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_freecount
= 0;
69 cur
->bc_rec
.i
.ir_free
= 0;
70 return xfs_btree_lookup(cur
, dir
, stat
);
74 * Update the record referred to by cur to the value given.
75 * This either works (return 0) or gets an EFSCORRUPTED error.
77 STATIC
int /* error */
79 struct xfs_btree_cur
*cur
, /* btree cursor */
80 xfs_inobt_rec_incore_t
*irec
) /* btree record */
82 union xfs_btree_rec rec
;
84 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
85 rec
.inobt
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
86 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
87 return xfs_btree_update(cur
, &rec
);
91 * Get the data from the pointed-to record.
95 struct xfs_btree_cur
*cur
, /* btree cursor */
96 xfs_inobt_rec_incore_t
*irec
, /* btree record */
97 int *stat
) /* output: success/failure */
99 union xfs_btree_rec
*rec
;
102 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
103 if (!error
&& *stat
== 1) {
104 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
105 irec
->ir_freecount
= be32_to_cpu(rec
->inobt
.ir_freecount
);
106 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
112 * Insert a single inobt record. Cursor must already point to desired location.
115 xfs_inobt_insert_rec(
116 struct xfs_btree_cur
*cur
,
121 cur
->bc_rec
.i
.ir_freecount
= freecount
;
122 cur
->bc_rec
.i
.ir_free
= free
;
123 return xfs_btree_insert(cur
, stat
);
127 * Insert records describing a newly allocated inode chunk into the inobt.
131 struct xfs_mount
*mp
,
132 struct xfs_trans
*tp
,
133 struct xfs_buf
*agbp
,
138 struct xfs_btree_cur
*cur
;
139 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
140 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
145 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, btnum
);
147 for (thisino
= newino
;
148 thisino
< newino
+ newlen
;
149 thisino
+= XFS_INODES_PER_CHUNK
) {
150 error
= xfs_inobt_lookup(cur
, thisino
, XFS_LOOKUP_EQ
, &i
);
152 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
157 error
= xfs_inobt_insert_rec(cur
, XFS_INODES_PER_CHUNK
,
158 XFS_INOBT_ALL_FREE
, &i
);
160 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
166 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
172 * Verify that the number of free inodes in the AGI is correct.
176 xfs_check_agi_freecount(
177 struct xfs_btree_cur
*cur
,
180 if (cur
->bc_nlevels
== 1) {
181 xfs_inobt_rec_incore_t rec
;
186 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
191 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
196 freecount
+= rec
.ir_freecount
;
197 error
= xfs_btree_increment(cur
, 0, &i
);
203 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
204 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
209 #define xfs_check_agi_freecount(cur, agi) 0
213 * Initialise a new set of inodes. When called without a transaction context
214 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
215 * than logging them (which in a transaction context puts them into the AIL
216 * for writeback rather than the xfsbufd queue).
219 xfs_ialloc_inode_init(
220 struct xfs_mount
*mp
,
221 struct xfs_trans
*tp
,
222 struct list_head
*buffer_list
,
225 xfs_agblock_t length
,
228 struct xfs_buf
*fbuf
;
229 struct xfs_dinode
*free
;
230 int nbufs
, blks_per_cluster
, inodes_per_cluster
;
237 * Loop over the new block(s), filling in the inodes. For small block
238 * sizes, manipulate the inodes in buffers which are multiples of the
241 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
242 inodes_per_cluster
= blks_per_cluster
<< mp
->m_sb
.sb_inopblog
;
243 nbufs
= length
/ blks_per_cluster
;
246 * Figure out what version number to use in the inodes we create. If
247 * the superblock version has caught up to the one that supports the new
248 * inode format, then use the new inode version. Otherwise use the old
249 * version so that old kernels will continue to be able to use the file
252 * For v3 inodes, we also need to write the inode number into the inode,
253 * so calculate the first inode number of the chunk here as
254 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
255 * across multiple filesystem blocks (such as a cluster) and so cannot
256 * be used in the cluster buffer loop below.
258 * Further, because we are writing the inode directly into the buffer
259 * and calculating a CRC on the entire inode, we have ot log the entire
260 * inode so that the entire range the CRC covers is present in the log.
261 * That means for v3 inode we log the entire buffer rather than just the
264 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
266 ino
= XFS_AGINO_TO_INO(mp
, agno
,
267 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
270 * log the initialisation that is about to take place as an
271 * logical operation. This means the transaction does not
272 * need to log the physical changes to the inode buffers as log
273 * recovery will know what initialisation is actually needed.
274 * Hence we only need to log the buffers as "ordered" buffers so
275 * they track in the AIL as if they were physically logged.
278 xfs_icreate_log(tp
, agno
, agbno
, mp
->m_ialloc_inos
,
279 mp
->m_sb
.sb_inodesize
, length
, gen
);
283 for (j
= 0; j
< nbufs
; j
++) {
287 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
288 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
289 mp
->m_bsize
* blks_per_cluster
,
294 /* Initialize the inode buffers and log them appropriately. */
295 fbuf
->b_ops
= &xfs_inode_buf_ops
;
296 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
297 for (i
= 0; i
< inodes_per_cluster
; i
++) {
298 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
299 uint isize
= xfs_dinode_size(version
);
301 free
= xfs_make_iptr(mp
, fbuf
, i
);
302 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
303 free
->di_version
= version
;
304 free
->di_gen
= cpu_to_be32(gen
);
305 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
308 free
->di_ino
= cpu_to_be64(ino
);
310 uuid_copy(&free
->di_uuid
, &mp
->m_sb
.sb_uuid
);
311 xfs_dinode_calc_crc(mp
, free
);
313 /* just log the inode core */
314 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
315 ioffset
+ isize
- 1);
321 * Mark the buffer as an inode allocation buffer so it
322 * sticks in AIL at the point of this allocation
323 * transaction. This ensures the they are on disk before
324 * the tail of the log can be moved past this
325 * transaction (i.e. by preventing relogging from moving
326 * it forward in the log).
328 xfs_trans_inode_alloc_buf(tp
, fbuf
);
331 * Mark the buffer as ordered so that they are
332 * not physically logged in the transaction but
333 * still tracked in the AIL as part of the
334 * transaction and pin the log appropriately.
336 xfs_trans_ordered_buf(tp
, fbuf
);
337 xfs_trans_log_buf(tp
, fbuf
, 0,
338 BBTOB(fbuf
->b_length
) - 1);
341 fbuf
->b_flags
|= XBF_DONE
;
342 xfs_buf_delwri_queue(fbuf
, buffer_list
);
350 * Allocate new inodes in the allocation group specified by agbp.
351 * Return 0 for success, else error code.
353 STATIC
int /* error code or 0 */
355 xfs_trans_t
*tp
, /* transaction pointer */
356 xfs_buf_t
*agbp
, /* alloc group buffer */
359 xfs_agi_t
*agi
; /* allocation group header */
360 xfs_alloc_arg_t args
; /* allocation argument structure */
363 xfs_agino_t newino
; /* new first inode's number */
364 xfs_agino_t newlen
; /* new number of inodes */
365 int isaligned
= 0; /* inode allocation at stripe unit */
367 struct xfs_perag
*pag
;
369 memset(&args
, 0, sizeof(args
));
371 args
.mp
= tp
->t_mountp
;
374 * Locking will ensure that we don't have two callers in here
377 newlen
= args
.mp
->m_ialloc_inos
;
378 if (args
.mp
->m_maxicount
&&
379 args
.mp
->m_sb
.sb_icount
+ newlen
> args
.mp
->m_maxicount
)
381 args
.minlen
= args
.maxlen
= args
.mp
->m_ialloc_blks
;
383 * First try to allocate inodes contiguous with the last-allocated
384 * chunk of inodes. If the filesystem is striped, this will fill
385 * an entire stripe unit with inodes.
387 agi
= XFS_BUF_TO_AGI(agbp
);
388 newino
= be32_to_cpu(agi
->agi_newino
);
389 agno
= be32_to_cpu(agi
->agi_seqno
);
390 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
391 args
.mp
->m_ialloc_blks
;
392 if (likely(newino
!= NULLAGINO
&&
393 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
394 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
395 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
399 * We need to take into account alignment here to ensure that
400 * we don't modify the free list if we fail to have an exact
401 * block. If we don't have an exact match, and every oher
402 * attempt allocation attempt fails, we'll end up cancelling
403 * a dirty transaction and shutting down.
405 * For an exact allocation, alignment must be 1,
406 * however we need to take cluster alignment into account when
407 * fixing up the freelist. Use the minalignslop field to
408 * indicate that extra blocks might be required for alignment,
409 * but not to use them in the actual exact allocation.
412 args
.minalignslop
= xfs_ialloc_cluster_alignment(args
.mp
) - 1;
414 /* Allow space for the inode btree to split. */
415 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
416 if ((error
= xfs_alloc_vextent(&args
)))
420 * This request might have dirtied the transaction if the AG can
421 * satisfy the request, but the exact block was not available.
422 * If the allocation did fail, subsequent requests will relax
423 * the exact agbno requirement and increase the alignment
424 * instead. It is critical that the total size of the request
425 * (len + alignment + slop) does not increase from this point
426 * on, so reset minalignslop to ensure it is not included in
427 * subsequent requests.
429 args
.minalignslop
= 0;
431 args
.fsbno
= NULLFSBLOCK
;
433 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
435 * Set the alignment for the allocation.
436 * If stripe alignment is turned on then align at stripe unit
438 * If the cluster size is smaller than a filesystem block
439 * then we're doing I/O for inodes in filesystem block size
440 * pieces, so don't need alignment anyway.
443 if (args
.mp
->m_sinoalign
) {
444 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
445 args
.alignment
= args
.mp
->m_dalign
;
448 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
450 * Need to figure out where to allocate the inode blocks.
451 * Ideally they should be spaced out through the a.g.
452 * For now, just allocate blocks up front.
454 args
.agbno
= be32_to_cpu(agi
->agi_root
);
455 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
457 * Allocate a fixed-size extent of inodes.
459 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
462 * Allow space for the inode btree to split.
464 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
465 if ((error
= xfs_alloc_vextent(&args
)))
470 * If stripe alignment is turned on, then try again with cluster
473 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
474 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
475 args
.agbno
= be32_to_cpu(agi
->agi_root
);
476 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
477 args
.alignment
= xfs_ialloc_cluster_alignment(args
.mp
);
478 if ((error
= xfs_alloc_vextent(&args
)))
482 if (args
.fsbno
== NULLFSBLOCK
) {
486 ASSERT(args
.len
== args
.minlen
);
489 * Stamp and write the inode buffers.
491 * Seed the new inode cluster with a random generation number. This
492 * prevents short-term reuse of generation numbers if a chunk is
493 * freed and then immediately reallocated. We use random numbers
494 * rather than a linear progression to prevent the next generation
495 * number from being easily guessable.
497 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, agno
, args
.agbno
,
498 args
.len
, prandom_u32());
503 * Convert the results.
505 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
506 be32_add_cpu(&agi
->agi_count
, newlen
);
507 be32_add_cpu(&agi
->agi_freecount
, newlen
);
508 pag
= xfs_perag_get(args
.mp
, agno
);
509 pag
->pagi_freecount
+= newlen
;
511 agi
->agi_newino
= cpu_to_be32(newino
);
514 * Insert records describing the new inode chunk into the btrees.
516 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
521 if (xfs_sb_version_hasfinobt(&args
.mp
->m_sb
)) {
522 error
= xfs_inobt_insert(args
.mp
, tp
, agbp
, newino
, newlen
,
528 * Log allocation group header fields
530 xfs_ialloc_log_agi(tp
, agbp
,
531 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
533 * Modify/log superblock values for inode count and inode free count.
535 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
536 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
541 STATIC xfs_agnumber_t
547 spin_lock(&mp
->m_agirotor_lock
);
548 agno
= mp
->m_agirotor
;
549 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
551 spin_unlock(&mp
->m_agirotor_lock
);
557 * Select an allocation group to look for a free inode in, based on the parent
558 * inode and the mode. Return the allocation group buffer.
560 STATIC xfs_agnumber_t
561 xfs_ialloc_ag_select(
562 xfs_trans_t
*tp
, /* transaction pointer */
563 xfs_ino_t parent
, /* parent directory inode number */
564 umode_t mode
, /* bits set to indicate file type */
565 int okalloc
) /* ok to allocate more space */
567 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
568 xfs_agnumber_t agno
; /* current ag number */
569 int flags
; /* alloc buffer locking flags */
570 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
571 xfs_extlen_t longest
= 0; /* longest extent available */
572 xfs_mount_t
*mp
; /* mount point structure */
573 int needspace
; /* file mode implies space allocated */
574 xfs_perag_t
*pag
; /* per allocation group data */
575 xfs_agnumber_t pagno
; /* parent (starting) ag number */
579 * Files of these types need at least one block if length > 0
580 * (and they won't fit in the inode, but that's hard to figure out).
582 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
584 agcount
= mp
->m_maxagi
;
586 pagno
= xfs_ialloc_next_ag(mp
);
588 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
589 if (pagno
>= agcount
)
593 ASSERT(pagno
< agcount
);
596 * Loop through allocation groups, looking for one with a little
597 * free space in it. Note we don't look for free inodes, exactly.
598 * Instead, we include whether there is a need to allocate inodes
599 * to mean that blocks must be allocated for them,
600 * if none are currently free.
603 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
605 pag
= xfs_perag_get(mp
, agno
);
606 if (!pag
->pagi_inodeok
) {
607 xfs_ialloc_next_ag(mp
);
611 if (!pag
->pagi_init
) {
612 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
617 if (pag
->pagi_freecount
) {
625 if (!pag
->pagf_init
) {
626 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
632 * Check that there is enough free space for the file plus a
633 * chunk of inodes if we need to allocate some. If this is the
634 * first pass across the AGs, take into account the potential
635 * space needed for alignment of inode chunks when checking the
636 * longest contiguous free space in the AG - this prevents us
637 * from getting ENOSPC because we have free space larger than
638 * m_ialloc_blks but alignment constraints prevent us from using
641 * If we can't find an AG with space for full alignment slack to
642 * be taken into account, we must be near ENOSPC in all AGs.
643 * Hence we don't include alignment for the second pass and so
644 * if we fail allocation due to alignment issues then it is most
645 * likely a real ENOSPC condition.
647 ineed
= mp
->m_ialloc_blks
;
648 if (flags
&& ineed
> 1)
649 ineed
+= xfs_ialloc_cluster_alignment(mp
);
650 longest
= pag
->pagf_longest
;
652 longest
= pag
->pagf_flcount
> 0;
654 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
662 * No point in iterating over the rest, if we're shutting
665 if (XFS_FORCED_SHUTDOWN(mp
))
679 * Try to retrieve the next record to the left/right from the current one.
683 struct xfs_btree_cur
*cur
,
684 xfs_inobt_rec_incore_t
*rec
,
692 error
= xfs_btree_decrement(cur
, 0, &i
);
694 error
= xfs_btree_increment(cur
, 0, &i
);
700 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
703 XFS_WANT_CORRUPTED_RETURN(i
== 1);
711 struct xfs_btree_cur
*cur
,
713 xfs_inobt_rec_incore_t
*rec
,
719 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
724 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
727 XFS_WANT_CORRUPTED_RETURN(i
== 1);
734 * Allocate an inode using the inobt-only algorithm.
737 xfs_dialloc_ag_inobt(
738 struct xfs_trans
*tp
,
739 struct xfs_buf
*agbp
,
743 struct xfs_mount
*mp
= tp
->t_mountp
;
744 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
745 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
746 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
747 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
748 struct xfs_perag
*pag
;
749 struct xfs_btree_cur
*cur
, *tcur
;
750 struct xfs_inobt_rec_incore rec
, trec
;
756 pag
= xfs_perag_get(mp
, agno
);
758 ASSERT(pag
->pagi_init
);
759 ASSERT(pag
->pagi_inodeok
);
760 ASSERT(pag
->pagi_freecount
> 0);
763 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
765 * If pagino is 0 (this is the root inode allocation) use newino.
766 * This must work because we've just allocated some.
769 pagino
= be32_to_cpu(agi
->agi_newino
);
771 error
= xfs_check_agi_freecount(cur
, agi
);
776 * If in the same AG as the parent, try to get near the parent.
779 int doneleft
; /* done, to the left */
780 int doneright
; /* done, to the right */
781 int searchdistance
= 10;
783 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
786 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
788 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
791 XFS_WANT_CORRUPTED_GOTO(j
== 1, error0
);
793 if (rec
.ir_freecount
> 0) {
795 * Found a free inode in the same chunk
796 * as the parent, done.
803 * In the same AG as parent, but parent's chunk is full.
806 /* duplicate the cursor, search left & right simultaneously */
807 error
= xfs_btree_dup_cursor(cur
, &tcur
);
812 * Skip to last blocks looked up if same parent inode.
814 if (pagino
!= NULLAGINO
&&
815 pag
->pagl_pagino
== pagino
&&
816 pag
->pagl_leftrec
!= NULLAGINO
&&
817 pag
->pagl_rightrec
!= NULLAGINO
) {
818 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
823 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
828 /* search left with tcur, back up 1 record */
829 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
833 /* search right with cur, go forward 1 record. */
834 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
840 * Loop until we find an inode chunk with a free inode.
842 while (!doneleft
|| !doneright
) {
843 int useleft
; /* using left inode chunk this time */
845 if (!--searchdistance
) {
847 * Not in range - save last search
848 * location and allocate a new inode
850 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
851 pag
->pagl_leftrec
= trec
.ir_startino
;
852 pag
->pagl_rightrec
= rec
.ir_startino
;
853 pag
->pagl_pagino
= pagino
;
857 /* figure out the closer block if both are valid. */
858 if (!doneleft
&& !doneright
) {
860 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
861 rec
.ir_startino
- pagino
;
866 /* free inodes to the left? */
867 if (useleft
&& trec
.ir_freecount
) {
869 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
872 pag
->pagl_leftrec
= trec
.ir_startino
;
873 pag
->pagl_rightrec
= rec
.ir_startino
;
874 pag
->pagl_pagino
= pagino
;
878 /* free inodes to the right? */
879 if (!useleft
&& rec
.ir_freecount
) {
880 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
882 pag
->pagl_leftrec
= trec
.ir_startino
;
883 pag
->pagl_rightrec
= rec
.ir_startino
;
884 pag
->pagl_pagino
= pagino
;
888 /* get next record to check */
890 error
= xfs_ialloc_next_rec(tcur
, &trec
,
893 error
= xfs_ialloc_next_rec(cur
, &rec
,
901 * We've reached the end of the btree. because
902 * we are only searching a small chunk of the
903 * btree each search, there is obviously free
904 * inodes closer to the parent inode than we
905 * are now. restart the search again.
907 pag
->pagl_pagino
= NULLAGINO
;
908 pag
->pagl_leftrec
= NULLAGINO
;
909 pag
->pagl_rightrec
= NULLAGINO
;
910 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
911 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
916 * In a different AG from the parent.
917 * See if the most recently allocated block has any free.
920 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
921 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
927 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
931 if (j
== 1 && rec
.ir_freecount
> 0) {
933 * The last chunk allocated in the group
934 * still has a free inode.
942 * None left in the last group, search the whole AG
944 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
947 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
950 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
953 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
954 if (rec
.ir_freecount
> 0)
956 error
= xfs_btree_increment(cur
, 0, &i
);
959 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
963 offset
= xfs_lowbit64(rec
.ir_free
);
965 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
966 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
967 XFS_INODES_PER_CHUNK
) == 0);
968 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
969 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
971 error
= xfs_inobt_update(cur
, &rec
);
974 be32_add_cpu(&agi
->agi_freecount
, -1);
975 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
976 pag
->pagi_freecount
--;
978 error
= xfs_check_agi_freecount(cur
, agi
);
982 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
983 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
988 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
990 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
996 * Use the free inode btree to allocate an inode based on distance from the
997 * parent. Note that the provided cursor may be deleted and replaced.
1000 xfs_dialloc_ag_finobt_near(
1002 struct xfs_btree_cur
**ocur
,
1003 struct xfs_inobt_rec_incore
*rec
)
1005 struct xfs_btree_cur
*lcur
= *ocur
; /* left search cursor */
1006 struct xfs_btree_cur
*rcur
; /* right search cursor */
1007 struct xfs_inobt_rec_incore rrec
;
1011 error
= xfs_inobt_lookup(lcur
, pagino
, XFS_LOOKUP_LE
, &i
);
1016 error
= xfs_inobt_get_rec(lcur
, rec
, &i
);
1019 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1022 * See if we've landed in the parent inode record. The finobt
1023 * only tracks chunks with at least one free inode, so record
1024 * existence is enough.
1026 if (pagino
>= rec
->ir_startino
&&
1027 pagino
< (rec
->ir_startino
+ XFS_INODES_PER_CHUNK
))
1031 error
= xfs_btree_dup_cursor(lcur
, &rcur
);
1035 error
= xfs_inobt_lookup(rcur
, pagino
, XFS_LOOKUP_GE
, &j
);
1039 error
= xfs_inobt_get_rec(rcur
, &rrec
, &j
);
1042 XFS_WANT_CORRUPTED_GOTO(j
== 1, error_rcur
);
1045 XFS_WANT_CORRUPTED_GOTO(i
== 1 || j
== 1, error_rcur
);
1046 if (i
== 1 && j
== 1) {
1048 * Both the left and right records are valid. Choose the closer
1049 * inode chunk to the target.
1051 if ((pagino
- rec
->ir_startino
+ XFS_INODES_PER_CHUNK
- 1) >
1052 (rrec
.ir_startino
- pagino
)) {
1054 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1057 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1059 } else if (j
== 1) {
1060 /* only the right record is valid */
1062 xfs_btree_del_cursor(lcur
, XFS_BTREE_NOERROR
);
1064 } else if (i
== 1) {
1065 /* only the left record is valid */
1066 xfs_btree_del_cursor(rcur
, XFS_BTREE_NOERROR
);
1072 xfs_btree_del_cursor(rcur
, XFS_BTREE_ERROR
);
1077 * Use the free inode btree to find a free inode based on a newino hint. If
1078 * the hint is NULL, find the first free inode in the AG.
1081 xfs_dialloc_ag_finobt_newino(
1082 struct xfs_agi
*agi
,
1083 struct xfs_btree_cur
*cur
,
1084 struct xfs_inobt_rec_incore
*rec
)
1089 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
1090 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
1095 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1098 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1104 * Find the first inode available in the AG.
1106 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
1109 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1111 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
1114 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1120 * Update the inobt based on a modification made to the finobt. Also ensure that
1121 * the records from both trees are equivalent post-modification.
1124 xfs_dialloc_ag_update_inobt(
1125 struct xfs_btree_cur
*cur
, /* inobt cursor */
1126 struct xfs_inobt_rec_incore
*frec
, /* finobt record */
1127 int offset
) /* inode offset */
1129 struct xfs_inobt_rec_incore rec
;
1133 error
= xfs_inobt_lookup(cur
, frec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1136 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1138 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1141 XFS_WANT_CORRUPTED_RETURN(i
== 1);
1142 ASSERT((XFS_AGINO_TO_OFFSET(cur
->bc_mp
, rec
.ir_startino
) %
1143 XFS_INODES_PER_CHUNK
) == 0);
1145 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1148 XFS_WANT_CORRUPTED_RETURN((rec
.ir_free
== frec
->ir_free
) &&
1149 (rec
.ir_freecount
== frec
->ir_freecount
));
1151 return xfs_inobt_update(cur
, &rec
);
1155 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1156 * back to the inobt search algorithm.
1158 * The caller selected an AG for us, and made sure that free inodes are
1163 struct xfs_trans
*tp
,
1164 struct xfs_buf
*agbp
,
1168 struct xfs_mount
*mp
= tp
->t_mountp
;
1169 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1170 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1171 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
1172 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
1173 struct xfs_perag
*pag
;
1174 struct xfs_btree_cur
*cur
; /* finobt cursor */
1175 struct xfs_btree_cur
*icur
; /* inobt cursor */
1176 struct xfs_inobt_rec_incore rec
;
1182 if (!xfs_sb_version_hasfinobt(&mp
->m_sb
))
1183 return xfs_dialloc_ag_inobt(tp
, agbp
, parent
, inop
);
1185 pag
= xfs_perag_get(mp
, agno
);
1188 * If pagino is 0 (this is the root inode allocation) use newino.
1189 * This must work because we've just allocated some.
1192 pagino
= be32_to_cpu(agi
->agi_newino
);
1194 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1196 error
= xfs_check_agi_freecount(cur
, agi
);
1201 * The search algorithm depends on whether we're in the same AG as the
1202 * parent. If so, find the closest available inode to the parent. If
1203 * not, consider the agi hint or find the first free inode in the AG.
1206 error
= xfs_dialloc_ag_finobt_near(pagino
, &cur
, &rec
);
1208 error
= xfs_dialloc_ag_finobt_newino(agi
, cur
, &rec
);
1212 offset
= xfs_lowbit64(rec
.ir_free
);
1213 ASSERT(offset
>= 0);
1214 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
1215 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
1216 XFS_INODES_PER_CHUNK
) == 0);
1217 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
1220 * Modify or remove the finobt record.
1222 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
1224 if (rec
.ir_freecount
)
1225 error
= xfs_inobt_update(cur
, &rec
);
1227 error
= xfs_btree_delete(cur
, &i
);
1232 * The finobt has now been updated appropriately. We haven't updated the
1233 * agi and superblock yet, so we can create an inobt cursor and validate
1234 * the original freecount. If all is well, make the equivalent update to
1235 * the inobt using the finobt record and offset information.
1237 icur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1239 error
= xfs_check_agi_freecount(icur
, agi
);
1243 error
= xfs_dialloc_ag_update_inobt(icur
, &rec
, offset
);
1248 * Both trees have now been updated. We must update the perag and
1249 * superblock before we can check the freecount for each btree.
1251 be32_add_cpu(&agi
->agi_freecount
, -1);
1252 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1253 pag
->pagi_freecount
--;
1255 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
1257 error
= xfs_check_agi_freecount(icur
, agi
);
1260 error
= xfs_check_agi_freecount(cur
, agi
);
1264 xfs_btree_del_cursor(icur
, XFS_BTREE_NOERROR
);
1265 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1271 xfs_btree_del_cursor(icur
, XFS_BTREE_ERROR
);
1273 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1279 * Allocate an inode on disk.
1281 * Mode is used to tell whether the new inode will need space, and whether it
1284 * This function is designed to be called twice if it has to do an allocation
1285 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1286 * If an inode is available without having to performn an allocation, an inode
1287 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1288 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1289 * The caller should then commit the current transaction, allocate a
1290 * new transaction, and call xfs_dialloc() again, passing in the previous value
1291 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1292 * buffer is locked across the two calls, the second call is guaranteed to have
1293 * a free inode available.
1295 * Once we successfully pick an inode its number is returned and the on-disk
1296 * data structures are updated. The inode itself is not read in, since doing so
1297 * would break ordering constraints with xfs_reclaim.
1301 struct xfs_trans
*tp
,
1305 struct xfs_buf
**IO_agbp
,
1308 struct xfs_mount
*mp
= tp
->t_mountp
;
1309 struct xfs_buf
*agbp
;
1310 xfs_agnumber_t agno
;
1314 xfs_agnumber_t start_agno
;
1315 struct xfs_perag
*pag
;
1319 * If the caller passes in a pointer to the AGI buffer,
1320 * continue where we left off before. In this case, we
1321 * know that the allocation group has free inodes.
1328 * We do not have an agbp, so select an initial allocation
1329 * group for inode allocation.
1331 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
, okalloc
);
1332 if (start_agno
== NULLAGNUMBER
) {
1338 * If we have already hit the ceiling of inode blocks then clear
1339 * okalloc so we scan all available agi structures for a free
1342 if (mp
->m_maxicount
&&
1343 mp
->m_sb
.sb_icount
+ mp
->m_ialloc_inos
> mp
->m_maxicount
) {
1349 * Loop until we find an allocation group that either has free inodes
1350 * or in which we can allocate some inodes. Iterate through the
1351 * allocation groups upward, wrapping at the end.
1355 pag
= xfs_perag_get(mp
, agno
);
1356 if (!pag
->pagi_inodeok
) {
1357 xfs_ialloc_next_ag(mp
);
1361 if (!pag
->pagi_init
) {
1362 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1368 * Do a first racy fast path check if this AG is usable.
1370 if (!pag
->pagi_freecount
&& !okalloc
)
1374 * Then read in the AGI buffer and recheck with the AGI buffer
1377 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1381 if (pag
->pagi_freecount
) {
1387 goto nextag_relse_buffer
;
1390 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1392 xfs_trans_brelse(tp
, agbp
);
1394 if (error
!= -ENOSPC
)
1404 * We successfully allocated some inodes, return
1405 * the current context to the caller so that it
1406 * can commit the current transaction and call
1407 * us again where we left off.
1409 ASSERT(pag
->pagi_freecount
> 0);
1417 nextag_relse_buffer
:
1418 xfs_trans_brelse(tp
, agbp
);
1421 if (++agno
== mp
->m_sb
.sb_agcount
)
1423 if (agno
== start_agno
) {
1425 return noroom
? -ENOSPC
: 0;
1431 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1439 struct xfs_mount
*mp
,
1440 struct xfs_trans
*tp
,
1441 struct xfs_buf
*agbp
,
1443 struct xfs_bmap_free
*flist
,
1445 xfs_ino_t
*first_ino
,
1446 struct xfs_inobt_rec_incore
*orec
)
1448 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1449 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1450 struct xfs_perag
*pag
;
1451 struct xfs_btree_cur
*cur
;
1452 struct xfs_inobt_rec_incore rec
;
1458 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1459 ASSERT(XFS_AGINO_TO_AGBNO(mp
, agino
) < be32_to_cpu(agi
->agi_length
));
1462 * Initialize the cursor.
1464 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1466 error
= xfs_check_agi_freecount(cur
, agi
);
1471 * Look for the entry describing this inode.
1473 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1474 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1478 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1479 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1481 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1485 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1487 * Get the offset in the inode chunk.
1489 off
= agino
- rec
.ir_startino
;
1490 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1491 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1493 * Mark the inode free & increment the count.
1495 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1499 * When an inode cluster is free, it becomes eligible for removal
1501 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1502 (rec
.ir_freecount
== mp
->m_ialloc_inos
)) {
1505 *first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1508 * Remove the inode cluster from the AGI B+Tree, adjust the
1509 * AGI and Superblock inode counts, and mark the disk space
1510 * to be freed when the transaction is committed.
1512 ilen
= mp
->m_ialloc_inos
;
1513 be32_add_cpu(&agi
->agi_count
, -ilen
);
1514 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1515 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1516 pag
= xfs_perag_get(mp
, agno
);
1517 pag
->pagi_freecount
-= ilen
- 1;
1519 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1520 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1522 if ((error
= xfs_btree_delete(cur
, &i
))) {
1523 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1528 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp
, agno
,
1529 XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
)),
1530 mp
->m_ialloc_blks
, flist
, mp
);
1534 error
= xfs_inobt_update(cur
, &rec
);
1536 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
1542 * Change the inode free counts and log the ag/sb changes.
1544 be32_add_cpu(&agi
->agi_freecount
, 1);
1545 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1546 pag
= xfs_perag_get(mp
, agno
);
1547 pag
->pagi_freecount
++;
1549 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
1552 error
= xfs_check_agi_freecount(cur
, agi
);
1557 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1561 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1566 * Free an inode in the free inode btree.
1570 struct xfs_mount
*mp
,
1571 struct xfs_trans
*tp
,
1572 struct xfs_buf
*agbp
,
1574 struct xfs_inobt_rec_incore
*ibtrec
) /* inobt record */
1576 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
1577 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
1578 struct xfs_btree_cur
*cur
;
1579 struct xfs_inobt_rec_incore rec
;
1580 int offset
= agino
- ibtrec
->ir_startino
;
1584 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_FINO
);
1586 error
= xfs_inobt_lookup(cur
, ibtrec
->ir_startino
, XFS_LOOKUP_EQ
, &i
);
1591 * If the record does not exist in the finobt, we must have just
1592 * freed an inode in a previously fully allocated chunk. If not,
1593 * something is out of sync.
1595 XFS_WANT_CORRUPTED_GOTO(ibtrec
->ir_freecount
== 1, error
);
1597 error
= xfs_inobt_insert_rec(cur
, ibtrec
->ir_freecount
,
1598 ibtrec
->ir_free
, &i
);
1607 * Read and update the existing record. We could just copy the ibtrec
1608 * across here, but that would defeat the purpose of having redundant
1609 * metadata. By making the modifications independently, we can catch
1610 * corruptions that we wouldn't see if we just copied from one record
1613 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1616 XFS_WANT_CORRUPTED_GOTO(i
== 1, error
);
1618 rec
.ir_free
|= XFS_INOBT_MASK(offset
);
1621 XFS_WANT_CORRUPTED_GOTO((rec
.ir_free
== ibtrec
->ir_free
) &&
1622 (rec
.ir_freecount
== ibtrec
->ir_freecount
),
1626 * The content of inobt records should always match between the inobt
1627 * and finobt. The lifecycle of records in the finobt is different from
1628 * the inobt in that the finobt only tracks records with at least one
1629 * free inode. Hence, if all of the inodes are free and we aren't
1630 * keeping inode chunks permanently on disk, remove the record.
1631 * Otherwise, update the record with the new information.
1633 if (rec
.ir_freecount
== mp
->m_ialloc_inos
&&
1634 !(mp
->m_flags
& XFS_MOUNT_IKEEP
)) {
1635 error
= xfs_btree_delete(cur
, &i
);
1640 error
= xfs_inobt_update(cur
, &rec
);
1646 error
= xfs_check_agi_freecount(cur
, agi
);
1650 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1654 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1659 * Free disk inode. Carefully avoids touching the incore inode, all
1660 * manipulations incore are the caller's responsibility.
1661 * The on-disk inode is not changed by this operation, only the
1662 * btree (free inode mask) is changed.
1666 struct xfs_trans
*tp
, /* transaction pointer */
1667 xfs_ino_t inode
, /* inode to be freed */
1668 struct xfs_bmap_free
*flist
, /* extents to free */
1669 int *deleted
,/* set if inode cluster was deleted */
1670 xfs_ino_t
*first_ino
)/* first inode in deleted cluster */
1673 xfs_agblock_t agbno
; /* block number containing inode */
1674 struct xfs_buf
*agbp
; /* buffer for allocation group header */
1675 xfs_agino_t agino
; /* allocation group inode number */
1676 xfs_agnumber_t agno
; /* allocation group number */
1677 int error
; /* error return value */
1678 struct xfs_mount
*mp
; /* mount structure for filesystem */
1679 struct xfs_inobt_rec_incore rec
;/* btree record */
1684 * Break up inode number into its components.
1686 agno
= XFS_INO_TO_AGNO(mp
, inode
);
1687 if (agno
>= mp
->m_sb
.sb_agcount
) {
1688 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1689 __func__
, agno
, mp
->m_sb
.sb_agcount
);
1693 agino
= XFS_INO_TO_AGINO(mp
, inode
);
1694 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1695 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1696 __func__
, (unsigned long long)inode
,
1697 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
1701 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
1702 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
1703 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1704 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
1709 * Get the allocation group header.
1711 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1713 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
1719 * Fix up the inode allocation btree.
1721 error
= xfs_difree_inobt(mp
, tp
, agbp
, agino
, flist
, deleted
, first_ino
,
1727 * Fix up the free inode btree.
1729 if (xfs_sb_version_hasfinobt(&mp
->m_sb
)) {
1730 error
= xfs_difree_finobt(mp
, tp
, agbp
, agino
, &rec
);
1743 struct xfs_mount
*mp
,
1744 struct xfs_trans
*tp
,
1745 xfs_agnumber_t agno
,
1747 xfs_agblock_t agbno
,
1748 xfs_agblock_t
*chunk_agbno
,
1749 xfs_agblock_t
*offset_agbno
,
1752 struct xfs_inobt_rec_incore rec
;
1753 struct xfs_btree_cur
*cur
;
1754 struct xfs_buf
*agbp
;
1758 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1761 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1762 __func__
, error
, agno
);
1767 * Lookup the inode record for the given agino. If the record cannot be
1768 * found, then it's an invalid inode number and we should abort. Once
1769 * we have a record, we need to ensure it contains the inode number
1770 * we are looking up.
1772 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
, XFS_BTNUM_INO
);
1773 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
1776 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1777 if (!error
&& i
== 0)
1781 xfs_trans_brelse(tp
, agbp
);
1782 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1786 /* check that the returned record contains the required inode */
1787 if (rec
.ir_startino
> agino
||
1788 rec
.ir_startino
+ mp
->m_ialloc_inos
<= agino
)
1791 /* for untrusted inodes check it is allocated first */
1792 if ((flags
& XFS_IGET_UNTRUSTED
) &&
1793 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
1796 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
1797 *offset_agbno
= agbno
- *chunk_agbno
;
1802 * Return the location of the inode in imap, for mapping it into a buffer.
1806 xfs_mount_t
*mp
, /* file system mount structure */
1807 xfs_trans_t
*tp
, /* transaction pointer */
1808 xfs_ino_t ino
, /* inode to locate */
1809 struct xfs_imap
*imap
, /* location map structure */
1810 uint flags
) /* flags for inode btree lookup */
1812 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
1813 xfs_agino_t agino
; /* inode number within alloc group */
1814 xfs_agnumber_t agno
; /* allocation group number */
1815 int blks_per_cluster
; /* num blocks per inode cluster */
1816 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
1817 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
1818 int error
; /* error code */
1819 int offset
; /* index of inode in its buffer */
1820 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
1822 ASSERT(ino
!= NULLFSINO
);
1825 * Split up the inode number into its parts.
1827 agno
= XFS_INO_TO_AGNO(mp
, ino
);
1828 agino
= XFS_INO_TO_AGINO(mp
, ino
);
1829 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
1830 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
1831 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1834 * Don't output diagnostic information for untrusted inodes
1835 * as they can be invalid without implying corruption.
1837 if (flags
& XFS_IGET_UNTRUSTED
)
1839 if (agno
>= mp
->m_sb
.sb_agcount
) {
1841 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1842 __func__
, agno
, mp
->m_sb
.sb_agcount
);
1844 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
1846 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1847 __func__
, (unsigned long long)agbno
,
1848 (unsigned long)mp
->m_sb
.sb_agblocks
);
1850 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1852 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1854 XFS_AGINO_TO_INO(mp
, agno
, agino
));
1861 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
1864 * For bulkstat and handle lookups, we have an untrusted inode number
1865 * that we have to verify is valid. We cannot do this just by reading
1866 * the inode buffer as it may have been unlinked and removed leaving
1867 * inodes in stale state on disk. Hence we have to do a btree lookup
1868 * in all cases where an untrusted inode number is passed.
1870 if (flags
& XFS_IGET_UNTRUSTED
) {
1871 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
1872 &chunk_agbno
, &offset_agbno
, flags
);
1879 * If the inode cluster size is the same as the blocksize or
1880 * smaller we get to the buffer by simple arithmetics.
1882 if (blks_per_cluster
== 1) {
1883 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
1884 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
1886 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
1887 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
1888 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
1893 * If the inode chunks are aligned then use simple maths to
1894 * find the location. Otherwise we have to do a btree
1895 * lookup to find the location.
1897 if (mp
->m_inoalign_mask
) {
1898 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
1899 chunk_agbno
= agbno
- offset_agbno
;
1901 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
1902 &chunk_agbno
, &offset_agbno
, flags
);
1908 ASSERT(agbno
>= chunk_agbno
);
1909 cluster_agbno
= chunk_agbno
+
1910 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
1911 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
1912 XFS_INO_TO_OFFSET(mp
, ino
);
1914 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
1915 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
1916 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
1919 * If the inode number maps to a block outside the bounds
1920 * of the file system then return NULL rather than calling
1921 * read_buf and panicing when we get an error from the
1924 if ((imap
->im_blkno
+ imap
->im_len
) >
1925 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
1927 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1928 __func__
, (unsigned long long) imap
->im_blkno
,
1929 (unsigned long long) imap
->im_len
,
1930 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
1937 * Compute and fill in value of m_in_maxlevels.
1940 xfs_ialloc_compute_maxlevels(
1941 xfs_mount_t
*mp
) /* file system mount structure */
1949 maxleafents
= (1LL << XFS_INO_AGINO_BITS(mp
)) >>
1950 XFS_INODES_PER_CHUNK_LOG
;
1951 minleafrecs
= mp
->m_alloc_mnr
[0];
1952 minnoderecs
= mp
->m_alloc_mnr
[1];
1953 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
1954 for (level
= 1; maxblocks
> 1; level
++)
1955 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
1956 mp
->m_in_maxlevels
= level
;
1960 * Log specified fields for the ag hdr (inode section). The growth of the agi
1961 * structure over time requires that we interpret the buffer as two logical
1962 * regions delineated by the end of the unlinked list. This is due to the size
1963 * of the hash table and its location in the middle of the agi.
1965 * For example, a request to log a field before agi_unlinked and a field after
1966 * agi_unlinked could cause us to log the entire hash table and use an excessive
1967 * amount of log space. To avoid this behavior, log the region up through
1968 * agi_unlinked in one call and the region after agi_unlinked through the end of
1969 * the structure in another.
1973 xfs_trans_t
*tp
, /* transaction pointer */
1974 xfs_buf_t
*bp
, /* allocation group header buffer */
1975 int fields
) /* bitmask of fields to log */
1977 int first
; /* first byte number */
1978 int last
; /* last byte number */
1979 static const short offsets
[] = { /* field starting offsets */
1980 /* keep in sync with bit definitions */
1981 offsetof(xfs_agi_t
, agi_magicnum
),
1982 offsetof(xfs_agi_t
, agi_versionnum
),
1983 offsetof(xfs_agi_t
, agi_seqno
),
1984 offsetof(xfs_agi_t
, agi_length
),
1985 offsetof(xfs_agi_t
, agi_count
),
1986 offsetof(xfs_agi_t
, agi_root
),
1987 offsetof(xfs_agi_t
, agi_level
),
1988 offsetof(xfs_agi_t
, agi_freecount
),
1989 offsetof(xfs_agi_t
, agi_newino
),
1990 offsetof(xfs_agi_t
, agi_dirino
),
1991 offsetof(xfs_agi_t
, agi_unlinked
),
1992 offsetof(xfs_agi_t
, agi_free_root
),
1993 offsetof(xfs_agi_t
, agi_free_level
),
1997 xfs_agi_t
*agi
; /* allocation group header */
1999 agi
= XFS_BUF_TO_AGI(bp
);
2000 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
2003 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_AGI_BUF
);
2006 * Compute byte offsets for the first and last fields in the first
2007 * region and log the agi buffer. This only logs up through
2010 if (fields
& XFS_AGI_ALL_BITS_R1
) {
2011 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R1
,
2013 xfs_trans_log_buf(tp
, bp
, first
, last
);
2017 * Mask off the bits in the first region and calculate the first and
2018 * last field offsets for any bits in the second region.
2020 fields
&= ~XFS_AGI_ALL_BITS_R1
;
2022 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS_R2
,
2024 xfs_trans_log_buf(tp
, bp
, first
, last
);
2030 xfs_check_agi_unlinked(
2031 struct xfs_agi
*agi
)
2035 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++)
2036 ASSERT(agi
->agi_unlinked
[i
]);
2039 #define xfs_check_agi_unlinked(agi)
2046 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2047 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
2049 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2050 !uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_uuid
))
2053 * Validate the magic number of the agi block.
2055 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
2057 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
2060 if (be32_to_cpu(agi
->agi_level
) > XFS_BTREE_MAXLEVELS
)
2063 * during growfs operations, the perag is not fully initialised,
2064 * so we can't use it for any useful checking. growfs ensures we can't
2065 * use it by using uncached buffers that don't have the perag attached
2066 * so we can detect and avoid this problem.
2068 if (bp
->b_pag
&& be32_to_cpu(agi
->agi_seqno
) != bp
->b_pag
->pag_agno
)
2071 xfs_check_agi_unlinked(agi
);
2076 xfs_agi_read_verify(
2079 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2081 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
2082 !xfs_buf_verify_cksum(bp
, XFS_AGI_CRC_OFF
))
2083 xfs_buf_ioerror(bp
, -EFSBADCRC
);
2084 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp
), mp
,
2085 XFS_ERRTAG_IALLOC_READ_AGI
,
2086 XFS_RANDOM_IALLOC_READ_AGI
))
2087 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2090 xfs_verifier_error(bp
);
2094 xfs_agi_write_verify(
2097 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
2098 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
2100 if (!xfs_agi_verify(bp
)) {
2101 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
2102 xfs_verifier_error(bp
);
2106 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2110 XFS_BUF_TO_AGI(bp
)->agi_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
2111 xfs_buf_update_cksum(bp
, XFS_AGI_CRC_OFF
);
2114 const struct xfs_buf_ops xfs_agi_buf_ops
= {
2115 .verify_read
= xfs_agi_read_verify
,
2116 .verify_write
= xfs_agi_write_verify
,
2120 * Read in the allocation group header (inode allocation section)
2124 struct xfs_mount
*mp
, /* file system mount structure */
2125 struct xfs_trans
*tp
, /* transaction pointer */
2126 xfs_agnumber_t agno
, /* allocation group number */
2127 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2131 trace_xfs_read_agi(mp
, agno
);
2133 ASSERT(agno
!= NULLAGNUMBER
);
2134 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
2135 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
2136 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
2140 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
2145 xfs_ialloc_read_agi(
2146 struct xfs_mount
*mp
, /* file system mount structure */
2147 struct xfs_trans
*tp
, /* transaction pointer */
2148 xfs_agnumber_t agno
, /* allocation group number */
2149 struct xfs_buf
**bpp
) /* allocation group hdr buf */
2151 struct xfs_agi
*agi
; /* allocation group header */
2152 struct xfs_perag
*pag
; /* per allocation group data */
2155 trace_xfs_ialloc_read_agi(mp
, agno
);
2157 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
2161 agi
= XFS_BUF_TO_AGI(*bpp
);
2162 pag
= xfs_perag_get(mp
, agno
);
2163 if (!pag
->pagi_init
) {
2164 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
2165 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
2170 * It's possible for these to be out of sync if
2171 * we are in the middle of a forced shutdown.
2173 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
2174 XFS_FORCED_SHUTDOWN(mp
));
2180 * Read in the agi to initialise the per-ag data in the mount structure
2183 xfs_ialloc_pagi_init(
2184 xfs_mount_t
*mp
, /* file system mount structure */
2185 xfs_trans_t
*tp
, /* transaction pointer */
2186 xfs_agnumber_t agno
) /* allocation group number */
2188 xfs_buf_t
*bp
= NULL
;
2191 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
2195 xfs_trans_brelse(tp
, bp
);