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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
37 #include "xfs_cksum.h"
38 #include "xfs_trans.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_icreate_item.h"
41 #include "xfs_icache.h"
42 #include "xfs_dinode.h"
43 #include "xfs_trace.h"
47 * Allocation group level functions.
50 xfs_ialloc_cluster_alignment(
51 xfs_alloc_arg_t *args)
53 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
54 args->mp->m_sb.sb_inoalignmt >=
55 XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size))
56 return args->mp->m_sb.sb_inoalignmt;
61 * Lookup a record by ino in the btree given by cur.
65 struct xfs_btree_cur *cur, /* btree cursor */
66 xfs_agino_t ino, /* starting inode of chunk */
67 xfs_lookup_t dir, /* <=, >=, == */
68 int *stat) /* success/failure */
70 cur->bc_rec.i.ir_startino = ino;
71 cur->bc_rec.i.ir_freecount = 0;
72 cur->bc_rec.i.ir_free = 0;
73 return xfs_btree_lookup(cur, dir, stat);
77 * Update the record referred to by cur to the value given.
78 * This either works (return 0) or gets an EFSCORRUPTED error.
80 STATIC int /* error */
82 struct xfs_btree_cur *cur, /* btree cursor */
83 xfs_inobt_rec_incore_t *irec) /* btree record */
85 union xfs_btree_rec rec;
87 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
88 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
89 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
90 return xfs_btree_update(cur, &rec);
94 * Get the data from the pointed-to record.
98 struct xfs_btree_cur *cur, /* btree cursor */
99 xfs_inobt_rec_incore_t *irec, /* btree record */
100 int *stat) /* output: success/failure */
102 union xfs_btree_rec *rec;
105 error = xfs_btree_get_rec(cur, &rec, stat);
106 if (!error && *stat == 1) {
107 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
108 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
109 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
115 * Insert a single inobt record. Cursor must already point to desired location.
118 xfs_inobt_insert_rec(
119 struct xfs_btree_cur *cur,
124 cur->bc_rec.i.ir_freecount = freecount;
125 cur->bc_rec.i.ir_free = free;
126 return xfs_btree_insert(cur, stat);
130 * Insert records describing a newly allocated inode chunk into the inobt.
134 struct xfs_mount *mp,
135 struct xfs_trans *tp,
136 struct xfs_buf *agbp,
141 struct xfs_btree_cur *cur;
142 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
143 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
148 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
150 for (thisino = newino;
151 thisino < newino + newlen;
152 thisino += XFS_INODES_PER_CHUNK) {
153 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
155 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
160 error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
161 XFS_INOBT_ALL_FREE, &i);
163 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
169 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
175 * Verify that the number of free inodes in the AGI is correct.
179 xfs_check_agi_freecount(
180 struct xfs_btree_cur *cur,
183 if (cur->bc_nlevels == 1) {
184 xfs_inobt_rec_incore_t rec;
189 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
194 error = xfs_inobt_get_rec(cur, &rec, &i);
199 freecount += rec.ir_freecount;
200 error = xfs_btree_increment(cur, 0, &i);
206 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
207 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
212 #define xfs_check_agi_freecount(cur, agi) 0
216 * Initialise a new set of inodes. When called without a transaction context
217 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
218 * than logging them (which in a transaction context puts them into the AIL
219 * for writeback rather than the xfsbufd queue).
222 xfs_ialloc_inode_init(
223 struct xfs_mount *mp,
224 struct xfs_trans *tp,
225 struct list_head *buffer_list,
228 xfs_agblock_t length,
231 struct xfs_buf *fbuf;
232 struct xfs_dinode *free;
233 int nbufs, blks_per_cluster, inodes_per_cluster;
240 * Loop over the new block(s), filling in the inodes. For small block
241 * sizes, manipulate the inodes in buffers which are multiples of the
244 blks_per_cluster = xfs_icluster_size_fsb(mp);
245 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
246 nbufs = length / blks_per_cluster;
249 * Figure out what version number to use in the inodes we create. If
250 * the superblock version has caught up to the one that supports the new
251 * inode format, then use the new inode version. Otherwise use the old
252 * version so that old kernels will continue to be able to use the file
255 * For v3 inodes, we also need to write the inode number into the inode,
256 * so calculate the first inode number of the chunk here as
257 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
258 * across multiple filesystem blocks (such as a cluster) and so cannot
259 * be used in the cluster buffer loop below.
261 * Further, because we are writing the inode directly into the buffer
262 * and calculating a CRC on the entire inode, we have ot log the entire
263 * inode so that the entire range the CRC covers is present in the log.
264 * That means for v3 inode we log the entire buffer rather than just the
267 if (xfs_sb_version_hascrc(&mp->m_sb)) {
269 ino = XFS_AGINO_TO_INO(mp, agno,
270 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
273 * log the initialisation that is about to take place as an
274 * logical operation. This means the transaction does not
275 * need to log the physical changes to the inode buffers as log
276 * recovery will know what initialisation is actually needed.
277 * Hence we only need to log the buffers as "ordered" buffers so
278 * they track in the AIL as if they were physically logged.
281 xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
282 mp->m_sb.sb_inodesize, length, gen);
283 } else if (xfs_sb_version_hasnlink(&mp->m_sb))
288 for (j = 0; j < nbufs; j++) {
292 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
293 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
294 mp->m_bsize * blks_per_cluster,
299 /* Initialize the inode buffers and log them appropriately. */
300 fbuf->b_ops = &xfs_inode_buf_ops;
301 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
302 for (i = 0; i < inodes_per_cluster; i++) {
303 int ioffset = i << mp->m_sb.sb_inodelog;
304 uint isize = xfs_dinode_size(version);
306 free = xfs_make_iptr(mp, fbuf, i);
307 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
308 free->di_version = version;
309 free->di_gen = cpu_to_be32(gen);
310 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
313 free->di_ino = cpu_to_be64(ino);
315 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
316 xfs_dinode_calc_crc(mp, free);
318 /* just log the inode core */
319 xfs_trans_log_buf(tp, fbuf, ioffset,
320 ioffset + isize - 1);
326 * Mark the buffer as an inode allocation buffer so it
327 * sticks in AIL at the point of this allocation
328 * transaction. This ensures the they are on disk before
329 * the tail of the log can be moved past this
330 * transaction (i.e. by preventing relogging from moving
331 * it forward in the log).
333 xfs_trans_inode_alloc_buf(tp, fbuf);
336 * Mark the buffer as ordered so that they are
337 * not physically logged in the transaction but
338 * still tracked in the AIL as part of the
339 * transaction and pin the log appropriately.
341 xfs_trans_ordered_buf(tp, fbuf);
342 xfs_trans_log_buf(tp, fbuf, 0,
343 BBTOB(fbuf->b_length) - 1);
346 fbuf->b_flags |= XBF_DONE;
347 xfs_buf_delwri_queue(fbuf, buffer_list);
355 * Allocate new inodes in the allocation group specified by agbp.
356 * Return 0 for success, else error code.
358 STATIC int /* error code or 0 */
360 xfs_trans_t *tp, /* transaction pointer */
361 xfs_buf_t *agbp, /* alloc group buffer */
364 xfs_agi_t *agi; /* allocation group header */
365 xfs_alloc_arg_t args; /* allocation argument structure */
368 xfs_agino_t newino; /* new first inode's number */
369 xfs_agino_t newlen; /* new number of inodes */
370 int isaligned = 0; /* inode allocation at stripe unit */
372 struct xfs_perag *pag;
374 memset(&args, 0, sizeof(args));
376 args.mp = tp->t_mountp;
379 * Locking will ensure that we don't have two callers in here
382 newlen = args.mp->m_ialloc_inos;
383 if (args.mp->m_maxicount &&
384 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
385 return XFS_ERROR(ENOSPC);
386 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
388 * First try to allocate inodes contiguous with the last-allocated
389 * chunk of inodes. If the filesystem is striped, this will fill
390 * an entire stripe unit with inodes.
392 agi = XFS_BUF_TO_AGI(agbp);
393 newino = be32_to_cpu(agi->agi_newino);
394 agno = be32_to_cpu(agi->agi_seqno);
395 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
396 args.mp->m_ialloc_blks;
397 if (likely(newino != NULLAGINO &&
398 (args.agbno < be32_to_cpu(agi->agi_length)))) {
399 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
400 args.type = XFS_ALLOCTYPE_THIS_BNO;
404 * We need to take into account alignment here to ensure that
405 * we don't modify the free list if we fail to have an exact
406 * block. If we don't have an exact match, and every oher
407 * attempt allocation attempt fails, we'll end up cancelling
408 * a dirty transaction and shutting down.
410 * For an exact allocation, alignment must be 1,
411 * however we need to take cluster alignment into account when
412 * fixing up the freelist. Use the minalignslop field to
413 * indicate that extra blocks might be required for alignment,
414 * but not to use them in the actual exact allocation.
417 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
419 /* Allow space for the inode btree to split. */
420 args.minleft = args.mp->m_in_maxlevels - 1;
421 if ((error = xfs_alloc_vextent(&args)))
425 * This request might have dirtied the transaction if the AG can
426 * satisfy the request, but the exact block was not available.
427 * If the allocation did fail, subsequent requests will relax
428 * the exact agbno requirement and increase the alignment
429 * instead. It is critical that the total size of the request
430 * (len + alignment + slop) does not increase from this point
431 * on, so reset minalignslop to ensure it is not included in
432 * subsequent requests.
434 args.minalignslop = 0;
436 args.fsbno = NULLFSBLOCK;
438 if (unlikely(args.fsbno == NULLFSBLOCK)) {
440 * Set the alignment for the allocation.
441 * If stripe alignment is turned on then align at stripe unit
443 * If the cluster size is smaller than a filesystem block
444 * then we're doing I/O for inodes in filesystem block size
445 * pieces, so don't need alignment anyway.
448 if (args.mp->m_sinoalign) {
449 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
450 args.alignment = args.mp->m_dalign;
453 args.alignment = xfs_ialloc_cluster_alignment(&args);
455 * Need to figure out where to allocate the inode blocks.
456 * Ideally they should be spaced out through the a.g.
457 * For now, just allocate blocks up front.
459 args.agbno = be32_to_cpu(agi->agi_root);
460 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
462 * Allocate a fixed-size extent of inodes.
464 args.type = XFS_ALLOCTYPE_NEAR_BNO;
467 * Allow space for the inode btree to split.
469 args.minleft = args.mp->m_in_maxlevels - 1;
470 if ((error = xfs_alloc_vextent(&args)))
475 * If stripe alignment is turned on, then try again with cluster
478 if (isaligned && args.fsbno == NULLFSBLOCK) {
479 args.type = XFS_ALLOCTYPE_NEAR_BNO;
480 args.agbno = be32_to_cpu(agi->agi_root);
481 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
482 args.alignment = xfs_ialloc_cluster_alignment(&args);
483 if ((error = xfs_alloc_vextent(&args)))
487 if (args.fsbno == NULLFSBLOCK) {
491 ASSERT(args.len == args.minlen);
494 * Stamp and write the inode buffers.
496 * Seed the new inode cluster with a random generation number. This
497 * prevents short-term reuse of generation numbers if a chunk is
498 * freed and then immediately reallocated. We use random numbers
499 * rather than a linear progression to prevent the next generation
500 * number from being easily guessable.
502 error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
503 args.len, prandom_u32());
508 * Convert the results.
510 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
511 be32_add_cpu(&agi->agi_count, newlen);
512 be32_add_cpu(&agi->agi_freecount, newlen);
513 pag = xfs_perag_get(args.mp, agno);
514 pag->pagi_freecount += newlen;
516 agi->agi_newino = cpu_to_be32(newino);
519 * Insert records describing the new inode chunk into the btrees.
521 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
526 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
527 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
533 * Log allocation group header fields
535 xfs_ialloc_log_agi(tp, agbp,
536 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
538 * Modify/log superblock values for inode count and inode free count.
540 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
541 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
546 STATIC xfs_agnumber_t
552 spin_lock(&mp->m_agirotor_lock);
553 agno = mp->m_agirotor;
554 if (++mp->m_agirotor >= mp->m_maxagi)
556 spin_unlock(&mp->m_agirotor_lock);
562 * Select an allocation group to look for a free inode in, based on the parent
563 * inode and the mode. Return the allocation group buffer.
565 STATIC xfs_agnumber_t
566 xfs_ialloc_ag_select(
567 xfs_trans_t *tp, /* transaction pointer */
568 xfs_ino_t parent, /* parent directory inode number */
569 umode_t mode, /* bits set to indicate file type */
570 int okalloc) /* ok to allocate more space */
572 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
573 xfs_agnumber_t agno; /* current ag number */
574 int flags; /* alloc buffer locking flags */
575 xfs_extlen_t ineed; /* blocks needed for inode allocation */
576 xfs_extlen_t longest = 0; /* longest extent available */
577 xfs_mount_t *mp; /* mount point structure */
578 int needspace; /* file mode implies space allocated */
579 xfs_perag_t *pag; /* per allocation group data */
580 xfs_agnumber_t pagno; /* parent (starting) ag number */
584 * Files of these types need at least one block if length > 0
585 * (and they won't fit in the inode, but that's hard to figure out).
587 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
589 agcount = mp->m_maxagi;
591 pagno = xfs_ialloc_next_ag(mp);
593 pagno = XFS_INO_TO_AGNO(mp, parent);
594 if (pagno >= agcount)
598 ASSERT(pagno < agcount);
601 * Loop through allocation groups, looking for one with a little
602 * free space in it. Note we don't look for free inodes, exactly.
603 * Instead, we include whether there is a need to allocate inodes
604 * to mean that blocks must be allocated for them,
605 * if none are currently free.
608 flags = XFS_ALLOC_FLAG_TRYLOCK;
610 pag = xfs_perag_get(mp, agno);
611 if (!pag->pagi_inodeok) {
612 xfs_ialloc_next_ag(mp);
616 if (!pag->pagi_init) {
617 error = xfs_ialloc_pagi_init(mp, tp, agno);
622 if (pag->pagi_freecount) {
630 if (!pag->pagf_init) {
631 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
637 * Is there enough free space for the file plus a block of
638 * inodes? (if we need to allocate some)?
640 ineed = mp->m_ialloc_blks;
641 longest = pag->pagf_longest;
643 longest = pag->pagf_flcount > 0;
645 if (pag->pagf_freeblks >= needspace + ineed &&
653 * No point in iterating over the rest, if we're shutting
656 if (XFS_FORCED_SHUTDOWN(mp))
670 * Try to retrieve the next record to the left/right from the current one.
674 struct xfs_btree_cur *cur,
675 xfs_inobt_rec_incore_t *rec,
683 error = xfs_btree_decrement(cur, 0, &i);
685 error = xfs_btree_increment(cur, 0, &i);
691 error = xfs_inobt_get_rec(cur, rec, &i);
694 XFS_WANT_CORRUPTED_RETURN(i == 1);
702 struct xfs_btree_cur *cur,
704 xfs_inobt_rec_incore_t *rec,
710 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
715 error = xfs_inobt_get_rec(cur, rec, &i);
718 XFS_WANT_CORRUPTED_RETURN(i == 1);
725 * Allocate an inode using the inobt-only algorithm.
728 xfs_dialloc_ag_inobt(
729 struct xfs_trans *tp,
730 struct xfs_buf *agbp,
734 struct xfs_mount *mp = tp->t_mountp;
735 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
736 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
737 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
738 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
739 struct xfs_perag *pag;
740 struct xfs_btree_cur *cur, *tcur;
741 struct xfs_inobt_rec_incore rec, trec;
747 pag = xfs_perag_get(mp, agno);
749 ASSERT(pag->pagi_init);
750 ASSERT(pag->pagi_inodeok);
751 ASSERT(pag->pagi_freecount > 0);
754 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
756 * If pagino is 0 (this is the root inode allocation) use newino.
757 * This must work because we've just allocated some.
760 pagino = be32_to_cpu(agi->agi_newino);
762 error = xfs_check_agi_freecount(cur, agi);
767 * If in the same AG as the parent, try to get near the parent.
770 int doneleft; /* done, to the left */
771 int doneright; /* done, to the right */
772 int searchdistance = 10;
774 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
777 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
779 error = xfs_inobt_get_rec(cur, &rec, &j);
782 XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
784 if (rec.ir_freecount > 0) {
786 * Found a free inode in the same chunk
787 * as the parent, done.
794 * In the same AG as parent, but parent's chunk is full.
797 /* duplicate the cursor, search left & right simultaneously */
798 error = xfs_btree_dup_cursor(cur, &tcur);
803 * Skip to last blocks looked up if same parent inode.
805 if (pagino != NULLAGINO &&
806 pag->pagl_pagino == pagino &&
807 pag->pagl_leftrec != NULLAGINO &&
808 pag->pagl_rightrec != NULLAGINO) {
809 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
814 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
819 /* search left with tcur, back up 1 record */
820 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
824 /* search right with cur, go forward 1 record. */
825 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
831 * Loop until we find an inode chunk with a free inode.
833 while (!doneleft || !doneright) {
834 int useleft; /* using left inode chunk this time */
836 if (!--searchdistance) {
838 * Not in range - save last search
839 * location and allocate a new inode
841 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
842 pag->pagl_leftrec = trec.ir_startino;
843 pag->pagl_rightrec = rec.ir_startino;
844 pag->pagl_pagino = pagino;
848 /* figure out the closer block if both are valid. */
849 if (!doneleft && !doneright) {
851 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
852 rec.ir_startino - pagino;
857 /* free inodes to the left? */
858 if (useleft && trec.ir_freecount) {
860 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
863 pag->pagl_leftrec = trec.ir_startino;
864 pag->pagl_rightrec = rec.ir_startino;
865 pag->pagl_pagino = pagino;
869 /* free inodes to the right? */
870 if (!useleft && rec.ir_freecount) {
871 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
873 pag->pagl_leftrec = trec.ir_startino;
874 pag->pagl_rightrec = rec.ir_startino;
875 pag->pagl_pagino = pagino;
879 /* get next record to check */
881 error = xfs_ialloc_next_rec(tcur, &trec,
884 error = xfs_ialloc_next_rec(cur, &rec,
892 * We've reached the end of the btree. because
893 * we are only searching a small chunk of the
894 * btree each search, there is obviously free
895 * inodes closer to the parent inode than we
896 * are now. restart the search again.
898 pag->pagl_pagino = NULLAGINO;
899 pag->pagl_leftrec = NULLAGINO;
900 pag->pagl_rightrec = NULLAGINO;
901 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
902 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
907 * In a different AG from the parent.
908 * See if the most recently allocated block has any free.
911 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
912 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
918 error = xfs_inobt_get_rec(cur, &rec, &j);
922 if (j == 1 && rec.ir_freecount > 0) {
924 * The last chunk allocated in the group
925 * still has a free inode.
933 * None left in the last group, search the whole AG
935 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
938 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
941 error = xfs_inobt_get_rec(cur, &rec, &i);
944 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
945 if (rec.ir_freecount > 0)
947 error = xfs_btree_increment(cur, 0, &i);
950 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
954 offset = xfs_lowbit64(rec.ir_free);
956 ASSERT(offset < XFS_INODES_PER_CHUNK);
957 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
958 XFS_INODES_PER_CHUNK) == 0);
959 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
960 rec.ir_free &= ~XFS_INOBT_MASK(offset);
962 error = xfs_inobt_update(cur, &rec);
965 be32_add_cpu(&agi->agi_freecount, -1);
966 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
967 pag->pagi_freecount--;
969 error = xfs_check_agi_freecount(cur, agi);
973 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
974 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
979 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
981 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
987 * Use the free inode btree to allocate an inode based on distance from the
988 * parent. Note that the provided cursor may be deleted and replaced.
991 xfs_dialloc_ag_finobt_near(
993 struct xfs_btree_cur **ocur,
994 struct xfs_inobt_rec_incore *rec)
996 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
997 struct xfs_btree_cur *rcur; /* right search cursor */
998 struct xfs_inobt_rec_incore rrec;
1002 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1007 error = xfs_inobt_get_rec(lcur, rec, &i);
1010 XFS_WANT_CORRUPTED_RETURN(i == 1);
1013 * See if we've landed in the parent inode record. The finobt
1014 * only tracks chunks with at least one free inode, so record
1015 * existence is enough.
1017 if (pagino >= rec->ir_startino &&
1018 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1022 error = xfs_btree_dup_cursor(lcur, &rcur);
1026 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1030 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1033 XFS_WANT_CORRUPTED_GOTO(j == 1, error_rcur);
1036 XFS_WANT_CORRUPTED_GOTO(i == 1 || j == 1, error_rcur);
1037 if (i == 1 && j == 1) {
1039 * Both the left and right records are valid. Choose the closer
1040 * inode chunk to the target.
1042 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1043 (rrec.ir_startino - pagino)) {
1045 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1048 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1050 } else if (j == 1) {
1051 /* only the right record is valid */
1053 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1055 } else if (i == 1) {
1056 /* only the left record is valid */
1057 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1063 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1068 * Use the free inode btree to find a free inode based on a newino hint. If
1069 * the hint is NULL, find the first free inode in the AG.
1072 xfs_dialloc_ag_finobt_newino(
1073 struct xfs_agi *agi,
1074 struct xfs_btree_cur *cur,
1075 struct xfs_inobt_rec_incore *rec)
1080 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1081 error = xfs_inobt_lookup(cur, agi->agi_newino, XFS_LOOKUP_EQ,
1086 error = xfs_inobt_get_rec(cur, rec, &i);
1089 XFS_WANT_CORRUPTED_RETURN(i == 1);
1096 * Find the first inode available in the AG.
1098 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1101 XFS_WANT_CORRUPTED_RETURN(i == 1);
1103 error = xfs_inobt_get_rec(cur, rec, &i);
1106 XFS_WANT_CORRUPTED_RETURN(i == 1);
1112 * Update the inobt based on a modification made to the finobt. Also ensure that
1113 * the records from both trees are equivalent post-modification.
1116 xfs_dialloc_ag_update_inobt(
1117 struct xfs_btree_cur *cur, /* inobt cursor */
1118 struct xfs_inobt_rec_incore *frec, /* finobt record */
1119 int offset) /* inode offset */
1121 struct xfs_inobt_rec_incore rec;
1125 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1128 XFS_WANT_CORRUPTED_RETURN(i == 1);
1130 error = xfs_inobt_get_rec(cur, &rec, &i);
1133 XFS_WANT_CORRUPTED_RETURN(i == 1);
1134 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1135 XFS_INODES_PER_CHUNK) == 0);
1137 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1140 XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) &&
1141 (rec.ir_freecount == frec->ir_freecount));
1143 error = xfs_inobt_update(cur, &rec);
1151 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1152 * back to the inobt search algorithm.
1154 * The caller selected an AG for us, and made sure that free inodes are
1159 struct xfs_trans *tp,
1160 struct xfs_buf *agbp,
1164 struct xfs_mount *mp = tp->t_mountp;
1165 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1166 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1167 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1168 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1169 struct xfs_perag *pag;
1170 struct xfs_btree_cur *cur; /* finobt cursor */
1171 struct xfs_btree_cur *icur; /* inobt cursor */
1172 struct xfs_inobt_rec_incore rec;
1178 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1179 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1181 pag = xfs_perag_get(mp, agno);
1184 * If pagino is 0 (this is the root inode allocation) use newino.
1185 * This must work because we've just allocated some.
1188 pagino = be32_to_cpu(agi->agi_newino);
1190 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1192 error = xfs_check_agi_freecount(cur, agi);
1197 * The search algorithm depends on whether we're in the same AG as the
1198 * parent. If so, find the closest available inode to the parent. If
1199 * not, consider the agi hint or find the first free inode in the AG.
1202 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1204 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1208 offset = xfs_lowbit64(rec.ir_free);
1209 ASSERT(offset >= 0);
1210 ASSERT(offset < XFS_INODES_PER_CHUNK);
1211 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1212 XFS_INODES_PER_CHUNK) == 0);
1213 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1216 * Modify or remove the finobt record.
1218 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1220 if (rec.ir_freecount)
1221 error = xfs_inobt_update(cur, &rec);
1223 error = xfs_btree_delete(cur, &i);
1228 * The finobt has now been updated appropriately. We haven't updated the
1229 * agi and superblock yet, so we can create an inobt cursor and validate
1230 * the original freecount. If all is well, make the equivalent update to
1231 * the inobt using the finobt record and offset information.
1233 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1235 error = xfs_check_agi_freecount(icur, agi);
1239 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1244 * Both trees have now been updated. We must update the perag and
1245 * superblock before we can check the freecount for each btree.
1247 be32_add_cpu(&agi->agi_freecount, -1);
1248 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1249 pag->pagi_freecount--;
1251 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1253 error = xfs_check_agi_freecount(icur, agi);
1256 error = xfs_check_agi_freecount(cur, agi);
1260 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1261 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1267 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1269 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1275 * Allocate an inode on disk.
1277 * Mode is used to tell whether the new inode will need space, and whether it
1280 * This function is designed to be called twice if it has to do an allocation
1281 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1282 * If an inode is available without having to performn an allocation, an inode
1283 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1284 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1285 * The caller should then commit the current transaction, allocate a
1286 * new transaction, and call xfs_dialloc() again, passing in the previous value
1287 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1288 * buffer is locked across the two calls, the second call is guaranteed to have
1289 * a free inode available.
1291 * Once we successfully pick an inode its number is returned and the on-disk
1292 * data structures are updated. The inode itself is not read in, since doing so
1293 * would break ordering constraints with xfs_reclaim.
1297 struct xfs_trans *tp,
1301 struct xfs_buf **IO_agbp,
1304 struct xfs_mount *mp = tp->t_mountp;
1305 struct xfs_buf *agbp;
1306 xfs_agnumber_t agno;
1310 xfs_agnumber_t start_agno;
1311 struct xfs_perag *pag;
1315 * If the caller passes in a pointer to the AGI buffer,
1316 * continue where we left off before. In this case, we
1317 * know that the allocation group has free inodes.
1324 * We do not have an agbp, so select an initial allocation
1325 * group for inode allocation.
1327 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1328 if (start_agno == NULLAGNUMBER) {
1334 * If we have already hit the ceiling of inode blocks then clear
1335 * okalloc so we scan all available agi structures for a free
1338 if (mp->m_maxicount &&
1339 mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) {
1345 * Loop until we find an allocation group that either has free inodes
1346 * or in which we can allocate some inodes. Iterate through the
1347 * allocation groups upward, wrapping at the end.
1351 pag = xfs_perag_get(mp, agno);
1352 if (!pag->pagi_inodeok) {
1353 xfs_ialloc_next_ag(mp);
1357 if (!pag->pagi_init) {
1358 error = xfs_ialloc_pagi_init(mp, tp, agno);
1364 * Do a first racy fast path check if this AG is usable.
1366 if (!pag->pagi_freecount && !okalloc)
1370 * Then read in the AGI buffer and recheck with the AGI buffer
1373 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1377 if (pag->pagi_freecount) {
1383 goto nextag_relse_buffer;
1386 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1388 xfs_trans_brelse(tp, agbp);
1390 if (error != ENOSPC)
1400 * We successfully allocated some inodes, return
1401 * the current context to the caller so that it
1402 * can commit the current transaction and call
1403 * us again where we left off.
1405 ASSERT(pag->pagi_freecount > 0);
1413 nextag_relse_buffer:
1414 xfs_trans_brelse(tp, agbp);
1417 if (++agno == mp->m_sb.sb_agcount)
1419 if (agno == start_agno) {
1421 return noroom ? ENOSPC : 0;
1427 return xfs_dialloc_ag(tp, agbp, parent, inop);
1430 return XFS_ERROR(error);
1435 struct xfs_mount *mp,
1436 struct xfs_trans *tp,
1437 struct xfs_buf *agbp,
1439 struct xfs_bmap_free *flist,
1441 xfs_ino_t *first_ino,
1442 struct xfs_inobt_rec_incore *orec)
1444 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1445 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1446 struct xfs_perag *pag;
1447 struct xfs_btree_cur *cur;
1448 struct xfs_inobt_rec_incore rec;
1454 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1455 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1458 * Initialize the cursor.
1460 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1462 error = xfs_check_agi_freecount(cur, agi);
1467 * Look for the entry describing this inode.
1469 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1470 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1474 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1475 error = xfs_inobt_get_rec(cur, &rec, &i);
1477 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1481 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1483 * Get the offset in the inode chunk.
1485 off = agino - rec.ir_startino;
1486 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1487 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1489 * Mark the inode free & increment the count.
1491 rec.ir_free |= XFS_INOBT_MASK(off);
1495 * When an inode cluster is free, it becomes eligible for removal
1497 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1498 (rec.ir_freecount == mp->m_ialloc_inos)) {
1501 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1504 * Remove the inode cluster from the AGI B+Tree, adjust the
1505 * AGI and Superblock inode counts, and mark the disk space
1506 * to be freed when the transaction is committed.
1508 ilen = mp->m_ialloc_inos;
1509 be32_add_cpu(&agi->agi_count, -ilen);
1510 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1511 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1512 pag = xfs_perag_get(mp, agno);
1513 pag->pagi_freecount -= ilen - 1;
1515 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1516 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1518 if ((error = xfs_btree_delete(cur, &i))) {
1519 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1524 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1525 XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1526 mp->m_ialloc_blks, flist, mp);
1530 error = xfs_inobt_update(cur, &rec);
1532 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1538 * Change the inode free counts and log the ag/sb changes.
1540 be32_add_cpu(&agi->agi_freecount, 1);
1541 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1542 pag = xfs_perag_get(mp, agno);
1543 pag->pagi_freecount++;
1545 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1548 error = xfs_check_agi_freecount(cur, agi);
1553 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1557 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1562 * Free an inode in the free inode btree.
1566 struct xfs_mount *mp,
1567 struct xfs_trans *tp,
1568 struct xfs_buf *agbp,
1570 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
1572 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1573 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1574 struct xfs_btree_cur *cur;
1575 struct xfs_inobt_rec_incore rec;
1576 int offset = agino - ibtrec->ir_startino;
1580 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1582 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
1587 * If the record does not exist in the finobt, we must have just
1588 * freed an inode in a previously fully allocated chunk. If not,
1589 * something is out of sync.
1591 XFS_WANT_CORRUPTED_GOTO(ibtrec->ir_freecount == 1, error);
1593 error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
1594 ibtrec->ir_free, &i);
1603 * Read and update the existing record. We could just copy the ibtrec
1604 * across here, but that would defeat the purpose of having redundant
1605 * metadata. By making the modifications independently, we can catch
1606 * corruptions that we wouldn't see if we just copied from one record
1609 error = xfs_inobt_get_rec(cur, &rec, &i);
1612 XFS_WANT_CORRUPTED_GOTO(i == 1, error);
1614 rec.ir_free |= XFS_INOBT_MASK(offset);
1617 XFS_WANT_CORRUPTED_GOTO((rec.ir_free == ibtrec->ir_free) &&
1618 (rec.ir_freecount == ibtrec->ir_freecount),
1622 * The content of inobt records should always match between the inobt
1623 * and finobt. The lifecycle of records in the finobt is different from
1624 * the inobt in that the finobt only tracks records with at least one
1625 * free inode. Hence, if all of the inodes are free and we aren't
1626 * keeping inode chunks permanently on disk, remove the record.
1627 * Otherwise, update the record with the new information.
1629 if (rec.ir_freecount == mp->m_ialloc_inos &&
1630 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
1631 error = xfs_btree_delete(cur, &i);
1636 error = xfs_inobt_update(cur, &rec);
1642 error = xfs_check_agi_freecount(cur, agi);
1646 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1650 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1655 * Free disk inode. Carefully avoids touching the incore inode, all
1656 * manipulations incore are the caller's responsibility.
1657 * The on-disk inode is not changed by this operation, only the
1658 * btree (free inode mask) is changed.
1662 struct xfs_trans *tp, /* transaction pointer */
1663 xfs_ino_t inode, /* inode to be freed */
1664 struct xfs_bmap_free *flist, /* extents to free */
1665 int *delete,/* set if inode cluster was deleted */
1666 xfs_ino_t *first_ino)/* first inode in deleted cluster */
1669 xfs_agblock_t agbno; /* block number containing inode */
1670 struct xfs_buf *agbp; /* buffer for allocation group header */
1671 xfs_agino_t agino; /* allocation group inode number */
1672 xfs_agnumber_t agno; /* allocation group number */
1673 int error; /* error return value */
1674 struct xfs_mount *mp; /* mount structure for filesystem */
1675 struct xfs_inobt_rec_incore rec;/* btree record */
1680 * Break up inode number into its components.
1682 agno = XFS_INO_TO_AGNO(mp, inode);
1683 if (agno >= mp->m_sb.sb_agcount) {
1684 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1685 __func__, agno, mp->m_sb.sb_agcount);
1687 return XFS_ERROR(EINVAL);
1689 agino = XFS_INO_TO_AGINO(mp, inode);
1690 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1691 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1692 __func__, (unsigned long long)inode,
1693 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1695 return XFS_ERROR(EINVAL);
1697 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1698 if (agbno >= mp->m_sb.sb_agblocks) {
1699 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1700 __func__, agbno, mp->m_sb.sb_agblocks);
1702 return XFS_ERROR(EINVAL);
1705 * Get the allocation group header.
1707 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1709 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1715 * Fix up the inode allocation btree.
1717 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, delete, first_ino,
1723 * Fix up the free inode btree.
1725 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
1726 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
1739 struct xfs_mount *mp,
1740 struct xfs_trans *tp,
1741 xfs_agnumber_t agno,
1743 xfs_agblock_t agbno,
1744 xfs_agblock_t *chunk_agbno,
1745 xfs_agblock_t *offset_agbno,
1748 struct xfs_inobt_rec_incore rec;
1749 struct xfs_btree_cur *cur;
1750 struct xfs_buf *agbp;
1754 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1757 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1758 __func__, error, agno);
1763 * Lookup the inode record for the given agino. If the record cannot be
1764 * found, then it's an invalid inode number and we should abort. Once
1765 * we have a record, we need to ensure it contains the inode number
1766 * we are looking up.
1768 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1769 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1772 error = xfs_inobt_get_rec(cur, &rec, &i);
1773 if (!error && i == 0)
1777 xfs_trans_brelse(tp, agbp);
1778 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1782 /* check that the returned record contains the required inode */
1783 if (rec.ir_startino > agino ||
1784 rec.ir_startino + mp->m_ialloc_inos <= agino)
1787 /* for untrusted inodes check it is allocated first */
1788 if ((flags & XFS_IGET_UNTRUSTED) &&
1789 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1792 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1793 *offset_agbno = agbno - *chunk_agbno;
1798 * Return the location of the inode in imap, for mapping it into a buffer.
1802 xfs_mount_t *mp, /* file system mount structure */
1803 xfs_trans_t *tp, /* transaction pointer */
1804 xfs_ino_t ino, /* inode to locate */
1805 struct xfs_imap *imap, /* location map structure */
1806 uint flags) /* flags for inode btree lookup */
1808 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1809 xfs_agino_t agino; /* inode number within alloc group */
1810 xfs_agnumber_t agno; /* allocation group number */
1811 int blks_per_cluster; /* num blocks per inode cluster */
1812 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1813 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1814 int error; /* error code */
1815 int offset; /* index of inode in its buffer */
1816 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
1818 ASSERT(ino != NULLFSINO);
1821 * Split up the inode number into its parts.
1823 agno = XFS_INO_TO_AGNO(mp, ino);
1824 agino = XFS_INO_TO_AGINO(mp, ino);
1825 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1826 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1827 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1830 * Don't output diagnostic information for untrusted inodes
1831 * as they can be invalid without implying corruption.
1833 if (flags & XFS_IGET_UNTRUSTED)
1834 return XFS_ERROR(EINVAL);
1835 if (agno >= mp->m_sb.sb_agcount) {
1837 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1838 __func__, agno, mp->m_sb.sb_agcount);
1840 if (agbno >= mp->m_sb.sb_agblocks) {
1842 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1843 __func__, (unsigned long long)agbno,
1844 (unsigned long)mp->m_sb.sb_agblocks);
1846 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1848 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1850 XFS_AGINO_TO_INO(mp, agno, agino));
1854 return XFS_ERROR(EINVAL);
1857 blks_per_cluster = xfs_icluster_size_fsb(mp);
1860 * For bulkstat and handle lookups, we have an untrusted inode number
1861 * that we have to verify is valid. We cannot do this just by reading
1862 * the inode buffer as it may have been unlinked and removed leaving
1863 * inodes in stale state on disk. Hence we have to do a btree lookup
1864 * in all cases where an untrusted inode number is passed.
1866 if (flags & XFS_IGET_UNTRUSTED) {
1867 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1868 &chunk_agbno, &offset_agbno, flags);
1875 * If the inode cluster size is the same as the blocksize or
1876 * smaller we get to the buffer by simple arithmetics.
1878 if (blks_per_cluster == 1) {
1879 offset = XFS_INO_TO_OFFSET(mp, ino);
1880 ASSERT(offset < mp->m_sb.sb_inopblock);
1882 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1883 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1884 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1889 * If the inode chunks are aligned then use simple maths to
1890 * find the location. Otherwise we have to do a btree
1891 * lookup to find the location.
1893 if (mp->m_inoalign_mask) {
1894 offset_agbno = agbno & mp->m_inoalign_mask;
1895 chunk_agbno = agbno - offset_agbno;
1897 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1898 &chunk_agbno, &offset_agbno, flags);
1904 ASSERT(agbno >= chunk_agbno);
1905 cluster_agbno = chunk_agbno +
1906 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1907 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1908 XFS_INO_TO_OFFSET(mp, ino);
1910 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1911 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1912 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1915 * If the inode number maps to a block outside the bounds
1916 * of the file system then return NULL rather than calling
1917 * read_buf and panicing when we get an error from the
1920 if ((imap->im_blkno + imap->im_len) >
1921 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1923 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1924 __func__, (unsigned long long) imap->im_blkno,
1925 (unsigned long long) imap->im_len,
1926 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1927 return XFS_ERROR(EINVAL);
1933 * Compute and fill in value of m_in_maxlevels.
1936 xfs_ialloc_compute_maxlevels(
1937 xfs_mount_t *mp) /* file system mount structure */
1945 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1946 XFS_INODES_PER_CHUNK_LOG;
1947 minleafrecs = mp->m_alloc_mnr[0];
1948 minnoderecs = mp->m_alloc_mnr[1];
1949 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1950 for (level = 1; maxblocks > 1; level++)
1951 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1952 mp->m_in_maxlevels = level;
1956 * Log specified fields for the ag hdr (inode section). The growth of the agi
1957 * structure over time requires that we interpret the buffer as two logical
1958 * regions delineated by the end of the unlinked list. This is due to the size
1959 * of the hash table and its location in the middle of the agi.
1961 * For example, a request to log a field before agi_unlinked and a field after
1962 * agi_unlinked could cause us to log the entire hash table and use an excessive
1963 * amount of log space. To avoid this behavior, log the region up through
1964 * agi_unlinked in one call and the region after agi_unlinked through the end of
1965 * the structure in another.
1969 xfs_trans_t *tp, /* transaction pointer */
1970 xfs_buf_t *bp, /* allocation group header buffer */
1971 int fields) /* bitmask of fields to log */
1973 int first; /* first byte number */
1974 int last; /* last byte number */
1975 static const short offsets[] = { /* field starting offsets */
1976 /* keep in sync with bit definitions */
1977 offsetof(xfs_agi_t, agi_magicnum),
1978 offsetof(xfs_agi_t, agi_versionnum),
1979 offsetof(xfs_agi_t, agi_seqno),
1980 offsetof(xfs_agi_t, agi_length),
1981 offsetof(xfs_agi_t, agi_count),
1982 offsetof(xfs_agi_t, agi_root),
1983 offsetof(xfs_agi_t, agi_level),
1984 offsetof(xfs_agi_t, agi_freecount),
1985 offsetof(xfs_agi_t, agi_newino),
1986 offsetof(xfs_agi_t, agi_dirino),
1987 offsetof(xfs_agi_t, agi_unlinked),
1988 offsetof(xfs_agi_t, agi_free_root),
1989 offsetof(xfs_agi_t, agi_free_level),
1993 xfs_agi_t *agi; /* allocation group header */
1995 agi = XFS_BUF_TO_AGI(bp);
1996 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1999 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
2002 * Compute byte offsets for the first and last fields in the first
2003 * region and log the agi buffer. This only logs up through
2006 if (fields & XFS_AGI_ALL_BITS_R1) {
2007 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2009 xfs_trans_log_buf(tp, bp, first, last);
2013 * Mask off the bits in the first region and calculate the first and
2014 * last field offsets for any bits in the second region.
2016 fields &= ~XFS_AGI_ALL_BITS_R1;
2018 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2020 xfs_trans_log_buf(tp, bp, first, last);
2026 xfs_check_agi_unlinked(
2027 struct xfs_agi *agi)
2031 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2032 ASSERT(agi->agi_unlinked[i]);
2035 #define xfs_check_agi_unlinked(agi)
2042 struct xfs_mount *mp = bp->b_target->bt_mount;
2043 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2045 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2046 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
2049 * Validate the magic number of the agi block.
2051 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2053 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2057 * during growfs operations, the perag is not fully initialised,
2058 * so we can't use it for any useful checking. growfs ensures we can't
2059 * use it by using uncached buffers that don't have the perag attached
2060 * so we can detect and avoid this problem.
2062 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2065 xfs_check_agi_unlinked(agi);
2070 xfs_agi_read_verify(
2073 struct xfs_mount *mp = bp->b_target->bt_mount;
2075 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2076 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2077 xfs_buf_ioerror(bp, EFSBADCRC);
2078 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2079 XFS_ERRTAG_IALLOC_READ_AGI,
2080 XFS_RANDOM_IALLOC_READ_AGI))
2081 xfs_buf_ioerror(bp, EFSCORRUPTED);
2084 xfs_verifier_error(bp);
2088 xfs_agi_write_verify(
2091 struct xfs_mount *mp = bp->b_target->bt_mount;
2092 struct xfs_buf_log_item *bip = bp->b_fspriv;
2094 if (!xfs_agi_verify(bp)) {
2095 xfs_buf_ioerror(bp, EFSCORRUPTED);
2096 xfs_verifier_error(bp);
2100 if (!xfs_sb_version_hascrc(&mp->m_sb))
2104 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2105 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2108 const struct xfs_buf_ops xfs_agi_buf_ops = {
2109 .verify_read = xfs_agi_read_verify,
2110 .verify_write = xfs_agi_write_verify,
2114 * Read in the allocation group header (inode allocation section)
2118 struct xfs_mount *mp, /* file system mount structure */
2119 struct xfs_trans *tp, /* transaction pointer */
2120 xfs_agnumber_t agno, /* allocation group number */
2121 struct xfs_buf **bpp) /* allocation group hdr buf */
2125 trace_xfs_read_agi(mp, agno);
2127 ASSERT(agno != NULLAGNUMBER);
2128 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2129 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2130 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2134 ASSERT(!xfs_buf_geterror(*bpp));
2135 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2140 xfs_ialloc_read_agi(
2141 struct xfs_mount *mp, /* file system mount structure */
2142 struct xfs_trans *tp, /* transaction pointer */
2143 xfs_agnumber_t agno, /* allocation group number */
2144 struct xfs_buf **bpp) /* allocation group hdr buf */
2146 struct xfs_agi *agi; /* allocation group header */
2147 struct xfs_perag *pag; /* per allocation group data */
2150 trace_xfs_ialloc_read_agi(mp, agno);
2152 error = xfs_read_agi(mp, tp, agno, bpp);
2156 agi = XFS_BUF_TO_AGI(*bpp);
2157 pag = xfs_perag_get(mp, agno);
2158 if (!pag->pagi_init) {
2159 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2160 pag->pagi_count = be32_to_cpu(agi->agi_count);
2165 * It's possible for these to be out of sync if
2166 * we are in the middle of a forced shutdown.
2168 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2169 XFS_FORCED_SHUTDOWN(mp));
2175 * Read in the agi to initialise the per-ag data in the mount structure
2178 xfs_ialloc_pagi_init(
2179 xfs_mount_t *mp, /* file system mount structure */
2180 xfs_trans_t *tp, /* transaction pointer */
2181 xfs_agnumber_t agno) /* allocation group number */
2183 xfs_buf_t *bp = NULL;
2186 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2190 xfs_trans_brelse(tp, bp);