2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_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"
33 #include "xfs_rtalloc.h"
34 #include "xfs_error.h"
36 #include "xfs_cksum.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_icreate_item.h"
40 #include "xfs_icache.h"
41 #include "xfs_trace.h"
47 * Allocation group level functions.
50 xfs_ialloc_cluster_alignment(
53 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
54 mp
->m_sb
.sb_inoalignmt
>= xfs_icluster_size_fsb(mp
))
55 return mp
->m_sb
.sb_inoalignmt
;
60 * Lookup a record by ino in the btree given by cur.
64 struct xfs_btree_cur
*cur
, /* btree cursor */
65 xfs_agino_t ino
, /* starting inode of chunk */
66 xfs_lookup_t dir
, /* <=, >=, == */
67 int *stat
) /* success/failure */
69 cur
->bc_rec
.i
.ir_startino
= ino
;
70 cur
->bc_rec
.i
.ir_holemask
= 0;
71 cur
->bc_rec
.i
.ir_count
= 0;
72 cur
->bc_rec
.i
.ir_freecount
= 0;
73 cur
->bc_rec
.i
.ir_free
= 0;
74 return xfs_btree_lookup(cur
, dir
, stat
);
78 * Update the record referred to by cur to the value given.
79 * This either works (return 0) or gets an EFSCORRUPTED error.
81 STATIC
int /* error */
83 struct xfs_btree_cur
*cur
, /* btree cursor */
84 xfs_inobt_rec_incore_t
*irec
) /* btree record */
86 union xfs_btree_rec rec
;
88 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
89 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
90 rec
.inobt
.ir_u
.sp
.ir_holemask
= cpu_to_be16(irec
->ir_holemask
);
91 rec
.inobt
.ir_u
.sp
.ir_count
= irec
->ir_count
;
92 rec
.inobt
.ir_u
.sp
.ir_freecount
= irec
->ir_freecount
;
94 /* ir_holemask/ir_count not supported on-disk */
95 rec
.inobt
.ir_u
.f
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
97 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
98 return xfs_btree_update(cur
, &rec
);
102 * Get the data from the pointed-to record.
106 struct xfs_btree_cur
*cur
, /* btree cursor */
107 xfs_inobt_rec_incore_t
*irec
, /* btree record */
108 int *stat
) /* output: success/failure */
110 union xfs_btree_rec
*rec
;
113 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
114 if (error
|| *stat
== 0)
117 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
118 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
119 irec
->ir_holemask
= be16_to_cpu(rec
->inobt
.ir_u
.sp
.ir_holemask
);
120 irec
->ir_count
= rec
->inobt
.ir_u
.sp
.ir_count
;
121 irec
->ir_freecount
= rec
->inobt
.ir_u
.sp
.ir_freecount
;
124 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
125 * values for full inode chunks.
127 irec
->ir_holemask
= XFS_INOBT_HOLEMASK_FULL
;
128 irec
->ir_count
= XFS_INODES_PER_CHUNK
;
130 be32_to_cpu(rec
->inobt
.ir_u
.f
.ir_freecount
);
132 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
138 * Insert a single inobt record. Cursor must already point to desired location.
141 xfs_inobt_insert_rec(
142 struct xfs_btree_cur
*cur
,
149 cur
->bc_rec
.i
.ir_holemask
= holemask
;
150 cur
->bc_rec
.i
.ir_count
= count
;
151 cur
->bc_rec
.i
.ir_freecount
= freecount
;
152 cur
->bc_rec
.i
.ir_free
= free
;
153 return xfs_btree_insert(cur
, stat
);
157 * Insert records describing a newly allocated inode chunk into the inobt.
161 struct xfs_mount
*mp
,
162 struct xfs_trans
*tp
,
163 struct xfs_buf
*agbp
,
168 struct xfs_btree_cur
*cur
;
169 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
170 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
175 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
177 for (thisino
= newino
;
178 thisino
< newino
+ newlen
;
179 thisino
+= XFS_INODES_PER_CHUNK
) {
180 error
= xfs_inobt_lookup(cur
, thisino
, XFS_LOOKUP_EQ
, &i
);
182 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
187 error
= xfs_inobt_insert_rec(cur
, XFS_INOBT_HOLEMASK_FULL
,
188 XFS_INODES_PER_CHUNK
,
189 XFS_INODES_PER_CHUNK
,
190 XFS_INOBT_ALL_FREE
, &i
);
192 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
198 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
204 * Verify that the number of free inodes in the AGI is correct.
208 xfs_check_agi_freecount(
209 struct xfs_btree_cur
*cur
,
212 if (cur
->bc_nlevels
== 1) {
213 xfs_inobt_rec_incore_t rec
;
218 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
223 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
228 freecount
+= rec
.ir_freecount
;
229 error
= xfs_btree_increment(cur
, 0, &i
);
235 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
236 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
241 #define xfs_check_agi_freecount(cur, agi) 0
245 * Initialise a new set of inodes. When called without a transaction context
246 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
247 * than logging them (which in a transaction context puts them into the AIL
248 * for writeback rather than the xfsbufd queue).
251 xfs_ialloc_inode_init(
252 struct xfs_mount
*mp
,
253 struct xfs_trans
*tp
,
254 struct list_head
*buffer_list
,
258 xfs_agblock_t length
,
261 struct xfs_buf
*fbuf
;
262 struct xfs_dinode
*free
;
263 int nbufs
, blks_per_cluster
, inodes_per_cluster
;
270 * Loop over the new block(s), filling in the inodes. For small block
271 * sizes, manipulate the inodes in buffers which are multiples of the
274 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
275 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
276 nbufs
= length
/ blks_per_cluster
;
279 * Figure out what version number to use in the inodes we create. If
280 * the superblock version has caught up to the one that supports the new
281 * inode format, then use the new inode version. Otherwise use the old
282 * version so that old kernels will continue to be able to use the file
285 * For v3 inodes, we also need to write the inode number into the inode,
286 * so calculate the first inode number of the chunk here as
287 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
288 * across multiple filesystem blocks (such as a cluster) and so cannot
289 * be used in the cluster buffer loop below.
291 * Further, because we are writing the inode directly into the buffer
292 * and calculating a CRC on the entire inode, we have ot log the entire
293 * inode so that the entire range the CRC covers is present in the log.
294 * That means for v3 inode we log the entire buffer rather than just the
297 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
299 ino
= XFS_AGINO_TO_INO(mp
, agno
,
300 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
303 * log the initialisation that is about to take place as an
304 * logical operation. This means the transaction does not
305 * need to log the physical changes to the inode buffers as log
306 * recovery will know what initialisation is actually needed.
307 * Hence we only need to log the buffers as "ordered" buffers so
308 * they track in the AIL as if they were physically logged.
311 xfs_icreate_log(tp
, agno
, agbno
, icount
,
312 mp
->m_sb
.sb_inodesize
, length
, gen
);
316 for (j
= 0; j
< nbufs
; j
++) {
320 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
321 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
322 mp
->m_bsize
* blks_per_cluster
,
327 /* Initialize the inode buffers and log them appropriately. */
328 fbuf
->b_ops
= &xfs_inode_buf_ops
;
329 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
330 for (i
= 0; i
< inodes_per_cluster
; i
++) {
331 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
332 uint isize
= xfs_dinode_size(version
);
334 free
= xfs_make_iptr(mp
, fbuf
, i
);
335 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
336 free
->di_version
= version
;
337 free
->di_gen
= cpu_to_be32(gen
);
338 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
341 free
->di_ino
= cpu_to_be64(ino
);
343 uuid_copy(&free
->di_uuid
,
344 &mp
->m_sb
.sb_meta_uuid
);
345 xfs_dinode_calc_crc(mp
, free
);
347 /* just log the inode core */
348 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
349 ioffset
+ isize
- 1);
355 * Mark the buffer as an inode allocation buffer so it
356 * sticks in AIL at the point of this allocation
357 * transaction. This ensures the they are on disk before
358 * the tail of the log can be moved past this
359 * transaction (i.e. by preventing relogging from moving
360 * it forward in the log).
362 xfs_trans_inode_alloc_buf(tp
, fbuf
);
365 * Mark the buffer as ordered so that they are
366 * not physically logged in the transaction but
367 * still tracked in the AIL as part of the
368 * transaction and pin the log appropriately.
370 xfs_trans_ordered_buf(tp
, fbuf
);
371 xfs_trans_log_buf(tp
, fbuf
, 0,
372 BBTOB(fbuf
->b_length
) - 1);
375 fbuf
->b_flags
|= XBF_DONE
;
376 xfs_buf_delwri_queue(fbuf
, buffer_list
);
384 * Align startino and allocmask for a recently allocated sparse chunk such that
385 * they are fit for insertion (or merge) into the on-disk inode btrees.
389 * When enabled, sparse inode support increases the inode alignment from cluster
390 * size to inode chunk size. This means that the minimum range between two
391 * non-adjacent inode records in the inobt is large enough for a full inode
392 * record. This allows for cluster sized, cluster aligned block allocation
393 * without need to worry about whether the resulting inode record overlaps with
394 * another record in the tree. Without this basic rule, we would have to deal
395 * with the consequences of overlap by potentially undoing recent allocations in
396 * the inode allocation codepath.
398 * Because of this alignment rule (which is enforced on mount), there are two
399 * inobt possibilities for newly allocated sparse chunks. One is that the
400 * aligned inode record for the chunk covers a range of inodes not already
401 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
402 * other is that a record already exists at the aligned startino that considers
403 * the newly allocated range as sparse. In the latter case, record content is
404 * merged in hope that sparse inode chunks fill to full chunks over time.
407 xfs_align_sparse_ino(
408 struct xfs_mount
*mp
,
409 xfs_agino_t
*startino
,
416 agbno
= XFS_AGINO_TO_AGBNO(mp
, *startino
);
417 mod
= agbno
% mp
->m_sb
.sb_inoalignmt
;
421 /* calculate the inode offset and align startino */
422 offset
= mod
<< mp
->m_sb
.sb_inopblog
;
426 * Since startino has been aligned down, left shift allocmask such that
427 * it continues to represent the same physical inodes relative to the
430 *allocmask
<<= offset
/ XFS_INODES_PER_HOLEMASK_BIT
;
434 * Determine whether the source inode record can merge into the target. Both
435 * records must be sparse, the inode ranges must match and there must be no
436 * allocation overlap between the records.
439 __xfs_inobt_can_merge(
440 struct xfs_inobt_rec_incore
*trec
, /* tgt record */
441 struct xfs_inobt_rec_incore
*srec
) /* src record */
446 /* records must cover the same inode range */
447 if (trec
->ir_startino
!= srec
->ir_startino
)
450 /* both records must be sparse */
451 if (!xfs_inobt_issparse(trec
->ir_holemask
) ||
452 !xfs_inobt_issparse(srec
->ir_holemask
))
455 /* both records must track some inodes */
456 if (!trec
->ir_count
|| !srec
->ir_count
)
459 /* can't exceed capacity of a full record */
460 if (trec
->ir_count
+ srec
->ir_count
> XFS_INODES_PER_CHUNK
)
463 /* verify there is no allocation overlap */
464 talloc
= xfs_inobt_irec_to_allocmask(trec
);
465 salloc
= xfs_inobt_irec_to_allocmask(srec
);
473 * Merge the source inode record into the target. The caller must call
474 * __xfs_inobt_can_merge() to ensure the merge is valid.
477 __xfs_inobt_rec_merge(
478 struct xfs_inobt_rec_incore
*trec
, /* target */
479 struct xfs_inobt_rec_incore
*srec
) /* src */
481 ASSERT(trec
->ir_startino
== srec
->ir_startino
);
483 /* combine the counts */
484 trec
->ir_count
+= srec
->ir_count
;
485 trec
->ir_freecount
+= srec
->ir_freecount
;
488 * Merge the holemask and free mask. For both fields, 0 bits refer to
489 * allocated inodes. We combine the allocated ranges with bitwise AND.
491 trec
->ir_holemask
&= srec
->ir_holemask
;
492 trec
->ir_free
&= srec
->ir_free
;
496 * Insert a new sparse inode chunk into the associated inode btree. The inode
497 * record for the sparse chunk is pre-aligned to a startino that should match
498 * any pre-existing sparse inode record in the tree. This allows sparse chunks
501 * This function supports two modes of handling preexisting records depending on
502 * the merge flag. If merge is true, the provided record is merged with the
503 * existing record and updated in place. The merged record is returned in nrec.
504 * If merge is false, an existing record is replaced with the provided record.
505 * If no preexisting record exists, the provided record is always inserted.
507 * It is considered corruption if a merge is requested and not possible. Given
508 * the sparse inode alignment constraints, this should never happen.
511 xfs_inobt_insert_sprec(
512 struct xfs_mount
*mp
,
513 struct xfs_trans
*tp
,
514 struct xfs_buf
*agbp
,
516 struct xfs_inobt_rec_incore
*nrec
, /* in/out: new/merged rec. */
517 bool merge
) /* merge or replace */
519 struct xfs_btree_cur
*cur
;
520 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
521 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
524 struct xfs_inobt_rec_incore rec
;
526 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
528 /* the new record is pre-aligned so we know where to look */
529 error
= xfs_inobt_lookup(cur
, nrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
532 /* if nothing there, insert a new record and return */
534 error
= xfs_inobt_insert_rec(cur
, nrec
->ir_holemask
,
535 nrec
->ir_count
, nrec
->ir_freecount
,
539 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
545 * A record exists at this startino. Merge or replace the record
546 * depending on what we've been asked to do.
549 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
552 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
553 XFS_WANT_CORRUPTED_GOTO(mp
,
554 rec
.ir_startino
== nrec
->ir_startino
,
558 * This should never fail. If we have coexisting records that
559 * cannot merge, something is seriously wrong.
561 XFS_WANT_CORRUPTED_GOTO(mp
, __xfs_inobt_can_merge(nrec
, &rec
),
564 trace_xfs_irec_merge_pre(mp
, agno
, rec
.ir_startino
,
565 rec
.ir_holemask
, nrec
->ir_startino
,
568 /* merge to nrec to output the updated record */
569 __xfs_inobt_rec_merge(nrec
, &rec
);
571 trace_xfs_irec_merge_post(mp
, agno
, nrec
->ir_startino
,
574 error
= xfs_inobt_rec_check_count(mp
, nrec
);
579 error
= xfs_inobt_update(cur
, nrec
);
584 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
587 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
592 * Allocate new inodes in the allocation group specified by agbp.
593 * Return 0 for success, else error code.
595 STATIC
int /* error code or 0 */
597 xfs_trans_t
*tp
, /* transaction pointer */
598 xfs_buf_t
*agbp
, /* alloc group buffer */
601 xfs_agi_t
*agi
; /* allocation group header */
602 xfs_alloc_arg_t args
; /* allocation argument structure */
605 xfs_agino_t newino
; /* new first inode's number */
606 xfs_agino_t newlen
; /* new number of inodes */
607 int isaligned
= 0; /* inode allocation at stripe unit */
609 uint16_t allocmask
= (uint16_t) -1; /* init. to full chunk */
610 struct xfs_inobt_rec_incore rec
;
611 struct xfs_perag
*pag
;
614 memset(&args
, 0, sizeof(args
));
616 args
.mp
= tp
->t_mountp
;
617 args
.fsbno
= NULLFSBLOCK
;
618 xfs_rmap_ag_owner(&args
.oinfo
, XFS_RMAP_OWN_INODES
);
621 /* randomly do sparse inode allocations */
622 if (xfs_sb_version_hassparseinodes(&tp
->t_mountp
->m_sb
) &&
623 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
)
624 do_sparse
= prandom_u32() & 1;
628 * Locking will ensure that we don't have two callers in here
631 newlen
= args
.mp
->m_ialloc_inos
;
632 if (args
.mp
->m_maxicount
&&
633 percpu_counter_read_positive(&args
.mp
->m_icount
) + newlen
>
634 args
.mp
->m_maxicount
)
636 args
.minlen
= args
.maxlen
= args
.mp
->m_ialloc_blks
;
638 * First try to allocate inodes contiguous with the last-allocated
639 * chunk of inodes. If the filesystem is striped, this will fill
640 * an entire stripe unit with inodes.
642 agi
= XFS_BUF_TO_AGI(agbp
);
643 newino
= be32_to_cpu(agi
->agi_newino
);
644 agno
= be32_to_cpu(agi
->agi_seqno
);
645 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
646 args
.mp
->m_ialloc_blks
;
649 if (likely(newino
!= NULLAGINO
&&
650 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
651 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
652 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
656 * We need to take into account alignment here to ensure that
657 * we don't modify the free list if we fail to have an exact
658 * block. If we don't have an exact match, and every oher
659 * attempt allocation attempt fails, we'll end up cancelling
660 * a dirty transaction and shutting down.
662 * For an exact allocation, alignment must be 1,
663 * however we need to take cluster alignment into account when
664 * fixing up the freelist. Use the minalignslop field to
665 * indicate that extra blocks might be required for alignment,
666 * but not to use them in the actual exact allocation.
669 args
.minalignslop
= xfs_ialloc_cluster_alignment(args
.mp
) - 1;
671 /* Allow space for the inode btree to split. */
672 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
673 if ((error
= xfs_alloc_vextent(&args
)))
677 * This request might have dirtied the transaction if the AG can
678 * satisfy the request, but the exact block was not available.
679 * If the allocation did fail, subsequent requests will relax
680 * the exact agbno requirement and increase the alignment
681 * instead. It is critical that the total size of the request
682 * (len + alignment + slop) does not increase from this point
683 * on, so reset minalignslop to ensure it is not included in
684 * subsequent requests.
686 args
.minalignslop
= 0;
689 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
691 * Set the alignment for the allocation.
692 * If stripe alignment is turned on then align at stripe unit
694 * If the cluster size is smaller than a filesystem block
695 * then we're doing I/O for inodes in filesystem block size
696 * pieces, so don't need alignment anyway.
699 if (args
.mp
->m_sinoalign
) {
700 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
701 args
.alignment
= args
.mp
->m_dalign
;
704 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
706 * Need to figure out where to allocate the inode blocks.
707 * Ideally they should be spaced out through the a.g.
708 * For now, just allocate blocks up front.
710 args
.agbno
= be32_to_cpu(agi
->agi_root
);
711 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
713 * Allocate a fixed-size extent of inodes.
715 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
718 * Allow space for the inode btree to split.
720 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
721 if ((error
= xfs_alloc_vextent(&args
)))
726 * If stripe alignment is turned on, then try again with cluster
729 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
730 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
731 args
.agbno
= be32_to_cpu(agi
->agi_root
);
732 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
733 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
734 if ((error
= xfs_alloc_vextent(&args
)))
739 * Finally, try a sparse allocation if the filesystem supports it and
740 * the sparse allocation length is smaller than a full chunk.
742 if (xfs_sb_version_hassparseinodes(&args
.mp
->m_sb
) &&
743 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
&&
744 args
.fsbno
== NULLFSBLOCK
) {
746 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
747 args
.agbno
= be32_to_cpu(agi
->agi_root
);
748 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
749 args
.alignment
= args
.mp
->m_sb
.sb_spino_align
;
752 args
.minlen
= args
.mp
->m_ialloc_min_blks
;
753 args
.maxlen
= args
.minlen
;
756 * The inode record will be aligned to full chunk size. We must
757 * prevent sparse allocation from AG boundaries that result in
758 * invalid inode records, such as records that start at agbno 0
759 * or extend beyond the AG.
761 * Set min agbno to the first aligned, non-zero agbno and max to
762 * the last aligned agbno that is at least one full chunk from
765 args
.min_agbno
= args
.mp
->m_sb
.sb_inoalignmt
;
766 args
.max_agbno
= round_down(args
.mp
->m_sb
.sb_agblocks
,
767 args
.mp
->m_sb
.sb_inoalignmt
) -
768 args
.mp
->m_ialloc_blks
;
770 error
= xfs_alloc_vextent(&args
);
774 newlen
= args
.len
<< args
.mp
->m_sb
.sb_inopblog
;
775 ASSERT(newlen
<= XFS_INODES_PER_CHUNK
);
776 allocmask
= (1 << (newlen
/ XFS_INODES_PER_HOLEMASK_BIT
)) - 1;
779 if (args
.fsbno
== NULLFSBLOCK
) {
783 ASSERT(args
.len
== args
.minlen
);
786 * Stamp and write the inode buffers.
788 * Seed the new inode cluster with a random generation number. This
789 * prevents short-term reuse of generation numbers if a chunk is
790 * freed and then immediately reallocated. We use random numbers
791 * rather than a linear progression to prevent the next generation
792 * number from being easily guessable.
794 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, newlen
, agno
,
795 args
.agbno
, args
.len
, prandom_u32());
800 * Convert the results.
802 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
804 if (xfs_inobt_issparse(~allocmask
)) {
806 * We've allocated a sparse chunk. Align the startino and mask.
808 xfs_align_sparse_ino(args
.mp
, &newino
, &allocmask
);
810 rec
.ir_startino
= newino
;
811 rec
.ir_holemask
= ~allocmask
;
812 rec
.ir_count
= newlen
;
813 rec
.ir_freecount
= newlen
;
814 rec
.ir_free
= XFS_INOBT_ALL_FREE
;
817 * Insert the sparse record into the inobt and allow for a merge
818 * if necessary. If a merge does occur, rec is updated to the
821 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
, XFS_BTNUM_INO
,
823 if (error
== -EFSCORRUPTED
) {
825 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
826 XFS_AGINO_TO_INO(args
.mp
, agno
,
828 rec
.ir_holemask
, rec
.ir_count
);
829 xfs_force_shutdown(args
.mp
, SHUTDOWN_CORRUPT_INCORE
);
835 * We can't merge the part we've just allocated as for the inobt
836 * due to finobt semantics. The original record may or may not
837 * exist independent of whether physical inodes exist in this
840 * We must update the finobt record based on the inobt record.
841 * rec contains the fully merged and up to date inobt record
842 * from the previous call. Set merge false to replace any
843 * existing record with this one.
845 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
846 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
,
847 XFS_BTNUM_FINO
, &rec
,
853 /* full chunk - insert new records to both btrees */
854 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
859 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
860 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
,
861 newlen
, XFS_BTNUM_FINO
);
868 * Update AGI counts and newino.
870 be32_add_cpu(&agi
->agi_count
, newlen
);
871 be32_add_cpu(&agi
->agi_freecount
, newlen
);
872 pag
= xfs_perag_get(args
.mp
, agno
);
873 pag
->pagi_freecount
+= newlen
;
875 agi
->agi_newino
= cpu_to_be32(newino
);
878 * Log allocation group header fields
880 xfs_ialloc_log_agi(tp
, agbp
,
881 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
883 * Modify/log superblock values for inode count and inode free count.
885 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
886 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
891 STATIC xfs_agnumber_t
897 spin_lock(&mp
->m_agirotor_lock
);
898 agno
= mp
->m_agirotor
;
899 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
901 spin_unlock(&mp
->m_agirotor_lock
);
907 * Select an allocation group to look for a free inode in, based on the parent
908 * inode and the mode. Return the allocation group buffer.
910 STATIC xfs_agnumber_t
911 xfs_ialloc_ag_select(
912 xfs_trans_t
*tp
, /* transaction pointer */
913 xfs_ino_t parent
, /* parent directory inode number */
914 umode_t mode
, /* bits set to indicate file type */
915 int okalloc
) /* ok to allocate more space */
917 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
918 xfs_agnumber_t agno
; /* current ag number */
919 int flags
; /* alloc buffer locking flags */
920 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
921 xfs_extlen_t longest
= 0; /* longest extent available */
922 xfs_mount_t
*mp
; /* mount point structure */
923 int needspace
; /* file mode implies space allocated */
924 xfs_perag_t
*pag
; /* per allocation group data */
925 xfs_agnumber_t pagno
; /* parent (starting) ag number */
929 * Files of these types need at least one block if length > 0
930 * (and they won't fit in the inode, but that's hard to figure out).
932 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
934 agcount
= mp
->m_maxagi
;
936 pagno
= xfs_ialloc_next_ag(mp
);
938 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
939 if (pagno
>= agcount
)
943 ASSERT(pagno
< agcount
);
946 * Loop through allocation groups, looking for one with a little
947 * free space in it. Note we don't look for free inodes, exactly.
948 * Instead, we include whether there is a need to allocate inodes
949 * to mean that blocks must be allocated for them,
950 * if none are currently free.
953 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
955 pag
= xfs_perag_get(mp
, agno
);
956 if (!pag
->pagi_inodeok
) {
957 xfs_ialloc_next_ag(mp
);
961 if (!pag
->pagi_init
) {
962 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
967 if (pag
->pagi_freecount
) {
975 if (!pag
->pagf_init
) {
976 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
982 * Check that there is enough free space for the file plus a
983 * chunk of inodes if we need to allocate some. If this is the
984 * first pass across the AGs, take into account the potential
985 * space needed for alignment of inode chunks when checking the
986 * longest contiguous free space in the AG - this prevents us
987 * from getting ENOSPC because we have free space larger than
988 * m_ialloc_blks but alignment constraints prevent us from using
991 * If we can't find an AG with space for full alignment slack to
992 * be taken into account, we must be near ENOSPC in all AGs.
993 * Hence we don't include alignment for the second pass and so
994 * if we fail allocation due to alignment issues then it is most
995 * likely a real ENOSPC condition.
997 ineed
= mp
->m_ialloc_min_blks
;
998 if (flags
&& ineed
> 1)
999 ineed
+= xfs_ialloc_cluster_alignment(mp
);
1000 longest
= pag
->pagf_longest
;
1002 longest
= pag
->pagf_flcount
> 0;
1004 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
1012 * No point in iterating over the rest, if we're shutting
1015 if (XFS_FORCED_SHUTDOWN(mp
))
1016 return NULLAGNUMBER
;
1018 if (agno
>= agcount
)
1020 if (agno
== pagno
) {
1022 return NULLAGNUMBER
;
1029 * Try to retrieve the next record to the left/right from the current one.
1032 xfs_ialloc_next_rec(
1033 struct xfs_btree_cur
*cur
,
1034 xfs_inobt_rec_incore_t
*rec
,
1042 error
= xfs_btree_decrement(cur
, 0, &i
);
1044 error
= xfs_btree_increment(cur
, 0, &i
);
1050 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1053 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1061 struct xfs_btree_cur
*cur
,
1063 xfs_inobt_rec_incore_t
*rec
,
1069 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
1074 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1077 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1084 * Return the offset of the first free inode in the record. If the inode chunk
1085 * is sparsely allocated, we convert the record holemask to inode granularity
1086 * and mask off the unallocated regions from the inode free mask.
1089 xfs_inobt_first_free_inode(
1090 struct xfs_inobt_rec_incore
*rec
)
1092 xfs_inofree_t realfree
;
1094 /* if there are no holes, return the first available offset */
1095 if (!xfs_inobt_issparse(rec
->ir_holemask
))
1096 return xfs_lowbit64(rec
->ir_free
);
1098 realfree
= xfs_inobt_irec_to_allocmask(rec
);
1099 realfree
&= rec
->ir_free
;
1101 return xfs_lowbit64(realfree
);
1105 * Allocate an inode using the inobt-only algorithm.
1108 xfs_dialloc_ag_inobt(
1109 struct xfs_trans
*tp
,
1110 struct xfs_buf
*agbp
,
1114 struct xfs_mount
*mp
= tp
->t_mountp
;
1115 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1116 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1117 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1118 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1119 struct xfs_perag
*pag
;
1120 struct xfs_btree_cur
*cur
, *tcur
;
1121 struct xfs_inobt_rec_incore rec
, trec
;
1127 pag
= xfs_perag_get(mp
, agno
);
1129 ASSERT(pag
->pagi_init
);
1130 ASSERT(pag
->pagi_inodeok
);
1131 ASSERT(pag
->pagi_freecount
> 0);
1134 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1136 * If pagino is 0 (this is the root inode allocation) use newino.
1137 * This must work because we've just allocated some.
1140 pagino
= be32_to_cpu(agi
->agi_newino
);
1142 error
= xfs_check_agi_freecount(cur
, agi
);
1147 * If in the same AG as the parent, try to get near the parent.
1149 if (pagno
== agno
) {
1150 int doneleft
; /* done, to the left */
1151 int doneright
; /* done, to the right */
1152 int searchdistance
= 10;
1154 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
1157 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1159 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1162 XFS_WANT_CORRUPTED_GOTO(mp
, j
== 1, error0
);
1164 if (rec
.ir_freecount
> 0) {
1166 * Found a free inode in the same chunk
1167 * as the parent, done.
1174 * In the same AG as parent, but parent's chunk is full.
1177 /* duplicate the cursor, search left & right simultaneously */
1178 error
= xfs_btree_dup_cursor(cur
, &tcur
);
1183 * Skip to last blocks looked up if same parent inode.
1185 if (pagino
!= NULLAGINO
&&
1186 pag
->pagl_pagino
== pagino
&&
1187 pag
->pagl_leftrec
!= NULLAGINO
&&
1188 pag
->pagl_rightrec
!= NULLAGINO
) {
1189 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
1194 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
1199 /* search left with tcur, back up 1 record */
1200 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
1204 /* search right with cur, go forward 1 record. */
1205 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
1211 * Loop until we find an inode chunk with a free inode.
1213 while (!doneleft
|| !doneright
) {
1214 int useleft
; /* using left inode chunk this time */
1216 if (!--searchdistance
) {
1218 * Not in range - save last search
1219 * location and allocate a new inode
1221 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1222 pag
->pagl_leftrec
= trec
.ir_startino
;
1223 pag
->pagl_rightrec
= rec
.ir_startino
;
1224 pag
->pagl_pagino
= pagino
;
1228 /* figure out the closer block if both are valid. */
1229 if (!doneleft
&& !doneright
) {
1231 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
1232 rec
.ir_startino
- pagino
;
1234 useleft
= !doneleft
;
1237 /* free inodes to the left? */
1238 if (useleft
&& trec
.ir_freecount
) {
1240 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1243 pag
->pagl_leftrec
= trec
.ir_startino
;
1244 pag
->pagl_rightrec
= rec
.ir_startino
;
1245 pag
->pagl_pagino
= pagino
;
1249 /* free inodes to the right? */
1250 if (!useleft
&& rec
.ir_freecount
) {
1251 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1253 pag
->pagl_leftrec
= trec
.ir_startino
;
1254 pag
->pagl_rightrec
= rec
.ir_startino
;
1255 pag
->pagl_pagino
= pagino
;
1259 /* get next record to check */
1261 error
= xfs_ialloc_next_rec(tcur
, &trec
,
1264 error
= xfs_ialloc_next_rec(cur
, &rec
,
1272 * We've reached the end of the btree. because
1273 * we are only searching a small chunk of the
1274 * btree each search, there is obviously free
1275 * inodes closer to the parent inode than we
1276 * are now. restart the search again.
1278 pag
->pagl_pagino
= NULLAGINO
;
1279 pag
->pagl_leftrec
= NULLAGINO
;
1280 pag
->pagl_rightrec
= NULLAGINO
;
1281 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1282 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1287 * In a different AG from the parent.
1288 * See if the most recently allocated block has any free.
1291 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1292 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1298 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1302 if (j
== 1 && rec
.ir_freecount
> 0) {
1304 * The last chunk allocated in the group
1305 * still has a free inode.
1313 * None left in the last group, search the whole AG
1315 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1318 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1321 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1324 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1325 if (rec
.ir_freecount
> 0)
1327 error
= xfs_btree_increment(cur
, 0, &i
);
1330 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1334 offset
= xfs_inobt_first_free_inode(&rec
);
1335 ASSERT(offset
>= 0);
1336 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1337 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1338 XFS_INODES_PER_CHUNK
) == 0);
1339 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1340 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1342 error
= xfs_inobt_update(cur
, &rec
);
1345 be32_add_cpu(&agi
->agi_freecount
, -1);
1346 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1347 pag
->pagi_freecount
--;
1349 error
= xfs_check_agi_freecount(cur
, agi
);
1353 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1354 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1359 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
1361 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1367 * Use the free inode btree to allocate an inode based on distance from the
1368 * parent. Note that the provided cursor may be deleted and replaced.
1371 xfs_dialloc_ag_finobt_near(
1373 struct xfs_btree_cur
**ocur
,
1374 struct xfs_inobt_rec_incore
*rec
)
1376 struct xfs_btree_cur
*lcur
= *ocur
; /* left search cursor */
1377 struct xfs_btree_cur
*rcur
; /* right search cursor */
1378 struct xfs_inobt_rec_incore rrec
;
1382 error
= xfs_inobt_lookup(lcur
, pagino
, XFS_LOOKUP_LE
, &i
);
1387 error
= xfs_inobt_get_rec(lcur
, rec
, &i
);
1390 XFS_WANT_CORRUPTED_RETURN(lcur
->bc_mp
, i
== 1);
1393 * See if we've landed in the parent inode record. The finobt
1394 * only tracks chunks with at least one free inode, so record
1395 * existence is enough.
1397 if (pagino
>= rec
->ir_startino
&&
1398 pagino
< (rec
->ir_startino
+ XFS_INODES_PER_CHUNK
))
1402 error
= xfs_btree_dup_cursor(lcur
, &rcur
);
1406 error
= xfs_inobt_lookup(rcur
, pagino
, XFS_LOOKUP_GE
, &j
);
1410 error
= xfs_inobt_get_rec(rcur
, &rrec
, &j
);
1413 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, j
== 1, error_rcur
);
1416 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, i
== 1 || j
== 1, error_rcur
);
1417 if (i
== 1 && j
== 1) {
1419 * Both the left and right records are valid. Choose the closer
1420 * inode chunk to the target.
1422 if ((pagino
- rec
->ir_startino
+ XFS_INODES_PER_CHUNK
- 1) >
1423 (rrec
.ir_startino
- pagino
)) {
1425 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1428 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1430 } else if (j
== 1) {
1431 /* only the right record is valid */
1433 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1435 } else if (i
== 1) {
1436 /* only the left record is valid */
1437 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1443 xfs_btree_del_cursor(rcur
, XFS_BTREE_ERROR
);
1448 * Use the free inode btree to find a free inode based on a newino hint. If
1449 * the hint is NULL, find the first free inode in the AG.
1452 xfs_dialloc_ag_finobt_newino(
1453 struct xfs_agi
*agi
,
1454 struct xfs_btree_cur
*cur
,
1455 struct xfs_inobt_rec_incore
*rec
)
1460 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1461 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1466 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1469 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1475 * Find the first inode available in the AG.
1477 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1480 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1482 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1485 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1491 * Update the inobt based on a modification made to the finobt. Also ensure that
1492 * the records from both trees are equivalent post-modification.
1495 xfs_dialloc_ag_update_inobt(
1496 struct xfs_btree_cur
*cur
, /* inobt cursor */
1497 struct xfs_inobt_rec_incore
*frec
, /* finobt record */
1498 int offset
) /* inode offset */
1500 struct xfs_inobt_rec_incore rec
;
1504 error
= xfs_inobt_lookup(cur
, frec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1507 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1509 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1512 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1513 ASSERT((XFS_AGINO_TO_OFFSET(cur
->bc_mp
, rec
.ir_startino
) %
1514 XFS_INODES_PER_CHUNK
) == 0);
1516 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1519 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, (rec
.ir_free
== frec
->ir_free
) &&
1520 (rec
.ir_freecount
== frec
->ir_freecount
));
1522 return xfs_inobt_update(cur
, &rec
);
1526 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1527 * back to the inobt search algorithm.
1529 * The caller selected an AG for us, and made sure that free inodes are
1534 struct xfs_trans
*tp
,
1535 struct xfs_buf
*agbp
,
1539 struct xfs_mount
*mp
= tp
->t_mountp
;
1540 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1541 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1542 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1543 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1544 struct xfs_perag
*pag
;
1545 struct xfs_btree_cur
*cur
; /* finobt cursor */
1546 struct xfs_btree_cur
*icur
; /* inobt cursor */
1547 struct xfs_inobt_rec_incore rec
;
1553 if (!xfs_sb_version_hasfinobt(&mp
->m_sb
))
1554 return xfs_dialloc_ag_inobt(tp
, agbp
, parent
, inop
);
1556 pag
= xfs_perag_get(mp
, agno
);
1559 * If pagino is 0 (this is the root inode allocation) use newino.
1560 * This must work because we've just allocated some.
1563 pagino
= be32_to_cpu(agi
->agi_newino
);
1565 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1567 error
= xfs_check_agi_freecount(cur
, agi
);
1572 * The search algorithm depends on whether we're in the same AG as the
1573 * parent. If so, find the closest available inode to the parent. If
1574 * not, consider the agi hint or find the first free inode in the AG.
1577 error
= xfs_dialloc_ag_finobt_near(pagino
, &cur
, &rec
);
1579 error
= xfs_dialloc_ag_finobt_newino(agi
, cur
, &rec
);
1583 offset
= xfs_inobt_first_free_inode(&rec
);
1584 ASSERT(offset
>= 0);
1585 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1586 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1587 XFS_INODES_PER_CHUNK
) == 0);
1588 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1591 * Modify or remove the finobt record.
1593 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1595 if (rec
.ir_freecount
)
1596 error
= xfs_inobt_update(cur
, &rec
);
1598 error
= xfs_btree_delete(cur
, &i
);
1603 * The finobt has now been updated appropriately. We haven't updated the
1604 * agi and superblock yet, so we can create an inobt cursor and validate
1605 * the original freecount. If all is well, make the equivalent update to
1606 * the inobt using the finobt record and offset information.
1608 icur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1610 error
= xfs_check_agi_freecount(icur
, agi
);
1614 error
= xfs_dialloc_ag_update_inobt(icur
, &rec
, offset
);
1619 * Both trees have now been updated. We must update the perag and
1620 * superblock before we can check the freecount for each btree.
1622 be32_add_cpu(&agi
->agi_freecount
, -1);
1623 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1624 pag
->pagi_freecount
--;
1626 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1628 error
= xfs_check_agi_freecount(icur
, agi
);
1631 error
= xfs_check_agi_freecount(cur
, agi
);
1635 xfs_btree_del_cursor(icur
, XFS_BTREE_NOERROR
);
1636 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1642 xfs_btree_del_cursor(icur
, XFS_BTREE_ERROR
);
1644 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1650 * Allocate an inode on disk.
1652 * Mode is used to tell whether the new inode will need space, and whether it
1655 * This function is designed to be called twice if it has to do an allocation
1656 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1657 * If an inode is available without having to performn an allocation, an inode
1658 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1659 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1660 * The caller should then commit the current transaction, allocate a
1661 * new transaction, and call xfs_dialloc() again, passing in the previous value
1662 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1663 * buffer is locked across the two calls, the second call is guaranteed to have
1664 * a free inode available.
1666 * Once we successfully pick an inode its number is returned and the on-disk
1667 * data structures are updated. The inode itself is not read in, since doing so
1668 * would break ordering constraints with xfs_reclaim.
1672 struct xfs_trans
*tp
,
1676 struct xfs_buf
**IO_agbp
,
1679 struct xfs_mount
*mp
= tp
->t_mountp
;
1680 struct xfs_buf
*agbp
;
1681 xfs_agnumber_t agno
;
1685 xfs_agnumber_t start_agno
;
1686 struct xfs_perag
*pag
;
1690 * If the caller passes in a pointer to the AGI buffer,
1691 * continue where we left off before. In this case, we
1692 * know that the allocation group has free inodes.
1699 * We do not have an agbp, so select an initial allocation
1700 * group for inode allocation.
1702 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
, okalloc
);
1703 if (start_agno
== NULLAGNUMBER
) {
1709 * If we have already hit the ceiling of inode blocks then clear
1710 * okalloc so we scan all available agi structures for a free
1713 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1714 * which will sacrifice the preciseness but improve the performance.
1716 if (mp
->m_maxicount
&&
1717 percpu_counter_read_positive(&mp
->m_icount
) + mp
->m_ialloc_inos
1718 > mp
->m_maxicount
) {
1724 * Loop until we find an allocation group that either has free inodes
1725 * or in which we can allocate some inodes. Iterate through the
1726 * allocation groups upward, wrapping at the end.
1730 pag
= xfs_perag_get(mp
, agno
);
1731 if (!pag
->pagi_inodeok
) {
1732 xfs_ialloc_next_ag(mp
);
1736 if (!pag
->pagi_init
) {
1737 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1743 * Do a first racy fast path check if this AG is usable.
1745 if (!pag
->pagi_freecount
&& !okalloc
)
1749 * Then read in the AGI buffer and recheck with the AGI buffer
1752 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1756 if (pag
->pagi_freecount
) {
1762 goto nextag_relse_buffer
;
1765 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1767 xfs_trans_brelse(tp
, agbp
);
1769 if (error
!= -ENOSPC
)
1779 * We successfully allocated some inodes, return
1780 * the current context to the caller so that it
1781 * can commit the current transaction and call
1782 * us again where we left off.
1784 ASSERT(pag
->pagi_freecount
> 0);
1792 nextag_relse_buffer
:
1793 xfs_trans_brelse(tp
, agbp
);
1796 if (++agno
== mp
->m_sb
.sb_agcount
)
1798 if (agno
== start_agno
) {
1800 return noroom
? -ENOSPC
: 0;
1806 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1813 * Free the blocks of an inode chunk. We must consider that the inode chunk
1814 * might be sparse and only free the regions that are allocated as part of the
1818 xfs_difree_inode_chunk(
1819 struct xfs_mount
*mp
,
1820 xfs_agnumber_t agno
,
1821 struct xfs_inobt_rec_incore
*rec
,
1822 struct xfs_defer_ops
*dfops
)
1824 xfs_agblock_t sagbno
= XFS_AGINO_TO_AGBNO(mp
, rec
->ir_startino
);
1825 int startidx
, endidx
;
1827 xfs_agblock_t agbno
;
1829 struct xfs_owner_info oinfo
;
1830 DECLARE_BITMAP(holemask
, XFS_INOBT_HOLEMASK_BITS
);
1831 xfs_rmap_ag_owner(&oinfo
, XFS_RMAP_OWN_INODES
);
1833 if (!xfs_inobt_issparse(rec
->ir_holemask
)) {
1834 /* not sparse, calculate extent info directly */
1835 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, sagbno
),
1836 mp
->m_ialloc_blks
, &oinfo
);
1840 /* holemask is only 16-bits (fits in an unsigned long) */
1841 ASSERT(sizeof(rec
->ir_holemask
) <= sizeof(holemask
[0]));
1842 holemask
[0] = rec
->ir_holemask
;
1845 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1846 * holemask and convert the start/end index of each range to an extent.
1847 * We start with the start and end index both pointing at the first 0 in
1850 startidx
= endidx
= find_first_zero_bit(holemask
,
1851 XFS_INOBT_HOLEMASK_BITS
);
1852 nextbit
= startidx
+ 1;
1853 while (startidx
< XFS_INOBT_HOLEMASK_BITS
) {
1854 nextbit
= find_next_zero_bit(holemask
, XFS_INOBT_HOLEMASK_BITS
,
1857 * If the next zero bit is contiguous, update the end index of
1858 * the current range and continue.
1860 if (nextbit
!= XFS_INOBT_HOLEMASK_BITS
&&
1861 nextbit
== endidx
+ 1) {
1867 * nextbit is not contiguous with the current end index. Convert
1868 * the current start/end to an extent and add it to the free
1871 agbno
= sagbno
+ (startidx
* XFS_INODES_PER_HOLEMASK_BIT
) /
1872 mp
->m_sb
.sb_inopblock
;
1873 contigblk
= ((endidx
- startidx
+ 1) *
1874 XFS_INODES_PER_HOLEMASK_BIT
) /
1875 mp
->m_sb
.sb_inopblock
;
1877 ASSERT(agbno
% mp
->m_sb
.sb_spino_align
== 0);
1878 ASSERT(contigblk
% mp
->m_sb
.sb_spino_align
== 0);
1879 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, agbno
),
1882 /* reset range to current bit and carry on... */
1883 startidx
= endidx
= nextbit
;
1892 struct xfs_mount
*mp
,
1893 struct xfs_trans
*tp
,
1894 struct xfs_buf
*agbp
,
1896 struct xfs_defer_ops
*dfops
,
1897 struct xfs_icluster
*xic
,
1898 struct xfs_inobt_rec_incore
*orec
)
1900 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1901 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1902 struct xfs_perag
*pag
;
1903 struct xfs_btree_cur
*cur
;
1904 struct xfs_inobt_rec_incore rec
;
1910 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1911 ASSERT(XFS_AGINO_TO_AGBNO(mp
, agino
) < be32_to_cpu(agi
->agi_length
));
1914 * Initialize the cursor.
1916 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1918 error
= xfs_check_agi_freecount(cur
, agi
);
1923 * Look for the entry describing this inode.
1925 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1926 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1930 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1931 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1933 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1937 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1939 * Get the offset in the inode chunk.
1941 off
= agino
- rec
.ir_startino
;
1942 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1943 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1945 * Mark the inode free & increment the count.
1947 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1951 * When an inode chunk is free, it becomes eligible for removal. Don't
1952 * remove the chunk if the block size is large enough for multiple inode
1953 * chunks (that might not be free).
1955 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1956 rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
1957 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
) {
1959 xic
->first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1960 xic
->alloc
= xfs_inobt_irec_to_allocmask(&rec
);
1963 * Remove the inode cluster from the AGI B+Tree, adjust the
1964 * AGI and Superblock inode counts, and mark the disk space
1965 * to be freed when the transaction is committed.
1967 ilen
= rec
.ir_freecount
;
1968 be32_add_cpu(&agi
->agi_count
, -ilen
);
1969 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1970 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1971 pag
= xfs_perag_get(mp
, agno
);
1972 pag
->pagi_freecount
-= ilen
- 1;
1974 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1975 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1977 if ((error
= xfs_btree_delete(cur
, &i
))) {
1978 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1983 xfs_difree_inode_chunk(mp
, agno
, &rec
, dfops
);
1987 error
= xfs_inobt_update(cur
, &rec
);
1989 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
1995 * Change the inode free counts and log the ag/sb changes.
1997 be32_add_cpu(&agi
->agi_freecount
, 1);
1998 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1999 pag
= xfs_perag_get(mp
, agno
);
2000 pag
->pagi_freecount
++;
2002 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
2005 error
= xfs_check_agi_freecount(cur
, agi
);
2010 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2014 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2019 * Free an inode in the free inode btree.
2023 struct xfs_mount
*mp
,
2024 struct xfs_trans
*tp
,
2025 struct xfs_buf
*agbp
,
2027 struct xfs_inobt_rec_incore
*ibtrec
) /* inobt record */
2029 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
2030 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
2031 struct xfs_btree_cur
*cur
;
2032 struct xfs_inobt_rec_incore rec
;
2033 int offset
= agino
- ibtrec
->ir_startino
;
2037 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
2039 error
= xfs_inobt_lookup(cur
, ibtrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
2044 * If the record does not exist in the finobt, we must have just
2045 * freed an inode in a previously fully allocated chunk. If not,
2046 * something is out of sync.
2048 XFS_WANT_CORRUPTED_GOTO(mp
, ibtrec
->ir_freecount
== 1, error
);
2050 error
= xfs_inobt_insert_rec(cur
, ibtrec
->ir_holemask
,
2052 ibtrec
->ir_freecount
,
2053 ibtrec
->ir_free
, &i
);
2062 * Read and update the existing record. We could just copy the ibtrec
2063 * across here, but that would defeat the purpose of having redundant
2064 * metadata. By making the modifications independently, we can catch
2065 * corruptions that we wouldn't see if we just copied from one record
2068 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2071 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
2073 rec
.ir_free
|= XFS_INOBT_MASK(offset
);
2076 XFS_WANT_CORRUPTED_GOTO(mp
, (rec
.ir_free
== ibtrec
->ir_free
) &&
2077 (rec
.ir_freecount
== ibtrec
->ir_freecount
),
2081 * The content of inobt records should always match between the inobt
2082 * and finobt. The lifecycle of records in the finobt is different from
2083 * the inobt in that the finobt only tracks records with at least one
2084 * free inode. Hence, if all of the inodes are free and we aren't
2085 * keeping inode chunks permanently on disk, remove the record.
2086 * Otherwise, update the record with the new information.
2088 * Note that we currently can't free chunks when the block size is large
2089 * enough for multiple chunks. Leave the finobt record to remain in sync
2092 if (rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
2093 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
&&
2094 !(mp
->m_flags
& XFS_MOUNT_IKEEP
)) {
2095 error
= xfs_btree_delete(cur
, &i
);
2100 error
= xfs_inobt_update(cur
, &rec
);
2106 error
= xfs_check_agi_freecount(cur
, agi
);
2110 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2114 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2119 * Free disk inode. Carefully avoids touching the incore inode, all
2120 * manipulations incore are the caller's responsibility.
2121 * The on-disk inode is not changed by this operation, only the
2122 * btree (free inode mask) is changed.
2126 struct xfs_trans
*tp
, /* transaction pointer */
2127 xfs_ino_t inode
, /* inode to be freed */
2128 struct xfs_defer_ops
*dfops
, /* extents to free */
2129 struct xfs_icluster
*xic
) /* cluster info if deleted */
2132 xfs_agblock_t agbno
; /* block number containing inode */
2133 struct xfs_buf
*agbp
; /* buffer for allocation group header */
2134 xfs_agino_t agino
; /* allocation group inode number */
2135 xfs_agnumber_t agno
; /* allocation group number */
2136 int error
; /* error return value */
2137 struct xfs_mount
*mp
; /* mount structure for filesystem */
2138 struct xfs_inobt_rec_incore rec
;/* btree record */
2143 * Break up inode number into its components.
2145 agno
= XFS_INO_TO_AGNO(mp
, inode
);
2146 if (agno
>= mp
->m_sb
.sb_agcount
) {
2147 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2148 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2152 agino
= XFS_INO_TO_AGINO(mp
, inode
);
2153 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2154 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2155 __func__
, (unsigned long long)inode
,
2156 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
2160 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2161 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2162 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2163 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
2168 * Get the allocation group header.
2170 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2172 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
2178 * Fix up the inode allocation btree.
2180 error
= xfs_difree_inobt(mp
, tp
, agbp
, agino
, dfops
, xic
, &rec
);
2185 * Fix up the free inode btree.
2187 if (xfs_sb_version_hasfinobt(&mp
->m_sb
)) {
2188 error
= xfs_difree_finobt(mp
, tp
, agbp
, agino
, &rec
);
2201 struct xfs_mount
*mp
,
2202 struct xfs_trans
*tp
,
2203 xfs_agnumber_t agno
,
2205 xfs_agblock_t agbno
,
2206 xfs_agblock_t
*chunk_agbno
,
2207 xfs_agblock_t
*offset_agbno
,
2210 struct xfs_inobt_rec_incore rec
;
2211 struct xfs_btree_cur
*cur
;
2212 struct xfs_buf
*agbp
;
2216 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2219 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2220 __func__
, error
, agno
);
2225 * Lookup the inode record for the given agino. If the record cannot be
2226 * found, then it's an invalid inode number and we should abort. Once
2227 * we have a record, we need to ensure it contains the inode number
2228 * we are looking up.
2230 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
2231 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
2234 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2235 if (!error
&& i
== 0)
2239 xfs_trans_brelse(tp
, agbp
);
2240 xfs_btree_del_cursor(cur
, error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
2244 /* check that the returned record contains the required inode */
2245 if (rec
.ir_startino
> agino
||
2246 rec
.ir_startino
+ mp
->m_ialloc_inos
<= agino
)
2249 /* for untrusted inodes check it is allocated first */
2250 if ((flags
& XFS_IGET_UNTRUSTED
) &&
2251 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
2254 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
2255 *offset_agbno
= agbno
- *chunk_agbno
;
2260 * Return the location of the inode in imap, for mapping it into a buffer.
2264 xfs_mount_t
*mp
, /* file system mount structure */
2265 xfs_trans_t
*tp
, /* transaction pointer */
2266 xfs_ino_t ino
, /* inode to locate */
2267 struct xfs_imap
*imap
, /* location map structure */
2268 uint flags
) /* flags for inode btree lookup */
2270 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
2271 xfs_agino_t agino
; /* inode number within alloc group */
2272 xfs_agnumber_t agno
; /* allocation group number */
2273 int blks_per_cluster
; /* num blocks per inode cluster */
2274 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
2275 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
2276 int error
; /* error code */
2277 int offset
; /* index of inode in its buffer */
2278 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
2280 ASSERT(ino
!= NULLFSINO
);
2283 * Split up the inode number into its parts.
2285 agno
= XFS_INO_TO_AGNO(mp
, ino
);
2286 agino
= XFS_INO_TO_AGINO(mp
, ino
);
2287 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2288 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
2289 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2292 * Don't output diagnostic information for untrusted inodes
2293 * as they can be invalid without implying corruption.
2295 if (flags
& XFS_IGET_UNTRUSTED
)
2297 if (agno
>= mp
->m_sb
.sb_agcount
) {
2299 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2300 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2302 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2304 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2305 __func__
, (unsigned long long)agbno
,
2306 (unsigned long)mp
->m_sb
.sb_agblocks
);
2308 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2310 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2312 XFS_AGINO_TO_INO(mp
, agno
, agino
));
2319 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2322 * For bulkstat and handle lookups, we have an untrusted inode number
2323 * that we have to verify is valid. We cannot do this just by reading
2324 * the inode buffer as it may have been unlinked and removed leaving
2325 * inodes in stale state on disk. Hence we have to do a btree lookup
2326 * in all cases where an untrusted inode number is passed.
2328 if (flags
& XFS_IGET_UNTRUSTED
) {
2329 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2330 &chunk_agbno
, &offset_agbno
, flags
);
2337 * If the inode cluster size is the same as the blocksize or
2338 * smaller we get to the buffer by simple arithmetics.
2340 if (blks_per_cluster
== 1) {
2341 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
2342 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
2344 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
2345 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
2346 imap
->im_boffset
= (unsigned short)(offset
<<
2347 mp
->m_sb
.sb_inodelog
);
2352 * If the inode chunks are aligned then use simple maths to
2353 * find the location. Otherwise we have to do a btree
2354 * lookup to find the location.
2356 if (mp
->m_inoalign_mask
) {
2357 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
2358 chunk_agbno
= agbno
- offset_agbno
;
2360 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2361 &chunk_agbno
, &offset_agbno
, flags
);
2367 ASSERT(agbno
>= chunk_agbno
);
2368 cluster_agbno
= chunk_agbno
+
2369 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
2370 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
2371 XFS_INO_TO_OFFSET(mp
, ino
);
2373 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
2374 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
2375 imap
->im_boffset
= (unsigned short)(offset
<< mp
->m_sb
.sb_inodelog
);
2378 * If the inode number maps to a block outside the bounds
2379 * of the file system then return NULL rather than calling
2380 * read_buf and panicing when we get an error from the
2383 if ((imap
->im_blkno
+ imap
->im_len
) >
2384 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
2386 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2387 __func__
, (unsigned long long) imap
->im_blkno
,
2388 (unsigned long long) imap
->im_len
,
2389 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
2396 * Compute and fill in value of m_in_maxlevels.
2399 xfs_ialloc_compute_maxlevels(
2400 xfs_mount_t
*mp
) /* file system mount structure */
2404 inodes
= (1LL << XFS_INO_AGINO_BITS(mp
)) >> XFS_INODES_PER_CHUNK_LOG
;
2405 mp
->m_in_maxlevels
= xfs_btree_compute_maxlevels(mp
, mp
->m_inobt_mnr
,
2410 * Log specified fields for the ag hdr (inode section). The growth of the agi
2411 * structure over time requires that we interpret the buffer as two logical
2412 * regions delineated by the end of the unlinked list. This is due to the size
2413 * of the hash table and its location in the middle of the agi.
2415 * For example, a request to log a field before agi_unlinked and a field after
2416 * agi_unlinked could cause us to log the entire hash table and use an excessive
2417 * amount of log space. To avoid this behavior, log the region up through
2418 * agi_unlinked in one call and the region after agi_unlinked through the end of
2419 * the structure in another.
2423 xfs_trans_t
*tp
, /* transaction pointer */
2424 xfs_buf_t
*bp
, /* allocation group header buffer */
2425 int fields
) /* bitmask of fields to log */
2427 int first
; /* first byte number */
2428 int last
; /* last byte number */
2429 static const short offsets
[] = { /* field starting offsets */
2430 /* keep in sync with bit definitions */
2431 offsetof(xfs_agi_t
, agi_magicnum
),
2432 offsetof(xfs_agi_t
, agi_versionnum
),
2433 offsetof(xfs_agi_t
, agi_seqno
),
2434 offsetof(xfs_agi_t
, agi_length
),
2435 offsetof(xfs_agi_t
, agi_count
),
2436 offsetof(xfs_agi_t
, agi_root
),
2437 offsetof(xfs_agi_t
, agi_level
),
2438 offsetof(xfs_agi_t
, agi_freecount
),
2439 offsetof(xfs_agi_t
, agi_newino
),
2440 offsetof(xfs_agi_t
, agi_dirino
),
2441 offsetof(xfs_agi_t
, agi_unlinked
),
2442 offsetof(xfs_agi_t
, agi_free_root
),
2443 offsetof(xfs_agi_t
, agi_free_level
),
2447 xfs_agi_t
*agi
; /* allocation group header */
2449 agi
= XFS_BUF_TO_AGI(bp
);
2450 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
2454 * Compute byte offsets for the first and last fields in the first
2455 * region and log the agi buffer. This only logs up through
2458 if (fields
& XFS_AGI_ALL_BITS_R1
) {
2459 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R1
,
2461 xfs_trans_log_buf(tp
, bp
, first
, last
);
2465 * Mask off the bits in the first region and calculate the first and
2466 * last field offsets for any bits in the second region.
2468 fields
&= ~XFS_AGI_ALL_BITS_R1
;
2470 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R2
,
2472 xfs_trans_log_buf(tp
, bp
, first
, last
);
2478 xfs_check_agi_unlinked(
2479 struct xfs_agi
*agi
)
2483 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++)
2484 ASSERT(agi
->agi_unlinked
[i
]);
2487 #define xfs_check_agi_unlinked(agi)
2494 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2495 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
2497 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2498 if (!uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_meta_uuid
))
2500 if (!xfs_log_check_lsn(mp
,
2501 be64_to_cpu(XFS_BUF_TO_AGI(bp
)->agi_lsn
)))
2506 * Validate the magic number of the agi block.
2508 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
2510 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
2513 if (be32_to_cpu(agi
->agi_level
) < 1 ||
2514 be32_to_cpu(agi
->agi_level
) > XFS_BTREE_MAXLEVELS
)
2517 if (xfs_sb_version_hasfinobt(&mp
->m_sb
) &&
2518 (be32_to_cpu(agi
->agi_free_level
) < 1 ||
2519 be32_to_cpu(agi
->agi_free_level
) > XFS_BTREE_MAXLEVELS
))
2523 * during growfs operations, the perag is not fully initialised,
2524 * so we can't use it for any useful checking. growfs ensures we can't
2525 * use it by using uncached buffers that don't have the perag attached
2526 * so we can detect and avoid this problem.
2528 if (bp
->b_pag
&& be32_to_cpu(agi
->agi_seqno
) != bp
->b_pag
->pag_agno
)
2531 xfs_check_agi_unlinked(agi
);
2536 xfs_agi_read_verify(
2539 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2541 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2542 !xfs_buf_verify_cksum(bp
, XFS_AGI_CRC_OFF
))
2543 xfs_buf_ioerror(bp
, -EFSBADCRC
);
2544 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp
), mp
,
2545 XFS_ERRTAG_IALLOC_READ_AGI
,
2546 XFS_RANDOM_IALLOC_READ_AGI
))
2547 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2550 xfs_verifier_error(bp
);
2554 xfs_agi_write_verify(
2557 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2558 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
2560 if (!xfs_agi_verify(bp
)) {
2561 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2562 xfs_verifier_error(bp
);
2566 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2570 XFS_BUF_TO_AGI(bp
)->agi_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2571 xfs_buf_update_cksum(bp
, XFS_AGI_CRC_OFF
);
2574 const struct xfs_buf_ops xfs_agi_buf_ops
= {
2576 .verify_read
= xfs_agi_read_verify
,
2577 .verify_write
= xfs_agi_write_verify
,
2581 * Read in the allocation group header (inode allocation section)
2585 struct xfs_mount
*mp
, /* file system mount structure */
2586 struct xfs_trans
*tp
, /* transaction pointer */
2587 xfs_agnumber_t agno
, /* allocation group number */
2588 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2592 trace_xfs_read_agi(mp
, agno
);
2594 ASSERT(agno
!= NULLAGNUMBER
);
2595 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
2596 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
2597 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
2601 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_AGI_BUF
);
2603 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
2608 xfs_ialloc_read_agi(
2609 struct xfs_mount
*mp
, /* file system mount structure */
2610 struct xfs_trans
*tp
, /* transaction pointer */
2611 xfs_agnumber_t agno
, /* allocation group number */
2612 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2614 struct xfs_agi
*agi
; /* allocation group header */
2615 struct xfs_perag
*pag
; /* per allocation group data */
2618 trace_xfs_ialloc_read_agi(mp
, agno
);
2620 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
2624 agi
= XFS_BUF_TO_AGI(*bpp
);
2625 pag
= xfs_perag_get(mp
, agno
);
2626 if (!pag
->pagi_init
) {
2627 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
2628 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
2633 * It's possible for these to be out of sync if
2634 * we are in the middle of a forced shutdown.
2636 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
2637 XFS_FORCED_SHUTDOWN(mp
));
2643 * Read in the agi to initialise the per-ag data in the mount structure
2646 xfs_ialloc_pagi_init(
2647 xfs_mount_t
*mp
, /* file system mount structure */
2648 xfs_trans_t
*tp
, /* transaction pointer */
2649 xfs_agnumber_t agno
) /* allocation group number */
2651 xfs_buf_t
*bp
= NULL
;
2654 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
2658 xfs_trans_brelse(tp
, bp
);