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,
208 * Initialize all inodes in this buffer and then log them.
210 * XXX: It would be much better if we had just one transaction
211 * to log a whole cluster of inodes instead of all the
212 * individual transactions causing a lot of log traffic.
214 fbuf->b_ops = &xfs_inode_buf_ops;
215 xfs_buf_zero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
216 for (i = 0; i < ninodes; i++) {
217 int ioffset = i << mp->m_sb.sb_inodelog;
218 uint isize = sizeof(struct xfs_dinode);
220 free = xfs_make_iptr(mp, fbuf, i);
221 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
222 free->di_version = version;
223 free->di_gen = cpu_to_be32(gen);
224 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
225 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
227 xfs_trans_inode_alloc_buf(tp, fbuf);
233 * Allocate new inodes in the allocation group specified by agbp.
234 * Return 0 for success, else error code.
236 STATIC int /* error code or 0 */
238 xfs_trans_t *tp, /* transaction pointer */
239 xfs_buf_t *agbp, /* alloc group buffer */
242 xfs_agi_t *agi; /* allocation group header */
243 xfs_alloc_arg_t args; /* allocation argument structure */
244 xfs_btree_cur_t *cur; /* inode btree cursor */
248 xfs_agino_t newino; /* new first inode's number */
249 xfs_agino_t newlen; /* new number of inodes */
250 xfs_agino_t thisino; /* current inode number, for loop */
251 int isaligned = 0; /* inode allocation at stripe unit */
253 struct xfs_perag *pag;
255 memset(&args, 0, sizeof(args));
257 args.mp = tp->t_mountp;
260 * Locking will ensure that we don't have two callers in here
263 newlen = XFS_IALLOC_INODES(args.mp);
264 if (args.mp->m_maxicount &&
265 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
266 return XFS_ERROR(ENOSPC);
267 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
269 * First try to allocate inodes contiguous with the last-allocated
270 * chunk of inodes. If the filesystem is striped, this will fill
271 * an entire stripe unit with inodes.
273 agi = XFS_BUF_TO_AGI(agbp);
274 newino = be32_to_cpu(agi->agi_newino);
275 agno = be32_to_cpu(agi->agi_seqno);
276 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
277 XFS_IALLOC_BLOCKS(args.mp);
278 if (likely(newino != NULLAGINO &&
279 (args.agbno < be32_to_cpu(agi->agi_length)))) {
280 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
281 args.type = XFS_ALLOCTYPE_THIS_BNO;
282 args.mod = args.total = args.wasdel = args.isfl =
283 args.userdata = args.minalignslop = 0;
287 * We need to take into account alignment here to ensure that
288 * we don't modify the free list if we fail to have an exact
289 * block. If we don't have an exact match, and every oher
290 * attempt allocation attempt fails, we'll end up cancelling
291 * a dirty transaction and shutting down.
293 * For an exact allocation, alignment must be 1,
294 * however we need to take cluster alignment into account when
295 * fixing up the freelist. Use the minalignslop field to
296 * indicate that extra blocks might be required for alignment,
297 * but not to use them in the actual exact allocation.
300 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
302 /* Allow space for the inode btree to split. */
303 args.minleft = args.mp->m_in_maxlevels - 1;
304 if ((error = xfs_alloc_vextent(&args)))
307 args.fsbno = NULLFSBLOCK;
309 if (unlikely(args.fsbno == NULLFSBLOCK)) {
311 * Set the alignment for the allocation.
312 * If stripe alignment is turned on then align at stripe unit
314 * If the cluster size is smaller than a filesystem block
315 * then we're doing I/O for inodes in filesystem block size
316 * pieces, so don't need alignment anyway.
319 if (args.mp->m_sinoalign) {
320 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
321 args.alignment = args.mp->m_dalign;
324 args.alignment = xfs_ialloc_cluster_alignment(&args);
326 * Need to figure out where to allocate the inode blocks.
327 * Ideally they should be spaced out through the a.g.
328 * For now, just allocate blocks up front.
330 args.agbno = be32_to_cpu(agi->agi_root);
331 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
333 * Allocate a fixed-size extent of inodes.
335 args.type = XFS_ALLOCTYPE_NEAR_BNO;
336 args.mod = args.total = args.wasdel = args.isfl =
337 args.userdata = args.minalignslop = 0;
340 * Allow space for the inode btree to split.
342 args.minleft = args.mp->m_in_maxlevels - 1;
343 if ((error = xfs_alloc_vextent(&args)))
348 * If stripe alignment is turned on, then try again with cluster
351 if (isaligned && args.fsbno == NULLFSBLOCK) {
352 args.type = XFS_ALLOCTYPE_NEAR_BNO;
353 args.agbno = be32_to_cpu(agi->agi_root);
354 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
355 args.alignment = xfs_ialloc_cluster_alignment(&args);
356 if ((error = xfs_alloc_vextent(&args)))
360 if (args.fsbno == NULLFSBLOCK) {
364 ASSERT(args.len == args.minlen);
367 * Stamp and write the inode buffers.
369 * Seed the new inode cluster with a random generation number. This
370 * prevents short-term reuse of generation numbers if a chunk is
371 * freed and then immediately reallocated. We use random numbers
372 * rather than a linear progression to prevent the next generation
373 * number from being easily guessable.
375 error = xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno,
376 args.len, random32());
381 * Convert the results.
383 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
384 be32_add_cpu(&agi->agi_count, newlen);
385 be32_add_cpu(&agi->agi_freecount, newlen);
386 pag = xfs_perag_get(args.mp, agno);
387 pag->pagi_freecount += newlen;
389 agi->agi_newino = cpu_to_be32(newino);
392 * Insert records describing the new inode chunk into the btree.
394 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
395 for (thisino = newino;
396 thisino < newino + newlen;
397 thisino += XFS_INODES_PER_CHUNK) {
398 cur->bc_rec.i.ir_startino = thisino;
399 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
400 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
401 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
403 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
407 error = xfs_btree_insert(cur, &i);
409 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
414 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
416 * Log allocation group header fields
418 xfs_ialloc_log_agi(tp, agbp,
419 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
421 * Modify/log superblock values for inode count and inode free count.
423 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
424 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
429 STATIC xfs_agnumber_t
435 spin_lock(&mp->m_agirotor_lock);
436 agno = mp->m_agirotor;
437 if (++mp->m_agirotor >= mp->m_maxagi)
439 spin_unlock(&mp->m_agirotor_lock);
445 * Select an allocation group to look for a free inode in, based on the parent
446 * inode and then mode. Return the allocation group buffer.
448 STATIC xfs_agnumber_t
449 xfs_ialloc_ag_select(
450 xfs_trans_t *tp, /* transaction pointer */
451 xfs_ino_t parent, /* parent directory inode number */
452 umode_t mode, /* bits set to indicate file type */
453 int okalloc) /* ok to allocate more space */
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 */
467 * Files of these types need at least one block if length > 0
468 * (and they won't fit in the inode, but that's hard to figure out).
470 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
472 agcount = mp->m_maxagi;
474 pagno = xfs_ialloc_next_ag(mp);
476 pagno = XFS_INO_TO_AGNO(mp, parent);
477 if (pagno >= agcount)
481 ASSERT(pagno < agcount);
484 * Loop through allocation groups, looking for one with a little
485 * free space in it. Note we don't look for free inodes, exactly.
486 * Instead, we include whether there is a need to allocate inodes
487 * to mean that blocks must be allocated for them,
488 * if none are currently free.
491 flags = XFS_ALLOC_FLAG_TRYLOCK;
493 pag = xfs_perag_get(mp, agno);
494 if (!pag->pagi_inodeok) {
495 xfs_ialloc_next_ag(mp);
499 if (!pag->pagi_init) {
500 error = xfs_ialloc_pagi_init(mp, tp, agno);
505 if (pag->pagi_freecount) {
513 if (!pag->pagf_init) {
514 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
520 * Is there enough free space for the file plus a block of
521 * inodes? (if we need to allocate some)?
523 ineed = XFS_IALLOC_BLOCKS(mp);
524 longest = pag->pagf_longest;
526 longest = pag->pagf_flcount > 0;
528 if (pag->pagf_freeblks >= needspace + ineed &&
536 * No point in iterating over the rest, if we're shutting
539 if (XFS_FORCED_SHUTDOWN(mp))
553 * Try to retrieve the next record to the left/right from the current one.
557 struct xfs_btree_cur *cur,
558 xfs_inobt_rec_incore_t *rec,
566 error = xfs_btree_decrement(cur, 0, &i);
568 error = xfs_btree_increment(cur, 0, &i);
574 error = xfs_inobt_get_rec(cur, rec, &i);
577 XFS_WANT_CORRUPTED_RETURN(i == 1);
585 struct xfs_btree_cur *cur,
587 xfs_inobt_rec_incore_t *rec,
594 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
599 error = xfs_inobt_get_rec(cur, rec, &i);
602 XFS_WANT_CORRUPTED_RETURN(i == 1);
611 * The caller selected an AG for us, and made sure that free inodes are
616 struct xfs_trans *tp,
617 struct xfs_buf *agbp,
621 struct xfs_mount *mp = tp->t_mountp;
622 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
623 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
624 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
625 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
626 struct xfs_perag *pag;
627 struct xfs_btree_cur *cur, *tcur;
628 struct xfs_inobt_rec_incore rec, trec;
634 pag = xfs_perag_get(mp, agno);
636 ASSERT(pag->pagi_init);
637 ASSERT(pag->pagi_inodeok);
638 ASSERT(pag->pagi_freecount > 0);
641 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
643 * If pagino is 0 (this is the root inode allocation) use newino.
644 * This must work because we've just allocated some.
647 pagino = be32_to_cpu(agi->agi_newino);
649 error = xfs_check_agi_freecount(cur, agi);
654 * If in the same AG as the parent, try to get near the parent.
657 int doneleft; /* done, to the left */
658 int doneright; /* done, to the right */
659 int searchdistance = 10;
661 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
664 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
666 error = xfs_inobt_get_rec(cur, &rec, &j);
669 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
671 if (rec.ir_freecount > 0) {
673 * Found a free inode in the same chunk
674 * as the parent, done.
681 * In the same AG as parent, but parent's chunk is full.
684 /* duplicate the cursor, search left & right simultaneously */
685 error = xfs_btree_dup_cursor(cur, &tcur);
690 * Skip to last blocks looked up if same parent inode.
692 if (pagino != NULLAGINO &&
693 pag->pagl_pagino == pagino &&
694 pag->pagl_leftrec != NULLAGINO &&
695 pag->pagl_rightrec != NULLAGINO) {
696 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
697 &trec, &doneleft, 1);
701 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
702 &rec, &doneright, 0);
706 /* search left with tcur, back up 1 record */
707 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
711 /* search right with cur, go forward 1 record. */
712 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
718 * Loop until we find an inode chunk with a free inode.
720 while (!doneleft || !doneright) {
721 int useleft; /* using left inode chunk this time */
723 if (!--searchdistance) {
725 * Not in range - save last search
726 * location and allocate a new inode
728 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
729 pag->pagl_leftrec = trec.ir_startino;
730 pag->pagl_rightrec = rec.ir_startino;
731 pag->pagl_pagino = pagino;
735 /* figure out the closer block if both are valid. */
736 if (!doneleft && !doneright) {
738 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
739 rec.ir_startino - pagino;
744 /* free inodes to the left? */
745 if (useleft && trec.ir_freecount) {
747 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
750 pag->pagl_leftrec = trec.ir_startino;
751 pag->pagl_rightrec = rec.ir_startino;
752 pag->pagl_pagino = pagino;
756 /* free inodes to the right? */
757 if (!useleft && rec.ir_freecount) {
758 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
760 pag->pagl_leftrec = trec.ir_startino;
761 pag->pagl_rightrec = rec.ir_startino;
762 pag->pagl_pagino = pagino;
766 /* get next record to check */
768 error = xfs_ialloc_next_rec(tcur, &trec,
771 error = xfs_ialloc_next_rec(cur, &rec,
779 * We've reached the end of the btree. because
780 * we are only searching a small chunk of the
781 * btree each search, there is obviously free
782 * inodes closer to the parent inode than we
783 * are now. restart the search again.
785 pag->pagl_pagino = NULLAGINO;
786 pag->pagl_leftrec = NULLAGINO;
787 pag->pagl_rightrec = NULLAGINO;
788 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
789 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
794 * In a different AG from the parent.
795 * See if the most recently allocated block has any free.
798 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
799 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
805 error = xfs_inobt_get_rec(cur, &rec, &j);
809 if (j == 1 && rec.ir_freecount > 0) {
811 * The last chunk allocated in the group
812 * still has a free inode.
820 * None left in the last group, search the whole AG
822 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
825 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
828 error = xfs_inobt_get_rec(cur, &rec, &i);
831 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
832 if (rec.ir_freecount > 0)
834 error = xfs_btree_increment(cur, 0, &i);
837 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
841 offset = xfs_lowbit64(rec.ir_free);
843 ASSERT(offset < XFS_INODES_PER_CHUNK);
844 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
845 XFS_INODES_PER_CHUNK) == 0);
846 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
847 rec.ir_free &= ~XFS_INOBT_MASK(offset);
849 error = xfs_inobt_update(cur, &rec);
852 be32_add_cpu(&agi->agi_freecount, -1);
853 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
854 pag->pagi_freecount--;
856 error = xfs_check_agi_freecount(cur, agi);
860 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
861 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
866 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
868 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
874 * Allocate an inode on disk.
876 * Mode is used to tell whether the new inode will need space, and whether it
879 * This function is designed to be called twice if it has to do an allocation
880 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
881 * If an inode is available without having to performn an allocation, an inode
882 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
883 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
884 * The caller should then commit the current transaction, allocate a
885 * new transaction, and call xfs_dialloc() again, passing in the previous value
886 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
887 * buffer is locked across the two calls, the second call is guaranteed to have
888 * a free inode available.
890 * Once we successfully pick an inode its number is returned and the on-disk
891 * data structures are updated. The inode itself is not read in, since doing so
892 * would break ordering constraints with xfs_reclaim.
896 struct xfs_trans *tp,
900 struct xfs_buf **IO_agbp,
903 struct xfs_mount *mp = tp->t_mountp;
904 struct xfs_buf *agbp;
909 xfs_agnumber_t start_agno;
910 struct xfs_perag *pag;
914 * If the caller passes in a pointer to the AGI buffer,
915 * continue where we left off before. In this case, we
916 * know that the allocation group has free inodes.
923 * We do not have an agbp, so select an initial allocation
924 * group for inode allocation.
926 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
927 if (start_agno == NULLAGNUMBER) {
933 * If we have already hit the ceiling of inode blocks then clear
934 * okalloc so we scan all available agi structures for a free
937 if (mp->m_maxicount &&
938 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
944 * Loop until we find an allocation group that either has free inodes
945 * or in which we can allocate some inodes. Iterate through the
946 * allocation groups upward, wrapping at the end.
950 pag = xfs_perag_get(mp, agno);
951 if (!pag->pagi_inodeok) {
952 xfs_ialloc_next_ag(mp);
956 if (!pag->pagi_init) {
957 error = xfs_ialloc_pagi_init(mp, tp, agno);
963 * Do a first racy fast path check if this AG is usable.
965 if (!pag->pagi_freecount && !okalloc)
969 * Then read in the AGI buffer and recheck with the AGI buffer
972 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
976 if (pag->pagi_freecount) {
982 goto nextag_relse_buffer;
985 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
987 xfs_trans_brelse(tp, agbp);
999 * We successfully allocated some inodes, return
1000 * the current context to the caller so that it
1001 * can commit the current transaction and call
1002 * us again where we left off.
1004 ASSERT(pag->pagi_freecount > 0);
1012 nextag_relse_buffer:
1013 xfs_trans_brelse(tp, agbp);
1016 if (++agno == mp->m_sb.sb_agcount)
1018 if (agno == start_agno) {
1020 return noroom ? ENOSPC : 0;
1026 return xfs_dialloc_ag(tp, agbp, parent, inop);
1029 return XFS_ERROR(error);
1033 * Free disk inode. Carefully avoids touching the incore inode, all
1034 * manipulations incore are the caller's responsibility.
1035 * The on-disk inode is not changed by this operation, only the
1036 * btree (free inode mask) is changed.
1040 xfs_trans_t *tp, /* transaction pointer */
1041 xfs_ino_t inode, /* inode to be freed */
1042 xfs_bmap_free_t *flist, /* extents to free */
1043 int *delete, /* set if inode cluster was deleted */
1044 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1047 xfs_agblock_t agbno; /* block number containing inode */
1048 xfs_buf_t *agbp; /* buffer containing allocation group header */
1049 xfs_agino_t agino; /* inode number relative to allocation group */
1050 xfs_agnumber_t agno; /* allocation group number */
1051 xfs_agi_t *agi; /* allocation group header */
1052 xfs_btree_cur_t *cur; /* inode btree cursor */
1053 int error; /* error return value */
1054 int i; /* result code */
1055 int ilen; /* inodes in an inode cluster */
1056 xfs_mount_t *mp; /* mount structure for filesystem */
1057 int off; /* offset of inode in inode chunk */
1058 xfs_inobt_rec_incore_t rec; /* btree record */
1059 struct xfs_perag *pag;
1064 * Break up inode number into its components.
1066 agno = XFS_INO_TO_AGNO(mp, inode);
1067 if (agno >= mp->m_sb.sb_agcount) {
1068 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1069 __func__, agno, mp->m_sb.sb_agcount);
1071 return XFS_ERROR(EINVAL);
1073 agino = XFS_INO_TO_AGINO(mp, inode);
1074 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1075 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1076 __func__, (unsigned long long)inode,
1077 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1079 return XFS_ERROR(EINVAL);
1081 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1082 if (agbno >= mp->m_sb.sb_agblocks) {
1083 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1084 __func__, agbno, mp->m_sb.sb_agblocks);
1086 return XFS_ERROR(EINVAL);
1089 * Get the allocation group header.
1091 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1093 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1097 agi = XFS_BUF_TO_AGI(agbp);
1098 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1099 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1101 * Initialize the cursor.
1103 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1105 error = xfs_check_agi_freecount(cur, agi);
1110 * Look for the entry describing this inode.
1112 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1113 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1117 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1118 error = xfs_inobt_get_rec(cur, &rec, &i);
1120 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1124 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1126 * Get the offset in the inode chunk.
1128 off = agino - rec.ir_startino;
1129 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1130 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1132 * Mark the inode free & increment the count.
1134 rec.ir_free |= XFS_INOBT_MASK(off);
1138 * When an inode cluster is free, it becomes eligible for removal
1140 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1141 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1144 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1147 * Remove the inode cluster from the AGI B+Tree, adjust the
1148 * AGI and Superblock inode counts, and mark the disk space
1149 * to be freed when the transaction is committed.
1151 ilen = XFS_IALLOC_INODES(mp);
1152 be32_add_cpu(&agi->agi_count, -ilen);
1153 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1154 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1155 pag = xfs_perag_get(mp, agno);
1156 pag->pagi_freecount -= ilen - 1;
1158 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1159 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1161 if ((error = xfs_btree_delete(cur, &i))) {
1162 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1167 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1168 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1169 XFS_IALLOC_BLOCKS(mp), flist, mp);
1173 error = xfs_inobt_update(cur, &rec);
1175 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1181 * Change the inode free counts and log the ag/sb changes.
1183 be32_add_cpu(&agi->agi_freecount, 1);
1184 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1185 pag = xfs_perag_get(mp, agno);
1186 pag->pagi_freecount++;
1188 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1191 error = xfs_check_agi_freecount(cur, agi);
1195 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1199 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1205 struct xfs_mount *mp,
1206 struct xfs_trans *tp,
1207 xfs_agnumber_t agno,
1209 xfs_agblock_t agbno,
1210 xfs_agblock_t *chunk_agbno,
1211 xfs_agblock_t *offset_agbno,
1214 struct xfs_inobt_rec_incore rec;
1215 struct xfs_btree_cur *cur;
1216 struct xfs_buf *agbp;
1220 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1223 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1224 __func__, error, agno);
1229 * Lookup the inode record for the given agino. If the record cannot be
1230 * found, then it's an invalid inode number and we should abort. Once
1231 * we have a record, we need to ensure it contains the inode number
1232 * we are looking up.
1234 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1235 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1238 error = xfs_inobt_get_rec(cur, &rec, &i);
1239 if (!error && i == 0)
1243 xfs_trans_brelse(tp, agbp);
1244 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1248 /* check that the returned record contains the required inode */
1249 if (rec.ir_startino > agino ||
1250 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1253 /* for untrusted inodes check it is allocated first */
1254 if ((flags & XFS_IGET_UNTRUSTED) &&
1255 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1258 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1259 *offset_agbno = agbno - *chunk_agbno;
1264 * Return the location of the inode in imap, for mapping it into a buffer.
1268 xfs_mount_t *mp, /* file system mount structure */
1269 xfs_trans_t *tp, /* transaction pointer */
1270 xfs_ino_t ino, /* inode to locate */
1271 struct xfs_imap *imap, /* location map structure */
1272 uint flags) /* flags for inode btree lookup */
1274 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1275 xfs_agino_t agino; /* inode number within alloc group */
1276 xfs_agnumber_t agno; /* allocation group number */
1277 int blks_per_cluster; /* num blocks per inode cluster */
1278 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1279 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1280 int error; /* error code */
1281 int offset; /* index of inode in its buffer */
1282 int offset_agbno; /* blks from chunk start to inode */
1284 ASSERT(ino != NULLFSINO);
1287 * Split up the inode number into its parts.
1289 agno = XFS_INO_TO_AGNO(mp, ino);
1290 agino = XFS_INO_TO_AGINO(mp, ino);
1291 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1292 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1293 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1296 * Don't output diagnostic information for untrusted inodes
1297 * as they can be invalid without implying corruption.
1299 if (flags & XFS_IGET_UNTRUSTED)
1300 return XFS_ERROR(EINVAL);
1301 if (agno >= mp->m_sb.sb_agcount) {
1303 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1304 __func__, agno, mp->m_sb.sb_agcount);
1306 if (agbno >= mp->m_sb.sb_agblocks) {
1308 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1309 __func__, (unsigned long long)agbno,
1310 (unsigned long)mp->m_sb.sb_agblocks);
1312 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1314 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1316 XFS_AGINO_TO_INO(mp, agno, agino));
1320 return XFS_ERROR(EINVAL);
1323 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1326 * For bulkstat and handle lookups, we have an untrusted inode number
1327 * that we have to verify is valid. We cannot do this just by reading
1328 * the inode buffer as it may have been unlinked and removed leaving
1329 * inodes in stale state on disk. Hence we have to do a btree lookup
1330 * in all cases where an untrusted inode number is passed.
1332 if (flags & XFS_IGET_UNTRUSTED) {
1333 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1334 &chunk_agbno, &offset_agbno, flags);
1341 * If the inode cluster size is the same as the blocksize or
1342 * smaller we get to the buffer by simple arithmetics.
1344 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1345 offset = XFS_INO_TO_OFFSET(mp, ino);
1346 ASSERT(offset < mp->m_sb.sb_inopblock);
1348 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1349 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1350 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1355 * If the inode chunks are aligned then use simple maths to
1356 * find the location. Otherwise we have to do a btree
1357 * lookup to find the location.
1359 if (mp->m_inoalign_mask) {
1360 offset_agbno = agbno & mp->m_inoalign_mask;
1361 chunk_agbno = agbno - offset_agbno;
1363 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1364 &chunk_agbno, &offset_agbno, flags);
1370 ASSERT(agbno >= chunk_agbno);
1371 cluster_agbno = chunk_agbno +
1372 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1373 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1374 XFS_INO_TO_OFFSET(mp, ino);
1376 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1377 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1378 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1381 * If the inode number maps to a block outside the bounds
1382 * of the file system then return NULL rather than calling
1383 * read_buf and panicing when we get an error from the
1386 if ((imap->im_blkno + imap->im_len) >
1387 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1389 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1390 __func__, (unsigned long long) imap->im_blkno,
1391 (unsigned long long) imap->im_len,
1392 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1393 return XFS_ERROR(EINVAL);
1399 * Compute and fill in value of m_in_maxlevels.
1402 xfs_ialloc_compute_maxlevels(
1403 xfs_mount_t *mp) /* file system mount structure */
1411 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1412 XFS_INODES_PER_CHUNK_LOG;
1413 minleafrecs = mp->m_alloc_mnr[0];
1414 minnoderecs = mp->m_alloc_mnr[1];
1415 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1416 for (level = 1; maxblocks > 1; level++)
1417 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1418 mp->m_in_maxlevels = level;
1422 * Log specified fields for the ag hdr (inode section)
1426 xfs_trans_t *tp, /* transaction pointer */
1427 xfs_buf_t *bp, /* allocation group header buffer */
1428 int fields) /* bitmask of fields to log */
1430 int first; /* first byte number */
1431 int last; /* last byte number */
1432 static const short offsets[] = { /* field starting offsets */
1433 /* keep in sync with bit definitions */
1434 offsetof(xfs_agi_t, agi_magicnum),
1435 offsetof(xfs_agi_t, agi_versionnum),
1436 offsetof(xfs_agi_t, agi_seqno),
1437 offsetof(xfs_agi_t, agi_length),
1438 offsetof(xfs_agi_t, agi_count),
1439 offsetof(xfs_agi_t, agi_root),
1440 offsetof(xfs_agi_t, agi_level),
1441 offsetof(xfs_agi_t, agi_freecount),
1442 offsetof(xfs_agi_t, agi_newino),
1443 offsetof(xfs_agi_t, agi_dirino),
1444 offsetof(xfs_agi_t, agi_unlinked),
1448 xfs_agi_t *agi; /* allocation group header */
1450 agi = XFS_BUF_TO_AGI(bp);
1451 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1454 * Compute byte offsets for the first and last fields.
1456 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1458 * Log the allocation group inode header buffer.
1460 xfs_trans_log_buf(tp, bp, first, last);
1465 xfs_check_agi_unlinked(
1466 struct xfs_agi *agi)
1470 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1471 ASSERT(agi->agi_unlinked[i]);
1474 #define xfs_check_agi_unlinked(agi)
1481 struct xfs_mount *mp = bp->b_target->bt_mount;
1482 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
1486 * Validate the magic number of the agi block.
1488 agi_ok = agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC) &&
1489 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1492 * during growfs operations, the perag is not fully initialised,
1493 * so we can't use it for any useful checking. growfs ensures we can't
1494 * use it by using uncached buffers that don't have the perag attached
1495 * so we can detect and avoid this problem.
1498 agi_ok = agi_ok && be32_to_cpu(agi->agi_seqno) ==
1499 bp->b_pag->pag_agno;
1501 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1502 XFS_RANDOM_IALLOC_READ_AGI))) {
1503 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, agi);
1504 xfs_buf_ioerror(bp, EFSCORRUPTED);
1506 xfs_check_agi_unlinked(agi);
1510 xfs_agi_read_verify(
1517 xfs_agi_write_verify(
1523 const struct xfs_buf_ops xfs_agi_buf_ops = {
1524 .verify_read = xfs_agi_read_verify,
1525 .verify_write = xfs_agi_write_verify,
1529 * Read in the allocation group header (inode allocation section)
1533 struct xfs_mount *mp, /* file system mount structure */
1534 struct xfs_trans *tp, /* transaction pointer */
1535 xfs_agnumber_t agno, /* allocation group number */
1536 struct xfs_buf **bpp) /* allocation group hdr buf */
1540 ASSERT(agno != NULLAGNUMBER);
1542 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1543 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1544 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
1548 ASSERT(!xfs_buf_geterror(*bpp));
1549 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1554 xfs_ialloc_read_agi(
1555 struct xfs_mount *mp, /* file system mount structure */
1556 struct xfs_trans *tp, /* transaction pointer */
1557 xfs_agnumber_t agno, /* allocation group number */
1558 struct xfs_buf **bpp) /* allocation group hdr buf */
1560 struct xfs_agi *agi; /* allocation group header */
1561 struct xfs_perag *pag; /* per allocation group data */
1564 error = xfs_read_agi(mp, tp, agno, bpp);
1568 agi = XFS_BUF_TO_AGI(*bpp);
1569 pag = xfs_perag_get(mp, agno);
1570 if (!pag->pagi_init) {
1571 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1572 pag->pagi_count = be32_to_cpu(agi->agi_count);
1577 * It's possible for these to be out of sync if
1578 * we are in the middle of a forced shutdown.
1580 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1581 XFS_FORCED_SHUTDOWN(mp));
1587 * Read in the agi to initialise the per-ag data in the mount structure
1590 xfs_ialloc_pagi_init(
1591 xfs_mount_t *mp, /* file system mount structure */
1592 xfs_trans_t *tp, /* transaction pointer */
1593 xfs_agnumber_t agno) /* allocation group number */
1595 xfs_buf_t *bp = NULL;
1598 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1602 xfs_trans_brelse(tp, bp);