1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_shared.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
19 #include "xfs_alloc_btree.h"
20 #include "xfs_alloc.h"
21 #include "xfs_extent_busy.h"
22 #include "xfs_errortag.h"
23 #include "xfs_error.h"
24 #include "xfs_cksum.h"
25 #include "xfs_trace.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
29 #include "xfs_ag_resv.h"
32 extern kmem_zone_t *xfs_bmap_free_item_zone;
34 struct workqueue_struct *xfs_alloc_wq;
36 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
38 #define XFSA_FIXUP_BNO_OK 1
39 #define XFSA_FIXUP_CNT_OK 2
41 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
42 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
43 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
44 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
45 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
48 * Size of the AGFL. For CRC-enabled filesystes we steal a couple of slots in
49 * the beginning of the block for a proper header with the location information
56 unsigned int size = mp->m_sb.sb_sectsize;
58 if (xfs_sb_version_hascrc(&mp->m_sb))
59 size -= sizeof(struct xfs_agfl);
61 return size / sizeof(xfs_agblock_t);
68 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
69 return XFS_RMAP_BLOCK(mp) + 1;
70 if (xfs_sb_version_hasfinobt(&mp->m_sb))
71 return XFS_FIBT_BLOCK(mp) + 1;
72 return XFS_IBT_BLOCK(mp) + 1;
79 if (xfs_sb_version_hasreflink(&mp->m_sb))
80 return xfs_refc_block(mp) + 1;
81 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
82 return XFS_RMAP_BLOCK(mp) + 1;
83 if (xfs_sb_version_hasfinobt(&mp->m_sb))
84 return XFS_FIBT_BLOCK(mp) + 1;
85 return XFS_IBT_BLOCK(mp) + 1;
89 * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of
90 * AGF buffer (PV 947395), we place constraints on the relationship among
91 * actual allocations for data blocks, freelist blocks, and potential file data
92 * bmap btree blocks. However, these restrictions may result in no actual space
93 * allocated for a delayed extent, for example, a data block in a certain AG is
94 * allocated but there is no additional block for the additional bmap btree
95 * block due to a split of the bmap btree of the file. The result of this may
96 * lead to an infinite loop when the file gets flushed to disk and all delayed
97 * extents need to be actually allocated. To get around this, we explicitly set
98 * aside a few blocks which will not be reserved in delayed allocation.
100 * We need to reserve 4 fsbs _per AG_ for the freelist and 4 more to handle a
101 * potential split of the file's bmap btree.
105 struct xfs_mount *mp)
107 return mp->m_sb.sb_agcount * (XFS_ALLOC_AGFL_RESERVE + 4);
111 * When deciding how much space to allocate out of an AG, we limit the
112 * allocation maximum size to the size the AG. However, we cannot use all the
113 * blocks in the AG - some are permanently used by metadata. These
114 * blocks are generally:
115 * - the AG superblock, AGF, AGI and AGFL
116 * - the AGF (bno and cnt) and AGI btree root blocks, and optionally
117 * the AGI free inode and rmap btree root blocks.
118 * - blocks on the AGFL according to xfs_alloc_set_aside() limits
119 * - the rmapbt root block
121 * The AG headers are sector sized, so the amount of space they take up is
122 * dependent on filesystem geometry. The others are all single blocks.
125 xfs_alloc_ag_max_usable(
126 struct xfs_mount *mp)
130 blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */
131 blocks += XFS_ALLOC_AGFL_RESERVE;
132 blocks += 3; /* AGF, AGI btree root blocks */
133 if (xfs_sb_version_hasfinobt(&mp->m_sb))
134 blocks++; /* finobt root block */
135 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
136 blocks++; /* rmap root block */
137 if (xfs_sb_version_hasreflink(&mp->m_sb))
138 blocks++; /* refcount root block */
140 return mp->m_sb.sb_agblocks - blocks;
144 * Lookup the record equal to [bno, len] in the btree given by cur.
146 STATIC int /* error */
148 struct xfs_btree_cur *cur, /* btree cursor */
149 xfs_agblock_t bno, /* starting block of extent */
150 xfs_extlen_t len, /* length of extent */
151 int *stat) /* success/failure */
153 cur->bc_rec.a.ar_startblock = bno;
154 cur->bc_rec.a.ar_blockcount = len;
155 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
159 * Lookup the first record greater than or equal to [bno, len]
160 * in the btree given by cur.
164 struct xfs_btree_cur *cur, /* btree cursor */
165 xfs_agblock_t bno, /* starting block of extent */
166 xfs_extlen_t len, /* length of extent */
167 int *stat) /* success/failure */
169 cur->bc_rec.a.ar_startblock = bno;
170 cur->bc_rec.a.ar_blockcount = len;
171 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
175 * Lookup the first record less than or equal to [bno, len]
176 * in the btree given by cur.
180 struct xfs_btree_cur *cur, /* btree cursor */
181 xfs_agblock_t bno, /* starting block of extent */
182 xfs_extlen_t len, /* length of extent */
183 int *stat) /* success/failure */
185 cur->bc_rec.a.ar_startblock = bno;
186 cur->bc_rec.a.ar_blockcount = len;
187 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
191 * Update the record referred to by cur to the value given
193 * This either works (return 0) or gets an EFSCORRUPTED error.
195 STATIC int /* error */
197 struct xfs_btree_cur *cur, /* btree cursor */
198 xfs_agblock_t bno, /* starting block of extent */
199 xfs_extlen_t len) /* length of extent */
201 union xfs_btree_rec rec;
203 rec.alloc.ar_startblock = cpu_to_be32(bno);
204 rec.alloc.ar_blockcount = cpu_to_be32(len);
205 return xfs_btree_update(cur, &rec);
209 * Get the data from the pointed-to record.
213 struct xfs_btree_cur *cur, /* btree cursor */
214 xfs_agblock_t *bno, /* output: starting block of extent */
215 xfs_extlen_t *len, /* output: length of extent */
216 int *stat) /* output: success/failure */
218 struct xfs_mount *mp = cur->bc_mp;
219 xfs_agnumber_t agno = cur->bc_private.a.agno;
220 union xfs_btree_rec *rec;
223 error = xfs_btree_get_rec(cur, &rec, stat);
224 if (error || !(*stat))
227 *bno = be32_to_cpu(rec->alloc.ar_startblock);
228 *len = be32_to_cpu(rec->alloc.ar_blockcount);
233 /* check for valid extent range, including overflow */
234 if (!xfs_verify_agbno(mp, agno, *bno))
236 if (*bno > *bno + *len)
238 if (!xfs_verify_agbno(mp, agno, *bno + *len - 1))
245 "%s Freespace BTree record corruption in AG %d detected!",
246 cur->bc_btnum == XFS_BTNUM_BNO ? "Block" : "Size", agno);
248 "start block 0x%x block count 0x%x", *bno, *len);
249 return -EFSCORRUPTED;
253 * Compute aligned version of the found extent.
254 * Takes alignment and min length into account.
257 xfs_alloc_compute_aligned(
258 xfs_alloc_arg_t *args, /* allocation argument structure */
259 xfs_agblock_t foundbno, /* starting block in found extent */
260 xfs_extlen_t foundlen, /* length in found extent */
261 xfs_agblock_t *resbno, /* result block number */
262 xfs_extlen_t *reslen, /* result length */
265 xfs_agblock_t bno = foundbno;
266 xfs_extlen_t len = foundlen;
270 /* Trim busy sections out of found extent */
271 busy = xfs_extent_busy_trim(args, &bno, &len, busy_gen);
274 * If we have a largish extent that happens to start before min_agbno,
275 * see if we can shift it into range...
277 if (bno < args->min_agbno && bno + len > args->min_agbno) {
278 diff = args->min_agbno - bno;
285 if (args->alignment > 1 && len >= args->minlen) {
286 xfs_agblock_t aligned_bno = roundup(bno, args->alignment);
288 diff = aligned_bno - bno;
290 *resbno = aligned_bno;
291 *reslen = diff >= len ? 0 : len - diff;
301 * Compute best start block and diff for "near" allocations.
302 * freelen >= wantlen already checked by caller.
304 STATIC xfs_extlen_t /* difference value (absolute) */
305 xfs_alloc_compute_diff(
306 xfs_agblock_t wantbno, /* target starting block */
307 xfs_extlen_t wantlen, /* target length */
308 xfs_extlen_t alignment, /* target alignment */
309 int datatype, /* are we allocating data? */
310 xfs_agblock_t freebno, /* freespace's starting block */
311 xfs_extlen_t freelen, /* freespace's length */
312 xfs_agblock_t *newbnop) /* result: best start block from free */
314 xfs_agblock_t freeend; /* end of freespace extent */
315 xfs_agblock_t newbno1; /* return block number */
316 xfs_agblock_t newbno2; /* other new block number */
317 xfs_extlen_t newlen1=0; /* length with newbno1 */
318 xfs_extlen_t newlen2=0; /* length with newbno2 */
319 xfs_agblock_t wantend; /* end of target extent */
320 bool userdata = xfs_alloc_is_userdata(datatype);
322 ASSERT(freelen >= wantlen);
323 freeend = freebno + freelen;
324 wantend = wantbno + wantlen;
326 * We want to allocate from the start of a free extent if it is past
327 * the desired block or if we are allocating user data and the free
328 * extent is before desired block. The second case is there to allow
329 * for contiguous allocation from the remaining free space if the file
330 * grows in the short term.
332 if (freebno >= wantbno || (userdata && freeend < wantend)) {
333 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
334 newbno1 = NULLAGBLOCK;
335 } else if (freeend >= wantend && alignment > 1) {
336 newbno1 = roundup(wantbno, alignment);
337 newbno2 = newbno1 - alignment;
338 if (newbno1 >= freeend)
339 newbno1 = NULLAGBLOCK;
341 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
342 if (newbno2 < freebno)
343 newbno2 = NULLAGBLOCK;
345 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
346 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
347 if (newlen1 < newlen2 ||
348 (newlen1 == newlen2 &&
349 XFS_ABSDIFF(newbno1, wantbno) >
350 XFS_ABSDIFF(newbno2, wantbno)))
352 } else if (newbno2 != NULLAGBLOCK)
354 } else if (freeend >= wantend) {
356 } else if (alignment > 1) {
357 newbno1 = roundup(freeend - wantlen, alignment);
358 if (newbno1 > freeend - wantlen &&
359 newbno1 - alignment >= freebno)
360 newbno1 -= alignment;
361 else if (newbno1 >= freeend)
362 newbno1 = NULLAGBLOCK;
364 newbno1 = freeend - wantlen;
366 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
370 * Fix up the length, based on mod and prod.
371 * len should be k * prod + mod for some k.
372 * If len is too small it is returned unchanged.
373 * If len hits maxlen it is left alone.
377 xfs_alloc_arg_t *args) /* allocation argument structure */
382 ASSERT(args->mod < args->prod);
384 ASSERT(rlen >= args->minlen);
385 ASSERT(rlen <= args->maxlen);
386 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
387 (args->mod == 0 && rlen < args->prod))
389 k = rlen % args->prod;
393 rlen = rlen - (k - args->mod);
395 rlen = rlen - args->prod + (args->mod - k);
396 /* casts to (int) catch length underflows */
397 if ((int)rlen < (int)args->minlen)
399 ASSERT(rlen >= args->minlen && rlen <= args->maxlen);
400 ASSERT(rlen % args->prod == args->mod);
401 ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >=
402 rlen + args->minleft);
407 * Update the two btrees, logically removing from freespace the extent
408 * starting at rbno, rlen blocks. The extent is contained within the
409 * actual (current) free extent fbno for flen blocks.
410 * Flags are passed in indicating whether the cursors are set to the
413 STATIC int /* error code */
414 xfs_alloc_fixup_trees(
415 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
416 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
417 xfs_agblock_t fbno, /* starting block of free extent */
418 xfs_extlen_t flen, /* length of free extent */
419 xfs_agblock_t rbno, /* starting block of returned extent */
420 xfs_extlen_t rlen, /* length of returned extent */
421 int flags) /* flags, XFSA_FIXUP_... */
423 int error; /* error code */
424 int i; /* operation results */
425 xfs_agblock_t nfbno1; /* first new free startblock */
426 xfs_agblock_t nfbno2; /* second new free startblock */
427 xfs_extlen_t nflen1=0; /* first new free length */
428 xfs_extlen_t nflen2=0; /* second new free length */
429 struct xfs_mount *mp;
434 * Look up the record in the by-size tree if necessary.
436 if (flags & XFSA_FIXUP_CNT_OK) {
438 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
440 XFS_WANT_CORRUPTED_RETURN(mp,
441 i == 1 && nfbno1 == fbno && nflen1 == flen);
444 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
446 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
449 * Look up the record in the by-block tree if necessary.
451 if (flags & XFSA_FIXUP_BNO_OK) {
453 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
455 XFS_WANT_CORRUPTED_RETURN(mp,
456 i == 1 && nfbno1 == fbno && nflen1 == flen);
459 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
461 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
465 if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) {
466 struct xfs_btree_block *bnoblock;
467 struct xfs_btree_block *cntblock;
469 bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
470 cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
472 XFS_WANT_CORRUPTED_RETURN(mp,
473 bnoblock->bb_numrecs == cntblock->bb_numrecs);
478 * Deal with all four cases: the allocated record is contained
479 * within the freespace record, so we can have new freespace
480 * at either (or both) end, or no freespace remaining.
482 if (rbno == fbno && rlen == flen)
483 nfbno1 = nfbno2 = NULLAGBLOCK;
484 else if (rbno == fbno) {
485 nfbno1 = rbno + rlen;
486 nflen1 = flen - rlen;
487 nfbno2 = NULLAGBLOCK;
488 } else if (rbno + rlen == fbno + flen) {
490 nflen1 = flen - rlen;
491 nfbno2 = NULLAGBLOCK;
494 nflen1 = rbno - fbno;
495 nfbno2 = rbno + rlen;
496 nflen2 = (fbno + flen) - nfbno2;
499 * Delete the entry from the by-size btree.
501 if ((error = xfs_btree_delete(cnt_cur, &i)))
503 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
505 * Add new by-size btree entry(s).
507 if (nfbno1 != NULLAGBLOCK) {
508 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
510 XFS_WANT_CORRUPTED_RETURN(mp, i == 0);
511 if ((error = xfs_btree_insert(cnt_cur, &i)))
513 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
515 if (nfbno2 != NULLAGBLOCK) {
516 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
518 XFS_WANT_CORRUPTED_RETURN(mp, i == 0);
519 if ((error = xfs_btree_insert(cnt_cur, &i)))
521 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
524 * Fix up the by-block btree entry(s).
526 if (nfbno1 == NULLAGBLOCK) {
528 * No remaining freespace, just delete the by-block tree entry.
530 if ((error = xfs_btree_delete(bno_cur, &i)))
532 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
535 * Update the by-block entry to start later|be shorter.
537 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
540 if (nfbno2 != NULLAGBLOCK) {
542 * 2 resulting free entries, need to add one.
544 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
546 XFS_WANT_CORRUPTED_RETURN(mp, i == 0);
547 if ((error = xfs_btree_insert(bno_cur, &i)))
549 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
554 static xfs_failaddr_t
558 struct xfs_mount *mp = bp->b_target->bt_mount;
559 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
563 * There is no verification of non-crc AGFLs because mkfs does not
564 * initialise the AGFL to zero or NULL. Hence the only valid part of the
565 * AGFL is what the AGF says is active. We can't get to the AGF, so we
566 * can't verify just those entries are valid.
568 if (!xfs_sb_version_hascrc(&mp->m_sb))
571 if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid))
572 return __this_address;
573 if (be32_to_cpu(agfl->agfl_magicnum) != XFS_AGFL_MAGIC)
574 return __this_address;
576 * during growfs operations, the perag is not fully initialised,
577 * so we can't use it for any useful checking. growfs ensures we can't
578 * use it by using uncached buffers that don't have the perag attached
579 * so we can detect and avoid this problem.
581 if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno)
582 return __this_address;
584 for (i = 0; i < xfs_agfl_size(mp); i++) {
585 if (be32_to_cpu(agfl->agfl_bno[i]) != NULLAGBLOCK &&
586 be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks)
587 return __this_address;
590 if (!xfs_log_check_lsn(mp, be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)))
591 return __this_address;
596 xfs_agfl_read_verify(
599 struct xfs_mount *mp = bp->b_target->bt_mount;
603 * There is no verification of non-crc AGFLs because mkfs does not
604 * initialise the AGFL to zero or NULL. Hence the only valid part of the
605 * AGFL is what the AGF says is active. We can't get to the AGF, so we
606 * can't verify just those entries are valid.
608 if (!xfs_sb_version_hascrc(&mp->m_sb))
611 if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF))
612 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
614 fa = xfs_agfl_verify(bp);
616 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
621 xfs_agfl_write_verify(
624 struct xfs_mount *mp = bp->b_target->bt_mount;
625 struct xfs_buf_log_item *bip = bp->b_log_item;
628 /* no verification of non-crc AGFLs */
629 if (!xfs_sb_version_hascrc(&mp->m_sb))
632 fa = xfs_agfl_verify(bp);
634 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
639 XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn);
641 xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF);
644 const struct xfs_buf_ops xfs_agfl_buf_ops = {
646 .verify_read = xfs_agfl_read_verify,
647 .verify_write = xfs_agfl_write_verify,
648 .verify_struct = xfs_agfl_verify,
652 * Read in the allocation group free block array.
656 xfs_mount_t *mp, /* mount point structure */
657 xfs_trans_t *tp, /* transaction pointer */
658 xfs_agnumber_t agno, /* allocation group number */
659 xfs_buf_t **bpp) /* buffer for the ag free block array */
661 xfs_buf_t *bp; /* return value */
664 ASSERT(agno != NULLAGNUMBER);
665 error = xfs_trans_read_buf(
666 mp, tp, mp->m_ddev_targp,
667 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
668 XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops);
671 xfs_buf_set_ref(bp, XFS_AGFL_REF);
677 xfs_alloc_update_counters(
678 struct xfs_trans *tp,
679 struct xfs_perag *pag,
680 struct xfs_buf *agbp,
683 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
685 pag->pagf_freeblks += len;
686 be32_add_cpu(&agf->agf_freeblks, len);
688 xfs_trans_agblocks_delta(tp, len);
689 if (unlikely(be32_to_cpu(agf->agf_freeblks) >
690 be32_to_cpu(agf->agf_length)))
691 return -EFSCORRUPTED;
693 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
698 * Allocation group level functions.
702 * Allocate a variable extent in the allocation group agno.
703 * Type and bno are used to determine where in the allocation group the
705 * Extent's length (returned in *len) will be between minlen and maxlen,
706 * and of the form k * prod + mod unless there's nothing that large.
707 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
709 STATIC int /* error */
710 xfs_alloc_ag_vextent(
711 xfs_alloc_arg_t *args) /* argument structure for allocation */
715 ASSERT(args->minlen > 0);
716 ASSERT(args->maxlen > 0);
717 ASSERT(args->minlen <= args->maxlen);
718 ASSERT(args->mod < args->prod);
719 ASSERT(args->alignment > 0);
722 * Branch to correct routine based on the type.
725 switch (args->type) {
726 case XFS_ALLOCTYPE_THIS_AG:
727 error = xfs_alloc_ag_vextent_size(args);
729 case XFS_ALLOCTYPE_NEAR_BNO:
730 error = xfs_alloc_ag_vextent_near(args);
732 case XFS_ALLOCTYPE_THIS_BNO:
733 error = xfs_alloc_ag_vextent_exact(args);
740 if (error || args->agbno == NULLAGBLOCK)
743 ASSERT(args->len >= args->minlen);
744 ASSERT(args->len <= args->maxlen);
745 ASSERT(!args->wasfromfl || args->resv != XFS_AG_RESV_AGFL);
746 ASSERT(args->agbno % args->alignment == 0);
748 /* if not file data, insert new block into the reverse map btree */
749 if (!xfs_rmap_should_skip_owner_update(&args->oinfo)) {
750 error = xfs_rmap_alloc(args->tp, args->agbp, args->agno,
751 args->agbno, args->len, &args->oinfo);
756 if (!args->wasfromfl) {
757 error = xfs_alloc_update_counters(args->tp, args->pag,
759 -((long)(args->len)));
763 ASSERT(!xfs_extent_busy_search(args->mp, args->agno,
764 args->agbno, args->len));
767 xfs_ag_resv_alloc_extent(args->pag, args->resv, args);
769 XFS_STATS_INC(args->mp, xs_allocx);
770 XFS_STATS_ADD(args->mp, xs_allocb, args->len);
775 * Allocate a variable extent at exactly agno/bno.
776 * Extent's length (returned in *len) will be between minlen and maxlen,
777 * and of the form k * prod + mod unless there's nothing that large.
778 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
780 STATIC int /* error */
781 xfs_alloc_ag_vextent_exact(
782 xfs_alloc_arg_t *args) /* allocation argument structure */
784 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
785 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
787 xfs_agblock_t fbno; /* start block of found extent */
788 xfs_extlen_t flen; /* length of found extent */
789 xfs_agblock_t tbno; /* start block of busy extent */
790 xfs_extlen_t tlen; /* length of busy extent */
791 xfs_agblock_t tend; /* end block of busy extent */
792 int i; /* success/failure of operation */
795 ASSERT(args->alignment == 1);
798 * Allocate/initialize a cursor for the by-number freespace btree.
800 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
801 args->agno, XFS_BTNUM_BNO);
804 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
805 * Look for the closest free block <= bno, it must contain bno
806 * if any free block does.
808 error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i);
815 * Grab the freespace record.
817 error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i);
820 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
821 ASSERT(fbno <= args->agbno);
824 * Check for overlapping busy extents.
828 xfs_extent_busy_trim(args, &tbno, &tlen, &busy_gen);
831 * Give up if the start of the extent is busy, or the freespace isn't
832 * long enough for the minimum request.
834 if (tbno > args->agbno)
836 if (tlen < args->minlen)
839 if (tend < args->agbno + args->minlen)
843 * End of extent will be smaller of the freespace end and the
844 * maximal requested end.
846 * Fix the length according to mod and prod if given.
848 args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen)
850 xfs_alloc_fix_len(args);
851 ASSERT(args->agbno + args->len <= tend);
854 * We are allocating agbno for args->len
855 * Allocate/initialize a cursor for the by-size btree.
857 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
858 args->agno, XFS_BTNUM_CNT);
859 ASSERT(args->agbno + args->len <=
860 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
861 error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno,
862 args->len, XFSA_FIXUP_BNO_OK);
864 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
868 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
869 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
872 trace_xfs_alloc_exact_done(args);
876 /* Didn't find it, return null. */
877 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
878 args->agbno = NULLAGBLOCK;
879 trace_xfs_alloc_exact_notfound(args);
883 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
884 trace_xfs_alloc_exact_error(args);
889 * Search the btree in a given direction via the search cursor and compare
890 * the records found against the good extent we've already found.
893 xfs_alloc_find_best_extent(
894 struct xfs_alloc_arg *args, /* allocation argument structure */
895 struct xfs_btree_cur **gcur, /* good cursor */
896 struct xfs_btree_cur **scur, /* searching cursor */
897 xfs_agblock_t gdiff, /* difference for search comparison */
898 xfs_agblock_t *sbno, /* extent found by search */
899 xfs_extlen_t *slen, /* extent length */
900 xfs_agblock_t *sbnoa, /* aligned extent found by search */
901 xfs_extlen_t *slena, /* aligned extent length */
902 int dir) /* 0 = search right, 1 = search left */
910 /* The good extent is perfect, no need to search. */
915 * Look until we find a better one, run out of space or run off the end.
918 error = xfs_alloc_get_rec(*scur, sbno, slen, &i);
921 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
922 xfs_alloc_compute_aligned(args, *sbno, *slen,
923 sbnoa, slena, &busy_gen);
926 * The good extent is closer than this one.
929 if (*sbnoa > args->max_agbno)
931 if (*sbnoa >= args->agbno + gdiff)
934 if (*sbnoa < args->min_agbno)
936 if (*sbnoa <= args->agbno - gdiff)
941 * Same distance, compare length and pick the best.
943 if (*slena >= args->minlen) {
944 args->len = XFS_EXTLEN_MIN(*slena, args->maxlen);
945 xfs_alloc_fix_len(args);
947 sdiff = xfs_alloc_compute_diff(args->agbno, args->len,
949 args->datatype, *sbnoa,
953 * Choose closer size and invalidate other cursor.
961 error = xfs_btree_increment(*scur, 0, &i);
963 error = xfs_btree_decrement(*scur, 0, &i);
969 xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR);
974 xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR);
979 /* caller invalidates cursors */
984 * Allocate a variable extent near bno in the allocation group agno.
985 * Extent's length (returned in len) will be between minlen and maxlen,
986 * and of the form k * prod + mod unless there's nothing that large.
987 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
989 STATIC int /* error */
990 xfs_alloc_ag_vextent_near(
991 xfs_alloc_arg_t *args) /* allocation argument structure */
993 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
994 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
995 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
996 xfs_agblock_t gtbno; /* start bno of right side entry */
997 xfs_agblock_t gtbnoa; /* aligned ... */
998 xfs_extlen_t gtdiff; /* difference to right side entry */
999 xfs_extlen_t gtlen; /* length of right side entry */
1000 xfs_extlen_t gtlena; /* aligned ... */
1001 xfs_agblock_t gtnew; /* useful start bno of right side */
1002 int error; /* error code */
1003 int i; /* result code, temporary */
1004 int j; /* result code, temporary */
1005 xfs_agblock_t ltbno; /* start bno of left side entry */
1006 xfs_agblock_t ltbnoa; /* aligned ... */
1007 xfs_extlen_t ltdiff; /* difference to left side entry */
1008 xfs_extlen_t ltlen; /* length of left side entry */
1009 xfs_extlen_t ltlena; /* aligned ... */
1010 xfs_agblock_t ltnew; /* useful start bno of left side */
1011 xfs_extlen_t rlen; /* length of returned extent */
1016 * Randomly don't execute the first algorithm.
1018 int dofirst; /* set to do first algorithm */
1020 dofirst = prandom_u32() & 1;
1023 /* handle unitialized agbno range so caller doesn't have to */
1024 if (!args->min_agbno && !args->max_agbno)
1025 args->max_agbno = args->mp->m_sb.sb_agblocks - 1;
1026 ASSERT(args->min_agbno <= args->max_agbno);
1028 /* clamp agbno to the range if it's outside */
1029 if (args->agbno < args->min_agbno)
1030 args->agbno = args->min_agbno;
1031 if (args->agbno > args->max_agbno)
1032 args->agbno = args->max_agbno;
1043 * Get a cursor for the by-size btree.
1045 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1046 args->agno, XFS_BTNUM_CNT);
1049 * See if there are any free extents as big as maxlen.
1051 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
1054 * If none, then pick up the last entry in the tree unless the
1058 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno,
1061 if (i == 0 || ltlen == 0) {
1062 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1063 trace_xfs_alloc_near_noentry(args);
1068 args->wasfromfl = 0;
1072 * If the requested extent is large wrt the freespaces available
1073 * in this a.g., then the cursor will be pointing to a btree entry
1074 * near the right edge of the tree. If it's in the last btree leaf
1075 * block, then we just examine all the entries in that block
1076 * that are big enough, and pick the best one.
1077 * This is written as a while loop so we can break out of it,
1078 * but we never loop back to the top.
1080 while (xfs_btree_islastblock(cnt_cur, 0)) {
1083 xfs_extlen_t blen=0;
1084 xfs_agblock_t bnew=0;
1091 * Start from the entry that lookup found, sequence through
1092 * all larger free blocks. If we're actually pointing at a
1093 * record smaller than maxlen, go to the start of this block,
1094 * and skip all those smaller than minlen.
1096 if (ltlen || args->alignment > 1) {
1097 cnt_cur->bc_ptrs[0] = 1;
1099 if ((error = xfs_alloc_get_rec(cnt_cur, <bno,
1102 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1103 if (ltlen >= args->minlen)
1105 if ((error = xfs_btree_increment(cnt_cur, 0, &i)))
1108 ASSERT(ltlen >= args->minlen);
1112 i = cnt_cur->bc_ptrs[0];
1113 for (j = 1, blen = 0, bdiff = 0;
1114 !error && j && (blen < args->maxlen || bdiff > 0);
1115 error = xfs_btree_increment(cnt_cur, 0, &j)) {
1117 * For each entry, decide if it's better than
1118 * the previous best entry.
1120 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
1122 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1123 busy = xfs_alloc_compute_aligned(args, ltbno, ltlen,
1124 <bnoa, <lena, &busy_gen);
1125 if (ltlena < args->minlen)
1127 if (ltbnoa < args->min_agbno || ltbnoa > args->max_agbno)
1129 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1130 xfs_alloc_fix_len(args);
1131 ASSERT(args->len >= args->minlen);
1132 if (args->len < blen)
1134 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1135 args->alignment, args->datatype, ltbnoa,
1137 if (ltnew != NULLAGBLOCK &&
1138 (args->len > blen || ltdiff < bdiff)) {
1142 besti = cnt_cur->bc_ptrs[0];
1146 * It didn't work. We COULD be in a case where
1147 * there's a good record somewhere, so try again.
1152 * Point at the best entry, and retrieve it again.
1154 cnt_cur->bc_ptrs[0] = besti;
1155 if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i)))
1157 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1158 ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1162 * We are allocating starting at bnew for blen blocks.
1165 ASSERT(bnew >= ltbno);
1166 ASSERT(bnew + blen <= ltbno + ltlen);
1168 * Set up a cursor for the by-bno tree.
1170 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp,
1171 args->agbp, args->agno, XFS_BTNUM_BNO);
1173 * Fix up the btree entries.
1175 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
1176 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
1178 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1179 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1181 trace_xfs_alloc_near_first(args);
1186 * Search in the by-bno tree to the left and to the right
1187 * simultaneously, until in each case we find a space big enough,
1188 * or run into the edge of the tree. When we run into the edge,
1189 * we deallocate that cursor.
1190 * If both searches succeed, we compare the two spaces and pick
1192 * With alignment, it's possible for both to fail; the upper
1193 * level algorithm that picks allocation groups for allocations
1194 * is not supposed to do this.
1197 * Allocate and initialize the cursor for the leftward search.
1199 bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1200 args->agno, XFS_BTNUM_BNO);
1202 * Lookup <= bno to find the leftward search's starting point.
1204 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
1208 * Didn't find anything; use this cursor for the rightward
1211 bno_cur_gt = bno_cur_lt;
1215 * Found something. Duplicate the cursor for the rightward search.
1217 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
1220 * Increment the cursor, so we will point at the entry just right
1221 * of the leftward entry if any, or to the leftmost entry.
1223 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1227 * It failed, there are no rightward entries.
1229 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
1233 * Loop going left with the leftward cursor, right with the
1234 * rightward cursor, until either both directions give up or
1235 * we find an entry at least as big as minlen.
1239 if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i)))
1241 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1242 busy |= xfs_alloc_compute_aligned(args, ltbno, ltlen,
1243 <bnoa, <lena, &busy_gen);
1244 if (ltlena >= args->minlen && ltbnoa >= args->min_agbno)
1246 if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i)))
1248 if (!i || ltbnoa < args->min_agbno) {
1249 xfs_btree_del_cursor(bno_cur_lt,
1255 if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i)))
1257 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1258 busy |= xfs_alloc_compute_aligned(args, gtbno, gtlen,
1259 >bnoa, >lena, &busy_gen);
1260 if (gtlena >= args->minlen && gtbnoa <= args->max_agbno)
1262 if ((error = xfs_btree_increment(bno_cur_gt, 0, &i)))
1264 if (!i || gtbnoa > args->max_agbno) {
1265 xfs_btree_del_cursor(bno_cur_gt,
1270 } while (bno_cur_lt || bno_cur_gt);
1273 * Got both cursors still active, need to find better entry.
1275 if (bno_cur_lt && bno_cur_gt) {
1276 if (ltlena >= args->minlen) {
1278 * Left side is good, look for a right side entry.
1280 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1281 xfs_alloc_fix_len(args);
1282 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1283 args->alignment, args->datatype, ltbnoa,
1286 error = xfs_alloc_find_best_extent(args,
1287 &bno_cur_lt, &bno_cur_gt,
1288 ltdiff, >bno, >len,
1290 0 /* search right */);
1292 ASSERT(gtlena >= args->minlen);
1295 * Right side is good, look for a left side entry.
1297 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1298 xfs_alloc_fix_len(args);
1299 gtdiff = xfs_alloc_compute_diff(args->agbno, args->len,
1300 args->alignment, args->datatype, gtbnoa,
1303 error = xfs_alloc_find_best_extent(args,
1304 &bno_cur_gt, &bno_cur_lt,
1305 gtdiff, <bno, <len,
1307 1 /* search left */);
1315 * If we couldn't get anything, give up.
1317 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1318 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1321 trace_xfs_alloc_near_busy(args);
1322 xfs_extent_busy_flush(args->mp, args->pag, busy_gen);
1325 trace_xfs_alloc_size_neither(args);
1326 args->agbno = NULLAGBLOCK;
1331 * At this point we have selected a freespace entry, either to the
1332 * left or to the right. If it's on the right, copy all the
1333 * useful variables to the "left" set so we only have one
1334 * copy of this code.
1337 bno_cur_lt = bno_cur_gt;
1348 * Fix up the length and compute the useful address.
1350 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1351 xfs_alloc_fix_len(args);
1353 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment,
1354 args->datatype, ltbnoa, ltlena, <new);
1355 ASSERT(ltnew >= ltbno);
1356 ASSERT(ltnew + rlen <= ltbnoa + ltlena);
1357 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1358 ASSERT(ltnew >= args->min_agbno && ltnew <= args->max_agbno);
1359 args->agbno = ltnew;
1361 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1362 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1366 trace_xfs_alloc_near_greater(args);
1368 trace_xfs_alloc_near_lesser(args);
1370 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1371 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1375 trace_xfs_alloc_near_error(args);
1376 if (cnt_cur != NULL)
1377 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1378 if (bno_cur_lt != NULL)
1379 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1380 if (bno_cur_gt != NULL)
1381 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1386 * Allocate a variable extent anywhere in the allocation group agno.
1387 * Extent's length (returned in len) will be between minlen and maxlen,
1388 * and of the form k * prod + mod unless there's nothing that large.
1389 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1391 STATIC int /* error */
1392 xfs_alloc_ag_vextent_size(
1393 xfs_alloc_arg_t *args) /* allocation argument structure */
1395 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1396 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1397 int error; /* error result */
1398 xfs_agblock_t fbno; /* start of found freespace */
1399 xfs_extlen_t flen; /* length of found freespace */
1400 int i; /* temp status variable */
1401 xfs_agblock_t rbno; /* returned block number */
1402 xfs_extlen_t rlen; /* length of returned extent */
1408 * Allocate and initialize a cursor for the by-size btree.
1410 cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1411 args->agno, XFS_BTNUM_CNT);
1416 * Look for an entry >= maxlen+alignment-1 blocks.
1418 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1419 args->maxlen + args->alignment - 1, &i)))
1423 * If none then we have to settle for a smaller extent. In the case that
1424 * there are no large extents, this will return the last entry in the
1425 * tree unless the tree is empty. In the case that there are only busy
1426 * large extents, this will return the largest small extent unless there
1427 * are no smaller extents available.
1430 error = xfs_alloc_ag_vextent_small(args, cnt_cur,
1434 if (i == 0 || flen == 0) {
1435 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1436 trace_xfs_alloc_size_noentry(args);
1440 busy = xfs_alloc_compute_aligned(args, fbno, flen, &rbno,
1444 * Search for a non-busy extent that is large enough.
1447 error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i);
1450 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1452 busy = xfs_alloc_compute_aligned(args, fbno, flen,
1453 &rbno, &rlen, &busy_gen);
1455 if (rlen >= args->maxlen)
1458 error = xfs_btree_increment(cnt_cur, 0, &i);
1463 * Our only valid extents must have been busy.
1464 * Make it unbusy by forcing the log out and
1467 xfs_btree_del_cursor(cnt_cur,
1469 trace_xfs_alloc_size_busy(args);
1470 xfs_extent_busy_flush(args->mp,
1471 args->pag, busy_gen);
1478 * In the first case above, we got the last entry in the
1479 * by-size btree. Now we check to see if the space hits maxlen
1480 * once aligned; if not, we search left for something better.
1481 * This can't happen in the second case above.
1483 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1484 XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 ||
1485 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1486 if (rlen < args->maxlen) {
1487 xfs_agblock_t bestfbno;
1488 xfs_extlen_t bestflen;
1489 xfs_agblock_t bestrbno;
1490 xfs_extlen_t bestrlen;
1497 if ((error = xfs_btree_decrement(cnt_cur, 0, &i)))
1501 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1504 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1505 if (flen < bestrlen)
1507 busy = xfs_alloc_compute_aligned(args, fbno, flen,
1508 &rbno, &rlen, &busy_gen);
1509 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1510 XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 ||
1511 (rlen <= flen && rbno + rlen <= fbno + flen),
1513 if (rlen > bestrlen) {
1518 if (rlen == args->maxlen)
1522 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1525 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1531 args->wasfromfl = 0;
1533 * Fix up the length.
1536 if (rlen < args->minlen) {
1538 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1539 trace_xfs_alloc_size_busy(args);
1540 xfs_extent_busy_flush(args->mp, args->pag, busy_gen);
1545 xfs_alloc_fix_len(args);
1548 XFS_WANT_CORRUPTED_GOTO(args->mp, rlen <= flen, error0);
1550 * Allocate and initialize a cursor for the by-block tree.
1552 bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp,
1553 args->agno, XFS_BTNUM_BNO);
1554 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1555 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1557 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1558 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1559 cnt_cur = bno_cur = NULL;
1562 XFS_WANT_CORRUPTED_GOTO(args->mp,
1563 args->agbno + args->len <=
1564 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1566 trace_xfs_alloc_size_done(args);
1570 trace_xfs_alloc_size_error(args);
1572 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1574 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1578 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1579 trace_xfs_alloc_size_nominleft(args);
1580 args->agbno = NULLAGBLOCK;
1585 * Deal with the case where only small freespaces remain.
1586 * Either return the contents of the last freespace record,
1587 * or allocate space from the freelist if there is nothing in the tree.
1589 STATIC int /* error */
1590 xfs_alloc_ag_vextent_small(
1591 xfs_alloc_arg_t *args, /* allocation argument structure */
1592 xfs_btree_cur_t *ccur, /* by-size cursor */
1593 xfs_agblock_t *fbnop, /* result block number */
1594 xfs_extlen_t *flenp, /* result length */
1595 int *stat) /* status: 0-freelist, 1-normal/none */
1597 struct xfs_owner_info oinfo;
1603 if ((error = xfs_btree_decrement(ccur, 0, &i)))
1606 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1608 XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0);
1611 * Nothing in the btree, try the freelist. Make sure
1612 * to respect minleft even when pulling from the
1615 else if (args->minlen == 1 && args->alignment == 1 &&
1616 args->resv != XFS_AG_RESV_AGFL &&
1617 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1619 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1622 if (fbno != NULLAGBLOCK) {
1623 xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1,
1624 xfs_alloc_allow_busy_reuse(args->datatype));
1626 if (xfs_alloc_is_userdata(args->datatype)) {
1629 bp = xfs_btree_get_bufs(args->mp, args->tp,
1630 args->agno, fbno, 0);
1632 error = -EFSCORRUPTED;
1635 xfs_trans_binval(args->tp, bp);
1639 XFS_WANT_CORRUPTED_GOTO(args->mp,
1640 args->agbno + args->len <=
1641 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1643 args->wasfromfl = 1;
1644 trace_xfs_alloc_small_freelist(args);
1647 * If we're feeding an AGFL block to something that
1648 * doesn't live in the free space, we need to clear
1649 * out the OWN_AG rmap.
1651 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_AG);
1652 error = xfs_rmap_free(args->tp, args->agbp, args->agno,
1661 * Nothing in the freelist.
1667 * Can't allocate from the freelist for some reason.
1674 * Can't do the allocation, give up.
1676 if (flen < args->minlen) {
1677 args->agbno = NULLAGBLOCK;
1678 trace_xfs_alloc_small_notenough(args);
1684 trace_xfs_alloc_small_done(args);
1688 trace_xfs_alloc_small_error(args);
1693 * Free the extent starting at agno/bno for length.
1699 xfs_agnumber_t agno,
1702 struct xfs_owner_info *oinfo,
1703 enum xfs_ag_resv_type type)
1705 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1706 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1707 int error; /* error return value */
1708 xfs_agblock_t gtbno; /* start of right neighbor block */
1709 xfs_extlen_t gtlen; /* length of right neighbor block */
1710 int haveleft; /* have a left neighbor block */
1711 int haveright; /* have a right neighbor block */
1712 int i; /* temp, result code */
1713 xfs_agblock_t ltbno; /* start of left neighbor block */
1714 xfs_extlen_t ltlen; /* length of left neighbor block */
1715 xfs_mount_t *mp; /* mount point struct for filesystem */
1716 xfs_agblock_t nbno; /* new starting block of freespace */
1717 xfs_extlen_t nlen; /* new length of freespace */
1718 xfs_perag_t *pag; /* per allocation group data */
1720 bno_cur = cnt_cur = NULL;
1723 if (!xfs_rmap_should_skip_owner_update(oinfo)) {
1724 error = xfs_rmap_free(tp, agbp, agno, bno, len, oinfo);
1730 * Allocate and initialize a cursor for the by-block btree.
1732 bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO);
1734 * Look for a neighboring block on the left (lower block numbers)
1735 * that is contiguous with this space.
1737 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1741 * There is a block to our left.
1743 if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i)))
1745 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1747 * It's not contiguous, though.
1749 if (ltbno + ltlen < bno)
1753 * If this failure happens the request to free this
1754 * space was invalid, it's (partly) already free.
1757 XFS_WANT_CORRUPTED_GOTO(mp,
1758 ltbno + ltlen <= bno, error0);
1762 * Look for a neighboring block on the right (higher block numbers)
1763 * that is contiguous with this space.
1765 if ((error = xfs_btree_increment(bno_cur, 0, &haveright)))
1769 * There is a block to our right.
1771 if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i)))
1773 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1775 * It's not contiguous, though.
1777 if (bno + len < gtbno)
1781 * If this failure happens the request to free this
1782 * space was invalid, it's (partly) already free.
1785 XFS_WANT_CORRUPTED_GOTO(mp, gtbno >= bno + len, error0);
1789 * Now allocate and initialize a cursor for the by-size tree.
1791 cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT);
1793 * Have both left and right contiguous neighbors.
1794 * Merge all three into a single free block.
1796 if (haveleft && haveright) {
1798 * Delete the old by-size entry on the left.
1800 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1802 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1803 if ((error = xfs_btree_delete(cnt_cur, &i)))
1805 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1807 * Delete the old by-size entry on the right.
1809 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1811 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1812 if ((error = xfs_btree_delete(cnt_cur, &i)))
1814 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1816 * Delete the old by-block entry for the right block.
1818 if ((error = xfs_btree_delete(bno_cur, &i)))
1820 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1822 * Move the by-block cursor back to the left neighbor.
1824 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1826 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1829 * Check that this is the right record: delete didn't
1830 * mangle the cursor.
1833 xfs_agblock_t xxbno;
1836 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1839 XFS_WANT_CORRUPTED_GOTO(mp,
1840 i == 1 && xxbno == ltbno && xxlen == ltlen,
1845 * Update remaining by-block entry to the new, joined block.
1848 nlen = len + ltlen + gtlen;
1849 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1853 * Have only a left contiguous neighbor.
1854 * Merge it together with the new freespace.
1856 else if (haveleft) {
1858 * Delete the old by-size entry on the left.
1860 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1862 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1863 if ((error = xfs_btree_delete(cnt_cur, &i)))
1865 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1867 * Back up the by-block cursor to the left neighbor, and
1868 * update its length.
1870 if ((error = xfs_btree_decrement(bno_cur, 0, &i)))
1872 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1875 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1879 * Have only a right contiguous neighbor.
1880 * Merge it together with the new freespace.
1882 else if (haveright) {
1884 * Delete the old by-size entry on the right.
1886 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1888 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1889 if ((error = xfs_btree_delete(cnt_cur, &i)))
1891 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1893 * Update the starting block and length of the right
1894 * neighbor in the by-block tree.
1898 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1902 * No contiguous neighbors.
1903 * Insert the new freespace into the by-block tree.
1908 if ((error = xfs_btree_insert(bno_cur, &i)))
1910 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1912 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1915 * In all cases we need to insert the new freespace in the by-size tree.
1917 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1919 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, error0);
1920 if ((error = xfs_btree_insert(cnt_cur, &i)))
1922 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1923 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1927 * Update the freespace totals in the ag and superblock.
1929 pag = xfs_perag_get(mp, agno);
1930 error = xfs_alloc_update_counters(tp, pag, agbp, len);
1931 xfs_ag_resv_free_extent(pag, type, tp, len);
1936 XFS_STATS_INC(mp, xs_freex);
1937 XFS_STATS_ADD(mp, xs_freeb, len);
1939 trace_xfs_free_extent(mp, agno, bno, len, type, haveleft, haveright);
1944 trace_xfs_free_extent(mp, agno, bno, len, type, -1, -1);
1946 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1948 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1953 * Visible (exported) allocation/free functions.
1954 * Some of these are used just by xfs_alloc_btree.c and this file.
1958 * Compute and fill in value of m_ag_maxlevels.
1961 xfs_alloc_compute_maxlevels(
1962 xfs_mount_t *mp) /* file system mount structure */
1964 mp->m_ag_maxlevels = xfs_btree_compute_maxlevels(mp->m_alloc_mnr,
1965 (mp->m_sb.sb_agblocks + 1) / 2);
1969 * Find the length of the longest extent in an AG. The 'need' parameter
1970 * specifies how much space we're going to need for the AGFL and the
1971 * 'reserved' parameter tells us how many blocks in this AG are reserved for
1975 xfs_alloc_longest_free_extent(
1976 struct xfs_perag *pag,
1978 xfs_extlen_t reserved)
1980 xfs_extlen_t delta = 0;
1983 * If the AGFL needs a recharge, we'll have to subtract that from the
1986 if (need > pag->pagf_flcount)
1987 delta = need - pag->pagf_flcount;
1990 * If we cannot maintain others' reservations with space from the
1991 * not-longest freesp extents, we'll have to subtract /that/ from
1992 * the longest extent too.
1994 if (pag->pagf_freeblks - pag->pagf_longest < reserved)
1995 delta += reserved - (pag->pagf_freeblks - pag->pagf_longest);
1998 * If the longest extent is long enough to satisfy all the
1999 * reservations and AGFL rules in place, we can return this extent.
2001 if (pag->pagf_longest > delta)
2002 return pag->pagf_longest - delta;
2004 /* Otherwise, let the caller try for 1 block if there's space. */
2005 return pag->pagf_flcount > 0 || pag->pagf_longest > 0;
2009 xfs_alloc_min_freelist(
2010 struct xfs_mount *mp,
2011 struct xfs_perag *pag)
2013 unsigned int min_free;
2015 /* space needed by-bno freespace btree */
2016 min_free = min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_BNOi] + 1,
2017 mp->m_ag_maxlevels);
2018 /* space needed by-size freespace btree */
2019 min_free += min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_CNTi] + 1,
2020 mp->m_ag_maxlevels);
2021 /* space needed reverse mapping used space btree */
2022 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2023 min_free += min_t(unsigned int,
2024 pag->pagf_levels[XFS_BTNUM_RMAPi] + 1,
2025 mp->m_rmap_maxlevels);
2031 * Check if the operation we are fixing up the freelist for should go ahead or
2032 * not. If we are freeing blocks, we always allow it, otherwise the allocation
2033 * is dependent on whether the size and shape of free space available will
2034 * permit the requested allocation to take place.
2037 xfs_alloc_space_available(
2038 struct xfs_alloc_arg *args,
2039 xfs_extlen_t min_free,
2042 struct xfs_perag *pag = args->pag;
2043 xfs_extlen_t alloc_len, longest;
2044 xfs_extlen_t reservation; /* blocks that are still reserved */
2047 if (flags & XFS_ALLOC_FLAG_FREEING)
2050 reservation = xfs_ag_resv_needed(pag, args->resv);
2052 /* do we have enough contiguous free space for the allocation? */
2053 alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop;
2054 longest = xfs_alloc_longest_free_extent(pag, min_free, reservation);
2055 if (longest < alloc_len)
2058 /* do we have enough free space remaining for the allocation? */
2059 available = (int)(pag->pagf_freeblks + pag->pagf_flcount -
2060 reservation - min_free - args->minleft);
2061 if (available < (int)max(args->total, alloc_len))
2065 * Clamp maxlen to the amount of free space available for the actual
2066 * extent allocation.
2068 if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) {
2069 args->maxlen = available;
2070 ASSERT(args->maxlen > 0);
2071 ASSERT(args->maxlen >= args->minlen);
2078 xfs_free_agfl_block(
2079 struct xfs_trans *tp,
2080 xfs_agnumber_t agno,
2081 xfs_agblock_t agbno,
2082 struct xfs_buf *agbp,
2083 struct xfs_owner_info *oinfo)
2088 error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo,
2093 bp = xfs_btree_get_bufs(tp->t_mountp, tp, agno, agbno, 0);
2095 return -EFSCORRUPTED;
2096 xfs_trans_binval(tp, bp);
2102 * Check the agfl fields of the agf for inconsistency or corruption. The purpose
2103 * is to detect an agfl header padding mismatch between current and early v5
2104 * kernels. This problem manifests as a 1-slot size difference between the
2105 * on-disk flcount and the active [first, last] range of a wrapped agfl. This
2106 * may also catch variants of agfl count corruption unrelated to padding. Either
2107 * way, we'll reset the agfl and warn the user.
2109 * Return true if a reset is required before the agfl can be used, false
2113 xfs_agfl_needs_reset(
2114 struct xfs_mount *mp,
2115 struct xfs_agf *agf)
2117 uint32_t f = be32_to_cpu(agf->agf_flfirst);
2118 uint32_t l = be32_to_cpu(agf->agf_fllast);
2119 uint32_t c = be32_to_cpu(agf->agf_flcount);
2120 int agfl_size = xfs_agfl_size(mp);
2123 /* no agfl header on v4 supers */
2124 if (!xfs_sb_version_hascrc(&mp->m_sb))
2128 * The agf read verifier catches severe corruption of these fields.
2129 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
2130 * the verifier allows it.
2132 if (f >= agfl_size || l >= agfl_size)
2138 * Check consistency between the on-disk count and the active range. An
2139 * agfl padding mismatch manifests as an inconsistent flcount.
2144 active = agfl_size - f + l + 1;
2152 * Reset the agfl to an empty state. Ignore/drop any existing blocks since the
2153 * agfl content cannot be trusted. Warn the user that a repair is required to
2154 * recover leaked blocks.
2156 * The purpose of this mechanism is to handle filesystems affected by the agfl
2157 * header padding mismatch problem. A reset keeps the filesystem online with a
2158 * relatively minor free space accounting inconsistency rather than suffer the
2159 * inevitable crash from use of an invalid agfl block.
2163 struct xfs_trans *tp,
2164 struct xfs_buf *agbp,
2165 struct xfs_perag *pag)
2167 struct xfs_mount *mp = tp->t_mountp;
2168 struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp);
2170 ASSERT(pag->pagf_agflreset);
2171 trace_xfs_agfl_reset(mp, agf, 0, _RET_IP_);
2174 "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. "
2175 "Please unmount and run xfs_repair.",
2176 pag->pag_agno, pag->pagf_flcount);
2178 agf->agf_flfirst = 0;
2179 agf->agf_fllast = cpu_to_be32(xfs_agfl_size(mp) - 1);
2180 agf->agf_flcount = 0;
2181 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLLAST |
2184 pag->pagf_flcount = 0;
2185 pag->pagf_agflreset = false;
2189 * Defer an AGFL block free. This is effectively equivalent to
2190 * xfs_bmap_add_free() with some special handling particular to AGFL blocks.
2192 * Deferring AGFL frees helps prevent log reservation overruns due to too many
2193 * allocation operations in a transaction. AGFL frees are prone to this problem
2194 * because for one they are always freed one at a time. Further, an immediate
2195 * AGFL block free can cause a btree join and require another block free before
2196 * the real allocation can proceed. Deferring the free disconnects freeing up
2197 * the AGFL slot from freeing the block.
2200 xfs_defer_agfl_block(
2201 struct xfs_trans *tp,
2202 xfs_agnumber_t agno,
2203 xfs_fsblock_t agbno,
2204 struct xfs_owner_info *oinfo)
2206 struct xfs_mount *mp = tp->t_mountp;
2207 struct xfs_extent_free_item *new; /* new element */
2209 ASSERT(xfs_bmap_free_item_zone != NULL);
2210 ASSERT(oinfo != NULL);
2212 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
2213 new->xefi_startblock = XFS_AGB_TO_FSB(mp, agno, agbno);
2214 new->xefi_blockcount = 1;
2215 new->xefi_oinfo = *oinfo;
2216 new->xefi_skip_discard = false;
2218 trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
2220 xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &new->xefi_list);
2224 * Decide whether to use this allocation group for this allocation.
2225 * If so, fix up the btree freelist's size.
2228 xfs_alloc_fix_freelist(
2229 struct xfs_alloc_arg *args, /* allocation argument structure */
2230 int flags) /* XFS_ALLOC_FLAG_... */
2232 struct xfs_mount *mp = args->mp;
2233 struct xfs_perag *pag = args->pag;
2234 struct xfs_trans *tp = args->tp;
2235 struct xfs_buf *agbp = NULL;
2236 struct xfs_buf *agflbp = NULL;
2237 struct xfs_alloc_arg targs; /* local allocation arguments */
2238 xfs_agblock_t bno; /* freelist block */
2239 xfs_extlen_t need; /* total blocks needed in freelist */
2242 if (!pag->pagf_init) {
2243 error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
2246 if (!pag->pagf_init) {
2247 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
2248 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
2249 goto out_agbp_relse;
2254 * If this is a metadata preferred pag and we are user data then try
2255 * somewhere else if we are not being asked to try harder at this
2258 if (pag->pagf_metadata && xfs_alloc_is_userdata(args->datatype) &&
2259 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
2260 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
2261 goto out_agbp_relse;
2264 need = xfs_alloc_min_freelist(mp, pag);
2265 if (!xfs_alloc_space_available(args, need, flags |
2266 XFS_ALLOC_FLAG_CHECK))
2267 goto out_agbp_relse;
2270 * Get the a.g. freespace buffer.
2271 * Can fail if we're not blocking on locks, and it's held.
2274 error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp);
2278 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
2279 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
2284 /* reset a padding mismatched agfl before final free space check */
2285 if (pag->pagf_agflreset)
2286 xfs_agfl_reset(tp, agbp, pag);
2288 /* If there isn't enough total space or single-extent, reject it. */
2289 need = xfs_alloc_min_freelist(mp, pag);
2290 if (!xfs_alloc_space_available(args, need, flags))
2291 goto out_agbp_relse;
2294 * Make the freelist shorter if it's too long.
2296 * Note that from this point onwards, we will always release the agf and
2297 * agfl buffers on error. This handles the case where we error out and
2298 * the buffers are clean or may not have been joined to the transaction
2299 * and hence need to be released manually. If they have been joined to
2300 * the transaction, then xfs_trans_brelse() will handle them
2301 * appropriately based on the recursion count and dirty state of the
2304 * XXX (dgc): When we have lots of free space, does this buy us
2305 * anything other than extra overhead when we need to put more blocks
2306 * back on the free list? Maybe we should only do this when space is
2307 * getting low or the AGFL is more than half full?
2309 * The NOSHRINK flag prevents the AGFL from being shrunk if it's too
2310 * big; the NORMAP flag prevents AGFL expand/shrink operations from
2311 * updating the rmapbt. Both flags are used in xfs_repair while we're
2312 * rebuilding the rmapbt, and neither are used by the kernel. They're
2313 * both required to ensure that rmaps are correctly recorded for the
2314 * regenerated AGFL, bnobt, and cntbt. See repair/phase5.c and
2315 * repair/rmap.c in xfsprogs for details.
2317 memset(&targs, 0, sizeof(targs));
2318 if (flags & XFS_ALLOC_FLAG_NORMAP)
2319 xfs_rmap_skip_owner_update(&targs.oinfo);
2321 xfs_rmap_ag_owner(&targs.oinfo, XFS_RMAP_OWN_AG);
2322 while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) {
2323 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
2325 goto out_agbp_relse;
2327 /* defer agfl frees */
2328 xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
2334 targs.agno = args->agno;
2335 targs.alignment = targs.minlen = targs.prod = 1;
2336 targs.type = XFS_ALLOCTYPE_THIS_AG;
2338 error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp);
2340 goto out_agbp_relse;
2342 /* Make the freelist longer if it's too short. */
2343 while (pag->pagf_flcount < need) {
2345 targs.maxlen = need - pag->pagf_flcount;
2346 targs.resv = XFS_AG_RESV_AGFL;
2348 /* Allocate as many blocks as possible at once. */
2349 error = xfs_alloc_ag_vextent(&targs);
2351 goto out_agflbp_relse;
2354 * Stop if we run out. Won't happen if callers are obeying
2355 * the restrictions correctly. Can happen for free calls
2356 * on a completely full ag.
2358 if (targs.agbno == NULLAGBLOCK) {
2359 if (flags & XFS_ALLOC_FLAG_FREEING)
2361 goto out_agflbp_relse;
2364 * Put each allocated block on the list.
2366 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
2367 error = xfs_alloc_put_freelist(tp, agbp,
2370 goto out_agflbp_relse;
2373 xfs_trans_brelse(tp, agflbp);
2378 xfs_trans_brelse(tp, agflbp);
2381 xfs_trans_brelse(tp, agbp);
2388 * Get a block from the freelist.
2389 * Returns with the buffer for the block gotten.
2392 xfs_alloc_get_freelist(
2393 xfs_trans_t *tp, /* transaction pointer */
2394 xfs_buf_t *agbp, /* buffer containing the agf structure */
2395 xfs_agblock_t *bnop, /* block address retrieved from freelist */
2396 int btreeblk) /* destination is a AGF btree */
2398 xfs_agf_t *agf; /* a.g. freespace structure */
2399 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
2400 xfs_agblock_t bno; /* block number returned */
2404 xfs_mount_t *mp = tp->t_mountp;
2405 xfs_perag_t *pag; /* per allocation group data */
2408 * Freelist is empty, give up.
2410 agf = XFS_BUF_TO_AGF(agbp);
2411 if (!agf->agf_flcount) {
2412 *bnop = NULLAGBLOCK;
2416 * Read the array of free blocks.
2418 error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno),
2425 * Get the block number and update the data structures.
2427 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2428 bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2429 be32_add_cpu(&agf->agf_flfirst, 1);
2430 xfs_trans_brelse(tp, agflbp);
2431 if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp))
2432 agf->agf_flfirst = 0;
2434 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2435 ASSERT(!pag->pagf_agflreset);
2436 be32_add_cpu(&agf->agf_flcount, -1);
2437 xfs_trans_agflist_delta(tp, -1);
2438 pag->pagf_flcount--;
2441 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2443 be32_add_cpu(&agf->agf_btreeblks, 1);
2444 pag->pagf_btreeblks++;
2445 logflags |= XFS_AGF_BTREEBLKS;
2448 xfs_alloc_log_agf(tp, agbp, logflags);
2455 * Log the given fields from the agf structure.
2459 xfs_trans_t *tp, /* transaction pointer */
2460 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2461 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2463 int first; /* first byte offset */
2464 int last; /* last byte offset */
2465 static const short offsets[] = {
2466 offsetof(xfs_agf_t, agf_magicnum),
2467 offsetof(xfs_agf_t, agf_versionnum),
2468 offsetof(xfs_agf_t, agf_seqno),
2469 offsetof(xfs_agf_t, agf_length),
2470 offsetof(xfs_agf_t, agf_roots[0]),
2471 offsetof(xfs_agf_t, agf_levels[0]),
2472 offsetof(xfs_agf_t, agf_flfirst),
2473 offsetof(xfs_agf_t, agf_fllast),
2474 offsetof(xfs_agf_t, agf_flcount),
2475 offsetof(xfs_agf_t, agf_freeblks),
2476 offsetof(xfs_agf_t, agf_longest),
2477 offsetof(xfs_agf_t, agf_btreeblks),
2478 offsetof(xfs_agf_t, agf_uuid),
2479 offsetof(xfs_agf_t, agf_rmap_blocks),
2480 offsetof(xfs_agf_t, agf_refcount_blocks),
2481 offsetof(xfs_agf_t, agf_refcount_root),
2482 offsetof(xfs_agf_t, agf_refcount_level),
2483 /* needed so that we don't log the whole rest of the structure: */
2484 offsetof(xfs_agf_t, agf_spare64),
2488 trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_);
2490 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF);
2492 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2493 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2497 * Interface for inode allocation to force the pag data to be initialized.
2500 xfs_alloc_pagf_init(
2501 xfs_mount_t *mp, /* file system mount structure */
2502 xfs_trans_t *tp, /* transaction pointer */
2503 xfs_agnumber_t agno, /* allocation group number */
2504 int flags) /* XFS_ALLOC_FLAGS_... */
2509 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2512 xfs_trans_brelse(tp, bp);
2517 * Put the block on the freelist for the allocation group.
2520 xfs_alloc_put_freelist(
2521 xfs_trans_t *tp, /* transaction pointer */
2522 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2523 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2524 xfs_agblock_t bno, /* block being freed */
2525 int btreeblk) /* block came from a AGF btree */
2527 xfs_agf_t *agf; /* a.g. freespace structure */
2528 __be32 *blockp;/* pointer to array entry */
2531 xfs_mount_t *mp; /* mount structure */
2532 xfs_perag_t *pag; /* per allocation group data */
2536 agf = XFS_BUF_TO_AGF(agbp);
2539 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2540 be32_to_cpu(agf->agf_seqno), &agflbp)))
2542 be32_add_cpu(&agf->agf_fllast, 1);
2543 if (be32_to_cpu(agf->agf_fllast) == xfs_agfl_size(mp))
2544 agf->agf_fllast = 0;
2546 pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno));
2547 ASSERT(!pag->pagf_agflreset);
2548 be32_add_cpu(&agf->agf_flcount, 1);
2549 xfs_trans_agflist_delta(tp, 1);
2550 pag->pagf_flcount++;
2552 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2554 be32_add_cpu(&agf->agf_btreeblks, -1);
2555 pag->pagf_btreeblks--;
2556 logflags |= XFS_AGF_BTREEBLKS;
2560 xfs_alloc_log_agf(tp, agbp, logflags);
2562 ASSERT(be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp));
2564 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
2565 blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)];
2566 *blockp = cpu_to_be32(bno);
2567 startoff = (char *)blockp - (char *)agflbp->b_addr;
2569 xfs_alloc_log_agf(tp, agbp, logflags);
2571 xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF);
2572 xfs_trans_log_buf(tp, agflbp, startoff,
2573 startoff + sizeof(xfs_agblock_t) - 1);
2577 static xfs_failaddr_t
2581 struct xfs_mount *mp = bp->b_target->bt_mount;
2582 struct xfs_agf *agf = XFS_BUF_TO_AGF(bp);
2584 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2585 if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid))
2586 return __this_address;
2587 if (!xfs_log_check_lsn(mp,
2588 be64_to_cpu(XFS_BUF_TO_AGF(bp)->agf_lsn)))
2589 return __this_address;
2592 if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) &&
2593 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2594 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2595 be32_to_cpu(agf->agf_flfirst) < xfs_agfl_size(mp) &&
2596 be32_to_cpu(agf->agf_fllast) < xfs_agfl_size(mp) &&
2597 be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp)))
2598 return __this_address;
2600 if (be32_to_cpu(agf->agf_length) > mp->m_sb.sb_dblocks)
2601 return __this_address;
2603 if (be32_to_cpu(agf->agf_freeblks) < be32_to_cpu(agf->agf_longest) ||
2604 be32_to_cpu(agf->agf_freeblks) > be32_to_cpu(agf->agf_length))
2605 return __this_address;
2607 if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 ||
2608 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 ||
2609 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > XFS_BTREE_MAXLEVELS ||
2610 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > XFS_BTREE_MAXLEVELS)
2611 return __this_address;
2613 if (xfs_sb_version_hasrmapbt(&mp->m_sb) &&
2614 (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 ||
2615 be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) > XFS_BTREE_MAXLEVELS))
2616 return __this_address;
2618 if (xfs_sb_version_hasrmapbt(&mp->m_sb) &&
2619 be32_to_cpu(agf->agf_rmap_blocks) > be32_to_cpu(agf->agf_length))
2620 return __this_address;
2623 * during growfs operations, the perag is not fully initialised,
2624 * so we can't use it for any useful checking. growfs ensures we can't
2625 * use it by using uncached buffers that don't have the perag attached
2626 * so we can detect and avoid this problem.
2628 if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno)
2629 return __this_address;
2631 if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
2632 be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length))
2633 return __this_address;
2635 if (xfs_sb_version_hasreflink(&mp->m_sb) &&
2636 be32_to_cpu(agf->agf_refcount_blocks) >
2637 be32_to_cpu(agf->agf_length))
2638 return __this_address;
2640 if (xfs_sb_version_hasreflink(&mp->m_sb) &&
2641 (be32_to_cpu(agf->agf_refcount_level) < 1 ||
2642 be32_to_cpu(agf->agf_refcount_level) > XFS_BTREE_MAXLEVELS))
2643 return __this_address;
2650 xfs_agf_read_verify(
2653 struct xfs_mount *mp = bp->b_target->bt_mount;
2656 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2657 !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF))
2658 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2660 fa = xfs_agf_verify(bp);
2661 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_ALLOC_READ_AGF))
2662 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2667 xfs_agf_write_verify(
2670 struct xfs_mount *mp = bp->b_target->bt_mount;
2671 struct xfs_buf_log_item *bip = bp->b_log_item;
2674 fa = xfs_agf_verify(bp);
2676 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2680 if (!xfs_sb_version_hascrc(&mp->m_sb))
2684 XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2686 xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF);
2689 const struct xfs_buf_ops xfs_agf_buf_ops = {
2691 .verify_read = xfs_agf_read_verify,
2692 .verify_write = xfs_agf_write_verify,
2693 .verify_struct = xfs_agf_verify,
2697 * Read in the allocation group header (free/alloc section).
2701 struct xfs_mount *mp, /* mount point structure */
2702 struct xfs_trans *tp, /* transaction pointer */
2703 xfs_agnumber_t agno, /* allocation group number */
2704 int flags, /* XFS_BUF_ */
2705 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2709 trace_xfs_read_agf(mp, agno);
2711 ASSERT(agno != NULLAGNUMBER);
2712 error = xfs_trans_read_buf(
2713 mp, tp, mp->m_ddev_targp,
2714 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2715 XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops);
2721 ASSERT(!(*bpp)->b_error);
2722 xfs_buf_set_ref(*bpp, XFS_AGF_REF);
2727 * Read in the allocation group header (free/alloc section).
2731 struct xfs_mount *mp, /* mount point structure */
2732 struct xfs_trans *tp, /* transaction pointer */
2733 xfs_agnumber_t agno, /* allocation group number */
2734 int flags, /* XFS_ALLOC_FLAG_... */
2735 struct xfs_buf **bpp) /* buffer for the ag freelist header */
2737 struct xfs_agf *agf; /* ag freelist header */
2738 struct xfs_perag *pag; /* per allocation group data */
2741 trace_xfs_alloc_read_agf(mp, agno);
2743 ASSERT(agno != NULLAGNUMBER);
2744 error = xfs_read_agf(mp, tp, agno,
2745 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2751 ASSERT(!(*bpp)->b_error);
2753 agf = XFS_BUF_TO_AGF(*bpp);
2754 pag = xfs_perag_get(mp, agno);
2755 if (!pag->pagf_init) {
2756 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2757 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2758 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2759 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2760 pag->pagf_levels[XFS_BTNUM_BNOi] =
2761 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2762 pag->pagf_levels[XFS_BTNUM_CNTi] =
2763 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2764 pag->pagf_levels[XFS_BTNUM_RMAPi] =
2765 be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]);
2766 pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level);
2768 pag->pagf_agflreset = xfs_agfl_needs_reset(mp, agf);
2771 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2772 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2773 ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks));
2774 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2775 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2776 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2777 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2778 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2779 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2787 * Allocate an extent (variable-size).
2788 * Depending on the allocation type, we either look in a single allocation
2789 * group or loop over the allocation groups to find the result.
2793 struct xfs_alloc_arg *args) /* allocation argument structure */
2795 xfs_agblock_t agsize; /* allocation group size */
2797 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2798 struct xfs_mount *mp; /* mount structure pointer */
2799 xfs_agnumber_t sagno; /* starting allocation group number */
2800 xfs_alloctype_t type; /* input allocation type */
2802 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2805 type = args->otype = args->type;
2806 args->agbno = NULLAGBLOCK;
2808 * Just fix this up, for the case where the last a.g. is shorter
2809 * (or there's only one a.g.) and the caller couldn't easily figure
2810 * that out (xfs_bmap_alloc).
2812 agsize = mp->m_sb.sb_agblocks;
2813 if (args->maxlen > agsize)
2814 args->maxlen = agsize;
2815 if (args->alignment == 0)
2816 args->alignment = 1;
2817 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2818 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2819 ASSERT(args->minlen <= args->maxlen);
2820 ASSERT(args->minlen <= agsize);
2821 ASSERT(args->mod < args->prod);
2822 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2823 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2824 args->minlen > args->maxlen || args->minlen > agsize ||
2825 args->mod >= args->prod) {
2826 args->fsbno = NULLFSBLOCK;
2827 trace_xfs_alloc_vextent_badargs(args);
2832 case XFS_ALLOCTYPE_THIS_AG:
2833 case XFS_ALLOCTYPE_NEAR_BNO:
2834 case XFS_ALLOCTYPE_THIS_BNO:
2836 * These three force us into a single a.g.
2838 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2839 args->pag = xfs_perag_get(mp, args->agno);
2840 error = xfs_alloc_fix_freelist(args, 0);
2842 trace_xfs_alloc_vextent_nofix(args);
2846 trace_xfs_alloc_vextent_noagbp(args);
2849 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2850 if ((error = xfs_alloc_ag_vextent(args)))
2853 case XFS_ALLOCTYPE_START_BNO:
2855 * Try near allocation first, then anywhere-in-ag after
2856 * the first a.g. fails.
2858 if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) &&
2859 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2860 args->fsbno = XFS_AGB_TO_FSB(mp,
2861 ((mp->m_agfrotor / rotorstep) %
2862 mp->m_sb.sb_agcount), 0);
2865 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2866 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2868 case XFS_ALLOCTYPE_FIRST_AG:
2870 * Rotate through the allocation groups looking for a winner.
2872 if (type == XFS_ALLOCTYPE_FIRST_AG) {
2874 * Start with allocation group given by bno.
2876 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2877 args->type = XFS_ALLOCTYPE_THIS_AG;
2882 * Start with the given allocation group.
2884 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2885 flags = XFS_ALLOC_FLAG_TRYLOCK;
2888 * Loop over allocation groups twice; first time with
2889 * trylock set, second time without.
2892 args->pag = xfs_perag_get(mp, args->agno);
2893 error = xfs_alloc_fix_freelist(args, flags);
2895 trace_xfs_alloc_vextent_nofix(args);
2899 * If we get a buffer back then the allocation will fly.
2902 if ((error = xfs_alloc_ag_vextent(args)))
2907 trace_xfs_alloc_vextent_loopfailed(args);
2910 * Didn't work, figure out the next iteration.
2912 if (args->agno == sagno &&
2913 type == XFS_ALLOCTYPE_START_BNO)
2914 args->type = XFS_ALLOCTYPE_THIS_AG;
2916 * For the first allocation, we can try any AG to get
2917 * space. However, if we already have allocated a
2918 * block, we don't want to try AGs whose number is below
2919 * sagno. Otherwise, we may end up with out-of-order
2920 * locking of AGF, which might cause deadlock.
2922 if (++(args->agno) == mp->m_sb.sb_agcount) {
2923 if (args->tp->t_firstblock != NULLFSBLOCK)
2929 * Reached the starting a.g., must either be done
2930 * or switch to non-trylock mode.
2932 if (args->agno == sagno) {
2934 args->agbno = NULLAGBLOCK;
2935 trace_xfs_alloc_vextent_allfailed(args);
2940 if (type == XFS_ALLOCTYPE_START_BNO) {
2941 args->agbno = XFS_FSB_TO_AGBNO(mp,
2943 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2946 xfs_perag_put(args->pag);
2949 if (args->agno == sagno)
2950 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2951 (mp->m_sb.sb_agcount * rotorstep);
2953 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2954 (mp->m_sb.sb_agcount * rotorstep);
2961 if (args->agbno == NULLAGBLOCK)
2962 args->fsbno = NULLFSBLOCK;
2964 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2966 ASSERT(args->len >= args->minlen);
2967 ASSERT(args->len <= args->maxlen);
2968 ASSERT(args->agbno % args->alignment == 0);
2969 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2973 /* Zero the extent if we were asked to do so */
2974 if (args->datatype & XFS_ALLOC_USERDATA_ZERO) {
2975 error = xfs_zero_extent(args->ip, args->fsbno, args->len);
2981 xfs_perag_put(args->pag);
2984 xfs_perag_put(args->pag);
2988 /* Ensure that the freelist is at full capacity. */
2990 xfs_free_extent_fix_freelist(
2991 struct xfs_trans *tp,
2992 xfs_agnumber_t agno,
2993 struct xfs_buf **agbp)
2995 struct xfs_alloc_arg args;
2998 memset(&args, 0, sizeof(struct xfs_alloc_arg));
3000 args.mp = tp->t_mountp;
3004 * validate that the block number is legal - the enables us to detect
3005 * and handle a silent filesystem corruption rather than crashing.
3007 if (args.agno >= args.mp->m_sb.sb_agcount)
3008 return -EFSCORRUPTED;
3010 args.pag = xfs_perag_get(args.mp, args.agno);
3013 error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
3019 xfs_perag_put(args.pag);
3025 * Just break up the extent address and hand off to xfs_free_ag_extent
3026 * after fixing up the freelist.
3030 struct xfs_trans *tp, /* transaction pointer */
3031 xfs_fsblock_t bno, /* starting block number of extent */
3032 xfs_extlen_t len, /* length of extent */
3033 struct xfs_owner_info *oinfo, /* extent owner */
3034 enum xfs_ag_resv_type type, /* block reservation type */
3037 struct xfs_mount *mp = tp->t_mountp;
3038 struct xfs_buf *agbp;
3039 xfs_agnumber_t agno = XFS_FSB_TO_AGNO(mp, bno);
3040 xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp, bno);
3042 unsigned int busy_flags = 0;
3045 ASSERT(type != XFS_AG_RESV_AGFL);
3047 if (XFS_TEST_ERROR(false, mp,
3048 XFS_ERRTAG_FREE_EXTENT))
3051 error = xfs_free_extent_fix_freelist(tp, agno, &agbp);
3055 XFS_WANT_CORRUPTED_GOTO(mp, agbno < mp->m_sb.sb_agblocks, err);
3057 /* validate the extent size is legal now we have the agf locked */
3058 XFS_WANT_CORRUPTED_GOTO(mp,
3059 agbno + len <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_length),
3062 error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, type);
3067 busy_flags |= XFS_EXTENT_BUSY_SKIP_DISCARD;
3068 xfs_extent_busy_insert(tp, agno, agbno, len, busy_flags);
3072 xfs_trans_brelse(tp, agbp);
3076 struct xfs_alloc_query_range_info {
3077 xfs_alloc_query_range_fn fn;
3081 /* Format btree record and pass to our callback. */
3083 xfs_alloc_query_range_helper(
3084 struct xfs_btree_cur *cur,
3085 union xfs_btree_rec *rec,
3088 struct xfs_alloc_query_range_info *query = priv;
3089 struct xfs_alloc_rec_incore irec;
3091 irec.ar_startblock = be32_to_cpu(rec->alloc.ar_startblock);
3092 irec.ar_blockcount = be32_to_cpu(rec->alloc.ar_blockcount);
3093 return query->fn(cur, &irec, query->priv);
3096 /* Find all free space within a given range of blocks. */
3098 xfs_alloc_query_range(
3099 struct xfs_btree_cur *cur,
3100 struct xfs_alloc_rec_incore *low_rec,
3101 struct xfs_alloc_rec_incore *high_rec,
3102 xfs_alloc_query_range_fn fn,
3105 union xfs_btree_irec low_brec;
3106 union xfs_btree_irec high_brec;
3107 struct xfs_alloc_query_range_info query;
3109 ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3110 low_brec.a = *low_rec;
3111 high_brec.a = *high_rec;
3114 return xfs_btree_query_range(cur, &low_brec, &high_brec,
3115 xfs_alloc_query_range_helper, &query);
3118 /* Find all free space records. */
3120 xfs_alloc_query_all(
3121 struct xfs_btree_cur *cur,
3122 xfs_alloc_query_range_fn fn,
3125 struct xfs_alloc_query_range_info query;
3127 ASSERT(cur->bc_btnum == XFS_BTNUM_BNO);
3130 return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query);
3133 /* Is there a record covering a given extent? */
3135 xfs_alloc_has_record(
3136 struct xfs_btree_cur *cur,
3141 union xfs_btree_irec low;
3142 union xfs_btree_irec high;
3144 memset(&low, 0, sizeof(low));
3145 low.a.ar_startblock = bno;
3146 memset(&high, 0xFF, sizeof(high));
3147 high.a.ar_startblock = bno + len - 1;
3149 return xfs_btree_has_record(cur, &low, &high, exists);
3153 * Walk all the blocks in the AGFL. The @walk_fn can return any negative
3154 * error code or XFS_BTREE_QUERY_RANGE_ABORT.
3158 struct xfs_mount *mp,
3159 struct xfs_agf *agf,
3160 struct xfs_buf *agflbp,
3161 xfs_agfl_walk_fn walk_fn,
3168 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp);
3169 i = be32_to_cpu(agf->agf_flfirst);
3171 /* Nothing to walk in an empty AGFL. */
3172 if (agf->agf_flcount == cpu_to_be32(0))
3175 /* Otherwise, walk from first to last, wrapping as needed. */
3177 error = walk_fn(mp, be32_to_cpu(agfl_bno[i]), priv);
3180 if (i == be32_to_cpu(agf->agf_fllast))
3182 if (++i == xfs_agfl_size(mp))