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_trace.h"
46 * Allocation group level functions.
49 xfs_ialloc_cluster_alignment(
50 xfs_alloc_arg_t *args)
52 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
53 args->mp->m_sb.sb_inoalignmt >=
54 XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size))
55 return args->mp->m_sb.sb_inoalignmt;
60 * Lookup a record by ino in the btree given by cur.
64 struct xfs_btree_cur *cur, /* btree cursor */
65 xfs_agino_t ino, /* starting inode of chunk */
66 xfs_lookup_t dir, /* <=, >=, == */
67 int *stat) /* success/failure */
69 cur->bc_rec.i.ir_startino = ino;
70 cur->bc_rec.i.ir_freecount = 0;
71 cur->bc_rec.i.ir_free = 0;
72 return xfs_btree_lookup(cur, dir, stat);
76 * Update the record referred to by cur to the value given.
77 * This either works (return 0) or gets an EFSCORRUPTED error.
79 STATIC int /* error */
81 struct xfs_btree_cur *cur, /* btree cursor */
82 xfs_inobt_rec_incore_t *irec) /* btree record */
84 union xfs_btree_rec rec;
86 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
87 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
88 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
89 return xfs_btree_update(cur, &rec);
93 * Get the data from the pointed-to record.
97 struct xfs_btree_cur *cur, /* btree cursor */
98 xfs_inobt_rec_incore_t *irec, /* btree record */
99 int *stat) /* output: success/failure */
101 union xfs_btree_rec *rec;
104 error = xfs_btree_get_rec(cur, &rec, stat);
105 if (!error && *stat == 1) {
106 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
107 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
108 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
114 * Insert a single inobt record. Cursor must already point to desired location.
117 xfs_inobt_insert_rec(
118 struct xfs_btree_cur *cur,
123 cur->bc_rec.i.ir_freecount = freecount;
124 cur->bc_rec.i.ir_free = free;
125 return xfs_btree_insert(cur, stat);
129 * Insert records describing a newly allocated inode chunk into the inobt.
133 struct xfs_mount *mp,
134 struct xfs_trans *tp,
135 struct xfs_buf *agbp,
140 struct xfs_btree_cur *cur;
141 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
142 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
147 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
149 for (thisino = newino;
150 thisino < newino + newlen;
151 thisino += XFS_INODES_PER_CHUNK) {
152 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
154 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
159 error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
160 XFS_INOBT_ALL_FREE, &i);
162 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
168 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
174 * Verify that the number of free inodes in the AGI is correct.
178 xfs_check_agi_freecount(
179 struct xfs_btree_cur *cur,
182 if (cur->bc_nlevels == 1) {
183 xfs_inobt_rec_incore_t rec;
188 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
193 error = xfs_inobt_get_rec(cur, &rec, &i);
198 freecount += rec.ir_freecount;
199 error = xfs_btree_increment(cur, 0, &i);
205 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
206 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
211 #define xfs_check_agi_freecount(cur, agi) 0
215 * Initialise a new set of inodes. When called without a transaction context
216 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
217 * than logging them (which in a transaction context puts them into the AIL
218 * for writeback rather than the xfsbufd queue).
221 xfs_ialloc_inode_init(
222 struct xfs_mount *mp,
223 struct xfs_trans *tp,
224 struct list_head *buffer_list,
227 xfs_agblock_t length,
230 struct xfs_buf *fbuf;
231 struct xfs_dinode *free;
232 int nbufs, blks_per_cluster, inodes_per_cluster;
239 * Loop over the new block(s), filling in the inodes. For small block
240 * sizes, manipulate the inodes in buffers which are multiples of the
243 blks_per_cluster = xfs_icluster_size_fsb(mp);
244 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
245 nbufs = length / blks_per_cluster;
248 * Figure out what version number to use in the inodes we create. If
249 * the superblock version has caught up to the one that supports the new
250 * inode format, then use the new inode version. Otherwise use the old
251 * version so that old kernels will continue to be able to use the file
254 * For v3 inodes, we also need to write the inode number into the inode,
255 * so calculate the first inode number of the chunk here as
256 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
257 * across multiple filesystem blocks (such as a cluster) and so cannot
258 * be used in the cluster buffer loop below.
260 * Further, because we are writing the inode directly into the buffer
261 * and calculating a CRC on the entire inode, we have ot log the entire
262 * inode so that the entire range the CRC covers is present in the log.
263 * That means for v3 inode we log the entire buffer rather than just the
266 if (xfs_sb_version_hascrc(&mp->m_sb)) {
268 ino = XFS_AGINO_TO_INO(mp, agno,
269 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
272 * log the initialisation that is about to take place as an
273 * logical operation. This means the transaction does not
274 * need to log the physical changes to the inode buffers as log
275 * recovery will know what initialisation is actually needed.
276 * Hence we only need to log the buffers as "ordered" buffers so
277 * they track in the AIL as if they were physically logged.
280 xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
281 mp->m_sb.sb_inodesize, length, gen);
285 for (j = 0; j < nbufs; j++) {
289 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
290 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
291 mp->m_bsize * blks_per_cluster,
296 /* Initialize the inode buffers and log them appropriately. */
297 fbuf->b_ops = &xfs_inode_buf_ops;
298 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
299 for (i = 0; i < inodes_per_cluster; i++) {
300 int ioffset = i << mp->m_sb.sb_inodelog;
301 uint isize = xfs_dinode_size(version);
303 free = xfs_make_iptr(mp, fbuf, i);
304 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
305 free->di_version = version;
306 free->di_gen = cpu_to_be32(gen);
307 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
310 free->di_ino = cpu_to_be64(ino);
312 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
313 xfs_dinode_calc_crc(mp, free);
315 /* just log the inode core */
316 xfs_trans_log_buf(tp, fbuf, ioffset,
317 ioffset + isize - 1);
323 * Mark the buffer as an inode allocation buffer so it
324 * sticks in AIL at the point of this allocation
325 * transaction. This ensures the they are on disk before
326 * the tail of the log can be moved past this
327 * transaction (i.e. by preventing relogging from moving
328 * it forward in the log).
330 xfs_trans_inode_alloc_buf(tp, fbuf);
333 * Mark the buffer as ordered so that they are
334 * not physically logged in the transaction but
335 * still tracked in the AIL as part of the
336 * transaction and pin the log appropriately.
338 xfs_trans_ordered_buf(tp, fbuf);
339 xfs_trans_log_buf(tp, fbuf, 0,
340 BBTOB(fbuf->b_length) - 1);
343 fbuf->b_flags |= XBF_DONE;
344 xfs_buf_delwri_queue(fbuf, buffer_list);
352 * Allocate new inodes in the allocation group specified by agbp.
353 * Return 0 for success, else error code.
355 STATIC int /* error code or 0 */
357 xfs_trans_t *tp, /* transaction pointer */
358 xfs_buf_t *agbp, /* alloc group buffer */
361 xfs_agi_t *agi; /* allocation group header */
362 xfs_alloc_arg_t args; /* allocation argument structure */
365 xfs_agino_t newino; /* new first inode's number */
366 xfs_agino_t newlen; /* new number of inodes */
367 int isaligned = 0; /* inode allocation at stripe unit */
369 struct xfs_perag *pag;
371 memset(&args, 0, sizeof(args));
373 args.mp = tp->t_mountp;
376 * Locking will ensure that we don't have two callers in here
379 newlen = args.mp->m_ialloc_inos;
380 if (args.mp->m_maxicount &&
381 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
383 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
385 * First try to allocate inodes contiguous with the last-allocated
386 * chunk of inodes. If the filesystem is striped, this will fill
387 * an entire stripe unit with inodes.
389 agi = XFS_BUF_TO_AGI(agbp);
390 newino = be32_to_cpu(agi->agi_newino);
391 agno = be32_to_cpu(agi->agi_seqno);
392 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
393 args.mp->m_ialloc_blks;
394 if (likely(newino != NULLAGINO &&
395 (args.agbno < be32_to_cpu(agi->agi_length)))) {
396 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
397 args.type = XFS_ALLOCTYPE_THIS_BNO;
401 * We need to take into account alignment here to ensure that
402 * we don't modify the free list if we fail to have an exact
403 * block. If we don't have an exact match, and every oher
404 * attempt allocation attempt fails, we'll end up cancelling
405 * a dirty transaction and shutting down.
407 * For an exact allocation, alignment must be 1,
408 * however we need to take cluster alignment into account when
409 * fixing up the freelist. Use the minalignslop field to
410 * indicate that extra blocks might be required for alignment,
411 * but not to use them in the actual exact allocation.
414 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
416 /* Allow space for the inode btree to split. */
417 args.minleft = args.mp->m_in_maxlevels - 1;
418 if ((error = xfs_alloc_vextent(&args)))
422 * This request might have dirtied the transaction if the AG can
423 * satisfy the request, but the exact block was not available.
424 * If the allocation did fail, subsequent requests will relax
425 * the exact agbno requirement and increase the alignment
426 * instead. It is critical that the total size of the request
427 * (len + alignment + slop) does not increase from this point
428 * on, so reset minalignslop to ensure it is not included in
429 * subsequent requests.
431 args.minalignslop = 0;
433 args.fsbno = NULLFSBLOCK;
435 if (unlikely(args.fsbno == NULLFSBLOCK)) {
437 * Set the alignment for the allocation.
438 * If stripe alignment is turned on then align at stripe unit
440 * If the cluster size is smaller than a filesystem block
441 * then we're doing I/O for inodes in filesystem block size
442 * pieces, so don't need alignment anyway.
445 if (args.mp->m_sinoalign) {
446 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
447 args.alignment = args.mp->m_dalign;
450 args.alignment = xfs_ialloc_cluster_alignment(&args);
452 * Need to figure out where to allocate the inode blocks.
453 * Ideally they should be spaced out through the a.g.
454 * For now, just allocate blocks up front.
456 args.agbno = be32_to_cpu(agi->agi_root);
457 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
459 * Allocate a fixed-size extent of inodes.
461 args.type = XFS_ALLOCTYPE_NEAR_BNO;
464 * Allow space for the inode btree to split.
466 args.minleft = args.mp->m_in_maxlevels - 1;
467 if ((error = xfs_alloc_vextent(&args)))
472 * If stripe alignment is turned on, then try again with cluster
475 if (isaligned && args.fsbno == NULLFSBLOCK) {
476 args.type = XFS_ALLOCTYPE_NEAR_BNO;
477 args.agbno = be32_to_cpu(agi->agi_root);
478 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
479 args.alignment = xfs_ialloc_cluster_alignment(&args);
480 if ((error = xfs_alloc_vextent(&args)))
484 if (args.fsbno == NULLFSBLOCK) {
488 ASSERT(args.len == args.minlen);
491 * Stamp and write the inode buffers.
493 * Seed the new inode cluster with a random generation number. This
494 * prevents short-term reuse of generation numbers if a chunk is
495 * freed and then immediately reallocated. We use random numbers
496 * rather than a linear progression to prevent the next generation
497 * number from being easily guessable.
499 error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
500 args.len, prandom_u32());
505 * Convert the results.
507 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
508 be32_add_cpu(&agi->agi_count, newlen);
509 be32_add_cpu(&agi->agi_freecount, newlen);
510 pag = xfs_perag_get(args.mp, agno);
511 pag->pagi_freecount += newlen;
513 agi->agi_newino = cpu_to_be32(newino);
516 * Insert records describing the new inode chunk into the btrees.
518 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
523 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
524 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
530 * Log allocation group header fields
532 xfs_ialloc_log_agi(tp, agbp,
533 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
535 * Modify/log superblock values for inode count and inode free count.
537 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
538 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
543 STATIC xfs_agnumber_t
549 spin_lock(&mp->m_agirotor_lock);
550 agno = mp->m_agirotor;
551 if (++mp->m_agirotor >= mp->m_maxagi)
553 spin_unlock(&mp->m_agirotor_lock);
559 * Select an allocation group to look for a free inode in, based on the parent
560 * inode and the mode. Return the allocation group buffer.
562 STATIC xfs_agnumber_t
563 xfs_ialloc_ag_select(
564 xfs_trans_t *tp, /* transaction pointer */
565 xfs_ino_t parent, /* parent directory inode number */
566 umode_t mode, /* bits set to indicate file type */
567 int okalloc) /* ok to allocate more space */
569 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
570 xfs_agnumber_t agno; /* current ag number */
571 int flags; /* alloc buffer locking flags */
572 xfs_extlen_t ineed; /* blocks needed for inode allocation */
573 xfs_extlen_t longest = 0; /* longest extent available */
574 xfs_mount_t *mp; /* mount point structure */
575 int needspace; /* file mode implies space allocated */
576 xfs_perag_t *pag; /* per allocation group data */
577 xfs_agnumber_t pagno; /* parent (starting) ag number */
581 * Files of these types need at least one block if length > 0
582 * (and they won't fit in the inode, but that's hard to figure out).
584 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
586 agcount = mp->m_maxagi;
588 pagno = xfs_ialloc_next_ag(mp);
590 pagno = XFS_INO_TO_AGNO(mp, parent);
591 if (pagno >= agcount)
595 ASSERT(pagno < agcount);
598 * Loop through allocation groups, looking for one with a little
599 * free space in it. Note we don't look for free inodes, exactly.
600 * Instead, we include whether there is a need to allocate inodes
601 * to mean that blocks must be allocated for them,
602 * if none are currently free.
605 flags = XFS_ALLOC_FLAG_TRYLOCK;
607 pag = xfs_perag_get(mp, agno);
608 if (!pag->pagi_inodeok) {
609 xfs_ialloc_next_ag(mp);
613 if (!pag->pagi_init) {
614 error = xfs_ialloc_pagi_init(mp, tp, agno);
619 if (pag->pagi_freecount) {
627 if (!pag->pagf_init) {
628 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
634 * Is there enough free space for the file plus a block of
635 * inodes? (if we need to allocate some)?
637 ineed = mp->m_ialloc_blks;
638 longest = pag->pagf_longest;
640 longest = pag->pagf_flcount > 0;
642 if (pag->pagf_freeblks >= needspace + ineed &&
650 * No point in iterating over the rest, if we're shutting
653 if (XFS_FORCED_SHUTDOWN(mp))
667 * Try to retrieve the next record to the left/right from the current one.
671 struct xfs_btree_cur *cur,
672 xfs_inobt_rec_incore_t *rec,
680 error = xfs_btree_decrement(cur, 0, &i);
682 error = xfs_btree_increment(cur, 0, &i);
688 error = xfs_inobt_get_rec(cur, rec, &i);
691 XFS_WANT_CORRUPTED_RETURN(i == 1);
699 struct xfs_btree_cur *cur,
701 xfs_inobt_rec_incore_t *rec,
707 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
712 error = xfs_inobt_get_rec(cur, rec, &i);
715 XFS_WANT_CORRUPTED_RETURN(i == 1);
722 * Allocate an inode using the inobt-only algorithm.
725 xfs_dialloc_ag_inobt(
726 struct xfs_trans *tp,
727 struct xfs_buf *agbp,
731 struct xfs_mount *mp = tp->t_mountp;
732 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
733 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
734 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
735 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
736 struct xfs_perag *pag;
737 struct xfs_btree_cur *cur, *tcur;
738 struct xfs_inobt_rec_incore rec, trec;
744 pag = xfs_perag_get(mp, agno);
746 ASSERT(pag->pagi_init);
747 ASSERT(pag->pagi_inodeok);
748 ASSERT(pag->pagi_freecount > 0);
751 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
753 * If pagino is 0 (this is the root inode allocation) use newino.
754 * This must work because we've just allocated some.
757 pagino = be32_to_cpu(agi->agi_newino);
759 error = xfs_check_agi_freecount(cur, agi);
764 * If in the same AG as the parent, try to get near the parent.
767 int doneleft; /* done, to the left */
768 int doneright; /* done, to the right */
769 int searchdistance = 10;
771 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
774 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
776 error = xfs_inobt_get_rec(cur, &rec, &j);
779 XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
781 if (rec.ir_freecount > 0) {
783 * Found a free inode in the same chunk
784 * as the parent, done.
791 * In the same AG as parent, but parent's chunk is full.
794 /* duplicate the cursor, search left & right simultaneously */
795 error = xfs_btree_dup_cursor(cur, &tcur);
800 * Skip to last blocks looked up if same parent inode.
802 if (pagino != NULLAGINO &&
803 pag->pagl_pagino == pagino &&
804 pag->pagl_leftrec != NULLAGINO &&
805 pag->pagl_rightrec != NULLAGINO) {
806 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
811 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
816 /* search left with tcur, back up 1 record */
817 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
821 /* search right with cur, go forward 1 record. */
822 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
828 * Loop until we find an inode chunk with a free inode.
830 while (!doneleft || !doneright) {
831 int useleft; /* using left inode chunk this time */
833 if (!--searchdistance) {
835 * Not in range - save last search
836 * location and allocate a new inode
838 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
839 pag->pagl_leftrec = trec.ir_startino;
840 pag->pagl_rightrec = rec.ir_startino;
841 pag->pagl_pagino = pagino;
845 /* figure out the closer block if both are valid. */
846 if (!doneleft && !doneright) {
848 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
849 rec.ir_startino - pagino;
854 /* free inodes to the left? */
855 if (useleft && trec.ir_freecount) {
857 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
860 pag->pagl_leftrec = trec.ir_startino;
861 pag->pagl_rightrec = rec.ir_startino;
862 pag->pagl_pagino = pagino;
866 /* free inodes to the right? */
867 if (!useleft && rec.ir_freecount) {
868 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
870 pag->pagl_leftrec = trec.ir_startino;
871 pag->pagl_rightrec = rec.ir_startino;
872 pag->pagl_pagino = pagino;
876 /* get next record to check */
878 error = xfs_ialloc_next_rec(tcur, &trec,
881 error = xfs_ialloc_next_rec(cur, &rec,
889 * We've reached the end of the btree. because
890 * we are only searching a small chunk of the
891 * btree each search, there is obviously free
892 * inodes closer to the parent inode than we
893 * are now. restart the search again.
895 pag->pagl_pagino = NULLAGINO;
896 pag->pagl_leftrec = NULLAGINO;
897 pag->pagl_rightrec = NULLAGINO;
898 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
899 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
904 * In a different AG from the parent.
905 * See if the most recently allocated block has any free.
908 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
909 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
915 error = xfs_inobt_get_rec(cur, &rec, &j);
919 if (j == 1 && rec.ir_freecount > 0) {
921 * The last chunk allocated in the group
922 * still has a free inode.
930 * None left in the last group, search the whole AG
932 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
935 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
938 error = xfs_inobt_get_rec(cur, &rec, &i);
941 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
942 if (rec.ir_freecount > 0)
944 error = xfs_btree_increment(cur, 0, &i);
947 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
951 offset = xfs_lowbit64(rec.ir_free);
953 ASSERT(offset < XFS_INODES_PER_CHUNK);
954 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
955 XFS_INODES_PER_CHUNK) == 0);
956 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
957 rec.ir_free &= ~XFS_INOBT_MASK(offset);
959 error = xfs_inobt_update(cur, &rec);
962 be32_add_cpu(&agi->agi_freecount, -1);
963 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
964 pag->pagi_freecount--;
966 error = xfs_check_agi_freecount(cur, agi);
970 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
971 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
976 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
978 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
984 * Use the free inode btree to allocate an inode based on distance from the
985 * parent. Note that the provided cursor may be deleted and replaced.
988 xfs_dialloc_ag_finobt_near(
990 struct xfs_btree_cur **ocur,
991 struct xfs_inobt_rec_incore *rec)
993 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
994 struct xfs_btree_cur *rcur; /* right search cursor */
995 struct xfs_inobt_rec_incore rrec;
999 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1004 error = xfs_inobt_get_rec(lcur, rec, &i);
1007 XFS_WANT_CORRUPTED_RETURN(i == 1);
1010 * See if we've landed in the parent inode record. The finobt
1011 * only tracks chunks with at least one free inode, so record
1012 * existence is enough.
1014 if (pagino >= rec->ir_startino &&
1015 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1019 error = xfs_btree_dup_cursor(lcur, &rcur);
1023 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1027 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1030 XFS_WANT_CORRUPTED_GOTO(j == 1, error_rcur);
1033 XFS_WANT_CORRUPTED_GOTO(i == 1 || j == 1, error_rcur);
1034 if (i == 1 && j == 1) {
1036 * Both the left and right records are valid. Choose the closer
1037 * inode chunk to the target.
1039 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1040 (rrec.ir_startino - pagino)) {
1042 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1045 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1047 } else if (j == 1) {
1048 /* only the right record is valid */
1050 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1052 } else if (i == 1) {
1053 /* only the left record is valid */
1054 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1060 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1065 * Use the free inode btree to find a free inode based on a newino hint. If
1066 * the hint is NULL, find the first free inode in the AG.
1069 xfs_dialloc_ag_finobt_newino(
1070 struct xfs_agi *agi,
1071 struct xfs_btree_cur *cur,
1072 struct xfs_inobt_rec_incore *rec)
1077 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1078 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1083 error = xfs_inobt_get_rec(cur, rec, &i);
1086 XFS_WANT_CORRUPTED_RETURN(i == 1);
1092 * Find the first inode available in the AG.
1094 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1097 XFS_WANT_CORRUPTED_RETURN(i == 1);
1099 error = xfs_inobt_get_rec(cur, rec, &i);
1102 XFS_WANT_CORRUPTED_RETURN(i == 1);
1108 * Update the inobt based on a modification made to the finobt. Also ensure that
1109 * the records from both trees are equivalent post-modification.
1112 xfs_dialloc_ag_update_inobt(
1113 struct xfs_btree_cur *cur, /* inobt cursor */
1114 struct xfs_inobt_rec_incore *frec, /* finobt record */
1115 int offset) /* inode offset */
1117 struct xfs_inobt_rec_incore rec;
1121 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1124 XFS_WANT_CORRUPTED_RETURN(i == 1);
1126 error = xfs_inobt_get_rec(cur, &rec, &i);
1129 XFS_WANT_CORRUPTED_RETURN(i == 1);
1130 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1131 XFS_INODES_PER_CHUNK) == 0);
1133 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1136 XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) &&
1137 (rec.ir_freecount == frec->ir_freecount));
1139 error = xfs_inobt_update(cur, &rec);
1147 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1148 * back to the inobt search algorithm.
1150 * The caller selected an AG for us, and made sure that free inodes are
1155 struct xfs_trans *tp,
1156 struct xfs_buf *agbp,
1160 struct xfs_mount *mp = tp->t_mountp;
1161 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1162 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1163 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1164 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1165 struct xfs_perag *pag;
1166 struct xfs_btree_cur *cur; /* finobt cursor */
1167 struct xfs_btree_cur *icur; /* inobt cursor */
1168 struct xfs_inobt_rec_incore rec;
1174 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1175 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1177 pag = xfs_perag_get(mp, agno);
1180 * If pagino is 0 (this is the root inode allocation) use newino.
1181 * This must work because we've just allocated some.
1184 pagino = be32_to_cpu(agi->agi_newino);
1186 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1188 error = xfs_check_agi_freecount(cur, agi);
1193 * The search algorithm depends on whether we're in the same AG as the
1194 * parent. If so, find the closest available inode to the parent. If
1195 * not, consider the agi hint or find the first free inode in the AG.
1198 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1200 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1204 offset = xfs_lowbit64(rec.ir_free);
1205 ASSERT(offset >= 0);
1206 ASSERT(offset < XFS_INODES_PER_CHUNK);
1207 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1208 XFS_INODES_PER_CHUNK) == 0);
1209 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1212 * Modify or remove the finobt record.
1214 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1216 if (rec.ir_freecount)
1217 error = xfs_inobt_update(cur, &rec);
1219 error = xfs_btree_delete(cur, &i);
1224 * The finobt has now been updated appropriately. We haven't updated the
1225 * agi and superblock yet, so we can create an inobt cursor and validate
1226 * the original freecount. If all is well, make the equivalent update to
1227 * the inobt using the finobt record and offset information.
1229 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1231 error = xfs_check_agi_freecount(icur, agi);
1235 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1240 * Both trees have now been updated. We must update the perag and
1241 * superblock before we can check the freecount for each btree.
1243 be32_add_cpu(&agi->agi_freecount, -1);
1244 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1245 pag->pagi_freecount--;
1247 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1249 error = xfs_check_agi_freecount(icur, agi);
1252 error = xfs_check_agi_freecount(cur, agi);
1256 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1257 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1263 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1265 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1271 * Allocate an inode on disk.
1273 * Mode is used to tell whether the new inode will need space, and whether it
1276 * This function is designed to be called twice if it has to do an allocation
1277 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1278 * If an inode is available without having to performn an allocation, an inode
1279 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1280 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1281 * The caller should then commit the current transaction, allocate a
1282 * new transaction, and call xfs_dialloc() again, passing in the previous value
1283 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1284 * buffer is locked across the two calls, the second call is guaranteed to have
1285 * a free inode available.
1287 * Once we successfully pick an inode its number is returned and the on-disk
1288 * data structures are updated. The inode itself is not read in, since doing so
1289 * would break ordering constraints with xfs_reclaim.
1293 struct xfs_trans *tp,
1297 struct xfs_buf **IO_agbp,
1300 struct xfs_mount *mp = tp->t_mountp;
1301 struct xfs_buf *agbp;
1302 xfs_agnumber_t agno;
1306 xfs_agnumber_t start_agno;
1307 struct xfs_perag *pag;
1311 * If the caller passes in a pointer to the AGI buffer,
1312 * continue where we left off before. In this case, we
1313 * know that the allocation group has free inodes.
1320 * We do not have an agbp, so select an initial allocation
1321 * group for inode allocation.
1323 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1324 if (start_agno == NULLAGNUMBER) {
1330 * If we have already hit the ceiling of inode blocks then clear
1331 * okalloc so we scan all available agi structures for a free
1334 if (mp->m_maxicount &&
1335 mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) {
1341 * Loop until we find an allocation group that either has free inodes
1342 * or in which we can allocate some inodes. Iterate through the
1343 * allocation groups upward, wrapping at the end.
1347 pag = xfs_perag_get(mp, agno);
1348 if (!pag->pagi_inodeok) {
1349 xfs_ialloc_next_ag(mp);
1353 if (!pag->pagi_init) {
1354 error = xfs_ialloc_pagi_init(mp, tp, agno);
1360 * Do a first racy fast path check if this AG is usable.
1362 if (!pag->pagi_freecount && !okalloc)
1366 * Then read in the AGI buffer and recheck with the AGI buffer
1369 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1373 if (pag->pagi_freecount) {
1379 goto nextag_relse_buffer;
1382 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1384 xfs_trans_brelse(tp, agbp);
1386 if (error != -ENOSPC)
1396 * We successfully allocated some inodes, return
1397 * the current context to the caller so that it
1398 * can commit the current transaction and call
1399 * us again where we left off.
1401 ASSERT(pag->pagi_freecount > 0);
1409 nextag_relse_buffer:
1410 xfs_trans_brelse(tp, agbp);
1413 if (++agno == mp->m_sb.sb_agcount)
1415 if (agno == start_agno) {
1417 return noroom ? -ENOSPC : 0;
1423 return xfs_dialloc_ag(tp, agbp, parent, inop);
1431 struct xfs_mount *mp,
1432 struct xfs_trans *tp,
1433 struct xfs_buf *agbp,
1435 struct xfs_bmap_free *flist,
1437 xfs_ino_t *first_ino,
1438 struct xfs_inobt_rec_incore *orec)
1440 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1441 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1442 struct xfs_perag *pag;
1443 struct xfs_btree_cur *cur;
1444 struct xfs_inobt_rec_incore rec;
1450 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1451 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1454 * Initialize the cursor.
1456 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1458 error = xfs_check_agi_freecount(cur, agi);
1463 * Look for the entry describing this inode.
1465 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1466 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1470 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1471 error = xfs_inobt_get_rec(cur, &rec, &i);
1473 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1477 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1479 * Get the offset in the inode chunk.
1481 off = agino - rec.ir_startino;
1482 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1483 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1485 * Mark the inode free & increment the count.
1487 rec.ir_free |= XFS_INOBT_MASK(off);
1491 * When an inode cluster is free, it becomes eligible for removal
1493 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1494 (rec.ir_freecount == mp->m_ialloc_inos)) {
1497 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1500 * Remove the inode cluster from the AGI B+Tree, adjust the
1501 * AGI and Superblock inode counts, and mark the disk space
1502 * to be freed when the transaction is committed.
1504 ilen = mp->m_ialloc_inos;
1505 be32_add_cpu(&agi->agi_count, -ilen);
1506 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1507 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1508 pag = xfs_perag_get(mp, agno);
1509 pag->pagi_freecount -= ilen - 1;
1511 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1512 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1514 if ((error = xfs_btree_delete(cur, &i))) {
1515 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1520 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1521 XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1522 mp->m_ialloc_blks, flist, mp);
1526 error = xfs_inobt_update(cur, &rec);
1528 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1534 * Change the inode free counts and log the ag/sb changes.
1536 be32_add_cpu(&agi->agi_freecount, 1);
1537 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1538 pag = xfs_perag_get(mp, agno);
1539 pag->pagi_freecount++;
1541 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1544 error = xfs_check_agi_freecount(cur, agi);
1549 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1553 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1558 * Free an inode in the free inode btree.
1562 struct xfs_mount *mp,
1563 struct xfs_trans *tp,
1564 struct xfs_buf *agbp,
1566 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
1568 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1569 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1570 struct xfs_btree_cur *cur;
1571 struct xfs_inobt_rec_incore rec;
1572 int offset = agino - ibtrec->ir_startino;
1576 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1578 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
1583 * If the record does not exist in the finobt, we must have just
1584 * freed an inode in a previously fully allocated chunk. If not,
1585 * something is out of sync.
1587 XFS_WANT_CORRUPTED_GOTO(ibtrec->ir_freecount == 1, error);
1589 error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
1590 ibtrec->ir_free, &i);
1599 * Read and update the existing record. We could just copy the ibtrec
1600 * across here, but that would defeat the purpose of having redundant
1601 * metadata. By making the modifications independently, we can catch
1602 * corruptions that we wouldn't see if we just copied from one record
1605 error = xfs_inobt_get_rec(cur, &rec, &i);
1608 XFS_WANT_CORRUPTED_GOTO(i == 1, error);
1610 rec.ir_free |= XFS_INOBT_MASK(offset);
1613 XFS_WANT_CORRUPTED_GOTO((rec.ir_free == ibtrec->ir_free) &&
1614 (rec.ir_freecount == ibtrec->ir_freecount),
1618 * The content of inobt records should always match between the inobt
1619 * and finobt. The lifecycle of records in the finobt is different from
1620 * the inobt in that the finobt only tracks records with at least one
1621 * free inode. Hence, if all of the inodes are free and we aren't
1622 * keeping inode chunks permanently on disk, remove the record.
1623 * Otherwise, update the record with the new information.
1625 if (rec.ir_freecount == mp->m_ialloc_inos &&
1626 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
1627 error = xfs_btree_delete(cur, &i);
1632 error = xfs_inobt_update(cur, &rec);
1638 error = xfs_check_agi_freecount(cur, agi);
1642 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1646 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1651 * Free disk inode. Carefully avoids touching the incore inode, all
1652 * manipulations incore are the caller's responsibility.
1653 * The on-disk inode is not changed by this operation, only the
1654 * btree (free inode mask) is changed.
1658 struct xfs_trans *tp, /* transaction pointer */
1659 xfs_ino_t inode, /* inode to be freed */
1660 struct xfs_bmap_free *flist, /* extents to free */
1661 int *deleted,/* set if inode cluster was deleted */
1662 xfs_ino_t *first_ino)/* first inode in deleted cluster */
1665 xfs_agblock_t agbno; /* block number containing inode */
1666 struct xfs_buf *agbp; /* buffer for allocation group header */
1667 xfs_agino_t agino; /* allocation group inode number */
1668 xfs_agnumber_t agno; /* allocation group number */
1669 int error; /* error return value */
1670 struct xfs_mount *mp; /* mount structure for filesystem */
1671 struct xfs_inobt_rec_incore rec;/* btree record */
1676 * Break up inode number into its components.
1678 agno = XFS_INO_TO_AGNO(mp, inode);
1679 if (agno >= mp->m_sb.sb_agcount) {
1680 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1681 __func__, agno, mp->m_sb.sb_agcount);
1685 agino = XFS_INO_TO_AGINO(mp, inode);
1686 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1687 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1688 __func__, (unsigned long long)inode,
1689 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1693 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1694 if (agbno >= mp->m_sb.sb_agblocks) {
1695 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1696 __func__, agbno, mp->m_sb.sb_agblocks);
1701 * Get the allocation group header.
1703 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1705 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1711 * Fix up the inode allocation btree.
1713 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino,
1719 * Fix up the free inode btree.
1721 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
1722 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
1735 struct xfs_mount *mp,
1736 struct xfs_trans *tp,
1737 xfs_agnumber_t agno,
1739 xfs_agblock_t agbno,
1740 xfs_agblock_t *chunk_agbno,
1741 xfs_agblock_t *offset_agbno,
1744 struct xfs_inobt_rec_incore rec;
1745 struct xfs_btree_cur *cur;
1746 struct xfs_buf *agbp;
1750 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1753 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1754 __func__, error, agno);
1759 * Lookup the inode record for the given agino. If the record cannot be
1760 * found, then it's an invalid inode number and we should abort. Once
1761 * we have a record, we need to ensure it contains the inode number
1762 * we are looking up.
1764 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1765 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1768 error = xfs_inobt_get_rec(cur, &rec, &i);
1769 if (!error && i == 0)
1773 xfs_trans_brelse(tp, agbp);
1774 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1778 /* check that the returned record contains the required inode */
1779 if (rec.ir_startino > agino ||
1780 rec.ir_startino + mp->m_ialloc_inos <= agino)
1783 /* for untrusted inodes check it is allocated first */
1784 if ((flags & XFS_IGET_UNTRUSTED) &&
1785 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1788 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1789 *offset_agbno = agbno - *chunk_agbno;
1794 * Return the location of the inode in imap, for mapping it into a buffer.
1798 xfs_mount_t *mp, /* file system mount structure */
1799 xfs_trans_t *tp, /* transaction pointer */
1800 xfs_ino_t ino, /* inode to locate */
1801 struct xfs_imap *imap, /* location map structure */
1802 uint flags) /* flags for inode btree lookup */
1804 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1805 xfs_agino_t agino; /* inode number within alloc group */
1806 xfs_agnumber_t agno; /* allocation group number */
1807 int blks_per_cluster; /* num blocks per inode cluster */
1808 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1809 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1810 int error; /* error code */
1811 int offset; /* index of inode in its buffer */
1812 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
1814 ASSERT(ino != NULLFSINO);
1817 * Split up the inode number into its parts.
1819 agno = XFS_INO_TO_AGNO(mp, ino);
1820 agino = XFS_INO_TO_AGINO(mp, ino);
1821 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1822 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1823 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1826 * Don't output diagnostic information for untrusted inodes
1827 * as they can be invalid without implying corruption.
1829 if (flags & XFS_IGET_UNTRUSTED)
1831 if (agno >= mp->m_sb.sb_agcount) {
1833 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1834 __func__, agno, mp->m_sb.sb_agcount);
1836 if (agbno >= mp->m_sb.sb_agblocks) {
1838 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1839 __func__, (unsigned long long)agbno,
1840 (unsigned long)mp->m_sb.sb_agblocks);
1842 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1844 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1846 XFS_AGINO_TO_INO(mp, agno, agino));
1853 blks_per_cluster = xfs_icluster_size_fsb(mp);
1856 * For bulkstat and handle lookups, we have an untrusted inode number
1857 * that we have to verify is valid. We cannot do this just by reading
1858 * the inode buffer as it may have been unlinked and removed leaving
1859 * inodes in stale state on disk. Hence we have to do a btree lookup
1860 * in all cases where an untrusted inode number is passed.
1862 if (flags & XFS_IGET_UNTRUSTED) {
1863 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1864 &chunk_agbno, &offset_agbno, flags);
1871 * If the inode cluster size is the same as the blocksize or
1872 * smaller we get to the buffer by simple arithmetics.
1874 if (blks_per_cluster == 1) {
1875 offset = XFS_INO_TO_OFFSET(mp, ino);
1876 ASSERT(offset < mp->m_sb.sb_inopblock);
1878 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1879 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1880 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1885 * If the inode chunks are aligned then use simple maths to
1886 * find the location. Otherwise we have to do a btree
1887 * lookup to find the location.
1889 if (mp->m_inoalign_mask) {
1890 offset_agbno = agbno & mp->m_inoalign_mask;
1891 chunk_agbno = agbno - offset_agbno;
1893 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1894 &chunk_agbno, &offset_agbno, flags);
1900 ASSERT(agbno >= chunk_agbno);
1901 cluster_agbno = chunk_agbno +
1902 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1903 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1904 XFS_INO_TO_OFFSET(mp, ino);
1906 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1907 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1908 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1911 * If the inode number maps to a block outside the bounds
1912 * of the file system then return NULL rather than calling
1913 * read_buf and panicing when we get an error from the
1916 if ((imap->im_blkno + imap->im_len) >
1917 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1919 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1920 __func__, (unsigned long long) imap->im_blkno,
1921 (unsigned long long) imap->im_len,
1922 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1929 * Compute and fill in value of m_in_maxlevels.
1932 xfs_ialloc_compute_maxlevels(
1933 xfs_mount_t *mp) /* file system mount structure */
1941 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1942 XFS_INODES_PER_CHUNK_LOG;
1943 minleafrecs = mp->m_alloc_mnr[0];
1944 minnoderecs = mp->m_alloc_mnr[1];
1945 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1946 for (level = 1; maxblocks > 1; level++)
1947 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1948 mp->m_in_maxlevels = level;
1952 * Log specified fields for the ag hdr (inode section). The growth of the agi
1953 * structure over time requires that we interpret the buffer as two logical
1954 * regions delineated by the end of the unlinked list. This is due to the size
1955 * of the hash table and its location in the middle of the agi.
1957 * For example, a request to log a field before agi_unlinked and a field after
1958 * agi_unlinked could cause us to log the entire hash table and use an excessive
1959 * amount of log space. To avoid this behavior, log the region up through
1960 * agi_unlinked in one call and the region after agi_unlinked through the end of
1961 * the structure in another.
1965 xfs_trans_t *tp, /* transaction pointer */
1966 xfs_buf_t *bp, /* allocation group header buffer */
1967 int fields) /* bitmask of fields to log */
1969 int first; /* first byte number */
1970 int last; /* last byte number */
1971 static const short offsets[] = { /* field starting offsets */
1972 /* keep in sync with bit definitions */
1973 offsetof(xfs_agi_t, agi_magicnum),
1974 offsetof(xfs_agi_t, agi_versionnum),
1975 offsetof(xfs_agi_t, agi_seqno),
1976 offsetof(xfs_agi_t, agi_length),
1977 offsetof(xfs_agi_t, agi_count),
1978 offsetof(xfs_agi_t, agi_root),
1979 offsetof(xfs_agi_t, agi_level),
1980 offsetof(xfs_agi_t, agi_freecount),
1981 offsetof(xfs_agi_t, agi_newino),
1982 offsetof(xfs_agi_t, agi_dirino),
1983 offsetof(xfs_agi_t, agi_unlinked),
1984 offsetof(xfs_agi_t, agi_free_root),
1985 offsetof(xfs_agi_t, agi_free_level),
1989 xfs_agi_t *agi; /* allocation group header */
1991 agi = XFS_BUF_TO_AGI(bp);
1992 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1995 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
1998 * Compute byte offsets for the first and last fields in the first
1999 * region and log the agi buffer. This only logs up through
2002 if (fields & XFS_AGI_ALL_BITS_R1) {
2003 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2005 xfs_trans_log_buf(tp, bp, first, last);
2009 * Mask off the bits in the first region and calculate the first and
2010 * last field offsets for any bits in the second region.
2012 fields &= ~XFS_AGI_ALL_BITS_R1;
2014 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2016 xfs_trans_log_buf(tp, bp, first, last);
2022 xfs_check_agi_unlinked(
2023 struct xfs_agi *agi)
2027 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2028 ASSERT(agi->agi_unlinked[i]);
2031 #define xfs_check_agi_unlinked(agi)
2038 struct xfs_mount *mp = bp->b_target->bt_mount;
2039 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2041 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2042 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
2045 * Validate the magic number of the agi block.
2047 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2049 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2052 if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2055 * during growfs operations, the perag is not fully initialised,
2056 * so we can't use it for any useful checking. growfs ensures we can't
2057 * use it by using uncached buffers that don't have the perag attached
2058 * so we can detect and avoid this problem.
2060 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2063 xfs_check_agi_unlinked(agi);
2068 xfs_agi_read_verify(
2071 struct xfs_mount *mp = bp->b_target->bt_mount;
2073 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2074 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2075 xfs_buf_ioerror(bp, -EFSBADCRC);
2076 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2077 XFS_ERRTAG_IALLOC_READ_AGI,
2078 XFS_RANDOM_IALLOC_READ_AGI))
2079 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2082 xfs_verifier_error(bp);
2086 xfs_agi_write_verify(
2089 struct xfs_mount *mp = bp->b_target->bt_mount;
2090 struct xfs_buf_log_item *bip = bp->b_fspriv;
2092 if (!xfs_agi_verify(bp)) {
2093 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2094 xfs_verifier_error(bp);
2098 if (!xfs_sb_version_hascrc(&mp->m_sb))
2102 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2103 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2106 const struct xfs_buf_ops xfs_agi_buf_ops = {
2107 .verify_read = xfs_agi_read_verify,
2108 .verify_write = xfs_agi_write_verify,
2112 * Read in the allocation group header (inode allocation section)
2116 struct xfs_mount *mp, /* file system mount structure */
2117 struct xfs_trans *tp, /* transaction pointer */
2118 xfs_agnumber_t agno, /* allocation group number */
2119 struct xfs_buf **bpp) /* allocation group hdr buf */
2123 trace_xfs_read_agi(mp, agno);
2125 ASSERT(agno != NULLAGNUMBER);
2126 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2127 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2128 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2132 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2137 xfs_ialloc_read_agi(
2138 struct xfs_mount *mp, /* file system mount structure */
2139 struct xfs_trans *tp, /* transaction pointer */
2140 xfs_agnumber_t agno, /* allocation group number */
2141 struct xfs_buf **bpp) /* allocation group hdr buf */
2143 struct xfs_agi *agi; /* allocation group header */
2144 struct xfs_perag *pag; /* per allocation group data */
2147 trace_xfs_ialloc_read_agi(mp, agno);
2149 error = xfs_read_agi(mp, tp, agno, bpp);
2153 agi = XFS_BUF_TO_AGI(*bpp);
2154 pag = xfs_perag_get(mp, agno);
2155 if (!pag->pagi_init) {
2156 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2157 pag->pagi_count = be32_to_cpu(agi->agi_count);
2162 * It's possible for these to be out of sync if
2163 * we are in the middle of a forced shutdown.
2165 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2166 XFS_FORCED_SHUTDOWN(mp));
2172 * Read in the agi to initialise the per-ag data in the mount structure
2175 xfs_ialloc_pagi_init(
2176 xfs_mount_t *mp, /* file system mount structure */
2177 xfs_trans_t *tp, /* transaction pointer */
2178 xfs_agnumber_t agno) /* allocation group number */
2180 xfs_buf_t *bp = NULL;
2183 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2187 xfs_trans_brelse(tp, bp);