1434 rval
= XFS_ILOG_CORE
;
1435 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1436 PREV
.br_startblock
, PREV
.br_blockcount
,
1439 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1440 /* new right extent - oldext */
1441 if ((error
= xfs_bmbt_update(cur
, r
[1].br_startoff
,
1442 r
[1].br_startblock
, r
[1].br_blockcount
,
1445 /* new left extent - oldext */
1446 cur
->bc_rec
.b
= PREV
;
1447 cur
->bc_rec
.b
.br_blockcount
=
1448 new->br_startoff
- PREV
.br_startoff
;
1449 if ((error
= xfs_btree_insert(cur
, &i
)))
1451 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1453 * Reset the cursor to the position of the new extent
1454 * we are about to insert as we can't trust it after
1455 * the previous insert.
1457 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1458 new->br_startblock
, new->br_blockcount
,
1461 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1462 /* new middle extent - newext */
1463 cur
->bc_rec
.b
.br_state
= new->br_state
;
1464 if ((error
= xfs_btree_insert(cur
, &i
)))
1466 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1470 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1471 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1472 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1473 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1474 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1475 case BMAP_LEFT_CONTIG
:
1476 case BMAP_RIGHT_CONTIG
:
1478 * These cases are all impossible.
1483 /* convert to a btree if necessary */
1484 if (xfs_bmap_needs_btree(ip
, XFS_DATA_FORK
)) {
1485 int tmp_logflags
; /* partial log flag return val */
1487 ASSERT(cur
== NULL
);
1488 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
, flist
, &cur
,
1489 0, &tmp_logflags
, XFS_DATA_FORK
);
1490 *logflagsp
|= tmp_logflags
;
1495 /* clear out the allocated field, done with it now in any case. */
1497 cur
->bc_private
.b
.allocated
= 0;
1501 xfs_bmap_check_leaf_extents(*curp
, ip
, XFS_DATA_FORK
);
1511 * Convert a hole to a delayed allocation.
1514 xfs_bmap_add_extent_hole_delay(
1515 xfs_inode_t
*ip
, /* incore inode pointer */
1516 xfs_extnum_t
*idx
, /* extent number to update/insert */
1517 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
1519 xfs_ifork_t
*ifp
; /* inode fork pointer */
1520 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
1521 xfs_filblks_t newlen
=0; /* new indirect size */
1522 xfs_filblks_t oldlen
=0; /* old indirect size */
1523 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
1524 int state
; /* state bits, accessed thru macros */
1525 xfs_filblks_t temp
=0; /* temp for indirect calculations */
1527 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1529 ASSERT(isnullstartblock(new->br_startblock
));
1532 * Check and set flags if this segment has a left neighbor
1535 state
|= BMAP_LEFT_VALID
;
1536 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
1538 if (isnullstartblock(left
.br_startblock
))
1539 state
|= BMAP_LEFT_DELAY
;
1543 * Check and set flags if the current (right) segment exists.
1544 * If it doesn't exist, we're converting the hole at end-of-file.
1546 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
1547 state
|= BMAP_RIGHT_VALID
;
1548 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
1550 if (isnullstartblock(right
.br_startblock
))
1551 state
|= BMAP_RIGHT_DELAY
;
1555 * Set contiguity flags on the left and right neighbors.
1556 * Don't let extents get too large, even if the pieces are contiguous.
1558 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
1559 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
1560 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1561 state
|= BMAP_LEFT_CONTIG
;
1563 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
1564 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
1565 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
1566 (!(state
& BMAP_LEFT_CONTIG
) ||
1567 (left
.br_blockcount
+ new->br_blockcount
+
1568 right
.br_blockcount
<= MAXEXTLEN
)))
1569 state
|= BMAP_RIGHT_CONTIG
;
1572 * Switch out based on the contiguity flags.
1574 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
1575 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1577 * New allocation is contiguous with delayed allocations
1578 * on the left and on the right.
1579 * Merge all three into a single extent record.
1582 temp
= left
.br_blockcount
+ new->br_blockcount
+
1583 right
.br_blockcount
;
1585 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1586 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
1587 oldlen
= startblockval(left
.br_startblock
) +
1588 startblockval(new->br_startblock
) +
1589 startblockval(right
.br_startblock
);
1590 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1591 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
1592 nullstartblock((int)newlen
));
1593 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1595 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1598 case BMAP_LEFT_CONTIG
:
1600 * New allocation is contiguous with a delayed allocation
1602 * Merge the new allocation with the left neighbor.
1605 temp
= left
.br_blockcount
+ new->br_blockcount
;
1607 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1608 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
1609 oldlen
= startblockval(left
.br_startblock
) +
1610 startblockval(new->br_startblock
);
1611 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1612 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
1613 nullstartblock((int)newlen
));
1614 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1617 case BMAP_RIGHT_CONTIG
:
1619 * New allocation is contiguous with a delayed allocation
1621 * Merge the new allocation with the right neighbor.
1623 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1624 temp
= new->br_blockcount
+ right
.br_blockcount
;
1625 oldlen
= startblockval(new->br_startblock
) +
1626 startblockval(right
.br_startblock
);
1627 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1628 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
1630 nullstartblock((int)newlen
), temp
, right
.br_state
);
1631 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1636 * New allocation is not contiguous with another
1637 * delayed allocation.
1638 * Insert a new entry.
1640 oldlen
= newlen
= 0;
1641 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1644 if (oldlen
!= newlen
) {
1645 ASSERT(oldlen
> newlen
);
1646 xfs_icsb_modify_counters(ip
->i_mount
, XFS_SBS_FDBLOCKS
,
1647 (int64_t)(oldlen
- newlen
), 0);
1649 * Nothing to do for disk quota accounting here.
1655 * Convert a hole to a real allocation.
1657 STATIC
int /* error */
1658 xfs_bmap_add_extent_hole_real(
1659 struct xfs_bmalloca
*bma
,
1662 struct xfs_bmbt_irec
*new = &bma
->got
;
1663 int error
; /* error return value */
1664 int i
; /* temp state */
1665 xfs_ifork_t
*ifp
; /* inode fork pointer */
1666 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
1667 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
1668 int rval
=0; /* return value (logging flags) */
1669 int state
; /* state bits, accessed thru macros */
1671 ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
1673 ASSERT(bma
->idx
>= 0);
1674 ASSERT(bma
->idx
<= ifp
->if_bytes
/ sizeof(struct xfs_bmbt_rec
));
1675 ASSERT(!isnullstartblock(new->br_startblock
));
1677 !(bma
->cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_WASDEL
));
1679 XFS_STATS_INC(xs_add_exlist
);
1682 if (whichfork
== XFS_ATTR_FORK
)
1683 state
|= BMAP_ATTRFORK
;
1686 * Check and set flags if this segment has a left neighbor.
1689 state
|= BMAP_LEFT_VALID
;
1690 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1), &left
);
1691 if (isnullstartblock(left
.br_startblock
))
1692 state
|= BMAP_LEFT_DELAY
;
1696 * Check and set flags if this segment has a current value.
1697 * Not true if we're inserting into the "hole" at eof.
1699 if (bma
->idx
< ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
1700 state
|= BMAP_RIGHT_VALID
;
1701 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &right
);
1702 if (isnullstartblock(right
.br_startblock
))
1703 state
|= BMAP_RIGHT_DELAY
;
1707 * We're inserting a real allocation between "left" and "right".
1708 * Set the contiguity flags. Don't let extents get too large.
1710 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1711 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
1712 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
1713 left
.br_state
== new->br_state
&&
1714 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1715 state
|= BMAP_LEFT_CONTIG
;
1717 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1718 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
1719 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
1720 new->br_state
== right
.br_state
&&
1721 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
1722 (!(state
& BMAP_LEFT_CONTIG
) ||
1723 left
.br_blockcount
+ new->br_blockcount
+
1724 right
.br_blockcount
<= MAXEXTLEN
))
1725 state
|= BMAP_RIGHT_CONTIG
;
1729 * Select which case we're in here, and implement it.
1731 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
1732 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1734 * New allocation is contiguous with real allocations on the
1735 * left and on the right.
1736 * Merge all three into a single extent record.
1739 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1740 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1741 left
.br_blockcount
+ new->br_blockcount
+
1742 right
.br_blockcount
);
1743 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1745 xfs_iext_remove(bma
->ip
, bma
->idx
+ 1, 1, state
);
1747 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
1748 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) - 1);
1749 if (bma
->cur
== NULL
) {
1750 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
1752 rval
= XFS_ILOG_CORE
;
1753 error
= xfs_bmbt_lookup_eq(bma
->cur
, right
.br_startoff
,
1754 right
.br_startblock
, right
.br_blockcount
,
1758 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1759 error
= xfs_btree_delete(bma
->cur
, &i
);
1762 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1763 error
= xfs_btree_decrement(bma
->cur
, 0, &i
);
1766 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1767 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
1769 left
.br_blockcount
+
1770 new->br_blockcount
+
1771 right
.br_blockcount
,
1778 case BMAP_LEFT_CONTIG
:
1780 * New allocation is contiguous with a real allocation
1782 * Merge the new allocation with the left neighbor.
1785 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1786 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, bma
->idx
),
1787 left
.br_blockcount
+ new->br_blockcount
);
1788 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1790 if (bma
->cur
== NULL
) {
1791 rval
= xfs_ilog_fext(whichfork
);
1794 error
= xfs_bmbt_lookup_eq(bma
->cur
, left
.br_startoff
,
1795 left
.br_startblock
, left
.br_blockcount
,
1799 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1800 error
= xfs_bmbt_update(bma
->cur
, left
.br_startoff
,
1802 left
.br_blockcount
+
1810 case BMAP_RIGHT_CONTIG
:
1812 * New allocation is contiguous with a real allocation
1814 * Merge the new allocation with the right neighbor.
1816 trace_xfs_bmap_pre_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1817 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, bma
->idx
),
1818 new->br_startoff
, new->br_startblock
,
1819 new->br_blockcount
+ right
.br_blockcount
,
1821 trace_xfs_bmap_post_update(bma
->ip
, bma
->idx
, state
, _THIS_IP_
);
1823 if (bma
->cur
== NULL
) {
1824 rval
= xfs_ilog_fext(whichfork
);
1827 error
= xfs_bmbt_lookup_eq(bma
->cur
,
1829 right
.br_startblock
,
1830 right
.br_blockcount
, &i
);
1833 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1834 error
= xfs_bmbt_update(bma
->cur
, new->br_startoff
,
1836 new->br_blockcount
+
1837 right
.br_blockcount
,
1846 * New allocation is not contiguous with another
1848 * Insert a new entry.
1850 xfs_iext_insert(bma
->ip
, bma
->idx
, 1, new, state
);
1851 XFS_IFORK_NEXT_SET(bma
->ip
, whichfork
,
1852 XFS_IFORK_NEXTENTS(bma
->ip
, whichfork
) + 1);
1853 if (bma
->cur
== NULL
) {
1854 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
1856 rval
= XFS_ILOG_CORE
;
1857 error
= xfs_bmbt_lookup_eq(bma
->cur
,
1860 new->br_blockcount
, &i
);
1863 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1864 bma
->cur
->bc_rec
.b
.br_state
= new->br_state
;
1865 error
= xfs_btree_insert(bma
->cur
, &i
);
1868 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1873 /* convert to a btree if necessary */
1874 if (xfs_bmap_needs_btree(bma
->ip
, whichfork
)) {
1875 int tmp_logflags
; /* partial log flag return val */
1877 ASSERT(bma
->cur
== NULL
);
1878 error
= xfs_bmap_extents_to_btree(bma
->tp
, bma
->ip
,
1879 bma
->firstblock
, bma
->flist
, &bma
->cur
,
1880 0, &tmp_logflags
, whichfork
);
1881 bma
->logflags
|= tmp_logflags
;
1886 /* clear out the allocated field, done with it now in any case. */
1888 bma
->cur
->bc_private
.b
.allocated
= 0;
1890 xfs_bmap_check_leaf_extents(bma
->cur
, bma
->ip
, whichfork
);
1892 bma
->logflags
|= rval
;
1897 * Adjust the size of the new extent based on di_extsize and rt extsize.
1900 xfs_bmap_extsize_align(
1902 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
1903 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
1904 xfs_extlen_t extsz
, /* align to this extent size */
1905 int rt
, /* is this a realtime inode? */
1906 int eof
, /* is extent at end-of-file? */
1907 int delay
, /* creating delalloc extent? */
1908 int convert
, /* overwriting unwritten extent? */
1909 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
1910 xfs_extlen_t
*lenp
) /* in/out: aligned length */
1912 xfs_fileoff_t orig_off
; /* original offset */
1913 xfs_extlen_t orig_alen
; /* original length */
1914 xfs_fileoff_t orig_end
; /* original off+len */
1915 xfs_fileoff_t nexto
; /* next file offset */
1916 xfs_fileoff_t prevo
; /* previous file offset */
1917 xfs_fileoff_t align_off
; /* temp for offset */
1918 xfs_extlen_t align_alen
; /* temp for length */
1919 xfs_extlen_t temp
; /* temp for calculations */
1924 orig_off
= align_off
= *offp
;
1925 orig_alen
= align_alen
= *lenp
;
1926 orig_end
= orig_off
+ orig_alen
;
1929 * If this request overlaps an existing extent, then don't
1930 * attempt to perform any additional alignment.
1932 if (!delay
&& !eof
&&
1933 (orig_off
>= gotp
->br_startoff
) &&
1934 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
1939 * If the file offset is unaligned vs. the extent size
1940 * we need to align it. This will be possible unless
1941 * the file was previously written with a kernel that didn't
1942 * perform this alignment, or if a truncate shot us in the
1945 temp
= do_mod(orig_off
, extsz
);
1951 * Same adjustment for the end of the requested area.
1953 if ((temp
= (align_alen
% extsz
))) {
1954 align_alen
+= extsz
- temp
;
1957 * If the previous block overlaps with this proposed allocation
1958 * then move the start forward without adjusting the length.
1960 if (prevp
->br_startoff
!= NULLFILEOFF
) {
1961 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
1962 prevo
= prevp
->br_startoff
;
1964 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
1967 if (align_off
!= orig_off
&& align_off
< prevo
)
1970 * If the next block overlaps with this proposed allocation
1971 * then move the start back without adjusting the length,
1972 * but not before offset 0.
1973 * This may of course make the start overlap previous block,
1974 * and if we hit the offset 0 limit then the next block
1975 * can still overlap too.
1977 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
1978 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
1979 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
1980 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
1982 nexto
= gotp
->br_startoff
;
1984 nexto
= NULLFILEOFF
;
1986 align_off
+ align_alen
!= orig_end
&&
1987 align_off
+ align_alen
> nexto
)
1988 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
1990 * If we're now overlapping the next or previous extent that
1991 * means we can't fit an extsz piece in this hole. Just move
1992 * the start forward to the first valid spot and set
1993 * the length so we hit the end.
1995 if (align_off
!= orig_off
&& align_off
< prevo
)
1997 if (align_off
+ align_alen
!= orig_end
&&
1998 align_off
+ align_alen
> nexto
&&
1999 nexto
!= NULLFILEOFF
) {
2000 ASSERT(nexto
> prevo
);
2001 align_alen
= nexto
- align_off
;
2005 * If realtime, and the result isn't a multiple of the realtime
2006 * extent size we need to remove blocks until it is.
2008 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
2010 * We're not covering the original request, or
2011 * we won't be able to once we fix the length.
2013 if (orig_off
< align_off
||
2014 orig_end
> align_off
+ align_alen
||
2015 align_alen
- temp
< orig_alen
)
2016 return XFS_ERROR(EINVAL
);
2018 * Try to fix it by moving the start up.
2020 if (align_off
+ temp
<= orig_off
) {
2025 * Try to fix it by moving the end in.
2027 else if (align_off
+ align_alen
- temp
>= orig_end
)
2030 * Set the start to the minimum then trim the length.
2033 align_alen
-= orig_off
- align_off
;
2034 align_off
= orig_off
;
2035 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
2038 * Result doesn't cover the request, fail it.
2040 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
2041 return XFS_ERROR(EINVAL
);
2043 ASSERT(orig_off
>= align_off
);
2044 ASSERT(orig_end
<= align_off
+ align_alen
);
2048 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
2049 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
2050 if (prevp
->br_startoff
!= NULLFILEOFF
)
2051 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
2059 #define XFS_ALLOC_GAP_UNITS 4
2063 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2065 xfs_fsblock_t adjust
; /* adjustment to block numbers */
2066 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
2067 xfs_mount_t
*mp
; /* mount point structure */
2068 int nullfb
; /* true if ap->firstblock isn't set */
2069 int rt
; /* true if inode is realtime */
2071 #define ISVALID(x,y) \
2073 (x) < mp->m_sb.sb_rblocks : \
2074 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2075 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2076 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2078 mp
= ap
->ip
->i_mount
;
2079 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
2080 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
;
2081 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
2083 * If allocating at eof, and there's a previous real block,
2084 * try to use its last block as our starting point.
2086 if (ap
->eof
&& ap
->prev
.br_startoff
!= NULLFILEOFF
&&
2087 !isnullstartblock(ap
->prev
.br_startblock
) &&
2088 ISVALID(ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
,
2089 ap
->prev
.br_startblock
)) {
2090 ap
->blkno
= ap
->prev
.br_startblock
+ ap
->prev
.br_blockcount
;
2092 * Adjust for the gap between prevp and us.
2094 adjust
= ap
->offset
-
2095 (ap
->prev
.br_startoff
+ ap
->prev
.br_blockcount
);
2097 ISVALID(ap
->blkno
+ adjust
, ap
->prev
.br_startblock
))
2098 ap
->blkno
+= adjust
;
2101 * If not at eof, then compare the two neighbor blocks.
2102 * Figure out whether either one gives us a good starting point,
2103 * and pick the better one.
2105 else if (!ap
->eof
) {
2106 xfs_fsblock_t gotbno
; /* right side block number */
2107 xfs_fsblock_t gotdiff
=0; /* right side difference */
2108 xfs_fsblock_t prevbno
; /* left side block number */
2109 xfs_fsblock_t prevdiff
=0; /* left side difference */
2112 * If there's a previous (left) block, select a requested
2113 * start block based on it.
2115 if (ap
->prev
.br_startoff
!= NULLFILEOFF
&&
2116 !isnullstartblock(ap
->prev
.br_startblock
) &&
2117 (prevbno
= ap
->prev
.br_startblock
+
2118 ap
->prev
.br_blockcount
) &&
2119 ISVALID(prevbno
, ap
->prev
.br_startblock
)) {
2121 * Calculate gap to end of previous block.
2123 adjust
= prevdiff
= ap
->offset
-
2124 (ap
->prev
.br_startoff
+
2125 ap
->prev
.br_blockcount
);
2127 * Figure the startblock based on the previous block's
2128 * end and the gap size.
2130 * If the gap is large relative to the piece we're
2131 * allocating, or using it gives us an invalid block
2132 * number, then just use the end of the previous block.
2134 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
2135 ISVALID(prevbno
+ prevdiff
,
2136 ap
->prev
.br_startblock
))
2141 * If the firstblock forbids it, can't use it,
2144 if (!rt
&& !nullfb
&&
2145 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
2146 prevbno
= NULLFSBLOCK
;
2149 * No previous block or can't follow it, just default.
2152 prevbno
= NULLFSBLOCK
;
2154 * If there's a following (right) block, select a requested
2155 * start block based on it.
2157 if (!isnullstartblock(ap
->got
.br_startblock
)) {
2159 * Calculate gap to start of next block.
2161 adjust
= gotdiff
= ap
->got
.br_startoff
- ap
->offset
;
2163 * Figure the startblock based on the next block's
2164 * start and the gap size.
2166 gotbno
= ap
->got
.br_startblock
;
2169 * If the gap is large relative to the piece we're
2170 * allocating, or using it gives us an invalid block
2171 * number, then just use the start of the next block
2172 * offset by our length.
2174 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->length
&&
2175 ISVALID(gotbno
- gotdiff
, gotbno
))
2177 else if (ISVALID(gotbno
- ap
->length
, gotbno
)) {
2178 gotbno
-= ap
->length
;
2179 gotdiff
+= adjust
- ap
->length
;
2183 * If the firstblock forbids it, can't use it,
2186 if (!rt
&& !nullfb
&&
2187 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
2188 gotbno
= NULLFSBLOCK
;
2191 * No next block, just default.
2194 gotbno
= NULLFSBLOCK
;
2196 * If both valid, pick the better one, else the only good
2197 * one, else ap->blkno is already set (to 0 or the inode block).
2199 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
2200 ap
->blkno
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
2201 else if (prevbno
!= NULLFSBLOCK
)
2202 ap
->blkno
= prevbno
;
2203 else if (gotbno
!= NULLFSBLOCK
)
2211 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2213 xfs_alloctype_t atype
= 0; /* type for allocation routines */
2214 int error
; /* error return value */
2215 xfs_mount_t
*mp
; /* mount point structure */
2216 xfs_extlen_t prod
= 0; /* product factor for allocators */
2217 xfs_extlen_t ralen
= 0; /* realtime allocation length */
2218 xfs_extlen_t align
; /* minimum allocation alignment */
2221 mp
= ap
->ip
->i_mount
;
2222 align
= xfs_get_extsz_hint(ap
->ip
);
2223 prod
= align
/ mp
->m_sb
.sb_rextsize
;
2224 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
2225 align
, 1, ap
->eof
, 0,
2226 ap
->conv
, &ap
->offset
, &ap
->length
);
2230 ASSERT(ap
->length
% mp
->m_sb
.sb_rextsize
== 0);
2233 * If the offset & length are not perfectly aligned
2234 * then kill prod, it will just get us in trouble.
2236 if (do_mod(ap
->offset
, align
) || ap
->length
% align
)
2239 * Set ralen to be the actual requested length in rtextents.
2241 ralen
= ap
->length
/ mp
->m_sb
.sb_rextsize
;
2243 * If the old value was close enough to MAXEXTLEN that
2244 * we rounded up to it, cut it back so it's valid again.
2245 * Note that if it's a really large request (bigger than
2246 * MAXEXTLEN), we don't hear about that number, and can't
2247 * adjust the starting point to match it.
2249 if (ralen
* mp
->m_sb
.sb_rextsize
>= MAXEXTLEN
)
2250 ralen
= MAXEXTLEN
/ mp
->m_sb
.sb_rextsize
;
2253 * Lock out other modifications to the RT bitmap inode.
2255 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
2256 xfs_trans_ijoin(ap
->tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
2259 * If it's an allocation to an empty file at offset 0,
2260 * pick an extent that will space things out in the rt area.
2262 if (ap
->eof
&& ap
->offset
== 0) {
2263 xfs_rtblock_t
uninitialized_var(rtx
); /* realtime extent no */
2265 error
= xfs_rtpick_extent(mp
, ap
->tp
, ralen
, &rtx
);
2268 ap
->blkno
= rtx
* mp
->m_sb
.sb_rextsize
;
2273 xfs_bmap_adjacent(ap
);
2276 * Realtime allocation, done through xfs_rtallocate_extent.
2278 atype
= ap
->blkno
== 0 ? XFS_ALLOCTYPE_ANY_AG
: XFS_ALLOCTYPE_NEAR_BNO
;
2279 do_div(ap
->blkno
, mp
->m_sb
.sb_rextsize
);
2282 if ((error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1, ap
->length
,
2283 &ralen
, atype
, ap
->wasdel
, prod
, &rtb
)))
2285 if (rtb
== NULLFSBLOCK
&& prod
> 1 &&
2286 (error
= xfs_rtallocate_extent(ap
->tp
, ap
->blkno
, 1,
2287 ap
->length
, &ralen
, atype
,
2288 ap
->wasdel
, 1, &rtb
)))
2291 if (ap
->blkno
!= NULLFSBLOCK
) {
2292 ap
->blkno
*= mp
->m_sb
.sb_rextsize
;
2293 ralen
*= mp
->m_sb
.sb_rextsize
;
2295 ap
->ip
->i_d
.di_nblocks
+= ralen
;
2296 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
2298 ap
->ip
->i_delayed_blks
-= ralen
;
2300 * Adjust the disk quota also. This was reserved
2303 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
2304 ap
->wasdel
? XFS_TRANS_DQ_DELRTBCOUNT
:
2305 XFS_TRANS_DQ_RTBCOUNT
, (long) ralen
);
2313 xfs_bmap_btalloc_nullfb(
2314 struct xfs_bmalloca
*ap
,
2315 struct xfs_alloc_arg
*args
,
2318 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
2319 struct xfs_perag
*pag
;
2320 xfs_agnumber_t ag
, startag
;
2324 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
))
2325 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2327 args
->type
= XFS_ALLOCTYPE_START_BNO
;
2328 args
->total
= ap
->total
;
2331 * Search for an allocation group with a single extent large enough
2332 * for the request. If one isn't found, then adjust the minimum
2333 * allocation size to the largest space found.
2335 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2336 if (startag
== NULLAGNUMBER
)
2339 pag
= xfs_perag_get(mp
, ag
);
2340 while (*blen
< args
->maxlen
) {
2341 if (!pag
->pagf_init
) {
2342 error
= xfs_alloc_pagf_init(mp
, args
->tp
, ag
,
2343 XFS_ALLOC_FLAG_TRYLOCK
);
2351 * See xfs_alloc_fix_freelist...
2353 if (pag
->pagf_init
) {
2354 xfs_extlen_t longest
;
2355 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
2356 if (*blen
< longest
)
2361 if (xfs_inode_is_filestream(ap
->ip
)) {
2362 if (*blen
>= args
->maxlen
)
2367 * If startag is an invalid AG, we've
2368 * come here once before and
2369 * xfs_filestream_new_ag picked the
2370 * best currently available.
2372 * Don't continue looping, since we
2373 * could loop forever.
2375 if (startag
== NULLAGNUMBER
)
2378 error
= xfs_filestream_new_ag(ap
, &ag
);
2383 /* loop again to set 'blen'*/
2384 startag
= NULLAGNUMBER
;
2385 pag
= xfs_perag_get(mp
, ag
);
2389 if (++ag
== mp
->m_sb
.sb_agcount
)
2394 pag
= xfs_perag_get(mp
, ag
);
2399 * Since the above loop did a BUF_TRYLOCK, it is
2400 * possible that there is space for this request.
2402 if (notinit
|| *blen
< ap
->minlen
)
2403 args
->minlen
= ap
->minlen
;
2405 * If the best seen length is less than the request
2406 * length, use the best as the minimum.
2408 else if (*blen
< args
->maxlen
)
2409 args
->minlen
= *blen
;
2411 * Otherwise we've seen an extent as big as maxlen,
2412 * use that as the minimum.
2415 args
->minlen
= args
->maxlen
;
2418 * set the failure fallback case to look in the selected
2419 * AG as the stream may have moved.
2421 if (xfs_inode_is_filestream(ap
->ip
))
2422 ap
->blkno
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
2429 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2431 xfs_mount_t
*mp
; /* mount point structure */
2432 xfs_alloctype_t atype
= 0; /* type for allocation routines */
2433 xfs_extlen_t align
; /* minimum allocation alignment */
2434 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
2436 xfs_alloc_arg_t args
;
2438 xfs_extlen_t nextminlen
= 0;
2439 int nullfb
; /* true if ap->firstblock isn't set */
2446 mp
= ap
->ip
->i_mount
;
2447 align
= ap
->userdata
? xfs_get_extsz_hint(ap
->ip
) : 0;
2448 if (unlikely(align
)) {
2449 error
= xfs_bmap_extsize_align(mp
, &ap
->got
, &ap
->prev
,
2450 align
, 0, ap
->eof
, 0, ap
->conv
,
2451 &ap
->offset
, &ap
->length
);
2455 nullfb
= *ap
->firstblock
== NULLFSBLOCK
;
2456 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
);
2458 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
)) {
2459 ag
= xfs_filestream_lookup_ag(ap
->ip
);
2460 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
2461 ap
->blkno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
2463 ap
->blkno
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
2466 ap
->blkno
= *ap
->firstblock
;
2468 xfs_bmap_adjacent(ap
);
2471 * If allowed, use ap->blkno; otherwise must use firstblock since
2472 * it's in the right allocation group.
2474 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->blkno
) == fb_agno
)
2477 ap
->blkno
= *ap
->firstblock
;
2479 * Normal allocation, done through xfs_alloc_vextent.
2481 tryagain
= isaligned
= 0;
2482 memset(&args
, 0, sizeof(args
));
2485 args
.fsbno
= ap
->blkno
;
2487 /* Trim the allocation back to the maximum an AG can fit. */
2488 args
.maxlen
= MIN(ap
->length
, XFS_ALLOC_AG_MAX_USABLE(mp
));
2489 args
.firstblock
= *ap
->firstblock
;
2492 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
2495 } else if (ap
->flist
->xbf_low
) {
2496 if (xfs_inode_is_filestream(ap
->ip
))
2497 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
2499 args
.type
= XFS_ALLOCTYPE_START_BNO
;
2500 args
.total
= args
.minlen
= ap
->minlen
;
2502 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
2503 args
.total
= ap
->total
;
2504 args
.minlen
= ap
->minlen
;
2506 /* apply extent size hints if obtained earlier */
2507 if (unlikely(align
)) {
2509 if ((args
.mod
= (xfs_extlen_t
)do_mod(ap
->offset
, args
.prod
)))
2510 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
2511 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_CACHE_SIZE
) {
2515 args
.prod
= PAGE_CACHE_SIZE
>> mp
->m_sb
.sb_blocklog
;
2516 if ((args
.mod
= (xfs_extlen_t
)(do_mod(ap
->offset
, args
.prod
))))
2517 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
2520 * If we are not low on available data blocks, and the
2521 * underlying logical volume manager is a stripe, and
2522 * the file offset is zero then try to allocate data
2523 * blocks on stripe unit boundary.
2524 * NOTE: ap->aeof is only set if the allocation length
2525 * is >= the stripe unit and the allocation offset is
2526 * at the end of file.
2528 if (!ap
->flist
->xbf_low
&& ap
->aeof
) {
2530 args
.alignment
= mp
->m_dalign
;
2534 * Adjust for alignment
2536 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
2537 args
.minlen
= blen
- args
.alignment
;
2538 args
.minalignslop
= 0;
2541 * First try an exact bno allocation.
2542 * If it fails then do a near or start bno
2543 * allocation with alignment turned on.
2547 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
2550 * Compute the minlen+alignment for the
2551 * next case. Set slop so that the value
2552 * of minlen+alignment+slop doesn't go up
2553 * between the calls.
2555 if (blen
> mp
->m_dalign
&& blen
<= args
.maxlen
)
2556 nextminlen
= blen
- mp
->m_dalign
;
2558 nextminlen
= args
.minlen
;
2559 if (nextminlen
+ mp
->m_dalign
> args
.minlen
+ 1)
2561 nextminlen
+ mp
->m_dalign
-
2564 args
.minalignslop
= 0;
2568 args
.minalignslop
= 0;
2570 args
.minleft
= ap
->minleft
;
2571 args
.wasdel
= ap
->wasdel
;
2573 args
.userdata
= ap
->userdata
;
2574 if ((error
= xfs_alloc_vextent(&args
)))
2576 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
2578 * Exact allocation failed. Now try with alignment
2582 args
.fsbno
= ap
->blkno
;
2583 args
.alignment
= mp
->m_dalign
;
2584 args
.minlen
= nextminlen
;
2585 args
.minalignslop
= 0;
2587 if ((error
= xfs_alloc_vextent(&args
)))
2590 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
2592 * allocation failed, so turn off alignment and
2596 args
.fsbno
= ap
->blkno
;
2598 if ((error
= xfs_alloc_vextent(&args
)))
2601 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
2602 args
.minlen
> ap
->minlen
) {
2603 args
.minlen
= ap
->minlen
;
2604 args
.type
= XFS_ALLOCTYPE_START_BNO
;
2605 args
.fsbno
= ap
->blkno
;
2606 if ((error
= xfs_alloc_vextent(&args
)))
2609 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
2611 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
2612 args
.total
= ap
->minlen
;
2614 if ((error
= xfs_alloc_vextent(&args
)))
2616 ap
->flist
->xbf_low
= 1;
2618 if (args
.fsbno
!= NULLFSBLOCK
) {
2620 * check the allocation happened at the same or higher AG than
2621 * the first block that was allocated.
2623 ASSERT(*ap
->firstblock
== NULLFSBLOCK
||
2624 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) ==
2625 XFS_FSB_TO_AGNO(mp
, args
.fsbno
) ||
2626 (ap
->flist
->xbf_low
&&
2627 XFS_FSB_TO_AGNO(mp
, *ap
->firstblock
) <
2628 XFS_FSB_TO_AGNO(mp
, args
.fsbno
)));
2630 ap
->blkno
= args
.fsbno
;
2631 if (*ap
->firstblock
== NULLFSBLOCK
)
2632 *ap
->firstblock
= args
.fsbno
;
2633 ASSERT(nullfb
|| fb_agno
== args
.agno
||
2634 (ap
->flist
->xbf_low
&& fb_agno
< args
.agno
));
2635 ap
->length
= args
.len
;
2636 ap
->ip
->i_d
.di_nblocks
+= args
.len
;
2637 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
2639 ap
->ip
->i_delayed_blks
-= args
.len
;
2641 * Adjust the disk quota also. This was reserved
2644 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
2645 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
:
2646 XFS_TRANS_DQ_BCOUNT
,
2649 ap
->blkno
= NULLFSBLOCK
;
2656 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2657 * It figures out where to ask the underlying allocator to put the new extent.
2661 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2663 if (XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
)
2664 return xfs_bmap_rtalloc(ap
);
2665 return xfs_bmap_btalloc(ap
);
2669 * Transform a btree format file with only one leaf node, where the
2670 * extents list will fit in the inode, into an extents format file.
2671 * Since the file extents are already in-core, all we have to do is
2672 * give up the space for the btree root and pitch the leaf block.
2674 STATIC
int /* error */
2675 xfs_bmap_btree_to_extents(
2676 xfs_trans_t
*tp
, /* transaction pointer */
2677 xfs_inode_t
*ip
, /* incore inode pointer */
2678 xfs_btree_cur_t
*cur
, /* btree cursor */
2679 int *logflagsp
, /* inode logging flags */
2680 int whichfork
) /* data or attr fork */
2683 struct xfs_btree_block
*cblock
;/* child btree block */
2684 xfs_fsblock_t cbno
; /* child block number */
2685 xfs_buf_t
*cbp
; /* child block's buffer */
2686 int error
; /* error return value */
2687 xfs_ifork_t
*ifp
; /* inode fork data */
2688 xfs_mount_t
*mp
; /* mount point structure */
2689 __be64
*pp
; /* ptr to block address */
2690 struct xfs_btree_block
*rblock
;/* root btree block */
2693 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2694 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
2695 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
2696 rblock
= ifp
->if_broot
;
2697 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
2698 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
2699 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
2700 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
2701 cbno
= be64_to_cpu(*pp
);
2704 if ((error
= xfs_btree_check_lptr(cur
, cbno
, 1)))
2707 error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
, XFS_BMAP_BTREE_REF
,
2711 cblock
= XFS_BUF_TO_BLOCK(cbp
);
2712 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
2714 xfs_bmap_add_free(cbno
, 1, cur
->bc_private
.b
.flist
, mp
);
2715 ip
->i_d
.di_nblocks
--;
2716 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
2717 xfs_trans_binval(tp
, cbp
);
2718 if (cur
->bc_bufs
[0] == cbp
)
2719 cur
->bc_bufs
[0] = NULL
;
2720 xfs_iroot_realloc(ip
, -1, whichfork
);
2721 ASSERT(ifp
->if_broot
== NULL
);
2722 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
2723 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
2724 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2729 * Called by xfs_bmapi to update file extent records and the btree
2730 * after removing space (or undoing a delayed allocation).
2732 STATIC
int /* error */
2733 xfs_bmap_del_extent(
2734 xfs_inode_t
*ip
, /* incore inode pointer */
2735 xfs_trans_t
*tp
, /* current transaction pointer */
2736 xfs_extnum_t
*idx
, /* extent number to update/delete */
2737 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
2738 xfs_btree_cur_t
*cur
, /* if null, not a btree */
2739 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
2740 int *logflagsp
, /* inode logging flags */
2741 int whichfork
) /* data or attr fork */
2743 xfs_filblks_t da_new
; /* new delay-alloc indirect blocks */
2744 xfs_filblks_t da_old
; /* old delay-alloc indirect blocks */
2745 xfs_fsblock_t del_endblock
=0; /* first block past del */
2746 xfs_fileoff_t del_endoff
; /* first offset past del */
2747 int delay
; /* current block is delayed allocated */
2748 int do_fx
; /* free extent at end of routine */
2749 xfs_bmbt_rec_host_t
*ep
; /* current extent entry pointer */
2750 int error
; /* error return value */
2751 int flags
; /* inode logging flags */
2752 xfs_bmbt_irec_t got
; /* current extent entry */
2753 xfs_fileoff_t got_endoff
; /* first offset past got */
2754 int i
; /* temp state */
2755 xfs_ifork_t
*ifp
; /* inode fork pointer */
2756 xfs_mount_t
*mp
; /* mount structure */
2757 xfs_filblks_t nblks
; /* quota/sb block count */
2758 xfs_bmbt_irec_t
new; /* new record to be inserted */
2760 uint qfield
; /* quota field to update */
2761 xfs_filblks_t temp
; /* for indirect length calculations */
2762 xfs_filblks_t temp2
; /* for indirect length calculations */
2765 XFS_STATS_INC(xs_del_exlist
);
2767 if (whichfork
== XFS_ATTR_FORK
)
2768 state
|= BMAP_ATTRFORK
;
2771 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2772 ASSERT((*idx
>= 0) && (*idx
< ifp
->if_bytes
/
2773 (uint
)sizeof(xfs_bmbt_rec_t
)));
2774 ASSERT(del
->br_blockcount
> 0);
2775 ep
= xfs_iext_get_ext(ifp
, *idx
);
2776 xfs_bmbt_get_all(ep
, &got
);
2777 ASSERT(got
.br_startoff
<= del
->br_startoff
);
2778 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
2779 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
2780 ASSERT(got_endoff
>= del_endoff
);
2781 delay
= isnullstartblock(got
.br_startblock
);
2782 ASSERT(isnullstartblock(del
->br_startblock
) == delay
);
2787 * If deleting a real allocation, must free up the disk space.
2790 flags
= XFS_ILOG_CORE
;
2792 * Realtime allocation. Free it and record di_nblocks update.
2794 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
2798 ASSERT(do_mod(del
->br_blockcount
,
2799 mp
->m_sb
.sb_rextsize
) == 0);
2800 ASSERT(do_mod(del
->br_startblock
,
2801 mp
->m_sb
.sb_rextsize
) == 0);
2802 bno
= del
->br_startblock
;
2803 len
= del
->br_blockcount
;
2804 do_div(bno
, mp
->m_sb
.sb_rextsize
);
2805 do_div(len
, mp
->m_sb
.sb_rextsize
);
2806 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
2810 nblks
= len
* mp
->m_sb
.sb_rextsize
;
2811 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
2814 * Ordinary allocation.
2818 nblks
= del
->br_blockcount
;
2819 qfield
= XFS_TRANS_DQ_BCOUNT
;
2822 * Set up del_endblock and cur for later.
2824 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
2826 if ((error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
2827 got
.br_startblock
, got
.br_blockcount
,
2830 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2832 da_old
= da_new
= 0;
2834 da_old
= startblockval(got
.br_startblock
);
2840 * Set flag value to use in switch statement.
2841 * Left-contig is 2, right-contig is 1.
2843 switch (((got
.br_startoff
== del
->br_startoff
) << 1) |
2844 (got_endoff
== del_endoff
)) {
2847 * Matches the whole extent. Delete the entry.
2849 xfs_iext_remove(ip
, *idx
, 1,
2850 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0);
2855 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2856 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2857 flags
|= XFS_ILOG_CORE
;
2859 flags
|= xfs_ilog_fext(whichfork
);
2862 if ((error
= xfs_btree_delete(cur
, &i
)))
2864 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2869 * Deleting the first part of the extent.
2871 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2872 xfs_bmbt_set_startoff(ep
, del_endoff
);
2873 temp
= got
.br_blockcount
- del
->br_blockcount
;
2874 xfs_bmbt_set_blockcount(ep
, temp
);
2876 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2878 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2879 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2883 xfs_bmbt_set_startblock(ep
, del_endblock
);
2884 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2886 flags
|= xfs_ilog_fext(whichfork
);
2889 if ((error
= xfs_bmbt_update(cur
, del_endoff
, del_endblock
,
2890 got
.br_blockcount
- del
->br_blockcount
,
2897 * Deleting the last part of the extent.
2899 temp
= got
.br_blockcount
- del
->br_blockcount
;
2900 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2901 xfs_bmbt_set_blockcount(ep
, temp
);
2903 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2905 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2906 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2910 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2912 flags
|= xfs_ilog_fext(whichfork
);
2915 if ((error
= xfs_bmbt_update(cur
, got
.br_startoff
,
2917 got
.br_blockcount
- del
->br_blockcount
,
2924 * Deleting the middle of the extent.
2926 temp
= del
->br_startoff
- got
.br_startoff
;
2927 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2928 xfs_bmbt_set_blockcount(ep
, temp
);
2929 new.br_startoff
= del_endoff
;
2930 temp2
= got_endoff
- del_endoff
;
2931 new.br_blockcount
= temp2
;
2932 new.br_state
= got
.br_state
;
2934 new.br_startblock
= del_endblock
;
2935 flags
|= XFS_ILOG_CORE
;
2937 if ((error
= xfs_bmbt_update(cur
,
2939 got
.br_startblock
, temp
,
2942 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
2944 cur
->bc_rec
.b
= new;
2945 error
= xfs_btree_insert(cur
, &i
);
2946 if (error
&& error
!= ENOSPC
)
2949 * If get no-space back from btree insert,
2950 * it tried a split, and we have a zero
2951 * block reservation.
2952 * Fix up our state and return the error.
2954 if (error
== ENOSPC
) {
2956 * Reset the cursor, don't trust
2957 * it after any insert operation.
2959 if ((error
= xfs_bmbt_lookup_eq(cur
,
2964 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2966 * Update the btree record back
2967 * to the original value.
2969 if ((error
= xfs_bmbt_update(cur
,
2976 * Reset the extent record back
2977 * to the original value.
2979 xfs_bmbt_set_blockcount(ep
,
2982 error
= XFS_ERROR(ENOSPC
);
2985 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2987 flags
|= xfs_ilog_fext(whichfork
);
2988 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2989 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2991 ASSERT(whichfork
== XFS_DATA_FORK
);
2992 temp
= xfs_bmap_worst_indlen(ip
, temp
);
2993 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2994 temp2
= xfs_bmap_worst_indlen(ip
, temp2
);
2995 new.br_startblock
= nullstartblock((int)temp2
);
2996 da_new
= temp
+ temp2
;
2997 while (da_new
> da_old
) {
3001 xfs_bmbt_set_startblock(ep
,
3002 nullstartblock((int)temp
));
3004 if (da_new
== da_old
)
3010 nullstartblock((int)temp2
);
3014 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
3015 xfs_iext_insert(ip
, *idx
+ 1, 1, &new, state
);
3020 * If we need to, add to list of extents to delete.
3023 xfs_bmap_add_free(del
->br_startblock
, del
->br_blockcount
, flist
,
3026 * Adjust inode # blocks in the file.
3029 ip
->i_d
.di_nblocks
-= nblks
;
3031 * Adjust quota data.
3034 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
3037 * Account for change in delayed indirect blocks.
3038 * Nothing to do for disk quota accounting here.
3040 ASSERT(da_old
>= da_new
);
3041 if (da_old
> da_new
) {
3042 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
3043 (int64_t)(da_old
- da_new
), 0);
3051 * Remove the entry "free" from the free item list. Prev points to the
3052 * previous entry, unless "free" is the head of the list.
3056 xfs_bmap_free_t
*flist
, /* free item list header */
3057 xfs_bmap_free_item_t
*prev
, /* previous item on list, if any */
3058 xfs_bmap_free_item_t
*free
) /* list item to be freed */
3061 prev
->xbfi_next
= free
->xbfi_next
;
3063 flist
->xbf_first
= free
->xbfi_next
;
3065 kmem_zone_free(xfs_bmap_free_item_zone
, free
);
3069 * Convert an extents-format file into a btree-format file.
3070 * The new file will have a root block (in the inode) and a single child block.
3072 STATIC
int /* error */
3073 xfs_bmap_extents_to_btree(
3074 xfs_trans_t
*tp
, /* transaction pointer */
3075 xfs_inode_t
*ip
, /* incore inode pointer */
3076 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
3077 xfs_bmap_free_t
*flist
, /* blocks freed in xaction */
3078 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
3079 int wasdel
, /* converting a delayed alloc */
3080 int *logflagsp
, /* inode logging flags */
3081 int whichfork
) /* data or attr fork */
3083 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
3084 xfs_buf_t
*abp
; /* buffer for ablock */
3085 xfs_alloc_arg_t args
; /* allocation arguments */
3086 xfs_bmbt_rec_t
*arp
; /* child record pointer */
3087 struct xfs_btree_block
*block
; /* btree root block */
3088 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
3089 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3090 int error
; /* error return value */
3091 xfs_extnum_t i
, cnt
; /* extent record index */
3092 xfs_ifork_t
*ifp
; /* inode fork pointer */
3093 xfs_bmbt_key_t
*kp
; /* root block key pointer */
3094 xfs_mount_t
*mp
; /* mount structure */
3095 xfs_extnum_t nextents
; /* number of file extents */
3096 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
3098 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3099 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
3102 * Make space in the inode incore.
3104 xfs_iroot_realloc(ip
, 1, whichfork
);
3105 ifp
->if_flags
|= XFS_IFBROOT
;
3110 block
= ifp
->if_broot
;
3111 block
->bb_magic
= cpu_to_be32(XFS_BMAP_MAGIC
);
3112 block
->bb_level
= cpu_to_be16(1);
3113 block
->bb_numrecs
= cpu_to_be16(1);
3114 block
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLDFSBNO
);
3115 block
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLDFSBNO
);
3118 * Need a cursor. Can't allocate until bb_level is filled in.
3121 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
3122 cur
->bc_private
.b
.firstblock
= *firstblock
;
3123 cur
->bc_private
.b
.flist
= flist
;
3124 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
3126 * Convert to a btree with two levels, one record in root.
3128 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
3129 memset(&args
, 0, sizeof(args
));
3132 args
.firstblock
= *firstblock
;
3133 if (*firstblock
== NULLFSBLOCK
) {
3134 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3135 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
3136 } else if (flist
->xbf_low
) {
3137 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3138 args
.fsbno
= *firstblock
;
3140 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3141 args
.fsbno
= *firstblock
;
3143 args
.minlen
= args
.maxlen
= args
.prod
= 1;
3144 args
.wasdel
= wasdel
;
3146 if ((error
= xfs_alloc_vextent(&args
))) {
3147 xfs_iroot_realloc(ip
, -1, whichfork
);
3148 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
3152 * Allocation can't fail, the space was reserved.
3154 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
3155 ASSERT(*firstblock
== NULLFSBLOCK
||
3156 args
.agno
== XFS_FSB_TO_AGNO(mp
, *firstblock
) ||
3158 args
.agno
> XFS_FSB_TO_AGNO(mp
, *firstblock
)));
3159 *firstblock
= cur
->bc_private
.b
.firstblock
= args
.fsbno
;
3160 cur
->bc_private
.b
.allocated
++;
3161 ip
->i_d
.di_nblocks
++;
3162 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
3163 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
3165 * Fill in the child block.
3167 abp
->b_ops
= &xfs_bmbt_buf_ops
;
3168 ablock
= XFS_BUF_TO_BLOCK(abp
);
3169 ablock
->bb_magic
= cpu_to_be32(XFS_BMAP_MAGIC
);
3170 ablock
->bb_level
= 0;
3171 ablock
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLDFSBNO
);
3172 ablock
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLDFSBNO
);
3173 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
3174 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3175 for (cnt
= i
= 0; i
< nextents
; i
++) {
3176 ep
= xfs_iext_get_ext(ifp
, i
);
3177 if (!isnullstartblock(xfs_bmbt_get_startblock(ep
))) {
3178 arp
->l0
= cpu_to_be64(ep
->l0
);
3179 arp
->l1
= cpu_to_be64(ep
->l1
);
3183 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
3184 xfs_btree_set_numrecs(ablock
, cnt
);
3187 * Fill in the root key and pointer.
3189 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
3190 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
3191 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
3192 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
3193 be16_to_cpu(block
->bb_level
)));
3194 *pp
= cpu_to_be64(args
.fsbno
);
3197 * Do all this logging at the end so that
3198 * the root is at the right level.
3200 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
3201 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
3202 ASSERT(*curp
== NULL
);
3204 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
3209 * Calculate the default attribute fork offset for newly created inodes.
3212 xfs_default_attroffset(
3213 struct xfs_inode
*ip
)
3215 struct xfs_mount
*mp
= ip
->i_mount
;
3218 if (mp
->m_sb
.sb_inodesize
== 256) {
3219 offset
= XFS_LITINO(mp
) -
3220 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
3222 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
3225 ASSERT(offset
< XFS_LITINO(mp
));
3230 * Helper routine to reset inode di_forkoff field when switching
3231 * attribute fork from local to extent format - we reset it where
3232 * possible to make space available for inline data fork extents.
3235 xfs_bmap_forkoff_reset(
3240 if (whichfork
== XFS_ATTR_FORK
&&
3241 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
3242 ip
->i_d
.di_format
!= XFS_DINODE_FMT_UUID
&&
3243 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
3244 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
3246 if (dfl_forkoff
> ip
->i_d
.di_forkoff
)
3247 ip
->i_d
.di_forkoff
= dfl_forkoff
;
3252 * Convert a local file to an extents file.
3253 * This code is out of bounds for data forks of regular files,
3254 * since the file data needs to get logged so things will stay consistent.
3255 * (The bmap-level manipulations are ok, though).
3257 STATIC
int /* error */
3258 xfs_bmap_local_to_extents(
3259 xfs_trans_t
*tp
, /* transaction pointer */
3260 xfs_inode_t
*ip
, /* incore inode pointer */
3261 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
3262 xfs_extlen_t total
, /* total blocks needed by transaction */
3263 int *logflagsp
, /* inode logging flags */
3265 void (*init_fn
)(struct xfs_buf
*bp
,
3266 struct xfs_inode
*ip
,
3267 struct xfs_ifork
*ifp
))
3269 int error
; /* error return value */
3270 int flags
; /* logging flags returned */
3271 xfs_ifork_t
*ifp
; /* inode fork pointer */
3274 * We don't want to deal with the case of keeping inode data inline yet.
3275 * So sending the data fork of a regular inode is invalid.
3277 ASSERT(!(S_ISREG(ip
->i_d
.di_mode
) && whichfork
== XFS_DATA_FORK
));
3278 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3279 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
3282 if (ifp
->if_bytes
) {
3283 xfs_alloc_arg_t args
; /* allocation arguments */
3284 xfs_buf_t
*bp
; /* buffer for extent block */
3285 xfs_bmbt_rec_host_t
*ep
;/* extent record pointer */
3287 ASSERT((ifp
->if_flags
&
3288 (XFS_IFINLINE
|XFS_IFEXTENTS
|XFS_IFEXTIREC
)) == XFS_IFINLINE
);
3289 memset(&args
, 0, sizeof(args
));
3291 args
.mp
= ip
->i_mount
;
3292 args
.firstblock
= *firstblock
;
3294 * Allocate a block. We know we need only one, since the
3295 * file currently fits in an inode.
3297 if (*firstblock
== NULLFSBLOCK
) {
3298 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
3299 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3301 args
.fsbno
= *firstblock
;
3302 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3305 args
.minlen
= args
.maxlen
= args
.prod
= 1;
3306 error
= xfs_alloc_vextent(&args
);
3310 /* Can't fail, the space was reserved. */
3311 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
3312 ASSERT(args
.len
== 1);
3313 *firstblock
= args
.fsbno
;
3314 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
3316 /* initialise the block and copy the data */
3317 init_fn(bp
, ip
, ifp
);
3319 /* account for the change in fork size and log everything */
3320 xfs_trans_log_buf(tp
, bp
, 0, ifp
->if_bytes
- 1);
3321 xfs_bmap_forkoff_reset(args
.mp
, ip
, whichfork
);
3322 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
3323 xfs_iext_add(ifp
, 0, 1);
3324 ep
= xfs_iext_get_ext(ifp
, 0);
3325 xfs_bmbt_set_allf(ep
, 0, args
.fsbno
, 1, XFS_EXT_NORM
);
3326 trace_xfs_bmap_post_update(ip
, 0,
3327 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0,
3329 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
3330 ip
->i_d
.di_nblocks
= 1;
3331 xfs_trans_mod_dquot_byino(tp
, ip
,
3332 XFS_TRANS_DQ_BCOUNT
, 1L);
3333 flags
|= xfs_ilog_fext(whichfork
);
3335 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
3336 xfs_bmap_forkoff_reset(ip
->i_mount
, ip
, whichfork
);
3338 ifp
->if_flags
&= ~XFS_IFINLINE
;
3339 ifp
->if_flags
|= XFS_IFEXTENTS
;
3340 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
3341 flags
|= XFS_ILOG_CORE
;
3348 * Search the extent records for the entry containing block bno.
3349 * If bno lies in a hole, point to the next entry. If bno lies
3350 * past eof, *eofp will be set, and *prevp will contain the last
3351 * entry (null if none). Else, *lastxp will be set to the index
3352 * of the found entry; *gotp will contain the entry.
3354 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
3355 xfs_bmap_search_multi_extents(
3356 xfs_ifork_t
*ifp
, /* inode fork pointer */
3357 xfs_fileoff_t bno
, /* block number searched for */
3358 int *eofp
, /* out: end of file found */
3359 xfs_extnum_t
*lastxp
, /* out: last extent index */
3360 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
3361 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
3363 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3364 xfs_extnum_t lastx
; /* last extent index */
3367 * Initialize the extent entry structure to catch access to
3368 * uninitialized br_startblock field.
3370 gotp
->br_startoff
= 0xffa5a5a5a5a5a5a5LL
;
3371 gotp
->br_blockcount
= 0xa55a5a5a5a5a5a5aLL
;
3372 gotp
->br_state
= XFS_EXT_INVALID
;
3374 gotp
->br_startblock
= 0xffffa5a5a5a5a5a5LL
;
3376 gotp
->br_startblock
= 0xffffa5a5;
3378 prevp
->br_startoff
= NULLFILEOFF
;
3380 ep
= xfs_iext_bno_to_ext(ifp
, bno
, &lastx
);
3382 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
- 1), prevp
);
3384 if (lastx
< (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))) {
3385 xfs_bmbt_get_all(ep
, gotp
);
3399 * Search the extents list for the inode, for the extent containing bno.
3400 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3401 * *eofp will be set, and *prevp will contain the last entry (null if none).
3402 * Else, *lastxp will be set to the index of the found
3403 * entry; *gotp will contain the entry.
3405 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
3406 xfs_bmap_search_extents(
3407 xfs_inode_t
*ip
, /* incore inode pointer */
3408 xfs_fileoff_t bno
, /* block number searched for */
3409 int fork
, /* data or attr fork */
3410 int *eofp
, /* out: end of file found */
3411 xfs_extnum_t
*lastxp
, /* out: last extent index */
3412 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
3413 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
3415 xfs_ifork_t
*ifp
; /* inode fork pointer */
3416 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3418 XFS_STATS_INC(xs_look_exlist
);
3419 ifp
= XFS_IFORK_PTR(ip
, fork
);
3421 ep
= xfs_bmap_search_multi_extents(ifp
, bno
, eofp
, lastxp
, gotp
, prevp
);
3423 if (unlikely(!(gotp
->br_startblock
) && (*lastxp
!= NULLEXTNUM
) &&
3424 !(XFS_IS_REALTIME_INODE(ip
) && fork
== XFS_DATA_FORK
))) {
3425 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
3426 "Access to block zero in inode %llu "
3427 "start_block: %llx start_off: %llx "
3428 "blkcnt: %llx extent-state: %x lastx: %x\n",
3429 (unsigned long long)ip
->i_ino
,
3430 (unsigned long long)gotp
->br_startblock
,
3431 (unsigned long long)gotp
->br_startoff
,
3432 (unsigned long long)gotp
->br_blockcount
,
3433 gotp
->br_state
, *lastxp
);
3434 *lastxp
= NULLEXTNUM
;
3442 * Compute the worst-case number of indirect blocks that will be used
3443 * for ip's delayed extent of length "len".
3445 STATIC xfs_filblks_t
3446 xfs_bmap_worst_indlen(
3447 xfs_inode_t
*ip
, /* incore inode pointer */
3448 xfs_filblks_t len
) /* delayed extent length */
3450 int level
; /* btree level number */
3451 int maxrecs
; /* maximum record count at this level */
3452 xfs_mount_t
*mp
; /* mount structure */
3453 xfs_filblks_t rval
; /* return value */
3456 maxrecs
= mp
->m_bmap_dmxr
[0];
3457 for (level
= 0, rval
= 0;
3458 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
3461 do_div(len
, maxrecs
);
3464 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
3467 maxrecs
= mp
->m_bmap_dmxr
[1];
3473 * Convert inode from non-attributed to attributed.
3474 * Must not be in a transaction, ip must not be locked.
3476 int /* error code */
3477 xfs_bmap_add_attrfork(
3478 xfs_inode_t
*ip
, /* incore inode pointer */
3479 int size
, /* space new attribute needs */
3480 int rsvd
) /* xact may use reserved blks */
3482 xfs_fsblock_t firstblock
; /* 1st block/ag allocated */
3483 xfs_bmap_free_t flist
; /* freed extent records */
3484 xfs_mount_t
*mp
; /* mount structure */
3485 xfs_trans_t
*tp
; /* transaction pointer */
3486 int blks
; /* space reservation */
3487 int version
= 1; /* superblock attr version */
3488 int committed
; /* xaction was committed */
3489 int logflags
; /* logging flags */
3490 int error
; /* error return value */
3492 ASSERT(XFS_IFORK_Q(ip
) == 0);
3495 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
3496 tp
= xfs_trans_alloc(mp
, XFS_TRANS_ADDAFORK
);
3497 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
3499 tp
->t_flags
|= XFS_TRANS_RESERVE
;
3500 if ((error
= xfs_trans_reserve(tp
, blks
, XFS_ADDAFORK_LOG_RES(mp
), 0,
3501 XFS_TRANS_PERM_LOG_RES
, XFS_ADDAFORK_LOG_COUNT
)))
3503 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3504 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
3505 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
3506 XFS_QMOPT_RES_REGBLKS
);
3508 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3509 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3512 if (XFS_IFORK_Q(ip
))
3514 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
3516 * For inodes coming from pre-6.2 filesystems.
3518 ASSERT(ip
->i_d
.di_aformat
== 0);
3519 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
3521 ASSERT(ip
->i_d
.di_anextents
== 0);
3523 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3524 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3526 switch (ip
->i_d
.di_format
) {
3527 case XFS_DINODE_FMT_DEV
:
3528 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
3530 case XFS_DINODE_FMT_UUID
:
3531 ip
->i_d
.di_forkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
3533 case XFS_DINODE_FMT_LOCAL
:
3534 case XFS_DINODE_FMT_EXTENTS
:
3535 case XFS_DINODE_FMT_BTREE
:
3536 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
3537 if (!ip
->i_d
.di_forkoff
)
3538 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
3539 else if (mp
->m_flags
& XFS_MOUNT_ATTR2
)
3544 error
= XFS_ERROR(EINVAL
);
3548 ASSERT(ip
->i_afp
== NULL
);
3549 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
3550 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
3552 xfs_bmap_init(&flist
, &firstblock
);
3553 switch (ip
->i_d
.di_format
) {
3554 case XFS_DINODE_FMT_LOCAL
:
3555 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &firstblock
, &flist
,
3558 case XFS_DINODE_FMT_EXTENTS
:
3559 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &firstblock
,
3562 case XFS_DINODE_FMT_BTREE
:
3563 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &firstblock
, &flist
,
3571 xfs_trans_log_inode(tp
, ip
, logflags
);
3574 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
3575 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
3576 __int64_t sbfields
= 0;
3578 spin_lock(&mp
->m_sb_lock
);
3579 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
3580 xfs_sb_version_addattr(&mp
->m_sb
);
3581 sbfields
|= XFS_SB_VERSIONNUM
;
3583 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
3584 xfs_sb_version_addattr2(&mp
->m_sb
);
3585 sbfields
|= (XFS_SB_VERSIONNUM
| XFS_SB_FEATURES2
);
3588 spin_unlock(&mp
->m_sb_lock
);
3589 xfs_mod_sb(tp
, sbfields
);
3591 spin_unlock(&mp
->m_sb_lock
);
3594 error
= xfs_bmap_finish(&tp
, &flist
, &committed
);
3597 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3599 xfs_bmap_cancel(&flist
);
3601 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3603 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
3608 * Add the extent to the list of extents to be free at transaction end.
3609 * The list is maintained sorted (by block number).
3614 xfs_fsblock_t bno
, /* fs block number of extent */
3615 xfs_filblks_t len
, /* length of extent */
3616 xfs_bmap_free_t
*flist
, /* list of extents */
3617 xfs_mount_t
*mp
) /* mount point structure */
3619 xfs_bmap_free_item_t
*cur
; /* current (next) element */
3620 xfs_bmap_free_item_t
*new; /* new element */
3621 xfs_bmap_free_item_t
*prev
; /* previous element */
3623 xfs_agnumber_t agno
;
3624 xfs_agblock_t agbno
;
3626 ASSERT(bno
!= NULLFSBLOCK
);
3628 ASSERT(len
<= MAXEXTLEN
);
3629 ASSERT(!isnullstartblock(bno
));
3630 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
3631 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
3632 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
3633 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
3634 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
3635 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
3637 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
3638 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
3639 new->xbfi_startblock
= bno
;
3640 new->xbfi_blockcount
= (xfs_extlen_t
)len
;
3641 for (prev
= NULL
, cur
= flist
->xbf_first
;
3643 prev
= cur
, cur
= cur
->xbfi_next
) {
3644 if (cur
->xbfi_startblock
>= bno
)
3648 prev
->xbfi_next
= new;
3650 flist
->xbf_first
= new;
3651 new->xbfi_next
= cur
;
3656 * Compute and fill in the value of the maximum depth of a bmap btree
3657 * in this filesystem. Done once, during mount.
3660 xfs_bmap_compute_maxlevels(
3661 xfs_mount_t
*mp
, /* file system mount structure */
3662 int whichfork
) /* data or attr fork */
3664 int level
; /* btree level */
3665 uint maxblocks
; /* max blocks at this level */
3666 uint maxleafents
; /* max leaf entries possible */
3667 int maxrootrecs
; /* max records in root block */
3668 int minleafrecs
; /* min records in leaf block */
3669 int minnoderecs
; /* min records in node block */
3670 int sz
; /* root block size */
3673 * The maximum number of extents in a file, hence the maximum
3674 * number of leaf entries, is controlled by the type of di_nextents
3675 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3676 * (a signed 16-bit number, xfs_aextnum_t).
3678 * Note that we can no longer assume that if we are in ATTR1 that
3679 * the fork offset of all the inodes will be
3680 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3681 * with ATTR2 and then mounted back with ATTR1, keeping the
3682 * di_forkoff's fixed but probably at various positions. Therefore,
3683 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3684 * of a minimum size available.
3686 if (whichfork
== XFS_DATA_FORK
) {
3687 maxleafents
= MAXEXTNUM
;
3688 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
3690 maxleafents
= MAXAEXTNUM
;
3691 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
3693 maxrootrecs
= xfs_bmdr_maxrecs(mp
, sz
, 0);
3694 minleafrecs
= mp
->m_bmap_dmnr
[0];
3695 minnoderecs
= mp
->m_bmap_dmnr
[1];
3696 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
3697 for (level
= 1; maxblocks
> 1; level
++) {
3698 if (maxblocks
<= maxrootrecs
)
3701 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
3703 mp
->m_bm_maxlevels
[whichfork
] = level
;
3707 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3708 * caller. Frees all the extents that need freeing, which must be done
3709 * last due to locking considerations. We never free any extents in
3710 * the first transaction.
3712 * Return 1 if the given transaction was committed and a new one
3713 * started, and 0 otherwise in the committed parameter.
3717 xfs_trans_t
**tp
, /* transaction pointer addr */
3718 xfs_bmap_free_t
*flist
, /* i/o: list extents to free */
3719 int *committed
) /* xact committed or not */
3721 xfs_efd_log_item_t
*efd
; /* extent free data */
3722 xfs_efi_log_item_t
*efi
; /* extent free intention */
3723 int error
; /* error return value */
3724 xfs_bmap_free_item_t
*free
; /* free extent item */
3725 unsigned int logres
; /* new log reservation */
3726 unsigned int logcount
; /* new log count */
3727 xfs_mount_t
*mp
; /* filesystem mount structure */
3728 xfs_bmap_free_item_t
*next
; /* next item on free list */
3729 xfs_trans_t
*ntp
; /* new transaction pointer */
3731 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
3732 if (flist
->xbf_count
== 0) {
3737 efi
= xfs_trans_get_efi(ntp
, flist
->xbf_count
);
3738 for (free
= flist
->xbf_first
; free
; free
= free
->xbfi_next
)
3739 xfs_trans_log_efi_extent(ntp
, efi
, free
->xbfi_startblock
,
3740 free
->xbfi_blockcount
);
3741 logres
= ntp
->t_log_res
;
3742 logcount
= ntp
->t_log_count
;
3743 ntp
= xfs_trans_dup(*tp
);
3744 error
= xfs_trans_commit(*tp
, 0);
3748 * We have a new transaction, so we should return committed=1,
3749 * even though we're returning an error.
3755 * transaction commit worked ok so we can drop the extra ticket
3756 * reference that we gained in xfs_trans_dup()
3758 xfs_log_ticket_put(ntp
->t_ticket
);
3760 if ((error
= xfs_trans_reserve(ntp
, 0, logres
, 0, XFS_TRANS_PERM_LOG_RES
,
3763 efd
= xfs_trans_get_efd(ntp
, efi
, flist
->xbf_count
);
3764 for (free
= flist
->xbf_first
; free
!= NULL
; free
= next
) {
3765 next
= free
->xbfi_next
;
3766 if ((error
= xfs_free_extent(ntp
, free
->xbfi_startblock
,
3767 free
->xbfi_blockcount
))) {
3769 * The bmap free list will be cleaned up at a
3770 * higher level. The EFI will be canceled when
3771 * this transaction is aborted.
3772 * Need to force shutdown here to make sure it
3773 * happens, since this transaction may not be
3777 if (!XFS_FORCED_SHUTDOWN(mp
))
3778 xfs_force_shutdown(mp
,
3779 (error
== EFSCORRUPTED
) ?
3780 SHUTDOWN_CORRUPT_INCORE
:
3781 SHUTDOWN_META_IO_ERROR
);
3784 xfs_trans_log_efd_extent(ntp
, efd
, free
->xbfi_startblock
,
3785 free
->xbfi_blockcount
);
3786 xfs_bmap_del_free(flist
, NULL
, free
);
3792 * Free up any items left in the list.
3796 xfs_bmap_free_t
*flist
) /* list of bmap_free_items */
3798 xfs_bmap_free_item_t
*free
; /* free list item */
3799 xfs_bmap_free_item_t
*next
;
3801 if (flist
->xbf_count
== 0)
3803 ASSERT(flist
->xbf_first
!= NULL
);
3804 for (free
= flist
->xbf_first
; free
; free
= next
) {
3805 next
= free
->xbfi_next
;
3806 xfs_bmap_del_free(flist
, NULL
, free
);
3808 ASSERT(flist
->xbf_count
== 0);
3812 * Returns the file-relative block number of the first unused block(s)
3813 * in the file with at least "len" logically contiguous blocks free.
3814 * This is the lowest-address hole if the file has holes, else the first block
3815 * past the end of file.
3816 * Return 0 if the file is currently local (in-inode).
3819 xfs_bmap_first_unused(
3820 xfs_trans_t
*tp
, /* transaction pointer */
3821 xfs_inode_t
*ip
, /* incore inode */
3822 xfs_extlen_t len
, /* size of hole to find */
3823 xfs_fileoff_t
*first_unused
, /* unused block */
3824 int whichfork
) /* data or attr fork */
3826 int error
; /* error return value */
3827 int idx
; /* extent record index */
3828 xfs_ifork_t
*ifp
; /* inode fork pointer */
3829 xfs_fileoff_t lastaddr
; /* last block number seen */
3830 xfs_fileoff_t lowest
; /* lowest useful block */
3831 xfs_fileoff_t max
; /* starting useful block */
3832 xfs_fileoff_t off
; /* offset for this block */
3833 xfs_extnum_t nextents
; /* number of extent entries */
3835 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
3836 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
3837 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
3838 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
3842 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3843 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
3844 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
3846 lowest
= *first_unused
;
3847 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3848 for (idx
= 0, lastaddr
= 0, max
= lowest
; idx
< nextents
; idx
++) {
3849 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, idx
);
3850 off
= xfs_bmbt_get_startoff(ep
);
3852 * See if the hole before this extent will work.
3854 if (off
>= lowest
+ len
&& off
- max
>= len
) {
3855 *first_unused
= max
;
3858 lastaddr
= off
+ xfs_bmbt_get_blockcount(ep
);
3859 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
3861 *first_unused
= max
;
3866 * Returns the file-relative block number of the last block + 1 before
3867 * last_block (input value) in the file.
3868 * This is not based on i_size, it is based on the extent records.
3869 * Returns 0 for local files, as they do not have extent records.
3872 xfs_bmap_last_before(
3873 xfs_trans_t
*tp
, /* transaction pointer */
3874 xfs_inode_t
*ip
, /* incore inode */
3875 xfs_fileoff_t
*last_block
, /* last block */
3876 int whichfork
) /* data or attr fork */
3878 xfs_fileoff_t bno
; /* input file offset */
3879 int eof
; /* hit end of file */
3880 xfs_bmbt_rec_host_t
*ep
; /* pointer to last extent */
3881 int error
; /* error return value */
3882 xfs_bmbt_irec_t got
; /* current extent value */
3883 xfs_ifork_t
*ifp
; /* inode fork pointer */
3884 xfs_extnum_t lastx
; /* last extent used */
3885 xfs_bmbt_irec_t prev
; /* previous extent value */
3887 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
3888 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
3889 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
)
3890 return XFS_ERROR(EIO
);
3891 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
3895 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3896 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
3897 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
3899 bno
= *last_block
- 1;
3900 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
3902 if (eof
|| xfs_bmbt_get_startoff(ep
) > bno
) {
3903 if (prev
.br_startoff
== NULLFILEOFF
)
3906 *last_block
= prev
.br_startoff
+ prev
.br_blockcount
;
3909 * Otherwise *last_block is already the right answer.
3915 xfs_bmap_last_extent(
3916 struct xfs_trans
*tp
,
3917 struct xfs_inode
*ip
,
3919 struct xfs_bmbt_irec
*rec
,
3922 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3926 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
3927 error
= xfs_iread_extents(tp
, ip
, whichfork
);
3932 nextents
= ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
);
3933 if (nextents
== 0) {
3938 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, nextents
- 1), rec
);
3944 * Check the last inode extent to determine whether this allocation will result
3945 * in blocks being allocated at the end of the file. When we allocate new data
3946 * blocks at the end of the file which do not start at the previous data block,
3947 * we will try to align the new blocks at stripe unit boundaries.
3949 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3950 * at, or past the EOF.
3954 struct xfs_bmalloca
*bma
,
3957 struct xfs_bmbt_irec rec
;
3962 error
= xfs_bmap_last_extent(NULL
, bma
->ip
, whichfork
, &rec
,
3964 if (error
|| is_empty
)
3968 * Check if we are allocation or past the last extent, or at least into
3969 * the last delayed allocated extent.
3971 bma
->aeof
= bma
->offset
>= rec
.br_startoff
+ rec
.br_blockcount
||
3972 (bma
->offset
>= rec
.br_startoff
&&
3973 isnullstartblock(rec
.br_startblock
));
3978 * Check if the endoff is outside the last extent. If so the caller will grow
3979 * the allocation to a stripe unit boundary. All offsets are considered outside
3980 * the end of file for an empty fork, so 1 is returned in *eof in that case.
3984 struct xfs_inode
*ip
,
3985 xfs_fileoff_t endoff
,
3989 struct xfs_bmbt_irec rec
;
3992 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, eof
);
3996 *eof
= endoff
>= rec
.br_startoff
+ rec
.br_blockcount
;
4001 * Returns the file-relative block number of the first block past eof in
4002 * the file. This is not based on i_size, it is based on the extent records.
4003 * Returns 0 for local files, as they do not have extent records.
4006 xfs_bmap_last_offset(
4007 struct xfs_trans
*tp
,
4008 struct xfs_inode
*ip
,
4009 xfs_fileoff_t
*last_block
,
4012 struct xfs_bmbt_irec rec
;
4018 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
)
4021 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
4022 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4023 return XFS_ERROR(EIO
);
4025 error
= xfs_bmap_last_extent(NULL
, ip
, whichfork
, &rec
, &is_empty
);
4026 if (error
|| is_empty
)
4029 *last_block
= rec
.br_startoff
+ rec
.br_blockcount
;
4034 * Returns whether the selected fork of the inode has exactly one
4035 * block or not. For the data fork we check this matches di_size,
4036 * implying the file's range is 0..bsize-1.
4038 int /* 1=>1 block, 0=>otherwise */
4040 xfs_inode_t
*ip
, /* incore inode */
4041 int whichfork
) /* data or attr fork */
4043 xfs_bmbt_rec_host_t
*ep
; /* ptr to fork's extent */
4044 xfs_ifork_t
*ifp
; /* inode fork pointer */
4045 int rval
; /* return value */
4046 xfs_bmbt_irec_t s
; /* internal version of extent */
4049 if (whichfork
== XFS_DATA_FORK
)
4050 return XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
;
4052 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
4054 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
4056 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4057 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
4058 ep
= xfs_iext_get_ext(ifp
, 0);
4059 xfs_bmbt_get_all(ep
, &s
);
4060 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
4061 if (rval
&& whichfork
== XFS_DATA_FORK
)
4062 ASSERT(XFS_ISIZE(ip
) == ip
->i_mount
->m_sb
.sb_blocksize
);
4067 xfs_bmap_sanity_check(
4068 struct xfs_mount
*mp
,
4072 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4074 if (block
->bb_magic
!= cpu_to_be32(XFS_BMAP_MAGIC
) ||
4075 be16_to_cpu(block
->bb_level
) != level
||
4076 be16_to_cpu(block
->bb_numrecs
) == 0 ||
4077 be16_to_cpu(block
->bb_numrecs
) > mp
->m_bmap_dmxr
[level
!= 0])
4083 * Read in the extents to if_extents.
4084 * All inode fields are set up by caller, we just traverse the btree
4085 * and copy the records in. If the file system cannot contain unwritten
4086 * extents, the records are checked for no "state" flags.
4089 xfs_bmap_read_extents(
4090 xfs_trans_t
*tp
, /* transaction pointer */
4091 xfs_inode_t
*ip
, /* incore inode */
4092 int whichfork
) /* data or attr fork */
4094 struct xfs_btree_block
*block
; /* current btree block */
4095 xfs_fsblock_t bno
; /* block # of "block" */
4096 xfs_buf_t
*bp
; /* buffer for "block" */
4097 int error
; /* error return value */
4098 xfs_exntfmt_t exntf
; /* XFS_EXTFMT_NOSTATE, if checking */
4099 xfs_extnum_t i
, j
; /* index into the extents list */
4100 xfs_ifork_t
*ifp
; /* fork structure */
4101 int level
; /* btree level, for checking */
4102 xfs_mount_t
*mp
; /* file system mount structure */
4103 __be64
*pp
; /* pointer to block address */
4105 xfs_extnum_t room
; /* number of entries there's room for */
4109 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4110 exntf
= (whichfork
!= XFS_DATA_FORK
) ? XFS_EXTFMT_NOSTATE
:
4111 XFS_EXTFMT_INODE(ip
);
4112 block
= ifp
->if_broot
;
4114 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4116 level
= be16_to_cpu(block
->bb_level
);
4118 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
4119 bno
= be64_to_cpu(*pp
);
4120 ASSERT(bno
!= NULLDFSBNO
);
4121 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
4122 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
4124 * Go down the tree until leaf level is reached, following the first
4125 * pointer (leftmost) at each level.
4127 while (level
-- > 0) {
4128 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
4129 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
4132 block
= XFS_BUF_TO_BLOCK(bp
);
4133 XFS_WANT_CORRUPTED_GOTO(
4134 xfs_bmap_sanity_check(mp
, bp
, level
),
4138 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
4139 bno
= be64_to_cpu(*pp
);
4140 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
4141 xfs_trans_brelse(tp
, bp
);
4144 * Here with bp and block set to the leftmost leaf node in the tree.
4146 room
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4149 * Loop over all leaf nodes. Copy information to the extent records.
4152 xfs_bmbt_rec_t
*frp
;
4153 xfs_fsblock_t nextbno
;
4154 xfs_extnum_t num_recs
;
4157 num_recs
= xfs_btree_get_numrecs(block
);
4158 if (unlikely(i
+ num_recs
> room
)) {
4159 ASSERT(i
+ num_recs
<= room
);
4160 xfs_warn(ip
->i_mount
,
4161 "corrupt dinode %Lu, (btree extents).",
4162 (unsigned long long) ip
->i_ino
);
4163 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4164 XFS_ERRLEVEL_LOW
, ip
->i_mount
, block
);
4167 XFS_WANT_CORRUPTED_GOTO(
4168 xfs_bmap_sanity_check(mp
, bp
, 0),
4171 * Read-ahead the next leaf block, if any.
4173 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
4174 if (nextbno
!= NULLFSBLOCK
)
4175 xfs_btree_reada_bufl(mp
, nextbno
, 1,
4178 * Copy records into the extent records.
4180 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
4182 for (j
= 0; j
< num_recs
; j
++, i
++, frp
++) {
4183 xfs_bmbt_rec_host_t
*trp
= xfs_iext_get_ext(ifp
, i
);
4184 trp
->l0
= be64_to_cpu(frp
->l0
);
4185 trp
->l1
= be64_to_cpu(frp
->l1
);
4187 if (exntf
== XFS_EXTFMT_NOSTATE
) {
4189 * Check all attribute bmap btree records and
4190 * any "older" data bmap btree records for a
4191 * set bit in the "extent flag" position.
4193 if (unlikely(xfs_check_nostate_extents(ifp
,
4194 start
, num_recs
))) {
4195 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4201 xfs_trans_brelse(tp
, bp
);
4204 * If we've reached the end, stop.
4206 if (bno
== NULLFSBLOCK
)
4208 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
4209 XFS_BMAP_BTREE_REF
, &xfs_bmbt_buf_ops
);
4212 block
= XFS_BUF_TO_BLOCK(bp
);
4214 ASSERT(i
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
4215 ASSERT(i
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
4216 XFS_BMAP_TRACE_EXLIST(ip
, i
, whichfork
);
4219 xfs_trans_brelse(tp
, bp
);
4220 return XFS_ERROR(EFSCORRUPTED
);
4225 * Add bmap trace insert entries for all the contents of the extent records.
4228 xfs_bmap_trace_exlist(
4229 xfs_inode_t
*ip
, /* incore inode pointer */
4230 xfs_extnum_t cnt
, /* count of entries in the list */
4231 int whichfork
, /* data or attr fork */
4232 unsigned long caller_ip
)
4234 xfs_extnum_t idx
; /* extent record index */
4235 xfs_ifork_t
*ifp
; /* inode fork pointer */
4238 if (whichfork
== XFS_ATTR_FORK
)
4239 state
|= BMAP_ATTRFORK
;
4241 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4242 ASSERT(cnt
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
4243 for (idx
= 0; idx
< cnt
; idx
++)
4244 trace_xfs_extlist(ip
, idx
, whichfork
, caller_ip
);
4248 * Validate that the bmbt_irecs being returned from bmapi are valid
4249 * given the callers original parameters. Specifically check the
4250 * ranges of the returned irecs to ensure that they only extent beyond
4251 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4254 xfs_bmap_validate_ret(
4258 xfs_bmbt_irec_t
*mval
,
4262 int i
; /* index to map values */
4264 ASSERT(ret_nmap
<= nmap
);
4266 for (i
= 0; i
< ret_nmap
; i
++) {
4267 ASSERT(mval
[i
].br_blockcount
> 0);
4268 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
4269 ASSERT(mval
[i
].br_startoff
>= bno
);
4270 ASSERT(mval
[i
].br_blockcount
<= len
);
4271 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
4274 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
4275 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
4279 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
4280 mval
[i
].br_startoff
);
4281 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
4282 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
4283 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
4284 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
4291 * Trim the returned map to the required bounds
4295 struct xfs_bmbt_irec
*mval
,
4296 struct xfs_bmbt_irec
*got
,
4304 if ((flags
& XFS_BMAPI_ENTIRE
) ||
4305 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
4307 if (isnullstartblock(got
->br_startblock
))
4308 mval
->br_startblock
= DELAYSTARTBLOCK
;
4314 ASSERT((*bno
>= obno
) || (n
== 0));
4316 mval
->br_startoff
= *bno
;
4317 if (isnullstartblock(got
->br_startblock
))
4318 mval
->br_startblock
= DELAYSTARTBLOCK
;
4320 mval
->br_startblock
= got
->br_startblock
+
4321 (*bno
- got
->br_startoff
);
4323 * Return the minimum of what we got and what we asked for for
4324 * the length. We can use the len variable here because it is
4325 * modified below and we could have been there before coming
4326 * here if the first part of the allocation didn't overlap what
4329 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
4330 got
->br_blockcount
- (*bno
- got
->br_startoff
));
4331 mval
->br_state
= got
->br_state
;
4332 ASSERT(mval
->br_blockcount
<= len
);
4337 * Update and validate the extent map to return
4340 xfs_bmapi_update_map(
4341 struct xfs_bmbt_irec
**map
,
4349 xfs_bmbt_irec_t
*mval
= *map
;
4351 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
4352 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
4353 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
4354 (mval
->br_startoff
< obno
));
4356 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
4358 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
4359 /* update previous map with new information */
4360 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
4361 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
4362 ASSERT(mval
->br_state
== mval
[-1].br_state
);
4363 mval
[-1].br_blockcount
= mval
->br_blockcount
;
4364 mval
[-1].br_state
= mval
->br_state
;
4365 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
4366 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
4367 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
4368 mval
->br_startblock
== mval
[-1].br_startblock
+
4369 mval
[-1].br_blockcount
&&
4370 ((flags
& XFS_BMAPI_IGSTATE
) ||
4371 mval
[-1].br_state
== mval
->br_state
)) {
4372 ASSERT(mval
->br_startoff
==
4373 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
4374 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4375 } else if (*n
> 0 &&
4376 mval
->br_startblock
== DELAYSTARTBLOCK
&&
4377 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
4378 mval
->br_startoff
==
4379 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
4380 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4381 mval
[-1].br_state
= mval
->br_state
;
4382 } else if (!((*n
== 0) &&
4383 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
4392 * Map file blocks to filesystem blocks without allocation.
4396 struct xfs_inode
*ip
,
4399 struct xfs_bmbt_irec
*mval
,
4403 struct xfs_mount
*mp
= ip
->i_mount
;
4404 struct xfs_ifork
*ifp
;
4405 struct xfs_bmbt_irec got
;
4406 struct xfs_bmbt_irec prev
;
4413 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4414 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4417 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
4418 XFS_BMAPI_IGSTATE
)));
4420 if (unlikely(XFS_TEST_ERROR(
4421 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4422 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4423 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4424 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
4425 return XFS_ERROR(EFSCORRUPTED
);
4428 if (XFS_FORCED_SHUTDOWN(mp
))
4429 return XFS_ERROR(EIO
);
4431 XFS_STATS_INC(xs_blk_mapr
);
4433 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4435 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4436 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
4441 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
, &prev
);
4445 while (bno
< end
&& n
< *nmap
) {
4446 /* Reading past eof, act as though there's a hole up to end. */
4448 got
.br_startoff
= end
;
4449 if (got
.br_startoff
> bno
) {
4450 /* Reading in a hole. */
4451 mval
->br_startoff
= bno
;
4452 mval
->br_startblock
= HOLESTARTBLOCK
;
4453 mval
->br_blockcount
=
4454 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4455 mval
->br_state
= XFS_EXT_NORM
;
4456 bno
+= mval
->br_blockcount
;
4457 len
-= mval
->br_blockcount
;
4463 /* set up the extent map to return. */
4464 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4465 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4467 /* If we're done, stop now. */
4468 if (bno
>= end
|| n
>= *nmap
)
4471 /* Else go on to the next record. */
4472 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4473 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4482 xfs_bmapi_reserve_delalloc(
4483 struct xfs_inode
*ip
,
4486 struct xfs_bmbt_irec
*got
,
4487 struct xfs_bmbt_irec
*prev
,
4488 xfs_extnum_t
*lastx
,
4491 struct xfs_mount
*mp
= ip
->i_mount
;
4492 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4494 xfs_extlen_t indlen
;
4495 char rt
= XFS_IS_REALTIME_INODE(ip
);
4499 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
);
4501 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4503 /* Figure out the extent size, adjust alen */
4504 extsz
= xfs_get_extsz_hint(ip
);
4507 * Make sure we don't exceed a single extent length when we
4508 * align the extent by reducing length we are going to
4509 * allocate by the maximum amount extent size aligment may
4512 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
- (2 * extsz
- 1));
4513 error
= xfs_bmap_extsize_align(mp
, got
, prev
, extsz
, rt
, eof
,
4514 1, 0, &aoff
, &alen
);
4519 extsz
= alen
/ mp
->m_sb
.sb_rextsize
;
4522 * Make a transaction-less quota reservation for delayed allocation
4523 * blocks. This number gets adjusted later. We return if we haven't
4524 * allocated blocks already inside this loop.
4526 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4527 rt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4532 * Split changing sb for alen and indlen since they could be coming
4533 * from different places.
4535 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4539 error
= xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
4540 -((int64_t)extsz
), 0);
4542 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4543 -((int64_t)alen
), 0);
4547 goto out_unreserve_quota
;
4549 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4550 -((int64_t)indlen
), 0);
4552 goto out_unreserve_blocks
;
4555 ip
->i_delayed_blks
+= alen
;
4557 got
->br_startoff
= aoff
;
4558 got
->br_startblock
= nullstartblock(indlen
);
4559 got
->br_blockcount
= alen
;
4560 got
->br_state
= XFS_EXT_NORM
;
4561 xfs_bmap_add_extent_hole_delay(ip
, lastx
, got
);
4564 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4565 * might have merged it into one of the neighbouring ones.
4567 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
), got
);
4569 ASSERT(got
->br_startoff
<= aoff
);
4570 ASSERT(got
->br_startoff
+ got
->br_blockcount
>= aoff
+ alen
);
4571 ASSERT(isnullstartblock(got
->br_startblock
));
4572 ASSERT(got
->br_state
== XFS_EXT_NORM
);
4575 out_unreserve_blocks
:
4577 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
, extsz
, 0);
4579 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, alen
, 0);
4580 out_unreserve_quota
:
4581 if (XFS_IS_QUOTA_ON(mp
))
4582 xfs_trans_unreserve_quota_nblks(NULL
, ip
, (long)alen
, 0, rt
?
4583 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4588 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4592 struct xfs_inode
*ip
, /* incore inode */
4593 xfs_fileoff_t bno
, /* starting file offs. mapped */
4594 xfs_filblks_t len
, /* length to map in file */
4595 struct xfs_bmbt_irec
*mval
, /* output: map values */
4596 int *nmap
, /* i/o: mval size/count */
4597 int flags
) /* XFS_BMAPI_... */
4599 struct xfs_mount
*mp
= ip
->i_mount
;
4600 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4601 struct xfs_bmbt_irec got
; /* current file extent record */
4602 struct xfs_bmbt_irec prev
; /* previous file extent record */
4603 xfs_fileoff_t obno
; /* old block number (offset) */
4604 xfs_fileoff_t end
; /* end of mapped file region */
4605 xfs_extnum_t lastx
; /* last useful extent number */
4606 int eof
; /* we've hit the end of extents */
4607 int n
= 0; /* current extent index */
4611 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4612 ASSERT(!(flags
& ~XFS_BMAPI_ENTIRE
));
4614 if (unlikely(XFS_TEST_ERROR(
4615 (XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_EXTENTS
&&
4616 XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_BTREE
),
4617 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4618 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW
, mp
);
4619 return XFS_ERROR(EFSCORRUPTED
);
4622 if (XFS_FORCED_SHUTDOWN(mp
))
4623 return XFS_ERROR(EIO
);
4625 XFS_STATS_INC(xs_blk_mapw
);
4627 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4628 error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
);
4633 xfs_bmap_search_extents(ip
, bno
, XFS_DATA_FORK
, &eof
, &lastx
, &got
, &prev
);
4637 while (bno
< end
&& n
< *nmap
) {
4638 if (eof
|| got
.br_startoff
> bno
) {
4639 error
= xfs_bmapi_reserve_delalloc(ip
, bno
, len
, &got
,
4640 &prev
, &lastx
, eof
);
4650 /* set up the extent map to return. */
4651 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4652 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4654 /* If we're done, stop now. */
4655 if (bno
>= end
|| n
>= *nmap
)
4658 /* Else go on to the next record. */
4660 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4661 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4672 __xfs_bmapi_allocate(
4673 struct xfs_bmalloca
*bma
)
4675 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4676 int whichfork
= (bma
->flags
& XFS_BMAPI_ATTRFORK
) ?
4677 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4678 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4679 int tmp_logflags
= 0;
4683 ASSERT(bma
->length
> 0);
4685 rt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(bma
->ip
);
4688 * For the wasdelay case, we could also just allocate the stuff asked
4689 * for in this bmap call but that wouldn't be as good.
4692 bma
->length
= (xfs_extlen_t
)bma
->got
.br_blockcount
;
4693 bma
->offset
= bma
->got
.br_startoff
;
4694 if (bma
->idx
!= NULLEXTNUM
&& bma
->idx
) {
4695 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
- 1),
4699 bma
->length
= XFS_FILBLKS_MIN(bma
->length
, MAXEXTLEN
);
4701 bma
->length
= XFS_FILBLKS_MIN(bma
->length
,
4702 bma
->got
.br_startoff
- bma
->offset
);
4706 * Indicate if this is the first user data in the file, or just any
4709 if (!(bma
->flags
& XFS_BMAPI_METADATA
)) {
4710 bma
->userdata
= (bma
->offset
== 0) ?
4711 XFS_ALLOC_INITIAL_USER_DATA
: XFS_ALLOC_USERDATA
;
4714 bma
->minlen
= (bma
->flags
& XFS_BMAPI_CONTIG
) ? bma
->length
: 1;
4717 * Only want to do the alignment at the eof if it is userdata and
4718 * allocation length is larger than a stripe unit.
4720 if (mp
->m_dalign
&& bma
->length
>= mp
->m_dalign
&&
4721 !(bma
->flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4722 error
= xfs_bmap_isaeof(bma
, whichfork
);
4727 error
= xfs_bmap_alloc(bma
);
4731 if (bma
->flist
->xbf_low
)
4734 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4735 if (bma
->blkno
== NULLFSBLOCK
)
4737 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4738 bma
->cur
= xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4739 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4740 bma
->cur
->bc_private
.b
.flist
= bma
->flist
;
4743 * Bump the number of extents we've allocated
4749 bma
->cur
->bc_private
.b
.flags
=
4750 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4752 bma
->got
.br_startoff
= bma
->offset
;
4753 bma
->got
.br_startblock
= bma
->blkno
;
4754 bma
->got
.br_blockcount
= bma
->length
;
4755 bma
->got
.br_state
= XFS_EXT_NORM
;
4758 * A wasdelay extent has been initialized, so shouldn't be flagged
4761 if (!bma
->wasdel
&& (bma
->flags
& XFS_BMAPI_PREALLOC
) &&
4762 xfs_sb_version_hasextflgbit(&mp
->m_sb
))
4763 bma
->got
.br_state
= XFS_EXT_UNWRITTEN
;
4766 error
= xfs_bmap_add_extent_delay_real(bma
);
4768 error
= xfs_bmap_add_extent_hole_real(bma
, whichfork
);
4770 bma
->logflags
|= tmp_logflags
;
4775 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4776 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4777 * the neighbouring ones.
4779 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4781 ASSERT(bma
->got
.br_startoff
<= bma
->offset
);
4782 ASSERT(bma
->got
.br_startoff
+ bma
->got
.br_blockcount
>=
4783 bma
->offset
+ bma
->length
);
4784 ASSERT(bma
->got
.br_state
== XFS_EXT_NORM
||
4785 bma
->got
.br_state
== XFS_EXT_UNWRITTEN
);
4790 xfs_bmapi_allocate_worker(
4791 struct work_struct
*work
)
4793 struct xfs_bmalloca
*args
= container_of(work
,
4794 struct xfs_bmalloca
, work
);
4795 unsigned long pflags
;
4797 /* we are in a transaction context here */
4798 current_set_flags_nested(&pflags
, PF_FSTRANS
);
4800 args
->result
= __xfs_bmapi_allocate(args
);
4801 complete(args
->done
);
4803 current_restore_flags_nested(&pflags
, PF_FSTRANS
);
4807 * Some allocation requests often come in with little stack to work on. Push
4808 * them off to a worker thread so there is lots of stack to use. Otherwise just
4809 * call directly to avoid the context switch overhead here.
4813 struct xfs_bmalloca
*args
)
4815 DECLARE_COMPLETION_ONSTACK(done
);
4817 if (!args
->stack_switch
)
4818 return __xfs_bmapi_allocate(args
);
4822 INIT_WORK_ONSTACK(&args
->work
, xfs_bmapi_allocate_worker
);
4823 queue_work(xfs_alloc_wq
, &args
->work
);
4824 wait_for_completion(&done
);
4825 return args
->result
;
4829 xfs_bmapi_convert_unwritten(
4830 struct xfs_bmalloca
*bma
,
4831 struct xfs_bmbt_irec
*mval
,
4835 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4836 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4837 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4838 int tmp_logflags
= 0;
4841 /* check if we need to do unwritten->real conversion */
4842 if (mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4843 (flags
& XFS_BMAPI_PREALLOC
))
4846 /* check if we need to do real->unwritten conversion */
4847 if (mval
->br_state
== XFS_EXT_NORM
&&
4848 (flags
& (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
)) !=
4849 (XFS_BMAPI_PREALLOC
| XFS_BMAPI_CONVERT
))
4853 * Modify (by adding) the state flag, if writing.
4855 ASSERT(mval
->br_blockcount
<= len
);
4856 if ((ifp
->if_flags
& XFS_IFBROOT
) && !bma
->cur
) {
4857 bma
->cur
= xfs_bmbt_init_cursor(bma
->ip
->i_mount
, bma
->tp
,
4858 bma
->ip
, whichfork
);
4859 bma
->cur
->bc_private
.b
.firstblock
= *bma
->firstblock
;
4860 bma
->cur
->bc_private
.b
.flist
= bma
->flist
;
4862 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4863 ? XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
4865 error
= xfs_bmap_add_extent_unwritten_real(bma
->tp
, bma
->ip
, &bma
->idx
,
4866 &bma
->cur
, mval
, bma
->firstblock
, bma
->flist
,
4868 bma
->logflags
|= tmp_logflags
;
4873 * Update our extent pointer, given that
4874 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4875 * of the neighbouring ones.
4877 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
->idx
), &bma
->got
);
4880 * We may have combined previously unwritten space with written space,
4881 * so generate another request.
4883 if (mval
->br_blockcount
< len
)
4889 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4890 * extent state if necessary. Details behaviour is controlled by the flags
4891 * parameter. Only allocates blocks from a single allocation group, to avoid
4894 * The returned value in "firstblock" from the first call in a transaction
4895 * must be remembered and presented to subsequent calls in "firstblock".
4896 * An upper bound for the number of blocks to be allocated is supplied to
4897 * the first call in "total"; if no allocation group has that many free
4898 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4902 struct xfs_trans
*tp
, /* transaction pointer */
4903 struct xfs_inode
*ip
, /* incore inode */
4904 xfs_fileoff_t bno
, /* starting file offs. mapped */
4905 xfs_filblks_t len
, /* length to map in file */
4906 int flags
, /* XFS_BMAPI_... */
4907 xfs_fsblock_t
*firstblock
, /* first allocated block
4908 controls a.g. for allocs */
4909 xfs_extlen_t total
, /* total blocks needed */
4910 struct xfs_bmbt_irec
*mval
, /* output: map values */
4911 int *nmap
, /* i/o: mval size/count */
4912 struct xfs_bmap_free
*flist
) /* i/o: list extents to free */
4914 struct xfs_mount
*mp
= ip
->i_mount
;
4915 struct xfs_ifork
*ifp
;
4916 struct xfs_bmalloca bma
= { 0 }; /* args for xfs_bmap_alloc */
4917 xfs_fileoff_t end
; /* end of mapped file region */
4918 int eof
; /* after the end of extents */
4919 int error
; /* error return */
4920 int n
; /* current extent index */
4921 xfs_fileoff_t obno
; /* old block number (offset) */
4922 int whichfork
; /* data or attr fork */
4923 char inhole
; /* current location is hole in file */
4924 char wasdelay
; /* old extent was delayed */
4927 xfs_fileoff_t orig_bno
; /* original block number value */
4928 int orig_flags
; /* original flags arg value */
4929 xfs_filblks_t orig_len
; /* original value of len arg */
4930 struct xfs_bmbt_irec
*orig_mval
; /* original value of mval */
4931 int orig_nmap
; /* original value of *nmap */
4941 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4942 ASSERT(!(flags
& XFS_BMAPI_IGSTATE
));
4946 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4947 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4949 if (unlikely(XFS_TEST_ERROR(
4950 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4951 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
4952 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
),
4953 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4954 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW
, mp
);
4955 return XFS_ERROR(EFSCORRUPTED
);
4958 if (XFS_FORCED_SHUTDOWN(mp
))
4959 return XFS_ERROR(EIO
);
4961 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4963 XFS_STATS_INC(xs_blk_mapw
);
4965 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
4967 * XXX (dgc): This assumes we are only called for inodes that
4968 * contain content neutral data in local format. Anything that
4969 * contains caller-specific data in local format that needs
4970 * transformation to move to a block format needs to do the
4971 * conversion to extent format itself.
4973 * Directory data forks and attribute forks handle this
4974 * themselves, but with the addition of metadata verifiers every
4975 * data fork in local format now contains caller specific data
4976 * and as such conversion through this function is likely to be
4979 * The only likely user of this branch is for remote symlinks,
4980 * but we cannot overwrite the data fork contents of the symlink
4981 * (EEXIST occurs higher up the stack) and so it will never go
4982 * from local format to extent format here. Hence I don't think
4983 * this branch is ever executed intentionally and we should
4984 * consider removing it and asserting that xfs_bmapi_write()
4985 * cannot be called directly on local format forks. i.e. callers
4986 * are completely responsible for local to extent format
4987 * conversion, not xfs_bmapi_write().
4989 error
= xfs_bmap_local_to_extents(tp
, ip
, firstblock
, total
,
4990 &bma
.logflags
, whichfork
,
4991 xfs_bmap_local_to_extents_init_fn
);
4996 if (*firstblock
== NULLFSBLOCK
) {
4997 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
)
4998 bma
.minleft
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
5005 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
5006 error
= xfs_iread_extents(tp
, ip
, whichfork
);
5011 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &bma
.idx
, &bma
.got
,
5022 bma
.firstblock
= firstblock
;
5024 if (flags
& XFS_BMAPI_STACK_SWITCH
)
5025 bma
.stack_switch
= 1;
5027 while (bno
< end
&& n
< *nmap
) {
5028 inhole
= eof
|| bma
.got
.br_startoff
> bno
;
5029 wasdelay
= !inhole
&& isnullstartblock(bma
.got
.br_startblock
);
5032 * First, deal with the hole before the allocated space
5033 * that we found, if any.
5035 if (inhole
|| wasdelay
) {
5037 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
5038 bma
.wasdel
= wasdelay
;
5043 * There's a 32/64 bit type mismatch between the
5044 * allocation length request (which can be 64 bits in
5045 * length) and the bma length request, which is
5046 * xfs_extlen_t and therefore 32 bits. Hence we have to
5047 * check for 32-bit overflows and handle them here.
5049 if (len
> (xfs_filblks_t
)MAXEXTLEN
)
5050 bma
.length
= MAXEXTLEN
;
5055 ASSERT(bma
.length
> 0);
5056 error
= xfs_bmapi_allocate(&bma
);
5059 if (bma
.blkno
== NULLFSBLOCK
)
5063 /* Deal with the allocated space we found. */
5064 xfs_bmapi_trim_map(mval
, &bma
.got
, &bno
, len
, obno
,
5067 /* Execute unwritten extent conversion if necessary */
5068 error
= xfs_bmapi_convert_unwritten(&bma
, mval
, len
, flags
);
5069 if (error
== EAGAIN
)
5074 /* update the extent map to return */
5075 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
5078 * If we're done, stop now. Stop when we've allocated
5079 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
5080 * the transaction may get too big.
5082 if (bno
>= end
|| n
>= *nmap
|| bma
.nallocs
>= *nmap
)
5085 /* Else go on to the next record. */
5087 if (++bma
.idx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
)) {
5088 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, bma
.idx
),
5096 * Transform from btree to extents, give it cur.
5098 if (xfs_bmap_wants_extents(ip
, whichfork
)) {
5099 int tmp_logflags
= 0;
5102 error
= xfs_bmap_btree_to_extents(tp
, ip
, bma
.cur
,
5103 &tmp_logflags
, whichfork
);
5104 bma
.logflags
|= tmp_logflags
;
5109 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
5110 XFS_IFORK_NEXTENTS(ip
, whichfork
) >
5111 XFS_IFORK_MAXEXT(ip
, whichfork
));
5115 * Log everything. Do this after conversion, there's no point in
5116 * logging the extent records if we've converted to btree format.
5118 if ((bma
.logflags
& xfs_ilog_fext(whichfork
)) &&
5119 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5120 bma
.logflags
&= ~xfs_ilog_fext(whichfork
);
5121 else if ((bma
.logflags
& xfs_ilog_fbroot(whichfork
)) &&
5122 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5123 bma
.logflags
&= ~xfs_ilog_fbroot(whichfork
);
5125 * Log whatever the flags say, even if error. Otherwise we might miss
5126 * detecting a case where the data is changed, there's an error,
5127 * and it's not logged so we don't shutdown when we should.
5130 xfs_trans_log_inode(tp
, ip
, bma
.logflags
);
5134 ASSERT(*firstblock
== NULLFSBLOCK
||
5135 XFS_FSB_TO_AGNO(mp
, *firstblock
) ==
5137 bma
.cur
->bc_private
.b
.firstblock
) ||
5139 XFS_FSB_TO_AGNO(mp
, *firstblock
) <
5141 bma
.cur
->bc_private
.b
.firstblock
)));
5142 *firstblock
= bma
.cur
->bc_private
.b
.firstblock
;
5144 xfs_btree_del_cursor(bma
.cur
,
5145 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5148 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
5154 * Unmap (remove) blocks from a file.
5155 * If nexts is nonzero then the number of extents to remove is limited to
5156 * that value. If not all extents in the block range can be removed then
5161 xfs_trans_t
*tp
, /* transaction pointer */
5162 struct xfs_inode
*ip
, /* incore inode */
5163 xfs_fileoff_t bno
, /* starting offset to unmap */
5164 xfs_filblks_t len
, /* length to unmap in file */
5165 int flags
, /* misc flags */
5166 xfs_extnum_t nexts
, /* number of extents max */
5167 xfs_fsblock_t
*firstblock
, /* first allocated block
5168 controls a.g. for allocs */
5169 xfs_bmap_free_t
*flist
, /* i/o: list extents to free */
5170 int *done
) /* set if not done yet */
5172 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
5173 xfs_bmbt_irec_t del
; /* extent being deleted */
5174 int eof
; /* is deleting at eof */
5175 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
5176 int error
; /* error return value */
5177 xfs_extnum_t extno
; /* extent number in list */
5178 xfs_bmbt_irec_t got
; /* current extent record */
5179 xfs_ifork_t
*ifp
; /* inode fork pointer */
5180 int isrt
; /* freeing in rt area */
5181 xfs_extnum_t lastx
; /* last extent index used */
5182 int logflags
; /* transaction logging flags */
5183 xfs_extlen_t mod
; /* rt extent offset */
5184 xfs_mount_t
*mp
; /* mount structure */
5185 xfs_extnum_t nextents
; /* number of file extents */
5186 xfs_bmbt_irec_t prev
; /* previous extent record */
5187 xfs_fileoff_t start
; /* first file offset deleted */
5188 int tmp_logflags
; /* partial logging flags */
5189 int wasdel
; /* was a delayed alloc extent */
5190 int whichfork
; /* data or attribute fork */
5193 trace_xfs_bunmap(ip
, bno
, len
, flags
, _RET_IP_
);
5195 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
5196 XFS_ATTR_FORK
: XFS_DATA_FORK
;
5197 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5199 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5200 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5201 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5203 return XFS_ERROR(EFSCORRUPTED
);
5206 if (XFS_FORCED_SHUTDOWN(mp
))
5207 return XFS_ERROR(EIO
);
5212 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5213 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5215 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
5216 if (nextents
== 0) {
5220 XFS_STATS_INC(xs_blk_unmap
);
5221 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5223 bno
= start
+ len
- 1;
5224 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
5228 * Check to see if the given block number is past the end of the
5229 * file, back up to the last block if so...
5232 ep
= xfs_iext_get_ext(ifp
, --lastx
);
5233 xfs_bmbt_get_all(ep
, &got
);
5234 bno
= got
.br_startoff
+ got
.br_blockcount
- 1;
5237 if (ifp
->if_flags
& XFS_IFBROOT
) {
5238 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5239 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5240 cur
->bc_private
.b
.firstblock
= *firstblock
;
5241 cur
->bc_private
.b
.flist
= flist
;
5242 cur
->bc_private
.b
.flags
= 0;
5248 * Synchronize by locking the bitmap inode.
5250 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5251 xfs_trans_ijoin(tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
5255 while (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
&& lastx
>= 0 &&
5256 (nexts
== 0 || extno
< nexts
)) {
5258 * Is the found extent after a hole in which bno lives?
5259 * Just back up to the previous extent, if so.
5261 if (got
.br_startoff
> bno
) {
5264 ep
= xfs_iext_get_ext(ifp
, lastx
);
5265 xfs_bmbt_get_all(ep
, &got
);
5268 * Is the last block of this extent before the range
5269 * we're supposed to delete? If so, we're done.
5271 bno
= XFS_FILEOFF_MIN(bno
,
5272 got
.br_startoff
+ got
.br_blockcount
- 1);
5276 * Then deal with the (possibly delayed) allocated space
5281 wasdel
= isnullstartblock(del
.br_startblock
);
5282 if (got
.br_startoff
< start
) {
5283 del
.br_startoff
= start
;
5284 del
.br_blockcount
-= start
- got
.br_startoff
;
5286 del
.br_startblock
+= start
- got
.br_startoff
;
5288 if (del
.br_startoff
+ del
.br_blockcount
> bno
+ 1)
5289 del
.br_blockcount
= bno
+ 1 - del
.br_startoff
;
5290 sum
= del
.br_startblock
+ del
.br_blockcount
;
5292 (mod
= do_mod(sum
, mp
->m_sb
.sb_rextsize
))) {
5294 * Realtime extent not lined up at the end.
5295 * The extent could have been split into written
5296 * and unwritten pieces, or we could just be
5297 * unmapping part of it. But we can't really
5298 * get rid of part of a realtime extent.
5300 if (del
.br_state
== XFS_EXT_UNWRITTEN
||
5301 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5303 * This piece is unwritten, or we're not
5304 * using unwritten extents. Skip over it.
5307 bno
-= mod
> del
.br_blockcount
?
5308 del
.br_blockcount
: mod
;
5309 if (bno
< got
.br_startoff
) {
5311 xfs_bmbt_get_all(xfs_iext_get_ext(
5317 * It's written, turn it unwritten.
5318 * This is better than zeroing it.
5320 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5321 ASSERT(xfs_trans_get_block_res(tp
) > 0);
5323 * If this spans a realtime extent boundary,
5324 * chop it back to the start of the one we end at.
5326 if (del
.br_blockcount
> mod
) {
5327 del
.br_startoff
+= del
.br_blockcount
- mod
;
5328 del
.br_startblock
+= del
.br_blockcount
- mod
;
5329 del
.br_blockcount
= mod
;
5331 del
.br_state
= XFS_EXT_UNWRITTEN
;
5332 error
= xfs_bmap_add_extent_unwritten_real(tp
, ip
,
5333 &lastx
, &cur
, &del
, firstblock
, flist
,
5339 if (isrt
&& (mod
= do_mod(del
.br_startblock
, mp
->m_sb
.sb_rextsize
))) {
5341 * Realtime extent is lined up at the end but not
5342 * at the front. We'll get rid of full extents if
5345 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5346 if (del
.br_blockcount
> mod
) {
5347 del
.br_blockcount
-= mod
;
5348 del
.br_startoff
+= mod
;
5349 del
.br_startblock
+= mod
;
5350 } else if ((del
.br_startoff
== start
&&
5351 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5352 xfs_trans_get_block_res(tp
) == 0)) ||
5353 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5355 * Can't make it unwritten. There isn't
5356 * a full extent here so just skip it.
5358 ASSERT(bno
>= del
.br_blockcount
);
5359 bno
-= del
.br_blockcount
;
5360 if (got
.br_startoff
> bno
) {
5362 ep
= xfs_iext_get_ext(ifp
,
5364 xfs_bmbt_get_all(ep
, &got
);
5368 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5370 * This one is already unwritten.
5371 * It must have a written left neighbor.
5372 * Unwrite the killed part of that one and
5376 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
,
5378 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5379 ASSERT(!isnullstartblock(prev
.br_startblock
));
5380 ASSERT(del
.br_startblock
==
5381 prev
.br_startblock
+ prev
.br_blockcount
);
5382 if (prev
.br_startoff
< start
) {
5383 mod
= start
- prev
.br_startoff
;
5384 prev
.br_blockcount
-= mod
;
5385 prev
.br_startblock
+= mod
;
5386 prev
.br_startoff
= start
;
5388 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5390 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5391 ip
, &lastx
, &cur
, &prev
,
5392 firstblock
, flist
, &logflags
);
5397 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5398 del
.br_state
= XFS_EXT_UNWRITTEN
;
5399 error
= xfs_bmap_add_extent_unwritten_real(tp
,
5400 ip
, &lastx
, &cur
, &del
,
5401 firstblock
, flist
, &logflags
);
5408 ASSERT(startblockval(del
.br_startblock
) > 0);
5409 /* Update realtime/data freespace, unreserve quota */
5411 xfs_filblks_t rtexts
;
5413 rtexts
= XFS_FSB_TO_B(mp
, del
.br_blockcount
);
5414 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
5415 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
5416 (int64_t)rtexts
, 0);
5417 (void)xfs_trans_reserve_quota_nblks(NULL
,
5418 ip
, -((long)del
.br_blockcount
), 0,
5419 XFS_QMOPT_RES_RTBLKS
);
5421 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
5422 (int64_t)del
.br_blockcount
, 0);
5423 (void)xfs_trans_reserve_quota_nblks(NULL
,
5424 ip
, -((long)del
.br_blockcount
), 0,
5425 XFS_QMOPT_RES_REGBLKS
);
5427 ip
->i_delayed_blks
-= del
.br_blockcount
;
5429 cur
->bc_private
.b
.flags
|=
5430 XFS_BTCUR_BPRV_WASDEL
;
5432 cur
->bc_private
.b
.flags
&= ~XFS_BTCUR_BPRV_WASDEL
;
5434 * If it's the case where the directory code is running
5435 * with no block reservation, and the deleted block is in
5436 * the middle of its extent, and the resulting insert
5437 * of an extent would cause transformation to btree format,
5438 * then reject it. The calling code will then swap
5439 * blocks around instead.
5440 * We have to do this now, rather than waiting for the
5441 * conversion to btree format, since the transaction
5444 if (!wasdel
&& xfs_trans_get_block_res(tp
) == 0 &&
5445 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5446 XFS_IFORK_NEXTENTS(ip
, whichfork
) >= /* Note the >= */
5447 XFS_IFORK_MAXEXT(ip
, whichfork
) &&
5448 del
.br_startoff
> got
.br_startoff
&&
5449 del
.br_startoff
+ del
.br_blockcount
<
5450 got
.br_startoff
+ got
.br_blockcount
) {
5451 error
= XFS_ERROR(ENOSPC
);
5454 error
= xfs_bmap_del_extent(ip
, tp
, &lastx
, flist
, cur
, &del
,
5455 &tmp_logflags
, whichfork
);
5456 logflags
|= tmp_logflags
;
5459 bno
= del
.br_startoff
- 1;
5462 * If not done go on to the next (previous) record.
5464 if (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
) {
5466 ep
= xfs_iext_get_ext(ifp
, lastx
);
5467 if (xfs_bmbt_get_startoff(ep
) > bno
) {
5469 ep
= xfs_iext_get_ext(ifp
,
5472 xfs_bmbt_get_all(ep
, &got
);
5477 *done
= bno
== (xfs_fileoff_t
)-1 || bno
< start
|| lastx
< 0;
5480 * Convert to a btree if necessary.
5482 if (xfs_bmap_needs_btree(ip
, whichfork
)) {
5483 ASSERT(cur
== NULL
);
5484 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
,
5485 &cur
, 0, &tmp_logflags
, whichfork
);
5486 logflags
|= tmp_logflags
;
5491 * transform from btree to extents, give it cur
5493 else if (xfs_bmap_wants_extents(ip
, whichfork
)) {
5494 ASSERT(cur
!= NULL
);
5495 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &tmp_logflags
,
5497 logflags
|= tmp_logflags
;
5502 * transform from extents to local?
5507 * Log everything. Do this after conversion, there's no point in
5508 * logging the extent records if we've converted to btree format.
5510 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5511 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5512 logflags
&= ~xfs_ilog_fext(whichfork
);
5513 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5514 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5515 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5517 * Log inode even in the error case, if the transaction
5518 * is dirty we'll need to shut down the filesystem.
5521 xfs_trans_log_inode(tp
, ip
, logflags
);
5524 *firstblock
= cur
->bc_private
.b
.firstblock
;
5525 cur
->bc_private
.b
.allocated
= 0;
5527 xfs_btree_del_cursor(cur
,
5528 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5534 * returns 1 for success, 0 if we failed to map the extent.
5537 xfs_getbmapx_fix_eof_hole(
5538 xfs_inode_t
*ip
, /* xfs incore inode pointer */
5539 struct getbmapx
*out
, /* output structure */
5540 int prealloced
, /* this is a file with
5541 * preallocated data space */
5542 __int64_t end
, /* last block requested */
5543 xfs_fsblock_t startblock
)
5546 xfs_mount_t
*mp
; /* file system mount point */
5547 xfs_ifork_t
*ifp
; /* inode fork pointer */
5548 xfs_extnum_t lastx
; /* last extent pointer */
5549 xfs_fileoff_t fileblock
;
5551 if (startblock
== HOLESTARTBLOCK
) {
5553 out
->bmv_block
= -1;
5554 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)));
5555 fixlen
-= out
->bmv_offset
;
5556 if (prealloced
&& out
->bmv_offset
+ out
->bmv_length
== end
) {
5557 /* Came to hole at EOF. Trim it. */
5560 out
->bmv_length
= fixlen
;
5563 if (startblock
== DELAYSTARTBLOCK
)
5564 out
->bmv_block
= -2;
5566 out
->bmv_block
= xfs_fsb_to_db(ip
, startblock
);
5567 fileblock
= XFS_BB_TO_FSB(ip
->i_mount
, out
->bmv_offset
);
5568 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
5569 if (xfs_iext_bno_to_ext(ifp
, fileblock
, &lastx
) &&
5570 (lastx
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))-1))
5571 out
->bmv_oflags
|= BMV_OF_LAST
;
5578 * Get inode's extents as described in bmv, and format for output.
5579 * Calls formatter to fill the user's buffer until all extents
5580 * are mapped, until the passed-in bmv->bmv_count slots have
5581 * been filled, or until the formatter short-circuits the loop,
5582 * if it is tracking filled-in extents on its own.
5584 int /* error code */
5587 struct getbmapx
*bmv
, /* user bmap structure */
5588 xfs_bmap_format_t formatter
, /* format to user */
5589 void *arg
) /* formatter arg */
5591 __int64_t bmvend
; /* last block requested */
5592 int error
= 0; /* return value */
5593 __int64_t fixlen
; /* length for -1 case */
5594 int i
; /* extent number */
5595 int lock
; /* lock state */
5596 xfs_bmbt_irec_t
*map
; /* buffer for user's data */
5597 xfs_mount_t
*mp
; /* file system mount point */
5598 int nex
; /* # of user extents can do */
5599 int nexleft
; /* # of user extents left */
5600 int subnex
; /* # of bmapi's can do */
5601 int nmap
; /* number of map entries */
5602 struct getbmapx
*out
; /* output structure */
5603 int whichfork
; /* data or attr fork */
5604 int prealloced
; /* this is a file with
5605 * preallocated data space */
5606 int iflags
; /* interface flags */
5607 int bmapi_flags
; /* flags for xfs_bmapi */
5611 iflags
= bmv
->bmv_iflags
;
5612 whichfork
= iflags
& BMV_IF_ATTRFORK
? XFS_ATTR_FORK
: XFS_DATA_FORK
;
5614 if (whichfork
== XFS_ATTR_FORK
) {
5615 if (XFS_IFORK_Q(ip
)) {
5616 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
&&
5617 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_BTREE
&&
5618 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)
5619 return XFS_ERROR(EINVAL
);
5620 } else if (unlikely(
5621 ip
->i_d
.di_aformat
!= 0 &&
5622 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
)) {
5623 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW
,
5625 return XFS_ERROR(EFSCORRUPTED
);
5631 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
&&
5632 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
&&
5633 ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
)
5634 return XFS_ERROR(EINVAL
);
5636 if (xfs_get_extsz_hint(ip
) ||
5637 ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
|XFS_DIFLAG_APPEND
)){
5639 fixlen
= mp
->m_super
->s_maxbytes
;
5642 fixlen
= XFS_ISIZE(ip
);
5646 if (bmv
->bmv_length
== -1) {
5647 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, fixlen
));
5649 max_t(__int64_t
, fixlen
- bmv
->bmv_offset
, 0);
5650 } else if (bmv
->bmv_length
== 0) {
5651 bmv
->bmv_entries
= 0;
5653 } else if (bmv
->bmv_length
< 0) {
5654 return XFS_ERROR(EINVAL
);
5657 nex
= bmv
->bmv_count
- 1;
5659 return XFS_ERROR(EINVAL
);
5660 bmvend
= bmv
->bmv_offset
+ bmv
->bmv_length
;
5663 if (bmv
->bmv_count
> ULONG_MAX
/ sizeof(struct getbmapx
))
5664 return XFS_ERROR(ENOMEM
);
5665 out
= kmem_zalloc(bmv
->bmv_count
* sizeof(struct getbmapx
), KM_MAYFAIL
);
5667 out
= kmem_zalloc_large(bmv
->bmv_count
*
5668 sizeof(struct getbmapx
));
5670 return XFS_ERROR(ENOMEM
);
5673 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
5674 if (whichfork
== XFS_DATA_FORK
&& !(iflags
& BMV_IF_DELALLOC
)) {
5675 if (ip
->i_delayed_blks
|| XFS_ISIZE(ip
) > ip
->i_d
.di_size
) {
5676 error
= -filemap_write_and_wait(VFS_I(ip
)->i_mapping
);
5678 goto out_unlock_iolock
;
5681 * even after flushing the inode, there can still be delalloc
5682 * blocks on the inode beyond EOF due to speculative
5683 * preallocation. These are not removed until the release
5684 * function is called or the inode is inactivated. Hence we
5685 * cannot assert here that ip->i_delayed_blks == 0.
5689 lock
= xfs_ilock_map_shared(ip
);
5692 * Don't let nex be bigger than the number of extents
5693 * we can have assuming alternating holes and real extents.
5695 if (nex
> XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1)
5696 nex
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1;
5698 bmapi_flags
= xfs_bmapi_aflag(whichfork
);
5699 if (!(iflags
& BMV_IF_PREALLOC
))
5700 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
5703 * Allocate enough space to handle "subnex" maps at a time.
5707 map
= kmem_alloc(subnex
* sizeof(*map
), KM_MAYFAIL
| KM_NOFS
);
5709 goto out_unlock_ilock
;
5711 bmv
->bmv_entries
= 0;
5713 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0 &&
5714 (whichfork
== XFS_ATTR_FORK
|| !(iflags
& BMV_IF_DELALLOC
))) {
5722 nmap
= (nexleft
> subnex
) ? subnex
: nexleft
;
5723 error
= xfs_bmapi_read(ip
, XFS_BB_TO_FSBT(mp
, bmv
->bmv_offset
),
5724 XFS_BB_TO_FSB(mp
, bmv
->bmv_length
),
5725 map
, &nmap
, bmapi_flags
);
5728 ASSERT(nmap
<= subnex
);
5730 for (i
= 0; i
< nmap
&& nexleft
&& bmv
->bmv_length
; i
++) {
5731 out
[cur_ext
].bmv_oflags
= 0;
5732 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
)
5733 out
[cur_ext
].bmv_oflags
|= BMV_OF_PREALLOC
;
5734 else if (map
[i
].br_startblock
== DELAYSTARTBLOCK
)
5735 out
[cur_ext
].bmv_oflags
|= BMV_OF_DELALLOC
;
5736 out
[cur_ext
].bmv_offset
=
5737 XFS_FSB_TO_BB(mp
, map
[i
].br_startoff
);
5738 out
[cur_ext
].bmv_length
=
5739 XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
5740 out
[cur_ext
].bmv_unused1
= 0;
5741 out
[cur_ext
].bmv_unused2
= 0;
5744 * delayed allocation extents that start beyond EOF can
5745 * occur due to speculative EOF allocation when the
5746 * delalloc extent is larger than the largest freespace
5747 * extent at conversion time. These extents cannot be
5748 * converted by data writeback, so can exist here even
5749 * if we are not supposed to be finding delalloc
5752 if (map
[i
].br_startblock
== DELAYSTARTBLOCK
&&
5753 map
[i
].br_startoff
<= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)))
5754 ASSERT((iflags
& BMV_IF_DELALLOC
) != 0);
5756 if (map
[i
].br_startblock
== HOLESTARTBLOCK
&&
5757 whichfork
== XFS_ATTR_FORK
) {
5758 /* came to the end of attribute fork */
5759 out
[cur_ext
].bmv_oflags
|= BMV_OF_LAST
;
5763 if (!xfs_getbmapx_fix_eof_hole(ip
, &out
[cur_ext
],
5765 map
[i
].br_startblock
))
5769 out
[cur_ext
].bmv_offset
+
5770 out
[cur_ext
].bmv_length
;
5772 max_t(__int64_t
, 0, bmvend
- bmv
->bmv_offset
);
5775 * In case we don't want to return the hole,
5776 * don't increase cur_ext so that we can reuse
5777 * it in the next loop.
5779 if ((iflags
& BMV_IF_NO_HOLES
) &&
5780 map
[i
].br_startblock
== HOLESTARTBLOCK
) {
5781 memset(&out
[cur_ext
], 0, sizeof(out
[cur_ext
]));
5789 } while (nmap
&& nexleft
&& bmv
->bmv_length
);
5794 xfs_iunlock_map_shared(ip
, lock
);
5796 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
5798 for (i
= 0; i
< cur_ext
; i
++) {
5799 int full
= 0; /* user array is full */
5801 /* format results & advance arg */
5802 error
= formatter(&arg
, &out
[i
], &full
);
5807 if (is_vmalloc_addr(out
))
5808 kmem_free_large(out
);
5815 STATIC
struct xfs_buf
*
5817 struct xfs_btree_cur
*cur
,
5820 struct xfs_log_item_desc
*lidp
;
5826 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
5827 if (!cur
->bc_bufs
[i
])
5829 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
5830 return cur
->bc_bufs
[i
];
5833 /* Chase down all the log items to see if the bp is there */
5834 list_for_each_entry(lidp
, &cur
->bc_tp
->t_items
, lid_trans
) {
5835 struct xfs_buf_log_item
*bip
;
5836 bip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
5837 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
5838 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
5839 return bip
->bli_buf
;
5847 struct xfs_btree_block
*block
,
5853 __be64
*pp
, *thispa
; /* pointer to block address */
5854 xfs_bmbt_key_t
*prevp
, *keyp
;
5856 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
5859 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
5860 dmxr
= mp
->m_bmap_dmxr
[0];
5861 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
5864 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
5865 be64_to_cpu(keyp
->br_startoff
));
5870 * Compare the block numbers to see if there are dups.
5873 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
5875 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
5877 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
5879 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
5881 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
5882 if (*thispa
== *pp
) {
5883 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
5885 (unsigned long long)be64_to_cpu(*thispa
));
5886 panic("%s: ptrs are equal in node\n",
5894 * Check that the extents for the inode ip are in the right order in all
5899 xfs_bmap_check_leaf_extents(
5900 xfs_btree_cur_t
*cur
, /* btree cursor or null */
5901 xfs_inode_t
*ip
, /* incore inode pointer */
5902 int whichfork
) /* data or attr fork */
5904 struct xfs_btree_block
*block
; /* current btree block */
5905 xfs_fsblock_t bno
; /* block # of "block" */
5906 xfs_buf_t
*bp
; /* buffer for "block" */
5907 int error
; /* error return value */
5908 xfs_extnum_t i
=0, j
; /* index into the extents list */
5909 xfs_ifork_t
*ifp
; /* fork structure */
5910 int level
; /* btree level, for checking */
5911 xfs_mount_t
*mp
; /* file system mount structure */
5912 __be64
*pp
; /* pointer to block address */
5913 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
5914 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
5915 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
5918 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
5924 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5925 block
= ifp
->if_broot
;
5927 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5929 level
= be16_to_cpu(block
->bb_level
);
5931 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
5932 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
5933 bno
= be64_to_cpu(*pp
);
5935 ASSERT(bno
!= NULLDFSBNO
);
5936 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
5937 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
5940 * Go down the tree until leaf level is reached, following the first
5941 * pointer (leftmost) at each level.
5943 while (level
-- > 0) {
5944 /* See if buf is in cur first */
5946 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
5949 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
5955 block
= XFS_BUF_TO_BLOCK(bp
);
5956 XFS_WANT_CORRUPTED_GOTO(
5957 xfs_bmap_sanity_check(mp
, bp
, level
),
5963 * Check this block for basic sanity (increasing keys and
5964 * no duplicate blocks).
5967 xfs_check_block(block
, mp
, 0, 0);
5968 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
5969 bno
= be64_to_cpu(*pp
);
5970 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
5973 xfs_trans_brelse(NULL
, bp
);
5978 * Here with bp and block set to the leftmost leaf node in the tree.
5983 * Loop over all leaf nodes checking that all extents are in the right order.
5986 xfs_fsblock_t nextbno
;
5987 xfs_extnum_t num_recs
;
5990 num_recs
= xfs_btree_get_numrecs(block
);
5993 * Read-ahead the next leaf block, if any.
5996 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
5999 * Check all the extents to make sure they are OK.
6000 * If we had a previous block, the last entry should
6001 * conform with the first entry in this one.
6004 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
6006 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
6007 xfs_bmbt_disk_get_blockcount(&last
) <=
6008 xfs_bmbt_disk_get_startoff(ep
));
6010 for (j
= 1; j
< num_recs
; j
++) {
6011 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
6012 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
6013 xfs_bmbt_disk_get_blockcount(ep
) <=
6014 xfs_bmbt_disk_get_startoff(nextp
));
6022 xfs_trans_brelse(NULL
, bp
);
6026 * If we've reached the end, stop.
6028 if (bno
== NULLFSBLOCK
)
6032 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
6035 error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
6041 block
= XFS_BUF_TO_BLOCK(bp
);
6045 xfs_trans_brelse(NULL
, bp
);
6050 xfs_warn(mp
, "%s: at error0", __func__
);
6052 xfs_trans_brelse(NULL
, bp
);
6054 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
6056 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__
);
6062 * Count fsblocks of the given fork.
6065 xfs_bmap_count_blocks(
6066 xfs_trans_t
*tp
, /* transaction pointer */
6067 xfs_inode_t
*ip
, /* incore inode */
6068 int whichfork
, /* data or attr fork */
6069 int *count
) /* out: count of blocks */
6071 struct xfs_btree_block
*block
; /* current btree block */
6072 xfs_fsblock_t bno
; /* block # of "block" */
6073 xfs_ifork_t
*ifp
; /* fork structure */
6074 int level
; /* btree level, for checking */
6075 xfs_mount_t
*mp
; /* file system mount structure */
6076 __be64
*pp
; /* pointer to block address */
6080 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6081 if ( XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
) {
6082 xfs_bmap_count_leaves(ifp
, 0,
6083 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
),
6089 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6091 block
= ifp
->if_broot
;
6092 level
= be16_to_cpu(block
->bb_level
);
6094 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
6095 bno
= be64_to_cpu(*pp
);
6096 ASSERT(bno
!= NULLDFSBNO
);
6097 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
6098 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
6100 if (unlikely(xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
) < 0)) {
6101 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW
,
6103 return XFS_ERROR(EFSCORRUPTED
);
6110 * Recursively walks each level of a btree
6111 * to count total fsblocks is use.
6113 STATIC
int /* error */
6114 xfs_bmap_count_tree(
6115 xfs_mount_t
*mp
, /* file system mount point */
6116 xfs_trans_t
*tp
, /* transaction pointer */
6117 xfs_ifork_t
*ifp
, /* inode fork pointer */
6118 xfs_fsblock_t blockno
, /* file system block number */
6119 int levelin
, /* level in btree */
6120 int *count
) /* Count of blocks */
6123 xfs_buf_t
*bp
, *nbp
;
6124 int level
= levelin
;
6126 xfs_fsblock_t bno
= blockno
;
6127 xfs_fsblock_t nextbno
;
6128 struct xfs_btree_block
*block
, *nextblock
;
6131 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
, XFS_BMAP_BTREE_REF
,
6136 block
= XFS_BUF_TO_BLOCK(bp
);
6139 /* Not at node above leaves, count this level of nodes */
6140 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
6141 while (nextbno
!= NULLFSBLOCK
) {
6142 error
= xfs_btree_read_bufl(mp
, tp
, nextbno
, 0, &nbp
,
6148 nextblock
= XFS_BUF_TO_BLOCK(nbp
);
6149 nextbno
= be64_to_cpu(nextblock
->bb_u
.l
.bb_rightsib
);
6150 xfs_trans_brelse(tp
, nbp
);
6153 /* Dive to the next level */
6154 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
6155 bno
= be64_to_cpu(*pp
);
6156 if (unlikely((error
=
6157 xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
)) < 0)) {
6158 xfs_trans_brelse(tp
, bp
);
6159 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6160 XFS_ERRLEVEL_LOW
, mp
);
6161 return XFS_ERROR(EFSCORRUPTED
);
6163 xfs_trans_brelse(tp
, bp
);
6165 /* count all level 1 nodes and their leaves */
6167 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
6168 numrecs
= be16_to_cpu(block
->bb_numrecs
);
6169 xfs_bmap_disk_count_leaves(mp
, block
, numrecs
, count
);
6170 xfs_trans_brelse(tp
, bp
);
6171 if (nextbno
== NULLFSBLOCK
)
6174 error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
6180 block
= XFS_BUF_TO_BLOCK(bp
);
6187 * Count leaf blocks given a range of extent records.
6190 xfs_bmap_count_leaves(
6198 for (b
= 0; b
< numrecs
; b
++) {
6199 xfs_bmbt_rec_host_t
*frp
= xfs_iext_get_ext(ifp
, idx
+ b
);
6200 *count
+= xfs_bmbt_get_blockcount(frp
);
6205 * Count leaf blocks given a range of extent records originally
6209 xfs_bmap_disk_count_leaves(
6210 struct xfs_mount
*mp
,
6211 struct xfs_btree_block
*block
,
6216 xfs_bmbt_rec_t
*frp
;
6218 for (b
= 1; b
<= numrecs
; b
++) {
6219 frp
= XFS_BMBT_REC_ADDR(mp
, block
, b
);
6220 *count
+= xfs_bmbt_disk_get_blockcount(frp
);
6225 * dead simple method of punching delalyed allocation blocks from a range in
6226 * the inode. Walks a block at a time so will be slow, but is only executed in
6227 * rare error cases so the overhead is not critical. This will alays punch out
6228 * both the start and end blocks, even if the ranges only partially overlap
6229 * them, so it is up to the caller to ensure that partial blocks are not
6233 xfs_bmap_punch_delalloc_range(
6234 struct xfs_inode
*ip
,
6235 xfs_fileoff_t start_fsb
,
6236 xfs_fileoff_t length
)
6238 xfs_fileoff_t remaining
= length
;
6241 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
6245 xfs_bmbt_irec_t imap
;
6247 xfs_fsblock_t firstblock
;
6248 xfs_bmap_free_t flist
;
6251 * Map the range first and check that it is a delalloc extent
6252 * before trying to unmap the range. Otherwise we will be
6253 * trying to remove a real extent (which requires a
6254 * transaction) or a hole, which is probably a bad idea...
6256 error
= xfs_bmapi_read(ip
, start_fsb
, 1, &imap
, &nimaps
,
6260 /* something screwed, just bail */
6261 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
6262 xfs_alert(ip
->i_mount
,
6263 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6264 ip
->i_ino
, start_fsb
);
6272 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
6273 /* been converted, ignore */
6276 WARN_ON(imap
.br_blockcount
== 0);
6279 * Note: while we initialise the firstblock/flist pair, they
6280 * should never be used because blocks should never be
6281 * allocated or freed for a delalloc extent and hence we need
6282 * don't cancel or finish them after the xfs_bunmapi() call.
6284 xfs_bmap_init(&flist
, &firstblock
);
6285 error
= xfs_bunmapi(NULL
, ip
, start_fsb
, 1, 0, 1, &firstblock
,
6290 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
6294 } while(remaining
> 0);
6300 * Convert the given file system block to a disk block. We have to treat it
6301 * differently based on whether the file is a real time file or not, because the
6305 xfs_fsb_to_db(struct xfs_inode
*ip
, xfs_fsblock_t fsb
)
6307 return (XFS_IS_REALTIME_INODE(ip
) ? \
6308 (xfs_daddr_t
)XFS_FSB_TO_BB((ip
)->i_mount
, (fsb
)) : \
6309 XFS_FSB_TO_DADDR((ip
)->i_mount
, (fsb
)));