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"
27 #include "xfs_mount.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_ialloc.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_error.h"
36 #include "xfs_cksum.h"
37 #include "xfs_trans.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_icreate_item.h"
40 #include "xfs_icache.h"
41 #include "xfs_trace.h"
45 * Allocation group level functions.
48 xfs_ialloc_cluster_alignment(
49 xfs_alloc_arg_t *args)
51 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
52 args->mp->m_sb.sb_inoalignmt >=
53 XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size))
54 return args->mp->m_sb.sb_inoalignmt;
59 * Lookup a record by ino in the btree given by cur.
63 struct xfs_btree_cur *cur, /* btree cursor */
64 xfs_agino_t ino, /* starting inode of chunk */
65 xfs_lookup_t dir, /* <=, >=, == */
66 int *stat) /* success/failure */
68 cur->bc_rec.i.ir_startino = ino;
69 cur->bc_rec.i.ir_freecount = 0;
70 cur->bc_rec.i.ir_free = 0;
71 return xfs_btree_lookup(cur, dir, stat);
75 * Update the record referred to by cur to the value given.
76 * This either works (return 0) or gets an EFSCORRUPTED error.
78 STATIC int /* error */
80 struct xfs_btree_cur *cur, /* btree cursor */
81 xfs_inobt_rec_incore_t *irec) /* btree record */
83 union xfs_btree_rec rec;
85 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
86 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
87 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
88 return xfs_btree_update(cur, &rec);
92 * Get the data from the pointed-to record.
96 struct xfs_btree_cur *cur, /* btree cursor */
97 xfs_inobt_rec_incore_t *irec, /* btree record */
98 int *stat) /* output: success/failure */
100 union xfs_btree_rec *rec;
103 error = xfs_btree_get_rec(cur, &rec, stat);
104 if (!error && *stat == 1) {
105 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
106 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
107 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
113 * Insert a single inobt record. Cursor must already point to desired location.
116 xfs_inobt_insert_rec(
117 struct xfs_btree_cur *cur,
122 cur->bc_rec.i.ir_freecount = freecount;
123 cur->bc_rec.i.ir_free = free;
124 return xfs_btree_insert(cur, stat);
128 * Insert records describing a newly allocated inode chunk into the inobt.
132 struct xfs_mount *mp,
133 struct xfs_trans *tp,
134 struct xfs_buf *agbp,
139 struct xfs_btree_cur *cur;
140 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
141 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
146 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
148 for (thisino = newino;
149 thisino < newino + newlen;
150 thisino += XFS_INODES_PER_CHUNK) {
151 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
153 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
158 error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
159 XFS_INOBT_ALL_FREE, &i);
161 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
167 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
173 * Verify that the number of free inodes in the AGI is correct.
177 xfs_check_agi_freecount(
178 struct xfs_btree_cur *cur,
181 if (cur->bc_nlevels == 1) {
182 xfs_inobt_rec_incore_t rec;
187 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
192 error = xfs_inobt_get_rec(cur, &rec, &i);
197 freecount += rec.ir_freecount;
198 error = xfs_btree_increment(cur, 0, &i);
204 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
205 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
210 #define xfs_check_agi_freecount(cur, agi) 0
214 * Initialise a new set of inodes. When called without a transaction context
215 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
216 * than logging them (which in a transaction context puts them into the AIL
217 * for writeback rather than the xfsbufd queue).
220 xfs_ialloc_inode_init(
221 struct xfs_mount *mp,
222 struct xfs_trans *tp,
223 struct list_head *buffer_list,
226 xfs_agblock_t length,
229 struct xfs_buf *fbuf;
230 struct xfs_dinode *free;
231 int nbufs, blks_per_cluster, inodes_per_cluster;
238 * Loop over the new block(s), filling in the inodes. For small block
239 * sizes, manipulate the inodes in buffers which are multiples of the
242 blks_per_cluster = xfs_icluster_size_fsb(mp);
243 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
244 nbufs = length / blks_per_cluster;
247 * Figure out what version number to use in the inodes we create. If
248 * the superblock version has caught up to the one that supports the new
249 * inode format, then use the new inode version. Otherwise use the old
250 * version so that old kernels will continue to be able to use the file
253 * For v3 inodes, we also need to write the inode number into the inode,
254 * so calculate the first inode number of the chunk here as
255 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
256 * across multiple filesystem blocks (such as a cluster) and so cannot
257 * be used in the cluster buffer loop below.
259 * Further, because we are writing the inode directly into the buffer
260 * and calculating a CRC on the entire inode, we have ot log the entire
261 * inode so that the entire range the CRC covers is present in the log.
262 * That means for v3 inode we log the entire buffer rather than just the
265 if (xfs_sb_version_hascrc(&mp->m_sb)) {
267 ino = XFS_AGINO_TO_INO(mp, agno,
268 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
271 * log the initialisation that is about to take place as an
272 * logical operation. This means the transaction does not
273 * need to log the physical changes to the inode buffers as log
274 * recovery will know what initialisation is actually needed.
275 * Hence we only need to log the buffers as "ordered" buffers so
276 * they track in the AIL as if they were physically logged.
279 xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
280 mp->m_sb.sb_inodesize, length, gen);
284 for (j = 0; j < nbufs; j++) {
288 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
289 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
290 mp->m_bsize * blks_per_cluster,
295 /* Initialize the inode buffers and log them appropriately. */
296 fbuf->b_ops = &xfs_inode_buf_ops;
297 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
298 for (i = 0; i < inodes_per_cluster; i++) {
299 int ioffset = i << mp->m_sb.sb_inodelog;
300 uint isize = xfs_dinode_size(version);
302 free = xfs_make_iptr(mp, fbuf, i);
303 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
304 free->di_version = version;
305 free->di_gen = cpu_to_be32(gen);
306 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
309 free->di_ino = cpu_to_be64(ino);
311 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
312 xfs_dinode_calc_crc(mp, free);
314 /* just log the inode core */
315 xfs_trans_log_buf(tp, fbuf, ioffset,
316 ioffset + isize - 1);
322 * Mark the buffer as an inode allocation buffer so it
323 * sticks in AIL at the point of this allocation
324 * transaction. This ensures the they are on disk before
325 * the tail of the log can be moved past this
326 * transaction (i.e. by preventing relogging from moving
327 * it forward in the log).
329 xfs_trans_inode_alloc_buf(tp, fbuf);
332 * Mark the buffer as ordered so that they are
333 * not physically logged in the transaction but
334 * still tracked in the AIL as part of the
335 * transaction and pin the log appropriately.
337 xfs_trans_ordered_buf(tp, fbuf);
338 xfs_trans_log_buf(tp, fbuf, 0,
339 BBTOB(fbuf->b_length) - 1);
342 fbuf->b_flags |= XBF_DONE;
343 xfs_buf_delwri_queue(fbuf, buffer_list);
351 * Allocate new inodes in the allocation group specified by agbp.
352 * Return 0 for success, else error code.
354 STATIC int /* error code or 0 */
356 xfs_trans_t *tp, /* transaction pointer */
357 xfs_buf_t *agbp, /* alloc group buffer */
360 xfs_agi_t *agi; /* allocation group header */
361 xfs_alloc_arg_t args; /* allocation argument structure */
364 xfs_agino_t newino; /* new first inode's number */
365 xfs_agino_t newlen; /* new number of inodes */
366 int isaligned = 0; /* inode allocation at stripe unit */
368 struct xfs_perag *pag;
370 memset(&args, 0, sizeof(args));
372 args.mp = tp->t_mountp;
375 * Locking will ensure that we don't have two callers in here
378 newlen = args.mp->m_ialloc_inos;
379 if (args.mp->m_maxicount &&
380 args.mp->m_sb.sb_icount + newlen > 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) - 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);
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);
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 * Is there enough free space for the file plus a block of
634 * inodes? (if we need to allocate some)?
636 ineed = mp->m_ialloc_blks;
637 longest = pag->pagf_longest;
639 longest = pag->pagf_flcount > 0;
641 if (pag->pagf_freeblks >= needspace + ineed &&
649 * No point in iterating over the rest, if we're shutting
652 if (XFS_FORCED_SHUTDOWN(mp))
666 * Try to retrieve the next record to the left/right from the current one.
670 struct xfs_btree_cur *cur,
671 xfs_inobt_rec_incore_t *rec,
679 error = xfs_btree_decrement(cur, 0, &i);
681 error = xfs_btree_increment(cur, 0, &i);
687 error = xfs_inobt_get_rec(cur, rec, &i);
690 XFS_WANT_CORRUPTED_RETURN(i == 1);
698 struct xfs_btree_cur *cur,
700 xfs_inobt_rec_incore_t *rec,
706 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
711 error = xfs_inobt_get_rec(cur, rec, &i);
714 XFS_WANT_CORRUPTED_RETURN(i == 1);
721 * Allocate an inode using the inobt-only algorithm.
724 xfs_dialloc_ag_inobt(
725 struct xfs_trans *tp,
726 struct xfs_buf *agbp,
730 struct xfs_mount *mp = tp->t_mountp;
731 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
732 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
733 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
734 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
735 struct xfs_perag *pag;
736 struct xfs_btree_cur *cur, *tcur;
737 struct xfs_inobt_rec_incore rec, trec;
743 pag = xfs_perag_get(mp, agno);
745 ASSERT(pag->pagi_init);
746 ASSERT(pag->pagi_inodeok);
747 ASSERT(pag->pagi_freecount > 0);
750 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
752 * If pagino is 0 (this is the root inode allocation) use newino.
753 * This must work because we've just allocated some.
756 pagino = be32_to_cpu(agi->agi_newino);
758 error = xfs_check_agi_freecount(cur, agi);
763 * If in the same AG as the parent, try to get near the parent.
766 int doneleft; /* done, to the left */
767 int doneright; /* done, to the right */
768 int searchdistance = 10;
770 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
773 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
775 error = xfs_inobt_get_rec(cur, &rec, &j);
778 XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
780 if (rec.ir_freecount > 0) {
782 * Found a free inode in the same chunk
783 * as the parent, done.
790 * In the same AG as parent, but parent's chunk is full.
793 /* duplicate the cursor, search left & right simultaneously */
794 error = xfs_btree_dup_cursor(cur, &tcur);
799 * Skip to last blocks looked up if same parent inode.
801 if (pagino != NULLAGINO &&
802 pag->pagl_pagino == pagino &&
803 pag->pagl_leftrec != NULLAGINO &&
804 pag->pagl_rightrec != NULLAGINO) {
805 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
810 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
815 /* search left with tcur, back up 1 record */
816 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
820 /* search right with cur, go forward 1 record. */
821 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
827 * Loop until we find an inode chunk with a free inode.
829 while (!doneleft || !doneright) {
830 int useleft; /* using left inode chunk this time */
832 if (!--searchdistance) {
834 * Not in range - save last search
835 * location and allocate a new inode
837 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
838 pag->pagl_leftrec = trec.ir_startino;
839 pag->pagl_rightrec = rec.ir_startino;
840 pag->pagl_pagino = pagino;
844 /* figure out the closer block if both are valid. */
845 if (!doneleft && !doneright) {
847 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
848 rec.ir_startino - pagino;
853 /* free inodes to the left? */
854 if (useleft && trec.ir_freecount) {
856 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
859 pag->pagl_leftrec = trec.ir_startino;
860 pag->pagl_rightrec = rec.ir_startino;
861 pag->pagl_pagino = pagino;
865 /* free inodes to the right? */
866 if (!useleft && rec.ir_freecount) {
867 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
869 pag->pagl_leftrec = trec.ir_startino;
870 pag->pagl_rightrec = rec.ir_startino;
871 pag->pagl_pagino = pagino;
875 /* get next record to check */
877 error = xfs_ialloc_next_rec(tcur, &trec,
880 error = xfs_ialloc_next_rec(cur, &rec,
888 * We've reached the end of the btree. because
889 * we are only searching a small chunk of the
890 * btree each search, there is obviously free
891 * inodes closer to the parent inode than we
892 * are now. restart the search again.
894 pag->pagl_pagino = NULLAGINO;
895 pag->pagl_leftrec = NULLAGINO;
896 pag->pagl_rightrec = NULLAGINO;
897 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
898 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
903 * In a different AG from the parent.
904 * See if the most recently allocated block has any free.
907 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
908 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
914 error = xfs_inobt_get_rec(cur, &rec, &j);
918 if (j == 1 && rec.ir_freecount > 0) {
920 * The last chunk allocated in the group
921 * still has a free inode.
929 * None left in the last group, search the whole AG
931 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
934 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
937 error = xfs_inobt_get_rec(cur, &rec, &i);
940 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
941 if (rec.ir_freecount > 0)
943 error = xfs_btree_increment(cur, 0, &i);
946 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
950 offset = xfs_lowbit64(rec.ir_free);
952 ASSERT(offset < XFS_INODES_PER_CHUNK);
953 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
954 XFS_INODES_PER_CHUNK) == 0);
955 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
956 rec.ir_free &= ~XFS_INOBT_MASK(offset);
958 error = xfs_inobt_update(cur, &rec);
961 be32_add_cpu(&agi->agi_freecount, -1);
962 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
963 pag->pagi_freecount--;
965 error = xfs_check_agi_freecount(cur, agi);
969 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
970 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
975 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
977 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
983 * Use the free inode btree to allocate an inode based on distance from the
984 * parent. Note that the provided cursor may be deleted and replaced.
987 xfs_dialloc_ag_finobt_near(
989 struct xfs_btree_cur **ocur,
990 struct xfs_inobt_rec_incore *rec)
992 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
993 struct xfs_btree_cur *rcur; /* right search cursor */
994 struct xfs_inobt_rec_incore rrec;
998 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1003 error = xfs_inobt_get_rec(lcur, rec, &i);
1006 XFS_WANT_CORRUPTED_RETURN(i == 1);
1009 * See if we've landed in the parent inode record. The finobt
1010 * only tracks chunks with at least one free inode, so record
1011 * existence is enough.
1013 if (pagino >= rec->ir_startino &&
1014 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1018 error = xfs_btree_dup_cursor(lcur, &rcur);
1022 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1026 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1029 XFS_WANT_CORRUPTED_GOTO(j == 1, error_rcur);
1032 XFS_WANT_CORRUPTED_GOTO(i == 1 || j == 1, error_rcur);
1033 if (i == 1 && j == 1) {
1035 * Both the left and right records are valid. Choose the closer
1036 * inode chunk to the target.
1038 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1039 (rrec.ir_startino - pagino)) {
1041 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1044 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1046 } else if (j == 1) {
1047 /* only the right record is valid */
1049 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1051 } else if (i == 1) {
1052 /* only the left record is valid */
1053 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1059 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1064 * Use the free inode btree to find a free inode based on a newino hint. If
1065 * the hint is NULL, find the first free inode in the AG.
1068 xfs_dialloc_ag_finobt_newino(
1069 struct xfs_agi *agi,
1070 struct xfs_btree_cur *cur,
1071 struct xfs_inobt_rec_incore *rec)
1076 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1077 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1082 error = xfs_inobt_get_rec(cur, rec, &i);
1085 XFS_WANT_CORRUPTED_RETURN(i == 1);
1091 * Find the first inode available in the AG.
1093 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1096 XFS_WANT_CORRUPTED_RETURN(i == 1);
1098 error = xfs_inobt_get_rec(cur, rec, &i);
1101 XFS_WANT_CORRUPTED_RETURN(i == 1);
1107 * Update the inobt based on a modification made to the finobt. Also ensure that
1108 * the records from both trees are equivalent post-modification.
1111 xfs_dialloc_ag_update_inobt(
1112 struct xfs_btree_cur *cur, /* inobt cursor */
1113 struct xfs_inobt_rec_incore *frec, /* finobt record */
1114 int offset) /* inode offset */
1116 struct xfs_inobt_rec_incore rec;
1120 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1123 XFS_WANT_CORRUPTED_RETURN(i == 1);
1125 error = xfs_inobt_get_rec(cur, &rec, &i);
1128 XFS_WANT_CORRUPTED_RETURN(i == 1);
1129 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1130 XFS_INODES_PER_CHUNK) == 0);
1132 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1135 XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) &&
1136 (rec.ir_freecount == frec->ir_freecount));
1138 error = xfs_inobt_update(cur, &rec);
1146 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1147 * back to the inobt search algorithm.
1149 * The caller selected an AG for us, and made sure that free inodes are
1154 struct xfs_trans *tp,
1155 struct xfs_buf *agbp,
1159 struct xfs_mount *mp = tp->t_mountp;
1160 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1161 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1162 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1163 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1164 struct xfs_perag *pag;
1165 struct xfs_btree_cur *cur; /* finobt cursor */
1166 struct xfs_btree_cur *icur; /* inobt cursor */
1167 struct xfs_inobt_rec_incore rec;
1173 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1174 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1176 pag = xfs_perag_get(mp, agno);
1179 * If pagino is 0 (this is the root inode allocation) use newino.
1180 * This must work because we've just allocated some.
1183 pagino = be32_to_cpu(agi->agi_newino);
1185 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1187 error = xfs_check_agi_freecount(cur, agi);
1192 * The search algorithm depends on whether we're in the same AG as the
1193 * parent. If so, find the closest available inode to the parent. If
1194 * not, consider the agi hint or find the first free inode in the AG.
1197 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1199 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1203 offset = xfs_lowbit64(rec.ir_free);
1204 ASSERT(offset >= 0);
1205 ASSERT(offset < XFS_INODES_PER_CHUNK);
1206 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1207 XFS_INODES_PER_CHUNK) == 0);
1208 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1211 * Modify or remove the finobt record.
1213 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1215 if (rec.ir_freecount)
1216 error = xfs_inobt_update(cur, &rec);
1218 error = xfs_btree_delete(cur, &i);
1223 * The finobt has now been updated appropriately. We haven't updated the
1224 * agi and superblock yet, so we can create an inobt cursor and validate
1225 * the original freecount. If all is well, make the equivalent update to
1226 * the inobt using the finobt record and offset information.
1228 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1230 error = xfs_check_agi_freecount(icur, agi);
1234 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1239 * Both trees have now been updated. We must update the perag and
1240 * superblock before we can check the freecount for each btree.
1242 be32_add_cpu(&agi->agi_freecount, -1);
1243 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1244 pag->pagi_freecount--;
1246 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1248 error = xfs_check_agi_freecount(icur, agi);
1251 error = xfs_check_agi_freecount(cur, agi);
1255 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1256 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1262 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1264 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1270 * Allocate an inode on disk.
1272 * Mode is used to tell whether the new inode will need space, and whether it
1275 * This function is designed to be called twice if it has to do an allocation
1276 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1277 * If an inode is available without having to performn an allocation, an inode
1278 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1279 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1280 * The caller should then commit the current transaction, allocate a
1281 * new transaction, and call xfs_dialloc() again, passing in the previous value
1282 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1283 * buffer is locked across the two calls, the second call is guaranteed to have
1284 * a free inode available.
1286 * Once we successfully pick an inode its number is returned and the on-disk
1287 * data structures are updated. The inode itself is not read in, since doing so
1288 * would break ordering constraints with xfs_reclaim.
1292 struct xfs_trans *tp,
1296 struct xfs_buf **IO_agbp,
1299 struct xfs_mount *mp = tp->t_mountp;
1300 struct xfs_buf *agbp;
1301 xfs_agnumber_t agno;
1305 xfs_agnumber_t start_agno;
1306 struct xfs_perag *pag;
1310 * If the caller passes in a pointer to the AGI buffer,
1311 * continue where we left off before. In this case, we
1312 * know that the allocation group has free inodes.
1319 * We do not have an agbp, so select an initial allocation
1320 * group for inode allocation.
1322 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1323 if (start_agno == NULLAGNUMBER) {
1329 * If we have already hit the ceiling of inode blocks then clear
1330 * okalloc so we scan all available agi structures for a free
1333 if (mp->m_maxicount &&
1334 mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) {
1340 * Loop until we find an allocation group that either has free inodes
1341 * or in which we can allocate some inodes. Iterate through the
1342 * allocation groups upward, wrapping at the end.
1346 pag = xfs_perag_get(mp, agno);
1347 if (!pag->pagi_inodeok) {
1348 xfs_ialloc_next_ag(mp);
1352 if (!pag->pagi_init) {
1353 error = xfs_ialloc_pagi_init(mp, tp, agno);
1359 * Do a first racy fast path check if this AG is usable.
1361 if (!pag->pagi_freecount && !okalloc)
1365 * Then read in the AGI buffer and recheck with the AGI buffer
1368 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1372 if (pag->pagi_freecount) {
1378 goto nextag_relse_buffer;
1381 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1383 xfs_trans_brelse(tp, agbp);
1385 if (error != -ENOSPC)
1395 * We successfully allocated some inodes, return
1396 * the current context to the caller so that it
1397 * can commit the current transaction and call
1398 * us again where we left off.
1400 ASSERT(pag->pagi_freecount > 0);
1408 nextag_relse_buffer:
1409 xfs_trans_brelse(tp, agbp);
1412 if (++agno == mp->m_sb.sb_agcount)
1414 if (agno == start_agno) {
1416 return noroom ? -ENOSPC : 0;
1422 return xfs_dialloc_ag(tp, agbp, parent, inop);
1430 struct xfs_mount *mp,
1431 struct xfs_trans *tp,
1432 struct xfs_buf *agbp,
1434 struct xfs_bmap_free *flist,
1436 xfs_ino_t *first_ino,
1437 struct xfs_inobt_rec_incore *orec)
1439 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1440 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1441 struct xfs_perag *pag;
1442 struct xfs_btree_cur *cur;
1443 struct xfs_inobt_rec_incore rec;
1449 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1450 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1453 * Initialize the cursor.
1455 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1457 error = xfs_check_agi_freecount(cur, agi);
1462 * Look for the entry describing this inode.
1464 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1465 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1469 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1470 error = xfs_inobt_get_rec(cur, &rec, &i);
1472 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1476 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1478 * Get the offset in the inode chunk.
1480 off = agino - rec.ir_startino;
1481 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1482 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1484 * Mark the inode free & increment the count.
1486 rec.ir_free |= XFS_INOBT_MASK(off);
1490 * When an inode cluster is free, it becomes eligible for removal
1492 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1493 (rec.ir_freecount == mp->m_ialloc_inos)) {
1496 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1499 * Remove the inode cluster from the AGI B+Tree, adjust the
1500 * AGI and Superblock inode counts, and mark the disk space
1501 * to be freed when the transaction is committed.
1503 ilen = mp->m_ialloc_inos;
1504 be32_add_cpu(&agi->agi_count, -ilen);
1505 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1506 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1507 pag = xfs_perag_get(mp, agno);
1508 pag->pagi_freecount -= ilen - 1;
1510 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1511 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1513 if ((error = xfs_btree_delete(cur, &i))) {
1514 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1519 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1520 XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1521 mp->m_ialloc_blks, flist, mp);
1525 error = xfs_inobt_update(cur, &rec);
1527 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1533 * Change the inode free counts and log the ag/sb changes.
1535 be32_add_cpu(&agi->agi_freecount, 1);
1536 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1537 pag = xfs_perag_get(mp, agno);
1538 pag->pagi_freecount++;
1540 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1543 error = xfs_check_agi_freecount(cur, agi);
1548 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1552 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1557 * Free an inode in the free inode btree.
1561 struct xfs_mount *mp,
1562 struct xfs_trans *tp,
1563 struct xfs_buf *agbp,
1565 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
1567 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1568 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1569 struct xfs_btree_cur *cur;
1570 struct xfs_inobt_rec_incore rec;
1571 int offset = agino - ibtrec->ir_startino;
1575 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1577 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
1582 * If the record does not exist in the finobt, we must have just
1583 * freed an inode in a previously fully allocated chunk. If not,
1584 * something is out of sync.
1586 XFS_WANT_CORRUPTED_GOTO(ibtrec->ir_freecount == 1, error);
1588 error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
1589 ibtrec->ir_free, &i);
1598 * Read and update the existing record. We could just copy the ibtrec
1599 * across here, but that would defeat the purpose of having redundant
1600 * metadata. By making the modifications independently, we can catch
1601 * corruptions that we wouldn't see if we just copied from one record
1604 error = xfs_inobt_get_rec(cur, &rec, &i);
1607 XFS_WANT_CORRUPTED_GOTO(i == 1, error);
1609 rec.ir_free |= XFS_INOBT_MASK(offset);
1612 XFS_WANT_CORRUPTED_GOTO((rec.ir_free == ibtrec->ir_free) &&
1613 (rec.ir_freecount == ibtrec->ir_freecount),
1617 * The content of inobt records should always match between the inobt
1618 * and finobt. The lifecycle of records in the finobt is different from
1619 * the inobt in that the finobt only tracks records with at least one
1620 * free inode. Hence, if all of the inodes are free and we aren't
1621 * keeping inode chunks permanently on disk, remove the record.
1622 * Otherwise, update the record with the new information.
1624 if (rec.ir_freecount == mp->m_ialloc_inos &&
1625 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
1626 error = xfs_btree_delete(cur, &i);
1631 error = xfs_inobt_update(cur, &rec);
1637 error = xfs_check_agi_freecount(cur, agi);
1641 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1645 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1650 * Free disk inode. Carefully avoids touching the incore inode, all
1651 * manipulations incore are the caller's responsibility.
1652 * The on-disk inode is not changed by this operation, only the
1653 * btree (free inode mask) is changed.
1657 struct xfs_trans *tp, /* transaction pointer */
1658 xfs_ino_t inode, /* inode to be freed */
1659 struct xfs_bmap_free *flist, /* extents to free */
1660 int *deleted,/* set if inode cluster was deleted */
1661 xfs_ino_t *first_ino)/* first inode in deleted cluster */
1664 xfs_agblock_t agbno; /* block number containing inode */
1665 struct xfs_buf *agbp; /* buffer for allocation group header */
1666 xfs_agino_t agino; /* allocation group inode number */
1667 xfs_agnumber_t agno; /* allocation group number */
1668 int error; /* error return value */
1669 struct xfs_mount *mp; /* mount structure for filesystem */
1670 struct xfs_inobt_rec_incore rec;/* btree record */
1675 * Break up inode number into its components.
1677 agno = XFS_INO_TO_AGNO(mp, inode);
1678 if (agno >= mp->m_sb.sb_agcount) {
1679 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1680 __func__, agno, mp->m_sb.sb_agcount);
1684 agino = XFS_INO_TO_AGINO(mp, inode);
1685 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1686 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1687 __func__, (unsigned long long)inode,
1688 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1692 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1693 if (agbno >= mp->m_sb.sb_agblocks) {
1694 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1695 __func__, agbno, mp->m_sb.sb_agblocks);
1700 * Get the allocation group header.
1702 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1704 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1710 * Fix up the inode allocation btree.
1712 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino,
1718 * Fix up the free inode btree.
1720 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
1721 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
1734 struct xfs_mount *mp,
1735 struct xfs_trans *tp,
1736 xfs_agnumber_t agno,
1738 xfs_agblock_t agbno,
1739 xfs_agblock_t *chunk_agbno,
1740 xfs_agblock_t *offset_agbno,
1743 struct xfs_inobt_rec_incore rec;
1744 struct xfs_btree_cur *cur;
1745 struct xfs_buf *agbp;
1749 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1752 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1753 __func__, error, agno);
1758 * Lookup the inode record for the given agino. If the record cannot be
1759 * found, then it's an invalid inode number and we should abort. Once
1760 * we have a record, we need to ensure it contains the inode number
1761 * we are looking up.
1763 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1764 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1767 error = xfs_inobt_get_rec(cur, &rec, &i);
1768 if (!error && i == 0)
1772 xfs_trans_brelse(tp, agbp);
1773 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1777 /* check that the returned record contains the required inode */
1778 if (rec.ir_startino > agino ||
1779 rec.ir_startino + mp->m_ialloc_inos <= agino)
1782 /* for untrusted inodes check it is allocated first */
1783 if ((flags & XFS_IGET_UNTRUSTED) &&
1784 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1787 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1788 *offset_agbno = agbno - *chunk_agbno;
1793 * Return the location of the inode in imap, for mapping it into a buffer.
1797 xfs_mount_t *mp, /* file system mount structure */
1798 xfs_trans_t *tp, /* transaction pointer */
1799 xfs_ino_t ino, /* inode to locate */
1800 struct xfs_imap *imap, /* location map structure */
1801 uint flags) /* flags for inode btree lookup */
1803 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1804 xfs_agino_t agino; /* inode number within alloc group */
1805 xfs_agnumber_t agno; /* allocation group number */
1806 int blks_per_cluster; /* num blocks per inode cluster */
1807 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1808 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1809 int error; /* error code */
1810 int offset; /* index of inode in its buffer */
1811 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
1813 ASSERT(ino != NULLFSINO);
1816 * Split up the inode number into its parts.
1818 agno = XFS_INO_TO_AGNO(mp, ino);
1819 agino = XFS_INO_TO_AGINO(mp, ino);
1820 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1821 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1822 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1825 * Don't output diagnostic information for untrusted inodes
1826 * as they can be invalid without implying corruption.
1828 if (flags & XFS_IGET_UNTRUSTED)
1830 if (agno >= mp->m_sb.sb_agcount) {
1832 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1833 __func__, agno, mp->m_sb.sb_agcount);
1835 if (agbno >= mp->m_sb.sb_agblocks) {
1837 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1838 __func__, (unsigned long long)agbno,
1839 (unsigned long)mp->m_sb.sb_agblocks);
1841 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1843 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1845 XFS_AGINO_TO_INO(mp, agno, agino));
1852 blks_per_cluster = xfs_icluster_size_fsb(mp);
1855 * For bulkstat and handle lookups, we have an untrusted inode number
1856 * that we have to verify is valid. We cannot do this just by reading
1857 * the inode buffer as it may have been unlinked and removed leaving
1858 * inodes in stale state on disk. Hence we have to do a btree lookup
1859 * in all cases where an untrusted inode number is passed.
1861 if (flags & XFS_IGET_UNTRUSTED) {
1862 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1863 &chunk_agbno, &offset_agbno, flags);
1870 * If the inode cluster size is the same as the blocksize or
1871 * smaller we get to the buffer by simple arithmetics.
1873 if (blks_per_cluster == 1) {
1874 offset = XFS_INO_TO_OFFSET(mp, ino);
1875 ASSERT(offset < mp->m_sb.sb_inopblock);
1877 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1878 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1879 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1884 * If the inode chunks are aligned then use simple maths to
1885 * find the location. Otherwise we have to do a btree
1886 * lookup to find the location.
1888 if (mp->m_inoalign_mask) {
1889 offset_agbno = agbno & mp->m_inoalign_mask;
1890 chunk_agbno = agbno - offset_agbno;
1892 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1893 &chunk_agbno, &offset_agbno, flags);
1899 ASSERT(agbno >= chunk_agbno);
1900 cluster_agbno = chunk_agbno +
1901 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1902 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1903 XFS_INO_TO_OFFSET(mp, ino);
1905 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1906 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1907 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1910 * If the inode number maps to a block outside the bounds
1911 * of the file system then return NULL rather than calling
1912 * read_buf and panicing when we get an error from the
1915 if ((imap->im_blkno + imap->im_len) >
1916 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1918 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1919 __func__, (unsigned long long) imap->im_blkno,
1920 (unsigned long long) imap->im_len,
1921 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1928 * Compute and fill in value of m_in_maxlevels.
1931 xfs_ialloc_compute_maxlevels(
1932 xfs_mount_t *mp) /* file system mount structure */
1940 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1941 XFS_INODES_PER_CHUNK_LOG;
1942 minleafrecs = mp->m_alloc_mnr[0];
1943 minnoderecs = mp->m_alloc_mnr[1];
1944 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1945 for (level = 1; maxblocks > 1; level++)
1946 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1947 mp->m_in_maxlevels = level;
1951 * Log specified fields for the ag hdr (inode section). The growth of the agi
1952 * structure over time requires that we interpret the buffer as two logical
1953 * regions delineated by the end of the unlinked list. This is due to the size
1954 * of the hash table and its location in the middle of the agi.
1956 * For example, a request to log a field before agi_unlinked and a field after
1957 * agi_unlinked could cause us to log the entire hash table and use an excessive
1958 * amount of log space. To avoid this behavior, log the region up through
1959 * agi_unlinked in one call and the region after agi_unlinked through the end of
1960 * the structure in another.
1964 xfs_trans_t *tp, /* transaction pointer */
1965 xfs_buf_t *bp, /* allocation group header buffer */
1966 int fields) /* bitmask of fields to log */
1968 int first; /* first byte number */
1969 int last; /* last byte number */
1970 static const short offsets[] = { /* field starting offsets */
1971 /* keep in sync with bit definitions */
1972 offsetof(xfs_agi_t, agi_magicnum),
1973 offsetof(xfs_agi_t, agi_versionnum),
1974 offsetof(xfs_agi_t, agi_seqno),
1975 offsetof(xfs_agi_t, agi_length),
1976 offsetof(xfs_agi_t, agi_count),
1977 offsetof(xfs_agi_t, agi_root),
1978 offsetof(xfs_agi_t, agi_level),
1979 offsetof(xfs_agi_t, agi_freecount),
1980 offsetof(xfs_agi_t, agi_newino),
1981 offsetof(xfs_agi_t, agi_dirino),
1982 offsetof(xfs_agi_t, agi_unlinked),
1983 offsetof(xfs_agi_t, agi_free_root),
1984 offsetof(xfs_agi_t, agi_free_level),
1988 xfs_agi_t *agi; /* allocation group header */
1990 agi = XFS_BUF_TO_AGI(bp);
1991 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1994 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
1997 * Compute byte offsets for the first and last fields in the first
1998 * region and log the agi buffer. This only logs up through
2001 if (fields & XFS_AGI_ALL_BITS_R1) {
2002 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2004 xfs_trans_log_buf(tp, bp, first, last);
2008 * Mask off the bits in the first region and calculate the first and
2009 * last field offsets for any bits in the second region.
2011 fields &= ~XFS_AGI_ALL_BITS_R1;
2013 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2015 xfs_trans_log_buf(tp, bp, first, last);
2021 xfs_check_agi_unlinked(
2022 struct xfs_agi *agi)
2026 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2027 ASSERT(agi->agi_unlinked[i]);
2030 #define xfs_check_agi_unlinked(agi)
2037 struct xfs_mount *mp = bp->b_target->bt_mount;
2038 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2040 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2041 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
2044 * Validate the magic number of the agi block.
2046 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2048 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2051 if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2054 * during growfs operations, the perag is not fully initialised,
2055 * so we can't use it for any useful checking. growfs ensures we can't
2056 * use it by using uncached buffers that don't have the perag attached
2057 * so we can detect and avoid this problem.
2059 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2062 xfs_check_agi_unlinked(agi);
2067 xfs_agi_read_verify(
2070 struct xfs_mount *mp = bp->b_target->bt_mount;
2072 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2073 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2074 xfs_buf_ioerror(bp, -EFSBADCRC);
2075 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2076 XFS_ERRTAG_IALLOC_READ_AGI,
2077 XFS_RANDOM_IALLOC_READ_AGI))
2078 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2081 xfs_verifier_error(bp);
2085 xfs_agi_write_verify(
2088 struct xfs_mount *mp = bp->b_target->bt_mount;
2089 struct xfs_buf_log_item *bip = bp->b_fspriv;
2091 if (!xfs_agi_verify(bp)) {
2092 xfs_buf_ioerror(bp, -EFSCORRUPTED);
2093 xfs_verifier_error(bp);
2097 if (!xfs_sb_version_hascrc(&mp->m_sb))
2101 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2102 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2105 const struct xfs_buf_ops xfs_agi_buf_ops = {
2106 .verify_read = xfs_agi_read_verify,
2107 .verify_write = xfs_agi_write_verify,
2111 * Read in the allocation group header (inode allocation section)
2115 struct xfs_mount *mp, /* file system mount structure */
2116 struct xfs_trans *tp, /* transaction pointer */
2117 xfs_agnumber_t agno, /* allocation group number */
2118 struct xfs_buf **bpp) /* allocation group hdr buf */
2122 trace_xfs_read_agi(mp, agno);
2124 ASSERT(agno != NULLAGNUMBER);
2125 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2126 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2127 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2131 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2136 xfs_ialloc_read_agi(
2137 struct xfs_mount *mp, /* file system mount structure */
2138 struct xfs_trans *tp, /* transaction pointer */
2139 xfs_agnumber_t agno, /* allocation group number */
2140 struct xfs_buf **bpp) /* allocation group hdr buf */
2142 struct xfs_agi *agi; /* allocation group header */
2143 struct xfs_perag *pag; /* per allocation group data */
2146 trace_xfs_ialloc_read_agi(mp, agno);
2148 error = xfs_read_agi(mp, tp, agno, bpp);
2152 agi = XFS_BUF_TO_AGI(*bpp);
2153 pag = xfs_perag_get(mp, agno);
2154 if (!pag->pagi_init) {
2155 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2156 pag->pagi_count = be32_to_cpu(agi->agi_count);
2161 * It's possible for these to be out of sync if
2162 * we are in the middle of a forced shutdown.
2164 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2165 XFS_FORCED_SHUTDOWN(mp));
2171 * Read in the agi to initialise the per-ag data in the mount structure
2174 xfs_ialloc_pagi_init(
2175 xfs_mount_t *mp, /* file system mount structure */
2176 xfs_trans_t *tp, /* transaction pointer */
2177 xfs_agnumber_t agno) /* allocation group number */
2179 xfs_buf_t *bp = NULL;
2182 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2186 xfs_trans_brelse(tp, bp);