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 a record by 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 xfs_lookup_t dir, /* <=, >=, == */
67 int *stat) /* success/failure */
69 cur->bc_rec.i.ir_startino = ino;
70 cur->bc_rec.i.ir_freecount = 0;
71 cur->bc_rec.i.ir_free = 0;
72 return xfs_btree_lookup(cur, dir, stat);
76 * Update the record referred to by cur to the value given.
77 * This either works (return 0) or gets an EFSCORRUPTED error.
79 STATIC int /* error */
81 struct xfs_btree_cur *cur, /* btree cursor */
82 xfs_inobt_rec_incore_t *irec) /* btree record */
84 union xfs_btree_rec rec;
86 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
87 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
88 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
89 return xfs_btree_update(cur, &rec);
93 * Get the data from the pointed-to record.
97 struct xfs_btree_cur *cur, /* btree cursor */
98 xfs_inobt_rec_incore_t *irec, /* btree record */
99 int *stat) /* output: success/failure */
101 union xfs_btree_rec *rec;
104 error = xfs_btree_get_rec(cur, &rec, stat);
105 if (!error && *stat == 1) {
106 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
107 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
108 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
114 * Verify that the number of free inodes in the AGI is correct.
118 xfs_check_agi_freecount(
119 struct xfs_btree_cur *cur,
122 if (cur->bc_nlevels == 1) {
123 xfs_inobt_rec_incore_t rec;
128 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
133 error = xfs_inobt_get_rec(cur, &rec, &i);
138 freecount += rec.ir_freecount;
139 error = xfs_btree_increment(cur, 0, &i);
145 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
146 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
151 #define xfs_check_agi_freecount(cur, agi) 0
155 * Initialise a new set of inodes.
158 xfs_ialloc_inode_init(
159 struct xfs_mount *mp,
160 struct xfs_trans *tp,
163 xfs_agblock_t length,
166 struct xfs_buf *fbuf;
167 struct xfs_dinode *free;
168 int blks_per_cluster, nbufs, ninodes;
174 * Loop over the new block(s), filling in the inodes.
175 * For small block sizes, manipulate the inodes in buffers
176 * which are multiples of the blocks size.
178 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
179 blks_per_cluster = 1;
181 ninodes = mp->m_sb.sb_inopblock;
183 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
184 mp->m_sb.sb_blocksize;
185 nbufs = length / blks_per_cluster;
186 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
190 * Figure out what version number to use in the inodes we create.
191 * If the superblock version has caught up to the one that supports
192 * the new inode format, then use the new inode version. Otherwise
193 * use the old version so that old kernels will continue to be
194 * able to use the file system.
196 if (xfs_sb_version_hasnlink(&mp->m_sb))
201 for (j = 0; j < nbufs; j++) {
205 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
206 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
207 mp->m_bsize * blks_per_cluster,
210 ASSERT(!XFS_BUF_GETERROR(fbuf));
213 * Initialize all inodes in this buffer and then log them.
215 * XXX: It would be much better if we had just one transaction
216 * to log a whole cluster of inodes instead of all the
217 * individual transactions causing a lot of log traffic.
219 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
220 for (i = 0; i < ninodes; i++) {
221 int ioffset = i << mp->m_sb.sb_inodelog;
222 uint isize = sizeof(struct xfs_dinode);
224 free = xfs_make_iptr(mp, fbuf, i);
225 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
226 free->di_version = version;
227 free->di_gen = cpu_to_be32(gen);
228 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
229 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
231 xfs_trans_inode_alloc_buf(tp, fbuf);
236 * Allocate new inodes in the allocation group specified by agbp.
237 * Return 0 for success, else error code.
239 STATIC int /* error code or 0 */
241 xfs_trans_t *tp, /* transaction pointer */
242 xfs_buf_t *agbp, /* alloc group buffer */
245 xfs_agi_t *agi; /* allocation group header */
246 xfs_alloc_arg_t args; /* allocation argument structure */
247 xfs_btree_cur_t *cur; /* inode btree cursor */
251 xfs_agino_t newino; /* new first inode's number */
252 xfs_agino_t newlen; /* new number of inodes */
253 xfs_agino_t thisino; /* current inode number, for loop */
254 int isaligned = 0; /* inode allocation at stripe unit */
258 args.mp = tp->t_mountp;
261 * Locking will ensure that we don't have two callers in here
264 newlen = XFS_IALLOC_INODES(args.mp);
265 if (args.mp->m_maxicount &&
266 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
267 return XFS_ERROR(ENOSPC);
268 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
270 * First try to allocate inodes contiguous with the last-allocated
271 * chunk of inodes. If the filesystem is striped, this will fill
272 * an entire stripe unit with inodes.
274 agi = XFS_BUF_TO_AGI(agbp);
275 newino = be32_to_cpu(agi->agi_newino);
276 agno = be32_to_cpu(agi->agi_seqno);
277 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
278 XFS_IALLOC_BLOCKS(args.mp);
279 if (likely(newino != NULLAGINO &&
280 (args.agbno < be32_to_cpu(agi->agi_length)))) {
281 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
282 args.type = XFS_ALLOCTYPE_THIS_BNO;
283 args.mod = args.total = args.wasdel = args.isfl =
284 args.userdata = args.minalignslop = 0;
288 * We need to take into account alignment here to ensure that
289 * we don't modify the free list if we fail to have an exact
290 * block. If we don't have an exact match, and every oher
291 * attempt allocation attempt fails, we'll end up cancelling
292 * a dirty transaction and shutting down.
294 * For an exact allocation, alignment must be 1,
295 * however we need to take cluster alignment into account when
296 * fixing up the freelist. Use the minalignslop field to
297 * indicate that extra blocks might be required for alignment,
298 * but not to use them in the actual exact allocation.
301 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
303 /* Allow space for the inode btree to split. */
304 args.minleft = args.mp->m_in_maxlevels - 1;
305 if ((error = xfs_alloc_vextent(&args)))
308 args.fsbno = NULLFSBLOCK;
310 if (unlikely(args.fsbno == NULLFSBLOCK)) {
312 * Set the alignment for the allocation.
313 * If stripe alignment is turned on then align at stripe unit
315 * If the cluster size is smaller than a filesystem block
316 * then we're doing I/O for inodes in filesystem block size
317 * pieces, so don't need alignment anyway.
320 if (args.mp->m_sinoalign) {
321 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
322 args.alignment = args.mp->m_dalign;
325 args.alignment = xfs_ialloc_cluster_alignment(&args);
327 * Need to figure out where to allocate the inode blocks.
328 * Ideally they should be spaced out through the a.g.
329 * For now, just allocate blocks up front.
331 args.agbno = be32_to_cpu(agi->agi_root);
332 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
334 * Allocate a fixed-size extent of inodes.
336 args.type = XFS_ALLOCTYPE_NEAR_BNO;
337 args.mod = args.total = args.wasdel = args.isfl =
338 args.userdata = args.minalignslop = 0;
341 * Allow space for the inode btree to split.
343 args.minleft = args.mp->m_in_maxlevels - 1;
344 if ((error = xfs_alloc_vextent(&args)))
349 * If stripe alignment is turned on, then try again with cluster
352 if (isaligned && args.fsbno == NULLFSBLOCK) {
353 args.type = XFS_ALLOCTYPE_NEAR_BNO;
354 args.agbno = be32_to_cpu(agi->agi_root);
355 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
356 args.alignment = xfs_ialloc_cluster_alignment(&args);
357 if ((error = xfs_alloc_vextent(&args)))
361 if (args.fsbno == NULLFSBLOCK) {
365 ASSERT(args.len == args.minlen);
368 * Stamp and write the inode buffers.
370 * Seed the new inode cluster with a random generation number. This
371 * prevents short-term reuse of generation numbers if a chunk is
372 * freed and then immediately reallocated. We use random numbers
373 * rather than a linear progression to prevent the next generation
374 * number from being easily guessable.
376 xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
380 * Convert the results.
382 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
383 be32_add_cpu(&agi->agi_count, newlen);
384 be32_add_cpu(&agi->agi_freecount, newlen);
385 down_read(&args.mp->m_peraglock);
386 args.mp->m_perag[agno].pagi_freecount += newlen;
387 up_read(&args.mp->m_peraglock);
388 agi->agi_newino = cpu_to_be32(newino);
391 * Insert records describing the new inode chunk into the btree.
393 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
394 for (thisino = newino;
395 thisino < newino + newlen;
396 thisino += XFS_INODES_PER_CHUNK) {
397 cur->bc_rec.i.ir_startino = thisino;
398 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
399 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
400 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
402 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
406 error = xfs_btree_insert(cur, &i);
408 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
413 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
415 * Log allocation group header fields
417 xfs_ialloc_log_agi(tp, agbp,
418 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
420 * Modify/log superblock values for inode count and inode free count.
422 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
423 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
428 STATIC_INLINE xfs_agnumber_t
434 spin_lock(&mp->m_agirotor_lock);
435 agno = mp->m_agirotor;
436 if (++mp->m_agirotor == mp->m_maxagi)
438 spin_unlock(&mp->m_agirotor_lock);
444 * Select an allocation group to look for a free inode in, based on the parent
445 * inode and then mode. Return the allocation group buffer.
447 STATIC xfs_buf_t * /* allocation group buffer */
448 xfs_ialloc_ag_select(
449 xfs_trans_t *tp, /* transaction pointer */
450 xfs_ino_t parent, /* parent directory inode number */
451 mode_t mode, /* bits set to indicate file type */
452 int okalloc) /* ok to allocate more space */
454 xfs_buf_t *agbp; /* allocation group header buffer */
455 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
456 xfs_agnumber_t agno; /* current ag number */
457 int flags; /* alloc buffer locking flags */
458 xfs_extlen_t ineed; /* blocks needed for inode allocation */
459 xfs_extlen_t longest = 0; /* longest extent available */
460 xfs_mount_t *mp; /* mount point structure */
461 int needspace; /* file mode implies space allocated */
462 xfs_perag_t *pag; /* per allocation group data */
463 xfs_agnumber_t pagno; /* parent (starting) ag number */
466 * Files of these types need at least one block if length > 0
467 * (and they won't fit in the inode, but that's hard to figure out).
469 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
471 agcount = mp->m_maxagi;
473 pagno = xfs_ialloc_next_ag(mp);
475 pagno = XFS_INO_TO_AGNO(mp, parent);
476 if (pagno >= agcount)
479 ASSERT(pagno < agcount);
481 * Loop through allocation groups, looking for one with a little
482 * free space in it. Note we don't look for free inodes, exactly.
483 * Instead, we include whether there is a need to allocate inodes
484 * to mean that blocks must be allocated for them,
485 * if none are currently free.
488 flags = XFS_ALLOC_FLAG_TRYLOCK;
489 down_read(&mp->m_peraglock);
491 pag = &mp->m_perag[agno];
492 if (!pag->pagi_init) {
493 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
500 if (!pag->pagi_inodeok) {
501 xfs_ialloc_next_ag(mp);
506 * Is there enough free space for the file plus a block
507 * of inodes (if we need to allocate some)?
509 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
510 if (ineed && !pag->pagf_init) {
512 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
516 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
518 if (!ineed || pag->pagf_init) {
519 if (ineed && !(longest = pag->pagf_longest))
520 longest = pag->pagf_flcount > 0;
522 (pag->pagf_freeblks >= needspace + ineed &&
526 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
530 up_read(&mp->m_peraglock);
536 xfs_trans_brelse(tp, agbp);
539 * No point in iterating over the rest, if we're shutting
542 if (XFS_FORCED_SHUTDOWN(mp)) {
543 up_read(&mp->m_peraglock);
551 up_read(&mp->m_peraglock);
560 * Try to retrieve the next record to the left/right from the current one.
564 struct xfs_btree_cur *cur,
565 xfs_inobt_rec_incore_t *rec,
573 error = xfs_btree_decrement(cur, 0, &i);
575 error = xfs_btree_increment(cur, 0, &i);
581 error = xfs_inobt_get_rec(cur, rec, &i);
584 XFS_WANT_CORRUPTED_RETURN(i == 1);
592 * Visible inode allocation functions.
596 * Allocate an inode on disk.
597 * Mode is used to tell whether the new inode will need space, and whether
600 * The arguments IO_agbp and alloc_done are defined to work within
601 * the constraint of one allocation per transaction.
602 * xfs_dialloc() is designed to be called twice if it has to do an
603 * allocation to make more free inodes. On the first call,
604 * IO_agbp should be set to NULL. If an inode is available,
605 * i.e., xfs_dialloc() did not need to do an allocation, an inode
606 * number is returned. In this case, IO_agbp would be set to the
607 * current ag_buf and alloc_done set to false.
608 * If an allocation needed to be done, xfs_dialloc would return
609 * the current ag_buf in IO_agbp and set alloc_done to true.
610 * The caller should then commit the current transaction, allocate a new
611 * transaction, and call xfs_dialloc() again, passing in the previous
612 * value of IO_agbp. IO_agbp should be held across the transactions.
613 * Since the agbp is locked across the two calls, the second call is
614 * guaranteed to have a free inode available.
616 * Once we successfully pick an inode its number is returned and the
617 * on-disk data structures are updated. The inode itself is not read
618 * in, since doing so would break ordering constraints with xfs_reclaim.
622 xfs_trans_t *tp, /* transaction pointer */
623 xfs_ino_t parent, /* parent inode (directory) */
624 mode_t mode, /* mode bits for new inode */
625 int okalloc, /* ok to allocate more space */
626 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
627 boolean_t *alloc_done, /* true if we needed to replenish
629 xfs_ino_t *inop) /* inode number allocated */
631 xfs_agnumber_t agcount; /* number of allocation groups */
632 xfs_buf_t *agbp; /* allocation group header's buffer */
633 xfs_agnumber_t agno; /* allocation group number */
634 xfs_agi_t *agi; /* allocation group header structure */
635 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
636 int error; /* error return value */
637 int i; /* result code */
638 int ialloced; /* inode allocation status */
639 int noroom = 0; /* no space for inode blk allocation */
640 xfs_ino_t ino; /* fs-relative inode to be returned */
642 int j; /* result code */
643 xfs_mount_t *mp; /* file system mount structure */
644 int offset; /* index of inode in chunk */
645 xfs_agino_t pagino; /* parent's AG relative inode # */
646 xfs_agnumber_t pagno; /* parent's AG number */
647 xfs_inobt_rec_incore_t rec; /* inode allocation record */
648 xfs_agnumber_t tagno; /* testing allocation group number */
649 xfs_btree_cur_t *tcur; /* temp cursor */
650 xfs_inobt_rec_incore_t trec; /* temp inode allocation record */
653 if (*IO_agbp == NULL) {
655 * We do not have an agbp, so select an initial allocation
656 * group for inode allocation.
658 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
660 * Couldn't find an allocation group satisfying the
667 agi = XFS_BUF_TO_AGI(agbp);
668 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
671 * Continue where we left off before. In this case, we
672 * know that the allocation group has free inodes.
675 agi = XFS_BUF_TO_AGI(agbp);
676 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
677 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
680 agcount = mp->m_sb.sb_agcount;
681 agno = be32_to_cpu(agi->agi_seqno);
683 pagno = XFS_INO_TO_AGNO(mp, parent);
684 pagino = XFS_INO_TO_AGINO(mp, parent);
687 * If we have already hit the ceiling of inode blocks then clear
688 * okalloc so we scan all available agi structures for a free
692 if (mp->m_maxicount &&
693 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
699 * Loop until we find an allocation group that either has free inodes
700 * or in which we can allocate some inodes. Iterate through the
701 * allocation groups upward, wrapping at the end.
703 *alloc_done = B_FALSE;
704 while (!agi->agi_freecount) {
706 * Don't do anything if we're not supposed to allocate
707 * any blocks, just go on to the next ag.
711 * Try to allocate some new inodes in the allocation
714 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
715 xfs_trans_brelse(tp, agbp);
716 if (error == ENOSPC) {
724 * We successfully allocated some inodes, return
725 * the current context to the caller so that it
726 * can commit the current transaction and call
727 * us again where we left off.
729 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
730 *alloc_done = B_TRUE;
737 * If it failed, give up on this ag.
739 xfs_trans_brelse(tp, agbp);
741 * Go on to the next ag: get its ag header.
744 if (++tagno == agcount)
748 return noroom ? ENOSPC : 0;
750 down_read(&mp->m_peraglock);
751 if (mp->m_perag[tagno].pagi_inodeok == 0) {
752 up_read(&mp->m_peraglock);
755 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
756 up_read(&mp->m_peraglock);
759 agi = XFS_BUF_TO_AGI(agbp);
760 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
763 * Here with an allocation group that has a free inode.
764 * Reset agno since we may have chosen a new ag in the
769 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
771 * If pagino is 0 (this is the root inode allocation) use newino.
772 * This must work because we've just allocated some.
775 pagino = be32_to_cpu(agi->agi_newino);
777 error = xfs_check_agi_freecount(cur, agi);
782 * If in the same AG as the parent, try to get near the parent.
785 int doneleft; /* done, to the left */
786 int doneright; /* done, to the right */
788 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
791 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
793 error = xfs_inobt_get_rec(cur, &rec, &j);
796 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
798 if (rec.ir_freecount > 0) {
800 * Found a free inode in the same chunk
801 * as the parent, done.
808 * In the same AG as parent, but parent's chunk is full.
811 /* duplicate the cursor, search left & right simultaneously */
812 error = xfs_btree_dup_cursor(cur, &tcur);
816 /* search left with tcur, back up 1 record */
817 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
821 /* search right with cur, go forward 1 record. */
822 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
827 * Loop until we find an inode chunk with a free inode.
829 while (!doneleft || !doneright) {
830 int useleft; /* using left inode chunk this time */
832 /* figure out the closer block if both are valid. */
833 if (!doneleft && !doneright) {
835 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
836 rec.ir_startino - pagino;
841 /* free inodes to the left? */
842 if (useleft && trec.ir_freecount) {
844 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
849 /* free inodes to the right? */
850 if (!useleft && rec.ir_freecount) {
851 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
855 /* get next record to check */
857 error = xfs_ialloc_next_rec(tcur, &trec,
860 error = xfs_ialloc_next_rec(cur, &rec,
866 ASSERT(!doneleft || !doneright);
870 * In a different AG from the parent.
871 * See if the most recently allocated block has any free.
873 else if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
874 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
880 error = xfs_inobt_get_rec(cur, &rec, &j);
884 if (j == 1 && rec.ir_freecount > 0) {
886 * The last chunk allocated in the group
887 * still has a free inode.
894 * None left in the last group, search the whole AG
896 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
899 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
902 error = xfs_inobt_get_rec(cur, &rec, &i);
905 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
906 if (rec.ir_freecount > 0)
908 error = xfs_btree_increment(cur, 0, &i);
911 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
916 offset = xfs_ialloc_find_free(&rec.ir_free);
918 ASSERT(offset < XFS_INODES_PER_CHUNK);
919 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
920 XFS_INODES_PER_CHUNK) == 0);
921 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
922 rec.ir_free &= ~XFS_INOBT_MASK(offset);
924 error = xfs_inobt_update(cur, &rec);
927 be32_add_cpu(&agi->agi_freecount, -1);
928 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
929 down_read(&mp->m_peraglock);
930 mp->m_perag[tagno].pagi_freecount--;
931 up_read(&mp->m_peraglock);
933 error = xfs_check_agi_freecount(cur, agi);
937 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
938 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
942 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
944 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
949 * Free disk inode. Carefully avoids touching the incore inode, all
950 * manipulations incore are the caller's responsibility.
951 * The on-disk inode is not changed by this operation, only the
952 * btree (free inode mask) is changed.
956 xfs_trans_t *tp, /* transaction pointer */
957 xfs_ino_t inode, /* inode to be freed */
958 xfs_bmap_free_t *flist, /* extents to free */
959 int *delete, /* set if inode cluster was deleted */
960 xfs_ino_t *first_ino) /* first inode in deleted cluster */
963 xfs_agblock_t agbno; /* block number containing inode */
964 xfs_buf_t *agbp; /* buffer containing allocation group header */
965 xfs_agino_t agino; /* inode number relative to allocation group */
966 xfs_agnumber_t agno; /* allocation group number */
967 xfs_agi_t *agi; /* allocation group header */
968 xfs_btree_cur_t *cur; /* inode btree cursor */
969 int error; /* error return value */
970 int i; /* result code */
971 int ilen; /* inodes in an inode cluster */
972 xfs_mount_t *mp; /* mount structure for filesystem */
973 int off; /* offset of inode in inode chunk */
974 xfs_inobt_rec_incore_t rec; /* btree record */
979 * Break up inode number into its components.
981 agno = XFS_INO_TO_AGNO(mp, inode);
982 if (agno >= mp->m_sb.sb_agcount) {
984 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
985 agno, mp->m_sb.sb_agcount, mp->m_fsname);
987 return XFS_ERROR(EINVAL);
989 agino = XFS_INO_TO_AGINO(mp, inode);
990 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
992 "xfs_difree: inode != XFS_AGINO_TO_INO() "
993 "(%llu != %llu) on %s. Returning EINVAL.",
994 (unsigned long long)inode,
995 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
998 return XFS_ERROR(EINVAL);
1000 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1001 if (agbno >= mp->m_sb.sb_agblocks) {
1003 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
1004 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
1006 return XFS_ERROR(EINVAL);
1009 * Get the allocation group header.
1011 down_read(&mp->m_peraglock);
1012 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1013 up_read(&mp->m_peraglock);
1016 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
1017 error, mp->m_fsname);
1020 agi = XFS_BUF_TO_AGI(agbp);
1021 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1022 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1024 * Initialize the cursor.
1026 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1028 error = xfs_check_agi_freecount(cur, agi);
1033 * Look for the entry describing this inode.
1035 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1037 "xfs_difree: xfs_inobt_lookup returned() an error %d on %s. Returning error.",
1038 error, mp->m_fsname);
1041 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1042 error = xfs_inobt_get_rec(cur, &rec, &i);
1045 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1046 error, mp->m_fsname);
1049 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1051 * Get the offset in the inode chunk.
1053 off = agino - rec.ir_startino;
1054 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1055 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1057 * Mark the inode free & increment the count.
1059 rec.ir_free |= XFS_INOBT_MASK(off);
1063 * When an inode cluster is free, it becomes eligible for removal
1065 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1066 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1069 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1072 * Remove the inode cluster from the AGI B+Tree, adjust the
1073 * AGI and Superblock inode counts, and mark the disk space
1074 * to be freed when the transaction is committed.
1076 ilen = XFS_IALLOC_INODES(mp);
1077 be32_add_cpu(&agi->agi_count, -ilen);
1078 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1079 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1080 down_read(&mp->m_peraglock);
1081 mp->m_perag[agno].pagi_freecount -= ilen - 1;
1082 up_read(&mp->m_peraglock);
1083 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1084 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1086 if ((error = xfs_btree_delete(cur, &i))) {
1087 cmn_err(CE_WARN, "xfs_difree: xfs_btree_delete returned an error %d on %s.\n",
1088 error, mp->m_fsname);
1092 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1093 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1094 XFS_IALLOC_BLOCKS(mp), flist, mp);
1098 error = xfs_inobt_update(cur, &rec);
1101 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1102 error, mp->m_fsname);
1107 * Change the inode free counts and log the ag/sb changes.
1109 be32_add_cpu(&agi->agi_freecount, 1);
1110 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1111 down_read(&mp->m_peraglock);
1112 mp->m_perag[agno].pagi_freecount++;
1113 up_read(&mp->m_peraglock);
1114 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1117 error = xfs_check_agi_freecount(cur, agi);
1121 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1125 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1130 * Return the location of the inode in imap, for mapping it into a buffer.
1134 xfs_mount_t *mp, /* file system mount structure */
1135 xfs_trans_t *tp, /* transaction pointer */
1136 xfs_ino_t ino, /* inode to locate */
1137 struct xfs_imap *imap, /* location map structure */
1138 uint flags) /* flags for inode btree lookup */
1140 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1141 xfs_agino_t agino; /* inode number within alloc group */
1142 xfs_agnumber_t agno; /* allocation group number */
1143 int blks_per_cluster; /* num blocks per inode cluster */
1144 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1145 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1146 int error; /* error code */
1147 int offset; /* index of inode in its buffer */
1148 int offset_agbno; /* blks from chunk start to inode */
1150 ASSERT(ino != NULLFSINO);
1153 * Split up the inode number into its parts.
1155 agno = XFS_INO_TO_AGNO(mp, ino);
1156 agino = XFS_INO_TO_AGINO(mp, ino);
1157 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1158 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1159 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1161 /* no diagnostics for bulkstat, ino comes from userspace */
1162 if (flags & XFS_IGET_BULKSTAT)
1163 return XFS_ERROR(EINVAL);
1164 if (agno >= mp->m_sb.sb_agcount) {
1165 xfs_fs_cmn_err(CE_ALERT, mp,
1166 "xfs_imap: agno (%d) >= "
1167 "mp->m_sb.sb_agcount (%d)",
1168 agno, mp->m_sb.sb_agcount);
1170 if (agbno >= mp->m_sb.sb_agblocks) {
1171 xfs_fs_cmn_err(CE_ALERT, mp,
1172 "xfs_imap: agbno (0x%llx) >= "
1173 "mp->m_sb.sb_agblocks (0x%lx)",
1174 (unsigned long long) agbno,
1175 (unsigned long) mp->m_sb.sb_agblocks);
1177 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1178 xfs_fs_cmn_err(CE_ALERT, mp,
1179 "xfs_imap: ino (0x%llx) != "
1180 "XFS_AGINO_TO_INO(mp, agno, agino) "
1182 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1186 return XFS_ERROR(EINVAL);
1190 * If the inode cluster size is the same as the blocksize or
1191 * smaller we get to the buffer by simple arithmetics.
1193 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1194 offset = XFS_INO_TO_OFFSET(mp, ino);
1195 ASSERT(offset < mp->m_sb.sb_inopblock);
1197 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1198 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1199 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1203 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1206 * If we get a block number passed from bulkstat we can use it to
1207 * find the buffer easily.
1209 if (imap->im_blkno) {
1210 offset = XFS_INO_TO_OFFSET(mp, ino);
1211 ASSERT(offset < mp->m_sb.sb_inopblock);
1213 cluster_agbno = xfs_daddr_to_agbno(mp, imap->im_blkno);
1214 offset += (agbno - cluster_agbno) * mp->m_sb.sb_inopblock;
1216 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1217 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1222 * If the inode chunks are aligned then use simple maths to
1223 * find the location. Otherwise we have to do a btree
1224 * lookup to find the location.
1226 if (mp->m_inoalign_mask) {
1227 offset_agbno = agbno & mp->m_inoalign_mask;
1228 chunk_agbno = agbno - offset_agbno;
1230 xfs_btree_cur_t *cur; /* inode btree cursor */
1231 xfs_inobt_rec_incore_t chunk_rec;
1232 xfs_buf_t *agbp; /* agi buffer */
1233 int i; /* temp state */
1235 down_read(&mp->m_peraglock);
1236 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1237 up_read(&mp->m_peraglock);
1239 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1240 "xfs_ialloc_read_agi() returned "
1241 "error %d, agno %d",
1246 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1247 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1249 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1250 "xfs_inobt_lookup() failed");
1254 error = xfs_inobt_get_rec(cur, &chunk_rec, &i);
1256 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1257 "xfs_inobt_get_rec() failed");
1262 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1263 "xfs_inobt_get_rec() failed");
1265 error = XFS_ERROR(EINVAL);
1268 xfs_trans_brelse(tp, agbp);
1269 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1272 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_rec.ir_startino);
1273 offset_agbno = agbno - chunk_agbno;
1276 ASSERT(agbno >= chunk_agbno);
1277 cluster_agbno = chunk_agbno +
1278 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1279 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1280 XFS_INO_TO_OFFSET(mp, ino);
1282 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1283 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1284 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1287 * If the inode number maps to a block outside the bounds
1288 * of the file system then return NULL rather than calling
1289 * read_buf and panicing when we get an error from the
1292 if ((imap->im_blkno + imap->im_len) >
1293 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1294 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1295 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
1296 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
1297 (unsigned long long) imap->im_blkno,
1298 (unsigned long long) imap->im_len,
1299 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1300 return XFS_ERROR(EINVAL);
1307 * Compute and fill in value of m_in_maxlevels.
1310 xfs_ialloc_compute_maxlevels(
1311 xfs_mount_t *mp) /* file system mount structure */
1319 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1320 XFS_INODES_PER_CHUNK_LOG;
1321 minleafrecs = mp->m_alloc_mnr[0];
1322 minnoderecs = mp->m_alloc_mnr[1];
1323 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1324 for (level = 1; maxblocks > 1; level++)
1325 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1326 mp->m_in_maxlevels = level;
1330 * Log specified fields for the ag hdr (inode section)
1334 xfs_trans_t *tp, /* transaction pointer */
1335 xfs_buf_t *bp, /* allocation group header buffer */
1336 int fields) /* bitmask of fields to log */
1338 int first; /* first byte number */
1339 int last; /* last byte number */
1340 static const short offsets[] = { /* field starting offsets */
1341 /* keep in sync with bit definitions */
1342 offsetof(xfs_agi_t, agi_magicnum),
1343 offsetof(xfs_agi_t, agi_versionnum),
1344 offsetof(xfs_agi_t, agi_seqno),
1345 offsetof(xfs_agi_t, agi_length),
1346 offsetof(xfs_agi_t, agi_count),
1347 offsetof(xfs_agi_t, agi_root),
1348 offsetof(xfs_agi_t, agi_level),
1349 offsetof(xfs_agi_t, agi_freecount),
1350 offsetof(xfs_agi_t, agi_newino),
1351 offsetof(xfs_agi_t, agi_dirino),
1352 offsetof(xfs_agi_t, agi_unlinked),
1356 xfs_agi_t *agi; /* allocation group header */
1358 agi = XFS_BUF_TO_AGI(bp);
1359 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
1362 * Compute byte offsets for the first and last fields.
1364 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1366 * Log the allocation group inode header buffer.
1368 xfs_trans_log_buf(tp, bp, first, last);
1373 xfs_check_agi_unlinked(
1374 struct xfs_agi *agi)
1378 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1379 ASSERT(agi->agi_unlinked[i]);
1382 #define xfs_check_agi_unlinked(agi)
1386 * Read in the allocation group header (inode allocation section)
1390 struct xfs_mount *mp, /* file system mount structure */
1391 struct xfs_trans *tp, /* transaction pointer */
1392 xfs_agnumber_t agno, /* allocation group number */
1393 struct xfs_buf **bpp) /* allocation group hdr buf */
1395 struct xfs_agi *agi; /* allocation group header */
1396 int agi_ok; /* agi is consistent */
1399 ASSERT(agno != NULLAGNUMBER);
1401 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1402 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1403 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1407 ASSERT(*bpp && !XFS_BUF_GETERROR(*bpp));
1408 agi = XFS_BUF_TO_AGI(*bpp);
1411 * Validate the magic number of the agi block.
1413 agi_ok = be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1414 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1415 be32_to_cpu(agi->agi_seqno) == agno;
1416 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1417 XFS_RANDOM_IALLOC_READ_AGI))) {
1418 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1420 xfs_trans_brelse(tp, *bpp);
1421 return XFS_ERROR(EFSCORRUPTED);
1424 XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGI, XFS_AGI_REF);
1426 xfs_check_agi_unlinked(agi);
1431 xfs_ialloc_read_agi(
1432 struct xfs_mount *mp, /* file system mount structure */
1433 struct xfs_trans *tp, /* transaction pointer */
1434 xfs_agnumber_t agno, /* allocation group number */
1435 struct xfs_buf **bpp) /* allocation group hdr buf */
1437 struct xfs_agi *agi; /* allocation group header */
1438 struct xfs_perag *pag; /* per allocation group data */
1441 error = xfs_read_agi(mp, tp, agno, bpp);
1445 agi = XFS_BUF_TO_AGI(*bpp);
1446 pag = &mp->m_perag[agno];
1448 if (!pag->pagi_init) {
1449 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1450 pag->pagi_count = be32_to_cpu(agi->agi_count);
1455 * It's possible for these to be out of sync if
1456 * we are in the middle of a forced shutdown.
1458 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1459 XFS_FORCED_SHUTDOWN(mp));
1464 * Read in the agi to initialise the per-ag data in the mount structure
1467 xfs_ialloc_pagi_init(
1468 xfs_mount_t *mp, /* file system mount structure */
1469 xfs_trans_t *tp, /* transaction pointer */
1470 xfs_agnumber_t agno) /* allocation group number */
1472 xfs_buf_t *bp = NULL;
1475 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1479 xfs_trans_brelse(tp, bp);