2 * Copyright (c) 2000-2002,2005 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"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc_btree.h"
29 #include "xfs_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_alloc.h"
34 #include "xfs_extent_busy.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
38 struct workqueue_struct
*xfs_alloc_wq
;
40 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
42 #define XFSA_FIXUP_BNO_OK 1
43 #define XFSA_FIXUP_CNT_OK 2
45 STATIC
int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t
*);
46 STATIC
int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t
*);
47 STATIC
int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t
*);
48 STATIC
int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t
*,
49 xfs_btree_cur_t
*, xfs_agblock_t
*, xfs_extlen_t
*, int *);
52 * Lookup the record equal to [bno, len] in the btree given by cur.
54 STATIC
int /* error */
56 struct xfs_btree_cur
*cur
, /* btree cursor */
57 xfs_agblock_t bno
, /* starting block of extent */
58 xfs_extlen_t len
, /* length of extent */
59 int *stat
) /* success/failure */
61 cur
->bc_rec
.a
.ar_startblock
= bno
;
62 cur
->bc_rec
.a
.ar_blockcount
= len
;
63 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
67 * Lookup the first record greater than or equal to [bno, len]
68 * in the btree given by cur.
72 struct xfs_btree_cur
*cur
, /* btree cursor */
73 xfs_agblock_t bno
, /* starting block of extent */
74 xfs_extlen_t len
, /* length of extent */
75 int *stat
) /* success/failure */
77 cur
->bc_rec
.a
.ar_startblock
= bno
;
78 cur
->bc_rec
.a
.ar_blockcount
= len
;
79 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
83 * Lookup the first record less than or equal to [bno, len]
84 * in the btree given by cur.
88 struct xfs_btree_cur
*cur
, /* btree cursor */
89 xfs_agblock_t bno
, /* starting block of extent */
90 xfs_extlen_t len
, /* length of extent */
91 int *stat
) /* success/failure */
93 cur
->bc_rec
.a
.ar_startblock
= bno
;
94 cur
->bc_rec
.a
.ar_blockcount
= len
;
95 return xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, stat
);
99 * Update the record referred to by cur to the value given
101 * This either works (return 0) or gets an EFSCORRUPTED error.
103 STATIC
int /* error */
105 struct xfs_btree_cur
*cur
, /* btree cursor */
106 xfs_agblock_t bno
, /* starting block of extent */
107 xfs_extlen_t len
) /* length of extent */
109 union xfs_btree_rec rec
;
111 rec
.alloc
.ar_startblock
= cpu_to_be32(bno
);
112 rec
.alloc
.ar_blockcount
= cpu_to_be32(len
);
113 return xfs_btree_update(cur
, &rec
);
117 * Get the data from the pointed-to record.
121 struct xfs_btree_cur
*cur
, /* btree cursor */
122 xfs_agblock_t
*bno
, /* output: starting block of extent */
123 xfs_extlen_t
*len
, /* output: length of extent */
124 int *stat
) /* output: success/failure */
126 union xfs_btree_rec
*rec
;
129 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
130 if (!error
&& *stat
== 1) {
131 *bno
= be32_to_cpu(rec
->alloc
.ar_startblock
);
132 *len
= be32_to_cpu(rec
->alloc
.ar_blockcount
);
138 * Compute aligned version of the found extent.
139 * Takes alignment and min length into account.
142 xfs_alloc_compute_aligned(
143 xfs_alloc_arg_t
*args
, /* allocation argument structure */
144 xfs_agblock_t foundbno
, /* starting block in found extent */
145 xfs_extlen_t foundlen
, /* length in found extent */
146 xfs_agblock_t
*resbno
, /* result block number */
147 xfs_extlen_t
*reslen
) /* result length */
152 /* Trim busy sections out of found extent */
153 xfs_extent_busy_trim(args
, foundbno
, foundlen
, &bno
, &len
);
155 if (args
->alignment
> 1 && len
>= args
->minlen
) {
156 xfs_agblock_t aligned_bno
= roundup(bno
, args
->alignment
);
157 xfs_extlen_t diff
= aligned_bno
- bno
;
159 *resbno
= aligned_bno
;
160 *reslen
= diff
>= len
? 0 : len
- diff
;
168 * Compute best start block and diff for "near" allocations.
169 * freelen >= wantlen already checked by caller.
171 STATIC xfs_extlen_t
/* difference value (absolute) */
172 xfs_alloc_compute_diff(
173 xfs_agblock_t wantbno
, /* target starting block */
174 xfs_extlen_t wantlen
, /* target length */
175 xfs_extlen_t alignment
, /* target alignment */
176 xfs_agblock_t freebno
, /* freespace's starting block */
177 xfs_extlen_t freelen
, /* freespace's length */
178 xfs_agblock_t
*newbnop
) /* result: best start block from free */
180 xfs_agblock_t freeend
; /* end of freespace extent */
181 xfs_agblock_t newbno1
; /* return block number */
182 xfs_agblock_t newbno2
; /* other new block number */
183 xfs_extlen_t newlen1
=0; /* length with newbno1 */
184 xfs_extlen_t newlen2
=0; /* length with newbno2 */
185 xfs_agblock_t wantend
; /* end of target extent */
187 ASSERT(freelen
>= wantlen
);
188 freeend
= freebno
+ freelen
;
189 wantend
= wantbno
+ wantlen
;
190 if (freebno
>= wantbno
) {
191 if ((newbno1
= roundup(freebno
, alignment
)) >= freeend
)
192 newbno1
= NULLAGBLOCK
;
193 } else if (freeend
>= wantend
&& alignment
> 1) {
194 newbno1
= roundup(wantbno
, alignment
);
195 newbno2
= newbno1
- alignment
;
196 if (newbno1
>= freeend
)
197 newbno1
= NULLAGBLOCK
;
199 newlen1
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno1
);
200 if (newbno2
< freebno
)
201 newbno2
= NULLAGBLOCK
;
203 newlen2
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno2
);
204 if (newbno1
!= NULLAGBLOCK
&& newbno2
!= NULLAGBLOCK
) {
205 if (newlen1
< newlen2
||
206 (newlen1
== newlen2
&&
207 XFS_ABSDIFF(newbno1
, wantbno
) >
208 XFS_ABSDIFF(newbno2
, wantbno
)))
210 } else if (newbno2
!= NULLAGBLOCK
)
212 } else if (freeend
>= wantend
) {
214 } else if (alignment
> 1) {
215 newbno1
= roundup(freeend
- wantlen
, alignment
);
216 if (newbno1
> freeend
- wantlen
&&
217 newbno1
- alignment
>= freebno
)
218 newbno1
-= alignment
;
219 else if (newbno1
>= freeend
)
220 newbno1
= NULLAGBLOCK
;
222 newbno1
= freeend
- wantlen
;
224 return newbno1
== NULLAGBLOCK
? 0 : XFS_ABSDIFF(newbno1
, wantbno
);
228 * Fix up the length, based on mod and prod.
229 * len should be k * prod + mod for some k.
230 * If len is too small it is returned unchanged.
231 * If len hits maxlen it is left alone.
235 xfs_alloc_arg_t
*args
) /* allocation argument structure */
240 ASSERT(args
->mod
< args
->prod
);
242 ASSERT(rlen
>= args
->minlen
);
243 ASSERT(rlen
<= args
->maxlen
);
244 if (args
->prod
<= 1 || rlen
< args
->mod
|| rlen
== args
->maxlen
||
245 (args
->mod
== 0 && rlen
< args
->prod
))
247 k
= rlen
% args
->prod
;
251 if ((int)(rlen
= rlen
- k
- args
->mod
) < (int)args
->minlen
)
254 if ((int)(rlen
= rlen
- args
->prod
- (args
->mod
- k
)) <
258 ASSERT(rlen
>= args
->minlen
);
259 ASSERT(rlen
<= args
->maxlen
);
264 * Fix up length if there is too little space left in the a.g.
265 * Return 1 if ok, 0 if too little, should give up.
268 xfs_alloc_fix_minleft(
269 xfs_alloc_arg_t
*args
) /* allocation argument structure */
271 xfs_agf_t
*agf
; /* a.g. freelist header */
272 int diff
; /* free space difference */
274 if (args
->minleft
== 0)
276 agf
= XFS_BUF_TO_AGF(args
->agbp
);
277 diff
= be32_to_cpu(agf
->agf_freeblks
)
278 - args
->len
- args
->minleft
;
281 args
->len
+= diff
; /* shrink the allocated space */
282 if (args
->len
>= args
->minlen
)
284 args
->agbno
= NULLAGBLOCK
;
289 * Update the two btrees, logically removing from freespace the extent
290 * starting at rbno, rlen blocks. The extent is contained within the
291 * actual (current) free extent fbno for flen blocks.
292 * Flags are passed in indicating whether the cursors are set to the
295 STATIC
int /* error code */
296 xfs_alloc_fixup_trees(
297 xfs_btree_cur_t
*cnt_cur
, /* cursor for by-size btree */
298 xfs_btree_cur_t
*bno_cur
, /* cursor for by-block btree */
299 xfs_agblock_t fbno
, /* starting block of free extent */
300 xfs_extlen_t flen
, /* length of free extent */
301 xfs_agblock_t rbno
, /* starting block of returned extent */
302 xfs_extlen_t rlen
, /* length of returned extent */
303 int flags
) /* flags, XFSA_FIXUP_... */
305 int error
; /* error code */
306 int i
; /* operation results */
307 xfs_agblock_t nfbno1
; /* first new free startblock */
308 xfs_agblock_t nfbno2
; /* second new free startblock */
309 xfs_extlen_t nflen1
=0; /* first new free length */
310 xfs_extlen_t nflen2
=0; /* second new free length */
313 * Look up the record in the by-size tree if necessary.
315 if (flags
& XFSA_FIXUP_CNT_OK
) {
317 if ((error
= xfs_alloc_get_rec(cnt_cur
, &nfbno1
, &nflen1
, &i
)))
319 XFS_WANT_CORRUPTED_RETURN(
320 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
323 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, fbno
, flen
, &i
)))
325 XFS_WANT_CORRUPTED_RETURN(i
== 1);
328 * Look up the record in the by-block tree if necessary.
330 if (flags
& XFSA_FIXUP_BNO_OK
) {
332 if ((error
= xfs_alloc_get_rec(bno_cur
, &nfbno1
, &nflen1
, &i
)))
334 XFS_WANT_CORRUPTED_RETURN(
335 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
338 if ((error
= xfs_alloc_lookup_eq(bno_cur
, fbno
, flen
, &i
)))
340 XFS_WANT_CORRUPTED_RETURN(i
== 1);
344 if (bno_cur
->bc_nlevels
== 1 && cnt_cur
->bc_nlevels
== 1) {
345 struct xfs_btree_block
*bnoblock
;
346 struct xfs_btree_block
*cntblock
;
348 bnoblock
= XFS_BUF_TO_BLOCK(bno_cur
->bc_bufs
[0]);
349 cntblock
= XFS_BUF_TO_BLOCK(cnt_cur
->bc_bufs
[0]);
351 XFS_WANT_CORRUPTED_RETURN(
352 bnoblock
->bb_numrecs
== cntblock
->bb_numrecs
);
357 * Deal with all four cases: the allocated record is contained
358 * within the freespace record, so we can have new freespace
359 * at either (or both) end, or no freespace remaining.
361 if (rbno
== fbno
&& rlen
== flen
)
362 nfbno1
= nfbno2
= NULLAGBLOCK
;
363 else if (rbno
== fbno
) {
364 nfbno1
= rbno
+ rlen
;
365 nflen1
= flen
- rlen
;
366 nfbno2
= NULLAGBLOCK
;
367 } else if (rbno
+ rlen
== fbno
+ flen
) {
369 nflen1
= flen
- rlen
;
370 nfbno2
= NULLAGBLOCK
;
373 nflen1
= rbno
- fbno
;
374 nfbno2
= rbno
+ rlen
;
375 nflen2
= (fbno
+ flen
) - nfbno2
;
378 * Delete the entry from the by-size btree.
380 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
382 XFS_WANT_CORRUPTED_RETURN(i
== 1);
384 * Add new by-size btree entry(s).
386 if (nfbno1
!= NULLAGBLOCK
) {
387 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno1
, nflen1
, &i
)))
389 XFS_WANT_CORRUPTED_RETURN(i
== 0);
390 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
392 XFS_WANT_CORRUPTED_RETURN(i
== 1);
394 if (nfbno2
!= NULLAGBLOCK
) {
395 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno2
, nflen2
, &i
)))
397 XFS_WANT_CORRUPTED_RETURN(i
== 0);
398 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
400 XFS_WANT_CORRUPTED_RETURN(i
== 1);
403 * Fix up the by-block btree entry(s).
405 if (nfbno1
== NULLAGBLOCK
) {
407 * No remaining freespace, just delete the by-block tree entry.
409 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
411 XFS_WANT_CORRUPTED_RETURN(i
== 1);
414 * Update the by-block entry to start later|be shorter.
416 if ((error
= xfs_alloc_update(bno_cur
, nfbno1
, nflen1
)))
419 if (nfbno2
!= NULLAGBLOCK
) {
421 * 2 resulting free entries, need to add one.
423 if ((error
= xfs_alloc_lookup_eq(bno_cur
, nfbno2
, nflen2
, &i
)))
425 XFS_WANT_CORRUPTED_RETURN(i
== 0);
426 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
428 XFS_WANT_CORRUPTED_RETURN(i
== 1);
437 #ifdef WHEN_CRCS_COME_ALONG
439 * we cannot actually do any verification of the AGFL because mkfs does
440 * not initialise the AGFL to zero or NULL. Hence the only valid part of
441 * the AGFL is what the AGF says is active. We can't get to the AGF, so
442 * we can't verify just those entries are valid.
444 * This problem goes away when the CRC format change comes along as that
445 * requires the AGFL to be initialised by mkfs. At that point, we can
446 * verify the blocks in the agfl -active or not- lie within the bounds
447 * of the AG. Until then, just leave this check ifdef'd out.
449 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
450 struct xfs_agfl
*agfl
= XFS_BUF_TO_AGFL(bp
);
455 for (i
= 0; i
< XFS_AGFL_SIZE(mp
); i
++) {
456 if (be32_to_cpu(agfl
->agfl_bno
[i
]) == NULLAGBLOCK
||
457 be32_to_cpu(agfl
->agfl_bno
[i
]) >= mp
->m_sb
.sb_agblocks
)
462 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, agfl
);
463 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
469 xfs_agfl_write_verify(
476 xfs_agfl_read_verify(
482 const struct xfs_buf_ops xfs_agfl_buf_ops
= {
483 .verify_read
= xfs_agfl_read_verify
,
484 .verify_write
= xfs_agfl_write_verify
,
488 * Read in the allocation group free block array.
490 STATIC
int /* error */
492 xfs_mount_t
*mp
, /* mount point structure */
493 xfs_trans_t
*tp
, /* transaction pointer */
494 xfs_agnumber_t agno
, /* allocation group number */
495 xfs_buf_t
**bpp
) /* buffer for the ag free block array */
497 xfs_buf_t
*bp
; /* return value */
500 ASSERT(agno
!= NULLAGNUMBER
);
501 error
= xfs_trans_read_buf(
502 mp
, tp
, mp
->m_ddev_targp
,
503 XFS_AG_DADDR(mp
, agno
, XFS_AGFL_DADDR(mp
)),
504 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, &xfs_agfl_buf_ops
);
507 ASSERT(!xfs_buf_geterror(bp
));
508 xfs_buf_set_ref(bp
, XFS_AGFL_REF
);
514 xfs_alloc_update_counters(
515 struct xfs_trans
*tp
,
516 struct xfs_perag
*pag
,
517 struct xfs_buf
*agbp
,
520 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
522 pag
->pagf_freeblks
+= len
;
523 be32_add_cpu(&agf
->agf_freeblks
, len
);
525 xfs_trans_agblocks_delta(tp
, len
);
526 if (unlikely(be32_to_cpu(agf
->agf_freeblks
) >
527 be32_to_cpu(agf
->agf_length
)))
530 xfs_alloc_log_agf(tp
, agbp
, XFS_AGF_FREEBLKS
);
535 * Allocation group level functions.
539 * Allocate a variable extent in the allocation group agno.
540 * Type and bno are used to determine where in the allocation group the
542 * Extent's length (returned in *len) will be between minlen and maxlen,
543 * and of the form k * prod + mod unless there's nothing that large.
544 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
546 STATIC
int /* error */
547 xfs_alloc_ag_vextent(
548 xfs_alloc_arg_t
*args
) /* argument structure for allocation */
552 ASSERT(args
->minlen
> 0);
553 ASSERT(args
->maxlen
> 0);
554 ASSERT(args
->minlen
<= args
->maxlen
);
555 ASSERT(args
->mod
< args
->prod
);
556 ASSERT(args
->alignment
> 0);
558 * Branch to correct routine based on the type.
561 switch (args
->type
) {
562 case XFS_ALLOCTYPE_THIS_AG
:
563 error
= xfs_alloc_ag_vextent_size(args
);
565 case XFS_ALLOCTYPE_NEAR_BNO
:
566 error
= xfs_alloc_ag_vextent_near(args
);
568 case XFS_ALLOCTYPE_THIS_BNO
:
569 error
= xfs_alloc_ag_vextent_exact(args
);
576 if (error
|| args
->agbno
== NULLAGBLOCK
)
579 ASSERT(args
->len
>= args
->minlen
);
580 ASSERT(args
->len
<= args
->maxlen
);
581 ASSERT(!args
->wasfromfl
|| !args
->isfl
);
582 ASSERT(args
->agbno
% args
->alignment
== 0);
584 if (!args
->wasfromfl
) {
585 error
= xfs_alloc_update_counters(args
->tp
, args
->pag
,
587 -((long)(args
->len
)));
591 ASSERT(!xfs_extent_busy_search(args
->mp
, args
->agno
,
592 args
->agbno
, args
->len
));
596 xfs_trans_mod_sb(args
->tp
, args
->wasdel
?
597 XFS_TRANS_SB_RES_FDBLOCKS
:
598 XFS_TRANS_SB_FDBLOCKS
,
599 -((long)(args
->len
)));
602 XFS_STATS_INC(xs_allocx
);
603 XFS_STATS_ADD(xs_allocb
, args
->len
);
608 * Allocate a variable extent at exactly agno/bno.
609 * Extent's length (returned in *len) will be between minlen and maxlen,
610 * and of the form k * prod + mod unless there's nothing that large.
611 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
613 STATIC
int /* error */
614 xfs_alloc_ag_vextent_exact(
615 xfs_alloc_arg_t
*args
) /* allocation argument structure */
617 xfs_btree_cur_t
*bno_cur
;/* by block-number btree cursor */
618 xfs_btree_cur_t
*cnt_cur
;/* by count btree cursor */
620 xfs_agblock_t fbno
; /* start block of found extent */
621 xfs_extlen_t flen
; /* length of found extent */
622 xfs_agblock_t tbno
; /* start block of trimmed extent */
623 xfs_extlen_t tlen
; /* length of trimmed extent */
624 xfs_agblock_t tend
; /* end block of trimmed extent */
625 int i
; /* success/failure of operation */
627 ASSERT(args
->alignment
== 1);
630 * Allocate/initialize a cursor for the by-number freespace btree.
632 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
633 args
->agno
, XFS_BTNUM_BNO
);
636 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
637 * Look for the closest free block <= bno, it must contain bno
638 * if any free block does.
640 error
= xfs_alloc_lookup_le(bno_cur
, args
->agbno
, args
->minlen
, &i
);
647 * Grab the freespace record.
649 error
= xfs_alloc_get_rec(bno_cur
, &fbno
, &flen
, &i
);
652 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
653 ASSERT(fbno
<= args
->agbno
);
656 * Check for overlapping busy extents.
658 xfs_extent_busy_trim(args
, fbno
, flen
, &tbno
, &tlen
);
661 * Give up if the start of the extent is busy, or the freespace isn't
662 * long enough for the minimum request.
664 if (tbno
> args
->agbno
)
666 if (tlen
< args
->minlen
)
669 if (tend
< args
->agbno
+ args
->minlen
)
673 * End of extent will be smaller of the freespace end and the
674 * maximal requested end.
676 * Fix the length according to mod and prod if given.
678 args
->len
= XFS_AGBLOCK_MIN(tend
, args
->agbno
+ args
->maxlen
)
680 xfs_alloc_fix_len(args
);
681 if (!xfs_alloc_fix_minleft(args
))
684 ASSERT(args
->agbno
+ args
->len
<= tend
);
687 * We are allocating agbno for args->len
688 * Allocate/initialize a cursor for the by-size btree.
690 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
691 args
->agno
, XFS_BTNUM_CNT
);
692 ASSERT(args
->agbno
+ args
->len
<=
693 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
694 error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
, args
->agbno
,
695 args
->len
, XFSA_FIXUP_BNO_OK
);
697 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
701 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
702 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
705 trace_xfs_alloc_exact_done(args
);
709 /* Didn't find it, return null. */
710 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
711 args
->agbno
= NULLAGBLOCK
;
712 trace_xfs_alloc_exact_notfound(args
);
716 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
717 trace_xfs_alloc_exact_error(args
);
722 * Search the btree in a given direction via the search cursor and compare
723 * the records found against the good extent we've already found.
726 xfs_alloc_find_best_extent(
727 struct xfs_alloc_arg
*args
, /* allocation argument structure */
728 struct xfs_btree_cur
**gcur
, /* good cursor */
729 struct xfs_btree_cur
**scur
, /* searching cursor */
730 xfs_agblock_t gdiff
, /* difference for search comparison */
731 xfs_agblock_t
*sbno
, /* extent found by search */
732 xfs_extlen_t
*slen
, /* extent length */
733 xfs_agblock_t
*sbnoa
, /* aligned extent found by search */
734 xfs_extlen_t
*slena
, /* aligned extent length */
735 int dir
) /* 0 = search right, 1 = search left */
742 /* The good extent is perfect, no need to search. */
747 * Look until we find a better one, run out of space or run off the end.
750 error
= xfs_alloc_get_rec(*scur
, sbno
, slen
, &i
);
753 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
754 xfs_alloc_compute_aligned(args
, *sbno
, *slen
, sbnoa
, slena
);
757 * The good extent is closer than this one.
760 if (*sbnoa
>= args
->agbno
+ gdiff
)
763 if (*sbnoa
<= args
->agbno
- gdiff
)
768 * Same distance, compare length and pick the best.
770 if (*slena
>= args
->minlen
) {
771 args
->len
= XFS_EXTLEN_MIN(*slena
, args
->maxlen
);
772 xfs_alloc_fix_len(args
);
774 sdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
775 args
->alignment
, *sbnoa
,
779 * Choose closer size and invalidate other cursor.
787 error
= xfs_btree_increment(*scur
, 0, &i
);
789 error
= xfs_btree_decrement(*scur
, 0, &i
);
795 xfs_btree_del_cursor(*scur
, XFS_BTREE_NOERROR
);
800 xfs_btree_del_cursor(*gcur
, XFS_BTREE_NOERROR
);
805 /* caller invalidates cursors */
810 * Allocate a variable extent near bno in the allocation group agno.
811 * Extent's length (returned in len) will be between minlen and maxlen,
812 * and of the form k * prod + mod unless there's nothing that large.
813 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
815 STATIC
int /* error */
816 xfs_alloc_ag_vextent_near(
817 xfs_alloc_arg_t
*args
) /* allocation argument structure */
819 xfs_btree_cur_t
*bno_cur_gt
; /* cursor for bno btree, right side */
820 xfs_btree_cur_t
*bno_cur_lt
; /* cursor for bno btree, left side */
821 xfs_btree_cur_t
*cnt_cur
; /* cursor for count btree */
822 xfs_agblock_t gtbno
; /* start bno of right side entry */
823 xfs_agblock_t gtbnoa
; /* aligned ... */
824 xfs_extlen_t gtdiff
; /* difference to right side entry */
825 xfs_extlen_t gtlen
; /* length of right side entry */
826 xfs_extlen_t gtlena
; /* aligned ... */
827 xfs_agblock_t gtnew
; /* useful start bno of right side */
828 int error
; /* error code */
829 int i
; /* result code, temporary */
830 int j
; /* result code, temporary */
831 xfs_agblock_t ltbno
; /* start bno of left side entry */
832 xfs_agblock_t ltbnoa
; /* aligned ... */
833 xfs_extlen_t ltdiff
; /* difference to left side entry */
834 xfs_extlen_t ltlen
; /* length of left side entry */
835 xfs_extlen_t ltlena
; /* aligned ... */
836 xfs_agblock_t ltnew
; /* useful start bno of left side */
837 xfs_extlen_t rlen
; /* length of returned extent */
839 #if defined(DEBUG) && defined(__KERNEL__)
841 * Randomly don't execute the first algorithm.
843 int dofirst
; /* set to do first algorithm */
845 dofirst
= random32() & 1;
856 * Get a cursor for the by-size btree.
858 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
859 args
->agno
, XFS_BTNUM_CNT
);
862 * See if there are any free extents as big as maxlen.
864 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0, args
->maxlen
, &i
)))
867 * If none, then pick up the last entry in the tree unless the
871 if ((error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
, <bno
,
874 if (i
== 0 || ltlen
== 0) {
875 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
876 trace_xfs_alloc_near_noentry(args
);
885 * If the requested extent is large wrt the freespaces available
886 * in this a.g., then the cursor will be pointing to a btree entry
887 * near the right edge of the tree. If it's in the last btree leaf
888 * block, then we just examine all the entries in that block
889 * that are big enough, and pick the best one.
890 * This is written as a while loop so we can break out of it,
891 * but we never loop back to the top.
893 while (xfs_btree_islastblock(cnt_cur
, 0)) {
897 xfs_agblock_t bnew
=0;
899 #if defined(DEBUG) && defined(__KERNEL__)
904 * Start from the entry that lookup found, sequence through
905 * all larger free blocks. If we're actually pointing at a
906 * record smaller than maxlen, go to the start of this block,
907 * and skip all those smaller than minlen.
909 if (ltlen
|| args
->alignment
> 1) {
910 cnt_cur
->bc_ptrs
[0] = 1;
912 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
,
915 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
916 if (ltlen
>= args
->minlen
)
918 if ((error
= xfs_btree_increment(cnt_cur
, 0, &i
)))
921 ASSERT(ltlen
>= args
->minlen
);
925 i
= cnt_cur
->bc_ptrs
[0];
926 for (j
= 1, blen
= 0, bdiff
= 0;
927 !error
&& j
&& (blen
< args
->maxlen
|| bdiff
> 0);
928 error
= xfs_btree_increment(cnt_cur
, 0, &j
)) {
930 * For each entry, decide if it's better than
931 * the previous best entry.
933 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
935 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
936 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
938 if (ltlena
< args
->minlen
)
940 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
941 xfs_alloc_fix_len(args
);
942 ASSERT(args
->len
>= args
->minlen
);
943 if (args
->len
< blen
)
945 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
946 args
->alignment
, ltbnoa
, ltlena
, <new
);
947 if (ltnew
!= NULLAGBLOCK
&&
948 (args
->len
> blen
|| ltdiff
< bdiff
)) {
952 besti
= cnt_cur
->bc_ptrs
[0];
956 * It didn't work. We COULD be in a case where
957 * there's a good record somewhere, so try again.
962 * Point at the best entry, and retrieve it again.
964 cnt_cur
->bc_ptrs
[0] = besti
;
965 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
967 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
968 ASSERT(ltbno
+ ltlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
970 if (!xfs_alloc_fix_minleft(args
)) {
971 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
972 trace_xfs_alloc_near_nominleft(args
);
977 * We are allocating starting at bnew for blen blocks.
980 ASSERT(bnew
>= ltbno
);
981 ASSERT(bnew
+ blen
<= ltbno
+ ltlen
);
983 * Set up a cursor for the by-bno tree.
985 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
,
986 args
->agbp
, args
->agno
, XFS_BTNUM_BNO
);
988 * Fix up the btree entries.
990 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
,
991 ltlen
, bnew
, blen
, XFSA_FIXUP_CNT_OK
)))
993 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
994 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
996 trace_xfs_alloc_near_first(args
);
1001 * Search in the by-bno tree to the left and to the right
1002 * simultaneously, until in each case we find a space big enough,
1003 * or run into the edge of the tree. When we run into the edge,
1004 * we deallocate that cursor.
1005 * If both searches succeed, we compare the two spaces and pick
1007 * With alignment, it's possible for both to fail; the upper
1008 * level algorithm that picks allocation groups for allocations
1009 * is not supposed to do this.
1012 * Allocate and initialize the cursor for the leftward search.
1014 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1015 args
->agno
, XFS_BTNUM_BNO
);
1017 * Lookup <= bno to find the leftward search's starting point.
1019 if ((error
= xfs_alloc_lookup_le(bno_cur_lt
, args
->agbno
, args
->maxlen
, &i
)))
1023 * Didn't find anything; use this cursor for the rightward
1026 bno_cur_gt
= bno_cur_lt
;
1030 * Found something. Duplicate the cursor for the rightward search.
1032 else if ((error
= xfs_btree_dup_cursor(bno_cur_lt
, &bno_cur_gt
)))
1035 * Increment the cursor, so we will point at the entry just right
1036 * of the leftward entry if any, or to the leftmost entry.
1038 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1042 * It failed, there are no rightward entries.
1044 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_NOERROR
);
1048 * Loop going left with the leftward cursor, right with the
1049 * rightward cursor, until either both directions give up or
1050 * we find an entry at least as big as minlen.
1054 if ((error
= xfs_alloc_get_rec(bno_cur_lt
, <bno
, <len
, &i
)))
1056 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1057 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
1059 if (ltlena
>= args
->minlen
)
1061 if ((error
= xfs_btree_decrement(bno_cur_lt
, 0, &i
)))
1064 xfs_btree_del_cursor(bno_cur_lt
,
1070 if ((error
= xfs_alloc_get_rec(bno_cur_gt
, >bno
, >len
, &i
)))
1072 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1073 xfs_alloc_compute_aligned(args
, gtbno
, gtlen
,
1075 if (gtlena
>= args
->minlen
)
1077 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1080 xfs_btree_del_cursor(bno_cur_gt
,
1085 } while (bno_cur_lt
|| bno_cur_gt
);
1088 * Got both cursors still active, need to find better entry.
1090 if (bno_cur_lt
&& bno_cur_gt
) {
1091 if (ltlena
>= args
->minlen
) {
1093 * Left side is good, look for a right side entry.
1095 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1096 xfs_alloc_fix_len(args
);
1097 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1098 args
->alignment
, ltbnoa
, ltlena
, <new
);
1100 error
= xfs_alloc_find_best_extent(args
,
1101 &bno_cur_lt
, &bno_cur_gt
,
1102 ltdiff
, >bno
, >len
,
1104 0 /* search right */);
1106 ASSERT(gtlena
>= args
->minlen
);
1109 * Right side is good, look for a left side entry.
1111 args
->len
= XFS_EXTLEN_MIN(gtlena
, args
->maxlen
);
1112 xfs_alloc_fix_len(args
);
1113 gtdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1114 args
->alignment
, gtbnoa
, gtlena
, >new
);
1116 error
= xfs_alloc_find_best_extent(args
,
1117 &bno_cur_gt
, &bno_cur_lt
,
1118 gtdiff
, <bno
, <len
,
1120 1 /* search left */);
1128 * If we couldn't get anything, give up.
1130 if (bno_cur_lt
== NULL
&& bno_cur_gt
== NULL
) {
1131 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1134 trace_xfs_alloc_near_busy(args
);
1135 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1138 trace_xfs_alloc_size_neither(args
);
1139 args
->agbno
= NULLAGBLOCK
;
1144 * At this point we have selected a freespace entry, either to the
1145 * left or to the right. If it's on the right, copy all the
1146 * useful variables to the "left" set so we only have one
1147 * copy of this code.
1150 bno_cur_lt
= bno_cur_gt
;
1161 * Fix up the length and compute the useful address.
1163 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1164 xfs_alloc_fix_len(args
);
1165 if (!xfs_alloc_fix_minleft(args
)) {
1166 trace_xfs_alloc_near_nominleft(args
);
1167 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1168 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1172 (void)xfs_alloc_compute_diff(args
->agbno
, rlen
, args
->alignment
,
1173 ltbnoa
, ltlena
, <new
);
1174 ASSERT(ltnew
>= ltbno
);
1175 ASSERT(ltnew
+ rlen
<= ltbnoa
+ ltlena
);
1176 ASSERT(ltnew
+ rlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1177 args
->agbno
= ltnew
;
1179 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
, ltlen
,
1180 ltnew
, rlen
, XFSA_FIXUP_BNO_OK
)))
1184 trace_xfs_alloc_near_greater(args
);
1186 trace_xfs_alloc_near_lesser(args
);
1188 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1189 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1193 trace_xfs_alloc_near_error(args
);
1194 if (cnt_cur
!= NULL
)
1195 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1196 if (bno_cur_lt
!= NULL
)
1197 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_ERROR
);
1198 if (bno_cur_gt
!= NULL
)
1199 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_ERROR
);
1204 * Allocate a variable extent anywhere in the allocation group agno.
1205 * Extent's length (returned in len) will be between minlen and maxlen,
1206 * and of the form k * prod + mod unless there's nothing that large.
1207 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1209 STATIC
int /* error */
1210 xfs_alloc_ag_vextent_size(
1211 xfs_alloc_arg_t
*args
) /* allocation argument structure */
1213 xfs_btree_cur_t
*bno_cur
; /* cursor for bno btree */
1214 xfs_btree_cur_t
*cnt_cur
; /* cursor for cnt btree */
1215 int error
; /* error result */
1216 xfs_agblock_t fbno
; /* start of found freespace */
1217 xfs_extlen_t flen
; /* length of found freespace */
1218 int i
; /* temp status variable */
1219 xfs_agblock_t rbno
; /* returned block number */
1220 xfs_extlen_t rlen
; /* length of returned extent */
1225 * Allocate and initialize a cursor for the by-size btree.
1227 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1228 args
->agno
, XFS_BTNUM_CNT
);
1232 * Look for an entry >= maxlen+alignment-1 blocks.
1234 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0,
1235 args
->maxlen
+ args
->alignment
- 1, &i
)))
1239 * If none or we have busy extents that we cannot allocate from, then
1240 * we have to settle for a smaller extent. In the case that there are
1241 * no large extents, this will return the last entry in the tree unless
1242 * the tree is empty. In the case that there are only busy large
1243 * extents, this will return the largest small extent unless there
1244 * are no smaller extents available.
1246 if (!i
|| forced
> 1) {
1247 error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
,
1251 if (i
== 0 || flen
== 0) {
1252 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1253 trace_xfs_alloc_size_noentry(args
);
1257 xfs_alloc_compute_aligned(args
, fbno
, flen
, &rbno
, &rlen
);
1260 * Search for a non-busy extent that is large enough.
1261 * If we are at low space, don't check, or if we fall of
1262 * the end of the btree, turn off the busy check and
1266 error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
, &i
);
1269 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1271 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1274 if (rlen
>= args
->maxlen
)
1277 error
= xfs_btree_increment(cnt_cur
, 0, &i
);
1282 * Our only valid extents must have been busy.
1283 * Make it unbusy by forcing the log out and
1284 * retrying. If we've been here before, forcing
1285 * the log isn't making the extents available,
1286 * which means they have probably been freed in
1287 * this transaction. In that case, we have to
1288 * give up on them and we'll attempt a minlen
1289 * allocation the next time around.
1291 xfs_btree_del_cursor(cnt_cur
,
1293 trace_xfs_alloc_size_busy(args
);
1295 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1302 * In the first case above, we got the last entry in the
1303 * by-size btree. Now we check to see if the space hits maxlen
1304 * once aligned; if not, we search left for something better.
1305 * This can't happen in the second case above.
1307 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1308 XFS_WANT_CORRUPTED_GOTO(rlen
== 0 ||
1309 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
), error0
);
1310 if (rlen
< args
->maxlen
) {
1311 xfs_agblock_t bestfbno
;
1312 xfs_extlen_t bestflen
;
1313 xfs_agblock_t bestrbno
;
1314 xfs_extlen_t bestrlen
;
1321 if ((error
= xfs_btree_decrement(cnt_cur
, 0, &i
)))
1325 if ((error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
,
1328 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1329 if (flen
< bestrlen
)
1331 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1333 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1334 XFS_WANT_CORRUPTED_GOTO(rlen
== 0 ||
1335 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
),
1337 if (rlen
> bestrlen
) {
1342 if (rlen
== args
->maxlen
)
1346 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, bestfbno
, bestflen
,
1349 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1355 args
->wasfromfl
= 0;
1357 * Fix up the length.
1360 if (rlen
< args
->minlen
) {
1362 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1363 trace_xfs_alloc_size_busy(args
);
1364 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1369 xfs_alloc_fix_len(args
);
1371 if (!xfs_alloc_fix_minleft(args
))
1374 XFS_WANT_CORRUPTED_GOTO(rlen
<= flen
, error0
);
1376 * Allocate and initialize a cursor for the by-block tree.
1378 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1379 args
->agno
, XFS_BTNUM_BNO
);
1380 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
,
1381 rbno
, rlen
, XFSA_FIXUP_CNT_OK
)))
1383 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1384 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1385 cnt_cur
= bno_cur
= NULL
;
1388 XFS_WANT_CORRUPTED_GOTO(
1389 args
->agbno
+ args
->len
<=
1390 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1392 trace_xfs_alloc_size_done(args
);
1396 trace_xfs_alloc_size_error(args
);
1398 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1400 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1404 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1405 trace_xfs_alloc_size_nominleft(args
);
1406 args
->agbno
= NULLAGBLOCK
;
1411 * Deal with the case where only small freespaces remain.
1412 * Either return the contents of the last freespace record,
1413 * or allocate space from the freelist if there is nothing in the tree.
1415 STATIC
int /* error */
1416 xfs_alloc_ag_vextent_small(
1417 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1418 xfs_btree_cur_t
*ccur
, /* by-size cursor */
1419 xfs_agblock_t
*fbnop
, /* result block number */
1420 xfs_extlen_t
*flenp
, /* result length */
1421 int *stat
) /* status: 0-freelist, 1-normal/none */
1428 if ((error
= xfs_btree_decrement(ccur
, 0, &i
)))
1431 if ((error
= xfs_alloc_get_rec(ccur
, &fbno
, &flen
, &i
)))
1433 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1436 * Nothing in the btree, try the freelist. Make sure
1437 * to respect minleft even when pulling from the
1440 else if (args
->minlen
== 1 && args
->alignment
== 1 && !args
->isfl
&&
1441 (be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_flcount
)
1443 error
= xfs_alloc_get_freelist(args
->tp
, args
->agbp
, &fbno
, 0);
1446 if (fbno
!= NULLAGBLOCK
) {
1447 xfs_extent_busy_reuse(args
->mp
, args
->agno
, fbno
, 1,
1450 if (args
->userdata
) {
1453 bp
= xfs_btree_get_bufs(args
->mp
, args
->tp
,
1454 args
->agno
, fbno
, 0);
1455 xfs_trans_binval(args
->tp
, bp
);
1459 XFS_WANT_CORRUPTED_GOTO(
1460 args
->agbno
+ args
->len
<=
1461 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1463 args
->wasfromfl
= 1;
1464 trace_xfs_alloc_small_freelist(args
);
1469 * Nothing in the freelist.
1475 * Can't allocate from the freelist for some reason.
1482 * Can't do the allocation, give up.
1484 if (flen
< args
->minlen
) {
1485 args
->agbno
= NULLAGBLOCK
;
1486 trace_xfs_alloc_small_notenough(args
);
1492 trace_xfs_alloc_small_done(args
);
1496 trace_xfs_alloc_small_error(args
);
1501 * Free the extent starting at agno/bno for length.
1503 STATIC
int /* error */
1505 xfs_trans_t
*tp
, /* transaction pointer */
1506 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
1507 xfs_agnumber_t agno
, /* allocation group number */
1508 xfs_agblock_t bno
, /* starting block number */
1509 xfs_extlen_t len
, /* length of extent */
1510 int isfl
) /* set if is freelist blocks - no sb acctg */
1512 xfs_btree_cur_t
*bno_cur
; /* cursor for by-block btree */
1513 xfs_btree_cur_t
*cnt_cur
; /* cursor for by-size btree */
1514 int error
; /* error return value */
1515 xfs_agblock_t gtbno
; /* start of right neighbor block */
1516 xfs_extlen_t gtlen
; /* length of right neighbor block */
1517 int haveleft
; /* have a left neighbor block */
1518 int haveright
; /* have a right neighbor block */
1519 int i
; /* temp, result code */
1520 xfs_agblock_t ltbno
; /* start of left neighbor block */
1521 xfs_extlen_t ltlen
; /* length of left neighbor block */
1522 xfs_mount_t
*mp
; /* mount point struct for filesystem */
1523 xfs_agblock_t nbno
; /* new starting block of freespace */
1524 xfs_extlen_t nlen
; /* new length of freespace */
1525 xfs_perag_t
*pag
; /* per allocation group data */
1529 * Allocate and initialize a cursor for the by-block btree.
1531 bno_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_BNO
);
1534 * Look for a neighboring block on the left (lower block numbers)
1535 * that is contiguous with this space.
1537 if ((error
= xfs_alloc_lookup_le(bno_cur
, bno
, len
, &haveleft
)))
1541 * There is a block to our left.
1543 if ((error
= xfs_alloc_get_rec(bno_cur
, <bno
, <len
, &i
)))
1545 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1547 * It's not contiguous, though.
1549 if (ltbno
+ ltlen
< bno
)
1553 * If this failure happens the request to free this
1554 * space was invalid, it's (partly) already free.
1557 XFS_WANT_CORRUPTED_GOTO(ltbno
+ ltlen
<= bno
, error0
);
1561 * Look for a neighboring block on the right (higher block numbers)
1562 * that is contiguous with this space.
1564 if ((error
= xfs_btree_increment(bno_cur
, 0, &haveright
)))
1568 * There is a block to our right.
1570 if ((error
= xfs_alloc_get_rec(bno_cur
, >bno
, >len
, &i
)))
1572 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1574 * It's not contiguous, though.
1576 if (bno
+ len
< gtbno
)
1580 * If this failure happens the request to free this
1581 * space was invalid, it's (partly) already free.
1584 XFS_WANT_CORRUPTED_GOTO(gtbno
>= bno
+ len
, error0
);
1588 * Now allocate and initialize a cursor for the by-size tree.
1590 cnt_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_CNT
);
1592 * Have both left and right contiguous neighbors.
1593 * Merge all three into a single free block.
1595 if (haveleft
&& haveright
) {
1597 * Delete the old by-size entry on the left.
1599 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1601 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1602 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1604 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1606 * Delete the old by-size entry on the right.
1608 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1610 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1611 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1613 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1615 * Delete the old by-block entry for the right block.
1617 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
1619 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1621 * Move the by-block cursor back to the left neighbor.
1623 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1625 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1628 * Check that this is the right record: delete didn't
1629 * mangle the cursor.
1632 xfs_agblock_t xxbno
;
1635 if ((error
= xfs_alloc_get_rec(bno_cur
, &xxbno
, &xxlen
,
1638 XFS_WANT_CORRUPTED_GOTO(
1639 i
== 1 && xxbno
== ltbno
&& xxlen
== ltlen
,
1644 * Update remaining by-block entry to the new, joined block.
1647 nlen
= len
+ ltlen
+ gtlen
;
1648 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1652 * Have only a left contiguous neighbor.
1653 * Merge it together with the new freespace.
1655 else if (haveleft
) {
1657 * Delete the old by-size entry on the left.
1659 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1661 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1662 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1664 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1666 * Back up the by-block cursor to the left neighbor, and
1667 * update its length.
1669 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1671 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1674 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1678 * Have only a right contiguous neighbor.
1679 * Merge it together with the new freespace.
1681 else if (haveright
) {
1683 * Delete the old by-size entry on the right.
1685 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1687 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1688 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1690 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1692 * Update the starting block and length of the right
1693 * neighbor in the by-block tree.
1697 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1701 * No contiguous neighbors.
1702 * Insert the new freespace into the by-block tree.
1707 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
1709 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1711 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1714 * In all cases we need to insert the new freespace in the by-size tree.
1716 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nbno
, nlen
, &i
)))
1718 XFS_WANT_CORRUPTED_GOTO(i
== 0, error0
);
1719 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
1721 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1722 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1726 * Update the freespace totals in the ag and superblock.
1728 pag
= xfs_perag_get(mp
, agno
);
1729 error
= xfs_alloc_update_counters(tp
, pag
, agbp
, len
);
1735 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_FDBLOCKS
, (long)len
);
1736 XFS_STATS_INC(xs_freex
);
1737 XFS_STATS_ADD(xs_freeb
, len
);
1739 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, haveleft
, haveright
);
1744 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, -1, -1);
1746 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1748 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1753 * Visible (exported) allocation/free functions.
1754 * Some of these are used just by xfs_alloc_btree.c and this file.
1758 * Compute and fill in value of m_ag_maxlevels.
1761 xfs_alloc_compute_maxlevels(
1762 xfs_mount_t
*mp
) /* file system mount structure */
1770 maxleafents
= (mp
->m_sb
.sb_agblocks
+ 1) / 2;
1771 minleafrecs
= mp
->m_alloc_mnr
[0];
1772 minnoderecs
= mp
->m_alloc_mnr
[1];
1773 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
1774 for (level
= 1; maxblocks
> 1; level
++)
1775 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
1776 mp
->m_ag_maxlevels
= level
;
1780 * Find the length of the longest extent in an AG.
1783 xfs_alloc_longest_free_extent(
1784 struct xfs_mount
*mp
,
1785 struct xfs_perag
*pag
)
1787 xfs_extlen_t need
, delta
= 0;
1789 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1790 if (need
> pag
->pagf_flcount
)
1791 delta
= need
- pag
->pagf_flcount
;
1793 if (pag
->pagf_longest
> delta
)
1794 return pag
->pagf_longest
- delta
;
1795 return pag
->pagf_flcount
> 0 || pag
->pagf_longest
> 0;
1799 * Decide whether to use this allocation group for this allocation.
1800 * If so, fix up the btree freelist's size.
1802 STATIC
int /* error */
1803 xfs_alloc_fix_freelist(
1804 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1805 int flags
) /* XFS_ALLOC_FLAG_... */
1807 xfs_buf_t
*agbp
; /* agf buffer pointer */
1808 xfs_agf_t
*agf
; /* a.g. freespace structure pointer */
1809 xfs_buf_t
*agflbp
;/* agfl buffer pointer */
1810 xfs_agblock_t bno
; /* freelist block */
1811 xfs_extlen_t delta
; /* new blocks needed in freelist */
1812 int error
; /* error result code */
1813 xfs_extlen_t longest
;/* longest extent in allocation group */
1814 xfs_mount_t
*mp
; /* file system mount point structure */
1815 xfs_extlen_t need
; /* total blocks needed in freelist */
1816 xfs_perag_t
*pag
; /* per-ag information structure */
1817 xfs_alloc_arg_t targs
; /* local allocation arguments */
1818 xfs_trans_t
*tp
; /* transaction pointer */
1824 if (!pag
->pagf_init
) {
1825 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1828 if (!pag
->pagf_init
) {
1829 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1830 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1838 * If this is a metadata preferred pag and we are user data
1839 * then try somewhere else if we are not being asked to
1840 * try harder at this point
1842 if (pag
->pagf_metadata
&& args
->userdata
&&
1843 (flags
& XFS_ALLOC_FLAG_TRYLOCK
)) {
1844 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1849 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1851 * If it looks like there isn't a long enough extent, or enough
1852 * total blocks, reject it.
1854 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1855 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
1856 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1858 ((int)(pag
->pagf_freeblks
+ pag
->pagf_flcount
-
1859 need
- args
->total
) < (int)args
->minleft
)) {
1861 xfs_trans_brelse(tp
, agbp
);
1868 * Get the a.g. freespace buffer.
1869 * Can fail if we're not blocking on locks, and it's held.
1872 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1876 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1877 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1883 * Figure out how many blocks we should have in the freelist.
1885 agf
= XFS_BUF_TO_AGF(agbp
);
1886 need
= XFS_MIN_FREELIST(agf
, mp
);
1888 * If there isn't enough total or single-extent, reject it.
1890 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1891 delta
= need
> be32_to_cpu(agf
->agf_flcount
) ?
1892 (need
- be32_to_cpu(agf
->agf_flcount
)) : 0;
1893 longest
= be32_to_cpu(agf
->agf_longest
);
1894 longest
= (longest
> delta
) ? (longest
- delta
) :
1895 (be32_to_cpu(agf
->agf_flcount
) > 0 || longest
> 0);
1896 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1898 ((int)(be32_to_cpu(agf
->agf_freeblks
) +
1899 be32_to_cpu(agf
->agf_flcount
) - need
- args
->total
) <
1900 (int)args
->minleft
)) {
1901 xfs_trans_brelse(tp
, agbp
);
1907 * Make the freelist shorter if it's too long.
1909 while (be32_to_cpu(agf
->agf_flcount
) > need
) {
1912 error
= xfs_alloc_get_freelist(tp
, agbp
, &bno
, 0);
1915 if ((error
= xfs_free_ag_extent(tp
, agbp
, args
->agno
, bno
, 1, 1)))
1917 bp
= xfs_btree_get_bufs(mp
, tp
, args
->agno
, bno
, 0);
1918 xfs_trans_binval(tp
, bp
);
1921 * Initialize the args structure.
1923 memset(&targs
, 0, sizeof(targs
));
1927 targs
.agno
= args
->agno
;
1928 targs
.mod
= targs
.minleft
= targs
.wasdel
= targs
.userdata
=
1929 targs
.minalignslop
= 0;
1930 targs
.alignment
= targs
.minlen
= targs
.prod
= targs
.isfl
= 1;
1931 targs
.type
= XFS_ALLOCTYPE_THIS_AG
;
1933 if ((error
= xfs_alloc_read_agfl(mp
, tp
, targs
.agno
, &agflbp
)))
1936 * Make the freelist longer if it's too short.
1938 while (be32_to_cpu(agf
->agf_flcount
) < need
) {
1940 targs
.maxlen
= need
- be32_to_cpu(agf
->agf_flcount
);
1942 * Allocate as many blocks as possible at once.
1944 if ((error
= xfs_alloc_ag_vextent(&targs
))) {
1945 xfs_trans_brelse(tp
, agflbp
);
1949 * Stop if we run out. Won't happen if callers are obeying
1950 * the restrictions correctly. Can happen for free calls
1951 * on a completely full ag.
1953 if (targs
.agbno
== NULLAGBLOCK
) {
1954 if (flags
& XFS_ALLOC_FLAG_FREEING
)
1956 xfs_trans_brelse(tp
, agflbp
);
1961 * Put each allocated block on the list.
1963 for (bno
= targs
.agbno
; bno
< targs
.agbno
+ targs
.len
; bno
++) {
1964 error
= xfs_alloc_put_freelist(tp
, agbp
,
1970 xfs_trans_brelse(tp
, agflbp
);
1976 * Get a block from the freelist.
1977 * Returns with the buffer for the block gotten.
1980 xfs_alloc_get_freelist(
1981 xfs_trans_t
*tp
, /* transaction pointer */
1982 xfs_buf_t
*agbp
, /* buffer containing the agf structure */
1983 xfs_agblock_t
*bnop
, /* block address retrieved from freelist */
1984 int btreeblk
) /* destination is a AGF btree */
1986 xfs_agf_t
*agf
; /* a.g. freespace structure */
1987 xfs_agfl_t
*agfl
; /* a.g. freelist structure */
1988 xfs_buf_t
*agflbp
;/* buffer for a.g. freelist structure */
1989 xfs_agblock_t bno
; /* block number returned */
1992 xfs_mount_t
*mp
; /* mount structure */
1993 xfs_perag_t
*pag
; /* per allocation group data */
1995 agf
= XFS_BUF_TO_AGF(agbp
);
1997 * Freelist is empty, give up.
1999 if (!agf
->agf_flcount
) {
2000 *bnop
= NULLAGBLOCK
;
2004 * Read the array of free blocks.
2007 if ((error
= xfs_alloc_read_agfl(mp
, tp
,
2008 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
2010 agfl
= XFS_BUF_TO_AGFL(agflbp
);
2012 * Get the block number and update the data structures.
2014 bno
= be32_to_cpu(agfl
->agfl_bno
[be32_to_cpu(agf
->agf_flfirst
)]);
2015 be32_add_cpu(&agf
->agf_flfirst
, 1);
2016 xfs_trans_brelse(tp
, agflbp
);
2017 if (be32_to_cpu(agf
->agf_flfirst
) == XFS_AGFL_SIZE(mp
))
2018 agf
->agf_flfirst
= 0;
2020 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2021 be32_add_cpu(&agf
->agf_flcount
, -1);
2022 xfs_trans_agflist_delta(tp
, -1);
2023 pag
->pagf_flcount
--;
2026 logflags
= XFS_AGF_FLFIRST
| XFS_AGF_FLCOUNT
;
2028 be32_add_cpu(&agf
->agf_btreeblks
, 1);
2029 pag
->pagf_btreeblks
++;
2030 logflags
|= XFS_AGF_BTREEBLKS
;
2033 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2040 * Log the given fields from the agf structure.
2044 xfs_trans_t
*tp
, /* transaction pointer */
2045 xfs_buf_t
*bp
, /* buffer for a.g. freelist header */
2046 int fields
) /* mask of fields to be logged (XFS_AGF_...) */
2048 int first
; /* first byte offset */
2049 int last
; /* last byte offset */
2050 static const short offsets
[] = {
2051 offsetof(xfs_agf_t
, agf_magicnum
),
2052 offsetof(xfs_agf_t
, agf_versionnum
),
2053 offsetof(xfs_agf_t
, agf_seqno
),
2054 offsetof(xfs_agf_t
, agf_length
),
2055 offsetof(xfs_agf_t
, agf_roots
[0]),
2056 offsetof(xfs_agf_t
, agf_levels
[0]),
2057 offsetof(xfs_agf_t
, agf_flfirst
),
2058 offsetof(xfs_agf_t
, agf_fllast
),
2059 offsetof(xfs_agf_t
, agf_flcount
),
2060 offsetof(xfs_agf_t
, agf_freeblks
),
2061 offsetof(xfs_agf_t
, agf_longest
),
2062 offsetof(xfs_agf_t
, agf_btreeblks
),
2066 trace_xfs_agf(tp
->t_mountp
, XFS_BUF_TO_AGF(bp
), fields
, _RET_IP_
);
2068 xfs_btree_offsets(fields
, offsets
, XFS_AGF_NUM_BITS
, &first
, &last
);
2069 xfs_trans_log_buf(tp
, bp
, (uint
)first
, (uint
)last
);
2073 * Interface for inode allocation to force the pag data to be initialized.
2076 xfs_alloc_pagf_init(
2077 xfs_mount_t
*mp
, /* file system mount structure */
2078 xfs_trans_t
*tp
, /* transaction pointer */
2079 xfs_agnumber_t agno
, /* allocation group number */
2080 int flags
) /* XFS_ALLOC_FLAGS_... */
2085 if ((error
= xfs_alloc_read_agf(mp
, tp
, agno
, flags
, &bp
)))
2088 xfs_trans_brelse(tp
, bp
);
2093 * Put the block on the freelist for the allocation group.
2096 xfs_alloc_put_freelist(
2097 xfs_trans_t
*tp
, /* transaction pointer */
2098 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
2099 xfs_buf_t
*agflbp
,/* buffer for a.g. free block array */
2100 xfs_agblock_t bno
, /* block being freed */
2101 int btreeblk
) /* block came from a AGF btree */
2103 xfs_agf_t
*agf
; /* a.g. freespace structure */
2104 xfs_agfl_t
*agfl
; /* a.g. free block array */
2105 __be32
*blockp
;/* pointer to array entry */
2108 xfs_mount_t
*mp
; /* mount structure */
2109 xfs_perag_t
*pag
; /* per allocation group data */
2111 agf
= XFS_BUF_TO_AGF(agbp
);
2114 if (!agflbp
&& (error
= xfs_alloc_read_agfl(mp
, tp
,
2115 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
2117 agfl
= XFS_BUF_TO_AGFL(agflbp
);
2118 be32_add_cpu(&agf
->agf_fllast
, 1);
2119 if (be32_to_cpu(agf
->agf_fllast
) == XFS_AGFL_SIZE(mp
))
2120 agf
->agf_fllast
= 0;
2122 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2123 be32_add_cpu(&agf
->agf_flcount
, 1);
2124 xfs_trans_agflist_delta(tp
, 1);
2125 pag
->pagf_flcount
++;
2127 logflags
= XFS_AGF_FLLAST
| XFS_AGF_FLCOUNT
;
2129 be32_add_cpu(&agf
->agf_btreeblks
, -1);
2130 pag
->pagf_btreeblks
--;
2131 logflags
|= XFS_AGF_BTREEBLKS
;
2135 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2137 ASSERT(be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
));
2138 blockp
= &agfl
->agfl_bno
[be32_to_cpu(agf
->agf_fllast
)];
2139 *blockp
= cpu_to_be32(bno
);
2140 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2141 xfs_trans_log_buf(tp
, agflbp
,
2142 (int)((xfs_caddr_t
)blockp
- (xfs_caddr_t
)agfl
),
2143 (int)((xfs_caddr_t
)blockp
- (xfs_caddr_t
)agfl
+
2144 sizeof(xfs_agblock_t
) - 1));
2152 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2153 struct xfs_agf
*agf
;
2156 agf
= XFS_BUF_TO_AGF(bp
);
2158 agf_ok
= agf
->agf_magicnum
== cpu_to_be32(XFS_AGF_MAGIC
) &&
2159 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf
->agf_versionnum
)) &&
2160 be32_to_cpu(agf
->agf_freeblks
) <= be32_to_cpu(agf
->agf_length
) &&
2161 be32_to_cpu(agf
->agf_flfirst
) < XFS_AGFL_SIZE(mp
) &&
2162 be32_to_cpu(agf
->agf_fllast
) < XFS_AGFL_SIZE(mp
) &&
2163 be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
);
2166 * during growfs operations, the perag is not fully initialised,
2167 * so we can't use it for any useful checking. growfs ensures we can't
2168 * use it by using uncached buffers that don't have the perag attached
2169 * so we can detect and avoid this problem.
2172 agf_ok
= agf_ok
&& be32_to_cpu(agf
->agf_seqno
) ==
2173 bp
->b_pag
->pag_agno
;
2175 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
2176 agf_ok
= agf_ok
&& be32_to_cpu(agf
->agf_btreeblks
) <=
2177 be32_to_cpu(agf
->agf_length
);
2179 if (unlikely(XFS_TEST_ERROR(!agf_ok
, mp
, XFS_ERRTAG_ALLOC_READ_AGF
,
2180 XFS_RANDOM_ALLOC_READ_AGF
))) {
2181 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, agf
);
2182 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
2187 xfs_agf_read_verify(
2194 xfs_agf_write_verify(
2200 const struct xfs_buf_ops xfs_agf_buf_ops
= {
2201 .verify_read
= xfs_agf_read_verify
,
2202 .verify_write
= xfs_agf_write_verify
,
2206 * Read in the allocation group header (free/alloc section).
2210 struct xfs_mount
*mp
, /* mount point structure */
2211 struct xfs_trans
*tp
, /* transaction pointer */
2212 xfs_agnumber_t agno
, /* allocation group number */
2213 int flags
, /* XFS_BUF_ */
2214 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2218 ASSERT(agno
!= NULLAGNUMBER
);
2219 error
= xfs_trans_read_buf(
2220 mp
, tp
, mp
->m_ddev_targp
,
2221 XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
)),
2222 XFS_FSS_TO_BB(mp
, 1), flags
, bpp
, &xfs_agf_buf_ops
);
2228 ASSERT(!(*bpp
)->b_error
);
2229 xfs_buf_set_ref(*bpp
, XFS_AGF_REF
);
2234 * Read in the allocation group header (free/alloc section).
2238 struct xfs_mount
*mp
, /* mount point structure */
2239 struct xfs_trans
*tp
, /* transaction pointer */
2240 xfs_agnumber_t agno
, /* allocation group number */
2241 int flags
, /* XFS_ALLOC_FLAG_... */
2242 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2244 struct xfs_agf
*agf
; /* ag freelist header */
2245 struct xfs_perag
*pag
; /* per allocation group data */
2248 ASSERT(agno
!= NULLAGNUMBER
);
2250 error
= xfs_read_agf(mp
, tp
, agno
,
2251 (flags
& XFS_ALLOC_FLAG_TRYLOCK
) ? XBF_TRYLOCK
: 0,
2257 ASSERT(!(*bpp
)->b_error
);
2259 agf
= XFS_BUF_TO_AGF(*bpp
);
2260 pag
= xfs_perag_get(mp
, agno
);
2261 if (!pag
->pagf_init
) {
2262 pag
->pagf_freeblks
= be32_to_cpu(agf
->agf_freeblks
);
2263 pag
->pagf_btreeblks
= be32_to_cpu(agf
->agf_btreeblks
);
2264 pag
->pagf_flcount
= be32_to_cpu(agf
->agf_flcount
);
2265 pag
->pagf_longest
= be32_to_cpu(agf
->agf_longest
);
2266 pag
->pagf_levels
[XFS_BTNUM_BNOi
] =
2267 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]);
2268 pag
->pagf_levels
[XFS_BTNUM_CNTi
] =
2269 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]);
2270 spin_lock_init(&pag
->pagb_lock
);
2271 pag
->pagb_count
= 0;
2272 pag
->pagb_tree
= RB_ROOT
;
2276 else if (!XFS_FORCED_SHUTDOWN(mp
)) {
2277 ASSERT(pag
->pagf_freeblks
== be32_to_cpu(agf
->agf_freeblks
));
2278 ASSERT(pag
->pagf_btreeblks
== be32_to_cpu(agf
->agf_btreeblks
));
2279 ASSERT(pag
->pagf_flcount
== be32_to_cpu(agf
->agf_flcount
));
2280 ASSERT(pag
->pagf_longest
== be32_to_cpu(agf
->agf_longest
));
2281 ASSERT(pag
->pagf_levels
[XFS_BTNUM_BNOi
] ==
2282 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]));
2283 ASSERT(pag
->pagf_levels
[XFS_BTNUM_CNTi
] ==
2284 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]));
2292 * Allocate an extent (variable-size).
2293 * Depending on the allocation type, we either look in a single allocation
2294 * group or loop over the allocation groups to find the result.
2298 xfs_alloc_arg_t
*args
) /* allocation argument structure */
2300 xfs_agblock_t agsize
; /* allocation group size */
2302 int flags
; /* XFS_ALLOC_FLAG_... locking flags */
2303 xfs_extlen_t minleft
;/* minimum left value, temp copy */
2304 xfs_mount_t
*mp
; /* mount structure pointer */
2305 xfs_agnumber_t sagno
; /* starting allocation group number */
2306 xfs_alloctype_t type
; /* input allocation type */
2309 xfs_agnumber_t rotorstep
= xfs_rotorstep
; /* inode32 agf stepper */
2312 type
= args
->otype
= args
->type
;
2313 args
->agbno
= NULLAGBLOCK
;
2315 * Just fix this up, for the case where the last a.g. is shorter
2316 * (or there's only one a.g.) and the caller couldn't easily figure
2317 * that out (xfs_bmap_alloc).
2319 agsize
= mp
->m_sb
.sb_agblocks
;
2320 if (args
->maxlen
> agsize
)
2321 args
->maxlen
= agsize
;
2322 if (args
->alignment
== 0)
2323 args
->alignment
= 1;
2324 ASSERT(XFS_FSB_TO_AGNO(mp
, args
->fsbno
) < mp
->m_sb
.sb_agcount
);
2325 ASSERT(XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) < agsize
);
2326 ASSERT(args
->minlen
<= args
->maxlen
);
2327 ASSERT(args
->minlen
<= agsize
);
2328 ASSERT(args
->mod
< args
->prod
);
2329 if (XFS_FSB_TO_AGNO(mp
, args
->fsbno
) >= mp
->m_sb
.sb_agcount
||
2330 XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) >= agsize
||
2331 args
->minlen
> args
->maxlen
|| args
->minlen
> agsize
||
2332 args
->mod
>= args
->prod
) {
2333 args
->fsbno
= NULLFSBLOCK
;
2334 trace_xfs_alloc_vextent_badargs(args
);
2337 minleft
= args
->minleft
;
2340 case XFS_ALLOCTYPE_THIS_AG
:
2341 case XFS_ALLOCTYPE_NEAR_BNO
:
2342 case XFS_ALLOCTYPE_THIS_BNO
:
2344 * These three force us into a single a.g.
2346 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2347 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2349 error
= xfs_alloc_fix_freelist(args
, 0);
2350 args
->minleft
= minleft
;
2352 trace_xfs_alloc_vextent_nofix(args
);
2356 trace_xfs_alloc_vextent_noagbp(args
);
2359 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2360 if ((error
= xfs_alloc_ag_vextent(args
)))
2363 case XFS_ALLOCTYPE_START_BNO
:
2365 * Try near allocation first, then anywhere-in-ag after
2366 * the first a.g. fails.
2368 if ((args
->userdata
== XFS_ALLOC_INITIAL_USER_DATA
) &&
2369 (mp
->m_flags
& XFS_MOUNT_32BITINODES
)) {
2370 args
->fsbno
= XFS_AGB_TO_FSB(mp
,
2371 ((mp
->m_agfrotor
/ rotorstep
) %
2372 mp
->m_sb
.sb_agcount
), 0);
2375 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2376 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2378 case XFS_ALLOCTYPE_ANY_AG
:
2379 case XFS_ALLOCTYPE_START_AG
:
2380 case XFS_ALLOCTYPE_FIRST_AG
:
2382 * Rotate through the allocation groups looking for a winner.
2384 if (type
== XFS_ALLOCTYPE_ANY_AG
) {
2386 * Start with the last place we left off.
2388 args
->agno
= sagno
= (mp
->m_agfrotor
/ rotorstep
) %
2389 mp
->m_sb
.sb_agcount
;
2390 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2391 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2392 } else if (type
== XFS_ALLOCTYPE_FIRST_AG
) {
2394 * Start with allocation group given by bno.
2396 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2397 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2401 if (type
== XFS_ALLOCTYPE_START_AG
)
2402 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2404 * Start with the given allocation group.
2406 args
->agno
= sagno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2407 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2410 * Loop over allocation groups twice; first time with
2411 * trylock set, second time without.
2414 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2415 if (no_min
) args
->minleft
= 0;
2416 error
= xfs_alloc_fix_freelist(args
, flags
);
2417 args
->minleft
= minleft
;
2419 trace_xfs_alloc_vextent_nofix(args
);
2423 * If we get a buffer back then the allocation will fly.
2426 if ((error
= xfs_alloc_ag_vextent(args
)))
2431 trace_xfs_alloc_vextent_loopfailed(args
);
2434 * Didn't work, figure out the next iteration.
2436 if (args
->agno
== sagno
&&
2437 type
== XFS_ALLOCTYPE_START_BNO
)
2438 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2440 * For the first allocation, we can try any AG to get
2441 * space. However, if we already have allocated a
2442 * block, we don't want to try AGs whose number is below
2443 * sagno. Otherwise, we may end up with out-of-order
2444 * locking of AGF, which might cause deadlock.
2446 if (++(args
->agno
) == mp
->m_sb
.sb_agcount
) {
2447 if (args
->firstblock
!= NULLFSBLOCK
)
2453 * Reached the starting a.g., must either be done
2454 * or switch to non-trylock mode.
2456 if (args
->agno
== sagno
) {
2458 args
->agbno
= NULLAGBLOCK
;
2459 trace_xfs_alloc_vextent_allfailed(args
);
2466 if (type
== XFS_ALLOCTYPE_START_BNO
) {
2467 args
->agbno
= XFS_FSB_TO_AGBNO(mp
,
2469 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2473 xfs_perag_put(args
->pag
);
2475 if (bump_rotor
|| (type
== XFS_ALLOCTYPE_ANY_AG
)) {
2476 if (args
->agno
== sagno
)
2477 mp
->m_agfrotor
= (mp
->m_agfrotor
+ 1) %
2478 (mp
->m_sb
.sb_agcount
* rotorstep
);
2480 mp
->m_agfrotor
= (args
->agno
* rotorstep
+ 1) %
2481 (mp
->m_sb
.sb_agcount
* rotorstep
);
2488 if (args
->agbno
== NULLAGBLOCK
)
2489 args
->fsbno
= NULLFSBLOCK
;
2491 args
->fsbno
= XFS_AGB_TO_FSB(mp
, args
->agno
, args
->agbno
);
2493 ASSERT(args
->len
>= args
->minlen
);
2494 ASSERT(args
->len
<= args
->maxlen
);
2495 ASSERT(args
->agbno
% args
->alignment
== 0);
2496 XFS_AG_CHECK_DADDR(mp
, XFS_FSB_TO_DADDR(mp
, args
->fsbno
),
2500 xfs_perag_put(args
->pag
);
2503 xfs_perag_put(args
->pag
);
2509 * Just break up the extent address and hand off to xfs_free_ag_extent
2510 * after fixing up the freelist.
2514 xfs_trans_t
*tp
, /* transaction pointer */
2515 xfs_fsblock_t bno
, /* starting block number of extent */
2516 xfs_extlen_t len
) /* length of extent */
2518 xfs_alloc_arg_t args
;
2522 memset(&args
, 0, sizeof(xfs_alloc_arg_t
));
2524 args
.mp
= tp
->t_mountp
;
2527 * validate that the block number is legal - the enables us to detect
2528 * and handle a silent filesystem corruption rather than crashing.
2530 args
.agno
= XFS_FSB_TO_AGNO(args
.mp
, bno
);
2531 if (args
.agno
>= args
.mp
->m_sb
.sb_agcount
)
2532 return EFSCORRUPTED
;
2534 args
.agbno
= XFS_FSB_TO_AGBNO(args
.mp
, bno
);
2535 if (args
.agbno
>= args
.mp
->m_sb
.sb_agblocks
)
2536 return EFSCORRUPTED
;
2538 args
.pag
= xfs_perag_get(args
.mp
, args
.agno
);
2541 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
2545 /* validate the extent size is legal now we have the agf locked */
2546 if (args
.agbno
+ len
>
2547 be32_to_cpu(XFS_BUF_TO_AGF(args
.agbp
)->agf_length
)) {
2548 error
= EFSCORRUPTED
;
2552 error
= xfs_free_ag_extent(tp
, args
.agbp
, args
.agno
, args
.agbno
, len
, 0);
2554 xfs_extent_busy_insert(tp
, args
.agno
, args
.agbno
, len
, 0);
2556 xfs_perag_put(args
.pag
);