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"
45 #include "xfs_quota.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_filestream.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_trace.h"
55 xfs_bmap_check_leaf_extents(xfs_btree_cur_t
*cur
, xfs_inode_t
*ip
, int whichfork
);
58 kmem_zone_t
*xfs_bmap_free_item_zone
;
61 * Prototypes for internal bmap routines.
66 * Called from xfs_bmap_add_attrfork to handle extents format files.
68 STATIC
int /* error */
69 xfs_bmap_add_attrfork_extents(
70 xfs_trans_t
*tp
, /* transaction pointer */
71 xfs_inode_t
*ip
, /* incore inode pointer */
72 xfs_fsblock_t
*firstblock
, /* first block allocated */
73 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
74 int *flags
); /* inode logging flags */
77 * Called from xfs_bmap_add_attrfork to handle local format files.
79 STATIC
int /* error */
80 xfs_bmap_add_attrfork_local(
81 xfs_trans_t
*tp
, /* transaction pointer */
82 xfs_inode_t
*ip
, /* incore inode pointer */
83 xfs_fsblock_t
*firstblock
, /* first block allocated */
84 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
85 int *flags
); /* inode logging flags */
88 * Called by xfs_bmap_add_extent to handle cases converting a delayed
89 * allocation to a real allocation.
91 STATIC
int /* error */
92 xfs_bmap_add_extent_delay_real(
93 struct xfs_trans
*tp
, /* transaction pointer */
94 xfs_inode_t
*ip
, /* incore inode pointer */
95 xfs_extnum_t
*idx
, /* extent number to update/insert */
96 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
97 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
98 xfs_filblks_t
*dnew
, /* new delayed-alloc indirect blocks */
99 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
100 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
101 int *logflagsp
); /* inode logging flags */
104 * Called by xfs_bmap_add_extent to handle cases converting a hole
105 * to a real allocation.
107 STATIC
int /* error */
108 xfs_bmap_add_extent_hole_real(
109 xfs_inode_t
*ip
, /* incore inode pointer */
110 xfs_extnum_t
*idx
, /* extent number to update/insert */
111 xfs_btree_cur_t
*cur
, /* if null, not a btree */
112 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
113 int *logflagsp
, /* inode logging flags */
114 int whichfork
); /* data or attr fork */
117 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
118 * allocation to a real allocation or vice versa.
120 STATIC
int /* error */
121 xfs_bmap_add_extent_unwritten_real(
122 xfs_inode_t
*ip
, /* incore inode pointer */
123 xfs_extnum_t
*idx
, /* extent number to update/insert */
124 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
125 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
126 int *logflagsp
); /* inode logging flags */
129 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
130 * It figures out where to ask the underlying allocator to put the new extent.
132 STATIC
int /* error */
134 xfs_bmalloca_t
*ap
); /* bmap alloc argument struct */
137 * Transform a btree format file with only one leaf node, where the
138 * extents list will fit in the inode, into an extents format file.
139 * Since the file extents are already in-core, all we have to do is
140 * give up the space for the btree root and pitch the leaf block.
142 STATIC
int /* error */
143 xfs_bmap_btree_to_extents(
144 xfs_trans_t
*tp
, /* transaction pointer */
145 xfs_inode_t
*ip
, /* incore inode pointer */
146 xfs_btree_cur_t
*cur
, /* btree cursor */
147 int *logflagsp
, /* inode logging flags */
148 int whichfork
); /* data or attr fork */
151 * Remove the entry "free" from the free item list. Prev points to the
152 * previous entry, unless "free" is the head of the list.
156 xfs_bmap_free_t
*flist
, /* free item list header */
157 xfs_bmap_free_item_t
*prev
, /* previous item on list, if any */
158 xfs_bmap_free_item_t
*free
); /* list item to be freed */
161 * Convert an extents-format file into a btree-format file.
162 * The new file will have a root block (in the inode) and a single child block.
164 STATIC
int /* error */
165 xfs_bmap_extents_to_btree(
166 xfs_trans_t
*tp
, /* transaction pointer */
167 xfs_inode_t
*ip
, /* incore inode pointer */
168 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
169 xfs_bmap_free_t
*flist
, /* blocks freed in xaction */
170 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
171 int wasdel
, /* converting a delayed alloc */
172 int *logflagsp
, /* inode logging flags */
173 int whichfork
); /* data or attr fork */
176 * Convert a local file to an extents file.
177 * This code is sort of bogus, since the file data needs to get
178 * logged so it won't be lost. The bmap-level manipulations are ok, though.
180 STATIC
int /* error */
181 xfs_bmap_local_to_extents(
182 xfs_trans_t
*tp
, /* transaction pointer */
183 xfs_inode_t
*ip
, /* incore inode pointer */
184 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
185 xfs_extlen_t total
, /* total blocks needed by transaction */
186 int *logflagsp
, /* inode logging flags */
187 int whichfork
); /* data or attr fork */
190 * Search the extents list for the inode, for the extent containing bno.
191 * If bno lies in a hole, point to the next entry. If bno lies past eof,
192 * *eofp will be set, and *prevp will contain the last entry (null if none).
193 * Else, *lastxp will be set to the index of the found
194 * entry; *gotp will contain the entry.
196 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
197 xfs_bmap_search_extents(
198 xfs_inode_t
*ip
, /* incore inode pointer */
199 xfs_fileoff_t bno
, /* block number searched for */
200 int whichfork
, /* data or attr fork */
201 int *eofp
, /* out: end of file found */
202 xfs_extnum_t
*lastxp
, /* out: last extent index */
203 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
204 xfs_bmbt_irec_t
*prevp
); /* out: previous extent entry found */
207 * Check the last inode extent to determine whether this allocation will result
208 * in blocks being allocated at the end of the file. When we allocate new data
209 * blocks at the end of the file which do not start at the previous data block,
210 * we will try to align the new blocks at stripe unit boundaries.
212 STATIC
int /* error */
214 xfs_inode_t
*ip
, /* incore inode pointer */
215 xfs_fileoff_t off
, /* file offset in fsblocks */
216 int whichfork
, /* data or attribute fork */
217 char *aeof
); /* return value */
220 * Compute the worst-case number of indirect blocks that will be used
221 * for ip's delayed extent of length "len".
224 xfs_bmap_worst_indlen(
225 xfs_inode_t
*ip
, /* incore inode pointer */
226 xfs_filblks_t len
); /* delayed extent length */
230 * Perform various validation checks on the values being returned
234 xfs_bmap_validate_ret(
238 xfs_bmbt_irec_t
*mval
,
242 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
250 xfs_fsblock_t blockno
,
255 xfs_bmap_count_leaves(
262 xfs_bmap_disk_count_leaves(
263 struct xfs_mount
*mp
,
264 struct xfs_btree_block
*block
,
269 * Bmap internal routines.
272 STATIC
int /* error */
274 struct xfs_btree_cur
*cur
,
278 int *stat
) /* success/failure */
280 cur
->bc_rec
.b
.br_startoff
= off
;
281 cur
->bc_rec
.b
.br_startblock
= bno
;
282 cur
->bc_rec
.b
.br_blockcount
= len
;
283 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
286 STATIC
int /* error */
288 struct xfs_btree_cur
*cur
,
292 int *stat
) /* success/failure */
294 cur
->bc_rec
.b
.br_startoff
= off
;
295 cur
->bc_rec
.b
.br_startblock
= bno
;
296 cur
->bc_rec
.b
.br_blockcount
= len
;
297 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
301 * Update the record referred to by cur to the value given
302 * by [off, bno, len, state].
303 * This either works (return 0) or gets an EFSCORRUPTED error.
307 struct xfs_btree_cur
*cur
,
313 union xfs_btree_rec rec
;
315 xfs_bmbt_disk_set_allf(&rec
.bmbt
, off
, bno
, len
, state
);
316 return xfs_btree_update(cur
, &rec
);
320 * Called from xfs_bmap_add_attrfork to handle btree format files.
322 STATIC
int /* error */
323 xfs_bmap_add_attrfork_btree(
324 xfs_trans_t
*tp
, /* transaction pointer */
325 xfs_inode_t
*ip
, /* incore inode pointer */
326 xfs_fsblock_t
*firstblock
, /* first block allocated */
327 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
328 int *flags
) /* inode logging flags */
330 xfs_btree_cur_t
*cur
; /* btree cursor */
331 int error
; /* error return value */
332 xfs_mount_t
*mp
; /* file system mount struct */
333 int stat
; /* newroot status */
336 if (ip
->i_df
.if_broot_bytes
<= XFS_IFORK_DSIZE(ip
))
337 *flags
|= XFS_ILOG_DBROOT
;
339 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, XFS_DATA_FORK
);
340 cur
->bc_private
.b
.flist
= flist
;
341 cur
->bc_private
.b
.firstblock
= *firstblock
;
342 if ((error
= xfs_bmbt_lookup_ge(cur
, 0, 0, 0, &stat
)))
344 /* must be at least one entry */
345 XFS_WANT_CORRUPTED_GOTO(stat
== 1, error0
);
346 if ((error
= xfs_btree_new_iroot(cur
, flags
, &stat
)))
349 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
350 return XFS_ERROR(ENOSPC
);
352 *firstblock
= cur
->bc_private
.b
.firstblock
;
353 cur
->bc_private
.b
.allocated
= 0;
354 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
358 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
363 * Called from xfs_bmap_add_attrfork to handle extents format files.
365 STATIC
int /* error */
366 xfs_bmap_add_attrfork_extents(
367 xfs_trans_t
*tp
, /* transaction pointer */
368 xfs_inode_t
*ip
, /* incore inode pointer */
369 xfs_fsblock_t
*firstblock
, /* first block allocated */
370 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
371 int *flags
) /* inode logging flags */
373 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
374 int error
; /* error return value */
376 if (ip
->i_d
.di_nextents
* sizeof(xfs_bmbt_rec_t
) <= XFS_IFORK_DSIZE(ip
))
379 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
, &cur
, 0,
380 flags
, XFS_DATA_FORK
);
382 cur
->bc_private
.b
.allocated
= 0;
383 xfs_btree_del_cursor(cur
,
384 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
390 * Called from xfs_bmap_add_attrfork to handle local format files.
392 STATIC
int /* error */
393 xfs_bmap_add_attrfork_local(
394 xfs_trans_t
*tp
, /* transaction pointer */
395 xfs_inode_t
*ip
, /* incore inode pointer */
396 xfs_fsblock_t
*firstblock
, /* first block allocated */
397 xfs_bmap_free_t
*flist
, /* blocks to free at commit */
398 int *flags
) /* inode logging flags */
400 xfs_da_args_t dargs
; /* args for dir/attr code */
401 int error
; /* error return value */
402 xfs_mount_t
*mp
; /* mount structure pointer */
404 if (ip
->i_df
.if_bytes
<= XFS_IFORK_DSIZE(ip
))
406 if (S_ISDIR(ip
->i_d
.di_mode
)) {
408 memset(&dargs
, 0, sizeof(dargs
));
410 dargs
.firstblock
= firstblock
;
412 dargs
.total
= mp
->m_dirblkfsbs
;
413 dargs
.whichfork
= XFS_DATA_FORK
;
415 error
= xfs_dir2_sf_to_block(&dargs
);
417 error
= xfs_bmap_local_to_extents(tp
, ip
, firstblock
, 1, flags
,
423 * Update file extent records and the btree after allocating space.
425 STATIC
int /* error */
427 struct xfs_trans
*tp
, /* transaction pointer */
428 xfs_inode_t
*ip
, /* incore inode pointer */
429 xfs_extnum_t
*idx
, /* extent number to update/insert */
430 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
431 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
432 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
433 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
434 int *logflagsp
, /* inode logging flags */
435 int whichfork
) /* data or attr fork */
437 xfs_btree_cur_t
*cur
; /* btree cursor or null */
438 xfs_filblks_t da_new
; /* new count del alloc blocks used */
439 xfs_filblks_t da_old
; /* old count del alloc blocks used */
440 int error
; /* error return value */
441 xfs_ifork_t
*ifp
; /* inode fork ptr */
442 int logflags
; /* returned value */
443 xfs_extnum_t nextents
; /* number of extents in file now */
445 XFS_STATS_INC(xs_add_exlist
);
448 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
449 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
454 ASSERT(*idx
<= nextents
);
455 ASSERT(!isnullstartblock(new->br_startblock
));
458 * Real allocation off the end of the file.
460 if (*idx
== nextents
) {
462 ASSERT((cur
->bc_private
.b
.flags
&
463 XFS_BTCUR_BPRV_WASDEL
) == 0);
464 error
= xfs_bmap_add_extent_hole_real(ip
, idx
, cur
, new,
465 &logflags
, whichfork
);
467 xfs_bmbt_irec_t prev
; /* old extent at offset idx */
470 * Get the record referred to by idx.
472 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &prev
);
474 * If it's a real allocation record, and the new allocation ends
475 * after the start of the referred to record, then we're filling
476 * in a delayed or unwritten allocation with a real one, or
477 * converting real back to unwritten.
479 if (!isnullstartblock(new->br_startblock
) &&
480 new->br_startoff
+ new->br_blockcount
> prev
.br_startoff
) {
481 if (prev
.br_state
!= XFS_EXT_UNWRITTEN
&&
482 isnullstartblock(prev
.br_startblock
)) {
483 da_old
= startblockval(prev
.br_startblock
);
485 ASSERT(cur
->bc_private
.b
.flags
&
486 XFS_BTCUR_BPRV_WASDEL
);
487 error
= xfs_bmap_add_extent_delay_real(tp
, ip
,
488 idx
, &cur
, new, &da_new
,
489 first
, flist
, &logflags
);
491 ASSERT(new->br_state
== XFS_EXT_NORM
||
492 new->br_state
== XFS_EXT_UNWRITTEN
);
494 error
= xfs_bmap_add_extent_unwritten_real(ip
,
495 idx
, &cur
, new, &logflags
);
501 * Otherwise we're filling in a hole with an allocation.
505 ASSERT((cur
->bc_private
.b
.flags
&
506 XFS_BTCUR_BPRV_WASDEL
) == 0);
507 error
= xfs_bmap_add_extent_hole_real(ip
, idx
, cur
,
508 new, &logflags
, whichfork
);
514 ASSERT(*curp
== cur
|| *curp
== NULL
);
517 * Convert to a btree if necessary.
519 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
520 XFS_IFORK_NEXTENTS(ip
, whichfork
) > ifp
->if_ext_max
) {
521 int tmp_logflags
; /* partial log flag return val */
524 error
= xfs_bmap_extents_to_btree(tp
, ip
, first
,
525 flist
, &cur
, da_old
> 0, &tmp_logflags
, whichfork
);
526 logflags
|= tmp_logflags
;
531 * Adjust for changes in reserved delayed indirect blocks.
532 * Nothing to do for disk quotas here.
534 if (da_old
|| da_new
) {
539 nblks
+= cur
->bc_private
.b
.allocated
;
540 ASSERT(nblks
<= da_old
);
542 xfs_icsb_modify_counters(ip
->i_mount
, XFS_SBS_FDBLOCKS
,
543 (int64_t)(da_old
- nblks
), 0);
546 * Clear out the allocated field, done with it now in any case.
549 cur
->bc_private
.b
.allocated
= 0;
555 xfs_bmap_check_leaf_extents(*curp
, ip
, whichfork
);
557 *logflagsp
= logflags
;
562 * Called by xfs_bmap_add_extent to handle cases converting a delayed
563 * allocation to a real allocation.
565 STATIC
int /* error */
566 xfs_bmap_add_extent_delay_real(
567 struct xfs_trans
*tp
, /* transaction pointer */
568 xfs_inode_t
*ip
, /* incore inode pointer */
569 xfs_extnum_t
*idx
, /* extent number to update/insert */
570 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
571 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
572 xfs_filblks_t
*dnew
, /* new delayed-alloc indirect blocks */
573 xfs_fsblock_t
*first
, /* pointer to firstblock variable */
574 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
575 int *logflagsp
) /* inode logging flags */
577 xfs_btree_cur_t
*cur
; /* btree cursor */
578 int diff
; /* temp value */
579 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
580 int error
; /* error return value */
581 int i
; /* temp state */
582 xfs_ifork_t
*ifp
; /* inode fork pointer */
583 xfs_fileoff_t new_endoff
; /* end offset of new entry */
584 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
585 /* left is 0, right is 1, prev is 2 */
586 int rval
=0; /* return value (logging flags) */
587 int state
= 0;/* state bits, accessed thru macros */
588 xfs_filblks_t temp
=0; /* value for dnew calculations */
589 xfs_filblks_t temp2
=0;/* value for dnew calculations */
590 int tmp_rval
; /* partial logging flags */
597 * Set up a bunch of variables to make the tests simpler.
600 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
601 ep
= xfs_iext_get_ext(ifp
, *idx
);
602 xfs_bmbt_get_all(ep
, &PREV
);
603 new_endoff
= new->br_startoff
+ new->br_blockcount
;
604 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
605 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
608 * Set flags determining what part of the previous delayed allocation
609 * extent is being replaced by a real allocation.
611 if (PREV
.br_startoff
== new->br_startoff
)
612 state
|= BMAP_LEFT_FILLING
;
613 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
614 state
|= BMAP_RIGHT_FILLING
;
617 * Check and set flags if this segment has a left neighbor.
618 * Don't set contiguous if the combined extent would be too large.
621 state
|= BMAP_LEFT_VALID
;
622 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
624 if (isnullstartblock(LEFT
.br_startblock
))
625 state
|= BMAP_LEFT_DELAY
;
628 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
629 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
630 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
631 LEFT
.br_state
== new->br_state
&&
632 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
633 state
|= BMAP_LEFT_CONTIG
;
636 * Check and set flags if this segment has a right neighbor.
637 * Don't set contiguous if the combined extent would be too large.
638 * Also check for all-three-contiguous being too large.
640 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
641 state
|= BMAP_RIGHT_VALID
;
642 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
644 if (isnullstartblock(RIGHT
.br_startblock
))
645 state
|= BMAP_RIGHT_DELAY
;
648 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
649 new_endoff
== RIGHT
.br_startoff
&&
650 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
651 new->br_state
== RIGHT
.br_state
&&
652 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
653 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
654 BMAP_RIGHT_FILLING
)) !=
655 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
656 BMAP_RIGHT_FILLING
) ||
657 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
659 state
|= BMAP_RIGHT_CONTIG
;
663 * Switch out based on the FILLING and CONTIG state bits.
665 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
666 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
667 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
668 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
670 * Filling in all of a previously delayed allocation extent.
671 * The left and right neighbors are both contiguous with new.
674 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
675 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
676 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
677 RIGHT
.br_blockcount
);
678 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
680 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
681 ip
->i_d
.di_nextents
--;
683 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
685 rval
= XFS_ILOG_CORE
;
686 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
688 RIGHT
.br_blockcount
, &i
)))
690 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
691 if ((error
= xfs_btree_delete(cur
, &i
)))
693 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
694 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
696 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
697 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
701 RIGHT
.br_blockcount
, LEFT
.br_state
)))
707 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
709 * Filling in all of a previously delayed allocation extent.
710 * The left neighbor is contiguous, the right is not.
714 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
715 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
716 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
717 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
719 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
721 rval
= XFS_ILOG_DEXT
;
724 if ((error
= xfs_bmbt_lookup_eq(cur
, LEFT
.br_startoff
,
725 LEFT
.br_startblock
, LEFT
.br_blockcount
,
728 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
729 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
732 PREV
.br_blockcount
, LEFT
.br_state
)))
738 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
740 * Filling in all of a previously delayed allocation extent.
741 * The right neighbor is contiguous, the left is not.
743 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
744 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
745 xfs_bmbt_set_blockcount(ep
,
746 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
747 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
749 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
751 rval
= XFS_ILOG_DEXT
;
754 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
756 RIGHT
.br_blockcount
, &i
)))
758 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
759 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
762 RIGHT
.br_blockcount
, PREV
.br_state
)))
769 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
771 * Filling in all of a previously delayed allocation extent.
772 * Neither the left nor right neighbors are contiguous with
775 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
776 xfs_bmbt_set_startblock(ep
, new->br_startblock
);
777 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
779 ip
->i_d
.di_nextents
++;
781 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
783 rval
= XFS_ILOG_CORE
;
784 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
785 new->br_startblock
, new->br_blockcount
,
788 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
789 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
790 if ((error
= xfs_btree_insert(cur
, &i
)))
792 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
798 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
800 * Filling in the first part of a previous delayed allocation.
801 * The left neighbor is contiguous.
803 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
804 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
805 LEFT
.br_blockcount
+ new->br_blockcount
);
806 xfs_bmbt_set_startoff(ep
,
807 PREV
.br_startoff
+ new->br_blockcount
);
808 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
810 temp
= PREV
.br_blockcount
- new->br_blockcount
;
811 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
812 xfs_bmbt_set_blockcount(ep
, temp
);
814 rval
= XFS_ILOG_DEXT
;
817 if ((error
= xfs_bmbt_lookup_eq(cur
, LEFT
.br_startoff
,
818 LEFT
.br_startblock
, LEFT
.br_blockcount
,
821 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
822 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
829 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
830 startblockval(PREV
.br_startblock
));
831 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
832 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
838 case BMAP_LEFT_FILLING
:
840 * Filling in the first part of a previous delayed allocation.
841 * The left neighbor is not contiguous.
843 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
844 xfs_bmbt_set_startoff(ep
, new_endoff
);
845 temp
= PREV
.br_blockcount
- new->br_blockcount
;
846 xfs_bmbt_set_blockcount(ep
, temp
);
847 xfs_iext_insert(ip
, *idx
, 1, new, state
);
848 ip
->i_d
.di_nextents
++;
850 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
852 rval
= XFS_ILOG_CORE
;
853 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
854 new->br_startblock
, new->br_blockcount
,
857 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
858 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
859 if ((error
= xfs_btree_insert(cur
, &i
)))
861 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
863 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
864 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
) {
865 error
= xfs_bmap_extents_to_btree(tp
, ip
,
866 first
, flist
, &cur
, 1, &tmp_rval
,
872 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
873 startblockval(PREV
.br_startblock
) -
874 (cur
? cur
->bc_private
.b
.allocated
: 0));
875 ep
= xfs_iext_get_ext(ifp
, *idx
+ 1);
876 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
877 trace_xfs_bmap_post_update(ip
, *idx
+ 1, state
, _THIS_IP_
);
882 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
884 * Filling in the last part of a previous delayed allocation.
885 * The right neighbor is contiguous with the new allocation.
887 temp
= PREV
.br_blockcount
- new->br_blockcount
;
888 trace_xfs_bmap_pre_update(ip
, *idx
+ 1, state
, _THIS_IP_
);
889 xfs_bmbt_set_blockcount(ep
, temp
);
890 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
+ 1),
891 new->br_startoff
, new->br_startblock
,
892 new->br_blockcount
+ RIGHT
.br_blockcount
,
894 trace_xfs_bmap_post_update(ip
, *idx
+ 1, state
, _THIS_IP_
);
896 rval
= XFS_ILOG_DEXT
;
899 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
901 RIGHT
.br_blockcount
, &i
)))
903 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
904 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
912 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
913 startblockval(PREV
.br_startblock
));
914 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
915 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
916 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
922 case BMAP_RIGHT_FILLING
:
924 * Filling in the last part of a previous delayed allocation.
925 * The right neighbor is not contiguous.
927 temp
= PREV
.br_blockcount
- new->br_blockcount
;
928 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
929 xfs_bmbt_set_blockcount(ep
, temp
);
930 xfs_iext_insert(ip
, *idx
+ 1, 1, new, state
);
931 ip
->i_d
.di_nextents
++;
933 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
935 rval
= XFS_ILOG_CORE
;
936 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
937 new->br_startblock
, new->br_blockcount
,
940 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
941 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
942 if ((error
= xfs_btree_insert(cur
, &i
)))
944 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
946 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
947 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
) {
948 error
= xfs_bmap_extents_to_btree(tp
, ip
,
949 first
, flist
, &cur
, 1, &tmp_rval
,
955 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
956 startblockval(PREV
.br_startblock
) -
957 (cur
? cur
->bc_private
.b
.allocated
: 0));
958 ep
= xfs_iext_get_ext(ifp
, *idx
);
959 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
960 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
968 * Filling in the middle part of a previous delayed allocation.
969 * Contiguity is impossible here.
970 * This case is avoided almost all the time.
972 * We start with a delayed allocation:
974 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
977 * and we are allocating:
978 * +rrrrrrrrrrrrrrrrr+
981 * and we set it up for insertion as:
982 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
984 * PREV @ idx LEFT RIGHT
985 * inserted at idx + 1
987 temp
= new->br_startoff
- PREV
.br_startoff
;
988 temp2
= PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
989 trace_xfs_bmap_pre_update(ip
, *idx
, 0, _THIS_IP_
);
990 xfs_bmbt_set_blockcount(ep
, temp
); /* truncate PREV */
992 RIGHT
.br_state
= PREV
.br_state
;
993 RIGHT
.br_startblock
= nullstartblock(
994 (int)xfs_bmap_worst_indlen(ip
, temp2
));
995 RIGHT
.br_startoff
= new_endoff
;
996 RIGHT
.br_blockcount
= temp2
;
997 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
998 xfs_iext_insert(ip
, *idx
+ 1, 2, &LEFT
, state
);
999 ip
->i_d
.di_nextents
++;
1001 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1003 rval
= XFS_ILOG_CORE
;
1004 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1005 new->br_startblock
, new->br_blockcount
,
1008 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1009 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1010 if ((error
= xfs_btree_insert(cur
, &i
)))
1012 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1014 if (ip
->i_d
.di_format
== XFS_DINODE_FMT_EXTENTS
&&
1015 ip
->i_d
.di_nextents
> ip
->i_df
.if_ext_max
) {
1016 error
= xfs_bmap_extents_to_btree(tp
, ip
,
1017 first
, flist
, &cur
, 1, &tmp_rval
,
1023 temp
= xfs_bmap_worst_indlen(ip
, temp
);
1024 temp2
= xfs_bmap_worst_indlen(ip
, temp2
);
1025 diff
= (int)(temp
+ temp2
- startblockval(PREV
.br_startblock
) -
1026 (cur
? cur
->bc_private
.b
.allocated
: 0));
1028 error
= xfs_icsb_modify_counters(ip
->i_mount
,
1030 -((int64_t)diff
), 0);
1036 ep
= xfs_iext_get_ext(ifp
, *idx
);
1037 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
1038 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1039 trace_xfs_bmap_pre_update(ip
, *idx
+ 2, state
, _THIS_IP_
);
1040 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
+ 2),
1041 nullstartblock((int)temp2
));
1042 trace_xfs_bmap_post_update(ip
, *idx
+ 2, state
, _THIS_IP_
);
1045 *dnew
= temp
+ temp2
;
1048 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1049 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1050 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1051 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1052 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1053 case BMAP_LEFT_CONTIG
:
1054 case BMAP_RIGHT_CONTIG
:
1056 * These cases are all impossible.
1070 * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1071 * allocation to a real allocation or vice versa.
1073 STATIC
int /* error */
1074 xfs_bmap_add_extent_unwritten_real(
1075 xfs_inode_t
*ip
, /* incore inode pointer */
1076 xfs_extnum_t
*idx
, /* extent number to update/insert */
1077 xfs_btree_cur_t
**curp
, /* if *curp is null, not a btree */
1078 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
1079 int *logflagsp
) /* inode logging flags */
1081 xfs_btree_cur_t
*cur
; /* btree cursor */
1082 xfs_bmbt_rec_host_t
*ep
; /* extent entry for idx */
1083 int error
; /* error return value */
1084 int i
; /* temp state */
1085 xfs_ifork_t
*ifp
; /* inode fork pointer */
1086 xfs_fileoff_t new_endoff
; /* end offset of new entry */
1087 xfs_exntst_t newext
; /* new extent state */
1088 xfs_exntst_t oldext
; /* old extent state */
1089 xfs_bmbt_irec_t r
[3]; /* neighbor extent entries */
1090 /* left is 0, right is 1, prev is 2 */
1091 int rval
=0; /* return value (logging flags) */
1092 int state
= 0;/* state bits, accessed thru macros */
1098 * Set up a bunch of variables to make the tests simpler.
1102 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1103 ep
= xfs_iext_get_ext(ifp
, *idx
);
1104 xfs_bmbt_get_all(ep
, &PREV
);
1105 newext
= new->br_state
;
1106 oldext
= (newext
== XFS_EXT_UNWRITTEN
) ?
1107 XFS_EXT_NORM
: XFS_EXT_UNWRITTEN
;
1108 ASSERT(PREV
.br_state
== oldext
);
1109 new_endoff
= new->br_startoff
+ new->br_blockcount
;
1110 ASSERT(PREV
.br_startoff
<= new->br_startoff
);
1111 ASSERT(PREV
.br_startoff
+ PREV
.br_blockcount
>= new_endoff
);
1114 * Set flags determining what part of the previous oldext allocation
1115 * extent is being replaced by a newext allocation.
1117 if (PREV
.br_startoff
== new->br_startoff
)
1118 state
|= BMAP_LEFT_FILLING
;
1119 if (PREV
.br_startoff
+ PREV
.br_blockcount
== new_endoff
)
1120 state
|= BMAP_RIGHT_FILLING
;
1123 * Check and set flags if this segment has a left neighbor.
1124 * Don't set contiguous if the combined extent would be too large.
1127 state
|= BMAP_LEFT_VALID
;
1128 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &LEFT
);
1130 if (isnullstartblock(LEFT
.br_startblock
))
1131 state
|= BMAP_LEFT_DELAY
;
1134 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1135 LEFT
.br_startoff
+ LEFT
.br_blockcount
== new->br_startoff
&&
1136 LEFT
.br_startblock
+ LEFT
.br_blockcount
== new->br_startblock
&&
1137 LEFT
.br_state
== newext
&&
1138 LEFT
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1139 state
|= BMAP_LEFT_CONTIG
;
1142 * Check and set flags if this segment has a right neighbor.
1143 * Don't set contiguous if the combined extent would be too large.
1144 * Also check for all-three-contiguous being too large.
1146 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
) - 1) {
1147 state
|= BMAP_RIGHT_VALID
;
1148 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
+ 1), &RIGHT
);
1149 if (isnullstartblock(RIGHT
.br_startblock
))
1150 state
|= BMAP_RIGHT_DELAY
;
1153 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1154 new_endoff
== RIGHT
.br_startoff
&&
1155 new->br_startblock
+ new->br_blockcount
== RIGHT
.br_startblock
&&
1156 newext
== RIGHT
.br_state
&&
1157 new->br_blockcount
+ RIGHT
.br_blockcount
<= MAXEXTLEN
&&
1158 ((state
& (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1159 BMAP_RIGHT_FILLING
)) !=
1160 (BMAP_LEFT_CONTIG
| BMAP_LEFT_FILLING
|
1161 BMAP_RIGHT_FILLING
) ||
1162 LEFT
.br_blockcount
+ new->br_blockcount
+ RIGHT
.br_blockcount
1164 state
|= BMAP_RIGHT_CONTIG
;
1167 * Switch out based on the FILLING and CONTIG state bits.
1169 switch (state
& (BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1170 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
)) {
1171 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
|
1172 BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1174 * Setting all of a previous oldext extent to newext.
1175 * The left and right neighbors are both contiguous with new.
1179 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1180 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1181 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1182 RIGHT
.br_blockcount
);
1183 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1185 xfs_iext_remove(ip
, *idx
+ 1, 2, state
);
1186 ip
->i_d
.di_nextents
-= 2;
1188 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1190 rval
= XFS_ILOG_CORE
;
1191 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
1192 RIGHT
.br_startblock
,
1193 RIGHT
.br_blockcount
, &i
)))
1195 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1196 if ((error
= xfs_btree_delete(cur
, &i
)))
1198 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1199 if ((error
= xfs_btree_decrement(cur
, 0, &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
, LEFT
.br_startoff
,
1210 LEFT
.br_blockcount
+ PREV
.br_blockcount
+
1211 RIGHT
.br_blockcount
, LEFT
.br_state
)))
1216 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1218 * Setting all of a previous oldext extent to newext.
1219 * The left neighbor is contiguous, the right is not.
1223 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1224 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1225 LEFT
.br_blockcount
+ PREV
.br_blockcount
);
1226 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1228 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1229 ip
->i_d
.di_nextents
--;
1231 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1233 rval
= XFS_ILOG_CORE
;
1234 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1235 PREV
.br_startblock
, PREV
.br_blockcount
,
1238 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1239 if ((error
= xfs_btree_delete(cur
, &i
)))
1241 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1242 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1244 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1245 if ((error
= xfs_bmbt_update(cur
, LEFT
.br_startoff
,
1247 LEFT
.br_blockcount
+ PREV
.br_blockcount
,
1253 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1255 * Setting all of a previous oldext extent to newext.
1256 * The right neighbor is contiguous, the left is not.
1258 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1259 xfs_bmbt_set_blockcount(ep
,
1260 PREV
.br_blockcount
+ RIGHT
.br_blockcount
);
1261 xfs_bmbt_set_state(ep
, newext
);
1262 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1263 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1264 ip
->i_d
.di_nextents
--;
1266 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1268 rval
= XFS_ILOG_CORE
;
1269 if ((error
= xfs_bmbt_lookup_eq(cur
, RIGHT
.br_startoff
,
1270 RIGHT
.br_startblock
,
1271 RIGHT
.br_blockcount
, &i
)))
1273 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1274 if ((error
= xfs_btree_delete(cur
, &i
)))
1276 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1277 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1279 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1280 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1282 new->br_blockcount
+ RIGHT
.br_blockcount
,
1288 case BMAP_LEFT_FILLING
| BMAP_RIGHT_FILLING
:
1290 * Setting all of a previous oldext extent to newext.
1291 * Neither the left nor right neighbors are contiguous with
1294 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1295 xfs_bmbt_set_state(ep
, newext
);
1296 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1299 rval
= XFS_ILOG_DEXT
;
1302 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1303 new->br_startblock
, new->br_blockcount
,
1306 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1307 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1308 new->br_startblock
, new->br_blockcount
,
1314 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
:
1316 * Setting the first part of a previous oldext extent to newext.
1317 * The left neighbor is contiguous.
1319 trace_xfs_bmap_pre_update(ip
, *idx
- 1, state
, _THIS_IP_
);
1320 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
- 1),
1321 LEFT
.br_blockcount
+ new->br_blockcount
);
1322 xfs_bmbt_set_startoff(ep
,
1323 PREV
.br_startoff
+ new->br_blockcount
);
1324 trace_xfs_bmap_post_update(ip
, *idx
- 1, state
, _THIS_IP_
);
1326 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1327 xfs_bmbt_set_startblock(ep
,
1328 new->br_startblock
+ new->br_blockcount
);
1329 xfs_bmbt_set_blockcount(ep
,
1330 PREV
.br_blockcount
- new->br_blockcount
);
1331 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1336 rval
= XFS_ILOG_DEXT
;
1339 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1340 PREV
.br_startblock
, PREV
.br_blockcount
,
1343 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1344 if ((error
= xfs_bmbt_update(cur
,
1345 PREV
.br_startoff
+ new->br_blockcount
,
1346 PREV
.br_startblock
+ new->br_blockcount
,
1347 PREV
.br_blockcount
- new->br_blockcount
,
1350 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1352 if (xfs_bmbt_update(cur
, LEFT
.br_startoff
,
1354 LEFT
.br_blockcount
+ new->br_blockcount
,
1360 case BMAP_LEFT_FILLING
:
1362 * Setting the first part of a previous oldext extent to newext.
1363 * The left neighbor is not contiguous.
1365 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1366 ASSERT(ep
&& xfs_bmbt_get_state(ep
) == oldext
);
1367 xfs_bmbt_set_startoff(ep
, new_endoff
);
1368 xfs_bmbt_set_blockcount(ep
,
1369 PREV
.br_blockcount
- new->br_blockcount
);
1370 xfs_bmbt_set_startblock(ep
,
1371 new->br_startblock
+ new->br_blockcount
);
1372 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1374 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1375 ip
->i_d
.di_nextents
++;
1377 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1379 rval
= XFS_ILOG_CORE
;
1380 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1381 PREV
.br_startblock
, PREV
.br_blockcount
,
1384 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1385 if ((error
= xfs_bmbt_update(cur
,
1386 PREV
.br_startoff
+ new->br_blockcount
,
1387 PREV
.br_startblock
+ new->br_blockcount
,
1388 PREV
.br_blockcount
- new->br_blockcount
,
1391 cur
->bc_rec
.b
= *new;
1392 if ((error
= xfs_btree_insert(cur
, &i
)))
1394 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1398 case BMAP_RIGHT_FILLING
| BMAP_RIGHT_CONTIG
:
1400 * Setting the last part of a previous oldext extent to newext.
1401 * The right neighbor is contiguous with the new allocation.
1403 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1404 xfs_bmbt_set_blockcount(ep
,
1405 PREV
.br_blockcount
- new->br_blockcount
);
1406 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1410 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1411 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
1412 new->br_startoff
, new->br_startblock
,
1413 new->br_blockcount
+ RIGHT
.br_blockcount
, newext
);
1414 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1417 rval
= XFS_ILOG_DEXT
;
1420 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1422 PREV
.br_blockcount
, &i
)))
1424 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1425 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
1427 PREV
.br_blockcount
- new->br_blockcount
,
1430 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
1432 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1434 new->br_blockcount
+ RIGHT
.br_blockcount
,
1440 case BMAP_RIGHT_FILLING
:
1442 * Setting the last part of a previous oldext extent to newext.
1443 * The right neighbor is not contiguous.
1445 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1446 xfs_bmbt_set_blockcount(ep
,
1447 PREV
.br_blockcount
- new->br_blockcount
);
1448 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1451 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1453 ip
->i_d
.di_nextents
++;
1455 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1457 rval
= XFS_ILOG_CORE
;
1458 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1459 PREV
.br_startblock
, PREV
.br_blockcount
,
1462 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1463 if ((error
= xfs_bmbt_update(cur
, PREV
.br_startoff
,
1465 PREV
.br_blockcount
- new->br_blockcount
,
1468 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1469 new->br_startblock
, new->br_blockcount
,
1472 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1473 cur
->bc_rec
.b
.br_state
= XFS_EXT_NORM
;
1474 if ((error
= xfs_btree_insert(cur
, &i
)))
1476 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1482 * Setting the middle part of a previous oldext extent to
1483 * newext. Contiguity is impossible here.
1484 * One extent becomes three extents.
1486 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1487 xfs_bmbt_set_blockcount(ep
,
1488 new->br_startoff
- PREV
.br_startoff
);
1489 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1492 r
[1].br_startoff
= new_endoff
;
1493 r
[1].br_blockcount
=
1494 PREV
.br_startoff
+ PREV
.br_blockcount
- new_endoff
;
1495 r
[1].br_startblock
= new->br_startblock
+ new->br_blockcount
;
1496 r
[1].br_state
= oldext
;
1499 xfs_iext_insert(ip
, *idx
, 2, &r
[0], state
);
1501 ip
->i_d
.di_nextents
+= 2;
1503 rval
= XFS_ILOG_CORE
| XFS_ILOG_DEXT
;
1505 rval
= XFS_ILOG_CORE
;
1506 if ((error
= xfs_bmbt_lookup_eq(cur
, PREV
.br_startoff
,
1507 PREV
.br_startblock
, PREV
.br_blockcount
,
1510 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1511 /* new right extent - oldext */
1512 if ((error
= xfs_bmbt_update(cur
, r
[1].br_startoff
,
1513 r
[1].br_startblock
, r
[1].br_blockcount
,
1516 /* new left extent - oldext */
1517 cur
->bc_rec
.b
= PREV
;
1518 cur
->bc_rec
.b
.br_blockcount
=
1519 new->br_startoff
- PREV
.br_startoff
;
1520 if ((error
= xfs_btree_insert(cur
, &i
)))
1522 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1524 * Reset the cursor to the position of the new extent
1525 * we are about to insert as we can't trust it after
1526 * the previous insert.
1528 if ((error
= xfs_bmbt_lookup_eq(cur
, new->br_startoff
,
1529 new->br_startblock
, new->br_blockcount
,
1532 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1533 /* new middle extent - newext */
1534 cur
->bc_rec
.b
.br_state
= new->br_state
;
1535 if ((error
= xfs_btree_insert(cur
, &i
)))
1537 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1541 case BMAP_LEFT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1542 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1543 case BMAP_LEFT_FILLING
| BMAP_RIGHT_CONTIG
:
1544 case BMAP_RIGHT_FILLING
| BMAP_LEFT_CONTIG
:
1545 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1546 case BMAP_LEFT_CONTIG
:
1547 case BMAP_RIGHT_CONTIG
:
1549 * These cases are all impossible.
1563 * Convert a hole to a delayed allocation.
1566 xfs_bmap_add_extent_hole_delay(
1567 xfs_inode_t
*ip
, /* incore inode pointer */
1568 xfs_extnum_t
*idx
, /* extent number to update/insert */
1569 xfs_bmbt_irec_t
*new) /* new data to add to file extents */
1571 xfs_ifork_t
*ifp
; /* inode fork pointer */
1572 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
1573 xfs_filblks_t newlen
=0; /* new indirect size */
1574 xfs_filblks_t oldlen
=0; /* old indirect size */
1575 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
1576 int state
; /* state bits, accessed thru macros */
1577 xfs_filblks_t temp
=0; /* temp for indirect calculations */
1579 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
1581 ASSERT(isnullstartblock(new->br_startblock
));
1584 * Check and set flags if this segment has a left neighbor
1587 state
|= BMAP_LEFT_VALID
;
1588 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
1590 if (isnullstartblock(left
.br_startblock
))
1591 state
|= BMAP_LEFT_DELAY
;
1595 * Check and set flags if the current (right) segment exists.
1596 * If it doesn't exist, we're converting the hole at end-of-file.
1598 if (*idx
< ip
->i_df
.if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
1599 state
|= BMAP_RIGHT_VALID
;
1600 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
1602 if (isnullstartblock(right
.br_startblock
))
1603 state
|= BMAP_RIGHT_DELAY
;
1607 * Set contiguity flags on the left and right neighbors.
1608 * Don't let extents get too large, even if the pieces are contiguous.
1610 if ((state
& BMAP_LEFT_VALID
) && (state
& BMAP_LEFT_DELAY
) &&
1611 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
1612 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1613 state
|= BMAP_LEFT_CONTIG
;
1615 if ((state
& BMAP_RIGHT_VALID
) && (state
& BMAP_RIGHT_DELAY
) &&
1616 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
1617 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
1618 (!(state
& BMAP_LEFT_CONTIG
) ||
1619 (left
.br_blockcount
+ new->br_blockcount
+
1620 right
.br_blockcount
<= MAXEXTLEN
)))
1621 state
|= BMAP_RIGHT_CONTIG
;
1624 * Switch out based on the contiguity flags.
1626 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
1627 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1629 * New allocation is contiguous with delayed allocations
1630 * on the left and on the right.
1631 * Merge all three into a single extent record.
1634 temp
= left
.br_blockcount
+ new->br_blockcount
+
1635 right
.br_blockcount
;
1637 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1638 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
1639 oldlen
= startblockval(left
.br_startblock
) +
1640 startblockval(new->br_startblock
) +
1641 startblockval(right
.br_startblock
);
1642 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1643 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
1644 nullstartblock((int)newlen
));
1645 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1647 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1650 case BMAP_LEFT_CONTIG
:
1652 * New allocation is contiguous with a delayed allocation
1654 * Merge the new allocation with the left neighbor.
1657 temp
= left
.br_blockcount
+ new->br_blockcount
;
1659 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1660 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
), temp
);
1661 oldlen
= startblockval(left
.br_startblock
) +
1662 startblockval(new->br_startblock
);
1663 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1664 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp
, *idx
),
1665 nullstartblock((int)newlen
));
1666 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1669 case BMAP_RIGHT_CONTIG
:
1671 * New allocation is contiguous with a delayed allocation
1673 * Merge the new allocation with the right neighbor.
1675 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1676 temp
= new->br_blockcount
+ right
.br_blockcount
;
1677 oldlen
= startblockval(new->br_startblock
) +
1678 startblockval(right
.br_startblock
);
1679 newlen
= xfs_bmap_worst_indlen(ip
, temp
);
1680 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
1682 nullstartblock((int)newlen
), temp
, right
.br_state
);
1683 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1688 * New allocation is not contiguous with another
1689 * delayed allocation.
1690 * Insert a new entry.
1692 oldlen
= newlen
= 0;
1693 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1696 if (oldlen
!= newlen
) {
1697 ASSERT(oldlen
> newlen
);
1698 xfs_icsb_modify_counters(ip
->i_mount
, XFS_SBS_FDBLOCKS
,
1699 (int64_t)(oldlen
- newlen
), 0);
1701 * Nothing to do for disk quota accounting here.
1707 * Called by xfs_bmap_add_extent to handle cases converting a hole
1708 * to a real allocation.
1710 STATIC
int /* error */
1711 xfs_bmap_add_extent_hole_real(
1712 xfs_inode_t
*ip
, /* incore inode pointer */
1713 xfs_extnum_t
*idx
, /* extent number to update/insert */
1714 xfs_btree_cur_t
*cur
, /* if null, not a btree */
1715 xfs_bmbt_irec_t
*new, /* new data to add to file extents */
1716 int *logflagsp
, /* inode logging flags */
1717 int whichfork
) /* data or attr fork */
1719 int error
; /* error return value */
1720 int i
; /* temp state */
1721 xfs_ifork_t
*ifp
; /* inode fork pointer */
1722 xfs_bmbt_irec_t left
; /* left neighbor extent entry */
1723 xfs_bmbt_irec_t right
; /* right neighbor extent entry */
1724 int rval
=0; /* return value (logging flags) */
1725 int state
; /* state bits, accessed thru macros */
1727 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
1728 ASSERT(*idx
<= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
));
1731 if (whichfork
== XFS_ATTR_FORK
)
1732 state
|= BMAP_ATTRFORK
;
1735 * Check and set flags if this segment has a left neighbor.
1738 state
|= BMAP_LEFT_VALID
;
1739 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
- 1), &left
);
1740 if (isnullstartblock(left
.br_startblock
))
1741 state
|= BMAP_LEFT_DELAY
;
1745 * Check and set flags if this segment has a current value.
1746 * Not true if we're inserting into the "hole" at eof.
1748 if (*idx
< ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)) {
1749 state
|= BMAP_RIGHT_VALID
;
1750 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *idx
), &right
);
1751 if (isnullstartblock(right
.br_startblock
))
1752 state
|= BMAP_RIGHT_DELAY
;
1756 * We're inserting a real allocation between "left" and "right".
1757 * Set the contiguity flags. Don't let extents get too large.
1759 if ((state
& BMAP_LEFT_VALID
) && !(state
& BMAP_LEFT_DELAY
) &&
1760 left
.br_startoff
+ left
.br_blockcount
== new->br_startoff
&&
1761 left
.br_startblock
+ left
.br_blockcount
== new->br_startblock
&&
1762 left
.br_state
== new->br_state
&&
1763 left
.br_blockcount
+ new->br_blockcount
<= MAXEXTLEN
)
1764 state
|= BMAP_LEFT_CONTIG
;
1766 if ((state
& BMAP_RIGHT_VALID
) && !(state
& BMAP_RIGHT_DELAY
) &&
1767 new->br_startoff
+ new->br_blockcount
== right
.br_startoff
&&
1768 new->br_startblock
+ new->br_blockcount
== right
.br_startblock
&&
1769 new->br_state
== right
.br_state
&&
1770 new->br_blockcount
+ right
.br_blockcount
<= MAXEXTLEN
&&
1771 (!(state
& BMAP_LEFT_CONTIG
) ||
1772 left
.br_blockcount
+ new->br_blockcount
+
1773 right
.br_blockcount
<= MAXEXTLEN
))
1774 state
|= BMAP_RIGHT_CONTIG
;
1778 * Select which case we're in here, and implement it.
1780 switch (state
& (BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
)) {
1781 case BMAP_LEFT_CONTIG
| BMAP_RIGHT_CONTIG
:
1783 * New allocation is contiguous with real allocations on the
1784 * left and on the right.
1785 * Merge all three into a single extent record.
1788 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1789 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1790 left
.br_blockcount
+ new->br_blockcount
+
1791 right
.br_blockcount
);
1792 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1794 xfs_iext_remove(ip
, *idx
+ 1, 1, state
);
1796 XFS_IFORK_NEXT_SET(ip
, whichfork
,
1797 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
1799 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
1801 rval
= XFS_ILOG_CORE
;
1802 if ((error
= xfs_bmbt_lookup_eq(cur
,
1804 right
.br_startblock
,
1805 right
.br_blockcount
, &i
)))
1807 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1808 if ((error
= xfs_btree_delete(cur
, &i
)))
1810 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1811 if ((error
= xfs_btree_decrement(cur
, 0, &i
)))
1813 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1814 if ((error
= xfs_bmbt_update(cur
, left
.br_startoff
,
1816 left
.br_blockcount
+
1817 new->br_blockcount
+
1818 right
.br_blockcount
,
1824 case BMAP_LEFT_CONTIG
:
1826 * New allocation is contiguous with a real allocation
1828 * Merge the new allocation with the left neighbor.
1831 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1832 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp
, *idx
),
1833 left
.br_blockcount
+ new->br_blockcount
);
1834 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1837 rval
= xfs_ilog_fext(whichfork
);
1840 if ((error
= xfs_bmbt_lookup_eq(cur
,
1843 left
.br_blockcount
, &i
)))
1845 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1846 if ((error
= xfs_bmbt_update(cur
, left
.br_startoff
,
1848 left
.br_blockcount
+
1855 case BMAP_RIGHT_CONTIG
:
1857 * New allocation is contiguous with a real allocation
1859 * Merge the new allocation with the right neighbor.
1861 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
1862 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp
, *idx
),
1863 new->br_startoff
, new->br_startblock
,
1864 new->br_blockcount
+ right
.br_blockcount
,
1866 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
1869 rval
= xfs_ilog_fext(whichfork
);
1872 if ((error
= xfs_bmbt_lookup_eq(cur
,
1874 right
.br_startblock
,
1875 right
.br_blockcount
, &i
)))
1877 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1878 if ((error
= xfs_bmbt_update(cur
, new->br_startoff
,
1880 new->br_blockcount
+
1881 right
.br_blockcount
,
1889 * New allocation is not contiguous with another
1891 * Insert a new entry.
1893 xfs_iext_insert(ip
, *idx
, 1, new, state
);
1894 XFS_IFORK_NEXT_SET(ip
, whichfork
,
1895 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
1897 rval
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
1899 rval
= XFS_ILOG_CORE
;
1900 if ((error
= xfs_bmbt_lookup_eq(cur
,
1903 new->br_blockcount
, &i
)))
1905 XFS_WANT_CORRUPTED_GOTO(i
== 0, done
);
1906 cur
->bc_rec
.b
.br_state
= new->br_state
;
1907 if ((error
= xfs_btree_insert(cur
, &i
)))
1909 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
1919 * Adjust the size of the new extent based on di_extsize and rt extsize.
1922 xfs_bmap_extsize_align(
1924 xfs_bmbt_irec_t
*gotp
, /* next extent pointer */
1925 xfs_bmbt_irec_t
*prevp
, /* previous extent pointer */
1926 xfs_extlen_t extsz
, /* align to this extent size */
1927 int rt
, /* is this a realtime inode? */
1928 int eof
, /* is extent at end-of-file? */
1929 int delay
, /* creating delalloc extent? */
1930 int convert
, /* overwriting unwritten extent? */
1931 xfs_fileoff_t
*offp
, /* in/out: aligned offset */
1932 xfs_extlen_t
*lenp
) /* in/out: aligned length */
1934 xfs_fileoff_t orig_off
; /* original offset */
1935 xfs_extlen_t orig_alen
; /* original length */
1936 xfs_fileoff_t orig_end
; /* original off+len */
1937 xfs_fileoff_t nexto
; /* next file offset */
1938 xfs_fileoff_t prevo
; /* previous file offset */
1939 xfs_fileoff_t align_off
; /* temp for offset */
1940 xfs_extlen_t align_alen
; /* temp for length */
1941 xfs_extlen_t temp
; /* temp for calculations */
1946 orig_off
= align_off
= *offp
;
1947 orig_alen
= align_alen
= *lenp
;
1948 orig_end
= orig_off
+ orig_alen
;
1951 * If this request overlaps an existing extent, then don't
1952 * attempt to perform any additional alignment.
1954 if (!delay
&& !eof
&&
1955 (orig_off
>= gotp
->br_startoff
) &&
1956 (orig_end
<= gotp
->br_startoff
+ gotp
->br_blockcount
)) {
1961 * If the file offset is unaligned vs. the extent size
1962 * we need to align it. This will be possible unless
1963 * the file was previously written with a kernel that didn't
1964 * perform this alignment, or if a truncate shot us in the
1967 temp
= do_mod(orig_off
, extsz
);
1973 * Same adjustment for the end of the requested area.
1975 if ((temp
= (align_alen
% extsz
))) {
1976 align_alen
+= extsz
- temp
;
1979 * If the previous block overlaps with this proposed allocation
1980 * then move the start forward without adjusting the length.
1982 if (prevp
->br_startoff
!= NULLFILEOFF
) {
1983 if (prevp
->br_startblock
== HOLESTARTBLOCK
)
1984 prevo
= prevp
->br_startoff
;
1986 prevo
= prevp
->br_startoff
+ prevp
->br_blockcount
;
1989 if (align_off
!= orig_off
&& align_off
< prevo
)
1992 * If the next block overlaps with this proposed allocation
1993 * then move the start back without adjusting the length,
1994 * but not before offset 0.
1995 * This may of course make the start overlap previous block,
1996 * and if we hit the offset 0 limit then the next block
1997 * can still overlap too.
1999 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
) {
2000 if ((delay
&& gotp
->br_startblock
== HOLESTARTBLOCK
) ||
2001 (!delay
&& gotp
->br_startblock
== DELAYSTARTBLOCK
))
2002 nexto
= gotp
->br_startoff
+ gotp
->br_blockcount
;
2004 nexto
= gotp
->br_startoff
;
2006 nexto
= NULLFILEOFF
;
2008 align_off
+ align_alen
!= orig_end
&&
2009 align_off
+ align_alen
> nexto
)
2010 align_off
= nexto
> align_alen
? nexto
- align_alen
: 0;
2012 * If we're now overlapping the next or previous extent that
2013 * means we can't fit an extsz piece in this hole. Just move
2014 * the start forward to the first valid spot and set
2015 * the length so we hit the end.
2017 if (align_off
!= orig_off
&& align_off
< prevo
)
2019 if (align_off
+ align_alen
!= orig_end
&&
2020 align_off
+ align_alen
> nexto
&&
2021 nexto
!= NULLFILEOFF
) {
2022 ASSERT(nexto
> prevo
);
2023 align_alen
= nexto
- align_off
;
2027 * If realtime, and the result isn't a multiple of the realtime
2028 * extent size we need to remove blocks until it is.
2030 if (rt
&& (temp
= (align_alen
% mp
->m_sb
.sb_rextsize
))) {
2032 * We're not covering the original request, or
2033 * we won't be able to once we fix the length.
2035 if (orig_off
< align_off
||
2036 orig_end
> align_off
+ align_alen
||
2037 align_alen
- temp
< orig_alen
)
2038 return XFS_ERROR(EINVAL
);
2040 * Try to fix it by moving the start up.
2042 if (align_off
+ temp
<= orig_off
) {
2047 * Try to fix it by moving the end in.
2049 else if (align_off
+ align_alen
- temp
>= orig_end
)
2052 * Set the start to the minimum then trim the length.
2055 align_alen
-= orig_off
- align_off
;
2056 align_off
= orig_off
;
2057 align_alen
-= align_alen
% mp
->m_sb
.sb_rextsize
;
2060 * Result doesn't cover the request, fail it.
2062 if (orig_off
< align_off
|| orig_end
> align_off
+ align_alen
)
2063 return XFS_ERROR(EINVAL
);
2065 ASSERT(orig_off
>= align_off
);
2066 ASSERT(orig_end
<= align_off
+ align_alen
);
2070 if (!eof
&& gotp
->br_startoff
!= NULLFILEOFF
)
2071 ASSERT(align_off
+ align_alen
<= gotp
->br_startoff
);
2072 if (prevp
->br_startoff
!= NULLFILEOFF
)
2073 ASSERT(align_off
>= prevp
->br_startoff
+ prevp
->br_blockcount
);
2081 #define XFS_ALLOC_GAP_UNITS 4
2085 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2087 xfs_fsblock_t adjust
; /* adjustment to block numbers */
2088 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
2089 xfs_mount_t
*mp
; /* mount point structure */
2090 int nullfb
; /* true if ap->firstblock isn't set */
2091 int rt
; /* true if inode is realtime */
2093 #define ISVALID(x,y) \
2095 (x) < mp->m_sb.sb_rblocks : \
2096 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2097 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2098 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2100 mp
= ap
->ip
->i_mount
;
2101 nullfb
= ap
->firstblock
== NULLFSBLOCK
;
2102 rt
= XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
;
2103 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, ap
->firstblock
);
2105 * If allocating at eof, and there's a previous real block,
2106 * try to use its last block as our starting point.
2108 if (ap
->eof
&& ap
->prevp
->br_startoff
!= NULLFILEOFF
&&
2109 !isnullstartblock(ap
->prevp
->br_startblock
) &&
2110 ISVALID(ap
->prevp
->br_startblock
+ ap
->prevp
->br_blockcount
,
2111 ap
->prevp
->br_startblock
)) {
2112 ap
->rval
= ap
->prevp
->br_startblock
+ ap
->prevp
->br_blockcount
;
2114 * Adjust for the gap between prevp and us.
2117 (ap
->prevp
->br_startoff
+ ap
->prevp
->br_blockcount
);
2119 ISVALID(ap
->rval
+ adjust
, ap
->prevp
->br_startblock
))
2123 * If not at eof, then compare the two neighbor blocks.
2124 * Figure out whether either one gives us a good starting point,
2125 * and pick the better one.
2127 else if (!ap
->eof
) {
2128 xfs_fsblock_t gotbno
; /* right side block number */
2129 xfs_fsblock_t gotdiff
=0; /* right side difference */
2130 xfs_fsblock_t prevbno
; /* left side block number */
2131 xfs_fsblock_t prevdiff
=0; /* left side difference */
2134 * If there's a previous (left) block, select a requested
2135 * start block based on it.
2137 if (ap
->prevp
->br_startoff
!= NULLFILEOFF
&&
2138 !isnullstartblock(ap
->prevp
->br_startblock
) &&
2139 (prevbno
= ap
->prevp
->br_startblock
+
2140 ap
->prevp
->br_blockcount
) &&
2141 ISVALID(prevbno
, ap
->prevp
->br_startblock
)) {
2143 * Calculate gap to end of previous block.
2145 adjust
= prevdiff
= ap
->off
-
2146 (ap
->prevp
->br_startoff
+
2147 ap
->prevp
->br_blockcount
);
2149 * Figure the startblock based on the previous block's
2150 * end and the gap size.
2152 * If the gap is large relative to the piece we're
2153 * allocating, or using it gives us an invalid block
2154 * number, then just use the end of the previous block.
2156 if (prevdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->alen
&&
2157 ISVALID(prevbno
+ prevdiff
,
2158 ap
->prevp
->br_startblock
))
2163 * If the firstblock forbids it, can't use it,
2166 if (!rt
&& !nullfb
&&
2167 XFS_FSB_TO_AGNO(mp
, prevbno
) != fb_agno
)
2168 prevbno
= NULLFSBLOCK
;
2171 * No previous block or can't follow it, just default.
2174 prevbno
= NULLFSBLOCK
;
2176 * If there's a following (right) block, select a requested
2177 * start block based on it.
2179 if (!isnullstartblock(ap
->gotp
->br_startblock
)) {
2181 * Calculate gap to start of next block.
2183 adjust
= gotdiff
= ap
->gotp
->br_startoff
- ap
->off
;
2185 * Figure the startblock based on the next block's
2186 * start and the gap size.
2188 gotbno
= ap
->gotp
->br_startblock
;
2191 * If the gap is large relative to the piece we're
2192 * allocating, or using it gives us an invalid block
2193 * number, then just use the start of the next block
2194 * offset by our length.
2196 if (gotdiff
<= XFS_ALLOC_GAP_UNITS
* ap
->alen
&&
2197 ISVALID(gotbno
- gotdiff
, gotbno
))
2199 else if (ISVALID(gotbno
- ap
->alen
, gotbno
)) {
2201 gotdiff
+= adjust
- ap
->alen
;
2205 * If the firstblock forbids it, can't use it,
2208 if (!rt
&& !nullfb
&&
2209 XFS_FSB_TO_AGNO(mp
, gotbno
) != fb_agno
)
2210 gotbno
= NULLFSBLOCK
;
2213 * No next block, just default.
2216 gotbno
= NULLFSBLOCK
;
2218 * If both valid, pick the better one, else the only good
2219 * one, else ap->rval is already set (to 0 or the inode block).
2221 if (prevbno
!= NULLFSBLOCK
&& gotbno
!= NULLFSBLOCK
)
2222 ap
->rval
= prevdiff
<= gotdiff
? prevbno
: gotbno
;
2223 else if (prevbno
!= NULLFSBLOCK
)
2225 else if (gotbno
!= NULLFSBLOCK
)
2233 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2235 xfs_alloctype_t atype
= 0; /* type for allocation routines */
2236 int error
; /* error return value */
2237 xfs_mount_t
*mp
; /* mount point structure */
2238 xfs_extlen_t prod
= 0; /* product factor for allocators */
2239 xfs_extlen_t ralen
= 0; /* realtime allocation length */
2240 xfs_extlen_t align
; /* minimum allocation alignment */
2243 mp
= ap
->ip
->i_mount
;
2244 align
= xfs_get_extsz_hint(ap
->ip
);
2245 prod
= align
/ mp
->m_sb
.sb_rextsize
;
2246 error
= xfs_bmap_extsize_align(mp
, ap
->gotp
, ap
->prevp
,
2247 align
, 1, ap
->eof
, 0,
2248 ap
->conv
, &ap
->off
, &ap
->alen
);
2252 ASSERT(ap
->alen
% mp
->m_sb
.sb_rextsize
== 0);
2255 * If the offset & length are not perfectly aligned
2256 * then kill prod, it will just get us in trouble.
2258 if (do_mod(ap
->off
, align
) || ap
->alen
% align
)
2261 * Set ralen to be the actual requested length in rtextents.
2263 ralen
= ap
->alen
/ mp
->m_sb
.sb_rextsize
;
2265 * If the old value was close enough to MAXEXTLEN that
2266 * we rounded up to it, cut it back so it's valid again.
2267 * Note that if it's a really large request (bigger than
2268 * MAXEXTLEN), we don't hear about that number, and can't
2269 * adjust the starting point to match it.
2271 if (ralen
* mp
->m_sb
.sb_rextsize
>= MAXEXTLEN
)
2272 ralen
= MAXEXTLEN
/ mp
->m_sb
.sb_rextsize
;
2275 * Lock out other modifications to the RT bitmap inode.
2277 xfs_ilock(mp
->m_rbmip
, XFS_ILOCK_EXCL
);
2278 xfs_trans_ijoin_ref(ap
->tp
, mp
->m_rbmip
, XFS_ILOCK_EXCL
);
2281 * If it's an allocation to an empty file at offset 0,
2282 * pick an extent that will space things out in the rt area.
2284 if (ap
->eof
&& ap
->off
== 0) {
2285 xfs_rtblock_t
uninitialized_var(rtx
); /* realtime extent no */
2287 error
= xfs_rtpick_extent(mp
, ap
->tp
, ralen
, &rtx
);
2290 ap
->rval
= rtx
* mp
->m_sb
.sb_rextsize
;
2295 xfs_bmap_adjacent(ap
);
2298 * Realtime allocation, done through xfs_rtallocate_extent.
2300 atype
= ap
->rval
== 0 ? XFS_ALLOCTYPE_ANY_AG
: XFS_ALLOCTYPE_NEAR_BNO
;
2301 do_div(ap
->rval
, mp
->m_sb
.sb_rextsize
);
2304 if ((error
= xfs_rtallocate_extent(ap
->tp
, ap
->rval
, 1, ap
->alen
,
2305 &ralen
, atype
, ap
->wasdel
, prod
, &rtb
)))
2307 if (rtb
== NULLFSBLOCK
&& prod
> 1 &&
2308 (error
= xfs_rtallocate_extent(ap
->tp
, ap
->rval
, 1,
2309 ap
->alen
, &ralen
, atype
,
2310 ap
->wasdel
, 1, &rtb
)))
2313 if (ap
->rval
!= NULLFSBLOCK
) {
2314 ap
->rval
*= mp
->m_sb
.sb_rextsize
;
2315 ralen
*= mp
->m_sb
.sb_rextsize
;
2317 ap
->ip
->i_d
.di_nblocks
+= ralen
;
2318 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
2320 ap
->ip
->i_delayed_blks
-= ralen
;
2322 * Adjust the disk quota also. This was reserved
2325 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
2326 ap
->wasdel
? XFS_TRANS_DQ_DELRTBCOUNT
:
2327 XFS_TRANS_DQ_RTBCOUNT
, (long) ralen
);
2335 xfs_bmap_btalloc_nullfb(
2336 struct xfs_bmalloca
*ap
,
2337 struct xfs_alloc_arg
*args
,
2340 struct xfs_mount
*mp
= ap
->ip
->i_mount
;
2341 struct xfs_perag
*pag
;
2342 xfs_agnumber_t ag
, startag
;
2346 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
))
2347 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2349 args
->type
= XFS_ALLOCTYPE_START_BNO
;
2350 args
->total
= ap
->total
;
2353 * Search for an allocation group with a single extent large enough
2354 * for the request. If one isn't found, then adjust the minimum
2355 * allocation size to the largest space found.
2357 startag
= ag
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2358 if (startag
== NULLAGNUMBER
)
2361 pag
= xfs_perag_get(mp
, ag
);
2362 while (*blen
< args
->maxlen
) {
2363 if (!pag
->pagf_init
) {
2364 error
= xfs_alloc_pagf_init(mp
, args
->tp
, ag
,
2365 XFS_ALLOC_FLAG_TRYLOCK
);
2373 * See xfs_alloc_fix_freelist...
2375 if (pag
->pagf_init
) {
2376 xfs_extlen_t longest
;
2377 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
2378 if (*blen
< longest
)
2383 if (xfs_inode_is_filestream(ap
->ip
)) {
2384 if (*blen
>= args
->maxlen
)
2389 * If startag is an invalid AG, we've
2390 * come here once before and
2391 * xfs_filestream_new_ag picked the
2392 * best currently available.
2394 * Don't continue looping, since we
2395 * could loop forever.
2397 if (startag
== NULLAGNUMBER
)
2400 error
= xfs_filestream_new_ag(ap
, &ag
);
2405 /* loop again to set 'blen'*/
2406 startag
= NULLAGNUMBER
;
2407 pag
= xfs_perag_get(mp
, ag
);
2411 if (++ag
== mp
->m_sb
.sb_agcount
)
2416 pag
= xfs_perag_get(mp
, ag
);
2421 * Since the above loop did a BUF_TRYLOCK, it is
2422 * possible that there is space for this request.
2424 if (notinit
|| *blen
< ap
->minlen
)
2425 args
->minlen
= ap
->minlen
;
2427 * If the best seen length is less than the request
2428 * length, use the best as the minimum.
2430 else if (*blen
< args
->maxlen
)
2431 args
->minlen
= *blen
;
2433 * Otherwise we've seen an extent as big as maxlen,
2434 * use that as the minimum.
2437 args
->minlen
= args
->maxlen
;
2440 * set the failure fallback case to look in the selected
2441 * AG as the stream may have moved.
2443 if (xfs_inode_is_filestream(ap
->ip
))
2444 ap
->rval
= args
->fsbno
= XFS_AGB_TO_FSB(mp
, ag
, 0);
2451 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2453 xfs_mount_t
*mp
; /* mount point structure */
2454 xfs_alloctype_t atype
= 0; /* type for allocation routines */
2455 xfs_extlen_t align
; /* minimum allocation alignment */
2456 xfs_agnumber_t fb_agno
; /* ag number of ap->firstblock */
2458 xfs_alloc_arg_t args
;
2460 xfs_extlen_t nextminlen
= 0;
2461 int nullfb
; /* true if ap->firstblock isn't set */
2466 mp
= ap
->ip
->i_mount
;
2467 align
= ap
->userdata
? xfs_get_extsz_hint(ap
->ip
) : 0;
2468 if (unlikely(align
)) {
2469 error
= xfs_bmap_extsize_align(mp
, ap
->gotp
, ap
->prevp
,
2470 align
, 0, ap
->eof
, 0, ap
->conv
,
2471 &ap
->off
, &ap
->alen
);
2475 nullfb
= ap
->firstblock
== NULLFSBLOCK
;
2476 fb_agno
= nullfb
? NULLAGNUMBER
: XFS_FSB_TO_AGNO(mp
, ap
->firstblock
);
2478 if (ap
->userdata
&& xfs_inode_is_filestream(ap
->ip
)) {
2479 ag
= xfs_filestream_lookup_ag(ap
->ip
);
2480 ag
= (ag
!= NULLAGNUMBER
) ? ag
: 0;
2481 ap
->rval
= XFS_AGB_TO_FSB(mp
, ag
, 0);
2483 ap
->rval
= XFS_INO_TO_FSB(mp
, ap
->ip
->i_ino
);
2486 ap
->rval
= ap
->firstblock
;
2488 xfs_bmap_adjacent(ap
);
2491 * If allowed, use ap->rval; otherwise must use firstblock since
2492 * it's in the right allocation group.
2494 if (nullfb
|| XFS_FSB_TO_AGNO(mp
, ap
->rval
) == fb_agno
)
2497 ap
->rval
= ap
->firstblock
;
2499 * Normal allocation, done through xfs_alloc_vextent.
2501 tryagain
= isaligned
= 0;
2504 args
.fsbno
= ap
->rval
;
2506 /* Trim the allocation back to the maximum an AG can fit. */
2507 args
.maxlen
= MIN(ap
->alen
, XFS_ALLOC_AG_MAX_USABLE(mp
));
2508 args
.firstblock
= ap
->firstblock
;
2511 error
= xfs_bmap_btalloc_nullfb(ap
, &args
, &blen
);
2514 } else if (ap
->low
) {
2515 if (xfs_inode_is_filestream(ap
->ip
))
2516 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
2518 args
.type
= XFS_ALLOCTYPE_START_BNO
;
2519 args
.total
= args
.minlen
= ap
->minlen
;
2521 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
2522 args
.total
= ap
->total
;
2523 args
.minlen
= ap
->minlen
;
2525 /* apply extent size hints if obtained earlier */
2526 if (unlikely(align
)) {
2528 if ((args
.mod
= (xfs_extlen_t
)do_mod(ap
->off
, args
.prod
)))
2529 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
2530 } else if (mp
->m_sb
.sb_blocksize
>= PAGE_CACHE_SIZE
) {
2534 args
.prod
= PAGE_CACHE_SIZE
>> mp
->m_sb
.sb_blocklog
;
2535 if ((args
.mod
= (xfs_extlen_t
)(do_mod(ap
->off
, args
.prod
))))
2536 args
.mod
= (xfs_extlen_t
)(args
.prod
- args
.mod
);
2539 * If we are not low on available data blocks, and the
2540 * underlying logical volume manager is a stripe, and
2541 * the file offset is zero then try to allocate data
2542 * blocks on stripe unit boundary.
2543 * NOTE: ap->aeof is only set if the allocation length
2544 * is >= the stripe unit and the allocation offset is
2545 * at the end of file.
2547 if (!ap
->low
&& ap
->aeof
) {
2549 args
.alignment
= mp
->m_dalign
;
2553 * Adjust for alignment
2555 if (blen
> args
.alignment
&& blen
<= args
.maxlen
)
2556 args
.minlen
= blen
- args
.alignment
;
2557 args
.minalignslop
= 0;
2560 * First try an exact bno allocation.
2561 * If it fails then do a near or start bno
2562 * allocation with alignment turned on.
2566 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
2569 * Compute the minlen+alignment for the
2570 * next case. Set slop so that the value
2571 * of minlen+alignment+slop doesn't go up
2572 * between the calls.
2574 if (blen
> mp
->m_dalign
&& blen
<= args
.maxlen
)
2575 nextminlen
= blen
- mp
->m_dalign
;
2577 nextminlen
= args
.minlen
;
2578 if (nextminlen
+ mp
->m_dalign
> args
.minlen
+ 1)
2580 nextminlen
+ mp
->m_dalign
-
2583 args
.minalignslop
= 0;
2587 args
.minalignslop
= 0;
2589 args
.minleft
= ap
->minleft
;
2590 args
.wasdel
= ap
->wasdel
;
2592 args
.userdata
= ap
->userdata
;
2593 if ((error
= xfs_alloc_vextent(&args
)))
2595 if (tryagain
&& args
.fsbno
== NULLFSBLOCK
) {
2597 * Exact allocation failed. Now try with alignment
2601 args
.fsbno
= ap
->rval
;
2602 args
.alignment
= mp
->m_dalign
;
2603 args
.minlen
= nextminlen
;
2604 args
.minalignslop
= 0;
2606 if ((error
= xfs_alloc_vextent(&args
)))
2609 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
2611 * allocation failed, so turn off alignment and
2615 args
.fsbno
= ap
->rval
;
2617 if ((error
= xfs_alloc_vextent(&args
)))
2620 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
&&
2621 args
.minlen
> ap
->minlen
) {
2622 args
.minlen
= ap
->minlen
;
2623 args
.type
= XFS_ALLOCTYPE_START_BNO
;
2624 args
.fsbno
= ap
->rval
;
2625 if ((error
= xfs_alloc_vextent(&args
)))
2628 if (args
.fsbno
== NULLFSBLOCK
&& nullfb
) {
2630 args
.type
= XFS_ALLOCTYPE_FIRST_AG
;
2631 args
.total
= ap
->minlen
;
2633 if ((error
= xfs_alloc_vextent(&args
)))
2637 if (args
.fsbno
!= NULLFSBLOCK
) {
2638 ap
->firstblock
= ap
->rval
= args
.fsbno
;
2639 ASSERT(nullfb
|| fb_agno
== args
.agno
||
2640 (ap
->low
&& fb_agno
< args
.agno
));
2641 ap
->alen
= args
.len
;
2642 ap
->ip
->i_d
.di_nblocks
+= args
.len
;
2643 xfs_trans_log_inode(ap
->tp
, ap
->ip
, XFS_ILOG_CORE
);
2645 ap
->ip
->i_delayed_blks
-= args
.len
;
2647 * Adjust the disk quota also. This was reserved
2650 xfs_trans_mod_dquot_byino(ap
->tp
, ap
->ip
,
2651 ap
->wasdel
? XFS_TRANS_DQ_DELBCOUNT
:
2652 XFS_TRANS_DQ_BCOUNT
,
2655 ap
->rval
= NULLFSBLOCK
;
2662 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2663 * It figures out where to ask the underlying allocator to put the new extent.
2667 xfs_bmalloca_t
*ap
) /* bmap alloc argument struct */
2669 if (XFS_IS_REALTIME_INODE(ap
->ip
) && ap
->userdata
)
2670 return xfs_bmap_rtalloc(ap
);
2671 return xfs_bmap_btalloc(ap
);
2675 * Transform a btree format file with only one leaf node, where the
2676 * extents list will fit in the inode, into an extents format file.
2677 * Since the file extents are already in-core, all we have to do is
2678 * give up the space for the btree root and pitch the leaf block.
2680 STATIC
int /* error */
2681 xfs_bmap_btree_to_extents(
2682 xfs_trans_t
*tp
, /* transaction pointer */
2683 xfs_inode_t
*ip
, /* incore inode pointer */
2684 xfs_btree_cur_t
*cur
, /* btree cursor */
2685 int *logflagsp
, /* inode logging flags */
2686 int whichfork
) /* data or attr fork */
2689 struct xfs_btree_block
*cblock
;/* child btree block */
2690 xfs_fsblock_t cbno
; /* child block number */
2691 xfs_buf_t
*cbp
; /* child block's buffer */
2692 int error
; /* error return value */
2693 xfs_ifork_t
*ifp
; /* inode fork data */
2694 xfs_mount_t
*mp
; /* mount point structure */
2695 __be64
*pp
; /* ptr to block address */
2696 struct xfs_btree_block
*rblock
;/* root btree block */
2699 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2700 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
2701 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
2702 rblock
= ifp
->if_broot
;
2703 ASSERT(be16_to_cpu(rblock
->bb_level
) == 1);
2704 ASSERT(be16_to_cpu(rblock
->bb_numrecs
) == 1);
2705 ASSERT(xfs_bmbt_maxrecs(mp
, ifp
->if_broot_bytes
, 0) == 1);
2706 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, rblock
, 1, ifp
->if_broot_bytes
);
2707 cbno
= be64_to_cpu(*pp
);
2710 if ((error
= xfs_btree_check_lptr(cur
, cbno
, 1)))
2713 if ((error
= xfs_btree_read_bufl(mp
, tp
, cbno
, 0, &cbp
,
2714 XFS_BMAP_BTREE_REF
)))
2716 cblock
= XFS_BUF_TO_BLOCK(cbp
);
2717 if ((error
= xfs_btree_check_block(cur
, cblock
, 0, cbp
)))
2719 xfs_bmap_add_free(cbno
, 1, cur
->bc_private
.b
.flist
, mp
);
2720 ip
->i_d
.di_nblocks
--;
2721 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, -1L);
2722 xfs_trans_binval(tp
, cbp
);
2723 if (cur
->bc_bufs
[0] == cbp
)
2724 cur
->bc_bufs
[0] = NULL
;
2725 xfs_iroot_realloc(ip
, -1, whichfork
);
2726 ASSERT(ifp
->if_broot
== NULL
);
2727 ASSERT((ifp
->if_flags
& XFS_IFBROOT
) == 0);
2728 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
2729 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fext(whichfork
);
2734 * Called by xfs_bmapi to update file extent records and the btree
2735 * after removing space (or undoing a delayed allocation).
2737 STATIC
int /* error */
2738 xfs_bmap_del_extent(
2739 xfs_inode_t
*ip
, /* incore inode pointer */
2740 xfs_trans_t
*tp
, /* current transaction pointer */
2741 xfs_extnum_t
*idx
, /* extent number to update/delete */
2742 xfs_bmap_free_t
*flist
, /* list of extents to be freed */
2743 xfs_btree_cur_t
*cur
, /* if null, not a btree */
2744 xfs_bmbt_irec_t
*del
, /* data to remove from extents */
2745 int *logflagsp
, /* inode logging flags */
2746 int whichfork
) /* data or attr fork */
2748 xfs_filblks_t da_new
; /* new delay-alloc indirect blocks */
2749 xfs_filblks_t da_old
; /* old delay-alloc indirect blocks */
2750 xfs_fsblock_t del_endblock
=0; /* first block past del */
2751 xfs_fileoff_t del_endoff
; /* first offset past del */
2752 int delay
; /* current block is delayed allocated */
2753 int do_fx
; /* free extent at end of routine */
2754 xfs_bmbt_rec_host_t
*ep
; /* current extent entry pointer */
2755 int error
; /* error return value */
2756 int flags
; /* inode logging flags */
2757 xfs_bmbt_irec_t got
; /* current extent entry */
2758 xfs_fileoff_t got_endoff
; /* first offset past got */
2759 int i
; /* temp state */
2760 xfs_ifork_t
*ifp
; /* inode fork pointer */
2761 xfs_mount_t
*mp
; /* mount structure */
2762 xfs_filblks_t nblks
; /* quota/sb block count */
2763 xfs_bmbt_irec_t
new; /* new record to be inserted */
2765 uint qfield
; /* quota field to update */
2766 xfs_filblks_t temp
; /* for indirect length calculations */
2767 xfs_filblks_t temp2
; /* for indirect length calculations */
2770 XFS_STATS_INC(xs_del_exlist
);
2772 if (whichfork
== XFS_ATTR_FORK
)
2773 state
|= BMAP_ATTRFORK
;
2776 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
2777 ASSERT((*idx
>= 0) && (*idx
< ifp
->if_bytes
/
2778 (uint
)sizeof(xfs_bmbt_rec_t
)));
2779 ASSERT(del
->br_blockcount
> 0);
2780 ep
= xfs_iext_get_ext(ifp
, *idx
);
2781 xfs_bmbt_get_all(ep
, &got
);
2782 ASSERT(got
.br_startoff
<= del
->br_startoff
);
2783 del_endoff
= del
->br_startoff
+ del
->br_blockcount
;
2784 got_endoff
= got
.br_startoff
+ got
.br_blockcount
;
2785 ASSERT(got_endoff
>= del_endoff
);
2786 delay
= isnullstartblock(got
.br_startblock
);
2787 ASSERT(isnullstartblock(del
->br_startblock
) == delay
);
2792 * If deleting a real allocation, must free up the disk space.
2795 flags
= XFS_ILOG_CORE
;
2797 * Realtime allocation. Free it and record di_nblocks update.
2799 if (whichfork
== XFS_DATA_FORK
&& XFS_IS_REALTIME_INODE(ip
)) {
2803 ASSERT(do_mod(del
->br_blockcount
,
2804 mp
->m_sb
.sb_rextsize
) == 0);
2805 ASSERT(do_mod(del
->br_startblock
,
2806 mp
->m_sb
.sb_rextsize
) == 0);
2807 bno
= del
->br_startblock
;
2808 len
= del
->br_blockcount
;
2809 do_div(bno
, mp
->m_sb
.sb_rextsize
);
2810 do_div(len
, mp
->m_sb
.sb_rextsize
);
2811 error
= xfs_rtfree_extent(tp
, bno
, (xfs_extlen_t
)len
);
2815 nblks
= len
* mp
->m_sb
.sb_rextsize
;
2816 qfield
= XFS_TRANS_DQ_RTBCOUNT
;
2819 * Ordinary allocation.
2823 nblks
= del
->br_blockcount
;
2824 qfield
= XFS_TRANS_DQ_BCOUNT
;
2827 * Set up del_endblock and cur for later.
2829 del_endblock
= del
->br_startblock
+ del
->br_blockcount
;
2831 if ((error
= xfs_bmbt_lookup_eq(cur
, got
.br_startoff
,
2832 got
.br_startblock
, got
.br_blockcount
,
2835 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2837 da_old
= da_new
= 0;
2839 da_old
= startblockval(got
.br_startblock
);
2845 * Set flag value to use in switch statement.
2846 * Left-contig is 2, right-contig is 1.
2848 switch (((got
.br_startoff
== del
->br_startoff
) << 1) |
2849 (got_endoff
== del_endoff
)) {
2852 * Matches the whole extent. Delete the entry.
2854 xfs_iext_remove(ip
, *idx
, 1,
2855 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0);
2860 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2861 XFS_IFORK_NEXTENTS(ip
, whichfork
) - 1);
2862 flags
|= XFS_ILOG_CORE
;
2864 flags
|= xfs_ilog_fext(whichfork
);
2867 if ((error
= xfs_btree_delete(cur
, &i
)))
2869 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2874 * Deleting the first part of the extent.
2876 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2877 xfs_bmbt_set_startoff(ep
, del_endoff
);
2878 temp
= got
.br_blockcount
- del
->br_blockcount
;
2879 xfs_bmbt_set_blockcount(ep
, temp
);
2881 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2883 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2884 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2888 xfs_bmbt_set_startblock(ep
, del_endblock
);
2889 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2891 flags
|= xfs_ilog_fext(whichfork
);
2894 if ((error
= xfs_bmbt_update(cur
, del_endoff
, del_endblock
,
2895 got
.br_blockcount
- del
->br_blockcount
,
2902 * Deleting the last part of the extent.
2904 temp
= got
.br_blockcount
- del
->br_blockcount
;
2905 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2906 xfs_bmbt_set_blockcount(ep
, temp
);
2908 temp
= XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip
, temp
),
2910 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2911 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2915 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
2917 flags
|= xfs_ilog_fext(whichfork
);
2920 if ((error
= xfs_bmbt_update(cur
, got
.br_startoff
,
2922 got
.br_blockcount
- del
->br_blockcount
,
2929 * Deleting the middle of the extent.
2931 temp
= del
->br_startoff
- got
.br_startoff
;
2932 trace_xfs_bmap_pre_update(ip
, *idx
, state
, _THIS_IP_
);
2933 xfs_bmbt_set_blockcount(ep
, temp
);
2934 new.br_startoff
= del_endoff
;
2935 temp2
= got_endoff
- del_endoff
;
2936 new.br_blockcount
= temp2
;
2937 new.br_state
= got
.br_state
;
2939 new.br_startblock
= del_endblock
;
2940 flags
|= XFS_ILOG_CORE
;
2942 if ((error
= xfs_bmbt_update(cur
,
2944 got
.br_startblock
, temp
,
2947 if ((error
= xfs_btree_increment(cur
, 0, &i
)))
2949 cur
->bc_rec
.b
= new;
2950 error
= xfs_btree_insert(cur
, &i
);
2951 if (error
&& error
!= ENOSPC
)
2954 * If get no-space back from btree insert,
2955 * it tried a split, and we have a zero
2956 * block reservation.
2957 * Fix up our state and return the error.
2959 if (error
== ENOSPC
) {
2961 * Reset the cursor, don't trust
2962 * it after any insert operation.
2964 if ((error
= xfs_bmbt_lookup_eq(cur
,
2969 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2971 * Update the btree record back
2972 * to the original value.
2974 if ((error
= xfs_bmbt_update(cur
,
2981 * Reset the extent record back
2982 * to the original value.
2984 xfs_bmbt_set_blockcount(ep
,
2987 error
= XFS_ERROR(ENOSPC
);
2990 XFS_WANT_CORRUPTED_GOTO(i
== 1, done
);
2992 flags
|= xfs_ilog_fext(whichfork
);
2993 XFS_IFORK_NEXT_SET(ip
, whichfork
,
2994 XFS_IFORK_NEXTENTS(ip
, whichfork
) + 1);
2996 ASSERT(whichfork
== XFS_DATA_FORK
);
2997 temp
= xfs_bmap_worst_indlen(ip
, temp
);
2998 xfs_bmbt_set_startblock(ep
, nullstartblock((int)temp
));
2999 temp2
= xfs_bmap_worst_indlen(ip
, temp2
);
3000 new.br_startblock
= nullstartblock((int)temp2
);
3001 da_new
= temp
+ temp2
;
3002 while (da_new
> da_old
) {
3006 xfs_bmbt_set_startblock(ep
,
3007 nullstartblock((int)temp
));
3009 if (da_new
== da_old
)
3015 nullstartblock((int)temp2
);
3019 trace_xfs_bmap_post_update(ip
, *idx
, state
, _THIS_IP_
);
3020 xfs_iext_insert(ip
, *idx
+ 1, 1, &new, state
);
3025 * If we need to, add to list of extents to delete.
3028 xfs_bmap_add_free(del
->br_startblock
, del
->br_blockcount
, flist
,
3031 * Adjust inode # blocks in the file.
3034 ip
->i_d
.di_nblocks
-= nblks
;
3036 * Adjust quota data.
3039 xfs_trans_mod_dquot_byino(tp
, ip
, qfield
, (long)-nblks
);
3042 * Account for change in delayed indirect blocks.
3043 * Nothing to do for disk quota accounting here.
3045 ASSERT(da_old
>= da_new
);
3046 if (da_old
> da_new
) {
3047 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
3048 (int64_t)(da_old
- da_new
), 0);
3056 * Remove the entry "free" from the free item list. Prev points to the
3057 * previous entry, unless "free" is the head of the list.
3061 xfs_bmap_free_t
*flist
, /* free item list header */
3062 xfs_bmap_free_item_t
*prev
, /* previous item on list, if any */
3063 xfs_bmap_free_item_t
*free
) /* list item to be freed */
3066 prev
->xbfi_next
= free
->xbfi_next
;
3068 flist
->xbf_first
= free
->xbfi_next
;
3070 kmem_zone_free(xfs_bmap_free_item_zone
, free
);
3074 * Convert an extents-format file into a btree-format file.
3075 * The new file will have a root block (in the inode) and a single child block.
3077 STATIC
int /* error */
3078 xfs_bmap_extents_to_btree(
3079 xfs_trans_t
*tp
, /* transaction pointer */
3080 xfs_inode_t
*ip
, /* incore inode pointer */
3081 xfs_fsblock_t
*firstblock
, /* first-block-allocated */
3082 xfs_bmap_free_t
*flist
, /* blocks freed in xaction */
3083 xfs_btree_cur_t
**curp
, /* cursor returned to caller */
3084 int wasdel
, /* converting a delayed alloc */
3085 int *logflagsp
, /* inode logging flags */
3086 int whichfork
) /* data or attr fork */
3088 struct xfs_btree_block
*ablock
; /* allocated (child) bt block */
3089 xfs_buf_t
*abp
; /* buffer for ablock */
3090 xfs_alloc_arg_t args
; /* allocation arguments */
3091 xfs_bmbt_rec_t
*arp
; /* child record pointer */
3092 struct xfs_btree_block
*block
; /* btree root block */
3093 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
3094 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3095 int error
; /* error return value */
3096 xfs_extnum_t i
, cnt
; /* extent record index */
3097 xfs_ifork_t
*ifp
; /* inode fork pointer */
3098 xfs_bmbt_key_t
*kp
; /* root block key pointer */
3099 xfs_mount_t
*mp
; /* mount structure */
3100 xfs_extnum_t nextents
; /* number of file extents */
3101 xfs_bmbt_ptr_t
*pp
; /* root block address pointer */
3103 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3104 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
);
3105 ASSERT(ifp
->if_ext_max
==
3106 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
3108 * Make space in the inode incore.
3110 xfs_iroot_realloc(ip
, 1, whichfork
);
3111 ifp
->if_flags
|= XFS_IFBROOT
;
3116 block
= ifp
->if_broot
;
3117 block
->bb_magic
= cpu_to_be32(XFS_BMAP_MAGIC
);
3118 block
->bb_level
= cpu_to_be16(1);
3119 block
->bb_numrecs
= cpu_to_be16(1);
3120 block
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLDFSBNO
);
3121 block
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLDFSBNO
);
3124 * Need a cursor. Can't allocate until bb_level is filled in.
3127 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
3128 cur
->bc_private
.b
.firstblock
= *firstblock
;
3129 cur
->bc_private
.b
.flist
= flist
;
3130 cur
->bc_private
.b
.flags
= wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
3132 * Convert to a btree with two levels, one record in root.
3134 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_BTREE
);
3137 args
.firstblock
= *firstblock
;
3138 if (*firstblock
== NULLFSBLOCK
) {
3139 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3140 args
.fsbno
= XFS_INO_TO_FSB(mp
, ip
->i_ino
);
3141 } else if (flist
->xbf_low
) {
3142 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3143 args
.fsbno
= *firstblock
;
3145 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3146 args
.fsbno
= *firstblock
;
3148 args
.minlen
= args
.maxlen
= args
.prod
= 1;
3149 args
.total
= args
.minleft
= args
.alignment
= args
.mod
= args
.isfl
=
3150 args
.minalignslop
= 0;
3151 args
.wasdel
= wasdel
;
3153 if ((error
= xfs_alloc_vextent(&args
))) {
3154 xfs_iroot_realloc(ip
, -1, whichfork
);
3155 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
3159 * Allocation can't fail, the space was reserved.
3161 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
3162 ASSERT(*firstblock
== NULLFSBLOCK
||
3163 args
.agno
== XFS_FSB_TO_AGNO(mp
, *firstblock
) ||
3165 args
.agno
> XFS_FSB_TO_AGNO(mp
, *firstblock
)));
3166 *firstblock
= cur
->bc_private
.b
.firstblock
= args
.fsbno
;
3167 cur
->bc_private
.b
.allocated
++;
3168 ip
->i_d
.di_nblocks
++;
3169 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_BCOUNT
, 1L);
3170 abp
= xfs_btree_get_bufl(mp
, tp
, args
.fsbno
, 0);
3172 * Fill in the child block.
3174 ablock
= XFS_BUF_TO_BLOCK(abp
);
3175 ablock
->bb_magic
= cpu_to_be32(XFS_BMAP_MAGIC
);
3176 ablock
->bb_level
= 0;
3177 ablock
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLDFSBNO
);
3178 ablock
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLDFSBNO
);
3179 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
3180 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3181 for (cnt
= i
= 0; i
< nextents
; i
++) {
3182 ep
= xfs_iext_get_ext(ifp
, i
);
3183 if (!isnullstartblock(xfs_bmbt_get_startblock(ep
))) {
3184 arp
->l0
= cpu_to_be64(ep
->l0
);
3185 arp
->l1
= cpu_to_be64(ep
->l1
);
3189 ASSERT(cnt
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
3190 xfs_btree_set_numrecs(ablock
, cnt
);
3193 * Fill in the root key and pointer.
3195 kp
= XFS_BMBT_KEY_ADDR(mp
, block
, 1);
3196 arp
= XFS_BMBT_REC_ADDR(mp
, ablock
, 1);
3197 kp
->br_startoff
= cpu_to_be64(xfs_bmbt_disk_get_startoff(arp
));
3198 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, xfs_bmbt_get_maxrecs(cur
,
3199 be16_to_cpu(block
->bb_level
)));
3200 *pp
= cpu_to_be64(args
.fsbno
);
3203 * Do all this logging at the end so that
3204 * the root is at the right level.
3206 xfs_btree_log_block(cur
, abp
, XFS_BB_ALL_BITS
);
3207 xfs_btree_log_recs(cur
, abp
, 1, be16_to_cpu(ablock
->bb_numrecs
));
3208 ASSERT(*curp
== NULL
);
3210 *logflagsp
= XFS_ILOG_CORE
| xfs_ilog_fbroot(whichfork
);
3215 * Calculate the default attribute fork offset for newly created inodes.
3218 xfs_default_attroffset(
3219 struct xfs_inode
*ip
)
3221 struct xfs_mount
*mp
= ip
->i_mount
;
3224 if (mp
->m_sb
.sb_inodesize
== 256) {
3225 offset
= XFS_LITINO(mp
) -
3226 XFS_BMDR_SPACE_CALC(MINABTPTRS
);
3228 offset
= XFS_BMDR_SPACE_CALC(6 * MINABTPTRS
);
3231 ASSERT(offset
< XFS_LITINO(mp
));
3236 * Helper routine to reset inode di_forkoff field when switching
3237 * attribute fork from local to extent format - we reset it where
3238 * possible to make space available for inline data fork extents.
3241 xfs_bmap_forkoff_reset(
3246 if (whichfork
== XFS_ATTR_FORK
&&
3247 ip
->i_d
.di_format
!= XFS_DINODE_FMT_DEV
&&
3248 ip
->i_d
.di_format
!= XFS_DINODE_FMT_UUID
&&
3249 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
) {
3250 uint dfl_forkoff
= xfs_default_attroffset(ip
) >> 3;
3252 if (dfl_forkoff
> ip
->i_d
.di_forkoff
) {
3253 ip
->i_d
.di_forkoff
= dfl_forkoff
;
3254 ip
->i_df
.if_ext_max
=
3255 XFS_IFORK_DSIZE(ip
) / sizeof(xfs_bmbt_rec_t
);
3256 ip
->i_afp
->if_ext_max
=
3257 XFS_IFORK_ASIZE(ip
) / sizeof(xfs_bmbt_rec_t
);
3263 * Convert a local file to an extents file.
3264 * This code is out of bounds for data forks of regular files,
3265 * since the file data needs to get logged so things will stay consistent.
3266 * (The bmap-level manipulations are ok, though).
3268 STATIC
int /* error */
3269 xfs_bmap_local_to_extents(
3270 xfs_trans_t
*tp
, /* transaction pointer */
3271 xfs_inode_t
*ip
, /* incore inode pointer */
3272 xfs_fsblock_t
*firstblock
, /* first block allocated in xaction */
3273 xfs_extlen_t total
, /* total blocks needed by transaction */
3274 int *logflagsp
, /* inode logging flags */
3275 int whichfork
) /* data or attr fork */
3277 int error
; /* error return value */
3278 int flags
; /* logging flags returned */
3279 xfs_ifork_t
*ifp
; /* inode fork pointer */
3282 * We don't want to deal with the case of keeping inode data inline yet.
3283 * So sending the data fork of a regular inode is invalid.
3285 ASSERT(!(S_ISREG(ip
->i_d
.di_mode
) && whichfork
== XFS_DATA_FORK
));
3286 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3287 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
3290 if (ifp
->if_bytes
) {
3291 xfs_alloc_arg_t args
; /* allocation arguments */
3292 xfs_buf_t
*bp
; /* buffer for extent block */
3293 xfs_bmbt_rec_host_t
*ep
;/* extent record pointer */
3296 args
.mp
= ip
->i_mount
;
3297 args
.firstblock
= *firstblock
;
3298 ASSERT((ifp
->if_flags
&
3299 (XFS_IFINLINE
|XFS_IFEXTENTS
|XFS_IFEXTIREC
)) == XFS_IFINLINE
);
3301 * Allocate a block. We know we need only one, since the
3302 * file currently fits in an inode.
3304 if (*firstblock
== NULLFSBLOCK
) {
3305 args
.fsbno
= XFS_INO_TO_FSB(args
.mp
, ip
->i_ino
);
3306 args
.type
= XFS_ALLOCTYPE_START_BNO
;
3308 args
.fsbno
= *firstblock
;
3309 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
3312 args
.mod
= args
.minleft
= args
.alignment
= args
.wasdel
=
3313 args
.isfl
= args
.minalignslop
= 0;
3314 args
.minlen
= args
.maxlen
= args
.prod
= 1;
3315 if ((error
= xfs_alloc_vextent(&args
)))
3318 * Can't fail, the space was reserved.
3320 ASSERT(args
.fsbno
!= NULLFSBLOCK
);
3321 ASSERT(args
.len
== 1);
3322 *firstblock
= args
.fsbno
;
3323 bp
= xfs_btree_get_bufl(args
.mp
, tp
, args
.fsbno
, 0);
3324 memcpy(bp
->b_addr
, ifp
->if_u1
.if_data
, ifp
->if_bytes
);
3325 xfs_trans_log_buf(tp
, bp
, 0, ifp
->if_bytes
- 1);
3326 xfs_bmap_forkoff_reset(args
.mp
, ip
, whichfork
);
3327 xfs_idata_realloc(ip
, -ifp
->if_bytes
, whichfork
);
3328 xfs_iext_add(ifp
, 0, 1);
3329 ep
= xfs_iext_get_ext(ifp
, 0);
3330 xfs_bmbt_set_allf(ep
, 0, args
.fsbno
, 1, XFS_EXT_NORM
);
3331 trace_xfs_bmap_post_update(ip
, 0,
3332 whichfork
== XFS_ATTR_FORK
? BMAP_ATTRFORK
: 0,
3334 XFS_IFORK_NEXT_SET(ip
, whichfork
, 1);
3335 ip
->i_d
.di_nblocks
= 1;
3336 xfs_trans_mod_dquot_byino(tp
, ip
,
3337 XFS_TRANS_DQ_BCOUNT
, 1L);
3338 flags
|= xfs_ilog_fext(whichfork
);
3340 ASSERT(XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0);
3341 xfs_bmap_forkoff_reset(ip
->i_mount
, ip
, whichfork
);
3343 ifp
->if_flags
&= ~XFS_IFINLINE
;
3344 ifp
->if_flags
|= XFS_IFEXTENTS
;
3345 XFS_IFORK_FMT_SET(ip
, whichfork
, XFS_DINODE_FMT_EXTENTS
);
3346 flags
|= XFS_ILOG_CORE
;
3353 * Search the extent records for the entry containing block bno.
3354 * If bno lies in a hole, point to the next entry. If bno lies
3355 * past eof, *eofp will be set, and *prevp will contain the last
3356 * entry (null if none). Else, *lastxp will be set to the index
3357 * of the found entry; *gotp will contain the entry.
3359 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
3360 xfs_bmap_search_multi_extents(
3361 xfs_ifork_t
*ifp
, /* inode fork pointer */
3362 xfs_fileoff_t bno
, /* block number searched for */
3363 int *eofp
, /* out: end of file found */
3364 xfs_extnum_t
*lastxp
, /* out: last extent index */
3365 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
3366 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
3368 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3369 xfs_extnum_t lastx
; /* last extent index */
3372 * Initialize the extent entry structure to catch access to
3373 * uninitialized br_startblock field.
3375 gotp
->br_startoff
= 0xffa5a5a5a5a5a5a5LL
;
3376 gotp
->br_blockcount
= 0xa55a5a5a5a5a5a5aLL
;
3377 gotp
->br_state
= XFS_EXT_INVALID
;
3379 gotp
->br_startblock
= 0xffffa5a5a5a5a5a5LL
;
3381 gotp
->br_startblock
= 0xffffa5a5;
3383 prevp
->br_startoff
= NULLFILEOFF
;
3385 ep
= xfs_iext_bno_to_ext(ifp
, bno
, &lastx
);
3387 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
- 1), prevp
);
3389 if (lastx
< (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))) {
3390 xfs_bmbt_get_all(ep
, gotp
);
3404 * Search the extents list for the inode, for the extent containing bno.
3405 * If bno lies in a hole, point to the next entry. If bno lies past eof,
3406 * *eofp will be set, and *prevp will contain the last entry (null if none).
3407 * Else, *lastxp will be set to the index of the found
3408 * entry; *gotp will contain the entry.
3410 STATIC xfs_bmbt_rec_host_t
* /* pointer to found extent entry */
3411 xfs_bmap_search_extents(
3412 xfs_inode_t
*ip
, /* incore inode pointer */
3413 xfs_fileoff_t bno
, /* block number searched for */
3414 int fork
, /* data or attr fork */
3415 int *eofp
, /* out: end of file found */
3416 xfs_extnum_t
*lastxp
, /* out: last extent index */
3417 xfs_bmbt_irec_t
*gotp
, /* out: extent entry found */
3418 xfs_bmbt_irec_t
*prevp
) /* out: previous extent entry found */
3420 xfs_ifork_t
*ifp
; /* inode fork pointer */
3421 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
3423 XFS_STATS_INC(xs_look_exlist
);
3424 ifp
= XFS_IFORK_PTR(ip
, fork
);
3426 ep
= xfs_bmap_search_multi_extents(ifp
, bno
, eofp
, lastxp
, gotp
, prevp
);
3428 if (unlikely(!(gotp
->br_startblock
) && (*lastxp
!= NULLEXTNUM
) &&
3429 !(XFS_IS_REALTIME_INODE(ip
) && fork
== XFS_DATA_FORK
))) {
3430 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
3431 "Access to block zero in inode %llu "
3432 "start_block: %llx start_off: %llx "
3433 "blkcnt: %llx extent-state: %x lastx: %x\n",
3434 (unsigned long long)ip
->i_ino
,
3435 (unsigned long long)gotp
->br_startblock
,
3436 (unsigned long long)gotp
->br_startoff
,
3437 (unsigned long long)gotp
->br_blockcount
,
3438 gotp
->br_state
, *lastxp
);
3439 *lastxp
= NULLEXTNUM
;
3447 * Compute the worst-case number of indirect blocks that will be used
3448 * for ip's delayed extent of length "len".
3450 STATIC xfs_filblks_t
3451 xfs_bmap_worst_indlen(
3452 xfs_inode_t
*ip
, /* incore inode pointer */
3453 xfs_filblks_t len
) /* delayed extent length */
3455 int level
; /* btree level number */
3456 int maxrecs
; /* maximum record count at this level */
3457 xfs_mount_t
*mp
; /* mount structure */
3458 xfs_filblks_t rval
; /* return value */
3461 maxrecs
= mp
->m_bmap_dmxr
[0];
3462 for (level
= 0, rval
= 0;
3463 level
< XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
);
3466 do_div(len
, maxrecs
);
3469 return rval
+ XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) -
3472 maxrecs
= mp
->m_bmap_dmxr
[1];
3478 * Convert inode from non-attributed to attributed.
3479 * Must not be in a transaction, ip must not be locked.
3481 int /* error code */
3482 xfs_bmap_add_attrfork(
3483 xfs_inode_t
*ip
, /* incore inode pointer */
3484 int size
, /* space new attribute needs */
3485 int rsvd
) /* xact may use reserved blks */
3487 xfs_fsblock_t firstblock
; /* 1st block/ag allocated */
3488 xfs_bmap_free_t flist
; /* freed extent records */
3489 xfs_mount_t
*mp
; /* mount structure */
3490 xfs_trans_t
*tp
; /* transaction pointer */
3491 int blks
; /* space reservation */
3492 int version
= 1; /* superblock attr version */
3493 int committed
; /* xaction was committed */
3494 int logflags
; /* logging flags */
3495 int error
; /* error return value */
3497 ASSERT(XFS_IFORK_Q(ip
) == 0);
3498 ASSERT(ip
->i_df
.if_ext_max
==
3499 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
));
3502 ASSERT(!XFS_NOT_DQATTACHED(mp
, ip
));
3503 tp
= xfs_trans_alloc(mp
, XFS_TRANS_ADDAFORK
);
3504 blks
= XFS_ADDAFORK_SPACE_RES(mp
);
3506 tp
->t_flags
|= XFS_TRANS_RESERVE
;
3507 if ((error
= xfs_trans_reserve(tp
, blks
, XFS_ADDAFORK_LOG_RES(mp
), 0,
3508 XFS_TRANS_PERM_LOG_RES
, XFS_ADDAFORK_LOG_COUNT
)))
3510 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3511 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, blks
, 0, rsvd
?
3512 XFS_QMOPT_RES_REGBLKS
| XFS_QMOPT_FORCE_RES
:
3513 XFS_QMOPT_RES_REGBLKS
);
3515 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3516 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3519 if (XFS_IFORK_Q(ip
))
3521 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
) {
3523 * For inodes coming from pre-6.2 filesystems.
3525 ASSERT(ip
->i_d
.di_aformat
== 0);
3526 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
3528 ASSERT(ip
->i_d
.di_anextents
== 0);
3530 xfs_trans_ijoin_ref(tp
, ip
, XFS_ILOCK_EXCL
);
3531 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3533 switch (ip
->i_d
.di_format
) {
3534 case XFS_DINODE_FMT_DEV
:
3535 ip
->i_d
.di_forkoff
= roundup(sizeof(xfs_dev_t
), 8) >> 3;
3537 case XFS_DINODE_FMT_UUID
:
3538 ip
->i_d
.di_forkoff
= roundup(sizeof(uuid_t
), 8) >> 3;
3540 case XFS_DINODE_FMT_LOCAL
:
3541 case XFS_DINODE_FMT_EXTENTS
:
3542 case XFS_DINODE_FMT_BTREE
:
3543 ip
->i_d
.di_forkoff
= xfs_attr_shortform_bytesfit(ip
, size
);
3544 if (!ip
->i_d
.di_forkoff
)
3545 ip
->i_d
.di_forkoff
= xfs_default_attroffset(ip
) >> 3;
3546 else if (mp
->m_flags
& XFS_MOUNT_ATTR2
)
3551 error
= XFS_ERROR(EINVAL
);
3554 ip
->i_df
.if_ext_max
=
3555 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
3556 ASSERT(ip
->i_afp
== NULL
);
3557 ip
->i_afp
= kmem_zone_zalloc(xfs_ifork_zone
, KM_SLEEP
);
3558 ip
->i_afp
->if_ext_max
=
3559 XFS_IFORK_ASIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
);
3560 ip
->i_afp
->if_flags
= XFS_IFEXTENTS
;
3562 xfs_bmap_init(&flist
, &firstblock
);
3563 switch (ip
->i_d
.di_format
) {
3564 case XFS_DINODE_FMT_LOCAL
:
3565 error
= xfs_bmap_add_attrfork_local(tp
, ip
, &firstblock
, &flist
,
3568 case XFS_DINODE_FMT_EXTENTS
:
3569 error
= xfs_bmap_add_attrfork_extents(tp
, ip
, &firstblock
,
3572 case XFS_DINODE_FMT_BTREE
:
3573 error
= xfs_bmap_add_attrfork_btree(tp
, ip
, &firstblock
, &flist
,
3581 xfs_trans_log_inode(tp
, ip
, logflags
);
3584 if (!xfs_sb_version_hasattr(&mp
->m_sb
) ||
3585 (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2)) {
3586 __int64_t sbfields
= 0;
3588 spin_lock(&mp
->m_sb_lock
);
3589 if (!xfs_sb_version_hasattr(&mp
->m_sb
)) {
3590 xfs_sb_version_addattr(&mp
->m_sb
);
3591 sbfields
|= XFS_SB_VERSIONNUM
;
3593 if (!xfs_sb_version_hasattr2(&mp
->m_sb
) && version
== 2) {
3594 xfs_sb_version_addattr2(&mp
->m_sb
);
3595 sbfields
|= (XFS_SB_VERSIONNUM
| XFS_SB_FEATURES2
);
3598 spin_unlock(&mp
->m_sb_lock
);
3599 xfs_mod_sb(tp
, sbfields
);
3601 spin_unlock(&mp
->m_sb_lock
);
3603 if ((error
= xfs_bmap_finish(&tp
, &flist
, &committed
)))
3605 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3606 ASSERT(ip
->i_df
.if_ext_max
==
3607 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
));
3610 xfs_bmap_cancel(&flist
);
3612 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3614 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
3615 ASSERT(ip
->i_df
.if_ext_max
==
3616 XFS_IFORK_DSIZE(ip
) / (uint
)sizeof(xfs_bmbt_rec_t
));
3621 * Add the extent to the list of extents to be free at transaction end.
3622 * The list is maintained sorted (by block number).
3627 xfs_fsblock_t bno
, /* fs block number of extent */
3628 xfs_filblks_t len
, /* length of extent */
3629 xfs_bmap_free_t
*flist
, /* list of extents */
3630 xfs_mount_t
*mp
) /* mount point structure */
3632 xfs_bmap_free_item_t
*cur
; /* current (next) element */
3633 xfs_bmap_free_item_t
*new; /* new element */
3634 xfs_bmap_free_item_t
*prev
; /* previous element */
3636 xfs_agnumber_t agno
;
3637 xfs_agblock_t agbno
;
3639 ASSERT(bno
!= NULLFSBLOCK
);
3641 ASSERT(len
<= MAXEXTLEN
);
3642 ASSERT(!isnullstartblock(bno
));
3643 agno
= XFS_FSB_TO_AGNO(mp
, bno
);
3644 agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
3645 ASSERT(agno
< mp
->m_sb
.sb_agcount
);
3646 ASSERT(agbno
< mp
->m_sb
.sb_agblocks
);
3647 ASSERT(len
< mp
->m_sb
.sb_agblocks
);
3648 ASSERT(agbno
+ len
<= mp
->m_sb
.sb_agblocks
);
3650 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
3651 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
3652 new->xbfi_startblock
= bno
;
3653 new->xbfi_blockcount
= (xfs_extlen_t
)len
;
3654 for (prev
= NULL
, cur
= flist
->xbf_first
;
3656 prev
= cur
, cur
= cur
->xbfi_next
) {
3657 if (cur
->xbfi_startblock
>= bno
)
3661 prev
->xbfi_next
= new;
3663 flist
->xbf_first
= new;
3664 new->xbfi_next
= cur
;
3669 * Compute and fill in the value of the maximum depth of a bmap btree
3670 * in this filesystem. Done once, during mount.
3673 xfs_bmap_compute_maxlevels(
3674 xfs_mount_t
*mp
, /* file system mount structure */
3675 int whichfork
) /* data or attr fork */
3677 int level
; /* btree level */
3678 uint maxblocks
; /* max blocks at this level */
3679 uint maxleafents
; /* max leaf entries possible */
3680 int maxrootrecs
; /* max records in root block */
3681 int minleafrecs
; /* min records in leaf block */
3682 int minnoderecs
; /* min records in node block */
3683 int sz
; /* root block size */
3686 * The maximum number of extents in a file, hence the maximum
3687 * number of leaf entries, is controlled by the type of di_nextents
3688 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3689 * (a signed 16-bit number, xfs_aextnum_t).
3691 * Note that we can no longer assume that if we are in ATTR1 that
3692 * the fork offset of all the inodes will be
3693 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3694 * with ATTR2 and then mounted back with ATTR1, keeping the
3695 * di_forkoff's fixed but probably at various positions. Therefore,
3696 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3697 * of a minimum size available.
3699 if (whichfork
== XFS_DATA_FORK
) {
3700 maxleafents
= MAXEXTNUM
;
3701 sz
= XFS_BMDR_SPACE_CALC(MINDBTPTRS
);
3703 maxleafents
= MAXAEXTNUM
;
3704 sz
= XFS_BMDR_SPACE_CALC(MINABTPTRS
);
3706 maxrootrecs
= xfs_bmdr_maxrecs(mp
, sz
, 0);
3707 minleafrecs
= mp
->m_bmap_dmnr
[0];
3708 minnoderecs
= mp
->m_bmap_dmnr
[1];
3709 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
3710 for (level
= 1; maxblocks
> 1; level
++) {
3711 if (maxblocks
<= maxrootrecs
)
3714 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
3716 mp
->m_bm_maxlevels
[whichfork
] = level
;
3720 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3721 * caller. Frees all the extents that need freeing, which must be done
3722 * last due to locking considerations. We never free any extents in
3723 * the first transaction.
3725 * Return 1 if the given transaction was committed and a new one
3726 * started, and 0 otherwise in the committed parameter.
3730 xfs_trans_t
**tp
, /* transaction pointer addr */
3731 xfs_bmap_free_t
*flist
, /* i/o: list extents to free */
3732 int *committed
) /* xact committed or not */
3734 xfs_efd_log_item_t
*efd
; /* extent free data */
3735 xfs_efi_log_item_t
*efi
; /* extent free intention */
3736 int error
; /* error return value */
3737 xfs_bmap_free_item_t
*free
; /* free extent item */
3738 unsigned int logres
; /* new log reservation */
3739 unsigned int logcount
; /* new log count */
3740 xfs_mount_t
*mp
; /* filesystem mount structure */
3741 xfs_bmap_free_item_t
*next
; /* next item on free list */
3742 xfs_trans_t
*ntp
; /* new transaction pointer */
3744 ASSERT((*tp
)->t_flags
& XFS_TRANS_PERM_LOG_RES
);
3745 if (flist
->xbf_count
== 0) {
3750 efi
= xfs_trans_get_efi(ntp
, flist
->xbf_count
);
3751 for (free
= flist
->xbf_first
; free
; free
= free
->xbfi_next
)
3752 xfs_trans_log_efi_extent(ntp
, efi
, free
->xbfi_startblock
,
3753 free
->xbfi_blockcount
);
3754 logres
= ntp
->t_log_res
;
3755 logcount
= ntp
->t_log_count
;
3756 ntp
= xfs_trans_dup(*tp
);
3757 error
= xfs_trans_commit(*tp
, 0);
3761 * We have a new transaction, so we should return committed=1,
3762 * even though we're returning an error.
3768 * transaction commit worked ok so we can drop the extra ticket
3769 * reference that we gained in xfs_trans_dup()
3771 xfs_log_ticket_put(ntp
->t_ticket
);
3773 if ((error
= xfs_trans_reserve(ntp
, 0, logres
, 0, XFS_TRANS_PERM_LOG_RES
,
3776 efd
= xfs_trans_get_efd(ntp
, efi
, flist
->xbf_count
);
3777 for (free
= flist
->xbf_first
; free
!= NULL
; free
= next
) {
3778 next
= free
->xbfi_next
;
3779 if ((error
= xfs_free_extent(ntp
, free
->xbfi_startblock
,
3780 free
->xbfi_blockcount
))) {
3782 * The bmap free list will be cleaned up at a
3783 * higher level. The EFI will be canceled when
3784 * this transaction is aborted.
3785 * Need to force shutdown here to make sure it
3786 * happens, since this transaction may not be
3790 if (!XFS_FORCED_SHUTDOWN(mp
))
3791 xfs_force_shutdown(mp
,
3792 (error
== EFSCORRUPTED
) ?
3793 SHUTDOWN_CORRUPT_INCORE
:
3794 SHUTDOWN_META_IO_ERROR
);
3797 xfs_trans_log_efd_extent(ntp
, efd
, free
->xbfi_startblock
,
3798 free
->xbfi_blockcount
);
3799 xfs_bmap_del_free(flist
, NULL
, free
);
3805 * Free up any items left in the list.
3809 xfs_bmap_free_t
*flist
) /* list of bmap_free_items */
3811 xfs_bmap_free_item_t
*free
; /* free list item */
3812 xfs_bmap_free_item_t
*next
;
3814 if (flist
->xbf_count
== 0)
3816 ASSERT(flist
->xbf_first
!= NULL
);
3817 for (free
= flist
->xbf_first
; free
; free
= next
) {
3818 next
= free
->xbfi_next
;
3819 xfs_bmap_del_free(flist
, NULL
, free
);
3821 ASSERT(flist
->xbf_count
== 0);
3825 * Returns the file-relative block number of the first unused block(s)
3826 * in the file with at least "len" logically contiguous blocks free.
3827 * This is the lowest-address hole if the file has holes, else the first block
3828 * past the end of file.
3829 * Return 0 if the file is currently local (in-inode).
3832 xfs_bmap_first_unused(
3833 xfs_trans_t
*tp
, /* transaction pointer */
3834 xfs_inode_t
*ip
, /* incore inode */
3835 xfs_extlen_t len
, /* size of hole to find */
3836 xfs_fileoff_t
*first_unused
, /* unused block */
3837 int whichfork
) /* data or attr fork */
3839 int error
; /* error return value */
3840 int idx
; /* extent record index */
3841 xfs_ifork_t
*ifp
; /* inode fork pointer */
3842 xfs_fileoff_t lastaddr
; /* last block number seen */
3843 xfs_fileoff_t lowest
; /* lowest useful block */
3844 xfs_fileoff_t max
; /* starting useful block */
3845 xfs_fileoff_t off
; /* offset for this block */
3846 xfs_extnum_t nextents
; /* number of extent entries */
3848 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
||
3849 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
||
3850 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
);
3851 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
3855 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3856 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
3857 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
3859 lowest
= *first_unused
;
3860 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3861 for (idx
= 0, lastaddr
= 0, max
= lowest
; idx
< nextents
; idx
++) {
3862 xfs_bmbt_rec_host_t
*ep
= xfs_iext_get_ext(ifp
, idx
);
3863 off
= xfs_bmbt_get_startoff(ep
);
3865 * See if the hole before this extent will work.
3867 if (off
>= lowest
+ len
&& off
- max
>= len
) {
3868 *first_unused
= max
;
3871 lastaddr
= off
+ xfs_bmbt_get_blockcount(ep
);
3872 max
= XFS_FILEOFF_MAX(lastaddr
, lowest
);
3874 *first_unused
= max
;
3879 * Returns the file-relative block number of the last block + 1 before
3880 * last_block (input value) in the file.
3881 * This is not based on i_size, it is based on the extent records.
3882 * Returns 0 for local files, as they do not have extent records.
3885 xfs_bmap_last_before(
3886 xfs_trans_t
*tp
, /* transaction pointer */
3887 xfs_inode_t
*ip
, /* incore inode */
3888 xfs_fileoff_t
*last_block
, /* last block */
3889 int whichfork
) /* data or attr fork */
3891 xfs_fileoff_t bno
; /* input file offset */
3892 int eof
; /* hit end of file */
3893 xfs_bmbt_rec_host_t
*ep
; /* pointer to last extent */
3894 int error
; /* error return value */
3895 xfs_bmbt_irec_t got
; /* current extent value */
3896 xfs_ifork_t
*ifp
; /* inode fork pointer */
3897 xfs_extnum_t lastx
; /* last extent used */
3898 xfs_bmbt_irec_t prev
; /* previous extent value */
3900 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
3901 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
3902 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
)
3903 return XFS_ERROR(EIO
);
3904 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
3908 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3909 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
3910 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
3912 bno
= *last_block
- 1;
3913 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
3915 if (eof
|| xfs_bmbt_get_startoff(ep
) > bno
) {
3916 if (prev
.br_startoff
== NULLFILEOFF
)
3919 *last_block
= prev
.br_startoff
+ prev
.br_blockcount
;
3922 * Otherwise *last_block is already the right answer.
3928 * Returns the file-relative block number of the first block past eof in
3929 * the file. This is not based on i_size, it is based on the extent records.
3930 * Returns 0 for local files, as they do not have extent records.
3933 xfs_bmap_last_offset(
3934 xfs_trans_t
*tp
, /* transaction pointer */
3935 xfs_inode_t
*ip
, /* incore inode */
3936 xfs_fileoff_t
*last_block
, /* last block */
3937 int whichfork
) /* data or attr fork */
3939 xfs_bmbt_rec_host_t
*ep
; /* pointer to last extent */
3940 int error
; /* error return value */
3941 xfs_ifork_t
*ifp
; /* inode fork pointer */
3942 xfs_extnum_t nextents
; /* number of extent entries */
3944 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
3945 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
3946 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
)
3947 return XFS_ERROR(EIO
);
3948 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
3952 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3953 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
3954 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
3956 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
3961 ep
= xfs_iext_get_ext(ifp
, nextents
- 1);
3962 *last_block
= xfs_bmbt_get_startoff(ep
) + xfs_bmbt_get_blockcount(ep
);
3967 * Returns whether the selected fork of the inode has exactly one
3968 * block or not. For the data fork we check this matches di_size,
3969 * implying the file's range is 0..bsize-1.
3971 int /* 1=>1 block, 0=>otherwise */
3973 xfs_inode_t
*ip
, /* incore inode */
3974 int whichfork
) /* data or attr fork */
3976 xfs_bmbt_rec_host_t
*ep
; /* ptr to fork's extent */
3977 xfs_ifork_t
*ifp
; /* inode fork pointer */
3978 int rval
; /* return value */
3979 xfs_bmbt_irec_t s
; /* internal version of extent */
3982 if (whichfork
== XFS_DATA_FORK
) {
3983 return S_ISREG(ip
->i_d
.di_mode
) ?
3984 (ip
->i_size
== ip
->i_mount
->m_sb
.sb_blocksize
) :
3985 (ip
->i_d
.di_size
== ip
->i_mount
->m_sb
.sb_blocksize
);
3988 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) != 1)
3990 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
3992 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3993 ASSERT(ifp
->if_flags
& XFS_IFEXTENTS
);
3994 ep
= xfs_iext_get_ext(ifp
, 0);
3995 xfs_bmbt_get_all(ep
, &s
);
3996 rval
= s
.br_startoff
== 0 && s
.br_blockcount
== 1;
3997 if (rval
&& whichfork
== XFS_DATA_FORK
)
3998 ASSERT(ip
->i_size
== ip
->i_mount
->m_sb
.sb_blocksize
);
4003 xfs_bmap_sanity_check(
4004 struct xfs_mount
*mp
,
4008 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4010 if (block
->bb_magic
!= cpu_to_be32(XFS_BMAP_MAGIC
) ||
4011 be16_to_cpu(block
->bb_level
) != level
||
4012 be16_to_cpu(block
->bb_numrecs
) == 0 ||
4013 be16_to_cpu(block
->bb_numrecs
) > mp
->m_bmap_dmxr
[level
!= 0])
4019 * Read in the extents to if_extents.
4020 * All inode fields are set up by caller, we just traverse the btree
4021 * and copy the records in. If the file system cannot contain unwritten
4022 * extents, the records are checked for no "state" flags.
4025 xfs_bmap_read_extents(
4026 xfs_trans_t
*tp
, /* transaction pointer */
4027 xfs_inode_t
*ip
, /* incore inode */
4028 int whichfork
) /* data or attr fork */
4030 struct xfs_btree_block
*block
; /* current btree block */
4031 xfs_fsblock_t bno
; /* block # of "block" */
4032 xfs_buf_t
*bp
; /* buffer for "block" */
4033 int error
; /* error return value */
4034 xfs_exntfmt_t exntf
; /* XFS_EXTFMT_NOSTATE, if checking */
4035 xfs_extnum_t i
, j
; /* index into the extents list */
4036 xfs_ifork_t
*ifp
; /* fork structure */
4037 int level
; /* btree level, for checking */
4038 xfs_mount_t
*mp
; /* file system mount structure */
4039 __be64
*pp
; /* pointer to block address */
4041 xfs_extnum_t room
; /* number of entries there's room for */
4045 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4046 exntf
= (whichfork
!= XFS_DATA_FORK
) ? XFS_EXTFMT_NOSTATE
:
4047 XFS_EXTFMT_INODE(ip
);
4048 block
= ifp
->if_broot
;
4050 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4052 level
= be16_to_cpu(block
->bb_level
);
4054 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
4055 bno
= be64_to_cpu(*pp
);
4056 ASSERT(bno
!= NULLDFSBNO
);
4057 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
4058 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
4060 * Go down the tree until leaf level is reached, following the first
4061 * pointer (leftmost) at each level.
4063 while (level
-- > 0) {
4064 if ((error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
4065 XFS_BMAP_BTREE_REF
)))
4067 block
= XFS_BUF_TO_BLOCK(bp
);
4068 XFS_WANT_CORRUPTED_GOTO(
4069 xfs_bmap_sanity_check(mp
, bp
, level
),
4073 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
4074 bno
= be64_to_cpu(*pp
);
4075 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
4076 xfs_trans_brelse(tp
, bp
);
4079 * Here with bp and block set to the leftmost leaf node in the tree.
4081 room
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
4084 * Loop over all leaf nodes. Copy information to the extent records.
4087 xfs_bmbt_rec_t
*frp
;
4088 xfs_fsblock_t nextbno
;
4089 xfs_extnum_t num_recs
;
4093 num_recs
= xfs_btree_get_numrecs(block
);
4094 if (unlikely(i
+ num_recs
> room
)) {
4095 ASSERT(i
+ num_recs
<= room
);
4096 xfs_warn(ip
->i_mount
,
4097 "corrupt dinode %Lu, (btree extents).",
4098 (unsigned long long) ip
->i_ino
);
4099 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4100 XFS_ERRLEVEL_LOW
, ip
->i_mount
, block
);
4103 XFS_WANT_CORRUPTED_GOTO(
4104 xfs_bmap_sanity_check(mp
, bp
, 0),
4107 * Read-ahead the next leaf block, if any.
4109 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
4110 if (nextbno
!= NULLFSBLOCK
)
4111 xfs_btree_reada_bufl(mp
, nextbno
, 1);
4113 * Copy records into the extent records.
4115 frp
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
4117 for (j
= 0; j
< num_recs
; j
++, i
++, frp
++) {
4118 xfs_bmbt_rec_host_t
*trp
= xfs_iext_get_ext(ifp
, i
);
4119 trp
->l0
= be64_to_cpu(frp
->l0
);
4120 trp
->l1
= be64_to_cpu(frp
->l1
);
4122 if (exntf
== XFS_EXTFMT_NOSTATE
) {
4124 * Check all attribute bmap btree records and
4125 * any "older" data bmap btree records for a
4126 * set bit in the "extent flag" position.
4128 if (unlikely(xfs_check_nostate_extents(ifp
,
4129 start
, num_recs
))) {
4130 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4136 xfs_trans_brelse(tp
, bp
);
4139 * If we've reached the end, stop.
4141 if (bno
== NULLFSBLOCK
)
4143 if ((error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
4144 XFS_BMAP_BTREE_REF
)))
4146 block
= XFS_BUF_TO_BLOCK(bp
);
4148 ASSERT(i
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
4149 ASSERT(i
== XFS_IFORK_NEXTENTS(ip
, whichfork
));
4150 XFS_BMAP_TRACE_EXLIST(ip
, i
, whichfork
);
4153 xfs_trans_brelse(tp
, bp
);
4154 return XFS_ERROR(EFSCORRUPTED
);
4159 * Add bmap trace insert entries for all the contents of the extent records.
4162 xfs_bmap_trace_exlist(
4163 xfs_inode_t
*ip
, /* incore inode pointer */
4164 xfs_extnum_t cnt
, /* count of entries in the list */
4165 int whichfork
, /* data or attr fork */
4166 unsigned long caller_ip
)
4168 xfs_extnum_t idx
; /* extent record index */
4169 xfs_ifork_t
*ifp
; /* inode fork pointer */
4172 if (whichfork
== XFS_ATTR_FORK
)
4173 state
|= BMAP_ATTRFORK
;
4175 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4176 ASSERT(cnt
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
)));
4177 for (idx
= 0; idx
< cnt
; idx
++)
4178 trace_xfs_extlist(ip
, idx
, whichfork
, caller_ip
);
4182 * Validate that the bmbt_irecs being returned from bmapi are valid
4183 * given the callers original parameters. Specifically check the
4184 * ranges of the returned irecs to ensure that they only extent beyond
4185 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4188 xfs_bmap_validate_ret(
4192 xfs_bmbt_irec_t
*mval
,
4196 int i
; /* index to map values */
4198 ASSERT(ret_nmap
<= nmap
);
4200 for (i
= 0; i
< ret_nmap
; i
++) {
4201 ASSERT(mval
[i
].br_blockcount
> 0);
4202 if (!(flags
& XFS_BMAPI_ENTIRE
)) {
4203 ASSERT(mval
[i
].br_startoff
>= bno
);
4204 ASSERT(mval
[i
].br_blockcount
<= len
);
4205 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
<=
4208 ASSERT(mval
[i
].br_startoff
< bno
+ len
);
4209 ASSERT(mval
[i
].br_startoff
+ mval
[i
].br_blockcount
>
4213 mval
[i
- 1].br_startoff
+ mval
[i
- 1].br_blockcount
==
4214 mval
[i
].br_startoff
);
4215 if (flags
& XFS_BMAPI_WRITE
)
4216 ASSERT(mval
[i
].br_startblock
!= DELAYSTARTBLOCK
&&
4217 mval
[i
].br_startblock
!= HOLESTARTBLOCK
);
4218 ASSERT(mval
[i
].br_state
== XFS_EXT_NORM
||
4219 mval
[i
].br_state
== XFS_EXT_UNWRITTEN
);
4226 * Trim the returned map to the required bounds
4230 struct xfs_bmbt_irec
*mval
,
4231 struct xfs_bmbt_irec
*got
,
4239 if ((flags
& XFS_BMAPI_ENTIRE
) ||
4240 got
->br_startoff
+ got
->br_blockcount
<= obno
) {
4242 if (isnullstartblock(got
->br_startblock
))
4243 mval
->br_startblock
= DELAYSTARTBLOCK
;
4249 ASSERT((*bno
>= obno
) || (n
== 0));
4251 mval
->br_startoff
= *bno
;
4252 if (isnullstartblock(got
->br_startblock
))
4253 mval
->br_startblock
= DELAYSTARTBLOCK
;
4255 mval
->br_startblock
= got
->br_startblock
+
4256 (*bno
- got
->br_startoff
);
4258 * Return the minimum of what we got and what we asked for for
4259 * the length. We can use the len variable here because it is
4260 * modified below and we could have been there before coming
4261 * here if the first part of the allocation didn't overlap what
4264 mval
->br_blockcount
= XFS_FILBLKS_MIN(end
- *bno
,
4265 got
->br_blockcount
- (*bno
- got
->br_startoff
));
4266 mval
->br_state
= got
->br_state
;
4267 ASSERT(mval
->br_blockcount
<= len
);
4272 * Update and validate the extent map to return
4275 xfs_bmapi_update_map(
4276 struct xfs_bmbt_irec
**map
,
4284 xfs_bmbt_irec_t
*mval
= *map
;
4286 ASSERT((flags
& XFS_BMAPI_ENTIRE
) ||
4287 ((mval
->br_startoff
+ mval
->br_blockcount
) <= end
));
4288 ASSERT((flags
& XFS_BMAPI_ENTIRE
) || (mval
->br_blockcount
<= *len
) ||
4289 (mval
->br_startoff
< obno
));
4291 *bno
= mval
->br_startoff
+ mval
->br_blockcount
;
4293 if (*n
> 0 && mval
->br_startoff
== mval
[-1].br_startoff
) {
4294 /* update previous map with new information */
4295 ASSERT(mval
->br_startblock
== mval
[-1].br_startblock
);
4296 ASSERT(mval
->br_blockcount
> mval
[-1].br_blockcount
);
4297 ASSERT(mval
->br_state
== mval
[-1].br_state
);
4298 mval
[-1].br_blockcount
= mval
->br_blockcount
;
4299 mval
[-1].br_state
= mval
->br_state
;
4300 } else if (*n
> 0 && mval
->br_startblock
!= DELAYSTARTBLOCK
&&
4301 mval
[-1].br_startblock
!= DELAYSTARTBLOCK
&&
4302 mval
[-1].br_startblock
!= HOLESTARTBLOCK
&&
4303 mval
->br_startblock
== mval
[-1].br_startblock
+
4304 mval
[-1].br_blockcount
&&
4305 ((flags
& XFS_BMAPI_IGSTATE
) ||
4306 mval
[-1].br_state
== mval
->br_state
)) {
4307 ASSERT(mval
->br_startoff
==
4308 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
);
4309 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4310 } else if (*n
> 0 &&
4311 mval
->br_startblock
== DELAYSTARTBLOCK
&&
4312 mval
[-1].br_startblock
== DELAYSTARTBLOCK
&&
4313 mval
->br_startoff
==
4314 mval
[-1].br_startoff
+ mval
[-1].br_blockcount
) {
4315 mval
[-1].br_blockcount
+= mval
->br_blockcount
;
4316 mval
[-1].br_state
= mval
->br_state
;
4317 } else if (!((*n
== 0) &&
4318 ((mval
->br_startoff
+ mval
->br_blockcount
) <=
4327 * Map file blocks to filesystem blocks without allocation.
4331 struct xfs_inode
*ip
,
4334 struct xfs_bmbt_irec
*mval
,
4338 struct xfs_mount
*mp
= ip
->i_mount
;
4339 struct xfs_ifork
*ifp
;
4340 struct xfs_bmbt_irec got
;
4341 struct xfs_bmbt_irec prev
;
4348 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4349 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4352 ASSERT(!(flags
& ~(XFS_BMAPI_ATTRFORK
|XFS_BMAPI_ENTIRE
|
4353 XFS_BMAPI_IGSTATE
)));
4355 if (unlikely(XFS_TEST_ERROR(
4356 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4357 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
),
4358 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4359 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW
, mp
);
4360 return XFS_ERROR(EFSCORRUPTED
);
4363 if (XFS_FORCED_SHUTDOWN(mp
))
4364 return XFS_ERROR(EIO
);
4366 XFS_STATS_INC(xs_blk_mapr
);
4368 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4369 ASSERT(ifp
->if_ext_max
==
4370 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
4372 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4373 error
= xfs_iread_extents(NULL
, ip
, whichfork
);
4378 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
, &prev
);
4382 while (bno
< end
&& n
< *nmap
) {
4383 /* Reading past eof, act as though there's a hole up to end. */
4385 got
.br_startoff
= end
;
4386 if (got
.br_startoff
> bno
) {
4387 /* Reading in a hole. */
4388 mval
->br_startoff
= bno
;
4389 mval
->br_startblock
= HOLESTARTBLOCK
;
4390 mval
->br_blockcount
=
4391 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4392 mval
->br_state
= XFS_EXT_NORM
;
4393 bno
+= mval
->br_blockcount
;
4394 len
-= mval
->br_blockcount
;
4400 /* set up the extent map to return. */
4401 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4402 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4404 /* If we're done, stop now. */
4405 if (bno
>= end
|| n
>= *nmap
)
4408 /* Else go on to the next record. */
4409 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4410 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4419 xfs_bmapi_reserve_delalloc(
4420 struct xfs_inode
*ip
,
4423 struct xfs_bmbt_irec
*got
,
4424 struct xfs_bmbt_irec
*prev
,
4425 xfs_extnum_t
*lastx
,
4428 struct xfs_mount
*mp
= ip
->i_mount
;
4429 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4431 xfs_extlen_t indlen
;
4432 char rt
= XFS_IS_REALTIME_INODE(ip
);
4436 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
);
4438 alen
= XFS_FILBLKS_MIN(alen
, got
->br_startoff
- aoff
);
4440 /* Figure out the extent size, adjust alen */
4441 extsz
= xfs_get_extsz_hint(ip
);
4444 * Make sure we don't exceed a single extent length when we
4445 * align the extent by reducing length we are going to
4446 * allocate by the maximum amount extent size aligment may
4449 alen
= XFS_FILBLKS_MIN(len
, MAXEXTLEN
- (2 * extsz
- 1));
4450 error
= xfs_bmap_extsize_align(mp
, got
, prev
, extsz
, rt
, eof
,
4451 1, 0, &aoff
, &alen
);
4456 extsz
= alen
/ mp
->m_sb
.sb_rextsize
;
4459 * Make a transaction-less quota reservation for delayed allocation
4460 * blocks. This number gets adjusted later. We return if we haven't
4461 * allocated blocks already inside this loop.
4463 error
= xfs_trans_reserve_quota_nblks(NULL
, ip
, (long)alen
, 0,
4464 rt
? XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4469 * Split changing sb for alen and indlen since they could be coming
4470 * from different places.
4472 indlen
= (xfs_extlen_t
)xfs_bmap_worst_indlen(ip
, alen
);
4476 error
= xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
4477 -((int64_t)extsz
), 0);
4479 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4480 -((int64_t)alen
), 0);
4484 goto out_unreserve_quota
;
4486 error
= xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
4487 -((int64_t)indlen
), 0);
4489 goto out_unreserve_blocks
;
4492 ip
->i_delayed_blks
+= alen
;
4494 got
->br_startoff
= aoff
;
4495 got
->br_startblock
= nullstartblock(indlen
);
4496 got
->br_blockcount
= alen
;
4497 got
->br_state
= XFS_EXT_NORM
;
4498 xfs_bmap_add_extent_hole_delay(ip
, lastx
, got
);
4501 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4502 * might have merged it into one of the neighbouring ones.
4504 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
), got
);
4506 ASSERT(got
->br_startoff
<= aoff
);
4507 ASSERT(got
->br_startoff
+ got
->br_blockcount
>= aoff
+ alen
);
4508 ASSERT(isnullstartblock(got
->br_startblock
));
4509 ASSERT(got
->br_state
== XFS_EXT_NORM
);
4512 out_unreserve_blocks
:
4514 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
, extsz
, 0);
4516 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
, alen
, 0);
4517 out_unreserve_quota
:
4518 if (XFS_IS_QUOTA_ON(mp
))
4519 xfs_trans_unreserve_quota_nblks(NULL
, ip
, alen
, 0, rt
?
4520 XFS_QMOPT_RES_RTBLKS
: XFS_QMOPT_RES_REGBLKS
);
4525 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4529 struct xfs_inode
*ip
, /* incore inode */
4530 xfs_fileoff_t bno
, /* starting file offs. mapped */
4531 xfs_filblks_t len
, /* length to map in file */
4532 struct xfs_bmbt_irec
*mval
, /* output: map values */
4533 int *nmap
, /* i/o: mval size/count */
4534 int flags
) /* XFS_BMAPI_... */
4536 struct xfs_mount
*mp
= ip
->i_mount
;
4537 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
4538 struct xfs_bmbt_irec got
; /* current file extent record */
4539 struct xfs_bmbt_irec prev
; /* previous file extent record */
4540 xfs_fileoff_t obno
; /* old block number (offset) */
4541 xfs_fileoff_t end
; /* end of mapped file region */
4542 xfs_extnum_t lastx
; /* last useful extent number */
4543 int eof
; /* we've hit the end of extents */
4544 int n
= 0; /* current extent index */
4548 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
);
4549 ASSERT(!(flags
& ~XFS_BMAPI_ENTIRE
));
4551 if (unlikely(XFS_TEST_ERROR(
4552 (XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_EXTENTS
&&
4553 XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_BTREE
),
4554 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4555 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW
, mp
);
4556 return XFS_ERROR(EFSCORRUPTED
);
4559 if (XFS_FORCED_SHUTDOWN(mp
))
4560 return XFS_ERROR(EIO
);
4562 XFS_STATS_INC(xs_blk_mapw
);
4564 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
4565 error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
);
4570 xfs_bmap_search_extents(ip
, bno
, XFS_DATA_FORK
, &eof
, &lastx
, &got
, &prev
);
4574 while (bno
< end
&& n
< *nmap
) {
4575 if (eof
|| got
.br_startoff
> bno
) {
4576 error
= xfs_bmapi_reserve_delalloc(ip
, bno
, len
, &got
,
4577 &prev
, &lastx
, eof
);
4587 /* set up the extent map to return. */
4588 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4589 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4591 /* If we're done, stop now. */
4592 if (bno
>= end
|| n
>= *nmap
)
4595 /* Else go on to the next record. */
4597 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
))
4598 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, lastx
), &got
);
4610 struct xfs_bmalloca
*bma
,
4611 xfs_extnum_t
*lastx
,
4612 struct xfs_btree_cur
**cur
,
4613 xfs_fsblock_t
*firstblock
,
4614 struct xfs_bmap_free
*flist
,
4619 struct xfs_mount
*mp
= bma
->ip
->i_mount
;
4620 int whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4621 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4622 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(bma
->ip
, whichfork
);
4629 rt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(bma
->ip
);
4632 * For the wasdelay case, we could also just allocate the stuff asked
4633 * for in this bmap call but that wouldn't be as good.
4636 alen
= (xfs_extlen_t
)bma
->gotp
->br_blockcount
;
4637 aoff
= bma
->gotp
->br_startoff
;
4638 if (*lastx
!= NULLEXTNUM
&& *lastx
) {
4639 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
- 1),
4643 alen
= (xfs_extlen_t
)XFS_FILBLKS_MIN(bma
->alen
, MAXEXTLEN
);
4645 alen
= (xfs_extlen_t
)XFS_FILBLKS_MIN(alen
,
4646 bma
->gotp
->br_startoff
- bma
->off
);
4651 * Indicate if this is the first user data in the file, or just any
4654 if (!(flags
& XFS_BMAPI_METADATA
)) {
4655 bma
->userdata
= (aoff
== 0) ?
4656 XFS_ALLOC_INITIAL_USER_DATA
: XFS_ALLOC_USERDATA
;
4660 * Fill in changeable bma fields.
4664 bma
->firstblock
= *firstblock
;
4665 bma
->minlen
= (flags
& XFS_BMAPI_CONTIG
) ? alen
: 1;
4666 bma
->low
= flist
->xbf_low
;
4670 * Only want to do the alignment at the eof if it is userdata and
4671 * allocation length is larger than a stripe unit.
4673 if (mp
->m_dalign
&& alen
>= mp
->m_dalign
&&
4674 !(flags
& XFS_BMAPI_METADATA
) && whichfork
== XFS_DATA_FORK
) {
4675 error
= xfs_bmap_isaeof(bma
->ip
, aoff
, whichfork
, &bma
->aeof
);
4680 error
= xfs_bmap_alloc(bma
);
4685 * Copy out result fields.
4688 flist
->xbf_low
= bma
->low
;
4691 ASSERT(*firstblock
== NULLFSBLOCK
||
4692 XFS_FSB_TO_AGNO(mp
, *firstblock
) ==
4693 XFS_FSB_TO_AGNO(mp
, bma
->firstblock
) ||
4695 XFS_FSB_TO_AGNO(mp
, *firstblock
) <
4696 XFS_FSB_TO_AGNO(mp
, bma
->firstblock
)));
4697 *firstblock
= bma
->firstblock
;
4699 (*cur
)->bc_private
.b
.firstblock
= *firstblock
;
4700 if (abno
== NULLFSBLOCK
)
4702 if ((ifp
->if_flags
& XFS_IFBROOT
) && !*cur
) {
4703 (*cur
) = xfs_bmbt_init_cursor(mp
, bma
->tp
, bma
->ip
, whichfork
);
4704 (*cur
)->bc_private
.b
.firstblock
= *firstblock
;
4705 (*cur
)->bc_private
.b
.flist
= flist
;
4708 * Bump the number of extents we've allocated
4714 (*cur
)->bc_private
.b
.flags
=
4715 bma
->wasdel
? XFS_BTCUR_BPRV_WASDEL
: 0;
4717 bma
->gotp
->br_startoff
= aoff
;
4718 bma
->gotp
->br_startblock
= abno
;
4719 bma
->gotp
->br_blockcount
= alen
;
4720 bma
->gotp
->br_state
= XFS_EXT_NORM
;
4723 * A wasdelay extent has been initialized, so shouldn't be flagged
4726 if (!bma
->wasdel
&& (flags
& XFS_BMAPI_PREALLOC
) &&
4727 xfs_sb_version_hasextflgbit(&mp
->m_sb
))
4728 bma
->gotp
->br_state
= XFS_EXT_UNWRITTEN
;
4730 error
= xfs_bmap_add_extent(bma
->tp
, bma
->ip
, lastx
, cur
, bma
->gotp
,
4731 firstblock
, flist
, logflags
, whichfork
);
4736 * Update our extent pointer, given that xfs_bmap_add_extent might
4737 * have merged it into one of the neighbouring ones.
4739 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
, *lastx
), bma
->gotp
);
4741 ASSERT(bma
->gotp
->br_startoff
<= aoff
);
4742 ASSERT(bma
->gotp
->br_startoff
+ bma
->gotp
->br_blockcount
>=
4744 ASSERT(bma
->gotp
->br_state
== XFS_EXT_NORM
||
4745 bma
->gotp
->br_state
== XFS_EXT_UNWRITTEN
);
4750 * Map file blocks to filesystem blocks.
4751 * File range is given by the bno/len pair.
4752 * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4753 * into a hole or past eof.
4754 * Only allocates blocks from a single allocation group,
4755 * to avoid locking problems.
4756 * The returned value in "firstblock" from the first call in a transaction
4757 * must be remembered and presented to subsequent calls in "firstblock".
4758 * An upper bound for the number of blocks to be allocated is supplied to
4759 * the first call in "total"; if no allocation group has that many free
4760 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4764 xfs_trans_t
*tp
, /* transaction pointer */
4765 xfs_inode_t
*ip
, /* incore inode */
4766 xfs_fileoff_t bno
, /* starting file offs. mapped */
4767 xfs_filblks_t len
, /* length to map in file */
4768 int flags
, /* XFS_BMAPI_... */
4769 xfs_fsblock_t
*firstblock
, /* first allocated block
4770 controls a.g. for allocs */
4771 xfs_extlen_t total
, /* total blocks needed */
4772 xfs_bmbt_irec_t
*mval
, /* output: map values */
4773 int *nmap
, /* i/o: mval size/count */
4774 xfs_bmap_free_t
*flist
) /* i/o: list extents to free */
4776 xfs_bmalloca_t bma
= { 0 }; /* args for xfs_bmap_alloc */
4777 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
4778 xfs_fileoff_t end
; /* end of mapped file region */
4779 int eof
; /* we've hit the end of extents */
4780 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
4781 int error
; /* error return */
4782 xfs_bmbt_irec_t got
; /* current file extent record */
4783 xfs_ifork_t
*ifp
; /* inode fork pointer */
4784 xfs_extnum_t lastx
; /* last useful extent number */
4785 int logflags
; /* flags for transaction logging */
4786 xfs_extlen_t minleft
; /* min blocks left after allocation */
4787 xfs_mount_t
*mp
; /* xfs mount structure */
4788 int n
; /* current extent index */
4789 int nallocs
; /* number of extents alloc'd */
4790 xfs_fileoff_t obno
; /* old block number (offset) */
4791 xfs_bmbt_irec_t prev
; /* previous file extent record */
4792 int tmp_logflags
; /* temp flags holder */
4793 int whichfork
; /* data or attr fork */
4794 char inhole
; /* current location is hole in file */
4795 char wasdelay
; /* old extent was delayed */
4796 char wr
; /* this is a write request */
4797 char rt
; /* this is a realtime file */
4799 xfs_fileoff_t orig_bno
; /* original block number value */
4800 int orig_flags
; /* original flags arg value */
4801 xfs_filblks_t orig_len
; /* original value of len arg */
4802 xfs_bmbt_irec_t
*orig_mval
; /* original value of mval */
4803 int orig_nmap
; /* original value of *nmap */
4812 ASSERT(*nmap
<= XFS_BMAP_MAX_NMAP
|| !(flags
& XFS_BMAPI_WRITE
));
4813 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
4814 XFS_ATTR_FORK
: XFS_DATA_FORK
;
4816 if (unlikely(XFS_TEST_ERROR(
4817 (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
4818 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
4819 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
),
4820 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
4821 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW
, mp
);
4822 return XFS_ERROR(EFSCORRUPTED
);
4824 if (XFS_FORCED_SHUTDOWN(mp
))
4825 return XFS_ERROR(EIO
);
4826 rt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
4827 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
4828 ASSERT(ifp
->if_ext_max
==
4829 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
4830 if ((wr
= (flags
& XFS_BMAPI_WRITE
)) != 0)
4831 XFS_STATS_INC(xs_blk_mapw
);
4833 XFS_STATS_INC(xs_blk_mapr
);
4835 * IGSTATE flag is used to combine extents which
4836 * differ only due to the state of the extents.
4837 * This technique is used from xfs_getbmap()
4838 * when the caller does not wish to see the
4839 * separation (which is the default).
4841 * This technique is also used when writing a
4842 * buffer which has been partially written,
4843 * (usually by being flushed during a chunkread),
4844 * to ensure one write takes place. This also
4845 * prevents a change in the xfs inode extents at
4846 * this time, intentionally. This change occurs
4847 * on completion of the write operation, in
4848 * xfs_strat_comp(), where the xfs_bmapi() call
4849 * is transactioned, and the extents combined.
4851 if ((flags
& XFS_BMAPI_IGSTATE
) && wr
) /* if writing unwritten space */
4852 wr
= 0; /* no allocations are allowed */
4856 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
4858 if ((error
= xfs_bmap_local_to_extents(tp
, ip
,
4859 firstblock
, total
, &logflags
, whichfork
)))
4862 if (wr
&& *firstblock
== NULLFSBLOCK
) {
4863 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
)
4864 minleft
= be16_to_cpu(ifp
->if_broot
->bb_level
) + 1;
4869 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
4870 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
4872 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
4885 while (bno
< end
&& n
< *nmap
) {
4887 * Reading past eof, act as though there's a hole
4891 got
.br_startoff
= end
;
4892 inhole
= eof
|| got
.br_startoff
> bno
;
4893 wasdelay
= wr
&& !inhole
&& isnullstartblock(got
.br_startblock
);
4895 * First, deal with the hole before the allocated space
4896 * that we found, if any.
4898 if (wr
&& (inhole
|| wasdelay
)) {
4900 bma
.conv
= !!(flags
& XFS_BMAPI_CONVERT
);
4901 bma
.wasdel
= wasdelay
;
4904 bma
.minleft
= minleft
;
4906 error
= xfs_bmapi_allocate(&bma
, &lastx
, &cur
,
4907 firstblock
, flist
, flags
, &nallocs
,
4909 logflags
|= tmp_logflags
;
4912 if (flist
&& flist
->xbf_low
)
4914 if (bma
.rval
== NULLFSBLOCK
)
4916 } else if (inhole
) {
4918 * Reading in a hole.
4920 mval
->br_startoff
= bno
;
4921 mval
->br_startblock
= HOLESTARTBLOCK
;
4922 mval
->br_blockcount
=
4923 XFS_FILBLKS_MIN(len
, got
.br_startoff
- bno
);
4924 mval
->br_state
= XFS_EXT_NORM
;
4925 bno
+= mval
->br_blockcount
;
4926 len
-= mval
->br_blockcount
;
4932 /* Deal with the allocated space we found. */
4933 xfs_bmapi_trim_map(mval
, &got
, &bno
, len
, obno
, end
, n
, flags
);
4936 * Check if writing previously allocated but
4937 * unwritten extents.
4940 ((mval
->br_state
== XFS_EXT_UNWRITTEN
&&
4941 ((flags
& XFS_BMAPI_PREALLOC
) == 0)) ||
4942 (mval
->br_state
== XFS_EXT_NORM
&&
4943 ((flags
& (XFS_BMAPI_PREALLOC
|XFS_BMAPI_CONVERT
)) ==
4944 (XFS_BMAPI_PREALLOC
|XFS_BMAPI_CONVERT
))))) {
4946 * Modify (by adding) the state flag, if writing.
4948 ASSERT(mval
->br_blockcount
<= len
);
4949 if ((ifp
->if_flags
& XFS_IFBROOT
) && !cur
) {
4950 cur
= xfs_bmbt_init_cursor(mp
,
4952 cur
->bc_private
.b
.firstblock
=
4954 cur
->bc_private
.b
.flist
= flist
;
4956 mval
->br_state
= (mval
->br_state
== XFS_EXT_UNWRITTEN
)
4958 : XFS_EXT_UNWRITTEN
;
4959 error
= xfs_bmap_add_extent(tp
, ip
, &lastx
, &cur
, mval
,
4960 firstblock
, flist
, &tmp_logflags
,
4962 logflags
|= tmp_logflags
;
4965 ep
= xfs_iext_get_ext(ifp
, lastx
);
4966 xfs_bmbt_get_all(ep
, &got
);
4968 * We may have combined previously unwritten
4969 * space with written space, so generate
4972 if (mval
->br_blockcount
< len
)
4976 /* update the extent map to return */
4977 xfs_bmapi_update_map(&mval
, &bno
, &len
, obno
, end
, &n
, flags
);
4980 * If we're done, stop now. Stop when we've allocated
4981 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4982 * the transaction may get too big.
4984 if (bno
>= end
|| n
>= *nmap
|| nallocs
>= *nmap
)
4987 * Else go on to the next record.
4990 if (++lastx
< ifp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
)) {
4991 ep
= xfs_iext_get_ext(ifp
, lastx
);
4992 xfs_bmbt_get_all(ep
, &got
);
4999 * Transform from btree to extents, give it cur.
5001 if (tp
&& XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
5002 XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
) {
5004 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
,
5005 &tmp_logflags
, whichfork
);
5006 logflags
|= tmp_logflags
;
5010 ASSERT(ifp
->if_ext_max
==
5011 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
5012 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
||
5013 XFS_IFORK_NEXTENTS(ip
, whichfork
) > ifp
->if_ext_max
);
5017 * Log everything. Do this after conversion, there's no point in
5018 * logging the extent records if we've converted to btree format.
5020 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5021 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5022 logflags
&= ~xfs_ilog_fext(whichfork
);
5023 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5024 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5025 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5027 * Log whatever the flags say, even if error. Otherwise we might miss
5028 * detecting a case where the data is changed, there's an error,
5029 * and it's not logged so we don't shutdown when we should.
5033 xfs_trans_log_inode(tp
, ip
, logflags
);
5037 ASSERT(*firstblock
== NULLFSBLOCK
||
5038 XFS_FSB_TO_AGNO(mp
, *firstblock
) ==
5040 cur
->bc_private
.b
.firstblock
) ||
5042 XFS_FSB_TO_AGNO(mp
, *firstblock
) <
5044 cur
->bc_private
.b
.firstblock
)));
5045 *firstblock
= cur
->bc_private
.b
.firstblock
;
5047 xfs_btree_del_cursor(cur
,
5048 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5051 xfs_bmap_validate_ret(orig_bno
, orig_len
, orig_flags
, orig_mval
,
5057 * Unmap (remove) blocks from a file.
5058 * If nexts is nonzero then the number of extents to remove is limited to
5059 * that value. If not all extents in the block range can be removed then
5064 xfs_trans_t
*tp
, /* transaction pointer */
5065 struct xfs_inode
*ip
, /* incore inode */
5066 xfs_fileoff_t bno
, /* starting offset to unmap */
5067 xfs_filblks_t len
, /* length to unmap in file */
5068 int flags
, /* misc flags */
5069 xfs_extnum_t nexts
, /* number of extents max */
5070 xfs_fsblock_t
*firstblock
, /* first allocated block
5071 controls a.g. for allocs */
5072 xfs_bmap_free_t
*flist
, /* i/o: list extents to free */
5073 int *done
) /* set if not done yet */
5075 xfs_btree_cur_t
*cur
; /* bmap btree cursor */
5076 xfs_bmbt_irec_t del
; /* extent being deleted */
5077 int eof
; /* is deleting at eof */
5078 xfs_bmbt_rec_host_t
*ep
; /* extent record pointer */
5079 int error
; /* error return value */
5080 xfs_extnum_t extno
; /* extent number in list */
5081 xfs_bmbt_irec_t got
; /* current extent record */
5082 xfs_ifork_t
*ifp
; /* inode fork pointer */
5083 int isrt
; /* freeing in rt area */
5084 xfs_extnum_t lastx
; /* last extent index used */
5085 int logflags
; /* transaction logging flags */
5086 xfs_extlen_t mod
; /* rt extent offset */
5087 xfs_mount_t
*mp
; /* mount structure */
5088 xfs_extnum_t nextents
; /* number of file extents */
5089 xfs_bmbt_irec_t prev
; /* previous extent record */
5090 xfs_fileoff_t start
; /* first file offset deleted */
5091 int tmp_logflags
; /* partial logging flags */
5092 int wasdel
; /* was a delayed alloc extent */
5093 int whichfork
; /* data or attribute fork */
5096 trace_xfs_bunmap(ip
, bno
, len
, flags
, _RET_IP_
);
5098 whichfork
= (flags
& XFS_BMAPI_ATTRFORK
) ?
5099 XFS_ATTR_FORK
: XFS_DATA_FORK
;
5100 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5102 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
5103 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)) {
5104 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW
,
5106 return XFS_ERROR(EFSCORRUPTED
);
5109 if (XFS_FORCED_SHUTDOWN(mp
))
5110 return XFS_ERROR(EIO
);
5114 ASSERT(ifp
->if_ext_max
==
5115 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
5116 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5117 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
5119 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
5120 if (nextents
== 0) {
5124 XFS_STATS_INC(xs_blk_unmap
);
5125 isrt
= (whichfork
== XFS_DATA_FORK
) && XFS_IS_REALTIME_INODE(ip
);
5127 bno
= start
+ len
- 1;
5128 ep
= xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
,
5132 * Check to see if the given block number is past the end of the
5133 * file, back up to the last block if so...
5136 ep
= xfs_iext_get_ext(ifp
, --lastx
);
5137 xfs_bmbt_get_all(ep
, &got
);
5138 bno
= got
.br_startoff
+ got
.br_blockcount
- 1;
5141 if (ifp
->if_flags
& XFS_IFBROOT
) {
5142 ASSERT(XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
);
5143 cur
= xfs_bmbt_init_cursor(mp
, tp
, ip
, whichfork
);
5144 cur
->bc_private
.b
.firstblock
= *firstblock
;
5145 cur
->bc_private
.b
.flist
= flist
;
5146 cur
->bc_private
.b
.flags
= 0;
5150 while (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
&& lastx
>= 0 &&
5151 (nexts
== 0 || extno
< nexts
)) {
5153 * Is the found extent after a hole in which bno lives?
5154 * Just back up to the previous extent, if so.
5156 if (got
.br_startoff
> bno
) {
5159 ep
= xfs_iext_get_ext(ifp
, lastx
);
5160 xfs_bmbt_get_all(ep
, &got
);
5163 * Is the last block of this extent before the range
5164 * we're supposed to delete? If so, we're done.
5166 bno
= XFS_FILEOFF_MIN(bno
,
5167 got
.br_startoff
+ got
.br_blockcount
- 1);
5171 * Then deal with the (possibly delayed) allocated space
5176 wasdel
= isnullstartblock(del
.br_startblock
);
5177 if (got
.br_startoff
< start
) {
5178 del
.br_startoff
= start
;
5179 del
.br_blockcount
-= start
- got
.br_startoff
;
5181 del
.br_startblock
+= start
- got
.br_startoff
;
5183 if (del
.br_startoff
+ del
.br_blockcount
> bno
+ 1)
5184 del
.br_blockcount
= bno
+ 1 - del
.br_startoff
;
5185 sum
= del
.br_startblock
+ del
.br_blockcount
;
5187 (mod
= do_mod(sum
, mp
->m_sb
.sb_rextsize
))) {
5189 * Realtime extent not lined up at the end.
5190 * The extent could have been split into written
5191 * and unwritten pieces, or we could just be
5192 * unmapping part of it. But we can't really
5193 * get rid of part of a realtime extent.
5195 if (del
.br_state
== XFS_EXT_UNWRITTEN
||
5196 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5198 * This piece is unwritten, or we're not
5199 * using unwritten extents. Skip over it.
5202 bno
-= mod
> del
.br_blockcount
?
5203 del
.br_blockcount
: mod
;
5204 if (bno
< got
.br_startoff
) {
5206 xfs_bmbt_get_all(xfs_iext_get_ext(
5212 * It's written, turn it unwritten.
5213 * This is better than zeroing it.
5215 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5216 ASSERT(xfs_trans_get_block_res(tp
) > 0);
5218 * If this spans a realtime extent boundary,
5219 * chop it back to the start of the one we end at.
5221 if (del
.br_blockcount
> mod
) {
5222 del
.br_startoff
+= del
.br_blockcount
- mod
;
5223 del
.br_startblock
+= del
.br_blockcount
- mod
;
5224 del
.br_blockcount
= mod
;
5226 del
.br_state
= XFS_EXT_UNWRITTEN
;
5227 error
= xfs_bmap_add_extent(tp
, ip
, &lastx
, &cur
, &del
,
5228 firstblock
, flist
, &logflags
,
5234 if (isrt
&& (mod
= do_mod(del
.br_startblock
, mp
->m_sb
.sb_rextsize
))) {
5236 * Realtime extent is lined up at the end but not
5237 * at the front. We'll get rid of full extents if
5240 mod
= mp
->m_sb
.sb_rextsize
- mod
;
5241 if (del
.br_blockcount
> mod
) {
5242 del
.br_blockcount
-= mod
;
5243 del
.br_startoff
+= mod
;
5244 del
.br_startblock
+= mod
;
5245 } else if ((del
.br_startoff
== start
&&
5246 (del
.br_state
== XFS_EXT_UNWRITTEN
||
5247 xfs_trans_get_block_res(tp
) == 0)) ||
5248 !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
5250 * Can't make it unwritten. There isn't
5251 * a full extent here so just skip it.
5253 ASSERT(bno
>= del
.br_blockcount
);
5254 bno
-= del
.br_blockcount
;
5255 if (got
.br_startoff
> bno
) {
5257 ep
= xfs_iext_get_ext(ifp
,
5259 xfs_bmbt_get_all(ep
, &got
);
5263 } else if (del
.br_state
== XFS_EXT_UNWRITTEN
) {
5265 * This one is already unwritten.
5266 * It must have a written left neighbor.
5267 * Unwrite the killed part of that one and
5271 xfs_bmbt_get_all(xfs_iext_get_ext(ifp
,
5273 ASSERT(prev
.br_state
== XFS_EXT_NORM
);
5274 ASSERT(!isnullstartblock(prev
.br_startblock
));
5275 ASSERT(del
.br_startblock
==
5276 prev
.br_startblock
+ prev
.br_blockcount
);
5277 if (prev
.br_startoff
< start
) {
5278 mod
= start
- prev
.br_startoff
;
5279 prev
.br_blockcount
-= mod
;
5280 prev
.br_startblock
+= mod
;
5281 prev
.br_startoff
= start
;
5283 prev
.br_state
= XFS_EXT_UNWRITTEN
;
5285 error
= xfs_bmap_add_extent(tp
, ip
, &lastx
,
5286 &cur
, &prev
, firstblock
, flist
,
5287 &logflags
, XFS_DATA_FORK
);
5292 ASSERT(del
.br_state
== XFS_EXT_NORM
);
5293 del
.br_state
= XFS_EXT_UNWRITTEN
;
5294 error
= xfs_bmap_add_extent(tp
, ip
, &lastx
,
5295 &cur
, &del
, firstblock
, flist
,
5296 &logflags
, XFS_DATA_FORK
);
5303 ASSERT(startblockval(del
.br_startblock
) > 0);
5304 /* Update realtime/data freespace, unreserve quota */
5306 xfs_filblks_t rtexts
;
5308 rtexts
= XFS_FSB_TO_B(mp
, del
.br_blockcount
);
5309 do_div(rtexts
, mp
->m_sb
.sb_rextsize
);
5310 xfs_mod_incore_sb(mp
, XFS_SBS_FREXTENTS
,
5311 (int64_t)rtexts
, 0);
5312 (void)xfs_trans_reserve_quota_nblks(NULL
,
5313 ip
, -((long)del
.br_blockcount
), 0,
5314 XFS_QMOPT_RES_RTBLKS
);
5316 xfs_icsb_modify_counters(mp
, XFS_SBS_FDBLOCKS
,
5317 (int64_t)del
.br_blockcount
, 0);
5318 (void)xfs_trans_reserve_quota_nblks(NULL
,
5319 ip
, -((long)del
.br_blockcount
), 0,
5320 XFS_QMOPT_RES_REGBLKS
);
5322 ip
->i_delayed_blks
-= del
.br_blockcount
;
5324 cur
->bc_private
.b
.flags
|=
5325 XFS_BTCUR_BPRV_WASDEL
;
5327 cur
->bc_private
.b
.flags
&= ~XFS_BTCUR_BPRV_WASDEL
;
5329 * If it's the case where the directory code is running
5330 * with no block reservation, and the deleted block is in
5331 * the middle of its extent, and the resulting insert
5332 * of an extent would cause transformation to btree format,
5333 * then reject it. The calling code will then swap
5334 * blocks around instead.
5335 * We have to do this now, rather than waiting for the
5336 * conversion to btree format, since the transaction
5339 if (!wasdel
&& xfs_trans_get_block_res(tp
) == 0 &&
5340 XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5341 XFS_IFORK_NEXTENTS(ip
, whichfork
) >= ifp
->if_ext_max
&&
5342 del
.br_startoff
> got
.br_startoff
&&
5343 del
.br_startoff
+ del
.br_blockcount
<
5344 got
.br_startoff
+ got
.br_blockcount
) {
5345 error
= XFS_ERROR(ENOSPC
);
5348 error
= xfs_bmap_del_extent(ip
, tp
, &lastx
, flist
, cur
, &del
,
5349 &tmp_logflags
, whichfork
);
5350 logflags
|= tmp_logflags
;
5353 bno
= del
.br_startoff
- 1;
5356 * If not done go on to the next (previous) record.
5358 if (bno
!= (xfs_fileoff_t
)-1 && bno
>= start
) {
5360 ep
= xfs_iext_get_ext(ifp
, lastx
);
5361 if (xfs_bmbt_get_startoff(ep
) > bno
) {
5363 ep
= xfs_iext_get_ext(ifp
,
5366 xfs_bmbt_get_all(ep
, &got
);
5371 *done
= bno
== (xfs_fileoff_t
)-1 || bno
< start
|| lastx
< 0;
5372 ASSERT(ifp
->if_ext_max
==
5373 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
5375 * Convert to a btree if necessary.
5377 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
&&
5378 XFS_IFORK_NEXTENTS(ip
, whichfork
) > ifp
->if_ext_max
) {
5379 ASSERT(cur
== NULL
);
5380 error
= xfs_bmap_extents_to_btree(tp
, ip
, firstblock
, flist
,
5381 &cur
, 0, &tmp_logflags
, whichfork
);
5382 logflags
|= tmp_logflags
;
5387 * transform from btree to extents, give it cur
5389 else if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_BTREE
&&
5390 XFS_IFORK_NEXTENTS(ip
, whichfork
) <= ifp
->if_ext_max
) {
5391 ASSERT(cur
!= NULL
);
5392 error
= xfs_bmap_btree_to_extents(tp
, ip
, cur
, &tmp_logflags
,
5394 logflags
|= tmp_logflags
;
5399 * transform from extents to local?
5401 ASSERT(ifp
->if_ext_max
==
5402 XFS_IFORK_SIZE(ip
, whichfork
) / (uint
)sizeof(xfs_bmbt_rec_t
));
5406 * Log everything. Do this after conversion, there's no point in
5407 * logging the extent records if we've converted to btree format.
5409 if ((logflags
& xfs_ilog_fext(whichfork
)) &&
5410 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
)
5411 logflags
&= ~xfs_ilog_fext(whichfork
);
5412 else if ((logflags
& xfs_ilog_fbroot(whichfork
)) &&
5413 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
)
5414 logflags
&= ~xfs_ilog_fbroot(whichfork
);
5416 * Log inode even in the error case, if the transaction
5417 * is dirty we'll need to shut down the filesystem.
5420 xfs_trans_log_inode(tp
, ip
, logflags
);
5423 *firstblock
= cur
->bc_private
.b
.firstblock
;
5424 cur
->bc_private
.b
.allocated
= 0;
5426 xfs_btree_del_cursor(cur
,
5427 error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
5433 * returns 1 for success, 0 if we failed to map the extent.
5436 xfs_getbmapx_fix_eof_hole(
5437 xfs_inode_t
*ip
, /* xfs incore inode pointer */
5438 struct getbmapx
*out
, /* output structure */
5439 int prealloced
, /* this is a file with
5440 * preallocated data space */
5441 __int64_t end
, /* last block requested */
5442 xfs_fsblock_t startblock
)
5445 xfs_mount_t
*mp
; /* file system mount point */
5446 xfs_ifork_t
*ifp
; /* inode fork pointer */
5447 xfs_extnum_t lastx
; /* last extent pointer */
5448 xfs_fileoff_t fileblock
;
5450 if (startblock
== HOLESTARTBLOCK
) {
5452 out
->bmv_block
= -1;
5453 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, ip
->i_size
));
5454 fixlen
-= out
->bmv_offset
;
5455 if (prealloced
&& out
->bmv_offset
+ out
->bmv_length
== end
) {
5456 /* Came to hole at EOF. Trim it. */
5459 out
->bmv_length
= fixlen
;
5462 if (startblock
== DELAYSTARTBLOCK
)
5463 out
->bmv_block
= -2;
5465 out
->bmv_block
= xfs_fsb_to_db(ip
, startblock
);
5466 fileblock
= XFS_BB_TO_FSB(ip
->i_mount
, out
->bmv_offset
);
5467 ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
5468 if (xfs_iext_bno_to_ext(ifp
, fileblock
, &lastx
) &&
5469 (lastx
== (ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
))-1))
5470 out
->bmv_oflags
|= BMV_OF_LAST
;
5477 * Get inode's extents as described in bmv, and format for output.
5478 * Calls formatter to fill the user's buffer until all extents
5479 * are mapped, until the passed-in bmv->bmv_count slots have
5480 * been filled, or until the formatter short-circuits the loop,
5481 * if it is tracking filled-in extents on its own.
5483 int /* error code */
5486 struct getbmapx
*bmv
, /* user bmap structure */
5487 xfs_bmap_format_t formatter
, /* format to user */
5488 void *arg
) /* formatter arg */
5490 __int64_t bmvend
; /* last block requested */
5491 int error
= 0; /* return value */
5492 __int64_t fixlen
; /* length for -1 case */
5493 int i
; /* extent number */
5494 int lock
; /* lock state */
5495 xfs_bmbt_irec_t
*map
; /* buffer for user's data */
5496 xfs_mount_t
*mp
; /* file system mount point */
5497 int nex
; /* # of user extents can do */
5498 int nexleft
; /* # of user extents left */
5499 int subnex
; /* # of bmapi's can do */
5500 int nmap
; /* number of map entries */
5501 struct getbmapx
*out
; /* output structure */
5502 int whichfork
; /* data or attr fork */
5503 int prealloced
; /* this is a file with
5504 * preallocated data space */
5505 int iflags
; /* interface flags */
5506 int bmapi_flags
; /* flags for xfs_bmapi */
5510 iflags
= bmv
->bmv_iflags
;
5511 whichfork
= iflags
& BMV_IF_ATTRFORK
? XFS_ATTR_FORK
: XFS_DATA_FORK
;
5513 if (whichfork
== XFS_ATTR_FORK
) {
5514 if (XFS_IFORK_Q(ip
)) {
5515 if (ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
&&
5516 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_BTREE
&&
5517 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
)
5518 return XFS_ERROR(EINVAL
);
5519 } else if (unlikely(
5520 ip
->i_d
.di_aformat
!= 0 &&
5521 ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_EXTENTS
)) {
5522 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW
,
5524 return XFS_ERROR(EFSCORRUPTED
);
5530 if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_EXTENTS
&&
5531 ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
&&
5532 ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
)
5533 return XFS_ERROR(EINVAL
);
5535 if (xfs_get_extsz_hint(ip
) ||
5536 ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
|XFS_DIFLAG_APPEND
)){
5538 fixlen
= XFS_MAXIOFFSET(mp
);
5541 fixlen
= ip
->i_size
;
5545 if (bmv
->bmv_length
== -1) {
5546 fixlen
= XFS_FSB_TO_BB(mp
, XFS_B_TO_FSB(mp
, fixlen
));
5548 max_t(__int64_t
, fixlen
- bmv
->bmv_offset
, 0);
5549 } else if (bmv
->bmv_length
== 0) {
5550 bmv
->bmv_entries
= 0;
5552 } else if (bmv
->bmv_length
< 0) {
5553 return XFS_ERROR(EINVAL
);
5556 nex
= bmv
->bmv_count
- 1;
5558 return XFS_ERROR(EINVAL
);
5559 bmvend
= bmv
->bmv_offset
+ bmv
->bmv_length
;
5562 if (bmv
->bmv_count
> ULONG_MAX
/ sizeof(struct getbmapx
))
5563 return XFS_ERROR(ENOMEM
);
5564 out
= kmem_zalloc(bmv
->bmv_count
* sizeof(struct getbmapx
), KM_MAYFAIL
);
5566 return XFS_ERROR(ENOMEM
);
5568 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
5569 if (whichfork
== XFS_DATA_FORK
&& !(iflags
& BMV_IF_DELALLOC
)) {
5570 if (ip
->i_delayed_blks
|| ip
->i_size
> ip
->i_d
.di_size
) {
5571 error
= xfs_flush_pages(ip
, 0, -1, 0, FI_REMAPF
);
5573 goto out_unlock_iolock
;
5576 * even after flushing the inode, there can still be delalloc
5577 * blocks on the inode beyond EOF due to speculative
5578 * preallocation. These are not removed until the release
5579 * function is called or the inode is inactivated. Hence we
5580 * cannot assert here that ip->i_delayed_blks == 0.
5584 lock
= xfs_ilock_map_shared(ip
);
5587 * Don't let nex be bigger than the number of extents
5588 * we can have assuming alternating holes and real extents.
5590 if (nex
> XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1)
5591 nex
= XFS_IFORK_NEXTENTS(ip
, whichfork
) * 2 + 1;
5593 bmapi_flags
= xfs_bmapi_aflag(whichfork
);
5594 if (!(iflags
& BMV_IF_PREALLOC
))
5595 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
5598 * Allocate enough space to handle "subnex" maps at a time.
5602 map
= kmem_alloc(subnex
* sizeof(*map
), KM_MAYFAIL
| KM_NOFS
);
5604 goto out_unlock_ilock
;
5606 bmv
->bmv_entries
= 0;
5608 if (XFS_IFORK_NEXTENTS(ip
, whichfork
) == 0 &&
5609 (whichfork
== XFS_ATTR_FORK
|| !(iflags
& BMV_IF_DELALLOC
))) {
5617 nmap
= (nexleft
> subnex
) ? subnex
: nexleft
;
5618 error
= xfs_bmapi_read(ip
, XFS_BB_TO_FSBT(mp
, bmv
->bmv_offset
),
5619 XFS_BB_TO_FSB(mp
, bmv
->bmv_length
),
5620 map
, &nmap
, bmapi_flags
);
5623 ASSERT(nmap
<= subnex
);
5625 for (i
= 0; i
< nmap
&& nexleft
&& bmv
->bmv_length
; i
++) {
5626 out
[cur_ext
].bmv_oflags
= 0;
5627 if (map
[i
].br_state
== XFS_EXT_UNWRITTEN
)
5628 out
[cur_ext
].bmv_oflags
|= BMV_OF_PREALLOC
;
5629 else if (map
[i
].br_startblock
== DELAYSTARTBLOCK
)
5630 out
[cur_ext
].bmv_oflags
|= BMV_OF_DELALLOC
;
5631 out
[cur_ext
].bmv_offset
=
5632 XFS_FSB_TO_BB(mp
, map
[i
].br_startoff
);
5633 out
[cur_ext
].bmv_length
=
5634 XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
5635 out
[cur_ext
].bmv_unused1
= 0;
5636 out
[cur_ext
].bmv_unused2
= 0;
5637 ASSERT(((iflags
& BMV_IF_DELALLOC
) != 0) ||
5638 (map
[i
].br_startblock
!= DELAYSTARTBLOCK
));
5639 if (map
[i
].br_startblock
== HOLESTARTBLOCK
&&
5640 whichfork
== XFS_ATTR_FORK
) {
5641 /* came to the end of attribute fork */
5642 out
[cur_ext
].bmv_oflags
|= BMV_OF_LAST
;
5646 if (!xfs_getbmapx_fix_eof_hole(ip
, &out
[cur_ext
],
5648 map
[i
].br_startblock
))
5652 out
[cur_ext
].bmv_offset
+
5653 out
[cur_ext
].bmv_length
;
5655 max_t(__int64_t
, 0, bmvend
- bmv
->bmv_offset
);
5658 * In case we don't want to return the hole,
5659 * don't increase cur_ext so that we can reuse
5660 * it in the next loop.
5662 if ((iflags
& BMV_IF_NO_HOLES
) &&
5663 map
[i
].br_startblock
== HOLESTARTBLOCK
) {
5664 memset(&out
[cur_ext
], 0, sizeof(out
[cur_ext
]));
5672 } while (nmap
&& nexleft
&& bmv
->bmv_length
);
5677 xfs_iunlock_map_shared(ip
, lock
);
5679 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
5681 for (i
= 0; i
< cur_ext
; i
++) {
5682 int full
= 0; /* user array is full */
5684 /* format results & advance arg */
5685 error
= formatter(&arg
, &out
[i
], &full
);
5695 * Check the last inode extent to determine whether this allocation will result
5696 * in blocks being allocated at the end of the file. When we allocate new data
5697 * blocks at the end of the file which do not start at the previous data block,
5698 * we will try to align the new blocks at stripe unit boundaries.
5700 STATIC
int /* error */
5702 xfs_inode_t
*ip
, /* incore inode pointer */
5703 xfs_fileoff_t off
, /* file offset in fsblocks */
5704 int whichfork
, /* data or attribute fork */
5705 char *aeof
) /* return value */
5707 int error
; /* error return value */
5708 xfs_ifork_t
*ifp
; /* inode fork pointer */
5709 xfs_bmbt_rec_host_t
*lastrec
; /* extent record pointer */
5710 xfs_extnum_t nextents
; /* number of file extents */
5711 xfs_bmbt_irec_t s
; /* expanded extent record */
5713 ASSERT(whichfork
== XFS_DATA_FORK
);
5714 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5715 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5716 (error
= xfs_iread_extents(NULL
, ip
, whichfork
)))
5718 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
5719 if (nextents
== 0) {
5724 * Go to the last extent
5726 lastrec
= xfs_iext_get_ext(ifp
, nextents
- 1);
5727 xfs_bmbt_get_all(lastrec
, &s
);
5729 * Check we are allocating in the last extent (for delayed allocations)
5730 * or past the last extent for non-delayed allocations.
5732 *aeof
= (off
>= s
.br_startoff
&&
5733 off
< s
.br_startoff
+ s
.br_blockcount
&&
5734 isnullstartblock(s
.br_startblock
)) ||
5735 off
>= s
.br_startoff
+ s
.br_blockcount
;
5740 * Check if the endoff is outside the last extent. If so the caller will grow
5741 * the allocation to a stripe unit boundary.
5745 xfs_inode_t
*ip
, /* incore inode pointer */
5746 xfs_fileoff_t endoff
, /* file offset in fsblocks */
5747 int whichfork
, /* data or attribute fork */
5748 int *eof
) /* result value */
5750 xfs_fsblock_t blockcount
; /* extent block count */
5751 int error
; /* error return value */
5752 xfs_ifork_t
*ifp
; /* inode fork pointer */
5753 xfs_bmbt_rec_host_t
*lastrec
; /* extent record pointer */
5754 xfs_extnum_t nextents
; /* number of file extents */
5755 xfs_fileoff_t startoff
; /* extent starting file offset */
5757 ASSERT(whichfork
== XFS_DATA_FORK
);
5758 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5759 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
5760 (error
= xfs_iread_extents(NULL
, ip
, whichfork
)))
5762 nextents
= ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
);
5763 if (nextents
== 0) {
5768 * Go to the last extent
5770 lastrec
= xfs_iext_get_ext(ifp
, nextents
- 1);
5771 startoff
= xfs_bmbt_get_startoff(lastrec
);
5772 blockcount
= xfs_bmbt_get_blockcount(lastrec
);
5773 *eof
= endoff
>= startoff
+ blockcount
;
5778 STATIC
struct xfs_buf
*
5780 struct xfs_btree_cur
*cur
,
5783 struct xfs_log_item_desc
*lidp
;
5789 for (i
= 0; i
< XFS_BTREE_MAXLEVELS
; i
++) {
5790 if (!cur
->bc_bufs
[i
])
5792 if (XFS_BUF_ADDR(cur
->bc_bufs
[i
]) == bno
)
5793 return cur
->bc_bufs
[i
];
5796 /* Chase down all the log items to see if the bp is there */
5797 list_for_each_entry(lidp
, &cur
->bc_tp
->t_items
, lid_trans
) {
5798 struct xfs_buf_log_item
*bip
;
5799 bip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
5800 if (bip
->bli_item
.li_type
== XFS_LI_BUF
&&
5801 XFS_BUF_ADDR(bip
->bli_buf
) == bno
)
5802 return bip
->bli_buf
;
5810 struct xfs_btree_block
*block
,
5816 __be64
*pp
, *thispa
; /* pointer to block address */
5817 xfs_bmbt_key_t
*prevp
, *keyp
;
5819 ASSERT(be16_to_cpu(block
->bb_level
) > 0);
5822 for( i
= 1; i
<= xfs_btree_get_numrecs(block
); i
++) {
5823 dmxr
= mp
->m_bmap_dmxr
[0];
5824 keyp
= XFS_BMBT_KEY_ADDR(mp
, block
, i
);
5827 ASSERT(be64_to_cpu(prevp
->br_startoff
) <
5828 be64_to_cpu(keyp
->br_startoff
));
5833 * Compare the block numbers to see if there are dups.
5836 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, i
, sz
);
5838 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, i
, dmxr
);
5840 for (j
= i
+1; j
<= be16_to_cpu(block
->bb_numrecs
); j
++) {
5842 thispa
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, j
, sz
);
5844 thispa
= XFS_BMBT_PTR_ADDR(mp
, block
, j
, dmxr
);
5845 if (*thispa
== *pp
) {
5846 xfs_warn(mp
, "%s: thispa(%d) == pp(%d) %Ld",
5848 (unsigned long long)be64_to_cpu(*thispa
));
5849 panic("%s: ptrs are equal in node\n",
5857 * Check that the extents for the inode ip are in the right order in all
5862 xfs_bmap_check_leaf_extents(
5863 xfs_btree_cur_t
*cur
, /* btree cursor or null */
5864 xfs_inode_t
*ip
, /* incore inode pointer */
5865 int whichfork
) /* data or attr fork */
5867 struct xfs_btree_block
*block
; /* current btree block */
5868 xfs_fsblock_t bno
; /* block # of "block" */
5869 xfs_buf_t
*bp
; /* buffer for "block" */
5870 int error
; /* error return value */
5871 xfs_extnum_t i
=0, j
; /* index into the extents list */
5872 xfs_ifork_t
*ifp
; /* fork structure */
5873 int level
; /* btree level, for checking */
5874 xfs_mount_t
*mp
; /* file system mount structure */
5875 __be64
*pp
; /* pointer to block address */
5876 xfs_bmbt_rec_t
*ep
; /* pointer to current extent */
5877 xfs_bmbt_rec_t last
= {0, 0}; /* last extent in prev block */
5878 xfs_bmbt_rec_t
*nextp
; /* pointer to next extent */
5881 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
) {
5887 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
5888 block
= ifp
->if_broot
;
5890 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5892 level
= be16_to_cpu(block
->bb_level
);
5894 xfs_check_block(block
, mp
, 1, ifp
->if_broot_bytes
);
5895 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
5896 bno
= be64_to_cpu(*pp
);
5898 ASSERT(bno
!= NULLDFSBNO
);
5899 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
5900 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
5903 * Go down the tree until leaf level is reached, following the first
5904 * pointer (leftmost) at each level.
5906 while (level
-- > 0) {
5907 /* See if buf is in cur first */
5908 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
5914 if (!bp
&& (error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
5915 XFS_BMAP_BTREE_REF
)))
5917 block
= XFS_BUF_TO_BLOCK(bp
);
5918 XFS_WANT_CORRUPTED_GOTO(
5919 xfs_bmap_sanity_check(mp
, bp
, level
),
5925 * Check this block for basic sanity (increasing keys and
5926 * no duplicate blocks).
5929 xfs_check_block(block
, mp
, 0, 0);
5930 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
5931 bno
= be64_to_cpu(*pp
);
5932 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp
, bno
), error0
);
5935 xfs_trans_brelse(NULL
, bp
);
5940 * Here with bp and block set to the leftmost leaf node in the tree.
5945 * Loop over all leaf nodes checking that all extents are in the right order.
5948 xfs_fsblock_t nextbno
;
5949 xfs_extnum_t num_recs
;
5952 num_recs
= xfs_btree_get_numrecs(block
);
5955 * Read-ahead the next leaf block, if any.
5958 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
5961 * Check all the extents to make sure they are OK.
5962 * If we had a previous block, the last entry should
5963 * conform with the first entry in this one.
5966 ep
= XFS_BMBT_REC_ADDR(mp
, block
, 1);
5968 ASSERT(xfs_bmbt_disk_get_startoff(&last
) +
5969 xfs_bmbt_disk_get_blockcount(&last
) <=
5970 xfs_bmbt_disk_get_startoff(ep
));
5972 for (j
= 1; j
< num_recs
; j
++) {
5973 nextp
= XFS_BMBT_REC_ADDR(mp
, block
, j
+ 1);
5974 ASSERT(xfs_bmbt_disk_get_startoff(ep
) +
5975 xfs_bmbt_disk_get_blockcount(ep
) <=
5976 xfs_bmbt_disk_get_startoff(nextp
));
5984 xfs_trans_brelse(NULL
, bp
);
5988 * If we've reached the end, stop.
5990 if (bno
== NULLFSBLOCK
)
5993 bp
= xfs_bmap_get_bp(cur
, XFS_FSB_TO_DADDR(mp
, bno
));
5999 if (!bp
&& (error
= xfs_btree_read_bufl(mp
, NULL
, bno
, 0, &bp
,
6000 XFS_BMAP_BTREE_REF
)))
6002 block
= XFS_BUF_TO_BLOCK(bp
);
6006 xfs_trans_brelse(NULL
, bp
);
6011 xfs_warn(mp
, "%s: at error0", __func__
);
6013 xfs_trans_brelse(NULL
, bp
);
6015 xfs_warn(mp
, "%s: BAD after btree leaves for %d extents",
6017 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__
);
6023 * Count fsblocks of the given fork.
6026 xfs_bmap_count_blocks(
6027 xfs_trans_t
*tp
, /* transaction pointer */
6028 xfs_inode_t
*ip
, /* incore inode */
6029 int whichfork
, /* data or attr fork */
6030 int *count
) /* out: count of blocks */
6032 struct xfs_btree_block
*block
; /* current btree block */
6033 xfs_fsblock_t bno
; /* block # of "block" */
6034 xfs_ifork_t
*ifp
; /* fork structure */
6035 int level
; /* btree level, for checking */
6036 xfs_mount_t
*mp
; /* file system mount structure */
6037 __be64
*pp
; /* pointer to block address */
6041 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
6042 if ( XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_EXTENTS
) {
6043 xfs_bmap_count_leaves(ifp
, 0,
6044 ifp
->if_bytes
/ (uint
)sizeof(xfs_bmbt_rec_t
),
6050 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6052 block
= ifp
->if_broot
;
6053 level
= be16_to_cpu(block
->bb_level
);
6055 pp
= XFS_BMAP_BROOT_PTR_ADDR(mp
, block
, 1, ifp
->if_broot_bytes
);
6056 bno
= be64_to_cpu(*pp
);
6057 ASSERT(bno
!= NULLDFSBNO
);
6058 ASSERT(XFS_FSB_TO_AGNO(mp
, bno
) < mp
->m_sb
.sb_agcount
);
6059 ASSERT(XFS_FSB_TO_AGBNO(mp
, bno
) < mp
->m_sb
.sb_agblocks
);
6061 if (unlikely(xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
) < 0)) {
6062 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW
,
6064 return XFS_ERROR(EFSCORRUPTED
);
6071 * Recursively walks each level of a btree
6072 * to count total fsblocks is use.
6074 STATIC
int /* error */
6075 xfs_bmap_count_tree(
6076 xfs_mount_t
*mp
, /* file system mount point */
6077 xfs_trans_t
*tp
, /* transaction pointer */
6078 xfs_ifork_t
*ifp
, /* inode fork pointer */
6079 xfs_fsblock_t blockno
, /* file system block number */
6080 int levelin
, /* level in btree */
6081 int *count
) /* Count of blocks */
6084 xfs_buf_t
*bp
, *nbp
;
6085 int level
= levelin
;
6087 xfs_fsblock_t bno
= blockno
;
6088 xfs_fsblock_t nextbno
;
6089 struct xfs_btree_block
*block
, *nextblock
;
6092 if ((error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
, XFS_BMAP_BTREE_REF
)))
6095 block
= XFS_BUF_TO_BLOCK(bp
);
6098 /* Not at node above leaves, count this level of nodes */
6099 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
6100 while (nextbno
!= NULLFSBLOCK
) {
6101 if ((error
= xfs_btree_read_bufl(mp
, tp
, nextbno
,
6102 0, &nbp
, XFS_BMAP_BTREE_REF
)))
6105 nextblock
= XFS_BUF_TO_BLOCK(nbp
);
6106 nextbno
= be64_to_cpu(nextblock
->bb_u
.l
.bb_rightsib
);
6107 xfs_trans_brelse(tp
, nbp
);
6110 /* Dive to the next level */
6111 pp
= XFS_BMBT_PTR_ADDR(mp
, block
, 1, mp
->m_bmap_dmxr
[1]);
6112 bno
= be64_to_cpu(*pp
);
6113 if (unlikely((error
=
6114 xfs_bmap_count_tree(mp
, tp
, ifp
, bno
, level
, count
)) < 0)) {
6115 xfs_trans_brelse(tp
, bp
);
6116 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6117 XFS_ERRLEVEL_LOW
, mp
);
6118 return XFS_ERROR(EFSCORRUPTED
);
6120 xfs_trans_brelse(tp
, bp
);
6122 /* count all level 1 nodes and their leaves */
6124 nextbno
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
6125 numrecs
= be16_to_cpu(block
->bb_numrecs
);
6126 xfs_bmap_disk_count_leaves(mp
, block
, numrecs
, count
);
6127 xfs_trans_brelse(tp
, bp
);
6128 if (nextbno
== NULLFSBLOCK
)
6131 if ((error
= xfs_btree_read_bufl(mp
, tp
, bno
, 0, &bp
,
6132 XFS_BMAP_BTREE_REF
)))
6135 block
= XFS_BUF_TO_BLOCK(bp
);
6142 * Count leaf blocks given a range of extent records.
6145 xfs_bmap_count_leaves(
6153 for (b
= 0; b
< numrecs
; b
++) {
6154 xfs_bmbt_rec_host_t
*frp
= xfs_iext_get_ext(ifp
, idx
+ b
);
6155 *count
+= xfs_bmbt_get_blockcount(frp
);
6160 * Count leaf blocks given a range of extent records originally
6164 xfs_bmap_disk_count_leaves(
6165 struct xfs_mount
*mp
,
6166 struct xfs_btree_block
*block
,
6171 xfs_bmbt_rec_t
*frp
;
6173 for (b
= 1; b
<= numrecs
; b
++) {
6174 frp
= XFS_BMBT_REC_ADDR(mp
, block
, b
);
6175 *count
+= xfs_bmbt_disk_get_blockcount(frp
);
6180 * dead simple method of punching delalyed allocation blocks from a range in
6181 * the inode. Walks a block at a time so will be slow, but is only executed in
6182 * rare error cases so the overhead is not critical. This will alays punch out
6183 * both the start and end blocks, even if the ranges only partially overlap
6184 * them, so it is up to the caller to ensure that partial blocks are not
6188 xfs_bmap_punch_delalloc_range(
6189 struct xfs_inode
*ip
,
6190 xfs_fileoff_t start_fsb
,
6191 xfs_fileoff_t length
)
6193 xfs_fileoff_t remaining
= length
;
6196 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
6200 xfs_bmbt_irec_t imap
;
6202 xfs_fsblock_t firstblock
;
6203 xfs_bmap_free_t flist
;
6206 * Map the range first and check that it is a delalloc extent
6207 * before trying to unmap the range. Otherwise we will be
6208 * trying to remove a real extent (which requires a
6209 * transaction) or a hole, which is probably a bad idea...
6211 error
= xfs_bmapi_read(ip
, start_fsb
, 1, &imap
, &nimaps
,
6215 /* something screwed, just bail */
6216 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
6217 xfs_alert(ip
->i_mount
,
6218 "Failed delalloc mapping lookup ino %lld fsb %lld.",
6219 ip
->i_ino
, start_fsb
);
6227 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
6228 /* been converted, ignore */
6231 WARN_ON(imap
.br_blockcount
== 0);
6234 * Note: while we initialise the firstblock/flist pair, they
6235 * should never be used because blocks should never be
6236 * allocated or freed for a delalloc extent and hence we need
6237 * don't cancel or finish them after the xfs_bunmapi() call.
6239 xfs_bmap_init(&flist
, &firstblock
);
6240 error
= xfs_bunmapi(NULL
, ip
, start_fsb
, 1, 0, 1, &firstblock
,
6245 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
6249 } while(remaining
> 0);