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
18 #include "libxfs_priv.h"
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_shared.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_alloc.h"
31 #include "xfs_cksum.h"
32 #include "xfs_trace.h"
33 #include "xfs_trans.h"
35 struct workqueue_struct
*xfs_alloc_wq
;
37 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
39 #define XFSA_FIXUP_BNO_OK 1
40 #define XFSA_FIXUP_CNT_OK 2
42 STATIC
int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t
*);
43 STATIC
int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t
*);
44 STATIC
int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t
*);
45 STATIC
int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t
*,
46 xfs_btree_cur_t
*, xfs_agblock_t
*, xfs_extlen_t
*, int *);
49 * Lookup the record equal to [bno, len] in the btree given by cur.
51 STATIC
int /* error */
53 struct xfs_btree_cur
*cur
, /* btree cursor */
54 xfs_agblock_t bno
, /* starting block of extent */
55 xfs_extlen_t len
, /* length of extent */
56 int *stat
) /* success/failure */
58 cur
->bc_rec
.a
.ar_startblock
= bno
;
59 cur
->bc_rec
.a
.ar_blockcount
= len
;
60 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
64 * Lookup the first record greater than or equal to [bno, len]
65 * in the btree given by cur.
69 struct xfs_btree_cur
*cur
, /* btree cursor */
70 xfs_agblock_t bno
, /* starting block of extent */
71 xfs_extlen_t len
, /* length of extent */
72 int *stat
) /* success/failure */
74 cur
->bc_rec
.a
.ar_startblock
= bno
;
75 cur
->bc_rec
.a
.ar_blockcount
= len
;
76 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
80 * Lookup the first record less than or equal to [bno, len]
81 * in the btree given by cur.
85 struct xfs_btree_cur
*cur
, /* btree cursor */
86 xfs_agblock_t bno
, /* starting block of extent */
87 xfs_extlen_t len
, /* length of extent */
88 int *stat
) /* success/failure */
90 cur
->bc_rec
.a
.ar_startblock
= bno
;
91 cur
->bc_rec
.a
.ar_blockcount
= len
;
92 return xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, stat
);
96 * Update the record referred to by cur to the value given
98 * This either works (return 0) or gets an EFSCORRUPTED error.
100 STATIC
int /* error */
102 struct xfs_btree_cur
*cur
, /* btree cursor */
103 xfs_agblock_t bno
, /* starting block of extent */
104 xfs_extlen_t len
) /* length of extent */
106 union xfs_btree_rec rec
;
108 rec
.alloc
.ar_startblock
= cpu_to_be32(bno
);
109 rec
.alloc
.ar_blockcount
= cpu_to_be32(len
);
110 return xfs_btree_update(cur
, &rec
);
114 * Get the data from the pointed-to record.
118 struct xfs_btree_cur
*cur
, /* btree cursor */
119 xfs_agblock_t
*bno
, /* output: starting block of extent */
120 xfs_extlen_t
*len
, /* output: length of extent */
121 int *stat
) /* output: success/failure */
123 union xfs_btree_rec
*rec
;
126 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
127 if (!error
&& *stat
== 1) {
128 *bno
= be32_to_cpu(rec
->alloc
.ar_startblock
);
129 *len
= be32_to_cpu(rec
->alloc
.ar_blockcount
);
135 * Compute aligned version of the found extent.
136 * Takes alignment and min length into account.
139 xfs_alloc_compute_aligned(
140 xfs_alloc_arg_t
*args
, /* allocation argument structure */
141 xfs_agblock_t foundbno
, /* starting block in found extent */
142 xfs_extlen_t foundlen
, /* length in found extent */
143 xfs_agblock_t
*resbno
, /* result block number */
144 xfs_extlen_t
*reslen
) /* result length */
149 /* Trim busy sections out of found extent */
150 xfs_extent_busy_trim(args
, foundbno
, foundlen
, &bno
, &len
);
152 if (args
->alignment
> 1 && len
>= args
->minlen
) {
153 xfs_agblock_t aligned_bno
= roundup(bno
, args
->alignment
);
154 xfs_extlen_t diff
= aligned_bno
- bno
;
156 *resbno
= aligned_bno
;
157 *reslen
= diff
>= len
? 0 : len
- diff
;
165 * Compute best start block and diff for "near" allocations.
166 * freelen >= wantlen already checked by caller.
168 STATIC xfs_extlen_t
/* difference value (absolute) */
169 xfs_alloc_compute_diff(
170 xfs_agblock_t wantbno
, /* target starting block */
171 xfs_extlen_t wantlen
, /* target length */
172 xfs_extlen_t alignment
, /* target alignment */
173 char userdata
, /* are we allocating data? */
174 xfs_agblock_t freebno
, /* freespace's starting block */
175 xfs_extlen_t freelen
, /* freespace's length */
176 xfs_agblock_t
*newbnop
) /* result: best start block from free */
178 xfs_agblock_t freeend
; /* end of freespace extent */
179 xfs_agblock_t newbno1
; /* return block number */
180 xfs_agblock_t newbno2
; /* other new block number */
181 xfs_extlen_t newlen1
=0; /* length with newbno1 */
182 xfs_extlen_t newlen2
=0; /* length with newbno2 */
183 xfs_agblock_t wantend
; /* end of target extent */
185 ASSERT(freelen
>= wantlen
);
186 freeend
= freebno
+ freelen
;
187 wantend
= wantbno
+ wantlen
;
189 * We want to allocate from the start of a free extent if it is past
190 * the desired block or if we are allocating user data and the free
191 * extent is before desired block. The second case is there to allow
192 * for contiguous allocation from the remaining free space if the file
193 * grows in the short term.
195 if (freebno
>= wantbno
|| (userdata
&& freeend
< wantend
)) {
196 if ((newbno1
= roundup(freebno
, alignment
)) >= freeend
)
197 newbno1
= NULLAGBLOCK
;
198 } else if (freeend
>= wantend
&& alignment
> 1) {
199 newbno1
= roundup(wantbno
, alignment
);
200 newbno2
= newbno1
- alignment
;
201 if (newbno1
>= freeend
)
202 newbno1
= NULLAGBLOCK
;
204 newlen1
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno1
);
205 if (newbno2
< freebno
)
206 newbno2
= NULLAGBLOCK
;
208 newlen2
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno2
);
209 if (newbno1
!= NULLAGBLOCK
&& newbno2
!= NULLAGBLOCK
) {
210 if (newlen1
< newlen2
||
211 (newlen1
== newlen2
&&
212 XFS_ABSDIFF(newbno1
, wantbno
) >
213 XFS_ABSDIFF(newbno2
, wantbno
)))
215 } else if (newbno2
!= NULLAGBLOCK
)
217 } else if (freeend
>= wantend
) {
219 } else if (alignment
> 1) {
220 newbno1
= roundup(freeend
- wantlen
, alignment
);
221 if (newbno1
> freeend
- wantlen
&&
222 newbno1
- alignment
>= freebno
)
223 newbno1
-= alignment
;
224 else if (newbno1
>= freeend
)
225 newbno1
= NULLAGBLOCK
;
227 newbno1
= freeend
- wantlen
;
229 return newbno1
== NULLAGBLOCK
? 0 : XFS_ABSDIFF(newbno1
, wantbno
);
233 * Fix up the length, based on mod and prod.
234 * len should be k * prod + mod for some k.
235 * If len is too small it is returned unchanged.
236 * If len hits maxlen it is left alone.
240 xfs_alloc_arg_t
*args
) /* allocation argument structure */
245 ASSERT(args
->mod
< args
->prod
);
247 ASSERT(rlen
>= args
->minlen
);
248 ASSERT(rlen
<= args
->maxlen
);
249 if (args
->prod
<= 1 || rlen
< args
->mod
|| rlen
== args
->maxlen
||
250 (args
->mod
== 0 && rlen
< args
->prod
))
252 k
= rlen
% args
->prod
;
256 rlen
= rlen
- (k
- args
->mod
);
258 rlen
= rlen
- args
->prod
+ (args
->mod
- k
);
259 /* casts to (int) catch length underflows */
260 if ((int)rlen
< (int)args
->minlen
)
262 ASSERT(rlen
>= args
->minlen
&& rlen
<= args
->maxlen
);
263 ASSERT(rlen
% args
->prod
== args
->mod
);
268 * Fix up length if there is too little space left in the a.g.
269 * Return 1 if ok, 0 if too little, should give up.
272 xfs_alloc_fix_minleft(
273 xfs_alloc_arg_t
*args
) /* allocation argument structure */
275 xfs_agf_t
*agf
; /* a.g. freelist header */
276 int diff
; /* free space difference */
278 if (args
->minleft
== 0)
280 agf
= XFS_BUF_TO_AGF(args
->agbp
);
281 diff
= be32_to_cpu(agf
->agf_freeblks
)
282 - args
->len
- args
->minleft
;
285 args
->len
+= diff
; /* shrink the allocated space */
286 /* casts to (int) catch length underflows */
287 if ((int)args
->len
>= (int)args
->minlen
)
289 args
->agbno
= NULLAGBLOCK
;
294 * Update the two btrees, logically removing from freespace the extent
295 * starting at rbno, rlen blocks. The extent is contained within the
296 * actual (current) free extent fbno for flen blocks.
297 * Flags are passed in indicating whether the cursors are set to the
300 STATIC
int /* error code */
301 xfs_alloc_fixup_trees(
302 xfs_btree_cur_t
*cnt_cur
, /* cursor for by-size btree */
303 xfs_btree_cur_t
*bno_cur
, /* cursor for by-block btree */
304 xfs_agblock_t fbno
, /* starting block of free extent */
305 xfs_extlen_t flen
, /* length of free extent */
306 xfs_agblock_t rbno
, /* starting block of returned extent */
307 xfs_extlen_t rlen
, /* length of returned extent */
308 int flags
) /* flags, XFSA_FIXUP_... */
310 int error
; /* error code */
311 int i
; /* operation results */
312 xfs_agblock_t nfbno1
; /* first new free startblock */
313 xfs_agblock_t nfbno2
; /* second new free startblock */
314 xfs_extlen_t nflen1
=0; /* first new free length */
315 xfs_extlen_t nflen2
=0; /* second new free length */
316 struct xfs_mount
*mp
;
321 * Look up the record in the by-size tree if necessary.
323 if (flags
& XFSA_FIXUP_CNT_OK
) {
325 if ((error
= xfs_alloc_get_rec(cnt_cur
, &nfbno1
, &nflen1
, &i
)))
327 XFS_WANT_CORRUPTED_RETURN(mp
,
328 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
331 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, fbno
, flen
, &i
)))
333 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
336 * Look up the record in the by-block tree if necessary.
338 if (flags
& XFSA_FIXUP_BNO_OK
) {
340 if ((error
= xfs_alloc_get_rec(bno_cur
, &nfbno1
, &nflen1
, &i
)))
342 XFS_WANT_CORRUPTED_RETURN(mp
,
343 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
346 if ((error
= xfs_alloc_lookup_eq(bno_cur
, fbno
, flen
, &i
)))
348 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
352 if (bno_cur
->bc_nlevels
== 1 && cnt_cur
->bc_nlevels
== 1) {
353 struct xfs_btree_block
*bnoblock
;
354 struct xfs_btree_block
*cntblock
;
356 bnoblock
= XFS_BUF_TO_BLOCK(bno_cur
->bc_bufs
[0]);
357 cntblock
= XFS_BUF_TO_BLOCK(cnt_cur
->bc_bufs
[0]);
359 XFS_WANT_CORRUPTED_RETURN(mp
,
360 bnoblock
->bb_numrecs
== cntblock
->bb_numrecs
);
365 * Deal with all four cases: the allocated record is contained
366 * within the freespace record, so we can have new freespace
367 * at either (or both) end, or no freespace remaining.
369 if (rbno
== fbno
&& rlen
== flen
)
370 nfbno1
= nfbno2
= NULLAGBLOCK
;
371 else if (rbno
== fbno
) {
372 nfbno1
= rbno
+ rlen
;
373 nflen1
= flen
- rlen
;
374 nfbno2
= NULLAGBLOCK
;
375 } else if (rbno
+ rlen
== fbno
+ flen
) {
377 nflen1
= flen
- rlen
;
378 nfbno2
= NULLAGBLOCK
;
381 nflen1
= rbno
- fbno
;
382 nfbno2
= rbno
+ rlen
;
383 nflen2
= (fbno
+ flen
) - nfbno2
;
386 * Delete the entry from the by-size btree.
388 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
390 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
392 * Add new by-size btree entry(s).
394 if (nfbno1
!= NULLAGBLOCK
) {
395 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno1
, nflen1
, &i
)))
397 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
398 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
400 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
402 if (nfbno2
!= NULLAGBLOCK
) {
403 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno2
, nflen2
, &i
)))
405 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
406 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
408 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
411 * Fix up the by-block btree entry(s).
413 if (nfbno1
== NULLAGBLOCK
) {
415 * No remaining freespace, just delete the by-block tree entry.
417 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
419 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
422 * Update the by-block entry to start later|be shorter.
424 if ((error
= xfs_alloc_update(bno_cur
, nfbno1
, nflen1
)))
427 if (nfbno2
!= NULLAGBLOCK
) {
429 * 2 resulting free entries, need to add one.
431 if ((error
= xfs_alloc_lookup_eq(bno_cur
, nfbno2
, nflen2
, &i
)))
433 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
434 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
436 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
445 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
446 struct xfs_agfl
*agfl
= XFS_BUF_TO_AGFL(bp
);
449 if (!uuid_equal(&agfl
->agfl_uuid
, &mp
->m_sb
.sb_uuid
))
451 if (be32_to_cpu(agfl
->agfl_magicnum
) != XFS_AGFL_MAGIC
)
454 * during growfs operations, the perag is not fully initialised,
455 * so we can't use it for any useful checking. growfs ensures we can't
456 * use it by using uncached buffers that don't have the perag attached
457 * so we can detect and avoid this problem.
459 if (bp
->b_pag
&& be32_to_cpu(agfl
->agfl_seqno
) != bp
->b_pag
->pag_agno
)
462 for (i
= 0; i
< XFS_AGFL_SIZE(mp
); i
++) {
463 if (be32_to_cpu(agfl
->agfl_bno
[i
]) != NULLAGBLOCK
&&
464 be32_to_cpu(agfl
->agfl_bno
[i
]) >= mp
->m_sb
.sb_agblocks
)
471 xfs_agfl_read_verify(
474 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
477 * There is no verification of non-crc AGFLs because mkfs does not
478 * initialise the AGFL to zero or NULL. Hence the only valid part of the
479 * AGFL is what the AGF says is active. We can't get to the AGF, so we
480 * can't verify just those entries are valid.
482 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
485 if (!xfs_buf_verify_cksum(bp
, XFS_AGFL_CRC_OFF
))
486 xfs_buf_ioerror(bp
, -EFSBADCRC
);
487 else if (!xfs_agfl_verify(bp
))
488 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
491 xfs_verifier_error(bp
);
495 xfs_agfl_write_verify(
498 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
499 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
501 /* no verification of non-crc AGFLs */
502 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
505 if (!xfs_agfl_verify(bp
)) {
506 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
507 xfs_verifier_error(bp
);
512 XFS_BUF_TO_AGFL(bp
)->agfl_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
514 xfs_buf_update_cksum(bp
, XFS_AGFL_CRC_OFF
);
517 const struct xfs_buf_ops xfs_agfl_buf_ops
= {
518 .verify_read
= xfs_agfl_read_verify
,
519 .verify_write
= xfs_agfl_write_verify
,
523 * Read in the allocation group free block array.
525 STATIC
int /* error */
527 xfs_mount_t
*mp
, /* mount point structure */
528 xfs_trans_t
*tp
, /* transaction pointer */
529 xfs_agnumber_t agno
, /* allocation group number */
530 xfs_buf_t
**bpp
) /* buffer for the ag free block array */
532 xfs_buf_t
*bp
; /* return value */
535 ASSERT(agno
!= NULLAGNUMBER
);
536 error
= xfs_trans_read_buf(
537 mp
, tp
, mp
->m_ddev_targp
,
538 XFS_AG_DADDR(mp
, agno
, XFS_AGFL_DADDR(mp
)),
539 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, &xfs_agfl_buf_ops
);
542 xfs_buf_set_ref(bp
, XFS_AGFL_REF
);
548 xfs_alloc_update_counters(
549 struct xfs_trans
*tp
,
550 struct xfs_perag
*pag
,
551 struct xfs_buf
*agbp
,
554 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
556 pag
->pagf_freeblks
+= len
;
557 be32_add_cpu(&agf
->agf_freeblks
, len
);
559 xfs_trans_agblocks_delta(tp
, len
);
560 if (unlikely(be32_to_cpu(agf
->agf_freeblks
) >
561 be32_to_cpu(agf
->agf_length
)))
562 return -EFSCORRUPTED
;
564 xfs_alloc_log_agf(tp
, agbp
, XFS_AGF_FREEBLKS
);
569 * Allocation group level functions.
573 * Allocate a variable extent in the allocation group agno.
574 * Type and bno are used to determine where in the allocation group the
576 * Extent's length (returned in *len) will be between minlen and maxlen,
577 * and of the form k * prod + mod unless there's nothing that large.
578 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
580 STATIC
int /* error */
581 xfs_alloc_ag_vextent(
582 xfs_alloc_arg_t
*args
) /* argument structure for allocation */
586 ASSERT(args
->minlen
> 0);
587 ASSERT(args
->maxlen
> 0);
588 ASSERT(args
->minlen
<= args
->maxlen
);
589 ASSERT(args
->mod
< args
->prod
);
590 ASSERT(args
->alignment
> 0);
592 * Branch to correct routine based on the type.
595 switch (args
->type
) {
596 case XFS_ALLOCTYPE_THIS_AG
:
597 error
= xfs_alloc_ag_vextent_size(args
);
599 case XFS_ALLOCTYPE_NEAR_BNO
:
600 error
= xfs_alloc_ag_vextent_near(args
);
602 case XFS_ALLOCTYPE_THIS_BNO
:
603 error
= xfs_alloc_ag_vextent_exact(args
);
610 if (error
|| args
->agbno
== NULLAGBLOCK
)
613 ASSERT(args
->len
>= args
->minlen
);
614 ASSERT(args
->len
<= args
->maxlen
);
615 ASSERT(!args
->wasfromfl
|| !args
->isfl
);
616 ASSERT(args
->agbno
% args
->alignment
== 0);
618 if (!args
->wasfromfl
) {
619 error
= xfs_alloc_update_counters(args
->tp
, args
->pag
,
621 -((long)(args
->len
)));
625 ASSERT(!xfs_extent_busy_search(args
->mp
, args
->agno
,
626 args
->agbno
, args
->len
));
630 xfs_trans_mod_sb(args
->tp
, args
->wasdel
?
631 XFS_TRANS_SB_RES_FDBLOCKS
:
632 XFS_TRANS_SB_FDBLOCKS
,
633 -((long)(args
->len
)));
636 XFS_STATS_INC(xs_allocx
);
637 XFS_STATS_ADD(xs_allocb
, args
->len
);
642 * Allocate a variable extent at exactly agno/bno.
643 * Extent's length (returned in *len) will be between minlen and maxlen,
644 * and of the form k * prod + mod unless there's nothing that large.
645 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
647 STATIC
int /* error */
648 xfs_alloc_ag_vextent_exact(
649 xfs_alloc_arg_t
*args
) /* allocation argument structure */
651 xfs_btree_cur_t
*bno_cur
;/* by block-number btree cursor */
652 xfs_btree_cur_t
*cnt_cur
;/* by count btree cursor */
654 xfs_agblock_t fbno
; /* start block of found extent */
655 xfs_extlen_t flen
; /* length of found extent */
656 xfs_agblock_t tbno
; /* start block of trimmed extent */
657 xfs_extlen_t tlen
; /* length of trimmed extent */
658 xfs_agblock_t tend
; /* end block of trimmed extent */
659 int i
; /* success/failure of operation */
661 ASSERT(args
->alignment
== 1);
664 * Allocate/initialize a cursor for the by-number freespace btree.
666 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
667 args
->agno
, XFS_BTNUM_BNO
);
670 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
671 * Look for the closest free block <= bno, it must contain bno
672 * if any free block does.
674 error
= xfs_alloc_lookup_le(bno_cur
, args
->agbno
, args
->minlen
, &i
);
681 * Grab the freespace record.
683 error
= xfs_alloc_get_rec(bno_cur
, &fbno
, &flen
, &i
);
686 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
687 ASSERT(fbno
<= args
->agbno
);
690 * Check for overlapping busy extents.
692 xfs_extent_busy_trim(args
, fbno
, flen
, &tbno
, &tlen
);
695 * Give up if the start of the extent is busy, or the freespace isn't
696 * long enough for the minimum request.
698 if (tbno
> args
->agbno
)
700 if (tlen
< args
->minlen
)
703 if (tend
< args
->agbno
+ args
->minlen
)
707 * End of extent will be smaller of the freespace end and the
708 * maximal requested end.
710 * Fix the length according to mod and prod if given.
712 args
->len
= XFS_AGBLOCK_MIN(tend
, args
->agbno
+ args
->maxlen
)
714 xfs_alloc_fix_len(args
);
715 if (!xfs_alloc_fix_minleft(args
))
718 ASSERT(args
->agbno
+ args
->len
<= tend
);
721 * We are allocating agbno for args->len
722 * Allocate/initialize a cursor for the by-size btree.
724 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
725 args
->agno
, XFS_BTNUM_CNT
);
726 ASSERT(args
->agbno
+ args
->len
<=
727 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
728 error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
, args
->agbno
,
729 args
->len
, XFSA_FIXUP_BNO_OK
);
731 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
735 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
736 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
739 trace_xfs_alloc_exact_done(args
);
743 /* Didn't find it, return null. */
744 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
745 args
->agbno
= NULLAGBLOCK
;
746 trace_xfs_alloc_exact_notfound(args
);
750 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
751 trace_xfs_alloc_exact_error(args
);
756 * Search the btree in a given direction via the search cursor and compare
757 * the records found against the good extent we've already found.
760 xfs_alloc_find_best_extent(
761 struct xfs_alloc_arg
*args
, /* allocation argument structure */
762 struct xfs_btree_cur
**gcur
, /* good cursor */
763 struct xfs_btree_cur
**scur
, /* searching cursor */
764 xfs_agblock_t gdiff
, /* difference for search comparison */
765 xfs_agblock_t
*sbno
, /* extent found by search */
766 xfs_extlen_t
*slen
, /* extent length */
767 xfs_agblock_t
*sbnoa
, /* aligned extent found by search */
768 xfs_extlen_t
*slena
, /* aligned extent length */
769 int dir
) /* 0 = search right, 1 = search left */
776 /* The good extent is perfect, no need to search. */
781 * Look until we find a better one, run out of space or run off the end.
784 error
= xfs_alloc_get_rec(*scur
, sbno
, slen
, &i
);
787 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
788 xfs_alloc_compute_aligned(args
, *sbno
, *slen
, sbnoa
, slena
);
791 * The good extent is closer than this one.
794 if (*sbnoa
>= args
->agbno
+ gdiff
)
797 if (*sbnoa
<= args
->agbno
- gdiff
)
802 * Same distance, compare length and pick the best.
804 if (*slena
>= args
->minlen
) {
805 args
->len
= XFS_EXTLEN_MIN(*slena
, args
->maxlen
);
806 xfs_alloc_fix_len(args
);
808 sdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
810 args
->userdata
, *sbnoa
,
814 * Choose closer size and invalidate other cursor.
822 error
= xfs_btree_increment(*scur
, 0, &i
);
824 error
= xfs_btree_decrement(*scur
, 0, &i
);
830 xfs_btree_del_cursor(*scur
, XFS_BTREE_NOERROR
);
835 xfs_btree_del_cursor(*gcur
, XFS_BTREE_NOERROR
);
840 /* caller invalidates cursors */
845 * Allocate a variable extent near bno in the allocation group agno.
846 * Extent's length (returned in len) will be between minlen and maxlen,
847 * and of the form k * prod + mod unless there's nothing that large.
848 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
850 STATIC
int /* error */
851 xfs_alloc_ag_vextent_near(
852 xfs_alloc_arg_t
*args
) /* allocation argument structure */
854 xfs_btree_cur_t
*bno_cur_gt
; /* cursor for bno btree, right side */
855 xfs_btree_cur_t
*bno_cur_lt
; /* cursor for bno btree, left side */
856 xfs_btree_cur_t
*cnt_cur
; /* cursor for count btree */
857 xfs_agblock_t gtbno
; /* start bno of right side entry */
858 xfs_agblock_t gtbnoa
; /* aligned ... */
859 xfs_extlen_t gtdiff
; /* difference to right side entry */
860 xfs_extlen_t gtlen
; /* length of right side entry */
861 xfs_extlen_t gtlena
; /* aligned ... */
862 xfs_agblock_t gtnew
; /* useful start bno of right side */
863 int error
; /* error code */
864 int i
; /* result code, temporary */
865 int j
; /* result code, temporary */
866 xfs_agblock_t ltbno
; /* start bno of left side entry */
867 xfs_agblock_t ltbnoa
; /* aligned ... */
868 xfs_extlen_t ltdiff
; /* difference to left side entry */
869 xfs_extlen_t ltlen
; /* length of left side entry */
870 xfs_extlen_t ltlena
; /* aligned ... */
871 xfs_agblock_t ltnew
; /* useful start bno of left side */
872 xfs_extlen_t rlen
; /* length of returned extent */
876 * Randomly don't execute the first algorithm.
878 int dofirst
; /* set to do first algorithm */
880 dofirst
= prandom_u32() & 1;
891 * Get a cursor for the by-size btree.
893 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
894 args
->agno
, XFS_BTNUM_CNT
);
897 * See if there are any free extents as big as maxlen.
899 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0, args
->maxlen
, &i
)))
902 * If none, then pick up the last entry in the tree unless the
906 if ((error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
, <bno
,
909 if (i
== 0 || ltlen
== 0) {
910 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
911 trace_xfs_alloc_near_noentry(args
);
920 * If the requested extent is large wrt the freespaces available
921 * in this a.g., then the cursor will be pointing to a btree entry
922 * near the right edge of the tree. If it's in the last btree leaf
923 * block, then we just examine all the entries in that block
924 * that are big enough, and pick the best one.
925 * This is written as a while loop so we can break out of it,
926 * but we never loop back to the top.
928 while (xfs_btree_islastblock(cnt_cur
, 0)) {
932 xfs_agblock_t bnew
=0;
939 * Start from the entry that lookup found, sequence through
940 * all larger free blocks. If we're actually pointing at a
941 * record smaller than maxlen, go to the start of this block,
942 * and skip all those smaller than minlen.
944 if (ltlen
|| args
->alignment
> 1) {
945 cnt_cur
->bc_ptrs
[0] = 1;
947 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
,
950 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
951 if (ltlen
>= args
->minlen
)
953 if ((error
= xfs_btree_increment(cnt_cur
, 0, &i
)))
956 ASSERT(ltlen
>= args
->minlen
);
960 i
= cnt_cur
->bc_ptrs
[0];
961 for (j
= 1, blen
= 0, bdiff
= 0;
962 !error
&& j
&& (blen
< args
->maxlen
|| bdiff
> 0);
963 error
= xfs_btree_increment(cnt_cur
, 0, &j
)) {
965 * For each entry, decide if it's better than
966 * the previous best entry.
968 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
970 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
971 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
973 if (ltlena
< args
->minlen
)
975 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
976 xfs_alloc_fix_len(args
);
977 ASSERT(args
->len
>= args
->minlen
);
978 if (args
->len
< blen
)
980 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
981 args
->alignment
, args
->userdata
, ltbnoa
,
983 if (ltnew
!= NULLAGBLOCK
&&
984 (args
->len
> blen
|| ltdiff
< bdiff
)) {
988 besti
= cnt_cur
->bc_ptrs
[0];
992 * It didn't work. We COULD be in a case where
993 * there's a good record somewhere, so try again.
998 * Point at the best entry, and retrieve it again.
1000 cnt_cur
->bc_ptrs
[0] = besti
;
1001 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
1003 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1004 ASSERT(ltbno
+ ltlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1006 if (!xfs_alloc_fix_minleft(args
)) {
1007 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1008 trace_xfs_alloc_near_nominleft(args
);
1013 * We are allocating starting at bnew for blen blocks.
1016 ASSERT(bnew
>= ltbno
);
1017 ASSERT(bnew
+ blen
<= ltbno
+ ltlen
);
1019 * Set up a cursor for the by-bno tree.
1021 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
,
1022 args
->agbp
, args
->agno
, XFS_BTNUM_BNO
);
1024 * Fix up the btree entries.
1026 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
,
1027 ltlen
, bnew
, blen
, XFSA_FIXUP_CNT_OK
)))
1029 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1030 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1032 trace_xfs_alloc_near_first(args
);
1037 * Search in the by-bno tree to the left and to the right
1038 * simultaneously, until in each case we find a space big enough,
1039 * or run into the edge of the tree. When we run into the edge,
1040 * we deallocate that cursor.
1041 * If both searches succeed, we compare the two spaces and pick
1043 * With alignment, it's possible for both to fail; the upper
1044 * level algorithm that picks allocation groups for allocations
1045 * is not supposed to do this.
1048 * Allocate and initialize the cursor for the leftward search.
1050 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1051 args
->agno
, XFS_BTNUM_BNO
);
1053 * Lookup <= bno to find the leftward search's starting point.
1055 if ((error
= xfs_alloc_lookup_le(bno_cur_lt
, args
->agbno
, args
->maxlen
, &i
)))
1059 * Didn't find anything; use this cursor for the rightward
1062 bno_cur_gt
= bno_cur_lt
;
1066 * Found something. Duplicate the cursor for the rightward search.
1068 else if ((error
= xfs_btree_dup_cursor(bno_cur_lt
, &bno_cur_gt
)))
1071 * Increment the cursor, so we will point at the entry just right
1072 * of the leftward entry if any, or to the leftmost entry.
1074 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1078 * It failed, there are no rightward entries.
1080 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_NOERROR
);
1084 * Loop going left with the leftward cursor, right with the
1085 * rightward cursor, until either both directions give up or
1086 * we find an entry at least as big as minlen.
1090 if ((error
= xfs_alloc_get_rec(bno_cur_lt
, <bno
, <len
, &i
)))
1092 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1093 xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
1095 if (ltlena
>= args
->minlen
)
1097 if ((error
= xfs_btree_decrement(bno_cur_lt
, 0, &i
)))
1100 xfs_btree_del_cursor(bno_cur_lt
,
1106 if ((error
= xfs_alloc_get_rec(bno_cur_gt
, >bno
, >len
, &i
)))
1108 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1109 xfs_alloc_compute_aligned(args
, gtbno
, gtlen
,
1111 if (gtlena
>= args
->minlen
)
1113 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1116 xfs_btree_del_cursor(bno_cur_gt
,
1121 } while (bno_cur_lt
|| bno_cur_gt
);
1124 * Got both cursors still active, need to find better entry.
1126 if (bno_cur_lt
&& bno_cur_gt
) {
1127 if (ltlena
>= args
->minlen
) {
1129 * Left side is good, look for a right side entry.
1131 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1132 xfs_alloc_fix_len(args
);
1133 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1134 args
->alignment
, args
->userdata
, ltbnoa
,
1137 error
= xfs_alloc_find_best_extent(args
,
1138 &bno_cur_lt
, &bno_cur_gt
,
1139 ltdiff
, >bno
, >len
,
1141 0 /* search right */);
1143 ASSERT(gtlena
>= args
->minlen
);
1146 * Right side is good, look for a left side entry.
1148 args
->len
= XFS_EXTLEN_MIN(gtlena
, args
->maxlen
);
1149 xfs_alloc_fix_len(args
);
1150 gtdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1151 args
->alignment
, args
->userdata
, gtbnoa
,
1154 error
= xfs_alloc_find_best_extent(args
,
1155 &bno_cur_gt
, &bno_cur_lt
,
1156 gtdiff
, <bno
, <len
,
1158 1 /* search left */);
1166 * If we couldn't get anything, give up.
1168 if (bno_cur_lt
== NULL
&& bno_cur_gt
== NULL
) {
1169 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1172 trace_xfs_alloc_near_busy(args
);
1173 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1176 trace_xfs_alloc_size_neither(args
);
1177 args
->agbno
= NULLAGBLOCK
;
1182 * At this point we have selected a freespace entry, either to the
1183 * left or to the right. If it's on the right, copy all the
1184 * useful variables to the "left" set so we only have one
1185 * copy of this code.
1188 bno_cur_lt
= bno_cur_gt
;
1199 * Fix up the length and compute the useful address.
1201 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1202 xfs_alloc_fix_len(args
);
1203 if (!xfs_alloc_fix_minleft(args
)) {
1204 trace_xfs_alloc_near_nominleft(args
);
1205 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1206 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1210 (void)xfs_alloc_compute_diff(args
->agbno
, rlen
, args
->alignment
,
1211 args
->userdata
, ltbnoa
, ltlena
, <new
);
1212 ASSERT(ltnew
>= ltbno
);
1213 ASSERT(ltnew
+ rlen
<= ltbnoa
+ ltlena
);
1214 ASSERT(ltnew
+ rlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1215 args
->agbno
= ltnew
;
1217 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
, ltlen
,
1218 ltnew
, rlen
, XFSA_FIXUP_BNO_OK
)))
1222 trace_xfs_alloc_near_greater(args
);
1224 trace_xfs_alloc_near_lesser(args
);
1226 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1227 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1231 trace_xfs_alloc_near_error(args
);
1232 if (cnt_cur
!= NULL
)
1233 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1234 if (bno_cur_lt
!= NULL
)
1235 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_ERROR
);
1236 if (bno_cur_gt
!= NULL
)
1237 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_ERROR
);
1242 * Allocate a variable extent anywhere in the allocation group agno.
1243 * Extent's length (returned in len) will be between minlen and maxlen,
1244 * and of the form k * prod + mod unless there's nothing that large.
1245 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1247 STATIC
int /* error */
1248 xfs_alloc_ag_vextent_size(
1249 xfs_alloc_arg_t
*args
) /* allocation argument structure */
1251 xfs_btree_cur_t
*bno_cur
; /* cursor for bno btree */
1252 xfs_btree_cur_t
*cnt_cur
; /* cursor for cnt btree */
1253 int error
; /* error result */
1254 xfs_agblock_t fbno
; /* start of found freespace */
1255 xfs_extlen_t flen
; /* length of found freespace */
1256 int i
; /* temp status variable */
1257 xfs_agblock_t rbno
; /* returned block number */
1258 xfs_extlen_t rlen
; /* length of returned extent */
1263 * Allocate and initialize a cursor for the by-size btree.
1265 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1266 args
->agno
, XFS_BTNUM_CNT
);
1270 * Look for an entry >= maxlen+alignment-1 blocks.
1272 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0,
1273 args
->maxlen
+ args
->alignment
- 1, &i
)))
1277 * If none or we have busy extents that we cannot allocate from, then
1278 * we have to settle for a smaller extent. In the case that there are
1279 * no large extents, this will return the last entry in the tree unless
1280 * the tree is empty. In the case that there are only busy large
1281 * extents, this will return the largest small extent unless there
1282 * are no smaller extents available.
1284 if (!i
|| forced
> 1) {
1285 error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
,
1289 if (i
== 0 || flen
== 0) {
1290 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1291 trace_xfs_alloc_size_noentry(args
);
1295 xfs_alloc_compute_aligned(args
, fbno
, flen
, &rbno
, &rlen
);
1298 * Search for a non-busy extent that is large enough.
1299 * If we are at low space, don't check, or if we fall of
1300 * the end of the btree, turn off the busy check and
1304 error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
, &i
);
1307 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1309 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1312 if (rlen
>= args
->maxlen
)
1315 error
= xfs_btree_increment(cnt_cur
, 0, &i
);
1320 * Our only valid extents must have been busy.
1321 * Make it unbusy by forcing the log out and
1322 * retrying. If we've been here before, forcing
1323 * the log isn't making the extents available,
1324 * which means they have probably been freed in
1325 * this transaction. In that case, we have to
1326 * give up on them and we'll attempt a minlen
1327 * allocation the next time around.
1329 xfs_btree_del_cursor(cnt_cur
,
1331 trace_xfs_alloc_size_busy(args
);
1333 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1340 * In the first case above, we got the last entry in the
1341 * by-size btree. Now we check to see if the space hits maxlen
1342 * once aligned; if not, we search left for something better.
1343 * This can't happen in the second case above.
1345 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1346 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
== 0 ||
1347 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
), error0
);
1348 if (rlen
< args
->maxlen
) {
1349 xfs_agblock_t bestfbno
;
1350 xfs_extlen_t bestflen
;
1351 xfs_agblock_t bestrbno
;
1352 xfs_extlen_t bestrlen
;
1359 if ((error
= xfs_btree_decrement(cnt_cur
, 0, &i
)))
1363 if ((error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
,
1366 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1367 if (flen
< bestrlen
)
1369 xfs_alloc_compute_aligned(args
, fbno
, flen
,
1371 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1372 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
== 0 ||
1373 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
),
1375 if (rlen
> bestrlen
) {
1380 if (rlen
== args
->maxlen
)
1384 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, bestfbno
, bestflen
,
1387 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1393 args
->wasfromfl
= 0;
1395 * Fix up the length.
1398 if (rlen
< args
->minlen
) {
1400 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1401 trace_xfs_alloc_size_busy(args
);
1402 xfs_log_force(args
->mp
, XFS_LOG_SYNC
);
1407 xfs_alloc_fix_len(args
);
1409 if (!xfs_alloc_fix_minleft(args
))
1412 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
<= flen
, error0
);
1414 * Allocate and initialize a cursor for the by-block tree.
1416 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1417 args
->agno
, XFS_BTNUM_BNO
);
1418 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
,
1419 rbno
, rlen
, XFSA_FIXUP_CNT_OK
)))
1421 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1422 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1423 cnt_cur
= bno_cur
= NULL
;
1426 XFS_WANT_CORRUPTED_GOTO(args
->mp
,
1427 args
->agbno
+ args
->len
<=
1428 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1430 trace_xfs_alloc_size_done(args
);
1434 trace_xfs_alloc_size_error(args
);
1436 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1438 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1442 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1443 trace_xfs_alloc_size_nominleft(args
);
1444 args
->agbno
= NULLAGBLOCK
;
1449 * Deal with the case where only small freespaces remain.
1450 * Either return the contents of the last freespace record,
1451 * or allocate space from the freelist if there is nothing in the tree.
1453 STATIC
int /* error */
1454 xfs_alloc_ag_vextent_small(
1455 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1456 xfs_btree_cur_t
*ccur
, /* by-size cursor */
1457 xfs_agblock_t
*fbnop
, /* result block number */
1458 xfs_extlen_t
*flenp
, /* result length */
1459 int *stat
) /* status: 0-freelist, 1-normal/none */
1466 if ((error
= xfs_btree_decrement(ccur
, 0, &i
)))
1469 if ((error
= xfs_alloc_get_rec(ccur
, &fbno
, &flen
, &i
)))
1471 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1474 * Nothing in the btree, try the freelist. Make sure
1475 * to respect minleft even when pulling from the
1478 else if (args
->minlen
== 1 && args
->alignment
== 1 && !args
->isfl
&&
1479 (be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_flcount
)
1481 error
= xfs_alloc_get_freelist(args
->tp
, args
->agbp
, &fbno
, 0);
1484 if (fbno
!= NULLAGBLOCK
) {
1485 xfs_extent_busy_reuse(args
->mp
, args
->agno
, fbno
, 1,
1488 if (args
->userdata
) {
1491 bp
= xfs_btree_get_bufs(args
->mp
, args
->tp
,
1492 args
->agno
, fbno
, 0);
1493 xfs_trans_binval(args
->tp
, bp
);
1497 XFS_WANT_CORRUPTED_GOTO(args
->mp
,
1498 args
->agbno
+ args
->len
<=
1499 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1501 args
->wasfromfl
= 1;
1502 trace_xfs_alloc_small_freelist(args
);
1507 * Nothing in the freelist.
1513 * Can't allocate from the freelist for some reason.
1520 * Can't do the allocation, give up.
1522 if (flen
< args
->minlen
) {
1523 args
->agbno
= NULLAGBLOCK
;
1524 trace_xfs_alloc_small_notenough(args
);
1530 trace_xfs_alloc_small_done(args
);
1534 trace_xfs_alloc_small_error(args
);
1539 * Free the extent starting at agno/bno for length.
1541 STATIC
int /* error */
1543 xfs_trans_t
*tp
, /* transaction pointer */
1544 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
1545 xfs_agnumber_t agno
, /* allocation group number */
1546 xfs_agblock_t bno
, /* starting block number */
1547 xfs_extlen_t len
, /* length of extent */
1548 int isfl
) /* set if is freelist blocks - no sb acctg */
1550 xfs_btree_cur_t
*bno_cur
; /* cursor for by-block btree */
1551 xfs_btree_cur_t
*cnt_cur
; /* cursor for by-size btree */
1552 int error
; /* error return value */
1553 xfs_agblock_t gtbno
; /* start of right neighbor block */
1554 xfs_extlen_t gtlen
; /* length of right neighbor block */
1555 int haveleft
; /* have a left neighbor block */
1556 int haveright
; /* have a right neighbor block */
1557 int i
; /* temp, result code */
1558 xfs_agblock_t ltbno
; /* start of left neighbor block */
1559 xfs_extlen_t ltlen
; /* length of left neighbor block */
1560 xfs_mount_t
*mp
; /* mount point struct for filesystem */
1561 xfs_agblock_t nbno
; /* new starting block of freespace */
1562 xfs_extlen_t nlen
; /* new length of freespace */
1563 xfs_perag_t
*pag
; /* per allocation group data */
1567 * Allocate and initialize a cursor for the by-block btree.
1569 bno_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_BNO
);
1572 * Look for a neighboring block on the left (lower block numbers)
1573 * that is contiguous with this space.
1575 if ((error
= xfs_alloc_lookup_le(bno_cur
, bno
, len
, &haveleft
)))
1579 * There is a block to our left.
1581 if ((error
= xfs_alloc_get_rec(bno_cur
, <bno
, <len
, &i
)))
1583 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1585 * It's not contiguous, though.
1587 if (ltbno
+ ltlen
< bno
)
1591 * If this failure happens the request to free this
1592 * space was invalid, it's (partly) already free.
1595 XFS_WANT_CORRUPTED_GOTO(mp
,
1596 ltbno
+ ltlen
<= bno
, error0
);
1600 * Look for a neighboring block on the right (higher block numbers)
1601 * that is contiguous with this space.
1603 if ((error
= xfs_btree_increment(bno_cur
, 0, &haveright
)))
1607 * There is a block to our right.
1609 if ((error
= xfs_alloc_get_rec(bno_cur
, >bno
, >len
, &i
)))
1611 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1613 * It's not contiguous, though.
1615 if (bno
+ len
< gtbno
)
1619 * If this failure happens the request to free this
1620 * space was invalid, it's (partly) already free.
1623 XFS_WANT_CORRUPTED_GOTO(mp
, gtbno
>= bno
+ len
, error0
);
1627 * Now allocate and initialize a cursor for the by-size tree.
1629 cnt_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_CNT
);
1631 * Have both left and right contiguous neighbors.
1632 * Merge all three into a single free block.
1634 if (haveleft
&& haveright
) {
1636 * Delete the old by-size entry on the left.
1638 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1640 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1641 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1643 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1645 * Delete the old by-size entry on the right.
1647 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1649 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1650 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1652 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1654 * Delete the old by-block entry for the right block.
1656 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
1658 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1660 * Move the by-block cursor back to the left neighbor.
1662 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1664 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1667 * Check that this is the right record: delete didn't
1668 * mangle the cursor.
1671 xfs_agblock_t xxbno
;
1674 if ((error
= xfs_alloc_get_rec(bno_cur
, &xxbno
, &xxlen
,
1677 XFS_WANT_CORRUPTED_GOTO(mp
,
1678 i
== 1 && xxbno
== ltbno
&& xxlen
== ltlen
,
1683 * Update remaining by-block entry to the new, joined block.
1686 nlen
= len
+ ltlen
+ gtlen
;
1687 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1691 * Have only a left contiguous neighbor.
1692 * Merge it together with the new freespace.
1694 else if (haveleft
) {
1696 * Delete the old by-size entry on the left.
1698 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1700 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1701 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1703 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1705 * Back up the by-block cursor to the left neighbor, and
1706 * update its length.
1708 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1710 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1713 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1717 * Have only a right contiguous neighbor.
1718 * Merge it together with the new freespace.
1720 else if (haveright
) {
1722 * Delete the old by-size entry on the right.
1724 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1726 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1727 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1729 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1731 * Update the starting block and length of the right
1732 * neighbor in the by-block tree.
1736 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1740 * No contiguous neighbors.
1741 * Insert the new freespace into the by-block tree.
1746 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
1748 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1750 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1753 * In all cases we need to insert the new freespace in the by-size tree.
1755 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nbno
, nlen
, &i
)))
1757 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, error0
);
1758 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
1760 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1761 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1765 * Update the freespace totals in the ag and superblock.
1767 pag
= xfs_perag_get(mp
, agno
);
1768 error
= xfs_alloc_update_counters(tp
, pag
, agbp
, len
);
1774 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_FDBLOCKS
, (long)len
);
1775 XFS_STATS_INC(xs_freex
);
1776 XFS_STATS_ADD(xs_freeb
, len
);
1778 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, haveleft
, haveright
);
1783 trace_xfs_free_extent(mp
, agno
, bno
, len
, isfl
, -1, -1);
1785 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1787 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1792 * Visible (exported) allocation/free functions.
1793 * Some of these are used just by xfs_alloc_btree.c and this file.
1797 * Compute and fill in value of m_ag_maxlevels.
1800 xfs_alloc_compute_maxlevels(
1801 xfs_mount_t
*mp
) /* file system mount structure */
1809 maxleafents
= (mp
->m_sb
.sb_agblocks
+ 1) / 2;
1810 minleafrecs
= mp
->m_alloc_mnr
[0];
1811 minnoderecs
= mp
->m_alloc_mnr
[1];
1812 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
1813 for (level
= 1; maxblocks
> 1; level
++)
1814 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
1815 mp
->m_ag_maxlevels
= level
;
1819 * Find the length of the longest extent in an AG.
1822 xfs_alloc_longest_free_extent(
1823 struct xfs_mount
*mp
,
1824 struct xfs_perag
*pag
)
1826 xfs_extlen_t need
, delta
= 0;
1828 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1829 if (need
> pag
->pagf_flcount
)
1830 delta
= need
- pag
->pagf_flcount
;
1832 if (pag
->pagf_longest
> delta
)
1833 return pag
->pagf_longest
- delta
;
1834 return pag
->pagf_flcount
> 0 || pag
->pagf_longest
> 0;
1838 * Decide whether to use this allocation group for this allocation.
1839 * If so, fix up the btree freelist's size.
1842 xfs_alloc_fix_freelist(
1843 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1844 int flags
) /* XFS_ALLOC_FLAG_... */
1846 xfs_buf_t
*agbp
; /* agf buffer pointer */
1847 xfs_agf_t
*agf
; /* a.g. freespace structure pointer */
1848 xfs_buf_t
*agflbp
;/* agfl buffer pointer */
1849 xfs_agblock_t bno
; /* freelist block */
1850 xfs_extlen_t delta
; /* new blocks needed in freelist */
1851 int error
; /* error result code */
1852 xfs_extlen_t longest
;/* longest extent in allocation group */
1853 xfs_mount_t
*mp
; /* file system mount point structure */
1854 xfs_extlen_t need
; /* total blocks needed in freelist */
1855 xfs_perag_t
*pag
; /* per-ag information structure */
1856 xfs_alloc_arg_t targs
; /* local allocation arguments */
1857 xfs_trans_t
*tp
; /* transaction pointer */
1863 if (!pag
->pagf_init
) {
1864 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1867 if (!pag
->pagf_init
) {
1868 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1869 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1877 * If this is a metadata preferred pag and we are user data
1878 * then try somewhere else if we are not being asked to
1879 * try harder at this point
1881 if (pag
->pagf_metadata
&& args
->userdata
&&
1882 (flags
& XFS_ALLOC_FLAG_TRYLOCK
)) {
1883 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1888 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1890 * If it looks like there isn't a long enough extent, or enough
1891 * total blocks, reject it.
1893 need
= XFS_MIN_FREELIST_PAG(pag
, mp
);
1894 longest
= xfs_alloc_longest_free_extent(mp
, pag
);
1895 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1897 ((int)(pag
->pagf_freeblks
+ pag
->pagf_flcount
-
1898 need
- args
->total
) < (int)args
->minleft
)) {
1900 xfs_trans_brelse(tp
, agbp
);
1907 * Get the a.g. freespace buffer.
1908 * Can fail if we're not blocking on locks, and it's held.
1911 if ((error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
,
1915 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
1916 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
1922 * Figure out how many blocks we should have in the freelist.
1924 agf
= XFS_BUF_TO_AGF(agbp
);
1925 need
= XFS_MIN_FREELIST(agf
, mp
);
1927 * If there isn't enough total or single-extent, reject it.
1929 if (!(flags
& XFS_ALLOC_FLAG_FREEING
)) {
1930 delta
= need
> be32_to_cpu(agf
->agf_flcount
) ?
1931 (need
- be32_to_cpu(agf
->agf_flcount
)) : 0;
1932 longest
= be32_to_cpu(agf
->agf_longest
);
1933 longest
= (longest
> delta
) ? (longest
- delta
) :
1934 (be32_to_cpu(agf
->agf_flcount
) > 0 || longest
> 0);
1935 if ((args
->minlen
+ args
->alignment
+ args
->minalignslop
- 1) >
1937 ((int)(be32_to_cpu(agf
->agf_freeblks
) +
1938 be32_to_cpu(agf
->agf_flcount
) - need
- args
->total
) <
1939 (int)args
->minleft
)) {
1940 xfs_trans_brelse(tp
, agbp
);
1946 * Make the freelist shorter if it's too long.
1948 while (be32_to_cpu(agf
->agf_flcount
) > need
) {
1951 error
= xfs_alloc_get_freelist(tp
, agbp
, &bno
, 0);
1954 if ((error
= xfs_free_ag_extent(tp
, agbp
, args
->agno
, bno
, 1, 1)))
1956 bp
= xfs_btree_get_bufs(mp
, tp
, args
->agno
, bno
, 0);
1957 xfs_trans_binval(tp
, bp
);
1960 * Initialize the args structure.
1962 memset(&targs
, 0, sizeof(targs
));
1966 targs
.agno
= args
->agno
;
1967 targs
.alignment
= targs
.minlen
= targs
.prod
= targs
.isfl
= 1;
1968 targs
.type
= XFS_ALLOCTYPE_THIS_AG
;
1970 if ((error
= xfs_alloc_read_agfl(mp
, tp
, targs
.agno
, &agflbp
)))
1973 * Make the freelist longer if it's too short.
1975 while (be32_to_cpu(agf
->agf_flcount
) < need
) {
1977 targs
.maxlen
= need
- be32_to_cpu(agf
->agf_flcount
);
1979 * Allocate as many blocks as possible at once.
1981 if ((error
= xfs_alloc_ag_vextent(&targs
))) {
1982 xfs_trans_brelse(tp
, agflbp
);
1986 * Stop if we run out. Won't happen if callers are obeying
1987 * the restrictions correctly. Can happen for free calls
1988 * on a completely full ag.
1990 if (targs
.agbno
== NULLAGBLOCK
) {
1991 if (flags
& XFS_ALLOC_FLAG_FREEING
)
1993 xfs_trans_brelse(tp
, agflbp
);
1998 * Put each allocated block on the list.
2000 for (bno
= targs
.agbno
; bno
< targs
.agbno
+ targs
.len
; bno
++) {
2001 error
= xfs_alloc_put_freelist(tp
, agbp
,
2007 xfs_trans_brelse(tp
, agflbp
);
2013 * Get a block from the freelist.
2014 * Returns with the buffer for the block gotten.
2017 xfs_alloc_get_freelist(
2018 xfs_trans_t
*tp
, /* transaction pointer */
2019 xfs_buf_t
*agbp
, /* buffer containing the agf structure */
2020 xfs_agblock_t
*bnop
, /* block address retrieved from freelist */
2021 int btreeblk
) /* destination is a AGF btree */
2023 xfs_agf_t
*agf
; /* a.g. freespace structure */
2024 xfs_buf_t
*agflbp
;/* buffer for a.g. freelist structure */
2025 xfs_agblock_t bno
; /* block number returned */
2029 xfs_mount_t
*mp
= tp
->t_mountp
;
2030 xfs_perag_t
*pag
; /* per allocation group data */
2033 * Freelist is empty, give up.
2035 agf
= XFS_BUF_TO_AGF(agbp
);
2036 if (!agf
->agf_flcount
) {
2037 *bnop
= NULLAGBLOCK
;
2041 * Read the array of free blocks.
2043 error
= xfs_alloc_read_agfl(mp
, tp
, be32_to_cpu(agf
->agf_seqno
),
2050 * Get the block number and update the data structures.
2052 agfl_bno
= XFS_BUF_TO_AGFL_BNO(mp
, agflbp
);
2053 bno
= be32_to_cpu(agfl_bno
[be32_to_cpu(agf
->agf_flfirst
)]);
2054 be32_add_cpu(&agf
->agf_flfirst
, 1);
2055 xfs_trans_brelse(tp
, agflbp
);
2056 if (be32_to_cpu(agf
->agf_flfirst
) == XFS_AGFL_SIZE(mp
))
2057 agf
->agf_flfirst
= 0;
2059 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2060 be32_add_cpu(&agf
->agf_flcount
, -1);
2061 xfs_trans_agflist_delta(tp
, -1);
2062 pag
->pagf_flcount
--;
2065 logflags
= XFS_AGF_FLFIRST
| XFS_AGF_FLCOUNT
;
2067 be32_add_cpu(&agf
->agf_btreeblks
, 1);
2068 pag
->pagf_btreeblks
++;
2069 logflags
|= XFS_AGF_BTREEBLKS
;
2072 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2079 * Log the given fields from the agf structure.
2083 xfs_trans_t
*tp
, /* transaction pointer */
2084 xfs_buf_t
*bp
, /* buffer for a.g. freelist header */
2085 int fields
) /* mask of fields to be logged (XFS_AGF_...) */
2087 int first
; /* first byte offset */
2088 int last
; /* last byte offset */
2089 static const short offsets
[] = {
2090 offsetof(xfs_agf_t
, agf_magicnum
),
2091 offsetof(xfs_agf_t
, agf_versionnum
),
2092 offsetof(xfs_agf_t
, agf_seqno
),
2093 offsetof(xfs_agf_t
, agf_length
),
2094 offsetof(xfs_agf_t
, agf_roots
[0]),
2095 offsetof(xfs_agf_t
, agf_levels
[0]),
2096 offsetof(xfs_agf_t
, agf_flfirst
),
2097 offsetof(xfs_agf_t
, agf_fllast
),
2098 offsetof(xfs_agf_t
, agf_flcount
),
2099 offsetof(xfs_agf_t
, agf_freeblks
),
2100 offsetof(xfs_agf_t
, agf_longest
),
2101 offsetof(xfs_agf_t
, agf_btreeblks
),
2102 offsetof(xfs_agf_t
, agf_uuid
),
2106 trace_xfs_agf(tp
->t_mountp
, XFS_BUF_TO_AGF(bp
), fields
, _RET_IP_
);
2108 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_AGF_BUF
);
2110 xfs_btree_offsets(fields
, offsets
, XFS_AGF_NUM_BITS
, &first
, &last
);
2111 xfs_trans_log_buf(tp
, bp
, (uint
)first
, (uint
)last
);
2115 * Interface for inode allocation to force the pag data to be initialized.
2118 xfs_alloc_pagf_init(
2119 xfs_mount_t
*mp
, /* file system mount structure */
2120 xfs_trans_t
*tp
, /* transaction pointer */
2121 xfs_agnumber_t agno
, /* allocation group number */
2122 int flags
) /* XFS_ALLOC_FLAGS_... */
2127 if ((error
= xfs_alloc_read_agf(mp
, tp
, agno
, flags
, &bp
)))
2130 xfs_trans_brelse(tp
, bp
);
2135 * Put the block on the freelist for the allocation group.
2138 xfs_alloc_put_freelist(
2139 xfs_trans_t
*tp
, /* transaction pointer */
2140 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
2141 xfs_buf_t
*agflbp
,/* buffer for a.g. free block array */
2142 xfs_agblock_t bno
, /* block being freed */
2143 int btreeblk
) /* block came from a AGF btree */
2145 xfs_agf_t
*agf
; /* a.g. freespace structure */
2146 __be32
*blockp
;/* pointer to array entry */
2149 xfs_mount_t
*mp
; /* mount structure */
2150 xfs_perag_t
*pag
; /* per allocation group data */
2154 agf
= XFS_BUF_TO_AGF(agbp
);
2157 if (!agflbp
&& (error
= xfs_alloc_read_agfl(mp
, tp
,
2158 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
2160 be32_add_cpu(&agf
->agf_fllast
, 1);
2161 if (be32_to_cpu(agf
->agf_fllast
) == XFS_AGFL_SIZE(mp
))
2162 agf
->agf_fllast
= 0;
2164 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2165 be32_add_cpu(&agf
->agf_flcount
, 1);
2166 xfs_trans_agflist_delta(tp
, 1);
2167 pag
->pagf_flcount
++;
2169 logflags
= XFS_AGF_FLLAST
| XFS_AGF_FLCOUNT
;
2171 be32_add_cpu(&agf
->agf_btreeblks
, -1);
2172 pag
->pagf_btreeblks
--;
2173 logflags
|= XFS_AGF_BTREEBLKS
;
2177 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2179 ASSERT(be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
));
2181 agfl_bno
= XFS_BUF_TO_AGFL_BNO(mp
, agflbp
);
2182 blockp
= &agfl_bno
[be32_to_cpu(agf
->agf_fllast
)];
2183 *blockp
= cpu_to_be32(bno
);
2184 startoff
= (char *)blockp
- (char *)agflbp
->b_addr
;
2186 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2188 xfs_trans_buf_set_type(tp
, agflbp
, XFS_BLFT_AGFL_BUF
);
2189 xfs_trans_log_buf(tp
, agflbp
, startoff
,
2190 startoff
+ sizeof(xfs_agblock_t
) - 1);
2196 struct xfs_mount
*mp
,
2199 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(bp
);
2201 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2202 !uuid_equal(&agf
->agf_uuid
, &mp
->m_sb
.sb_uuid
))
2205 if (!(agf
->agf_magicnum
== cpu_to_be32(XFS_AGF_MAGIC
) &&
2206 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf
->agf_versionnum
)) &&
2207 be32_to_cpu(agf
->agf_freeblks
) <= be32_to_cpu(agf
->agf_length
) &&
2208 be32_to_cpu(agf
->agf_flfirst
) < XFS_AGFL_SIZE(mp
) &&
2209 be32_to_cpu(agf
->agf_fllast
) < XFS_AGFL_SIZE(mp
) &&
2210 be32_to_cpu(agf
->agf_flcount
) <= XFS_AGFL_SIZE(mp
)))
2213 if (be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNO
]) > XFS_BTREE_MAXLEVELS
||
2214 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNT
]) > XFS_BTREE_MAXLEVELS
)
2218 * during growfs operations, the perag is not fully initialised,
2219 * so we can't use it for any useful checking. growfs ensures we can't
2220 * use it by using uncached buffers that don't have the perag attached
2221 * so we can detect and avoid this problem.
2223 if (bp
->b_pag
&& be32_to_cpu(agf
->agf_seqno
) != bp
->b_pag
->pag_agno
)
2226 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
2227 be32_to_cpu(agf
->agf_btreeblks
) > be32_to_cpu(agf
->agf_length
))
2235 xfs_agf_read_verify(
2238 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2240 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2241 !xfs_buf_verify_cksum(bp
, XFS_AGF_CRC_OFF
))
2242 xfs_buf_ioerror(bp
, -EFSBADCRC
);
2243 else if (XFS_TEST_ERROR(!xfs_agf_verify(mp
, bp
), mp
,
2244 XFS_ERRTAG_ALLOC_READ_AGF
,
2245 XFS_RANDOM_ALLOC_READ_AGF
))
2246 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2249 xfs_verifier_error(bp
);
2253 xfs_agf_write_verify(
2256 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2257 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
2259 if (!xfs_agf_verify(mp
, bp
)) {
2260 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2261 xfs_verifier_error(bp
);
2265 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2269 XFS_BUF_TO_AGF(bp
)->agf_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2271 xfs_buf_update_cksum(bp
, XFS_AGF_CRC_OFF
);
2274 const struct xfs_buf_ops xfs_agf_buf_ops
= {
2275 .verify_read
= xfs_agf_read_verify
,
2276 .verify_write
= xfs_agf_write_verify
,
2280 * Read in the allocation group header (free/alloc section).
2284 struct xfs_mount
*mp
, /* mount point structure */
2285 struct xfs_trans
*tp
, /* transaction pointer */
2286 xfs_agnumber_t agno
, /* allocation group number */
2287 int flags
, /* XFS_BUF_ */
2288 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2292 trace_xfs_read_agf(mp
, agno
);
2294 ASSERT(agno
!= NULLAGNUMBER
);
2295 error
= xfs_trans_read_buf(
2296 mp
, tp
, mp
->m_ddev_targp
,
2297 XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
)),
2298 XFS_FSS_TO_BB(mp
, 1), flags
, bpp
, &xfs_agf_buf_ops
);
2304 ASSERT(!(*bpp
)->b_error
);
2305 xfs_buf_set_ref(*bpp
, XFS_AGF_REF
);
2310 * Read in the allocation group header (free/alloc section).
2314 struct xfs_mount
*mp
, /* mount point structure */
2315 struct xfs_trans
*tp
, /* transaction pointer */
2316 xfs_agnumber_t agno
, /* allocation group number */
2317 int flags
, /* XFS_ALLOC_FLAG_... */
2318 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2320 struct xfs_agf
*agf
; /* ag freelist header */
2321 struct xfs_perag
*pag
; /* per allocation group data */
2324 trace_xfs_alloc_read_agf(mp
, agno
);
2326 ASSERT(agno
!= NULLAGNUMBER
);
2327 error
= xfs_read_agf(mp
, tp
, agno
,
2328 (flags
& XFS_ALLOC_FLAG_TRYLOCK
) ? XBF_TRYLOCK
: 0,
2334 ASSERT(!(*bpp
)->b_error
);
2336 agf
= XFS_BUF_TO_AGF(*bpp
);
2337 pag
= xfs_perag_get(mp
, agno
);
2338 if (!pag
->pagf_init
) {
2339 pag
->pagf_freeblks
= be32_to_cpu(agf
->agf_freeblks
);
2340 pag
->pagf_btreeblks
= be32_to_cpu(agf
->agf_btreeblks
);
2341 pag
->pagf_flcount
= be32_to_cpu(agf
->agf_flcount
);
2342 pag
->pagf_longest
= be32_to_cpu(agf
->agf_longest
);
2343 pag
->pagf_levels
[XFS_BTNUM_BNOi
] =
2344 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]);
2345 pag
->pagf_levels
[XFS_BTNUM_CNTi
] =
2346 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]);
2347 spin_lock_init(&pag
->pagb_lock
);
2348 pag
->pagb_count
= 0;
2349 /* XXX: pagb_tree doesn't exist in userspace */
2350 //pag->pagb_tree = RB_ROOT;
2354 else if (!XFS_FORCED_SHUTDOWN(mp
)) {
2355 ASSERT(pag
->pagf_freeblks
== be32_to_cpu(agf
->agf_freeblks
));
2356 ASSERT(pag
->pagf_btreeblks
== be32_to_cpu(agf
->agf_btreeblks
));
2357 ASSERT(pag
->pagf_flcount
== be32_to_cpu(agf
->agf_flcount
));
2358 ASSERT(pag
->pagf_longest
== be32_to_cpu(agf
->agf_longest
));
2359 ASSERT(pag
->pagf_levels
[XFS_BTNUM_BNOi
] ==
2360 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]));
2361 ASSERT(pag
->pagf_levels
[XFS_BTNUM_CNTi
] ==
2362 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]));
2370 * Allocate an extent (variable-size).
2371 * Depending on the allocation type, we either look in a single allocation
2372 * group or loop over the allocation groups to find the result.
2376 xfs_alloc_arg_t
*args
) /* allocation argument structure */
2378 xfs_agblock_t agsize
; /* allocation group size */
2380 int flags
; /* XFS_ALLOC_FLAG_... locking flags */
2381 xfs_extlen_t minleft
;/* minimum left value, temp copy */
2382 xfs_mount_t
*mp
; /* mount structure pointer */
2383 xfs_agnumber_t sagno
; /* starting allocation group number */
2384 xfs_alloctype_t type
; /* input allocation type */
2387 xfs_agnumber_t rotorstep
= xfs_rotorstep
; /* inode32 agf stepper */
2390 type
= args
->otype
= args
->type
;
2391 args
->agbno
= NULLAGBLOCK
;
2393 * Just fix this up, for the case where the last a.g. is shorter
2394 * (or there's only one a.g.) and the caller couldn't easily figure
2395 * that out (xfs_bmap_alloc).
2397 agsize
= mp
->m_sb
.sb_agblocks
;
2398 if (args
->maxlen
> agsize
)
2399 args
->maxlen
= agsize
;
2400 if (args
->alignment
== 0)
2401 args
->alignment
= 1;
2402 ASSERT(XFS_FSB_TO_AGNO(mp
, args
->fsbno
) < mp
->m_sb
.sb_agcount
);
2403 ASSERT(XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) < agsize
);
2404 ASSERT(args
->minlen
<= args
->maxlen
);
2405 ASSERT(args
->minlen
<= agsize
);
2406 ASSERT(args
->mod
< args
->prod
);
2407 if (XFS_FSB_TO_AGNO(mp
, args
->fsbno
) >= mp
->m_sb
.sb_agcount
||
2408 XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) >= agsize
||
2409 args
->minlen
> args
->maxlen
|| args
->minlen
> agsize
||
2410 args
->mod
>= args
->prod
) {
2411 args
->fsbno
= NULLFSBLOCK
;
2412 trace_xfs_alloc_vextent_badargs(args
);
2415 minleft
= args
->minleft
;
2418 case XFS_ALLOCTYPE_THIS_AG
:
2419 case XFS_ALLOCTYPE_NEAR_BNO
:
2420 case XFS_ALLOCTYPE_THIS_BNO
:
2422 * These three force us into a single a.g.
2424 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2425 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2427 error
= xfs_alloc_fix_freelist(args
, 0);
2428 args
->minleft
= minleft
;
2430 trace_xfs_alloc_vextent_nofix(args
);
2434 trace_xfs_alloc_vextent_noagbp(args
);
2437 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2438 if ((error
= xfs_alloc_ag_vextent(args
)))
2441 case XFS_ALLOCTYPE_START_BNO
:
2443 * Try near allocation first, then anywhere-in-ag after
2444 * the first a.g. fails.
2446 if ((args
->userdata
== XFS_ALLOC_INITIAL_USER_DATA
) &&
2447 (mp
->m_flags
& XFS_MOUNT_32BITINODES
)) {
2448 args
->fsbno
= XFS_AGB_TO_FSB(mp
,
2449 ((mp
->m_agfrotor
/ rotorstep
) %
2450 mp
->m_sb
.sb_agcount
), 0);
2453 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2454 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2456 case XFS_ALLOCTYPE_ANY_AG
:
2457 case XFS_ALLOCTYPE_START_AG
:
2458 case XFS_ALLOCTYPE_FIRST_AG
:
2460 * Rotate through the allocation groups looking for a winner.
2462 if (type
== XFS_ALLOCTYPE_ANY_AG
) {
2464 * Start with the last place we left off.
2466 args
->agno
= sagno
= (mp
->m_agfrotor
/ rotorstep
) %
2467 mp
->m_sb
.sb_agcount
;
2468 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2469 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2470 } else if (type
== XFS_ALLOCTYPE_FIRST_AG
) {
2472 * Start with allocation group given by bno.
2474 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2475 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2479 if (type
== XFS_ALLOCTYPE_START_AG
)
2480 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2482 * Start with the given allocation group.
2484 args
->agno
= sagno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2485 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2488 * Loop over allocation groups twice; first time with
2489 * trylock set, second time without.
2492 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2493 if (no_min
) args
->minleft
= 0;
2494 error
= xfs_alloc_fix_freelist(args
, flags
);
2495 args
->minleft
= minleft
;
2497 trace_xfs_alloc_vextent_nofix(args
);
2501 * If we get a buffer back then the allocation will fly.
2504 if ((error
= xfs_alloc_ag_vextent(args
)))
2509 trace_xfs_alloc_vextent_loopfailed(args
);
2512 * Didn't work, figure out the next iteration.
2514 if (args
->agno
== sagno
&&
2515 type
== XFS_ALLOCTYPE_START_BNO
)
2516 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2518 * For the first allocation, we can try any AG to get
2519 * space. However, if we already have allocated a
2520 * block, we don't want to try AGs whose number is below
2521 * sagno. Otherwise, we may end up with out-of-order
2522 * locking of AGF, which might cause deadlock.
2524 if (++(args
->agno
) == mp
->m_sb
.sb_agcount
) {
2525 if (args
->firstblock
!= NULLFSBLOCK
)
2531 * Reached the starting a.g., must either be done
2532 * or switch to non-trylock mode.
2534 if (args
->agno
== sagno
) {
2536 args
->agbno
= NULLAGBLOCK
;
2537 trace_xfs_alloc_vextent_allfailed(args
);
2544 if (type
== XFS_ALLOCTYPE_START_BNO
) {
2545 args
->agbno
= XFS_FSB_TO_AGBNO(mp
,
2547 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2551 xfs_perag_put(args
->pag
);
2553 if (bump_rotor
|| (type
== XFS_ALLOCTYPE_ANY_AG
)) {
2554 if (args
->agno
== sagno
)
2555 mp
->m_agfrotor
= (mp
->m_agfrotor
+ 1) %
2556 (mp
->m_sb
.sb_agcount
* rotorstep
);
2558 mp
->m_agfrotor
= (args
->agno
* rotorstep
+ 1) %
2559 (mp
->m_sb
.sb_agcount
* rotorstep
);
2566 if (args
->agbno
== NULLAGBLOCK
)
2567 args
->fsbno
= NULLFSBLOCK
;
2569 args
->fsbno
= XFS_AGB_TO_FSB(mp
, args
->agno
, args
->agbno
);
2571 ASSERT(args
->len
>= args
->minlen
);
2572 ASSERT(args
->len
<= args
->maxlen
);
2573 ASSERT(args
->agbno
% args
->alignment
== 0);
2574 XFS_AG_CHECK_DADDR(mp
, XFS_FSB_TO_DADDR(mp
, args
->fsbno
),
2578 xfs_perag_put(args
->pag
);
2581 xfs_perag_put(args
->pag
);
2587 * Just break up the extent address and hand off to xfs_free_ag_extent
2588 * after fixing up the freelist.
2592 xfs_trans_t
*tp
, /* transaction pointer */
2593 xfs_fsblock_t bno
, /* starting block number of extent */
2594 xfs_extlen_t len
) /* length of extent */
2596 xfs_alloc_arg_t args
;
2600 memset(&args
, 0, sizeof(xfs_alloc_arg_t
));
2602 args
.mp
= tp
->t_mountp
;
2605 * validate that the block number is legal - the enables us to detect
2606 * and handle a silent filesystem corruption rather than crashing.
2608 args
.agno
= XFS_FSB_TO_AGNO(args
.mp
, bno
);
2609 if (args
.agno
>= args
.mp
->m_sb
.sb_agcount
)
2610 return -EFSCORRUPTED
;
2612 args
.agbno
= XFS_FSB_TO_AGBNO(args
.mp
, bno
);
2613 if (args
.agbno
>= args
.mp
->m_sb
.sb_agblocks
)
2614 return -EFSCORRUPTED
;
2616 args
.pag
= xfs_perag_get(args
.mp
, args
.agno
);
2619 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
2623 /* validate the extent size is legal now we have the agf locked */
2624 if (args
.agbno
+ len
>
2625 be32_to_cpu(XFS_BUF_TO_AGF(args
.agbp
)->agf_length
)) {
2626 error
= -EFSCORRUPTED
;
2630 error
= xfs_free_ag_extent(tp
, args
.agbp
, args
.agno
, args
.agbno
, len
, 0);
2632 xfs_extent_busy_insert(tp
, args
.agno
, args
.agbno
, len
, 0);
2634 xfs_perag_put(args
.pag
);