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"
39 #include "xfs_cksum.h"
40 #include "xfs_buf_item.h"
44 * Allocation group level functions.
47 xfs_ialloc_cluster_alignment(
48 xfs_alloc_arg_t *args)
50 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
51 args->mp->m_sb.sb_inoalignmt >=
52 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
53 return args->mp->m_sb.sb_inoalignmt;
58 * Lookup a record by ino in the btree given by cur.
62 struct xfs_btree_cur *cur, /* btree cursor */
63 xfs_agino_t ino, /* starting inode of chunk */
64 xfs_lookup_t dir, /* <=, >=, == */
65 int *stat) /* success/failure */
67 cur->bc_rec.i.ir_startino = ino;
68 cur->bc_rec.i.ir_freecount = 0;
69 cur->bc_rec.i.ir_free = 0;
70 return xfs_btree_lookup(cur, dir, stat);
74 * Update the record referred to by cur to the value given.
75 * This either works (return 0) or gets an EFSCORRUPTED error.
77 STATIC int /* error */
79 struct xfs_btree_cur *cur, /* btree cursor */
80 xfs_inobt_rec_incore_t *irec) /* btree record */
82 union xfs_btree_rec rec;
84 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
85 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
86 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
87 return xfs_btree_update(cur, &rec);
91 * Get the data from the pointed-to record.
95 struct xfs_btree_cur *cur, /* btree cursor */
96 xfs_inobt_rec_incore_t *irec, /* btree record */
97 int *stat) /* output: success/failure */
99 union xfs_btree_rec *rec;
102 error = xfs_btree_get_rec(cur, &rec, stat);
103 if (!error && *stat == 1) {
104 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
105 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
106 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
112 * Verify that the number of free inodes in the AGI is correct.
116 xfs_check_agi_freecount(
117 struct xfs_btree_cur *cur,
120 if (cur->bc_nlevels == 1) {
121 xfs_inobt_rec_incore_t rec;
126 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
131 error = xfs_inobt_get_rec(cur, &rec, &i);
136 freecount += rec.ir_freecount;
137 error = xfs_btree_increment(cur, 0, &i);
143 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
144 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
149 #define xfs_check_agi_freecount(cur, agi) 0
153 * Initialise a new set of inodes.
156 xfs_ialloc_inode_init(
157 struct xfs_mount *mp,
158 struct xfs_trans *tp,
161 xfs_agblock_t length,
164 struct xfs_buf *fbuf;
165 struct xfs_dinode *free;
166 int blks_per_cluster, nbufs, ninodes;
173 * Loop over the new block(s), filling in the inodes.
174 * For small block sizes, manipulate the inodes in buffers
175 * which are multiples of the blocks size.
177 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
178 blks_per_cluster = 1;
180 ninodes = mp->m_sb.sb_inopblock;
182 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
183 mp->m_sb.sb_blocksize;
184 nbufs = length / blks_per_cluster;
185 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
189 * Figure out what version number to use in the inodes we create. If
190 * the superblock version has caught up to the one that supports the new
191 * inode format, then use the new inode version. Otherwise use the old
192 * version so that old kernels will continue to be able to use the file
195 * For v3 inodes, we also need to write the inode number into the inode,
196 * so calculate the first inode number of the chunk here as
197 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
198 * across multiple filesystem blocks (such as a cluster) and so cannot
199 * be used in the cluster buffer loop below.
201 * Further, because we are writing the inode directly into the buffer
202 * and calculating a CRC on the entire inode, we have ot log the entire
203 * inode so that the entire range the CRC covers is present in the log.
204 * That means for v3 inode we log the entire buffer rather than just the
207 if (xfs_sb_version_hascrc(&mp->m_sb)) {
209 ino = XFS_AGINO_TO_INO(mp, agno,
210 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
211 } else if (xfs_sb_version_hasnlink(&mp->m_sb))
216 for (j = 0; j < nbufs; j++) {
220 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
221 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
222 mp->m_bsize * blks_per_cluster,
227 * Initialize all inodes in this buffer and then log them.
229 * XXX: It would be much better if we had just one transaction
230 * to log a whole cluster of inodes instead of all the
231 * individual transactions causing a lot of log traffic.
233 fbuf->b_ops = &xfs_inode_buf_ops;
234 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
235 for (i = 0; i < ninodes; i++) {
236 int ioffset = i << mp->m_sb.sb_inodelog;
237 uint isize = xfs_dinode_size(version);
239 free = xfs_make_iptr(mp, fbuf, i);
240 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
241 free->di_version = version;
242 free->di_gen = cpu_to_be32(gen);
243 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
246 free->di_ino = cpu_to_be64(ino);
248 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
249 xfs_dinode_calc_crc(mp, free);
251 /* just log the inode core */
252 xfs_trans_log_buf(tp, fbuf, ioffset,
253 ioffset + isize - 1);
257 /* need to log the entire buffer */
258 xfs_trans_log_buf(tp, fbuf, 0,
259 BBTOB(fbuf->b_length) - 1);
261 xfs_trans_inode_alloc_buf(tp, fbuf);
267 * Allocate new inodes in the allocation group specified by agbp.
268 * Return 0 for success, else error code.
270 STATIC int /* error code or 0 */
272 xfs_trans_t *tp, /* transaction pointer */
273 xfs_buf_t *agbp, /* alloc group buffer */
276 xfs_agi_t *agi; /* allocation group header */
277 xfs_alloc_arg_t args; /* allocation argument structure */
278 xfs_btree_cur_t *cur; /* inode btree cursor */
282 xfs_agino_t newino; /* new first inode's number */
283 xfs_agino_t newlen; /* new number of inodes */
284 xfs_agino_t thisino; /* current inode number, for loop */
285 int isaligned = 0; /* inode allocation at stripe unit */
287 struct xfs_perag *pag;
289 memset(&args, 0, sizeof(args));
291 args.mp = tp->t_mountp;
294 * Locking will ensure that we don't have two callers in here
297 newlen = XFS_IALLOC_INODES(args.mp);
298 if (args.mp->m_maxicount &&
299 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
300 return XFS_ERROR(ENOSPC);
301 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
303 * First try to allocate inodes contiguous with the last-allocated
304 * chunk of inodes. If the filesystem is striped, this will fill
305 * an entire stripe unit with inodes.
307 agi = XFS_BUF_TO_AGI(agbp);
308 newino = be32_to_cpu(agi->agi_newino);
309 agno = be32_to_cpu(agi->agi_seqno);
310 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
311 XFS_IALLOC_BLOCKS(args.mp);
312 if (likely(newino != NULLAGINO &&
313 (args.agbno < be32_to_cpu(agi->agi_length)))) {
314 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
315 args.type = XFS_ALLOCTYPE_THIS_BNO;
319 * We need to take into account alignment here to ensure that
320 * we don't modify the free list if we fail to have an exact
321 * block. If we don't have an exact match, and every oher
322 * attempt allocation attempt fails, we'll end up cancelling
323 * a dirty transaction and shutting down.
325 * For an exact allocation, alignment must be 1,
326 * however we need to take cluster alignment into account when
327 * fixing up the freelist. Use the minalignslop field to
328 * indicate that extra blocks might be required for alignment,
329 * but not to use them in the actual exact allocation.
332 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
334 /* Allow space for the inode btree to split. */
335 args.minleft = args.mp->m_in_maxlevels - 1;
336 if ((error = xfs_alloc_vextent(&args)))
339 args.fsbno = NULLFSBLOCK;
341 if (unlikely(args.fsbno == NULLFSBLOCK)) {
343 * Set the alignment for the allocation.
344 * If stripe alignment is turned on then align at stripe unit
346 * If the cluster size is smaller than a filesystem block
347 * then we're doing I/O for inodes in filesystem block size
348 * pieces, so don't need alignment anyway.
351 if (args.mp->m_sinoalign) {
352 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
353 args.alignment = args.mp->m_dalign;
356 args.alignment = xfs_ialloc_cluster_alignment(&args);
358 * Need to figure out where to allocate the inode blocks.
359 * Ideally they should be spaced out through the a.g.
360 * For now, just allocate blocks up front.
362 args.agbno = be32_to_cpu(agi->agi_root);
363 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
365 * Allocate a fixed-size extent of inodes.
367 args.type = XFS_ALLOCTYPE_NEAR_BNO;
370 * Allow space for the inode btree to split.
372 args.minleft = args.mp->m_in_maxlevels - 1;
373 if ((error = xfs_alloc_vextent(&args)))
378 * If stripe alignment is turned on, then try again with cluster
381 if (isaligned && args.fsbno == NULLFSBLOCK) {
382 args.type = XFS_ALLOCTYPE_NEAR_BNO;
383 args.agbno = be32_to_cpu(agi->agi_root);
384 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
385 args.alignment = xfs_ialloc_cluster_alignment(&args);
386 if ((error = xfs_alloc_vextent(&args)))
390 if (args.fsbno == NULLFSBLOCK) {
394 ASSERT(args.len == args.minlen);
397 * Stamp and write the inode buffers.
399 * Seed the new inode cluster with a random generation number. This
400 * prevents short-term reuse of generation numbers if a chunk is
401 * freed and then immediately reallocated. We use random numbers
402 * rather than a linear progression to prevent the next generation
403 * number from being easily guessable.
405 error = xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno,
406 args.len, prandom_u32());
411 * Convert the results.
413 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
414 be32_add_cpu(&agi->agi_count, newlen);
415 be32_add_cpu(&agi->agi_freecount, newlen);
416 pag = xfs_perag_get(args.mp, agno);
417 pag->pagi_freecount += newlen;
419 agi->agi_newino = cpu_to_be32(newino);
422 * Insert records describing the new inode chunk into the btree.
424 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
425 for (thisino = newino;
426 thisino < newino + newlen;
427 thisino += XFS_INODES_PER_CHUNK) {
428 cur->bc_rec.i.ir_startino = thisino;
429 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
430 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
431 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
433 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
437 error = xfs_btree_insert(cur, &i);
439 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
444 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
446 * Log allocation group header fields
448 xfs_ialloc_log_agi(tp, agbp,
449 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
451 * Modify/log superblock values for inode count and inode free count.
453 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
454 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
459 STATIC xfs_agnumber_t
465 spin_lock(&mp->m_agirotor_lock);
466 agno = mp->m_agirotor;
467 if (++mp->m_agirotor >= mp->m_maxagi)
469 spin_unlock(&mp->m_agirotor_lock);
475 * Select an allocation group to look for a free inode in, based on the parent
476 * inode and then mode. Return the allocation group buffer.
478 STATIC xfs_agnumber_t
479 xfs_ialloc_ag_select(
480 xfs_trans_t *tp, /* transaction pointer */
481 xfs_ino_t parent, /* parent directory inode number */
482 umode_t mode, /* bits set to indicate file type */
483 int okalloc) /* ok to allocate more space */
485 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
486 xfs_agnumber_t agno; /* current ag number */
487 int flags; /* alloc buffer locking flags */
488 xfs_extlen_t ineed; /* blocks needed for inode allocation */
489 xfs_extlen_t longest = 0; /* longest extent available */
490 xfs_mount_t *mp; /* mount point structure */
491 int needspace; /* file mode implies space allocated */
492 xfs_perag_t *pag; /* per allocation group data */
493 xfs_agnumber_t pagno; /* parent (starting) ag number */
497 * Files of these types need at least one block if length > 0
498 * (and they won't fit in the inode, but that's hard to figure out).
500 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
502 agcount = mp->m_maxagi;
504 pagno = xfs_ialloc_next_ag(mp);
506 pagno = XFS_INO_TO_AGNO(mp, parent);
507 if (pagno >= agcount)
511 ASSERT(pagno < agcount);
514 * Loop through allocation groups, looking for one with a little
515 * free space in it. Note we don't look for free inodes, exactly.
516 * Instead, we include whether there is a need to allocate inodes
517 * to mean that blocks must be allocated for them,
518 * if none are currently free.
521 flags = XFS_ALLOC_FLAG_TRYLOCK;
523 pag = xfs_perag_get(mp, agno);
524 if (!pag->pagi_inodeok) {
525 xfs_ialloc_next_ag(mp);
529 if (!pag->pagi_init) {
530 error = xfs_ialloc_pagi_init(mp, tp, agno);
535 if (pag->pagi_freecount) {
543 if (!pag->pagf_init) {
544 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
550 * Is there enough free space for the file plus a block of
551 * inodes? (if we need to allocate some)?
553 ineed = XFS_IALLOC_BLOCKS(mp);
554 longest = pag->pagf_longest;
556 longest = pag->pagf_flcount > 0;
558 if (pag->pagf_freeblks >= needspace + ineed &&
566 * No point in iterating over the rest, if we're shutting
569 if (XFS_FORCED_SHUTDOWN(mp))
583 * Try to retrieve the next record to the left/right from the current one.
587 struct xfs_btree_cur *cur,
588 xfs_inobt_rec_incore_t *rec,
596 error = xfs_btree_decrement(cur, 0, &i);
598 error = xfs_btree_increment(cur, 0, &i);
604 error = xfs_inobt_get_rec(cur, rec, &i);
607 XFS_WANT_CORRUPTED_RETURN(i == 1);
615 struct xfs_btree_cur *cur,
617 xfs_inobt_rec_incore_t *rec,
624 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
629 error = xfs_inobt_get_rec(cur, rec, &i);
632 XFS_WANT_CORRUPTED_RETURN(i == 1);
641 * The caller selected an AG for us, and made sure that free inodes are
646 struct xfs_trans *tp,
647 struct xfs_buf *agbp,
651 struct xfs_mount *mp = tp->t_mountp;
652 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
653 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
654 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
655 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
656 struct xfs_perag *pag;
657 struct xfs_btree_cur *cur, *tcur;
658 struct xfs_inobt_rec_incore rec, trec;
664 pag = xfs_perag_get(mp, agno);
666 ASSERT(pag->pagi_init);
667 ASSERT(pag->pagi_inodeok);
668 ASSERT(pag->pagi_freecount > 0);
671 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
673 * If pagino is 0 (this is the root inode allocation) use newino.
674 * This must work because we've just allocated some.
677 pagino = be32_to_cpu(agi->agi_newino);
679 error = xfs_check_agi_freecount(cur, agi);
684 * If in the same AG as the parent, try to get near the parent.
687 int doneleft; /* done, to the left */
688 int doneright; /* done, to the right */
689 int searchdistance = 10;
691 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
694 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
696 error = xfs_inobt_get_rec(cur, &rec, &j);
699 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
701 if (rec.ir_freecount > 0) {
703 * Found a free inode in the same chunk
704 * as the parent, done.
711 * In the same AG as parent, but parent's chunk is full.
714 /* duplicate the cursor, search left & right simultaneously */
715 error = xfs_btree_dup_cursor(cur, &tcur);
720 * Skip to last blocks looked up if same parent inode.
722 if (pagino != NULLAGINO &&
723 pag->pagl_pagino == pagino &&
724 pag->pagl_leftrec != NULLAGINO &&
725 pag->pagl_rightrec != NULLAGINO) {
726 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
727 &trec, &doneleft, 1);
731 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
732 &rec, &doneright, 0);
736 /* search left with tcur, back up 1 record */
737 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
741 /* search right with cur, go forward 1 record. */
742 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
748 * Loop until we find an inode chunk with a free inode.
750 while (!doneleft || !doneright) {
751 int useleft; /* using left inode chunk this time */
753 if (!--searchdistance) {
755 * Not in range - save last search
756 * location and allocate a new inode
758 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
759 pag->pagl_leftrec = trec.ir_startino;
760 pag->pagl_rightrec = rec.ir_startino;
761 pag->pagl_pagino = pagino;
765 /* figure out the closer block if both are valid. */
766 if (!doneleft && !doneright) {
768 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
769 rec.ir_startino - pagino;
774 /* free inodes to the left? */
775 if (useleft && trec.ir_freecount) {
777 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
780 pag->pagl_leftrec = trec.ir_startino;
781 pag->pagl_rightrec = rec.ir_startino;
782 pag->pagl_pagino = pagino;
786 /* free inodes to the right? */
787 if (!useleft && rec.ir_freecount) {
788 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
790 pag->pagl_leftrec = trec.ir_startino;
791 pag->pagl_rightrec = rec.ir_startino;
792 pag->pagl_pagino = pagino;
796 /* get next record to check */
798 error = xfs_ialloc_next_rec(tcur, &trec,
801 error = xfs_ialloc_next_rec(cur, &rec,
809 * We've reached the end of the btree. because
810 * we are only searching a small chunk of the
811 * btree each search, there is obviously free
812 * inodes closer to the parent inode than we
813 * are now. restart the search again.
815 pag->pagl_pagino = NULLAGINO;
816 pag->pagl_leftrec = NULLAGINO;
817 pag->pagl_rightrec = NULLAGINO;
818 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
819 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
824 * In a different AG from the parent.
825 * See if the most recently allocated block has any free.
828 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
829 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
835 error = xfs_inobt_get_rec(cur, &rec, &j);
839 if (j == 1 && rec.ir_freecount > 0) {
841 * The last chunk allocated in the group
842 * still has a free inode.
850 * None left in the last group, search the whole AG
852 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
855 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
858 error = xfs_inobt_get_rec(cur, &rec, &i);
861 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
862 if (rec.ir_freecount > 0)
864 error = xfs_btree_increment(cur, 0, &i);
867 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
871 offset = xfs_lowbit64(rec.ir_free);
873 ASSERT(offset < XFS_INODES_PER_CHUNK);
874 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
875 XFS_INODES_PER_CHUNK) == 0);
876 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
877 rec.ir_free &= ~XFS_INOBT_MASK(offset);
879 error = xfs_inobt_update(cur, &rec);
882 be32_add_cpu(&agi->agi_freecount, -1);
883 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
884 pag->pagi_freecount--;
886 error = xfs_check_agi_freecount(cur, agi);
890 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
891 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
896 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
898 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
904 * Allocate an inode on disk.
906 * Mode is used to tell whether the new inode will need space, and whether it
909 * This function is designed to be called twice if it has to do an allocation
910 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
911 * If an inode is available without having to performn an allocation, an inode
912 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
913 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
914 * The caller should then commit the current transaction, allocate a
915 * new transaction, and call xfs_dialloc() again, passing in the previous value
916 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
917 * buffer is locked across the two calls, the second call is guaranteed to have
918 * a free inode available.
920 * Once we successfully pick an inode its number is returned and the on-disk
921 * data structures are updated. The inode itself is not read in, since doing so
922 * would break ordering constraints with xfs_reclaim.
926 struct xfs_trans *tp,
930 struct xfs_buf **IO_agbp,
933 struct xfs_mount *mp = tp->t_mountp;
934 struct xfs_buf *agbp;
939 xfs_agnumber_t start_agno;
940 struct xfs_perag *pag;
944 * If the caller passes in a pointer to the AGI buffer,
945 * continue where we left off before. In this case, we
946 * know that the allocation group has free inodes.
953 * We do not have an agbp, so select an initial allocation
954 * group for inode allocation.
956 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
957 if (start_agno == NULLAGNUMBER) {
963 * If we have already hit the ceiling of inode blocks then clear
964 * okalloc so we scan all available agi structures for a free
967 if (mp->m_maxicount &&
968 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
974 * Loop until we find an allocation group that either has free inodes
975 * or in which we can allocate some inodes. Iterate through the
976 * allocation groups upward, wrapping at the end.
980 pag = xfs_perag_get(mp, agno);
981 if (!pag->pagi_inodeok) {
982 xfs_ialloc_next_ag(mp);
986 if (!pag->pagi_init) {
987 error = xfs_ialloc_pagi_init(mp, tp, agno);
993 * Do a first racy fast path check if this AG is usable.
995 if (!pag->pagi_freecount && !okalloc)
999 * Then read in the AGI buffer and recheck with the AGI buffer
1002 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1006 if (pag->pagi_freecount) {
1012 goto nextag_relse_buffer;
1015 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1017 xfs_trans_brelse(tp, agbp);
1019 if (error != ENOSPC)
1029 * We successfully allocated some inodes, return
1030 * the current context to the caller so that it
1031 * can commit the current transaction and call
1032 * us again where we left off.
1034 ASSERT(pag->pagi_freecount > 0);
1042 nextag_relse_buffer:
1043 xfs_trans_brelse(tp, agbp);
1046 if (++agno == mp->m_sb.sb_agcount)
1048 if (agno == start_agno) {
1050 return noroom ? ENOSPC : 0;
1056 return xfs_dialloc_ag(tp, agbp, parent, inop);
1059 return XFS_ERROR(error);
1063 * Free disk inode. Carefully avoids touching the incore inode, all
1064 * manipulations incore are the caller's responsibility.
1065 * The on-disk inode is not changed by this operation, only the
1066 * btree (free inode mask) is changed.
1070 xfs_trans_t *tp, /* transaction pointer */
1071 xfs_ino_t inode, /* inode to be freed */
1072 xfs_bmap_free_t *flist, /* extents to free */
1073 int *delete, /* set if inode cluster was deleted */
1074 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1077 xfs_agblock_t agbno; /* block number containing inode */
1078 xfs_buf_t *agbp; /* buffer containing allocation group header */
1079 xfs_agino_t agino; /* inode number relative to allocation group */
1080 xfs_agnumber_t agno; /* allocation group number */
1081 xfs_agi_t *agi; /* allocation group header */
1082 xfs_btree_cur_t *cur; /* inode btree cursor */
1083 int error; /* error return value */
1084 int i; /* result code */
1085 int ilen; /* inodes in an inode cluster */
1086 xfs_mount_t *mp; /* mount structure for filesystem */
1087 int off; /* offset of inode in inode chunk */
1088 xfs_inobt_rec_incore_t rec; /* btree record */
1089 struct xfs_perag *pag;
1094 * Break up inode number into its components.
1096 agno = XFS_INO_TO_AGNO(mp, inode);
1097 if (agno >= mp->m_sb.sb_agcount) {
1098 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1099 __func__, agno, mp->m_sb.sb_agcount);
1101 return XFS_ERROR(EINVAL);
1103 agino = XFS_INO_TO_AGINO(mp, inode);
1104 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1105 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1106 __func__, (unsigned long long)inode,
1107 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1109 return XFS_ERROR(EINVAL);
1111 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1112 if (agbno >= mp->m_sb.sb_agblocks) {
1113 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1114 __func__, agbno, mp->m_sb.sb_agblocks);
1116 return XFS_ERROR(EINVAL);
1119 * Get the allocation group header.
1121 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1123 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1127 agi = XFS_BUF_TO_AGI(agbp);
1128 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1129 ASSERT(agbno < be32_to_cpu(agi->agi_length));
1131 * Initialize the cursor.
1133 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1135 error = xfs_check_agi_freecount(cur, agi);
1140 * Look for the entry describing this inode.
1142 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1143 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1147 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1148 error = xfs_inobt_get_rec(cur, &rec, &i);
1150 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1154 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1156 * Get the offset in the inode chunk.
1158 off = agino - rec.ir_startino;
1159 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1160 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1162 * Mark the inode free & increment the count.
1164 rec.ir_free |= XFS_INOBT_MASK(off);
1168 * When an inode cluster is free, it becomes eligible for removal
1170 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1171 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1174 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1177 * Remove the inode cluster from the AGI B+Tree, adjust the
1178 * AGI and Superblock inode counts, and mark the disk space
1179 * to be freed when the transaction is committed.
1181 ilen = XFS_IALLOC_INODES(mp);
1182 be32_add_cpu(&agi->agi_count, -ilen);
1183 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1184 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1185 pag = xfs_perag_get(mp, agno);
1186 pag->pagi_freecount -= ilen - 1;
1188 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1189 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1191 if ((error = xfs_btree_delete(cur, &i))) {
1192 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1197 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1198 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1199 XFS_IALLOC_BLOCKS(mp), flist, mp);
1203 error = xfs_inobt_update(cur, &rec);
1205 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1211 * Change the inode free counts and log the ag/sb changes.
1213 be32_add_cpu(&agi->agi_freecount, 1);
1214 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1215 pag = xfs_perag_get(mp, agno);
1216 pag->pagi_freecount++;
1218 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1221 error = xfs_check_agi_freecount(cur, agi);
1225 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1229 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1235 struct xfs_mount *mp,
1236 struct xfs_trans *tp,
1237 xfs_agnumber_t agno,
1239 xfs_agblock_t agbno,
1240 xfs_agblock_t *chunk_agbno,
1241 xfs_agblock_t *offset_agbno,
1244 struct xfs_inobt_rec_incore rec;
1245 struct xfs_btree_cur *cur;
1246 struct xfs_buf *agbp;
1250 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1253 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1254 __func__, error, agno);
1259 * Lookup the inode record for the given agino. If the record cannot be
1260 * found, then it's an invalid inode number and we should abort. Once
1261 * we have a record, we need to ensure it contains the inode number
1262 * we are looking up.
1264 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1265 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1268 error = xfs_inobt_get_rec(cur, &rec, &i);
1269 if (!error && i == 0)
1273 xfs_trans_brelse(tp, agbp);
1274 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1278 /* check that the returned record contains the required inode */
1279 if (rec.ir_startino > agino ||
1280 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1283 /* for untrusted inodes check it is allocated first */
1284 if ((flags & XFS_IGET_UNTRUSTED) &&
1285 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1288 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1289 *offset_agbno = agbno - *chunk_agbno;
1294 * Return the location of the inode in imap, for mapping it into a buffer.
1298 xfs_mount_t *mp, /* file system mount structure */
1299 xfs_trans_t *tp, /* transaction pointer */
1300 xfs_ino_t ino, /* inode to locate */
1301 struct xfs_imap *imap, /* location map structure */
1302 uint flags) /* flags for inode btree lookup */
1304 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1305 xfs_agino_t agino; /* inode number within alloc group */
1306 xfs_agnumber_t agno; /* allocation group number */
1307 int blks_per_cluster; /* num blocks per inode cluster */
1308 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1309 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1310 int error; /* error code */
1311 int offset; /* index of inode in its buffer */
1312 int offset_agbno; /* blks from chunk start to inode */
1314 ASSERT(ino != NULLFSINO);
1317 * Split up the inode number into its parts.
1319 agno = XFS_INO_TO_AGNO(mp, ino);
1320 agino = XFS_INO_TO_AGINO(mp, ino);
1321 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1322 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1323 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1326 * Don't output diagnostic information for untrusted inodes
1327 * as they can be invalid without implying corruption.
1329 if (flags & XFS_IGET_UNTRUSTED)
1330 return XFS_ERROR(EINVAL);
1331 if (agno >= mp->m_sb.sb_agcount) {
1333 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1334 __func__, agno, mp->m_sb.sb_agcount);
1336 if (agbno >= mp->m_sb.sb_agblocks) {
1338 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1339 __func__, (unsigned long long)agbno,
1340 (unsigned long)mp->m_sb.sb_agblocks);
1342 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1344 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1346 XFS_AGINO_TO_INO(mp, agno, agino));
1350 return XFS_ERROR(EINVAL);
1353 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1356 * For bulkstat and handle lookups, we have an untrusted inode number
1357 * that we have to verify is valid. We cannot do this just by reading
1358 * the inode buffer as it may have been unlinked and removed leaving
1359 * inodes in stale state on disk. Hence we have to do a btree lookup
1360 * in all cases where an untrusted inode number is passed.
1362 if (flags & XFS_IGET_UNTRUSTED) {
1363 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1364 &chunk_agbno, &offset_agbno, flags);
1371 * If the inode cluster size is the same as the blocksize or
1372 * smaller we get to the buffer by simple arithmetics.
1374 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1375 offset = XFS_INO_TO_OFFSET(mp, ino);
1376 ASSERT(offset < mp->m_sb.sb_inopblock);
1378 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1379 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1380 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1385 * If the inode chunks are aligned then use simple maths to
1386 * find the location. Otherwise we have to do a btree
1387 * lookup to find the location.
1389 if (mp->m_inoalign_mask) {
1390 offset_agbno = agbno & mp->m_inoalign_mask;
1391 chunk_agbno = agbno - offset_agbno;
1393 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1394 &chunk_agbno, &offset_agbno, flags);
1400 ASSERT(agbno >= chunk_agbno);
1401 cluster_agbno = chunk_agbno +
1402 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1403 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1404 XFS_INO_TO_OFFSET(mp, ino);
1406 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1407 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1408 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1411 * If the inode number maps to a block outside the bounds
1412 * of the file system then return NULL rather than calling
1413 * read_buf and panicing when we get an error from the
1416 if ((imap->im_blkno + imap->im_len) >
1417 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1419 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1420 __func__, (unsigned long long) imap->im_blkno,
1421 (unsigned long long) imap->im_len,
1422 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1423 return XFS_ERROR(EINVAL);
1429 * Compute and fill in value of m_in_maxlevels.
1432 xfs_ialloc_compute_maxlevels(
1433 xfs_mount_t *mp) /* file system mount structure */
1441 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1442 XFS_INODES_PER_CHUNK_LOG;
1443 minleafrecs = mp->m_alloc_mnr[0];
1444 minnoderecs = mp->m_alloc_mnr[1];
1445 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1446 for (level = 1; maxblocks > 1; level++)
1447 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1448 mp->m_in_maxlevels = level;
1452 * Log specified fields for the ag hdr (inode section)
1456 xfs_trans_t *tp, /* transaction pointer */
1457 xfs_buf_t *bp, /* allocation group header buffer */
1458 int fields) /* bitmask of fields to log */
1460 int first; /* first byte number */
1461 int last; /* last byte number */
1462 static const short offsets[] = { /* field starting offsets */
1463 /* keep in sync with bit definitions */
1464 offsetof(xfs_agi_t, agi_magicnum),
1465 offsetof(xfs_agi_t, agi_versionnum),
1466 offsetof(xfs_agi_t, agi_seqno),
1467 offsetof(xfs_agi_t, agi_length),
1468 offsetof(xfs_agi_t, agi_count),
1469 offsetof(xfs_agi_t, agi_root),
1470 offsetof(xfs_agi_t, agi_level),
1471 offsetof(xfs_agi_t, agi_freecount),
1472 offsetof(xfs_agi_t, agi_newino),
1473 offsetof(xfs_agi_t, agi_dirino),
1474 offsetof(xfs_agi_t, agi_unlinked),
1478 xfs_agi_t *agi; /* allocation group header */
1480 agi = XFS_BUF_TO_AGI(bp);
1481 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1484 * Compute byte offsets for the first and last fields.
1486 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1488 * Log the allocation group inode header buffer.
1490 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
1491 xfs_trans_log_buf(tp, bp, first, last);
1496 xfs_check_agi_unlinked(
1497 struct xfs_agi *agi)
1501 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1502 ASSERT(agi->agi_unlinked[i]);
1505 #define xfs_check_agi_unlinked(agi)
1512 struct xfs_mount *mp = bp->b_target->bt_mount;
1513 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
1515 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1516 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
1519 * Validate the magic number of the agi block.
1521 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
1523 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
1527 * during growfs operations, the perag is not fully initialised,
1528 * so we can't use it for any useful checking. growfs ensures we can't
1529 * use it by using uncached buffers that don't have the perag attached
1530 * so we can detect and avoid this problem.
1532 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
1535 xfs_check_agi_unlinked(agi);
1540 xfs_agi_read_verify(
1543 struct xfs_mount *mp = bp->b_target->bt_mount;
1546 if (xfs_sb_version_hascrc(&mp->m_sb))
1547 agi_ok = xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
1548 offsetof(struct xfs_agi, agi_crc));
1549 agi_ok = agi_ok && xfs_agi_verify(bp);
1551 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1552 XFS_RANDOM_IALLOC_READ_AGI))) {
1553 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
1554 xfs_buf_ioerror(bp, EFSCORRUPTED);
1559 xfs_agi_write_verify(
1562 struct xfs_mount *mp = bp->b_target->bt_mount;
1563 struct xfs_buf_log_item *bip = bp->b_fspriv;
1565 if (!xfs_agi_verify(bp)) {
1566 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
1567 xfs_buf_ioerror(bp, EFSCORRUPTED);
1571 if (!xfs_sb_version_hascrc(&mp->m_sb))
1575 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
1576 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
1577 offsetof(struct xfs_agi, agi_crc));
1580 const struct xfs_buf_ops xfs_agi_buf_ops = {
1581 .verify_read = xfs_agi_read_verify,
1582 .verify_write = xfs_agi_write_verify,
1586 * Read in the allocation group header (inode allocation section)
1590 struct xfs_mount *mp, /* file system mount structure */
1591 struct xfs_trans *tp, /* transaction pointer */
1592 xfs_agnumber_t agno, /* allocation group number */
1593 struct xfs_buf **bpp) /* allocation group hdr buf */
1597 ASSERT(agno != NULLAGNUMBER);
1599 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1600 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1601 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
1605 ASSERT(!xfs_buf_geterror(*bpp));
1606 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1611 xfs_ialloc_read_agi(
1612 struct xfs_mount *mp, /* file system mount structure */
1613 struct xfs_trans *tp, /* transaction pointer */
1614 xfs_agnumber_t agno, /* allocation group number */
1615 struct xfs_buf **bpp) /* allocation group hdr buf */
1617 struct xfs_agi *agi; /* allocation group header */
1618 struct xfs_perag *pag; /* per allocation group data */
1621 error = xfs_read_agi(mp, tp, agno, bpp);
1625 agi = XFS_BUF_TO_AGI(*bpp);
1626 pag = xfs_perag_get(mp, agno);
1627 if (!pag->pagi_init) {
1628 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1629 pag->pagi_count = be32_to_cpu(agi->agi_count);
1634 * It's possible for these to be out of sync if
1635 * we are in the middle of a forced shutdown.
1637 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1638 XFS_FORCED_SHUTDOWN(mp));
1644 * Read in the agi to initialise the per-ag data in the mount structure
1647 xfs_ialloc_pagi_init(
1648 xfs_mount_t *mp, /* file system mount structure */
1649 xfs_trans_t *tp, /* transaction pointer */
1650 xfs_agnumber_t agno) /* allocation group number */
1652 xfs_buf_t *bp = NULL;
1655 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1659 xfs_trans_brelse(tp, bp);