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_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_ialloc.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_alloc.h"
32 #include "xfs_rtalloc.h"
33 #include "xfs_error.h"
35 #include "xfs_cksum.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_icreate_item.h"
39 #include "xfs_icache.h"
40 #include "xfs_trace.h"
44 * Allocation group level functions.
47 xfs_ialloc_cluster_alignment(
50 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
51 mp
->m_sb
.sb_inoalignmt
>=
52 XFS_B_TO_FSBT(mp
, mp
->m_inode_cluster_size
))
53 return mp
->m_sb
.sb_inoalignmt
;
58 * Lookup a record by ino in the btree given by cur.
62 struct xfs_btree_cur
*cur
, /* btree cursor */
63 xfs_agino_t ino
, /* starting inode of chunk */
64 xfs_lookup_t dir
, /* <=, >=, == */
65 int *stat
) /* success/failure */
67 cur
->bc_rec
.i
.ir_startino
= ino
;
68 cur
->bc_rec
.i
.ir_holemask
= 0;
69 cur
->bc_rec
.i
.ir_count
= 0;
70 cur
->bc_rec
.i
.ir_freecount
= 0;
71 cur
->bc_rec
.i
.ir_free
= 0;
72 return xfs_btree_lookup(cur
, dir
, stat
);
76 * Update the record referred to by cur to the value given.
77 * This either works (return 0) or gets an EFSCORRUPTED error.
79 STATIC
int /* error */
81 struct xfs_btree_cur
*cur
, /* btree cursor */
82 xfs_inobt_rec_incore_t
*irec
) /* btree record */
84 union xfs_btree_rec rec
;
86 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
87 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
88 rec
.inobt
.ir_u
.sp
.ir_holemask
= cpu_to_be16(irec
->ir_holemask
);
89 rec
.inobt
.ir_u
.sp
.ir_count
= irec
->ir_count
;
90 rec
.inobt
.ir_u
.sp
.ir_freecount
= irec
->ir_freecount
;
92 /* ir_holemask/ir_count not supported on-disk */
93 rec
.inobt
.ir_u
.f
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
95 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
96 return xfs_btree_update(cur
, &rec
);
100 * Get the data from the pointed-to record.
104 struct xfs_btree_cur
*cur
, /* btree cursor */
105 xfs_inobt_rec_incore_t
*irec
, /* btree record */
106 int *stat
) /* output: success/failure */
108 union xfs_btree_rec
*rec
;
111 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
112 if (error
|| *stat
== 0)
115 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
116 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
117 irec
->ir_holemask
= be16_to_cpu(rec
->inobt
.ir_u
.sp
.ir_holemask
);
118 irec
->ir_count
= rec
->inobt
.ir_u
.sp
.ir_count
;
119 irec
->ir_freecount
= rec
->inobt
.ir_u
.sp
.ir_freecount
;
122 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
123 * values for full inode chunks.
125 irec
->ir_holemask
= XFS_INOBT_HOLEMASK_FULL
;
126 irec
->ir_count
= XFS_INODES_PER_CHUNK
;
128 be32_to_cpu(rec
->inobt
.ir_u
.f
.ir_freecount
);
130 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
136 * Insert a single inobt record. Cursor must already point to desired location.
139 xfs_inobt_insert_rec(
140 struct xfs_btree_cur
*cur
,
147 cur
->bc_rec
.i
.ir_holemask
= holemask
;
148 cur
->bc_rec
.i
.ir_count
= count
;
149 cur
->bc_rec
.i
.ir_freecount
= freecount
;
150 cur
->bc_rec
.i
.ir_free
= free
;
151 return xfs_btree_insert(cur
, stat
);
155 * Insert records describing a newly allocated inode chunk into the inobt.
159 struct xfs_mount
*mp
,
160 struct xfs_trans
*tp
,
161 struct xfs_buf
*agbp
,
166 struct xfs_btree_cur
*cur
;
167 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
168 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
173 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
175 for (thisino
= newino
;
176 thisino
< newino
+ newlen
;
177 thisino
+= XFS_INODES_PER_CHUNK
) {
178 error
= xfs_inobt_lookup(cur
, thisino
, XFS_LOOKUP_EQ
, &i
);
180 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
185 error
= xfs_inobt_insert_rec(cur
, XFS_INOBT_HOLEMASK_FULL
,
186 XFS_INODES_PER_CHUNK
,
187 XFS_INODES_PER_CHUNK
,
188 XFS_INOBT_ALL_FREE
, &i
);
190 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
196 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
202 * Verify that the number of free inodes in the AGI is correct.
206 xfs_check_agi_freecount(
207 struct xfs_btree_cur
*cur
,
210 if (cur
->bc_nlevels
== 1) {
211 xfs_inobt_rec_incore_t rec
;
216 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
221 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
226 freecount
+= rec
.ir_freecount
;
227 error
= xfs_btree_increment(cur
, 0, &i
);
233 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
234 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
239 #define xfs_check_agi_freecount(cur, agi) 0
243 * Initialise a new set of inodes. When called without a transaction context
244 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
245 * than logging them (which in a transaction context puts them into the AIL
246 * for writeback rather than the xfsbufd queue).
249 xfs_ialloc_inode_init(
250 struct xfs_mount
*mp
,
251 struct xfs_trans
*tp
,
252 struct list_head
*buffer_list
,
256 xfs_agblock_t length
,
259 struct xfs_buf
*fbuf
;
260 struct xfs_dinode
*free
;
261 int nbufs
, blks_per_cluster
, inodes_per_cluster
;
268 * Loop over the new block(s), filling in the inodes. For small block
269 * sizes, manipulate the inodes in buffers which are multiples of the
272 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
273 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
274 nbufs
= length
/ blks_per_cluster
;
277 * Figure out what version number to use in the inodes we create. If
278 * the superblock version has caught up to the one that supports the new
279 * inode format, then use the new inode version. Otherwise use the old
280 * version so that old kernels will continue to be able to use the file
283 * For v3 inodes, we also need to write the inode number into the inode,
284 * so calculate the first inode number of the chunk here as
285 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
286 * across multiple filesystem blocks (such as a cluster) and so cannot
287 * be used in the cluster buffer loop below.
289 * Further, because we are writing the inode directly into the buffer
290 * and calculating a CRC on the entire inode, we have ot log the entire
291 * inode so that the entire range the CRC covers is present in the log.
292 * That means for v3 inode we log the entire buffer rather than just the
295 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
297 ino
= XFS_AGINO_TO_INO(mp
, agno
,
298 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
301 * log the initialisation that is about to take place as an
302 * logical operation. This means the transaction does not
303 * need to log the physical changes to the inode buffers as log
304 * recovery will know what initialisation is actually needed.
305 * Hence we only need to log the buffers as "ordered" buffers so
306 * they track in the AIL as if they were physically logged.
309 xfs_icreate_log(tp
, agno
, agbno
, icount
,
310 mp
->m_sb
.sb_inodesize
, length
, gen
);
314 for (j
= 0; j
< nbufs
; j
++) {
318 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
319 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
320 mp
->m_bsize
* blks_per_cluster
,
325 /* Initialize the inode buffers and log them appropriately. */
326 fbuf
->b_ops
= &xfs_inode_buf_ops
;
327 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
328 for (i
= 0; i
< inodes_per_cluster
; i
++) {
329 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
330 uint isize
= xfs_dinode_size(version
);
332 free
= xfs_make_iptr(mp
, fbuf
, i
);
333 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
334 free
->di_version
= version
;
335 free
->di_gen
= cpu_to_be32(gen
);
336 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
339 free
->di_ino
= cpu_to_be64(ino
);
341 uuid_copy(&free
->di_uuid
,
342 &mp
->m_sb
.sb_meta_uuid
);
343 xfs_dinode_calc_crc(mp
, free
);
345 /* just log the inode core */
346 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
347 ioffset
+ isize
- 1);
353 * Mark the buffer as an inode allocation buffer so it
354 * sticks in AIL at the point of this allocation
355 * transaction. This ensures the they are on disk before
356 * the tail of the log can be moved past this
357 * transaction (i.e. by preventing relogging from moving
358 * it forward in the log).
360 xfs_trans_inode_alloc_buf(tp
, fbuf
);
363 * Mark the buffer as ordered so that they are
364 * not physically logged in the transaction but
365 * still tracked in the AIL as part of the
366 * transaction and pin the log appropriately.
368 xfs_trans_ordered_buf(tp
, fbuf
);
369 xfs_trans_log_buf(tp
, fbuf
, 0,
370 BBTOB(fbuf
->b_length
) - 1);
373 fbuf
->b_flags
|= XBF_DONE
;
374 xfs_buf_delwri_queue(fbuf
, buffer_list
);
382 * Align startino and allocmask for a recently allocated sparse chunk such that
383 * they are fit for insertion (or merge) into the on-disk inode btrees.
387 * When enabled, sparse inode support increases the inode alignment from cluster
388 * size to inode chunk size. This means that the minimum range between two
389 * non-adjacent inode records in the inobt is large enough for a full inode
390 * record. This allows for cluster sized, cluster aligned block allocation
391 * without need to worry about whether the resulting inode record overlaps with
392 * another record in the tree. Without this basic rule, we would have to deal
393 * with the consequences of overlap by potentially undoing recent allocations in
394 * the inode allocation codepath.
396 * Because of this alignment rule (which is enforced on mount), there are two
397 * inobt possibilities for newly allocated sparse chunks. One is that the
398 * aligned inode record for the chunk covers a range of inodes not already
399 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
400 * other is that a record already exists at the aligned startino that considers
401 * the newly allocated range as sparse. In the latter case, record content is
402 * merged in hope that sparse inode chunks fill to full chunks over time.
405 xfs_align_sparse_ino(
406 struct xfs_mount
*mp
,
407 xfs_agino_t
*startino
,
414 agbno
= XFS_AGINO_TO_AGBNO(mp
, *startino
);
415 mod
= agbno
% mp
->m_sb
.sb_inoalignmt
;
419 /* calculate the inode offset and align startino */
420 offset
= mod
<< mp
->m_sb
.sb_inopblog
;
424 * Since startino has been aligned down, left shift allocmask such that
425 * it continues to represent the same physical inodes relative to the
428 *allocmask
<<= offset
/ XFS_INODES_PER_HOLEMASK_BIT
;
432 * Determine whether the source inode record can merge into the target. Both
433 * records must be sparse, the inode ranges must match and there must be no
434 * allocation overlap between the records.
437 __xfs_inobt_can_merge(
438 struct xfs_inobt_rec_incore
*trec
, /* tgt record */
439 struct xfs_inobt_rec_incore
*srec
) /* src record */
444 /* records must cover the same inode range */
445 if (trec
->ir_startino
!= srec
->ir_startino
)
448 /* both records must be sparse */
449 if (!xfs_inobt_issparse(trec
->ir_holemask
) ||
450 !xfs_inobt_issparse(srec
->ir_holemask
))
453 /* both records must track some inodes */
454 if (!trec
->ir_count
|| !srec
->ir_count
)
457 /* can't exceed capacity of a full record */
458 if (trec
->ir_count
+ srec
->ir_count
> XFS_INODES_PER_CHUNK
)
461 /* verify there is no allocation overlap */
462 talloc
= xfs_inobt_irec_to_allocmask(trec
);
463 salloc
= xfs_inobt_irec_to_allocmask(srec
);
471 * Merge the source inode record into the target. The caller must call
472 * __xfs_inobt_can_merge() to ensure the merge is valid.
475 __xfs_inobt_rec_merge(
476 struct xfs_inobt_rec_incore
*trec
, /* target */
477 struct xfs_inobt_rec_incore
*srec
) /* src */
479 ASSERT(trec
->ir_startino
== srec
->ir_startino
);
481 /* combine the counts */
482 trec
->ir_count
+= srec
->ir_count
;
483 trec
->ir_freecount
+= srec
->ir_freecount
;
486 * Merge the holemask and free mask. For both fields, 0 bits refer to
487 * allocated inodes. We combine the allocated ranges with bitwise AND.
489 trec
->ir_holemask
&= srec
->ir_holemask
;
490 trec
->ir_free
&= srec
->ir_free
;
494 * Insert a new sparse inode chunk into the associated inode btree. The inode
495 * record for the sparse chunk is pre-aligned to a startino that should match
496 * any pre-existing sparse inode record in the tree. This allows sparse chunks
499 * This function supports two modes of handling preexisting records depending on
500 * the merge flag. If merge is true, the provided record is merged with the
501 * existing record and updated in place. The merged record is returned in nrec.
502 * If merge is false, an existing record is replaced with the provided record.
503 * If no preexisting record exists, the provided record is always inserted.
505 * It is considered corruption if a merge is requested and not possible. Given
506 * the sparse inode alignment constraints, this should never happen.
509 xfs_inobt_insert_sprec(
510 struct xfs_mount
*mp
,
511 struct xfs_trans
*tp
,
512 struct xfs_buf
*agbp
,
514 struct xfs_inobt_rec_incore
*nrec
, /* in/out: new/merged rec. */
515 bool merge
) /* merge or replace */
517 struct xfs_btree_cur
*cur
;
518 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
519 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
522 struct xfs_inobt_rec_incore rec
;
524 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
526 /* the new record is pre-aligned so we know where to look */
527 error
= xfs_inobt_lookup(cur
, nrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
530 /* if nothing there, insert a new record and return */
532 error
= xfs_inobt_insert_rec(cur
, nrec
->ir_holemask
,
533 nrec
->ir_count
, nrec
->ir_freecount
,
537 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
543 * A record exists at this startino. Merge or replace the record
544 * depending on what we've been asked to do.
547 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
550 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
551 XFS_WANT_CORRUPTED_GOTO(mp
,
552 rec
.ir_startino
== nrec
->ir_startino
,
556 * This should never fail. If we have coexisting records that
557 * cannot merge, something is seriously wrong.
559 XFS_WANT_CORRUPTED_GOTO(mp
, __xfs_inobt_can_merge(nrec
, &rec
),
562 trace_xfs_irec_merge_pre(mp
, agno
, rec
.ir_startino
,
563 rec
.ir_holemask
, nrec
->ir_startino
,
566 /* merge to nrec to output the updated record */
567 __xfs_inobt_rec_merge(nrec
, &rec
);
569 trace_xfs_irec_merge_post(mp
, agno
, nrec
->ir_startino
,
572 error
= xfs_inobt_rec_check_count(mp
, nrec
);
577 error
= xfs_inobt_update(cur
, nrec
);
582 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
585 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
590 * Allocate new inodes in the allocation group specified by agbp.
591 * Return 0 for success, else error code.
593 STATIC
int /* error code or 0 */
595 xfs_trans_t
*tp
, /* transaction pointer */
596 xfs_buf_t
*agbp
, /* alloc group buffer */
599 xfs_agi_t
*agi
; /* allocation group header */
600 xfs_alloc_arg_t args
; /* allocation argument structure */
603 xfs_agino_t newino
; /* new first inode's number */
604 xfs_agino_t newlen
; /* new number of inodes */
605 int isaligned
= 0; /* inode allocation at stripe unit */
607 uint16_t allocmask
= (uint16_t) -1; /* init. to full chunk */
608 struct xfs_inobt_rec_incore rec
;
609 struct xfs_perag
*pag
;
612 memset(&args
, 0, sizeof(args
));
614 args
.mp
= tp
->t_mountp
;
615 args
.fsbno
= NULLFSBLOCK
;
618 /* randomly do sparse inode allocations */
619 if (xfs_sb_version_hassparseinodes(&tp
->t_mountp
->m_sb
) &&
620 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
)
621 do_sparse
= prandom_u32() & 1;
625 * Locking will ensure that we don't have two callers in here
628 newlen
= args
.mp
->m_ialloc_inos
;
629 if (args
.mp
->m_maxicount
&&
630 percpu_counter_read_positive(&args
.mp
->m_icount
) + newlen
>
631 args
.mp
->m_maxicount
)
633 args
.minlen
= args
.maxlen
= args
.mp
->m_ialloc_blks
;
635 * First try to allocate inodes contiguous with the last-allocated
636 * chunk of inodes. If the filesystem is striped, this will fill
637 * an entire stripe unit with inodes.
639 agi
= XFS_BUF_TO_AGI(agbp
);
640 newino
= be32_to_cpu(agi
->agi_newino
);
641 agno
= be32_to_cpu(agi
->agi_seqno
);
642 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
643 args
.mp
->m_ialloc_blks
;
646 if (likely(newino
!= NULLAGINO
&&
647 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
648 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
649 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
653 * We need to take into account alignment here to ensure that
654 * we don't modify the free list if we fail to have an exact
655 * block. If we don't have an exact match, and every oher
656 * attempt allocation attempt fails, we'll end up cancelling
657 * a dirty transaction and shutting down.
659 * For an exact allocation, alignment must be 1,
660 * however we need to take cluster alignment into account when
661 * fixing up the freelist. Use the minalignslop field to
662 * indicate that extra blocks might be required for alignment,
663 * but not to use them in the actual exact allocation.
666 args
.minalignslop
= xfs_ialloc_cluster_alignment(args
.mp
) - 1;
668 /* Allow space for the inode btree to split. */
669 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
670 if ((error
= xfs_alloc_vextent(&args
)))
674 * This request might have dirtied the transaction if the AG can
675 * satisfy the request, but the exact block was not available.
676 * If the allocation did fail, subsequent requests will relax
677 * the exact agbno requirement and increase the alignment
678 * instead. It is critical that the total size of the request
679 * (len + alignment + slop) does not increase from this point
680 * on, so reset minalignslop to ensure it is not included in
681 * subsequent requests.
683 args
.minalignslop
= 0;
686 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
688 * Set the alignment for the allocation.
689 * If stripe alignment is turned on then align at stripe unit
691 * If the cluster size is smaller than a filesystem block
692 * then we're doing I/O for inodes in filesystem block size
693 * pieces, so don't need alignment anyway.
696 if (args
.mp
->m_sinoalign
) {
697 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
698 args
.alignment
= args
.mp
->m_dalign
;
701 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
703 * Need to figure out where to allocate the inode blocks.
704 * Ideally they should be spaced out through the a.g.
705 * For now, just allocate blocks up front.
707 args
.agbno
= be32_to_cpu(agi
->agi_root
);
708 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
710 * Allocate a fixed-size extent of inodes.
712 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
715 * Allow space for the inode btree to split.
717 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
718 if ((error
= xfs_alloc_vextent(&args
)))
723 * If stripe alignment is turned on, then try again with cluster
726 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
727 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
728 args
.agbno
= be32_to_cpu(agi
->agi_root
);
729 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
730 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
731 if ((error
= xfs_alloc_vextent(&args
)))
736 * Finally, try a sparse allocation if the filesystem supports it and
737 * the sparse allocation length is smaller than a full chunk.
739 if (xfs_sb_version_hassparseinodes(&args
.mp
->m_sb
) &&
740 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
&&
741 args
.fsbno
== NULLFSBLOCK
) {
743 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
744 args
.agbno
= be32_to_cpu(agi
->agi_root
);
745 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
746 args
.alignment
= args
.mp
->m_sb
.sb_spino_align
;
749 args
.minlen
= args
.mp
->m_ialloc_min_blks
;
750 args
.maxlen
= args
.minlen
;
753 * The inode record will be aligned to full chunk size. We must
754 * prevent sparse allocation from AG boundaries that result in
755 * invalid inode records, such as records that start at agbno 0
756 * or extend beyond the AG.
758 * Set min agbno to the first aligned, non-zero agbno and max to
759 * the last aligned agbno that is at least one full chunk from
762 args
.min_agbno
= args
.mp
->m_sb
.sb_inoalignmt
;
763 args
.max_agbno
= round_down(args
.mp
->m_sb
.sb_agblocks
,
764 args
.mp
->m_sb
.sb_inoalignmt
) -
765 args
.mp
->m_ialloc_blks
;
767 error
= xfs_alloc_vextent(&args
);
771 newlen
= args
.len
<< args
.mp
->m_sb
.sb_inopblog
;
772 ASSERT(newlen
<= XFS_INODES_PER_CHUNK
);
773 allocmask
= (1 << (newlen
/ XFS_INODES_PER_HOLEMASK_BIT
)) - 1;
776 if (args
.fsbno
== NULLFSBLOCK
) {
780 ASSERT(args
.len
== args
.minlen
);
783 * Stamp and write the inode buffers.
785 * Seed the new inode cluster with a random generation number. This
786 * prevents short-term reuse of generation numbers if a chunk is
787 * freed and then immediately reallocated. We use random numbers
788 * rather than a linear progression to prevent the next generation
789 * number from being easily guessable.
791 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, newlen
, agno
,
792 args
.agbno
, args
.len
, prandom_u32());
797 * Convert the results.
799 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
801 if (xfs_inobt_issparse(~allocmask
)) {
803 * We've allocated a sparse chunk. Align the startino and mask.
805 xfs_align_sparse_ino(args
.mp
, &newino
, &allocmask
);
807 rec
.ir_startino
= newino
;
808 rec
.ir_holemask
= ~allocmask
;
809 rec
.ir_count
= newlen
;
810 rec
.ir_freecount
= newlen
;
811 rec
.ir_free
= XFS_INOBT_ALL_FREE
;
814 * Insert the sparse record into the inobt and allow for a merge
815 * if necessary. If a merge does occur, rec is updated to the
818 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
, XFS_BTNUM_INO
,
820 if (error
== -EFSCORRUPTED
) {
822 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
823 XFS_AGINO_TO_INO(args
.mp
, agno
,
825 rec
.ir_holemask
, rec
.ir_count
);
826 xfs_force_shutdown(args
.mp
, SHUTDOWN_CORRUPT_INCORE
);
832 * We can't merge the part we've just allocated as for the inobt
833 * due to finobt semantics. The original record may or may not
834 * exist independent of whether physical inodes exist in this
837 * We must update the finobt record based on the inobt record.
838 * rec contains the fully merged and up to date inobt record
839 * from the previous call. Set merge false to replace any
840 * existing record with this one.
842 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
843 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
,
844 XFS_BTNUM_FINO
, &rec
,
850 /* full chunk - insert new records to both btrees */
851 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
856 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
857 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
,
858 newlen
, XFS_BTNUM_FINO
);
865 * Update AGI counts and newino.
867 be32_add_cpu(&agi
->agi_count
, newlen
);
868 be32_add_cpu(&agi
->agi_freecount
, newlen
);
869 pag
= xfs_perag_get(args
.mp
, agno
);
870 pag
->pagi_freecount
+= newlen
;
872 agi
->agi_newino
= cpu_to_be32(newino
);
875 * Log allocation group header fields
877 xfs_ialloc_log_agi(tp
, agbp
,
878 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
880 * Modify/log superblock values for inode count and inode free count.
882 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
883 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
888 STATIC xfs_agnumber_t
894 spin_lock(&mp
->m_agirotor_lock
);
895 agno
= mp
->m_agirotor
;
896 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
898 spin_unlock(&mp
->m_agirotor_lock
);
904 * Select an allocation group to look for a free inode in, based on the parent
905 * inode and the mode. Return the allocation group buffer.
907 STATIC xfs_agnumber_t
908 xfs_ialloc_ag_select(
909 xfs_trans_t
*tp
, /* transaction pointer */
910 xfs_ino_t parent
, /* parent directory inode number */
911 umode_t mode
, /* bits set to indicate file type */
912 int okalloc
) /* ok to allocate more space */
914 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
915 xfs_agnumber_t agno
; /* current ag number */
916 int flags
; /* alloc buffer locking flags */
917 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
918 xfs_extlen_t longest
= 0; /* longest extent available */
919 xfs_mount_t
*mp
; /* mount point structure */
920 int needspace
; /* file mode implies space allocated */
921 xfs_perag_t
*pag
; /* per allocation group data */
922 xfs_agnumber_t pagno
; /* parent (starting) ag number */
926 * Files of these types need at least one block if length > 0
927 * (and they won't fit in the inode, but that's hard to figure out).
929 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
931 agcount
= mp
->m_maxagi
;
933 pagno
= xfs_ialloc_next_ag(mp
);
935 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
936 if (pagno
>= agcount
)
940 ASSERT(pagno
< agcount
);
943 * Loop through allocation groups, looking for one with a little
944 * free space in it. Note we don't look for free inodes, exactly.
945 * Instead, we include whether there is a need to allocate inodes
946 * to mean that blocks must be allocated for them,
947 * if none are currently free.
950 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
952 pag
= xfs_perag_get(mp
, agno
);
953 if (!pag
->pagi_inodeok
) {
954 xfs_ialloc_next_ag(mp
);
958 if (!pag
->pagi_init
) {
959 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
964 if (pag
->pagi_freecount
) {
972 if (!pag
->pagf_init
) {
973 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
979 * Check that there is enough free space for the file plus a
980 * chunk of inodes if we need to allocate some. If this is the
981 * first pass across the AGs, take into account the potential
982 * space needed for alignment of inode chunks when checking the
983 * longest contiguous free space in the AG - this prevents us
984 * from getting ENOSPC because we have free space larger than
985 * m_ialloc_blks but alignment constraints prevent us from using
988 * If we can't find an AG with space for full alignment slack to
989 * be taken into account, we must be near ENOSPC in all AGs.
990 * Hence we don't include alignment for the second pass and so
991 * if we fail allocation due to alignment issues then it is most
992 * likely a real ENOSPC condition.
994 ineed
= mp
->m_ialloc_min_blks
;
995 if (flags
&& ineed
> 1)
996 ineed
+= xfs_ialloc_cluster_alignment(mp
);
997 longest
= pag
->pagf_longest
;
999 longest
= pag
->pagf_flcount
> 0;
1001 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
1009 * No point in iterating over the rest, if we're shutting
1012 if (XFS_FORCED_SHUTDOWN(mp
))
1013 return NULLAGNUMBER
;
1015 if (agno
>= agcount
)
1017 if (agno
== pagno
) {
1019 return NULLAGNUMBER
;
1026 * Try to retrieve the next record to the left/right from the current one.
1029 xfs_ialloc_next_rec(
1030 struct xfs_btree_cur
*cur
,
1031 xfs_inobt_rec_incore_t
*rec
,
1039 error
= xfs_btree_decrement(cur
, 0, &i
);
1041 error
= xfs_btree_increment(cur
, 0, &i
);
1047 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1050 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1058 struct xfs_btree_cur
*cur
,
1060 xfs_inobt_rec_incore_t
*rec
,
1066 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
1071 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1074 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1081 * Return the offset of the first free inode in the record. If the inode chunk
1082 * is sparsely allocated, we convert the record holemask to inode granularity
1083 * and mask off the unallocated regions from the inode free mask.
1086 xfs_inobt_first_free_inode(
1087 struct xfs_inobt_rec_incore
*rec
)
1089 xfs_inofree_t realfree
;
1091 /* if there are no holes, return the first available offset */
1092 if (!xfs_inobt_issparse(rec
->ir_holemask
))
1093 return xfs_lowbit64(rec
->ir_free
);
1095 realfree
= xfs_inobt_irec_to_allocmask(rec
);
1096 realfree
&= rec
->ir_free
;
1098 return xfs_lowbit64(realfree
);
1102 * Allocate an inode using the inobt-only algorithm.
1105 xfs_dialloc_ag_inobt(
1106 struct xfs_trans
*tp
,
1107 struct xfs_buf
*agbp
,
1111 struct xfs_mount
*mp
= tp
->t_mountp
;
1112 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1113 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1114 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1115 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1116 struct xfs_perag
*pag
;
1117 struct xfs_btree_cur
*cur
, *tcur
;
1118 struct xfs_inobt_rec_incore rec
, trec
;
1124 pag
= xfs_perag_get(mp
, agno
);
1126 ASSERT(pag
->pagi_init
);
1127 ASSERT(pag
->pagi_inodeok
);
1128 ASSERT(pag
->pagi_freecount
> 0);
1131 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1133 * If pagino is 0 (this is the root inode allocation) use newino.
1134 * This must work because we've just allocated some.
1137 pagino
= be32_to_cpu(agi
->agi_newino
);
1139 error
= xfs_check_agi_freecount(cur
, agi
);
1144 * If in the same AG as the parent, try to get near the parent.
1146 if (pagno
== agno
) {
1147 int doneleft
; /* done, to the left */
1148 int doneright
; /* done, to the right */
1149 int searchdistance
= 10;
1151 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
1154 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1156 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1159 XFS_WANT_CORRUPTED_GOTO(mp
, j
== 1, error0
);
1161 if (rec
.ir_freecount
> 0) {
1163 * Found a free inode in the same chunk
1164 * as the parent, done.
1171 * In the same AG as parent, but parent's chunk is full.
1174 /* duplicate the cursor, search left & right simultaneously */
1175 error
= xfs_btree_dup_cursor(cur
, &tcur
);
1180 * Skip to last blocks looked up if same parent inode.
1182 if (pagino
!= NULLAGINO
&&
1183 pag
->pagl_pagino
== pagino
&&
1184 pag
->pagl_leftrec
!= NULLAGINO
&&
1185 pag
->pagl_rightrec
!= NULLAGINO
) {
1186 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
1191 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
1196 /* search left with tcur, back up 1 record */
1197 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
1201 /* search right with cur, go forward 1 record. */
1202 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
1208 * Loop until we find an inode chunk with a free inode.
1210 while (!doneleft
|| !doneright
) {
1211 int useleft
; /* using left inode chunk this time */
1213 if (!--searchdistance
) {
1215 * Not in range - save last search
1216 * location and allocate a new inode
1218 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1219 pag
->pagl_leftrec
= trec
.ir_startino
;
1220 pag
->pagl_rightrec
= rec
.ir_startino
;
1221 pag
->pagl_pagino
= pagino
;
1225 /* figure out the closer block if both are valid. */
1226 if (!doneleft
&& !doneright
) {
1228 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
1229 rec
.ir_startino
- pagino
;
1231 useleft
= !doneleft
;
1234 /* free inodes to the left? */
1235 if (useleft
&& trec
.ir_freecount
) {
1237 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1240 pag
->pagl_leftrec
= trec
.ir_startino
;
1241 pag
->pagl_rightrec
= rec
.ir_startino
;
1242 pag
->pagl_pagino
= pagino
;
1246 /* free inodes to the right? */
1247 if (!useleft
&& rec
.ir_freecount
) {
1248 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1250 pag
->pagl_leftrec
= trec
.ir_startino
;
1251 pag
->pagl_rightrec
= rec
.ir_startino
;
1252 pag
->pagl_pagino
= pagino
;
1256 /* get next record to check */
1258 error
= xfs_ialloc_next_rec(tcur
, &trec
,
1261 error
= xfs_ialloc_next_rec(cur
, &rec
,
1269 * We've reached the end of the btree. because
1270 * we are only searching a small chunk of the
1271 * btree each search, there is obviously free
1272 * inodes closer to the parent inode than we
1273 * are now. restart the search again.
1275 pag
->pagl_pagino
= NULLAGINO
;
1276 pag
->pagl_leftrec
= NULLAGINO
;
1277 pag
->pagl_rightrec
= NULLAGINO
;
1278 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1279 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1284 * In a different AG from the parent.
1285 * See if the most recently allocated block has any free.
1288 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1289 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1295 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1299 if (j
== 1 && rec
.ir_freecount
> 0) {
1301 * The last chunk allocated in the group
1302 * still has a free inode.
1310 * None left in the last group, search the whole AG
1312 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1315 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1318 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1321 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1322 if (rec
.ir_freecount
> 0)
1324 error
= xfs_btree_increment(cur
, 0, &i
);
1327 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1331 offset
= xfs_inobt_first_free_inode(&rec
);
1332 ASSERT(offset
>= 0);
1333 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1334 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1335 XFS_INODES_PER_CHUNK
) == 0);
1336 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1337 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1339 error
= xfs_inobt_update(cur
, &rec
);
1342 be32_add_cpu(&agi
->agi_freecount
, -1);
1343 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1344 pag
->pagi_freecount
--;
1346 error
= xfs_check_agi_freecount(cur
, agi
);
1350 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1351 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1356 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
1358 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1364 * Use the free inode btree to allocate an inode based on distance from the
1365 * parent. Note that the provided cursor may be deleted and replaced.
1368 xfs_dialloc_ag_finobt_near(
1370 struct xfs_btree_cur
**ocur
,
1371 struct xfs_inobt_rec_incore
*rec
)
1373 struct xfs_btree_cur
*lcur
= *ocur
; /* left search cursor */
1374 struct xfs_btree_cur
*rcur
; /* right search cursor */
1375 struct xfs_inobt_rec_incore rrec
;
1379 error
= xfs_inobt_lookup(lcur
, pagino
, XFS_LOOKUP_LE
, &i
);
1384 error
= xfs_inobt_get_rec(lcur
, rec
, &i
);
1387 XFS_WANT_CORRUPTED_RETURN(lcur
->bc_mp
, i
== 1);
1390 * See if we've landed in the parent inode record. The finobt
1391 * only tracks chunks with at least one free inode, so record
1392 * existence is enough.
1394 if (pagino
>= rec
->ir_startino
&&
1395 pagino
< (rec
->ir_startino
+ XFS_INODES_PER_CHUNK
))
1399 error
= xfs_btree_dup_cursor(lcur
, &rcur
);
1403 error
= xfs_inobt_lookup(rcur
, pagino
, XFS_LOOKUP_GE
, &j
);
1407 error
= xfs_inobt_get_rec(rcur
, &rrec
, &j
);
1410 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, j
== 1, error_rcur
);
1413 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, i
== 1 || j
== 1, error_rcur
);
1414 if (i
== 1 && j
== 1) {
1416 * Both the left and right records are valid. Choose the closer
1417 * inode chunk to the target.
1419 if ((pagino
- rec
->ir_startino
+ XFS_INODES_PER_CHUNK
- 1) >
1420 (rrec
.ir_startino
- pagino
)) {
1422 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1425 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1427 } else if (j
== 1) {
1428 /* only the right record is valid */
1430 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1432 } else if (i
== 1) {
1433 /* only the left record is valid */
1434 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1440 xfs_btree_del_cursor(rcur
, XFS_BTREE_ERROR
);
1445 * Use the free inode btree to find a free inode based on a newino hint. If
1446 * the hint is NULL, find the first free inode in the AG.
1449 xfs_dialloc_ag_finobt_newino(
1450 struct xfs_agi
*agi
,
1451 struct xfs_btree_cur
*cur
,
1452 struct xfs_inobt_rec_incore
*rec
)
1457 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1458 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1463 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1466 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1472 * Find the first inode available in the AG.
1474 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1477 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1479 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1482 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1488 * Update the inobt based on a modification made to the finobt. Also ensure that
1489 * the records from both trees are equivalent post-modification.
1492 xfs_dialloc_ag_update_inobt(
1493 struct xfs_btree_cur
*cur
, /* inobt cursor */
1494 struct xfs_inobt_rec_incore
*frec
, /* finobt record */
1495 int offset
) /* inode offset */
1497 struct xfs_inobt_rec_incore rec
;
1501 error
= xfs_inobt_lookup(cur
, frec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1504 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1506 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1509 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1510 ASSERT((XFS_AGINO_TO_OFFSET(cur
->bc_mp
, rec
.ir_startino
) %
1511 XFS_INODES_PER_CHUNK
) == 0);
1513 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1516 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, (rec
.ir_free
== frec
->ir_free
) &&
1517 (rec
.ir_freecount
== frec
->ir_freecount
));
1519 return xfs_inobt_update(cur
, &rec
);
1523 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1524 * back to the inobt search algorithm.
1526 * The caller selected an AG for us, and made sure that free inodes are
1531 struct xfs_trans
*tp
,
1532 struct xfs_buf
*agbp
,
1536 struct xfs_mount
*mp
= tp
->t_mountp
;
1537 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1538 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1539 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1540 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1541 struct xfs_perag
*pag
;
1542 struct xfs_btree_cur
*cur
; /* finobt cursor */
1543 struct xfs_btree_cur
*icur
; /* inobt cursor */
1544 struct xfs_inobt_rec_incore rec
;
1550 if (!xfs_sb_version_hasfinobt(&mp
->m_sb
))
1551 return xfs_dialloc_ag_inobt(tp
, agbp
, parent
, inop
);
1553 pag
= xfs_perag_get(mp
, agno
);
1556 * If pagino is 0 (this is the root inode allocation) use newino.
1557 * This must work because we've just allocated some.
1560 pagino
= be32_to_cpu(agi
->agi_newino
);
1562 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1564 error
= xfs_check_agi_freecount(cur
, agi
);
1569 * The search algorithm depends on whether we're in the same AG as the
1570 * parent. If so, find the closest available inode to the parent. If
1571 * not, consider the agi hint or find the first free inode in the AG.
1574 error
= xfs_dialloc_ag_finobt_near(pagino
, &cur
, &rec
);
1576 error
= xfs_dialloc_ag_finobt_newino(agi
, cur
, &rec
);
1580 offset
= xfs_inobt_first_free_inode(&rec
);
1581 ASSERT(offset
>= 0);
1582 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1583 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1584 XFS_INODES_PER_CHUNK
) == 0);
1585 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1588 * Modify or remove the finobt record.
1590 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1592 if (rec
.ir_freecount
)
1593 error
= xfs_inobt_update(cur
, &rec
);
1595 error
= xfs_btree_delete(cur
, &i
);
1600 * The finobt has now been updated appropriately. We haven't updated the
1601 * agi and superblock yet, so we can create an inobt cursor and validate
1602 * the original freecount. If all is well, make the equivalent update to
1603 * the inobt using the finobt record and offset information.
1605 icur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1607 error
= xfs_check_agi_freecount(icur
, agi
);
1611 error
= xfs_dialloc_ag_update_inobt(icur
, &rec
, offset
);
1616 * Both trees have now been updated. We must update the perag and
1617 * superblock before we can check the freecount for each btree.
1619 be32_add_cpu(&agi
->agi_freecount
, -1);
1620 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1621 pag
->pagi_freecount
--;
1623 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1625 error
= xfs_check_agi_freecount(icur
, agi
);
1628 error
= xfs_check_agi_freecount(cur
, agi
);
1632 xfs_btree_del_cursor(icur
, XFS_BTREE_NOERROR
);
1633 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1639 xfs_btree_del_cursor(icur
, XFS_BTREE_ERROR
);
1641 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1647 * Allocate an inode on disk.
1649 * Mode is used to tell whether the new inode will need space, and whether it
1652 * This function is designed to be called twice if it has to do an allocation
1653 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1654 * If an inode is available without having to performn an allocation, an inode
1655 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1656 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1657 * The caller should then commit the current transaction, allocate a
1658 * new transaction, and call xfs_dialloc() again, passing in the previous value
1659 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1660 * buffer is locked across the two calls, the second call is guaranteed to have
1661 * a free inode available.
1663 * Once we successfully pick an inode its number is returned and the on-disk
1664 * data structures are updated. The inode itself is not read in, since doing so
1665 * would break ordering constraints with xfs_reclaim.
1669 struct xfs_trans
*tp
,
1673 struct xfs_buf
**IO_agbp
,
1676 struct xfs_mount
*mp
= tp
->t_mountp
;
1677 struct xfs_buf
*agbp
;
1678 xfs_agnumber_t agno
;
1682 xfs_agnumber_t start_agno
;
1683 struct xfs_perag
*pag
;
1687 * If the caller passes in a pointer to the AGI buffer,
1688 * continue where we left off before. In this case, we
1689 * know that the allocation group has free inodes.
1696 * We do not have an agbp, so select an initial allocation
1697 * group for inode allocation.
1699 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
, okalloc
);
1700 if (start_agno
== NULLAGNUMBER
) {
1706 * If we have already hit the ceiling of inode blocks then clear
1707 * okalloc so we scan all available agi structures for a free
1710 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1711 * which will sacrifice the preciseness but improve the performance.
1713 if (mp
->m_maxicount
&&
1714 percpu_counter_read_positive(&mp
->m_icount
) + mp
->m_ialloc_inos
1715 > mp
->m_maxicount
) {
1721 * Loop until we find an allocation group that either has free inodes
1722 * or in which we can allocate some inodes. Iterate through the
1723 * allocation groups upward, wrapping at the end.
1727 pag
= xfs_perag_get(mp
, agno
);
1728 if (!pag
->pagi_inodeok
) {
1729 xfs_ialloc_next_ag(mp
);
1733 if (!pag
->pagi_init
) {
1734 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1740 * Do a first racy fast path check if this AG is usable.
1742 if (!pag
->pagi_freecount
&& !okalloc
)
1746 * Then read in the AGI buffer and recheck with the AGI buffer
1749 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1753 if (pag
->pagi_freecount
) {
1759 goto nextag_relse_buffer
;
1762 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1764 xfs_trans_brelse(tp
, agbp
);
1766 if (error
!= -ENOSPC
)
1776 * We successfully allocated some inodes, return
1777 * the current context to the caller so that it
1778 * can commit the current transaction and call
1779 * us again where we left off.
1781 ASSERT(pag
->pagi_freecount
> 0);
1789 nextag_relse_buffer
:
1790 xfs_trans_brelse(tp
, agbp
);
1793 if (++agno
== mp
->m_sb
.sb_agcount
)
1795 if (agno
== start_agno
) {
1797 return noroom
? -ENOSPC
: 0;
1803 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1810 * Free the blocks of an inode chunk. We must consider that the inode chunk
1811 * might be sparse and only free the regions that are allocated as part of the
1815 xfs_difree_inode_chunk(
1816 struct xfs_mount
*mp
,
1817 xfs_agnumber_t agno
,
1818 struct xfs_inobt_rec_incore
*rec
,
1819 struct xfs_bmap_free
*flist
)
1821 xfs_agblock_t sagbno
= XFS_AGINO_TO_AGBNO(mp
, rec
->ir_startino
);
1822 int startidx
, endidx
;
1824 xfs_agblock_t agbno
;
1826 DECLARE_BITMAP(holemask
, XFS_INOBT_HOLEMASK_BITS
);
1828 if (!xfs_inobt_issparse(rec
->ir_holemask
)) {
1829 /* not sparse, calculate extent info directly */
1830 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp
, agno
,
1831 XFS_AGINO_TO_AGBNO(mp
, rec
->ir_startino
)),
1832 mp
->m_ialloc_blks
, flist
, mp
);
1836 /* holemask is only 16-bits (fits in an unsigned long) */
1837 ASSERT(sizeof(rec
->ir_holemask
) <= sizeof(holemask
[0]));
1838 holemask
[0] = rec
->ir_holemask
;
1841 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1842 * holemask and convert the start/end index of each range to an extent.
1843 * We start with the start and end index both pointing at the first 0 in
1846 startidx
= endidx
= find_first_zero_bit(holemask
,
1847 XFS_INOBT_HOLEMASK_BITS
);
1848 nextbit
= startidx
+ 1;
1849 while (startidx
< XFS_INOBT_HOLEMASK_BITS
) {
1850 nextbit
= find_next_zero_bit(holemask
, XFS_INOBT_HOLEMASK_BITS
,
1853 * If the next zero bit is contiguous, update the end index of
1854 * the current range and continue.
1856 if (nextbit
!= XFS_INOBT_HOLEMASK_BITS
&&
1857 nextbit
== endidx
+ 1) {
1863 * nextbit is not contiguous with the current end index. Convert
1864 * the current start/end to an extent and add it to the free
1867 agbno
= sagbno
+ (startidx
* XFS_INODES_PER_HOLEMASK_BIT
) /
1868 mp
->m_sb
.sb_inopblock
;
1869 contigblk
= ((endidx
- startidx
+ 1) *
1870 XFS_INODES_PER_HOLEMASK_BIT
) /
1871 mp
->m_sb
.sb_inopblock
;
1873 ASSERT(agbno
% mp
->m_sb
.sb_spino_align
== 0);
1874 ASSERT(contigblk
% mp
->m_sb
.sb_spino_align
== 0);
1875 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp
, agno
, agbno
), contigblk
,
1878 /* reset range to current bit and carry on... */
1879 startidx
= endidx
= nextbit
;
1888 struct xfs_mount
*mp
,
1889 struct xfs_trans
*tp
,
1890 struct xfs_buf
*agbp
,
1892 struct xfs_bmap_free
*flist
,
1893 struct xfs_icluster
*xic
,
1894 struct xfs_inobt_rec_incore
*orec
)
1896 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1897 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1898 struct xfs_perag
*pag
;
1899 struct xfs_btree_cur
*cur
;
1900 struct xfs_inobt_rec_incore rec
;
1906 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1907 ASSERT(XFS_AGINO_TO_AGBNO(mp
, agino
) < be32_to_cpu(agi
->agi_length
));
1910 * Initialize the cursor.
1912 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1914 error
= xfs_check_agi_freecount(cur
, agi
);
1919 * Look for the entry describing this inode.
1921 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1922 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1926 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1927 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1929 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1933 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1935 * Get the offset in the inode chunk.
1937 off
= agino
- rec
.ir_startino
;
1938 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1939 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1941 * Mark the inode free & increment the count.
1943 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1947 * When an inode chunk is free, it becomes eligible for removal. Don't
1948 * remove the chunk if the block size is large enough for multiple inode
1949 * chunks (that might not be free).
1951 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1952 rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
1953 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
) {
1955 xic
->first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1956 xic
->alloc
= xfs_inobt_irec_to_allocmask(&rec
);
1959 * Remove the inode cluster from the AGI B+Tree, adjust the
1960 * AGI and Superblock inode counts, and mark the disk space
1961 * to be freed when the transaction is committed.
1963 ilen
= rec
.ir_freecount
;
1964 be32_add_cpu(&agi
->agi_count
, -ilen
);
1965 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1966 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1967 pag
= xfs_perag_get(mp
, agno
);
1968 pag
->pagi_freecount
-= ilen
- 1;
1970 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1971 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1973 if ((error
= xfs_btree_delete(cur
, &i
))) {
1974 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1979 xfs_difree_inode_chunk(mp
, agno
, &rec
, flist
);
1983 error
= xfs_inobt_update(cur
, &rec
);
1985 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
1991 * Change the inode free counts and log the ag/sb changes.
1993 be32_add_cpu(&agi
->agi_freecount
, 1);
1994 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1995 pag
= xfs_perag_get(mp
, agno
);
1996 pag
->pagi_freecount
++;
1998 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
2001 error
= xfs_check_agi_freecount(cur
, agi
);
2006 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2010 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2015 * Free an inode in the free inode btree.
2019 struct xfs_mount
*mp
,
2020 struct xfs_trans
*tp
,
2021 struct xfs_buf
*agbp
,
2023 struct xfs_inobt_rec_incore
*ibtrec
) /* inobt record */
2025 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
2026 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
2027 struct xfs_btree_cur
*cur
;
2028 struct xfs_inobt_rec_incore rec
;
2029 int offset
= agino
- ibtrec
->ir_startino
;
2033 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
2035 error
= xfs_inobt_lookup(cur
, ibtrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
2040 * If the record does not exist in the finobt, we must have just
2041 * freed an inode in a previously fully allocated chunk. If not,
2042 * something is out of sync.
2044 XFS_WANT_CORRUPTED_GOTO(mp
, ibtrec
->ir_freecount
== 1, error
);
2046 error
= xfs_inobt_insert_rec(cur
, ibtrec
->ir_holemask
,
2048 ibtrec
->ir_freecount
,
2049 ibtrec
->ir_free
, &i
);
2058 * Read and update the existing record. We could just copy the ibtrec
2059 * across here, but that would defeat the purpose of having redundant
2060 * metadata. By making the modifications independently, we can catch
2061 * corruptions that we wouldn't see if we just copied from one record
2064 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2067 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
2069 rec
.ir_free
|= XFS_INOBT_MASK(offset
);
2072 XFS_WANT_CORRUPTED_GOTO(mp
, (rec
.ir_free
== ibtrec
->ir_free
) &&
2073 (rec
.ir_freecount
== ibtrec
->ir_freecount
),
2077 * The content of inobt records should always match between the inobt
2078 * and finobt. The lifecycle of records in the finobt is different from
2079 * the inobt in that the finobt only tracks records with at least one
2080 * free inode. Hence, if all of the inodes are free and we aren't
2081 * keeping inode chunks permanently on disk, remove the record.
2082 * Otherwise, update the record with the new information.
2084 * Note that we currently can't free chunks when the block size is large
2085 * enough for multiple chunks. Leave the finobt record to remain in sync
2088 if (rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
2089 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
&&
2090 !(mp
->m_flags
& XFS_MOUNT_IKEEP
)) {
2091 error
= xfs_btree_delete(cur
, &i
);
2096 error
= xfs_inobt_update(cur
, &rec
);
2102 error
= xfs_check_agi_freecount(cur
, agi
);
2106 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2110 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2115 * Free disk inode. Carefully avoids touching the incore inode, all
2116 * manipulations incore are the caller's responsibility.
2117 * The on-disk inode is not changed by this operation, only the
2118 * btree (free inode mask) is changed.
2122 struct xfs_trans
*tp
, /* transaction pointer */
2123 xfs_ino_t inode
, /* inode to be freed */
2124 struct xfs_bmap_free
*flist
, /* extents to free */
2125 struct xfs_icluster
*xic
) /* cluster info if deleted */
2128 xfs_agblock_t agbno
; /* block number containing inode */
2129 struct xfs_buf
*agbp
; /* buffer for allocation group header */
2130 xfs_agino_t agino
; /* allocation group inode number */
2131 xfs_agnumber_t agno
; /* allocation group number */
2132 int error
; /* error return value */
2133 struct xfs_mount
*mp
; /* mount structure for filesystem */
2134 struct xfs_inobt_rec_incore rec
;/* btree record */
2139 * Break up inode number into its components.
2141 agno
= XFS_INO_TO_AGNO(mp
, inode
);
2142 if (agno
>= mp
->m_sb
.sb_agcount
) {
2143 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2144 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2148 agino
= XFS_INO_TO_AGINO(mp
, inode
);
2149 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2150 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2151 __func__
, (unsigned long long)inode
,
2152 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
2156 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2157 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2158 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2159 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
2164 * Get the allocation group header.
2166 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2168 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
2174 * Fix up the inode allocation btree.
2176 error
= xfs_difree_inobt(mp
, tp
, agbp
, agino
, flist
, xic
, &rec
);
2181 * Fix up the free inode btree.
2183 if (xfs_sb_version_hasfinobt(&mp
->m_sb
)) {
2184 error
= xfs_difree_finobt(mp
, tp
, agbp
, agino
, &rec
);
2197 struct xfs_mount
*mp
,
2198 struct xfs_trans
*tp
,
2199 xfs_agnumber_t agno
,
2201 xfs_agblock_t agbno
,
2202 xfs_agblock_t
*chunk_agbno
,
2203 xfs_agblock_t
*offset_agbno
,
2206 struct xfs_inobt_rec_incore rec
;
2207 struct xfs_btree_cur
*cur
;
2208 struct xfs_buf
*agbp
;
2212 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2215 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2216 __func__
, error
, agno
);
2221 * Lookup the inode record for the given agino. If the record cannot be
2222 * found, then it's an invalid inode number and we should abort. Once
2223 * we have a record, we need to ensure it contains the inode number
2224 * we are looking up.
2226 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
2227 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
2230 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2231 if (!error
&& i
== 0)
2235 xfs_trans_brelse(tp
, agbp
);
2236 xfs_btree_del_cursor(cur
, error
? XFS_BTREE_ERROR
: XFS_BTREE_NOERROR
);
2240 /* check that the returned record contains the required inode */
2241 if (rec
.ir_startino
> agino
||
2242 rec
.ir_startino
+ mp
->m_ialloc_inos
<= agino
)
2245 /* for untrusted inodes check it is allocated first */
2246 if ((flags
& XFS_IGET_UNTRUSTED
) &&
2247 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
2250 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
2251 *offset_agbno
= agbno
- *chunk_agbno
;
2256 * Return the location of the inode in imap, for mapping it into a buffer.
2260 xfs_mount_t
*mp
, /* file system mount structure */
2261 xfs_trans_t
*tp
, /* transaction pointer */
2262 xfs_ino_t ino
, /* inode to locate */
2263 struct xfs_imap
*imap
, /* location map structure */
2264 uint flags
) /* flags for inode btree lookup */
2266 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
2267 xfs_agino_t agino
; /* inode number within alloc group */
2268 xfs_agnumber_t agno
; /* allocation group number */
2269 int blks_per_cluster
; /* num blocks per inode cluster */
2270 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
2271 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
2272 int error
; /* error code */
2273 int offset
; /* index of inode in its buffer */
2274 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
2276 ASSERT(ino
!= NULLFSINO
);
2279 * Split up the inode number into its parts.
2281 agno
= XFS_INO_TO_AGNO(mp
, ino
);
2282 agino
= XFS_INO_TO_AGINO(mp
, ino
);
2283 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2284 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
2285 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2288 * Don't output diagnostic information for untrusted inodes
2289 * as they can be invalid without implying corruption.
2291 if (flags
& XFS_IGET_UNTRUSTED
)
2293 if (agno
>= mp
->m_sb
.sb_agcount
) {
2295 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2296 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2298 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2300 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2301 __func__
, (unsigned long long)agbno
,
2302 (unsigned long)mp
->m_sb
.sb_agblocks
);
2304 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2306 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2308 XFS_AGINO_TO_INO(mp
, agno
, agino
));
2315 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2318 * For bulkstat and handle lookups, we have an untrusted inode number
2319 * that we have to verify is valid. We cannot do this just by reading
2320 * the inode buffer as it may have been unlinked and removed leaving
2321 * inodes in stale state on disk. Hence we have to do a btree lookup
2322 * in all cases where an untrusted inode number is passed.
2324 if (flags
& XFS_IGET_UNTRUSTED
) {
2325 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2326 &chunk_agbno
, &offset_agbno
, flags
);
2333 * If the inode cluster size is the same as the blocksize or
2334 * smaller we get to the buffer by simple arithmetics.
2336 if (blks_per_cluster
== 1) {
2337 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
2338 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
2340 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
2341 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
2342 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
2347 * If the inode chunks are aligned then use simple maths to
2348 * find the location. Otherwise we have to do a btree
2349 * lookup to find the location.
2351 if (mp
->m_inoalign_mask
) {
2352 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
2353 chunk_agbno
= agbno
- offset_agbno
;
2355 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2356 &chunk_agbno
, &offset_agbno
, flags
);
2362 ASSERT(agbno
>= chunk_agbno
);
2363 cluster_agbno
= chunk_agbno
+
2364 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
2365 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
2366 XFS_INO_TO_OFFSET(mp
, ino
);
2368 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
2369 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
2370 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
2373 * If the inode number maps to a block outside the bounds
2374 * of the file system then return NULL rather than calling
2375 * read_buf and panicing when we get an error from the
2378 if ((imap
->im_blkno
+ imap
->im_len
) >
2379 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
2381 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2382 __func__
, (unsigned long long) imap
->im_blkno
,
2383 (unsigned long long) imap
->im_len
,
2384 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
2391 * Compute and fill in value of m_in_maxlevels.
2394 xfs_ialloc_compute_maxlevels(
2395 xfs_mount_t
*mp
) /* file system mount structure */
2403 maxleafents
= (1LL << XFS_INO_AGINO_BITS(mp
)) >>
2404 XFS_INODES_PER_CHUNK_LOG
;
2405 minleafrecs
= mp
->m_alloc_mnr
[0];
2406 minnoderecs
= mp
->m_alloc_mnr
[1];
2407 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
2408 for (level
= 1; maxblocks
> 1; level
++)
2409 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
2410 mp
->m_in_maxlevels
= level
;
2414 * Log specified fields for the ag hdr (inode section). The growth of the agi
2415 * structure over time requires that we interpret the buffer as two logical
2416 * regions delineated by the end of the unlinked list. This is due to the size
2417 * of the hash table and its location in the middle of the agi.
2419 * For example, a request to log a field before agi_unlinked and a field after
2420 * agi_unlinked could cause us to log the entire hash table and use an excessive
2421 * amount of log space. To avoid this behavior, log the region up through
2422 * agi_unlinked in one call and the region after agi_unlinked through the end of
2423 * the structure in another.
2427 xfs_trans_t
*tp
, /* transaction pointer */
2428 xfs_buf_t
*bp
, /* allocation group header buffer */
2429 int fields
) /* bitmask of fields to log */
2431 int first
; /* first byte number */
2432 int last
; /* last byte number */
2433 static const short offsets
[] = { /* field starting offsets */
2434 /* keep in sync with bit definitions */
2435 offsetof(xfs_agi_t
, agi_magicnum
),
2436 offsetof(xfs_agi_t
, agi_versionnum
),
2437 offsetof(xfs_agi_t
, agi_seqno
),
2438 offsetof(xfs_agi_t
, agi_length
),
2439 offsetof(xfs_agi_t
, agi_count
),
2440 offsetof(xfs_agi_t
, agi_root
),
2441 offsetof(xfs_agi_t
, agi_level
),
2442 offsetof(xfs_agi_t
, agi_freecount
),
2443 offsetof(xfs_agi_t
, agi_newino
),
2444 offsetof(xfs_agi_t
, agi_dirino
),
2445 offsetof(xfs_agi_t
, agi_unlinked
),
2446 offsetof(xfs_agi_t
, agi_free_root
),
2447 offsetof(xfs_agi_t
, agi_free_level
),
2451 xfs_agi_t
*agi
; /* allocation group header */
2453 agi
= XFS_BUF_TO_AGI(bp
);
2454 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
2457 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_AGI_BUF
);
2460 * Compute byte offsets for the first and last fields in the first
2461 * region and log the agi buffer. This only logs up through
2464 if (fields
& XFS_AGI_ALL_BITS_R1
) {
2465 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R1
,
2467 xfs_trans_log_buf(tp
, bp
, first
, last
);
2471 * Mask off the bits in the first region and calculate the first and
2472 * last field offsets for any bits in the second region.
2474 fields
&= ~XFS_AGI_ALL_BITS_R1
;
2476 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R2
,
2478 xfs_trans_log_buf(tp
, bp
, first
, last
);
2484 xfs_check_agi_unlinked(
2485 struct xfs_agi
*agi
)
2489 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++)
2490 ASSERT(agi
->agi_unlinked
[i
]);
2493 #define xfs_check_agi_unlinked(agi)
2500 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2501 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
2503 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2504 !uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_meta_uuid
))
2507 * Validate the magic number of the agi block.
2509 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
2511 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
2514 if (be32_to_cpu(agi
->agi_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
= {
2569 .verify_read
= xfs_agi_read_verify
,
2570 .verify_write
= xfs_agi_write_verify
,
2574 * Read in the allocation group header (inode allocation section)
2578 struct xfs_mount
*mp
, /* file system mount structure */
2579 struct xfs_trans
*tp
, /* transaction pointer */
2580 xfs_agnumber_t agno
, /* allocation group number */
2581 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2585 trace_xfs_read_agi(mp
, agno
);
2587 ASSERT(agno
!= NULLAGNUMBER
);
2588 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
2589 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
2590 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
2594 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
2599 xfs_ialloc_read_agi(
2600 struct xfs_mount
*mp
, /* file system mount structure */
2601 struct xfs_trans
*tp
, /* transaction pointer */
2602 xfs_agnumber_t agno
, /* allocation group number */
2603 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2605 struct xfs_agi
*agi
; /* allocation group header */
2606 struct xfs_perag
*pag
; /* per allocation group data */
2609 trace_xfs_ialloc_read_agi(mp
, agno
);
2611 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
2615 agi
= XFS_BUF_TO_AGI(*bpp
);
2616 pag
= xfs_perag_get(mp
, agno
);
2617 if (!pag
->pagi_init
) {
2618 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
2619 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
2624 * It's possible for these to be out of sync if
2625 * we are in the middle of a forced shutdown.
2627 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
2628 XFS_FORCED_SHUTDOWN(mp
));
2634 * Read in the agi to initialise the per-ag data in the mount structure
2637 xfs_ialloc_pagi_init(
2638 xfs_mount_t
*mp
, /* file system mount structure */
2639 xfs_trans_t
*tp
, /* transaction pointer */
2640 xfs_agnumber_t agno
) /* allocation group number */
2642 xfs_buf_t
*bp
= NULL
;
2645 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
2649 xfs_trans_brelse(tp
, bp
);