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);
434 * Read in the allocation group free block array.
436 STATIC
int /* error */
438 xfs_mount_t
*mp
, /* mount point structure */
439 xfs_trans_t
*tp
, /* transaction pointer */
440 xfs_agnumber_t agno
, /* allocation group number */
441 xfs_buf_t
**bpp
) /* buffer for the ag free block array */
443 xfs_buf_t
*bp
; /* return value */
446 ASSERT(agno
!= NULLAGNUMBER
);
447 error
= xfs_trans_read_buf(
448 mp
, tp
, mp
->m_ddev_targp
,
449 XFS_AG_DADDR(mp
, agno
, XFS_AGFL_DADDR(mp
)),
450 XFS_FSS_TO_BB(mp
, 1), 0, &bp
);
453 ASSERT(!xfs_buf_geterror(bp
));
454 xfs_buf_set_ref(bp
, XFS_AGFL_REF
);
460 xfs_alloc_update_counters(
461 struct xfs_trans
*tp
,
462 struct xfs_perag
*pag
,
463 struct xfs_buf
*agbp
,
466 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
468 pag
->pagf_freeblks
+= len
;
469 be32_add_cpu(&agf
->agf_freeblks
, len
);
471 xfs_trans_agblocks_delta(tp
, len
);
472 if (unlikely(be32_to_cpu(agf
->agf_freeblks
) >
473 be32_to_cpu(agf
->agf_length
)))
476 xfs_alloc_log_agf(tp
, agbp
, XFS_AGF_FREEBLKS
);
481 * Allocation group level functions.
485 * Allocate a variable extent in the allocation group agno.
486 * Type and bno are used to determine where in the allocation group the
488 * Extent's length (returned in *len) will be between minlen and maxlen,
489 * and of the form k * prod + mod unless there's nothing that large.
490 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
492 STATIC
int /* error */
493 xfs_alloc_ag_vextent(
494 xfs_alloc_arg_t
*args
) /* argument structure for allocation */
498 ASSERT(args
->minlen
> 0);
499 ASSERT(args
->maxlen
> 0);
500 ASSERT(args
->minlen
<= args
->maxlen
);
501 ASSERT(args
->mod
< args
->prod
);
502 ASSERT(args
->alignment
> 0);
504 * Branch to correct routine based on the type.
507 switch (args
->type
) {
508 case XFS_ALLOCTYPE_THIS_AG
:
509 error
= xfs_alloc_ag_vextent_size(args
);
511 case XFS_ALLOCTYPE_NEAR_BNO
:
512 error
= xfs_alloc_ag_vextent_near(args
);
514 case XFS_ALLOCTYPE_THIS_BNO
:
515 error
= xfs_alloc_ag_vextent_exact(args
);
522 if (error
|| args
->agbno
== NULLAGBLOCK
)
525 ASSERT(args
->len
>= args
->minlen
);
526 ASSERT(args
->len
<= args
->maxlen
);
527 ASSERT(!args
->wasfromfl
|| !args
->isfl
);
528 ASSERT(args
->agbno
% args
->alignment
== 0);
530 if (!args
->wasfromfl
) {
531 error
= xfs_alloc_update_counters(args
->tp
, args
->pag
,
533 -((long)(args
->len
)));
537 ASSERT(!xfs_extent_busy_search(args
->mp
, args
->agno
,
538 args
->agbno
, args
->len
));
542 xfs_trans_mod_sb(args
->tp
, args
->wasdel
?
543 XFS_TRANS_SB_RES_FDBLOCKS
:
544 XFS_TRANS_SB_FDBLOCKS
,
545 -((long)(args
->len
)));
548 XFS_STATS_INC(xs_allocx
);
549 XFS_STATS_ADD(xs_allocb
, args
->len
);
554 * Allocate a variable extent at exactly agno/bno.
555 * Extent's length (returned in *len) will be between minlen and maxlen,
556 * and of the form k * prod + mod unless there's nothing that large.
557 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
559 STATIC
int /* error */
560 xfs_alloc_ag_vextent_exact(
561 xfs_alloc_arg_t
*args
) /* allocation argument structure */
563 xfs_btree_cur_t
*bno_cur
;/* by block-number btree cursor */
564 xfs_btree_cur_t
*cnt_cur
;/* by count btree cursor */
566 xfs_agblock_t fbno
; /* start block of found extent */
567 xfs_extlen_t flen
; /* length of found extent */
568 xfs_agblock_t tbno
; /* start block of trimmed extent */
569 xfs_extlen_t tlen
; /* length of trimmed extent */
570 xfs_agblock_t tend
; /* end block of trimmed extent */
571 int i
; /* success/failure of operation */
573 ASSERT(args
->alignment
== 1);
576 * Allocate/initialize a cursor for the by-number freespace btree.
578 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
579 args
->agno
, XFS_BTNUM_BNO
);
582 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
583 * Look for the closest free block <= bno, it must contain bno
584 * if any free block does.
586 error
= xfs_alloc_lookup_le(bno_cur
, args
->agbno
, args
->minlen
, &i
);
593 * Grab the freespace record.
595 error
= xfs_alloc_get_rec(bno_cur
, &fbno
, &flen
, &i
);
598 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
599 ASSERT(fbno
<= args
->agbno
);
602 * Check for overlapping busy extents.
604 xfs_extent_busy_trim(args
, fbno
, flen
, &tbno
, &tlen
);
607 * Give up if the start of the extent is busy, or the freespace isn't
608 * long enough for the minimum request.
610 if (tbno
> args
->agbno
)
612 if (tlen
< args
->minlen
)
615 if (tend
< args
->agbno
+ args
->minlen
)
619 * End of extent will be smaller of the freespace end and the
620 * maximal requested end.
622 * Fix the length according to mod and prod if given.
624 args
->len
= XFS_AGBLOCK_MIN(tend
, args
->agbno
+ args
->maxlen
)
626 xfs_alloc_fix_len(args
);
627 if (!xfs_alloc_fix_minleft(args
))
630 ASSERT(args
->agbno
+ args
->len
<= tend
);
633 * We are allocating agbno for args->len
634 * Allocate/initialize a cursor for the by-size btree.
636 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
637 args
->agno
, XFS_BTNUM_CNT
);
638 ASSERT(args
->agbno
+ args
->len
<=
639 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
640 error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
, args
->agbno
,
641 args
->len
, XFSA_FIXUP_BNO_OK
);
643 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
647 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
648 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
651 trace_xfs_alloc_exact_done(args
);
655 /* Didn't find it, return null. */
656 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
657 args
->agbno
= NULLAGBLOCK
;
658 trace_xfs_alloc_exact_notfound(args
);
662 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
663 trace_xfs_alloc_exact_error(args
);
668 * Search the btree in a given direction via the search cursor and compare
669 * the records found against the good extent we've already found.
672 xfs_alloc_find_best_extent(
673 struct xfs_alloc_arg
*args
, /* allocation argument structure */
674 struct xfs_btree_cur
**gcur
, /* good cursor */
675 struct xfs_btree_cur
**scur
, /* searching cursor */
676 xfs_agblock_t gdiff
, /* difference for search comparison */
677 xfs_agblock_t
*sbno
, /* extent found by search */
678 xfs_extlen_t
*slen
, /* extent length */
679 xfs_agblock_t
*sbnoa
, /* aligned extent found by search */
680 xfs_extlen_t
*slena
, /* aligned extent length */
681 int dir
) /* 0 = search right, 1 = search left */
688 /* The good extent is perfect, no need to search. */
693 * Look until we find a better one, run out of space or run off the end.
696 error
= xfs_alloc_get_rec(*scur
, sbno
, slen
, &i
);
699 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
700 xfs_alloc_compute_aligned(args
, *sbno
, *slen
, sbnoa
, slena
);
703 * The good extent is closer than this one.
706 if (*sbnoa
>= args
->agbno
+ gdiff
)
709 if (*sbnoa
<= args
->agbno
- gdiff
)
714 * Same distance, compare length and pick the best.
716 if (*slena
>= args
->minlen
) {
717 args
->len
= XFS_EXTLEN_MIN(*slena
, args
->maxlen
);
718 xfs_alloc_fix_len(args
);
720 sdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
721 args
->alignment
, *sbnoa
,
725 * Choose closer size and invalidate other cursor.
733 error
= xfs_btree_increment(*scur
, 0, &i
);
735 error
= xfs_btree_decrement(*scur
, 0, &i
);
741 xfs_btree_del_cursor(*scur
, XFS_BTREE_NOERROR
);
746 xfs_btree_del_cursor(*gcur
, XFS_BTREE_NOERROR
);
751 /* caller invalidates cursors */
756 * Allocate a variable extent near bno in the allocation group agno.
757 * Extent's length (returned in len) will be between minlen and maxlen,
758 * and of the form k * prod + mod unless there's nothing that large.
759 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
761 STATIC
int /* error */
762 xfs_alloc_ag_vextent_near(
763 xfs_alloc_arg_t
*args
) /* allocation argument structure */
765 xfs_btree_cur_t
*bno_cur_gt
; /* cursor for bno btree, right side */
766 xfs_btree_cur_t
*bno_cur_lt
; /* cursor for bno btree, left side */
767 xfs_btree_cur_t
*cnt_cur
; /* cursor for count btree */
768 xfs_agblock_t gtbno
; /* start bno of right side entry */
769 xfs_agblock_t gtbnoa
; /* aligned ... */
770 xfs_extlen_t gtdiff
; /* difference to right side entry */
771 xfs_extlen_t gtlen
; /* length of right side entry */
772 xfs_extlen_t gtlena
; /* aligned ... */
773 xfs_agblock_t gtnew
; /* useful start bno of right side */
774 int error
; /* error code */
775 int i
; /* result code, temporary */
776 int j
; /* result code, temporary */
777 xfs_agblock_t ltbno
; /* start bno of left side entry */
778 xfs_agblock_t ltbnoa
; /* aligned ... */
779 xfs_extlen_t ltdiff
; /* difference to left side entry */
780 xfs_extlen_t ltlen
; /* length of left side entry */
781 xfs_extlen_t ltlena
; /* aligned ... */
782 xfs_agblock_t ltnew
; /* useful start bno of left side */
783 xfs_extlen_t rlen
; /* length of returned extent */
785 #if defined(DEBUG) && defined(__KERNEL__)
787 * Randomly don't execute the first algorithm.
789 int dofirst
; /* set to do first algorithm */
791 dofirst
= random32() & 1;
802 * Get a cursor for the by-size btree.
804 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
805 args
->agno
, XFS_BTNUM_CNT
);
808 * See if there are any free extents as big as maxlen.
810 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0, args
->maxlen
, &i
)))
813 * If none, then pick up the last entry in the tree unless the
817 if ((error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
, <bno
,
820 if (i
== 0 || ltlen
== 0) {
821 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
822 trace_xfs_alloc_near_noentry(args
);
831 * If the requested extent is large wrt the freespaces available
832 * in this a.g., then the cursor will be pointing to a btree entry
833 * near the right edge of the tree. If it's in the last btree leaf
834 * block, then we just examine all the entries in that block
835 * that are big enough, and pick the best one.
836 * This is written as a while loop so we can break out of it,
837 * but we never loop back to the top.
839 while (xfs_btree_islastblock(cnt_cur
, 0)) {
843 xfs_agblock_t bnew
=0;
845 #if defined(DEBUG) && defined(__KERNEL__)
850 * Start from the entry that lookup found, sequence through
851 * all larger free blocks. If we're actually pointing at a
852 * record smaller than maxlen, go to the start of this block,
853 * and skip all those smaller than minlen.
855 if (ltlen
|| args
->alignment
> 1) {
856 cnt_cur
->bc_ptrs
[0] = 1;
858 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
,
861 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
862 if (ltlen
>= args
->minlen
)
864 if ((error
= xfs_btree_increment(cnt_cur
, 0, &i
)))
867 ASSERT(ltlen
>= args
->minlen
);
871 i
= cnt_cur
->bc_ptrs
[0];
872 for (j
= 1, blen
= 0, bdiff
= 0;
873 !error
&& j
&& (blen
< args
->maxlen
|| bdiff
> 0);
874 error
= xfs_btree_increment(cnt_cur
, 0, &j
)) {
876 * For each entry, decide if it's better than
877 * the previous best entry.
879 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
881 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
882 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
884 if (ltlena
< args
->minlen
)
886 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
887 xfs_alloc_fix_len(args
);
888 ASSERT(args
->len
>= args
->minlen
);
889 if (args
->len
< blen
)
891 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
892 args
->alignment
, ltbnoa
, ltlena
, <new
);
893 if (ltnew
!= NULLAGBLOCK
&&
894 (args
->len
> blen
|| ltdiff
< bdiff
)) {
898 besti
= cnt_cur
->bc_ptrs
[0];
902 * It didn't work. We COULD be in a case where
903 * there's a good record somewhere, so try again.
908 * Point at the best entry, and retrieve it again.
910 cnt_cur
->bc_ptrs
[0] = besti
;
911 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
913 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
914 ASSERT(ltbno
+ ltlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
916 if (!xfs_alloc_fix_minleft(args
)) {
917 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
918 trace_xfs_alloc_near_nominleft(args
);
923 * We are allocating starting at bnew for blen blocks.
926 ASSERT(bnew
>= ltbno
);
927 ASSERT(bnew
+ blen
<= ltbno
+ ltlen
);
929 * Set up a cursor for the by-bno tree.
931 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
,
932 args
->agbp
, args
->agno
, XFS_BTNUM_BNO
);
934 * Fix up the btree entries.
936 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
,
937 ltlen
, bnew
, blen
, XFSA_FIXUP_CNT_OK
)))
939 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
940 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
942 trace_xfs_alloc_near_first(args
);
947 * Search in the by-bno tree to the left and to the right
948 * simultaneously, until in each case we find a space big enough,
949 * or run into the edge of the tree. When we run into the edge,
950 * we deallocate that cursor.
951 * If both searches succeed, we compare the two spaces and pick
953 * With alignment, it's possible for both to fail; the upper
954 * level algorithm that picks allocation groups for allocations
955 * is not supposed to do this.
958 * Allocate and initialize the cursor for the leftward search.
960 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
961 args
->agno
, XFS_BTNUM_BNO
);
963 * Lookup <= bno to find the leftward search's starting point.
965 if ((error
= xfs_alloc_lookup_le(bno_cur_lt
, args
->agbno
, args
->maxlen
, &i
)))
969 * Didn't find anything; use this cursor for the rightward
972 bno_cur_gt
= bno_cur_lt
;
976 * Found something. Duplicate the cursor for the rightward search.
978 else if ((error
= xfs_btree_dup_cursor(bno_cur_lt
, &bno_cur_gt
)))
981 * Increment the cursor, so we will point at the entry just right
982 * of the leftward entry if any, or to the leftmost entry.
984 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
988 * It failed, there are no rightward entries.
990 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_NOERROR
);
994 * Loop going left with the leftward cursor, right with the
995 * rightward cursor, until either both directions give up or
996 * we find an entry at least as big as minlen.
1000 if ((error
= xfs_alloc_get_rec(bno_cur_lt
, <bno
, <len
, &i
)))
1002 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1003 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
1005 if (ltlena
>= args
->minlen
)
1007 if ((error
= xfs_btree_decrement(bno_cur_lt
, 0, &i
)))
1010 xfs_btree_del_cursor(bno_cur_lt
,
1016 if ((error
= xfs_alloc_get_rec(bno_cur_gt
, >bno
, >len
, &i
)))
1018 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1019 xfs_alloc_compute_aligned(args
, gtbno
, gtlen
,
1021 if (gtlena
>= args
->minlen
)
1023 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1026 xfs_btree_del_cursor(bno_cur_gt
,
1031 } while (bno_cur_lt
|| bno_cur_gt
);
1034 * Got both cursors still active, need to find better entry.
1036 if (bno_cur_lt
&& bno_cur_gt
) {
1037 if (ltlena
>= args
->minlen
) {
1039 * Left side is good, look for a right side entry.
1041 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1042 xfs_alloc_fix_len(args
);
1043 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1044 args
->alignment
, ltbnoa
, ltlena
, <new
);
1046 error
= xfs_alloc_find_best_extent(args
,
1047 &bno_cur_lt
, &bno_cur_gt
,
1048 ltdiff
, >bno
, >len
,
1050 0 /* search right */);
1052 ASSERT(gtlena
>= args
->minlen
);
1055 * Right side is good, look for a left side entry.
1057 args
->len
= XFS_EXTLEN_MIN(gtlena
, args
->maxlen
);
1058 xfs_alloc_fix_len(args
);
1059 gtdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1060 args
->alignment
, gtbnoa
, gtlena
, >new
);
1062 error
= xfs_alloc_find_best_extent(args
,
1063 &bno_cur_gt
, &bno_cur_lt
,
1064 gtdiff
, <bno
, <len
,
1066 1 /* search left */);
1074 * If we couldn't get anything, give up.
1076 if (bno_cur_lt
== NULL
&& bno_cur_gt
== NULL
) {
1077 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1080 trace_xfs_alloc_near_busy(args
);
1081 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1084 trace_xfs_alloc_size_neither(args
);
1085 args
->agbno
= NULLAGBLOCK
;
1090 * At this point we have selected a freespace entry, either to the
1091 * left or to the right. If it's on the right, copy all the
1092 * useful variables to the "left" set so we only have one
1093 * copy of this code.
1096 bno_cur_lt
= bno_cur_gt
;
1107 * Fix up the length and compute the useful address.
1109 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1110 xfs_alloc_fix_len(args
);
1111 if (!xfs_alloc_fix_minleft(args
)) {
1112 trace_xfs_alloc_near_nominleft(args
);
1113 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1114 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1118 (void)xfs_alloc_compute_diff(args
->agbno
, rlen
, args
->alignment
,
1119 ltbnoa
, ltlena
, <new
);
1120 ASSERT(ltnew
>= ltbno
);
1121 ASSERT(ltnew
+ rlen
<= ltbnoa
+ ltlena
);
1122 ASSERT(ltnew
+ rlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1123 args
->agbno
= ltnew
;
1125 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
, ltlen
,
1126 ltnew
, rlen
, XFSA_FIXUP_BNO_OK
)))
1130 trace_xfs_alloc_near_greater(args
);
1132 trace_xfs_alloc_near_lesser(args
);
1134 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1135 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1139 trace_xfs_alloc_near_error(args
);
1140 if (cnt_cur
!= NULL
)
1141 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1142 if (bno_cur_lt
!= NULL
)
1143 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_ERROR
);
1144 if (bno_cur_gt
!= NULL
)
1145 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_ERROR
);
1150 * Allocate a variable extent anywhere in the allocation group agno.
1151 * Extent's length (returned in len) will be between minlen and maxlen,
1152 * and of the form k * prod + mod unless there's nothing that large.
1153 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1155 STATIC
int /* error */
1156 xfs_alloc_ag_vextent_size(
1157 xfs_alloc_arg_t
*args
) /* allocation argument structure */
1159 xfs_btree_cur_t
*bno_cur
; /* cursor for bno btree */
1160 xfs_btree_cur_t
*cnt_cur
; /* cursor for cnt btree */
1161 int error
; /* error result */
1162 xfs_agblock_t fbno
; /* start of found freespace */
1163 xfs_extlen_t flen
; /* length of found freespace */
1164 int i
; /* temp status variable */
1165 xfs_agblock_t rbno
; /* returned block number */
1166 xfs_extlen_t rlen
; /* length of returned extent */
1171 * Allocate and initialize a cursor for the by-size btree.
1173 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1174 args
->agno
, XFS_BTNUM_CNT
);
1178 * Look for an entry >= maxlen+alignment-1 blocks.
1180 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0,
1181 args
->maxlen
+ args
->alignment
- 1, &i
)))
1185 * If none or we have busy extents that we cannot allocate from, then
1186 * we have to settle for a smaller extent. In the case that there are
1187 * no large extents, this will return the last entry in the tree unless
1188 * the tree is empty. In the case that there are only busy large
1189 * extents, this will return the largest small extent unless there
1190 * are no smaller extents available.
1192 if (!i
|| forced
> 1) {
1193 error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
,
1197 if (i
== 0 || flen
== 0) {
1198 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1199 trace_xfs_alloc_size_noentry(args
);
1203 xfs_alloc_compute_aligned(args
, fbno
, flen
, &rbno
, &rlen
);
1206 * Search for a non-busy extent that is large enough.
1207 * If we are at low space, don't check, or if we fall of
1208 * the end of the btree, turn off the busy check and
1212 error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
, &i
);
1215 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1217 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1220 if (rlen
>= args
->maxlen
)
1223 error
= xfs_btree_increment(cnt_cur
, 0, &i
);
1228 * Our only valid extents must have been busy.
1229 * Make it unbusy by forcing the log out and
1230 * retrying. If we've been here before, forcing
1231 * the log isn't making the extents available,
1232 * which means they have probably been freed in
1233 * this transaction. In that case, we have to
1234 * give up on them and we'll attempt a minlen
1235 * allocation the next time around.
1237 xfs_btree_del_cursor(cnt_cur
,
1239 trace_xfs_alloc_size_busy(args
);
1241 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1248 * In the first case above, we got the last entry in the
1249 * by-size btree. Now we check to see if the space hits maxlen
1250 * once aligned; if not, we search left for something better.
1251 * This can't happen in the second case above.
1253 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1254 XFS_WANT_CORRUPTED_GOTO(rlen
== 0 ||
1255 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
), error0
);
1256 if (rlen
< args
->maxlen
) {
1257 xfs_agblock_t bestfbno
;
1258 xfs_extlen_t bestflen
;
1259 xfs_agblock_t bestrbno
;
1260 xfs_extlen_t bestrlen
;
1267 if ((error
= xfs_btree_decrement(cnt_cur
, 0, &i
)))
1271 if ((error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
,
1274 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1275 if (flen
< bestrlen
)
1277 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1279 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1280 XFS_WANT_CORRUPTED_GOTO(rlen
== 0 ||
1281 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
),
1283 if (rlen
> bestrlen
) {
1288 if (rlen
== args
->maxlen
)
1292 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, bestfbno
, bestflen
,
1295 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1301 args
->wasfromfl
= 0;
1303 * Fix up the length.
1306 if (rlen
< args
->minlen
) {
1308 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1309 trace_xfs_alloc_size_busy(args
);
1310 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1315 xfs_alloc_fix_len(args
);
1317 if (!xfs_alloc_fix_minleft(args
))
1320 XFS_WANT_CORRUPTED_GOTO(rlen
<= flen
, error0
);
1322 * Allocate and initialize a cursor for the by-block tree.
1324 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1325 args
->agno
, XFS_BTNUM_BNO
);
1326 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
,
1327 rbno
, rlen
, XFSA_FIXUP_CNT_OK
)))
1329 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1330 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1331 cnt_cur
= bno_cur
= NULL
;
1334 XFS_WANT_CORRUPTED_GOTO(
1335 args
->agbno
+ args
->len
<=
1336 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1338 trace_xfs_alloc_size_done(args
);
1342 trace_xfs_alloc_size_error(args
);
1344 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1346 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1350 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1351 trace_xfs_alloc_size_nominleft(args
);
1352 args
->agbno
= NULLAGBLOCK
;
1357 * Deal with the case where only small freespaces remain.
1358 * Either return the contents of the last freespace record,
1359 * or allocate space from the freelist if there is nothing in the tree.
1361 STATIC
int /* error */
1362 xfs_alloc_ag_vextent_small(
1363 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1364 xfs_btree_cur_t
*ccur
, /* by-size cursor */
1365 xfs_agblock_t
*fbnop
, /* result block number */
1366 xfs_extlen_t
*flenp
, /* result length */
1367 int *stat
) /* status: 0-freelist, 1-normal/none */
1374 if ((error
= xfs_btree_decrement(ccur
, 0, &i
)))
1377 if ((error
= xfs_alloc_get_rec(ccur
, &fbno
, &flen
, &i
)))
1379 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1382 * Nothing in the btree, try the freelist. Make sure
1383 * to respect minleft even when pulling from the
1386 else if (args
->minlen
== 1 && args
->alignment
== 1 && !args
->isfl
&&
1387 (be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_flcount
)
1389 error
= xfs_alloc_get_freelist(args
->tp
, args
->agbp
, &fbno
, 0);
1392 if (fbno
!= NULLAGBLOCK
) {
1393 xfs_extent_busy_reuse(args
->mp
, args
->agno
, fbno
, 1,
1396 if (args
->userdata
) {
1399 bp
= xfs_btree_get_bufs(args
->mp
, args
->tp
,
1400 args
->agno
, fbno
, 0);
1401 xfs_trans_binval(args
->tp
, bp
);
1405 XFS_WANT_CORRUPTED_GOTO(
1406 args
->agbno
+ args
->len
<=
1407 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1409 args
->wasfromfl
= 1;
1410 trace_xfs_alloc_small_freelist(args
);
1415 * Nothing in the freelist.
1421 * Can't allocate from the freelist for some reason.
1428 * Can't do the allocation, give up.
1430 if (flen
< args
->minlen
) {
1431 args
->agbno
= NULLAGBLOCK
;
1432 trace_xfs_alloc_small_notenough(args
);
1438 trace_xfs_alloc_small_done(args
);
1442 trace_xfs_alloc_small_error(args
);
1447 * Free the extent starting at agno/bno for length.
1449 STATIC
int /* error */
1451 xfs_trans_t
*tp
, /* transaction pointer */
1452 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
1453 xfs_agnumber_t agno
, /* allocation group number */
1454 xfs_agblock_t bno
, /* starting block number */
1455 xfs_extlen_t len
, /* length of extent */
1456 int isfl
) /* set if is freelist blocks - no sb acctg */
1458 xfs_btree_cur_t
*bno_cur
; /* cursor for by-block btree */
1459 xfs_btree_cur_t
*cnt_cur
; /* cursor for by-size btree */
1460 int error
; /* error return value */
1461 xfs_agblock_t gtbno
; /* start of right neighbor block */
1462 xfs_extlen_t gtlen
; /* length of right neighbor block */
1463 int haveleft
; /* have a left neighbor block */
1464 int haveright
; /* have a right neighbor block */
1465 int i
; /* temp, result code */
1466 xfs_agblock_t ltbno
; /* start of left neighbor block */
1467 xfs_extlen_t ltlen
; /* length of left neighbor block */
1468 xfs_mount_t
*mp
; /* mount point struct for filesystem */
1469 xfs_agblock_t nbno
; /* new starting block of freespace */
1470 xfs_extlen_t nlen
; /* new length of freespace */
1471 xfs_perag_t
*pag
; /* per allocation group data */
1475 * Allocate and initialize a cursor for the by-block btree.
1477 bno_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_BNO
);
1480 * Look for a neighboring block on the left (lower block numbers)
1481 * that is contiguous with this space.
1483 if ((error
= xfs_alloc_lookup_le(bno_cur
, bno
, len
, &haveleft
)))
1487 * There is a block to our left.
1489 if ((error
= xfs_alloc_get_rec(bno_cur
, <bno
, <len
, &i
)))
1491 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1493 * It's not contiguous, though.
1495 if (ltbno
+ ltlen
< bno
)
1499 * If this failure happens the request to free this
1500 * space was invalid, it's (partly) already free.
1503 XFS_WANT_CORRUPTED_GOTO(ltbno
+ ltlen
<= bno
, error0
);
1507 * Look for a neighboring block on the right (higher block numbers)
1508 * that is contiguous with this space.
1510 if ((error
= xfs_btree_increment(bno_cur
, 0, &haveright
)))
1514 * There is a block to our right.
1516 if ((error
= xfs_alloc_get_rec(bno_cur
, >bno
, >len
, &i
)))
1518 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1520 * It's not contiguous, though.
1522 if (bno
+ len
< gtbno
)
1526 * If this failure happens the request to free this
1527 * space was invalid, it's (partly) already free.
1530 XFS_WANT_CORRUPTED_GOTO(gtbno
>= bno
+ len
, error0
);
1534 * Now allocate and initialize a cursor for the by-size tree.
1536 cnt_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_CNT
);
1538 * Have both left and right contiguous neighbors.
1539 * Merge all three into a single free block.
1541 if (haveleft
&& haveright
) {
1543 * Delete the old by-size entry on the left.
1545 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1547 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1548 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1550 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1552 * Delete the old by-size entry on the right.
1554 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1556 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1557 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1559 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1561 * Delete the old by-block entry for the right block.
1563 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
1565 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1567 * Move the by-block cursor back to the left neighbor.
1569 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1571 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1574 * Check that this is the right record: delete didn't
1575 * mangle the cursor.
1578 xfs_agblock_t xxbno
;
1581 if ((error
= xfs_alloc_get_rec(bno_cur
, &xxbno
, &xxlen
,
1584 XFS_WANT_CORRUPTED_GOTO(
1585 i
== 1 && xxbno
== ltbno
&& xxlen
== ltlen
,
1590 * Update remaining by-block entry to the new, joined block.
1593 nlen
= len
+ ltlen
+ gtlen
;
1594 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1598 * Have only a left contiguous neighbor.
1599 * Merge it together with the new freespace.
1601 else if (haveleft
) {
1603 * Delete the old by-size entry on the left.
1605 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1607 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1608 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1610 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1612 * Back up the by-block cursor to the left neighbor, and
1613 * update its length.
1615 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1617 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1620 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1624 * Have only a right contiguous neighbor.
1625 * Merge it together with the new freespace.
1627 else if (haveright
) {
1629 * Delete the old by-size entry on the right.
1631 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1633 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1634 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1636 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1638 * Update the starting block and length of the right
1639 * neighbor in the by-block tree.
1643 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1647 * No contiguous neighbors.
1648 * Insert the new freespace into the by-block tree.
1653 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
1655 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1657 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1660 * In all cases we need to insert the new freespace in the by-size tree.
1662 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nbno
, nlen
, &i
)))
1664 XFS_WANT_CORRUPTED_GOTO(i
== 0, error0
);
1665 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
1667 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1668 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1672 * Update the freespace totals in the ag and superblock.
1674 pag
= xfs_perag_get(mp
, agno
);
1675 error
= xfs_alloc_update_counters(tp
, pag
, agbp
, len
);
1681 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_FDBLOCKS
, (long)len
);
1682 XFS_STATS_INC(xs_freex
);
1683 XFS_STATS_ADD(xs_freeb
, len
);
1685 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, haveleft
, haveright
);
1690 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, -1, -1);
1692 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1694 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1699 * Visible (exported) allocation/free functions.
1700 * Some of these are used just by xfs_alloc_btree.c and this file.
1704 * Compute and fill in value of m_ag_maxlevels.
1707 xfs_alloc_compute_maxlevels(
1708 xfs_mount_t
*mp
) /* file system mount structure */
1716 maxleafents
= (mp
->m_sb
.sb_agblocks
+ 1) / 2;
1717 minleafrecs
= mp
->m_alloc_mnr
[0];
1718 minnoderecs
= mp
->m_alloc_mnr
[1];
1719 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
1720 for (level
= 1; maxblocks
> 1; level
++)
1721 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
1722 mp
->m_ag_maxlevels
= level
;
1726 * Find the length of the longest extent in an AG.
1729 xfs_alloc_longest_free_extent(
1730 struct xfs_mount
*mp
,
1731 struct xfs_perag
*pag
)
1733 xfs_extlen_t need
, delta
= 0;
1735 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1736 if (need
> pag
->pagf_flcount
)
1737 delta
= need
- pag
->pagf_flcount
;
1739 if (pag
->pagf_longest
> delta
)
1740 return pag
->pagf_longest
- delta
;
1741 return pag
->pagf_flcount
> 0 || pag
->pagf_longest
> 0;
1745 * Decide whether to use this allocation group for this allocation.
1746 * If so, fix up the btree freelist's size.
1748 STATIC
int /* error */
1749 xfs_alloc_fix_freelist(
1750 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1751 int flags
) /* XFS_ALLOC_FLAG_... */
1753 xfs_buf_t
*agbp
; /* agf buffer pointer */
1754 xfs_agf_t
*agf
; /* a.g. freespace structure pointer */
1755 xfs_buf_t
*agflbp
;/* agfl buffer pointer */
1756 xfs_agblock_t bno
; /* freelist block */
1757 xfs_extlen_t delta
; /* new blocks needed in freelist */
1758 int error
; /* error result code */
1759 xfs_extlen_t longest
;/* longest extent in allocation group */
1760 xfs_mount_t
*mp
; /* file system mount point structure */
1761 xfs_extlen_t need
; /* total blocks needed in freelist */
1762 xfs_perag_t
*pag
; /* per-ag information structure */
1763 xfs_alloc_arg_t targs
; /* local allocation arguments */
1764 xfs_trans_t
*tp
; /* transaction pointer */
1770 if (!pag
->pagf_init
) {
1771 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1774 if (!pag
->pagf_init
) {
1775 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1776 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1784 * If this is a metadata preferred pag and we are user data
1785 * then try somewhere else if we are not being asked to
1786 * try harder at this point
1788 if (pag
->pagf_metadata
&& args
->userdata
&&
1789 (flags
& XFS_ALLOC_FLAG_TRYLOCK
)) {
1790 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1795 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1797 * If it looks like there isn't a long enough extent, or enough
1798 * total blocks, reject it.
1800 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1801 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
1802 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1804 ((int)(pag
->pagf_freeblks
+ pag
->pagf_flcount
-
1805 need
- args
->total
) < (int)args
->minleft
)) {
1807 xfs_trans_brelse(tp
, agbp
);
1814 * Get the a.g. freespace buffer.
1815 * Can fail if we're not blocking on locks, and it's held.
1818 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1822 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1823 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1829 * Figure out how many blocks we should have in the freelist.
1831 agf
= XFS_BUF_TO_AGF(agbp
);
1832 need
= XFS_MIN_FREELIST(agf
, mp
);
1834 * If there isn't enough total or single-extent, reject it.
1836 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1837 delta
= need
> be32_to_cpu(agf
->agf_flcount
) ?
1838 (need
- be32_to_cpu(agf
->agf_flcount
)) : 0;
1839 longest
= be32_to_cpu(agf
->agf_longest
);
1840 longest
= (longest
> delta
) ? (longest
- delta
) :
1841 (be32_to_cpu(agf
->agf_flcount
) > 0 || longest
> 0);
1842 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1844 ((int)(be32_to_cpu(agf
->agf_freeblks
) +
1845 be32_to_cpu(agf
->agf_flcount
) - need
- args
->total
) <
1846 (int)args
->minleft
)) {
1847 xfs_trans_brelse(tp
, agbp
);
1853 * Make the freelist shorter if it's too long.
1855 while (be32_to_cpu(agf
->agf_flcount
) > need
) {
1858 error
= xfs_alloc_get_freelist(tp
, agbp
, &bno
, 0);
1861 if ((error
= xfs_free_ag_extent(tp
, agbp
, args
->agno
, bno
, 1, 1)))
1863 bp
= xfs_btree_get_bufs(mp
, tp
, args
->agno
, bno
, 0);
1864 xfs_trans_binval(tp
, bp
);
1867 * Initialize the args structure.
1872 targs
.agno
= args
->agno
;
1873 targs
.mod
= targs
.minleft
= targs
.wasdel
= targs
.userdata
=
1874 targs
.minalignslop
= 0;
1875 targs
.alignment
= targs
.minlen
= targs
.prod
= targs
.isfl
= 1;
1876 targs
.type
= XFS_ALLOCTYPE_THIS_AG
;
1878 if ((error
= xfs_alloc_read_agfl(mp
, tp
, targs
.agno
, &agflbp
)))
1881 * Make the freelist longer if it's too short.
1883 while (be32_to_cpu(agf
->agf_flcount
) < need
) {
1885 targs
.maxlen
= need
- be32_to_cpu(agf
->agf_flcount
);
1887 * Allocate as many blocks as possible at once.
1889 if ((error
= xfs_alloc_ag_vextent(&targs
))) {
1890 xfs_trans_brelse(tp
, agflbp
);
1894 * Stop if we run out. Won't happen if callers are obeying
1895 * the restrictions correctly. Can happen for free calls
1896 * on a completely full ag.
1898 if (targs
.agbno
== NULLAGBLOCK
) {
1899 if (flags
& XFS_ALLOC_FLAG_FREEING
)
1901 xfs_trans_brelse(tp
, agflbp
);
1906 * Put each allocated block on the list.
1908 for (bno
= targs
.agbno
; bno
< targs
.agbno
+ targs
.len
; bno
++) {
1909 error
= xfs_alloc_put_freelist(tp
, agbp
,
1915 xfs_trans_brelse(tp
, agflbp
);
1921 * Get a block from the freelist.
1922 * Returns with the buffer for the block gotten.
1925 xfs_alloc_get_freelist(
1926 xfs_trans_t
*tp
, /* transaction pointer */
1927 xfs_buf_t
*agbp
, /* buffer containing the agf structure */
1928 xfs_agblock_t
*bnop
, /* block address retrieved from freelist */
1929 int btreeblk
) /* destination is a AGF btree */
1931 xfs_agf_t
*agf
; /* a.g. freespace structure */
1932 xfs_agfl_t
*agfl
; /* a.g. freelist structure */
1933 xfs_buf_t
*agflbp
;/* buffer for a.g. freelist structure */
1934 xfs_agblock_t bno
; /* block number returned */
1937 xfs_mount_t
*mp
; /* mount structure */
1938 xfs_perag_t
*pag
; /* per allocation group data */
1940 agf
= XFS_BUF_TO_AGF(agbp
);
1942 * Freelist is empty, give up.
1944 if (!agf
->agf_flcount
) {
1945 *bnop
= NULLAGBLOCK
;
1949 * Read the array of free blocks.
1952 if ((error
= xfs_alloc_read_agfl(mp
, tp
,
1953 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
1955 agfl
= XFS_BUF_TO_AGFL(agflbp
);
1957 * Get the block number and update the data structures.
1959 bno
= be32_to_cpu(agfl
->agfl_bno
[be32_to_cpu(agf
->agf_flfirst
)]);
1960 be32_add_cpu(&agf
->agf_flfirst
, 1);
1961 xfs_trans_brelse(tp
, agflbp
);
1962 if (be32_to_cpu(agf
->agf_flfirst
) == XFS_AGFL_SIZE(mp
))
1963 agf
->agf_flfirst
= 0;
1965 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
1966 be32_add_cpu(&agf
->agf_flcount
, -1);
1967 xfs_trans_agflist_delta(tp
, -1);
1968 pag
->pagf_flcount
--;
1971 logflags
= XFS_AGF_FLFIRST
| XFS_AGF_FLCOUNT
;
1973 be32_add_cpu(&agf
->agf_btreeblks
, 1);
1974 pag
->pagf_btreeblks
++;
1975 logflags
|= XFS_AGF_BTREEBLKS
;
1978 xfs_alloc_log_agf(tp
, agbp
, logflags
);
1985 * Log the given fields from the agf structure.
1989 xfs_trans_t
*tp
, /* transaction pointer */
1990 xfs_buf_t
*bp
, /* buffer for a.g. freelist header */
1991 int fields
) /* mask of fields to be logged (XFS_AGF_...) */
1993 int first
; /* first byte offset */
1994 int last
; /* last byte offset */
1995 static const short offsets
[] = {
1996 offsetof(xfs_agf_t
, agf_magicnum
),
1997 offsetof(xfs_agf_t
, agf_versionnum
),
1998 offsetof(xfs_agf_t
, agf_seqno
),
1999 offsetof(xfs_agf_t
, agf_length
),
2000 offsetof(xfs_agf_t
, agf_roots
[0]),
2001 offsetof(xfs_agf_t
, agf_levels
[0]),
2002 offsetof(xfs_agf_t
, agf_flfirst
),
2003 offsetof(xfs_agf_t
, agf_fllast
),
2004 offsetof(xfs_agf_t
, agf_flcount
),
2005 offsetof(xfs_agf_t
, agf_freeblks
),
2006 offsetof(xfs_agf_t
, agf_longest
),
2007 offsetof(xfs_agf_t
, agf_btreeblks
),
2011 trace_xfs_agf(tp
->t_mountp
, XFS_BUF_TO_AGF(bp
), fields
, _RET_IP_
);
2013 xfs_btree_offsets(fields
, offsets
, XFS_AGF_NUM_BITS
, &first
, &last
);
2014 xfs_trans_log_buf(tp
, bp
, (uint
)first
, (uint
)last
);
2018 * Interface for inode allocation to force the pag data to be initialized.
2021 xfs_alloc_pagf_init(
2022 xfs_mount_t
*mp
, /* file system mount structure */
2023 xfs_trans_t
*tp
, /* transaction pointer */
2024 xfs_agnumber_t agno
, /* allocation group number */
2025 int flags
) /* XFS_ALLOC_FLAGS_... */
2030 if ((error
= xfs_alloc_read_agf(mp
, tp
, agno
, flags
, &bp
)))
2033 xfs_trans_brelse(tp
, bp
);
2038 * Put the block on the freelist for the allocation group.
2041 xfs_alloc_put_freelist(
2042 xfs_trans_t
*tp
, /* transaction pointer */
2043 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
2044 xfs_buf_t
*agflbp
,/* buffer for a.g. free block array */
2045 xfs_agblock_t bno
, /* block being freed */
2046 int btreeblk
) /* block came from a AGF btree */
2048 xfs_agf_t
*agf
; /* a.g. freespace structure */
2049 xfs_agfl_t
*agfl
; /* a.g. free block array */
2050 __be32
*blockp
;/* pointer to array entry */
2053 xfs_mount_t
*mp
; /* mount structure */
2054 xfs_perag_t
*pag
; /* per allocation group data */
2056 agf
= XFS_BUF_TO_AGF(agbp
);
2059 if (!agflbp
&& (error
= xfs_alloc_read_agfl(mp
, tp
,
2060 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
2062 agfl
= XFS_BUF_TO_AGFL(agflbp
);
2063 be32_add_cpu(&agf
->agf_fllast
, 1);
2064 if (be32_to_cpu(agf
->agf_fllast
) == XFS_AGFL_SIZE(mp
))
2065 agf
->agf_fllast
= 0;
2067 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2068 be32_add_cpu(&agf
->agf_flcount
, 1);
2069 xfs_trans_agflist_delta(tp
, 1);
2070 pag
->pagf_flcount
++;
2072 logflags
= XFS_AGF_FLLAST
| XFS_AGF_FLCOUNT
;
2074 be32_add_cpu(&agf
->agf_btreeblks
, -1);
2075 pag
->pagf_btreeblks
--;
2076 logflags
|= XFS_AGF_BTREEBLKS
;
2080 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2082 ASSERT(be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
));
2083 blockp
= &agfl
->agfl_bno
[be32_to_cpu(agf
->agf_fllast
)];
2084 *blockp
= cpu_to_be32(bno
);
2085 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2086 xfs_trans_log_buf(tp
, agflbp
,
2087 (int)((xfs_caddr_t
)blockp
- (xfs_caddr_t
)agfl
),
2088 (int)((xfs_caddr_t
)blockp
- (xfs_caddr_t
)agfl
+
2089 sizeof(xfs_agblock_t
) - 1));
2094 * Read in the allocation group header (free/alloc section).
2098 struct xfs_mount
*mp
, /* mount point structure */
2099 struct xfs_trans
*tp
, /* transaction pointer */
2100 xfs_agnumber_t agno
, /* allocation group number */
2101 int flags
, /* XFS_BUF_ */
2102 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2104 struct xfs_agf
*agf
; /* ag freelist header */
2105 int agf_ok
; /* set if agf is consistent */
2108 ASSERT(agno
!= NULLAGNUMBER
);
2109 error
= xfs_trans_read_buf(
2110 mp
, tp
, mp
->m_ddev_targp
,
2111 XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
)),
2112 XFS_FSS_TO_BB(mp
, 1), flags
, bpp
);
2118 ASSERT(!(*bpp
)->b_error
);
2119 agf
= XFS_BUF_TO_AGF(*bpp
);
2122 * Validate the magic number of the agf block.
2125 agf
->agf_magicnum
== cpu_to_be32(XFS_AGF_MAGIC
) &&
2126 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf
->agf_versionnum
)) &&
2127 be32_to_cpu(agf
->agf_freeblks
) <= be32_to_cpu(agf
->agf_length
) &&
2128 be32_to_cpu(agf
->agf_flfirst
) < XFS_AGFL_SIZE(mp
) &&
2129 be32_to_cpu(agf
->agf_fllast
) < XFS_AGFL_SIZE(mp
) &&
2130 be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
) &&
2131 be32_to_cpu(agf
->agf_seqno
) == agno
;
2132 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
))
2133 agf_ok
= agf_ok
&& be32_to_cpu(agf
->agf_btreeblks
) <=
2134 be32_to_cpu(agf
->agf_length
);
2135 if (unlikely(XFS_TEST_ERROR(!agf_ok
, mp
, XFS_ERRTAG_ALLOC_READ_AGF
,
2136 XFS_RANDOM_ALLOC_READ_AGF
))) {
2137 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2138 XFS_ERRLEVEL_LOW
, mp
, agf
);
2139 xfs_trans_brelse(tp
, *bpp
);
2140 return XFS_ERROR(EFSCORRUPTED
);
2142 xfs_buf_set_ref(*bpp
, XFS_AGF_REF
);
2147 * Read in the allocation group header (free/alloc section).
2151 struct xfs_mount
*mp
, /* mount point structure */
2152 struct xfs_trans
*tp
, /* transaction pointer */
2153 xfs_agnumber_t agno
, /* allocation group number */
2154 int flags
, /* XFS_ALLOC_FLAG_... */
2155 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2157 struct xfs_agf
*agf
; /* ag freelist header */
2158 struct xfs_perag
*pag
; /* per allocation group data */
2161 ASSERT(agno
!= NULLAGNUMBER
);
2163 error
= xfs_read_agf(mp
, tp
, agno
,
2164 (flags
& XFS_ALLOC_FLAG_TRYLOCK
) ? XBF_TRYLOCK
: 0,
2170 ASSERT(!(*bpp
)->b_error
);
2172 agf
= XFS_BUF_TO_AGF(*bpp
);
2173 pag
= xfs_perag_get(mp
, agno
);
2174 if (!pag
->pagf_init
) {
2175 pag
->pagf_freeblks
= be32_to_cpu(agf
->agf_freeblks
);
2176 pag
->pagf_btreeblks
= be32_to_cpu(agf
->agf_btreeblks
);
2177 pag
->pagf_flcount
= be32_to_cpu(agf
->agf_flcount
);
2178 pag
->pagf_longest
= be32_to_cpu(agf
->agf_longest
);
2179 pag
->pagf_levels
[XFS_BTNUM_BNOi
] =
2180 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]);
2181 pag
->pagf_levels
[XFS_BTNUM_CNTi
] =
2182 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]);
2183 spin_lock_init(&pag
->pagb_lock
);
2184 pag
->pagb_count
= 0;
2185 pag
->pagb_tree
= RB_ROOT
;
2189 else if (!XFS_FORCED_SHUTDOWN(mp
)) {
2190 ASSERT(pag
->pagf_freeblks
== be32_to_cpu(agf
->agf_freeblks
));
2191 ASSERT(pag
->pagf_btreeblks
== be32_to_cpu(agf
->agf_btreeblks
));
2192 ASSERT(pag
->pagf_flcount
== be32_to_cpu(agf
->agf_flcount
));
2193 ASSERT(pag
->pagf_longest
== be32_to_cpu(agf
->agf_longest
));
2194 ASSERT(pag
->pagf_levels
[XFS_BTNUM_BNOi
] ==
2195 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]));
2196 ASSERT(pag
->pagf_levels
[XFS_BTNUM_CNTi
] ==
2197 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]));
2205 * Allocate an extent (variable-size).
2206 * Depending on the allocation type, we either look in a single allocation
2207 * group or loop over the allocation groups to find the result.
2210 __xfs_alloc_vextent(
2211 xfs_alloc_arg_t
*args
) /* allocation argument structure */
2213 xfs_agblock_t agsize
; /* allocation group size */
2215 int flags
; /* XFS_ALLOC_FLAG_... locking flags */
2216 xfs_extlen_t minleft
;/* minimum left value, temp copy */
2217 xfs_mount_t
*mp
; /* mount structure pointer */
2218 xfs_agnumber_t sagno
; /* starting allocation group number */
2219 xfs_alloctype_t type
; /* input allocation type */
2222 xfs_agnumber_t rotorstep
= xfs_rotorstep
; /* inode32 agf stepper */
2225 type
= args
->otype
= args
->type
;
2226 args
->agbno
= NULLAGBLOCK
;
2228 * Just fix this up, for the case where the last a.g. is shorter
2229 * (or there's only one a.g.) and the caller couldn't easily figure
2230 * that out (xfs_bmap_alloc).
2232 agsize
= mp
->m_sb
.sb_agblocks
;
2233 if (args
->maxlen
> agsize
)
2234 args
->maxlen
= agsize
;
2235 if (args
->alignment
== 0)
2236 args
->alignment
= 1;
2237 ASSERT(XFS_FSB_TO_AGNO(mp
, args
->fsbno
) < mp
->m_sb
.sb_agcount
);
2238 ASSERT(XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) < agsize
);
2239 ASSERT(args
->minlen
<= args
->maxlen
);
2240 ASSERT(args
->minlen
<= agsize
);
2241 ASSERT(args
->mod
< args
->prod
);
2242 if (XFS_FSB_TO_AGNO(mp
, args
->fsbno
) >= mp
->m_sb
.sb_agcount
||
2243 XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) >= agsize
||
2244 args
->minlen
> args
->maxlen
|| args
->minlen
> agsize
||
2245 args
->mod
>= args
->prod
) {
2246 args
->fsbno
= NULLFSBLOCK
;
2247 trace_xfs_alloc_vextent_badargs(args
);
2250 minleft
= args
->minleft
;
2253 case XFS_ALLOCTYPE_THIS_AG
:
2254 case XFS_ALLOCTYPE_NEAR_BNO
:
2255 case XFS_ALLOCTYPE_THIS_BNO
:
2257 * These three force us into a single a.g.
2259 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2260 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2262 error
= xfs_alloc_fix_freelist(args
, 0);
2263 args
->minleft
= minleft
;
2265 trace_xfs_alloc_vextent_nofix(args
);
2269 trace_xfs_alloc_vextent_noagbp(args
);
2272 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2273 if ((error
= xfs_alloc_ag_vextent(args
)))
2276 case XFS_ALLOCTYPE_START_BNO
:
2278 * Try near allocation first, then anywhere-in-ag after
2279 * the first a.g. fails.
2281 if ((args
->userdata
== XFS_ALLOC_INITIAL_USER_DATA
) &&
2282 (mp
->m_flags
& XFS_MOUNT_32BITINODES
)) {
2283 args
->fsbno
= XFS_AGB_TO_FSB(mp
,
2284 ((mp
->m_agfrotor
/ rotorstep
) %
2285 mp
->m_sb
.sb_agcount
), 0);
2288 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2289 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2291 case XFS_ALLOCTYPE_ANY_AG
:
2292 case XFS_ALLOCTYPE_START_AG
:
2293 case XFS_ALLOCTYPE_FIRST_AG
:
2295 * Rotate through the allocation groups looking for a winner.
2297 if (type
== XFS_ALLOCTYPE_ANY_AG
) {
2299 * Start with the last place we left off.
2301 args
->agno
= sagno
= (mp
->m_agfrotor
/ rotorstep
) %
2302 mp
->m_sb
.sb_agcount
;
2303 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2304 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2305 } else if (type
== XFS_ALLOCTYPE_FIRST_AG
) {
2307 * Start with allocation group given by bno.
2309 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2310 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2314 if (type
== XFS_ALLOCTYPE_START_AG
)
2315 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2317 * Start with the given allocation group.
2319 args
->agno
= sagno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2320 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2323 * Loop over allocation groups twice; first time with
2324 * trylock set, second time without.
2327 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2328 if (no_min
) args
->minleft
= 0;
2329 error
= xfs_alloc_fix_freelist(args
, flags
);
2330 args
->minleft
= minleft
;
2332 trace_xfs_alloc_vextent_nofix(args
);
2336 * If we get a buffer back then the allocation will fly.
2339 if ((error
= xfs_alloc_ag_vextent(args
)))
2344 trace_xfs_alloc_vextent_loopfailed(args
);
2347 * Didn't work, figure out the next iteration.
2349 if (args
->agno
== sagno
&&
2350 type
== XFS_ALLOCTYPE_START_BNO
)
2351 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2353 * For the first allocation, we can try any AG to get
2354 * space. However, if we already have allocated a
2355 * block, we don't want to try AGs whose number is below
2356 * sagno. Otherwise, we may end up with out-of-order
2357 * locking of AGF, which might cause deadlock.
2359 if (++(args
->agno
) == mp
->m_sb
.sb_agcount
) {
2360 if (args
->firstblock
!= NULLFSBLOCK
)
2366 * Reached the starting a.g., must either be done
2367 * or switch to non-trylock mode.
2369 if (args
->agno
== sagno
) {
2371 args
->agbno
= NULLAGBLOCK
;
2372 trace_xfs_alloc_vextent_allfailed(args
);
2379 if (type
== XFS_ALLOCTYPE_START_BNO
) {
2380 args
->agbno
= XFS_FSB_TO_AGBNO(mp
,
2382 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2386 xfs_perag_put(args
->pag
);
2388 if (bump_rotor
|| (type
== XFS_ALLOCTYPE_ANY_AG
)) {
2389 if (args
->agno
== sagno
)
2390 mp
->m_agfrotor
= (mp
->m_agfrotor
+ 1) %
2391 (mp
->m_sb
.sb_agcount
* rotorstep
);
2393 mp
->m_agfrotor
= (args
->agno
* rotorstep
+ 1) %
2394 (mp
->m_sb
.sb_agcount
* rotorstep
);
2401 if (args
->agbno
== NULLAGBLOCK
)
2402 args
->fsbno
= NULLFSBLOCK
;
2404 args
->fsbno
= XFS_AGB_TO_FSB(mp
, args
->agno
, args
->agbno
);
2406 ASSERT(args
->len
>= args
->minlen
);
2407 ASSERT(args
->len
<= args
->maxlen
);
2408 ASSERT(args
->agbno
% args
->alignment
== 0);
2409 XFS_AG_CHECK_DADDR(mp
, XFS_FSB_TO_DADDR(mp
, args
->fsbno
),
2413 xfs_perag_put(args
->pag
);
2416 xfs_perag_put(args
->pag
);
2421 xfs_alloc_vextent_worker(
2422 struct work_struct
*work
)
2424 struct xfs_alloc_arg
*args
= container_of(work
,
2425 struct xfs_alloc_arg
, work
);
2426 unsigned long pflags
;
2428 /* we are in a transaction context here */
2429 current_set_flags_nested(&pflags
, PF_FSTRANS
);
2431 args
->result
= __xfs_alloc_vextent(args
);
2432 complete(args
->done
);
2434 current_restore_flags_nested(&pflags
, PF_FSTRANS
);
2438 * Data allocation requests often come in with little stack to work on. Push
2439 * them off to a worker thread so there is lots of stack to use. Metadata
2440 * requests, OTOH, are generally from low stack usage paths, so avoid the
2441 * context switch overhead here.
2445 struct xfs_alloc_arg
*args
)
2447 DECLARE_COMPLETION_ONSTACK(done
);
2449 if (!args
->userdata
)
2450 return __xfs_alloc_vextent(args
);
2454 INIT_WORK_ONSTACK(&args
->work
, xfs_alloc_vextent_worker
);
2455 queue_work(xfs_alloc_wq
, &args
->work
);
2456 wait_for_completion(&done
);
2457 return args
->result
;
2462 * Just break up the extent address and hand off to xfs_free_ag_extent
2463 * after fixing up the freelist.
2467 xfs_trans_t
*tp
, /* transaction pointer */
2468 xfs_fsblock_t bno
, /* starting block number of extent */
2469 xfs_extlen_t len
) /* length of extent */
2471 xfs_alloc_arg_t args
;
2475 memset(&args
, 0, sizeof(xfs_alloc_arg_t
));
2477 args
.mp
= tp
->t_mountp
;
2480 * validate that the block number is legal - the enables us to detect
2481 * and handle a silent filesystem corruption rather than crashing.
2483 args
.agno
= XFS_FSB_TO_AGNO(args
.mp
, bno
);
2484 if (args
.agno
>= args
.mp
->m_sb
.sb_agcount
)
2485 return EFSCORRUPTED
;
2487 args
.agbno
= XFS_FSB_TO_AGBNO(args
.mp
, bno
);
2488 if (args
.agbno
>= args
.mp
->m_sb
.sb_agblocks
)
2489 return EFSCORRUPTED
;
2491 args
.pag
= xfs_perag_get(args
.mp
, args
.agno
);
2494 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
2498 /* validate the extent size is legal now we have the agf locked */
2499 if (args
.agbno
+ len
>
2500 be32_to_cpu(XFS_BUF_TO_AGF(args
.agbp
)->agf_length
)) {
2501 error
= EFSCORRUPTED
;
2505 error
= xfs_free_ag_extent(tp
, args
.agbp
, args
.agno
, args
.agbno
, len
, 0);
2507 xfs_extent_busy_insert(tp
, args
.agno
, args
.agbno
, len
, 0);
2509 xfs_perag_put(args
.pag
);