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 ASSERT(!xfs_buf_geterror(bp));
545 xfs_buf_set_ref(bp, XFS_AGFL_REF);
551 xfs_alloc_update_counters(
552 struct xfs_trans *tp,
553 struct xfs_perag *pag,
554 struct xfs_buf *agbp,
557 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
559 pag->pagf_freeblks += len;
560 be32_add_cpu(&agf->agf_freeblks, len);
562 xfs_trans_agblocks_delta(tp, len);
563 if (unlikely(be32_to_cpu(agf->agf_freeblks) >
564 be32_to_cpu(agf->agf_length)))
567 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
572 * Allocation group level functions.
576 * Allocate a variable extent in the allocation group agno.
577 * Type and bno are used to determine where in the allocation group the
579 * Extent's length (returned in *len) will be between minlen and maxlen,
580 * and of the form k * prod + mod unless there's nothing that large.
581 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
583 STATIC int /* error */
584 xfs_alloc_ag_vextent(
585 xfs_alloc_arg_t *args) /* argument structure for allocation */
589 ASSERT(args->minlen > 0);
590 ASSERT(args->maxlen > 0);
591 ASSERT(args->minlen <= args->maxlen);
592 ASSERT(args->mod < args->prod);
593 ASSERT(args->alignment > 0);
595 * Branch to correct routine based on the type.
598 switch (args->type) {
599 case XFS_ALLOCTYPE_THIS_AG:
600 error = xfs_alloc_ag_vextent_size(args);
602 case XFS_ALLOCTYPE_NEAR_BNO:
603 error = xfs_alloc_ag_vextent_near(args);
605 case XFS_ALLOCTYPE_THIS_BNO:
606 error = xfs_alloc_ag_vextent_exact(args);
613 if (error || args->agbno == NULLAGBLOCK)
616 ASSERT(args->len >= args->minlen);
617 ASSERT(args->len <= args->maxlen);
618 ASSERT(!args->wasfromfl || !args->isfl);
619 ASSERT(args->agbno % args->alignment == 0);
621 if (!args->wasfromfl) {
622 error = xfs_alloc_update_counters(args->tp, args->pag,
624 -((long)(args->len)));
628 ASSERT(!xfs_extent_busy_search(args->mp, args->agno,
629 args->agbno, args->len));
633 xfs_trans_mod_sb(args->tp, args->wasdel ?
634 XFS_TRANS_SB_RES_FDBLOCKS :
635 XFS_TRANS_SB_FDBLOCKS,
636 -((long)(args->len)));
639 XFS_STATS_INC(xs_allocx);
640 XFS_STATS_ADD(xs_allocb, args->len);
645 * Allocate a variable extent at exactly agno/bno.
646 * Extent's length (returned in *len) will be between minlen and maxlen,
647 * and of the form k * prod + mod unless there's nothing that large.
648 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
650 STATIC int /* error */
651 xfs_alloc_ag_vextent_exact(
652 xfs_alloc_arg_t *args) /* allocation argument structure */
654 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
655 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
657 xfs_agblock_t fbno; /* start block of found extent */
658 xfs_extlen_t flen; /* length of found extent */
659 xfs_agblock_t tbno; /* start block of trimmed extent */
660 xfs_extlen_t tlen; /* length of trimmed extent */
661 xfs_agblock_t tend; /* end block of trimmed extent */
662 int i; /* success/failure of operation */
664 ASSERT(args->alignment == 1);
667 * Allocate/initialize a cursor for the by-number freespace btree.
669 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
670 args->agno, XFS_BTNUM_BNO);
673 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
674 * Look for the closest free block <= bno, it must contain bno
675 * if any free block does.
677 error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
684 * Grab the freespace record.
686 error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
689 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
690 ASSERT(fbno <= args->agbno);
693 * Check for overlapping busy extents.
695 xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen);
698 * Give up if the start of the extent is busy, or the freespace isn't
699 * long enough for the minimum request.
701 if (tbno > args->agbno)
703 if (tlen < args->minlen)
706 if (tend < args->agbno + args->minlen)
710 * End of extent will be smaller of the freespace end and the
711 * maximal requested end.
713 * Fix the length according to mod and prod if given.
715 args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
717 xfs_alloc_fix_len(args);
718 if (!xfs_alloc_fix_minleft(args))
721 ASSERT(args->agbno + args->len <= tend);
724 * We are allocating agbno for args->len
725 * Allocate/initialize a cursor for the by-size btree.
727 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
728 args->agno, XFS_BTNUM_CNT);
729 ASSERT(args->agbno + args->len <=
730 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
731 error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
732 args->len, XFSA_FIXUP_BNO_OK);
734 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
738 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
739 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
742 trace_xfs_alloc_exact_done(args);
746 /* Didn't find it, return null. */
747 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
748 args->agbno = NULLAGBLOCK;
749 trace_xfs_alloc_exact_notfound(args);
753 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
754 trace_xfs_alloc_exact_error(args);
759 * Search the btree in a given direction via the search cursor and compare
760 * the records found against the good extent we've already found.
763 xfs_alloc_find_best_extent(
764 struct xfs_alloc_arg *args, /* allocation argument structure */
765 struct xfs_btree_cur **gcur, /* good cursor */
766 struct xfs_btree_cur **scur, /* searching cursor */
767 xfs_agblock_t gdiff, /* difference for search comparison */
768 xfs_agblock_t *sbno, /* extent found by search */
769 xfs_extlen_t *slen, /* extent length */
770 xfs_agblock_t *sbnoa, /* aligned extent found by search */
771 xfs_extlen_t *slena, /* aligned extent length */
772 int dir) /* 0 = search right, 1 = search left */
779 /* The good extent is perfect, no need to search. */
784 * Look until we find a better one, run out of space or run off the end.
787 error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
790 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
791 xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena);
794 * The good extent is closer than this one.
797 if (*sbnoa >= args->agbno + gdiff)
800 if (*sbnoa <= args->agbno - gdiff)
805 * Same distance, compare length and pick the best.
807 if (*slena >= args->minlen) {
808 args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
809 xfs_alloc_fix_len(args);
811 sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
813 args->userdata, *sbnoa,
817 * Choose closer size and invalidate other cursor.
825 error = xfs_btree_increment(*scur, 0, &i);
827 error = xfs_btree_decrement(*scur, 0, &i);
833 xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
838 xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
843 /* caller invalidates cursors */
848 * Allocate a variable extent near bno in the allocation group agno.
849 * Extent's length (returned in len) will be between minlen and maxlen,
850 * and of the form k * prod + mod unless there's nothing that large.
851 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
853 STATIC int /* error */
854 xfs_alloc_ag_vextent_near(
855 xfs_alloc_arg_t *args) /* allocation argument structure */
857 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
858 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
859 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
860 xfs_agblock_t gtbno; /* start bno of right side entry */
861 xfs_agblock_t gtbnoa; /* aligned ... */
862 xfs_extlen_t gtdiff; /* difference to right side entry */
863 xfs_extlen_t gtlen; /* length of right side entry */
864 xfs_extlen_t gtlena; /* aligned ... */
865 xfs_agblock_t gtnew; /* useful start bno of right side */
866 int error; /* error code */
867 int i; /* result code, temporary */
868 int j; /* result code, temporary */
869 xfs_agblock_t ltbno; /* start bno of left side entry */
870 xfs_agblock_t ltbnoa; /* aligned ... */
871 xfs_extlen_t ltdiff; /* difference to left side entry */
872 xfs_extlen_t ltlen; /* length of left side entry */
873 xfs_extlen_t ltlena; /* aligned ... */
874 xfs_agblock_t ltnew; /* useful start bno of left side */
875 xfs_extlen_t rlen; /* length of returned extent */
879 * Randomly don't execute the first algorithm.
881 int dofirst; /* set to do first algorithm */
883 dofirst = prandom_u32() & 1;
894 * Get a cursor for the by-size btree.
896 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
897 args->agno, XFS_BTNUM_CNT);
900 * See if there are any free extents as big as maxlen.
902 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
905 * If none, then pick up the last entry in the tree unless the
909 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
912 if (i == 0 || ltlen == 0) {
913 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
914 trace_xfs_alloc_near_noentry(args);
923 * If the requested extent is large wrt the freespaces available
924 * in this a.g., then the cursor will be pointing to a btree entry
925 * near the right edge of the tree. If it's in the last btree leaf
926 * block, then we just examine all the entries in that block
927 * that are big enough, and pick the best one.
928 * This is written as a while loop so we can break out of it,
929 * but we never loop back to the top.
931 while (xfs_btree_islastblock(cnt_cur, 0)) {
935 xfs_agblock_t bnew=0;
942 * Start from the entry that lookup found, sequence through
943 * all larger free blocks. If we're actually pointing at a
944 * record smaller than maxlen, go to the start of this block,
945 * and skip all those smaller than minlen.
947 if (ltlen || args->alignment > 1) {
948 cnt_cur->bc_ptrs[0] = 1;
950 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
953 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
954 if (ltlen >= args->minlen)
956 if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
959 ASSERT(ltlen >= args->minlen);
963 i = cnt_cur->bc_ptrs[0];
964 for (j = 1, blen = 0, bdiff = 0;
965 !error && j && (blen < args->maxlen || bdiff > 0);
966 error = xfs_btree_increment(cnt_cur, 0, &j)) {
968 * For each entry, decide if it's better than
969 * the previous best entry.
971 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
973 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
974 xfs_alloc_compute_aligned(args, ltbno, ltlen,
976 if (ltlena < args->minlen)
978 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
979 xfs_alloc_fix_len(args);
980 ASSERT(args->len >= args->minlen);
981 if (args->len < blen)
983 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
984 args->alignment, args->userdata, ltbnoa,
986 if (ltnew != NULLAGBLOCK &&
987 (args->len > blen || ltdiff < bdiff)) {
991 besti = cnt_cur->bc_ptrs[0];
995 * It didn't work. We COULD be in a case where
996 * there's a good record somewhere, so try again.
1001 * Point at the best entry, and retrieve it again.
1003 cnt_cur->bc_ptrs[0] = besti;
1004 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
1006 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1007 ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1009 if (!xfs_alloc_fix_minleft(args)) {
1010 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1011 trace_xfs_alloc_near_nominleft(args);
1016 * We are allocating starting at bnew for blen blocks.
1019 ASSERT(bnew >= ltbno);
1020 ASSERT(bnew + blen <= ltbno + ltlen);
1022 * Set up a cursor for the by-bno tree.
1024 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
1025 args->agbp, args->agno, XFS_BTNUM_BNO);
1027 * Fix up the btree entries.
1029 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
1030 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
1032 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1033 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1035 trace_xfs_alloc_near_first(args);
1040 * Search in the by-bno tree to the left and to the right
1041 * simultaneously, until in each case we find a space big enough,
1042 * or run into the edge of the tree. When we run into the edge,
1043 * we deallocate that cursor.
1044 * If both searches succeed, we compare the two spaces and pick
1046 * With alignment, it's possible for both to fail; the upper
1047 * level algorithm that picks allocation groups for allocations
1048 * is not supposed to do this.
1051 * Allocate and initialize the cursor for the leftward search.
1053 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1054 args->agno, XFS_BTNUM_BNO);
1056 * Lookup <= bno to find the leftward search's starting point.
1058 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
1062 * Didn't find anything; use this cursor for the rightward
1065 bno_cur_gt = bno_cur_lt;
1069 * Found something. Duplicate the cursor for the rightward search.
1071 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
1074 * Increment the cursor, so we will point at the entry just right
1075 * of the leftward entry if any, or to the leftmost entry.
1077 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1081 * It failed, there are no rightward entries.
1083 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
1087 * Loop going left with the leftward cursor, right with the
1088 * rightward cursor, until either both directions give up or
1089 * we find an entry at least as big as minlen.
1093 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1095 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1096 xfs_alloc_compute_aligned(args, ltbno, ltlen,
1098 if (ltlena >= args->minlen)
1100 if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1103 xfs_btree_del_cursor(bno_cur_lt,
1109 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1111 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1112 xfs_alloc_compute_aligned(args, gtbno, gtlen,
1114 if (gtlena >= args->minlen)
1116 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1119 xfs_btree_del_cursor(bno_cur_gt,
1124 } while (bno_cur_lt || bno_cur_gt);
1127 * Got both cursors still active, need to find better entry.
1129 if (bno_cur_lt && bno_cur_gt) {
1130 if (ltlena >= args->minlen) {
1132 * Left side is good, look for a right side entry.
1134 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1135 xfs_alloc_fix_len(args);
1136 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1137 args->alignment, args->userdata, ltbnoa,
1140 error = xfs_alloc_find_best_extent(args,
1141 &bno_cur_lt, &bno_cur_gt,
1142 ltdiff, >bno, >len,
1144 0 /* search right */);
1146 ASSERT(gtlena >= args->minlen);
1149 * Right side is good, look for a left side entry.
1151 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1152 xfs_alloc_fix_len(args);
1153 gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1154 args->alignment, args->userdata, gtbnoa,
1157 error = xfs_alloc_find_best_extent(args,
1158 &bno_cur_gt, &bno_cur_lt,
1159 gtdiff, <bno, <len,
1161 1 /* search left */);
1169 * If we couldn't get anything, give up.
1171 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1172 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1175 trace_xfs_alloc_near_busy(args);
1176 xfs_log_force(args->mp, XFS_LOG_SYNC);
1179 trace_xfs_alloc_size_neither(args);
1180 args->agbno = NULLAGBLOCK;
1185 * At this point we have selected a freespace entry, either to the
1186 * left or to the right. If it's on the right, copy all the
1187 * useful variables to the "left" set so we only have one
1188 * copy of this code.
1191 bno_cur_lt = bno_cur_gt;
1202 * Fix up the length and compute the useful address.
1204 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1205 xfs_alloc_fix_len(args);
1206 if (!xfs_alloc_fix_minleft(args)) {
1207 trace_xfs_alloc_near_nominleft(args);
1208 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1209 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1213 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1214 args->userdata, ltbnoa, ltlena, <new);
1215 ASSERT(ltnew >= ltbno);
1216 ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1217 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1218 args->agbno = ltnew;
1220 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1221 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1225 trace_xfs_alloc_near_greater(args);
1227 trace_xfs_alloc_near_lesser(args);
1229 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1230 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1234 trace_xfs_alloc_near_error(args);
1235 if (cnt_cur != NULL)
1236 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1237 if (bno_cur_lt != NULL)
1238 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1239 if (bno_cur_gt != NULL)
1240 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1245 * Allocate a variable extent anywhere in the allocation group agno.
1246 * Extent's length (returned in len) will be between minlen and maxlen,
1247 * and of the form k * prod + mod unless there's nothing that large.
1248 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1250 STATIC int /* error */
1251 xfs_alloc_ag_vextent_size(
1252 xfs_alloc_arg_t *args) /* allocation argument structure */
1254 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1255 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1256 int error; /* error result */
1257 xfs_agblock_t fbno; /* start of found freespace */
1258 xfs_extlen_t flen; /* length of found freespace */
1259 int i; /* temp status variable */
1260 xfs_agblock_t rbno; /* returned block number */
1261 xfs_extlen_t rlen; /* length of returned extent */
1266 * Allocate and initialize a cursor for the by-size btree.
1268 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1269 args->agno, XFS_BTNUM_CNT);
1273 * Look for an entry >= maxlen+alignment-1 blocks.
1275 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1276 args->maxlen + args->alignment - 1, &i)))
1280 * If none or we have busy extents that we cannot allocate from, then
1281 * we have to settle for a smaller extent. In the case that there are
1282 * no large extents, this will return the last entry in the tree unless
1283 * the tree is empty. In the case that there are only busy large
1284 * extents, this will return the largest small extent unless there
1285 * are no smaller extents available.
1287 if (!i || forced > 1) {
1288 error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1292 if (i == 0 || flen == 0) {
1293 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1294 trace_xfs_alloc_size_noentry(args);
1298 xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen);
1301 * Search for a non-busy extent that is large enough.
1302 * If we are at low space, don't check, or if we fall of
1303 * the end of the btree, turn off the busy check and
1307 error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1310 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1312 xfs_alloc_compute_aligned(args, fbno, flen,
1315 if (rlen >= args->maxlen)
1318 error = xfs_btree_increment(cnt_cur, 0, &i);
1323 * Our only valid extents must have been busy.
1324 * Make it unbusy by forcing the log out and
1325 * retrying. If we've been here before, forcing
1326 * the log isn't making the extents available,
1327 * which means they have probably been freed in
1328 * this transaction. In that case, we have to
1329 * give up on them and we'll attempt a minlen
1330 * allocation the next time around.
1332 xfs_btree_del_cursor(cnt_cur,
1334 trace_xfs_alloc_size_busy(args);
1336 xfs_log_force(args->mp, XFS_LOG_SYNC);
1343 * In the first case above, we got the last entry in the
1344 * by-size btree. Now we check to see if the space hits maxlen
1345 * once aligned; if not, we search left for something better.
1346 * This can't happen in the second case above.
1348 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1349 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1350 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1351 if (rlen < args->maxlen) {
1352 xfs_agblock_t bestfbno;
1353 xfs_extlen_t bestflen;
1354 xfs_agblock_t bestrbno;
1355 xfs_extlen_t bestrlen;
1362 if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1366 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1369 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1370 if (flen < bestrlen)
1372 xfs_alloc_compute_aligned(args, fbno, flen,
1374 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1375 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1376 (rlen <= flen && rbno + rlen <= fbno + flen),
1378 if (rlen > bestrlen) {
1383 if (rlen == args->maxlen)
1387 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1390 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1396 args->wasfromfl = 0;
1398 * Fix up the length.
1401 if (rlen < args->minlen) {
1403 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1404 trace_xfs_alloc_size_busy(args);
1405 xfs_log_force(args->mp, XFS_LOG_SYNC);
1410 xfs_alloc_fix_len(args);
1412 if (!xfs_alloc_fix_minleft(args))
1415 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1417 * Allocate and initialize a cursor for the by-block tree.
1419 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1420 args->agno, XFS_BTNUM_BNO);
1421 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1422 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1424 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1425 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1426 cnt_cur = bno_cur = NULL;
1429 XFS_WANT_CORRUPTED_GOTO(
1430 args->agbno + args->len <=
1431 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1433 trace_xfs_alloc_size_done(args);
1437 trace_xfs_alloc_size_error(args);
1439 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1441 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1445 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1446 trace_xfs_alloc_size_nominleft(args);
1447 args->agbno = NULLAGBLOCK;
1452 * Deal with the case where only small freespaces remain.
1453 * Either return the contents of the last freespace record,
1454 * or allocate space from the freelist if there is nothing in the tree.
1456 STATIC int /* error */
1457 xfs_alloc_ag_vextent_small(
1458 xfs_alloc_arg_t *args, /* allocation argument structure */
1459 xfs_btree_cur_t *ccur, /* by-size cursor */
1460 xfs_agblock_t *fbnop, /* result block number */
1461 xfs_extlen_t *flenp, /* result length */
1462 int *stat) /* status: 0-freelist, 1-normal/none */
1469 if ((error = xfs_btree_decrement(ccur, 0, &i)))
1472 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1474 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1477 * Nothing in the btree, try the freelist. Make sure
1478 * to respect minleft even when pulling from the
1481 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1482 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1484 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1487 if (fbno != NULLAGBLOCK) {
1488 xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1,
1491 if (args->userdata) {
1494 bp = xfs_btree_get_bufs(args->mp, args->tp,
1495 args->agno, fbno, 0);
1496 xfs_trans_binval(args->tp, bp);
1500 XFS_WANT_CORRUPTED_GOTO(
1501 args->agbno + args->len <=
1502 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1504 args->wasfromfl = 1;
1505 trace_xfs_alloc_small_freelist(args);
1510 * Nothing in the freelist.
1516 * Can't allocate from the freelist for some reason.
1523 * Can't do the allocation, give up.
1525 if (flen < args->minlen) {
1526 args->agbno = NULLAGBLOCK;
1527 trace_xfs_alloc_small_notenough(args);
1533 trace_xfs_alloc_small_done(args);
1537 trace_xfs_alloc_small_error(args);
1542 * Free the extent starting at agno/bno for length.
1544 STATIC int /* error */
1546 xfs_trans_t *tp, /* transaction pointer */
1547 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1548 xfs_agnumber_t agno, /* allocation group number */
1549 xfs_agblock_t bno, /* starting block number */
1550 xfs_extlen_t len, /* length of extent */
1551 int isfl) /* set if is freelist blocks - no sb acctg */
1553 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1554 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1555 int error; /* error return value */
1556 xfs_agblock_t gtbno; /* start of right neighbor block */
1557 xfs_extlen_t gtlen; /* length of right neighbor block */
1558 int haveleft; /* have a left neighbor block */
1559 int haveright; /* have a right neighbor block */
1560 int i; /* temp, result code */
1561 xfs_agblock_t ltbno; /* start of left neighbor block */
1562 xfs_extlen_t ltlen; /* length of left neighbor block */
1563 xfs_mount_t *mp; /* mount point struct for filesystem */
1564 xfs_agblock_t nbno; /* new starting block of freespace */
1565 xfs_extlen_t nlen; /* new length of freespace */
1566 xfs_perag_t *pag; /* per allocation group data */
1570 * Allocate and initialize a cursor for the by-block btree.
1572 bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1575 * Look for a neighboring block on the left (lower block numbers)
1576 * that is contiguous with this space.
1578 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1582 * There is a block to our left.
1584 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1586 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1588 * It's not contiguous, though.
1590 if (ltbno + ltlen < bno)
1594 * If this failure happens the request to free this
1595 * space was invalid, it's (partly) already free.
1598 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1602 * Look for a neighboring block on the right (higher block numbers)
1603 * that is contiguous with this space.
1605 if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1609 * There is a block to our right.
1611 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1613 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1615 * It's not contiguous, though.
1617 if (bno + len < gtbno)
1621 * If this failure happens the request to free this
1622 * space was invalid, it's (partly) already free.
1625 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1629 * Now allocate and initialize a cursor for the by-size tree.
1631 cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1633 * Have both left and right contiguous neighbors.
1634 * Merge all three into a single free block.
1636 if (haveleft && haveright) {
1638 * Delete the old by-size entry on the left.
1640 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1642 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1643 if ((error = xfs_btree_delete(cnt_cur, &i)))
1645 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1647 * Delete the old by-size entry on the right.
1649 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1651 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1652 if ((error = xfs_btree_delete(cnt_cur, &i)))
1654 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1656 * Delete the old by-block entry for the right block.
1658 if ((error = xfs_btree_delete(bno_cur, &i)))
1660 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1662 * Move the by-block cursor back to the left neighbor.
1664 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1666 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1669 * Check that this is the right record: delete didn't
1670 * mangle the cursor.
1673 xfs_agblock_t xxbno;
1676 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1679 XFS_WANT_CORRUPTED_GOTO(
1680 i == 1 && xxbno == ltbno && xxlen == ltlen,
1685 * Update remaining by-block entry to the new, joined block.
1688 nlen = len + ltlen + gtlen;
1689 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1693 * Have only a left contiguous neighbor.
1694 * Merge it together with the new freespace.
1696 else if (haveleft) {
1698 * Delete the old by-size entry on the left.
1700 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1702 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1703 if ((error = xfs_btree_delete(cnt_cur, &i)))
1705 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1707 * Back up the by-block cursor to the left neighbor, and
1708 * update its length.
1710 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1712 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1715 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1719 * Have only a right contiguous neighbor.
1720 * Merge it together with the new freespace.
1722 else if (haveright) {
1724 * Delete the old by-size entry on the right.
1726 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1728 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1729 if ((error = xfs_btree_delete(cnt_cur, &i)))
1731 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1733 * Update the starting block and length of the right
1734 * neighbor in the by-block tree.
1738 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1742 * No contiguous neighbors.
1743 * Insert the new freespace into the by-block tree.
1748 if ((error = xfs_btree_insert(bno_cur, &i)))
1750 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1752 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1755 * In all cases we need to insert the new freespace in the by-size tree.
1757 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1759 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1760 if ((error = xfs_btree_insert(cnt_cur, &i)))
1762 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1763 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1767 * Update the freespace totals in the ag and superblock.
1769 pag = xfs_perag_get(mp, agno);
1770 error = xfs_alloc_update_counters(tp, pag, agbp, len);
1776 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1777 XFS_STATS_INC(xs_freex);
1778 XFS_STATS_ADD(xs_freeb, len);
1780 trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright);
1785 trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1);
1787 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1789 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1794 * Visible (exported) allocation/free functions.
1795 * Some of these are used just by xfs_alloc_btree.c and this file.
1799 * Compute and fill in value of m_ag_maxlevels.
1802 xfs_alloc_compute_maxlevels(
1803 xfs_mount_t *mp) /* file system mount structure */
1811 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1812 minleafrecs = mp->m_alloc_mnr[0];
1813 minnoderecs = mp->m_alloc_mnr[1];
1814 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1815 for (level = 1; maxblocks > 1; level++)
1816 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1817 mp->m_ag_maxlevels = level;
1821 * Find the length of the longest extent in an AG.
1824 xfs_alloc_longest_free_extent(
1825 struct xfs_mount *mp,
1826 struct xfs_perag *pag)
1828 xfs_extlen_t need, delta = 0;
1830 need = XFS_MIN_FREELIST_PAG(pag, mp);
1831 if (need > pag->pagf_flcount)
1832 delta = need - pag->pagf_flcount;
1834 if (pag->pagf_longest > delta)
1835 return pag->pagf_longest - delta;
1836 return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
1840 * Decide whether to use this allocation group for this allocation.
1841 * If so, fix up the btree freelist's size.
1843 STATIC int /* error */
1844 xfs_alloc_fix_freelist(
1845 xfs_alloc_arg_t *args, /* allocation argument structure */
1846 int flags) /* XFS_ALLOC_FLAG_... */
1848 xfs_buf_t *agbp; /* agf buffer pointer */
1849 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1850 xfs_buf_t *agflbp;/* agfl buffer pointer */
1851 xfs_agblock_t bno; /* freelist block */
1852 xfs_extlen_t delta; /* new blocks needed in freelist */
1853 int error; /* error result code */
1854 xfs_extlen_t longest;/* longest extent in allocation group */
1855 xfs_mount_t *mp; /* file system mount point structure */
1856 xfs_extlen_t need; /* total blocks needed in freelist */
1857 xfs_perag_t *pag; /* per-ag information structure */
1858 xfs_alloc_arg_t targs; /* local allocation arguments */
1859 xfs_trans_t *tp; /* transaction pointer */
1865 if (!pag->pagf_init) {
1866 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1869 if (!pag->pagf_init) {
1870 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1871 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1879 * If this is a metadata preferred pag and we are user data
1880 * then try somewhere else if we are not being asked to
1881 * try harder at this point
1883 if (pag->pagf_metadata && args->userdata &&
1884 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1885 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1890 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1892 * If it looks like there isn't a long enough extent, or enough
1893 * total blocks, reject it.
1895 need = XFS_MIN_FREELIST_PAG(pag, mp);
1896 longest = xfs_alloc_longest_free_extent(mp, pag);
1897 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1899 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1900 need - args->total) < (int)args->minleft)) {
1902 xfs_trans_brelse(tp, agbp);
1909 * Get the a.g. freespace buffer.
1910 * Can fail if we're not blocking on locks, and it's held.
1913 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1917 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1918 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1924 * Figure out how many blocks we should have in the freelist.
1926 agf = XFS_BUF_TO_AGF(agbp);
1927 need = XFS_MIN_FREELIST(agf, mp);
1929 * If there isn't enough total or single-extent, reject it.
1931 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1932 delta = need > be32_to_cpu(agf->agf_flcount) ?
1933 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1934 longest = be32_to_cpu(agf->agf_longest);
1935 longest = (longest > delta) ? (longest - delta) :
1936 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1937 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1939 ((int)(be32_to_cpu(agf->agf_freeblks) +
1940 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1941 (int)args->minleft)) {
1942 xfs_trans_brelse(tp, agbp);
1948 * Make the freelist shorter if it's too long.
1950 while (be32_to_cpu(agf->agf_flcount) > need) {
1953 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1956 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1958 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1959 xfs_trans_binval(tp, bp);
1962 * Initialize the args structure.
1964 memset(&targs, 0, sizeof(targs));
1968 targs.agno = args->agno;
1969 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1970 targs.type = XFS_ALLOCTYPE_THIS_AG;
1972 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1975 * Make the freelist longer if it's too short.
1977 while (be32_to_cpu(agf->agf_flcount) < need) {
1979 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1981 * Allocate as many blocks as possible at once.
1983 if ((error = xfs_alloc_ag_vextent(&targs))) {
1984 xfs_trans_brelse(tp, agflbp);
1988 * Stop if we run out. Won't happen if callers are obeying
1989 * the restrictions correctly. Can happen for free calls
1990 * on a completely full ag.
1992 if (targs.agbno == NULLAGBLOCK) {
1993 if (flags & XFS_ALLOC_FLAG_FREEING)
1995 xfs_trans_brelse(tp, agflbp);
2000 * Put each allocated block on the list.
2002 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2003 error = xfs_alloc_put_freelist(tp, agbp,
2009 xfs_trans_brelse(tp, agflbp);
2015 * Get a block from the freelist.
2016 * Returns with the buffer for the block gotten.
2019 xfs_alloc_get_freelist(
2020 xfs_trans_t *tp, /* transaction pointer */
2021 xfs_buf_t *agbp, /* buffer containing the agf structure */
2022 xfs_agblock_t *bnop, /* block address retrieved from freelist */
2023 int btreeblk) /* destination is a AGF btree */
2025 xfs_agf_t *agf; /* a.g. freespace structure */
2026 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
2027 xfs_agblock_t bno; /* block number returned */
2031 xfs_mount_t *mp = tp->t_mountp;
2032 xfs_perag_t *pag; /* per allocation group data */
2035 * Freelist is empty, give up.
2037 agf = XFS_BUF_TO_AGF(agbp);
2038 if (!agf->agf_flcount) {
2039 *bnop = NULLAGBLOCK;
2043 * Read the array of free blocks.
2045 error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno),
2052 * Get the block number and update the data structures.
2054 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2055 bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2056 be32_add_cpu(&agf->agf_flfirst, 1);
2057 xfs_trans_brelse(tp, agflbp);
2058 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
2059 agf->agf_flfirst = 0;
2061 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2062 be32_add_cpu(&agf->agf_flcount, -1);
2063 xfs_trans_agflist_delta(tp, -1);
2064 pag->pagf_flcount--;
2067 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2069 be32_add_cpu(&agf->agf_btreeblks, 1);
2070 pag->pagf_btreeblks++;
2071 logflags |= XFS_AGF_BTREEBLKS;
2074 xfs_alloc_log_agf(tp, agbp, logflags);
2081 * Log the given fields from the agf structure.
2085 xfs_trans_t *tp, /* transaction pointer */
2086 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2087 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2089 int first; /* first byte offset */
2090 int last; /* last byte offset */
2091 static const short offsets[] = {
2092 offsetof(xfs_agf_t, agf_magicnum),
2093 offsetof(xfs_agf_t, agf_versionnum),
2094 offsetof(xfs_agf_t, agf_seqno),
2095 offsetof(xfs_agf_t, agf_length),
2096 offsetof(xfs_agf_t, agf_roots[0]),
2097 offsetof(xfs_agf_t, agf_levels[0]),
2098 offsetof(xfs_agf_t, agf_flfirst),
2099 offsetof(xfs_agf_t, agf_fllast),
2100 offsetof(xfs_agf_t, agf_flcount),
2101 offsetof(xfs_agf_t, agf_freeblks),
2102 offsetof(xfs_agf_t, agf_longest),
2103 offsetof(xfs_agf_t, agf_btreeblks),
2104 offsetof(xfs_agf_t, agf_uuid),
2108 trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2110 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2112 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2113 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2117 * Interface for inode allocation to force the pag data to be initialized.
2120 xfs_alloc_pagf_init(
2121 xfs_mount_t *mp, /* file system mount structure */
2122 xfs_trans_t *tp, /* transaction pointer */
2123 xfs_agnumber_t agno, /* allocation group number */
2124 int flags) /* XFS_ALLOC_FLAGS_... */
2129 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2132 xfs_trans_brelse(tp, bp);
2137 * Put the block on the freelist for the allocation group.
2140 xfs_alloc_put_freelist(
2141 xfs_trans_t *tp, /* transaction pointer */
2142 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2143 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2144 xfs_agblock_t bno, /* block being freed */
2145 int btreeblk) /* block came from a AGF btree */
2147 xfs_agf_t *agf; /* a.g. freespace structure */
2148 __be32 *blockp;/* pointer to array entry */
2151 xfs_mount_t *mp; /* mount structure */
2152 xfs_perag_t *pag; /* per allocation group data */
2156 agf = XFS_BUF_TO_AGF(agbp);
2159 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2160 be32_to_cpu(agf->agf_seqno), &agflbp)))
2162 be32_add_cpu(&agf->agf_fllast, 1);
2163 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2164 agf->agf_fllast = 0;
2166 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2167 be32_add_cpu(&agf->agf_flcount, 1);
2168 xfs_trans_agflist_delta(tp, 1);
2169 pag->pagf_flcount++;
2171 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2173 be32_add_cpu(&agf->agf_btreeblks, -1);
2174 pag->pagf_btreeblks--;
2175 logflags |= XFS_AGF_BTREEBLKS;
2179 xfs_alloc_log_agf(tp, agbp, logflags);
2181 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2183 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2184 blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2185 *blockp = cpu_to_be32(bno);
2186 startoff = (char *)blockp - (char *)agflbp->b_addr;
2188 xfs_alloc_log_agf(tp, agbp, logflags);
2190 xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2191 xfs_trans_log_buf(tp, agflbp, startoff,
2192 startoff + sizeof(xfs_agblock_t) - 1);
2198 struct xfs_mount *mp,
2201 struct xfs_agf *agf = XFS_BUF_TO_AGF(bp);
2203 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2204 !uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_uuid))
2207 if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) &&
2208 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2209 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2210 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2211 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2212 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)))
2216 * during growfs operations, the perag is not fully initialised,
2217 * so we can't use it for any useful checking. growfs ensures we can't
2218 * use it by using uncached buffers that don't have the perag attached
2219 * so we can detect and avoid this problem.
2221 if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno)
2224 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
2225 be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
2233 xfs_agf_read_verify(
2236 struct xfs_mount *mp = bp->b_target->bt_mount;
2238 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2239 !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
2240 xfs_buf_ioerror(bp, EFSBADCRC);
2241 else if (XFS_TEST_ERROR(!xfs_agf_verify(mp, bp), mp,
2242 XFS_ERRTAG_ALLOC_READ_AGF,
2243 XFS_RANDOM_ALLOC_READ_AGF))
2244 xfs_buf_ioerror(bp, EFSCORRUPTED);
2247 xfs_verifier_error(bp);
2251 xfs_agf_write_verify(
2254 struct xfs_mount *mp = bp->b_target->bt_mount;
2255 struct xfs_buf_log_item *bip = bp->b_fspriv;
2257 if (!xfs_agf_verify(mp, bp)) {
2258 xfs_buf_ioerror(bp, EFSCORRUPTED);
2259 xfs_verifier_error(bp);
2263 if (!xfs_sb_version_hascrc(&mp->m_sb))
2267 XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2269 xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
2272 const struct xfs_buf_ops xfs_agf_buf_ops = {
2273 .verify_read = xfs_agf_read_verify,
2274 .verify_write = xfs_agf_write_verify,
2278 * Read in the allocation group header (free/alloc section).
2282 struct xfs_mount *mp, /* mount point structure */
2283 struct xfs_trans *tp, /* transaction pointer */
2284 xfs_agnumber_t agno, /* allocation group number */
2285 int flags, /* XFS_BUF_ */
2286 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2290 trace_xfs_read_agf(mp, agno);
2292 ASSERT(agno != NULLAGNUMBER);
2293 error = xfs_trans_read_buf(
2294 mp, tp, mp->m_ddev_targp,
2295 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2296 XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops);
2302 ASSERT(!(*bpp)->b_error);
2303 xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2308 * Read in the allocation group header (free/alloc section).
2312 struct xfs_mount *mp, /* mount point structure */
2313 struct xfs_trans *tp, /* transaction pointer */
2314 xfs_agnumber_t agno, /* allocation group number */
2315 int flags, /* XFS_ALLOC_FLAG_... */
2316 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2318 struct xfs_agf *agf; /* ag freelist header */
2319 struct xfs_perag *pag; /* per allocation group data */
2322 trace_xfs_alloc_read_agf(mp, agno);
2324 ASSERT(agno != NULLAGNUMBER);
2325 error = xfs_read_agf(mp, tp, agno,
2326 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2332 ASSERT(!(*bpp)->b_error);
2334 agf = XFS_BUF_TO_AGF(*bpp);
2335 pag = xfs_perag_get(mp, agno);
2336 if (!pag->pagf_init) {
2337 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2338 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2339 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2340 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2341 pag->pagf_levels[XFS_BTNUM_BNOi] =
2342 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2343 pag->pagf_levels[XFS_BTNUM_CNTi] =
2344 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2345 spin_lock_init(&pag->pagb_lock);
2346 pag->pagb_count = 0;
2347 pag->pagb_tree = RB_ROOT;
2351 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2352 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2353 ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2354 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2355 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2356 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2357 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2358 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2359 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2367 * Allocate an extent (variable-size).
2368 * Depending on the allocation type, we either look in a single allocation
2369 * group or loop over the allocation groups to find the result.
2373 xfs_alloc_arg_t *args) /* allocation argument structure */
2375 xfs_agblock_t agsize; /* allocation group size */
2377 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2378 xfs_extlen_t minleft;/* minimum left value, temp copy */
2379 xfs_mount_t *mp; /* mount structure pointer */
2380 xfs_agnumber_t sagno; /* starting allocation group number */
2381 xfs_alloctype_t type; /* input allocation type */
2384 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2387 type = args->otype = args->type;
2388 args->agbno = NULLAGBLOCK;
2390 * Just fix this up, for the case where the last a.g. is shorter
2391 * (or there's only one a.g.) and the caller couldn't easily figure
2392 * that out (xfs_bmap_alloc).
2394 agsize = mp->m_sb.sb_agblocks;
2395 if (args->maxlen > agsize)
2396 args->maxlen = agsize;
2397 if (args->alignment == 0)
2398 args->alignment = 1;
2399 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2400 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2401 ASSERT(args->minlen <= args->maxlen);
2402 ASSERT(args->minlen <= agsize);
2403 ASSERT(args->mod < args->prod);
2404 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2405 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2406 args->minlen > args->maxlen || args->minlen > agsize ||
2407 args->mod >= args->prod) {
2408 args->fsbno = NULLFSBLOCK;
2409 trace_xfs_alloc_vextent_badargs(args);
2412 minleft = args->minleft;
2415 case XFS_ALLOCTYPE_THIS_AG:
2416 case XFS_ALLOCTYPE_NEAR_BNO:
2417 case XFS_ALLOCTYPE_THIS_BNO:
2419 * These three force us into a single a.g.
2421 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2422 args->pag = xfs_perag_get(mp, args->agno);
2424 error = xfs_alloc_fix_freelist(args, 0);
2425 args->minleft = minleft;
2427 trace_xfs_alloc_vextent_nofix(args);
2431 trace_xfs_alloc_vextent_noagbp(args);
2434 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2435 if ((error = xfs_alloc_ag_vextent(args)))
2438 case XFS_ALLOCTYPE_START_BNO:
2440 * Try near allocation first, then anywhere-in-ag after
2441 * the first a.g. fails.
2443 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2444 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2445 args->fsbno = XFS_AGB_TO_FSB(mp,
2446 ((mp->m_agfrotor / rotorstep) %
2447 mp->m_sb.sb_agcount), 0);
2450 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2451 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2453 case XFS_ALLOCTYPE_ANY_AG:
2454 case XFS_ALLOCTYPE_START_AG:
2455 case XFS_ALLOCTYPE_FIRST_AG:
2457 * Rotate through the allocation groups looking for a winner.
2459 if (type == XFS_ALLOCTYPE_ANY_AG) {
2461 * Start with the last place we left off.
2463 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2464 mp->m_sb.sb_agcount;
2465 args->type = XFS_ALLOCTYPE_THIS_AG;
2466 flags = XFS_ALLOC_FLAG_TRYLOCK;
2467 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2469 * Start with allocation group given by bno.
2471 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2472 args->type = XFS_ALLOCTYPE_THIS_AG;
2476 if (type == XFS_ALLOCTYPE_START_AG)
2477 args->type = XFS_ALLOCTYPE_THIS_AG;
2479 * Start with the given allocation group.
2481 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2482 flags = XFS_ALLOC_FLAG_TRYLOCK;
2485 * Loop over allocation groups twice; first time with
2486 * trylock set, second time without.
2489 args->pag = xfs_perag_get(mp, args->agno);
2490 if (no_min) args->minleft = 0;
2491 error = xfs_alloc_fix_freelist(args, flags);
2492 args->minleft = minleft;
2494 trace_xfs_alloc_vextent_nofix(args);
2498 * If we get a buffer back then the allocation will fly.
2501 if ((error = xfs_alloc_ag_vextent(args)))
2506 trace_xfs_alloc_vextent_loopfailed(args);
2509 * Didn't work, figure out the next iteration.
2511 if (args->agno == sagno &&
2512 type == XFS_ALLOCTYPE_START_BNO)
2513 args->type = XFS_ALLOCTYPE_THIS_AG;
2515 * For the first allocation, we can try any AG to get
2516 * space. However, if we already have allocated a
2517 * block, we don't want to try AGs whose number is below
2518 * sagno. Otherwise, we may end up with out-of-order
2519 * locking of AGF, which might cause deadlock.
2521 if (++(args->agno) == mp->m_sb.sb_agcount) {
2522 if (args->firstblock != NULLFSBLOCK)
2528 * Reached the starting a.g., must either be done
2529 * or switch to non-trylock mode.
2531 if (args->agno == sagno) {
2533 args->agbno = NULLAGBLOCK;
2534 trace_xfs_alloc_vextent_allfailed(args);
2541 if (type == XFS_ALLOCTYPE_START_BNO) {
2542 args->agbno = XFS_FSB_TO_AGBNO(mp,
2544 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2548 xfs_perag_put(args->pag);
2550 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2551 if (args->agno == sagno)
2552 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2553 (mp->m_sb.sb_agcount * rotorstep);
2555 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2556 (mp->m_sb.sb_agcount * rotorstep);
2563 if (args->agbno == NULLAGBLOCK)
2564 args->fsbno = NULLFSBLOCK;
2566 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2568 ASSERT(args->len >= args->minlen);
2569 ASSERT(args->len <= args->maxlen);
2570 ASSERT(args->agbno % args->alignment == 0);
2571 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2575 xfs_perag_put(args->pag);
2578 xfs_perag_put(args->pag);
2584 * Just break up the extent address and hand off to xfs_free_ag_extent
2585 * after fixing up the freelist.
2589 xfs_trans_t *tp, /* transaction pointer */
2590 xfs_fsblock_t bno, /* starting block number of extent */
2591 xfs_extlen_t len) /* length of extent */
2593 xfs_alloc_arg_t args;
2597 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2599 args.mp = tp->t_mountp;
2602 * validate that the block number is legal - the enables us to detect
2603 * and handle a silent filesystem corruption rather than crashing.
2605 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2606 if (args.agno >= args.mp->m_sb.sb_agcount)
2607 return EFSCORRUPTED;
2609 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2610 if (args.agbno >= args.mp->m_sb.sb_agblocks)
2611 return EFSCORRUPTED;
2613 args.pag = xfs_perag_get(args.mp, args.agno);
2616 error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
2620 /* validate the extent size is legal now we have the agf locked */
2621 if (args.agbno + len >
2622 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) {
2623 error = EFSCORRUPTED;
2627 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2629 xfs_extent_busy_insert(tp, args.agno, args.agbno, len, 0);
2631 xfs_perag_put(args.pag);