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 xfs_agfl_read_verify(
437 #ifdef WHEN_CRCS_COME_ALONG
439 * we cannot actually do any verification of the AGFL because mkfs does
440 * not initialise the AGFL to zero or NULL. Hence the only valid part of
441 * the AGFL is what the AGF says is active. We can't get to the AGF, so
442 * we can't verify just those entries are valid.
444 * This problem goes away when the CRC format change comes along as that
445 * requires the AGFL to be initialised by mkfs. At that point, we can
446 * verify the blocks in the agfl -active or not- lie within the bounds
447 * of the AG. Until then, just leave this check ifdef'd out.
449 struct xfs_mount *mp = bp->b_target->bt_mount;
450 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
455 for (i = 0; i < XFS_AGFL_SIZE(mp); i++) {
456 if (be32_to_cpu(agfl->agfl_bno[i]) == NULLAGBLOCK ||
457 be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks)
462 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, agfl);
463 xfs_buf_ioerror(bp, EFSCORRUPTED);
467 xfs_buf_ioend(bp, 0);
471 * Read in the allocation group free block array.
473 STATIC int /* error */
475 xfs_mount_t *mp, /* mount point structure */
476 xfs_trans_t *tp, /* transaction pointer */
477 xfs_agnumber_t agno, /* allocation group number */
478 xfs_buf_t **bpp) /* buffer for the ag free block array */
480 xfs_buf_t *bp; /* return value */
483 ASSERT(agno != NULLAGNUMBER);
484 error = xfs_trans_read_buf(
485 mp, tp, mp->m_ddev_targp,
486 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
487 XFS_FSS_TO_BB(mp, 1), 0, &bp, xfs_agfl_read_verify);
490 ASSERT(!xfs_buf_geterror(bp));
491 xfs_buf_set_ref(bp, XFS_AGFL_REF);
497 xfs_alloc_update_counters(
498 struct xfs_trans *tp,
499 struct xfs_perag *pag,
500 struct xfs_buf *agbp,
503 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
505 pag->pagf_freeblks += len;
506 be32_add_cpu(&agf->agf_freeblks, len);
508 xfs_trans_agblocks_delta(tp, len);
509 if (unlikely(be32_to_cpu(agf->agf_freeblks) >
510 be32_to_cpu(agf->agf_length)))
513 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
518 * Allocation group level functions.
522 * Allocate a variable extent in the allocation group agno.
523 * Type and bno are used to determine where in the allocation group the
525 * Extent's length (returned in *len) will be between minlen and maxlen,
526 * and of the form k * prod + mod unless there's nothing that large.
527 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
529 STATIC int /* error */
530 xfs_alloc_ag_vextent(
531 xfs_alloc_arg_t *args) /* argument structure for allocation */
535 ASSERT(args->minlen > 0);
536 ASSERT(args->maxlen > 0);
537 ASSERT(args->minlen <= args->maxlen);
538 ASSERT(args->mod < args->prod);
539 ASSERT(args->alignment > 0);
541 * Branch to correct routine based on the type.
544 switch (args->type) {
545 case XFS_ALLOCTYPE_THIS_AG:
546 error = xfs_alloc_ag_vextent_size(args);
548 case XFS_ALLOCTYPE_NEAR_BNO:
549 error = xfs_alloc_ag_vextent_near(args);
551 case XFS_ALLOCTYPE_THIS_BNO:
552 error = xfs_alloc_ag_vextent_exact(args);
559 if (error || args->agbno == NULLAGBLOCK)
562 ASSERT(args->len >= args->minlen);
563 ASSERT(args->len <= args->maxlen);
564 ASSERT(!args->wasfromfl || !args->isfl);
565 ASSERT(args->agbno % args->alignment == 0);
567 if (!args->wasfromfl) {
568 error = xfs_alloc_update_counters(args->tp, args->pag,
570 -((long)(args->len)));
574 ASSERT(!xfs_extent_busy_search(args->mp, args->agno,
575 args->agbno, args->len));
579 xfs_trans_mod_sb(args->tp, args->wasdel ?
580 XFS_TRANS_SB_RES_FDBLOCKS :
581 XFS_TRANS_SB_FDBLOCKS,
582 -((long)(args->len)));
585 XFS_STATS_INC(xs_allocx);
586 XFS_STATS_ADD(xs_allocb, args->len);
591 * Allocate a variable extent at exactly agno/bno.
592 * Extent's length (returned in *len) will be between minlen and maxlen,
593 * and of the form k * prod + mod unless there's nothing that large.
594 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
596 STATIC int /* error */
597 xfs_alloc_ag_vextent_exact(
598 xfs_alloc_arg_t *args) /* allocation argument structure */
600 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
601 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
603 xfs_agblock_t fbno; /* start block of found extent */
604 xfs_extlen_t flen; /* length of found extent */
605 xfs_agblock_t tbno; /* start block of trimmed extent */
606 xfs_extlen_t tlen; /* length of trimmed extent */
607 xfs_agblock_t tend; /* end block of trimmed extent */
608 int i; /* success/failure of operation */
610 ASSERT(args->alignment == 1);
613 * Allocate/initialize a cursor for the by-number freespace btree.
615 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
616 args->agno, XFS_BTNUM_BNO);
619 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
620 * Look for the closest free block <= bno, it must contain bno
621 * if any free block does.
623 error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
630 * Grab the freespace record.
632 error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
635 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
636 ASSERT(fbno <= args->agbno);
639 * Check for overlapping busy extents.
641 xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen);
644 * Give up if the start of the extent is busy, or the freespace isn't
645 * long enough for the minimum request.
647 if (tbno > args->agbno)
649 if (tlen < args->minlen)
652 if (tend < args->agbno + args->minlen)
656 * End of extent will be smaller of the freespace end and the
657 * maximal requested end.
659 * Fix the length according to mod and prod if given.
661 args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
663 xfs_alloc_fix_len(args);
664 if (!xfs_alloc_fix_minleft(args))
667 ASSERT(args->agbno + args->len <= tend);
670 * We are allocating agbno for args->len
671 * Allocate/initialize a cursor for the by-size btree.
673 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
674 args->agno, XFS_BTNUM_CNT);
675 ASSERT(args->agbno + args->len <=
676 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
677 error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
678 args->len, XFSA_FIXUP_BNO_OK);
680 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
684 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
685 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
688 trace_xfs_alloc_exact_done(args);
692 /* Didn't find it, return null. */
693 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
694 args->agbno = NULLAGBLOCK;
695 trace_xfs_alloc_exact_notfound(args);
699 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
700 trace_xfs_alloc_exact_error(args);
705 * Search the btree in a given direction via the search cursor and compare
706 * the records found against the good extent we've already found.
709 xfs_alloc_find_best_extent(
710 struct xfs_alloc_arg *args, /* allocation argument structure */
711 struct xfs_btree_cur **gcur, /* good cursor */
712 struct xfs_btree_cur **scur, /* searching cursor */
713 xfs_agblock_t gdiff, /* difference for search comparison */
714 xfs_agblock_t *sbno, /* extent found by search */
715 xfs_extlen_t *slen, /* extent length */
716 xfs_agblock_t *sbnoa, /* aligned extent found by search */
717 xfs_extlen_t *slena, /* aligned extent length */
718 int dir) /* 0 = search right, 1 = search left */
725 /* The good extent is perfect, no need to search. */
730 * Look until we find a better one, run out of space or run off the end.
733 error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
736 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
737 xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena);
740 * The good extent is closer than this one.
743 if (*sbnoa >= args->agbno + gdiff)
746 if (*sbnoa <= args->agbno - gdiff)
751 * Same distance, compare length and pick the best.
753 if (*slena >= args->minlen) {
754 args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
755 xfs_alloc_fix_len(args);
757 sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
758 args->alignment, *sbnoa,
762 * Choose closer size and invalidate other cursor.
770 error = xfs_btree_increment(*scur, 0, &i);
772 error = xfs_btree_decrement(*scur, 0, &i);
778 xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
783 xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
788 /* caller invalidates cursors */
793 * Allocate a variable extent near bno in the allocation group agno.
794 * Extent's length (returned in len) will be between minlen and maxlen,
795 * and of the form k * prod + mod unless there's nothing that large.
796 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
798 STATIC int /* error */
799 xfs_alloc_ag_vextent_near(
800 xfs_alloc_arg_t *args) /* allocation argument structure */
802 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
803 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
804 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
805 xfs_agblock_t gtbno; /* start bno of right side entry */
806 xfs_agblock_t gtbnoa; /* aligned ... */
807 xfs_extlen_t gtdiff; /* difference to right side entry */
808 xfs_extlen_t gtlen; /* length of right side entry */
809 xfs_extlen_t gtlena; /* aligned ... */
810 xfs_agblock_t gtnew; /* useful start bno of right side */
811 int error; /* error code */
812 int i; /* result code, temporary */
813 int j; /* result code, temporary */
814 xfs_agblock_t ltbno; /* start bno of left side entry */
815 xfs_agblock_t ltbnoa; /* aligned ... */
816 xfs_extlen_t ltdiff; /* difference to left side entry */
817 xfs_extlen_t ltlen; /* length of left side entry */
818 xfs_extlen_t ltlena; /* aligned ... */
819 xfs_agblock_t ltnew; /* useful start bno of left side */
820 xfs_extlen_t rlen; /* length of returned extent */
822 #if defined(DEBUG) && defined(__KERNEL__)
824 * Randomly don't execute the first algorithm.
826 int dofirst; /* set to do first algorithm */
828 dofirst = random32() & 1;
839 * Get a cursor for the by-size btree.
841 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
842 args->agno, XFS_BTNUM_CNT);
845 * See if there are any free extents as big as maxlen.
847 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
850 * If none, then pick up the last entry in the tree unless the
854 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
857 if (i == 0 || ltlen == 0) {
858 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
859 trace_xfs_alloc_near_noentry(args);
868 * If the requested extent is large wrt the freespaces available
869 * in this a.g., then the cursor will be pointing to a btree entry
870 * near the right edge of the tree. If it's in the last btree leaf
871 * block, then we just examine all the entries in that block
872 * that are big enough, and pick the best one.
873 * This is written as a while loop so we can break out of it,
874 * but we never loop back to the top.
876 while (xfs_btree_islastblock(cnt_cur, 0)) {
880 xfs_agblock_t bnew=0;
882 #if defined(DEBUG) && defined(__KERNEL__)
887 * Start from the entry that lookup found, sequence through
888 * all larger free blocks. If we're actually pointing at a
889 * record smaller than maxlen, go to the start of this block,
890 * and skip all those smaller than minlen.
892 if (ltlen || args->alignment > 1) {
893 cnt_cur->bc_ptrs[0] = 1;
895 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
898 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
899 if (ltlen >= args->minlen)
901 if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
904 ASSERT(ltlen >= args->minlen);
908 i = cnt_cur->bc_ptrs[0];
909 for (j = 1, blen = 0, bdiff = 0;
910 !error && j && (blen < args->maxlen || bdiff > 0);
911 error = xfs_btree_increment(cnt_cur, 0, &j)) {
913 * For each entry, decide if it's better than
914 * the previous best entry.
916 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
918 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
919 xfs_alloc_compute_aligned(args, ltbno, ltlen,
921 if (ltlena < args->minlen)
923 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
924 xfs_alloc_fix_len(args);
925 ASSERT(args->len >= args->minlen);
926 if (args->len < blen)
928 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
929 args->alignment, ltbnoa, ltlena, <new);
930 if (ltnew != NULLAGBLOCK &&
931 (args->len > blen || ltdiff < bdiff)) {
935 besti = cnt_cur->bc_ptrs[0];
939 * It didn't work. We COULD be in a case where
940 * there's a good record somewhere, so try again.
945 * Point at the best entry, and retrieve it again.
947 cnt_cur->bc_ptrs[0] = besti;
948 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
950 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
951 ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
953 if (!xfs_alloc_fix_minleft(args)) {
954 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
955 trace_xfs_alloc_near_nominleft(args);
960 * We are allocating starting at bnew for blen blocks.
963 ASSERT(bnew >= ltbno);
964 ASSERT(bnew + blen <= ltbno + ltlen);
966 * Set up a cursor for the by-bno tree.
968 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
969 args->agbp, args->agno, XFS_BTNUM_BNO);
971 * Fix up the btree entries.
973 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
974 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
976 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
977 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
979 trace_xfs_alloc_near_first(args);
984 * Search in the by-bno tree to the left and to the right
985 * simultaneously, until in each case we find a space big enough,
986 * or run into the edge of the tree. When we run into the edge,
987 * we deallocate that cursor.
988 * If both searches succeed, we compare the two spaces and pick
990 * With alignment, it's possible for both to fail; the upper
991 * level algorithm that picks allocation groups for allocations
992 * is not supposed to do this.
995 * Allocate and initialize the cursor for the leftward search.
997 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
998 args->agno, XFS_BTNUM_BNO);
1000 * Lookup <= bno to find the leftward search's starting point.
1002 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
1006 * Didn't find anything; use this cursor for the rightward
1009 bno_cur_gt = bno_cur_lt;
1013 * Found something. Duplicate the cursor for the rightward search.
1015 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
1018 * Increment the cursor, so we will point at the entry just right
1019 * of the leftward entry if any, or to the leftmost entry.
1021 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1025 * It failed, there are no rightward entries.
1027 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
1031 * Loop going left with the leftward cursor, right with the
1032 * rightward cursor, until either both directions give up or
1033 * we find an entry at least as big as minlen.
1037 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1039 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1040 xfs_alloc_compute_aligned(args, ltbno, ltlen,
1042 if (ltlena >= args->minlen)
1044 if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1047 xfs_btree_del_cursor(bno_cur_lt,
1053 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1055 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1056 xfs_alloc_compute_aligned(args, gtbno, gtlen,
1058 if (gtlena >= args->minlen)
1060 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1063 xfs_btree_del_cursor(bno_cur_gt,
1068 } while (bno_cur_lt || bno_cur_gt);
1071 * Got both cursors still active, need to find better entry.
1073 if (bno_cur_lt && bno_cur_gt) {
1074 if (ltlena >= args->minlen) {
1076 * Left side is good, look for a right side entry.
1078 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1079 xfs_alloc_fix_len(args);
1080 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1081 args->alignment, ltbnoa, ltlena, <new);
1083 error = xfs_alloc_find_best_extent(args,
1084 &bno_cur_lt, &bno_cur_gt,
1085 ltdiff, >bno, >len,
1087 0 /* search right */);
1089 ASSERT(gtlena >= args->minlen);
1092 * Right side is good, look for a left side entry.
1094 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1095 xfs_alloc_fix_len(args);
1096 gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1097 args->alignment, gtbnoa, gtlena, >new);
1099 error = xfs_alloc_find_best_extent(args,
1100 &bno_cur_gt, &bno_cur_lt,
1101 gtdiff, <bno, <len,
1103 1 /* search left */);
1111 * If we couldn't get anything, give up.
1113 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1114 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1117 trace_xfs_alloc_near_busy(args);
1118 xfs_log_force(args->mp, XFS_LOG_SYNC);
1121 trace_xfs_alloc_size_neither(args);
1122 args->agbno = NULLAGBLOCK;
1127 * At this point we have selected a freespace entry, either to the
1128 * left or to the right. If it's on the right, copy all the
1129 * useful variables to the "left" set so we only have one
1130 * copy of this code.
1133 bno_cur_lt = bno_cur_gt;
1144 * Fix up the length and compute the useful address.
1146 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1147 xfs_alloc_fix_len(args);
1148 if (!xfs_alloc_fix_minleft(args)) {
1149 trace_xfs_alloc_near_nominleft(args);
1150 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1151 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1155 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1156 ltbnoa, ltlena, <new);
1157 ASSERT(ltnew >= ltbno);
1158 ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1159 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1160 args->agbno = ltnew;
1162 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1163 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1167 trace_xfs_alloc_near_greater(args);
1169 trace_xfs_alloc_near_lesser(args);
1171 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1172 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1176 trace_xfs_alloc_near_error(args);
1177 if (cnt_cur != NULL)
1178 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1179 if (bno_cur_lt != NULL)
1180 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1181 if (bno_cur_gt != NULL)
1182 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1187 * Allocate a variable extent anywhere in the allocation group agno.
1188 * Extent's length (returned in len) will be between minlen and maxlen,
1189 * and of the form k * prod + mod unless there's nothing that large.
1190 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1192 STATIC int /* error */
1193 xfs_alloc_ag_vextent_size(
1194 xfs_alloc_arg_t *args) /* allocation argument structure */
1196 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1197 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1198 int error; /* error result */
1199 xfs_agblock_t fbno; /* start of found freespace */
1200 xfs_extlen_t flen; /* length of found freespace */
1201 int i; /* temp status variable */
1202 xfs_agblock_t rbno; /* returned block number */
1203 xfs_extlen_t rlen; /* length of returned extent */
1208 * Allocate and initialize a cursor for the by-size btree.
1210 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1211 args->agno, XFS_BTNUM_CNT);
1215 * Look for an entry >= maxlen+alignment-1 blocks.
1217 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1218 args->maxlen + args->alignment - 1, &i)))
1222 * If none or we have busy extents that we cannot allocate from, then
1223 * we have to settle for a smaller extent. In the case that there are
1224 * no large extents, this will return the last entry in the tree unless
1225 * the tree is empty. In the case that there are only busy large
1226 * extents, this will return the largest small extent unless there
1227 * are no smaller extents available.
1229 if (!i || forced > 1) {
1230 error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1234 if (i == 0 || flen == 0) {
1235 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1236 trace_xfs_alloc_size_noentry(args);
1240 xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen);
1243 * Search for a non-busy extent that is large enough.
1244 * If we are at low space, don't check, or if we fall of
1245 * the end of the btree, turn off the busy check and
1249 error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1252 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1254 xfs_alloc_compute_aligned(args, fbno, flen,
1257 if (rlen >= args->maxlen)
1260 error = xfs_btree_increment(cnt_cur, 0, &i);
1265 * Our only valid extents must have been busy.
1266 * Make it unbusy by forcing the log out and
1267 * retrying. If we've been here before, forcing
1268 * the log isn't making the extents available,
1269 * which means they have probably been freed in
1270 * this transaction. In that case, we have to
1271 * give up on them and we'll attempt a minlen
1272 * allocation the next time around.
1274 xfs_btree_del_cursor(cnt_cur,
1276 trace_xfs_alloc_size_busy(args);
1278 xfs_log_force(args->mp, XFS_LOG_SYNC);
1285 * In the first case above, we got the last entry in the
1286 * by-size btree. Now we check to see if the space hits maxlen
1287 * once aligned; if not, we search left for something better.
1288 * This can't happen in the second case above.
1290 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1291 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1292 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1293 if (rlen < args->maxlen) {
1294 xfs_agblock_t bestfbno;
1295 xfs_extlen_t bestflen;
1296 xfs_agblock_t bestrbno;
1297 xfs_extlen_t bestrlen;
1304 if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1308 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1311 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1312 if (flen < bestrlen)
1314 xfs_alloc_compute_aligned(args, fbno, flen,
1316 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1317 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1318 (rlen <= flen && rbno + rlen <= fbno + flen),
1320 if (rlen > bestrlen) {
1325 if (rlen == args->maxlen)
1329 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1332 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1338 args->wasfromfl = 0;
1340 * Fix up the length.
1343 if (rlen < args->minlen) {
1345 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1346 trace_xfs_alloc_size_busy(args);
1347 xfs_log_force(args->mp, XFS_LOG_SYNC);
1352 xfs_alloc_fix_len(args);
1354 if (!xfs_alloc_fix_minleft(args))
1357 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1359 * Allocate and initialize a cursor for the by-block tree.
1361 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1362 args->agno, XFS_BTNUM_BNO);
1363 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1364 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1366 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1367 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1368 cnt_cur = bno_cur = NULL;
1371 XFS_WANT_CORRUPTED_GOTO(
1372 args->agbno + args->len <=
1373 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1375 trace_xfs_alloc_size_done(args);
1379 trace_xfs_alloc_size_error(args);
1381 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1383 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1387 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1388 trace_xfs_alloc_size_nominleft(args);
1389 args->agbno = NULLAGBLOCK;
1394 * Deal with the case where only small freespaces remain.
1395 * Either return the contents of the last freespace record,
1396 * or allocate space from the freelist if there is nothing in the tree.
1398 STATIC int /* error */
1399 xfs_alloc_ag_vextent_small(
1400 xfs_alloc_arg_t *args, /* allocation argument structure */
1401 xfs_btree_cur_t *ccur, /* by-size cursor */
1402 xfs_agblock_t *fbnop, /* result block number */
1403 xfs_extlen_t *flenp, /* result length */
1404 int *stat) /* status: 0-freelist, 1-normal/none */
1411 if ((error = xfs_btree_decrement(ccur, 0, &i)))
1414 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1416 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1419 * Nothing in the btree, try the freelist. Make sure
1420 * to respect minleft even when pulling from the
1423 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1424 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1426 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1429 if (fbno != NULLAGBLOCK) {
1430 xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1,
1433 if (args->userdata) {
1436 bp = xfs_btree_get_bufs(args->mp, args->tp,
1437 args->agno, fbno, 0);
1438 xfs_trans_binval(args->tp, bp);
1442 XFS_WANT_CORRUPTED_GOTO(
1443 args->agbno + args->len <=
1444 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1446 args->wasfromfl = 1;
1447 trace_xfs_alloc_small_freelist(args);
1452 * Nothing in the freelist.
1458 * Can't allocate from the freelist for some reason.
1465 * Can't do the allocation, give up.
1467 if (flen < args->minlen) {
1468 args->agbno = NULLAGBLOCK;
1469 trace_xfs_alloc_small_notenough(args);
1475 trace_xfs_alloc_small_done(args);
1479 trace_xfs_alloc_small_error(args);
1484 * Free the extent starting at agno/bno for length.
1486 STATIC int /* error */
1488 xfs_trans_t *tp, /* transaction pointer */
1489 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1490 xfs_agnumber_t agno, /* allocation group number */
1491 xfs_agblock_t bno, /* starting block number */
1492 xfs_extlen_t len, /* length of extent */
1493 int isfl) /* set if is freelist blocks - no sb acctg */
1495 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1496 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1497 int error; /* error return value */
1498 xfs_agblock_t gtbno; /* start of right neighbor block */
1499 xfs_extlen_t gtlen; /* length of right neighbor block */
1500 int haveleft; /* have a left neighbor block */
1501 int haveright; /* have a right neighbor block */
1502 int i; /* temp, result code */
1503 xfs_agblock_t ltbno; /* start of left neighbor block */
1504 xfs_extlen_t ltlen; /* length of left neighbor block */
1505 xfs_mount_t *mp; /* mount point struct for filesystem */
1506 xfs_agblock_t nbno; /* new starting block of freespace */
1507 xfs_extlen_t nlen; /* new length of freespace */
1508 xfs_perag_t *pag; /* per allocation group data */
1512 * Allocate and initialize a cursor for the by-block btree.
1514 bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1517 * Look for a neighboring block on the left (lower block numbers)
1518 * that is contiguous with this space.
1520 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1524 * There is a block to our left.
1526 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1528 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1530 * It's not contiguous, though.
1532 if (ltbno + ltlen < bno)
1536 * If this failure happens the request to free this
1537 * space was invalid, it's (partly) already free.
1540 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1544 * Look for a neighboring block on the right (higher block numbers)
1545 * that is contiguous with this space.
1547 if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1551 * There is a block to our right.
1553 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1555 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1557 * It's not contiguous, though.
1559 if (bno + len < gtbno)
1563 * If this failure happens the request to free this
1564 * space was invalid, it's (partly) already free.
1567 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1571 * Now allocate and initialize a cursor for the by-size tree.
1573 cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1575 * Have both left and right contiguous neighbors.
1576 * Merge all three into a single free block.
1578 if (haveleft && haveright) {
1580 * Delete the old by-size entry on the left.
1582 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1584 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1585 if ((error = xfs_btree_delete(cnt_cur, &i)))
1587 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1589 * Delete the old by-size entry on the right.
1591 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1593 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1594 if ((error = xfs_btree_delete(cnt_cur, &i)))
1596 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1598 * Delete the old by-block entry for the right block.
1600 if ((error = xfs_btree_delete(bno_cur, &i)))
1602 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1604 * Move the by-block cursor back to the left neighbor.
1606 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1608 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1611 * Check that this is the right record: delete didn't
1612 * mangle the cursor.
1615 xfs_agblock_t xxbno;
1618 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1621 XFS_WANT_CORRUPTED_GOTO(
1622 i == 1 && xxbno == ltbno && xxlen == ltlen,
1627 * Update remaining by-block entry to the new, joined block.
1630 nlen = len + ltlen + gtlen;
1631 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1635 * Have only a left contiguous neighbor.
1636 * Merge it together with the new freespace.
1638 else if (haveleft) {
1640 * Delete the old by-size entry on the left.
1642 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1644 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1645 if ((error = xfs_btree_delete(cnt_cur, &i)))
1647 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1649 * Back up the by-block cursor to the left neighbor, and
1650 * update its length.
1652 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1654 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1657 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1661 * Have only a right contiguous neighbor.
1662 * Merge it together with the new freespace.
1664 else if (haveright) {
1666 * Delete the old by-size entry on the right.
1668 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1670 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1671 if ((error = xfs_btree_delete(cnt_cur, &i)))
1673 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1675 * Update the starting block and length of the right
1676 * neighbor in the by-block tree.
1680 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1684 * No contiguous neighbors.
1685 * Insert the new freespace into the by-block tree.
1690 if ((error = xfs_btree_insert(bno_cur, &i)))
1692 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1694 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1697 * In all cases we need to insert the new freespace in the by-size tree.
1699 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1701 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1702 if ((error = xfs_btree_insert(cnt_cur, &i)))
1704 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1705 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1709 * Update the freespace totals in the ag and superblock.
1711 pag = xfs_perag_get(mp, agno);
1712 error = xfs_alloc_update_counters(tp, pag, agbp, len);
1718 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1719 XFS_STATS_INC(xs_freex);
1720 XFS_STATS_ADD(xs_freeb, len);
1722 trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright);
1727 trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1);
1729 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1731 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1736 * Visible (exported) allocation/free functions.
1737 * Some of these are used just by xfs_alloc_btree.c and this file.
1741 * Compute and fill in value of m_ag_maxlevels.
1744 xfs_alloc_compute_maxlevels(
1745 xfs_mount_t *mp) /* file system mount structure */
1753 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1754 minleafrecs = mp->m_alloc_mnr[0];
1755 minnoderecs = mp->m_alloc_mnr[1];
1756 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1757 for (level = 1; maxblocks > 1; level++)
1758 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1759 mp->m_ag_maxlevels = level;
1763 * Find the length of the longest extent in an AG.
1766 xfs_alloc_longest_free_extent(
1767 struct xfs_mount *mp,
1768 struct xfs_perag *pag)
1770 xfs_extlen_t need, delta = 0;
1772 need = XFS_MIN_FREELIST_PAG(pag, mp);
1773 if (need > pag->pagf_flcount)
1774 delta = need - pag->pagf_flcount;
1776 if (pag->pagf_longest > delta)
1777 return pag->pagf_longest - delta;
1778 return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
1782 * Decide whether to use this allocation group for this allocation.
1783 * If so, fix up the btree freelist's size.
1785 STATIC int /* error */
1786 xfs_alloc_fix_freelist(
1787 xfs_alloc_arg_t *args, /* allocation argument structure */
1788 int flags) /* XFS_ALLOC_FLAG_... */
1790 xfs_buf_t *agbp; /* agf buffer pointer */
1791 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1792 xfs_buf_t *agflbp;/* agfl buffer pointer */
1793 xfs_agblock_t bno; /* freelist block */
1794 xfs_extlen_t delta; /* new blocks needed in freelist */
1795 int error; /* error result code */
1796 xfs_extlen_t longest;/* longest extent in allocation group */
1797 xfs_mount_t *mp; /* file system mount point structure */
1798 xfs_extlen_t need; /* total blocks needed in freelist */
1799 xfs_perag_t *pag; /* per-ag information structure */
1800 xfs_alloc_arg_t targs; /* local allocation arguments */
1801 xfs_trans_t *tp; /* transaction pointer */
1807 if (!pag->pagf_init) {
1808 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1811 if (!pag->pagf_init) {
1812 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1813 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1821 * If this is a metadata preferred pag and we are user data
1822 * then try somewhere else if we are not being asked to
1823 * try harder at this point
1825 if (pag->pagf_metadata && args->userdata &&
1826 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1827 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1832 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1834 * If it looks like there isn't a long enough extent, or enough
1835 * total blocks, reject it.
1837 need = XFS_MIN_FREELIST_PAG(pag, mp);
1838 longest = xfs_alloc_longest_free_extent(mp, pag);
1839 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1841 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1842 need - args->total) < (int)args->minleft)) {
1844 xfs_trans_brelse(tp, agbp);
1851 * Get the a.g. freespace buffer.
1852 * Can fail if we're not blocking on locks, and it's held.
1855 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1859 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1860 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1866 * Figure out how many blocks we should have in the freelist.
1868 agf = XFS_BUF_TO_AGF(agbp);
1869 need = XFS_MIN_FREELIST(agf, mp);
1871 * If there isn't enough total or single-extent, reject it.
1873 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1874 delta = need > be32_to_cpu(agf->agf_flcount) ?
1875 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1876 longest = be32_to_cpu(agf->agf_longest);
1877 longest = (longest > delta) ? (longest - delta) :
1878 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1879 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1881 ((int)(be32_to_cpu(agf->agf_freeblks) +
1882 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1883 (int)args->minleft)) {
1884 xfs_trans_brelse(tp, agbp);
1890 * Make the freelist shorter if it's too long.
1892 while (be32_to_cpu(agf->agf_flcount) > need) {
1895 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1898 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1900 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1901 xfs_trans_binval(tp, bp);
1904 * Initialize the args structure.
1906 memset(&targs, 0, sizeof(targs));
1910 targs.agno = args->agno;
1911 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1912 targs.minalignslop = 0;
1913 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1914 targs.type = XFS_ALLOCTYPE_THIS_AG;
1916 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1919 * Make the freelist longer if it's too short.
1921 while (be32_to_cpu(agf->agf_flcount) < need) {
1923 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1925 * Allocate as many blocks as possible at once.
1927 if ((error = xfs_alloc_ag_vextent(&targs))) {
1928 xfs_trans_brelse(tp, agflbp);
1932 * Stop if we run out. Won't happen if callers are obeying
1933 * the restrictions correctly. Can happen for free calls
1934 * on a completely full ag.
1936 if (targs.agbno == NULLAGBLOCK) {
1937 if (flags & XFS_ALLOC_FLAG_FREEING)
1939 xfs_trans_brelse(tp, agflbp);
1944 * Put each allocated block on the list.
1946 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1947 error = xfs_alloc_put_freelist(tp, agbp,
1953 xfs_trans_brelse(tp, agflbp);
1959 * Get a block from the freelist.
1960 * Returns with the buffer for the block gotten.
1963 xfs_alloc_get_freelist(
1964 xfs_trans_t *tp, /* transaction pointer */
1965 xfs_buf_t *agbp, /* buffer containing the agf structure */
1966 xfs_agblock_t *bnop, /* block address retrieved from freelist */
1967 int btreeblk) /* destination is a AGF btree */
1969 xfs_agf_t *agf; /* a.g. freespace structure */
1970 xfs_agfl_t *agfl; /* a.g. freelist structure */
1971 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1972 xfs_agblock_t bno; /* block number returned */
1975 xfs_mount_t *mp; /* mount structure */
1976 xfs_perag_t *pag; /* per allocation group data */
1978 agf = XFS_BUF_TO_AGF(agbp);
1980 * Freelist is empty, give up.
1982 if (!agf->agf_flcount) {
1983 *bnop = NULLAGBLOCK;
1987 * Read the array of free blocks.
1990 if ((error = xfs_alloc_read_agfl(mp, tp,
1991 be32_to_cpu(agf->agf_seqno), &agflbp)))
1993 agfl = XFS_BUF_TO_AGFL(agflbp);
1995 * Get the block number and update the data structures.
1997 bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
1998 be32_add_cpu(&agf->agf_flfirst, 1);
1999 xfs_trans_brelse(tp, agflbp);
2000 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
2001 agf->agf_flfirst = 0;
2003 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2004 be32_add_cpu(&agf->agf_flcount, -1);
2005 xfs_trans_agflist_delta(tp, -1);
2006 pag->pagf_flcount--;
2009 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2011 be32_add_cpu(&agf->agf_btreeblks, 1);
2012 pag->pagf_btreeblks++;
2013 logflags |= XFS_AGF_BTREEBLKS;
2016 xfs_alloc_log_agf(tp, agbp, logflags);
2023 * Log the given fields from the agf structure.
2027 xfs_trans_t *tp, /* transaction pointer */
2028 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2029 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2031 int first; /* first byte offset */
2032 int last; /* last byte offset */
2033 static const short offsets[] = {
2034 offsetof(xfs_agf_t, agf_magicnum),
2035 offsetof(xfs_agf_t, agf_versionnum),
2036 offsetof(xfs_agf_t, agf_seqno),
2037 offsetof(xfs_agf_t, agf_length),
2038 offsetof(xfs_agf_t, agf_roots[0]),
2039 offsetof(xfs_agf_t, agf_levels[0]),
2040 offsetof(xfs_agf_t, agf_flfirst),
2041 offsetof(xfs_agf_t, agf_fllast),
2042 offsetof(xfs_agf_t, agf_flcount),
2043 offsetof(xfs_agf_t, agf_freeblks),
2044 offsetof(xfs_agf_t, agf_longest),
2045 offsetof(xfs_agf_t, agf_btreeblks),
2049 trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2051 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2052 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2056 * Interface for inode allocation to force the pag data to be initialized.
2059 xfs_alloc_pagf_init(
2060 xfs_mount_t *mp, /* file system mount structure */
2061 xfs_trans_t *tp, /* transaction pointer */
2062 xfs_agnumber_t agno, /* allocation group number */
2063 int flags) /* XFS_ALLOC_FLAGS_... */
2068 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2071 xfs_trans_brelse(tp, bp);
2076 * Put the block on the freelist for the allocation group.
2079 xfs_alloc_put_freelist(
2080 xfs_trans_t *tp, /* transaction pointer */
2081 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2082 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2083 xfs_agblock_t bno, /* block being freed */
2084 int btreeblk) /* block came from a AGF btree */
2086 xfs_agf_t *agf; /* a.g. freespace structure */
2087 xfs_agfl_t *agfl; /* a.g. free block array */
2088 __be32 *blockp;/* pointer to array entry */
2091 xfs_mount_t *mp; /* mount structure */
2092 xfs_perag_t *pag; /* per allocation group data */
2094 agf = XFS_BUF_TO_AGF(agbp);
2097 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2098 be32_to_cpu(agf->agf_seqno), &agflbp)))
2100 agfl = XFS_BUF_TO_AGFL(agflbp);
2101 be32_add_cpu(&agf->agf_fllast, 1);
2102 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2103 agf->agf_fllast = 0;
2105 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2106 be32_add_cpu(&agf->agf_flcount, 1);
2107 xfs_trans_agflist_delta(tp, 1);
2108 pag->pagf_flcount++;
2110 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2112 be32_add_cpu(&agf->agf_btreeblks, -1);
2113 pag->pagf_btreeblks--;
2114 logflags |= XFS_AGF_BTREEBLKS;
2118 xfs_alloc_log_agf(tp, agbp, logflags);
2120 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2121 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
2122 *blockp = cpu_to_be32(bno);
2123 xfs_alloc_log_agf(tp, agbp, logflags);
2124 xfs_trans_log_buf(tp, agflbp,
2125 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2126 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2127 sizeof(xfs_agblock_t) - 1));
2132 xfs_agf_read_verify(
2135 struct xfs_mount *mp = bp->b_target->bt_mount;
2136 struct xfs_agf *agf;
2139 agf = XFS_BUF_TO_AGF(bp);
2141 agf_ok = agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) &&
2142 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2143 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2144 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2145 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2146 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp);
2149 * during growfs operations, the perag is not fully initialised,
2150 * so we can't use it for any useful checking. growfs ensures we can't
2151 * use it by using uncached buffers that don't have the perag attached
2152 * so we can detect and avoid this problem.
2155 agf_ok = agf_ok && be32_to_cpu(agf->agf_seqno) ==
2156 bp->b_pag->pag_agno;
2158 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
2159 agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <=
2160 be32_to_cpu(agf->agf_length);
2162 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2163 XFS_RANDOM_ALLOC_READ_AGF))) {
2164 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, agf);
2165 xfs_buf_ioerror(bp, EFSCORRUPTED);
2168 bp->b_iodone = NULL;
2169 xfs_buf_ioend(bp, 0);
2173 * Read in the allocation group header (free/alloc section).
2177 struct xfs_mount *mp, /* mount point structure */
2178 struct xfs_trans *tp, /* transaction pointer */
2179 xfs_agnumber_t agno, /* allocation group number */
2180 int flags, /* XFS_BUF_ */
2181 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2185 ASSERT(agno != NULLAGNUMBER);
2186 error = xfs_trans_read_buf(
2187 mp, tp, mp->m_ddev_targp,
2188 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2189 XFS_FSS_TO_BB(mp, 1), flags, bpp, xfs_agf_read_verify);
2195 ASSERT(!(*bpp)->b_error);
2196 xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2201 * Read in the allocation group header (free/alloc section).
2205 struct xfs_mount *mp, /* mount point structure */
2206 struct xfs_trans *tp, /* transaction pointer */
2207 xfs_agnumber_t agno, /* allocation group number */
2208 int flags, /* XFS_ALLOC_FLAG_... */
2209 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2211 struct xfs_agf *agf; /* ag freelist header */
2212 struct xfs_perag *pag; /* per allocation group data */
2215 ASSERT(agno != NULLAGNUMBER);
2217 error = xfs_read_agf(mp, tp, agno,
2218 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2224 ASSERT(!(*bpp)->b_error);
2226 agf = XFS_BUF_TO_AGF(*bpp);
2227 pag = xfs_perag_get(mp, agno);
2228 if (!pag->pagf_init) {
2229 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2230 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2231 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2232 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2233 pag->pagf_levels[XFS_BTNUM_BNOi] =
2234 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2235 pag->pagf_levels[XFS_BTNUM_CNTi] =
2236 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2237 spin_lock_init(&pag->pagb_lock);
2238 pag->pagb_count = 0;
2239 pag->pagb_tree = RB_ROOT;
2243 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2244 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2245 ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2246 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2247 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2248 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2249 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2250 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2251 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2259 * Allocate an extent (variable-size).
2260 * Depending on the allocation type, we either look in a single allocation
2261 * group or loop over the allocation groups to find the result.
2265 xfs_alloc_arg_t *args) /* allocation argument structure */
2267 xfs_agblock_t agsize; /* allocation group size */
2269 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2270 xfs_extlen_t minleft;/* minimum left value, temp copy */
2271 xfs_mount_t *mp; /* mount structure pointer */
2272 xfs_agnumber_t sagno; /* starting allocation group number */
2273 xfs_alloctype_t type; /* input allocation type */
2276 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2279 type = args->otype = args->type;
2280 args->agbno = NULLAGBLOCK;
2282 * Just fix this up, for the case where the last a.g. is shorter
2283 * (or there's only one a.g.) and the caller couldn't easily figure
2284 * that out (xfs_bmap_alloc).
2286 agsize = mp->m_sb.sb_agblocks;
2287 if (args->maxlen > agsize)
2288 args->maxlen = agsize;
2289 if (args->alignment == 0)
2290 args->alignment = 1;
2291 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2292 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2293 ASSERT(args->minlen <= args->maxlen);
2294 ASSERT(args->minlen <= agsize);
2295 ASSERT(args->mod < args->prod);
2296 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2297 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2298 args->minlen > args->maxlen || args->minlen > agsize ||
2299 args->mod >= args->prod) {
2300 args->fsbno = NULLFSBLOCK;
2301 trace_xfs_alloc_vextent_badargs(args);
2304 minleft = args->minleft;
2307 case XFS_ALLOCTYPE_THIS_AG:
2308 case XFS_ALLOCTYPE_NEAR_BNO:
2309 case XFS_ALLOCTYPE_THIS_BNO:
2311 * These three force us into a single a.g.
2313 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2314 args->pag = xfs_perag_get(mp, args->agno);
2316 error = xfs_alloc_fix_freelist(args, 0);
2317 args->minleft = minleft;
2319 trace_xfs_alloc_vextent_nofix(args);
2323 trace_xfs_alloc_vextent_noagbp(args);
2326 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2327 if ((error = xfs_alloc_ag_vextent(args)))
2330 case XFS_ALLOCTYPE_START_BNO:
2332 * Try near allocation first, then anywhere-in-ag after
2333 * the first a.g. fails.
2335 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2336 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2337 args->fsbno = XFS_AGB_TO_FSB(mp,
2338 ((mp->m_agfrotor / rotorstep) %
2339 mp->m_sb.sb_agcount), 0);
2342 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2343 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2345 case XFS_ALLOCTYPE_ANY_AG:
2346 case XFS_ALLOCTYPE_START_AG:
2347 case XFS_ALLOCTYPE_FIRST_AG:
2349 * Rotate through the allocation groups looking for a winner.
2351 if (type == XFS_ALLOCTYPE_ANY_AG) {
2353 * Start with the last place we left off.
2355 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2356 mp->m_sb.sb_agcount;
2357 args->type = XFS_ALLOCTYPE_THIS_AG;
2358 flags = XFS_ALLOC_FLAG_TRYLOCK;
2359 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2361 * Start with allocation group given by bno.
2363 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2364 args->type = XFS_ALLOCTYPE_THIS_AG;
2368 if (type == XFS_ALLOCTYPE_START_AG)
2369 args->type = XFS_ALLOCTYPE_THIS_AG;
2371 * Start with the given allocation group.
2373 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2374 flags = XFS_ALLOC_FLAG_TRYLOCK;
2377 * Loop over allocation groups twice; first time with
2378 * trylock set, second time without.
2381 args->pag = xfs_perag_get(mp, args->agno);
2382 if (no_min) args->minleft = 0;
2383 error = xfs_alloc_fix_freelist(args, flags);
2384 args->minleft = minleft;
2386 trace_xfs_alloc_vextent_nofix(args);
2390 * If we get a buffer back then the allocation will fly.
2393 if ((error = xfs_alloc_ag_vextent(args)))
2398 trace_xfs_alloc_vextent_loopfailed(args);
2401 * Didn't work, figure out the next iteration.
2403 if (args->agno == sagno &&
2404 type == XFS_ALLOCTYPE_START_BNO)
2405 args->type = XFS_ALLOCTYPE_THIS_AG;
2407 * For the first allocation, we can try any AG to get
2408 * space. However, if we already have allocated a
2409 * block, we don't want to try AGs whose number is below
2410 * sagno. Otherwise, we may end up with out-of-order
2411 * locking of AGF, which might cause deadlock.
2413 if (++(args->agno) == mp->m_sb.sb_agcount) {
2414 if (args->firstblock != NULLFSBLOCK)
2420 * Reached the starting a.g., must either be done
2421 * or switch to non-trylock mode.
2423 if (args->agno == sagno) {
2425 args->agbno = NULLAGBLOCK;
2426 trace_xfs_alloc_vextent_allfailed(args);
2433 if (type == XFS_ALLOCTYPE_START_BNO) {
2434 args->agbno = XFS_FSB_TO_AGBNO(mp,
2436 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2440 xfs_perag_put(args->pag);
2442 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2443 if (args->agno == sagno)
2444 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2445 (mp->m_sb.sb_agcount * rotorstep);
2447 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2448 (mp->m_sb.sb_agcount * rotorstep);
2455 if (args->agbno == NULLAGBLOCK)
2456 args->fsbno = NULLFSBLOCK;
2458 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2460 ASSERT(args->len >= args->minlen);
2461 ASSERT(args->len <= args->maxlen);
2462 ASSERT(args->agbno % args->alignment == 0);
2463 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2467 xfs_perag_put(args->pag);
2470 xfs_perag_put(args->pag);
2476 * Just break up the extent address and hand off to xfs_free_ag_extent
2477 * after fixing up the freelist.
2481 xfs_trans_t *tp, /* transaction pointer */
2482 xfs_fsblock_t bno, /* starting block number of extent */
2483 xfs_extlen_t len) /* length of extent */
2485 xfs_alloc_arg_t args;
2489 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2491 args.mp = tp->t_mountp;
2494 * validate that the block number is legal - the enables us to detect
2495 * and handle a silent filesystem corruption rather than crashing.
2497 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2498 if (args.agno >= args.mp->m_sb.sb_agcount)
2499 return EFSCORRUPTED;
2501 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2502 if (args.agbno >= args.mp->m_sb.sb_agblocks)
2503 return EFSCORRUPTED;
2505 args.pag = xfs_perag_get(args.mp, args.agno);
2508 error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
2512 /* validate the extent size is legal now we have the agf locked */
2513 if (args.agbno + len >
2514 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) {
2515 error = EFSCORRUPTED;
2519 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2521 xfs_extent_busy_insert(tp, args.agno, args.agbno, len, 0);
2523 xfs_perag_put(args.pag);