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"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
39 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
41 #define XFSA_FIXUP_BNO_OK 1
42 #define XFSA_FIXUP_CNT_OK 2
44 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
45 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
46 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
47 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
48 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
49 STATIC void xfs_alloc_busy_trim(struct xfs_alloc_arg *,
50 xfs_agblock_t, xfs_extlen_t, xfs_agblock_t *, xfs_extlen_t *);
53 * Lookup the record equal to [bno, len] in the btree given by cur.
55 STATIC int /* error */
57 struct xfs_btree_cur *cur, /* btree cursor */
58 xfs_agblock_t bno, /* starting block of extent */
59 xfs_extlen_t len, /* length of extent */
60 int *stat) /* success/failure */
62 cur->bc_rec.a.ar_startblock = bno;
63 cur->bc_rec.a.ar_blockcount = len;
64 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
68 * Lookup the first record greater than or equal to [bno, len]
69 * in the btree given by cur.
71 STATIC int /* error */
73 struct xfs_btree_cur *cur, /* btree cursor */
74 xfs_agblock_t bno, /* starting block of extent */
75 xfs_extlen_t len, /* length of extent */
76 int *stat) /* success/failure */
78 cur->bc_rec.a.ar_startblock = bno;
79 cur->bc_rec.a.ar_blockcount = len;
80 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
84 * Lookup the first record less than or equal to [bno, len]
85 * in the btree given by cur.
89 struct xfs_btree_cur *cur, /* btree cursor */
90 xfs_agblock_t bno, /* starting block of extent */
91 xfs_extlen_t len, /* length of extent */
92 int *stat) /* success/failure */
94 cur->bc_rec.a.ar_startblock = bno;
95 cur->bc_rec.a.ar_blockcount = len;
96 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
100 * Update the record referred to by cur to the value given
102 * This either works (return 0) or gets an EFSCORRUPTED error.
104 STATIC int /* error */
106 struct xfs_btree_cur *cur, /* btree cursor */
107 xfs_agblock_t bno, /* starting block of extent */
108 xfs_extlen_t len) /* length of extent */
110 union xfs_btree_rec rec;
112 rec.alloc.ar_startblock = cpu_to_be32(bno);
113 rec.alloc.ar_blockcount = cpu_to_be32(len);
114 return xfs_btree_update(cur, &rec);
118 * Get the data from the pointed-to record.
122 struct xfs_btree_cur *cur, /* btree cursor */
123 xfs_agblock_t *bno, /* output: starting block of extent */
124 xfs_extlen_t *len, /* output: length of extent */
125 int *stat) /* output: success/failure */
127 union xfs_btree_rec *rec;
130 error = xfs_btree_get_rec(cur, &rec, stat);
131 if (!error && *stat == 1) {
132 *bno = be32_to_cpu(rec->alloc.ar_startblock);
133 *len = be32_to_cpu(rec->alloc.ar_blockcount);
139 * Compute aligned version of the found extent.
140 * Takes alignment and min length into account.
143 xfs_alloc_compute_aligned(
144 xfs_alloc_arg_t *args, /* allocation argument structure */
145 xfs_agblock_t foundbno, /* starting block in found extent */
146 xfs_extlen_t foundlen, /* length in found extent */
147 xfs_agblock_t *resbno, /* result block number */
148 xfs_extlen_t *reslen) /* result length */
153 /* Trim busy sections out of found extent */
154 xfs_alloc_busy_trim(args, foundbno, foundlen, &bno, &len);
156 if (args->alignment > 1 && len >= args->minlen) {
157 xfs_agblock_t aligned_bno = roundup(bno, args->alignment);
158 xfs_extlen_t diff = aligned_bno - bno;
160 *resbno = aligned_bno;
161 *reslen = diff >= len ? 0 : len - diff;
169 * Compute best start block and diff for "near" allocations.
170 * freelen >= wantlen already checked by caller.
172 STATIC xfs_extlen_t /* difference value (absolute) */
173 xfs_alloc_compute_diff(
174 xfs_agblock_t wantbno, /* target starting block */
175 xfs_extlen_t wantlen, /* target length */
176 xfs_extlen_t alignment, /* target alignment */
177 xfs_agblock_t freebno, /* freespace's starting block */
178 xfs_extlen_t freelen, /* freespace's length */
179 xfs_agblock_t *newbnop) /* result: best start block from free */
181 xfs_agblock_t freeend; /* end of freespace extent */
182 xfs_agblock_t newbno1; /* return block number */
183 xfs_agblock_t newbno2; /* other new block number */
184 xfs_extlen_t newlen1=0; /* length with newbno1 */
185 xfs_extlen_t newlen2=0; /* length with newbno2 */
186 xfs_agblock_t wantend; /* end of target extent */
188 ASSERT(freelen >= wantlen);
189 freeend = freebno + freelen;
190 wantend = wantbno + wantlen;
191 if (freebno >= wantbno) {
192 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
193 newbno1 = NULLAGBLOCK;
194 } else if (freeend >= wantend && alignment > 1) {
195 newbno1 = roundup(wantbno, alignment);
196 newbno2 = newbno1 - alignment;
197 if (newbno1 >= freeend)
198 newbno1 = NULLAGBLOCK;
200 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
201 if (newbno2 < freebno)
202 newbno2 = NULLAGBLOCK;
204 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
205 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
206 if (newlen1 < newlen2 ||
207 (newlen1 == newlen2 &&
208 XFS_ABSDIFF(newbno1, wantbno) >
209 XFS_ABSDIFF(newbno2, wantbno)))
211 } else if (newbno2 != NULLAGBLOCK)
213 } else if (freeend >= wantend) {
215 } else if (alignment > 1) {
216 newbno1 = roundup(freeend - wantlen, alignment);
217 if (newbno1 > freeend - wantlen &&
218 newbno1 - alignment >= freebno)
219 newbno1 -= alignment;
220 else if (newbno1 >= freeend)
221 newbno1 = NULLAGBLOCK;
223 newbno1 = freeend - wantlen;
225 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
229 * Fix up the length, based on mod and prod.
230 * len should be k * prod + mod for some k.
231 * If len is too small it is returned unchanged.
232 * If len hits maxlen it is left alone.
236 xfs_alloc_arg_t *args) /* allocation argument structure */
241 ASSERT(args->mod < args->prod);
243 ASSERT(rlen >= args->minlen);
244 ASSERT(rlen <= args->maxlen);
245 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
246 (args->mod == 0 && rlen < args->prod))
248 k = rlen % args->prod;
252 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
255 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
259 ASSERT(rlen >= args->minlen);
260 ASSERT(rlen <= args->maxlen);
265 * Fix up length if there is too little space left in the a.g.
266 * Return 1 if ok, 0 if too little, should give up.
269 xfs_alloc_fix_minleft(
270 xfs_alloc_arg_t *args) /* allocation argument structure */
272 xfs_agf_t *agf; /* a.g. freelist header */
273 int diff; /* free space difference */
275 if (args->minleft == 0)
277 agf = XFS_BUF_TO_AGF(args->agbp);
278 diff = be32_to_cpu(agf->agf_freeblks)
279 - args->len - args->minleft;
282 args->len += diff; /* shrink the allocated space */
283 if (args->len >= args->minlen)
285 args->agbno = NULLAGBLOCK;
290 * Update the two btrees, logically removing from freespace the extent
291 * starting at rbno, rlen blocks. The extent is contained within the
292 * actual (current) free extent fbno for flen blocks.
293 * Flags are passed in indicating whether the cursors are set to the
296 STATIC int /* error code */
297 xfs_alloc_fixup_trees(
298 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
299 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
300 xfs_agblock_t fbno, /* starting block of free extent */
301 xfs_extlen_t flen, /* length of free extent */
302 xfs_agblock_t rbno, /* starting block of returned extent */
303 xfs_extlen_t rlen, /* length of returned extent */
304 int flags) /* flags, XFSA_FIXUP_... */
306 int error; /* error code */
307 int i; /* operation results */
308 xfs_agblock_t nfbno1; /* first new free startblock */
309 xfs_agblock_t nfbno2; /* second new free startblock */
310 xfs_extlen_t nflen1=0; /* first new free length */
311 xfs_extlen_t nflen2=0; /* second new free length */
314 * Look up the record in the by-size tree if necessary.
316 if (flags & XFSA_FIXUP_CNT_OK) {
318 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
320 XFS_WANT_CORRUPTED_RETURN(
321 i == 1 && nfbno1 == fbno && nflen1 == flen);
324 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
326 XFS_WANT_CORRUPTED_RETURN(i == 1);
329 * Look up the record in the by-block tree if necessary.
331 if (flags & XFSA_FIXUP_BNO_OK) {
333 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
335 XFS_WANT_CORRUPTED_RETURN(
336 i == 1 && nfbno1 == fbno && nflen1 == flen);
339 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
341 XFS_WANT_CORRUPTED_RETURN(i == 1);
345 if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
346 struct xfs_btree_block *bnoblock;
347 struct xfs_btree_block *cntblock;
349 bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
350 cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
352 XFS_WANT_CORRUPTED_RETURN(
353 bnoblock->bb_numrecs == cntblock->bb_numrecs);
358 * Deal with all four cases: the allocated record is contained
359 * within the freespace record, so we can have new freespace
360 * at either (or both) end, or no freespace remaining.
362 if (rbno == fbno && rlen == flen)
363 nfbno1 = nfbno2 = NULLAGBLOCK;
364 else if (rbno == fbno) {
365 nfbno1 = rbno + rlen;
366 nflen1 = flen - rlen;
367 nfbno2 = NULLAGBLOCK;
368 } else if (rbno + rlen == fbno + flen) {
370 nflen1 = flen - rlen;
371 nfbno2 = NULLAGBLOCK;
374 nflen1 = rbno - fbno;
375 nfbno2 = rbno + rlen;
376 nflen2 = (fbno + flen) - nfbno2;
379 * Delete the entry from the by-size btree.
381 if ((error = xfs_btree_delete(cnt_cur, &i)))
383 XFS_WANT_CORRUPTED_RETURN(i == 1);
385 * Add new by-size btree entry(s).
387 if (nfbno1 != NULLAGBLOCK) {
388 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
390 XFS_WANT_CORRUPTED_RETURN(i == 0);
391 if ((error = xfs_btree_insert(cnt_cur, &i)))
393 XFS_WANT_CORRUPTED_RETURN(i == 1);
395 if (nfbno2 != NULLAGBLOCK) {
396 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
398 XFS_WANT_CORRUPTED_RETURN(i == 0);
399 if ((error = xfs_btree_insert(cnt_cur, &i)))
401 XFS_WANT_CORRUPTED_RETURN(i == 1);
404 * Fix up the by-block btree entry(s).
406 if (nfbno1 == NULLAGBLOCK) {
408 * No remaining freespace, just delete the by-block tree entry.
410 if ((error = xfs_btree_delete(bno_cur, &i)))
412 XFS_WANT_CORRUPTED_RETURN(i == 1);
415 * Update the by-block entry to start later|be shorter.
417 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
420 if (nfbno2 != NULLAGBLOCK) {
422 * 2 resulting free entries, need to add one.
424 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
426 XFS_WANT_CORRUPTED_RETURN(i == 0);
427 if ((error = xfs_btree_insert(bno_cur, &i)))
429 XFS_WANT_CORRUPTED_RETURN(i == 1);
435 * Read in the allocation group free block array.
437 STATIC int /* error */
439 xfs_mount_t *mp, /* mount point structure */
440 xfs_trans_t *tp, /* transaction pointer */
441 xfs_agnumber_t agno, /* allocation group number */
442 xfs_buf_t **bpp) /* buffer for the ag free block array */
444 xfs_buf_t *bp; /* return value */
447 ASSERT(agno != NULLAGNUMBER);
448 error = xfs_trans_read_buf(
449 mp, tp, mp->m_ddev_targp,
450 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
451 XFS_FSS_TO_BB(mp, 1), 0, &bp);
455 ASSERT(!XFS_BUF_GETERROR(bp));
456 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF);
462 xfs_alloc_update_counters(
463 struct xfs_trans *tp,
464 struct xfs_perag *pag,
465 struct xfs_buf *agbp,
468 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
470 pag->pagf_freeblks += len;
471 be32_add_cpu(&agf->agf_freeblks, len);
473 xfs_trans_agblocks_delta(tp, len);
474 if (unlikely(be32_to_cpu(agf->agf_freeblks) >
475 be32_to_cpu(agf->agf_length)))
478 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
483 * Allocation group level functions.
487 * Allocate a variable extent in the allocation group agno.
488 * Type and bno are used to determine where in the allocation group the
490 * Extent's length (returned in *len) will be between minlen and maxlen,
491 * and of the form k * prod + mod unless there's nothing that large.
492 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
494 STATIC int /* error */
495 xfs_alloc_ag_vextent(
496 xfs_alloc_arg_t *args) /* argument structure for allocation */
500 ASSERT(args->minlen > 0);
501 ASSERT(args->maxlen > 0);
502 ASSERT(args->minlen <= args->maxlen);
503 ASSERT(args->mod < args->prod);
504 ASSERT(args->alignment > 0);
506 * Branch to correct routine based on the type.
509 switch (args->type) {
510 case XFS_ALLOCTYPE_THIS_AG:
511 error = xfs_alloc_ag_vextent_size(args);
513 case XFS_ALLOCTYPE_NEAR_BNO:
514 error = xfs_alloc_ag_vextent_near(args);
516 case XFS_ALLOCTYPE_THIS_BNO:
517 error = xfs_alloc_ag_vextent_exact(args);
524 if (error || args->agbno == NULLAGBLOCK)
527 ASSERT(args->len >= args->minlen);
528 ASSERT(args->len <= args->maxlen);
529 ASSERT(!args->wasfromfl || !args->isfl);
530 ASSERT(args->agbno % args->alignment == 0);
532 if (!args->wasfromfl) {
533 error = xfs_alloc_update_counters(args->tp, args->pag,
535 -((long)(args->len)));
539 ASSERT(!xfs_alloc_busy_search(args->mp, args->agno,
540 args->agbno, args->len));
544 xfs_trans_mod_sb(args->tp, args->wasdel ?
545 XFS_TRANS_SB_RES_FDBLOCKS :
546 XFS_TRANS_SB_FDBLOCKS,
547 -((long)(args->len)));
550 XFS_STATS_INC(xs_allocx);
551 XFS_STATS_ADD(xs_allocb, args->len);
556 * Allocate a variable extent at exactly agno/bno.
557 * Extent's length (returned in *len) will be between minlen and maxlen,
558 * and of the form k * prod + mod unless there's nothing that large.
559 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
561 STATIC int /* error */
562 xfs_alloc_ag_vextent_exact(
563 xfs_alloc_arg_t *args) /* allocation argument structure */
565 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
566 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
568 xfs_agblock_t fbno; /* start block of found extent */
569 xfs_extlen_t flen; /* length of found extent */
570 xfs_agblock_t tbno; /* start block of trimmed extent */
571 xfs_extlen_t tlen; /* length of trimmed extent */
572 xfs_agblock_t tend; /* end block of trimmed extent */
573 xfs_agblock_t end; /* end of allocated extent */
574 int i; /* success/failure of operation */
575 xfs_extlen_t rlen; /* length of returned extent */
577 ASSERT(args->alignment == 1);
580 * Allocate/initialize a cursor for the by-number freespace btree.
582 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
583 args->agno, XFS_BTNUM_BNO);
586 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
587 * Look for the closest free block <= bno, it must contain bno
588 * if any free block does.
590 error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
597 * Grab the freespace record.
599 error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
602 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
603 ASSERT(fbno <= args->agbno);
606 * Check for overlapping busy extents.
608 xfs_alloc_busy_trim(args, fbno, flen, &tbno, &tlen);
611 * Give up if the start of the extent is busy, or the freespace isn't
612 * long enough for the minimum request.
614 if (tbno > args->agbno)
616 if (tlen < args->minlen)
619 if (tend < args->agbno + args->minlen)
623 * End of extent will be smaller of the freespace end and the
624 * maximal requested end.
626 * Fix the length according to mod and prod if given.
628 end = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen);
629 args->len = end - args->agbno;
630 xfs_alloc_fix_len(args);
631 if (!xfs_alloc_fix_minleft(args))
635 ASSERT(args->agbno + rlen <= tend);
636 end = args->agbno + rlen;
639 * We are allocating agbno for rlen [agbno .. end]
640 * Allocate/initialize a cursor for the by-size btree.
642 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
643 args->agno, XFS_BTNUM_CNT);
644 ASSERT(args->agbno + args->len <=
645 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
646 error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
647 args->len, XFSA_FIXUP_BNO_OK);
649 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
653 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
654 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
657 trace_xfs_alloc_exact_done(args);
661 /* Didn't find it, return null. */
662 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
663 args->agbno = NULLAGBLOCK;
664 trace_xfs_alloc_exact_notfound(args);
668 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
669 trace_xfs_alloc_exact_error(args);
674 * Search the btree in a given direction via the search cursor and compare
675 * the records found against the good extent we've already found.
678 xfs_alloc_find_best_extent(
679 struct xfs_alloc_arg *args, /* allocation argument structure */
680 struct xfs_btree_cur **gcur, /* good cursor */
681 struct xfs_btree_cur **scur, /* searching cursor */
682 xfs_agblock_t gdiff, /* difference for search comparison */
683 xfs_agblock_t *sbno, /* extent found by search */
684 xfs_extlen_t *slen, /* extent length */
685 xfs_agblock_t *sbnoa, /* aligned extent found by search */
686 xfs_extlen_t *slena, /* aligned extent length */
687 int dir) /* 0 = search right, 1 = search left */
694 /* The good extent is perfect, no need to search. */
699 * Look until we find a better one, run out of space or run off the end.
702 error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
705 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
706 xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena);
709 * The good extent is closer than this one.
712 if (*sbnoa >= args->agbno + gdiff)
715 if (*sbnoa <= args->agbno - gdiff)
720 * Same distance, compare length and pick the best.
722 if (*slena >= args->minlen) {
723 args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
724 xfs_alloc_fix_len(args);
726 sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
727 args->alignment, *sbnoa,
731 * Choose closer size and invalidate other cursor.
739 error = xfs_btree_increment(*scur, 0, &i);
741 error = xfs_btree_decrement(*scur, 0, &i);
747 xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
752 xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
757 /* caller invalidates cursors */
762 * Allocate a variable extent near bno in the allocation group agno.
763 * Extent's length (returned in len) will be between minlen and maxlen,
764 * and of the form k * prod + mod unless there's nothing that large.
765 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
767 STATIC int /* error */
768 xfs_alloc_ag_vextent_near(
769 xfs_alloc_arg_t *args) /* allocation argument structure */
771 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
772 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
773 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
774 xfs_agblock_t gtbno; /* start bno of right side entry */
775 xfs_agblock_t gtbnoa; /* aligned ... */
776 xfs_extlen_t gtdiff; /* difference to right side entry */
777 xfs_extlen_t gtlen; /* length of right side entry */
778 xfs_extlen_t gtlena; /* aligned ... */
779 xfs_agblock_t gtnew; /* useful start bno of right side */
780 int error; /* error code */
781 int i; /* result code, temporary */
782 int j; /* result code, temporary */
783 xfs_agblock_t ltbno; /* start bno of left side entry */
784 xfs_agblock_t ltbnoa; /* aligned ... */
785 xfs_extlen_t ltdiff; /* difference to left side entry */
786 xfs_extlen_t ltlen; /* length of left side entry */
787 xfs_extlen_t ltlena; /* aligned ... */
788 xfs_agblock_t ltnew; /* useful start bno of left side */
789 xfs_extlen_t rlen; /* length of returned extent */
791 #if defined(DEBUG) && defined(__KERNEL__)
793 * Randomly don't execute the first algorithm.
795 int dofirst; /* set to do first algorithm */
797 dofirst = random32() & 1;
808 * Get a cursor for the by-size btree.
810 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
811 args->agno, XFS_BTNUM_CNT);
814 * See if there are any free extents as big as maxlen.
816 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
819 * If none, then pick up the last entry in the tree unless the
823 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
826 if (i == 0 || ltlen == 0) {
827 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
828 trace_xfs_alloc_near_noentry(args);
837 * If the requested extent is large wrt the freespaces available
838 * in this a.g., then the cursor will be pointing to a btree entry
839 * near the right edge of the tree. If it's in the last btree leaf
840 * block, then we just examine all the entries in that block
841 * that are big enough, and pick the best one.
842 * This is written as a while loop so we can break out of it,
843 * but we never loop back to the top.
845 while (xfs_btree_islastblock(cnt_cur, 0)) {
849 xfs_agblock_t bnew=0;
851 #if defined(DEBUG) && defined(__KERNEL__)
856 * Start from the entry that lookup found, sequence through
857 * all larger free blocks. If we're actually pointing at a
858 * record smaller than maxlen, go to the start of this block,
859 * and skip all those smaller than minlen.
861 if (ltlen || args->alignment > 1) {
862 cnt_cur->bc_ptrs[0] = 1;
864 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
867 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
868 if (ltlen >= args->minlen)
870 if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
873 ASSERT(ltlen >= args->minlen);
877 i = cnt_cur->bc_ptrs[0];
878 for (j = 1, blen = 0, bdiff = 0;
879 !error && j && (blen < args->maxlen || bdiff > 0);
880 error = xfs_btree_increment(cnt_cur, 0, &j)) {
882 * For each entry, decide if it's better than
883 * the previous best entry.
885 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
887 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
888 xfs_alloc_compute_aligned(args, ltbno, ltlen,
890 if (ltlena < args->minlen)
892 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
893 xfs_alloc_fix_len(args);
894 ASSERT(args->len >= args->minlen);
895 if (args->len < blen)
897 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
898 args->alignment, ltbnoa, ltlena, <new);
899 if (ltnew != NULLAGBLOCK &&
900 (args->len > blen || ltdiff < bdiff)) {
904 besti = cnt_cur->bc_ptrs[0];
908 * It didn't work. We COULD be in a case where
909 * there's a good record somewhere, so try again.
914 * Point at the best entry, and retrieve it again.
916 cnt_cur->bc_ptrs[0] = besti;
917 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
919 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
920 ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
922 if (!xfs_alloc_fix_minleft(args)) {
923 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
924 trace_xfs_alloc_near_nominleft(args);
929 * We are allocating starting at bnew for blen blocks.
932 ASSERT(bnew >= ltbno);
933 ASSERT(bnew + blen <= ltbno + ltlen);
935 * Set up a cursor for the by-bno tree.
937 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
938 args->agbp, args->agno, XFS_BTNUM_BNO);
940 * Fix up the btree entries.
942 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
943 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
945 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
946 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
948 trace_xfs_alloc_near_first(args);
953 * Search in the by-bno tree to the left and to the right
954 * simultaneously, until in each case we find a space big enough,
955 * or run into the edge of the tree. When we run into the edge,
956 * we deallocate that cursor.
957 * If both searches succeed, we compare the two spaces and pick
959 * With alignment, it's possible for both to fail; the upper
960 * level algorithm that picks allocation groups for allocations
961 * is not supposed to do this.
964 * Allocate and initialize the cursor for the leftward search.
966 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
967 args->agno, XFS_BTNUM_BNO);
969 * Lookup <= bno to find the leftward search's starting point.
971 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
975 * Didn't find anything; use this cursor for the rightward
978 bno_cur_gt = bno_cur_lt;
982 * Found something. Duplicate the cursor for the rightward search.
984 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
987 * Increment the cursor, so we will point at the entry just right
988 * of the leftward entry if any, or to the leftmost entry.
990 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
994 * It failed, there are no rightward entries.
996 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
1000 * Loop going left with the leftward cursor, right with the
1001 * rightward cursor, until either both directions give up or
1002 * we find an entry at least as big as minlen.
1006 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1008 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1009 xfs_alloc_compute_aligned(args, ltbno, ltlen,
1011 if (ltlena >= args->minlen)
1013 if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1016 xfs_btree_del_cursor(bno_cur_lt,
1022 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1024 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1025 xfs_alloc_compute_aligned(args, gtbno, gtlen,
1027 if (gtlena >= args->minlen)
1029 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1032 xfs_btree_del_cursor(bno_cur_gt,
1037 } while (bno_cur_lt || bno_cur_gt);
1040 * Got both cursors still active, need to find better entry.
1042 if (bno_cur_lt && bno_cur_gt) {
1043 if (ltlena >= args->minlen) {
1045 * Left side is good, look for a right side entry.
1047 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1048 xfs_alloc_fix_len(args);
1049 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1050 args->alignment, ltbnoa, ltlena, <new);
1052 error = xfs_alloc_find_best_extent(args,
1053 &bno_cur_lt, &bno_cur_gt,
1054 ltdiff, >bno, >len,
1056 0 /* search right */);
1058 ASSERT(gtlena >= args->minlen);
1061 * Right side is good, look for a left side entry.
1063 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1064 xfs_alloc_fix_len(args);
1065 gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1066 args->alignment, gtbnoa, gtlena, >new);
1068 error = xfs_alloc_find_best_extent(args,
1069 &bno_cur_gt, &bno_cur_lt,
1070 gtdiff, <bno, <len,
1072 1 /* search left */);
1080 * If we couldn't get anything, give up.
1082 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1084 trace_xfs_alloc_near_busy(args);
1085 xfs_log_force(args->mp, XFS_LOG_SYNC);
1089 trace_xfs_alloc_size_neither(args);
1090 args->agbno = NULLAGBLOCK;
1095 * At this point we have selected a freespace entry, either to the
1096 * left or to the right. If it's on the right, copy all the
1097 * useful variables to the "left" set so we only have one
1098 * copy of this code.
1101 bno_cur_lt = bno_cur_gt;
1112 * Fix up the length and compute the useful address.
1114 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1115 xfs_alloc_fix_len(args);
1116 if (!xfs_alloc_fix_minleft(args)) {
1117 trace_xfs_alloc_near_nominleft(args);
1118 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1119 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1123 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1124 ltbnoa, ltlena, <new);
1125 ASSERT(ltnew >= ltbno);
1126 ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1127 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1128 args->agbno = ltnew;
1130 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1131 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1135 trace_xfs_alloc_near_greater(args);
1137 trace_xfs_alloc_near_lesser(args);
1139 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1140 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1144 trace_xfs_alloc_near_error(args);
1145 if (cnt_cur != NULL)
1146 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1147 if (bno_cur_lt != NULL)
1148 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1149 if (bno_cur_gt != NULL)
1150 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1155 * Allocate a variable extent anywhere in the allocation group agno.
1156 * Extent's length (returned in len) will be between minlen and maxlen,
1157 * and of the form k * prod + mod unless there's nothing that large.
1158 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1160 STATIC int /* error */
1161 xfs_alloc_ag_vextent_size(
1162 xfs_alloc_arg_t *args) /* allocation argument structure */
1164 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1165 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1166 int error; /* error result */
1167 xfs_agblock_t fbno; /* start of found freespace */
1168 xfs_extlen_t flen; /* length of found freespace */
1169 int i; /* temp status variable */
1170 xfs_agblock_t rbno; /* returned block number */
1171 xfs_extlen_t rlen; /* length of returned extent */
1176 * Allocate and initialize a cursor for the by-size btree.
1178 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1179 args->agno, XFS_BTNUM_CNT);
1183 * Look for an entry >= maxlen+alignment-1 blocks.
1185 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1186 args->maxlen + args->alignment - 1, &i)))
1190 * If none or we have busy extents that we cannot allocate from, then
1191 * we have to settle for a smaller extent. In the case that there are
1192 * no large extents, this will return the last entry in the tree unless
1193 * the tree is empty. In the case that there are only busy large
1194 * extents, this will return the largest small extent unless there
1195 * are no smaller extents available.
1197 if (!i || forced > 1) {
1198 error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1202 if (i == 0 || flen == 0) {
1203 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1204 trace_xfs_alloc_size_noentry(args);
1208 xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen);
1211 * Search for a non-busy extent that is large enough.
1212 * If we are at low space, don't check, or if we fall of
1213 * the end of the btree, turn off the busy check and
1217 error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1220 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1222 xfs_alloc_compute_aligned(args, fbno, flen,
1225 if (rlen >= args->maxlen)
1228 error = xfs_btree_increment(cnt_cur, 0, &i);
1233 * Our only valid extents must have been busy.
1234 * Make it unbusy by forcing the log out and
1235 * retrying. If we've been here before, forcing
1236 * the log isn't making the extents available,
1237 * which means they have probably been freed in
1238 * this transaction. In that case, we have to
1239 * give up on them and we'll attempt a minlen
1240 * allocation the next time around.
1242 xfs_btree_del_cursor(cnt_cur,
1244 trace_xfs_alloc_size_busy(args);
1246 xfs_log_force(args->mp, XFS_LOG_SYNC);
1253 * In the first case above, we got the last entry in the
1254 * by-size btree. Now we check to see if the space hits maxlen
1255 * once aligned; if not, we search left for something better.
1256 * This can't happen in the second case above.
1258 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1259 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1260 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1261 if (rlen < args->maxlen) {
1262 xfs_agblock_t bestfbno;
1263 xfs_extlen_t bestflen;
1264 xfs_agblock_t bestrbno;
1265 xfs_extlen_t bestrlen;
1272 if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1276 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1279 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1280 if (flen < bestrlen)
1282 xfs_alloc_compute_aligned(args, fbno, flen,
1284 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1285 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1286 (rlen <= flen && rbno + rlen <= fbno + flen),
1288 if (rlen > bestrlen) {
1293 if (rlen == args->maxlen)
1297 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1300 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1306 args->wasfromfl = 0;
1308 * Fix up the length.
1311 if (rlen < args->minlen) {
1313 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1314 trace_xfs_alloc_size_busy(args);
1315 xfs_log_force(args->mp, XFS_LOG_SYNC);
1320 xfs_alloc_fix_len(args);
1322 if (!xfs_alloc_fix_minleft(args))
1325 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1327 * Allocate and initialize a cursor for the by-block tree.
1329 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1330 args->agno, XFS_BTNUM_BNO);
1331 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1332 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1334 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1335 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1336 cnt_cur = bno_cur = NULL;
1339 XFS_WANT_CORRUPTED_GOTO(
1340 args->agbno + args->len <=
1341 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1343 trace_xfs_alloc_size_done(args);
1347 trace_xfs_alloc_size_error(args);
1349 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1351 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1355 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1356 trace_xfs_alloc_size_nominleft(args);
1357 args->agbno = NULLAGBLOCK;
1362 * Deal with the case where only small freespaces remain.
1363 * Either return the contents of the last freespace record,
1364 * or allocate space from the freelist if there is nothing in the tree.
1366 STATIC int /* error */
1367 xfs_alloc_ag_vextent_small(
1368 xfs_alloc_arg_t *args, /* allocation argument structure */
1369 xfs_btree_cur_t *ccur, /* by-size cursor */
1370 xfs_agblock_t *fbnop, /* result block number */
1371 xfs_extlen_t *flenp, /* result length */
1372 int *stat) /* status: 0-freelist, 1-normal/none */
1379 if ((error = xfs_btree_decrement(ccur, 0, &i)))
1382 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1384 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1387 * Nothing in the btree, try the freelist. Make sure
1388 * to respect minleft even when pulling from the
1391 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1392 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1394 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1397 if (fbno != NULLAGBLOCK) {
1398 xfs_alloc_busy_reuse(args->mp, args->agno, fbno, 1,
1401 if (args->userdata) {
1404 bp = xfs_btree_get_bufs(args->mp, args->tp,
1405 args->agno, fbno, 0);
1406 xfs_trans_binval(args->tp, bp);
1410 XFS_WANT_CORRUPTED_GOTO(
1411 args->agbno + args->len <=
1412 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1414 args->wasfromfl = 1;
1415 trace_xfs_alloc_small_freelist(args);
1420 * Nothing in the freelist.
1426 * Can't allocate from the freelist for some reason.
1433 * Can't do the allocation, give up.
1435 if (flen < args->minlen) {
1436 args->agbno = NULLAGBLOCK;
1437 trace_xfs_alloc_small_notenough(args);
1443 trace_xfs_alloc_small_done(args);
1447 trace_xfs_alloc_small_error(args);
1452 * Free the extent starting at agno/bno for length.
1454 STATIC int /* error */
1456 xfs_trans_t *tp, /* transaction pointer */
1457 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1458 xfs_agnumber_t agno, /* allocation group number */
1459 xfs_agblock_t bno, /* starting block number */
1460 xfs_extlen_t len, /* length of extent */
1461 int isfl) /* set if is freelist blocks - no sb acctg */
1463 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1464 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1465 int error; /* error return value */
1466 xfs_agblock_t gtbno; /* start of right neighbor block */
1467 xfs_extlen_t gtlen; /* length of right neighbor block */
1468 int haveleft; /* have a left neighbor block */
1469 int haveright; /* have a right neighbor block */
1470 int i; /* temp, result code */
1471 xfs_agblock_t ltbno; /* start of left neighbor block */
1472 xfs_extlen_t ltlen; /* length of left neighbor block */
1473 xfs_mount_t *mp; /* mount point struct for filesystem */
1474 xfs_agblock_t nbno; /* new starting block of freespace */
1475 xfs_extlen_t nlen; /* new length of freespace */
1476 xfs_perag_t *pag; /* per allocation group data */
1480 * Allocate and initialize a cursor for the by-block btree.
1482 bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1485 * Look for a neighboring block on the left (lower block numbers)
1486 * that is contiguous with this space.
1488 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1492 * There is a block to our left.
1494 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1496 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1498 * It's not contiguous, though.
1500 if (ltbno + ltlen < bno)
1504 * If this failure happens the request to free this
1505 * space was invalid, it's (partly) already free.
1508 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1512 * Look for a neighboring block on the right (higher block numbers)
1513 * that is contiguous with this space.
1515 if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1519 * There is a block to our right.
1521 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1523 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1525 * It's not contiguous, though.
1527 if (bno + len < gtbno)
1531 * If this failure happens the request to free this
1532 * space was invalid, it's (partly) already free.
1535 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1539 * Now allocate and initialize a cursor for the by-size tree.
1541 cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1543 * Have both left and right contiguous neighbors.
1544 * Merge all three into a single free block.
1546 if (haveleft && haveright) {
1548 * Delete the old by-size entry on the left.
1550 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1552 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1553 if ((error = xfs_btree_delete(cnt_cur, &i)))
1555 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1557 * Delete the old by-size entry on the right.
1559 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1561 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1562 if ((error = xfs_btree_delete(cnt_cur, &i)))
1564 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1566 * Delete the old by-block entry for the right block.
1568 if ((error = xfs_btree_delete(bno_cur, &i)))
1570 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1572 * Move the by-block cursor back to the left neighbor.
1574 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1576 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1579 * Check that this is the right record: delete didn't
1580 * mangle the cursor.
1583 xfs_agblock_t xxbno;
1586 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1589 XFS_WANT_CORRUPTED_GOTO(
1590 i == 1 && xxbno == ltbno && xxlen == ltlen,
1595 * Update remaining by-block entry to the new, joined block.
1598 nlen = len + ltlen + gtlen;
1599 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1603 * Have only a left contiguous neighbor.
1604 * Merge it together with the new freespace.
1606 else if (haveleft) {
1608 * Delete the old by-size entry on the left.
1610 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1612 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1613 if ((error = xfs_btree_delete(cnt_cur, &i)))
1615 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1617 * Back up the by-block cursor to the left neighbor, and
1618 * update its length.
1620 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1622 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1625 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1629 * Have only a right contiguous neighbor.
1630 * Merge it together with the new freespace.
1632 else if (haveright) {
1634 * Delete the old by-size entry on the right.
1636 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1638 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1639 if ((error = xfs_btree_delete(cnt_cur, &i)))
1641 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1643 * Update the starting block and length of the right
1644 * neighbor in the by-block tree.
1648 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1652 * No contiguous neighbors.
1653 * Insert the new freespace into the by-block tree.
1658 if ((error = xfs_btree_insert(bno_cur, &i)))
1660 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1662 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1665 * In all cases we need to insert the new freespace in the by-size tree.
1667 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1669 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1670 if ((error = xfs_btree_insert(cnt_cur, &i)))
1672 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1673 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1677 * Update the freespace totals in the ag and superblock.
1679 pag = xfs_perag_get(mp, agno);
1680 error = xfs_alloc_update_counters(tp, pag, agbp, len);
1686 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1687 XFS_STATS_INC(xs_freex);
1688 XFS_STATS_ADD(xs_freeb, len);
1690 trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright);
1695 trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1);
1697 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1699 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1704 * Visible (exported) allocation/free functions.
1705 * Some of these are used just by xfs_alloc_btree.c and this file.
1709 * Compute and fill in value of m_ag_maxlevels.
1712 xfs_alloc_compute_maxlevels(
1713 xfs_mount_t *mp) /* file system mount structure */
1721 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1722 minleafrecs = mp->m_alloc_mnr[0];
1723 minnoderecs = mp->m_alloc_mnr[1];
1724 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1725 for (level = 1; maxblocks > 1; level++)
1726 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1727 mp->m_ag_maxlevels = level;
1731 * Find the length of the longest extent in an AG.
1734 xfs_alloc_longest_free_extent(
1735 struct xfs_mount *mp,
1736 struct xfs_perag *pag)
1738 xfs_extlen_t need, delta = 0;
1740 need = XFS_MIN_FREELIST_PAG(pag, mp);
1741 if (need > pag->pagf_flcount)
1742 delta = need - pag->pagf_flcount;
1744 if (pag->pagf_longest > delta)
1745 return pag->pagf_longest - delta;
1746 return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
1750 * Decide whether to use this allocation group for this allocation.
1751 * If so, fix up the btree freelist's size.
1753 STATIC int /* error */
1754 xfs_alloc_fix_freelist(
1755 xfs_alloc_arg_t *args, /* allocation argument structure */
1756 int flags) /* XFS_ALLOC_FLAG_... */
1758 xfs_buf_t *agbp; /* agf buffer pointer */
1759 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1760 xfs_buf_t *agflbp;/* agfl buffer pointer */
1761 xfs_agblock_t bno; /* freelist block */
1762 xfs_extlen_t delta; /* new blocks needed in freelist */
1763 int error; /* error result code */
1764 xfs_extlen_t longest;/* longest extent in allocation group */
1765 xfs_mount_t *mp; /* file system mount point structure */
1766 xfs_extlen_t need; /* total blocks needed in freelist */
1767 xfs_perag_t *pag; /* per-ag information structure */
1768 xfs_alloc_arg_t targs; /* local allocation arguments */
1769 xfs_trans_t *tp; /* transaction pointer */
1775 if (!pag->pagf_init) {
1776 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1779 if (!pag->pagf_init) {
1780 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1781 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1789 * If this is a metadata preferred pag and we are user data
1790 * then try somewhere else if we are not being asked to
1791 * try harder at this point
1793 if (pag->pagf_metadata && args->userdata &&
1794 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1795 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1800 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1802 * If it looks like there isn't a long enough extent, or enough
1803 * total blocks, reject it.
1805 need = XFS_MIN_FREELIST_PAG(pag, mp);
1806 longest = xfs_alloc_longest_free_extent(mp, pag);
1807 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1809 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1810 need - args->total) < (int)args->minleft)) {
1812 xfs_trans_brelse(tp, agbp);
1819 * Get the a.g. freespace buffer.
1820 * Can fail if we're not blocking on locks, and it's held.
1823 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1827 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1828 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1834 * Figure out how many blocks we should have in the freelist.
1836 agf = XFS_BUF_TO_AGF(agbp);
1837 need = XFS_MIN_FREELIST(agf, mp);
1839 * If there isn't enough total or single-extent, reject it.
1841 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1842 delta = need > be32_to_cpu(agf->agf_flcount) ?
1843 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1844 longest = be32_to_cpu(agf->agf_longest);
1845 longest = (longest > delta) ? (longest - delta) :
1846 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1847 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1849 ((int)(be32_to_cpu(agf->agf_freeblks) +
1850 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1851 (int)args->minleft)) {
1852 xfs_trans_brelse(tp, agbp);
1858 * Make the freelist shorter if it's too long.
1860 while (be32_to_cpu(agf->agf_flcount) > need) {
1863 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1866 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1868 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1869 xfs_trans_binval(tp, bp);
1872 * Initialize the args structure.
1877 targs.agno = args->agno;
1878 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1879 targs.minalignslop = 0;
1880 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1881 targs.type = XFS_ALLOCTYPE_THIS_AG;
1883 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1886 * Make the freelist longer if it's too short.
1888 while (be32_to_cpu(agf->agf_flcount) < need) {
1890 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1892 * Allocate as many blocks as possible at once.
1894 if ((error = xfs_alloc_ag_vextent(&targs))) {
1895 xfs_trans_brelse(tp, agflbp);
1899 * Stop if we run out. Won't happen if callers are obeying
1900 * the restrictions correctly. Can happen for free calls
1901 * on a completely full ag.
1903 if (targs.agbno == NULLAGBLOCK) {
1904 if (flags & XFS_ALLOC_FLAG_FREEING)
1906 xfs_trans_brelse(tp, agflbp);
1911 * Put each allocated block on the list.
1913 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1914 error = xfs_alloc_put_freelist(tp, agbp,
1920 xfs_trans_brelse(tp, agflbp);
1926 * Get a block from the freelist.
1927 * Returns with the buffer for the block gotten.
1930 xfs_alloc_get_freelist(
1931 xfs_trans_t *tp, /* transaction pointer */
1932 xfs_buf_t *agbp, /* buffer containing the agf structure */
1933 xfs_agblock_t *bnop, /* block address retrieved from freelist */
1934 int btreeblk) /* destination is a AGF btree */
1936 xfs_agf_t *agf; /* a.g. freespace structure */
1937 xfs_agfl_t *agfl; /* a.g. freelist structure */
1938 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1939 xfs_agblock_t bno; /* block number returned */
1942 xfs_mount_t *mp; /* mount structure */
1943 xfs_perag_t *pag; /* per allocation group data */
1945 agf = XFS_BUF_TO_AGF(agbp);
1947 * Freelist is empty, give up.
1949 if (!agf->agf_flcount) {
1950 *bnop = NULLAGBLOCK;
1954 * Read the array of free blocks.
1957 if ((error = xfs_alloc_read_agfl(mp, tp,
1958 be32_to_cpu(agf->agf_seqno), &agflbp)))
1960 agfl = XFS_BUF_TO_AGFL(agflbp);
1962 * Get the block number and update the data structures.
1964 bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
1965 be32_add_cpu(&agf->agf_flfirst, 1);
1966 xfs_trans_brelse(tp, agflbp);
1967 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
1968 agf->agf_flfirst = 0;
1970 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
1971 be32_add_cpu(&agf->agf_flcount, -1);
1972 xfs_trans_agflist_delta(tp, -1);
1973 pag->pagf_flcount--;
1976 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
1978 be32_add_cpu(&agf->agf_btreeblks, 1);
1979 pag->pagf_btreeblks++;
1980 logflags |= XFS_AGF_BTREEBLKS;
1983 xfs_alloc_log_agf(tp, agbp, logflags);
1990 * Log the given fields from the agf structure.
1994 xfs_trans_t *tp, /* transaction pointer */
1995 xfs_buf_t *bp, /* buffer for a.g. freelist header */
1996 int fields) /* mask of fields to be logged (XFS_AGF_...) */
1998 int first; /* first byte offset */
1999 int last; /* last byte offset */
2000 static const short offsets[] = {
2001 offsetof(xfs_agf_t, agf_magicnum),
2002 offsetof(xfs_agf_t, agf_versionnum),
2003 offsetof(xfs_agf_t, agf_seqno),
2004 offsetof(xfs_agf_t, agf_length),
2005 offsetof(xfs_agf_t, agf_roots[0]),
2006 offsetof(xfs_agf_t, agf_levels[0]),
2007 offsetof(xfs_agf_t, agf_flfirst),
2008 offsetof(xfs_agf_t, agf_fllast),
2009 offsetof(xfs_agf_t, agf_flcount),
2010 offsetof(xfs_agf_t, agf_freeblks),
2011 offsetof(xfs_agf_t, agf_longest),
2012 offsetof(xfs_agf_t, agf_btreeblks),
2016 trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2018 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2019 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2023 * Interface for inode allocation to force the pag data to be initialized.
2026 xfs_alloc_pagf_init(
2027 xfs_mount_t *mp, /* file system mount structure */
2028 xfs_trans_t *tp, /* transaction pointer */
2029 xfs_agnumber_t agno, /* allocation group number */
2030 int flags) /* XFS_ALLOC_FLAGS_... */
2035 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2038 xfs_trans_brelse(tp, bp);
2043 * Put the block on the freelist for the allocation group.
2046 xfs_alloc_put_freelist(
2047 xfs_trans_t *tp, /* transaction pointer */
2048 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2049 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2050 xfs_agblock_t bno, /* block being freed */
2051 int btreeblk) /* block came from a AGF btree */
2053 xfs_agf_t *agf; /* a.g. freespace structure */
2054 xfs_agfl_t *agfl; /* a.g. free block array */
2055 __be32 *blockp;/* pointer to array entry */
2058 xfs_mount_t *mp; /* mount structure */
2059 xfs_perag_t *pag; /* per allocation group data */
2061 agf = XFS_BUF_TO_AGF(agbp);
2064 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2065 be32_to_cpu(agf->agf_seqno), &agflbp)))
2067 agfl = XFS_BUF_TO_AGFL(agflbp);
2068 be32_add_cpu(&agf->agf_fllast, 1);
2069 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2070 agf->agf_fllast = 0;
2072 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2073 be32_add_cpu(&agf->agf_flcount, 1);
2074 xfs_trans_agflist_delta(tp, 1);
2075 pag->pagf_flcount++;
2077 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2079 be32_add_cpu(&agf->agf_btreeblks, -1);
2080 pag->pagf_btreeblks--;
2081 logflags |= XFS_AGF_BTREEBLKS;
2085 xfs_alloc_log_agf(tp, agbp, logflags);
2087 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2088 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
2089 *blockp = cpu_to_be32(bno);
2090 xfs_alloc_log_agf(tp, agbp, logflags);
2091 xfs_trans_log_buf(tp, agflbp,
2092 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2093 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2094 sizeof(xfs_agblock_t) - 1));
2099 * Read in the allocation group header (free/alloc section).
2103 struct xfs_mount *mp, /* mount point structure */
2104 struct xfs_trans *tp, /* transaction pointer */
2105 xfs_agnumber_t agno, /* allocation group number */
2106 int flags, /* XFS_BUF_ */
2107 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2109 struct xfs_agf *agf; /* ag freelist header */
2110 int agf_ok; /* set if agf is consistent */
2113 ASSERT(agno != NULLAGNUMBER);
2114 error = xfs_trans_read_buf(
2115 mp, tp, mp->m_ddev_targp,
2116 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2117 XFS_FSS_TO_BB(mp, 1), flags, bpp);
2123 ASSERT(!XFS_BUF_GETERROR(*bpp));
2124 agf = XFS_BUF_TO_AGF(*bpp);
2127 * Validate the magic number of the agf block.
2130 be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC &&
2131 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2132 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2133 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2134 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2135 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp) &&
2136 be32_to_cpu(agf->agf_seqno) == agno;
2137 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
2138 agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <=
2139 be32_to_cpu(agf->agf_length);
2140 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2141 XFS_RANDOM_ALLOC_READ_AGF))) {
2142 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2143 XFS_ERRLEVEL_LOW, mp, agf);
2144 xfs_trans_brelse(tp, *bpp);
2145 return XFS_ERROR(EFSCORRUPTED);
2147 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGF, XFS_AGF_REF);
2152 * Read in the allocation group header (free/alloc section).
2156 struct xfs_mount *mp, /* mount point structure */
2157 struct xfs_trans *tp, /* transaction pointer */
2158 xfs_agnumber_t agno, /* allocation group number */
2159 int flags, /* XFS_ALLOC_FLAG_... */
2160 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2162 struct xfs_agf *agf; /* ag freelist header */
2163 struct xfs_perag *pag; /* per allocation group data */
2166 ASSERT(agno != NULLAGNUMBER);
2168 error = xfs_read_agf(mp, tp, agno,
2169 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2175 ASSERT(!XFS_BUF_GETERROR(*bpp));
2177 agf = XFS_BUF_TO_AGF(*bpp);
2178 pag = xfs_perag_get(mp, agno);
2179 if (!pag->pagf_init) {
2180 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2181 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2182 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2183 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2184 pag->pagf_levels[XFS_BTNUM_BNOi] =
2185 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2186 pag->pagf_levels[XFS_BTNUM_CNTi] =
2187 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2188 spin_lock_init(&pag->pagb_lock);
2189 pag->pagb_count = 0;
2190 pag->pagb_tree = RB_ROOT;
2194 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2195 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2196 ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2197 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2198 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2199 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2200 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2201 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2202 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2210 * Allocate an extent (variable-size).
2211 * Depending on the allocation type, we either look in a single allocation
2212 * group or loop over the allocation groups to find the result.
2216 xfs_alloc_arg_t *args) /* allocation argument structure */
2218 xfs_agblock_t agsize; /* allocation group size */
2220 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2221 xfs_extlen_t minleft;/* minimum left value, temp copy */
2222 xfs_mount_t *mp; /* mount structure pointer */
2223 xfs_agnumber_t sagno; /* starting allocation group number */
2224 xfs_alloctype_t type; /* input allocation type */
2227 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2230 type = args->otype = args->type;
2231 args->agbno = NULLAGBLOCK;
2233 * Just fix this up, for the case where the last a.g. is shorter
2234 * (or there's only one a.g.) and the caller couldn't easily figure
2235 * that out (xfs_bmap_alloc).
2237 agsize = mp->m_sb.sb_agblocks;
2238 if (args->maxlen > agsize)
2239 args->maxlen = agsize;
2240 if (args->alignment == 0)
2241 args->alignment = 1;
2242 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2243 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2244 ASSERT(args->minlen <= args->maxlen);
2245 ASSERT(args->minlen <= agsize);
2246 ASSERT(args->mod < args->prod);
2247 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2248 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2249 args->minlen > args->maxlen || args->minlen > agsize ||
2250 args->mod >= args->prod) {
2251 args->fsbno = NULLFSBLOCK;
2252 trace_xfs_alloc_vextent_badargs(args);
2255 minleft = args->minleft;
2258 case XFS_ALLOCTYPE_THIS_AG:
2259 case XFS_ALLOCTYPE_NEAR_BNO:
2260 case XFS_ALLOCTYPE_THIS_BNO:
2262 * These three force us into a single a.g.
2264 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2265 args->pag = xfs_perag_get(mp, args->agno);
2267 error = xfs_alloc_fix_freelist(args, 0);
2268 args->minleft = minleft;
2270 trace_xfs_alloc_vextent_nofix(args);
2274 trace_xfs_alloc_vextent_noagbp(args);
2277 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2278 if ((error = xfs_alloc_ag_vextent(args)))
2281 case XFS_ALLOCTYPE_START_BNO:
2283 * Try near allocation first, then anywhere-in-ag after
2284 * the first a.g. fails.
2286 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2287 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2288 args->fsbno = XFS_AGB_TO_FSB(mp,
2289 ((mp->m_agfrotor / rotorstep) %
2290 mp->m_sb.sb_agcount), 0);
2293 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2294 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2296 case XFS_ALLOCTYPE_ANY_AG:
2297 case XFS_ALLOCTYPE_START_AG:
2298 case XFS_ALLOCTYPE_FIRST_AG:
2300 * Rotate through the allocation groups looking for a winner.
2302 if (type == XFS_ALLOCTYPE_ANY_AG) {
2304 * Start with the last place we left off.
2306 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2307 mp->m_sb.sb_agcount;
2308 args->type = XFS_ALLOCTYPE_THIS_AG;
2309 flags = XFS_ALLOC_FLAG_TRYLOCK;
2310 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2312 * Start with allocation group given by bno.
2314 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2315 args->type = XFS_ALLOCTYPE_THIS_AG;
2319 if (type == XFS_ALLOCTYPE_START_AG)
2320 args->type = XFS_ALLOCTYPE_THIS_AG;
2322 * Start with the given allocation group.
2324 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2325 flags = XFS_ALLOC_FLAG_TRYLOCK;
2328 * Loop over allocation groups twice; first time with
2329 * trylock set, second time without.
2332 args->pag = xfs_perag_get(mp, args->agno);
2333 if (no_min) args->minleft = 0;
2334 error = xfs_alloc_fix_freelist(args, flags);
2335 args->minleft = minleft;
2337 trace_xfs_alloc_vextent_nofix(args);
2341 * If we get a buffer back then the allocation will fly.
2344 if ((error = xfs_alloc_ag_vextent(args)))
2349 trace_xfs_alloc_vextent_loopfailed(args);
2352 * Didn't work, figure out the next iteration.
2354 if (args->agno == sagno &&
2355 type == XFS_ALLOCTYPE_START_BNO)
2356 args->type = XFS_ALLOCTYPE_THIS_AG;
2358 * For the first allocation, we can try any AG to get
2359 * space. However, if we already have allocated a
2360 * block, we don't want to try AGs whose number is below
2361 * sagno. Otherwise, we may end up with out-of-order
2362 * locking of AGF, which might cause deadlock.
2364 if (++(args->agno) == mp->m_sb.sb_agcount) {
2365 if (args->firstblock != NULLFSBLOCK)
2371 * Reached the starting a.g., must either be done
2372 * or switch to non-trylock mode.
2374 if (args->agno == sagno) {
2376 args->agbno = NULLAGBLOCK;
2377 trace_xfs_alloc_vextent_allfailed(args);
2384 if (type == XFS_ALLOCTYPE_START_BNO) {
2385 args->agbno = XFS_FSB_TO_AGBNO(mp,
2387 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2391 xfs_perag_put(args->pag);
2393 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2394 if (args->agno == sagno)
2395 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2396 (mp->m_sb.sb_agcount * rotorstep);
2398 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2399 (mp->m_sb.sb_agcount * rotorstep);
2406 if (args->agbno == NULLAGBLOCK)
2407 args->fsbno = NULLFSBLOCK;
2409 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2411 ASSERT(args->len >= args->minlen);
2412 ASSERT(args->len <= args->maxlen);
2413 ASSERT(args->agbno % args->alignment == 0);
2414 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2418 xfs_perag_put(args->pag);
2421 xfs_perag_put(args->pag);
2427 * Just break up the extent address and hand off to xfs_free_ag_extent
2428 * after fixing up the freelist.
2432 xfs_trans_t *tp, /* transaction pointer */
2433 xfs_fsblock_t bno, /* starting block number of extent */
2434 xfs_extlen_t len) /* length of extent */
2436 xfs_alloc_arg_t args;
2440 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2442 args.mp = tp->t_mountp;
2445 * validate that the block number is legal - the enables us to detect
2446 * and handle a silent filesystem corruption rather than crashing.
2448 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2449 if (args.agno >= args.mp->m_sb.sb_agcount)
2450 return EFSCORRUPTED;
2452 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2453 if (args.agbno >= args.mp->m_sb.sb_agblocks)
2454 return EFSCORRUPTED;
2456 args.pag = xfs_perag_get(args.mp, args.agno);
2459 error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
2463 /* validate the extent size is legal now we have the agf locked */
2464 if (args.agbno + len >
2465 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) {
2466 error = EFSCORRUPTED;
2470 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2472 xfs_alloc_busy_insert(tp, args.agno, args.agbno, len);
2474 xfs_perag_put(args.pag);
2479 xfs_alloc_busy_insert(
2480 struct xfs_trans *tp,
2481 xfs_agnumber_t agno,
2485 struct xfs_busy_extent *new;
2486 struct xfs_busy_extent *busyp;
2487 struct xfs_perag *pag;
2488 struct rb_node **rbp;
2489 struct rb_node *parent = NULL;
2491 new = kmem_zalloc(sizeof(struct xfs_busy_extent), KM_MAYFAIL);
2494 * No Memory! Since it is now not possible to track the free
2495 * block, make this a synchronous transaction to insure that
2496 * the block is not reused before this transaction commits.
2498 trace_xfs_alloc_busy_enomem(tp->t_mountp, agno, bno, len);
2499 xfs_trans_set_sync(tp);
2506 INIT_LIST_HEAD(&new->list);
2508 /* trace before insert to be able to see failed inserts */
2509 trace_xfs_alloc_busy(tp->t_mountp, agno, bno, len);
2511 pag = xfs_perag_get(tp->t_mountp, new->agno);
2512 spin_lock(&pag->pagb_lock);
2513 rbp = &pag->pagb_tree.rb_node;
2516 busyp = rb_entry(parent, struct xfs_busy_extent, rb_node);
2518 if (new->bno < busyp->bno) {
2519 rbp = &(*rbp)->rb_left;
2520 ASSERT(new->bno + new->length <= busyp->bno);
2521 } else if (new->bno > busyp->bno) {
2522 rbp = &(*rbp)->rb_right;
2523 ASSERT(bno >= busyp->bno + busyp->length);
2529 rb_link_node(&new->rb_node, parent, rbp);
2530 rb_insert_color(&new->rb_node, &pag->pagb_tree);
2532 list_add(&new->list, &tp->t_busy);
2533 spin_unlock(&pag->pagb_lock);
2538 * Search for a busy extent within the range of the extent we are about to
2539 * allocate. You need to be holding the busy extent tree lock when calling
2540 * xfs_alloc_busy_search(). This function returns 0 for no overlapping busy
2541 * extent, -1 for an overlapping but not exact busy extent, and 1 for an exact
2542 * match. This is done so that a non-zero return indicates an overlap that
2543 * will require a synchronous transaction, but it can still be
2544 * used to distinguish between a partial or exact match.
2547 xfs_alloc_busy_search(
2548 struct xfs_mount *mp,
2549 xfs_agnumber_t agno,
2553 struct xfs_perag *pag;
2554 struct rb_node *rbp;
2555 struct xfs_busy_extent *busyp;
2558 pag = xfs_perag_get(mp, agno);
2559 spin_lock(&pag->pagb_lock);
2561 rbp = pag->pagb_tree.rb_node;
2563 /* find closest start bno overlap */
2565 busyp = rb_entry(rbp, struct xfs_busy_extent, rb_node);
2566 if (bno < busyp->bno) {
2567 /* may overlap, but exact start block is lower */
2568 if (bno + len > busyp->bno)
2571 } else if (bno > busyp->bno) {
2572 /* may overlap, but exact start block is higher */
2573 if (bno < busyp->bno + busyp->length)
2575 rbp = rbp->rb_right;
2577 /* bno matches busyp, length determines exact match */
2578 match = (busyp->length == len) ? 1 : -1;
2582 spin_unlock(&pag->pagb_lock);
2588 * The found free extent [fbno, fend] overlaps part or all of the given busy
2589 * extent. If the overlap covers the beginning, the end, or all of the busy
2590 * extent, the overlapping portion can be made unbusy and used for the
2591 * allocation. We can't split a busy extent because we can't modify a
2592 * transaction/CIL context busy list, but we can update an entries block
2595 * Returns true if the extent can safely be reused, or false if the search
2596 * needs to be restarted.
2599 xfs_alloc_busy_update_extent(
2600 struct xfs_mount *mp,
2601 struct xfs_perag *pag,
2602 struct xfs_busy_extent *busyp,
2607 xfs_agblock_t fend = fbno + flen;
2608 xfs_agblock_t bbno = busyp->bno;
2609 xfs_agblock_t bend = bbno + busyp->length;
2612 * If there is a busy extent overlapping a user allocation, we have
2613 * no choice but to force the log and retry the search.
2615 * Fortunately this does not happen during normal operation, but
2616 * only if the filesystem is very low on space and has to dip into
2617 * the AGFL for normal allocations.
2622 if (bbno < fbno && bend > fend) {
2626 * +BBBBBBBBBBBBBBBBB+
2632 * We would have to split the busy extent to be able to track
2633 * it correct, which we cannot do because we would have to
2634 * modify the list of busy extents attached to the transaction
2635 * or CIL context, which is immutable.
2637 * Force out the log to clear the busy extent and retry the
2641 } else if (bbno >= fbno && bend <= fend) {
2645 * +BBBBBBBBBBBBBBBBB+
2646 * +-----------------+
2651 * +BBBBBBBBBBBBBBBBB+
2652 * +--------------------------+
2657 * +BBBBBBBBBBBBBBBBB+
2658 * +--------------------------+
2663 * +BBBBBBBBBBBBBBBBB+
2664 * +-----------------------------------+
2670 * The busy extent is fully covered by the extent we are
2671 * allocating, and can simply be removed from the rbtree.
2672 * However we cannot remove it from the immutable list
2673 * tracking busy extents in the transaction or CIL context,
2674 * so set the length to zero to mark it invalid.
2676 * We also need to restart the busy extent search from the
2677 * tree root, because erasing the node can rearrange the
2680 rb_erase(&busyp->rb_node, &pag->pagb_tree);
2683 } else if (fend < bend) {
2687 * +BBBBBBBBBBBBBBBBB+
2693 * +BBBBBBBBBBBBBBBBB+
2694 * +------------------+
2699 } else if (bbno < fbno) {
2703 * +BBBBBBBBBBBBBBBBB+
2709 * +BBBBBBBBBBBBBBBBB+
2710 * +----------------------+
2713 busyp->length = fbno - busyp->bno;
2718 trace_xfs_alloc_busy_reuse(mp, pag->pag_agno, fbno, flen);
2722 spin_unlock(&pag->pagb_lock);
2723 xfs_log_force(mp, XFS_LOG_SYNC);
2724 trace_xfs_alloc_busy_force(mp, pag->pag_agno, fbno, flen);
2725 spin_lock(&pag->pagb_lock);
2731 * For a given extent [fbno, flen], make sure we can reuse it safely.
2734 xfs_alloc_busy_reuse(
2735 struct xfs_mount *mp,
2736 xfs_agnumber_t agno,
2741 struct xfs_perag *pag;
2742 struct rb_node *rbp;
2746 pag = xfs_perag_get(mp, agno);
2747 spin_lock(&pag->pagb_lock);
2749 rbp = pag->pagb_tree.rb_node;
2751 struct xfs_busy_extent *busyp =
2752 rb_entry(rbp, struct xfs_busy_extent, rb_node);
2753 xfs_agblock_t bbno = busyp->bno;
2754 xfs_agblock_t bend = bbno + busyp->length;
2756 if (fbno + flen <= bbno) {
2759 } else if (fbno >= bend) {
2760 rbp = rbp->rb_right;
2764 if (!xfs_alloc_busy_update_extent(mp, pag, busyp, fbno, flen,
2768 spin_unlock(&pag->pagb_lock);
2773 * For a given extent [fbno, flen], search the busy extent list to find a
2774 * subset of the extent that is not busy. If *rlen is smaller than
2775 * args->minlen no suitable extent could be found, and the higher level
2776 * code needs to force out the log and retry the allocation.
2779 xfs_alloc_busy_trim(
2780 struct xfs_alloc_arg *args,
2783 xfs_agblock_t *rbno,
2788 struct rb_node *rbp;
2792 spin_lock(&args->pag->pagb_lock);
2796 rbp = args->pag->pagb_tree.rb_node;
2797 while (rbp && flen >= args->minlen) {
2798 struct xfs_busy_extent *busyp =
2799 rb_entry(rbp, struct xfs_busy_extent, rb_node);
2800 xfs_agblock_t fend = fbno + flen;
2801 xfs_agblock_t bbno = busyp->bno;
2802 xfs_agblock_t bend = bbno + busyp->length;
2807 } else if (fbno >= bend) {
2808 rbp = rbp->rb_right;
2813 * If this is a metadata allocation, try to reuse the busy
2814 * extent instead of trimming the allocation.
2816 if (!args->userdata) {
2817 if (!xfs_alloc_busy_update_extent(args->mp, args->pag,
2830 * +BBBBBBBBBBBBBBBBB+
2836 * +BBBBBBBBBBBBBBBBB+
2842 * +BBBBBBBBBBBBBBBBB+
2848 * +BBBBBBBBBBBBBBBBB+
2849 * +-----------------+
2852 * No unbusy region in extent, return failure.
2860 * +BBBBBBBBBBBBBBBBB+
2861 * +----------------------+
2866 * +BBBBBBBBBBBBBBBBB+
2867 * +--------------------------+
2870 * Needs to be trimmed to:
2875 } else if (bend >= fend) {
2881 * +BBBBBBBBBBBBBBBBB+
2882 * +------------------+
2887 * +BBBBBBBBBBBBBBBBB+
2888 * +--------------------------+
2891 * Needs to be trimmed to:
2897 /* middle overlap */
2902 * +BBBBBBBBBBBBBBBBB+
2903 * +-----------------------------------+
2906 * Can be trimmed to:
2907 * +-------+ OR +-------+
2908 * fbno fend fbno fend
2910 * Backward allocation leads to significant
2911 * fragmentation of directories, which degrades
2912 * directory performance, therefore we always want to
2913 * choose the option that produces forward allocation
2915 * Preferring the lower bno extent will make the next
2916 * request use "fend" as the start of the next
2917 * allocation; if the segment is no longer busy at
2918 * that point, we'll get a contiguous allocation, but
2919 * even if it is still busy, we will get a forward
2921 * We try to avoid choosing the segment at "bend",
2922 * because that can lead to the next allocation
2923 * taking the segment at "fbno", which would be a
2924 * backward allocation. We only use the segment at
2925 * "fbno" if it is much larger than the current
2926 * requested size, because in that case there's a
2927 * good chance subsequent allocations will be
2930 if (bbno - fbno >= args->maxlen) {
2931 /* left candidate fits perfect */
2933 } else if (fend - bend >= args->maxlen * 4) {
2934 /* right candidate has enough free space */
2936 } else if (bbno - fbno >= args->minlen) {
2937 /* left candidate fits minimum requirement */
2946 spin_unlock(&args->pag->pagb_lock);
2948 if (fbno != bno || flen != len) {
2949 trace_xfs_alloc_busy_trim(args->mp, args->agno, bno, len,
2957 * Return a zero extent length as failure indications. All callers
2958 * re-check if the trimmed extent satisfies the minlen requirement.
2960 spin_unlock(&args->pag->pagb_lock);
2961 trace_xfs_alloc_busy_trim(args->mp, args->agno, bno, len, fbno, 0);
2967 xfs_alloc_busy_clear_one(
2968 struct xfs_mount *mp,
2969 struct xfs_perag *pag,
2970 struct xfs_busy_extent *busyp)
2972 if (busyp->length) {
2973 trace_xfs_alloc_busy_clear(mp, busyp->agno, busyp->bno,
2975 rb_erase(&busyp->rb_node, &pag->pagb_tree);
2978 list_del_init(&busyp->list);
2983 xfs_alloc_busy_clear(
2984 struct xfs_mount *mp,
2985 struct list_head *list)
2987 struct xfs_busy_extent *busyp, *n;
2988 struct xfs_perag *pag = NULL;
2989 xfs_agnumber_t agno = NULLAGNUMBER;
2991 list_for_each_entry_safe(busyp, n, list, list) {
2992 if (busyp->agno != agno) {
2994 spin_unlock(&pag->pagb_lock);
2997 pag = xfs_perag_get(mp, busyp->agno);
2998 spin_lock(&pag->pagb_lock);
3002 xfs_alloc_busy_clear_one(mp, pag, busyp);
3006 spin_unlock(&pag->pagb_lock);
3012 * Callback for list_sort to sort busy extents by the AG they reside in.
3015 xfs_busy_extent_ag_cmp(
3017 struct list_head *a,
3018 struct list_head *b)
3020 return container_of(a, struct xfs_busy_extent, list)->agno -
3021 container_of(b, struct xfs_busy_extent, list)->agno;