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"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_ialloc.h"
35 #include "xfs_alloc.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
42 * Allocation group level functions.
45 xfs_ialloc_cluster_alignment(
46 xfs_alloc_arg_t *args)
48 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
49 args->mp->m_sb.sb_inoalignmt >=
50 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
51 return args->mp->m_sb.sb_inoalignmt;
56 * Lookup a record by ino in the btree given by cur.
60 struct xfs_btree_cur *cur, /* btree cursor */
61 xfs_agino_t ino, /* starting inode of chunk */
62 xfs_lookup_t dir, /* <=, >=, == */
63 int *stat) /* success/failure */
65 cur->bc_rec.i.ir_startino = ino;
66 cur->bc_rec.i.ir_freecount = 0;
67 cur->bc_rec.i.ir_free = 0;
68 return xfs_btree_lookup(cur, dir, stat);
72 * Update the record referred to by cur to the value given.
73 * This either works (return 0) or gets an EFSCORRUPTED error.
75 STATIC int /* error */
77 struct xfs_btree_cur *cur, /* btree cursor */
78 xfs_inobt_rec_incore_t *irec) /* btree record */
80 union xfs_btree_rec rec;
82 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
83 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
84 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
85 return xfs_btree_update(cur, &rec);
89 * Get the data from the pointed-to record.
93 struct xfs_btree_cur *cur, /* btree cursor */
94 xfs_inobt_rec_incore_t *irec, /* btree record */
95 int *stat) /* output: success/failure */
97 union xfs_btree_rec *rec;
100 error = xfs_btree_get_rec(cur, &rec, stat);
101 if (!error && *stat == 1) {
102 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
103 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
104 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
110 * Verify that the number of free inodes in the AGI is correct.
114 xfs_check_agi_freecount(
115 struct xfs_btree_cur *cur,
118 if (cur->bc_nlevels == 1) {
119 xfs_inobt_rec_incore_t rec;
124 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
129 error = xfs_inobt_get_rec(cur, &rec, &i);
134 freecount += rec.ir_freecount;
135 error = xfs_btree_increment(cur, 0, &i);
141 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
142 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
147 #define xfs_check_agi_freecount(cur, agi) 0
151 * Initialise a new set of inodes.
154 xfs_ialloc_inode_init(
155 struct xfs_mount *mp,
156 struct xfs_trans *tp,
159 xfs_agblock_t length,
162 struct xfs_buf *fbuf;
163 struct xfs_dinode *free;
164 int blks_per_cluster, nbufs, ninodes;
170 * Loop over the new block(s), filling in the inodes.
171 * For small block sizes, manipulate the inodes in buffers
172 * which are multiples of the blocks size.
174 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
175 blks_per_cluster = 1;
177 ninodes = mp->m_sb.sb_inopblock;
179 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
180 mp->m_sb.sb_blocksize;
181 nbufs = length / blks_per_cluster;
182 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
186 * Figure out what version number to use in the inodes we create.
187 * If the superblock version has caught up to the one that supports
188 * the new inode format, then use the new inode version. Otherwise
189 * use the old version so that old kernels will continue to be
190 * able to use the file system.
192 if (xfs_sb_version_hasnlink(&mp->m_sb))
197 for (j = 0; j < nbufs; j++) {
201 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
202 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
203 mp->m_bsize * blks_per_cluster, 0);
207 * Initialize all inodes in this buffer and then log them.
209 * XXX: It would be much better if we had just one transaction
210 * to log a whole cluster of inodes instead of all the
211 * individual transactions causing a lot of log traffic.
213 xfs_buf_zero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
214 for (i = 0; i < ninodes; i++) {
215 int ioffset = i << mp->m_sb.sb_inodelog;
216 uint isize = sizeof(struct xfs_dinode);
218 free = xfs_make_iptr(mp, fbuf, i);
219 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
220 free->di_version = version;
221 free->di_gen = cpu_to_be32(gen);
222 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
223 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
225 xfs_trans_inode_alloc_buf(tp, fbuf);
231 * Allocate new inodes in the allocation group specified by agbp.
232 * Return 0 for success, else error code.
234 STATIC int /* error code or 0 */
236 xfs_trans_t *tp, /* transaction pointer */
237 xfs_buf_t *agbp, /* alloc group buffer */
240 xfs_agi_t *agi; /* allocation group header */
241 xfs_alloc_arg_t args; /* allocation argument structure */
242 xfs_btree_cur_t *cur; /* inode btree cursor */
246 xfs_agino_t newino; /* new first inode's number */
247 xfs_agino_t newlen; /* new number of inodes */
248 xfs_agino_t thisino; /* current inode number, for loop */
249 int isaligned = 0; /* inode allocation at stripe unit */
251 struct xfs_perag *pag;
254 args.mp = tp->t_mountp;
257 * Locking will ensure that we don't have two callers in here
260 newlen = XFS_IALLOC_INODES(args.mp);
261 if (args.mp->m_maxicount &&
262 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
263 return XFS_ERROR(ENOSPC);
264 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
266 * First try to allocate inodes contiguous with the last-allocated
267 * chunk of inodes. If the filesystem is striped, this will fill
268 * an entire stripe unit with inodes.
270 agi = XFS_BUF_TO_AGI(agbp);
271 newino = be32_to_cpu(agi->agi_newino);
272 agno = be32_to_cpu(agi->agi_seqno);
273 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
274 XFS_IALLOC_BLOCKS(args.mp);
275 if (likely(newino != NULLAGINO &&
276 (args.agbno < be32_to_cpu(agi->agi_length)))) {
277 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
278 args.type = XFS_ALLOCTYPE_THIS_BNO;
279 args.mod = args.total = args.wasdel = args.isfl =
280 args.userdata = args.minalignslop = 0;
284 * We need to take into account alignment here to ensure that
285 * we don't modify the free list if we fail to have an exact
286 * block. If we don't have an exact match, and every oher
287 * attempt allocation attempt fails, we'll end up cancelling
288 * a dirty transaction and shutting down.
290 * For an exact allocation, alignment must be 1,
291 * however we need to take cluster alignment into account when
292 * fixing up the freelist. Use the minalignslop field to
293 * indicate that extra blocks might be required for alignment,
294 * but not to use them in the actual exact allocation.
297 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
299 /* Allow space for the inode btree to split. */
300 args.minleft = args.mp->m_in_maxlevels - 1;
301 if ((error = xfs_alloc_vextent(&args)))
304 args.fsbno = NULLFSBLOCK;
306 if (unlikely(args.fsbno == NULLFSBLOCK)) {
308 * Set the alignment for the allocation.
309 * If stripe alignment is turned on then align at stripe unit
311 * If the cluster size is smaller than a filesystem block
312 * then we're doing I/O for inodes in filesystem block size
313 * pieces, so don't need alignment anyway.
316 if (args.mp->m_sinoalign) {
317 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
318 args.alignment = args.mp->m_dalign;
321 args.alignment = xfs_ialloc_cluster_alignment(&args);
323 * Need to figure out where to allocate the inode blocks.
324 * Ideally they should be spaced out through the a.g.
325 * For now, just allocate blocks up front.
327 args.agbno = be32_to_cpu(agi->agi_root);
328 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
330 * Allocate a fixed-size extent of inodes.
332 args.type = XFS_ALLOCTYPE_NEAR_BNO;
333 args.mod = args.total = args.wasdel = args.isfl =
334 args.userdata = args.minalignslop = 0;
337 * Allow space for the inode btree to split.
339 args.minleft = args.mp->m_in_maxlevels - 1;
340 if ((error = xfs_alloc_vextent(&args)))
345 * If stripe alignment is turned on, then try again with cluster
348 if (isaligned && args.fsbno == NULLFSBLOCK) {
349 args.type = XFS_ALLOCTYPE_NEAR_BNO;
350 args.agbno = be32_to_cpu(agi->agi_root);
351 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
352 args.alignment = xfs_ialloc_cluster_alignment(&args);
353 if ((error = xfs_alloc_vextent(&args)))
357 if (args.fsbno == NULLFSBLOCK) {
361 ASSERT(args.len == args.minlen);
364 * Stamp and write the inode buffers.
366 * Seed the new inode cluster with a random generation number. This
367 * prevents short-term reuse of generation numbers if a chunk is
368 * freed and then immediately reallocated. We use random numbers
369 * rather than a linear progression to prevent the next generation
370 * number from being easily guessable.
372 error = xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno,
373 args.len, random32());
378 * Convert the results.
380 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
381 be32_add_cpu(&agi->agi_count, newlen);
382 be32_add_cpu(&agi->agi_freecount, newlen);
383 pag = xfs_perag_get(args.mp, agno);
384 pag->pagi_freecount += newlen;
386 agi->agi_newino = cpu_to_be32(newino);
389 * Insert records describing the new inode chunk into the btree.
391 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
392 for (thisino = newino;
393 thisino < newino + newlen;
394 thisino += XFS_INODES_PER_CHUNK) {
395 cur->bc_rec.i.ir_startino = thisino;
396 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
397 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
398 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
400 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
404 error = xfs_btree_insert(cur, &i);
406 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
411 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
413 * Log allocation group header fields
415 xfs_ialloc_log_agi(tp, agbp,
416 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
418 * Modify/log superblock values for inode count and inode free count.
420 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
421 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
426 STATIC xfs_agnumber_t
432 spin_lock(&mp->m_agirotor_lock);
433 agno = mp->m_agirotor;
434 if (++mp->m_agirotor == mp->m_maxagi)
436 spin_unlock(&mp->m_agirotor_lock);
442 * Select an allocation group to look for a free inode in, based on the parent
443 * inode and then mode. Return the allocation group buffer.
445 STATIC xfs_buf_t * /* allocation group buffer */
446 xfs_ialloc_ag_select(
447 xfs_trans_t *tp, /* transaction pointer */
448 xfs_ino_t parent, /* parent directory inode number */
449 umode_t mode, /* bits set to indicate file type */
450 int okalloc) /* ok to allocate more space */
452 xfs_buf_t *agbp; /* allocation group header buffer */
453 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
454 xfs_agnumber_t agno; /* current ag number */
455 int flags; /* alloc buffer locking flags */
456 xfs_extlen_t ineed; /* blocks needed for inode allocation */
457 xfs_extlen_t longest = 0; /* longest extent available */
458 xfs_mount_t *mp; /* mount point structure */
459 int needspace; /* file mode implies space allocated */
460 xfs_perag_t *pag; /* per allocation group data */
461 xfs_agnumber_t pagno; /* parent (starting) ag number */
464 * Files of these types need at least one block if length > 0
465 * (and they won't fit in the inode, but that's hard to figure out).
467 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
469 agcount = mp->m_maxagi;
471 pagno = xfs_ialloc_next_ag(mp);
473 pagno = XFS_INO_TO_AGNO(mp, parent);
474 if (pagno >= agcount)
477 ASSERT(pagno < agcount);
479 * Loop through allocation groups, looking for one with a little
480 * free space in it. Note we don't look for free inodes, exactly.
481 * Instead, we include whether there is a need to allocate inodes
482 * to mean that blocks must be allocated for them,
483 * if none are currently free.
486 flags = XFS_ALLOC_FLAG_TRYLOCK;
488 pag = xfs_perag_get(mp, agno);
489 if (!pag->pagi_init) {
490 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
497 if (!pag->pagi_inodeok) {
498 xfs_ialloc_next_ag(mp);
503 * Is there enough free space for the file plus a block
504 * of inodes (if we need to allocate some)?
506 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
507 if (ineed && !pag->pagf_init) {
509 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
513 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
515 if (!ineed || pag->pagf_init) {
516 if (ineed && !(longest = pag->pagf_longest))
517 longest = pag->pagf_flcount > 0;
519 (pag->pagf_freeblks >= needspace + ineed &&
523 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
533 xfs_trans_brelse(tp, agbp);
537 * No point in iterating over the rest, if we're shutting
540 if (XFS_FORCED_SHUTDOWN(mp))
554 * Try to retrieve the next record to the left/right from the current one.
558 struct xfs_btree_cur *cur,
559 xfs_inobt_rec_incore_t *rec,
567 error = xfs_btree_decrement(cur, 0, &i);
569 error = xfs_btree_increment(cur, 0, &i);
575 error = xfs_inobt_get_rec(cur, rec, &i);
578 XFS_WANT_CORRUPTED_RETURN(i == 1);
586 struct xfs_btree_cur *cur,
588 xfs_inobt_rec_incore_t *rec,
595 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
600 error = xfs_inobt_get_rec(cur, rec, &i);
603 XFS_WANT_CORRUPTED_RETURN(i == 1);
610 * Visible inode allocation functions.
614 * Allocate an inode on disk.
615 * Mode is used to tell whether the new inode will need space, and whether
618 * The arguments IO_agbp and alloc_done are defined to work within
619 * the constraint of one allocation per transaction.
620 * xfs_dialloc() is designed to be called twice if it has to do an
621 * allocation to make more free inodes. On the first call,
622 * IO_agbp should be set to NULL. If an inode is available,
623 * i.e., xfs_dialloc() did not need to do an allocation, an inode
624 * number is returned. In this case, IO_agbp would be set to the
625 * current ag_buf and alloc_done set to false.
626 * If an allocation needed to be done, xfs_dialloc would return
627 * the current ag_buf in IO_agbp and set alloc_done to true.
628 * The caller should then commit the current transaction, allocate a new
629 * transaction, and call xfs_dialloc() again, passing in the previous
630 * value of IO_agbp. IO_agbp should be held across the transactions.
631 * Since the agbp is locked across the two calls, the second call is
632 * guaranteed to have a free inode available.
634 * Once we successfully pick an inode its number is returned and the
635 * on-disk data structures are updated. The inode itself is not read
636 * in, since doing so would break ordering constraints with xfs_reclaim.
640 xfs_trans_t *tp, /* transaction pointer */
641 xfs_ino_t parent, /* parent inode (directory) */
642 umode_t mode, /* mode bits for new inode */
643 int okalloc, /* ok to allocate more space */
644 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
645 boolean_t *alloc_done, /* true if we needed to replenish
647 xfs_ino_t *inop) /* inode number allocated */
649 xfs_agnumber_t agcount; /* number of allocation groups */
650 xfs_buf_t *agbp; /* allocation group header's buffer */
651 xfs_agnumber_t agno; /* allocation group number */
652 xfs_agi_t *agi; /* allocation group header structure */
653 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
654 int error; /* error return value */
655 int i; /* result code */
656 int ialloced; /* inode allocation status */
657 int noroom = 0; /* no space for inode blk allocation */
658 xfs_ino_t ino; /* fs-relative inode to be returned */
660 int j; /* result code */
661 xfs_mount_t *mp; /* file system mount structure */
662 int offset; /* index of inode in chunk */
663 xfs_agino_t pagino; /* parent's AG relative inode # */
664 xfs_agnumber_t pagno; /* parent's AG number */
665 xfs_inobt_rec_incore_t rec; /* inode allocation record */
666 xfs_agnumber_t tagno; /* testing allocation group number */
667 xfs_btree_cur_t *tcur; /* temp cursor */
668 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
669 struct xfs_perag *pag;
672 if (*IO_agbp == NULL) {
674 * We do not have an agbp, so select an initial allocation
675 * group for inode allocation.
677 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
679 * Couldn't find an allocation group satisfying the
686 agi = XFS_BUF_TO_AGI(agbp);
687 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
690 * Continue where we left off before. In this case, we
691 * know that the allocation group has free inodes.
694 agi = XFS_BUF_TO_AGI(agbp);
695 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
696 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
699 agcount = mp->m_sb.sb_agcount;
700 agno = be32_to_cpu(agi->agi_seqno);
702 pagno = XFS_INO_TO_AGNO(mp, parent);
703 pagino = XFS_INO_TO_AGINO(mp, parent);
706 * If we have already hit the ceiling of inode blocks then clear
707 * okalloc so we scan all available agi structures for a free
711 if (mp->m_maxicount &&
712 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
718 * Loop until we find an allocation group that either has free inodes
719 * or in which we can allocate some inodes. Iterate through the
720 * allocation groups upward, wrapping at the end.
722 *alloc_done = B_FALSE;
723 while (!agi->agi_freecount) {
725 * Don't do anything if we're not supposed to allocate
726 * any blocks, just go on to the next ag.
730 * Try to allocate some new inodes in the allocation
733 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
734 xfs_trans_brelse(tp, agbp);
735 if (error == ENOSPC) {
743 * We successfully allocated some inodes, return
744 * the current context to the caller so that it
745 * can commit the current transaction and call
746 * us again where we left off.
748 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
749 *alloc_done = B_TRUE;
756 * If it failed, give up on this ag.
758 xfs_trans_brelse(tp, agbp);
760 * Go on to the next ag: get its ag header.
763 if (++tagno == agcount)
767 return noroom ? ENOSPC : 0;
769 pag = xfs_perag_get(mp, tagno);
770 if (pag->pagi_inodeok == 0) {
774 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
778 agi = XFS_BUF_TO_AGI(agbp);
779 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
782 * Here with an allocation group that has a free inode.
783 * Reset agno since we may have chosen a new ag in the
788 pag = xfs_perag_get(mp, agno);
791 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
793 * If pagino is 0 (this is the root inode allocation) use newino.
794 * This must work because we've just allocated some.
797 pagino = be32_to_cpu(agi->agi_newino);
799 error = xfs_check_agi_freecount(cur, agi);
804 * If in the same AG as the parent, try to get near the parent.
807 int doneleft; /* done, to the left */
808 int doneright; /* done, to the right */
809 int searchdistance = 10;
811 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
814 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
816 error = xfs_inobt_get_rec(cur, &rec, &j);
819 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
821 if (rec.ir_freecount > 0) {
823 * Found a free inode in the same chunk
824 * as the parent, done.
831 * In the same AG as parent, but parent's chunk is full.
834 /* duplicate the cursor, search left & right simultaneously */
835 error = xfs_btree_dup_cursor(cur, &tcur);
840 * Skip to last blocks looked up if same parent inode.
842 if (pagino != NULLAGINO &&
843 pag->pagl_pagino == pagino &&
844 pag->pagl_leftrec != NULLAGINO &&
845 pag->pagl_rightrec != NULLAGINO) {
846 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
847 &trec, &doneleft, 1);
851 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
852 &rec, &doneright, 0);
856 /* search left with tcur, back up 1 record */
857 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
861 /* search right with cur, go forward 1 record. */
862 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
868 * Loop until we find an inode chunk with a free inode.
870 while (!doneleft || !doneright) {
871 int useleft; /* using left inode chunk this time */
873 if (!--searchdistance) {
875 * Not in range - save last search
876 * location and allocate a new inode
878 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
879 pag->pagl_leftrec = trec.ir_startino;
880 pag->pagl_rightrec = rec.ir_startino;
881 pag->pagl_pagino = pagino;
885 /* figure out the closer block if both are valid. */
886 if (!doneleft && !doneright) {
888 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
889 rec.ir_startino - pagino;
894 /* free inodes to the left? */
895 if (useleft && trec.ir_freecount) {
897 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
900 pag->pagl_leftrec = trec.ir_startino;
901 pag->pagl_rightrec = rec.ir_startino;
902 pag->pagl_pagino = pagino;
906 /* free inodes to the right? */
907 if (!useleft && rec.ir_freecount) {
908 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
910 pag->pagl_leftrec = trec.ir_startino;
911 pag->pagl_rightrec = rec.ir_startino;
912 pag->pagl_pagino = pagino;
916 /* get next record to check */
918 error = xfs_ialloc_next_rec(tcur, &trec,
921 error = xfs_ialloc_next_rec(cur, &rec,
929 * We've reached the end of the btree. because
930 * we are only searching a small chunk of the
931 * btree each search, there is obviously free
932 * inodes closer to the parent inode than we
933 * are now. restart the search again.
935 pag->pagl_pagino = NULLAGINO;
936 pag->pagl_leftrec = NULLAGINO;
937 pag->pagl_rightrec = NULLAGINO;
938 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
939 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
944 * In a different AG from the parent.
945 * See if the most recently allocated block has any free.
948 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
949 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
955 error = xfs_inobt_get_rec(cur, &rec, &j);
959 if (j == 1 && rec.ir_freecount > 0) {
961 * The last chunk allocated in the group
962 * still has a free inode.
970 * None left in the last group, search the whole AG
972 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
975 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
978 error = xfs_inobt_get_rec(cur, &rec, &i);
981 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
982 if (rec.ir_freecount > 0)
984 error = xfs_btree_increment(cur, 0, &i);
987 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
991 offset = xfs_ialloc_find_free(&rec.ir_free);
993 ASSERT(offset < XFS_INODES_PER_CHUNK);
994 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
995 XFS_INODES_PER_CHUNK) == 0);
996 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
997 rec.ir_free &= ~XFS_INOBT_MASK(offset);
999 error = xfs_inobt_update(cur, &rec);
1002 be32_add_cpu(&agi->agi_freecount, -1);
1003 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1004 pag->pagi_freecount--;
1006 error = xfs_check_agi_freecount(cur, agi);
1010 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1011 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1016 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1018 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1024 * Free disk inode. Carefully avoids touching the incore inode, all
1025 * manipulations incore are the caller's responsibility.
1026 * The on-disk inode is not changed by this operation, only the
1027 * btree (free inode mask) is changed.
1031 xfs_trans_t *tp, /* transaction pointer */
1032 xfs_ino_t inode, /* inode to be freed */
1033 xfs_bmap_free_t *flist, /* extents to free */
1034 int *delete, /* set if inode cluster was deleted */
1035 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1038 xfs_agblock_t agbno; /* block number containing inode */
1039 xfs_buf_t *agbp; /* buffer containing allocation group header */
1040 xfs_agino_t agino; /* inode number relative to allocation group */
1041 xfs_agnumber_t agno; /* allocation group number */
1042 xfs_agi_t *agi; /* allocation group header */
1043 xfs_btree_cur_t *cur; /* inode btree cursor */
1044 int error; /* error return value */
1045 int i; /* result code */
1046 int ilen; /* inodes in an inode cluster */
1047 xfs_mount_t *mp; /* mount structure for filesystem */
1048 int off; /* offset of inode in inode chunk */
1049 xfs_inobt_rec_incore_t rec; /* btree record */
1050 struct xfs_perag *pag;
1055 * Break up inode number into its components.
1057 agno = XFS_INO_TO_AGNO(mp, inode);
1058 if (agno >= mp->m_sb.sb_agcount) {
1059 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1060 __func__, agno, mp->m_sb.sb_agcount);
1062 return XFS_ERROR(EINVAL);
1064 agino = XFS_INO_TO_AGINO(mp, inode);
1065 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1066 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1067 __func__, (unsigned long long)inode,
1068 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1070 return XFS_ERROR(EINVAL);
1072 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1073 if (agbno >= mp->m_sb.sb_agblocks) {
1074 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1075 __func__, agbno, mp->m_sb.sb_agblocks);
1077 return XFS_ERROR(EINVAL);
1080 * Get the allocation group header.
1082 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1084 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1088 agi = XFS_BUF_TO_AGI(agbp);
1089 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1090 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1092 * Initialize the cursor.
1094 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1096 error = xfs_check_agi_freecount(cur, agi);
1101 * Look for the entry describing this inode.
1103 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1104 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1108 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1109 error = xfs_inobt_get_rec(cur, &rec, &i);
1111 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1115 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1117 * Get the offset in the inode chunk.
1119 off = agino - rec.ir_startino;
1120 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1121 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1123 * Mark the inode free & increment the count.
1125 rec.ir_free |= XFS_INOBT_MASK(off);
1129 * When an inode cluster is free, it becomes eligible for removal
1131 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1132 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1135 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1138 * Remove the inode cluster from the AGI B+Tree, adjust the
1139 * AGI and Superblock inode counts, and mark the disk space
1140 * to be freed when the transaction is committed.
1142 ilen = XFS_IALLOC_INODES(mp);
1143 be32_add_cpu(&agi->agi_count, -ilen);
1144 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1145 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1146 pag = xfs_perag_get(mp, agno);
1147 pag->pagi_freecount -= ilen - 1;
1149 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1150 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1152 if ((error = xfs_btree_delete(cur, &i))) {
1153 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1158 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1159 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1160 XFS_IALLOC_BLOCKS(mp), flist, mp);
1164 error = xfs_inobt_update(cur, &rec);
1166 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1172 * Change the inode free counts and log the ag/sb changes.
1174 be32_add_cpu(&agi->agi_freecount, 1);
1175 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1176 pag = xfs_perag_get(mp, agno);
1177 pag->pagi_freecount++;
1179 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1182 error = xfs_check_agi_freecount(cur, agi);
1186 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1190 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1196 struct xfs_mount *mp,
1197 struct xfs_trans *tp,
1198 xfs_agnumber_t agno,
1200 xfs_agblock_t agbno,
1201 xfs_agblock_t *chunk_agbno,
1202 xfs_agblock_t *offset_agbno,
1205 struct xfs_inobt_rec_incore rec;
1206 struct xfs_btree_cur *cur;
1207 struct xfs_buf *agbp;
1211 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1214 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1215 __func__, error, agno);
1220 * Lookup the inode record for the given agino. If the record cannot be
1221 * found, then it's an invalid inode number and we should abort. Once
1222 * we have a record, we need to ensure it contains the inode number
1223 * we are looking up.
1225 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1226 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1229 error = xfs_inobt_get_rec(cur, &rec, &i);
1230 if (!error && i == 0)
1234 xfs_trans_brelse(tp, agbp);
1235 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1239 /* check that the returned record contains the required inode */
1240 if (rec.ir_startino > agino ||
1241 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1244 /* for untrusted inodes check it is allocated first */
1245 if ((flags & XFS_IGET_UNTRUSTED) &&
1246 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1249 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1250 *offset_agbno = agbno - *chunk_agbno;
1255 * Return the location of the inode in imap, for mapping it into a buffer.
1259 xfs_mount_t *mp, /* file system mount structure */
1260 xfs_trans_t *tp, /* transaction pointer */
1261 xfs_ino_t ino, /* inode to locate */
1262 struct xfs_imap *imap, /* location map structure */
1263 uint flags) /* flags for inode btree lookup */
1265 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1266 xfs_agino_t agino; /* inode number within alloc group */
1267 xfs_agnumber_t agno; /* allocation group number */
1268 int blks_per_cluster; /* num blocks per inode cluster */
1269 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1270 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1271 int error; /* error code */
1272 int offset; /* index of inode in its buffer */
1273 int offset_agbno; /* blks from chunk start to inode */
1275 ASSERT(ino != NULLFSINO);
1278 * Split up the inode number into its parts.
1280 agno = XFS_INO_TO_AGNO(mp, ino);
1281 agino = XFS_INO_TO_AGINO(mp, ino);
1282 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1283 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1284 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1287 * Don't output diagnostic information for untrusted inodes
1288 * as they can be invalid without implying corruption.
1290 if (flags & XFS_IGET_UNTRUSTED)
1291 return XFS_ERROR(EINVAL);
1292 if (agno >= mp->m_sb.sb_agcount) {
1294 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1295 __func__, agno, mp->m_sb.sb_agcount);
1297 if (agbno >= mp->m_sb.sb_agblocks) {
1299 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1300 __func__, (unsigned long long)agbno,
1301 (unsigned long)mp->m_sb.sb_agblocks);
1303 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1305 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1307 XFS_AGINO_TO_INO(mp, agno, agino));
1311 return XFS_ERROR(EINVAL);
1314 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1317 * For bulkstat and handle lookups, we have an untrusted inode number
1318 * that we have to verify is valid. We cannot do this just by reading
1319 * the inode buffer as it may have been unlinked and removed leaving
1320 * inodes in stale state on disk. Hence we have to do a btree lookup
1321 * in all cases where an untrusted inode number is passed.
1323 if (flags & XFS_IGET_UNTRUSTED) {
1324 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1325 &chunk_agbno, &offset_agbno, flags);
1332 * If the inode cluster size is the same as the blocksize or
1333 * smaller we get to the buffer by simple arithmetics.
1335 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1336 offset = XFS_INO_TO_OFFSET(mp, ino);
1337 ASSERT(offset < mp->m_sb.sb_inopblock);
1339 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1340 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1341 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1346 * If the inode chunks are aligned then use simple maths to
1347 * find the location. Otherwise we have to do a btree
1348 * lookup to find the location.
1350 if (mp->m_inoalign_mask) {
1351 offset_agbno = agbno & mp->m_inoalign_mask;
1352 chunk_agbno = agbno - offset_agbno;
1354 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1355 &chunk_agbno, &offset_agbno, flags);
1361 ASSERT(agbno >= chunk_agbno);
1362 cluster_agbno = chunk_agbno +
1363 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1364 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1365 XFS_INO_TO_OFFSET(mp, ino);
1367 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1368 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1369 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1372 * If the inode number maps to a block outside the bounds
1373 * of the file system then return NULL rather than calling
1374 * read_buf and panicing when we get an error from the
1377 if ((imap->im_blkno + imap->im_len) >
1378 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1380 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1381 __func__, (unsigned long long) imap->im_blkno,
1382 (unsigned long long) imap->im_len,
1383 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1384 return XFS_ERROR(EINVAL);
1390 * Compute and fill in value of m_in_maxlevels.
1393 xfs_ialloc_compute_maxlevels(
1394 xfs_mount_t *mp) /* file system mount structure */
1402 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1403 XFS_INODES_PER_CHUNK_LOG;
1404 minleafrecs = mp->m_alloc_mnr[0];
1405 minnoderecs = mp->m_alloc_mnr[1];
1406 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1407 for (level = 1; maxblocks > 1; level++)
1408 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1409 mp->m_in_maxlevels = level;
1413 * Log specified fields for the ag hdr (inode section)
1417 xfs_trans_t *tp, /* transaction pointer */
1418 xfs_buf_t *bp, /* allocation group header buffer */
1419 int fields) /* bitmask of fields to log */
1421 int first; /* first byte number */
1422 int last; /* last byte number */
1423 static const short offsets[] = { /* field starting offsets */
1424 /* keep in sync with bit definitions */
1425 offsetof(xfs_agi_t, agi_magicnum),
1426 offsetof(xfs_agi_t, agi_versionnum),
1427 offsetof(xfs_agi_t, agi_seqno),
1428 offsetof(xfs_agi_t, agi_length),
1429 offsetof(xfs_agi_t, agi_count),
1430 offsetof(xfs_agi_t, agi_root),
1431 offsetof(xfs_agi_t, agi_level),
1432 offsetof(xfs_agi_t, agi_freecount),
1433 offsetof(xfs_agi_t, agi_newino),
1434 offsetof(xfs_agi_t, agi_dirino),
1435 offsetof(xfs_agi_t, agi_unlinked),
1439 xfs_agi_t *agi; /* allocation group header */
1441 agi = XFS_BUF_TO_AGI(bp);
1442 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1445 * Compute byte offsets for the first and last fields.
1447 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1449 * Log the allocation group inode header buffer.
1451 xfs_trans_log_buf(tp, bp, first, last);
1456 xfs_check_agi_unlinked(
1457 struct xfs_agi *agi)
1461 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1462 ASSERT(agi->agi_unlinked[i]);
1465 #define xfs_check_agi_unlinked(agi)
1469 * Read in the allocation group header (inode allocation section)
1473 struct xfs_mount *mp, /* file system mount structure */
1474 struct xfs_trans *tp, /* transaction pointer */
1475 xfs_agnumber_t agno, /* allocation group number */
1476 struct xfs_buf **bpp) /* allocation group hdr buf */
1478 struct xfs_agi *agi; /* allocation group header */
1479 int agi_ok; /* agi is consistent */
1482 ASSERT(agno != NULLAGNUMBER);
1484 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1485 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1486 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1490 ASSERT(!xfs_buf_geterror(*bpp));
1491 agi = XFS_BUF_TO_AGI(*bpp);
1494 * Validate the magic number of the agi block.
1496 agi_ok = agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC) &&
1497 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1498 be32_to_cpu(agi->agi_seqno) == agno;
1499 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1500 XFS_RANDOM_IALLOC_READ_AGI))) {
1501 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1503 xfs_trans_brelse(tp, *bpp);
1504 return XFS_ERROR(EFSCORRUPTED);
1507 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1509 xfs_check_agi_unlinked(agi);
1514 xfs_ialloc_read_agi(
1515 struct xfs_mount *mp, /* file system mount structure */
1516 struct xfs_trans *tp, /* transaction pointer */
1517 xfs_agnumber_t agno, /* allocation group number */
1518 struct xfs_buf **bpp) /* allocation group hdr buf */
1520 struct xfs_agi *agi; /* allocation group header */
1521 struct xfs_perag *pag; /* per allocation group data */
1524 error = xfs_read_agi(mp, tp, agno, bpp);
1528 agi = XFS_BUF_TO_AGI(*bpp);
1529 pag = xfs_perag_get(mp, agno);
1530 if (!pag->pagi_init) {
1531 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1532 pag->pagi_count = be32_to_cpu(agi->agi_count);
1537 * It's possible for these to be out of sync if
1538 * we are in the middle of a forced shutdown.
1540 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1541 XFS_FORCED_SHUTDOWN(mp));
1547 * Read in the agi to initialise the per-ag data in the mount structure
1550 xfs_ialloc_pagi_init(
1551 xfs_mount_t *mp, /* file system mount structure */
1552 xfs_trans_t *tp, /* transaction pointer */
1553 xfs_agnumber_t agno) /* allocation group number */
1555 xfs_buf_t *bp = NULL;
1558 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1562 xfs_trans_brelse(tp, bp);