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"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_ialloc.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_alloc.h"
32 #include "xfs_rtalloc.h"
33 #include "xfs_error.h"
35 #include "xfs_cksum.h"
36 #include "xfs_trans.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_icreate_item.h"
39 #include "xfs_icache.h"
40 #include "xfs_trace.h"
44 * Allocation group level functions.
47 xfs_ialloc_cluster_alignment(
50 if (xfs_sb_version_hasalign(&mp->m_sb) &&
51 mp->m_sb.sb_inoalignmt >=
52 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
53 return mp->m_sb.sb_inoalignmt;
58 * Lookup a record by ino in the btree given by cur.
62 struct xfs_btree_cur *cur, /* btree cursor */
63 xfs_agino_t ino, /* starting inode of chunk */
64 xfs_lookup_t dir, /* <=, >=, == */
65 int *stat) /* success/failure */
67 cur->bc_rec.i.ir_startino = ino;
68 cur->bc_rec.i.ir_freecount = 0;
69 cur->bc_rec.i.ir_free = 0;
70 return xfs_btree_lookup(cur, dir, stat);
74 * Update the record referred to by cur to the value given.
75 * This either works (return 0) or gets an EFSCORRUPTED error.
77 STATIC int /* error */
79 struct xfs_btree_cur *cur, /* btree cursor */
80 xfs_inobt_rec_incore_t *irec) /* btree record */
82 union xfs_btree_rec rec;
84 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
85 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
86 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
87 return xfs_btree_update(cur, &rec);
91 * Get the data from the pointed-to record.
95 struct xfs_btree_cur *cur, /* btree cursor */
96 xfs_inobt_rec_incore_t *irec, /* btree record */
97 int *stat) /* output: success/failure */
99 union xfs_btree_rec *rec;
102 error = xfs_btree_get_rec(cur, &rec, stat);
103 if (!error && *stat == 1) {
104 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
105 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
106 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
112 * Insert a single inobt record. Cursor must already point to desired location.
115 xfs_inobt_insert_rec(
116 struct xfs_btree_cur *cur,
121 cur->bc_rec.i.ir_freecount = freecount;
122 cur->bc_rec.i.ir_free = free;
123 return xfs_btree_insert(cur, stat);
127 * Insert records describing a newly allocated inode chunk into the inobt.
131 struct xfs_mount *mp,
132 struct xfs_trans *tp,
133 struct xfs_buf *agbp,
138 struct xfs_btree_cur *cur;
139 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
140 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
145 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
147 for (thisino = newino;
148 thisino < newino + newlen;
149 thisino += XFS_INODES_PER_CHUNK) {
150 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
152 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
157 error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
158 XFS_INOBT_ALL_FREE, &i);
160 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
166 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
172 * Verify that the number of free inodes in the AGI is correct.
176 xfs_check_agi_freecount(
177 struct xfs_btree_cur *cur,
180 if (cur->bc_nlevels == 1) {
181 xfs_inobt_rec_incore_t rec;
186 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
191 error = xfs_inobt_get_rec(cur, &rec, &i);
196 freecount += rec.ir_freecount;
197 error = xfs_btree_increment(cur, 0, &i);
203 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
204 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
209 #define xfs_check_agi_freecount(cur, agi) 0
213 * Initialise a new set of inodes. When called without a transaction context
214 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
215 * than logging them (which in a transaction context puts them into the AIL
216 * for writeback rather than the xfsbufd queue).
219 xfs_ialloc_inode_init(
220 struct xfs_mount *mp,
221 struct xfs_trans *tp,
222 struct list_head *buffer_list,
225 xfs_agblock_t length,
228 struct xfs_buf *fbuf;
229 struct xfs_dinode *free;
230 int nbufs, blks_per_cluster, inodes_per_cluster;
237 * Loop over the new block(s), filling in the inodes. For small block
238 * sizes, manipulate the inodes in buffers which are multiples of the
241 blks_per_cluster = xfs_icluster_size_fsb(mp);
242 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
243 nbufs = length / blks_per_cluster;
246 * Figure out what version number to use in the inodes we create. If
247 * the superblock version has caught up to the one that supports the new
248 * inode format, then use the new inode version. Otherwise use the old
249 * version so that old kernels will continue to be able to use the file
252 * For v3 inodes, we also need to write the inode number into the inode,
253 * so calculate the first inode number of the chunk here as
254 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
255 * across multiple filesystem blocks (such as a cluster) and so cannot
256 * be used in the cluster buffer loop below.
258 * Further, because we are writing the inode directly into the buffer
259 * and calculating a CRC on the entire inode, we have ot log the entire
260 * inode so that the entire range the CRC covers is present in the log.
261 * That means for v3 inode we log the entire buffer rather than just the
264 if (xfs_sb_version_hascrc(&mp->m_sb)) {
266 ino = XFS_AGINO_TO_INO(mp, agno,
267 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
270 * log the initialisation that is about to take place as an
271 * logical operation. This means the transaction does not
272 * need to log the physical changes to the inode buffers as log
273 * recovery will know what initialisation is actually needed.
274 * Hence we only need to log the buffers as "ordered" buffers so
275 * they track in the AIL as if they were physically logged.
278 xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
279 mp->m_sb.sb_inodesize, length, gen);
283 for (j = 0; j < nbufs; j++) {
287 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
288 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
289 mp->m_bsize * blks_per_cluster,
294 /* Initialize the inode buffers and log them appropriately. */
295 fbuf->b_ops = &xfs_inode_buf_ops;
296 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
297 for (i = 0; i < inodes_per_cluster; i++) {
298 int ioffset = i << mp->m_sb.sb_inodelog;
299 uint isize = xfs_dinode_size(version);
301 free = xfs_make_iptr(mp, fbuf, i);
302 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
303 free->di_version = version;
304 free->di_gen = cpu_to_be32(gen);
305 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
308 free->di_ino = cpu_to_be64(ino);
310 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
311 xfs_dinode_calc_crc(mp, free);
313 /* just log the inode core */
314 xfs_trans_log_buf(tp, fbuf, ioffset,
315 ioffset + isize - 1);
321 * Mark the buffer as an inode allocation buffer so it
322 * sticks in AIL at the point of this allocation
323 * transaction. This ensures the they are on disk before
324 * the tail of the log can be moved past this
325 * transaction (i.e. by preventing relogging from moving
326 * it forward in the log).
328 xfs_trans_inode_alloc_buf(tp, fbuf);
331 * Mark the buffer as ordered so that they are
332 * not physically logged in the transaction but
333 * still tracked in the AIL as part of the
334 * transaction and pin the log appropriately.
336 xfs_trans_ordered_buf(tp, fbuf);
337 xfs_trans_log_buf(tp, fbuf, 0,
338 BBTOB(fbuf->b_length) - 1);
341 fbuf->b_flags |= XBF_DONE;
342 xfs_buf_delwri_queue(fbuf, buffer_list);
350 * Allocate new inodes in the allocation group specified by agbp.
351 * Return 0 for success, else error code.
353 STATIC int /* error code or 0 */
355 xfs_trans_t *tp, /* transaction pointer */
356 xfs_buf_t *agbp, /* alloc group buffer */
359 xfs_agi_t *agi; /* allocation group header */
360 xfs_alloc_arg_t args; /* allocation argument structure */
363 xfs_agino_t newino; /* new first inode's number */
364 xfs_agino_t newlen; /* new number of inodes */
365 int isaligned = 0; /* inode allocation at stripe unit */
367 struct xfs_perag *pag;
369 memset(&args, 0, sizeof(args));
371 args.mp = tp->t_mountp;
374 * Locking will ensure that we don't have two callers in here
377 newlen = args.mp->m_ialloc_inos;
378 if (args.mp->m_maxicount &&
379 percpu_counter_read(&args.mp->m_icount) + newlen >
380 args.mp->m_maxicount)
382 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
384 * First try to allocate inodes contiguous with the last-allocated
385 * chunk of inodes. If the filesystem is striped, this will fill
386 * an entire stripe unit with inodes.
388 agi = XFS_BUF_TO_AGI(agbp);
389 newino = be32_to_cpu(agi->agi_newino);
390 agno = be32_to_cpu(agi->agi_seqno);
391 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
392 args.mp->m_ialloc_blks;
393 if (likely(newino != NULLAGINO &&
394 (args.agbno < be32_to_cpu(agi->agi_length)))) {
395 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
396 args.type = XFS_ALLOCTYPE_THIS_BNO;
400 * We need to take into account alignment here to ensure that
401 * we don't modify the free list if we fail to have an exact
402 * block. If we don't have an exact match, and every oher
403 * attempt allocation attempt fails, we'll end up cancelling
404 * a dirty transaction and shutting down.
406 * For an exact allocation, alignment must be 1,
407 * however we need to take cluster alignment into account when
408 * fixing up the freelist. Use the minalignslop field to
409 * indicate that extra blocks might be required for alignment,
410 * but not to use them in the actual exact allocation.
413 args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1;
415 /* Allow space for the inode btree to split. */
416 args.minleft = args.mp->m_in_maxlevels - 1;
417 if ((error = xfs_alloc_vextent(&args)))
421 * This request might have dirtied the transaction if the AG can
422 * satisfy the request, but the exact block was not available.
423 * If the allocation did fail, subsequent requests will relax
424 * the exact agbno requirement and increase the alignment
425 * instead. It is critical that the total size of the request
426 * (len + alignment + slop) does not increase from this point
427 * on, so reset minalignslop to ensure it is not included in
428 * subsequent requests.
430 args.minalignslop = 0;
432 args.fsbno = NULLFSBLOCK;
434 if (unlikely(args.fsbno == NULLFSBLOCK)) {
436 * Set the alignment for the allocation.
437 * If stripe alignment is turned on then align at stripe unit
439 * If the cluster size is smaller than a filesystem block
440 * then we're doing I/O for inodes in filesystem block size
441 * pieces, so don't need alignment anyway.
444 if (args.mp->m_sinoalign) {
445 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
446 args.alignment = args.mp->m_dalign;
449 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
451 * Need to figure out where to allocate the inode blocks.
452 * Ideally they should be spaced out through the a.g.
453 * For now, just allocate blocks up front.
455 args.agbno = be32_to_cpu(agi->agi_root);
456 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
458 * Allocate a fixed-size extent of inodes.
460 args.type = XFS_ALLOCTYPE_NEAR_BNO;
463 * Allow space for the inode btree to split.
465 args.minleft = args.mp->m_in_maxlevels - 1;
466 if ((error = xfs_alloc_vextent(&args)))
471 * If stripe alignment is turned on, then try again with cluster
474 if (isaligned && args.fsbno == NULLFSBLOCK) {
475 args.type = XFS_ALLOCTYPE_NEAR_BNO;
476 args.agbno = be32_to_cpu(agi->agi_root);
477 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
478 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
479 if ((error = xfs_alloc_vextent(&args)))
483 if (args.fsbno == NULLFSBLOCK) {
487 ASSERT(args.len == args.minlen);
490 * Stamp and write the inode buffers.
492 * Seed the new inode cluster with a random generation number. This
493 * prevents short-term reuse of generation numbers if a chunk is
494 * freed and then immediately reallocated. We use random numbers
495 * rather than a linear progression to prevent the next generation
496 * number from being easily guessable.
498 error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
499 args.len, prandom_u32());
504 * Convert the results.
506 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
507 be32_add_cpu(&agi->agi_count, newlen);
508 be32_add_cpu(&agi->agi_freecount, newlen);
509 pag = xfs_perag_get(args.mp, agno);
510 pag->pagi_freecount += newlen;
512 agi->agi_newino = cpu_to_be32(newino);
515 * Insert records describing the new inode chunk into the btrees.
517 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
522 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
523 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
529 * Log allocation group header fields
531 xfs_ialloc_log_agi(tp, agbp,
532 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
534 * Modify/log superblock values for inode count and inode free count.
536 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
537 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
542 STATIC xfs_agnumber_t
548 spin_lock(&mp->m_agirotor_lock);
549 agno = mp->m_agirotor;
550 if (++mp->m_agirotor >= mp->m_maxagi)
552 spin_unlock(&mp->m_agirotor_lock);
558 * Select an allocation group to look for a free inode in, based on the parent
559 * inode and the mode. Return the allocation group buffer.
561 STATIC xfs_agnumber_t
562 xfs_ialloc_ag_select(
563 xfs_trans_t *tp, /* transaction pointer */
564 xfs_ino_t parent, /* parent directory inode number */
565 umode_t mode, /* bits set to indicate file type */
566 int okalloc) /* ok to allocate more space */
568 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
569 xfs_agnumber_t agno; /* current ag number */
570 int flags; /* alloc buffer locking flags */
571 xfs_extlen_t ineed; /* blocks needed for inode allocation */
572 xfs_extlen_t longest = 0; /* longest extent available */
573 xfs_mount_t *mp; /* mount point structure */
574 int needspace; /* file mode implies space allocated */
575 xfs_perag_t *pag; /* per allocation group data */
576 xfs_agnumber_t pagno; /* parent (starting) ag number */
580 * Files of these types need at least one block if length > 0
581 * (and they won't fit in the inode, but that's hard to figure out).
583 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
585 agcount = mp->m_maxagi;
587 pagno = xfs_ialloc_next_ag(mp);
589 pagno = XFS_INO_TO_AGNO(mp, parent);
590 if (pagno >= agcount)
594 ASSERT(pagno < agcount);
597 * Loop through allocation groups, looking for one with a little
598 * free space in it. Note we don't look for free inodes, exactly.
599 * Instead, we include whether there is a need to allocate inodes
600 * to mean that blocks must be allocated for them,
601 * if none are currently free.
604 flags = XFS_ALLOC_FLAG_TRYLOCK;
606 pag = xfs_perag_get(mp, agno);
607 if (!pag->pagi_inodeok) {
608 xfs_ialloc_next_ag(mp);
612 if (!pag->pagi_init) {
613 error = xfs_ialloc_pagi_init(mp, tp, agno);
618 if (pag->pagi_freecount) {
626 if (!pag->pagf_init) {
627 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
633 * Check that there is enough free space for the file plus a
634 * chunk of inodes if we need to allocate some. If this is the
635 * first pass across the AGs, take into account the potential
636 * space needed for alignment of inode chunks when checking the
637 * longest contiguous free space in the AG - this prevents us
638 * from getting ENOSPC because we have free space larger than
639 * m_ialloc_blks but alignment constraints prevent us from using
642 * If we can't find an AG with space for full alignment slack to
643 * be taken into account, we must be near ENOSPC in all AGs.
644 * Hence we don't include alignment for the second pass and so
645 * if we fail allocation due to alignment issues then it is most
646 * likely a real ENOSPC condition.
648 ineed = mp->m_ialloc_blks;
649 if (flags && ineed > 1)
650 ineed += xfs_ialloc_cluster_alignment(mp);
651 longest = pag->pagf_longest;
653 longest = pag->pagf_flcount > 0;
655 if (pag->pagf_freeblks >= needspace + ineed &&
663 * No point in iterating over the rest, if we're shutting
666 if (XFS_FORCED_SHUTDOWN(mp))
680 * Try to retrieve the next record to the left/right from the current one.
684 struct xfs_btree_cur *cur,
685 xfs_inobt_rec_incore_t *rec,
693 error = xfs_btree_decrement(cur, 0, &i);
695 error = xfs_btree_increment(cur, 0, &i);
701 error = xfs_inobt_get_rec(cur, rec, &i);
704 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
712 struct xfs_btree_cur *cur,
714 xfs_inobt_rec_incore_t *rec,
720 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
725 error = xfs_inobt_get_rec(cur, rec, &i);
728 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
735 * Allocate an inode using the inobt-only algorithm.
738 xfs_dialloc_ag_inobt(
739 struct xfs_trans *tp,
740 struct xfs_buf *agbp,
744 struct xfs_mount *mp = tp->t_mountp;
745 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
746 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
747 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
748 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
749 struct xfs_perag *pag;
750 struct xfs_btree_cur *cur, *tcur;
751 struct xfs_inobt_rec_incore rec, trec;
757 pag = xfs_perag_get(mp, agno);
759 ASSERT(pag->pagi_init);
760 ASSERT(pag->pagi_inodeok);
761 ASSERT(pag->pagi_freecount > 0);
764 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
766 * If pagino is 0 (this is the root inode allocation) use newino.
767 * This must work because we've just allocated some.
770 pagino = be32_to_cpu(agi->agi_newino);
772 error = xfs_check_agi_freecount(cur, agi);
777 * If in the same AG as the parent, try to get near the parent.
780 int doneleft; /* done, to the left */
781 int doneright; /* done, to the right */
782 int searchdistance = 10;
784 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
787 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
789 error = xfs_inobt_get_rec(cur, &rec, &j);
792 XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
794 if (rec.ir_freecount > 0) {
796 * Found a free inode in the same chunk
797 * as the parent, done.
804 * In the same AG as parent, but parent's chunk is full.
807 /* duplicate the cursor, search left & right simultaneously */
808 error = xfs_btree_dup_cursor(cur, &tcur);
813 * Skip to last blocks looked up if same parent inode.
815 if (pagino != NULLAGINO &&
816 pag->pagl_pagino == pagino &&
817 pag->pagl_leftrec != NULLAGINO &&
818 pag->pagl_rightrec != NULLAGINO) {
819 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
824 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
829 /* search left with tcur, back up 1 record */
830 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
834 /* search right with cur, go forward 1 record. */
835 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
841 * Loop until we find an inode chunk with a free inode.
843 while (!doneleft || !doneright) {
844 int useleft; /* using left inode chunk this time */
846 if (!--searchdistance) {
848 * Not in range - save last search
849 * location and allocate a new inode
851 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
852 pag->pagl_leftrec = trec.ir_startino;
853 pag->pagl_rightrec = rec.ir_startino;
854 pag->pagl_pagino = pagino;
858 /* figure out the closer block if both are valid. */
859 if (!doneleft && !doneright) {
861 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
862 rec.ir_startino - pagino;
867 /* free inodes to the left? */
868 if (useleft && trec.ir_freecount) {
870 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
873 pag->pagl_leftrec = trec.ir_startino;
874 pag->pagl_rightrec = rec.ir_startino;
875 pag->pagl_pagino = pagino;
879 /* free inodes to the right? */
880 if (!useleft && rec.ir_freecount) {
881 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
883 pag->pagl_leftrec = trec.ir_startino;
884 pag->pagl_rightrec = rec.ir_startino;
885 pag->pagl_pagino = pagino;
889 /* get next record to check */
891 error = xfs_ialloc_next_rec(tcur, &trec,
894 error = xfs_ialloc_next_rec(cur, &rec,
902 * We've reached the end of the btree. because
903 * we are only searching a small chunk of the
904 * btree each search, there is obviously free
905 * inodes closer to the parent inode than we
906 * are now. restart the search again.
908 pag->pagl_pagino = NULLAGINO;
909 pag->pagl_leftrec = NULLAGINO;
910 pag->pagl_rightrec = NULLAGINO;
911 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
912 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
917 * In a different AG from the parent.
918 * See if the most recently allocated block has any free.
921 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
922 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
928 error = xfs_inobt_get_rec(cur, &rec, &j);
932 if (j == 1 && rec.ir_freecount > 0) {
934 * The last chunk allocated in the group
935 * still has a free inode.
943 * None left in the last group, search the whole AG
945 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
948 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
951 error = xfs_inobt_get_rec(cur, &rec, &i);
954 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
955 if (rec.ir_freecount > 0)
957 error = xfs_btree_increment(cur, 0, &i);
960 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
964 offset = xfs_lowbit64(rec.ir_free);
966 ASSERT(offset < XFS_INODES_PER_CHUNK);
967 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
968 XFS_INODES_PER_CHUNK) == 0);
969 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
970 rec.ir_free &= ~XFS_INOBT_MASK(offset);
972 error = xfs_inobt_update(cur, &rec);
975 be32_add_cpu(&agi->agi_freecount, -1);
976 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
977 pag->pagi_freecount--;
979 error = xfs_check_agi_freecount(cur, agi);
983 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
984 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
989 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
991 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
997 * Use the free inode btree to allocate an inode based on distance from the
998 * parent. Note that the provided cursor may be deleted and replaced.
1001 xfs_dialloc_ag_finobt_near(
1003 struct xfs_btree_cur **ocur,
1004 struct xfs_inobt_rec_incore *rec)
1006 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1007 struct xfs_btree_cur *rcur; /* right search cursor */
1008 struct xfs_inobt_rec_incore rrec;
1012 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1017 error = xfs_inobt_get_rec(lcur, rec, &i);
1020 XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1023 * See if we've landed in the parent inode record. The finobt
1024 * only tracks chunks with at least one free inode, so record
1025 * existence is enough.
1027 if (pagino >= rec->ir_startino &&
1028 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1032 error = xfs_btree_dup_cursor(lcur, &rcur);
1036 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1040 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1043 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1046 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1047 if (i == 1 && j == 1) {
1049 * Both the left and right records are valid. Choose the closer
1050 * inode chunk to the target.
1052 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1053 (rrec.ir_startino - pagino)) {
1055 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1058 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1060 } else if (j == 1) {
1061 /* only the right record is valid */
1063 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1065 } else if (i == 1) {
1066 /* only the left record is valid */
1067 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1073 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1078 * Use the free inode btree to find a free inode based on a newino hint. If
1079 * the hint is NULL, find the first free inode in the AG.
1082 xfs_dialloc_ag_finobt_newino(
1083 struct xfs_agi *agi,
1084 struct xfs_btree_cur *cur,
1085 struct xfs_inobt_rec_incore *rec)
1090 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1091 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1096 error = xfs_inobt_get_rec(cur, rec, &i);
1099 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1105 * Find the first inode available in the AG.
1107 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1110 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1112 error = xfs_inobt_get_rec(cur, rec, &i);
1115 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1121 * Update the inobt based on a modification made to the finobt. Also ensure that
1122 * the records from both trees are equivalent post-modification.
1125 xfs_dialloc_ag_update_inobt(
1126 struct xfs_btree_cur *cur, /* inobt cursor */
1127 struct xfs_inobt_rec_incore *frec, /* finobt record */
1128 int offset) /* inode offset */
1130 struct xfs_inobt_rec_incore rec;
1134 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1137 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1139 error = xfs_inobt_get_rec(cur, &rec, &i);
1142 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1143 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1144 XFS_INODES_PER_CHUNK) == 0);
1146 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1149 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1150 (rec.ir_freecount == frec->ir_freecount));
1152 return xfs_inobt_update(cur, &rec);
1156 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1157 * back to the inobt search algorithm.
1159 * The caller selected an AG for us, and made sure that free inodes are
1164 struct xfs_trans *tp,
1165 struct xfs_buf *agbp,
1169 struct xfs_mount *mp = tp->t_mountp;
1170 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1171 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1172 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1173 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1174 struct xfs_perag *pag;
1175 struct xfs_btree_cur *cur; /* finobt cursor */
1176 struct xfs_btree_cur *icur; /* inobt cursor */
1177 struct xfs_inobt_rec_incore rec;
1183 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1184 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1186 pag = xfs_perag_get(mp, agno);
1189 * If pagino is 0 (this is the root inode allocation) use newino.
1190 * This must work because we've just allocated some.
1193 pagino = be32_to_cpu(agi->agi_newino);
1195 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1197 error = xfs_check_agi_freecount(cur, agi);
1202 * The search algorithm depends on whether we're in the same AG as the
1203 * parent. If so, find the closest available inode to the parent. If
1204 * not, consider the agi hint or find the first free inode in the AG.
1207 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1209 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1213 offset = xfs_lowbit64(rec.ir_free);
1214 ASSERT(offset >= 0);
1215 ASSERT(offset < XFS_INODES_PER_CHUNK);
1216 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1217 XFS_INODES_PER_CHUNK) == 0);
1218 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1221 * Modify or remove the finobt record.
1223 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1225 if (rec.ir_freecount)
1226 error = xfs_inobt_update(cur, &rec);
1228 error = xfs_btree_delete(cur, &i);
1233 * The finobt has now been updated appropriately. We haven't updated the
1234 * agi and superblock yet, so we can create an inobt cursor and validate
1235 * the original freecount. If all is well, make the equivalent update to
1236 * the inobt using the finobt record and offset information.
1238 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1240 error = xfs_check_agi_freecount(icur, agi);
1244 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1249 * Both trees have now been updated. We must update the perag and
1250 * superblock before we can check the freecount for each btree.
1252 be32_add_cpu(&agi->agi_freecount, -1);
1253 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1254 pag->pagi_freecount--;
1256 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1258 error = xfs_check_agi_freecount(icur, agi);
1261 error = xfs_check_agi_freecount(cur, agi);
1265 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1266 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1272 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1274 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1280 * Allocate an inode on disk.
1282 * Mode is used to tell whether the new inode will need space, and whether it
1285 * This function is designed to be called twice if it has to do an allocation
1286 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1287 * If an inode is available without having to performn an allocation, an inode
1288 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1289 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1290 * The caller should then commit the current transaction, allocate a
1291 * new transaction, and call xfs_dialloc() again, passing in the previous value
1292 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1293 * buffer is locked across the two calls, the second call is guaranteed to have
1294 * a free inode available.
1296 * Once we successfully pick an inode its number is returned and the on-disk
1297 * data structures are updated. The inode itself is not read in, since doing so
1298 * would break ordering constraints with xfs_reclaim.
1302 struct xfs_trans *tp,
1306 struct xfs_buf **IO_agbp,
1309 struct xfs_mount *mp = tp->t_mountp;
1310 struct xfs_buf *agbp;
1311 xfs_agnumber_t agno;
1315 xfs_agnumber_t start_agno;
1316 struct xfs_perag *pag;
1320 * If the caller passes in a pointer to the AGI buffer,
1321 * continue where we left off before. In this case, we
1322 * know that the allocation group has free inodes.
1329 * We do not have an agbp, so select an initial allocation
1330 * group for inode allocation.
1332 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1333 if (start_agno == NULLAGNUMBER) {
1339 * If we have already hit the ceiling of inode blocks then clear
1340 * okalloc so we scan all available agi structures for a free
1343 if (mp->m_maxicount &&
1344 percpu_counter_read(&mp->m_icount) + mp->m_ialloc_inos >
1351 * Loop until we find an allocation group that either has free inodes
1352 * or in which we can allocate some inodes. Iterate through the
1353 * allocation groups upward, wrapping at the end.
1357 pag = xfs_perag_get(mp, agno);
1358 if (!pag->pagi_inodeok) {
1359 xfs_ialloc_next_ag(mp);
1363 if (!pag->pagi_init) {
1364 error = xfs_ialloc_pagi_init(mp, tp, agno);
1370 * Do a first racy fast path check if this AG is usable.
1372 if (!pag->pagi_freecount && !okalloc)
1376 * Then read in the AGI buffer and recheck with the AGI buffer
1379 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1383 if (pag->pagi_freecount) {
1389 goto nextag_relse_buffer;
1392 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1394 xfs_trans_brelse(tp, agbp);
1396 if (error != -ENOSPC)
1406 * We successfully allocated some inodes, return
1407 * the current context to the caller so that it
1408 * can commit the current transaction and call
1409 * us again where we left off.
1411 ASSERT(pag->pagi_freecount > 0);
1419 nextag_relse_buffer:
1420 xfs_trans_brelse(tp, agbp);
1423 if (++agno == mp->m_sb.sb_agcount)
1425 if (agno == start_agno) {
1427 return noroom ? -ENOSPC : 0;
1433 return xfs_dialloc_ag(tp, agbp, parent, inop);
1441 struct xfs_mount *mp,
1442 struct xfs_trans *tp,
1443 struct xfs_buf *agbp,
1445 struct xfs_bmap_free *flist,
1447 xfs_ino_t *first_ino,
1448 struct xfs_inobt_rec_incore *orec)
1450 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1451 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1452 struct xfs_perag *pag;
1453 struct xfs_btree_cur *cur;
1454 struct xfs_inobt_rec_incore rec;
1460 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1461 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1464 * Initialize the cursor.
1466 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1468 error = xfs_check_agi_freecount(cur, agi);
1473 * Look for the entry describing this inode.
1475 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1476 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1480 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1481 error = xfs_inobt_get_rec(cur, &rec, &i);
1483 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1487 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1489 * Get the offset in the inode chunk.
1491 off = agino - rec.ir_startino;
1492 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1493 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1495 * Mark the inode free & increment the count.
1497 rec.ir_free |= XFS_INOBT_MASK(off);
1501 * When an inode cluster is free, it becomes eligible for removal
1503 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1504 (rec.ir_freecount == mp->m_ialloc_inos)) {
1507 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1510 * Remove the inode cluster from the AGI B+Tree, adjust the
1511 * AGI and Superblock inode counts, and mark the disk space
1512 * to be freed when the transaction is committed.
1514 ilen = mp->m_ialloc_inos;
1515 be32_add_cpu(&agi->agi_count, -ilen);
1516 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1517 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1518 pag = xfs_perag_get(mp, agno);
1519 pag->pagi_freecount -= ilen - 1;
1521 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1522 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1524 if ((error = xfs_btree_delete(cur, &i))) {
1525 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1530 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1531 XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1532 mp->m_ialloc_blks, flist, mp);
1536 error = xfs_inobt_update(cur, &rec);
1538 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1544 * Change the inode free counts and log the ag/sb changes.
1546 be32_add_cpu(&agi->agi_freecount, 1);
1547 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1548 pag = xfs_perag_get(mp, agno);
1549 pag->pagi_freecount++;
1551 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1554 error = xfs_check_agi_freecount(cur, agi);
1559 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1563 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1568 * Free an inode in the free inode btree.
1572 struct xfs_mount *mp,
1573 struct xfs_trans *tp,
1574 struct xfs_buf *agbp,
1576 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
1578 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1579 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1580 struct xfs_btree_cur *cur;
1581 struct xfs_inobt_rec_incore rec;
1582 int offset = agino - ibtrec->ir_startino;
1586 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1588 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
1593 * If the record does not exist in the finobt, we must have just
1594 * freed an inode in a previously fully allocated chunk. If not,
1595 * something is out of sync.
1597 XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
1599 error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
1600 ibtrec->ir_free, &i);
1609 * Read and update the existing record. We could just copy the ibtrec
1610 * across here, but that would defeat the purpose of having redundant
1611 * metadata. By making the modifications independently, we can catch
1612 * corruptions that we wouldn't see if we just copied from one record
1615 error = xfs_inobt_get_rec(cur, &rec, &i);
1618 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
1620 rec.ir_free |= XFS_INOBT_MASK(offset);
1623 XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
1624 (rec.ir_freecount == ibtrec->ir_freecount),
1628 * The content of inobt records should always match between the inobt
1629 * and finobt. The lifecycle of records in the finobt is different from
1630 * the inobt in that the finobt only tracks records with at least one
1631 * free inode. Hence, if all of the inodes are free and we aren't
1632 * keeping inode chunks permanently on disk, remove the record.
1633 * Otherwise, update the record with the new information.
1635 if (rec.ir_freecount == mp->m_ialloc_inos &&
1636 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
1637 error = xfs_btree_delete(cur, &i);
1642 error = xfs_inobt_update(cur, &rec);
1648 error = xfs_check_agi_freecount(cur, agi);
1652 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1656 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1661 * Free disk inode. Carefully avoids touching the incore inode, all
1662 * manipulations incore are the caller's responsibility.
1663 * The on-disk inode is not changed by this operation, only the
1664 * btree (free inode mask) is changed.
1668 struct xfs_trans *tp, /* transaction pointer */
1669 xfs_ino_t inode, /* inode to be freed */
1670 struct xfs_bmap_free *flist, /* extents to free */
1671 int *deleted,/* set if inode cluster was deleted */
1672 xfs_ino_t *first_ino)/* first inode in deleted cluster */
1675 xfs_agblock_t agbno; /* block number containing inode */
1676 struct xfs_buf *agbp; /* buffer for allocation group header */
1677 xfs_agino_t agino; /* allocation group inode number */
1678 xfs_agnumber_t agno; /* allocation group number */
1679 int error; /* error return value */
1680 struct xfs_mount *mp; /* mount structure for filesystem */
1681 struct xfs_inobt_rec_incore rec;/* btree record */
1686 * Break up inode number into its components.
1688 agno = XFS_INO_TO_AGNO(mp, inode);
1689 if (agno >= mp->m_sb.sb_agcount) {
1690 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1691 __func__, agno, mp->m_sb.sb_agcount);
1695 agino = XFS_INO_TO_AGINO(mp, inode);
1696 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1697 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1698 __func__, (unsigned long long)inode,
1699 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1703 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1704 if (agbno >= mp->m_sb.sb_agblocks) {
1705 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1706 __func__, agbno, mp->m_sb.sb_agblocks);
1711 * Get the allocation group header.
1713 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1715 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1721 * Fix up the inode allocation btree.
1723 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino,
1729 * Fix up the free inode btree.
1731 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
1732 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
1745 struct xfs_mount *mp,
1746 struct xfs_trans *tp,
1747 xfs_agnumber_t agno,
1749 xfs_agblock_t agbno,
1750 xfs_agblock_t *chunk_agbno,
1751 xfs_agblock_t *offset_agbno,
1754 struct xfs_inobt_rec_incore rec;
1755 struct xfs_btree_cur *cur;
1756 struct xfs_buf *agbp;
1760 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1763 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1764 __func__, error, agno);
1769 * Lookup the inode record for the given agino. If the record cannot be
1770 * found, then it's an invalid inode number and we should abort. Once
1771 * we have a record, we need to ensure it contains the inode number
1772 * we are looking up.
1774 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1775 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1778 error = xfs_inobt_get_rec(cur, &rec, &i);
1779 if (!error && i == 0)
1783 xfs_trans_brelse(tp, agbp);
1784 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1788 /* check that the returned record contains the required inode */
1789 if (rec.ir_startino > agino ||
1790 rec.ir_startino + mp->m_ialloc_inos <= agino)
1793 /* for untrusted inodes check it is allocated first */
1794 if ((flags & XFS_IGET_UNTRUSTED) &&
1795 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1798 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1799 *offset_agbno = agbno - *chunk_agbno;
1804 * Return the location of the inode in imap, for mapping it into a buffer.
1808 xfs_mount_t *mp, /* file system mount structure */
1809 xfs_trans_t *tp, /* transaction pointer */
1810 xfs_ino_t ino, /* inode to locate */
1811 struct xfs_imap *imap, /* location map structure */
1812 uint flags) /* flags for inode btree lookup */
1814 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1815 xfs_agino_t agino; /* inode number within alloc group */
1816 xfs_agnumber_t agno; /* allocation group number */
1817 int blks_per_cluster; /* num blocks per inode cluster */
1818 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1819 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1820 int error; /* error code */
1821 int offset; /* index of inode in its buffer */
1822 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
1824 ASSERT(ino != NULLFSINO);
1827 * Split up the inode number into its parts.
1829 agno = XFS_INO_TO_AGNO(mp, ino);
1830 agino = XFS_INO_TO_AGINO(mp, ino);
1831 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1832 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1833 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1836 * Don't output diagnostic information for untrusted inodes
1837 * as they can be invalid without implying corruption.
1839 if (flags & XFS_IGET_UNTRUSTED)
1841 if (agno >= mp->m_sb.sb_agcount) {
1843 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1844 __func__, agno, mp->m_sb.sb_agcount);
1846 if (agbno >= mp->m_sb.sb_agblocks) {
1848 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1849 __func__, (unsigned long long)agbno,
1850 (unsigned long)mp->m_sb.sb_agblocks);
1852 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1854 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1856 XFS_AGINO_TO_INO(mp, agno, agino));
1863 blks_per_cluster = xfs_icluster_size_fsb(mp);
1866 * For bulkstat and handle lookups, we have an untrusted inode number
1867 * that we have to verify is valid. We cannot do this just by reading
1868 * the inode buffer as it may have been unlinked and removed leaving
1869 * inodes in stale state on disk. Hence we have to do a btree lookup
1870 * in all cases where an untrusted inode number is passed.
1872 if (flags & XFS_IGET_UNTRUSTED) {
1873 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1874 &chunk_agbno, &offset_agbno, flags);
1881 * If the inode cluster size is the same as the blocksize or
1882 * smaller we get to the buffer by simple arithmetics.
1884 if (blks_per_cluster == 1) {
1885 offset = XFS_INO_TO_OFFSET(mp, ino);
1886 ASSERT(offset < mp->m_sb.sb_inopblock);
1888 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1889 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1890 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1895 * If the inode chunks are aligned then use simple maths to
1896 * find the location. Otherwise we have to do a btree
1897 * lookup to find the location.
1899 if (mp->m_inoalign_mask) {
1900 offset_agbno = agbno & mp->m_inoalign_mask;
1901 chunk_agbno = agbno - offset_agbno;
1903 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1904 &chunk_agbno, &offset_agbno, flags);
1910 ASSERT(agbno >= chunk_agbno);
1911 cluster_agbno = chunk_agbno +
1912 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1913 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1914 XFS_INO_TO_OFFSET(mp, ino);
1916 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1917 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1918 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1921 * If the inode number maps to a block outside the bounds
1922 * of the file system then return NULL rather than calling
1923 * read_buf and panicing when we get an error from the
1926 if ((imap->im_blkno + imap->im_len) >
1927 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1929 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1930 __func__, (unsigned long long) imap->im_blkno,
1931 (unsigned long long) imap->im_len,
1932 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1939 * Compute and fill in value of m_in_maxlevels.
1942 xfs_ialloc_compute_maxlevels(
1943 xfs_mount_t *mp) /* file system mount structure */
1951 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1952 XFS_INODES_PER_CHUNK_LOG;
1953 minleafrecs = mp->m_alloc_mnr[0];
1954 minnoderecs = mp->m_alloc_mnr[1];
1955 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1956 for (level = 1; maxblocks > 1; level++)
1957 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1958 mp->m_in_maxlevels = level;
1962 * Log specified fields for the ag hdr (inode section). The growth of the agi
1963 * structure over time requires that we interpret the buffer as two logical
1964 * regions delineated by the end of the unlinked list. This is due to the size
1965 * of the hash table and its location in the middle of the agi.
1967 * For example, a request to log a field before agi_unlinked and a field after
1968 * agi_unlinked could cause us to log the entire hash table and use an excessive
1969 * amount of log space. To avoid this behavior, log the region up through
1970 * agi_unlinked in one call and the region after agi_unlinked through the end of
1971 * the structure in another.
1975 xfs_trans_t *tp, /* transaction pointer */
1976 xfs_buf_t *bp, /* allocation group header buffer */
1977 int fields) /* bitmask of fields to log */
1979 int first; /* first byte number */
1980 int last; /* last byte number */
1981 static const short offsets[] = { /* field starting offsets */
1982 /* keep in sync with bit definitions */
1983 offsetof(xfs_agi_t, agi_magicnum),
1984 offsetof(xfs_agi_t, agi_versionnum),
1985 offsetof(xfs_agi_t, agi_seqno),
1986 offsetof(xfs_agi_t, agi_length),
1987 offsetof(xfs_agi_t, agi_count),
1988 offsetof(xfs_agi_t, agi_root),
1989 offsetof(xfs_agi_t, agi_level),
1990 offsetof(xfs_agi_t, agi_freecount),
1991 offsetof(xfs_agi_t, agi_newino),
1992 offsetof(xfs_agi_t, agi_dirino),
1993 offsetof(xfs_agi_t, agi_unlinked),
1994 offsetof(xfs_agi_t, agi_free_root),
1995 offsetof(xfs_agi_t, agi_free_level),
1999 xfs_agi_t *agi; /* allocation group header */
2001 agi = XFS_BUF_TO_AGI(bp);
2002 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2005 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
2008 * Compute byte offsets for the first and last fields in the first
2009 * region and log the agi buffer. This only logs up through
2012 if (fields & XFS_AGI_ALL_BITS_R1) {
2013 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2015 xfs_trans_log_buf(tp, bp, first, last);
2019 * Mask off the bits in the first region and calculate the first and
2020 * last field offsets for any bits in the second region.
2022 fields &= ~XFS_AGI_ALL_BITS_R1;
2024 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2026 xfs_trans_log_buf(tp, bp, first, last);
2032 xfs_check_agi_unlinked(
2033 struct xfs_agi *agi)
2037 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2038 ASSERT(agi->agi_unlinked[i]);
2041 #define xfs_check_agi_unlinked(agi)
2048 struct xfs_mount *mp = bp->b_target->bt_mount;
2049 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2051 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2052 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
2055 * Validate the magic number of the agi block.
2057 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2059 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2062 if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2065 * during growfs operations, the perag is not fully initialised,
2066 * so we can't use it for any useful checking. growfs ensures we can't
2067 * use it by using uncached buffers that don't have the perag attached
2068 * so we can detect and avoid this problem.
2070 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2073 xfs_check_agi_unlinked(agi);
2078 xfs_agi_read_verify(
2081 struct xfs_mount *mp = bp->b_target->bt_mount;
2083 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2084 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2085 xfs_buf_ioerror(bp, -EFSBADCRC);
2086 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2087 XFS_ERRTAG_IALLOC_READ_AGI,
2088 XFS_RANDOM_IALLOC_READ_AGI))
2089 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2092 xfs_verifier_error(bp);
2096 xfs_agi_write_verify(
2099 struct xfs_mount *mp = bp->b_target->bt_mount;
2100 struct xfs_buf_log_item *bip = bp->b_fspriv;
2102 if (!xfs_agi_verify(bp)) {
2103 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2104 xfs_verifier_error(bp);
2108 if (!xfs_sb_version_hascrc(&mp->m_sb))
2112 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2113 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2116 const struct xfs_buf_ops xfs_agi_buf_ops = {
2117 .verify_read = xfs_agi_read_verify,
2118 .verify_write = xfs_agi_write_verify,
2122 * Read in the allocation group header (inode allocation section)
2126 struct xfs_mount *mp, /* file system mount structure */
2127 struct xfs_trans *tp, /* transaction pointer */
2128 xfs_agnumber_t agno, /* allocation group number */
2129 struct xfs_buf **bpp) /* allocation group hdr buf */
2133 trace_xfs_read_agi(mp, agno);
2135 ASSERT(agno != NULLAGNUMBER);
2136 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2137 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2138 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2142 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2147 xfs_ialloc_read_agi(
2148 struct xfs_mount *mp, /* file system mount structure */
2149 struct xfs_trans *tp, /* transaction pointer */
2150 xfs_agnumber_t agno, /* allocation group number */
2151 struct xfs_buf **bpp) /* allocation group hdr buf */
2153 struct xfs_agi *agi; /* allocation group header */
2154 struct xfs_perag *pag; /* per allocation group data */
2157 trace_xfs_ialloc_read_agi(mp, agno);
2159 error = xfs_read_agi(mp, tp, agno, bpp);
2163 agi = XFS_BUF_TO_AGI(*bpp);
2164 pag = xfs_perag_get(mp, agno);
2165 if (!pag->pagi_init) {
2166 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2167 pag->pagi_count = be32_to_cpu(agi->agi_count);
2172 * It's possible for these to be out of sync if
2173 * we are in the middle of a forced shutdown.
2175 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2176 XFS_FORCED_SHUTDOWN(mp));
2182 * Read in the agi to initialise the per-ag data in the mount structure
2185 xfs_ialloc_pagi_init(
2186 xfs_mount_t *mp, /* file system mount structure */
2187 xfs_trans_t *tp, /* transaction pointer */
2188 xfs_agnumber_t agno) /* allocation group number */
2190 xfs_buf_t *bp = NULL;
2193 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2197 xfs_trans_brelse(tp, bp);