2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "libxfs_priv.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_ialloc.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
34 #include "xfs_cksum.h"
35 #include "xfs_trans.h"
36 #include "xfs_trace.h"
41 * Allocation group level functions.
44 xfs_ialloc_cluster_alignment(
47 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
48 mp
->m_sb
.sb_inoalignmt
>= xfs_icluster_size_fsb(mp
))
49 return mp
->m_sb
.sb_inoalignmt
;
54 * Lookup a record by ino in the btree given by cur.
58 struct xfs_btree_cur
*cur
, /* btree cursor */
59 xfs_agino_t ino
, /* starting inode of chunk */
60 xfs_lookup_t dir
, /* <=, >=, == */
61 int *stat
) /* success/failure */
63 cur
->bc_rec
.i
.ir_startino
= ino
;
64 cur
->bc_rec
.i
.ir_holemask
= 0;
65 cur
->bc_rec
.i
.ir_count
= 0;
66 cur
->bc_rec
.i
.ir_freecount
= 0;
67 cur
->bc_rec
.i
.ir_free
= 0;
68 return xfs_btree_lookup(cur
, dir
, stat
);
72 * Update the record referred to by cur to the value given.
73 * This either works (return 0) or gets an EFSCORRUPTED error.
75 STATIC
int /* error */
77 struct xfs_btree_cur
*cur
, /* btree cursor */
78 xfs_inobt_rec_incore_t
*irec
) /* btree record */
80 union xfs_btree_rec rec
;
82 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
83 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
84 rec
.inobt
.ir_u
.sp
.ir_holemask
= cpu_to_be16(irec
->ir_holemask
);
85 rec
.inobt
.ir_u
.sp
.ir_count
= irec
->ir_count
;
86 rec
.inobt
.ir_u
.sp
.ir_freecount
= irec
->ir_freecount
;
88 /* ir_holemask/ir_count not supported on-disk */
89 rec
.inobt
.ir_u
.f
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
91 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
92 return xfs_btree_update(cur
, &rec
);
96 * Get the data from the pointed-to record.
100 struct xfs_btree_cur
*cur
, /* btree cursor */
101 xfs_inobt_rec_incore_t
*irec
, /* btree record */
102 int *stat
) /* output: success/failure */
104 union xfs_btree_rec
*rec
;
107 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
108 if (error
|| *stat
== 0)
111 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
112 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
113 irec
->ir_holemask
= be16_to_cpu(rec
->inobt
.ir_u
.sp
.ir_holemask
);
114 irec
->ir_count
= rec
->inobt
.ir_u
.sp
.ir_count
;
115 irec
->ir_freecount
= rec
->inobt
.ir_u
.sp
.ir_freecount
;
118 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
119 * values for full inode chunks.
121 irec
->ir_holemask
= XFS_INOBT_HOLEMASK_FULL
;
122 irec
->ir_count
= XFS_INODES_PER_CHUNK
;
124 be32_to_cpu(rec
->inobt
.ir_u
.f
.ir_freecount
);
126 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
132 * Insert a single inobt record. Cursor must already point to desired location.
135 xfs_inobt_insert_rec(
136 struct xfs_btree_cur
*cur
,
143 cur
->bc_rec
.i
.ir_holemask
= holemask
;
144 cur
->bc_rec
.i
.ir_count
= count
;
145 cur
->bc_rec
.i
.ir_freecount
= freecount
;
146 cur
->bc_rec
.i
.ir_free
= free
;
147 return xfs_btree_insert(cur
, stat
);
151 * Insert records describing a newly allocated inode chunk into the inobt.
155 struct xfs_mount
*mp
,
156 struct xfs_trans
*tp
,
157 struct xfs_buf
*agbp
,
162 struct xfs_btree_cur
*cur
;
163 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
164 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
169 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
171 for (thisino
= newino
;
172 thisino
< newino
+ newlen
;
173 thisino
+= XFS_INODES_PER_CHUNK
) {
174 error
= xfs_inobt_lookup(cur
, thisino
, XFS_LOOKUP_EQ
, &i
);
176 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
181 error
= xfs_inobt_insert_rec(cur
, XFS_INOBT_HOLEMASK_FULL
,
182 XFS_INODES_PER_CHUNK
,
183 XFS_INODES_PER_CHUNK
,
184 XFS_INOBT_ALL_FREE
, &i
);
186 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
192 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
198 * Verify that the number of free inodes in the AGI is correct.
202 xfs_check_agi_freecount(
203 struct xfs_btree_cur
*cur
,
206 if (cur
->bc_nlevels
== 1) {
207 xfs_inobt_rec_incore_t rec
;
212 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
217 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
222 freecount
+= rec
.ir_freecount
;
223 error
= xfs_btree_increment(cur
, 0, &i
);
229 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
230 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
235 #define xfs_check_agi_freecount(cur, agi) 0
239 * Initialise a new set of inodes. When called without a transaction context
240 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
241 * than logging them (which in a transaction context puts them into the AIL
242 * for writeback rather than the xfsbufd queue).
245 xfs_ialloc_inode_init(
246 struct xfs_mount
*mp
,
247 struct xfs_trans
*tp
,
248 struct list_head
*buffer_list
,
252 xfs_agblock_t length
,
255 struct xfs_buf
*fbuf
;
256 struct xfs_dinode
*free
;
257 int nbufs
, blks_per_cluster
, inodes_per_cluster
;
264 * Loop over the new block(s), filling in the inodes. For small block
265 * sizes, manipulate the inodes in buffers which are multiples of the
268 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
269 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
270 nbufs
= length
/ blks_per_cluster
;
273 * Figure out what version number to use in the inodes we create. If
274 * the superblock version has caught up to the one that supports the new
275 * inode format, then use the new inode version. Otherwise use the old
276 * version so that old kernels will continue to be able to use the file
279 * For v3 inodes, we also need to write the inode number into the inode,
280 * so calculate the first inode number of the chunk here as
281 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
282 * across multiple filesystem blocks (such as a cluster) and so cannot
283 * be used in the cluster buffer loop below.
285 * Further, because we are writing the inode directly into the buffer
286 * and calculating a CRC on the entire inode, we have ot log the entire
287 * inode so that the entire range the CRC covers is present in the log.
288 * That means for v3 inode we log the entire buffer rather than just the
291 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
293 ino
= XFS_AGINO_TO_INO(mp
, agno
,
294 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
297 * log the initialisation that is about to take place as an
298 * logical operation. This means the transaction does not
299 * need to log the physical changes to the inode buffers as log
300 * recovery will know what initialisation is actually needed.
301 * Hence we only need to log the buffers as "ordered" buffers so
302 * they track in the AIL as if they were physically logged.
305 xfs_icreate_log(tp
, agno
, agbno
, icount
,
306 mp
->m_sb
.sb_inodesize
, length
, gen
);
310 for (j
= 0; j
< nbufs
; j
++) {
314 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
315 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
316 mp
->m_bsize
* blks_per_cluster
,
321 /* Initialize the inode buffers and log them appropriately. */
322 fbuf
->b_ops
= &xfs_inode_buf_ops
;
323 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
324 for (i
= 0; i
< inodes_per_cluster
; i
++) {
325 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
326 uint isize
= xfs_dinode_size(version
);
328 free
= xfs_make_iptr(mp
, fbuf
, i
);
329 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
330 free
->di_version
= version
;
331 free
->di_gen
= cpu_to_be32(gen
);
332 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
335 free
->di_ino
= cpu_to_be64(ino
);
337 uuid_copy(&free
->di_uuid
,
338 &mp
->m_sb
.sb_meta_uuid
);
339 xfs_dinode_calc_crc(mp
, free
);
341 /* just log the inode core */
342 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
343 ioffset
+ isize
- 1);
349 * Mark the buffer as an inode allocation buffer so it
350 * sticks in AIL at the point of this allocation
351 * transaction. This ensures the they are on disk before
352 * the tail of the log can be moved past this
353 * transaction (i.e. by preventing relogging from moving
354 * it forward in the log).
356 xfs_trans_inode_alloc_buf(tp
, fbuf
);
359 * Mark the buffer as ordered so that they are
360 * not physically logged in the transaction but
361 * still tracked in the AIL as part of the
362 * transaction and pin the log appropriately.
364 xfs_trans_ordered_buf(tp
, fbuf
);
365 xfs_trans_log_buf(tp
, fbuf
, 0,
366 BBTOB(fbuf
->b_length
) - 1);
369 fbuf
->b_flags
|= XBF_DONE
;
370 xfs_buf_delwri_queue(fbuf
, buffer_list
);
378 * Align startino and allocmask for a recently allocated sparse chunk such that
379 * they are fit for insertion (or merge) into the on-disk inode btrees.
383 * When enabled, sparse inode support increases the inode alignment from cluster
384 * size to inode chunk size. This means that the minimum range between two
385 * non-adjacent inode records in the inobt is large enough for a full inode
386 * record. This allows for cluster sized, cluster aligned block allocation
387 * without need to worry about whether the resulting inode record overlaps with
388 * another record in the tree. Without this basic rule, we would have to deal
389 * with the consequences of overlap by potentially undoing recent allocations in
390 * the inode allocation codepath.
392 * Because of this alignment rule (which is enforced on mount), there are two
393 * inobt possibilities for newly allocated sparse chunks. One is that the
394 * aligned inode record for the chunk covers a range of inodes not already
395 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
396 * other is that a record already exists at the aligned startino that considers
397 * the newly allocated range as sparse. In the latter case, record content is
398 * merged in hope that sparse inode chunks fill to full chunks over time.
401 xfs_align_sparse_ino(
402 struct xfs_mount
*mp
,
403 xfs_agino_t
*startino
,
410 agbno
= XFS_AGINO_TO_AGBNO(mp
, *startino
);
411 mod
= agbno
% mp
->m_sb
.sb_inoalignmt
;
415 /* calculate the inode offset and align startino */
416 offset
= mod
<< mp
->m_sb
.sb_inopblog
;
420 * Since startino has been aligned down, left shift allocmask such that
421 * it continues to represent the same physical inodes relative to the
424 *allocmask
<<= offset
/ XFS_INODES_PER_HOLEMASK_BIT
;
428 * Determine whether the source inode record can merge into the target. Both
429 * records must be sparse, the inode ranges must match and there must be no
430 * allocation overlap between the records.
433 __xfs_inobt_can_merge(
434 struct xfs_inobt_rec_incore
*trec
, /* tgt record */
435 struct xfs_inobt_rec_incore
*srec
) /* src record */
440 /* records must cover the same inode range */
441 if (trec
->ir_startino
!= srec
->ir_startino
)
444 /* both records must be sparse */
445 if (!xfs_inobt_issparse(trec
->ir_holemask
) ||
446 !xfs_inobt_issparse(srec
->ir_holemask
))
449 /* both records must track some inodes */
450 if (!trec
->ir_count
|| !srec
->ir_count
)
453 /* can't exceed capacity of a full record */
454 if (trec
->ir_count
+ srec
->ir_count
> XFS_INODES_PER_CHUNK
)
457 /* verify there is no allocation overlap */
458 talloc
= xfs_inobt_irec_to_allocmask(trec
);
459 salloc
= xfs_inobt_irec_to_allocmask(srec
);
467 * Merge the source inode record into the target. The caller must call
468 * __xfs_inobt_can_merge() to ensure the merge is valid.
471 __xfs_inobt_rec_merge(
472 struct xfs_inobt_rec_incore
*trec
, /* target */
473 struct xfs_inobt_rec_incore
*srec
) /* src */
475 ASSERT(trec
->ir_startino
== srec
->ir_startino
);
477 /* combine the counts */
478 trec
->ir_count
+= srec
->ir_count
;
479 trec
->ir_freecount
+= srec
->ir_freecount
;
482 * Merge the holemask and free mask. For both fields, 0 bits refer to
483 * allocated inodes. We combine the allocated ranges with bitwise AND.
485 trec
->ir_holemask
&= srec
->ir_holemask
;
486 trec
->ir_free
&= srec
->ir_free
;
490 * Insert a new sparse inode chunk into the associated inode btree. The inode
491 * record for the sparse chunk is pre-aligned to a startino that should match
492 * any pre-existing sparse inode record in the tree. This allows sparse chunks
495 * This function supports two modes of handling preexisting records depending on
496 * the merge flag. If merge is true, the provided record is merged with the
497 * existing record and updated in place. The merged record is returned in nrec.
498 * If merge is false, an existing record is replaced with the provided record.
499 * If no preexisting record exists, the provided record is always inserted.
501 * It is considered corruption if a merge is requested and not possible. Given
502 * the sparse inode alignment constraints, this should never happen.
505 xfs_inobt_insert_sprec(
506 struct xfs_mount
*mp
,
507 struct xfs_trans
*tp
,
508 struct xfs_buf
*agbp
,
510 struct xfs_inobt_rec_incore
*nrec
, /* in/out: new/merged rec. */
511 bool merge
) /* merge or replace */
513 struct xfs_btree_cur
*cur
;
514 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
515 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
518 struct xfs_inobt_rec_incore rec
;
520 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
522 /* the new record is pre-aligned so we know where to look */
523 error
= xfs_inobt_lookup(cur
, nrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
526 /* if nothing there, insert a new record and return */
528 error
= xfs_inobt_insert_rec(cur
, nrec
->ir_holemask
,
529 nrec
->ir_count
, nrec
->ir_freecount
,
533 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
539 * A record exists at this startino. Merge or replace the record
540 * depending on what we've been asked to do.
543 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
546 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
547 XFS_WANT_CORRUPTED_GOTO(mp
,
548 rec
.ir_startino
== nrec
->ir_startino
,
552 * This should never fail. If we have coexisting records that
553 * cannot merge, something is seriously wrong.
555 XFS_WANT_CORRUPTED_GOTO(mp
, __xfs_inobt_can_merge(nrec
, &rec
),
558 trace_xfs_irec_merge_pre(mp
, agno
, rec
.ir_startino
,
559 rec
.ir_holemask
, nrec
->ir_startino
,
562 /* merge to nrec to output the updated record */
563 __xfs_inobt_rec_merge(nrec
, &rec
);
565 trace_xfs_irec_merge_post(mp
, agno
, nrec
->ir_startino
,
568 error
= xfs_inobt_rec_check_count(mp
, nrec
);
573 error
= xfs_inobt_update(cur
, nrec
);
578 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
581 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
586 * Allocate new inodes in the allocation group specified by agbp.
587 * Return 0 for success, else error code.
589 STATIC
int /* error code or 0 */
591 xfs_trans_t
*tp
, /* transaction pointer */
592 xfs_buf_t
*agbp
, /* alloc group buffer */
595 xfs_agi_t
*agi
; /* allocation group header */
596 xfs_alloc_arg_t args
; /* allocation argument structure */
599 xfs_agino_t newino
; /* new first inode's number */
600 xfs_agino_t newlen
; /* new number of inodes */
601 int isaligned
= 0; /* inode allocation at stripe unit */
603 uint16_t allocmask
= (uint16_t) -1; /* init. to full chunk */
604 struct xfs_inobt_rec_incore rec
;
605 struct xfs_perag
*pag
;
608 memset(&args
, 0, sizeof(args
));
610 args
.mp
= tp
->t_mountp
;
611 args
.fsbno
= NULLFSBLOCK
;
612 xfs_rmap_ag_owner(&args
.oinfo
, XFS_RMAP_OWN_INODES
);
615 /* randomly do sparse inode allocations */
616 if (xfs_sb_version_hassparseinodes(&tp
->t_mountp
->m_sb
) &&
617 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
)
618 do_sparse
= prandom_u32() & 1;
622 * Locking will ensure that we don't have two callers in here
625 newlen
= args
.mp
->m_ialloc_inos
;
626 if (args
.mp
->m_maxicount
&&
627 percpu_counter_read_positive(&args
.mp
->m_icount
) + newlen
>
628 args
.mp
->m_maxicount
)
630 args
.minlen
= args
.maxlen
= args
.mp
->m_ialloc_blks
;
632 * First try to allocate inodes contiguous with the last-allocated
633 * chunk of inodes. If the filesystem is striped, this will fill
634 * an entire stripe unit with inodes.
636 agi
= XFS_BUF_TO_AGI(agbp
);
637 newino
= be32_to_cpu(agi
->agi_newino
);
638 agno
= be32_to_cpu(agi
->agi_seqno
);
639 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
640 args
.mp
->m_ialloc_blks
;
643 if (likely(newino
!= NULLAGINO
&&
644 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
645 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
646 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
650 * We need to take into account alignment here to ensure that
651 * we don't modify the free list if we fail to have an exact
652 * block. If we don't have an exact match, and every oher
653 * attempt allocation attempt fails, we'll end up cancelling
654 * a dirty transaction and shutting down.
656 * For an exact allocation, alignment must be 1,
657 * however we need to take cluster alignment into account when
658 * fixing up the freelist. Use the minalignslop field to
659 * indicate that extra blocks might be required for alignment,
660 * but not to use them in the actual exact allocation.
663 args
.minalignslop
= xfs_ialloc_cluster_alignment(args
.mp
) - 1;
665 /* Allow space for the inode btree to split. */
666 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
667 if ((error
= xfs_alloc_vextent(&args
)))
671 * This request might have dirtied the transaction if the AG can
672 * satisfy the request, but the exact block was not available.
673 * If the allocation did fail, subsequent requests will relax
674 * the exact agbno requirement and increase the alignment
675 * instead. It is critical that the total size of the request
676 * (len + alignment + slop) does not increase from this point
677 * on, so reset minalignslop to ensure it is not included in
678 * subsequent requests.
680 args
.minalignslop
= 0;
683 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
685 * Set the alignment for the allocation.
686 * If stripe alignment is turned on then align at stripe unit
688 * If the cluster size is smaller than a filesystem block
689 * then we're doing I/O for inodes in filesystem block size
690 * pieces, so don't need alignment anyway.
693 if (args
.mp
->m_sinoalign
) {
694 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
695 args
.alignment
= args
.mp
->m_dalign
;
698 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
700 * Need to figure out where to allocate the inode blocks.
701 * Ideally they should be spaced out through the a.g.
702 * For now, just allocate blocks up front.
704 args
.agbno
= be32_to_cpu(agi
->agi_root
);
705 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
707 * Allocate a fixed-size extent of inodes.
709 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
712 * Allow space for the inode btree to split.
714 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
715 if ((error
= xfs_alloc_vextent(&args
)))
720 * If stripe alignment is turned on, then try again with cluster
723 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
724 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
725 args
.agbno
= be32_to_cpu(agi
->agi_root
);
726 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
727 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
728 if ((error
= xfs_alloc_vextent(&args
)))
733 * Finally, try a sparse allocation if the filesystem supports it and
734 * the sparse allocation length is smaller than a full chunk.
736 if (xfs_sb_version_hassparseinodes(&args
.mp
->m_sb
) &&
737 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
&&
738 args
.fsbno
== NULLFSBLOCK
) {
740 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
741 args
.agbno
= be32_to_cpu(agi
->agi_root
);
742 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
743 args
.alignment
= args
.mp
->m_sb
.sb_spino_align
;
746 args
.minlen
= args
.mp
->m_ialloc_min_blks
;
747 args
.maxlen
= args
.minlen
;
750 * The inode record will be aligned to full chunk size. We must
751 * prevent sparse allocation from AG boundaries that result in
752 * invalid inode records, such as records that start at agbno 0
753 * or extend beyond the AG.
755 * Set min agbno to the first aligned, non-zero agbno and max to
756 * the last aligned agbno that is at least one full chunk from
759 args
.min_agbno
= args
.mp
->m_sb
.sb_inoalignmt
;
760 args
.max_agbno
= round_down(args
.mp
->m_sb
.sb_agblocks
,
761 args
.mp
->m_sb
.sb_inoalignmt
) -
762 args
.mp
->m_ialloc_blks
;
764 error
= xfs_alloc_vextent(&args
);
768 newlen
= args
.len
<< args
.mp
->m_sb
.sb_inopblog
;
769 ASSERT(newlen
<= XFS_INODES_PER_CHUNK
);
770 allocmask
= (1 << (newlen
/ XFS_INODES_PER_HOLEMASK_BIT
)) - 1;
773 if (args
.fsbno
== NULLFSBLOCK
) {
777 ASSERT(args
.len
== args
.minlen
);
780 * Stamp and write the inode buffers.
782 * Seed the new inode cluster with a random generation number. This
783 * prevents short-term reuse of generation numbers if a chunk is
784 * freed and then immediately reallocated. We use random numbers
785 * rather than a linear progression to prevent the next generation
786 * number from being easily guessable.
788 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, newlen
, agno
,
789 args
.agbno
, args
.len
, prandom_u32());
794 * Convert the results.
796 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
798 if (xfs_inobt_issparse(~allocmask
)) {
800 * We've allocated a sparse chunk. Align the startino and mask.
802 xfs_align_sparse_ino(args
.mp
, &newino
, &allocmask
);
804 rec
.ir_startino
= newino
;
805 rec
.ir_holemask
= ~allocmask
;
806 rec
.ir_count
= newlen
;
807 rec
.ir_freecount
= newlen
;
808 rec
.ir_free
= XFS_INOBT_ALL_FREE
;
811 * Insert the sparse record into the inobt and allow for a merge
812 * if necessary. If a merge does occur, rec is updated to the
815 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
, XFS_BTNUM_INO
,
817 if (error
== -EFSCORRUPTED
) {
819 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
820 XFS_AGINO_TO_INO(args
.mp
, agno
,
822 rec
.ir_holemask
, rec
.ir_count
);
823 xfs_force_shutdown(args
.mp
, SHUTDOWN_CORRUPT_INCORE
);
829 * We can't merge the part we've just allocated as for the inobt
830 * due to finobt semantics. The original record may or may not
831 * exist independent of whether physical inodes exist in this
834 * We must update the finobt record based on the inobt record.
835 * rec contains the fully merged and up to date inobt record
836 * from the previous call. Set merge false to replace any
837 * existing record with this one.
839 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
840 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
,
841 XFS_BTNUM_FINO
, &rec
,
847 /* full chunk - insert new records to both btrees */
848 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
853 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
854 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
,
855 newlen
, XFS_BTNUM_FINO
);
862 * Update AGI counts and newino.
864 be32_add_cpu(&agi
->agi_count
, newlen
);
865 be32_add_cpu(&agi
->agi_freecount
, newlen
);
866 pag
= xfs_perag_get(args
.mp
, agno
);
867 pag
->pagi_freecount
+= newlen
;
869 agi
->agi_newino
= cpu_to_be32(newino
);
872 * Log allocation group header fields
874 xfs_ialloc_log_agi(tp
, agbp
,
875 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
877 * Modify/log superblock values for inode count and inode free count.
879 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
880 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
885 STATIC xfs_agnumber_t
891 spin_lock(&mp
->m_agirotor_lock
);
892 agno
= mp
->m_agirotor
;
893 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
895 spin_unlock(&mp
->m_agirotor_lock
);
901 * Select an allocation group to look for a free inode in, based on the parent
902 * inode and the mode. Return the allocation group buffer.
904 STATIC xfs_agnumber_t
905 xfs_ialloc_ag_select(
906 xfs_trans_t
*tp
, /* transaction pointer */
907 xfs_ino_t parent
, /* parent directory inode number */
908 umode_t mode
, /* bits set to indicate file type */
909 int okalloc
) /* ok to allocate more space */
911 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
912 xfs_agnumber_t agno
; /* current ag number */
913 int flags
; /* alloc buffer locking flags */
914 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
915 xfs_extlen_t longest
= 0; /* longest extent available */
916 xfs_mount_t
*mp
; /* mount point structure */
917 int needspace
; /* file mode implies space allocated */
918 xfs_perag_t
*pag
; /* per allocation group data */
919 xfs_agnumber_t pagno
; /* parent (starting) ag number */
923 * Files of these types need at least one block if length > 0
924 * (and they won't fit in the inode, but that's hard to figure out).
926 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
928 agcount
= mp
->m_maxagi
;
930 pagno
= xfs_ialloc_next_ag(mp
);
932 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
933 if (pagno
>= agcount
)
937 ASSERT(pagno
< agcount
);
940 * Loop through allocation groups, looking for one with a little
941 * free space in it. Note we don't look for free inodes, exactly.
942 * Instead, we include whether there is a need to allocate inodes
943 * to mean that blocks must be allocated for them,
944 * if none are currently free.
947 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
949 pag
= xfs_perag_get(mp
, agno
);
950 if (!pag
->pagi_inodeok
) {
951 xfs_ialloc_next_ag(mp
);
955 if (!pag
->pagi_init
) {
956 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
961 if (pag
->pagi_freecount
) {
969 if (!pag
->pagf_init
) {
970 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
976 * Check that there is enough free space for the file plus a
977 * chunk of inodes if we need to allocate some. If this is the
978 * first pass across the AGs, take into account the potential
979 * space needed for alignment of inode chunks when checking the
980 * longest contiguous free space in the AG - this prevents us
981 * from getting ENOSPC because we have free space larger than
982 * m_ialloc_blks but alignment constraints prevent us from using
985 * If we can't find an AG with space for full alignment slack to
986 * be taken into account, we must be near ENOSPC in all AGs.
987 * Hence we don't include alignment for the second pass and so
988 * if we fail allocation due to alignment issues then it is most
989 * likely a real ENOSPC condition.
991 ineed
= mp
->m_ialloc_min_blks
;
992 if (flags
&& ineed
> 1)
993 ineed
+= xfs_ialloc_cluster_alignment(mp
);
994 longest
= pag
->pagf_longest
;
996 longest
= pag
->pagf_flcount
> 0;
998 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
1006 * No point in iterating over the rest, if we're shutting
1009 if (XFS_FORCED_SHUTDOWN(mp
))
1010 return NULLAGNUMBER
;
1012 if (agno
>= agcount
)
1014 if (agno
== pagno
) {
1016 return NULLAGNUMBER
;
1023 * Try to retrieve the next record to the left/right from the current one.
1026 xfs_ialloc_next_rec(
1027 struct xfs_btree_cur
*cur
,
1028 xfs_inobt_rec_incore_t
*rec
,
1036 error
= xfs_btree_decrement(cur
, 0, &i
);
1038 error
= xfs_btree_increment(cur
, 0, &i
);
1044 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1047 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1055 struct xfs_btree_cur
*cur
,
1057 xfs_inobt_rec_incore_t
*rec
,
1063 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
1068 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1071 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1078 * Return the offset of the first free inode in the record. If the inode chunk
1079 * is sparsely allocated, we convert the record holemask to inode granularity
1080 * and mask off the unallocated regions from the inode free mask.
1083 xfs_inobt_first_free_inode(
1084 struct xfs_inobt_rec_incore
*rec
)
1086 xfs_inofree_t realfree
;
1088 /* if there are no holes, return the first available offset */
1089 if (!xfs_inobt_issparse(rec
->ir_holemask
))
1090 return xfs_lowbit64(rec
->ir_free
);
1092 realfree
= xfs_inobt_irec_to_allocmask(rec
);
1093 realfree
&= rec
->ir_free
;
1095 return xfs_lowbit64(realfree
);
1099 * Allocate an inode using the inobt-only algorithm.
1102 xfs_dialloc_ag_inobt(
1103 struct xfs_trans
*tp
,
1104 struct xfs_buf
*agbp
,
1108 struct xfs_mount
*mp
= tp
->t_mountp
;
1109 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1110 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1111 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1112 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1113 struct xfs_perag
*pag
;
1114 struct xfs_btree_cur
*cur
, *tcur
;
1115 struct xfs_inobt_rec_incore rec
, trec
;
1121 pag
= xfs_perag_get(mp
, agno
);
1123 ASSERT(pag
->pagi_init
);
1124 ASSERT(pag
->pagi_inodeok
);
1125 ASSERT(pag
->pagi_freecount
> 0);
1128 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1130 * If pagino is 0 (this is the root inode allocation) use newino.
1131 * This must work because we've just allocated some.
1134 pagino
= be32_to_cpu(agi
->agi_newino
);
1136 error
= xfs_check_agi_freecount(cur
, agi
);
1141 * If in the same AG as the parent, try to get near the parent.
1143 if (pagno
== agno
) {
1144 int doneleft
; /* done, to the left */
1145 int doneright
; /* done, to the right */
1146 int searchdistance
= 10;
1148 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
1151 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1153 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1156 XFS_WANT_CORRUPTED_GOTO(mp
, j
== 1, error0
);
1158 if (rec
.ir_freecount
> 0) {
1160 * Found a free inode in the same chunk
1161 * as the parent, done.
1168 * In the same AG as parent, but parent's chunk is full.
1171 /* duplicate the cursor, search left & right simultaneously */
1172 error
= xfs_btree_dup_cursor(cur
, &tcur
);
1177 * Skip to last blocks looked up if same parent inode.
1179 if (pagino
!= NULLAGINO
&&
1180 pag
->pagl_pagino
== pagino
&&
1181 pag
->pagl_leftrec
!= NULLAGINO
&&
1182 pag
->pagl_rightrec
!= NULLAGINO
) {
1183 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
1188 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
1193 /* search left with tcur, back up 1 record */
1194 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
1198 /* search right with cur, go forward 1 record. */
1199 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
1205 * Loop until we find an inode chunk with a free inode.
1207 while (!doneleft
|| !doneright
) {
1208 int useleft
; /* using left inode chunk this time */
1210 if (!--searchdistance
) {
1212 * Not in range - save last search
1213 * location and allocate a new inode
1215 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1216 pag
->pagl_leftrec
= trec
.ir_startino
;
1217 pag
->pagl_rightrec
= rec
.ir_startino
;
1218 pag
->pagl_pagino
= pagino
;
1222 /* figure out the closer block if both are valid. */
1223 if (!doneleft
&& !doneright
) {
1225 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
1226 rec
.ir_startino
- pagino
;
1228 useleft
= !doneleft
;
1231 /* free inodes to the left? */
1232 if (useleft
&& trec
.ir_freecount
) {
1234 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1237 pag
->pagl_leftrec
= trec
.ir_startino
;
1238 pag
->pagl_rightrec
= rec
.ir_startino
;
1239 pag
->pagl_pagino
= pagino
;
1243 /* free inodes to the right? */
1244 if (!useleft
&& rec
.ir_freecount
) {
1245 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1247 pag
->pagl_leftrec
= trec
.ir_startino
;
1248 pag
->pagl_rightrec
= rec
.ir_startino
;
1249 pag
->pagl_pagino
= pagino
;
1253 /* get next record to check */
1255 error
= xfs_ialloc_next_rec(tcur
, &trec
,
1258 error
= xfs_ialloc_next_rec(cur
, &rec
,
1266 * We've reached the end of the btree. because
1267 * we are only searching a small chunk of the
1268 * btree each search, there is obviously free
1269 * inodes closer to the parent inode than we
1270 * are now. restart the search again.
1272 pag
->pagl_pagino
= NULLAGINO
;
1273 pag
->pagl_leftrec
= NULLAGINO
;
1274 pag
->pagl_rightrec
= NULLAGINO
;
1275 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1276 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1281 * In a different AG from the parent.
1282 * See if the most recently allocated block has any free.
1285 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1286 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1292 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1296 if (j
== 1 && rec
.ir_freecount
> 0) {
1298 * The last chunk allocated in the group
1299 * still has a free inode.
1307 * None left in the last group, search the whole AG
1309 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1312 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1315 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1318 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1319 if (rec
.ir_freecount
> 0)
1321 error
= xfs_btree_increment(cur
, 0, &i
);
1324 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1328 offset
= xfs_inobt_first_free_inode(&rec
);
1329 ASSERT(offset
>= 0);
1330 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1331 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1332 XFS_INODES_PER_CHUNK
) == 0);
1333 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1334 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1336 error
= xfs_inobt_update(cur
, &rec
);
1339 be32_add_cpu(&agi
->agi_freecount
, -1);
1340 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1341 pag
->pagi_freecount
--;
1343 error
= xfs_check_agi_freecount(cur
, agi
);
1347 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1348 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1353 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
1355 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1361 * Use the free inode btree to allocate an inode based on distance from the
1362 * parent. Note that the provided cursor may be deleted and replaced.
1365 xfs_dialloc_ag_finobt_near(
1367 struct xfs_btree_cur
**ocur
,
1368 struct xfs_inobt_rec_incore
*rec
)
1370 struct xfs_btree_cur
*lcur
= *ocur
; /* left search cursor */
1371 struct xfs_btree_cur
*rcur
; /* right search cursor */
1372 struct xfs_inobt_rec_incore rrec
;
1376 error
= xfs_inobt_lookup(lcur
, pagino
, XFS_LOOKUP_LE
, &i
);
1381 error
= xfs_inobt_get_rec(lcur
, rec
, &i
);
1384 XFS_WANT_CORRUPTED_RETURN(lcur
->bc_mp
, i
== 1);
1387 * See if we've landed in the parent inode record. The finobt
1388 * only tracks chunks with at least one free inode, so record
1389 * existence is enough.
1391 if (pagino
>= rec
->ir_startino
&&
1392 pagino
< (rec
->ir_startino
+ XFS_INODES_PER_CHUNK
))
1396 error
= xfs_btree_dup_cursor(lcur
, &rcur
);
1400 error
= xfs_inobt_lookup(rcur
, pagino
, XFS_LOOKUP_GE
, &j
);
1404 error
= xfs_inobt_get_rec(rcur
, &rrec
, &j
);
1407 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, j
== 1, error_rcur
);
1410 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, i
== 1 || j
== 1, error_rcur
);
1411 if (i
== 1 && j
== 1) {
1413 * Both the left and right records are valid. Choose the closer
1414 * inode chunk to the target.
1416 if ((pagino
- rec
->ir_startino
+ XFS_INODES_PER_CHUNK
- 1) >
1417 (rrec
.ir_startino
- pagino
)) {
1419 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1422 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1424 } else if (j
== 1) {
1425 /* only the right record is valid */
1427 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1429 } else if (i
== 1) {
1430 /* only the left record is valid */
1431 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1437 xfs_btree_del_cursor(rcur
, XFS_BTREE_ERROR
);
1442 * Use the free inode btree to find a free inode based on a newino hint. If
1443 * the hint is NULL, find the first free inode in the AG.
1446 xfs_dialloc_ag_finobt_newino(
1447 struct xfs_agi
*agi
,
1448 struct xfs_btree_cur
*cur
,
1449 struct xfs_inobt_rec_incore
*rec
)
1454 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1455 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1460 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1463 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1469 * Find the first inode available in the AG.
1471 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1474 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1476 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1479 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1485 * Update the inobt based on a modification made to the finobt. Also ensure that
1486 * the records from both trees are equivalent post-modification.
1489 xfs_dialloc_ag_update_inobt(
1490 struct xfs_btree_cur
*cur
, /* inobt cursor */
1491 struct xfs_inobt_rec_incore
*frec
, /* finobt record */
1492 int offset
) /* inode offset */
1494 struct xfs_inobt_rec_incore rec
;
1498 error
= xfs_inobt_lookup(cur
, frec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1501 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1503 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1506 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1507 ASSERT((XFS_AGINO_TO_OFFSET(cur
->bc_mp
, rec
.ir_startino
) %
1508 XFS_INODES_PER_CHUNK
) == 0);
1510 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1513 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, (rec
.ir_free
== frec
->ir_free
) &&
1514 (rec
.ir_freecount
== frec
->ir_freecount
));
1516 return xfs_inobt_update(cur
, &rec
);
1520 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1521 * back to the inobt search algorithm.
1523 * The caller selected an AG for us, and made sure that free inodes are
1528 struct xfs_trans
*tp
,
1529 struct xfs_buf
*agbp
,
1533 struct xfs_mount
*mp
= tp
->t_mountp
;
1534 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1535 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1536 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1537 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1538 struct xfs_perag
*pag
;
1539 struct xfs_btree_cur
*cur
; /* finobt cursor */
1540 struct xfs_btree_cur
*icur
; /* inobt cursor */
1541 struct xfs_inobt_rec_incore rec
;
1547 if (!xfs_sb_version_hasfinobt(&mp
->m_sb
))
1548 return xfs_dialloc_ag_inobt(tp
, agbp
, parent
, inop
);
1550 pag
= xfs_perag_get(mp
, agno
);
1553 * If pagino is 0 (this is the root inode allocation) use newino.
1554 * This must work because we've just allocated some.
1557 pagino
= be32_to_cpu(agi
->agi_newino
);
1559 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1561 error
= xfs_check_agi_freecount(cur
, agi
);
1566 * The search algorithm depends on whether we're in the same AG as the
1567 * parent. If so, find the closest available inode to the parent. If
1568 * not, consider the agi hint or find the first free inode in the AG.
1571 error
= xfs_dialloc_ag_finobt_near(pagino
, &cur
, &rec
);
1573 error
= xfs_dialloc_ag_finobt_newino(agi
, cur
, &rec
);
1577 offset
= xfs_inobt_first_free_inode(&rec
);
1578 ASSERT(offset
>= 0);
1579 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1580 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1581 XFS_INODES_PER_CHUNK
) == 0);
1582 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1585 * Modify or remove the finobt record.
1587 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1589 if (rec
.ir_freecount
)
1590 error
= xfs_inobt_update(cur
, &rec
);
1592 error
= xfs_btree_delete(cur
, &i
);
1597 * The finobt has now been updated appropriately. We haven't updated the
1598 * agi and superblock yet, so we can create an inobt cursor and validate
1599 * the original freecount. If all is well, make the equivalent update to
1600 * the inobt using the finobt record and offset information.
1602 icur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1604 error
= xfs_check_agi_freecount(icur
, agi
);
1608 error
= xfs_dialloc_ag_update_inobt(icur
, &rec
, offset
);
1613 * Both trees have now been updated. We must update the perag and
1614 * superblock before we can check the freecount for each btree.
1616 be32_add_cpu(&agi
->agi_freecount
, -1);
1617 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1618 pag
->pagi_freecount
--;
1620 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1622 error
= xfs_check_agi_freecount(icur
, agi
);
1625 error
= xfs_check_agi_freecount(cur
, agi
);
1629 xfs_btree_del_cursor(icur
, XFS_BTREE_NOERROR
);
1630 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1636 xfs_btree_del_cursor(icur
, XFS_BTREE_ERROR
);
1638 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1644 * Allocate an inode on disk.
1646 * Mode is used to tell whether the new inode will need space, and whether it
1649 * This function is designed to be called twice if it has to do an allocation
1650 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1651 * If an inode is available without having to performn an allocation, an inode
1652 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1653 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1654 * The caller should then commit the current transaction, allocate a
1655 * new transaction, and call xfs_dialloc() again, passing in the previous value
1656 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1657 * buffer is locked across the two calls, the second call is guaranteed to have
1658 * a free inode available.
1660 * Once we successfully pick an inode its number is returned and the on-disk
1661 * data structures are updated. The inode itself is not read in, since doing so
1662 * would break ordering constraints with xfs_reclaim.
1666 struct xfs_trans
*tp
,
1670 struct xfs_buf
**IO_agbp
,
1673 struct xfs_mount
*mp
= tp
->t_mountp
;
1674 struct xfs_buf
*agbp
;
1675 xfs_agnumber_t agno
;
1679 xfs_agnumber_t start_agno
;
1680 struct xfs_perag
*pag
;
1684 * If the caller passes in a pointer to the AGI buffer,
1685 * continue where we left off before. In this case, we
1686 * know that the allocation group has free inodes.
1693 * We do not have an agbp, so select an initial allocation
1694 * group for inode allocation.
1696 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
, okalloc
);
1697 if (start_agno
== NULLAGNUMBER
) {
1703 * If we have already hit the ceiling of inode blocks then clear
1704 * okalloc so we scan all available agi structures for a free
1707 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1708 * which will sacrifice the preciseness but improve the performance.
1710 if (mp
->m_maxicount
&&
1711 percpu_counter_read_positive(&mp
->m_icount
) + mp
->m_ialloc_inos
1712 > mp
->m_maxicount
) {
1718 * Loop until we find an allocation group that either has free inodes
1719 * or in which we can allocate some inodes. Iterate through the
1720 * allocation groups upward, wrapping at the end.
1724 pag
= xfs_perag_get(mp
, agno
);
1725 if (!pag
->pagi_inodeok
) {
1726 xfs_ialloc_next_ag(mp
);
1730 if (!pag
->pagi_init
) {
1731 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1737 * Do a first racy fast path check if this AG is usable.
1739 if (!pag
->pagi_freecount
&& !okalloc
)
1743 * Then read in the AGI buffer and recheck with the AGI buffer
1746 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1750 if (pag
->pagi_freecount
) {
1756 goto nextag_relse_buffer
;
1759 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1761 xfs_trans_brelse(tp
, agbp
);
1763 if (error
!= -ENOSPC
)
1773 * We successfully allocated some inodes, return
1774 * the current context to the caller so that it
1775 * can commit the current transaction and call
1776 * us again where we left off.
1778 ASSERT(pag
->pagi_freecount
> 0);
1786 nextag_relse_buffer
:
1787 xfs_trans_brelse(tp
, agbp
);
1790 if (++agno
== mp
->m_sb
.sb_agcount
)
1792 if (agno
== start_agno
) {
1794 return noroom
? -ENOSPC
: 0;
1800 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1807 * Free the blocks of an inode chunk. We must consider that the inode chunk
1808 * might be sparse and only free the regions that are allocated as part of the
1812 xfs_difree_inode_chunk(
1813 struct xfs_mount
*mp
,
1814 xfs_agnumber_t agno
,
1815 struct xfs_inobt_rec_incore
*rec
,
1816 struct xfs_defer_ops
*dfops
)
1818 xfs_agblock_t sagbno
= XFS_AGINO_TO_AGBNO(mp
, rec
->ir_startino
);
1819 int startidx
, endidx
;
1821 xfs_agblock_t agbno
;
1823 struct xfs_owner_info oinfo
;
1824 DECLARE_BITMAP(holemask
, XFS_INOBT_HOLEMASK_BITS
);
1825 xfs_rmap_ag_owner(&oinfo
, XFS_RMAP_OWN_INODES
);
1827 if (!xfs_inobt_issparse(rec
->ir_holemask
)) {
1828 /* not sparse, calculate extent info directly */
1829 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, sagbno
),
1830 mp
->m_ialloc_blks
, &oinfo
);
1834 /* holemask is only 16-bits (fits in an unsigned long) */
1835 ASSERT(sizeof(rec
->ir_holemask
) <= sizeof(holemask
[0]));
1836 holemask
[0] = rec
->ir_holemask
;
1839 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1840 * holemask and convert the start/end index of each range to an extent.
1841 * We start with the start and end index both pointing at the first 0 in
1844 startidx
= endidx
= find_first_zero_bit(holemask
,
1845 XFS_INOBT_HOLEMASK_BITS
);
1846 nextbit
= startidx
+ 1;
1847 while (startidx
< XFS_INOBT_HOLEMASK_BITS
) {
1848 nextbit
= find_next_zero_bit(holemask
, XFS_INOBT_HOLEMASK_BITS
,
1851 * If the next zero bit is contiguous, update the end index of
1852 * the current range and continue.
1854 if (nextbit
!= XFS_INOBT_HOLEMASK_BITS
&&
1855 nextbit
== endidx
+ 1) {
1861 * nextbit is not contiguous with the current end index. Convert
1862 * the current start/end to an extent and add it to the free
1865 agbno
= sagbno
+ (startidx
* XFS_INODES_PER_HOLEMASK_BIT
) /
1866 mp
->m_sb
.sb_inopblock
;
1867 contigblk
= ((endidx
- startidx
+ 1) *
1868 XFS_INODES_PER_HOLEMASK_BIT
) /
1869 mp
->m_sb
.sb_inopblock
;
1871 ASSERT(agbno
% mp
->m_sb
.sb_spino_align
== 0);
1872 ASSERT(contigblk
% mp
->m_sb
.sb_spino_align
== 0);
1873 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, agbno
),
1876 /* reset range to current bit and carry on... */
1877 startidx
= endidx
= nextbit
;
1886 struct xfs_mount
*mp
,
1887 struct xfs_trans
*tp
,
1888 struct xfs_buf
*agbp
,
1890 struct xfs_defer_ops
*dfops
,
1891 struct xfs_icluster
*xic
,
1892 struct xfs_inobt_rec_incore
*orec
)
1894 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1895 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1896 struct xfs_perag
*pag
;
1897 struct xfs_btree_cur
*cur
;
1898 struct xfs_inobt_rec_incore rec
;
1904 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1905 ASSERT(XFS_AGINO_TO_AGBNO(mp
, agino
) < be32_to_cpu(agi
->agi_length
));
1908 * Initialize the cursor.
1910 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1912 error
= xfs_check_agi_freecount(cur
, agi
);
1917 * Look for the entry describing this inode.
1919 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1920 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1924 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1925 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1927 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1931 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1933 * Get the offset in the inode chunk.
1935 off
= agino
- rec
.ir_startino
;
1936 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1937 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1939 * Mark the inode free & increment the count.
1941 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1945 * When an inode chunk is free, it becomes eligible for removal. Don't
1946 * remove the chunk if the block size is large enough for multiple inode
1947 * chunks (that might not be free).
1949 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1950 rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
1951 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
) {
1953 xic
->first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1954 xic
->alloc
= xfs_inobt_irec_to_allocmask(&rec
);
1957 * Remove the inode cluster from the AGI B+Tree, adjust the
1958 * AGI and Superblock inode counts, and mark the disk space
1959 * to be freed when the transaction is committed.
1961 ilen
= rec
.ir_freecount
;
1962 be32_add_cpu(&agi
->agi_count
, -ilen
);
1963 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1964 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1965 pag
= xfs_perag_get(mp
, agno
);
1966 pag
->pagi_freecount
-= ilen
- 1;
1968 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1969 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1971 if ((error
= xfs_btree_delete(cur
, &i
))) {
1972 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1977 xfs_difree_inode_chunk(mp
, agno
, &rec
, dfops
);
1981 error
= xfs_inobt_update(cur
, &rec
);
1983 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
1989 * Change the inode free counts and log the ag/sb changes.
1991 be32_add_cpu(&agi
->agi_freecount
, 1);
1992 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1993 pag
= xfs_perag_get(mp
, agno
);
1994 pag
->pagi_freecount
++;
1996 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
1999 error
= xfs_check_agi_freecount(cur
, agi
);
2004 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2008 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2013 * Free an inode in the free inode btree.
2017 struct xfs_mount
*mp
,
2018 struct xfs_trans
*tp
,
2019 struct xfs_buf
*agbp
,
2021 struct xfs_inobt_rec_incore
*ibtrec
) /* inobt record */
2023 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
2024 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
2025 struct xfs_btree_cur
*cur
;
2026 struct xfs_inobt_rec_incore rec
;
2027 int offset
= agino
- ibtrec
->ir_startino
;
2031 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
2033 error
= xfs_inobt_lookup(cur
, ibtrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
2038 * If the record does not exist in the finobt, we must have just
2039 * freed an inode in a previously fully allocated chunk. If not,
2040 * something is out of sync.
2042 XFS_WANT_CORRUPTED_GOTO(mp
, ibtrec
->ir_freecount
== 1, error
);
2044 error
= xfs_inobt_insert_rec(cur
, ibtrec
->ir_holemask
,
2046 ibtrec
->ir_freecount
,
2047 ibtrec
->ir_free
, &i
);
2056 * Read and update the existing record. We could just copy the ibtrec
2057 * across here, but that would defeat the purpose of having redundant
2058 * metadata. By making the modifications independently, we can catch
2059 * corruptions that we wouldn't see if we just copied from one record
2062 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2065 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
2067 rec
.ir_free
|= XFS_INOBT_MASK(offset
);
2070 XFS_WANT_CORRUPTED_GOTO(mp
, (rec
.ir_free
== ibtrec
->ir_free
) &&
2071 (rec
.ir_freecount
== ibtrec
->ir_freecount
),
2075 * The content of inobt records should always match between the inobt
2076 * and finobt. The lifecycle of records in the finobt is different from
2077 * the inobt in that the finobt only tracks records with at least one
2078 * free inode. Hence, if all of the inodes are free and we aren't
2079 * keeping inode chunks permanently on disk, remove the record.
2080 * Otherwise, update the record with the new information.
2082 * Note that we currently can't free chunks when the block size is large
2083 * enough for multiple chunks. Leave the finobt record to remain in sync
2086 if (rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
2087 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
&&
2088 !(mp
->m_flags
& XFS_MOUNT_IKEEP
)) {
2089 error
= xfs_btree_delete(cur
, &i
);
2094 error
= xfs_inobt_update(cur
, &rec
);
2100 error
= xfs_check_agi_freecount(cur
, agi
);
2104 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2108 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2113 * Free disk inode. Carefully avoids touching the incore inode, all
2114 * manipulations incore are the caller's responsibility.
2115 * The on-disk inode is not changed by this operation, only the
2116 * btree (free inode mask) is changed.
2120 struct xfs_trans
*tp
, /* transaction pointer */
2121 xfs_ino_t inode
, /* inode to be freed */
2122 struct xfs_defer_ops
*dfops
, /* extents to free */
2123 struct xfs_icluster
*xic
) /* cluster info if deleted */
2126 xfs_agblock_t agbno
; /* block number containing inode */
2127 struct xfs_buf
*agbp
; /* buffer for allocation group header */
2128 xfs_agino_t agino
; /* allocation group inode number */
2129 xfs_agnumber_t agno
; /* allocation group number */
2130 int error
; /* error return value */
2131 struct xfs_mount
*mp
; /* mount structure for filesystem */
2132 struct xfs_inobt_rec_incore rec
;/* btree record */
2137 * Break up inode number into its components.
2139 agno
= XFS_INO_TO_AGNO(mp
, inode
);
2140 if (agno
>= mp
->m_sb
.sb_agcount
) {
2141 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2142 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2146 agino
= XFS_INO_TO_AGINO(mp
, inode
);
2147 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2148 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2149 __func__
, (unsigned long long)inode
,
2150 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
2154 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2155 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2156 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2157 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
2162 * Get the allocation group header.
2164 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2166 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
2172 * Fix up the inode allocation btree.
2174 error
= xfs_difree_inobt(mp
, tp
, agbp
, agino
, dfops
, xic
, &rec
);
2179 * Fix up the free inode btree.
2181 if (xfs_sb_version_hasfinobt(&mp
->m_sb
)) {
2182 error
= xfs_difree_finobt(mp
, tp
, agbp
, agino
, &rec
);
2195 struct xfs_mount
*mp
,
2196 struct xfs_trans
*tp
,
2197 xfs_agnumber_t agno
,
2199 xfs_agblock_t agbno
,
2200 xfs_agblock_t
*chunk_agbno
,
2201 xfs_agblock_t
*offset_agbno
,
2204 struct xfs_inobt_rec_incore rec
;
2205 struct xfs_btree_cur
*cur
;
2206 struct xfs_buf
*agbp
;
2210 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2213 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2214 __func__
, error
, agno
);
2219 * Lookup the inode record for the given agino. If the record cannot be
2220 * found, then it's an invalid inode number and we should abort. Once
2221 * we have a record, we need to ensure it contains the inode number
2222 * we are looking up.
2224 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
2225 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
2228 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2229 if (!error
&& i
== 0)
2233 xfs_trans_brelse(tp
, agbp
);
2234 xfs_btree_del_cursor(cur
, error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
2238 /* check that the returned record contains the required inode */
2239 if (rec
.ir_startino
> agino
||
2240 rec
.ir_startino
+ mp
->m_ialloc_inos
<= agino
)
2243 /* for untrusted inodes check it is allocated first */
2244 if ((flags
& XFS_IGET_UNTRUSTED
) &&
2245 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
2248 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
2249 *offset_agbno
= agbno
- *chunk_agbno
;
2254 * Return the location of the inode in imap, for mapping it into a buffer.
2258 xfs_mount_t
*mp
, /* file system mount structure */
2259 xfs_trans_t
*tp
, /* transaction pointer */
2260 xfs_ino_t ino
, /* inode to locate */
2261 struct xfs_imap
*imap
, /* location map structure */
2262 uint flags
) /* flags for inode btree lookup */
2264 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
2265 xfs_agino_t agino
; /* inode number within alloc group */
2266 xfs_agnumber_t agno
; /* allocation group number */
2267 int blks_per_cluster
; /* num blocks per inode cluster */
2268 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
2269 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
2270 int error
; /* error code */
2271 int offset
; /* index of inode in its buffer */
2272 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
2274 ASSERT(ino
!= NULLFSINO
);
2277 * Split up the inode number into its parts.
2279 agno
= XFS_INO_TO_AGNO(mp
, ino
);
2280 agino
= XFS_INO_TO_AGINO(mp
, ino
);
2281 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2282 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
2283 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2286 * Don't output diagnostic information for untrusted inodes
2287 * as they can be invalid without implying corruption.
2289 if (flags
& XFS_IGET_UNTRUSTED
)
2291 if (agno
>= mp
->m_sb
.sb_agcount
) {
2293 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2294 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2296 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2298 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2299 __func__
, (unsigned long long)agbno
,
2300 (unsigned long)mp
->m_sb
.sb_agblocks
);
2302 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2304 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2306 XFS_AGINO_TO_INO(mp
, agno
, agino
));
2313 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2316 * For bulkstat and handle lookups, we have an untrusted inode number
2317 * that we have to verify is valid. We cannot do this just by reading
2318 * the inode buffer as it may have been unlinked and removed leaving
2319 * inodes in stale state on disk. Hence we have to do a btree lookup
2320 * in all cases where an untrusted inode number is passed.
2322 if (flags
& XFS_IGET_UNTRUSTED
) {
2323 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2324 &chunk_agbno
, &offset_agbno
, flags
);
2331 * If the inode cluster size is the same as the blocksize or
2332 * smaller we get to the buffer by simple arithmetics.
2334 if (blks_per_cluster
== 1) {
2335 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
2336 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
2338 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
2339 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
2340 imap
->im_boffset
= (unsigned short)(offset
<<
2341 mp
->m_sb
.sb_inodelog
);
2346 * If the inode chunks are aligned then use simple maths to
2347 * find the location. Otherwise we have to do a btree
2348 * lookup to find the location.
2350 if (mp
->m_inoalign_mask
) {
2351 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
2352 chunk_agbno
= agbno
- offset_agbno
;
2354 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2355 &chunk_agbno
, &offset_agbno
, flags
);
2361 ASSERT(agbno
>= chunk_agbno
);
2362 cluster_agbno
= chunk_agbno
+
2363 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
2364 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
2365 XFS_INO_TO_OFFSET(mp
, ino
);
2367 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
2368 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
2369 imap
->im_boffset
= (unsigned short)(offset
<< mp
->m_sb
.sb_inodelog
);
2372 * If the inode number maps to a block outside the bounds
2373 * of the file system then return NULL rather than calling
2374 * read_buf and panicing when we get an error from the
2377 if ((imap
->im_blkno
+ imap
->im_len
) >
2378 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
2380 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2381 __func__
, (unsigned long long) imap
->im_blkno
,
2382 (unsigned long long) imap
->im_len
,
2383 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
2390 * Compute and fill in value of m_in_maxlevels.
2393 xfs_ialloc_compute_maxlevels(
2394 xfs_mount_t
*mp
) /* file system mount structure */
2398 inodes
= (1LL << XFS_INO_AGINO_BITS(mp
)) >> XFS_INODES_PER_CHUNK_LOG
;
2399 mp
->m_in_maxlevels
= xfs_btree_compute_maxlevels(mp
, mp
->m_inobt_mnr
,
2404 * Log specified fields for the ag hdr (inode section). The growth of the agi
2405 * structure over time requires that we interpret the buffer as two logical
2406 * regions delineated by the end of the unlinked list. This is due to the size
2407 * of the hash table and its location in the middle of the agi.
2409 * For example, a request to log a field before agi_unlinked and a field after
2410 * agi_unlinked could cause us to log the entire hash table and use an excessive
2411 * amount of log space. To avoid this behavior, log the region up through
2412 * agi_unlinked in one call and the region after agi_unlinked through the end of
2413 * the structure in another.
2417 xfs_trans_t
*tp
, /* transaction pointer */
2418 xfs_buf_t
*bp
, /* allocation group header buffer */
2419 int fields
) /* bitmask of fields to log */
2421 int first
; /* first byte number */
2422 int last
; /* last byte number */
2423 static const short offsets
[] = { /* field starting offsets */
2424 /* keep in sync with bit definitions */
2425 offsetof(xfs_agi_t
, agi_magicnum
),
2426 offsetof(xfs_agi_t
, agi_versionnum
),
2427 offsetof(xfs_agi_t
, agi_seqno
),
2428 offsetof(xfs_agi_t
, agi_length
),
2429 offsetof(xfs_agi_t
, agi_count
),
2430 offsetof(xfs_agi_t
, agi_root
),
2431 offsetof(xfs_agi_t
, agi_level
),
2432 offsetof(xfs_agi_t
, agi_freecount
),
2433 offsetof(xfs_agi_t
, agi_newino
),
2434 offsetof(xfs_agi_t
, agi_dirino
),
2435 offsetof(xfs_agi_t
, agi_unlinked
),
2436 offsetof(xfs_agi_t
, agi_free_root
),
2437 offsetof(xfs_agi_t
, agi_free_level
),
2441 xfs_agi_t
*agi
; /* allocation group header */
2443 agi
= XFS_BUF_TO_AGI(bp
);
2444 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
2448 * Compute byte offsets for the first and last fields in the first
2449 * region and log the agi buffer. This only logs up through
2452 if (fields
& XFS_AGI_ALL_BITS_R1
) {
2453 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R1
,
2455 xfs_trans_log_buf(tp
, bp
, first
, last
);
2459 * Mask off the bits in the first region and calculate the first and
2460 * last field offsets for any bits in the second region.
2462 fields
&= ~XFS_AGI_ALL_BITS_R1
;
2464 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R2
,
2466 xfs_trans_log_buf(tp
, bp
, first
, last
);
2472 xfs_check_agi_unlinked(
2473 struct xfs_agi
*agi
)
2477 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++)
2478 ASSERT(agi
->agi_unlinked
[i
]);
2481 #define xfs_check_agi_unlinked(agi)
2488 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2489 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
2491 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2492 if (!uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_meta_uuid
))
2494 if (!xfs_log_check_lsn(mp
,
2495 be64_to_cpu(XFS_BUF_TO_AGI(bp
)->agi_lsn
)))
2500 * Validate the magic number of the agi block.
2502 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
2504 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
2507 if (be32_to_cpu(agi
->agi_level
) < 1 ||
2508 be32_to_cpu(agi
->agi_level
) > XFS_BTREE_MAXLEVELS
)
2511 if (xfs_sb_version_hasfinobt(&mp
->m_sb
) &&
2512 (be32_to_cpu(agi
->agi_free_level
) < 1 ||
2513 be32_to_cpu(agi
->agi_free_level
) > XFS_BTREE_MAXLEVELS
))
2517 * during growfs operations, the perag is not fully initialised,
2518 * so we can't use it for any useful checking. growfs ensures we can't
2519 * use it by using uncached buffers that don't have the perag attached
2520 * so we can detect and avoid this problem.
2522 if (bp
->b_pag
&& be32_to_cpu(agi
->agi_seqno
) != bp
->b_pag
->pag_agno
)
2525 xfs_check_agi_unlinked(agi
);
2530 xfs_agi_read_verify(
2533 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2535 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2536 !xfs_buf_verify_cksum(bp
, XFS_AGI_CRC_OFF
))
2537 xfs_buf_ioerror(bp
, -EFSBADCRC
);
2538 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp
), mp
,
2539 XFS_ERRTAG_IALLOC_READ_AGI
,
2540 XFS_RANDOM_IALLOC_READ_AGI
))
2541 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2544 xfs_verifier_error(bp
);
2548 xfs_agi_write_verify(
2551 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2552 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
2554 if (!xfs_agi_verify(bp
)) {
2555 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2556 xfs_verifier_error(bp
);
2560 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2564 XFS_BUF_TO_AGI(bp
)->agi_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2565 xfs_buf_update_cksum(bp
, XFS_AGI_CRC_OFF
);
2568 const struct xfs_buf_ops xfs_agi_buf_ops
= {
2570 .verify_read
= xfs_agi_read_verify
,
2571 .verify_write
= xfs_agi_write_verify
,
2575 * Read in the allocation group header (inode allocation section)
2579 struct xfs_mount
*mp
, /* file system mount structure */
2580 struct xfs_trans
*tp
, /* transaction pointer */
2581 xfs_agnumber_t agno
, /* allocation group number */
2582 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2586 trace_xfs_read_agi(mp
, agno
);
2588 ASSERT(agno
!= NULLAGNUMBER
);
2589 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
2590 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
2591 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
2595 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_AGI_BUF
);
2597 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
2602 xfs_ialloc_read_agi(
2603 struct xfs_mount
*mp
, /* file system mount structure */
2604 struct xfs_trans
*tp
, /* transaction pointer */
2605 xfs_agnumber_t agno
, /* allocation group number */
2606 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2608 struct xfs_agi
*agi
; /* allocation group header */
2609 struct xfs_perag
*pag
; /* per allocation group data */
2612 trace_xfs_ialloc_read_agi(mp
, agno
);
2614 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
2618 agi
= XFS_BUF_TO_AGI(*bpp
);
2619 pag
= xfs_perag_get(mp
, agno
);
2620 if (!pag
->pagi_init
) {
2621 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
2622 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
2627 * It's possible for these to be out of sync if
2628 * we are in the middle of a forced shutdown.
2630 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
2631 XFS_FORCED_SHUTDOWN(mp
));
2637 * Read in the agi to initialise the per-ag data in the mount structure
2640 xfs_ialloc_pagi_init(
2641 xfs_mount_t
*mp
, /* file system mount structure */
2642 xfs_trans_t
*tp
, /* transaction pointer */
2643 xfs_agnumber_t agno
) /* allocation group number */
2645 xfs_buf_t
*bp
= NULL
;
2648 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
2652 xfs_trans_brelse(tp
, bp
);