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"
28 #include "xfs_mount.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
37 #include "xfs_cksum.h"
38 #include "xfs_trans.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_icreate_item.h"
41 #include "xfs_icache.h"
42 #include "xfs_dinode.h"
43 #include "xfs_trace.h"
47 * Allocation group level functions.
50 xfs_ialloc_cluster_alignment(
51 xfs_alloc_arg_t
*args
)
53 if (xfs_sb_version_hasalign(&args
->mp
->m_sb
) &&
54 args
->mp
->m_sb
.sb_inoalignmt
>=
55 XFS_B_TO_FSBT(args
->mp
, XFS_INODE_CLUSTER_SIZE(args
->mp
)))
56 return args
->mp
->m_sb
.sb_inoalignmt
;
61 * Lookup a record by ino in the btree given by cur.
65 struct xfs_btree_cur
*cur
, /* btree cursor */
66 xfs_agino_t ino
, /* starting inode of chunk */
67 xfs_lookup_t dir
, /* <=, >=, == */
68 int *stat
) /* success/failure */
70 cur
->bc_rec
.i
.ir_startino
= ino
;
71 cur
->bc_rec
.i
.ir_freecount
= 0;
72 cur
->bc_rec
.i
.ir_free
= 0;
73 return xfs_btree_lookup(cur
, dir
, stat
);
77 * Update the record referred to by cur to the value given.
78 * This either works (return 0) or gets an EFSCORRUPTED error.
80 STATIC
int /* error */
82 struct xfs_btree_cur
*cur
, /* btree cursor */
83 xfs_inobt_rec_incore_t
*irec
) /* btree record */
85 union xfs_btree_rec rec
;
87 rec
.inobt
.ir_startino
= cpu_to_be32(irec
->ir_startino
);
88 rec
.inobt
.ir_freecount
= cpu_to_be32(irec
->ir_freecount
);
89 rec
.inobt
.ir_free
= cpu_to_be64(irec
->ir_free
);
90 return xfs_btree_update(cur
, &rec
);
94 * Get the data from the pointed-to record.
98 struct xfs_btree_cur
*cur
, /* btree cursor */
99 xfs_inobt_rec_incore_t
*irec
, /* btree record */
100 int *stat
) /* output: success/failure */
102 union xfs_btree_rec
*rec
;
105 error
= xfs_btree_get_rec(cur
, &rec
, stat
);
106 if (!error
&& *stat
== 1) {
107 irec
->ir_startino
= be32_to_cpu(rec
->inobt
.ir_startino
);
108 irec
->ir_freecount
= be32_to_cpu(rec
->inobt
.ir_freecount
);
109 irec
->ir_free
= be64_to_cpu(rec
->inobt
.ir_free
);
115 * Verify that the number of free inodes in the AGI is correct.
119 xfs_check_agi_freecount(
120 struct xfs_btree_cur
*cur
,
123 if (cur
->bc_nlevels
== 1) {
124 xfs_inobt_rec_incore_t rec
;
129 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
134 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
139 freecount
+= rec
.ir_freecount
;
140 error
= xfs_btree_increment(cur
, 0, &i
);
146 if (!XFS_FORCED_SHUTDOWN(cur
->bc_mp
))
147 ASSERT(freecount
== be32_to_cpu(agi
->agi_freecount
));
152 #define xfs_check_agi_freecount(cur, agi) 0
156 * Initialise a new set of inodes. When called without a transaction context
157 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
158 * than logging them (which in a transaction context puts them into the AIL
159 * for writeback rather than the xfsbufd queue).
162 xfs_ialloc_inode_init(
163 struct xfs_mount
*mp
,
164 struct xfs_trans
*tp
,
165 struct list_head
*buffer_list
,
168 xfs_agblock_t length
,
171 struct xfs_buf
*fbuf
;
172 struct xfs_dinode
*free
;
173 int blks_per_cluster
, nbufs
, ninodes
;
180 * Loop over the new block(s), filling in the inodes.
181 * For small block sizes, manipulate the inodes in buffers
182 * which are multiples of the blocks size.
184 if (mp
->m_sb
.sb_blocksize
>= XFS_INODE_CLUSTER_SIZE(mp
)) {
185 blks_per_cluster
= 1;
187 ninodes
= mp
->m_sb
.sb_inopblock
;
189 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) /
190 mp
->m_sb
.sb_blocksize
;
191 nbufs
= length
/ blks_per_cluster
;
192 ninodes
= blks_per_cluster
* mp
->m_sb
.sb_inopblock
;
196 * Figure out what version number to use in the inodes we create. If
197 * the superblock version has caught up to the one that supports the new
198 * inode format, then use the new inode version. Otherwise use the old
199 * version so that old kernels will continue to be able to use the file
202 * For v3 inodes, we also need to write the inode number into the inode,
203 * so calculate the first inode number of the chunk here as
204 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
205 * across multiple filesystem blocks (such as a cluster) and so cannot
206 * be used in the cluster buffer loop below.
208 * Further, because we are writing the inode directly into the buffer
209 * and calculating a CRC on the entire inode, we have ot log the entire
210 * inode so that the entire range the CRC covers is present in the log.
211 * That means for v3 inode we log the entire buffer rather than just the
214 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
216 ino
= XFS_AGINO_TO_INO(mp
, agno
,
217 XFS_OFFBNO_TO_AGINO(mp
, agbno
, 0));
220 * log the initialisation that is about to take place as an
221 * logical operation. This means the transaction does not
222 * need to log the physical changes to the inode buffers as log
223 * recovery will know what initialisation is actually needed.
224 * Hence we only need to log the buffers as "ordered" buffers so
225 * they track in the AIL as if they were physically logged.
228 xfs_icreate_log(tp
, agno
, agbno
, XFS_IALLOC_INODES(mp
),
229 mp
->m_sb
.sb_inodesize
, length
, gen
);
230 } else if (xfs_sb_version_hasnlink(&mp
->m_sb
))
235 for (j
= 0; j
< nbufs
; j
++) {
239 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
+ (j
* blks_per_cluster
));
240 fbuf
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
241 mp
->m_bsize
* blks_per_cluster
,
246 /* Initialize the inode buffers and log them appropriately. */
247 fbuf
->b_ops
= &xfs_inode_buf_ops
;
248 xfs_buf_zero(fbuf
, 0, BBTOB(fbuf
->b_length
));
249 for (i
= 0; i
< ninodes
; i
++) {
250 int ioffset
= i
<< mp
->m_sb
.sb_inodelog
;
251 uint isize
= xfs_dinode_size(version
);
253 free
= xfs_make_iptr(mp
, fbuf
, i
);
254 free
->di_magic
= cpu_to_be16(XFS_DINODE_MAGIC
);
255 free
->di_version
= version
;
256 free
->di_gen
= cpu_to_be32(gen
);
257 free
->di_next_unlinked
= cpu_to_be32(NULLAGINO
);
260 free
->di_ino
= cpu_to_be64(ino
);
262 uuid_copy(&free
->di_uuid
, &mp
->m_sb
.sb_uuid
);
263 xfs_dinode_calc_crc(mp
, free
);
265 /* just log the inode core */
266 xfs_trans_log_buf(tp
, fbuf
, ioffset
,
267 ioffset
+ isize
- 1);
273 * Mark the buffer as an inode allocation buffer so it
274 * sticks in AIL at the point of this allocation
275 * transaction. This ensures the they are on disk before
276 * the tail of the log can be moved past this
277 * transaction (i.e. by preventing relogging from moving
278 * it forward in the log).
280 xfs_trans_inode_alloc_buf(tp
, fbuf
);
283 * Mark the buffer as ordered so that they are
284 * not physically logged in the transaction but
285 * still tracked in the AIL as part of the
286 * transaction and pin the log appropriately.
288 xfs_trans_ordered_buf(tp
, fbuf
);
289 xfs_trans_log_buf(tp
, fbuf
, 0,
290 BBTOB(fbuf
->b_length
) - 1);
293 fbuf
->b_flags
|= XBF_DONE
;
294 xfs_buf_delwri_queue(fbuf
, buffer_list
);
302 * Allocate new inodes in the allocation group specified by agbp.
303 * Return 0 for success, else error code.
305 STATIC
int /* error code or 0 */
307 xfs_trans_t
*tp
, /* transaction pointer */
308 xfs_buf_t
*agbp
, /* alloc group buffer */
311 xfs_agi_t
*agi
; /* allocation group header */
312 xfs_alloc_arg_t args
; /* allocation argument structure */
313 xfs_btree_cur_t
*cur
; /* inode btree cursor */
317 xfs_agino_t newino
; /* new first inode's number */
318 xfs_agino_t newlen
; /* new number of inodes */
319 xfs_agino_t thisino
; /* current inode number, for loop */
320 int isaligned
= 0; /* inode allocation at stripe unit */
322 struct xfs_perag
*pag
;
324 memset(&args
, 0, sizeof(args
));
326 args
.mp
= tp
->t_mountp
;
329 * Locking will ensure that we don't have two callers in here
332 newlen
= XFS_IALLOC_INODES(args
.mp
);
333 if (args
.mp
->m_maxicount
&&
334 args
.mp
->m_sb
.sb_icount
+ newlen
> args
.mp
->m_maxicount
)
335 return XFS_ERROR(ENOSPC
);
336 args
.minlen
= args
.maxlen
= XFS_IALLOC_BLOCKS(args
.mp
);
338 * First try to allocate inodes contiguous with the last-allocated
339 * chunk of inodes. If the filesystem is striped, this will fill
340 * an entire stripe unit with inodes.
342 agi
= XFS_BUF_TO_AGI(agbp
);
343 newino
= be32_to_cpu(agi
->agi_newino
);
344 agno
= be32_to_cpu(agi
->agi_seqno
);
345 args
.agbno
= XFS_AGINO_TO_AGBNO(args
.mp
, newino
) +
346 XFS_IALLOC_BLOCKS(args
.mp
);
347 if (likely(newino
!= NULLAGINO
&&
348 (args
.agbno
< be32_to_cpu(agi
->agi_length
)))) {
349 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
350 args
.type
= XFS_ALLOCTYPE_THIS_BNO
;
354 * We need to take into account alignment here to ensure that
355 * we don't modify the free list if we fail to have an exact
356 * block. If we don't have an exact match, and every oher
357 * attempt allocation attempt fails, we'll end up cancelling
358 * a dirty transaction and shutting down.
360 * For an exact allocation, alignment must be 1,
361 * however we need to take cluster alignment into account when
362 * fixing up the freelist. Use the minalignslop field to
363 * indicate that extra blocks might be required for alignment,
364 * but not to use them in the actual exact allocation.
367 args
.minalignslop
= xfs_ialloc_cluster_alignment(&args
) - 1;
369 /* Allow space for the inode btree to split. */
370 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
371 if ((error
= xfs_alloc_vextent(&args
)))
374 args
.fsbno
= NULLFSBLOCK
;
376 if (unlikely(args
.fsbno
== NULLFSBLOCK
)) {
378 * Set the alignment for the allocation.
379 * If stripe alignment is turned on then align at stripe unit
381 * If the cluster size is smaller than a filesystem block
382 * then we're doing I/O for inodes in filesystem block size
383 * pieces, so don't need alignment anyway.
386 if (args
.mp
->m_sinoalign
) {
387 ASSERT(!(args
.mp
->m_flags
& XFS_MOUNT_NOALIGN
));
388 args
.alignment
= args
.mp
->m_dalign
;
391 args
.alignment
= xfs_ialloc_cluster_alignment(&args
);
393 * Need to figure out where to allocate the inode blocks.
394 * Ideally they should be spaced out through the a.g.
395 * For now, just allocate blocks up front.
397 args
.agbno
= be32_to_cpu(agi
->agi_root
);
398 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
400 * Allocate a fixed-size extent of inodes.
402 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
405 * Allow space for the inode btree to split.
407 args
.minleft
= args
.mp
->m_in_maxlevels
- 1;
408 if ((error
= xfs_alloc_vextent(&args
)))
413 * If stripe alignment is turned on, then try again with cluster
416 if (isaligned
&& args
.fsbno
== NULLFSBLOCK
) {
417 args
.type
= XFS_ALLOCTYPE_NEAR_BNO
;
418 args
.agbno
= be32_to_cpu(agi
->agi_root
);
419 args
.fsbno
= XFS_AGB_TO_FSB(args
.mp
, agno
, args
.agbno
);
420 args
.alignment
= xfs_ialloc_cluster_alignment(&args
);
421 if ((error
= xfs_alloc_vextent(&args
)))
425 if (args
.fsbno
== NULLFSBLOCK
) {
429 ASSERT(args
.len
== args
.minlen
);
432 * Stamp and write the inode buffers.
434 * Seed the new inode cluster with a random generation number. This
435 * prevents short-term reuse of generation numbers if a chunk is
436 * freed and then immediately reallocated. We use random numbers
437 * rather than a linear progression to prevent the next generation
438 * number from being easily guessable.
440 error
= xfs_ialloc_inode_init(args
.mp
, tp
, NULL
, agno
, args
.agbno
,
441 args
.len
, prandom_u32());
446 * Convert the results.
448 newino
= XFS_OFFBNO_TO_AGINO(args
.mp
, args
.agbno
, 0);
449 be32_add_cpu(&agi
->agi_count
, newlen
);
450 be32_add_cpu(&agi
->agi_freecount
, newlen
);
451 pag
= xfs_perag_get(args
.mp
, agno
);
452 pag
->pagi_freecount
+= newlen
;
454 agi
->agi_newino
= cpu_to_be32(newino
);
457 * Insert records describing the new inode chunk into the btree.
459 cur
= xfs_inobt_init_cursor(args
.mp
, tp
, agbp
, agno
);
460 for (thisino
= newino
;
461 thisino
< newino
+ newlen
;
462 thisino
+= XFS_INODES_PER_CHUNK
) {
463 cur
->bc_rec
.i
.ir_startino
= thisino
;
464 cur
->bc_rec
.i
.ir_freecount
= XFS_INODES_PER_CHUNK
;
465 cur
->bc_rec
.i
.ir_free
= XFS_INOBT_ALL_FREE
;
466 error
= xfs_btree_lookup(cur
, XFS_LOOKUP_EQ
, &i
);
468 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
472 error
= xfs_btree_insert(cur
, &i
);
474 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
479 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
481 * Log allocation group header fields
483 xfs_ialloc_log_agi(tp
, agbp
,
484 XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
| XFS_AGI_NEWINO
);
486 * Modify/log superblock values for inode count and inode free count.
488 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, (long)newlen
);
489 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, (long)newlen
);
494 STATIC xfs_agnumber_t
500 spin_lock(&mp
->m_agirotor_lock
);
501 agno
= mp
->m_agirotor
;
502 if (++mp
->m_agirotor
>= mp
->m_maxagi
)
504 spin_unlock(&mp
->m_agirotor_lock
);
510 * Select an allocation group to look for a free inode in, based on the parent
511 * inode and the mode. Return the allocation group buffer.
513 STATIC xfs_agnumber_t
514 xfs_ialloc_ag_select(
515 xfs_trans_t
*tp
, /* transaction pointer */
516 xfs_ino_t parent
, /* parent directory inode number */
517 umode_t mode
, /* bits set to indicate file type */
518 int okalloc
) /* ok to allocate more space */
520 xfs_agnumber_t agcount
; /* number of ag's in the filesystem */
521 xfs_agnumber_t agno
; /* current ag number */
522 int flags
; /* alloc buffer locking flags */
523 xfs_extlen_t ineed
; /* blocks needed for inode allocation */
524 xfs_extlen_t longest
= 0; /* longest extent available */
525 xfs_mount_t
*mp
; /* mount point structure */
526 int needspace
; /* file mode implies space allocated */
527 xfs_perag_t
*pag
; /* per allocation group data */
528 xfs_agnumber_t pagno
; /* parent (starting) ag number */
532 * Files of these types need at least one block if length > 0
533 * (and they won't fit in the inode, but that's hard to figure out).
535 needspace
= S_ISDIR(mode
) || S_ISREG(mode
) || S_ISLNK(mode
);
537 agcount
= mp
->m_maxagi
;
539 pagno
= xfs_ialloc_next_ag(mp
);
541 pagno
= XFS_INO_TO_AGNO(mp
, parent
);
542 if (pagno
>= agcount
)
546 ASSERT(pagno
< agcount
);
549 * Loop through allocation groups, looking for one with a little
550 * free space in it. Note we don't look for free inodes, exactly.
551 * Instead, we include whether there is a need to allocate inodes
552 * to mean that blocks must be allocated for them,
553 * if none are currently free.
556 flags
= XFS_ALLOC_FLAG_TRYLOCK
;
558 pag
= xfs_perag_get(mp
, agno
);
559 if (!pag
->pagi_inodeok
) {
560 xfs_ialloc_next_ag(mp
);
564 if (!pag
->pagi_init
) {
565 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
570 if (pag
->pagi_freecount
) {
578 if (!pag
->pagf_init
) {
579 error
= xfs_alloc_pagf_init(mp
, tp
, agno
, flags
);
585 * Is there enough free space for the file plus a block of
586 * inodes? (if we need to allocate some)?
588 ineed
= XFS_IALLOC_BLOCKS(mp
);
589 longest
= pag
->pagf_longest
;
591 longest
= pag
->pagf_flcount
> 0;
593 if (pag
->pagf_freeblks
>= needspace
+ ineed
&&
601 * No point in iterating over the rest, if we're shutting
604 if (XFS_FORCED_SHUTDOWN(mp
))
618 * Try to retrieve the next record to the left/right from the current one.
622 struct xfs_btree_cur
*cur
,
623 xfs_inobt_rec_incore_t
*rec
,
631 error
= xfs_btree_decrement(cur
, 0, &i
);
633 error
= xfs_btree_increment(cur
, 0, &i
);
639 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
642 XFS_WANT_CORRUPTED_RETURN(i
== 1);
650 struct xfs_btree_cur
*cur
,
652 xfs_inobt_rec_incore_t
*rec
,
658 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_EQ
, &i
);
663 error
= xfs_inobt_get_rec(cur
, rec
, &i
);
666 XFS_WANT_CORRUPTED_RETURN(i
== 1);
675 * The caller selected an AG for us, and made sure that free inodes are
680 struct xfs_trans
*tp
,
681 struct xfs_buf
*agbp
,
685 struct xfs_mount
*mp
= tp
->t_mountp
;
686 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agbp
);
687 xfs_agnumber_t agno
= be32_to_cpu(agi
->agi_seqno
);
688 xfs_agnumber_t pagno
= XFS_INO_TO_AGNO(mp
, parent
);
689 xfs_agino_t pagino
= XFS_INO_TO_AGINO(mp
, parent
);
690 struct xfs_perag
*pag
;
691 struct xfs_btree_cur
*cur
, *tcur
;
692 struct xfs_inobt_rec_incore rec
, trec
;
698 pag
= xfs_perag_get(mp
, agno
);
700 ASSERT(pag
->pagi_init
);
701 ASSERT(pag
->pagi_inodeok
);
702 ASSERT(pag
->pagi_freecount
> 0);
705 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
);
707 * If pagino is 0 (this is the root inode allocation) use newino.
708 * This must work because we've just allocated some.
711 pagino
= be32_to_cpu(agi
->agi_newino
);
713 error
= xfs_check_agi_freecount(cur
, agi
);
718 * If in the same AG as the parent, try to get near the parent.
721 int doneleft
; /* done, to the left */
722 int doneright
; /* done, to the right */
723 int searchdistance
= 10;
725 error
= xfs_inobt_lookup(cur
, pagino
, XFS_LOOKUP_LE
, &i
);
728 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
730 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
733 XFS_WANT_CORRUPTED_GOTO(j
== 1, error0
);
735 if (rec
.ir_freecount
> 0) {
737 * Found a free inode in the same chunk
738 * as the parent, done.
745 * In the same AG as parent, but parent's chunk is full.
748 /* duplicate the cursor, search left & right simultaneously */
749 error
= xfs_btree_dup_cursor(cur
, &tcur
);
754 * Skip to last blocks looked up if same parent inode.
756 if (pagino
!= NULLAGINO
&&
757 pag
->pagl_pagino
== pagino
&&
758 pag
->pagl_leftrec
!= NULLAGINO
&&
759 pag
->pagl_rightrec
!= NULLAGINO
) {
760 error
= xfs_ialloc_get_rec(tcur
, pag
->pagl_leftrec
,
765 error
= xfs_ialloc_get_rec(cur
, pag
->pagl_rightrec
,
770 /* search left with tcur, back up 1 record */
771 error
= xfs_ialloc_next_rec(tcur
, &trec
, &doneleft
, 1);
775 /* search right with cur, go forward 1 record. */
776 error
= xfs_ialloc_next_rec(cur
, &rec
, &doneright
, 0);
782 * Loop until we find an inode chunk with a free inode.
784 while (!doneleft
|| !doneright
) {
785 int useleft
; /* using left inode chunk this time */
787 if (!--searchdistance
) {
789 * Not in range - save last search
790 * location and allocate a new inode
792 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
793 pag
->pagl_leftrec
= trec
.ir_startino
;
794 pag
->pagl_rightrec
= rec
.ir_startino
;
795 pag
->pagl_pagino
= pagino
;
799 /* figure out the closer block if both are valid. */
800 if (!doneleft
&& !doneright
) {
802 (trec
.ir_startino
+ XFS_INODES_PER_CHUNK
- 1) <
803 rec
.ir_startino
- pagino
;
808 /* free inodes to the left? */
809 if (useleft
&& trec
.ir_freecount
) {
811 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
814 pag
->pagl_leftrec
= trec
.ir_startino
;
815 pag
->pagl_rightrec
= rec
.ir_startino
;
816 pag
->pagl_pagino
= pagino
;
820 /* free inodes to the right? */
821 if (!useleft
&& rec
.ir_freecount
) {
822 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
824 pag
->pagl_leftrec
= trec
.ir_startino
;
825 pag
->pagl_rightrec
= rec
.ir_startino
;
826 pag
->pagl_pagino
= pagino
;
830 /* get next record to check */
832 error
= xfs_ialloc_next_rec(tcur
, &trec
,
835 error
= xfs_ialloc_next_rec(cur
, &rec
,
843 * We've reached the end of the btree. because
844 * we are only searching a small chunk of the
845 * btree each search, there is obviously free
846 * inodes closer to the parent inode than we
847 * are now. restart the search again.
849 pag
->pagl_pagino
= NULLAGINO
;
850 pag
->pagl_leftrec
= NULLAGINO
;
851 pag
->pagl_rightrec
= NULLAGINO
;
852 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
853 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
858 * In a different AG from the parent.
859 * See if the most recently allocated block has any free.
862 if (agi
->agi_newino
!= cpu_to_be32(NULLAGINO
)) {
863 error
= xfs_inobt_lookup(cur
, be32_to_cpu(agi
->agi_newino
),
869 error
= xfs_inobt_get_rec(cur
, &rec
, &j
);
873 if (j
== 1 && rec
.ir_freecount
> 0) {
875 * The last chunk allocated in the group
876 * still has a free inode.
884 * None left in the last group, search the whole AG
886 error
= xfs_inobt_lookup(cur
, 0, XFS_LOOKUP_GE
, &i
);
889 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
892 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
895 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
896 if (rec
.ir_freecount
> 0)
898 error
= xfs_btree_increment(cur
, 0, &i
);
901 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
905 offset
= xfs_lowbit64(rec
.ir_free
);
907 ASSERT(offset
< XFS_INODES_PER_CHUNK
);
908 ASSERT((XFS_AGINO_TO_OFFSET(mp
, rec
.ir_startino
) %
909 XFS_INODES_PER_CHUNK
) == 0);
910 ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
+ offset
);
911 rec
.ir_free
&= ~XFS_INOBT_MASK(offset
);
913 error
= xfs_inobt_update(cur
, &rec
);
916 be32_add_cpu(&agi
->agi_freecount
, -1);
917 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
918 pag
->pagi_freecount
--;
920 error
= xfs_check_agi_freecount(cur
, agi
);
924 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
925 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -1);
930 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
932 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
938 * Allocate an inode on disk.
940 * Mode is used to tell whether the new inode will need space, and whether it
943 * This function is designed to be called twice if it has to do an allocation
944 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
945 * If an inode is available without having to performn an allocation, an inode
946 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
947 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
948 * The caller should then commit the current transaction, allocate a
949 * new transaction, and call xfs_dialloc() again, passing in the previous value
950 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
951 * buffer is locked across the two calls, the second call is guaranteed to have
952 * a free inode available.
954 * Once we successfully pick an inode its number is returned and the on-disk
955 * data structures are updated. The inode itself is not read in, since doing so
956 * would break ordering constraints with xfs_reclaim.
960 struct xfs_trans
*tp
,
964 struct xfs_buf
**IO_agbp
,
967 struct xfs_mount
*mp
= tp
->t_mountp
;
968 struct xfs_buf
*agbp
;
973 xfs_agnumber_t start_agno
;
974 struct xfs_perag
*pag
;
978 * If the caller passes in a pointer to the AGI buffer,
979 * continue where we left off before. In this case, we
980 * know that the allocation group has free inodes.
987 * We do not have an agbp, so select an initial allocation
988 * group for inode allocation.
990 start_agno
= xfs_ialloc_ag_select(tp
, parent
, mode
, okalloc
);
991 if (start_agno
== NULLAGNUMBER
) {
997 * If we have already hit the ceiling of inode blocks then clear
998 * okalloc so we scan all available agi structures for a free
1001 if (mp
->m_maxicount
&&
1002 mp
->m_sb
.sb_icount
+ XFS_IALLOC_INODES(mp
) > mp
->m_maxicount
) {
1008 * Loop until we find an allocation group that either has free inodes
1009 * or in which we can allocate some inodes. Iterate through the
1010 * allocation groups upward, wrapping at the end.
1014 pag
= xfs_perag_get(mp
, agno
);
1015 if (!pag
->pagi_inodeok
) {
1016 xfs_ialloc_next_ag(mp
);
1020 if (!pag
->pagi_init
) {
1021 error
= xfs_ialloc_pagi_init(mp
, tp
, agno
);
1027 * Do a first racy fast path check if this AG is usable.
1029 if (!pag
->pagi_freecount
&& !okalloc
)
1033 * Then read in the AGI buffer and recheck with the AGI buffer
1036 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1040 if (pag
->pagi_freecount
) {
1046 goto nextag_relse_buffer
;
1049 error
= xfs_ialloc_ag_alloc(tp
, agbp
, &ialloced
);
1051 xfs_trans_brelse(tp
, agbp
);
1053 if (error
!= ENOSPC
)
1063 * We successfully allocated some inodes, return
1064 * the current context to the caller so that it
1065 * can commit the current transaction and call
1066 * us again where we left off.
1068 ASSERT(pag
->pagi_freecount
> 0);
1076 nextag_relse_buffer
:
1077 xfs_trans_brelse(tp
, agbp
);
1080 if (++agno
== mp
->m_sb
.sb_agcount
)
1082 if (agno
== start_agno
) {
1084 return noroom
? ENOSPC
: 0;
1090 return xfs_dialloc_ag(tp
, agbp
, parent
, inop
);
1093 return XFS_ERROR(error
);
1097 * Free disk inode. Carefully avoids touching the incore inode, all
1098 * manipulations incore are the caller's responsibility.
1099 * The on-disk inode is not changed by this operation, only the
1100 * btree (free inode mask) is changed.
1104 xfs_trans_t
*tp
, /* transaction pointer */
1105 xfs_ino_t inode
, /* inode to be freed */
1106 xfs_bmap_free_t
*flist
, /* extents to free */
1107 int *delete, /* set if inode cluster was deleted */
1108 xfs_ino_t
*first_ino
) /* first inode in deleted cluster */
1111 xfs_agblock_t agbno
; /* block number containing inode */
1112 xfs_buf_t
*agbp
; /* buffer containing allocation group header */
1113 xfs_agino_t agino
; /* inode number relative to allocation group */
1114 xfs_agnumber_t agno
; /* allocation group number */
1115 xfs_agi_t
*agi
; /* allocation group header */
1116 xfs_btree_cur_t
*cur
; /* inode btree cursor */
1117 int error
; /* error return value */
1118 int i
; /* result code */
1119 int ilen
; /* inodes in an inode cluster */
1120 xfs_mount_t
*mp
; /* mount structure for filesystem */
1121 int off
; /* offset of inode in inode chunk */
1122 xfs_inobt_rec_incore_t rec
; /* btree record */
1123 struct xfs_perag
*pag
;
1128 * Break up inode number into its components.
1130 agno
= XFS_INO_TO_AGNO(mp
, inode
);
1131 if (agno
>= mp
->m_sb
.sb_agcount
) {
1132 xfs_warn(mp
, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1133 __func__
, agno
, mp
->m_sb
.sb_agcount
);
1135 return XFS_ERROR(EINVAL
);
1137 agino
= XFS_INO_TO_AGINO(mp
, inode
);
1138 if (inode
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1139 xfs_warn(mp
, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1140 __func__
, (unsigned long long)inode
,
1141 (unsigned long long)XFS_AGINO_TO_INO(mp
, agno
, agino
));
1143 return XFS_ERROR(EINVAL
);
1145 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
1146 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
1147 xfs_warn(mp
, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1148 __func__
, agbno
, mp
->m_sb
.sb_agblocks
);
1150 return XFS_ERROR(EINVAL
);
1153 * Get the allocation group header.
1155 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1157 xfs_warn(mp
, "%s: xfs_ialloc_read_agi() returned error %d.",
1161 agi
= XFS_BUF_TO_AGI(agbp
);
1162 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1163 ASSERT(agbno
< be32_to_cpu(agi
->agi_length
));
1165 * Initialize the cursor.
1167 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
);
1169 error
= xfs_check_agi_freecount(cur
, agi
);
1174 * Look for the entry describing this inode.
1176 if ((error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
))) {
1177 xfs_warn(mp
, "%s: xfs_inobt_lookup() returned error %d.",
1181 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1182 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1184 xfs_warn(mp
, "%s: xfs_inobt_get_rec() returned error %d.",
1188 XFS_WANT_CORRUPTED_GOTO(i
== 1, error0
);
1190 * Get the offset in the inode chunk.
1192 off
= agino
- rec
.ir_startino
;
1193 ASSERT(off
>= 0 && off
< XFS_INODES_PER_CHUNK
);
1194 ASSERT(!(rec
.ir_free
& XFS_INOBT_MASK(off
)));
1196 * Mark the inode free & increment the count.
1198 rec
.ir_free
|= XFS_INOBT_MASK(off
);
1202 * When an inode cluster is free, it becomes eligible for removal
1204 if (!(mp
->m_flags
& XFS_MOUNT_IKEEP
) &&
1205 (rec
.ir_freecount
== XFS_IALLOC_INODES(mp
))) {
1208 *first_ino
= XFS_AGINO_TO_INO(mp
, agno
, rec
.ir_startino
);
1211 * Remove the inode cluster from the AGI B+Tree, adjust the
1212 * AGI and Superblock inode counts, and mark the disk space
1213 * to be freed when the transaction is committed.
1215 ilen
= XFS_IALLOC_INODES(mp
);
1216 be32_add_cpu(&agi
->agi_count
, -ilen
);
1217 be32_add_cpu(&agi
->agi_freecount
, -(ilen
- 1));
1218 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_COUNT
| XFS_AGI_FREECOUNT
);
1219 pag
= xfs_perag_get(mp
, agno
);
1220 pag
->pagi_freecount
-= ilen
- 1;
1222 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_ICOUNT
, -ilen
);
1223 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, -(ilen
- 1));
1225 if ((error
= xfs_btree_delete(cur
, &i
))) {
1226 xfs_warn(mp
, "%s: xfs_btree_delete returned error %d.",
1231 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp
,
1232 agno
, XFS_INO_TO_AGBNO(mp
,rec
.ir_startino
)),
1233 XFS_IALLOC_BLOCKS(mp
), flist
, mp
);
1237 error
= xfs_inobt_update(cur
, &rec
);
1239 xfs_warn(mp
, "%s: xfs_inobt_update returned error %d.",
1245 * Change the inode free counts and log the ag/sb changes.
1247 be32_add_cpu(&agi
->agi_freecount
, 1);
1248 xfs_ialloc_log_agi(tp
, agbp
, XFS_AGI_FREECOUNT
);
1249 pag
= xfs_perag_get(mp
, agno
);
1250 pag
->pagi_freecount
++;
1252 xfs_trans_mod_sb(tp
, XFS_TRANS_SB_IFREE
, 1);
1255 error
= xfs_check_agi_freecount(cur
, agi
);
1259 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1263 xfs_btree_del_cursor(cur
, XFS_BTREE_ERROR
);
1269 struct xfs_mount
*mp
,
1270 struct xfs_trans
*tp
,
1271 xfs_agnumber_t agno
,
1273 xfs_agblock_t agbno
,
1274 xfs_agblock_t
*chunk_agbno
,
1275 xfs_agblock_t
*offset_agbno
,
1278 struct xfs_inobt_rec_incore rec
;
1279 struct xfs_btree_cur
*cur
;
1280 struct xfs_buf
*agbp
;
1284 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &agbp
);
1287 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1288 __func__
, error
, agno
);
1293 * Lookup the inode record for the given agino. If the record cannot be
1294 * found, then it's an invalid inode number and we should abort. Once
1295 * we have a record, we need to ensure it contains the inode number
1296 * we are looking up.
1298 cur
= xfs_inobt_init_cursor(mp
, tp
, agbp
, agno
);
1299 error
= xfs_inobt_lookup(cur
, agino
, XFS_LOOKUP_LE
, &i
);
1302 error
= xfs_inobt_get_rec(cur
, &rec
, &i
);
1303 if (!error
&& i
== 0)
1307 xfs_trans_brelse(tp
, agbp
);
1308 xfs_btree_del_cursor(cur
, XFS_BTREE_NOERROR
);
1312 /* check that the returned record contains the required inode */
1313 if (rec
.ir_startino
> agino
||
1314 rec
.ir_startino
+ XFS_IALLOC_INODES(mp
) <= agino
)
1317 /* for untrusted inodes check it is allocated first */
1318 if ((flags
& XFS_IGET_UNTRUSTED
) &&
1319 (rec
.ir_free
& XFS_INOBT_MASK(agino
- rec
.ir_startino
)))
1322 *chunk_agbno
= XFS_AGINO_TO_AGBNO(mp
, rec
.ir_startino
);
1323 *offset_agbno
= agbno
- *chunk_agbno
;
1328 * Return the location of the inode in imap, for mapping it into a buffer.
1332 xfs_mount_t
*mp
, /* file system mount structure */
1333 xfs_trans_t
*tp
, /* transaction pointer */
1334 xfs_ino_t ino
, /* inode to locate */
1335 struct xfs_imap
*imap
, /* location map structure */
1336 uint flags
) /* flags for inode btree lookup */
1338 xfs_agblock_t agbno
; /* block number of inode in the alloc group */
1339 xfs_agino_t agino
; /* inode number within alloc group */
1340 xfs_agnumber_t agno
; /* allocation group number */
1341 int blks_per_cluster
; /* num blocks per inode cluster */
1342 xfs_agblock_t chunk_agbno
; /* first block in inode chunk */
1343 xfs_agblock_t cluster_agbno
; /* first block in inode cluster */
1344 int error
; /* error code */
1345 int offset
; /* index of inode in its buffer */
1346 xfs_agblock_t offset_agbno
; /* blks from chunk start to inode */
1348 ASSERT(ino
!= NULLFSINO
);
1351 * Split up the inode number into its parts.
1353 agno
= XFS_INO_TO_AGNO(mp
, ino
);
1354 agino
= XFS_INO_TO_AGINO(mp
, ino
);
1355 agbno
= XFS_AGINO_TO_AGBNO(mp
, agino
);
1356 if (agno
>= mp
->m_sb
.sb_agcount
|| agbno
>= mp
->m_sb
.sb_agblocks
||
1357 ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1360 * Don't output diagnostic information for untrusted inodes
1361 * as they can be invalid without implying corruption.
1363 if (flags
& XFS_IGET_UNTRUSTED
)
1364 return XFS_ERROR(EINVAL
);
1365 if (agno
>= mp
->m_sb
.sb_agcount
) {
1367 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1368 __func__
, agno
, mp
->m_sb
.sb_agcount
);
1370 if (agbno
>= mp
->m_sb
.sb_agblocks
) {
1372 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1373 __func__
, (unsigned long long)agbno
,
1374 (unsigned long)mp
->m_sb
.sb_agblocks
);
1376 if (ino
!= XFS_AGINO_TO_INO(mp
, agno
, agino
)) {
1378 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1380 XFS_AGINO_TO_INO(mp
, agno
, agino
));
1384 return XFS_ERROR(EINVAL
);
1387 blks_per_cluster
= XFS_INODE_CLUSTER_SIZE(mp
) >> mp
->m_sb
.sb_blocklog
;
1390 * For bulkstat and handle lookups, we have an untrusted inode number
1391 * that we have to verify is valid. We cannot do this just by reading
1392 * the inode buffer as it may have been unlinked and removed leaving
1393 * inodes in stale state on disk. Hence we have to do a btree lookup
1394 * in all cases where an untrusted inode number is passed.
1396 if (flags
& XFS_IGET_UNTRUSTED
) {
1397 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
1398 &chunk_agbno
, &offset_agbno
, flags
);
1405 * If the inode cluster size is the same as the blocksize or
1406 * smaller we get to the buffer by simple arithmetics.
1408 if (XFS_INODE_CLUSTER_SIZE(mp
) <= mp
->m_sb
.sb_blocksize
) {
1409 offset
= XFS_INO_TO_OFFSET(mp
, ino
);
1410 ASSERT(offset
< mp
->m_sb
.sb_inopblock
);
1412 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
1413 imap
->im_len
= XFS_FSB_TO_BB(mp
, 1);
1414 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
1419 * If the inode chunks are aligned then use simple maths to
1420 * find the location. Otherwise we have to do a btree
1421 * lookup to find the location.
1423 if (mp
->m_inoalign_mask
) {
1424 offset_agbno
= agbno
& mp
->m_inoalign_mask
;
1425 chunk_agbno
= agbno
- offset_agbno
;
1427 error
= xfs_imap_lookup(mp
, tp
, agno
, agino
, agbno
,
1428 &chunk_agbno
, &offset_agbno
, flags
);
1434 ASSERT(agbno
>= chunk_agbno
);
1435 cluster_agbno
= chunk_agbno
+
1436 ((offset_agbno
/ blks_per_cluster
) * blks_per_cluster
);
1437 offset
= ((agbno
- cluster_agbno
) * mp
->m_sb
.sb_inopblock
) +
1438 XFS_INO_TO_OFFSET(mp
, ino
);
1440 imap
->im_blkno
= XFS_AGB_TO_DADDR(mp
, agno
, cluster_agbno
);
1441 imap
->im_len
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
1442 imap
->im_boffset
= (ushort
)(offset
<< mp
->m_sb
.sb_inodelog
);
1445 * If the inode number maps to a block outside the bounds
1446 * of the file system then return NULL rather than calling
1447 * read_buf and panicing when we get an error from the
1450 if ((imap
->im_blkno
+ imap
->im_len
) >
1451 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)) {
1453 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1454 __func__
, (unsigned long long) imap
->im_blkno
,
1455 (unsigned long long) imap
->im_len
,
1456 XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
));
1457 return XFS_ERROR(EINVAL
);
1463 * Compute and fill in value of m_in_maxlevels.
1466 xfs_ialloc_compute_maxlevels(
1467 xfs_mount_t
*mp
) /* file system mount structure */
1475 maxleafents
= (1LL << XFS_INO_AGINO_BITS(mp
)) >>
1476 XFS_INODES_PER_CHUNK_LOG
;
1477 minleafrecs
= mp
->m_alloc_mnr
[0];
1478 minnoderecs
= mp
->m_alloc_mnr
[1];
1479 maxblocks
= (maxleafents
+ minleafrecs
- 1) / minleafrecs
;
1480 for (level
= 1; maxblocks
> 1; level
++)
1481 maxblocks
= (maxblocks
+ minnoderecs
- 1) / minnoderecs
;
1482 mp
->m_in_maxlevels
= level
;
1486 * Log specified fields for the ag hdr (inode section)
1490 xfs_trans_t
*tp
, /* transaction pointer */
1491 xfs_buf_t
*bp
, /* allocation group header buffer */
1492 int fields
) /* bitmask of fields to log */
1494 int first
; /* first byte number */
1495 int last
; /* last byte number */
1496 static const short offsets
[] = { /* field starting offsets */
1497 /* keep in sync with bit definitions */
1498 offsetof(xfs_agi_t
, agi_magicnum
),
1499 offsetof(xfs_agi_t
, agi_versionnum
),
1500 offsetof(xfs_agi_t
, agi_seqno
),
1501 offsetof(xfs_agi_t
, agi_length
),
1502 offsetof(xfs_agi_t
, agi_count
),
1503 offsetof(xfs_agi_t
, agi_root
),
1504 offsetof(xfs_agi_t
, agi_level
),
1505 offsetof(xfs_agi_t
, agi_freecount
),
1506 offsetof(xfs_agi_t
, agi_newino
),
1507 offsetof(xfs_agi_t
, agi_dirino
),
1508 offsetof(xfs_agi_t
, agi_unlinked
),
1512 xfs_agi_t
*agi
; /* allocation group header */
1514 agi
= XFS_BUF_TO_AGI(bp
);
1515 ASSERT(agi
->agi_magicnum
== cpu_to_be32(XFS_AGI_MAGIC
));
1518 * Compute byte offsets for the first and last fields.
1520 xfs_btree_offsets(fields
, offsets
, XFS_AGI_NUM_BITS
, &first
, &last
);
1522 * Log the allocation group inode header buffer.
1524 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_AGI_BUF
);
1525 xfs_trans_log_buf(tp
, bp
, first
, last
);
1530 xfs_check_agi_unlinked(
1531 struct xfs_agi
*agi
)
1535 for (i
= 0; i
< XFS_AGI_UNLINKED_BUCKETS
; i
++)
1536 ASSERT(agi
->agi_unlinked
[i
]);
1539 #define xfs_check_agi_unlinked(agi)
1546 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
1547 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(bp
);
1549 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
1550 !uuid_equal(&agi
->agi_uuid
, &mp
->m_sb
.sb_uuid
))
1553 * Validate the magic number of the agi block.
1555 if (agi
->agi_magicnum
!= cpu_to_be32(XFS_AGI_MAGIC
))
1557 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi
->agi_versionnum
)))
1561 * during growfs operations, the perag is not fully initialised,
1562 * so we can't use it for any useful checking. growfs ensures we can't
1563 * use it by using uncached buffers that don't have the perag attached
1564 * so we can detect and avoid this problem.
1566 if (bp
->b_pag
&& be32_to_cpu(agi
->agi_seqno
) != bp
->b_pag
->pag_agno
)
1569 xfs_check_agi_unlinked(agi
);
1574 xfs_agi_read_verify(
1577 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
1580 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1581 agi_ok
= xfs_verify_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
1582 offsetof(struct xfs_agi
, agi_crc
));
1583 agi_ok
= agi_ok
&& xfs_agi_verify(bp
);
1585 if (unlikely(XFS_TEST_ERROR(!agi_ok
, mp
, XFS_ERRTAG_IALLOC_READ_AGI
,
1586 XFS_RANDOM_IALLOC_READ_AGI
))) {
1587 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
1588 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
1593 xfs_agi_write_verify(
1596 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
1597 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
1599 if (!xfs_agi_verify(bp
)) {
1600 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
1601 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
1605 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1609 XFS_BUF_TO_AGI(bp
)->agi_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
1610 xfs_update_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
1611 offsetof(struct xfs_agi
, agi_crc
));
1614 const struct xfs_buf_ops xfs_agi_buf_ops
= {
1615 .verify_read
= xfs_agi_read_verify
,
1616 .verify_write
= xfs_agi_write_verify
,
1620 * Read in the allocation group header (inode allocation section)
1624 struct xfs_mount
*mp
, /* file system mount structure */
1625 struct xfs_trans
*tp
, /* transaction pointer */
1626 xfs_agnumber_t agno
, /* allocation group number */
1627 struct xfs_buf
**bpp
) /* allocation group hdr buf */
1631 trace_xfs_read_agi(mp
, agno
);
1633 ASSERT(agno
!= NULLAGNUMBER
);
1634 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
1635 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
1636 XFS_FSS_TO_BB(mp
, 1), 0, bpp
, &xfs_agi_buf_ops
);
1640 ASSERT(!xfs_buf_geterror(*bpp
));
1641 xfs_buf_set_ref(*bpp
, XFS_AGI_REF
);
1646 xfs_ialloc_read_agi(
1647 struct xfs_mount
*mp
, /* file system mount structure */
1648 struct xfs_trans
*tp
, /* transaction pointer */
1649 xfs_agnumber_t agno
, /* allocation group number */
1650 struct xfs_buf
**bpp
) /* allocation group hdr buf */
1652 struct xfs_agi
*agi
; /* allocation group header */
1653 struct xfs_perag
*pag
; /* per allocation group data */
1656 trace_xfs_ialloc_read_agi(mp
, agno
);
1658 error
= xfs_read_agi(mp
, tp
, agno
, bpp
);
1662 agi
= XFS_BUF_TO_AGI(*bpp
);
1663 pag
= xfs_perag_get(mp
, agno
);
1664 if (!pag
->pagi_init
) {
1665 pag
->pagi_freecount
= be32_to_cpu(agi
->agi_freecount
);
1666 pag
->pagi_count
= be32_to_cpu(agi
->agi_count
);
1671 * It's possible for these to be out of sync if
1672 * we are in the middle of a forced shutdown.
1674 ASSERT(pag
->pagi_freecount
== be32_to_cpu(agi
->agi_freecount
) ||
1675 XFS_FORCED_SHUTDOWN(mp
));
1681 * Read in the agi to initialise the per-ag data in the mount structure
1684 xfs_ialloc_pagi_init(
1685 xfs_mount_t
*mp
, /* file system mount structure */
1686 xfs_trans_t
*tp
, /* transaction pointer */
1687 xfs_agnumber_t agno
) /* allocation group number */
1689 xfs_buf_t
*bp
= NULL
;
1692 error
= xfs_ialloc_read_agi(mp
, tp
, agno
, &bp
);
1696 xfs_trans_brelse(tp
, bp
);