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_agnumber_t
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_agnumber_t agcount; /* number of ag's in the filesystem */
453 xfs_agnumber_t agno; /* current ag number */
454 int flags; /* alloc buffer locking flags */
455 xfs_extlen_t ineed; /* blocks needed for inode allocation */
456 xfs_extlen_t longest = 0; /* longest extent available */
457 xfs_mount_t *mp; /* mount point structure */
458 int needspace; /* file mode implies space allocated */
459 xfs_perag_t *pag; /* per allocation group data */
460 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)
478 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;
490 pag = xfs_perag_get(mp, agno);
491 if (!pag->pagi_inodeok) {
492 xfs_ialloc_next_ag(mp);
496 if (!pag->pagi_init) {
497 error = xfs_ialloc_pagi_init(mp, tp, agno);
502 if (pag->pagi_freecount) {
510 if (!pag->pagf_init) {
511 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
517 * Is there enough free space for the file plus a block of
518 * inodes? (if we need to allocate some)?
520 ineed = XFS_IALLOC_BLOCKS(mp);
521 longest = pag->pagf_longest;
523 longest = pag->pagf_flcount > 0;
525 if (pag->pagf_freeblks >= needspace + ineed &&
533 * No point in iterating over the rest, if we're shutting
536 if (XFS_FORCED_SHUTDOWN(mp))
550 * Try to retrieve the next record to the left/right from the current one.
554 struct xfs_btree_cur *cur,
555 xfs_inobt_rec_incore_t *rec,
563 error = xfs_btree_decrement(cur, 0, &i);
565 error = xfs_btree_increment(cur, 0, &i);
571 error = xfs_inobt_get_rec(cur, rec, &i);
574 XFS_WANT_CORRUPTED_RETURN(i == 1);
582 struct xfs_btree_cur *cur,
584 xfs_inobt_rec_incore_t *rec,
591 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
596 error = xfs_inobt_get_rec(cur, rec, &i);
599 XFS_WANT_CORRUPTED_RETURN(i == 1);
608 * The caller selected an AG for us, and made sure that free inodes are
613 struct xfs_trans *tp,
614 struct xfs_buf *agbp,
618 struct xfs_mount *mp = tp->t_mountp;
619 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
620 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
621 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
622 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
623 struct xfs_perag *pag;
624 struct xfs_btree_cur *cur, *tcur;
625 struct xfs_inobt_rec_incore rec, trec;
631 pag = xfs_perag_get(mp, agno);
633 ASSERT(pag->pagi_init);
634 ASSERT(pag->pagi_inodeok);
635 ASSERT(pag->pagi_freecount > 0);
638 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
640 * If pagino is 0 (this is the root inode allocation) use newino.
641 * This must work because we've just allocated some.
644 pagino = be32_to_cpu(agi->agi_newino);
646 error = xfs_check_agi_freecount(cur, agi);
651 * If in the same AG as the parent, try to get near the parent.
654 int doneleft; /* done, to the left */
655 int doneright; /* done, to the right */
656 int searchdistance = 10;
658 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
661 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
663 error = xfs_inobt_get_rec(cur, &rec, &j);
666 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
668 if (rec.ir_freecount > 0) {
670 * Found a free inode in the same chunk
671 * as the parent, done.
678 * In the same AG as parent, but parent's chunk is full.
681 /* duplicate the cursor, search left & right simultaneously */
682 error = xfs_btree_dup_cursor(cur, &tcur);
687 * Skip to last blocks looked up if same parent inode.
689 if (pagino != NULLAGINO &&
690 pag->pagl_pagino == pagino &&
691 pag->pagl_leftrec != NULLAGINO &&
692 pag->pagl_rightrec != NULLAGINO) {
693 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
694 &trec, &doneleft, 1);
698 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
699 &rec, &doneright, 0);
703 /* search left with tcur, back up 1 record */
704 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
708 /* search right with cur, go forward 1 record. */
709 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
715 * Loop until we find an inode chunk with a free inode.
717 while (!doneleft || !doneright) {
718 int useleft; /* using left inode chunk this time */
720 if (!--searchdistance) {
722 * Not in range - save last search
723 * location and allocate a new inode
725 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
726 pag->pagl_leftrec = trec.ir_startino;
727 pag->pagl_rightrec = rec.ir_startino;
728 pag->pagl_pagino = pagino;
732 /* figure out the closer block if both are valid. */
733 if (!doneleft && !doneright) {
735 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
736 rec.ir_startino - pagino;
741 /* free inodes to the left? */
742 if (useleft && trec.ir_freecount) {
744 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
747 pag->pagl_leftrec = trec.ir_startino;
748 pag->pagl_rightrec = rec.ir_startino;
749 pag->pagl_pagino = pagino;
753 /* free inodes to the right? */
754 if (!useleft && rec.ir_freecount) {
755 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
757 pag->pagl_leftrec = trec.ir_startino;
758 pag->pagl_rightrec = rec.ir_startino;
759 pag->pagl_pagino = pagino;
763 /* get next record to check */
765 error = xfs_ialloc_next_rec(tcur, &trec,
768 error = xfs_ialloc_next_rec(cur, &rec,
776 * We've reached the end of the btree. because
777 * we are only searching a small chunk of the
778 * btree each search, there is obviously free
779 * inodes closer to the parent inode than we
780 * are now. restart the search again.
782 pag->pagl_pagino = NULLAGINO;
783 pag->pagl_leftrec = NULLAGINO;
784 pag->pagl_rightrec = NULLAGINO;
785 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
786 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
791 * In a different AG from the parent.
792 * See if the most recently allocated block has any free.
795 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
796 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
802 error = xfs_inobt_get_rec(cur, &rec, &j);
806 if (j == 1 && rec.ir_freecount > 0) {
808 * The last chunk allocated in the group
809 * still has a free inode.
817 * None left in the last group, search the whole AG
819 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
822 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
825 error = xfs_inobt_get_rec(cur, &rec, &i);
828 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
829 if (rec.ir_freecount > 0)
831 error = xfs_btree_increment(cur, 0, &i);
834 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
838 offset = xfs_lowbit64(rec.ir_free);
840 ASSERT(offset < XFS_INODES_PER_CHUNK);
841 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
842 XFS_INODES_PER_CHUNK) == 0);
843 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
844 rec.ir_free &= ~XFS_INOBT_MASK(offset);
846 error = xfs_inobt_update(cur, &rec);
849 be32_add_cpu(&agi->agi_freecount, -1);
850 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
851 pag->pagi_freecount--;
853 error = xfs_check_agi_freecount(cur, agi);
857 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
858 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
863 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
865 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
871 * Allocate an inode on disk.
873 * Mode is used to tell whether the new inode will need space, and whether it
876 * This function is designed to be called twice if it has to do an allocation
877 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
878 * If an inode is available without having to performn an allocation, an inode
879 * number is returned. In this case, *IO_agbp would be NULL. If an allocation
880 * needes to be done, xfs_dialloc would return the current AGI buffer in
881 * *IO_agbp. The caller should then commit the current transaction, allocate a
882 * new transaction, and call xfs_dialloc() again, passing in the previous value
883 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
884 * buffer is locked across the two calls, the second call is guaranteed to have
885 * a free inode available.
887 * Once we successfully pick an inode its number is returned and the on-disk
888 * data structures are updated. The inode itself is not read in, since doing so
889 * would break ordering constraints with xfs_reclaim.
893 struct xfs_trans *tp,
897 struct xfs_buf **IO_agbp,
900 struct xfs_mount *mp = tp->t_mountp;
901 struct xfs_buf *agbp;
906 xfs_agnumber_t start_agno;
907 struct xfs_perag *pag;
911 * If the caller passes in a pointer to the AGI buffer,
912 * continue where we left off before. In this case, we
913 * know that the allocation group has free inodes.
920 * We do not have an agbp, so select an initial allocation
921 * group for inode allocation.
923 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
924 if (start_agno == NULLAGNUMBER) {
930 * If we have already hit the ceiling of inode blocks then clear
931 * okalloc so we scan all available agi structures for a free
934 if (mp->m_maxicount &&
935 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
941 * Loop until we find an allocation group that either has free inodes
942 * or in which we can allocate some inodes. Iterate through the
943 * allocation groups upward, wrapping at the end.
947 pag = xfs_perag_get(mp, agno);
948 if (!pag->pagi_inodeok) {
949 xfs_ialloc_next_ag(mp);
953 if (!pag->pagi_init) {
954 error = xfs_ialloc_pagi_init(mp, tp, agno);
960 * Do a first racy fast path check if this AG is usable.
962 if (!pag->pagi_freecount && !okalloc)
965 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
970 * Once the AGI has been read in we have to recheck
971 * pagi_freecount with the AGI buffer lock held.
973 if (pag->pagi_freecount) {
979 xfs_trans_brelse(tp, agbp);
983 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
985 xfs_trans_brelse(tp, agbp);
997 * We successfully allocated some inodes, return
998 * the current context to the caller so that it
999 * can commit the current transaction and call
1000 * us again where we left off.
1002 ASSERT(pag->pagi_freecount > 0);
1012 if (++agno == mp->m_sb.sb_agcount)
1014 if (agno == start_agno) {
1016 return noroom ? ENOSPC : 0;
1022 return xfs_dialloc_ag(tp, agbp, parent, inop);
1025 return XFS_ERROR(error);
1029 * Free disk inode. Carefully avoids touching the incore inode, all
1030 * manipulations incore are the caller's responsibility.
1031 * The on-disk inode is not changed by this operation, only the
1032 * btree (free inode mask) is changed.
1036 xfs_trans_t *tp, /* transaction pointer */
1037 xfs_ino_t inode, /* inode to be freed */
1038 xfs_bmap_free_t *flist, /* extents to free */
1039 int *delete, /* set if inode cluster was deleted */
1040 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1043 xfs_agblock_t agbno; /* block number containing inode */
1044 xfs_buf_t *agbp; /* buffer containing allocation group header */
1045 xfs_agino_t agino; /* inode number relative to allocation group */
1046 xfs_agnumber_t agno; /* allocation group number */
1047 xfs_agi_t *agi; /* allocation group header */
1048 xfs_btree_cur_t *cur; /* inode btree cursor */
1049 int error; /* error return value */
1050 int i; /* result code */
1051 int ilen; /* inodes in an inode cluster */
1052 xfs_mount_t *mp; /* mount structure for filesystem */
1053 int off; /* offset of inode in inode chunk */
1054 xfs_inobt_rec_incore_t rec; /* btree record */
1055 struct xfs_perag *pag;
1060 * Break up inode number into its components.
1062 agno = XFS_INO_TO_AGNO(mp, inode);
1063 if (agno >= mp->m_sb.sb_agcount) {
1064 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1065 __func__, agno, mp->m_sb.sb_agcount);
1067 return XFS_ERROR(EINVAL);
1069 agino = XFS_INO_TO_AGINO(mp, inode);
1070 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1071 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1072 __func__, (unsigned long long)inode,
1073 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1075 return XFS_ERROR(EINVAL);
1077 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1078 if (agbno >= mp->m_sb.sb_agblocks) {
1079 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1080 __func__, agbno, mp->m_sb.sb_agblocks);
1082 return XFS_ERROR(EINVAL);
1085 * Get the allocation group header.
1087 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1089 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1093 agi = XFS_BUF_TO_AGI(agbp);
1094 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1095 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1097 * Initialize the cursor.
1099 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1101 error = xfs_check_agi_freecount(cur, agi);
1106 * Look for the entry describing this inode.
1108 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1109 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1113 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1114 error = xfs_inobt_get_rec(cur, &rec, &i);
1116 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1120 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1122 * Get the offset in the inode chunk.
1124 off = agino - rec.ir_startino;
1125 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1126 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1128 * Mark the inode free & increment the count.
1130 rec.ir_free |= XFS_INOBT_MASK(off);
1134 * When an inode cluster is free, it becomes eligible for removal
1136 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1137 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1140 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1143 * Remove the inode cluster from the AGI B+Tree, adjust the
1144 * AGI and Superblock inode counts, and mark the disk space
1145 * to be freed when the transaction is committed.
1147 ilen = XFS_IALLOC_INODES(mp);
1148 be32_add_cpu(&agi->agi_count, -ilen);
1149 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1150 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1151 pag = xfs_perag_get(mp, agno);
1152 pag->pagi_freecount -= ilen - 1;
1154 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1155 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1157 if ((error = xfs_btree_delete(cur, &i))) {
1158 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1163 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1164 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1165 XFS_IALLOC_BLOCKS(mp), flist, mp);
1169 error = xfs_inobt_update(cur, &rec);
1171 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1177 * Change the inode free counts and log the ag/sb changes.
1179 be32_add_cpu(&agi->agi_freecount, 1);
1180 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1181 pag = xfs_perag_get(mp, agno);
1182 pag->pagi_freecount++;
1184 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1187 error = xfs_check_agi_freecount(cur, agi);
1191 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1195 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1201 struct xfs_mount *mp,
1202 struct xfs_trans *tp,
1203 xfs_agnumber_t agno,
1205 xfs_agblock_t agbno,
1206 xfs_agblock_t *chunk_agbno,
1207 xfs_agblock_t *offset_agbno,
1210 struct xfs_inobt_rec_incore rec;
1211 struct xfs_btree_cur *cur;
1212 struct xfs_buf *agbp;
1216 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1219 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1220 __func__, error, agno);
1225 * Lookup the inode record for the given agino. If the record cannot be
1226 * found, then it's an invalid inode number and we should abort. Once
1227 * we have a record, we need to ensure it contains the inode number
1228 * we are looking up.
1230 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1231 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1234 error = xfs_inobt_get_rec(cur, &rec, &i);
1235 if (!error && i == 0)
1239 xfs_trans_brelse(tp, agbp);
1240 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1244 /* check that the returned record contains the required inode */
1245 if (rec.ir_startino > agino ||
1246 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1249 /* for untrusted inodes check it is allocated first */
1250 if ((flags & XFS_IGET_UNTRUSTED) &&
1251 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1254 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1255 *offset_agbno = agbno - *chunk_agbno;
1260 * Return the location of the inode in imap, for mapping it into a buffer.
1264 xfs_mount_t *mp, /* file system mount structure */
1265 xfs_trans_t *tp, /* transaction pointer */
1266 xfs_ino_t ino, /* inode to locate */
1267 struct xfs_imap *imap, /* location map structure */
1268 uint flags) /* flags for inode btree lookup */
1270 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1271 xfs_agino_t agino; /* inode number within alloc group */
1272 xfs_agnumber_t agno; /* allocation group number */
1273 int blks_per_cluster; /* num blocks per inode cluster */
1274 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1275 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1276 int error; /* error code */
1277 int offset; /* index of inode in its buffer */
1278 int offset_agbno; /* blks from chunk start to inode */
1280 ASSERT(ino != NULLFSINO);
1283 * Split up the inode number into its parts.
1285 agno = XFS_INO_TO_AGNO(mp, ino);
1286 agino = XFS_INO_TO_AGINO(mp, ino);
1287 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1288 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1289 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1292 * Don't output diagnostic information for untrusted inodes
1293 * as they can be invalid without implying corruption.
1295 if (flags & XFS_IGET_UNTRUSTED)
1296 return XFS_ERROR(EINVAL);
1297 if (agno >= mp->m_sb.sb_agcount) {
1299 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1300 __func__, agno, mp->m_sb.sb_agcount);
1302 if (agbno >= mp->m_sb.sb_agblocks) {
1304 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1305 __func__, (unsigned long long)agbno,
1306 (unsigned long)mp->m_sb.sb_agblocks);
1308 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1310 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1312 XFS_AGINO_TO_INO(mp, agno, agino));
1316 return XFS_ERROR(EINVAL);
1319 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1322 * For bulkstat and handle lookups, we have an untrusted inode number
1323 * that we have to verify is valid. We cannot do this just by reading
1324 * the inode buffer as it may have been unlinked and removed leaving
1325 * inodes in stale state on disk. Hence we have to do a btree lookup
1326 * in all cases where an untrusted inode number is passed.
1328 if (flags & XFS_IGET_UNTRUSTED) {
1329 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1330 &chunk_agbno, &offset_agbno, flags);
1337 * If the inode cluster size is the same as the blocksize or
1338 * smaller we get to the buffer by simple arithmetics.
1340 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1341 offset = XFS_INO_TO_OFFSET(mp, ino);
1342 ASSERT(offset < mp->m_sb.sb_inopblock);
1344 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1345 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1346 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1351 * If the inode chunks are aligned then use simple maths to
1352 * find the location. Otherwise we have to do a btree
1353 * lookup to find the location.
1355 if (mp->m_inoalign_mask) {
1356 offset_agbno = agbno & mp->m_inoalign_mask;
1357 chunk_agbno = agbno - offset_agbno;
1359 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1360 &chunk_agbno, &offset_agbno, flags);
1366 ASSERT(agbno >= chunk_agbno);
1367 cluster_agbno = chunk_agbno +
1368 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1369 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1370 XFS_INO_TO_OFFSET(mp, ino);
1372 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1373 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1374 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1377 * If the inode number maps to a block outside the bounds
1378 * of the file system then return NULL rather than calling
1379 * read_buf and panicing when we get an error from the
1382 if ((imap->im_blkno + imap->im_len) >
1383 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1385 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1386 __func__, (unsigned long long) imap->im_blkno,
1387 (unsigned long long) imap->im_len,
1388 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1389 return XFS_ERROR(EINVAL);
1395 * Compute and fill in value of m_in_maxlevels.
1398 xfs_ialloc_compute_maxlevels(
1399 xfs_mount_t *mp) /* file system mount structure */
1407 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1408 XFS_INODES_PER_CHUNK_LOG;
1409 minleafrecs = mp->m_alloc_mnr[0];
1410 minnoderecs = mp->m_alloc_mnr[1];
1411 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1412 for (level = 1; maxblocks > 1; level++)
1413 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1414 mp->m_in_maxlevels = level;
1418 * Log specified fields for the ag hdr (inode section)
1422 xfs_trans_t *tp, /* transaction pointer */
1423 xfs_buf_t *bp, /* allocation group header buffer */
1424 int fields) /* bitmask of fields to log */
1426 int first; /* first byte number */
1427 int last; /* last byte number */
1428 static const short offsets[] = { /* field starting offsets */
1429 /* keep in sync with bit definitions */
1430 offsetof(xfs_agi_t, agi_magicnum),
1431 offsetof(xfs_agi_t, agi_versionnum),
1432 offsetof(xfs_agi_t, agi_seqno),
1433 offsetof(xfs_agi_t, agi_length),
1434 offsetof(xfs_agi_t, agi_count),
1435 offsetof(xfs_agi_t, agi_root),
1436 offsetof(xfs_agi_t, agi_level),
1437 offsetof(xfs_agi_t, agi_freecount),
1438 offsetof(xfs_agi_t, agi_newino),
1439 offsetof(xfs_agi_t, agi_dirino),
1440 offsetof(xfs_agi_t, agi_unlinked),
1444 xfs_agi_t *agi; /* allocation group header */
1446 agi = XFS_BUF_TO_AGI(bp);
1447 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1450 * Compute byte offsets for the first and last fields.
1452 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1454 * Log the allocation group inode header buffer.
1456 xfs_trans_log_buf(tp, bp, first, last);
1461 xfs_check_agi_unlinked(
1462 struct xfs_agi *agi)
1466 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1467 ASSERT(agi->agi_unlinked[i]);
1470 #define xfs_check_agi_unlinked(agi)
1474 * Read in the allocation group header (inode allocation section)
1478 struct xfs_mount *mp, /* file system mount structure */
1479 struct xfs_trans *tp, /* transaction pointer */
1480 xfs_agnumber_t agno, /* allocation group number */
1481 struct xfs_buf **bpp) /* allocation group hdr buf */
1483 struct xfs_agi *agi; /* allocation group header */
1484 int agi_ok; /* agi is consistent */
1487 ASSERT(agno != NULLAGNUMBER);
1489 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1490 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1491 XFS_FSS_TO_BB(mp, 1), 0, bpp);
1495 ASSERT(!xfs_buf_geterror(*bpp));
1496 agi = XFS_BUF_TO_AGI(*bpp);
1499 * Validate the magic number of the agi block.
1501 agi_ok = agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC) &&
1502 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1503 be32_to_cpu(agi->agi_seqno) == agno;
1504 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1505 XFS_RANDOM_IALLOC_READ_AGI))) {
1506 XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1508 xfs_trans_brelse(tp, *bpp);
1509 return XFS_ERROR(EFSCORRUPTED);
1512 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1514 xfs_check_agi_unlinked(agi);
1519 xfs_ialloc_read_agi(
1520 struct xfs_mount *mp, /* file system mount structure */
1521 struct xfs_trans *tp, /* transaction pointer */
1522 xfs_agnumber_t agno, /* allocation group number */
1523 struct xfs_buf **bpp) /* allocation group hdr buf */
1525 struct xfs_agi *agi; /* allocation group header */
1526 struct xfs_perag *pag; /* per allocation group data */
1529 error = xfs_read_agi(mp, tp, agno, bpp);
1533 agi = XFS_BUF_TO_AGI(*bpp);
1534 pag = xfs_perag_get(mp, agno);
1535 if (!pag->pagi_init) {
1536 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1537 pag->pagi_count = be32_to_cpu(agi->agi_count);
1542 * It's possible for these to be out of sync if
1543 * we are in the middle of a forced shutdown.
1545 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1546 XFS_FORCED_SHUTDOWN(mp));
1552 * Read in the agi to initialise the per-ag data in the mount structure
1555 xfs_ialloc_pagi_init(
1556 xfs_mount_t *mp, /* file system mount structure */
1557 xfs_trans_t *tp, /* transaction pointer */
1558 xfs_agnumber_t agno) /* allocation group number */
1560 xfs_buf_t *bp = NULL;
1563 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1567 xfs_trans_brelse(tp, bp);