1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
6 #include "libxfs_priv.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_ialloc_btree.h"
20 #include "xfs_alloc.h"
21 #include "xfs_errortag.h"
23 #include "xfs_cksum.h"
24 #include "xfs_trans.h"
25 #include "xfs_trace.h"
30 * Allocation group level functions.
33 xfs_ialloc_cluster_alignment(
36 if (xfs_sb_version_hasalign(&mp
->m_sb
) &&
37 mp
->m_sb
.sb_inoalignmt
>= xfs_icluster_size_fsb(mp
))
38 return mp
->m_sb
.sb_inoalignmt
;
43 * Lookup a record by ino in the btree given by cur.
47 struct xfs_btree_cur
*cur
, /* btree cursor */
48 xfs_agino_t ino
, /* starting inode of chunk */
49 xfs_lookup_t dir
, /* <=, >=, == */
50 int *stat
) /* success/failure */
52 cur
->bc_rec
.i
.ir_startino
= ino
;
53 cur
->bc_rec
.i
.ir_holemask
= 0;
54 cur
->bc_rec
.i
.ir_count
= 0;
55 cur
->bc_rec
.i
.ir_freecount
= 0;
56 cur
->bc_rec
.i
.ir_free
= 0;
57 return xfs_btree_lookup(cur
, dir
, stat
);
61 * Update the record referred to by cur to the value given.
62 * This either works (return 0) or gets an EFSCORRUPTED error.
64 STATIC
int /* error */
66 struct xfs_btree_cur
*cur
, /* btree cursor */
67 xfs_inobt_rec_incore_t
*irec
) /* btree record */
69 union xfs_btree_rec rec
;
71 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
72 if (xfs_sb_version_hassparseinodes(&cur
->bc_mp
->m_sb
)) {
73 rec
.inobt
.ir_u
.sp
.ir_holemask
= cpu_to_be16(irec
->ir_holemask
);
74 rec
.inobt
.ir_u
.sp
.ir_count
= irec
->ir_count
;
75 rec
.inobt
.ir_u
.sp
.ir_freecount
= irec
->ir_freecount
;
77 /* ir_holemask/ir_count not supported on-disk */
78 rec
.inobt
.ir_u
.f
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
80 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
81 return xfs_btree_update(cur
, &rec
);
84 /* Convert on-disk btree record to incore inobt record. */
86 xfs_inobt_btrec_to_irec(
88 union xfs_btree_rec
*rec
,
89 struct xfs_inobt_rec_incore
*irec
)
91 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
92 if (xfs_sb_version_hassparseinodes(&mp
->m_sb
)) {
93 irec
->ir_holemask
= be16_to_cpu(rec
->inobt
.ir_u
.sp
.ir_holemask
);
94 irec
->ir_count
= rec
->inobt
.ir_u
.sp
.ir_count
;
95 irec
->ir_freecount
= rec
->inobt
.ir_u
.sp
.ir_freecount
;
98 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
99 * values for full inode chunks.
101 irec
->ir_holemask
= XFS_INOBT_HOLEMASK_FULL
;
102 irec
->ir_count
= XFS_INODES_PER_CHUNK
;
104 be32_to_cpu(rec
->inobt
.ir_u
.f
.ir_freecount
);
106 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
110 * Get the data from the pointed-to record.
114 struct xfs_btree_cur
*cur
,
115 struct xfs_inobt_rec_incore
*irec
,
118 struct xfs_mount
*mp
= cur
->bc_mp
;
119 xfs_agnumber_t agno
= cur
->bc_private
.a
.agno
;
120 union xfs_btree_rec
*rec
;
124 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
125 if (error
|| *stat
== 0)
128 xfs_inobt_btrec_to_irec(mp
, rec
, irec
);
130 if (!xfs_verify_agino(mp
, agno
, irec
->ir_startino
))
132 if (irec
->ir_count
< XFS_INODES_PER_HOLEMASK_BIT
||
133 irec
->ir_count
> XFS_INODES_PER_CHUNK
)
135 if (irec
->ir_freecount
> XFS_INODES_PER_CHUNK
)
138 /* if there are no holes, return the first available offset */
139 if (!xfs_inobt_issparse(irec
->ir_holemask
))
140 realfree
= irec
->ir_free
;
142 realfree
= irec
->ir_free
& xfs_inobt_irec_to_allocmask(irec
);
143 if (hweight64(realfree
) != irec
->ir_freecount
)
150 "%s Inode BTree record corruption in AG %d detected!",
151 cur
->bc_btnum
== XFS_BTNUM_INO
? "Used" : "Free", agno
);
153 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
154 irec
->ir_startino
, irec
->ir_count
, irec
->ir_freecount
,
155 irec
->ir_free
, irec
->ir_holemask
);
156 return -EFSCORRUPTED
;
160 * Insert a single inobt record. Cursor must already point to desired location.
163 xfs_inobt_insert_rec(
164 struct xfs_btree_cur
*cur
,
171 cur
->bc_rec
.i
.ir_holemask
= holemask
;
172 cur
->bc_rec
.i
.ir_count
= count
;
173 cur
->bc_rec
.i
.ir_freecount
= freecount
;
174 cur
->bc_rec
.i
.ir_free
= free
;
175 return xfs_btree_insert(cur
, stat
);
179 * Insert records describing a newly allocated inode chunk into the inobt.
183 struct xfs_mount
*mp
,
184 struct xfs_trans
*tp
,
185 struct xfs_buf
*agbp
,
190 struct xfs_btree_cur
*cur
;
191 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
192 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
197 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
199 for (thisino
= newino
;
200 thisino
< newino
+ newlen
;
201 thisino
+= XFS_INODES_PER_CHUNK
) {
202 error
= xfs_inobt_lookup(cur
, thisino
, XFS_LOOKUP_EQ
, &i
);
204 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
209 error
= xfs_inobt_insert_rec(cur
, XFS_INOBT_HOLEMASK_FULL
,
210 XFS_INODES_PER_CHUNK
,
211 XFS_INODES_PER_CHUNK
,
212 XFS_INOBT_ALL_FREE
, &i
);
214 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
220 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
226 * Verify that the number of free inodes in the AGI is correct.
230 xfs_check_agi_freecount(
231 struct xfs_btree_cur
*cur
,
234 if (cur
->bc_nlevels
== 1) {
235 xfs_inobt_rec_incore_t rec
;
240 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
245 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
250 freecount
+= rec
.ir_freecount
;
251 error
= xfs_btree_increment(cur
, 0, &i
);
257 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
258 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
263 #define xfs_check_agi_freecount(cur, agi) 0
267 * Initialise a new set of inodes. When called without a transaction context
268 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
269 * than logging them (which in a transaction context puts them into the AIL
270 * for writeback rather than the xfsbufd queue).
273 xfs_ialloc_inode_init(
274 struct xfs_mount
*mp
,
275 struct xfs_trans
*tp
,
276 struct list_head
*buffer_list
,
280 xfs_agblock_t length
,
283 struct xfs_buf
*fbuf
;
284 struct xfs_dinode
*free
;
285 int nbufs
, blks_per_cluster
, inodes_per_cluster
;
292 * Loop over the new block(s), filling in the inodes. For small block
293 * sizes, manipulate the inodes in buffers which are multiples of the
296 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
297 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
298 nbufs
= length
/ blks_per_cluster
;
301 * Figure out what version number to use in the inodes we create. If
302 * the superblock version has caught up to the one that supports the new
303 * inode format, then use the new inode version. Otherwise use the old
304 * version so that old kernels will continue to be able to use the file
307 * For v3 inodes, we also need to write the inode number into the inode,
308 * so calculate the first inode number of the chunk here as
309 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
310 * across multiple filesystem blocks (such as a cluster) and so cannot
311 * be used in the cluster buffer loop below.
313 * Further, because we are writing the inode directly into the buffer
314 * and calculating a CRC on the entire inode, we have ot log the entire
315 * inode so that the entire range the CRC covers is present in the log.
316 * That means for v3 inode we log the entire buffer rather than just the
319 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
321 ino
= XFS_AGINO_TO_INO(mp
, agno
,
322 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
325 * log the initialisation that is about to take place as an
326 * logical operation. This means the transaction does not
327 * need to log the physical changes to the inode buffers as log
328 * recovery will know what initialisation is actually needed.
329 * Hence we only need to log the buffers as "ordered" buffers so
330 * they track in the AIL as if they were physically logged.
333 xfs_icreate_log(tp
, agno
, agbno
, icount
,
334 mp
->m_sb
.sb_inodesize
, length
, gen
);
338 for (j
= 0; j
< nbufs
; j
++) {
342 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
343 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
344 mp
->m_bsize
* blks_per_cluster
,
349 /* Initialize the inode buffers and log them appropriately. */
350 fbuf
->b_ops
= &xfs_inode_buf_ops
;
351 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
352 for (i
= 0; i
< inodes_per_cluster
; i
++) {
353 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
354 uint isize
= xfs_dinode_size(version
);
356 free
= xfs_make_iptr(mp
, fbuf
, i
);
357 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
358 free
->di_version
= version
;
359 free
->di_gen
= cpu_to_be32(gen
);
360 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
363 free
->di_ino
= cpu_to_be64(ino
);
365 uuid_copy(&free
->di_uuid
,
366 &mp
->m_sb
.sb_meta_uuid
);
367 xfs_dinode_calc_crc(mp
, free
);
369 /* just log the inode core */
370 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
371 ioffset
+ isize
- 1);
377 * Mark the buffer as an inode allocation buffer so it
378 * sticks in AIL at the point of this allocation
379 * transaction. This ensures the they are on disk before
380 * the tail of the log can be moved past this
381 * transaction (i.e. by preventing relogging from moving
382 * it forward in the log).
384 xfs_trans_inode_alloc_buf(tp
, fbuf
);
387 * Mark the buffer as ordered so that they are
388 * not physically logged in the transaction but
389 * still tracked in the AIL as part of the
390 * transaction and pin the log appropriately.
392 xfs_trans_ordered_buf(tp
, fbuf
);
395 fbuf
->b_flags
|= XBF_DONE
;
396 xfs_buf_delwri_queue(fbuf
, buffer_list
);
404 * Align startino and allocmask for a recently allocated sparse chunk such that
405 * they are fit for insertion (or merge) into the on-disk inode btrees.
409 * When enabled, sparse inode support increases the inode alignment from cluster
410 * size to inode chunk size. This means that the minimum range between two
411 * non-adjacent inode records in the inobt is large enough for a full inode
412 * record. This allows for cluster sized, cluster aligned block allocation
413 * without need to worry about whether the resulting inode record overlaps with
414 * another record in the tree. Without this basic rule, we would have to deal
415 * with the consequences of overlap by potentially undoing recent allocations in
416 * the inode allocation codepath.
418 * Because of this alignment rule (which is enforced on mount), there are two
419 * inobt possibilities for newly allocated sparse chunks. One is that the
420 * aligned inode record for the chunk covers a range of inodes not already
421 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
422 * other is that a record already exists at the aligned startino that considers
423 * the newly allocated range as sparse. In the latter case, record content is
424 * merged in hope that sparse inode chunks fill to full chunks over time.
427 xfs_align_sparse_ino(
428 struct xfs_mount
*mp
,
429 xfs_agino_t
*startino
,
436 agbno
= XFS_AGINO_TO_AGBNO(mp
, *startino
);
437 mod
= agbno
% mp
->m_sb
.sb_inoalignmt
;
441 /* calculate the inode offset and align startino */
442 offset
= mod
<< mp
->m_sb
.sb_inopblog
;
446 * Since startino has been aligned down, left shift allocmask such that
447 * it continues to represent the same physical inodes relative to the
450 *allocmask
<<= offset
/ XFS_INODES_PER_HOLEMASK_BIT
;
454 * Determine whether the source inode record can merge into the target. Both
455 * records must be sparse, the inode ranges must match and there must be no
456 * allocation overlap between the records.
459 __xfs_inobt_can_merge(
460 struct xfs_inobt_rec_incore
*trec
, /* tgt record */
461 struct xfs_inobt_rec_incore
*srec
) /* src record */
466 /* records must cover the same inode range */
467 if (trec
->ir_startino
!= srec
->ir_startino
)
470 /* both records must be sparse */
471 if (!xfs_inobt_issparse(trec
->ir_holemask
) ||
472 !xfs_inobt_issparse(srec
->ir_holemask
))
475 /* both records must track some inodes */
476 if (!trec
->ir_count
|| !srec
->ir_count
)
479 /* can't exceed capacity of a full record */
480 if (trec
->ir_count
+ srec
->ir_count
> XFS_INODES_PER_CHUNK
)
483 /* verify there is no allocation overlap */
484 talloc
= xfs_inobt_irec_to_allocmask(trec
);
485 salloc
= xfs_inobt_irec_to_allocmask(srec
);
493 * Merge the source inode record into the target. The caller must call
494 * __xfs_inobt_can_merge() to ensure the merge is valid.
497 __xfs_inobt_rec_merge(
498 struct xfs_inobt_rec_incore
*trec
, /* target */
499 struct xfs_inobt_rec_incore
*srec
) /* src */
501 ASSERT(trec
->ir_startino
== srec
->ir_startino
);
503 /* combine the counts */
504 trec
->ir_count
+= srec
->ir_count
;
505 trec
->ir_freecount
+= srec
->ir_freecount
;
508 * Merge the holemask and free mask. For both fields, 0 bits refer to
509 * allocated inodes. We combine the allocated ranges with bitwise AND.
511 trec
->ir_holemask
&= srec
->ir_holemask
;
512 trec
->ir_free
&= srec
->ir_free
;
516 * Insert a new sparse inode chunk into the associated inode btree. The inode
517 * record for the sparse chunk is pre-aligned to a startino that should match
518 * any pre-existing sparse inode record in the tree. This allows sparse chunks
521 * This function supports two modes of handling preexisting records depending on
522 * the merge flag. If merge is true, the provided record is merged with the
523 * existing record and updated in place. The merged record is returned in nrec.
524 * If merge is false, an existing record is replaced with the provided record.
525 * If no preexisting record exists, the provided record is always inserted.
527 * It is considered corruption if a merge is requested and not possible. Given
528 * the sparse inode alignment constraints, this should never happen.
531 xfs_inobt_insert_sprec(
532 struct xfs_mount
*mp
,
533 struct xfs_trans
*tp
,
534 struct xfs_buf
*agbp
,
536 struct xfs_inobt_rec_incore
*nrec
, /* in/out: new/merged rec. */
537 bool merge
) /* merge or replace */
539 struct xfs_btree_cur
*cur
;
540 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
541 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
544 struct xfs_inobt_rec_incore rec
;
546 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
548 /* the new record is pre-aligned so we know where to look */
549 error
= xfs_inobt_lookup(cur
, nrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
552 /* if nothing there, insert a new record and return */
554 error
= xfs_inobt_insert_rec(cur
, nrec
->ir_holemask
,
555 nrec
->ir_count
, nrec
->ir_freecount
,
559 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
565 * A record exists at this startino. Merge or replace the record
566 * depending on what we've been asked to do.
569 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
572 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
573 XFS_WANT_CORRUPTED_GOTO(mp
,
574 rec
.ir_startino
== nrec
->ir_startino
,
578 * This should never fail. If we have coexisting records that
579 * cannot merge, something is seriously wrong.
581 XFS_WANT_CORRUPTED_GOTO(mp
, __xfs_inobt_can_merge(nrec
, &rec
),
584 trace_xfs_irec_merge_pre(mp
, agno
, rec
.ir_startino
,
585 rec
.ir_holemask
, nrec
->ir_startino
,
588 /* merge to nrec to output the updated record */
589 __xfs_inobt_rec_merge(nrec
, &rec
);
591 trace_xfs_irec_merge_post(mp
, agno
, nrec
->ir_startino
,
594 error
= xfs_inobt_rec_check_count(mp
, nrec
);
599 error
= xfs_inobt_update(cur
, nrec
);
604 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
607 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
612 * Allocate new inodes in the allocation group specified by agbp.
613 * Return 0 for success, else error code.
615 STATIC
int /* error code or 0 */
617 xfs_trans_t
*tp
, /* transaction pointer */
618 xfs_buf_t
*agbp
, /* alloc group buffer */
621 xfs_agi_t
*agi
; /* allocation group header */
622 xfs_alloc_arg_t args
; /* allocation argument structure */
625 xfs_agino_t newino
; /* new first inode's number */
626 xfs_agino_t newlen
; /* new number of inodes */
627 int isaligned
= 0; /* inode allocation at stripe unit */
629 uint16_t allocmask
= (uint16_t) -1; /* init. to full chunk */
630 struct xfs_inobt_rec_incore rec
;
631 struct xfs_perag
*pag
;
634 memset(&args
, 0, sizeof(args
));
636 args
.mp
= tp
->t_mountp
;
637 args
.fsbno
= NULLFSBLOCK
;
638 xfs_rmap_ag_owner(&args
.oinfo
, XFS_RMAP_OWN_INODES
);
641 /* randomly do sparse inode allocations */
642 if (xfs_sb_version_hassparseinodes(&tp
->t_mountp
->m_sb
) &&
643 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
)
644 do_sparse
= prandom_u32() & 1;
648 * Locking will ensure that we don't have two callers in here
651 newlen
= args
.mp
->m_ialloc_inos
;
652 if (args
.mp
->m_maxicount
&&
653 percpu_counter_read_positive(&args
.mp
->m_icount
) + newlen
>
654 args
.mp
->m_maxicount
)
656 args
.minlen
= args
.maxlen
= args
.mp
->m_ialloc_blks
;
658 * First try to allocate inodes contiguous with the last-allocated
659 * chunk of inodes. If the filesystem is striped, this will fill
660 * an entire stripe unit with inodes.
662 agi
= XFS_BUF_TO_AGI(agbp
);
663 newino
= be32_to_cpu(agi
->agi_newino
);
664 agno
= be32_to_cpu(agi
->agi_seqno
);
665 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
666 args
.mp
->m_ialloc_blks
;
669 if (likely(newino
!= NULLAGINO
&&
670 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
671 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
672 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
676 * We need to take into account alignment here to ensure that
677 * we don't modify the free list if we fail to have an exact
678 * block. If we don't have an exact match, and every oher
679 * attempt allocation attempt fails, we'll end up cancelling
680 * a dirty transaction and shutting down.
682 * For an exact allocation, alignment must be 1,
683 * however we need to take cluster alignment into account when
684 * fixing up the freelist. Use the minalignslop field to
685 * indicate that extra blocks might be required for alignment,
686 * but not to use them in the actual exact allocation.
689 args
.minalignslop
= xfs_ialloc_cluster_alignment(args
.mp
) - 1;
691 /* Allow space for the inode btree to split. */
692 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
693 if ((error
= xfs_alloc_vextent(&args
)))
697 * This request might have dirtied the transaction if the AG can
698 * satisfy the request, but the exact block was not available.
699 * If the allocation did fail, subsequent requests will relax
700 * the exact agbno requirement and increase the alignment
701 * instead. It is critical that the total size of the request
702 * (len + alignment + slop) does not increase from this point
703 * on, so reset minalignslop to ensure it is not included in
704 * subsequent requests.
706 args
.minalignslop
= 0;
709 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
711 * Set the alignment for the allocation.
712 * If stripe alignment is turned on then align at stripe unit
714 * If the cluster size is smaller than a filesystem block
715 * then we're doing I/O for inodes in filesystem block size
716 * pieces, so don't need alignment anyway.
719 if (args
.mp
->m_sinoalign
) {
720 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
721 args
.alignment
= args
.mp
->m_dalign
;
724 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
726 * Need to figure out where to allocate the inode blocks.
727 * Ideally they should be spaced out through the a.g.
728 * For now, just allocate blocks up front.
730 args
.agbno
= be32_to_cpu(agi
->agi_root
);
731 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
733 * Allocate a fixed-size extent of inodes.
735 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
738 * Allow space for the inode btree to split.
740 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
741 if ((error
= xfs_alloc_vextent(&args
)))
746 * If stripe alignment is turned on, then try again with cluster
749 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
750 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
751 args
.agbno
= be32_to_cpu(agi
->agi_root
);
752 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
753 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
754 if ((error
= xfs_alloc_vextent(&args
)))
759 * Finally, try a sparse allocation if the filesystem supports it and
760 * the sparse allocation length is smaller than a full chunk.
762 if (xfs_sb_version_hassparseinodes(&args
.mp
->m_sb
) &&
763 args
.mp
->m_ialloc_min_blks
< args
.mp
->m_ialloc_blks
&&
764 args
.fsbno
== NULLFSBLOCK
) {
766 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
767 args
.agbno
= be32_to_cpu(agi
->agi_root
);
768 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
769 args
.alignment
= args
.mp
->m_sb
.sb_spino_align
;
772 args
.minlen
= args
.mp
->m_ialloc_min_blks
;
773 args
.maxlen
= args
.minlen
;
776 * The inode record will be aligned to full chunk size. We must
777 * prevent sparse allocation from AG boundaries that result in
778 * invalid inode records, such as records that start at agbno 0
779 * or extend beyond the AG.
781 * Set min agbno to the first aligned, non-zero agbno and max to
782 * the last aligned agbno that is at least one full chunk from
785 args
.min_agbno
= args
.mp
->m_sb
.sb_inoalignmt
;
786 args
.max_agbno
= round_down(args
.mp
->m_sb
.sb_agblocks
,
787 args
.mp
->m_sb
.sb_inoalignmt
) -
788 args
.mp
->m_ialloc_blks
;
790 error
= xfs_alloc_vextent(&args
);
794 newlen
= args
.len
<< args
.mp
->m_sb
.sb_inopblog
;
795 ASSERT(newlen
<= XFS_INODES_PER_CHUNK
);
796 allocmask
= (1 << (newlen
/ XFS_INODES_PER_HOLEMASK_BIT
)) - 1;
799 if (args
.fsbno
== NULLFSBLOCK
) {
803 ASSERT(args
.len
== args
.minlen
);
806 * Stamp and write the inode buffers.
808 * Seed the new inode cluster with a random generation number. This
809 * prevents short-term reuse of generation numbers if a chunk is
810 * freed and then immediately reallocated. We use random numbers
811 * rather than a linear progression to prevent the next generation
812 * number from being easily guessable.
814 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, newlen
, agno
,
815 args
.agbno
, args
.len
, prandom_u32());
820 * Convert the results.
822 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
824 if (xfs_inobt_issparse(~allocmask
)) {
826 * We've allocated a sparse chunk. Align the startino and mask.
828 xfs_align_sparse_ino(args
.mp
, &newino
, &allocmask
);
830 rec
.ir_startino
= newino
;
831 rec
.ir_holemask
= ~allocmask
;
832 rec
.ir_count
= newlen
;
833 rec
.ir_freecount
= newlen
;
834 rec
.ir_free
= XFS_INOBT_ALL_FREE
;
837 * Insert the sparse record into the inobt and allow for a merge
838 * if necessary. If a merge does occur, rec is updated to the
841 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
, XFS_BTNUM_INO
,
843 if (error
== -EFSCORRUPTED
) {
845 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
846 XFS_AGINO_TO_INO(args
.mp
, agno
,
848 rec
.ir_holemask
, rec
.ir_count
);
849 xfs_force_shutdown(args
.mp
, SHUTDOWN_CORRUPT_INCORE
);
855 * We can't merge the part we've just allocated as for the inobt
856 * due to finobt semantics. The original record may or may not
857 * exist independent of whether physical inodes exist in this
860 * We must update the finobt record based on the inobt record.
861 * rec contains the fully merged and up to date inobt record
862 * from the previous call. Set merge false to replace any
863 * existing record with this one.
865 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
866 error
= xfs_inobt_insert_sprec(args
.mp
, tp
, agbp
,
867 XFS_BTNUM_FINO
, &rec
,
873 /* full chunk - insert new records to both btrees */
874 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
879 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
880 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
,
881 newlen
, XFS_BTNUM_FINO
);
888 * Update AGI counts and newino.
890 be32_add_cpu(&agi
->agi_count
, newlen
);
891 be32_add_cpu(&agi
->agi_freecount
, newlen
);
892 pag
= xfs_perag_get(args
.mp
, agno
);
893 pag
->pagi_freecount
+= newlen
;
894 pag
->pagi_count
+= newlen
;
896 agi
->agi_newino
= cpu_to_be32(newino
);
899 * Log allocation group header fields
901 xfs_ialloc_log_agi(tp
, agbp
,
902 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
904 * Modify/log superblock values for inode count and inode free count.
906 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
907 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
912 STATIC xfs_agnumber_t
918 spin_lock(&mp
->m_agirotor_lock
);
919 agno
= mp
->m_agirotor
;
920 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
922 spin_unlock(&mp
->m_agirotor_lock
);
928 * Select an allocation group to look for a free inode in, based on the parent
929 * inode and the mode. Return the allocation group buffer.
931 STATIC xfs_agnumber_t
932 xfs_ialloc_ag_select(
933 xfs_trans_t
*tp
, /* transaction pointer */
934 xfs_ino_t parent
, /* parent directory inode number */
935 umode_t mode
) /* bits set to indicate file type */
937 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
938 xfs_agnumber_t agno
; /* current ag number */
939 int flags
; /* alloc buffer locking flags */
940 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
941 xfs_extlen_t longest
= 0; /* longest extent available */
942 xfs_mount_t
*mp
; /* mount point structure */
943 int needspace
; /* file mode implies space allocated */
944 xfs_perag_t
*pag
; /* per allocation group data */
945 xfs_agnumber_t pagno
; /* parent (starting) ag number */
949 * Files of these types need at least one block if length > 0
950 * (and they won't fit in the inode, but that's hard to figure out).
952 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
954 agcount
= mp
->m_maxagi
;
956 pagno
= xfs_ialloc_next_ag(mp
);
958 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
959 if (pagno
>= agcount
)
963 ASSERT(pagno
< agcount
);
966 * Loop through allocation groups, looking for one with a little
967 * free space in it. Note we don't look for free inodes, exactly.
968 * Instead, we include whether there is a need to allocate inodes
969 * to mean that blocks must be allocated for them,
970 * if none are currently free.
973 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
975 pag
= xfs_perag_get(mp
, agno
);
976 if (!pag
->pagi_inodeok
) {
977 xfs_ialloc_next_ag(mp
);
981 if (!pag
->pagi_init
) {
982 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
987 if (pag
->pagi_freecount
) {
992 if (!pag
->pagf_init
) {
993 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
999 * Check that there is enough free space for the file plus a
1000 * chunk of inodes if we need to allocate some. If this is the
1001 * first pass across the AGs, take into account the potential
1002 * space needed for alignment of inode chunks when checking the
1003 * longest contiguous free space in the AG - this prevents us
1004 * from getting ENOSPC because we have free space larger than
1005 * m_ialloc_blks but alignment constraints prevent us from using
1008 * If we can't find an AG with space for full alignment slack to
1009 * be taken into account, we must be near ENOSPC in all AGs.
1010 * Hence we don't include alignment for the second pass and so
1011 * if we fail allocation due to alignment issues then it is most
1012 * likely a real ENOSPC condition.
1014 ineed
= mp
->m_ialloc_min_blks
;
1015 if (flags
&& ineed
> 1)
1016 ineed
+= xfs_ialloc_cluster_alignment(mp
);
1017 longest
= pag
->pagf_longest
;
1019 longest
= pag
->pagf_flcount
> 0;
1021 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
1029 * No point in iterating over the rest, if we're shutting
1032 if (XFS_FORCED_SHUTDOWN(mp
))
1033 return NULLAGNUMBER
;
1035 if (agno
>= agcount
)
1037 if (agno
== pagno
) {
1039 return NULLAGNUMBER
;
1046 * Try to retrieve the next record to the left/right from the current one.
1049 xfs_ialloc_next_rec(
1050 struct xfs_btree_cur
*cur
,
1051 xfs_inobt_rec_incore_t
*rec
,
1059 error
= xfs_btree_decrement(cur
, 0, &i
);
1061 error
= xfs_btree_increment(cur
, 0, &i
);
1067 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1070 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1078 struct xfs_btree_cur
*cur
,
1080 xfs_inobt_rec_incore_t
*rec
,
1086 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
1091 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1094 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1101 * Return the offset of the first free inode in the record. If the inode chunk
1102 * is sparsely allocated, we convert the record holemask to inode granularity
1103 * and mask off the unallocated regions from the inode free mask.
1106 xfs_inobt_first_free_inode(
1107 struct xfs_inobt_rec_incore
*rec
)
1109 xfs_inofree_t realfree
;
1111 /* if there are no holes, return the first available offset */
1112 if (!xfs_inobt_issparse(rec
->ir_holemask
))
1113 return xfs_lowbit64(rec
->ir_free
);
1115 realfree
= xfs_inobt_irec_to_allocmask(rec
);
1116 realfree
&= rec
->ir_free
;
1118 return xfs_lowbit64(realfree
);
1122 * Allocate an inode using the inobt-only algorithm.
1125 xfs_dialloc_ag_inobt(
1126 struct xfs_trans
*tp
,
1127 struct xfs_buf
*agbp
,
1131 struct xfs_mount
*mp
= tp
->t_mountp
;
1132 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1133 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1134 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1135 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1136 struct xfs_perag
*pag
;
1137 struct xfs_btree_cur
*cur
, *tcur
;
1138 struct xfs_inobt_rec_incore rec
, trec
;
1143 int searchdistance
= 10;
1145 pag
= xfs_perag_get(mp
, agno
);
1147 ASSERT(pag
->pagi_init
);
1148 ASSERT(pag
->pagi_inodeok
);
1149 ASSERT(pag
->pagi_freecount
> 0);
1152 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1154 * If pagino is 0 (this is the root inode allocation) use newino.
1155 * This must work because we've just allocated some.
1158 pagino
= be32_to_cpu(agi
->agi_newino
);
1160 error
= xfs_check_agi_freecount(cur
, agi
);
1165 * If in the same AG as the parent, try to get near the parent.
1167 if (pagno
== agno
) {
1168 int doneleft
; /* done, to the left */
1169 int doneright
; /* done, to the right */
1171 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
1174 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1176 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1179 XFS_WANT_CORRUPTED_GOTO(mp
, j
== 1, error0
);
1181 if (rec
.ir_freecount
> 0) {
1183 * Found a free inode in the same chunk
1184 * as the parent, done.
1191 * In the same AG as parent, but parent's chunk is full.
1194 /* duplicate the cursor, search left & right simultaneously */
1195 error
= xfs_btree_dup_cursor(cur
, &tcur
);
1200 * Skip to last blocks looked up if same parent inode.
1202 if (pagino
!= NULLAGINO
&&
1203 pag
->pagl_pagino
== pagino
&&
1204 pag
->pagl_leftrec
!= NULLAGINO
&&
1205 pag
->pagl_rightrec
!= NULLAGINO
) {
1206 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
1211 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
1216 /* search left with tcur, back up 1 record */
1217 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
1221 /* search right with cur, go forward 1 record. */
1222 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
1228 * Loop until we find an inode chunk with a free inode.
1230 while (--searchdistance
> 0 && (!doneleft
|| !doneright
)) {
1231 int useleft
; /* using left inode chunk this time */
1233 /* figure out the closer block if both are valid. */
1234 if (!doneleft
&& !doneright
) {
1236 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
1237 rec
.ir_startino
- pagino
;
1239 useleft
= !doneleft
;
1242 /* free inodes to the left? */
1243 if (useleft
&& trec
.ir_freecount
) {
1244 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1247 pag
->pagl_leftrec
= trec
.ir_startino
;
1248 pag
->pagl_rightrec
= rec
.ir_startino
;
1249 pag
->pagl_pagino
= pagino
;
1254 /* free inodes to the right? */
1255 if (!useleft
&& rec
.ir_freecount
) {
1256 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1258 pag
->pagl_leftrec
= trec
.ir_startino
;
1259 pag
->pagl_rightrec
= rec
.ir_startino
;
1260 pag
->pagl_pagino
= pagino
;
1264 /* get next record to check */
1266 error
= xfs_ialloc_next_rec(tcur
, &trec
,
1269 error
= xfs_ialloc_next_rec(cur
, &rec
,
1276 if (searchdistance
<= 0) {
1278 * Not in range - save last search
1279 * location and allocate a new inode
1281 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1282 pag
->pagl_leftrec
= trec
.ir_startino
;
1283 pag
->pagl_rightrec
= rec
.ir_startino
;
1284 pag
->pagl_pagino
= pagino
;
1288 * We've reached the end of the btree. because
1289 * we are only searching a small chunk of the
1290 * btree each search, there is obviously free
1291 * inodes closer to the parent inode than we
1292 * are now. restart the search again.
1294 pag
->pagl_pagino
= NULLAGINO
;
1295 pag
->pagl_leftrec
= NULLAGINO
;
1296 pag
->pagl_rightrec
= NULLAGINO
;
1297 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
1298 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1304 * In a different AG from the parent.
1305 * See if the most recently allocated block has any free.
1307 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1308 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1314 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
1318 if (j
== 1 && rec
.ir_freecount
> 0) {
1320 * The last chunk allocated in the group
1321 * still has a free inode.
1329 * None left in the last group, search the whole AG
1331 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1334 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1337 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1340 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1341 if (rec
.ir_freecount
> 0)
1343 error
= xfs_btree_increment(cur
, 0, &i
);
1346 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1350 offset
= xfs_inobt_first_free_inode(&rec
);
1351 ASSERT(offset
>= 0);
1352 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1353 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1354 XFS_INODES_PER_CHUNK
) == 0);
1355 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1356 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1358 error
= xfs_inobt_update(cur
, &rec
);
1361 be32_add_cpu(&agi
->agi_freecount
, -1);
1362 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1363 pag
->pagi_freecount
--;
1365 error
= xfs_check_agi_freecount(cur
, agi
);
1369 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1370 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1375 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
1377 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1383 * Use the free inode btree to allocate an inode based on distance from the
1384 * parent. Note that the provided cursor may be deleted and replaced.
1387 xfs_dialloc_ag_finobt_near(
1389 struct xfs_btree_cur
**ocur
,
1390 struct xfs_inobt_rec_incore
*rec
)
1392 struct xfs_btree_cur
*lcur
= *ocur
; /* left search cursor */
1393 struct xfs_btree_cur
*rcur
; /* right search cursor */
1394 struct xfs_inobt_rec_incore rrec
;
1398 error
= xfs_inobt_lookup(lcur
, pagino
, XFS_LOOKUP_LE
, &i
);
1403 error
= xfs_inobt_get_rec(lcur
, rec
, &i
);
1406 XFS_WANT_CORRUPTED_RETURN(lcur
->bc_mp
, i
== 1);
1409 * See if we've landed in the parent inode record. The finobt
1410 * only tracks chunks with at least one free inode, so record
1411 * existence is enough.
1413 if (pagino
>= rec
->ir_startino
&&
1414 pagino
< (rec
->ir_startino
+ XFS_INODES_PER_CHUNK
))
1418 error
= xfs_btree_dup_cursor(lcur
, &rcur
);
1422 error
= xfs_inobt_lookup(rcur
, pagino
, XFS_LOOKUP_GE
, &j
);
1426 error
= xfs_inobt_get_rec(rcur
, &rrec
, &j
);
1429 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, j
== 1, error_rcur
);
1432 XFS_WANT_CORRUPTED_GOTO(lcur
->bc_mp
, i
== 1 || j
== 1, error_rcur
);
1433 if (i
== 1 && j
== 1) {
1435 * Both the left and right records are valid. Choose the closer
1436 * inode chunk to the target.
1438 if ((pagino
- rec
->ir_startino
+ XFS_INODES_PER_CHUNK
- 1) >
1439 (rrec
.ir_startino
- pagino
)) {
1441 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1444 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1446 } else if (j
== 1) {
1447 /* only the right record is valid */
1449 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1451 } else if (i
== 1) {
1452 /* only the left record is valid */
1453 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1459 xfs_btree_del_cursor(rcur
, XFS_BTREE_ERROR
);
1464 * Use the free inode btree to find a free inode based on a newino hint. If
1465 * the hint is NULL, find the first free inode in the AG.
1468 xfs_dialloc_ag_finobt_newino(
1469 struct xfs_agi
*agi
,
1470 struct xfs_btree_cur
*cur
,
1471 struct xfs_inobt_rec_incore
*rec
)
1476 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1477 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1482 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1485 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1491 * Find the first inode available in the AG.
1493 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1496 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1498 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1501 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1507 * Update the inobt based on a modification made to the finobt. Also ensure that
1508 * the records from both trees are equivalent post-modification.
1511 xfs_dialloc_ag_update_inobt(
1512 struct xfs_btree_cur
*cur
, /* inobt cursor */
1513 struct xfs_inobt_rec_incore
*frec
, /* finobt record */
1514 int offset
) /* inode offset */
1516 struct xfs_inobt_rec_incore rec
;
1520 error
= xfs_inobt_lookup(cur
, frec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1523 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1525 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1528 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1529 ASSERT((XFS_AGINO_TO_OFFSET(cur
->bc_mp
, rec
.ir_startino
) %
1530 XFS_INODES_PER_CHUNK
) == 0);
1532 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1535 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, (rec
.ir_free
== frec
->ir_free
) &&
1536 (rec
.ir_freecount
== frec
->ir_freecount
));
1538 return xfs_inobt_update(cur
, &rec
);
1542 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1543 * back to the inobt search algorithm.
1545 * The caller selected an AG for us, and made sure that free inodes are
1550 struct xfs_trans
*tp
,
1551 struct xfs_buf
*agbp
,
1555 struct xfs_mount
*mp
= tp
->t_mountp
;
1556 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1557 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1558 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1559 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1560 struct xfs_perag
*pag
;
1561 struct xfs_btree_cur
*cur
; /* finobt cursor */
1562 struct xfs_btree_cur
*icur
; /* inobt cursor */
1563 struct xfs_inobt_rec_incore rec
;
1569 if (!xfs_sb_version_hasfinobt(&mp
->m_sb
))
1570 return xfs_dialloc_ag_inobt(tp
, agbp
, parent
, inop
);
1572 pag
= xfs_perag_get(mp
, agno
);
1575 * If pagino is 0 (this is the root inode allocation) use newino.
1576 * This must work because we've just allocated some.
1579 pagino
= be32_to_cpu(agi
->agi_newino
);
1581 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1583 error
= xfs_check_agi_freecount(cur
, agi
);
1588 * The search algorithm depends on whether we're in the same AG as the
1589 * parent. If so, find the closest available inode to the parent. If
1590 * not, consider the agi hint or find the first free inode in the AG.
1593 error
= xfs_dialloc_ag_finobt_near(pagino
, &cur
, &rec
);
1595 error
= xfs_dialloc_ag_finobt_newino(agi
, cur
, &rec
);
1599 offset
= xfs_inobt_first_free_inode(&rec
);
1600 ASSERT(offset
>= 0);
1601 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1602 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1603 XFS_INODES_PER_CHUNK
) == 0);
1604 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1607 * Modify or remove the finobt record.
1609 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1611 if (rec
.ir_freecount
)
1612 error
= xfs_inobt_update(cur
, &rec
);
1614 error
= xfs_btree_delete(cur
, &i
);
1619 * The finobt has now been updated appropriately. We haven't updated the
1620 * agi and superblock yet, so we can create an inobt cursor and validate
1621 * the original freecount. If all is well, make the equivalent update to
1622 * the inobt using the finobt record and offset information.
1624 icur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1626 error
= xfs_check_agi_freecount(icur
, agi
);
1630 error
= xfs_dialloc_ag_update_inobt(icur
, &rec
, offset
);
1635 * Both trees have now been updated. We must update the perag and
1636 * superblock before we can check the freecount for each btree.
1638 be32_add_cpu(&agi
->agi_freecount
, -1);
1639 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1640 pag
->pagi_freecount
--;
1642 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1644 error
= xfs_check_agi_freecount(icur
, agi
);
1647 error
= xfs_check_agi_freecount(cur
, agi
);
1651 xfs_btree_del_cursor(icur
, XFS_BTREE_NOERROR
);
1652 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1658 xfs_btree_del_cursor(icur
, XFS_BTREE_ERROR
);
1660 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1666 * Allocate an inode on disk.
1668 * Mode is used to tell whether the new inode will need space, and whether it
1671 * This function is designed to be called twice if it has to do an allocation
1672 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1673 * If an inode is available without having to performn an allocation, an inode
1674 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1675 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1676 * The caller should then commit the current transaction, allocate a
1677 * new transaction, and call xfs_dialloc() again, passing in the previous value
1678 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1679 * buffer is locked across the two calls, the second call is guaranteed to have
1680 * a free inode available.
1682 * Once we successfully pick an inode its number is returned and the on-disk
1683 * data structures are updated. The inode itself is not read in, since doing so
1684 * would break ordering constraints with xfs_reclaim.
1688 struct xfs_trans
*tp
,
1691 struct xfs_buf
**IO_agbp
,
1694 struct xfs_mount
*mp
= tp
->t_mountp
;
1695 struct xfs_buf
*agbp
;
1696 xfs_agnumber_t agno
;
1700 xfs_agnumber_t start_agno
;
1701 struct xfs_perag
*pag
;
1706 * If the caller passes in a pointer to the AGI buffer,
1707 * continue where we left off before. In this case, we
1708 * know that the allocation group has free inodes.
1715 * We do not have an agbp, so select an initial allocation
1716 * group for inode allocation.
1718 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
);
1719 if (start_agno
== NULLAGNUMBER
) {
1725 * If we have already hit the ceiling of inode blocks then clear
1726 * okalloc so we scan all available agi structures for a free
1729 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1730 * which will sacrifice the preciseness but improve the performance.
1732 if (mp
->m_maxicount
&&
1733 percpu_counter_read_positive(&mp
->m_icount
) + mp
->m_ialloc_inos
1734 > mp
->m_maxicount
) {
1740 * Loop until we find an allocation group that either has free inodes
1741 * or in which we can allocate some inodes. Iterate through the
1742 * allocation groups upward, wrapping at the end.
1746 pag
= xfs_perag_get(mp
, agno
);
1747 if (!pag
->pagi_inodeok
) {
1748 xfs_ialloc_next_ag(mp
);
1752 if (!pag
->pagi_init
) {
1753 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1759 * Do a first racy fast path check if this AG is usable.
1761 if (!pag
->pagi_freecount
&& !okalloc
)
1765 * Then read in the AGI buffer and recheck with the AGI buffer
1768 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1772 if (pag
->pagi_freecount
) {
1778 goto nextag_relse_buffer
;
1781 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1783 xfs_trans_brelse(tp
, agbp
);
1785 if (error
!= -ENOSPC
)
1795 * We successfully allocated some inodes, return
1796 * the current context to the caller so that it
1797 * can commit the current transaction and call
1798 * us again where we left off.
1800 ASSERT(pag
->pagi_freecount
> 0);
1808 nextag_relse_buffer
:
1809 xfs_trans_brelse(tp
, agbp
);
1812 if (++agno
== mp
->m_sb
.sb_agcount
)
1814 if (agno
== start_agno
) {
1816 return noroom
? -ENOSPC
: 0;
1822 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1829 * Free the blocks of an inode chunk. We must consider that the inode chunk
1830 * might be sparse and only free the regions that are allocated as part of the
1834 xfs_difree_inode_chunk(
1835 struct xfs_mount
*mp
,
1836 xfs_agnumber_t agno
,
1837 struct xfs_inobt_rec_incore
*rec
,
1838 struct xfs_defer_ops
*dfops
)
1840 xfs_agblock_t sagbno
= XFS_AGINO_TO_AGBNO(mp
, rec
->ir_startino
);
1841 int startidx
, endidx
;
1843 xfs_agblock_t agbno
;
1845 struct xfs_owner_info oinfo
;
1846 DECLARE_BITMAP(holemask
, XFS_INOBT_HOLEMASK_BITS
);
1847 xfs_rmap_ag_owner(&oinfo
, XFS_RMAP_OWN_INODES
);
1849 if (!xfs_inobt_issparse(rec
->ir_holemask
)) {
1850 /* not sparse, calculate extent info directly */
1851 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, sagbno
),
1852 mp
->m_ialloc_blks
, &oinfo
);
1856 /* holemask is only 16-bits (fits in an unsigned long) */
1857 ASSERT(sizeof(rec
->ir_holemask
) <= sizeof(holemask
[0]));
1858 holemask
[0] = rec
->ir_holemask
;
1861 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1862 * holemask and convert the start/end index of each range to an extent.
1863 * We start with the start and end index both pointing at the first 0 in
1866 startidx
= endidx
= find_first_zero_bit(holemask
,
1867 XFS_INOBT_HOLEMASK_BITS
);
1868 nextbit
= startidx
+ 1;
1869 while (startidx
< XFS_INOBT_HOLEMASK_BITS
) {
1870 nextbit
= find_next_zero_bit(holemask
, XFS_INOBT_HOLEMASK_BITS
,
1873 * If the next zero bit is contiguous, update the end index of
1874 * the current range and continue.
1876 if (nextbit
!= XFS_INOBT_HOLEMASK_BITS
&&
1877 nextbit
== endidx
+ 1) {
1883 * nextbit is not contiguous with the current end index. Convert
1884 * the current start/end to an extent and add it to the free
1887 agbno
= sagbno
+ (startidx
* XFS_INODES_PER_HOLEMASK_BIT
) /
1888 mp
->m_sb
.sb_inopblock
;
1889 contigblk
= ((endidx
- startidx
+ 1) *
1890 XFS_INODES_PER_HOLEMASK_BIT
) /
1891 mp
->m_sb
.sb_inopblock
;
1893 ASSERT(agbno
% mp
->m_sb
.sb_spino_align
== 0);
1894 ASSERT(contigblk
% mp
->m_sb
.sb_spino_align
== 0);
1895 xfs_bmap_add_free(mp
, dfops
, XFS_AGB_TO_FSB(mp
, agno
, agbno
),
1898 /* reset range to current bit and carry on... */
1899 startidx
= endidx
= nextbit
;
1908 struct xfs_mount
*mp
,
1909 struct xfs_trans
*tp
,
1910 struct xfs_buf
*agbp
,
1912 struct xfs_icluster
*xic
,
1913 struct xfs_inobt_rec_incore
*orec
)
1915 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1916 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1917 struct xfs_perag
*pag
;
1918 struct xfs_btree_cur
*cur
;
1919 struct xfs_inobt_rec_incore rec
;
1925 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1926 ASSERT(XFS_AGINO_TO_AGBNO(mp
, agino
) < be32_to_cpu(agi
->agi_length
));
1929 * Initialize the cursor.
1931 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1933 error
= xfs_check_agi_freecount(cur
, agi
);
1938 * Look for the entry describing this inode.
1940 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1941 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1945 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1946 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1948 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1952 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error0
);
1954 * Get the offset in the inode chunk.
1956 off
= agino
- rec
.ir_startino
;
1957 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1958 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1960 * Mark the inode free & increment the count.
1962 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1966 * When an inode chunk is free, it becomes eligible for removal. Don't
1967 * remove the chunk if the block size is large enough for multiple inode
1968 * chunks (that might not be free).
1970 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1971 rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
1972 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
) {
1973 xic
->deleted
= true;
1974 xic
->first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1975 xic
->alloc
= xfs_inobt_irec_to_allocmask(&rec
);
1978 * Remove the inode cluster from the AGI B+Tree, adjust the
1979 * AGI and Superblock inode counts, and mark the disk space
1980 * to be freed when the transaction is committed.
1982 ilen
= rec
.ir_freecount
;
1983 be32_add_cpu(&agi
->agi_count
, -ilen
);
1984 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1985 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1986 pag
= xfs_perag_get(mp
, agno
);
1987 pag
->pagi_freecount
-= ilen
- 1;
1988 pag
->pagi_count
-= ilen
;
1990 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1991 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1993 if ((error
= xfs_btree_delete(cur
, &i
))) {
1994 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1999 xfs_difree_inode_chunk(mp
, agno
, &rec
, tp
->t_dfops
);
2001 xic
->deleted
= false;
2003 error
= xfs_inobt_update(cur
, &rec
);
2005 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
2011 * Change the inode free counts and log the ag/sb changes.
2013 be32_add_cpu(&agi
->agi_freecount
, 1);
2014 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
2015 pag
= xfs_perag_get(mp
, agno
);
2016 pag
->pagi_freecount
++;
2018 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
2021 error
= xfs_check_agi_freecount(cur
, agi
);
2026 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2030 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2035 * Free an inode in the free inode btree.
2039 struct xfs_mount
*mp
,
2040 struct xfs_trans
*tp
,
2041 struct xfs_buf
*agbp
,
2043 struct xfs_inobt_rec_incore
*ibtrec
) /* inobt record */
2045 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
2046 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
2047 struct xfs_btree_cur
*cur
;
2048 struct xfs_inobt_rec_incore rec
;
2049 int offset
= agino
- ibtrec
->ir_startino
;
2053 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
2055 error
= xfs_inobt_lookup(cur
, ibtrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
2060 * If the record does not exist in the finobt, we must have just
2061 * freed an inode in a previously fully allocated chunk. If not,
2062 * something is out of sync.
2064 XFS_WANT_CORRUPTED_GOTO(mp
, ibtrec
->ir_freecount
== 1, error
);
2066 error
= xfs_inobt_insert_rec(cur
, ibtrec
->ir_holemask
,
2068 ibtrec
->ir_freecount
,
2069 ibtrec
->ir_free
, &i
);
2078 * Read and update the existing record. We could just copy the ibtrec
2079 * across here, but that would defeat the purpose of having redundant
2080 * metadata. By making the modifications independently, we can catch
2081 * corruptions that we wouldn't see if we just copied from one record
2084 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2087 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, error
);
2089 rec
.ir_free
|= XFS_INOBT_MASK(offset
);
2092 XFS_WANT_CORRUPTED_GOTO(mp
, (rec
.ir_free
== ibtrec
->ir_free
) &&
2093 (rec
.ir_freecount
== ibtrec
->ir_freecount
),
2097 * The content of inobt records should always match between the inobt
2098 * and finobt. The lifecycle of records in the finobt is different from
2099 * the inobt in that the finobt only tracks records with at least one
2100 * free inode. Hence, if all of the inodes are free and we aren't
2101 * keeping inode chunks permanently on disk, remove the record.
2102 * Otherwise, update the record with the new information.
2104 * Note that we currently can't free chunks when the block size is large
2105 * enough for multiple chunks. Leave the finobt record to remain in sync
2108 if (rec
.ir_free
== XFS_INOBT_ALL_FREE
&&
2109 mp
->m_sb
.sb_inopblock
<= XFS_INODES_PER_CHUNK
&&
2110 !(mp
->m_flags
& XFS_MOUNT_IKEEP
)) {
2111 error
= xfs_btree_delete(cur
, &i
);
2116 error
= xfs_inobt_update(cur
, &rec
);
2122 error
= xfs_check_agi_freecount(cur
, agi
);
2126 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
2130 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
2135 * Free disk inode. Carefully avoids touching the incore inode, all
2136 * manipulations incore are the caller's responsibility.
2137 * The on-disk inode is not changed by this operation, only the
2138 * btree (free inode mask) is changed.
2142 struct xfs_trans
*tp
, /* transaction pointer */
2143 xfs_ino_t inode
, /* inode to be freed */
2144 struct xfs_icluster
*xic
) /* cluster info if deleted */
2147 xfs_agblock_t agbno
; /* block number containing inode */
2148 struct xfs_buf
*agbp
; /* buffer for allocation group header */
2149 xfs_agino_t agino
; /* allocation group inode number */
2150 xfs_agnumber_t agno
; /* allocation group number */
2151 int error
; /* error return value */
2152 struct xfs_mount
*mp
; /* mount structure for filesystem */
2153 struct xfs_inobt_rec_incore rec
;/* btree record */
2158 * Break up inode number into its components.
2160 agno
= XFS_INO_TO_AGNO(mp
, inode
);
2161 if (agno
>= mp
->m_sb
.sb_agcount
) {
2162 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2163 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2167 agino
= XFS_INO_TO_AGINO(mp
, inode
);
2168 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2169 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2170 __func__
, (unsigned long long)inode
,
2171 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
2175 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2176 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2177 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2178 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
2183 * Get the allocation group header.
2185 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2187 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
2193 * Fix up the inode allocation btree.
2195 error
= xfs_difree_inobt(mp
, tp
, agbp
, agino
, xic
, &rec
);
2200 * Fix up the free inode btree.
2202 if (xfs_sb_version_hasfinobt(&mp
->m_sb
)) {
2203 error
= xfs_difree_finobt(mp
, tp
, agbp
, agino
, &rec
);
2216 struct xfs_mount
*mp
,
2217 struct xfs_trans
*tp
,
2218 xfs_agnumber_t agno
,
2220 xfs_agblock_t agbno
,
2221 xfs_agblock_t
*chunk_agbno
,
2222 xfs_agblock_t
*offset_agbno
,
2225 struct xfs_inobt_rec_incore rec
;
2226 struct xfs_btree_cur
*cur
;
2227 struct xfs_buf
*agbp
;
2231 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
2234 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2235 __func__
, error
, agno
);
2240 * Lookup the inode record for the given agino. If the record cannot be
2241 * found, then it's an invalid inode number and we should abort. Once
2242 * we have a record, we need to ensure it contains the inode number
2243 * we are looking up.
2245 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
2246 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
2249 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
2250 if (!error
&& i
== 0)
2254 xfs_trans_brelse(tp
, agbp
);
2255 xfs_btree_del_cursor(cur
, error
);
2259 /* check that the returned record contains the required inode */
2260 if (rec
.ir_startino
> agino
||
2261 rec
.ir_startino
+ mp
->m_ialloc_inos
<= agino
)
2264 /* for untrusted inodes check it is allocated first */
2265 if ((flags
& XFS_IGET_UNTRUSTED
) &&
2266 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
2269 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
2270 *offset_agbno
= agbno
- *chunk_agbno
;
2275 * Return the location of the inode in imap, for mapping it into a buffer.
2279 xfs_mount_t
*mp
, /* file system mount structure */
2280 xfs_trans_t
*tp
, /* transaction pointer */
2281 xfs_ino_t ino
, /* inode to locate */
2282 struct xfs_imap
*imap
, /* location map structure */
2283 uint flags
) /* flags for inode btree lookup */
2285 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
2286 xfs_agino_t agino
; /* inode number within alloc group */
2287 xfs_agnumber_t agno
; /* allocation group number */
2288 int blks_per_cluster
; /* num blocks per inode cluster */
2289 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
2290 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
2291 int error
; /* error code */
2292 int offset
; /* index of inode in its buffer */
2293 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
2295 ASSERT(ino
!= NULLFSINO
);
2298 * Split up the inode number into its parts.
2300 agno
= XFS_INO_TO_AGNO(mp
, ino
);
2301 agino
= XFS_INO_TO_AGINO(mp
, ino
);
2302 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
2303 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
2304 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2307 * Don't output diagnostic information for untrusted inodes
2308 * as they can be invalid without implying corruption.
2310 if (flags
& XFS_IGET_UNTRUSTED
)
2312 if (agno
>= mp
->m_sb
.sb_agcount
) {
2314 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2315 __func__
, agno
, mp
->m_sb
.sb_agcount
);
2317 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
2319 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2320 __func__
, (unsigned long long)agbno
,
2321 (unsigned long)mp
->m_sb
.sb_agblocks
);
2323 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
2325 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2327 XFS_AGINO_TO_INO(mp
, agno
, agino
));
2334 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
2337 * For bulkstat and handle lookups, we have an untrusted inode number
2338 * that we have to verify is valid. We cannot do this just by reading
2339 * the inode buffer as it may have been unlinked and removed leaving
2340 * inodes in stale state on disk. Hence we have to do a btree lookup
2341 * in all cases where an untrusted inode number is passed.
2343 if (flags
& XFS_IGET_UNTRUSTED
) {
2344 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2345 &chunk_agbno
, &offset_agbno
, flags
);
2352 * If the inode cluster size is the same as the blocksize or
2353 * smaller we get to the buffer by simple arithmetics.
2355 if (blks_per_cluster
== 1) {
2356 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
2357 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
2359 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
2360 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
2361 imap
->im_boffset
= (unsigned short)(offset
<<
2362 mp
->m_sb
.sb_inodelog
);
2367 * If the inode chunks are aligned then use simple maths to
2368 * find the location. Otherwise we have to do a btree
2369 * lookup to find the location.
2371 if (mp
->m_inoalign_mask
) {
2372 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
2373 chunk_agbno
= agbno
- offset_agbno
;
2375 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
2376 &chunk_agbno
, &offset_agbno
, flags
);
2382 ASSERT(agbno
>= chunk_agbno
);
2383 cluster_agbno
= chunk_agbno
+
2384 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
2385 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
2386 XFS_INO_TO_OFFSET(mp
, ino
);
2388 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
2389 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
2390 imap
->im_boffset
= (unsigned short)(offset
<< mp
->m_sb
.sb_inodelog
);
2393 * If the inode number maps to a block outside the bounds
2394 * of the file system then return NULL rather than calling
2395 * read_buf and panicing when we get an error from the
2398 if ((imap
->im_blkno
+ imap
->im_len
) >
2399 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
2401 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2402 __func__
, (unsigned long long) imap
->im_blkno
,
2403 (unsigned long long) imap
->im_len
,
2404 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
2411 * Compute and fill in value of m_in_maxlevels.
2414 xfs_ialloc_compute_maxlevels(
2415 xfs_mount_t
*mp
) /* file system mount structure */
2419 inodes
= (1LL << XFS_INO_AGINO_BITS(mp
)) >> XFS_INODES_PER_CHUNK_LOG
;
2420 mp
->m_in_maxlevels
= xfs_btree_compute_maxlevels(mp
->m_inobt_mnr
,
2425 * Log specified fields for the ag hdr (inode section). The growth of the agi
2426 * structure over time requires that we interpret the buffer as two logical
2427 * regions delineated by the end of the unlinked list. This is due to the size
2428 * of the hash table and its location in the middle of the agi.
2430 * For example, a request to log a field before agi_unlinked and a field after
2431 * agi_unlinked could cause us to log the entire hash table and use an excessive
2432 * amount of log space. To avoid this behavior, log the region up through
2433 * agi_unlinked in one call and the region after agi_unlinked through the end of
2434 * the structure in another.
2438 xfs_trans_t
*tp
, /* transaction pointer */
2439 xfs_buf_t
*bp
, /* allocation group header buffer */
2440 int fields
) /* bitmask of fields to log */
2442 int first
; /* first byte number */
2443 int last
; /* last byte number */
2444 static const short offsets
[] = { /* field starting offsets */
2445 /* keep in sync with bit definitions */
2446 offsetof(xfs_agi_t
, agi_magicnum
),
2447 offsetof(xfs_agi_t
, agi_versionnum
),
2448 offsetof(xfs_agi_t
, agi_seqno
),
2449 offsetof(xfs_agi_t
, agi_length
),
2450 offsetof(xfs_agi_t
, agi_count
),
2451 offsetof(xfs_agi_t
, agi_root
),
2452 offsetof(xfs_agi_t
, agi_level
),
2453 offsetof(xfs_agi_t
, agi_freecount
),
2454 offsetof(xfs_agi_t
, agi_newino
),
2455 offsetof(xfs_agi_t
, agi_dirino
),
2456 offsetof(xfs_agi_t
, agi_unlinked
),
2457 offsetof(xfs_agi_t
, agi_free_root
),
2458 offsetof(xfs_agi_t
, agi_free_level
),
2462 xfs_agi_t
*agi
; /* allocation group header */
2464 agi
= XFS_BUF_TO_AGI(bp
);
2465 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
2469 * Compute byte offsets for the first and last fields in the first
2470 * region and log the agi buffer. This only logs up through
2473 if (fields
& XFS_AGI_ALL_BITS_R1
) {
2474 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R1
,
2476 xfs_trans_log_buf(tp
, bp
, first
, last
);
2480 * Mask off the bits in the first region and calculate the first and
2481 * last field offsets for any bits in the second region.
2483 fields
&= ~XFS_AGI_ALL_BITS_R1
;
2485 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R2
,
2487 xfs_trans_log_buf(tp
, bp
, first
, last
);
2491 static xfs_failaddr_t
2495 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2496 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
2499 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2500 if (!uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_meta_uuid
))
2501 return __this_address
;
2502 if (!xfs_log_check_lsn(mp
,
2503 be64_to_cpu(XFS_BUF_TO_AGI(bp
)->agi_lsn
)))
2504 return __this_address
;
2508 * Validate the magic number of the agi block.
2510 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
2511 return __this_address
;
2512 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
2513 return __this_address
;
2515 if (be32_to_cpu(agi
->agi_level
) < 1 ||
2516 be32_to_cpu(agi
->agi_level
) > XFS_BTREE_MAXLEVELS
)
2517 return __this_address
;
2519 if (xfs_sb_version_hasfinobt(&mp
->m_sb
) &&
2520 (be32_to_cpu(agi
->agi_free_level
) < 1 ||
2521 be32_to_cpu(agi
->agi_free_level
) > XFS_BTREE_MAXLEVELS
))
2522 return __this_address
;
2525 * during growfs operations, the perag is not fully initialised,
2526 * so we can't use it for any useful checking. growfs ensures we can't
2527 * use it by using uncached buffers that don't have the perag attached
2528 * so we can detect and avoid this problem.
2530 if (bp
->b_pag
&& be32_to_cpu(agi
->agi_seqno
) != bp
->b_pag
->pag_agno
)
2531 return __this_address
;
2533 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++) {
2534 if (agi
->agi_unlinked
[i
] == cpu_to_be32(NULLAGINO
))
2536 if (!xfs_verify_ino(mp
, be32_to_cpu(agi
->agi_unlinked
[i
])))
2537 return __this_address
;
2544 xfs_agi_read_verify(
2547 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2550 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2551 !xfs_buf_verify_cksum(bp
, XFS_AGI_CRC_OFF
))
2552 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
2554 fa
= xfs_agi_verify(bp
);
2555 if (XFS_TEST_ERROR(fa
, mp
, XFS_ERRTAG_IALLOC_READ_AGI
))
2556 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
2561 xfs_agi_write_verify(
2564 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2565 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
2568 fa
= xfs_agi_verify(bp
);
2570 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
2574 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2578 XFS_BUF_TO_AGI(bp
)->agi_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2579 xfs_buf_update_cksum(bp
, XFS_AGI_CRC_OFF
);
2582 const struct xfs_buf_ops xfs_agi_buf_ops
= {
2584 .verify_read
= xfs_agi_read_verify
,
2585 .verify_write
= xfs_agi_write_verify
,
2586 .verify_struct
= xfs_agi_verify
,
2590 * Read in the allocation group header (inode allocation section)
2594 struct xfs_mount
*mp
, /* file system mount structure */
2595 struct xfs_trans
*tp
, /* transaction pointer */
2596 xfs_agnumber_t agno
, /* allocation group number */
2597 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2601 trace_xfs_read_agi(mp
, agno
);
2603 ASSERT(agno
!= NULLAGNUMBER
);
2604 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
2605 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
2606 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
2610 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_AGI_BUF
);
2612 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
2617 xfs_ialloc_read_agi(
2618 struct xfs_mount
*mp
, /* file system mount structure */
2619 struct xfs_trans
*tp
, /* transaction pointer */
2620 xfs_agnumber_t agno
, /* allocation group number */
2621 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2623 struct xfs_agi
*agi
; /* allocation group header */
2624 struct xfs_perag
*pag
; /* per allocation group data */
2627 trace_xfs_ialloc_read_agi(mp
, agno
);
2629 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
2633 agi
= XFS_BUF_TO_AGI(*bpp
);
2634 pag
= xfs_perag_get(mp
, agno
);
2635 if (!pag
->pagi_init
) {
2636 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
2637 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
2642 * It's possible for these to be out of sync if
2643 * we are in the middle of a forced shutdown.
2645 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
2646 XFS_FORCED_SHUTDOWN(mp
));
2652 * Read in the agi to initialise the per-ag data in the mount structure
2655 xfs_ialloc_pagi_init(
2656 xfs_mount_t
*mp
, /* file system mount structure */
2657 xfs_trans_t
*tp
, /* transaction pointer */
2658 xfs_agnumber_t agno
) /* allocation group number */
2660 xfs_buf_t
*bp
= NULL
;
2663 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
2667 xfs_trans_brelse(tp
, bp
);
2671 /* Is there an inode record covering a given range of inode numbers? */
2673 xfs_ialloc_has_inode_record(
2674 struct xfs_btree_cur
*cur
,
2679 struct xfs_inobt_rec_incore irec
;
2687 error
= xfs_inobt_lookup(cur
, low
, XFS_LOOKUP_LE
, &has_record
);
2688 while (error
== 0 && has_record
) {
2689 error
= xfs_inobt_get_rec(cur
, &irec
, &has_record
);
2690 if (error
|| irec
.ir_startino
> high
)
2693 agino
= irec
.ir_startino
;
2694 holemask
= irec
.ir_holemask
;
2695 for (i
= 0; i
< XFS_INOBT_HOLEMASK_BITS
; holemask
>>= 1,
2696 i
++, agino
+= XFS_INODES_PER_HOLEMASK_BIT
) {
2699 if (agino
+ XFS_INODES_PER_HOLEMASK_BIT
> low
&&
2706 error
= xfs_btree_increment(cur
, 0, &has_record
);
2711 /* Is there an inode record covering a given extent? */
2713 xfs_ialloc_has_inodes_at_extent(
2714 struct xfs_btree_cur
*cur
,
2722 low
= XFS_OFFBNO_TO_AGINO(cur
->bc_mp
, bno
, 0);
2723 high
= XFS_OFFBNO_TO_AGINO(cur
->bc_mp
, bno
+ len
, 0) - 1;
2725 return xfs_ialloc_has_inode_record(cur
, low
, high
, exists
);
2728 struct xfs_ialloc_count_inodes
{
2730 xfs_agino_t freecount
;
2733 /* Record inode counts across all inobt records. */
2735 xfs_ialloc_count_inodes_rec(
2736 struct xfs_btree_cur
*cur
,
2737 union xfs_btree_rec
*rec
,
2740 struct xfs_inobt_rec_incore irec
;
2741 struct xfs_ialloc_count_inodes
*ci
= priv
;
2743 xfs_inobt_btrec_to_irec(cur
->bc_mp
, rec
, &irec
);
2744 ci
->count
+= irec
.ir_count
;
2745 ci
->freecount
+= irec
.ir_freecount
;
2750 /* Count allocated and free inodes under an inobt. */
2752 xfs_ialloc_count_inodes(
2753 struct xfs_btree_cur
*cur
,
2755 xfs_agino_t
*freecount
)
2757 struct xfs_ialloc_count_inodes ci
= {0};
2760 ASSERT(cur
->bc_btnum
== XFS_BTNUM_INO
);
2761 error
= xfs_btree_query_all(cur
, xfs_ialloc_count_inodes_rec
, &ci
);
2766 *freecount
= ci
.freecount
;