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_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_shared.h"
23 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_alloc.h"
32 #include "xfs_extent_busy.h"
33 #include "xfs_error.h"
34 #include "xfs_cksum.h"
35 #include "xfs_trace.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
40 struct workqueue_struct *xfs_alloc_wq;
42 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
44 #define XFSA_FIXUP_BNO_OK 1
45 #define XFSA_FIXUP_CNT_OK 2
47 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
48 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
49 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
50 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
51 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
54 * Lookup the record equal to [bno, len] in the btree given by cur.
56 STATIC int /* error */
58 struct xfs_btree_cur *cur, /* btree cursor */
59 xfs_agblock_t bno, /* starting block of extent */
60 xfs_extlen_t len, /* length of extent */
61 int *stat) /* success/failure */
63 cur->bc_rec.a.ar_startblock = bno;
64 cur->bc_rec.a.ar_blockcount = len;
65 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
69 * Lookup the first record greater than or equal to [bno, len]
70 * in the btree given by cur.
74 struct xfs_btree_cur *cur, /* btree cursor */
75 xfs_agblock_t bno, /* starting block of extent */
76 xfs_extlen_t len, /* length of extent */
77 int *stat) /* success/failure */
79 cur->bc_rec.a.ar_startblock = bno;
80 cur->bc_rec.a.ar_blockcount = len;
81 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
85 * Lookup the first record less than or equal to [bno, len]
86 * in the btree given by cur.
90 struct xfs_btree_cur *cur, /* btree cursor */
91 xfs_agblock_t bno, /* starting block of extent */
92 xfs_extlen_t len, /* length of extent */
93 int *stat) /* success/failure */
95 cur->bc_rec.a.ar_startblock = bno;
96 cur->bc_rec.a.ar_blockcount = len;
97 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
101 * Update the record referred to by cur to the value given
103 * This either works (return 0) or gets an EFSCORRUPTED error.
105 STATIC int /* error */
107 struct xfs_btree_cur *cur, /* btree cursor */
108 xfs_agblock_t bno, /* starting block of extent */
109 xfs_extlen_t len) /* length of extent */
111 union xfs_btree_rec rec;
113 rec.alloc.ar_startblock = cpu_to_be32(bno);
114 rec.alloc.ar_blockcount = cpu_to_be32(len);
115 return xfs_btree_update(cur, &rec);
119 * Get the data from the pointed-to record.
123 struct xfs_btree_cur *cur, /* btree cursor */
124 xfs_agblock_t *bno, /* output: starting block of extent */
125 xfs_extlen_t *len, /* output: length of extent */
126 int *stat) /* output: success/failure */
128 union xfs_btree_rec *rec;
131 error = xfs_btree_get_rec(cur, &rec, stat);
132 if (!error && *stat == 1) {
133 *bno = be32_to_cpu(rec->alloc.ar_startblock);
134 *len = be32_to_cpu(rec->alloc.ar_blockcount);
140 * Compute aligned version of the found extent.
141 * Takes alignment and min length into account.
144 xfs_alloc_compute_aligned(
145 xfs_alloc_arg_t *args, /* allocation argument structure */
146 xfs_agblock_t foundbno, /* starting block in found extent */
147 xfs_extlen_t foundlen, /* length in found extent */
148 xfs_agblock_t *resbno, /* result block number */
149 xfs_extlen_t *reslen) /* result length */
154 /* Trim busy sections out of found extent */
155 xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len);
157 if (args->alignment > 1 && len >= args->minlen) {
158 xfs_agblock_t aligned_bno = roundup(bno, args->alignment);
159 xfs_extlen_t diff = aligned_bno - bno;
161 *resbno = aligned_bno;
162 *reslen = diff >= len ? 0 : len - diff;
170 * Compute best start block and diff for "near" allocations.
171 * freelen >= wantlen already checked by caller.
173 STATIC xfs_extlen_t /* difference value (absolute) */
174 xfs_alloc_compute_diff(
175 xfs_agblock_t wantbno, /* target starting block */
176 xfs_extlen_t wantlen, /* target length */
177 xfs_extlen_t alignment, /* target alignment */
178 char userdata, /* are we allocating data? */
179 xfs_agblock_t freebno, /* freespace's starting block */
180 xfs_extlen_t freelen, /* freespace's length */
181 xfs_agblock_t *newbnop) /* result: best start block from free */
183 xfs_agblock_t freeend; /* end of freespace extent */
184 xfs_agblock_t newbno1; /* return block number */
185 xfs_agblock_t newbno2; /* other new block number */
186 xfs_extlen_t newlen1=0; /* length with newbno1 */
187 xfs_extlen_t newlen2=0; /* length with newbno2 */
188 xfs_agblock_t wantend; /* end of target extent */
190 ASSERT(freelen >= wantlen);
191 freeend = freebno + freelen;
192 wantend = wantbno + wantlen;
194 * We want to allocate from the start of a free extent if it is past
195 * the desired block or if we are allocating user data and the free
196 * extent is before desired block. The second case is there to allow
197 * for contiguous allocation from the remaining free space if the file
198 * grows in the short term.
200 if (freebno >= wantbno || (userdata && freeend < wantend)) {
201 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
202 newbno1 = NULLAGBLOCK;
203 } else if (freeend >= wantend && alignment > 1) {
204 newbno1 = roundup(wantbno, alignment);
205 newbno2 = newbno1 - alignment;
206 if (newbno1 >= freeend)
207 newbno1 = NULLAGBLOCK;
209 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
210 if (newbno2 < freebno)
211 newbno2 = NULLAGBLOCK;
213 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
214 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
215 if (newlen1 < newlen2 ||
216 (newlen1 == newlen2 &&
217 XFS_ABSDIFF(newbno1, wantbno) >
218 XFS_ABSDIFF(newbno2, wantbno)))
220 } else if (newbno2 != NULLAGBLOCK)
222 } else if (freeend >= wantend) {
224 } else if (alignment > 1) {
225 newbno1 = roundup(freeend - wantlen, alignment);
226 if (newbno1 > freeend - wantlen &&
227 newbno1 - alignment >= freebno)
228 newbno1 -= alignment;
229 else if (newbno1 >= freeend)
230 newbno1 = NULLAGBLOCK;
232 newbno1 = freeend - wantlen;
234 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
238 * Fix up the length, based on mod and prod.
239 * len should be k * prod + mod for some k.
240 * If len is too small it is returned unchanged.
241 * If len hits maxlen it is left alone.
245 xfs_alloc_arg_t *args) /* allocation argument structure */
250 ASSERT(args->mod < args->prod);
252 ASSERT(rlen >= args->minlen);
253 ASSERT(rlen <= args->maxlen);
254 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
255 (args->mod == 0 && rlen < args->prod))
257 k = rlen % args->prod;
261 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
264 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
268 ASSERT(rlen >= args->minlen);
269 ASSERT(rlen <= args->maxlen);
274 * Fix up length if there is too little space left in the a.g.
275 * Return 1 if ok, 0 if too little, should give up.
278 xfs_alloc_fix_minleft(
279 xfs_alloc_arg_t *args) /* allocation argument structure */
281 xfs_agf_t *agf; /* a.g. freelist header */
282 int diff; /* free space difference */
284 if (args->minleft == 0)
286 agf = XFS_BUF_TO_AGF(args->agbp);
287 diff = be32_to_cpu(agf->agf_freeblks)
288 - args->len - args->minleft;
291 args->len += diff; /* shrink the allocated space */
292 if (args->len >= args->minlen)
294 args->agbno = NULLAGBLOCK;
299 * Update the two btrees, logically removing from freespace the extent
300 * starting at rbno, rlen blocks. The extent is contained within the
301 * actual (current) free extent fbno for flen blocks.
302 * Flags are passed in indicating whether the cursors are set to the
305 STATIC int /* error code */
306 xfs_alloc_fixup_trees(
307 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
308 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
309 xfs_agblock_t fbno, /* starting block of free extent */
310 xfs_extlen_t flen, /* length of free extent */
311 xfs_agblock_t rbno, /* starting block of returned extent */
312 xfs_extlen_t rlen, /* length of returned extent */
313 int flags) /* flags, XFSA_FIXUP_... */
315 int error; /* error code */
316 int i; /* operation results */
317 xfs_agblock_t nfbno1; /* first new free startblock */
318 xfs_agblock_t nfbno2; /* second new free startblock */
319 xfs_extlen_t nflen1=0; /* first new free length */
320 xfs_extlen_t nflen2=0; /* second new free length */
323 * Look up the record in the by-size tree if necessary.
325 if (flags & XFSA_FIXUP_CNT_OK) {
327 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
329 XFS_WANT_CORRUPTED_RETURN(
330 i == 1 && nfbno1 == fbno && nflen1 == flen);
333 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
335 XFS_WANT_CORRUPTED_RETURN(i == 1);
338 * Look up the record in the by-block tree if necessary.
340 if (flags & XFSA_FIXUP_BNO_OK) {
342 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
344 XFS_WANT_CORRUPTED_RETURN(
345 i == 1 && nfbno1 == fbno && nflen1 == flen);
348 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
350 XFS_WANT_CORRUPTED_RETURN(i == 1);
354 if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
355 struct xfs_btree_block *bnoblock;
356 struct xfs_btree_block *cntblock;
358 bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
359 cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
361 XFS_WANT_CORRUPTED_RETURN(
362 bnoblock->bb_numrecs == cntblock->bb_numrecs);
367 * Deal with all four cases: the allocated record is contained
368 * within the freespace record, so we can have new freespace
369 * at either (or both) end, or no freespace remaining.
371 if (rbno == fbno && rlen == flen)
372 nfbno1 = nfbno2 = NULLAGBLOCK;
373 else if (rbno == fbno) {
374 nfbno1 = rbno + rlen;
375 nflen1 = flen - rlen;
376 nfbno2 = NULLAGBLOCK;
377 } else if (rbno + rlen == fbno + flen) {
379 nflen1 = flen - rlen;
380 nfbno2 = NULLAGBLOCK;
383 nflen1 = rbno - fbno;
384 nfbno2 = rbno + rlen;
385 nflen2 = (fbno + flen) - nfbno2;
388 * Delete the entry from the by-size btree.
390 if ((error = xfs_btree_delete(cnt_cur, &i)))
392 XFS_WANT_CORRUPTED_RETURN(i == 1);
394 * Add new by-size btree entry(s).
396 if (nfbno1 != NULLAGBLOCK) {
397 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
399 XFS_WANT_CORRUPTED_RETURN(i == 0);
400 if ((error = xfs_btree_insert(cnt_cur, &i)))
402 XFS_WANT_CORRUPTED_RETURN(i == 1);
404 if (nfbno2 != NULLAGBLOCK) {
405 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
407 XFS_WANT_CORRUPTED_RETURN(i == 0);
408 if ((error = xfs_btree_insert(cnt_cur, &i)))
410 XFS_WANT_CORRUPTED_RETURN(i == 1);
413 * Fix up the by-block btree entry(s).
415 if (nfbno1 == NULLAGBLOCK) {
417 * No remaining freespace, just delete the by-block tree entry.
419 if ((error = xfs_btree_delete(bno_cur, &i)))
421 XFS_WANT_CORRUPTED_RETURN(i == 1);
424 * Update the by-block entry to start later|be shorter.
426 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
429 if (nfbno2 != NULLAGBLOCK) {
431 * 2 resulting free entries, need to add one.
433 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
435 XFS_WANT_CORRUPTED_RETURN(i == 0);
436 if ((error = xfs_btree_insert(bno_cur, &i)))
438 XFS_WANT_CORRUPTED_RETURN(i == 1);
447 struct xfs_mount *mp = bp->b_target->bt_mount;
448 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
451 if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_uuid))
453 if (be32_to_cpu(agfl->agfl_magicnum) != XFS_AGFL_MAGIC)
456 * during growfs operations, the perag is not fully initialised,
457 * so we can't use it for any useful checking. growfs ensures we can't
458 * use it by using uncached buffers that don't have the perag attached
459 * so we can detect and avoid this problem.
461 if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno)
464 for (i = 0; i < XFS_AGFL_SIZE(mp); i++) {
465 if (be32_to_cpu(agfl->agfl_bno[i]) != NULLAGBLOCK &&
466 be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks)
473 xfs_agfl_read_verify(
476 struct xfs_mount *mp = bp->b_target->bt_mount;
479 * There is no verification of non-crc AGFLs because mkfs does not
480 * initialise the AGFL to zero or NULL. Hence the only valid part of the
481 * AGFL is what the AGF says is active. We can't get to the AGF, so we
482 * can't verify just those entries are valid.
484 if (!xfs_sb_version_hascrc(&mp->m_sb))
487 if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF))
488 xfs_buf_ioerror(bp, EFSBADCRC);
489 else if (!xfs_agfl_verify(bp))
490 xfs_buf_ioerror(bp, EFSCORRUPTED);
493 xfs_verifier_error(bp);
497 xfs_agfl_write_verify(
500 struct xfs_mount *mp = bp->b_target->bt_mount;
501 struct xfs_buf_log_item *bip = bp->b_fspriv;
503 /* no verification of non-crc AGFLs */
504 if (!xfs_sb_version_hascrc(&mp->m_sb))
507 if (!xfs_agfl_verify(bp)) {
508 xfs_buf_ioerror(bp, EFSCORRUPTED);
509 xfs_verifier_error(bp);
514 XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn);
516 xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF);
519 const struct xfs_buf_ops xfs_agfl_buf_ops = {
520 .verify_read = xfs_agfl_read_verify,
521 .verify_write = xfs_agfl_write_verify,
525 * Read in the allocation group free block array.
527 STATIC int /* error */
529 xfs_mount_t *mp, /* mount point structure */
530 xfs_trans_t *tp, /* transaction pointer */
531 xfs_agnumber_t agno, /* allocation group number */
532 xfs_buf_t **bpp) /* buffer for the ag free block array */
534 xfs_buf_t *bp; /* return value */
537 ASSERT(agno != NULLAGNUMBER);
538 error = xfs_trans_read_buf(
539 mp, tp, mp->m_ddev_targp,
540 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
541 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops);
544 xfs_buf_set_ref(bp, XFS_AGFL_REF);
550 xfs_alloc_update_counters(
551 struct xfs_trans *tp,
552 struct xfs_perag *pag,
553 struct xfs_buf *agbp,
556 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
558 pag->pagf_freeblks += len;
559 be32_add_cpu(&agf->agf_freeblks, len);
561 xfs_trans_agblocks_delta(tp, len);
562 if (unlikely(be32_to_cpu(agf->agf_freeblks) >
563 be32_to_cpu(agf->agf_length)))
566 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
571 * Allocation group level functions.
575 * Allocate a variable extent in the allocation group agno.
576 * Type and bno are used to determine where in the allocation group the
578 * Extent's length (returned in *len) will be between minlen and maxlen,
579 * and of the form k * prod + mod unless there's nothing that large.
580 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
582 STATIC int /* error */
583 xfs_alloc_ag_vextent(
584 xfs_alloc_arg_t *args) /* argument structure for allocation */
588 ASSERT(args->minlen > 0);
589 ASSERT(args->maxlen > 0);
590 ASSERT(args->minlen <= args->maxlen);
591 ASSERT(args->mod < args->prod);
592 ASSERT(args->alignment > 0);
594 * Branch to correct routine based on the type.
597 switch (args->type) {
598 case XFS_ALLOCTYPE_THIS_AG:
599 error = xfs_alloc_ag_vextent_size(args);
601 case XFS_ALLOCTYPE_NEAR_BNO:
602 error = xfs_alloc_ag_vextent_near(args);
604 case XFS_ALLOCTYPE_THIS_BNO:
605 error = xfs_alloc_ag_vextent_exact(args);
612 if (error || args->agbno == NULLAGBLOCK)
615 ASSERT(args->len >= args->minlen);
616 ASSERT(args->len <= args->maxlen);
617 ASSERT(!args->wasfromfl || !args->isfl);
618 ASSERT(args->agbno % args->alignment == 0);
620 if (!args->wasfromfl) {
621 error = xfs_alloc_update_counters(args->tp, args->pag,
623 -((long)(args->len)));
627 ASSERT(!xfs_extent_busy_search(args->mp, args->agno,
628 args->agbno, args->len));
632 xfs_trans_mod_sb(args->tp, args->wasdel ?
633 XFS_TRANS_SB_RES_FDBLOCKS :
634 XFS_TRANS_SB_FDBLOCKS,
635 -((long)(args->len)));
638 XFS_STATS_INC(xs_allocx);
639 XFS_STATS_ADD(xs_allocb, args->len);
644 * Allocate a variable extent at exactly agno/bno.
645 * Extent's length (returned in *len) will be between minlen and maxlen,
646 * and of the form k * prod + mod unless there's nothing that large.
647 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
649 STATIC int /* error */
650 xfs_alloc_ag_vextent_exact(
651 xfs_alloc_arg_t *args) /* allocation argument structure */
653 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
654 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
656 xfs_agblock_t fbno; /* start block of found extent */
657 xfs_extlen_t flen; /* length of found extent */
658 xfs_agblock_t tbno; /* start block of trimmed extent */
659 xfs_extlen_t tlen; /* length of trimmed extent */
660 xfs_agblock_t tend; /* end block of trimmed extent */
661 int i; /* success/failure of operation */
663 ASSERT(args->alignment == 1);
666 * Allocate/initialize a cursor for the by-number freespace btree.
668 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
669 args->agno, XFS_BTNUM_BNO);
672 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
673 * Look for the closest free block <= bno, it must contain bno
674 * if any free block does.
676 error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
683 * Grab the freespace record.
685 error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
688 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
689 ASSERT(fbno <= args->agbno);
692 * Check for overlapping busy extents.
694 xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen);
697 * Give up if the start of the extent is busy, or the freespace isn't
698 * long enough for the minimum request.
700 if (tbno > args->agbno)
702 if (tlen < args->minlen)
705 if (tend < args->agbno + args->minlen)
709 * End of extent will be smaller of the freespace end and the
710 * maximal requested end.
712 * Fix the length according to mod and prod if given.
714 args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
716 xfs_alloc_fix_len(args);
717 if (!xfs_alloc_fix_minleft(args))
720 ASSERT(args->agbno + args->len <= tend);
723 * We are allocating agbno for args->len
724 * Allocate/initialize a cursor for the by-size btree.
726 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
727 args->agno, XFS_BTNUM_CNT);
728 ASSERT(args->agbno + args->len <=
729 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
730 error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
731 args->len, XFSA_FIXUP_BNO_OK);
733 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
737 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
738 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
741 trace_xfs_alloc_exact_done(args);
745 /* Didn't find it, return null. */
746 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
747 args->agbno = NULLAGBLOCK;
748 trace_xfs_alloc_exact_notfound(args);
752 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
753 trace_xfs_alloc_exact_error(args);
758 * Search the btree in a given direction via the search cursor and compare
759 * the records found against the good extent we've already found.
762 xfs_alloc_find_best_extent(
763 struct xfs_alloc_arg *args, /* allocation argument structure */
764 struct xfs_btree_cur **gcur, /* good cursor */
765 struct xfs_btree_cur **scur, /* searching cursor */
766 xfs_agblock_t gdiff, /* difference for search comparison */
767 xfs_agblock_t *sbno, /* extent found by search */
768 xfs_extlen_t *slen, /* extent length */
769 xfs_agblock_t *sbnoa, /* aligned extent found by search */
770 xfs_extlen_t *slena, /* aligned extent length */
771 int dir) /* 0 = search right, 1 = search left */
778 /* The good extent is perfect, no need to search. */
783 * Look until we find a better one, run out of space or run off the end.
786 error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
789 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
790 xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena);
793 * The good extent is closer than this one.
796 if (*sbnoa >= args->agbno + gdiff)
799 if (*sbnoa <= args->agbno - gdiff)
804 * Same distance, compare length and pick the best.
806 if (*slena >= args->minlen) {
807 args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
808 xfs_alloc_fix_len(args);
810 sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
812 args->userdata, *sbnoa,
816 * Choose closer size and invalidate other cursor.
824 error = xfs_btree_increment(*scur, 0, &i);
826 error = xfs_btree_decrement(*scur, 0, &i);
832 xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
837 xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
842 /* caller invalidates cursors */
847 * Allocate a variable extent near bno in the allocation group agno.
848 * Extent's length (returned in len) will be between minlen and maxlen,
849 * and of the form k * prod + mod unless there's nothing that large.
850 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
852 STATIC int /* error */
853 xfs_alloc_ag_vextent_near(
854 xfs_alloc_arg_t *args) /* allocation argument structure */
856 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
857 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
858 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
859 xfs_agblock_t gtbno; /* start bno of right side entry */
860 xfs_agblock_t gtbnoa; /* aligned ... */
861 xfs_extlen_t gtdiff; /* difference to right side entry */
862 xfs_extlen_t gtlen; /* length of right side entry */
863 xfs_extlen_t gtlena; /* aligned ... */
864 xfs_agblock_t gtnew; /* useful start bno of right side */
865 int error; /* error code */
866 int i; /* result code, temporary */
867 int j; /* result code, temporary */
868 xfs_agblock_t ltbno; /* start bno of left side entry */
869 xfs_agblock_t ltbnoa; /* aligned ... */
870 xfs_extlen_t ltdiff; /* difference to left side entry */
871 xfs_extlen_t ltlen; /* length of left side entry */
872 xfs_extlen_t ltlena; /* aligned ... */
873 xfs_agblock_t ltnew; /* useful start bno of left side */
874 xfs_extlen_t rlen; /* length of returned extent */
878 * Randomly don't execute the first algorithm.
880 int dofirst; /* set to do first algorithm */
882 dofirst = prandom_u32() & 1;
893 * Get a cursor for the by-size btree.
895 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
896 args->agno, XFS_BTNUM_CNT);
899 * See if there are any free extents as big as maxlen.
901 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
904 * If none, then pick up the last entry in the tree unless the
908 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
911 if (i == 0 || ltlen == 0) {
912 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
913 trace_xfs_alloc_near_noentry(args);
922 * If the requested extent is large wrt the freespaces available
923 * in this a.g., then the cursor will be pointing to a btree entry
924 * near the right edge of the tree. If it's in the last btree leaf
925 * block, then we just examine all the entries in that block
926 * that are big enough, and pick the best one.
927 * This is written as a while loop so we can break out of it,
928 * but we never loop back to the top.
930 while (xfs_btree_islastblock(cnt_cur, 0)) {
934 xfs_agblock_t bnew=0;
941 * Start from the entry that lookup found, sequence through
942 * all larger free blocks. If we're actually pointing at a
943 * record smaller than maxlen, go to the start of this block,
944 * and skip all those smaller than minlen.
946 if (ltlen || args->alignment > 1) {
947 cnt_cur->bc_ptrs[0] = 1;
949 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
952 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
953 if (ltlen >= args->minlen)
955 if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
958 ASSERT(ltlen >= args->minlen);
962 i = cnt_cur->bc_ptrs[0];
963 for (j = 1, blen = 0, bdiff = 0;
964 !error && j && (blen < args->maxlen || bdiff > 0);
965 error = xfs_btree_increment(cnt_cur, 0, &j)) {
967 * For each entry, decide if it's better than
968 * the previous best entry.
970 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
972 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
973 xfs_alloc_compute_aligned(args, ltbno, ltlen,
975 if (ltlena < args->minlen)
977 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
978 xfs_alloc_fix_len(args);
979 ASSERT(args->len >= args->minlen);
980 if (args->len < blen)
982 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
983 args->alignment, args->userdata, ltbnoa,
985 if (ltnew != NULLAGBLOCK &&
986 (args->len > blen || ltdiff < bdiff)) {
990 besti = cnt_cur->bc_ptrs[0];
994 * It didn't work. We COULD be in a case where
995 * there's a good record somewhere, so try again.
1000 * Point at the best entry, and retrieve it again.
1002 cnt_cur->bc_ptrs[0] = besti;
1003 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
1005 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1006 ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1008 if (!xfs_alloc_fix_minleft(args)) {
1009 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1010 trace_xfs_alloc_near_nominleft(args);
1015 * We are allocating starting at bnew for blen blocks.
1018 ASSERT(bnew >= ltbno);
1019 ASSERT(bnew + blen <= ltbno + ltlen);
1021 * Set up a cursor for the by-bno tree.
1023 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
1024 args->agbp, args->agno, XFS_BTNUM_BNO);
1026 * Fix up the btree entries.
1028 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
1029 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
1031 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1032 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1034 trace_xfs_alloc_near_first(args);
1039 * Search in the by-bno tree to the left and to the right
1040 * simultaneously, until in each case we find a space big enough,
1041 * or run into the edge of the tree. When we run into the edge,
1042 * we deallocate that cursor.
1043 * If both searches succeed, we compare the two spaces and pick
1045 * With alignment, it's possible for both to fail; the upper
1046 * level algorithm that picks allocation groups for allocations
1047 * is not supposed to do this.
1050 * Allocate and initialize the cursor for the leftward search.
1052 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1053 args->agno, XFS_BTNUM_BNO);
1055 * Lookup <= bno to find the leftward search's starting point.
1057 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
1061 * Didn't find anything; use this cursor for the rightward
1064 bno_cur_gt = bno_cur_lt;
1068 * Found something. Duplicate the cursor for the rightward search.
1070 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
1073 * Increment the cursor, so we will point at the entry just right
1074 * of the leftward entry if any, or to the leftmost entry.
1076 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1080 * It failed, there are no rightward entries.
1082 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
1086 * Loop going left with the leftward cursor, right with the
1087 * rightward cursor, until either both directions give up or
1088 * we find an entry at least as big as minlen.
1092 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1094 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1095 xfs_alloc_compute_aligned(args, ltbno, ltlen,
1097 if (ltlena >= args->minlen)
1099 if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1102 xfs_btree_del_cursor(bno_cur_lt,
1108 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1110 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1111 xfs_alloc_compute_aligned(args, gtbno, gtlen,
1113 if (gtlena >= args->minlen)
1115 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1118 xfs_btree_del_cursor(bno_cur_gt,
1123 } while (bno_cur_lt || bno_cur_gt);
1126 * Got both cursors still active, need to find better entry.
1128 if (bno_cur_lt && bno_cur_gt) {
1129 if (ltlena >= args->minlen) {
1131 * Left side is good, look for a right side entry.
1133 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1134 xfs_alloc_fix_len(args);
1135 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1136 args->alignment, args->userdata, ltbnoa,
1139 error = xfs_alloc_find_best_extent(args,
1140 &bno_cur_lt, &bno_cur_gt,
1141 ltdiff, >bno, >len,
1143 0 /* search right */);
1145 ASSERT(gtlena >= args->minlen);
1148 * Right side is good, look for a left side entry.
1150 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1151 xfs_alloc_fix_len(args);
1152 gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1153 args->alignment, args->userdata, gtbnoa,
1156 error = xfs_alloc_find_best_extent(args,
1157 &bno_cur_gt, &bno_cur_lt,
1158 gtdiff, <bno, <len,
1160 1 /* search left */);
1168 * If we couldn't get anything, give up.
1170 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1171 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1174 trace_xfs_alloc_near_busy(args);
1175 xfs_log_force(args->mp, XFS_LOG_SYNC);
1178 trace_xfs_alloc_size_neither(args);
1179 args->agbno = NULLAGBLOCK;
1184 * At this point we have selected a freespace entry, either to the
1185 * left or to the right. If it's on the right, copy all the
1186 * useful variables to the "left" set so we only have one
1187 * copy of this code.
1190 bno_cur_lt = bno_cur_gt;
1201 * Fix up the length and compute the useful address.
1203 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1204 xfs_alloc_fix_len(args);
1205 if (!xfs_alloc_fix_minleft(args)) {
1206 trace_xfs_alloc_near_nominleft(args);
1207 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1208 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1212 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1213 args->userdata, ltbnoa, ltlena, <new);
1214 ASSERT(ltnew >= ltbno);
1215 ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1216 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1217 args->agbno = ltnew;
1219 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1220 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1224 trace_xfs_alloc_near_greater(args);
1226 trace_xfs_alloc_near_lesser(args);
1228 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1229 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1233 trace_xfs_alloc_near_error(args);
1234 if (cnt_cur != NULL)
1235 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1236 if (bno_cur_lt != NULL)
1237 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1238 if (bno_cur_gt != NULL)
1239 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1244 * Allocate a variable extent anywhere in the allocation group agno.
1245 * Extent's length (returned in len) will be between minlen and maxlen,
1246 * and of the form k * prod + mod unless there's nothing that large.
1247 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1249 STATIC int /* error */
1250 xfs_alloc_ag_vextent_size(
1251 xfs_alloc_arg_t *args) /* allocation argument structure */
1253 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1254 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1255 int error; /* error result */
1256 xfs_agblock_t fbno; /* start of found freespace */
1257 xfs_extlen_t flen; /* length of found freespace */
1258 int i; /* temp status variable */
1259 xfs_agblock_t rbno; /* returned block number */
1260 xfs_extlen_t rlen; /* length of returned extent */
1265 * Allocate and initialize a cursor for the by-size btree.
1267 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1268 args->agno, XFS_BTNUM_CNT);
1272 * Look for an entry >= maxlen+alignment-1 blocks.
1274 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1275 args->maxlen + args->alignment - 1, &i)))
1279 * If none or we have busy extents that we cannot allocate from, then
1280 * we have to settle for a smaller extent. In the case that there are
1281 * no large extents, this will return the last entry in the tree unless
1282 * the tree is empty. In the case that there are only busy large
1283 * extents, this will return the largest small extent unless there
1284 * are no smaller extents available.
1286 if (!i || forced > 1) {
1287 error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1291 if (i == 0 || flen == 0) {
1292 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1293 trace_xfs_alloc_size_noentry(args);
1297 xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen);
1300 * Search for a non-busy extent that is large enough.
1301 * If we are at low space, don't check, or if we fall of
1302 * the end of the btree, turn off the busy check and
1306 error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1309 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1311 xfs_alloc_compute_aligned(args, fbno, flen,
1314 if (rlen >= args->maxlen)
1317 error = xfs_btree_increment(cnt_cur, 0, &i);
1322 * Our only valid extents must have been busy.
1323 * Make it unbusy by forcing the log out and
1324 * retrying. If we've been here before, forcing
1325 * the log isn't making the extents available,
1326 * which means they have probably been freed in
1327 * this transaction. In that case, we have to
1328 * give up on them and we'll attempt a minlen
1329 * allocation the next time around.
1331 xfs_btree_del_cursor(cnt_cur,
1333 trace_xfs_alloc_size_busy(args);
1335 xfs_log_force(args->mp, XFS_LOG_SYNC);
1342 * In the first case above, we got the last entry in the
1343 * by-size btree. Now we check to see if the space hits maxlen
1344 * once aligned; if not, we search left for something better.
1345 * This can't happen in the second case above.
1347 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1348 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1349 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1350 if (rlen < args->maxlen) {
1351 xfs_agblock_t bestfbno;
1352 xfs_extlen_t bestflen;
1353 xfs_agblock_t bestrbno;
1354 xfs_extlen_t bestrlen;
1361 if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1365 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1368 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1369 if (flen < bestrlen)
1371 xfs_alloc_compute_aligned(args, fbno, flen,
1373 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1374 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1375 (rlen <= flen && rbno + rlen <= fbno + flen),
1377 if (rlen > bestrlen) {
1382 if (rlen == args->maxlen)
1386 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1389 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1395 args->wasfromfl = 0;
1397 * Fix up the length.
1400 if (rlen < args->minlen) {
1402 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1403 trace_xfs_alloc_size_busy(args);
1404 xfs_log_force(args->mp, XFS_LOG_SYNC);
1409 xfs_alloc_fix_len(args);
1411 if (!xfs_alloc_fix_minleft(args))
1414 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1416 * Allocate and initialize a cursor for the by-block tree.
1418 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1419 args->agno, XFS_BTNUM_BNO);
1420 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1421 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1423 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1424 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1425 cnt_cur = bno_cur = NULL;
1428 XFS_WANT_CORRUPTED_GOTO(
1429 args->agbno + args->len <=
1430 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1432 trace_xfs_alloc_size_done(args);
1436 trace_xfs_alloc_size_error(args);
1438 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1440 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1444 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1445 trace_xfs_alloc_size_nominleft(args);
1446 args->agbno = NULLAGBLOCK;
1451 * Deal with the case where only small freespaces remain.
1452 * Either return the contents of the last freespace record,
1453 * or allocate space from the freelist if there is nothing in the tree.
1455 STATIC int /* error */
1456 xfs_alloc_ag_vextent_small(
1457 xfs_alloc_arg_t *args, /* allocation argument structure */
1458 xfs_btree_cur_t *ccur, /* by-size cursor */
1459 xfs_agblock_t *fbnop, /* result block number */
1460 xfs_extlen_t *flenp, /* result length */
1461 int *stat) /* status: 0-freelist, 1-normal/none */
1468 if ((error = xfs_btree_decrement(ccur, 0, &i)))
1471 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1473 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1476 * Nothing in the btree, try the freelist. Make sure
1477 * to respect minleft even when pulling from the
1480 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1481 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1483 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1486 if (fbno != NULLAGBLOCK) {
1487 xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1,
1490 if (args->userdata) {
1493 bp = xfs_btree_get_bufs(args->mp, args->tp,
1494 args->agno, fbno, 0);
1495 xfs_trans_binval(args->tp, bp);
1499 XFS_WANT_CORRUPTED_GOTO(
1500 args->agbno + args->len <=
1501 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1503 args->wasfromfl = 1;
1504 trace_xfs_alloc_small_freelist(args);
1509 * Nothing in the freelist.
1515 * Can't allocate from the freelist for some reason.
1522 * Can't do the allocation, give up.
1524 if (flen < args->minlen) {
1525 args->agbno = NULLAGBLOCK;
1526 trace_xfs_alloc_small_notenough(args);
1532 trace_xfs_alloc_small_done(args);
1536 trace_xfs_alloc_small_error(args);
1541 * Free the extent starting at agno/bno for length.
1543 STATIC int /* error */
1545 xfs_trans_t *tp, /* transaction pointer */
1546 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1547 xfs_agnumber_t agno, /* allocation group number */
1548 xfs_agblock_t bno, /* starting block number */
1549 xfs_extlen_t len, /* length of extent */
1550 int isfl) /* set if is freelist blocks - no sb acctg */
1552 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1553 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1554 int error; /* error return value */
1555 xfs_agblock_t gtbno; /* start of right neighbor block */
1556 xfs_extlen_t gtlen; /* length of right neighbor block */
1557 int haveleft; /* have a left neighbor block */
1558 int haveright; /* have a right neighbor block */
1559 int i; /* temp, result code */
1560 xfs_agblock_t ltbno; /* start of left neighbor block */
1561 xfs_extlen_t ltlen; /* length of left neighbor block */
1562 xfs_mount_t *mp; /* mount point struct for filesystem */
1563 xfs_agblock_t nbno; /* new starting block of freespace */
1564 xfs_extlen_t nlen; /* new length of freespace */
1565 xfs_perag_t *pag; /* per allocation group data */
1569 * Allocate and initialize a cursor for the by-block btree.
1571 bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1574 * Look for a neighboring block on the left (lower block numbers)
1575 * that is contiguous with this space.
1577 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1581 * There is a block to our left.
1583 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1585 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1587 * It's not contiguous, though.
1589 if (ltbno + ltlen < bno)
1593 * If this failure happens the request to free this
1594 * space was invalid, it's (partly) already free.
1597 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1601 * Look for a neighboring block on the right (higher block numbers)
1602 * that is contiguous with this space.
1604 if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1608 * There is a block to our right.
1610 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1612 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1614 * It's not contiguous, though.
1616 if (bno + len < gtbno)
1620 * If this failure happens the request to free this
1621 * space was invalid, it's (partly) already free.
1624 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1628 * Now allocate and initialize a cursor for the by-size tree.
1630 cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1632 * Have both left and right contiguous neighbors.
1633 * Merge all three into a single free block.
1635 if (haveleft && haveright) {
1637 * Delete the old by-size entry on the left.
1639 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1641 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1642 if ((error = xfs_btree_delete(cnt_cur, &i)))
1644 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1646 * Delete the old by-size entry on the right.
1648 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1650 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1651 if ((error = xfs_btree_delete(cnt_cur, &i)))
1653 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1655 * Delete the old by-block entry for the right block.
1657 if ((error = xfs_btree_delete(bno_cur, &i)))
1659 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1661 * Move the by-block cursor back to the left neighbor.
1663 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1665 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1668 * Check that this is the right record: delete didn't
1669 * mangle the cursor.
1672 xfs_agblock_t xxbno;
1675 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1678 XFS_WANT_CORRUPTED_GOTO(
1679 i == 1 && xxbno == ltbno && xxlen == ltlen,
1684 * Update remaining by-block entry to the new, joined block.
1687 nlen = len + ltlen + gtlen;
1688 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1692 * Have only a left contiguous neighbor.
1693 * Merge it together with the new freespace.
1695 else if (haveleft) {
1697 * Delete the old by-size entry on the left.
1699 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1701 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1702 if ((error = xfs_btree_delete(cnt_cur, &i)))
1704 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1706 * Back up the by-block cursor to the left neighbor, and
1707 * update its length.
1709 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1711 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1714 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1718 * Have only a right contiguous neighbor.
1719 * Merge it together with the new freespace.
1721 else if (haveright) {
1723 * Delete the old by-size entry on the right.
1725 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1727 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1728 if ((error = xfs_btree_delete(cnt_cur, &i)))
1730 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1732 * Update the starting block and length of the right
1733 * neighbor in the by-block tree.
1737 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1741 * No contiguous neighbors.
1742 * Insert the new freespace into the by-block tree.
1747 if ((error = xfs_btree_insert(bno_cur, &i)))
1749 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1751 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1754 * In all cases we need to insert the new freespace in the by-size tree.
1756 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1758 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1759 if ((error = xfs_btree_insert(cnt_cur, &i)))
1761 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1762 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1766 * Update the freespace totals in the ag and superblock.
1768 pag = xfs_perag_get(mp, agno);
1769 error = xfs_alloc_update_counters(tp, pag, agbp, len);
1775 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1776 XFS_STATS_INC(xs_freex);
1777 XFS_STATS_ADD(xs_freeb, len);
1779 trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright);
1784 trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1);
1786 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1788 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1793 * Visible (exported) allocation/free functions.
1794 * Some of these are used just by xfs_alloc_btree.c and this file.
1798 * Compute and fill in value of m_ag_maxlevels.
1801 xfs_alloc_compute_maxlevels(
1802 xfs_mount_t *mp) /* file system mount structure */
1810 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1811 minleafrecs = mp->m_alloc_mnr[0];
1812 minnoderecs = mp->m_alloc_mnr[1];
1813 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1814 for (level = 1; maxblocks > 1; level++)
1815 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1816 mp->m_ag_maxlevels = level;
1820 * Find the length of the longest extent in an AG.
1823 xfs_alloc_longest_free_extent(
1824 struct xfs_mount *mp,
1825 struct xfs_perag *pag)
1827 xfs_extlen_t need, delta = 0;
1829 need = XFS_MIN_FREELIST_PAG(pag, mp);
1830 if (need > pag->pagf_flcount)
1831 delta = need - pag->pagf_flcount;
1833 if (pag->pagf_longest > delta)
1834 return pag->pagf_longest - delta;
1835 return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
1839 * Decide whether to use this allocation group for this allocation.
1840 * If so, fix up the btree freelist's size.
1842 STATIC int /* error */
1843 xfs_alloc_fix_freelist(
1844 xfs_alloc_arg_t *args, /* allocation argument structure */
1845 int flags) /* XFS_ALLOC_FLAG_... */
1847 xfs_buf_t *agbp; /* agf buffer pointer */
1848 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1849 xfs_buf_t *agflbp;/* agfl buffer pointer */
1850 xfs_agblock_t bno; /* freelist block */
1851 xfs_extlen_t delta; /* new blocks needed in freelist */
1852 int error; /* error result code */
1853 xfs_extlen_t longest;/* longest extent in allocation group */
1854 xfs_mount_t *mp; /* file system mount point structure */
1855 xfs_extlen_t need; /* total blocks needed in freelist */
1856 xfs_perag_t *pag; /* per-ag information structure */
1857 xfs_alloc_arg_t targs; /* local allocation arguments */
1858 xfs_trans_t *tp; /* transaction pointer */
1864 if (!pag->pagf_init) {
1865 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1868 if (!pag->pagf_init) {
1869 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1870 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1878 * If this is a metadata preferred pag and we are user data
1879 * then try somewhere else if we are not being asked to
1880 * try harder at this point
1882 if (pag->pagf_metadata && args->userdata &&
1883 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1884 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1889 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1891 * If it looks like there isn't a long enough extent, or enough
1892 * total blocks, reject it.
1894 need = XFS_MIN_FREELIST_PAG(pag, mp);
1895 longest = xfs_alloc_longest_free_extent(mp, pag);
1896 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1898 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1899 need - args->total) < (int)args->minleft)) {
1901 xfs_trans_brelse(tp, agbp);
1908 * Get the a.g. freespace buffer.
1909 * Can fail if we're not blocking on locks, and it's held.
1912 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1916 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1917 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1923 * Figure out how many blocks we should have in the freelist.
1925 agf = XFS_BUF_TO_AGF(agbp);
1926 need = XFS_MIN_FREELIST(agf, mp);
1928 * If there isn't enough total or single-extent, reject it.
1930 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1931 delta = need > be32_to_cpu(agf->agf_flcount) ?
1932 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1933 longest = be32_to_cpu(agf->agf_longest);
1934 longest = (longest > delta) ? (longest - delta) :
1935 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1936 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1938 ((int)(be32_to_cpu(agf->agf_freeblks) +
1939 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1940 (int)args->minleft)) {
1941 xfs_trans_brelse(tp, agbp);
1947 * Make the freelist shorter if it's too long.
1949 while (be32_to_cpu(agf->agf_flcount) > need) {
1952 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1955 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1957 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1958 xfs_trans_binval(tp, bp);
1961 * Initialize the args structure.
1963 memset(&targs, 0, sizeof(targs));
1967 targs.agno = args->agno;
1968 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1969 targs.type = XFS_ALLOCTYPE_THIS_AG;
1971 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1974 * Make the freelist longer if it's too short.
1976 while (be32_to_cpu(agf->agf_flcount) < need) {
1978 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1980 * Allocate as many blocks as possible at once.
1982 if ((error = xfs_alloc_ag_vextent(&targs))) {
1983 xfs_trans_brelse(tp, agflbp);
1987 * Stop if we run out. Won't happen if callers are obeying
1988 * the restrictions correctly. Can happen for free calls
1989 * on a completely full ag.
1991 if (targs.agbno == NULLAGBLOCK) {
1992 if (flags & XFS_ALLOC_FLAG_FREEING)
1994 xfs_trans_brelse(tp, agflbp);
1999 * Put each allocated block on the list.
2001 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2002 error = xfs_alloc_put_freelist(tp, agbp,
2008 xfs_trans_brelse(tp, agflbp);
2014 * Get a block from the freelist.
2015 * Returns with the buffer for the block gotten.
2018 xfs_alloc_get_freelist(
2019 xfs_trans_t *tp, /* transaction pointer */
2020 xfs_buf_t *agbp, /* buffer containing the agf structure */
2021 xfs_agblock_t *bnop, /* block address retrieved from freelist */
2022 int btreeblk) /* destination is a AGF btree */
2024 xfs_agf_t *agf; /* a.g. freespace structure */
2025 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
2026 xfs_agblock_t bno; /* block number returned */
2030 xfs_mount_t *mp = tp->t_mountp;
2031 xfs_perag_t *pag; /* per allocation group data */
2034 * Freelist is empty, give up.
2036 agf = XFS_BUF_TO_AGF(agbp);
2037 if (!agf->agf_flcount) {
2038 *bnop = NULLAGBLOCK;
2042 * Read the array of free blocks.
2044 error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno),
2051 * Get the block number and update the data structures.
2053 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2054 bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2055 be32_add_cpu(&agf->agf_flfirst, 1);
2056 xfs_trans_brelse(tp, agflbp);
2057 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
2058 agf->agf_flfirst = 0;
2060 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2061 be32_add_cpu(&agf->agf_flcount, -1);
2062 xfs_trans_agflist_delta(tp, -1);
2063 pag->pagf_flcount--;
2066 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2068 be32_add_cpu(&agf->agf_btreeblks, 1);
2069 pag->pagf_btreeblks++;
2070 logflags |= XFS_AGF_BTREEBLKS;
2073 xfs_alloc_log_agf(tp, agbp, logflags);
2080 * Log the given fields from the agf structure.
2084 xfs_trans_t *tp, /* transaction pointer */
2085 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2086 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2088 int first; /* first byte offset */
2089 int last; /* last byte offset */
2090 static const short offsets[] = {
2091 offsetof(xfs_agf_t, agf_magicnum),
2092 offsetof(xfs_agf_t, agf_versionnum),
2093 offsetof(xfs_agf_t, agf_seqno),
2094 offsetof(xfs_agf_t, agf_length),
2095 offsetof(xfs_agf_t, agf_roots[0]),
2096 offsetof(xfs_agf_t, agf_levels[0]),
2097 offsetof(xfs_agf_t, agf_flfirst),
2098 offsetof(xfs_agf_t, agf_fllast),
2099 offsetof(xfs_agf_t, agf_flcount),
2100 offsetof(xfs_agf_t, agf_freeblks),
2101 offsetof(xfs_agf_t, agf_longest),
2102 offsetof(xfs_agf_t, agf_btreeblks),
2103 offsetof(xfs_agf_t, agf_uuid),
2107 trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2109 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2111 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2112 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2116 * Interface for inode allocation to force the pag data to be initialized.
2119 xfs_alloc_pagf_init(
2120 xfs_mount_t *mp, /* file system mount structure */
2121 xfs_trans_t *tp, /* transaction pointer */
2122 xfs_agnumber_t agno, /* allocation group number */
2123 int flags) /* XFS_ALLOC_FLAGS_... */
2128 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2131 xfs_trans_brelse(tp, bp);
2136 * Put the block on the freelist for the allocation group.
2139 xfs_alloc_put_freelist(
2140 xfs_trans_t *tp, /* transaction pointer */
2141 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2142 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2143 xfs_agblock_t bno, /* block being freed */
2144 int btreeblk) /* block came from a AGF btree */
2146 xfs_agf_t *agf; /* a.g. freespace structure */
2147 __be32 *blockp;/* pointer to array entry */
2150 xfs_mount_t *mp; /* mount structure */
2151 xfs_perag_t *pag; /* per allocation group data */
2155 agf = XFS_BUF_TO_AGF(agbp);
2158 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2159 be32_to_cpu(agf->agf_seqno), &agflbp)))
2161 be32_add_cpu(&agf->agf_fllast, 1);
2162 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2163 agf->agf_fllast = 0;
2165 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2166 be32_add_cpu(&agf->agf_flcount, 1);
2167 xfs_trans_agflist_delta(tp, 1);
2168 pag->pagf_flcount++;
2170 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2172 be32_add_cpu(&agf->agf_btreeblks, -1);
2173 pag->pagf_btreeblks--;
2174 logflags |= XFS_AGF_BTREEBLKS;
2178 xfs_alloc_log_agf(tp, agbp, logflags);
2180 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2182 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2183 blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2184 *blockp = cpu_to_be32(bno);
2185 startoff = (char *)blockp - (char *)agflbp->b_addr;
2187 xfs_alloc_log_agf(tp, agbp, logflags);
2189 xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2190 xfs_trans_log_buf(tp, agflbp, startoff,
2191 startoff + sizeof(xfs_agblock_t) - 1);
2197 struct xfs_mount *mp,
2200 struct xfs_agf *agf = XFS_BUF_TO_AGF(bp);
2202 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2203 !uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_uuid))
2206 if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) &&
2207 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2208 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2209 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2210 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2211 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)))
2215 * during growfs operations, the perag is not fully initialised,
2216 * so we can't use it for any useful checking. growfs ensures we can't
2217 * use it by using uncached buffers that don't have the perag attached
2218 * so we can detect and avoid this problem.
2220 if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno)
2223 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
2224 be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
2232 xfs_agf_read_verify(
2235 struct xfs_mount *mp = bp->b_target->bt_mount;
2237 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2238 !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
2239 xfs_buf_ioerror(bp, EFSBADCRC);
2240 else if (XFS_TEST_ERROR(!xfs_agf_verify(mp, bp), mp,
2241 XFS_ERRTAG_ALLOC_READ_AGF,
2242 XFS_RANDOM_ALLOC_READ_AGF))
2243 xfs_buf_ioerror(bp, EFSCORRUPTED);
2246 xfs_verifier_error(bp);
2250 xfs_agf_write_verify(
2253 struct xfs_mount *mp = bp->b_target->bt_mount;
2254 struct xfs_buf_log_item *bip = bp->b_fspriv;
2256 if (!xfs_agf_verify(mp, bp)) {
2257 xfs_buf_ioerror(bp, EFSCORRUPTED);
2258 xfs_verifier_error(bp);
2262 if (!xfs_sb_version_hascrc(&mp->m_sb))
2266 XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2268 xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
2271 const struct xfs_buf_ops xfs_agf_buf_ops = {
2272 .verify_read = xfs_agf_read_verify,
2273 .verify_write = xfs_agf_write_verify,
2277 * Read in the allocation group header (free/alloc section).
2281 struct xfs_mount *mp, /* mount point structure */
2282 struct xfs_trans *tp, /* transaction pointer */
2283 xfs_agnumber_t agno, /* allocation group number */
2284 int flags, /* XFS_BUF_ */
2285 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2289 trace_xfs_read_agf(mp, agno);
2291 ASSERT(agno != NULLAGNUMBER);
2292 error = xfs_trans_read_buf(
2293 mp, tp, mp->m_ddev_targp,
2294 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2295 XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops);
2301 ASSERT(!(*bpp)->b_error);
2302 xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2307 * Read in the allocation group header (free/alloc section).
2311 struct xfs_mount *mp, /* mount point structure */
2312 struct xfs_trans *tp, /* transaction pointer */
2313 xfs_agnumber_t agno, /* allocation group number */
2314 int flags, /* XFS_ALLOC_FLAG_... */
2315 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2317 struct xfs_agf *agf; /* ag freelist header */
2318 struct xfs_perag *pag; /* per allocation group data */
2321 trace_xfs_alloc_read_agf(mp, agno);
2323 ASSERT(agno != NULLAGNUMBER);
2324 error = xfs_read_agf(mp, tp, agno,
2325 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2331 ASSERT(!(*bpp)->b_error);
2333 agf = XFS_BUF_TO_AGF(*bpp);
2334 pag = xfs_perag_get(mp, agno);
2335 if (!pag->pagf_init) {
2336 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2337 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2338 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2339 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2340 pag->pagf_levels[XFS_BTNUM_BNOi] =
2341 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2342 pag->pagf_levels[XFS_BTNUM_CNTi] =
2343 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2344 spin_lock_init(&pag->pagb_lock);
2345 pag->pagb_count = 0;
2346 pag->pagb_tree = RB_ROOT;
2350 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2351 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2352 ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2353 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2354 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2355 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2356 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2357 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2358 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2366 * Allocate an extent (variable-size).
2367 * Depending on the allocation type, we either look in a single allocation
2368 * group or loop over the allocation groups to find the result.
2372 xfs_alloc_arg_t *args) /* allocation argument structure */
2374 xfs_agblock_t agsize; /* allocation group size */
2376 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2377 xfs_extlen_t minleft;/* minimum left value, temp copy */
2378 xfs_mount_t *mp; /* mount structure pointer */
2379 xfs_agnumber_t sagno; /* starting allocation group number */
2380 xfs_alloctype_t type; /* input allocation type */
2383 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2386 type = args->otype = args->type;
2387 args->agbno = NULLAGBLOCK;
2389 * Just fix this up, for the case where the last a.g. is shorter
2390 * (or there's only one a.g.) and the caller couldn't easily figure
2391 * that out (xfs_bmap_alloc).
2393 agsize = mp->m_sb.sb_agblocks;
2394 if (args->maxlen > agsize)
2395 args->maxlen = agsize;
2396 if (args->alignment == 0)
2397 args->alignment = 1;
2398 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2399 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2400 ASSERT(args->minlen <= args->maxlen);
2401 ASSERT(args->minlen <= agsize);
2402 ASSERT(args->mod < args->prod);
2403 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2404 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2405 args->minlen > args->maxlen || args->minlen > agsize ||
2406 args->mod >= args->prod) {
2407 args->fsbno = NULLFSBLOCK;
2408 trace_xfs_alloc_vextent_badargs(args);
2411 minleft = args->minleft;
2414 case XFS_ALLOCTYPE_THIS_AG:
2415 case XFS_ALLOCTYPE_NEAR_BNO:
2416 case XFS_ALLOCTYPE_THIS_BNO:
2418 * These three force us into a single a.g.
2420 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2421 args->pag = xfs_perag_get(mp, args->agno);
2423 error = xfs_alloc_fix_freelist(args, 0);
2424 args->minleft = minleft;
2426 trace_xfs_alloc_vextent_nofix(args);
2430 trace_xfs_alloc_vextent_noagbp(args);
2433 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2434 if ((error = xfs_alloc_ag_vextent(args)))
2437 case XFS_ALLOCTYPE_START_BNO:
2439 * Try near allocation first, then anywhere-in-ag after
2440 * the first a.g. fails.
2442 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2443 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2444 args->fsbno = XFS_AGB_TO_FSB(mp,
2445 ((mp->m_agfrotor / rotorstep) %
2446 mp->m_sb.sb_agcount), 0);
2449 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2450 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2452 case XFS_ALLOCTYPE_ANY_AG:
2453 case XFS_ALLOCTYPE_START_AG:
2454 case XFS_ALLOCTYPE_FIRST_AG:
2456 * Rotate through the allocation groups looking for a winner.
2458 if (type == XFS_ALLOCTYPE_ANY_AG) {
2460 * Start with the last place we left off.
2462 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2463 mp->m_sb.sb_agcount;
2464 args->type = XFS_ALLOCTYPE_THIS_AG;
2465 flags = XFS_ALLOC_FLAG_TRYLOCK;
2466 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2468 * Start with allocation group given by bno.
2470 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2471 args->type = XFS_ALLOCTYPE_THIS_AG;
2475 if (type == XFS_ALLOCTYPE_START_AG)
2476 args->type = XFS_ALLOCTYPE_THIS_AG;
2478 * Start with the given allocation group.
2480 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2481 flags = XFS_ALLOC_FLAG_TRYLOCK;
2484 * Loop over allocation groups twice; first time with
2485 * trylock set, second time without.
2488 args->pag = xfs_perag_get(mp, args->agno);
2489 if (no_min) args->minleft = 0;
2490 error = xfs_alloc_fix_freelist(args, flags);
2491 args->minleft = minleft;
2493 trace_xfs_alloc_vextent_nofix(args);
2497 * If we get a buffer back then the allocation will fly.
2500 if ((error = xfs_alloc_ag_vextent(args)))
2505 trace_xfs_alloc_vextent_loopfailed(args);
2508 * Didn't work, figure out the next iteration.
2510 if (args->agno == sagno &&
2511 type == XFS_ALLOCTYPE_START_BNO)
2512 args->type = XFS_ALLOCTYPE_THIS_AG;
2514 * For the first allocation, we can try any AG to get
2515 * space. However, if we already have allocated a
2516 * block, we don't want to try AGs whose number is below
2517 * sagno. Otherwise, we may end up with out-of-order
2518 * locking of AGF, which might cause deadlock.
2520 if (++(args->agno) == mp->m_sb.sb_agcount) {
2521 if (args->firstblock != NULLFSBLOCK)
2527 * Reached the starting a.g., must either be done
2528 * or switch to non-trylock mode.
2530 if (args->agno == sagno) {
2532 args->agbno = NULLAGBLOCK;
2533 trace_xfs_alloc_vextent_allfailed(args);
2540 if (type == XFS_ALLOCTYPE_START_BNO) {
2541 args->agbno = XFS_FSB_TO_AGBNO(mp,
2543 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2547 xfs_perag_put(args->pag);
2549 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2550 if (args->agno == sagno)
2551 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2552 (mp->m_sb.sb_agcount * rotorstep);
2554 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2555 (mp->m_sb.sb_agcount * rotorstep);
2562 if (args->agbno == NULLAGBLOCK)
2563 args->fsbno = NULLFSBLOCK;
2565 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2567 ASSERT(args->len >= args->minlen);
2568 ASSERT(args->len <= args->maxlen);
2569 ASSERT(args->agbno % args->alignment == 0);
2570 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2574 xfs_perag_put(args->pag);
2577 xfs_perag_put(args->pag);
2583 * Just break up the extent address and hand off to xfs_free_ag_extent
2584 * after fixing up the freelist.
2588 xfs_trans_t *tp, /* transaction pointer */
2589 xfs_fsblock_t bno, /* starting block number of extent */
2590 xfs_extlen_t len) /* length of extent */
2592 xfs_alloc_arg_t args;
2596 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2598 args.mp = tp->t_mountp;
2601 * validate that the block number is legal - the enables us to detect
2602 * and handle a silent filesystem corruption rather than crashing.
2604 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2605 if (args.agno >= args.mp->m_sb.sb_agcount)
2606 return EFSCORRUPTED;
2608 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2609 if (args.agbno >= args.mp->m_sb.sb_agblocks)
2610 return EFSCORRUPTED;
2612 args.pag = xfs_perag_get(args.mp, args.agno);
2615 error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
2619 /* validate the extent size is legal now we have the agf locked */
2620 if (args.agbno + len >
2621 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) {
2622 error = EFSCORRUPTED;
2626 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2628 xfs_extent_busy_insert(tp, args.agno, args.agbno, len, 0);
2630 xfs_perag_put(args.pag);