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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_error.h"
43 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
45 #define XFSA_FIXUP_BNO_OK 1
46 #define XFSA_FIXUP_CNT_OK 2
49 xfs_alloc_search_busy(xfs_trans_t *tp,
54 #if defined(XFS_ALLOC_TRACE)
55 ktrace_t *xfs_alloc_trace_buf;
57 #define TRACE_ALLOC(s,a) \
58 xfs_alloc_trace_alloc(__func__, s, a, __LINE__)
59 #define TRACE_FREE(s,a,b,x,f) \
60 xfs_alloc_trace_free(__func__, s, mp, a, b, x, f, __LINE__)
61 #define TRACE_MODAGF(s,a,f) \
62 xfs_alloc_trace_modagf(__func__, s, mp, a, f, __LINE__)
63 #define TRACE_BUSY(__func__,s,ag,agb,l,sl,tp) \
64 xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSY, __LINE__)
65 #define TRACE_UNBUSY(__func__,s,ag,sl,tp) \
66 xfs_alloc_trace_busy(__func__, s, mp, ag, -1, -1, sl, tp, XFS_ALLOC_KTRACE_UNBUSY, __LINE__)
67 #define TRACE_BUSYSEARCH(__func__,s,ag,agb,l,tp) \
68 xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, 0, tp, XFS_ALLOC_KTRACE_BUSYSEARCH, __LINE__)
70 #define TRACE_ALLOC(s,a)
71 #define TRACE_FREE(s,a,b,x,f)
72 #define TRACE_MODAGF(s,a,f)
73 #define TRACE_BUSY(s,a,ag,agb,l,sl,tp)
74 #define TRACE_UNBUSY(fname,s,ag,sl,tp)
75 #define TRACE_BUSYSEARCH(fname,s,ag,agb,l,tp)
76 #endif /* XFS_ALLOC_TRACE */
79 * Prototypes for per-ag allocation routines
82 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
83 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
84 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
85 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
86 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
93 * Compute aligned version of the found extent.
94 * Takes alignment and min length into account.
97 xfs_alloc_compute_aligned(
98 xfs_agblock_t foundbno, /* starting block in found extent */
99 xfs_extlen_t foundlen, /* length in found extent */
100 xfs_extlen_t alignment, /* alignment for allocation */
101 xfs_extlen_t minlen, /* minimum length for allocation */
102 xfs_agblock_t *resbno, /* result block number */
103 xfs_extlen_t *reslen) /* result length */
109 if (alignment > 1 && foundlen >= minlen) {
110 bno = roundup(foundbno, alignment);
111 diff = bno - foundbno;
112 len = diff >= foundlen ? 0 : foundlen - diff;
122 * Compute best start block and diff for "near" allocations.
123 * freelen >= wantlen already checked by caller.
125 STATIC xfs_extlen_t /* difference value (absolute) */
126 xfs_alloc_compute_diff(
127 xfs_agblock_t wantbno, /* target starting block */
128 xfs_extlen_t wantlen, /* target length */
129 xfs_extlen_t alignment, /* target alignment */
130 xfs_agblock_t freebno, /* freespace's starting block */
131 xfs_extlen_t freelen, /* freespace's length */
132 xfs_agblock_t *newbnop) /* result: best start block from free */
134 xfs_agblock_t freeend; /* end of freespace extent */
135 xfs_agblock_t newbno1; /* return block number */
136 xfs_agblock_t newbno2; /* other new block number */
137 xfs_extlen_t newlen1=0; /* length with newbno1 */
138 xfs_extlen_t newlen2=0; /* length with newbno2 */
139 xfs_agblock_t wantend; /* end of target extent */
141 ASSERT(freelen >= wantlen);
142 freeend = freebno + freelen;
143 wantend = wantbno + wantlen;
144 if (freebno >= wantbno) {
145 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
146 newbno1 = NULLAGBLOCK;
147 } else if (freeend >= wantend && alignment > 1) {
148 newbno1 = roundup(wantbno, alignment);
149 newbno2 = newbno1 - alignment;
150 if (newbno1 >= freeend)
151 newbno1 = NULLAGBLOCK;
153 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
154 if (newbno2 < freebno)
155 newbno2 = NULLAGBLOCK;
157 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
158 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
159 if (newlen1 < newlen2 ||
160 (newlen1 == newlen2 &&
161 XFS_ABSDIFF(newbno1, wantbno) >
162 XFS_ABSDIFF(newbno2, wantbno)))
164 } else if (newbno2 != NULLAGBLOCK)
166 } else if (freeend >= wantend) {
168 } else if (alignment > 1) {
169 newbno1 = roundup(freeend - wantlen, alignment);
170 if (newbno1 > freeend - wantlen &&
171 newbno1 - alignment >= freebno)
172 newbno1 -= alignment;
173 else if (newbno1 >= freeend)
174 newbno1 = NULLAGBLOCK;
176 newbno1 = freeend - wantlen;
178 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
182 * Fix up the length, based on mod and prod.
183 * len should be k * prod + mod for some k.
184 * If len is too small it is returned unchanged.
185 * If len hits maxlen it is left alone.
189 xfs_alloc_arg_t *args) /* allocation argument structure */
194 ASSERT(args->mod < args->prod);
196 ASSERT(rlen >= args->minlen);
197 ASSERT(rlen <= args->maxlen);
198 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
199 (args->mod == 0 && rlen < args->prod))
201 k = rlen % args->prod;
205 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
208 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
212 ASSERT(rlen >= args->minlen);
213 ASSERT(rlen <= args->maxlen);
218 * Fix up length if there is too little space left in the a.g.
219 * Return 1 if ok, 0 if too little, should give up.
222 xfs_alloc_fix_minleft(
223 xfs_alloc_arg_t *args) /* allocation argument structure */
225 xfs_agf_t *agf; /* a.g. freelist header */
226 int diff; /* free space difference */
228 if (args->minleft == 0)
230 agf = XFS_BUF_TO_AGF(args->agbp);
231 diff = be32_to_cpu(agf->agf_freeblks)
232 + be32_to_cpu(agf->agf_flcount)
233 - args->len - args->minleft;
236 args->len += diff; /* shrink the allocated space */
237 if (args->len >= args->minlen)
239 args->agbno = NULLAGBLOCK;
244 * Update the two btrees, logically removing from freespace the extent
245 * starting at rbno, rlen blocks. The extent is contained within the
246 * actual (current) free extent fbno for flen blocks.
247 * Flags are passed in indicating whether the cursors are set to the
250 STATIC int /* error code */
251 xfs_alloc_fixup_trees(
252 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
253 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
254 xfs_agblock_t fbno, /* starting block of free extent */
255 xfs_extlen_t flen, /* length of free extent */
256 xfs_agblock_t rbno, /* starting block of returned extent */
257 xfs_extlen_t rlen, /* length of returned extent */
258 int flags) /* flags, XFSA_FIXUP_... */
260 int error; /* error code */
261 int i; /* operation results */
262 xfs_agblock_t nfbno1; /* first new free startblock */
263 xfs_agblock_t nfbno2; /* second new free startblock */
264 xfs_extlen_t nflen1=0; /* first new free length */
265 xfs_extlen_t nflen2=0; /* second new free length */
268 * Look up the record in the by-size tree if necessary.
270 if (flags & XFSA_FIXUP_CNT_OK) {
272 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
274 XFS_WANT_CORRUPTED_RETURN(
275 i == 1 && nfbno1 == fbno && nflen1 == flen);
278 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
280 XFS_WANT_CORRUPTED_RETURN(i == 1);
283 * Look up the record in the by-block tree if necessary.
285 if (flags & XFSA_FIXUP_BNO_OK) {
287 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
289 XFS_WANT_CORRUPTED_RETURN(
290 i == 1 && nfbno1 == fbno && nflen1 == flen);
293 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
295 XFS_WANT_CORRUPTED_RETURN(i == 1);
299 xfs_alloc_block_t *bnoblock;
300 xfs_alloc_block_t *cntblock;
302 if (bno_cur->bc_nlevels == 1 &&
303 cnt_cur->bc_nlevels == 1) {
304 bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]);
305 cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]);
306 XFS_WANT_CORRUPTED_RETURN(
307 be16_to_cpu(bnoblock->bb_numrecs) ==
308 be16_to_cpu(cntblock->bb_numrecs));
313 * Deal with all four cases: the allocated record is contained
314 * within the freespace record, so we can have new freespace
315 * at either (or both) end, or no freespace remaining.
317 if (rbno == fbno && rlen == flen)
318 nfbno1 = nfbno2 = NULLAGBLOCK;
319 else if (rbno == fbno) {
320 nfbno1 = rbno + rlen;
321 nflen1 = flen - rlen;
322 nfbno2 = NULLAGBLOCK;
323 } else if (rbno + rlen == fbno + flen) {
325 nflen1 = flen - rlen;
326 nfbno2 = NULLAGBLOCK;
329 nflen1 = rbno - fbno;
330 nfbno2 = rbno + rlen;
331 nflen2 = (fbno + flen) - nfbno2;
334 * Delete the entry from the by-size btree.
336 if ((error = xfs_alloc_delete(cnt_cur, &i)))
338 XFS_WANT_CORRUPTED_RETURN(i == 1);
340 * Add new by-size btree entry(s).
342 if (nfbno1 != NULLAGBLOCK) {
343 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
345 XFS_WANT_CORRUPTED_RETURN(i == 0);
346 if ((error = xfs_alloc_insert(cnt_cur, &i)))
348 XFS_WANT_CORRUPTED_RETURN(i == 1);
350 if (nfbno2 != NULLAGBLOCK) {
351 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
353 XFS_WANT_CORRUPTED_RETURN(i == 0);
354 if ((error = xfs_alloc_insert(cnt_cur, &i)))
356 XFS_WANT_CORRUPTED_RETURN(i == 1);
359 * Fix up the by-block btree entry(s).
361 if (nfbno1 == NULLAGBLOCK) {
363 * No remaining freespace, just delete the by-block tree entry.
365 if ((error = xfs_alloc_delete(bno_cur, &i)))
367 XFS_WANT_CORRUPTED_RETURN(i == 1);
370 * Update the by-block entry to start later|be shorter.
372 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
375 if (nfbno2 != NULLAGBLOCK) {
377 * 2 resulting free entries, need to add one.
379 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
381 XFS_WANT_CORRUPTED_RETURN(i == 0);
382 if ((error = xfs_alloc_insert(bno_cur, &i)))
384 XFS_WANT_CORRUPTED_RETURN(i == 1);
390 * Read in the allocation group free block array.
392 STATIC int /* error */
394 xfs_mount_t *mp, /* mount point structure */
395 xfs_trans_t *tp, /* transaction pointer */
396 xfs_agnumber_t agno, /* allocation group number */
397 xfs_buf_t **bpp) /* buffer for the ag free block array */
399 xfs_buf_t *bp; /* return value */
402 ASSERT(agno != NULLAGNUMBER);
403 error = xfs_trans_read_buf(
404 mp, tp, mp->m_ddev_targp,
405 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
406 XFS_FSS_TO_BB(mp, 1), 0, &bp);
410 ASSERT(!XFS_BUF_GETERROR(bp));
411 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF);
416 #if defined(XFS_ALLOC_TRACE)
418 * Add an allocation trace entry for an alloc call.
421 xfs_alloc_trace_alloc(
422 const char *name, /* function tag string */
423 char *str, /* additional string */
424 xfs_alloc_arg_t *args, /* allocation argument structure */
425 int line) /* source line number */
427 ktrace_enter(xfs_alloc_trace_buf,
428 (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)),
432 (void *)(__psunsigned_t)args->agno,
433 (void *)(__psunsigned_t)args->agbno,
434 (void *)(__psunsigned_t)args->minlen,
435 (void *)(__psunsigned_t)args->maxlen,
436 (void *)(__psunsigned_t)args->mod,
437 (void *)(__psunsigned_t)args->prod,
438 (void *)(__psunsigned_t)args->minleft,
439 (void *)(__psunsigned_t)args->total,
440 (void *)(__psunsigned_t)args->alignment,
441 (void *)(__psunsigned_t)args->len,
442 (void *)((((__psint_t)args->type) << 16) |
443 (__psint_t)args->otype),
444 (void *)(__psint_t)((args->wasdel << 3) |
445 (args->wasfromfl << 2) |
447 (args->userdata << 0)));
451 * Add an allocation trace entry for a free call.
454 xfs_alloc_trace_free(
455 const char *name, /* function tag string */
456 char *str, /* additional string */
457 xfs_mount_t *mp, /* file system mount point */
458 xfs_agnumber_t agno, /* allocation group number */
459 xfs_agblock_t agbno, /* a.g. relative block number */
460 xfs_extlen_t len, /* length of extent */
461 int isfl, /* set if is freelist allocation/free */
462 int line) /* source line number */
464 ktrace_enter(xfs_alloc_trace_buf,
465 (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)),
469 (void *)(__psunsigned_t)agno,
470 (void *)(__psunsigned_t)agbno,
471 (void *)(__psunsigned_t)len,
472 (void *)(__psint_t)isfl,
473 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL);
477 * Add an allocation trace entry for modifying an agf.
480 xfs_alloc_trace_modagf(
481 const char *name, /* function tag string */
482 char *str, /* additional string */
483 xfs_mount_t *mp, /* file system mount point */
484 xfs_agf_t *agf, /* new agf value */
485 int flags, /* logging flags for agf */
486 int line) /* source line number */
488 ktrace_enter(xfs_alloc_trace_buf,
489 (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)),
493 (void *)(__psint_t)flags,
494 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno),
495 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length),
496 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]),
497 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]),
498 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]),
499 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]),
500 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst),
501 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast),
502 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount),
503 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks),
504 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest));
508 xfs_alloc_trace_busy(
509 const char *name, /* function tag string */
510 char *str, /* additional string */
511 xfs_mount_t *mp, /* file system mount point */
512 xfs_agnumber_t agno, /* allocation group number */
513 xfs_agblock_t agbno, /* a.g. relative block number */
514 xfs_extlen_t len, /* length of extent */
515 int slot, /* perag Busy slot */
517 int trtype, /* type: add, delete, search */
518 int line) /* source line number */
520 ktrace_enter(xfs_alloc_trace_buf,
521 (void *)(__psint_t)(trtype | (line << 16)),
525 (void *)(__psunsigned_t)agno,
526 (void *)(__psunsigned_t)agbno,
527 (void *)(__psunsigned_t)len,
528 (void *)(__psint_t)slot,
530 NULL, NULL, NULL, NULL, NULL, NULL, NULL);
532 #endif /* XFS_ALLOC_TRACE */
535 * Allocation group level functions.
539 * Allocate a variable extent in the allocation group agno.
540 * Type and bno are used to determine where in the allocation group the
542 * Extent's length (returned in *len) will be between minlen and maxlen,
543 * and of the form k * prod + mod unless there's nothing that large.
544 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
546 STATIC int /* error */
547 xfs_alloc_ag_vextent(
548 xfs_alloc_arg_t *args) /* argument structure for allocation */
552 ASSERT(args->minlen > 0);
553 ASSERT(args->maxlen > 0);
554 ASSERT(args->minlen <= args->maxlen);
555 ASSERT(args->mod < args->prod);
556 ASSERT(args->alignment > 0);
558 * Branch to correct routine based on the type.
561 switch (args->type) {
562 case XFS_ALLOCTYPE_THIS_AG:
563 error = xfs_alloc_ag_vextent_size(args);
565 case XFS_ALLOCTYPE_NEAR_BNO:
566 error = xfs_alloc_ag_vextent_near(args);
568 case XFS_ALLOCTYPE_THIS_BNO:
569 error = xfs_alloc_ag_vextent_exact(args);
578 * If the allocation worked, need to change the agf structure
579 * (and log it), and the superblock.
581 if (args->agbno != NULLAGBLOCK) {
582 xfs_agf_t *agf; /* allocation group freelist header */
583 #ifdef XFS_ALLOC_TRACE
584 xfs_mount_t *mp = args->mp;
586 long slen = (long)args->len;
588 ASSERT(args->len >= args->minlen && args->len <= args->maxlen);
589 ASSERT(!(args->wasfromfl) || !args->isfl);
590 ASSERT(args->agbno % args->alignment == 0);
591 if (!(args->wasfromfl)) {
593 agf = XFS_BUF_TO_AGF(args->agbp);
594 be32_add_cpu(&agf->agf_freeblks, -(args->len));
595 xfs_trans_agblocks_delta(args->tp,
596 -((long)(args->len)));
597 args->pag->pagf_freeblks -= args->len;
598 ASSERT(be32_to_cpu(agf->agf_freeblks) <=
599 be32_to_cpu(agf->agf_length));
600 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
601 xfs_alloc_log_agf(args->tp, args->agbp,
603 /* search the busylist for these blocks */
604 xfs_alloc_search_busy(args->tp, args->agno,
605 args->agbno, args->len);
608 xfs_trans_mod_sb(args->tp,
609 args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS :
610 XFS_TRANS_SB_FDBLOCKS, -slen);
611 XFS_STATS_INC(xs_allocx);
612 XFS_STATS_ADD(xs_allocb, args->len);
618 * Allocate a variable extent at exactly agno/bno.
619 * Extent's length (returned in *len) will be between minlen and maxlen,
620 * and of the form k * prod + mod unless there's nothing that large.
621 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
623 STATIC int /* error */
624 xfs_alloc_ag_vextent_exact(
625 xfs_alloc_arg_t *args) /* allocation argument structure */
627 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
628 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
629 xfs_agblock_t end; /* end of allocated extent */
631 xfs_agblock_t fbno; /* start block of found extent */
632 xfs_agblock_t fend; /* end block of found extent */
633 xfs_extlen_t flen; /* length of found extent */
634 int i; /* success/failure of operation */
635 xfs_agblock_t maxend; /* end of maximal extent */
636 xfs_agblock_t minend; /* end of minimal extent */
637 xfs_extlen_t rlen; /* length of returned extent */
639 ASSERT(args->alignment == 1);
641 * Allocate/initialize a cursor for the by-number freespace btree.
643 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
644 args->agno, XFS_BTNUM_BNO, NULL, 0);
646 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
647 * Look for the closest free block <= bno, it must contain bno
648 * if any free block does.
650 if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i)))
654 * Didn't find it, return null.
656 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
657 args->agbno = NULLAGBLOCK;
661 * Grab the freespace record.
663 if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i)))
665 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
666 ASSERT(fbno <= args->agbno);
667 minend = args->agbno + args->minlen;
668 maxend = args->agbno + args->maxlen;
671 * Give up if the freespace isn't long enough for the minimum request.
674 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
675 args->agbno = NULLAGBLOCK;
679 * End of extent will be smaller of the freespace end and the
680 * maximal requested end.
682 end = XFS_AGBLOCK_MIN(fend, maxend);
684 * Fix the length according to mod and prod if given.
686 args->len = end - args->agbno;
687 xfs_alloc_fix_len(args);
688 if (!xfs_alloc_fix_minleft(args)) {
689 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
693 ASSERT(args->agbno + rlen <= fend);
694 end = args->agbno + rlen;
696 * We are allocating agbno for rlen [agbno .. end]
697 * Allocate/initialize a cursor for the by-size btree.
699 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
700 args->agno, XFS_BTNUM_CNT, NULL, 0);
701 ASSERT(args->agbno + args->len <=
702 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
703 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
704 args->agbno, args->len, XFSA_FIXUP_BNO_OK))) {
705 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
708 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
709 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
710 TRACE_ALLOC("normal", args);
715 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
716 TRACE_ALLOC("error", args);
721 * Allocate a variable extent near bno in the allocation group agno.
722 * Extent's length (returned in len) will be between minlen and maxlen,
723 * and of the form k * prod + mod unless there's nothing that large.
724 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
726 STATIC int /* error */
727 xfs_alloc_ag_vextent_near(
728 xfs_alloc_arg_t *args) /* allocation argument structure */
730 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
731 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
732 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
733 xfs_agblock_t gtbno; /* start bno of right side entry */
734 xfs_agblock_t gtbnoa; /* aligned ... */
735 xfs_extlen_t gtdiff; /* difference to right side entry */
736 xfs_extlen_t gtlen; /* length of right side entry */
737 xfs_extlen_t gtlena; /* aligned ... */
738 xfs_agblock_t gtnew; /* useful start bno of right side */
739 int error; /* error code */
740 int i; /* result code, temporary */
741 int j; /* result code, temporary */
742 xfs_agblock_t ltbno; /* start bno of left side entry */
743 xfs_agblock_t ltbnoa; /* aligned ... */
744 xfs_extlen_t ltdiff; /* difference to left side entry */
746 xfs_agblock_t ltend; /* end bno of left side entry */
747 xfs_extlen_t ltlen; /* length of left side entry */
748 xfs_extlen_t ltlena; /* aligned ... */
749 xfs_agblock_t ltnew; /* useful start bno of left side */
750 xfs_extlen_t rlen; /* length of returned extent */
751 #if defined(DEBUG) && defined(__KERNEL__)
753 * Randomly don't execute the first algorithm.
755 int dofirst; /* set to do first algorithm */
757 dofirst = random32() & 1;
760 * Get a cursor for the by-size btree.
762 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
763 args->agno, XFS_BTNUM_CNT, NULL, 0);
765 bno_cur_lt = bno_cur_gt = NULL;
767 * See if there are any free extents as big as maxlen.
769 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
772 * If none, then pick up the last entry in the tree unless the
776 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
779 if (i == 0 || ltlen == 0) {
780 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
788 * If the requested extent is large wrt the freespaces available
789 * in this a.g., then the cursor will be pointing to a btree entry
790 * near the right edge of the tree. If it's in the last btree leaf
791 * block, then we just examine all the entries in that block
792 * that are big enough, and pick the best one.
793 * This is written as a while loop so we can break out of it,
794 * but we never loop back to the top.
796 while (xfs_btree_islastblock(cnt_cur, 0)) {
800 xfs_agblock_t bnew=0;
802 #if defined(DEBUG) && defined(__KERNEL__)
807 * Start from the entry that lookup found, sequence through
808 * all larger free blocks. If we're actually pointing at a
809 * record smaller than maxlen, go to the start of this block,
810 * and skip all those smaller than minlen.
812 if (ltlen || args->alignment > 1) {
813 cnt_cur->bc_ptrs[0] = 1;
815 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
818 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
819 if (ltlen >= args->minlen)
821 if ((error = xfs_alloc_increment(cnt_cur, 0, &i)))
824 ASSERT(ltlen >= args->minlen);
828 i = cnt_cur->bc_ptrs[0];
829 for (j = 1, blen = 0, bdiff = 0;
830 !error && j && (blen < args->maxlen || bdiff > 0);
831 error = xfs_alloc_increment(cnt_cur, 0, &j)) {
833 * For each entry, decide if it's better than
834 * the previous best entry.
836 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
838 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
839 xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment,
840 args->minlen, <bnoa, <lena);
841 if (ltlena < args->minlen)
843 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
844 xfs_alloc_fix_len(args);
845 ASSERT(args->len >= args->minlen);
846 if (args->len < blen)
848 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
849 args->alignment, ltbno, ltlen, <new);
850 if (ltnew != NULLAGBLOCK &&
851 (args->len > blen || ltdiff < bdiff)) {
855 besti = cnt_cur->bc_ptrs[0];
859 * It didn't work. We COULD be in a case where
860 * there's a good record somewhere, so try again.
865 * Point at the best entry, and retrieve it again.
867 cnt_cur->bc_ptrs[0] = besti;
868 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
870 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
871 ltend = ltbno + ltlen;
872 ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
874 if (!xfs_alloc_fix_minleft(args)) {
875 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
876 TRACE_ALLOC("nominleft", args);
881 * We are allocating starting at bnew for blen blocks.
884 ASSERT(bnew >= ltbno);
885 ASSERT(bnew + blen <= ltend);
887 * Set up a cursor for the by-bno tree.
889 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp,
890 args->agbp, args->agno, XFS_BTNUM_BNO, NULL, 0);
892 * Fix up the btree entries.
894 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
895 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
897 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
898 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
899 TRACE_ALLOC("first", args);
904 * Search in the by-bno tree to the left and to the right
905 * simultaneously, until in each case we find a space big enough,
906 * or run into the edge of the tree. When we run into the edge,
907 * we deallocate that cursor.
908 * If both searches succeed, we compare the two spaces and pick
910 * With alignment, it's possible for both to fail; the upper
911 * level algorithm that picks allocation groups for allocations
912 * is not supposed to do this.
915 * Allocate and initialize the cursor for the leftward search.
917 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
918 args->agno, XFS_BTNUM_BNO, NULL, 0);
920 * Lookup <= bno to find the leftward search's starting point.
922 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
926 * Didn't find anything; use this cursor for the rightward
929 bno_cur_gt = bno_cur_lt;
933 * Found something. Duplicate the cursor for the rightward search.
935 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
938 * Increment the cursor, so we will point at the entry just right
939 * of the leftward entry if any, or to the leftmost entry.
941 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
945 * It failed, there are no rightward entries.
947 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
951 * Loop going left with the leftward cursor, right with the
952 * rightward cursor, until either both directions give up or
953 * we find an entry at least as big as minlen.
957 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
959 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
960 xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment,
961 args->minlen, <bnoa, <lena);
962 if (ltlena >= args->minlen)
964 if ((error = xfs_alloc_decrement(bno_cur_lt, 0, &i)))
967 xfs_btree_del_cursor(bno_cur_lt,
973 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
975 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
976 xfs_alloc_compute_aligned(gtbno, gtlen, args->alignment,
977 args->minlen, >bnoa, >lena);
978 if (gtlena >= args->minlen)
980 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
983 xfs_btree_del_cursor(bno_cur_gt,
988 } while (bno_cur_lt || bno_cur_gt);
990 * Got both cursors still active, need to find better entry.
992 if (bno_cur_lt && bno_cur_gt) {
994 * Left side is long enough, look for a right side entry.
996 if (ltlena >= args->minlen) {
1000 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1001 xfs_alloc_fix_len(args);
1003 ltdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1004 args->alignment, ltbno, ltlen, <new);
1010 * Look until we find a better one, run out of
1011 * space, or run off the end.
1013 while (bno_cur_lt && bno_cur_gt) {
1014 if ((error = xfs_alloc_get_rec(
1018 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1019 xfs_alloc_compute_aligned(gtbno, gtlen,
1020 args->alignment, args->minlen,
1023 * The left one is clearly better.
1025 if (gtbnoa >= args->agbno + ltdiff) {
1026 xfs_btree_del_cursor(
1033 * If we reach a big enough entry,
1034 * compare the two and pick the best.
1036 if (gtlena >= args->minlen) {
1038 XFS_EXTLEN_MIN(gtlena,
1040 xfs_alloc_fix_len(args);
1042 gtdiff = xfs_alloc_compute_diff(
1045 gtbno, gtlen, >new);
1047 * Right side is better.
1049 if (gtdiff < ltdiff) {
1050 xfs_btree_del_cursor(
1056 * Left side is better.
1059 xfs_btree_del_cursor(
1067 * Fell off the right end.
1069 if ((error = xfs_alloc_increment(
1070 bno_cur_gt, 0, &i)))
1073 xfs_btree_del_cursor(
1082 * The left side is perfect, trash the right side.
1085 xfs_btree_del_cursor(bno_cur_gt,
1091 * It's the right side that was found first, look left.
1095 * Fix up the length.
1097 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1098 xfs_alloc_fix_len(args);
1100 gtdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1101 args->alignment, gtbno, gtlen, >new);
1103 * Right side entry isn't perfect.
1107 * Look until we find a better one, run out of
1108 * space, or run off the end.
1110 while (bno_cur_lt && bno_cur_gt) {
1111 if ((error = xfs_alloc_get_rec(
1115 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1116 xfs_alloc_compute_aligned(ltbno, ltlen,
1117 args->alignment, args->minlen,
1120 * The right one is clearly better.
1122 if (ltbnoa <= args->agbno - gtdiff) {
1123 xfs_btree_del_cursor(
1130 * If we reach a big enough entry,
1131 * compare the two and pick the best.
1133 if (ltlena >= args->minlen) {
1134 args->len = XFS_EXTLEN_MIN(
1135 ltlena, args->maxlen);
1136 xfs_alloc_fix_len(args);
1138 ltdiff = xfs_alloc_compute_diff(
1141 ltbno, ltlen, <new);
1143 * Left side is better.
1145 if (ltdiff < gtdiff) {
1146 xfs_btree_del_cursor(
1152 * Right side is better.
1155 xfs_btree_del_cursor(
1163 * Fell off the left end.
1165 if ((error = xfs_alloc_decrement(
1166 bno_cur_lt, 0, &i)))
1169 xfs_btree_del_cursor(bno_cur_lt,
1177 * The right side is perfect, trash the left side.
1180 xfs_btree_del_cursor(bno_cur_lt,
1187 * If we couldn't get anything, give up.
1189 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1190 TRACE_ALLOC("neither", args);
1191 args->agbno = NULLAGBLOCK;
1195 * At this point we have selected a freespace entry, either to the
1196 * left or to the right. If it's on the right, copy all the
1197 * useful variables to the "left" set so we only have one
1198 * copy of this code.
1201 bno_cur_lt = bno_cur_gt;
1211 * Fix up the length and compute the useful address.
1213 ltend = ltbno + ltlen;
1214 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1215 xfs_alloc_fix_len(args);
1216 if (!xfs_alloc_fix_minleft(args)) {
1217 TRACE_ALLOC("nominleft", args);
1218 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1219 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1223 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno,
1225 ASSERT(ltnew >= ltbno);
1226 ASSERT(ltnew + rlen <= ltend);
1227 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1228 args->agbno = ltnew;
1229 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1230 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1232 TRACE_ALLOC(j ? "gt" : "lt", args);
1233 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1234 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1238 TRACE_ALLOC("error", args);
1239 if (cnt_cur != NULL)
1240 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1241 if (bno_cur_lt != NULL)
1242 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1243 if (bno_cur_gt != NULL)
1244 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1249 * Allocate a variable extent anywhere in the allocation group agno.
1250 * Extent's length (returned in len) will be between minlen and maxlen,
1251 * and of the form k * prod + mod unless there's nothing that large.
1252 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1254 STATIC int /* error */
1255 xfs_alloc_ag_vextent_size(
1256 xfs_alloc_arg_t *args) /* allocation argument structure */
1258 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1259 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1260 int error; /* error result */
1261 xfs_agblock_t fbno; /* start of found freespace */
1262 xfs_extlen_t flen; /* length of found freespace */
1263 int i; /* temp status variable */
1264 xfs_agblock_t rbno; /* returned block number */
1265 xfs_extlen_t rlen; /* length of returned extent */
1268 * Allocate and initialize a cursor for the by-size btree.
1270 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1271 args->agno, XFS_BTNUM_CNT, NULL, 0);
1274 * Look for an entry >= maxlen+alignment-1 blocks.
1276 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1277 args->maxlen + args->alignment - 1, &i)))
1280 * If none, then pick up the last entry in the tree unless the
1284 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno,
1287 if (i == 0 || flen == 0) {
1288 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1289 TRACE_ALLOC("noentry", args);
1295 * There's a freespace as big as maxlen+alignment-1, get it.
1298 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i)))
1300 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1303 * In the first case above, we got the last entry in the
1304 * by-size btree. Now we check to see if the space hits maxlen
1305 * once aligned; if not, we search left for something better.
1306 * This can't happen in the second case above.
1308 xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen,
1310 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1311 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1312 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1313 if (rlen < args->maxlen) {
1314 xfs_agblock_t bestfbno;
1315 xfs_extlen_t bestflen;
1316 xfs_agblock_t bestrbno;
1317 xfs_extlen_t bestrlen;
1324 if ((error = xfs_alloc_decrement(cnt_cur, 0, &i)))
1328 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1331 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1332 if (flen < bestrlen)
1334 xfs_alloc_compute_aligned(fbno, flen, args->alignment,
1335 args->minlen, &rbno, &rlen);
1336 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1337 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1338 (rlen <= flen && rbno + rlen <= fbno + flen),
1340 if (rlen > bestrlen) {
1345 if (rlen == args->maxlen)
1349 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1352 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1358 args->wasfromfl = 0;
1360 * Fix up the length.
1363 xfs_alloc_fix_len(args);
1364 if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) {
1365 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1366 TRACE_ALLOC("nominleft", args);
1367 args->agbno = NULLAGBLOCK;
1371 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1373 * Allocate and initialize a cursor for the by-block tree.
1375 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1376 args->agno, XFS_BTNUM_BNO, NULL, 0);
1377 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1378 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1380 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1381 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1382 cnt_cur = bno_cur = NULL;
1385 XFS_WANT_CORRUPTED_GOTO(
1386 args->agbno + args->len <=
1387 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1389 TRACE_ALLOC("normal", args);
1393 TRACE_ALLOC("error", args);
1395 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1397 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1402 * Deal with the case where only small freespaces remain.
1403 * Either return the contents of the last freespace record,
1404 * or allocate space from the freelist if there is nothing in the tree.
1406 STATIC int /* error */
1407 xfs_alloc_ag_vextent_small(
1408 xfs_alloc_arg_t *args, /* allocation argument structure */
1409 xfs_btree_cur_t *ccur, /* by-size cursor */
1410 xfs_agblock_t *fbnop, /* result block number */
1411 xfs_extlen_t *flenp, /* result length */
1412 int *stat) /* status: 0-freelist, 1-normal/none */
1419 if ((error = xfs_alloc_decrement(ccur, 0, &i)))
1422 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1424 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1427 * Nothing in the btree, try the freelist. Make sure
1428 * to respect minleft even when pulling from the
1431 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1432 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1434 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1437 if (fbno != NULLAGBLOCK) {
1438 if (args->userdata) {
1441 bp = xfs_btree_get_bufs(args->mp, args->tp,
1442 args->agno, fbno, 0);
1443 xfs_trans_binval(args->tp, bp);
1447 XFS_WANT_CORRUPTED_GOTO(
1448 args->agbno + args->len <=
1449 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1451 args->wasfromfl = 1;
1452 TRACE_ALLOC("freelist", args);
1457 * Nothing in the freelist.
1463 * Can't allocate from the freelist for some reason.
1470 * Can't do the allocation, give up.
1472 if (flen < args->minlen) {
1473 args->agbno = NULLAGBLOCK;
1474 TRACE_ALLOC("notenough", args);
1480 TRACE_ALLOC("normal", args);
1484 TRACE_ALLOC("error", args);
1489 * Free the extent starting at agno/bno for length.
1491 STATIC int /* error */
1493 xfs_trans_t *tp, /* transaction pointer */
1494 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1495 xfs_agnumber_t agno, /* allocation group number */
1496 xfs_agblock_t bno, /* starting block number */
1497 xfs_extlen_t len, /* length of extent */
1498 int isfl) /* set if is freelist blocks - no sb acctg */
1500 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1501 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1502 int error; /* error return value */
1503 xfs_agblock_t gtbno; /* start of right neighbor block */
1504 xfs_extlen_t gtlen; /* length of right neighbor block */
1505 int haveleft; /* have a left neighbor block */
1506 int haveright; /* have a right neighbor block */
1507 int i; /* temp, result code */
1508 xfs_agblock_t ltbno; /* start of left neighbor block */
1509 xfs_extlen_t ltlen; /* length of left neighbor block */
1510 xfs_mount_t *mp; /* mount point struct for filesystem */
1511 xfs_agblock_t nbno; /* new starting block of freespace */
1512 xfs_extlen_t nlen; /* new length of freespace */
1516 * Allocate and initialize a cursor for the by-block btree.
1518 bno_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO, NULL,
1522 * Look for a neighboring block on the left (lower block numbers)
1523 * that is contiguous with this space.
1525 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1529 * There is a block to our left.
1531 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1533 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1535 * It's not contiguous, though.
1537 if (ltbno + ltlen < bno)
1541 * If this failure happens the request to free this
1542 * space was invalid, it's (partly) already free.
1545 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1549 * Look for a neighboring block on the right (higher block numbers)
1550 * that is contiguous with this space.
1552 if ((error = xfs_alloc_increment(bno_cur, 0, &haveright)))
1556 * There is a block to our right.
1558 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1560 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1562 * It's not contiguous, though.
1564 if (bno + len < gtbno)
1568 * If this failure happens the request to free this
1569 * space was invalid, it's (partly) already free.
1572 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1576 * Now allocate and initialize a cursor for the by-size tree.
1578 cnt_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT, NULL,
1581 * Have both left and right contiguous neighbors.
1582 * Merge all three into a single free block.
1584 if (haveleft && haveright) {
1586 * Delete the old by-size entry on the left.
1588 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1590 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1591 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1593 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1595 * Delete the old by-size entry on the right.
1597 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1599 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1600 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1602 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1604 * Delete the old by-block entry for the right block.
1606 if ((error = xfs_alloc_delete(bno_cur, &i)))
1608 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1610 * Move the by-block cursor back to the left neighbor.
1612 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1614 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1617 * Check that this is the right record: delete didn't
1618 * mangle the cursor.
1621 xfs_agblock_t xxbno;
1624 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1627 XFS_WANT_CORRUPTED_GOTO(
1628 i == 1 && xxbno == ltbno && xxlen == ltlen,
1633 * Update remaining by-block entry to the new, joined block.
1636 nlen = len + ltlen + gtlen;
1637 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1641 * Have only a left contiguous neighbor.
1642 * Merge it together with the new freespace.
1644 else if (haveleft) {
1646 * Delete the old by-size entry on the left.
1648 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1650 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1651 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1653 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1655 * Back up the by-block cursor to the left neighbor, and
1656 * update its length.
1658 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1660 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1663 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1667 * Have only a right contiguous neighbor.
1668 * Merge it together with the new freespace.
1670 else if (haveright) {
1672 * Delete the old by-size entry on the right.
1674 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1676 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1677 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1679 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1681 * Update the starting block and length of the right
1682 * neighbor in the by-block tree.
1686 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1690 * No contiguous neighbors.
1691 * Insert the new freespace into the by-block tree.
1696 if ((error = xfs_alloc_insert(bno_cur, &i)))
1698 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1700 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1703 * In all cases we need to insert the new freespace in the by-size tree.
1705 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1707 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1708 if ((error = xfs_alloc_insert(cnt_cur, &i)))
1710 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1711 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1714 * Update the freespace totals in the ag and superblock.
1718 xfs_perag_t *pag; /* per allocation group data */
1720 agf = XFS_BUF_TO_AGF(agbp);
1721 pag = &mp->m_perag[agno];
1722 be32_add_cpu(&agf->agf_freeblks, len);
1723 xfs_trans_agblocks_delta(tp, len);
1724 pag->pagf_freeblks += len;
1725 XFS_WANT_CORRUPTED_GOTO(
1726 be32_to_cpu(agf->agf_freeblks) <=
1727 be32_to_cpu(agf->agf_length),
1729 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
1730 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
1732 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1733 XFS_STATS_INC(xs_freex);
1734 XFS_STATS_ADD(xs_freeb, len);
1736 TRACE_FREE(haveleft ?
1737 (haveright ? "both" : "left") :
1738 (haveright ? "right" : "none"),
1739 agno, bno, len, isfl);
1742 * Since blocks move to the free list without the coordination
1743 * used in xfs_bmap_finish, we can't allow block to be available
1744 * for reallocation and non-transaction writing (user data)
1745 * until we know that the transaction that moved it to the free
1746 * list is permanently on disk. We track the blocks by declaring
1747 * these blocks as "busy"; the busy list is maintained on a per-ag
1748 * basis and each transaction records which entries should be removed
1749 * when the iclog commits to disk. If a busy block is allocated,
1750 * the iclog is pushed up to the LSN that freed the block.
1752 xfs_alloc_mark_busy(tp, agno, bno, len);
1756 TRACE_FREE("error", agno, bno, len, isfl);
1758 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1760 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1765 * Visible (exported) allocation/free functions.
1766 * Some of these are used just by xfs_alloc_btree.c and this file.
1770 * Compute and fill in value of m_ag_maxlevels.
1773 xfs_alloc_compute_maxlevels(
1774 xfs_mount_t *mp) /* file system mount structure */
1782 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1783 minleafrecs = mp->m_alloc_mnr[0];
1784 minnoderecs = mp->m_alloc_mnr[1];
1785 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1786 for (level = 1; maxblocks > 1; level++)
1787 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1788 mp->m_ag_maxlevels = level;
1792 * Decide whether to use this allocation group for this allocation.
1793 * If so, fix up the btree freelist's size.
1795 STATIC int /* error */
1796 xfs_alloc_fix_freelist(
1797 xfs_alloc_arg_t *args, /* allocation argument structure */
1798 int flags) /* XFS_ALLOC_FLAG_... */
1800 xfs_buf_t *agbp; /* agf buffer pointer */
1801 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1802 xfs_buf_t *agflbp;/* agfl buffer pointer */
1803 xfs_agblock_t bno; /* freelist block */
1804 xfs_extlen_t delta; /* new blocks needed in freelist */
1805 int error; /* error result code */
1806 xfs_extlen_t longest;/* longest extent in allocation group */
1807 xfs_mount_t *mp; /* file system mount point structure */
1808 xfs_extlen_t need; /* total blocks needed in freelist */
1809 xfs_perag_t *pag; /* per-ag information structure */
1810 xfs_alloc_arg_t targs; /* local allocation arguments */
1811 xfs_trans_t *tp; /* transaction pointer */
1817 if (!pag->pagf_init) {
1818 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1821 if (!pag->pagf_init) {
1822 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1823 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1831 * If this is a metadata preferred pag and we are user data
1832 * then try somewhere else if we are not being asked to
1833 * try harder at this point
1835 if (pag->pagf_metadata && args->userdata &&
1836 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1837 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1842 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1843 need = XFS_MIN_FREELIST_PAG(pag, mp);
1844 delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0;
1846 * If it looks like there isn't a long enough extent, or enough
1847 * total blocks, reject it.
1849 longest = (pag->pagf_longest > delta) ?
1850 (pag->pagf_longest - delta) :
1851 (pag->pagf_flcount > 0 || pag->pagf_longest > 0);
1852 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1854 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1855 need - args->total) < (int)args->minleft)) {
1857 xfs_trans_brelse(tp, agbp);
1864 * Get the a.g. freespace buffer.
1865 * Can fail if we're not blocking on locks, and it's held.
1868 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1872 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1873 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1879 * Figure out how many blocks we should have in the freelist.
1881 agf = XFS_BUF_TO_AGF(agbp);
1882 need = XFS_MIN_FREELIST(agf, mp);
1884 * If there isn't enough total or single-extent, reject it.
1886 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1887 delta = need > be32_to_cpu(agf->agf_flcount) ?
1888 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1889 longest = be32_to_cpu(agf->agf_longest);
1890 longest = (longest > delta) ? (longest - delta) :
1891 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1892 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1894 ((int)(be32_to_cpu(agf->agf_freeblks) +
1895 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1896 (int)args->minleft)) {
1897 xfs_trans_brelse(tp, agbp);
1903 * Make the freelist shorter if it's too long.
1905 while (be32_to_cpu(agf->agf_flcount) > need) {
1908 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1911 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1913 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1914 xfs_trans_binval(tp, bp);
1917 * Initialize the args structure.
1922 targs.agno = args->agno;
1923 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1924 targs.minalignslop = 0;
1925 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1926 targs.type = XFS_ALLOCTYPE_THIS_AG;
1928 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1931 * Make the freelist longer if it's too short.
1933 while (be32_to_cpu(agf->agf_flcount) < need) {
1935 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1937 * Allocate as many blocks as possible at once.
1939 if ((error = xfs_alloc_ag_vextent(&targs))) {
1940 xfs_trans_brelse(tp, agflbp);
1944 * Stop if we run out. Won't happen if callers are obeying
1945 * the restrictions correctly. Can happen for free calls
1946 * on a completely full ag.
1948 if (targs.agbno == NULLAGBLOCK) {
1949 if (flags & XFS_ALLOC_FLAG_FREEING)
1951 xfs_trans_brelse(tp, agflbp);
1956 * Put each allocated block on the list.
1958 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1959 error = xfs_alloc_put_freelist(tp, agbp,
1965 xfs_trans_brelse(tp, agflbp);
1971 * Get a block from the freelist.
1972 * Returns with the buffer for the block gotten.
1975 xfs_alloc_get_freelist(
1976 xfs_trans_t *tp, /* transaction pointer */
1977 xfs_buf_t *agbp, /* buffer containing the agf structure */
1978 xfs_agblock_t *bnop, /* block address retrieved from freelist */
1979 int btreeblk) /* destination is a AGF btree */
1981 xfs_agf_t *agf; /* a.g. freespace structure */
1982 xfs_agfl_t *agfl; /* a.g. freelist structure */
1983 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1984 xfs_agblock_t bno; /* block number returned */
1987 xfs_mount_t *mp; /* mount structure */
1988 xfs_perag_t *pag; /* per allocation group data */
1990 agf = XFS_BUF_TO_AGF(agbp);
1992 * Freelist is empty, give up.
1994 if (!agf->agf_flcount) {
1995 *bnop = NULLAGBLOCK;
1999 * Read the array of free blocks.
2002 if ((error = xfs_alloc_read_agfl(mp, tp,
2003 be32_to_cpu(agf->agf_seqno), &agflbp)))
2005 agfl = XFS_BUF_TO_AGFL(agflbp);
2007 * Get the block number and update the data structures.
2009 bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2010 be32_add_cpu(&agf->agf_flfirst, 1);
2011 xfs_trans_brelse(tp, agflbp);
2012 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
2013 agf->agf_flfirst = 0;
2014 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2015 be32_add_cpu(&agf->agf_flcount, -1);
2016 xfs_trans_agflist_delta(tp, -1);
2017 pag->pagf_flcount--;
2019 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2021 be32_add_cpu(&agf->agf_btreeblks, 1);
2022 pag->pagf_btreeblks++;
2023 logflags |= XFS_AGF_BTREEBLKS;
2026 TRACE_MODAGF(NULL, agf, logflags);
2027 xfs_alloc_log_agf(tp, agbp, logflags);
2031 * As blocks are freed, they are added to the per-ag busy list
2032 * and remain there until the freeing transaction is committed to
2033 * disk. Now that we have allocated blocks, this list must be
2034 * searched to see if a block is being reused. If one is, then
2035 * the freeing transaction must be pushed to disk NOW by forcing
2036 * to disk all iclogs up that transaction's LSN.
2038 xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1);
2043 * Log the given fields from the agf structure.
2047 xfs_trans_t *tp, /* transaction pointer */
2048 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2049 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2051 int first; /* first byte offset */
2052 int last; /* last byte offset */
2053 static const short offsets[] = {
2054 offsetof(xfs_agf_t, agf_magicnum),
2055 offsetof(xfs_agf_t, agf_versionnum),
2056 offsetof(xfs_agf_t, agf_seqno),
2057 offsetof(xfs_agf_t, agf_length),
2058 offsetof(xfs_agf_t, agf_roots[0]),
2059 offsetof(xfs_agf_t, agf_levels[0]),
2060 offsetof(xfs_agf_t, agf_flfirst),
2061 offsetof(xfs_agf_t, agf_fllast),
2062 offsetof(xfs_agf_t, agf_flcount),
2063 offsetof(xfs_agf_t, agf_freeblks),
2064 offsetof(xfs_agf_t, agf_longest),
2065 offsetof(xfs_agf_t, agf_btreeblks),
2069 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2070 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2074 * Interface for inode allocation to force the pag data to be initialized.
2077 xfs_alloc_pagf_init(
2078 xfs_mount_t *mp, /* file system mount structure */
2079 xfs_trans_t *tp, /* transaction pointer */
2080 xfs_agnumber_t agno, /* allocation group number */
2081 int flags) /* XFS_ALLOC_FLAGS_... */
2086 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2089 xfs_trans_brelse(tp, bp);
2094 * Put the block on the freelist for the allocation group.
2097 xfs_alloc_put_freelist(
2098 xfs_trans_t *tp, /* transaction pointer */
2099 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2100 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2101 xfs_agblock_t bno, /* block being freed */
2102 int btreeblk) /* block came from a AGF btree */
2104 xfs_agf_t *agf; /* a.g. freespace structure */
2105 xfs_agfl_t *agfl; /* a.g. free block array */
2106 __be32 *blockp;/* pointer to array entry */
2109 xfs_mount_t *mp; /* mount structure */
2110 xfs_perag_t *pag; /* per allocation group data */
2112 agf = XFS_BUF_TO_AGF(agbp);
2115 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2116 be32_to_cpu(agf->agf_seqno), &agflbp)))
2118 agfl = XFS_BUF_TO_AGFL(agflbp);
2119 be32_add_cpu(&agf->agf_fllast, 1);
2120 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2121 agf->agf_fllast = 0;
2122 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2123 be32_add_cpu(&agf->agf_flcount, 1);
2124 xfs_trans_agflist_delta(tp, 1);
2125 pag->pagf_flcount++;
2127 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2129 be32_add_cpu(&agf->agf_btreeblks, -1);
2130 pag->pagf_btreeblks--;
2131 logflags |= XFS_AGF_BTREEBLKS;
2134 TRACE_MODAGF(NULL, agf, logflags);
2135 xfs_alloc_log_agf(tp, agbp, logflags);
2137 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2138 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
2139 *blockp = cpu_to_be32(bno);
2140 TRACE_MODAGF(NULL, agf, logflags);
2141 xfs_alloc_log_agf(tp, agbp, logflags);
2142 xfs_trans_log_buf(tp, agflbp,
2143 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2144 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2145 sizeof(xfs_agblock_t) - 1));
2150 * Read in the allocation group header (free/alloc section).
2154 xfs_mount_t *mp, /* mount point structure */
2155 xfs_trans_t *tp, /* transaction pointer */
2156 xfs_agnumber_t agno, /* allocation group number */
2157 int flags, /* XFS_ALLOC_FLAG_... */
2158 xfs_buf_t **bpp) /* buffer for the ag freelist header */
2160 xfs_agf_t *agf; /* ag freelist header */
2161 int agf_ok; /* set if agf is consistent */
2162 xfs_buf_t *bp; /* return value */
2163 xfs_perag_t *pag; /* per allocation group data */
2166 ASSERT(agno != NULLAGNUMBER);
2167 error = xfs_trans_read_buf(
2168 mp, tp, mp->m_ddev_targp,
2169 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2170 XFS_FSS_TO_BB(mp, 1),
2171 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U,
2175 ASSERT(!bp || !XFS_BUF_GETERROR(bp));
2181 * Validate the magic number of the agf block.
2183 agf = XFS_BUF_TO_AGF(bp);
2185 be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC &&
2186 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2187 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2188 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2189 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2190 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp);
2191 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2192 XFS_RANDOM_ALLOC_READ_AGF))) {
2193 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2194 XFS_ERRLEVEL_LOW, mp, agf);
2195 xfs_trans_brelse(tp, bp);
2196 return XFS_ERROR(EFSCORRUPTED);
2198 pag = &mp->m_perag[agno];
2199 if (!pag->pagf_init) {
2200 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2201 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2202 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2203 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2204 pag->pagf_levels[XFS_BTNUM_BNOi] =
2205 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2206 pag->pagf_levels[XFS_BTNUM_CNTi] =
2207 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2208 spin_lock_init(&pag->pagb_lock);
2209 pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS *
2210 sizeof(xfs_perag_busy_t), KM_SLEEP);
2214 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2215 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2216 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2217 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2218 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2219 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2220 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2221 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2224 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF);
2230 * Allocate an extent (variable-size).
2231 * Depending on the allocation type, we either look in a single allocation
2232 * group or loop over the allocation groups to find the result.
2236 xfs_alloc_arg_t *args) /* allocation argument structure */
2238 xfs_agblock_t agsize; /* allocation group size */
2240 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2241 xfs_extlen_t minleft;/* minimum left value, temp copy */
2242 xfs_mount_t *mp; /* mount structure pointer */
2243 xfs_agnumber_t sagno; /* starting allocation group number */
2244 xfs_alloctype_t type; /* input allocation type */
2247 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2250 type = args->otype = args->type;
2251 args->agbno = NULLAGBLOCK;
2253 * Just fix this up, for the case where the last a.g. is shorter
2254 * (or there's only one a.g.) and the caller couldn't easily figure
2255 * that out (xfs_bmap_alloc).
2257 agsize = mp->m_sb.sb_agblocks;
2258 if (args->maxlen > agsize)
2259 args->maxlen = agsize;
2260 if (args->alignment == 0)
2261 args->alignment = 1;
2262 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2263 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2264 ASSERT(args->minlen <= args->maxlen);
2265 ASSERT(args->minlen <= agsize);
2266 ASSERT(args->mod < args->prod);
2267 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2268 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2269 args->minlen > args->maxlen || args->minlen > agsize ||
2270 args->mod >= args->prod) {
2271 args->fsbno = NULLFSBLOCK;
2272 TRACE_ALLOC("badargs", args);
2275 minleft = args->minleft;
2278 case XFS_ALLOCTYPE_THIS_AG:
2279 case XFS_ALLOCTYPE_NEAR_BNO:
2280 case XFS_ALLOCTYPE_THIS_BNO:
2282 * These three force us into a single a.g.
2284 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2285 down_read(&mp->m_peraglock);
2286 args->pag = &mp->m_perag[args->agno];
2288 error = xfs_alloc_fix_freelist(args, 0);
2289 args->minleft = minleft;
2291 TRACE_ALLOC("nofix", args);
2295 up_read(&mp->m_peraglock);
2296 TRACE_ALLOC("noagbp", args);
2299 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2300 if ((error = xfs_alloc_ag_vextent(args)))
2302 up_read(&mp->m_peraglock);
2304 case XFS_ALLOCTYPE_START_BNO:
2306 * Try near allocation first, then anywhere-in-ag after
2307 * the first a.g. fails.
2309 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2310 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2311 args->fsbno = XFS_AGB_TO_FSB(mp,
2312 ((mp->m_agfrotor / rotorstep) %
2313 mp->m_sb.sb_agcount), 0);
2316 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2317 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2319 case XFS_ALLOCTYPE_ANY_AG:
2320 case XFS_ALLOCTYPE_START_AG:
2321 case XFS_ALLOCTYPE_FIRST_AG:
2323 * Rotate through the allocation groups looking for a winner.
2325 if (type == XFS_ALLOCTYPE_ANY_AG) {
2327 * Start with the last place we left off.
2329 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2330 mp->m_sb.sb_agcount;
2331 args->type = XFS_ALLOCTYPE_THIS_AG;
2332 flags = XFS_ALLOC_FLAG_TRYLOCK;
2333 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2335 * Start with allocation group given by bno.
2337 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2338 args->type = XFS_ALLOCTYPE_THIS_AG;
2342 if (type == XFS_ALLOCTYPE_START_AG)
2343 args->type = XFS_ALLOCTYPE_THIS_AG;
2345 * Start with the given allocation group.
2347 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2348 flags = XFS_ALLOC_FLAG_TRYLOCK;
2351 * Loop over allocation groups twice; first time with
2352 * trylock set, second time without.
2354 down_read(&mp->m_peraglock);
2356 args->pag = &mp->m_perag[args->agno];
2357 if (no_min) args->minleft = 0;
2358 error = xfs_alloc_fix_freelist(args, flags);
2359 args->minleft = minleft;
2361 TRACE_ALLOC("nofix", args);
2365 * If we get a buffer back then the allocation will fly.
2368 if ((error = xfs_alloc_ag_vextent(args)))
2372 TRACE_ALLOC("loopfailed", args);
2374 * Didn't work, figure out the next iteration.
2376 if (args->agno == sagno &&
2377 type == XFS_ALLOCTYPE_START_BNO)
2378 args->type = XFS_ALLOCTYPE_THIS_AG;
2380 * For the first allocation, we can try any AG to get
2381 * space. However, if we already have allocated a
2382 * block, we don't want to try AGs whose number is below
2383 * sagno. Otherwise, we may end up with out-of-order
2384 * locking of AGF, which might cause deadlock.
2386 if (++(args->agno) == mp->m_sb.sb_agcount) {
2387 if (args->firstblock != NULLFSBLOCK)
2393 * Reached the starting a.g., must either be done
2394 * or switch to non-trylock mode.
2396 if (args->agno == sagno) {
2398 args->agbno = NULLAGBLOCK;
2399 TRACE_ALLOC("allfailed", args);
2406 if (type == XFS_ALLOCTYPE_START_BNO) {
2407 args->agbno = XFS_FSB_TO_AGBNO(mp,
2409 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2414 up_read(&mp->m_peraglock);
2415 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2416 if (args->agno == sagno)
2417 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2418 (mp->m_sb.sb_agcount * rotorstep);
2420 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2421 (mp->m_sb.sb_agcount * rotorstep);
2428 if (args->agbno == NULLAGBLOCK)
2429 args->fsbno = NULLFSBLOCK;
2431 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2433 ASSERT(args->len >= args->minlen);
2434 ASSERT(args->len <= args->maxlen);
2435 ASSERT(args->agbno % args->alignment == 0);
2436 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2442 up_read(&mp->m_peraglock);
2448 * Just break up the extent address and hand off to xfs_free_ag_extent
2449 * after fixing up the freelist.
2453 xfs_trans_t *tp, /* transaction pointer */
2454 xfs_fsblock_t bno, /* starting block number of extent */
2455 xfs_extlen_t len) /* length of extent */
2457 xfs_alloc_arg_t args;
2461 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2463 args.mp = tp->t_mountp;
2464 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2465 ASSERT(args.agno < args.mp->m_sb.sb_agcount);
2466 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2467 down_read(&args.mp->m_peraglock);
2468 args.pag = &args.mp->m_perag[args.agno];
2469 if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING)))
2472 ASSERT(args.agbp != NULL);
2473 ASSERT((args.agbno + len) <=
2474 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length));
2476 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2478 up_read(&args.mp->m_peraglock);
2484 * AG Busy list management
2485 * The busy list contains block ranges that have been freed but whose
2486 * transactions have not yet hit disk. If any block listed in a busy
2487 * list is reused, the transaction that freed it must be forced to disk
2488 * before continuing to use the block.
2490 * xfs_alloc_mark_busy - add to the per-ag busy list
2491 * xfs_alloc_clear_busy - remove an item from the per-ag busy list
2494 xfs_alloc_mark_busy(xfs_trans_t *tp,
2495 xfs_agnumber_t agno,
2500 xfs_perag_busy_t *bsy;
2504 spin_lock(&mp->m_perag[agno].pagb_lock);
2506 /* search pagb_list for an open slot */
2507 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2508 n < XFS_PAGB_NUM_SLOTS;
2510 if (bsy->busy_tp == NULL) {
2515 if (n < XFS_PAGB_NUM_SLOTS) {
2516 bsy = &mp->m_perag[agno].pagb_list[n];
2517 mp->m_perag[agno].pagb_count++;
2518 TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp);
2519 bsy->busy_start = bno;
2520 bsy->busy_length = len;
2522 xfs_trans_add_busy(tp, agno, n);
2524 TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp);
2526 * The busy list is full! Since it is now not possible to
2527 * track the free block, make this a synchronous transaction
2528 * to insure that the block is not reused before this
2529 * transaction commits.
2531 xfs_trans_set_sync(tp);
2534 spin_unlock(&mp->m_perag[agno].pagb_lock);
2538 xfs_alloc_clear_busy(xfs_trans_t *tp,
2539 xfs_agnumber_t agno,
2543 xfs_perag_busy_t *list;
2547 spin_lock(&mp->m_perag[agno].pagb_lock);
2548 list = mp->m_perag[agno].pagb_list;
2550 ASSERT(idx < XFS_PAGB_NUM_SLOTS);
2551 if (list[idx].busy_tp == tp) {
2552 TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp);
2553 list[idx].busy_tp = NULL;
2554 mp->m_perag[agno].pagb_count--;
2556 TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp);
2559 spin_unlock(&mp->m_perag[agno].pagb_lock);
2564 * If we find the extent in the busy list, force the log out to get the
2565 * extent out of the busy list so the caller can use it straight away.
2568 xfs_alloc_search_busy(xfs_trans_t *tp,
2569 xfs_agnumber_t agno,
2574 xfs_perag_busy_t *bsy;
2575 xfs_agblock_t uend, bend;
2581 spin_lock(&mp->m_perag[agno].pagb_lock);
2582 cnt = mp->m_perag[agno].pagb_count;
2584 uend = bno + len - 1;
2586 /* search pagb_list for this slot, skipping open slots */
2587 for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) {
2590 * (start1,length1) within (start2, length2)
2592 if (bsy->busy_tp != NULL) {
2593 bend = bsy->busy_start + bsy->busy_length - 1;
2594 if ((bno > bend) || (uend < bsy->busy_start)) {
2597 TRACE_BUSYSEARCH("xfs_alloc_search_busy",
2598 "found1", agno, bno, len, tp);
2605 * If a block was found, force the log through the LSN of the
2606 * transaction that freed the block
2609 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, tp);
2610 lsn = bsy->busy_tp->t_commit_lsn;
2611 spin_unlock(&mp->m_perag[agno].pagb_lock);
2612 xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC);
2614 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, tp);
2615 spin_unlock(&mp->m_perag[agno].pagb_lock);