1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
6 #include "libxfs_priv.h"
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_errortag.h"
22 #include "xfs_cksum.h"
23 #include "xfs_trace.h"
24 #include "xfs_trans.h"
25 #include "xfs_ag_resv.h"
28 extern kmem_zone_t
*xfs_bmap_free_item_zone
;
30 struct workqueue_struct
*xfs_alloc_wq
;
32 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
34 #define XFSA_FIXUP_BNO_OK 1
35 #define XFSA_FIXUP_CNT_OK 2
37 STATIC
int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t
*);
38 STATIC
int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t
*);
39 STATIC
int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t
*);
40 STATIC
int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t
*,
41 xfs_btree_cur_t
*, xfs_agblock_t
*, xfs_extlen_t
*, int *);
44 * Size of the AGFL. For CRC-enabled filesystes we steal a couple of slots in
45 * the beginning of the block for a proper header with the location information
52 unsigned int size
= mp
->m_sb
.sb_sectsize
;
54 if (xfs_sb_version_hascrc(&mp
->m_sb
))
55 size
-= sizeof(struct xfs_agfl
);
57 return size
/ sizeof(xfs_agblock_t
);
64 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
65 return XFS_RMAP_BLOCK(mp
) + 1;
66 if (xfs_sb_version_hasfinobt(&mp
->m_sb
))
67 return XFS_FIBT_BLOCK(mp
) + 1;
68 return XFS_IBT_BLOCK(mp
) + 1;
75 if (xfs_sb_version_hasreflink(&mp
->m_sb
))
76 return xfs_refc_block(mp
) + 1;
77 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
78 return XFS_RMAP_BLOCK(mp
) + 1;
79 if (xfs_sb_version_hasfinobt(&mp
->m_sb
))
80 return XFS_FIBT_BLOCK(mp
) + 1;
81 return XFS_IBT_BLOCK(mp
) + 1;
85 * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of
86 * AGF buffer (PV 947395), we place constraints on the relationship among
87 * actual allocations for data blocks, freelist blocks, and potential file data
88 * bmap btree blocks. However, these restrictions may result in no actual space
89 * allocated for a delayed extent, for example, a data block in a certain AG is
90 * allocated but there is no additional block for the additional bmap btree
91 * block due to a split of the bmap btree of the file. The result of this may
92 * lead to an infinite loop when the file gets flushed to disk and all delayed
93 * extents need to be actually allocated. To get around this, we explicitly set
94 * aside a few blocks which will not be reserved in delayed allocation.
96 * We need to reserve 4 fsbs _per AG_ for the freelist and 4 more to handle a
97 * potential split of the file's bmap btree.
101 struct xfs_mount
*mp
)
103 return mp
->m_sb
.sb_agcount
* (XFS_ALLOC_AGFL_RESERVE
+ 4);
107 * When deciding how much space to allocate out of an AG, we limit the
108 * allocation maximum size to the size the AG. However, we cannot use all the
109 * blocks in the AG - some are permanently used by metadata. These
110 * blocks are generally:
111 * - the AG superblock, AGF, AGI and AGFL
112 * - the AGF (bno and cnt) and AGI btree root blocks, and optionally
113 * the AGI free inode and rmap btree root blocks.
114 * - blocks on the AGFL according to xfs_alloc_set_aside() limits
115 * - the rmapbt root block
117 * The AG headers are sector sized, so the amount of space they take up is
118 * dependent on filesystem geometry. The others are all single blocks.
121 xfs_alloc_ag_max_usable(
122 struct xfs_mount
*mp
)
126 blocks
= XFS_BB_TO_FSB(mp
, XFS_FSS_TO_BB(mp
, 4)); /* ag headers */
127 blocks
+= XFS_ALLOC_AGFL_RESERVE
;
128 blocks
+= 3; /* AGF, AGI btree root blocks */
129 if (xfs_sb_version_hasfinobt(&mp
->m_sb
))
130 blocks
++; /* finobt root block */
131 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
132 blocks
++; /* rmap root block */
133 if (xfs_sb_version_hasreflink(&mp
->m_sb
))
134 blocks
++; /* refcount root block */
136 return mp
->m_sb
.sb_agblocks
- blocks
;
140 * Lookup the record equal to [bno, len] in the btree given by cur.
142 STATIC
int /* error */
144 struct xfs_btree_cur
*cur
, /* btree cursor */
145 xfs_agblock_t bno
, /* starting block of extent */
146 xfs_extlen_t len
, /* length of extent */
147 int *stat
) /* success/failure */
149 cur
->bc_rec
.a
.ar_startblock
= bno
;
150 cur
->bc_rec
.a
.ar_blockcount
= len
;
151 return xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, stat
);
155 * Lookup the first record greater than or equal to [bno, len]
156 * in the btree given by cur.
160 struct xfs_btree_cur
*cur
, /* btree cursor */
161 xfs_agblock_t bno
, /* starting block of extent */
162 xfs_extlen_t len
, /* length of extent */
163 int *stat
) /* success/failure */
165 cur
->bc_rec
.a
.ar_startblock
= bno
;
166 cur
->bc_rec
.a
.ar_blockcount
= len
;
167 return xfs_btree_lookup(cur
, XFS_LOOKUP_GE
, stat
);
171 * Lookup the first record less than or equal to [bno, len]
172 * in the btree given by cur.
176 struct xfs_btree_cur
*cur
, /* btree cursor */
177 xfs_agblock_t bno
, /* starting block of extent */
178 xfs_extlen_t len
, /* length of extent */
179 int *stat
) /* success/failure */
181 cur
->bc_rec
.a
.ar_startblock
= bno
;
182 cur
->bc_rec
.a
.ar_blockcount
= len
;
183 return xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, stat
);
187 * Update the record referred to by cur to the value given
189 * This either works (return 0) or gets an EFSCORRUPTED error.
191 STATIC
int /* error */
193 struct xfs_btree_cur
*cur
, /* btree cursor */
194 xfs_agblock_t bno
, /* starting block of extent */
195 xfs_extlen_t len
) /* length of extent */
197 union xfs_btree_rec rec
;
199 rec
.alloc
.ar_startblock
= cpu_to_be32(bno
);
200 rec
.alloc
.ar_blockcount
= cpu_to_be32(len
);
201 return xfs_btree_update(cur
, &rec
);
205 * Get the data from the pointed-to record.
209 struct xfs_btree_cur
*cur
, /* btree cursor */
210 xfs_agblock_t
*bno
, /* output: starting block of extent */
211 xfs_extlen_t
*len
, /* output: length of extent */
212 int *stat
) /* output: success/failure */
214 struct xfs_mount
*mp
= cur
->bc_mp
;
215 xfs_agnumber_t agno
= cur
->bc_private
.a
.agno
;
216 union xfs_btree_rec
*rec
;
219 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
220 if (error
|| !(*stat
))
223 *bno
= be32_to_cpu(rec
->alloc
.ar_startblock
);
224 *len
= be32_to_cpu(rec
->alloc
.ar_blockcount
);
229 /* check for valid extent range, including overflow */
230 if (!xfs_verify_agbno(mp
, agno
, *bno
))
232 if (*bno
> *bno
+ *len
)
234 if (!xfs_verify_agbno(mp
, agno
, *bno
+ *len
- 1))
241 "%s Freespace BTree record corruption in AG %d detected!",
242 cur
->bc_btnum
== XFS_BTNUM_BNO
? "Block" : "Size", agno
);
244 "start block 0x%x block count 0x%x", *bno
, *len
);
245 return -EFSCORRUPTED
;
249 * Compute aligned version of the found extent.
250 * Takes alignment and min length into account.
253 xfs_alloc_compute_aligned(
254 xfs_alloc_arg_t
*args
, /* allocation argument structure */
255 xfs_agblock_t foundbno
, /* starting block in found extent */
256 xfs_extlen_t foundlen
, /* length in found extent */
257 xfs_agblock_t
*resbno
, /* result block number */
258 xfs_extlen_t
*reslen
, /* result length */
261 xfs_agblock_t bno
= foundbno
;
262 xfs_extlen_t len
= foundlen
;
266 /* Trim busy sections out of found extent */
267 busy
= xfs_extent_busy_trim(args
, &bno
, &len
, busy_gen
);
270 * If we have a largish extent that happens to start before min_agbno,
271 * see if we can shift it into range...
273 if (bno
< args
->min_agbno
&& bno
+ len
> args
->min_agbno
) {
274 diff
= args
->min_agbno
- bno
;
281 if (args
->alignment
> 1 && len
>= args
->minlen
) {
282 xfs_agblock_t aligned_bno
= roundup(bno
, args
->alignment
);
284 diff
= aligned_bno
- bno
;
286 *resbno
= aligned_bno
;
287 *reslen
= diff
>= len
? 0 : len
- diff
;
297 * Compute best start block and diff for "near" allocations.
298 * freelen >= wantlen already checked by caller.
300 STATIC xfs_extlen_t
/* difference value (absolute) */
301 xfs_alloc_compute_diff(
302 xfs_agblock_t wantbno
, /* target starting block */
303 xfs_extlen_t wantlen
, /* target length */
304 xfs_extlen_t alignment
, /* target alignment */
305 int datatype
, /* are we allocating data? */
306 xfs_agblock_t freebno
, /* freespace's starting block */
307 xfs_extlen_t freelen
, /* freespace's length */
308 xfs_agblock_t
*newbnop
) /* result: best start block from free */
310 xfs_agblock_t freeend
; /* end of freespace extent */
311 xfs_agblock_t newbno1
; /* return block number */
312 xfs_agblock_t newbno2
; /* other new block number */
313 xfs_extlen_t newlen1
=0; /* length with newbno1 */
314 xfs_extlen_t newlen2
=0; /* length with newbno2 */
315 xfs_agblock_t wantend
; /* end of target extent */
316 bool userdata
= xfs_alloc_is_userdata(datatype
);
318 ASSERT(freelen
>= wantlen
);
319 freeend
= freebno
+ freelen
;
320 wantend
= wantbno
+ wantlen
;
322 * We want to allocate from the start of a free extent if it is past
323 * the desired block or if we are allocating user data and the free
324 * extent is before desired block. The second case is there to allow
325 * for contiguous allocation from the remaining free space if the file
326 * grows in the short term.
328 if (freebno
>= wantbno
|| (userdata
&& freeend
< wantend
)) {
329 if ((newbno1
= roundup(freebno
, alignment
)) >= freeend
)
330 newbno1
= NULLAGBLOCK
;
331 } else if (freeend
>= wantend
&& alignment
> 1) {
332 newbno1
= roundup(wantbno
, alignment
);
333 newbno2
= newbno1
- alignment
;
334 if (newbno1
>= freeend
)
335 newbno1
= NULLAGBLOCK
;
337 newlen1
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno1
);
338 if (newbno2
< freebno
)
339 newbno2
= NULLAGBLOCK
;
341 newlen2
= XFS_EXTLEN_MIN(wantlen
, freeend
- newbno2
);
342 if (newbno1
!= NULLAGBLOCK
&& newbno2
!= NULLAGBLOCK
) {
343 if (newlen1
< newlen2
||
344 (newlen1
== newlen2
&&
345 XFS_ABSDIFF(newbno1
, wantbno
) >
346 XFS_ABSDIFF(newbno2
, wantbno
)))
348 } else if (newbno2
!= NULLAGBLOCK
)
350 } else if (freeend
>= wantend
) {
352 } else if (alignment
> 1) {
353 newbno1
= roundup(freeend
- wantlen
, alignment
);
354 if (newbno1
> freeend
- wantlen
&&
355 newbno1
- alignment
>= freebno
)
356 newbno1
-= alignment
;
357 else if (newbno1
>= freeend
)
358 newbno1
= NULLAGBLOCK
;
360 newbno1
= freeend
- wantlen
;
362 return newbno1
== NULLAGBLOCK
? 0 : XFS_ABSDIFF(newbno1
, wantbno
);
366 * Fix up the length, based on mod and prod.
367 * len should be k * prod + mod for some k.
368 * If len is too small it is returned unchanged.
369 * If len hits maxlen it is left alone.
373 xfs_alloc_arg_t
*args
) /* allocation argument structure */
378 ASSERT(args
->mod
< args
->prod
);
380 ASSERT(rlen
>= args
->minlen
);
381 ASSERT(rlen
<= args
->maxlen
);
382 if (args
->prod
<= 1 || rlen
< args
->mod
|| rlen
== args
->maxlen
||
383 (args
->mod
== 0 && rlen
< args
->prod
))
385 k
= rlen
% args
->prod
;
389 rlen
= rlen
- (k
- args
->mod
);
391 rlen
= rlen
- args
->prod
+ (args
->mod
- k
);
392 /* casts to (int) catch length underflows */
393 if ((int)rlen
< (int)args
->minlen
)
395 ASSERT(rlen
>= args
->minlen
&& rlen
<= args
->maxlen
);
396 ASSERT(rlen
% args
->prod
== args
->mod
);
397 ASSERT(args
->pag
->pagf_freeblks
+ args
->pag
->pagf_flcount
>=
398 rlen
+ args
->minleft
);
403 * Update the two btrees, logically removing from freespace the extent
404 * starting at rbno, rlen blocks. The extent is contained within the
405 * actual (current) free extent fbno for flen blocks.
406 * Flags are passed in indicating whether the cursors are set to the
409 STATIC
int /* error code */
410 xfs_alloc_fixup_trees(
411 xfs_btree_cur_t
*cnt_cur
, /* cursor for by-size btree */
412 xfs_btree_cur_t
*bno_cur
, /* cursor for by-block btree */
413 xfs_agblock_t fbno
, /* starting block of free extent */
414 xfs_extlen_t flen
, /* length of free extent */
415 xfs_agblock_t rbno
, /* starting block of returned extent */
416 xfs_extlen_t rlen
, /* length of returned extent */
417 int flags
) /* flags, XFSA_FIXUP_... */
419 int error
; /* error code */
420 int i
; /* operation results */
421 xfs_agblock_t nfbno1
; /* first new free startblock */
422 xfs_agblock_t nfbno2
; /* second new free startblock */
423 xfs_extlen_t nflen1
=0; /* first new free length */
424 xfs_extlen_t nflen2
=0; /* second new free length */
425 struct xfs_mount
*mp
;
430 * Look up the record in the by-size tree if necessary.
432 if (flags
& XFSA_FIXUP_CNT_OK
) {
434 if ((error
= xfs_alloc_get_rec(cnt_cur
, &nfbno1
, &nflen1
, &i
)))
436 XFS_WANT_CORRUPTED_RETURN(mp
,
437 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
440 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, fbno
, flen
, &i
)))
442 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
445 * Look up the record in the by-block tree if necessary.
447 if (flags
& XFSA_FIXUP_BNO_OK
) {
449 if ((error
= xfs_alloc_get_rec(bno_cur
, &nfbno1
, &nflen1
, &i
)))
451 XFS_WANT_CORRUPTED_RETURN(mp
,
452 i
== 1 && nfbno1
== fbno
&& nflen1
== flen
);
455 if ((error
= xfs_alloc_lookup_eq(bno_cur
, fbno
, flen
, &i
)))
457 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
461 if (bno_cur
->bc_nlevels
== 1 && cnt_cur
->bc_nlevels
== 1) {
462 struct xfs_btree_block
*bnoblock
;
463 struct xfs_btree_block
*cntblock
;
465 bnoblock
= XFS_BUF_TO_BLOCK(bno_cur
->bc_bufs
[0]);
466 cntblock
= XFS_BUF_TO_BLOCK(cnt_cur
->bc_bufs
[0]);
468 XFS_WANT_CORRUPTED_RETURN(mp
,
469 bnoblock
->bb_numrecs
== cntblock
->bb_numrecs
);
474 * Deal with all four cases: the allocated record is contained
475 * within the freespace record, so we can have new freespace
476 * at either (or both) end, or no freespace remaining.
478 if (rbno
== fbno
&& rlen
== flen
)
479 nfbno1
= nfbno2
= NULLAGBLOCK
;
480 else if (rbno
== fbno
) {
481 nfbno1
= rbno
+ rlen
;
482 nflen1
= flen
- rlen
;
483 nfbno2
= NULLAGBLOCK
;
484 } else if (rbno
+ rlen
== fbno
+ flen
) {
486 nflen1
= flen
- rlen
;
487 nfbno2
= NULLAGBLOCK
;
490 nflen1
= rbno
- fbno
;
491 nfbno2
= rbno
+ rlen
;
492 nflen2
= (fbno
+ flen
) - nfbno2
;
495 * Delete the entry from the by-size btree.
497 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
499 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
501 * Add new by-size btree entry(s).
503 if (nfbno1
!= NULLAGBLOCK
) {
504 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno1
, nflen1
, &i
)))
506 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
507 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
509 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
511 if (nfbno2
!= NULLAGBLOCK
) {
512 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nfbno2
, nflen2
, &i
)))
514 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
515 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
517 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
520 * Fix up the by-block btree entry(s).
522 if (nfbno1
== NULLAGBLOCK
) {
524 * No remaining freespace, just delete the by-block tree entry.
526 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
528 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
531 * Update the by-block entry to start later|be shorter.
533 if ((error
= xfs_alloc_update(bno_cur
, nfbno1
, nflen1
)))
536 if (nfbno2
!= NULLAGBLOCK
) {
538 * 2 resulting free entries, need to add one.
540 if ((error
= xfs_alloc_lookup_eq(bno_cur
, nfbno2
, nflen2
, &i
)))
542 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 0);
543 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
545 XFS_WANT_CORRUPTED_RETURN(mp
, i
== 1);
550 static xfs_failaddr_t
554 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
555 struct xfs_agfl
*agfl
= XFS_BUF_TO_AGFL(bp
);
559 * There is no verification of non-crc AGFLs because mkfs does not
560 * initialise the AGFL to zero or NULL. Hence the only valid part of the
561 * AGFL is what the AGF says is active. We can't get to the AGF, so we
562 * can't verify just those entries are valid.
564 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
567 if (!uuid_equal(&agfl
->agfl_uuid
, &mp
->m_sb
.sb_meta_uuid
))
568 return __this_address
;
569 if (be32_to_cpu(agfl
->agfl_magicnum
) != XFS_AGFL_MAGIC
)
570 return __this_address
;
572 * during growfs operations, the perag is not fully initialised,
573 * so we can't use it for any useful checking. growfs ensures we can't
574 * use it by using uncached buffers that don't have the perag attached
575 * so we can detect and avoid this problem.
577 if (bp
->b_pag
&& be32_to_cpu(agfl
->agfl_seqno
) != bp
->b_pag
->pag_agno
)
578 return __this_address
;
580 for (i
= 0; i
< xfs_agfl_size(mp
); i
++) {
581 if (be32_to_cpu(agfl
->agfl_bno
[i
]) != NULLAGBLOCK
&&
582 be32_to_cpu(agfl
->agfl_bno
[i
]) >= mp
->m_sb
.sb_agblocks
)
583 return __this_address
;
586 if (!xfs_log_check_lsn(mp
, be64_to_cpu(XFS_BUF_TO_AGFL(bp
)->agfl_lsn
)))
587 return __this_address
;
592 xfs_agfl_read_verify(
595 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
599 * There is no verification of non-crc AGFLs because mkfs does not
600 * initialise the AGFL to zero or NULL. Hence the only valid part of the
601 * AGFL is what the AGF says is active. We can't get to the AGF, so we
602 * can't verify just those entries are valid.
604 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
607 if (!xfs_buf_verify_cksum(bp
, XFS_AGFL_CRC_OFF
))
608 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
610 fa
= xfs_agfl_verify(bp
);
612 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
617 xfs_agfl_write_verify(
620 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
621 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
624 /* no verification of non-crc AGFLs */
625 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
628 fa
= xfs_agfl_verify(bp
);
630 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
635 XFS_BUF_TO_AGFL(bp
)->agfl_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
637 xfs_buf_update_cksum(bp
, XFS_AGFL_CRC_OFF
);
640 const struct xfs_buf_ops xfs_agfl_buf_ops
= {
642 .verify_read
= xfs_agfl_read_verify
,
643 .verify_write
= xfs_agfl_write_verify
,
644 .verify_struct
= xfs_agfl_verify
,
648 * Read in the allocation group free block array.
652 xfs_mount_t
*mp
, /* mount point structure */
653 xfs_trans_t
*tp
, /* transaction pointer */
654 xfs_agnumber_t agno
, /* allocation group number */
655 xfs_buf_t
**bpp
) /* buffer for the ag free block array */
657 xfs_buf_t
*bp
; /* return value */
660 ASSERT(agno
!= NULLAGNUMBER
);
661 error
= xfs_trans_read_buf(
662 mp
, tp
, mp
->m_ddev_targp
,
663 XFS_AG_DADDR(mp
, agno
, XFS_AGFL_DADDR(mp
)),
664 XFS_FSS_TO_BB(mp
, 1), 0, &bp
, &xfs_agfl_buf_ops
);
667 xfs_buf_set_ref(bp
, XFS_AGFL_REF
);
673 xfs_alloc_update_counters(
674 struct xfs_trans
*tp
,
675 struct xfs_perag
*pag
,
676 struct xfs_buf
*agbp
,
679 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
681 pag
->pagf_freeblks
+= len
;
682 be32_add_cpu(&agf
->agf_freeblks
, len
);
684 xfs_trans_agblocks_delta(tp
, len
);
685 if (unlikely(be32_to_cpu(agf
->agf_freeblks
) >
686 be32_to_cpu(agf
->agf_length
)))
687 return -EFSCORRUPTED
;
689 xfs_alloc_log_agf(tp
, agbp
, XFS_AGF_FREEBLKS
);
694 * Allocation group level functions.
698 * Allocate a variable extent in the allocation group agno.
699 * Type and bno are used to determine where in the allocation group the
701 * Extent's length (returned in *len) will be between minlen and maxlen,
702 * and of the form k * prod + mod unless there's nothing that large.
703 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
705 STATIC
int /* error */
706 xfs_alloc_ag_vextent(
707 xfs_alloc_arg_t
*args
) /* argument structure for allocation */
711 ASSERT(args
->minlen
> 0);
712 ASSERT(args
->maxlen
> 0);
713 ASSERT(args
->minlen
<= args
->maxlen
);
714 ASSERT(args
->mod
< args
->prod
);
715 ASSERT(args
->alignment
> 0);
718 * Branch to correct routine based on the type.
721 switch (args
->type
) {
722 case XFS_ALLOCTYPE_THIS_AG
:
723 error
= xfs_alloc_ag_vextent_size(args
);
725 case XFS_ALLOCTYPE_NEAR_BNO
:
726 error
= xfs_alloc_ag_vextent_near(args
);
728 case XFS_ALLOCTYPE_THIS_BNO
:
729 error
= xfs_alloc_ag_vextent_exact(args
);
736 if (error
|| args
->agbno
== NULLAGBLOCK
)
739 ASSERT(args
->len
>= args
->minlen
);
740 ASSERT(args
->len
<= args
->maxlen
);
741 ASSERT(!args
->wasfromfl
|| args
->resv
!= XFS_AG_RESV_AGFL
);
742 ASSERT(args
->agbno
% args
->alignment
== 0);
744 /* if not file data, insert new block into the reverse map btree */
745 if (!xfs_rmap_should_skip_owner_update(&args
->oinfo
)) {
746 error
= xfs_rmap_alloc(args
->tp
, args
->agbp
, args
->agno
,
747 args
->agbno
, args
->len
, &args
->oinfo
);
752 if (!args
->wasfromfl
) {
753 error
= xfs_alloc_update_counters(args
->tp
, args
->pag
,
755 -((long)(args
->len
)));
759 ASSERT(!xfs_extent_busy_search(args
->mp
, args
->agno
,
760 args
->agbno
, args
->len
));
763 xfs_ag_resv_alloc_extent(args
->pag
, args
->resv
, args
);
765 XFS_STATS_INC(args
->mp
, xs_allocx
);
766 XFS_STATS_ADD(args
->mp
, xs_allocb
, args
->len
);
771 * Allocate a variable extent at exactly agno/bno.
772 * Extent's length (returned in *len) will be between minlen and maxlen,
773 * and of the form k * prod + mod unless there's nothing that large.
774 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
776 STATIC
int /* error */
777 xfs_alloc_ag_vextent_exact(
778 xfs_alloc_arg_t
*args
) /* allocation argument structure */
780 xfs_btree_cur_t
*bno_cur
;/* by block-number btree cursor */
781 xfs_btree_cur_t
*cnt_cur
;/* by count btree cursor */
783 xfs_agblock_t fbno
; /* start block of found extent */
784 xfs_extlen_t flen
; /* length of found extent */
785 xfs_agblock_t tbno
; /* start block of busy extent */
786 xfs_extlen_t tlen
; /* length of busy extent */
787 xfs_agblock_t tend
; /* end block of busy extent */
788 int i
; /* success/failure of operation */
791 ASSERT(args
->alignment
== 1);
794 * Allocate/initialize a cursor for the by-number freespace btree.
796 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
797 args
->agno
, XFS_BTNUM_BNO
);
800 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
801 * Look for the closest free block <= bno, it must contain bno
802 * if any free block does.
804 error
= xfs_alloc_lookup_le(bno_cur
, args
->agbno
, args
->minlen
, &i
);
811 * Grab the freespace record.
813 error
= xfs_alloc_get_rec(bno_cur
, &fbno
, &flen
, &i
);
816 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
817 ASSERT(fbno
<= args
->agbno
);
820 * Check for overlapping busy extents.
824 xfs_extent_busy_trim(args
, &tbno
, &tlen
, &busy_gen
);
827 * Give up if the start of the extent is busy, or the freespace isn't
828 * long enough for the minimum request.
830 if (tbno
> args
->agbno
)
832 if (tlen
< args
->minlen
)
835 if (tend
< args
->agbno
+ args
->minlen
)
839 * End of extent will be smaller of the freespace end and the
840 * maximal requested end.
842 * Fix the length according to mod and prod if given.
844 args
->len
= XFS_AGBLOCK_MIN(tend
, args
->agbno
+ args
->maxlen
)
846 xfs_alloc_fix_len(args
);
847 ASSERT(args
->agbno
+ args
->len
<= tend
);
850 * We are allocating agbno for args->len
851 * Allocate/initialize a cursor for the by-size btree.
853 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
854 args
->agno
, XFS_BTNUM_CNT
);
855 ASSERT(args
->agbno
+ args
->len
<=
856 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
857 error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
, args
->agbno
,
858 args
->len
, XFSA_FIXUP_BNO_OK
);
860 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
864 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
865 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
868 trace_xfs_alloc_exact_done(args
);
872 /* Didn't find it, return null. */
873 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
874 args
->agbno
= NULLAGBLOCK
;
875 trace_xfs_alloc_exact_notfound(args
);
879 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
880 trace_xfs_alloc_exact_error(args
);
885 * Search the btree in a given direction via the search cursor and compare
886 * the records found against the good extent we've already found.
889 xfs_alloc_find_best_extent(
890 struct xfs_alloc_arg
*args
, /* allocation argument structure */
891 struct xfs_btree_cur
**gcur
, /* good cursor */
892 struct xfs_btree_cur
**scur
, /* searching cursor */
893 xfs_agblock_t gdiff
, /* difference for search comparison */
894 xfs_agblock_t
*sbno
, /* extent found by search */
895 xfs_extlen_t
*slen
, /* extent length */
896 xfs_agblock_t
*sbnoa
, /* aligned extent found by search */
897 xfs_extlen_t
*slena
, /* aligned extent length */
898 int dir
) /* 0 = search right, 1 = search left */
906 /* The good extent is perfect, no need to search. */
911 * Look until we find a better one, run out of space or run off the end.
914 error
= xfs_alloc_get_rec(*scur
, sbno
, slen
, &i
);
917 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
918 xfs_alloc_compute_aligned(args
, *sbno
, *slen
,
919 sbnoa
, slena
, &busy_gen
);
922 * The good extent is closer than this one.
925 if (*sbnoa
> args
->max_agbno
)
927 if (*sbnoa
>= args
->agbno
+ gdiff
)
930 if (*sbnoa
< args
->min_agbno
)
932 if (*sbnoa
<= args
->agbno
- gdiff
)
937 * Same distance, compare length and pick the best.
939 if (*slena
>= args
->minlen
) {
940 args
->len
= XFS_EXTLEN_MIN(*slena
, args
->maxlen
);
941 xfs_alloc_fix_len(args
);
943 sdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
945 args
->datatype
, *sbnoa
,
949 * Choose closer size and invalidate other cursor.
957 error
= xfs_btree_increment(*scur
, 0, &i
);
959 error
= xfs_btree_decrement(*scur
, 0, &i
);
965 xfs_btree_del_cursor(*scur
, XFS_BTREE_NOERROR
);
970 xfs_btree_del_cursor(*gcur
, XFS_BTREE_NOERROR
);
975 /* caller invalidates cursors */
980 * Allocate a variable extent near bno in the allocation group agno.
981 * Extent's length (returned in len) will be between minlen and maxlen,
982 * and of the form k * prod + mod unless there's nothing that large.
983 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
985 STATIC
int /* error */
986 xfs_alloc_ag_vextent_near(
987 xfs_alloc_arg_t
*args
) /* allocation argument structure */
989 xfs_btree_cur_t
*bno_cur_gt
; /* cursor for bno btree, right side */
990 xfs_btree_cur_t
*bno_cur_lt
; /* cursor for bno btree, left side */
991 xfs_btree_cur_t
*cnt_cur
; /* cursor for count btree */
992 xfs_agblock_t gtbno
; /* start bno of right side entry */
993 xfs_agblock_t gtbnoa
; /* aligned ... */
994 xfs_extlen_t gtdiff
; /* difference to right side entry */
995 xfs_extlen_t gtlen
; /* length of right side entry */
996 xfs_extlen_t gtlena
; /* aligned ... */
997 xfs_agblock_t gtnew
; /* useful start bno of right side */
998 int error
; /* error code */
999 int i
; /* result code, temporary */
1000 int j
; /* result code, temporary */
1001 xfs_agblock_t ltbno
; /* start bno of left side entry */
1002 xfs_agblock_t ltbnoa
; /* aligned ... */
1003 xfs_extlen_t ltdiff
; /* difference to left side entry */
1004 xfs_extlen_t ltlen
; /* length of left side entry */
1005 xfs_extlen_t ltlena
; /* aligned ... */
1006 xfs_agblock_t ltnew
; /* useful start bno of left side */
1007 xfs_extlen_t rlen
; /* length of returned extent */
1012 * Randomly don't execute the first algorithm.
1014 int dofirst
; /* set to do first algorithm */
1016 dofirst
= prandom_u32() & 1;
1019 /* handle unitialized agbno range so caller doesn't have to */
1020 if (!args
->min_agbno
&& !args
->max_agbno
)
1021 args
->max_agbno
= args
->mp
->m_sb
.sb_agblocks
- 1;
1022 ASSERT(args
->min_agbno
<= args
->max_agbno
);
1024 /* clamp agbno to the range if it's outside */
1025 if (args
->agbno
< args
->min_agbno
)
1026 args
->agbno
= args
->min_agbno
;
1027 if (args
->agbno
> args
->max_agbno
)
1028 args
->agbno
= args
->max_agbno
;
1039 * Get a cursor for the by-size btree.
1041 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1042 args
->agno
, XFS_BTNUM_CNT
);
1045 * See if there are any free extents as big as maxlen.
1047 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0, args
->maxlen
, &i
)))
1050 * If none, then pick up the last entry in the tree unless the
1054 if ((error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
, <bno
,
1057 if (i
== 0 || ltlen
== 0) {
1058 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1059 trace_xfs_alloc_near_noentry(args
);
1064 args
->wasfromfl
= 0;
1068 * If the requested extent is large wrt the freespaces available
1069 * in this a.g., then the cursor will be pointing to a btree entry
1070 * near the right edge of the tree. If it's in the last btree leaf
1071 * block, then we just examine all the entries in that block
1072 * that are big enough, and pick the best one.
1073 * This is written as a while loop so we can break out of it,
1074 * but we never loop back to the top.
1076 while (xfs_btree_islastblock(cnt_cur
, 0)) {
1079 xfs_extlen_t blen
=0;
1080 xfs_agblock_t bnew
=0;
1087 * Start from the entry that lookup found, sequence through
1088 * all larger free blocks. If we're actually pointing at a
1089 * record smaller than maxlen, go to the start of this block,
1090 * and skip all those smaller than minlen.
1092 if (ltlen
|| args
->alignment
> 1) {
1093 cnt_cur
->bc_ptrs
[0] = 1;
1095 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
,
1098 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1099 if (ltlen
>= args
->minlen
)
1101 if ((error
= xfs_btree_increment(cnt_cur
, 0, &i
)))
1104 ASSERT(ltlen
>= args
->minlen
);
1108 i
= cnt_cur
->bc_ptrs
[0];
1109 for (j
= 1, blen
= 0, bdiff
= 0;
1110 !error
&& j
&& (blen
< args
->maxlen
|| bdiff
> 0);
1111 error
= xfs_btree_increment(cnt_cur
, 0, &j
)) {
1113 * For each entry, decide if it's better than
1114 * the previous best entry.
1116 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
1118 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1119 busy
= xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
1120 <bnoa
, <lena
, &busy_gen
);
1121 if (ltlena
< args
->minlen
)
1123 if (ltbnoa
< args
->min_agbno
|| ltbnoa
> args
->max_agbno
)
1125 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1126 xfs_alloc_fix_len(args
);
1127 ASSERT(args
->len
>= args
->minlen
);
1128 if (args
->len
< blen
)
1130 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1131 args
->alignment
, args
->datatype
, ltbnoa
,
1133 if (ltnew
!= NULLAGBLOCK
&&
1134 (args
->len
> blen
|| ltdiff
< bdiff
)) {
1138 besti
= cnt_cur
->bc_ptrs
[0];
1142 * It didn't work. We COULD be in a case where
1143 * there's a good record somewhere, so try again.
1148 * Point at the best entry, and retrieve it again.
1150 cnt_cur
->bc_ptrs
[0] = besti
;
1151 if ((error
= xfs_alloc_get_rec(cnt_cur
, <bno
, <len
, &i
)))
1153 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1154 ASSERT(ltbno
+ ltlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1158 * We are allocating starting at bnew for blen blocks.
1161 ASSERT(bnew
>= ltbno
);
1162 ASSERT(bnew
+ blen
<= ltbno
+ ltlen
);
1164 * Set up a cursor for the by-bno tree.
1166 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
,
1167 args
->agbp
, args
->agno
, XFS_BTNUM_BNO
);
1169 * Fix up the btree entries.
1171 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
,
1172 ltlen
, bnew
, blen
, XFSA_FIXUP_CNT_OK
)))
1174 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1175 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1177 trace_xfs_alloc_near_first(args
);
1182 * Search in the by-bno tree to the left and to the right
1183 * simultaneously, until in each case we find a space big enough,
1184 * or run into the edge of the tree. When we run into the edge,
1185 * we deallocate that cursor.
1186 * If both searches succeed, we compare the two spaces and pick
1188 * With alignment, it's possible for both to fail; the upper
1189 * level algorithm that picks allocation groups for allocations
1190 * is not supposed to do this.
1193 * Allocate and initialize the cursor for the leftward search.
1195 bno_cur_lt
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1196 args
->agno
, XFS_BTNUM_BNO
);
1198 * Lookup <= bno to find the leftward search's starting point.
1200 if ((error
= xfs_alloc_lookup_le(bno_cur_lt
, args
->agbno
, args
->maxlen
, &i
)))
1204 * Didn't find anything; use this cursor for the rightward
1207 bno_cur_gt
= bno_cur_lt
;
1211 * Found something. Duplicate the cursor for the rightward search.
1213 else if ((error
= xfs_btree_dup_cursor(bno_cur_lt
, &bno_cur_gt
)))
1216 * Increment the cursor, so we will point at the entry just right
1217 * of the leftward entry if any, or to the leftmost entry.
1219 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1223 * It failed, there are no rightward entries.
1225 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_NOERROR
);
1229 * Loop going left with the leftward cursor, right with the
1230 * rightward cursor, until either both directions give up or
1231 * we find an entry at least as big as minlen.
1235 if ((error
= xfs_alloc_get_rec(bno_cur_lt
, <bno
, <len
, &i
)))
1237 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1238 busy
|= xfs_alloc_compute_aligned(args
, ltbno
, ltlen
,
1239 <bnoa
, <lena
, &busy_gen
);
1240 if (ltlena
>= args
->minlen
&& ltbnoa
>= args
->min_agbno
)
1242 if ((error
= xfs_btree_decrement(bno_cur_lt
, 0, &i
)))
1244 if (!i
|| ltbnoa
< args
->min_agbno
) {
1245 xfs_btree_del_cursor(bno_cur_lt
,
1251 if ((error
= xfs_alloc_get_rec(bno_cur_gt
, >bno
, >len
, &i
)))
1253 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1254 busy
|= xfs_alloc_compute_aligned(args
, gtbno
, gtlen
,
1255 >bnoa
, >lena
, &busy_gen
);
1256 if (gtlena
>= args
->minlen
&& gtbnoa
<= args
->max_agbno
)
1258 if ((error
= xfs_btree_increment(bno_cur_gt
, 0, &i
)))
1260 if (!i
|| gtbnoa
> args
->max_agbno
) {
1261 xfs_btree_del_cursor(bno_cur_gt
,
1266 } while (bno_cur_lt
|| bno_cur_gt
);
1269 * Got both cursors still active, need to find better entry.
1271 if (bno_cur_lt
&& bno_cur_gt
) {
1272 if (ltlena
>= args
->minlen
) {
1274 * Left side is good, look for a right side entry.
1276 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1277 xfs_alloc_fix_len(args
);
1278 ltdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1279 args
->alignment
, args
->datatype
, ltbnoa
,
1282 error
= xfs_alloc_find_best_extent(args
,
1283 &bno_cur_lt
, &bno_cur_gt
,
1284 ltdiff
, >bno
, >len
,
1286 0 /* search right */);
1288 ASSERT(gtlena
>= args
->minlen
);
1291 * Right side is good, look for a left side entry.
1293 args
->len
= XFS_EXTLEN_MIN(gtlena
, args
->maxlen
);
1294 xfs_alloc_fix_len(args
);
1295 gtdiff
= xfs_alloc_compute_diff(args
->agbno
, args
->len
,
1296 args
->alignment
, args
->datatype
, gtbnoa
,
1299 error
= xfs_alloc_find_best_extent(args
,
1300 &bno_cur_gt
, &bno_cur_lt
,
1301 gtdiff
, <bno
, <len
,
1303 1 /* search left */);
1311 * If we couldn't get anything, give up.
1313 if (bno_cur_lt
== NULL
&& bno_cur_gt
== NULL
) {
1314 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1317 trace_xfs_alloc_near_busy(args
);
1318 xfs_extent_busy_flush(args
->mp
, args
->pag
, busy_gen
);
1321 trace_xfs_alloc_size_neither(args
);
1322 args
->agbno
= NULLAGBLOCK
;
1327 * At this point we have selected a freespace entry, either to the
1328 * left or to the right. If it's on the right, copy all the
1329 * useful variables to the "left" set so we only have one
1330 * copy of this code.
1333 bno_cur_lt
= bno_cur_gt
;
1344 * Fix up the length and compute the useful address.
1346 args
->len
= XFS_EXTLEN_MIN(ltlena
, args
->maxlen
);
1347 xfs_alloc_fix_len(args
);
1349 (void)xfs_alloc_compute_diff(args
->agbno
, rlen
, args
->alignment
,
1350 args
->datatype
, ltbnoa
, ltlena
, <new
);
1351 ASSERT(ltnew
>= ltbno
);
1352 ASSERT(ltnew
+ rlen
<= ltbnoa
+ ltlena
);
1353 ASSERT(ltnew
+ rlen
<= be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
));
1354 ASSERT(ltnew
>= args
->min_agbno
&& ltnew
<= args
->max_agbno
);
1355 args
->agbno
= ltnew
;
1357 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur_lt
, ltbno
, ltlen
,
1358 ltnew
, rlen
, XFSA_FIXUP_BNO_OK
)))
1362 trace_xfs_alloc_near_greater(args
);
1364 trace_xfs_alloc_near_lesser(args
);
1366 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1367 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_NOERROR
);
1371 trace_xfs_alloc_near_error(args
);
1372 if (cnt_cur
!= NULL
)
1373 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1374 if (bno_cur_lt
!= NULL
)
1375 xfs_btree_del_cursor(bno_cur_lt
, XFS_BTREE_ERROR
);
1376 if (bno_cur_gt
!= NULL
)
1377 xfs_btree_del_cursor(bno_cur_gt
, XFS_BTREE_ERROR
);
1382 * Allocate a variable extent anywhere in the allocation group agno.
1383 * Extent's length (returned in len) will be between minlen and maxlen,
1384 * and of the form k * prod + mod unless there's nothing that large.
1385 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1387 STATIC
int /* error */
1388 xfs_alloc_ag_vextent_size(
1389 xfs_alloc_arg_t
*args
) /* allocation argument structure */
1391 xfs_btree_cur_t
*bno_cur
; /* cursor for bno btree */
1392 xfs_btree_cur_t
*cnt_cur
; /* cursor for cnt btree */
1393 int error
; /* error result */
1394 xfs_agblock_t fbno
; /* start of found freespace */
1395 xfs_extlen_t flen
; /* length of found freespace */
1396 int i
; /* temp status variable */
1397 xfs_agblock_t rbno
; /* returned block number */
1398 xfs_extlen_t rlen
; /* length of returned extent */
1404 * Allocate and initialize a cursor for the by-size btree.
1406 cnt_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1407 args
->agno
, XFS_BTNUM_CNT
);
1412 * Look for an entry >= maxlen+alignment-1 blocks.
1414 if ((error
= xfs_alloc_lookup_ge(cnt_cur
, 0,
1415 args
->maxlen
+ args
->alignment
- 1, &i
)))
1419 * If none then we have to settle for a smaller extent. In the case that
1420 * there are no large extents, this will return the last entry in the
1421 * tree unless the tree is empty. In the case that there are only busy
1422 * large extents, this will return the largest small extent unless there
1423 * are no smaller extents available.
1426 error
= xfs_alloc_ag_vextent_small(args
, cnt_cur
,
1430 if (i
== 0 || flen
== 0) {
1431 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1432 trace_xfs_alloc_size_noentry(args
);
1436 busy
= xfs_alloc_compute_aligned(args
, fbno
, flen
, &rbno
,
1440 * Search for a non-busy extent that is large enough.
1443 error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
, &i
);
1446 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1448 busy
= xfs_alloc_compute_aligned(args
, fbno
, flen
,
1449 &rbno
, &rlen
, &busy_gen
);
1451 if (rlen
>= args
->maxlen
)
1454 error
= xfs_btree_increment(cnt_cur
, 0, &i
);
1459 * Our only valid extents must have been busy.
1460 * Make it unbusy by forcing the log out and
1463 xfs_btree_del_cursor(cnt_cur
,
1465 trace_xfs_alloc_size_busy(args
);
1466 xfs_extent_busy_flush(args
->mp
,
1467 args
->pag
, busy_gen
);
1474 * In the first case above, we got the last entry in the
1475 * by-size btree. Now we check to see if the space hits maxlen
1476 * once aligned; if not, we search left for something better.
1477 * This can't happen in the second case above.
1479 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1480 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
== 0 ||
1481 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
), error0
);
1482 if (rlen
< args
->maxlen
) {
1483 xfs_agblock_t bestfbno
;
1484 xfs_extlen_t bestflen
;
1485 xfs_agblock_t bestrbno
;
1486 xfs_extlen_t bestrlen
;
1493 if ((error
= xfs_btree_decrement(cnt_cur
, 0, &i
)))
1497 if ((error
= xfs_alloc_get_rec(cnt_cur
, &fbno
, &flen
,
1500 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1501 if (flen
< bestrlen
)
1503 busy
= xfs_alloc_compute_aligned(args
, fbno
, flen
,
1504 &rbno
, &rlen
, &busy_gen
);
1505 rlen
= XFS_EXTLEN_MIN(args
->maxlen
, rlen
);
1506 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
== 0 ||
1507 (rlen
<= flen
&& rbno
+ rlen
<= fbno
+ flen
),
1509 if (rlen
> bestrlen
) {
1514 if (rlen
== args
->maxlen
)
1518 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, bestfbno
, bestflen
,
1521 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1527 args
->wasfromfl
= 0;
1529 * Fix up the length.
1532 if (rlen
< args
->minlen
) {
1534 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1535 trace_xfs_alloc_size_busy(args
);
1536 xfs_extent_busy_flush(args
->mp
, args
->pag
, busy_gen
);
1541 xfs_alloc_fix_len(args
);
1544 XFS_WANT_CORRUPTED_GOTO(args
->mp
, rlen
<= flen
, error0
);
1546 * Allocate and initialize a cursor for the by-block tree.
1548 bno_cur
= xfs_allocbt_init_cursor(args
->mp
, args
->tp
, args
->agbp
,
1549 args
->agno
, XFS_BTNUM_BNO
);
1550 if ((error
= xfs_alloc_fixup_trees(cnt_cur
, bno_cur
, fbno
, flen
,
1551 rbno
, rlen
, XFSA_FIXUP_CNT_OK
)))
1553 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1554 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1555 cnt_cur
= bno_cur
= NULL
;
1558 XFS_WANT_CORRUPTED_GOTO(args
->mp
,
1559 args
->agbno
+ args
->len
<=
1560 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1562 trace_xfs_alloc_size_done(args
);
1566 trace_xfs_alloc_size_error(args
);
1568 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1570 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1574 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1575 trace_xfs_alloc_size_nominleft(args
);
1576 args
->agbno
= NULLAGBLOCK
;
1581 * Deal with the case where only small freespaces remain.
1582 * Either return the contents of the last freespace record,
1583 * or allocate space from the freelist if there is nothing in the tree.
1585 STATIC
int /* error */
1586 xfs_alloc_ag_vextent_small(
1587 xfs_alloc_arg_t
*args
, /* allocation argument structure */
1588 xfs_btree_cur_t
*ccur
, /* by-size cursor */
1589 xfs_agblock_t
*fbnop
, /* result block number */
1590 xfs_extlen_t
*flenp
, /* result length */
1591 int *stat
) /* status: 0-freelist, 1-normal/none */
1593 struct xfs_owner_info oinfo
;
1599 if ((error
= xfs_btree_decrement(ccur
, 0, &i
)))
1602 if ((error
= xfs_alloc_get_rec(ccur
, &fbno
, &flen
, &i
)))
1604 XFS_WANT_CORRUPTED_GOTO(args
->mp
, i
== 1, error0
);
1607 * Nothing in the btree, try the freelist. Make sure
1608 * to respect minleft even when pulling from the
1611 else if (args
->minlen
== 1 && args
->alignment
== 1 &&
1612 args
->resv
!= XFS_AG_RESV_AGFL
&&
1613 (be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_flcount
)
1615 error
= xfs_alloc_get_freelist(args
->tp
, args
->agbp
, &fbno
, 0);
1618 if (fbno
!= NULLAGBLOCK
) {
1619 xfs_extent_busy_reuse(args
->mp
, args
->agno
, fbno
, 1,
1620 xfs_alloc_allow_busy_reuse(args
->datatype
));
1622 if (xfs_alloc_is_userdata(args
->datatype
)) {
1625 bp
= xfs_btree_get_bufs(args
->mp
, args
->tp
,
1626 args
->agno
, fbno
, 0);
1628 error
= -EFSCORRUPTED
;
1631 xfs_trans_binval(args
->tp
, bp
);
1635 XFS_WANT_CORRUPTED_GOTO(args
->mp
,
1636 args
->agbno
+ args
->len
<=
1637 be32_to_cpu(XFS_BUF_TO_AGF(args
->agbp
)->agf_length
),
1639 args
->wasfromfl
= 1;
1640 trace_xfs_alloc_small_freelist(args
);
1643 * If we're feeding an AGFL block to something that
1644 * doesn't live in the free space, we need to clear
1645 * out the OWN_AG rmap.
1647 xfs_rmap_ag_owner(&oinfo
, XFS_RMAP_OWN_AG
);
1648 error
= xfs_rmap_free(args
->tp
, args
->agbp
, args
->agno
,
1657 * Nothing in the freelist.
1663 * Can't allocate from the freelist for some reason.
1670 * Can't do the allocation, give up.
1672 if (flen
< args
->minlen
) {
1673 args
->agbno
= NULLAGBLOCK
;
1674 trace_xfs_alloc_small_notenough(args
);
1680 trace_xfs_alloc_small_done(args
);
1684 trace_xfs_alloc_small_error(args
);
1689 * Free the extent starting at agno/bno for length.
1695 xfs_agnumber_t agno
,
1698 struct xfs_owner_info
*oinfo
,
1699 enum xfs_ag_resv_type type
)
1701 xfs_btree_cur_t
*bno_cur
; /* cursor for by-block btree */
1702 xfs_btree_cur_t
*cnt_cur
; /* cursor for by-size btree */
1703 int error
; /* error return value */
1704 xfs_agblock_t gtbno
; /* start of right neighbor block */
1705 xfs_extlen_t gtlen
; /* length of right neighbor block */
1706 int haveleft
; /* have a left neighbor block */
1707 int haveright
; /* have a right neighbor block */
1708 int i
; /* temp, result code */
1709 xfs_agblock_t ltbno
; /* start of left neighbor block */
1710 xfs_extlen_t ltlen
; /* length of left neighbor block */
1711 xfs_mount_t
*mp
; /* mount point struct for filesystem */
1712 xfs_agblock_t nbno
; /* new starting block of freespace */
1713 xfs_extlen_t nlen
; /* new length of freespace */
1714 xfs_perag_t
*pag
; /* per allocation group data */
1716 bno_cur
= cnt_cur
= NULL
;
1719 if (!xfs_rmap_should_skip_owner_update(oinfo
)) {
1720 error
= xfs_rmap_free(tp
, agbp
, agno
, bno
, len
, oinfo
);
1726 * Allocate and initialize a cursor for the by-block btree.
1728 bno_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_BNO
);
1730 * Look for a neighboring block on the left (lower block numbers)
1731 * that is contiguous with this space.
1733 if ((error
= xfs_alloc_lookup_le(bno_cur
, bno
, len
, &haveleft
)))
1737 * There is a block to our left.
1739 if ((error
= xfs_alloc_get_rec(bno_cur
, <bno
, <len
, &i
)))
1741 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1743 * It's not contiguous, though.
1745 if (ltbno
+ ltlen
< bno
)
1749 * If this failure happens the request to free this
1750 * space was invalid, it's (partly) already free.
1753 XFS_WANT_CORRUPTED_GOTO(mp
,
1754 ltbno
+ ltlen
<= bno
, error0
);
1758 * Look for a neighboring block on the right (higher block numbers)
1759 * that is contiguous with this space.
1761 if ((error
= xfs_btree_increment(bno_cur
, 0, &haveright
)))
1765 * There is a block to our right.
1767 if ((error
= xfs_alloc_get_rec(bno_cur
, >bno
, >len
, &i
)))
1769 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1771 * It's not contiguous, though.
1773 if (bno
+ len
< gtbno
)
1777 * If this failure happens the request to free this
1778 * space was invalid, it's (partly) already free.
1781 XFS_WANT_CORRUPTED_GOTO(mp
, gtbno
>= bno
+ len
, error0
);
1785 * Now allocate and initialize a cursor for the by-size tree.
1787 cnt_cur
= xfs_allocbt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_CNT
);
1789 * Have both left and right contiguous neighbors.
1790 * Merge all three into a single free block.
1792 if (haveleft
&& haveright
) {
1794 * Delete the old by-size entry on the left.
1796 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1798 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1799 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1801 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1803 * Delete the old by-size entry on the right.
1805 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1807 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1808 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1810 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1812 * Delete the old by-block entry for the right block.
1814 if ((error
= xfs_btree_delete(bno_cur
, &i
)))
1816 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1818 * Move the by-block cursor back to the left neighbor.
1820 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1822 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1825 * Check that this is the right record: delete didn't
1826 * mangle the cursor.
1829 xfs_agblock_t xxbno
;
1832 if ((error
= xfs_alloc_get_rec(bno_cur
, &xxbno
, &xxlen
,
1835 XFS_WANT_CORRUPTED_GOTO(mp
,
1836 i
== 1 && xxbno
== ltbno
&& xxlen
== ltlen
,
1841 * Update remaining by-block entry to the new, joined block.
1844 nlen
= len
+ ltlen
+ gtlen
;
1845 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1849 * Have only a left contiguous neighbor.
1850 * Merge it together with the new freespace.
1852 else if (haveleft
) {
1854 * Delete the old by-size entry on the left.
1856 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, ltbno
, ltlen
, &i
)))
1858 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1859 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1861 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1863 * Back up the by-block cursor to the left neighbor, and
1864 * update its length.
1866 if ((error
= xfs_btree_decrement(bno_cur
, 0, &i
)))
1868 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1871 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1875 * Have only a right contiguous neighbor.
1876 * Merge it together with the new freespace.
1878 else if (haveright
) {
1880 * Delete the old by-size entry on the right.
1882 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, gtbno
, gtlen
, &i
)))
1884 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1885 if ((error
= xfs_btree_delete(cnt_cur
, &i
)))
1887 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1889 * Update the starting block and length of the right
1890 * neighbor in the by-block tree.
1894 if ((error
= xfs_alloc_update(bno_cur
, nbno
, nlen
)))
1898 * No contiguous neighbors.
1899 * Insert the new freespace into the by-block tree.
1904 if ((error
= xfs_btree_insert(bno_cur
, &i
)))
1906 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1908 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_NOERROR
);
1911 * In all cases we need to insert the new freespace in the by-size tree.
1913 if ((error
= xfs_alloc_lookup_eq(cnt_cur
, nbno
, nlen
, &i
)))
1915 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 0, error0
);
1916 if ((error
= xfs_btree_insert(cnt_cur
, &i
)))
1918 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1919 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_NOERROR
);
1923 * Update the freespace totals in the ag and superblock.
1925 pag
= xfs_perag_get(mp
, agno
);
1926 error
= xfs_alloc_update_counters(tp
, pag
, agbp
, len
);
1927 xfs_ag_resv_free_extent(pag
, type
, tp
, len
);
1932 XFS_STATS_INC(mp
, xs_freex
);
1933 XFS_STATS_ADD(mp
, xs_freeb
, len
);
1935 trace_xfs_free_extent(mp
, agno
, bno
, len
, type
, haveleft
, haveright
);
1940 trace_xfs_free_extent(mp
, agno
, bno
, len
, type
, -1, -1);
1942 xfs_btree_del_cursor(bno_cur
, XFS_BTREE_ERROR
);
1944 xfs_btree_del_cursor(cnt_cur
, XFS_BTREE_ERROR
);
1949 * Visible (exported) allocation/free functions.
1950 * Some of these are used just by xfs_alloc_btree.c and this file.
1954 * Compute and fill in value of m_ag_maxlevels.
1957 xfs_alloc_compute_maxlevels(
1958 xfs_mount_t
*mp
) /* file system mount structure */
1960 mp
->m_ag_maxlevels
= xfs_btree_compute_maxlevels(mp
->m_alloc_mnr
,
1961 (mp
->m_sb
.sb_agblocks
+ 1) / 2);
1965 * Find the length of the longest extent in an AG. The 'need' parameter
1966 * specifies how much space we're going to need for the AGFL and the
1967 * 'reserved' parameter tells us how many blocks in this AG are reserved for
1971 xfs_alloc_longest_free_extent(
1972 struct xfs_perag
*pag
,
1974 xfs_extlen_t reserved
)
1976 xfs_extlen_t delta
= 0;
1979 * If the AGFL needs a recharge, we'll have to subtract that from the
1982 if (need
> pag
->pagf_flcount
)
1983 delta
= need
- pag
->pagf_flcount
;
1986 * If we cannot maintain others' reservations with space from the
1987 * not-longest freesp extents, we'll have to subtract /that/ from
1988 * the longest extent too.
1990 if (pag
->pagf_freeblks
- pag
->pagf_longest
< reserved
)
1991 delta
+= reserved
- (pag
->pagf_freeblks
- pag
->pagf_longest
);
1994 * If the longest extent is long enough to satisfy all the
1995 * reservations and AGFL rules in place, we can return this extent.
1997 if (pag
->pagf_longest
> delta
)
1998 return pag
->pagf_longest
- delta
;
2000 /* Otherwise, let the caller try for 1 block if there's space. */
2001 return pag
->pagf_flcount
> 0 || pag
->pagf_longest
> 0;
2005 xfs_alloc_min_freelist(
2006 struct xfs_mount
*mp
,
2007 struct xfs_perag
*pag
)
2009 unsigned int min_free
;
2011 /* space needed by-bno freespace btree */
2012 min_free
= min_t(unsigned int, pag
->pagf_levels
[XFS_BTNUM_BNOi
] + 1,
2013 mp
->m_ag_maxlevels
);
2014 /* space needed by-size freespace btree */
2015 min_free
+= min_t(unsigned int, pag
->pagf_levels
[XFS_BTNUM_CNTi
] + 1,
2016 mp
->m_ag_maxlevels
);
2017 /* space needed reverse mapping used space btree */
2018 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
))
2019 min_free
+= min_t(unsigned int,
2020 pag
->pagf_levels
[XFS_BTNUM_RMAPi
] + 1,
2021 mp
->m_rmap_maxlevels
);
2027 * Check if the operation we are fixing up the freelist for should go ahead or
2028 * not. If we are freeing blocks, we always allow it, otherwise the allocation
2029 * is dependent on whether the size and shape of free space available will
2030 * permit the requested allocation to take place.
2033 xfs_alloc_space_available(
2034 struct xfs_alloc_arg
*args
,
2035 xfs_extlen_t min_free
,
2038 struct xfs_perag
*pag
= args
->pag
;
2039 xfs_extlen_t alloc_len
, longest
;
2040 xfs_extlen_t reservation
; /* blocks that are still reserved */
2043 if (flags
& XFS_ALLOC_FLAG_FREEING
)
2046 reservation
= xfs_ag_resv_needed(pag
, args
->resv
);
2048 /* do we have enough contiguous free space for the allocation? */
2049 alloc_len
= args
->minlen
+ (args
->alignment
- 1) + args
->minalignslop
;
2050 longest
= xfs_alloc_longest_free_extent(pag
, min_free
, reservation
);
2051 if (longest
< alloc_len
)
2054 /* do we have enough free space remaining for the allocation? */
2055 available
= (int)(pag
->pagf_freeblks
+ pag
->pagf_flcount
-
2056 reservation
- min_free
- args
->minleft
);
2057 if (available
< (int)max(args
->total
, alloc_len
))
2061 * Clamp maxlen to the amount of free space available for the actual
2062 * extent allocation.
2064 if (available
< (int)args
->maxlen
&& !(flags
& XFS_ALLOC_FLAG_CHECK
)) {
2065 args
->maxlen
= available
;
2066 ASSERT(args
->maxlen
> 0);
2067 ASSERT(args
->maxlen
>= args
->minlen
);
2074 xfs_free_agfl_block(
2075 struct xfs_trans
*tp
,
2076 xfs_agnumber_t agno
,
2077 xfs_agblock_t agbno
,
2078 struct xfs_buf
*agbp
,
2079 struct xfs_owner_info
*oinfo
)
2084 error
= xfs_free_ag_extent(tp
, agbp
, agno
, agbno
, 1, oinfo
,
2089 bp
= xfs_btree_get_bufs(tp
->t_mountp
, tp
, agno
, agbno
, 0);
2091 return -EFSCORRUPTED
;
2092 xfs_trans_binval(tp
, bp
);
2098 * Check the agfl fields of the agf for inconsistency or corruption. The purpose
2099 * is to detect an agfl header padding mismatch between current and early v5
2100 * kernels. This problem manifests as a 1-slot size difference between the
2101 * on-disk flcount and the active [first, last] range of a wrapped agfl. This
2102 * may also catch variants of agfl count corruption unrelated to padding. Either
2103 * way, we'll reset the agfl and warn the user.
2105 * Return true if a reset is required before the agfl can be used, false
2109 xfs_agfl_needs_reset(
2110 struct xfs_mount
*mp
,
2111 struct xfs_agf
*agf
)
2113 uint32_t f
= be32_to_cpu(agf
->agf_flfirst
);
2114 uint32_t l
= be32_to_cpu(agf
->agf_fllast
);
2115 uint32_t c
= be32_to_cpu(agf
->agf_flcount
);
2116 int agfl_size
= xfs_agfl_size(mp
);
2119 /* no agfl header on v4 supers */
2120 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2124 * The agf read verifier catches severe corruption of these fields.
2125 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
2126 * the verifier allows it.
2128 if (f
>= agfl_size
|| l
>= agfl_size
)
2134 * Check consistency between the on-disk count and the active range. An
2135 * agfl padding mismatch manifests as an inconsistent flcount.
2140 active
= agfl_size
- f
+ l
+ 1;
2148 * Reset the agfl to an empty state. Ignore/drop any existing blocks since the
2149 * agfl content cannot be trusted. Warn the user that a repair is required to
2150 * recover leaked blocks.
2152 * The purpose of this mechanism is to handle filesystems affected by the agfl
2153 * header padding mismatch problem. A reset keeps the filesystem online with a
2154 * relatively minor free space accounting inconsistency rather than suffer the
2155 * inevitable crash from use of an invalid agfl block.
2159 struct xfs_trans
*tp
,
2160 struct xfs_buf
*agbp
,
2161 struct xfs_perag
*pag
)
2163 struct xfs_mount
*mp
= tp
->t_mountp
;
2164 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(agbp
);
2166 ASSERT(pag
->pagf_agflreset
);
2167 trace_xfs_agfl_reset(mp
, agf
, 0, _RET_IP_
);
2170 "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. "
2171 "Please unmount and run xfs_repair.",
2172 pag
->pag_agno
, pag
->pagf_flcount
);
2174 agf
->agf_flfirst
= 0;
2175 agf
->agf_fllast
= cpu_to_be32(xfs_agfl_size(mp
) - 1);
2176 agf
->agf_flcount
= 0;
2177 xfs_alloc_log_agf(tp
, agbp
, XFS_AGF_FLFIRST
| XFS_AGF_FLLAST
|
2180 pag
->pagf_flcount
= 0;
2181 pag
->pagf_agflreset
= false;
2185 * Defer an AGFL block free. This is effectively equivalent to
2186 * xfs_bmap_add_free() with some special handling particular to AGFL blocks.
2188 * Deferring AGFL frees helps prevent log reservation overruns due to too many
2189 * allocation operations in a transaction. AGFL frees are prone to this problem
2190 * because for one they are always freed one at a time. Further, an immediate
2191 * AGFL block free can cause a btree join and require another block free before
2192 * the real allocation can proceed. Deferring the free disconnects freeing up
2193 * the AGFL slot from freeing the block.
2196 xfs_defer_agfl_block(
2197 struct xfs_trans
*tp
,
2198 xfs_agnumber_t agno
,
2199 xfs_fsblock_t agbno
,
2200 struct xfs_owner_info
*oinfo
)
2202 struct xfs_mount
*mp
= tp
->t_mountp
;
2203 struct xfs_extent_free_item
*new; /* new element */
2205 ASSERT(xfs_bmap_free_item_zone
!= NULL
);
2206 ASSERT(oinfo
!= NULL
);
2208 new = kmem_zone_alloc(xfs_bmap_free_item_zone
, KM_SLEEP
);
2209 new->xefi_startblock
= XFS_AGB_TO_FSB(mp
, agno
, agbno
);
2210 new->xefi_blockcount
= 1;
2211 new->xefi_oinfo
= *oinfo
;
2213 trace_xfs_agfl_free_defer(mp
, agno
, 0, agbno
, 1);
2215 xfs_defer_add(tp
, XFS_DEFER_OPS_TYPE_AGFL_FREE
, &new->xefi_list
);
2219 * Decide whether to use this allocation group for this allocation.
2220 * If so, fix up the btree freelist's size.
2223 xfs_alloc_fix_freelist(
2224 struct xfs_alloc_arg
*args
, /* allocation argument structure */
2225 int flags
) /* XFS_ALLOC_FLAG_... */
2227 struct xfs_mount
*mp
= args
->mp
;
2228 struct xfs_perag
*pag
= args
->pag
;
2229 struct xfs_trans
*tp
= args
->tp
;
2230 struct xfs_buf
*agbp
= NULL
;
2231 struct xfs_buf
*agflbp
= NULL
;
2232 struct xfs_alloc_arg targs
; /* local allocation arguments */
2233 xfs_agblock_t bno
; /* freelist block */
2234 xfs_extlen_t need
; /* total blocks needed in freelist */
2237 if (!pag
->pagf_init
) {
2238 error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
, &agbp
);
2241 if (!pag
->pagf_init
) {
2242 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
2243 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
2244 goto out_agbp_relse
;
2249 * If this is a metadata preferred pag and we are user data then try
2250 * somewhere else if we are not being asked to try harder at this
2253 if (pag
->pagf_metadata
&& xfs_alloc_is_userdata(args
->datatype
) &&
2254 (flags
& XFS_ALLOC_FLAG_TRYLOCK
)) {
2255 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
2256 goto out_agbp_relse
;
2259 need
= xfs_alloc_min_freelist(mp
, pag
);
2260 if (!xfs_alloc_space_available(args
, need
, flags
|
2261 XFS_ALLOC_FLAG_CHECK
))
2262 goto out_agbp_relse
;
2265 * Get the a.g. freespace buffer.
2266 * Can fail if we're not blocking on locks, and it's held.
2269 error
= xfs_alloc_read_agf(mp
, tp
, args
->agno
, flags
, &agbp
);
2273 ASSERT(flags
& XFS_ALLOC_FLAG_TRYLOCK
);
2274 ASSERT(!(flags
& XFS_ALLOC_FLAG_FREEING
));
2279 /* reset a padding mismatched agfl before final free space check */
2280 if (pag
->pagf_agflreset
)
2281 xfs_agfl_reset(tp
, agbp
, pag
);
2283 /* If there isn't enough total space or single-extent, reject it. */
2284 need
= xfs_alloc_min_freelist(mp
, pag
);
2285 if (!xfs_alloc_space_available(args
, need
, flags
))
2286 goto out_agbp_relse
;
2289 * Make the freelist shorter if it's too long.
2291 * Note that from this point onwards, we will always release the agf and
2292 * agfl buffers on error. This handles the case where we error out and
2293 * the buffers are clean or may not have been joined to the transaction
2294 * and hence need to be released manually. If they have been joined to
2295 * the transaction, then xfs_trans_brelse() will handle them
2296 * appropriately based on the recursion count and dirty state of the
2299 * XXX (dgc): When we have lots of free space, does this buy us
2300 * anything other than extra overhead when we need to put more blocks
2301 * back on the free list? Maybe we should only do this when space is
2302 * getting low or the AGFL is more than half full?
2304 * The NOSHRINK flag prevents the AGFL from being shrunk if it's too
2305 * big; the NORMAP flag prevents AGFL expand/shrink operations from
2306 * updating the rmapbt. Both flags are used in xfs_repair while we're
2307 * rebuilding the rmapbt, and neither are used by the kernel. They're
2308 * both required to ensure that rmaps are correctly recorded for the
2309 * regenerated AGFL, bnobt, and cntbt. See repair/phase5.c and
2310 * repair/rmap.c in xfsprogs for details.
2312 memset(&targs
, 0, sizeof(targs
));
2313 if (flags
& XFS_ALLOC_FLAG_NORMAP
)
2314 xfs_rmap_skip_owner_update(&targs
.oinfo
);
2316 xfs_rmap_ag_owner(&targs
.oinfo
, XFS_RMAP_OWN_AG
);
2317 while (!(flags
& XFS_ALLOC_FLAG_NOSHRINK
) && pag
->pagf_flcount
> need
) {
2318 error
= xfs_alloc_get_freelist(tp
, agbp
, &bno
, 0);
2320 goto out_agbp_relse
;
2322 /* defer agfl frees */
2323 xfs_defer_agfl_block(tp
, args
->agno
, bno
, &targs
.oinfo
);
2329 targs
.agno
= args
->agno
;
2330 targs
.alignment
= targs
.minlen
= targs
.prod
= 1;
2331 targs
.type
= XFS_ALLOCTYPE_THIS_AG
;
2333 error
= xfs_alloc_read_agfl(mp
, tp
, targs
.agno
, &agflbp
);
2335 goto out_agbp_relse
;
2337 /* Make the freelist longer if it's too short. */
2338 while (pag
->pagf_flcount
< need
) {
2340 targs
.maxlen
= need
- pag
->pagf_flcount
;
2341 targs
.resv
= XFS_AG_RESV_AGFL
;
2343 /* Allocate as many blocks as possible at once. */
2344 error
= xfs_alloc_ag_vextent(&targs
);
2346 goto out_agflbp_relse
;
2349 * Stop if we run out. Won't happen if callers are obeying
2350 * the restrictions correctly. Can happen for free calls
2351 * on a completely full ag.
2353 if (targs
.agbno
== NULLAGBLOCK
) {
2354 if (flags
& XFS_ALLOC_FLAG_FREEING
)
2356 goto out_agflbp_relse
;
2359 * Put each allocated block on the list.
2361 for (bno
= targs
.agbno
; bno
< targs
.agbno
+ targs
.len
; bno
++) {
2362 error
= xfs_alloc_put_freelist(tp
, agbp
,
2365 goto out_agflbp_relse
;
2368 xfs_trans_brelse(tp
, agflbp
);
2373 xfs_trans_brelse(tp
, agflbp
);
2376 xfs_trans_brelse(tp
, agbp
);
2383 * Get a block from the freelist.
2384 * Returns with the buffer for the block gotten.
2387 xfs_alloc_get_freelist(
2388 xfs_trans_t
*tp
, /* transaction pointer */
2389 xfs_buf_t
*agbp
, /* buffer containing the agf structure */
2390 xfs_agblock_t
*bnop
, /* block address retrieved from freelist */
2391 int btreeblk
) /* destination is a AGF btree */
2393 xfs_agf_t
*agf
; /* a.g. freespace structure */
2394 xfs_buf_t
*agflbp
;/* buffer for a.g. freelist structure */
2395 xfs_agblock_t bno
; /* block number returned */
2399 xfs_mount_t
*mp
= tp
->t_mountp
;
2400 xfs_perag_t
*pag
; /* per allocation group data */
2403 * Freelist is empty, give up.
2405 agf
= XFS_BUF_TO_AGF(agbp
);
2406 if (!agf
->agf_flcount
) {
2407 *bnop
= NULLAGBLOCK
;
2411 * Read the array of free blocks.
2413 error
= xfs_alloc_read_agfl(mp
, tp
, be32_to_cpu(agf
->agf_seqno
),
2420 * Get the block number and update the data structures.
2422 agfl_bno
= XFS_BUF_TO_AGFL_BNO(mp
, agflbp
);
2423 bno
= be32_to_cpu(agfl_bno
[be32_to_cpu(agf
->agf_flfirst
)]);
2424 be32_add_cpu(&agf
->agf_flfirst
, 1);
2425 xfs_trans_brelse(tp
, agflbp
);
2426 if (be32_to_cpu(agf
->agf_flfirst
) == xfs_agfl_size(mp
))
2427 agf
->agf_flfirst
= 0;
2429 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2430 ASSERT(!pag
->pagf_agflreset
);
2431 be32_add_cpu(&agf
->agf_flcount
, -1);
2432 xfs_trans_agflist_delta(tp
, -1);
2433 pag
->pagf_flcount
--;
2436 logflags
= XFS_AGF_FLFIRST
| XFS_AGF_FLCOUNT
;
2438 be32_add_cpu(&agf
->agf_btreeblks
, 1);
2439 pag
->pagf_btreeblks
++;
2440 logflags
|= XFS_AGF_BTREEBLKS
;
2443 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2450 * Log the given fields from the agf structure.
2454 xfs_trans_t
*tp
, /* transaction pointer */
2455 xfs_buf_t
*bp
, /* buffer for a.g. freelist header */
2456 int fields
) /* mask of fields to be logged (XFS_AGF_...) */
2458 int first
; /* first byte offset */
2459 int last
; /* last byte offset */
2460 static const short offsets
[] = {
2461 offsetof(xfs_agf_t
, agf_magicnum
),
2462 offsetof(xfs_agf_t
, agf_versionnum
),
2463 offsetof(xfs_agf_t
, agf_seqno
),
2464 offsetof(xfs_agf_t
, agf_length
),
2465 offsetof(xfs_agf_t
, agf_roots
[0]),
2466 offsetof(xfs_agf_t
, agf_levels
[0]),
2467 offsetof(xfs_agf_t
, agf_flfirst
),
2468 offsetof(xfs_agf_t
, agf_fllast
),
2469 offsetof(xfs_agf_t
, agf_flcount
),
2470 offsetof(xfs_agf_t
, agf_freeblks
),
2471 offsetof(xfs_agf_t
, agf_longest
),
2472 offsetof(xfs_agf_t
, agf_btreeblks
),
2473 offsetof(xfs_agf_t
, agf_uuid
),
2474 offsetof(xfs_agf_t
, agf_rmap_blocks
),
2475 offsetof(xfs_agf_t
, agf_refcount_blocks
),
2476 offsetof(xfs_agf_t
, agf_refcount_root
),
2477 offsetof(xfs_agf_t
, agf_refcount_level
),
2478 /* needed so that we don't log the whole rest of the structure: */
2479 offsetof(xfs_agf_t
, agf_spare64
),
2483 trace_xfs_agf(tp
->t_mountp
, XFS_BUF_TO_AGF(bp
), fields
, _RET_IP_
);
2485 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_AGF_BUF
);
2487 xfs_btree_offsets(fields
, offsets
, XFS_AGF_NUM_BITS
, &first
, &last
);
2488 xfs_trans_log_buf(tp
, bp
, (uint
)first
, (uint
)last
);
2492 * Interface for inode allocation to force the pag data to be initialized.
2495 xfs_alloc_pagf_init(
2496 xfs_mount_t
*mp
, /* file system mount structure */
2497 xfs_trans_t
*tp
, /* transaction pointer */
2498 xfs_agnumber_t agno
, /* allocation group number */
2499 int flags
) /* XFS_ALLOC_FLAGS_... */
2504 if ((error
= xfs_alloc_read_agf(mp
, tp
, agno
, flags
, &bp
)))
2507 xfs_trans_brelse(tp
, bp
);
2512 * Put the block on the freelist for the allocation group.
2515 xfs_alloc_put_freelist(
2516 xfs_trans_t
*tp
, /* transaction pointer */
2517 xfs_buf_t
*agbp
, /* buffer for a.g. freelist header */
2518 xfs_buf_t
*agflbp
,/* buffer for a.g. free block array */
2519 xfs_agblock_t bno
, /* block being freed */
2520 int btreeblk
) /* block came from a AGF btree */
2522 xfs_agf_t
*agf
; /* a.g. freespace structure */
2523 __be32
*blockp
;/* pointer to array entry */
2526 xfs_mount_t
*mp
; /* mount structure */
2527 xfs_perag_t
*pag
; /* per allocation group data */
2531 agf
= XFS_BUF_TO_AGF(agbp
);
2534 if (!agflbp
&& (error
= xfs_alloc_read_agfl(mp
, tp
,
2535 be32_to_cpu(agf
->agf_seqno
), &agflbp
)))
2537 be32_add_cpu(&agf
->agf_fllast
, 1);
2538 if (be32_to_cpu(agf
->agf_fllast
) == xfs_agfl_size(mp
))
2539 agf
->agf_fllast
= 0;
2541 pag
= xfs_perag_get(mp
, be32_to_cpu(agf
->agf_seqno
));
2542 ASSERT(!pag
->pagf_agflreset
);
2543 be32_add_cpu(&agf
->agf_flcount
, 1);
2544 xfs_trans_agflist_delta(tp
, 1);
2545 pag
->pagf_flcount
++;
2547 logflags
= XFS_AGF_FLLAST
| XFS_AGF_FLCOUNT
;
2549 be32_add_cpu(&agf
->agf_btreeblks
, -1);
2550 pag
->pagf_btreeblks
--;
2551 logflags
|= XFS_AGF_BTREEBLKS
;
2555 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2557 ASSERT(be32_to_cpu(agf
->agf_flcount
) <= xfs_agfl_size(mp
));
2559 agfl_bno
= XFS_BUF_TO_AGFL_BNO(mp
, agflbp
);
2560 blockp
= &agfl_bno
[be32_to_cpu(agf
->agf_fllast
)];
2561 *blockp
= cpu_to_be32(bno
);
2562 startoff
= (char *)blockp
- (char *)agflbp
->b_addr
;
2564 xfs_alloc_log_agf(tp
, agbp
, logflags
);
2566 xfs_trans_buf_set_type(tp
, agflbp
, XFS_BLFT_AGFL_BUF
);
2567 xfs_trans_log_buf(tp
, agflbp
, startoff
,
2568 startoff
+ sizeof(xfs_agblock_t
) - 1);
2572 static xfs_failaddr_t
2576 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2577 struct xfs_agf
*agf
= XFS_BUF_TO_AGF(bp
);
2579 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2580 if (!uuid_equal(&agf
->agf_uuid
, &mp
->m_sb
.sb_meta_uuid
))
2581 return __this_address
;
2582 if (!xfs_log_check_lsn(mp
,
2583 be64_to_cpu(XFS_BUF_TO_AGF(bp
)->agf_lsn
)))
2584 return __this_address
;
2587 if (!(agf
->agf_magicnum
== cpu_to_be32(XFS_AGF_MAGIC
) &&
2588 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf
->agf_versionnum
)) &&
2589 be32_to_cpu(agf
->agf_freeblks
) <= be32_to_cpu(agf
->agf_length
) &&
2590 be32_to_cpu(agf
->agf_flfirst
) < xfs_agfl_size(mp
) &&
2591 be32_to_cpu(agf
->agf_fllast
) < xfs_agfl_size(mp
) &&
2592 be32_to_cpu(agf
->agf_flcount
) <= xfs_agfl_size(mp
)))
2593 return __this_address
;
2595 if (be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNO
]) < 1 ||
2596 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNT
]) < 1 ||
2597 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNO
]) > XFS_BTREE_MAXLEVELS
||
2598 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNT
]) > XFS_BTREE_MAXLEVELS
)
2599 return __this_address
;
2601 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
) &&
2602 (be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_RMAP
]) < 1 ||
2603 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_RMAP
]) > XFS_BTREE_MAXLEVELS
))
2604 return __this_address
;
2607 * during growfs operations, the perag is not fully initialised,
2608 * so we can't use it for any useful checking. growfs ensures we can't
2609 * use it by using uncached buffers that don't have the perag attached
2610 * so we can detect and avoid this problem.
2612 if (bp
->b_pag
&& be32_to_cpu(agf
->agf_seqno
) != bp
->b_pag
->pag_agno
)
2613 return __this_address
;
2615 if (xfs_sb_version_haslazysbcount(&mp
->m_sb
) &&
2616 be32_to_cpu(agf
->agf_btreeblks
) > be32_to_cpu(agf
->agf_length
))
2617 return __this_address
;
2619 if (xfs_sb_version_hasreflink(&mp
->m_sb
) &&
2620 (be32_to_cpu(agf
->agf_refcount_level
) < 1 ||
2621 be32_to_cpu(agf
->agf_refcount_level
) > XFS_BTREE_MAXLEVELS
))
2622 return __this_address
;
2629 xfs_agf_read_verify(
2632 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2635 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2636 !xfs_buf_verify_cksum(bp
, XFS_AGF_CRC_OFF
))
2637 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
2639 fa
= xfs_agf_verify(bp
);
2640 if (XFS_TEST_ERROR(fa
, mp
, XFS_ERRTAG_ALLOC_READ_AGF
))
2641 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
2646 xfs_agf_write_verify(
2649 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2650 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
2653 fa
= xfs_agf_verify(bp
);
2655 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
2659 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2663 XFS_BUF_TO_AGF(bp
)->agf_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2665 xfs_buf_update_cksum(bp
, XFS_AGF_CRC_OFF
);
2668 const struct xfs_buf_ops xfs_agf_buf_ops
= {
2670 .verify_read
= xfs_agf_read_verify
,
2671 .verify_write
= xfs_agf_write_verify
,
2672 .verify_struct
= xfs_agf_verify
,
2676 * Read in the allocation group header (free/alloc section).
2680 struct xfs_mount
*mp
, /* mount point structure */
2681 struct xfs_trans
*tp
, /* transaction pointer */
2682 xfs_agnumber_t agno
, /* allocation group number */
2683 int flags
, /* XFS_BUF_ */
2684 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2688 trace_xfs_read_agf(mp
, agno
);
2690 ASSERT(agno
!= NULLAGNUMBER
);
2691 error
= xfs_trans_read_buf(
2692 mp
, tp
, mp
->m_ddev_targp
,
2693 XFS_AG_DADDR(mp
, agno
, XFS_AGF_DADDR(mp
)),
2694 XFS_FSS_TO_BB(mp
, 1), flags
, bpp
, &xfs_agf_buf_ops
);
2700 ASSERT(!(*bpp
)->b_error
);
2701 xfs_buf_set_ref(*bpp
, XFS_AGF_REF
);
2706 * Read in the allocation group header (free/alloc section).
2710 struct xfs_mount
*mp
, /* mount point structure */
2711 struct xfs_trans
*tp
, /* transaction pointer */
2712 xfs_agnumber_t agno
, /* allocation group number */
2713 int flags
, /* XFS_ALLOC_FLAG_... */
2714 struct xfs_buf
**bpp
) /* buffer for the ag freelist header */
2716 struct xfs_agf
*agf
; /* ag freelist header */
2717 struct xfs_perag
*pag
; /* per allocation group data */
2720 trace_xfs_alloc_read_agf(mp
, agno
);
2722 ASSERT(agno
!= NULLAGNUMBER
);
2723 error
= xfs_read_agf(mp
, tp
, agno
,
2724 (flags
& XFS_ALLOC_FLAG_TRYLOCK
) ? XBF_TRYLOCK
: 0,
2730 ASSERT(!(*bpp
)->b_error
);
2732 agf
= XFS_BUF_TO_AGF(*bpp
);
2733 pag
= xfs_perag_get(mp
, agno
);
2734 if (!pag
->pagf_init
) {
2735 pag
->pagf_freeblks
= be32_to_cpu(agf
->agf_freeblks
);
2736 pag
->pagf_btreeblks
= be32_to_cpu(agf
->agf_btreeblks
);
2737 pag
->pagf_flcount
= be32_to_cpu(agf
->agf_flcount
);
2738 pag
->pagf_longest
= be32_to_cpu(agf
->agf_longest
);
2739 pag
->pagf_levels
[XFS_BTNUM_BNOi
] =
2740 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]);
2741 pag
->pagf_levels
[XFS_BTNUM_CNTi
] =
2742 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]);
2743 pag
->pagf_levels
[XFS_BTNUM_RMAPi
] =
2744 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_RMAPi
]);
2745 pag
->pagf_refcount_level
= be32_to_cpu(agf
->agf_refcount_level
);
2747 pag
->pagf_agflreset
= xfs_agfl_needs_reset(mp
, agf
);
2750 else if (!XFS_FORCED_SHUTDOWN(mp
)) {
2751 ASSERT(pag
->pagf_freeblks
== be32_to_cpu(agf
->agf_freeblks
));
2752 ASSERT(pag
->pagf_btreeblks
== be32_to_cpu(agf
->agf_btreeblks
));
2753 ASSERT(pag
->pagf_flcount
== be32_to_cpu(agf
->agf_flcount
));
2754 ASSERT(pag
->pagf_longest
== be32_to_cpu(agf
->agf_longest
));
2755 ASSERT(pag
->pagf_levels
[XFS_BTNUM_BNOi
] ==
2756 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_BNOi
]));
2757 ASSERT(pag
->pagf_levels
[XFS_BTNUM_CNTi
] ==
2758 be32_to_cpu(agf
->agf_levels
[XFS_BTNUM_CNTi
]));
2766 * Allocate an extent (variable-size).
2767 * Depending on the allocation type, we either look in a single allocation
2768 * group or loop over the allocation groups to find the result.
2772 struct xfs_alloc_arg
*args
) /* allocation argument structure */
2774 xfs_agblock_t agsize
; /* allocation group size */
2776 int flags
; /* XFS_ALLOC_FLAG_... locking flags */
2777 struct xfs_mount
*mp
; /* mount structure pointer */
2778 xfs_agnumber_t sagno
; /* starting allocation group number */
2779 xfs_alloctype_t type
; /* input allocation type */
2781 xfs_agnumber_t rotorstep
= xfs_rotorstep
; /* inode32 agf stepper */
2784 type
= args
->otype
= args
->type
;
2785 args
->agbno
= NULLAGBLOCK
;
2787 * Just fix this up, for the case where the last a.g. is shorter
2788 * (or there's only one a.g.) and the caller couldn't easily figure
2789 * that out (xfs_bmap_alloc).
2791 agsize
= mp
->m_sb
.sb_agblocks
;
2792 if (args
->maxlen
> agsize
)
2793 args
->maxlen
= agsize
;
2794 if (args
->alignment
== 0)
2795 args
->alignment
= 1;
2796 ASSERT(XFS_FSB_TO_AGNO(mp
, args
->fsbno
) < mp
->m_sb
.sb_agcount
);
2797 ASSERT(XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) < agsize
);
2798 ASSERT(args
->minlen
<= args
->maxlen
);
2799 ASSERT(args
->minlen
<= agsize
);
2800 ASSERT(args
->mod
< args
->prod
);
2801 if (XFS_FSB_TO_AGNO(mp
, args
->fsbno
) >= mp
->m_sb
.sb_agcount
||
2802 XFS_FSB_TO_AGBNO(mp
, args
->fsbno
) >= agsize
||
2803 args
->minlen
> args
->maxlen
|| args
->minlen
> agsize
||
2804 args
->mod
>= args
->prod
) {
2805 args
->fsbno
= NULLFSBLOCK
;
2806 trace_xfs_alloc_vextent_badargs(args
);
2811 case XFS_ALLOCTYPE_THIS_AG
:
2812 case XFS_ALLOCTYPE_NEAR_BNO
:
2813 case XFS_ALLOCTYPE_THIS_BNO
:
2815 * These three force us into a single a.g.
2817 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2818 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2819 error
= xfs_alloc_fix_freelist(args
, 0);
2821 trace_xfs_alloc_vextent_nofix(args
);
2825 trace_xfs_alloc_vextent_noagbp(args
);
2828 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2829 if ((error
= xfs_alloc_ag_vextent(args
)))
2832 case XFS_ALLOCTYPE_START_BNO
:
2834 * Try near allocation first, then anywhere-in-ag after
2835 * the first a.g. fails.
2837 if ((args
->datatype
& XFS_ALLOC_INITIAL_USER_DATA
) &&
2838 (mp
->m_flags
& XFS_MOUNT_32BITINODES
)) {
2839 args
->fsbno
= XFS_AGB_TO_FSB(mp
,
2840 ((mp
->m_agfrotor
/ rotorstep
) %
2841 mp
->m_sb
.sb_agcount
), 0);
2844 args
->agbno
= XFS_FSB_TO_AGBNO(mp
, args
->fsbno
);
2845 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2847 case XFS_ALLOCTYPE_FIRST_AG
:
2849 * Rotate through the allocation groups looking for a winner.
2851 if (type
== XFS_ALLOCTYPE_FIRST_AG
) {
2853 * Start with allocation group given by bno.
2855 args
->agno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2856 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2861 * Start with the given allocation group.
2863 args
->agno
= sagno
= XFS_FSB_TO_AGNO(mp
, args
->fsbno
);
2864 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
2867 * Loop over allocation groups twice; first time with
2868 * trylock set, second time without.
2871 args
->pag
= xfs_perag_get(mp
, args
->agno
);
2872 error
= xfs_alloc_fix_freelist(args
, flags
);
2874 trace_xfs_alloc_vextent_nofix(args
);
2878 * If we get a buffer back then the allocation will fly.
2881 if ((error
= xfs_alloc_ag_vextent(args
)))
2886 trace_xfs_alloc_vextent_loopfailed(args
);
2889 * Didn't work, figure out the next iteration.
2891 if (args
->agno
== sagno
&&
2892 type
== XFS_ALLOCTYPE_START_BNO
)
2893 args
->type
= XFS_ALLOCTYPE_THIS_AG
;
2895 * For the first allocation, we can try any AG to get
2896 * space. However, if we already have allocated a
2897 * block, we don't want to try AGs whose number is below
2898 * sagno. Otherwise, we may end up with out-of-order
2899 * locking of AGF, which might cause deadlock.
2901 if (++(args
->agno
) == mp
->m_sb
.sb_agcount
) {
2902 if (args
->tp
->t_firstblock
!= NULLFSBLOCK
)
2908 * Reached the starting a.g., must either be done
2909 * or switch to non-trylock mode.
2911 if (args
->agno
== sagno
) {
2913 args
->agbno
= NULLAGBLOCK
;
2914 trace_xfs_alloc_vextent_allfailed(args
);
2919 if (type
== XFS_ALLOCTYPE_START_BNO
) {
2920 args
->agbno
= XFS_FSB_TO_AGBNO(mp
,
2922 args
->type
= XFS_ALLOCTYPE_NEAR_BNO
;
2925 xfs_perag_put(args
->pag
);
2928 if (args
->agno
== sagno
)
2929 mp
->m_agfrotor
= (mp
->m_agfrotor
+ 1) %
2930 (mp
->m_sb
.sb_agcount
* rotorstep
);
2932 mp
->m_agfrotor
= (args
->agno
* rotorstep
+ 1) %
2933 (mp
->m_sb
.sb_agcount
* rotorstep
);
2940 if (args
->agbno
== NULLAGBLOCK
)
2941 args
->fsbno
= NULLFSBLOCK
;
2943 args
->fsbno
= XFS_AGB_TO_FSB(mp
, args
->agno
, args
->agbno
);
2945 ASSERT(args
->len
>= args
->minlen
);
2946 ASSERT(args
->len
<= args
->maxlen
);
2947 ASSERT(args
->agbno
% args
->alignment
== 0);
2948 XFS_AG_CHECK_DADDR(mp
, XFS_FSB_TO_DADDR(mp
, args
->fsbno
),
2952 /* Zero the extent if we were asked to do so */
2953 if (args
->datatype
& XFS_ALLOC_USERDATA_ZERO
) {
2954 error
= xfs_zero_extent(args
->ip
, args
->fsbno
, args
->len
);
2960 xfs_perag_put(args
->pag
);
2963 xfs_perag_put(args
->pag
);
2967 /* Ensure that the freelist is at full capacity. */
2969 xfs_free_extent_fix_freelist(
2970 struct xfs_trans
*tp
,
2971 xfs_agnumber_t agno
,
2972 struct xfs_buf
**agbp
)
2974 struct xfs_alloc_arg args
;
2977 memset(&args
, 0, sizeof(struct xfs_alloc_arg
));
2979 args
.mp
= tp
->t_mountp
;
2983 * validate that the block number is legal - the enables us to detect
2984 * and handle a silent filesystem corruption rather than crashing.
2986 if (args
.agno
>= args
.mp
->m_sb
.sb_agcount
)
2987 return -EFSCORRUPTED
;
2989 args
.pag
= xfs_perag_get(args
.mp
, args
.agno
);
2992 error
= xfs_alloc_fix_freelist(&args
, XFS_ALLOC_FLAG_FREEING
);
2998 xfs_perag_put(args
.pag
);
3004 * Just break up the extent address and hand off to xfs_free_ag_extent
3005 * after fixing up the freelist.
3009 struct xfs_trans
*tp
, /* transaction pointer */
3010 xfs_fsblock_t bno
, /* starting block number of extent */
3011 xfs_extlen_t len
, /* length of extent */
3012 struct xfs_owner_info
*oinfo
, /* extent owner */
3013 enum xfs_ag_resv_type type
, /* block reservation type */
3016 struct xfs_mount
*mp
= tp
->t_mountp
;
3017 struct xfs_buf
*agbp
;
3018 xfs_agnumber_t agno
= XFS_FSB_TO_AGNO(mp
, bno
);
3019 xfs_agblock_t agbno
= XFS_FSB_TO_AGBNO(mp
, bno
);
3021 unsigned int busy_flags
= 0;
3024 ASSERT(type
!= XFS_AG_RESV_AGFL
);
3026 if (XFS_TEST_ERROR(false, mp
,
3027 XFS_ERRTAG_FREE_EXTENT
))
3030 error
= xfs_free_extent_fix_freelist(tp
, agno
, &agbp
);
3034 XFS_WANT_CORRUPTED_GOTO(mp
, agbno
< mp
->m_sb
.sb_agblocks
, err
);
3036 /* validate the extent size is legal now we have the agf locked */
3037 XFS_WANT_CORRUPTED_GOTO(mp
,
3038 agbno
+ len
<= be32_to_cpu(XFS_BUF_TO_AGF(agbp
)->agf_length
),
3041 error
= xfs_free_ag_extent(tp
, agbp
, agno
, agbno
, len
, oinfo
, type
);
3046 busy_flags
|= XFS_EXTENT_BUSY_SKIP_DISCARD
;
3047 xfs_extent_busy_insert(tp
, agno
, agbno
, len
, busy_flags
);
3051 xfs_trans_brelse(tp
, agbp
);
3055 struct xfs_alloc_query_range_info
{
3056 xfs_alloc_query_range_fn fn
;
3060 /* Format btree record and pass to our callback. */
3062 xfs_alloc_query_range_helper(
3063 struct xfs_btree_cur
*cur
,
3064 union xfs_btree_rec
*rec
,
3067 struct xfs_alloc_query_range_info
*query
= priv
;
3068 struct xfs_alloc_rec_incore irec
;
3070 irec
.ar_startblock
= be32_to_cpu(rec
->alloc
.ar_startblock
);
3071 irec
.ar_blockcount
= be32_to_cpu(rec
->alloc
.ar_blockcount
);
3072 return query
->fn(cur
, &irec
, query
->priv
);
3075 /* Find all free space within a given range of blocks. */
3077 xfs_alloc_query_range(
3078 struct xfs_btree_cur
*cur
,
3079 struct xfs_alloc_rec_incore
*low_rec
,
3080 struct xfs_alloc_rec_incore
*high_rec
,
3081 xfs_alloc_query_range_fn fn
,
3084 union xfs_btree_irec low_brec
;
3085 union xfs_btree_irec high_brec
;
3086 struct xfs_alloc_query_range_info query
;
3088 ASSERT(cur
->bc_btnum
== XFS_BTNUM_BNO
);
3089 low_brec
.a
= *low_rec
;
3090 high_brec
.a
= *high_rec
;
3093 return xfs_btree_query_range(cur
, &low_brec
, &high_brec
,
3094 xfs_alloc_query_range_helper
, &query
);
3097 /* Find all free space records. */
3099 xfs_alloc_query_all(
3100 struct xfs_btree_cur
*cur
,
3101 xfs_alloc_query_range_fn fn
,
3104 struct xfs_alloc_query_range_info query
;
3106 ASSERT(cur
->bc_btnum
== XFS_BTNUM_BNO
);
3109 return xfs_btree_query_all(cur
, xfs_alloc_query_range_helper
, &query
);
3112 /* Is there a record covering a given extent? */
3114 xfs_alloc_has_record(
3115 struct xfs_btree_cur
*cur
,
3120 union xfs_btree_irec low
;
3121 union xfs_btree_irec high
;
3123 memset(&low
, 0, sizeof(low
));
3124 low
.a
.ar_startblock
= bno
;
3125 memset(&high
, 0xFF, sizeof(high
));
3126 high
.a
.ar_startblock
= bno
+ len
- 1;
3128 return xfs_btree_has_record(cur
, &low
, &high
, exists
);
3132 * Walk all the blocks in the AGFL. The @walk_fn can return any negative
3133 * error code or XFS_BTREE_QUERY_RANGE_ABORT.
3137 struct xfs_mount
*mp
,
3138 struct xfs_agf
*agf
,
3139 struct xfs_buf
*agflbp
,
3140 xfs_agfl_walk_fn walk_fn
,
3147 agfl_bno
= XFS_BUF_TO_AGFL_BNO(mp
, agflbp
);
3148 i
= be32_to_cpu(agf
->agf_flfirst
);
3150 /* Nothing to walk in an empty AGFL. */
3151 if (agf
->agf_flcount
== cpu_to_be32(0))
3154 /* Otherwise, walk from first to last, wrapping as needed. */
3156 error
= walk_fn(mp
, be32_to_cpu(agfl_bno
[i
]), priv
);
3159 if (i
== be32_to_cpu(agf
->agf_fllast
))
3161 if (++i
== xfs_agfl_size(mp
))