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_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
46 * Allocation group level functions.
49 xfs_ialloc_cluster_alignment(
50 xfs_alloc_arg_t *args)
52 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
53 args->mp->m_sb.sb_inoalignmt >=
54 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
55 return args->mp->m_sb.sb_inoalignmt;
60 * Lookup the record equal to ino in the btree given by cur.
62 STATIC int /* error */
64 struct xfs_btree_cur *cur, /* btree cursor */
65 xfs_agino_t ino, /* starting inode of chunk */
66 __int32_t fcnt, /* free inode count */
67 xfs_inofree_t free, /* free inode mask */
68 int *stat) /* success/failure */
70 cur->bc_rec.i.ir_startino = ino;
71 cur->bc_rec.i.ir_freecount = fcnt;
72 cur->bc_rec.i.ir_free = free;
73 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
77 * Lookup the first record greater than or equal to ino
78 * in the btree given by cur.
82 struct xfs_btree_cur *cur, /* btree cursor */
83 xfs_agino_t ino, /* starting inode of chunk */
84 __int32_t fcnt, /* free inode count */
85 xfs_inofree_t free, /* free inode mask */
86 int *stat) /* success/failure */
88 cur->bc_rec.i.ir_startino = ino;
89 cur->bc_rec.i.ir_freecount = fcnt;
90 cur->bc_rec.i.ir_free = free;
91 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
95 * Lookup the first record less than or equal to ino
96 * in the btree given by cur.
100 struct xfs_btree_cur *cur, /* btree cursor */
101 xfs_agino_t ino, /* starting inode of chunk */
102 __int32_t fcnt, /* free inode count */
103 xfs_inofree_t free, /* free inode mask */
104 int *stat) /* success/failure */
106 cur->bc_rec.i.ir_startino = ino;
107 cur->bc_rec.i.ir_freecount = fcnt;
108 cur->bc_rec.i.ir_free = free;
109 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
113 * Update the record referred to by cur to the value given.
114 * This either works (return 0) or gets an EFSCORRUPTED error.
116 STATIC int /* error */
118 struct xfs_btree_cur *cur, /* btree cursor */
119 xfs_inobt_rec_incore_t *irec) /* btree record */
121 union xfs_btree_rec rec;
123 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
124 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
125 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
126 return xfs_btree_update(cur, &rec);
130 * Get the data from the pointed-to record.
134 struct xfs_btree_cur *cur, /* btree cursor */
135 xfs_inobt_rec_incore_t *irec, /* btree record */
136 int *stat) /* output: success/failure */
138 union xfs_btree_rec *rec;
141 error = xfs_btree_get_rec(cur, &rec, stat);
142 if (!error && *stat == 1) {
143 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
144 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
145 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
151 * Verify that the number of free inodes in the AGI is correct.
155 xfs_check_agi_freecount(
156 struct xfs_btree_cur *cur,
159 if (cur->bc_nlevels == 1) {
160 xfs_inobt_rec_incore_t rec;
165 error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i);
170 error = xfs_inobt_get_rec(cur, &rec, &i);
175 freecount += rec.ir_freecount;
176 error = xfs_btree_increment(cur, 0, &i);
182 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
183 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
188 #define xfs_check_agi_freecount(cur, agi) 0
192 * Initialise a new set of inodes.
195 xfs_ialloc_inode_init(
196 struct xfs_mount *mp,
197 struct xfs_trans *tp,
200 xfs_agblock_t length,
203 struct xfs_buf *fbuf;
204 struct xfs_dinode *free;
205 int blks_per_cluster, nbufs, ninodes;
211 * Loop over the new block(s), filling in the inodes.
212 * For small block sizes, manipulate the inodes in buffers
213 * which are multiples of the blocks size.
215 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
216 blks_per_cluster = 1;
218 ninodes = mp->m_sb.sb_inopblock;
220 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
221 mp->m_sb.sb_blocksize;
222 nbufs = length / blks_per_cluster;
223 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
227 * Figure out what version number to use in the inodes we create.
228 * If the superblock version has caught up to the one that supports
229 * the new inode format, then use the new inode version. Otherwise
230 * use the old version so that old kernels will continue to be
231 * able to use the file system.
233 if (xfs_sb_version_hasnlink(&mp->m_sb))
238 for (j = 0; j < nbufs; j++) {
242 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
243 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
244 mp->m_bsize * blks_per_cluster,
247 ASSERT(!XFS_BUF_GETERROR(fbuf));
250 * Initialize all inodes in this buffer and then log them.
252 * XXX: It would be much better if we had just one transaction
253 * to log a whole cluster of inodes instead of all the
254 * individual transactions causing a lot of log traffic.
256 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
257 for (i = 0; i < ninodes; i++) {
258 int ioffset = i << mp->m_sb.sb_inodelog;
259 uint isize = sizeof(struct xfs_dinode);
261 free = xfs_make_iptr(mp, fbuf, i);
262 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
263 free->di_version = version;
264 free->di_gen = cpu_to_be32(gen);
265 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
266 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
268 xfs_trans_inode_alloc_buf(tp, fbuf);
273 * Allocate new inodes in the allocation group specified by agbp.
274 * Return 0 for success, else error code.
276 STATIC int /* error code or 0 */
278 xfs_trans_t *tp, /* transaction pointer */
279 xfs_buf_t *agbp, /* alloc group buffer */
282 xfs_agi_t *agi; /* allocation group header */
283 xfs_alloc_arg_t args; /* allocation argument structure */
284 xfs_btree_cur_t *cur; /* inode btree cursor */
288 xfs_agino_t newino; /* new first inode's number */
289 xfs_agino_t newlen; /* new number of inodes */
290 xfs_agino_t thisino; /* current inode number, for loop */
291 int isaligned = 0; /* inode allocation at stripe unit */
295 args.mp = tp->t_mountp;
298 * Locking will ensure that we don't have two callers in here
301 newlen = XFS_IALLOC_INODES(args.mp);
302 if (args.mp->m_maxicount &&
303 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
304 return XFS_ERROR(ENOSPC);
305 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
307 * First try to allocate inodes contiguous with the last-allocated
308 * chunk of inodes. If the filesystem is striped, this will fill
309 * an entire stripe unit with inodes.
311 agi = XFS_BUF_TO_AGI(agbp);
312 newino = be32_to_cpu(agi->agi_newino);
313 agno = be32_to_cpu(agi->agi_seqno);
314 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
315 XFS_IALLOC_BLOCKS(args.mp);
316 if (likely(newino != NULLAGINO &&
317 (args.agbno < be32_to_cpu(agi->agi_length)))) {
318 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
319 args.type = XFS_ALLOCTYPE_THIS_BNO;
320 args.mod = args.total = args.wasdel = args.isfl =
321 args.userdata = args.minalignslop = 0;
325 * We need to take into account alignment here to ensure that
326 * we don't modify the free list if we fail to have an exact
327 * block. If we don't have an exact match, and every oher
328 * attempt allocation attempt fails, we'll end up cancelling
329 * a dirty transaction and shutting down.
331 * For an exact allocation, alignment must be 1,
332 * however we need to take cluster alignment into account when
333 * fixing up the freelist. Use the minalignslop field to
334 * indicate that extra blocks might be required for alignment,
335 * but not to use them in the actual exact allocation.
338 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
340 /* Allow space for the inode btree to split. */
341 args.minleft = args.mp->m_in_maxlevels - 1;
342 if ((error = xfs_alloc_vextent(&args)))
345 args.fsbno = NULLFSBLOCK;
347 if (unlikely(args.fsbno == NULLFSBLOCK)) {
349 * Set the alignment for the allocation.
350 * If stripe alignment is turned on then align at stripe unit
352 * If the cluster size is smaller than a filesystem block
353 * then we're doing I/O for inodes in filesystem block size
354 * pieces, so don't need alignment anyway.
357 if (args.mp->m_sinoalign) {
358 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
359 args.alignment = args.mp->m_dalign;
362 args.alignment = xfs_ialloc_cluster_alignment(&args);
364 * Need to figure out where to allocate the inode blocks.
365 * Ideally they should be spaced out through the a.g.
366 * For now, just allocate blocks up front.
368 args.agbno = be32_to_cpu(agi->agi_root);
369 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
371 * Allocate a fixed-size extent of inodes.
373 args.type = XFS_ALLOCTYPE_NEAR_BNO;
374 args.mod = args.total = args.wasdel = args.isfl =
375 args.userdata = args.minalignslop = 0;
378 * Allow space for the inode btree to split.
380 args.minleft = args.mp->m_in_maxlevels - 1;
381 if ((error = xfs_alloc_vextent(&args)))
386 * If stripe alignment is turned on, then try again with cluster
389 if (isaligned && args.fsbno == NULLFSBLOCK) {
390 args.type = XFS_ALLOCTYPE_NEAR_BNO;
391 args.agbno = be32_to_cpu(agi->agi_root);
392 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
393 args.alignment = xfs_ialloc_cluster_alignment(&args);
394 if ((error = xfs_alloc_vextent(&args)))
398 if (args.fsbno == NULLFSBLOCK) {
402 ASSERT(args.len == args.minlen);
405 * Stamp and write the inode buffers.
407 * Seed the new inode cluster with a random generation number. This
408 * prevents short-term reuse of generation numbers if a chunk is
409 * freed and then immediately reallocated. We use random numbers
410 * rather than a linear progression to prevent the next generation
411 * number from being easily guessable.
413 xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
417 * Convert the results.
419 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
420 be32_add_cpu(&agi->agi_count, newlen);
421 be32_add_cpu(&agi->agi_freecount, newlen);
422 down_read(&args.mp->m_peraglock);
423 args.mp->m_perag[agno].pagi_freecount += newlen;
424 up_read(&args.mp->m_peraglock);
425 agi->agi_newino = cpu_to_be32(newino);
428 * Insert records describing the new inode chunk into the btree.
430 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
431 for (thisino = newino;
432 thisino < newino + newlen;
433 thisino += XFS_INODES_PER_CHUNK) {
434 if ((error = xfs_inobt_lookup_eq(cur, thisino,
435 XFS_INODES_PER_CHUNK, XFS_INOBT_ALL_FREE, &i))) {
436 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
440 if ((error = xfs_btree_insert(cur, &i))) {
441 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
446 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
448 * Log allocation group header fields
450 xfs_ialloc_log_agi(tp, agbp,
451 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
453 * Modify/log superblock values for inode count and inode free count.
455 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
456 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
461 STATIC_INLINE xfs_agnumber_t
467 spin_lock(&mp->m_agirotor_lock);
468 agno = mp->m_agirotor;
469 if (++mp->m_agirotor == mp->m_maxagi)
471 spin_unlock(&mp->m_agirotor_lock);
477 * Select an allocation group to look for a free inode in, based on the parent
478 * inode and then mode. Return the allocation group buffer.
480 STATIC xfs_buf_t * /* allocation group buffer */
481 xfs_ialloc_ag_select(
482 xfs_trans_t *tp, /* transaction pointer */
483 xfs_ino_t parent, /* parent directory inode number */
484 mode_t mode, /* bits set to indicate file type */
485 int okalloc) /* ok to allocate more space */
487 xfs_buf_t *agbp; /* allocation group header buffer */
488 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
489 xfs_agnumber_t agno; /* current ag number */
490 int flags; /* alloc buffer locking flags */
491 xfs_extlen_t ineed; /* blocks needed for inode allocation */
492 xfs_extlen_t longest = 0; /* longest extent available */
493 xfs_mount_t *mp; /* mount point structure */
494 int needspace; /* file mode implies space allocated */
495 xfs_perag_t *pag; /* per allocation group data */
496 xfs_agnumber_t pagno; /* parent (starting) ag number */
499 * Files of these types need at least one block if length > 0
500 * (and they won't fit in the inode, but that's hard to figure out).
502 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
504 agcount = mp->m_maxagi;
506 pagno = xfs_ialloc_next_ag(mp);
508 pagno = XFS_INO_TO_AGNO(mp, parent);
509 if (pagno >= agcount)
512 ASSERT(pagno < agcount);
514 * Loop through allocation groups, looking for one with a little
515 * free space in it. Note we don't look for free inodes, exactly.
516 * Instead, we include whether there is a need to allocate inodes
517 * to mean that blocks must be allocated for them,
518 * if none are currently free.
521 flags = XFS_ALLOC_FLAG_TRYLOCK;
522 down_read(&mp->m_peraglock);
524 pag = &mp->m_perag[agno];
525 if (!pag->pagi_init) {
526 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
533 if (!pag->pagi_inodeok) {
534 xfs_ialloc_next_ag(mp);
539 * Is there enough free space for the file plus a block
540 * of inodes (if we need to allocate some)?
542 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
543 if (ineed && !pag->pagf_init) {
545 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
549 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
551 if (!ineed || pag->pagf_init) {
552 if (ineed && !(longest = pag->pagf_longest))
553 longest = pag->pagf_flcount > 0;
555 (pag->pagf_freeblks >= needspace + ineed &&
559 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
563 up_read(&mp->m_peraglock);
569 xfs_trans_brelse(tp, agbp);
572 * No point in iterating over the rest, if we're shutting
575 if (XFS_FORCED_SHUTDOWN(mp)) {
576 up_read(&mp->m_peraglock);
584 up_read(&mp->m_peraglock);
593 * Try to retrieve the next record to the left/right from the current one.
597 struct xfs_btree_cur *cur,
598 xfs_inobt_rec_incore_t *rec,
606 error = xfs_btree_decrement(cur, 0, &i);
608 error = xfs_btree_increment(cur, 0, &i);
614 error = xfs_inobt_get_rec(cur, rec, &i);
617 XFS_WANT_CORRUPTED_RETURN(i == 1);
625 * Visible inode allocation functions.
629 * Allocate an inode on disk.
630 * Mode is used to tell whether the new inode will need space, and whether
633 * The arguments IO_agbp and alloc_done are defined to work within
634 * the constraint of one allocation per transaction.
635 * xfs_dialloc() is designed to be called twice if it has to do an
636 * allocation to make more free inodes. On the first call,
637 * IO_agbp should be set to NULL. If an inode is available,
638 * i.e., xfs_dialloc() did not need to do an allocation, an inode
639 * number is returned. In this case, IO_agbp would be set to the
640 * current ag_buf and alloc_done set to false.
641 * If an allocation needed to be done, xfs_dialloc would return
642 * the current ag_buf in IO_agbp and set alloc_done to true.
643 * The caller should then commit the current transaction, allocate a new
644 * transaction, and call xfs_dialloc() again, passing in the previous
645 * value of IO_agbp. IO_agbp should be held across the transactions.
646 * Since the agbp is locked across the two calls, the second call is
647 * guaranteed to have a free inode available.
649 * Once we successfully pick an inode its number is returned and the
650 * on-disk data structures are updated. The inode itself is not read
651 * in, since doing so would break ordering constraints with xfs_reclaim.
655 xfs_trans_t *tp, /* transaction pointer */
656 xfs_ino_t parent, /* parent inode (directory) */
657 mode_t mode, /* mode bits for new inode */
658 int okalloc, /* ok to allocate more space */
659 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
660 boolean_t *alloc_done, /* true if we needed to replenish
662 xfs_ino_t *inop) /* inode number allocated */
664 xfs_agnumber_t agcount; /* number of allocation groups */
665 xfs_buf_t *agbp; /* allocation group header's buffer */
666 xfs_agnumber_t agno; /* allocation group number */
667 xfs_agi_t *agi; /* allocation group header structure */
668 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
669 int error; /* error return value */
670 int i; /* result code */
671 int ialloced; /* inode allocation status */
672 int noroom = 0; /* no space for inode blk allocation */
673 xfs_ino_t ino; /* fs-relative inode to be returned */
675 int j; /* result code */
676 xfs_mount_t *mp; /* file system mount structure */
677 int offset; /* index of inode in chunk */
678 xfs_agino_t pagino; /* parent's AG relative inode # */
679 xfs_agnumber_t pagno; /* parent's AG number */
680 xfs_inobt_rec_incore_t rec; /* inode allocation record */
681 xfs_agnumber_t tagno; /* testing allocation group number */
682 xfs_btree_cur_t *tcur; /* temp cursor */
683 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
686 if (*IO_agbp == NULL) {
688 * We do not have an agbp, so select an initial allocation
689 * group for inode allocation.
691 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
693 * Couldn't find an allocation group satisfying the
700 agi = XFS_BUF_TO_AGI(agbp);
701 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
704 * Continue where we left off before. In this case, we
705 * know that the allocation group has free inodes.
708 agi = XFS_BUF_TO_AGI(agbp);
709 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
710 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
713 agcount = mp->m_sb.sb_agcount;
714 agno = be32_to_cpu(agi->agi_seqno);
716 pagno = XFS_INO_TO_AGNO(mp, parent);
717 pagino = XFS_INO_TO_AGINO(mp, parent);
720 * If we have already hit the ceiling of inode blocks then clear
721 * okalloc so we scan all available agi structures for a free
725 if (mp->m_maxicount &&
726 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
732 * Loop until we find an allocation group that either has free inodes
733 * or in which we can allocate some inodes. Iterate through the
734 * allocation groups upward, wrapping at the end.
736 *alloc_done = B_FALSE;
737 while (!agi->agi_freecount) {
739 * Don't do anything if we're not supposed to allocate
740 * any blocks, just go on to the next ag.
744 * Try to allocate some new inodes in the allocation
747 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
748 xfs_trans_brelse(tp, agbp);
749 if (error == ENOSPC) {
757 * We successfully allocated some inodes, return
758 * the current context to the caller so that it
759 * can commit the current transaction and call
760 * us again where we left off.
762 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
763 *alloc_done = B_TRUE;
770 * If it failed, give up on this ag.
772 xfs_trans_brelse(tp, agbp);
774 * Go on to the next ag: get its ag header.
777 if (++tagno == agcount)
781 return noroom ? ENOSPC : 0;
783 down_read(&mp->m_peraglock);
784 if (mp->m_perag[tagno].pagi_inodeok == 0) {
785 up_read(&mp->m_peraglock);
788 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
789 up_read(&mp->m_peraglock);
792 agi = XFS_BUF_TO_AGI(agbp);
793 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
796 * Here with an allocation group that has a free inode.
797 * Reset agno since we may have chosen a new ag in the
802 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
804 * If pagino is 0 (this is the root inode allocation) use newino.
805 * This must work because we've just allocated some.
808 pagino = be32_to_cpu(agi->agi_newino);
810 error = xfs_check_agi_freecount(cur, agi);
815 * If in the same AG as the parent, try to get near the parent.
818 int doneleft; /* done, to the left */
819 int doneright; /* done, to the right */
821 error = xfs_inobt_lookup_le(cur, pagino, 0, 0, &i);
824 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
826 error = xfs_inobt_get_rec(cur, &rec, &j);
829 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
831 if (rec.ir_freecount > 0) {
833 * Found a free inode in the same chunk
834 * as the parent, done.
841 * In the same AG as parent, but parent's chunk is full.
844 /* duplicate the cursor, search left & right simultaneously */
845 error = xfs_btree_dup_cursor(cur, &tcur);
849 /* search left with tcur, back up 1 record */
850 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
854 /* search right with cur, go forward 1 record. */
855 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
860 * Loop until we find an inode chunk with a free inode.
862 while (!doneleft || !doneright) {
863 int useleft; /* using left inode chunk this time */
865 /* figure out the closer block if both are valid. */
866 if (!doneleft && !doneright) {
868 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
869 rec.ir_startino - pagino;
874 /* free inodes to the left? */
875 if (useleft && trec.ir_freecount) {
877 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
882 /* free inodes to the right? */
883 if (!useleft && rec.ir_freecount) {
884 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
888 /* get next record to check */
890 error = xfs_ialloc_next_rec(tcur, &trec,
893 error = xfs_ialloc_next_rec(cur, &rec,
899 ASSERT(!doneleft || !doneright);
903 * In a different AG from the parent.
904 * See if the most recently allocated block has any free.
906 else if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
907 error = xfs_inobt_lookup_eq(cur, be32_to_cpu(agi->agi_newino),
913 error = xfs_inobt_get_rec(cur, &rec, &j);
917 if (j == 1 && rec.ir_freecount > 0) {
919 * The last chunk allocated in the group
920 * still has a free inode.
927 * None left in the last group, search the whole AG
929 error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i);
932 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
935 error = xfs_inobt_get_rec(cur, &rec, &i);
938 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
939 if (rec.ir_freecount > 0)
941 error = xfs_btree_increment(cur, 0, &i);
944 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
949 offset = xfs_ialloc_find_free(&rec.ir_free);
951 ASSERT(offset < XFS_INODES_PER_CHUNK);
952 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
953 XFS_INODES_PER_CHUNK) == 0);
954 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
955 rec.ir_free &= ~XFS_INOBT_MASK(offset);
957 error = xfs_inobt_update(cur, &rec);
960 be32_add_cpu(&agi->agi_freecount, -1);
961 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
962 down_read(&mp->m_peraglock);
963 mp->m_perag[tagno].pagi_freecount--;
964 up_read(&mp->m_peraglock);
966 error = xfs_check_agi_freecount(cur, agi);
970 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
971 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
975 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
977 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
982 * Free disk inode. Carefully avoids touching the incore inode, all
983 * manipulations incore are the caller's responsibility.
984 * The on-disk inode is not changed by this operation, only the
985 * btree (free inode mask) is changed.
989 xfs_trans_t *tp, /* transaction pointer */
990 xfs_ino_t inode, /* inode to be freed */
991 xfs_bmap_free_t *flist, /* extents to free */
992 int *delete, /* set if inode cluster was deleted */
993 xfs_ino_t *first_ino) /* first inode in deleted cluster */
996 xfs_agblock_t agbno; /* block number containing inode */
997 xfs_buf_t *agbp; /* buffer containing allocation group header */
998 xfs_agino_t agino; /* inode number relative to allocation group */
999 xfs_agnumber_t agno; /* allocation group number */
1000 xfs_agi_t *agi; /* allocation group header */
1001 xfs_btree_cur_t *cur; /* inode btree cursor */
1002 int error; /* error return value */
1003 int i; /* result code */
1004 int ilen; /* inodes in an inode cluster */
1005 xfs_mount_t *mp; /* mount structure for filesystem */
1006 int off; /* offset of inode in inode chunk */
1007 xfs_inobt_rec_incore_t rec; /* btree record */
1012 * Break up inode number into its components.
1014 agno = XFS_INO_TO_AGNO(mp, inode);
1015 if (agno >= mp->m_sb.sb_agcount) {
1017 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
1018 agno, mp->m_sb.sb_agcount, mp->m_fsname);
1020 return XFS_ERROR(EINVAL);
1022 agino = XFS_INO_TO_AGINO(mp, inode);
1023 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1025 "xfs_difree: inode != XFS_AGINO_TO_INO() "
1026 "(%llu != %llu) on %s. Returning EINVAL.",
1027 (unsigned long long)inode,
1028 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
1031 return XFS_ERROR(EINVAL);
1033 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1034 if (agbno >= mp->m_sb.sb_agblocks) {
1036 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1037 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1039 return XFS_ERROR(EINVAL);
1042 * Get the allocation group header.
1044 down_read(&mp->m_peraglock);
1045 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1046 up_read(&mp->m_peraglock);
1049 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1050 error, mp->m_fsname);
1053 agi = XFS_BUF_TO_AGI(agbp);
1054 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1055 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1057 * Initialize the cursor.
1059 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1061 error = xfs_check_agi_freecount(cur, agi);
1066 * Look for the entry describing this inode.
1068 if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) {
1070 "xfs_difree: xfs_inobt_lookup_le returned() an error %d on %s. Returning error.",
1071 error, mp->m_fsname);
1074 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1075 error = xfs_inobt_get_rec(cur, &rec, &i);
1078 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1079 error, mp->m_fsname);
1082 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1084 * Get the offset in the inode chunk.
1086 off = agino - rec.ir_startino;
1087 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1088 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1090 * Mark the inode free & increment the count.
1092 rec.ir_free |= XFS_INOBT_MASK(off);
1096 * When an inode cluster is free, it becomes eligible for removal
1098 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1099 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1102 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1105 * Remove the inode cluster from the AGI B+Tree, adjust the
1106 * AGI and Superblock inode counts, and mark the disk space
1107 * to be freed when the transaction is committed.
1109 ilen = XFS_IALLOC_INODES(mp);
1110 be32_add_cpu(&agi->agi_count, -ilen);
1111 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1112 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1113 down_read(&mp->m_peraglock);
1114 mp->m_perag[agno].pagi_freecount -= ilen - 1;
1115 up_read(&mp->m_peraglock);
1116 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1117 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1119 if ((error = xfs_btree_delete(cur, &i))) {
1120 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1121 error, mp->m_fsname);
1125 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1126 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1127 XFS_IALLOC_BLOCKS(mp), flist, mp);
1131 error = xfs_inobt_update(cur, &rec);
1134 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1135 error, mp->m_fsname);
1140 * Change the inode free counts and log the ag/sb changes.
1142 be32_add_cpu(&agi->agi_freecount, 1);
1143 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1144 down_read(&mp->m_peraglock);
1145 mp->m_perag[agno].pagi_freecount++;
1146 up_read(&mp->m_peraglock);
1147 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1150 error = xfs_check_agi_freecount(cur, agi);
1154 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1158 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1163 * Return the location of the inode in imap, for mapping it into a buffer.
1167 xfs_mount_t *mp, /* file system mount structure */
1168 xfs_trans_t *tp, /* transaction pointer */
1169 xfs_ino_t ino, /* inode to locate */
1170 struct xfs_imap *imap, /* location map structure */
1171 uint flags) /* flags for inode btree lookup */
1173 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1174 xfs_agino_t agino; /* inode number within alloc group */
1175 xfs_agnumber_t agno; /* allocation group number */
1176 int blks_per_cluster; /* num blocks per inode cluster */
1177 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1178 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1179 int error; /* error code */
1180 int offset; /* index of inode in its buffer */
1181 int offset_agbno; /* blks from chunk start to inode */
1183 ASSERT(ino != NULLFSINO);
1186 * Split up the inode number into its parts.
1188 agno = XFS_INO_TO_AGNO(mp, ino);
1189 agino = XFS_INO_TO_AGINO(mp, ino);
1190 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1191 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1192 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1194 /* no diagnostics for bulkstat, ino comes from userspace */
1195 if (flags & XFS_IGET_BULKSTAT)
1196 return XFS_ERROR(EINVAL);
1197 if (agno >= mp->m_sb.sb_agcount) {
1198 xfs_fs_cmn_err(CE_ALERT, mp,
1199 "xfs_imap: agno (%d) >= "
1200 "mp->m_sb.sb_agcount (%d)",
1201 agno, mp->m_sb.sb_agcount);
1203 if (agbno >= mp->m_sb.sb_agblocks) {
1204 xfs_fs_cmn_err(CE_ALERT, mp,
1205 "xfs_imap: agbno (0x%llx) >= "
1206 "mp->m_sb.sb_agblocks (0x%lx)",
1207 (unsigned long long) agbno,
1208 (unsigned long) mp->m_sb.sb_agblocks);
1210 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1211 xfs_fs_cmn_err(CE_ALERT, mp,
1212 "xfs_imap: ino (0x%llx) != "
1213 "XFS_AGINO_TO_INO(mp, agno, agino) "
1215 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1219 return XFS_ERROR(EINVAL);
1223 * If the inode cluster size is the same as the blocksize or
1224 * smaller we get to the buffer by simple arithmetics.
1226 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1227 offset = XFS_INO_TO_OFFSET(mp, ino);
1228 ASSERT(offset < mp->m_sb.sb_inopblock);
1230 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1231 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1232 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1236 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1239 * If we get a block number passed from bulkstat we can use it to
1240 * find the buffer easily.
1242 if (imap->im_blkno) {
1243 offset = XFS_INO_TO_OFFSET(mp, ino);
1244 ASSERT(offset < mp->m_sb.sb_inopblock);
1246 cluster_agbno = xfs_daddr_to_agbno(mp, imap->im_blkno);
1247 offset += (agbno - cluster_agbno) * mp->m_sb.sb_inopblock;
1249 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1250 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1255 * If the inode chunks are aligned then use simple maths to
1256 * find the location. Otherwise we have to do a btree
1257 * lookup to find the location.
1259 if (mp->m_inoalign_mask) {
1260 offset_agbno = agbno & mp->m_inoalign_mask;
1261 chunk_agbno = agbno - offset_agbno;
1263 xfs_btree_cur_t *cur; /* inode btree cursor */
1264 xfs_inobt_rec_incore_t chunk_rec;
1265 xfs_buf_t *agbp; /* agi buffer */
1266 int i; /* temp state */
1268 down_read(&mp->m_peraglock);
1269 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1270 up_read(&mp->m_peraglock);
1272 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1273 "xfs_ialloc_read_agi() returned "
1274 "error %d, agno %d",
1279 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1280 error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i);
1282 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1283 "xfs_inobt_lookup_le() failed");
1287 error = xfs_inobt_get_rec(cur, &chunk_rec, &i);
1289 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1290 "xfs_inobt_get_rec() failed");
1295 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1296 "xfs_inobt_get_rec() failed");
1298 error = XFS_ERROR(EINVAL);
1301 xfs_trans_brelse(tp, agbp);
1302 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1305 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_rec.ir_startino);
1306 offset_agbno = agbno - chunk_agbno;
1309 ASSERT(agbno >= chunk_agbno);
1310 cluster_agbno = chunk_agbno +
1311 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1312 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1313 XFS_INO_TO_OFFSET(mp, ino);
1315 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1316 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1317 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1320 * If the inode number maps to a block outside the bounds
1321 * of the file system then return NULL rather than calling
1322 * read_buf and panicing when we get an error from the
1325 if ((imap->im_blkno + imap->im_len) >
1326 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1327 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1328 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1329 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1330 (unsigned long long) imap->im_blkno,
1331 (unsigned long long) imap->im_len,
1332 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1333 return XFS_ERROR(EINVAL);
1340 * Compute and fill in value of m_in_maxlevels.
1343 xfs_ialloc_compute_maxlevels(
1344 xfs_mount_t *mp) /* file system mount structure */
1352 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1353 XFS_INODES_PER_CHUNK_LOG;
1354 minleafrecs = mp->m_alloc_mnr[0];
1355 minnoderecs = mp->m_alloc_mnr[1];
1356 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1357 for (level = 1; maxblocks > 1; level++)
1358 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1359 mp->m_in_maxlevels = level;
1363 * Log specified fields for the ag hdr (inode section)
1367 xfs_trans_t *tp, /* transaction pointer */
1368 xfs_buf_t *bp, /* allocation group header buffer */
1369 int fields) /* bitmask of fields to log */
1371 int first; /* first byte number */
1372 int last; /* last byte number */
1373 static const short offsets[] = { /* field starting offsets */
1374 /* keep in sync with bit definitions */
1375 offsetof(xfs_agi_t, agi_magicnum),
1376 offsetof(xfs_agi_t, agi_versionnum),
1377 offsetof(xfs_agi_t, agi_seqno),
1378 offsetof(xfs_agi_t, agi_length),
1379 offsetof(xfs_agi_t, agi_count),
1380 offsetof(xfs_agi_t, agi_root),
1381 offsetof(xfs_agi_t, agi_level),
1382 offsetof(xfs_agi_t, agi_freecount),
1383 offsetof(xfs_agi_t, agi_newino),
1384 offsetof(xfs_agi_t, agi_dirino),
1385 offsetof(xfs_agi_t, agi_unlinked),
1389 xfs_agi_t *agi; /* allocation group header */
1391 agi = XFS_BUF_TO_AGI(bp);
1392 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1395 * Compute byte offsets for the first and last fields.
1397 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1399 * Log the allocation group inode header buffer.
1401 xfs_trans_log_buf(tp, bp, first, last);
1406 xfs_check_agi_unlinked(
1407 struct xfs_agi *agi)
1411 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1412 ASSERT(agi->agi_unlinked[i]);
1415 #define xfs_check_agi_unlinked(agi)
1419 * Read in the allocation group header (inode allocation section)
1423 struct xfs_mount *mp, /* file system mount structure */
1424 struct xfs_trans *tp, /* transaction pointer */
1425 xfs_agnumber_t agno, /* allocation group number */
1426 struct xfs_buf **bpp) /* allocation group hdr buf */
1428 struct xfs_agi *agi; /* allocation group header */
1429 int agi_ok; /* agi is consistent */
1432 ASSERT(agno != NULLAGNUMBER);
1434 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1435 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1436 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1440 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1441 agi = XFS_BUF_TO_AGI(*bpp);
1444 * Validate the magic number of the agi block.
1446 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1447 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1448 be32_to_cpu(agi->agi_seqno) == agno;
1449 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1450 XFS_RANDOM_IALLOC_READ_AGI))) {
1451 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1453 xfs_trans_brelse(tp, *bpp);
1454 return XFS_ERROR(EFSCORRUPTED);
1457 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1459 xfs_check_agi_unlinked(agi);
1464 xfs_ialloc_read_agi(
1465 struct xfs_mount *mp, /* file system mount structure */
1466 struct xfs_trans *tp, /* transaction pointer */
1467 xfs_agnumber_t agno, /* allocation group number */
1468 struct xfs_buf **bpp) /* allocation group hdr buf */
1470 struct xfs_agi *agi; /* allocation group header */
1471 struct xfs_perag *pag; /* per allocation group data */
1474 error = xfs_read_agi(mp, tp, agno, bpp);
1478 agi = XFS_BUF_TO_AGI(*bpp);
1479 pag = &mp->m_perag[agno];
1481 if (!pag->pagi_init) {
1482 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1483 pag->pagi_count = be32_to_cpu(agi->agi_count);
1488 * It's possible for these to be out of sync if
1489 * we are in the middle of a forced shutdown.
1491 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1492 XFS_FORCED_SHUTDOWN(mp));
1497 * Read in the agi to initialise the per-ag data in the mount structure
1500 xfs_ialloc_pagi_init(
1501 xfs_mount_t *mp, /* file system mount structure */
1502 xfs_trans_t *tp, /* transaction pointer */
1503 xfs_agnumber_t agno) /* allocation group number */
1505 xfs_buf_t *bp = NULL;
1508 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1512 xfs_trans_brelse(tp, bp);