2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_ialloc.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
37 #include "xfs_cksum.h"
38 #include "xfs_trans.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_icreate_item.h"
41 #include "xfs_icache.h"
42 #include "xfs_dinode.h"
43 #include "xfs_trace.h"
47 * Allocation group level functions.
50 xfs_ialloc_cluster_alignment(
51 xfs_alloc_arg_t *args)
53 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
54 args->mp->m_sb.sb_inoalignmt >=
55 XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size))
56 return args->mp->m_sb.sb_inoalignmt;
61 * Lookup a record by ino in the btree given by cur.
65 struct xfs_btree_cur *cur, /* btree cursor */
66 xfs_agino_t ino, /* starting inode of chunk */
67 xfs_lookup_t dir, /* <=, >=, == */
68 int *stat) /* success/failure */
70 cur->bc_rec.i.ir_startino = ino;
71 cur->bc_rec.i.ir_freecount = 0;
72 cur->bc_rec.i.ir_free = 0;
73 return xfs_btree_lookup(cur, dir, stat);
77 * Update the record referred to by cur to the value given.
78 * This either works (return 0) or gets an EFSCORRUPTED error.
80 STATIC int /* error */
82 struct xfs_btree_cur *cur, /* btree cursor */
83 xfs_inobt_rec_incore_t *irec) /* btree record */
85 union xfs_btree_rec rec;
87 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
88 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
89 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
90 return xfs_btree_update(cur, &rec);
94 * Get the data from the pointed-to record.
98 struct xfs_btree_cur *cur, /* btree cursor */
99 xfs_inobt_rec_incore_t *irec, /* btree record */
100 int *stat) /* output: success/failure */
102 union xfs_btree_rec *rec;
105 error = xfs_btree_get_rec(cur, &rec, stat);
106 if (!error && *stat == 1) {
107 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
108 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
109 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
115 * Insert a single inobt record. Cursor must already point to desired location.
118 xfs_inobt_insert_rec(
119 struct xfs_btree_cur *cur,
124 cur->bc_rec.i.ir_freecount = freecount;
125 cur->bc_rec.i.ir_free = free;
126 return xfs_btree_insert(cur, stat);
130 * Insert records describing a newly allocated inode chunk into the inobt.
134 struct xfs_mount *mp,
135 struct xfs_trans *tp,
136 struct xfs_buf *agbp,
141 struct xfs_btree_cur *cur;
142 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
143 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
148 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
150 for (thisino = newino;
151 thisino < newino + newlen;
152 thisino += XFS_INODES_PER_CHUNK) {
153 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
155 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
160 error = xfs_inobt_insert_rec(cur, XFS_INODES_PER_CHUNK,
161 XFS_INOBT_ALL_FREE, &i);
163 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
169 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
175 * Verify that the number of free inodes in the AGI is correct.
179 xfs_check_agi_freecount(
180 struct xfs_btree_cur *cur,
183 if (cur->bc_nlevels == 1) {
184 xfs_inobt_rec_incore_t rec;
189 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
194 error = xfs_inobt_get_rec(cur, &rec, &i);
199 freecount += rec.ir_freecount;
200 error = xfs_btree_increment(cur, 0, &i);
206 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
207 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
212 #define xfs_check_agi_freecount(cur, agi) 0
216 * Initialise a new set of inodes. When called without a transaction context
217 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
218 * than logging them (which in a transaction context puts them into the AIL
219 * for writeback rather than the xfsbufd queue).
222 xfs_ialloc_inode_init(
223 struct xfs_mount *mp,
224 struct xfs_trans *tp,
225 struct list_head *buffer_list,
228 xfs_agblock_t length,
231 struct xfs_buf *fbuf;
232 struct xfs_dinode *free;
233 int nbufs, blks_per_cluster, inodes_per_cluster;
240 * Loop over the new block(s), filling in the inodes. For small block
241 * sizes, manipulate the inodes in buffers which are multiples of the
244 blks_per_cluster = xfs_icluster_size_fsb(mp);
245 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
246 nbufs = length / blks_per_cluster;
249 * Figure out what version number to use in the inodes we create. If
250 * the superblock version has caught up to the one that supports the new
251 * inode format, then use the new inode version. Otherwise use the old
252 * version so that old kernels will continue to be able to use the file
255 * For v3 inodes, we also need to write the inode number into the inode,
256 * so calculate the first inode number of the chunk here as
257 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
258 * across multiple filesystem blocks (such as a cluster) and so cannot
259 * be used in the cluster buffer loop below.
261 * Further, because we are writing the inode directly into the buffer
262 * and calculating a CRC on the entire inode, we have ot log the entire
263 * inode so that the entire range the CRC covers is present in the log.
264 * That means for v3 inode we log the entire buffer rather than just the
267 if (xfs_sb_version_hascrc(&mp->m_sb)) {
269 ino = XFS_AGINO_TO_INO(mp, agno,
270 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
273 * log the initialisation that is about to take place as an
274 * logical operation. This means the transaction does not
275 * need to log the physical changes to the inode buffers as log
276 * recovery will know what initialisation is actually needed.
277 * Hence we only need to log the buffers as "ordered" buffers so
278 * they track in the AIL as if they were physically logged.
281 xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
282 mp->m_sb.sb_inodesize, length, gen);
286 for (j = 0; j < nbufs; j++) {
290 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
291 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
292 mp->m_bsize * blks_per_cluster,
297 /* Initialize the inode buffers and log them appropriately. */
298 fbuf->b_ops = &xfs_inode_buf_ops;
299 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
300 for (i = 0; i < inodes_per_cluster; i++) {
301 int ioffset = i << mp->m_sb.sb_inodelog;
302 uint isize = xfs_dinode_size(version);
304 free = xfs_make_iptr(mp, fbuf, i);
305 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
306 free->di_version = version;
307 free->di_gen = cpu_to_be32(gen);
308 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
311 free->di_ino = cpu_to_be64(ino);
313 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
314 xfs_dinode_calc_crc(mp, free);
316 /* just log the inode core */
317 xfs_trans_log_buf(tp, fbuf, ioffset,
318 ioffset + isize - 1);
324 * Mark the buffer as an inode allocation buffer so it
325 * sticks in AIL at the point of this allocation
326 * transaction. This ensures the they are on disk before
327 * the tail of the log can be moved past this
328 * transaction (i.e. by preventing relogging from moving
329 * it forward in the log).
331 xfs_trans_inode_alloc_buf(tp, fbuf);
334 * Mark the buffer as ordered so that they are
335 * not physically logged in the transaction but
336 * still tracked in the AIL as part of the
337 * transaction and pin the log appropriately.
339 xfs_trans_ordered_buf(tp, fbuf);
340 xfs_trans_log_buf(tp, fbuf, 0,
341 BBTOB(fbuf->b_length) - 1);
344 fbuf->b_flags |= XBF_DONE;
345 xfs_buf_delwri_queue(fbuf, buffer_list);
353 * Allocate new inodes in the allocation group specified by agbp.
354 * Return 0 for success, else error code.
356 STATIC int /* error code or 0 */
358 xfs_trans_t *tp, /* transaction pointer */
359 xfs_buf_t *agbp, /* alloc group buffer */
362 xfs_agi_t *agi; /* allocation group header */
363 xfs_alloc_arg_t args; /* allocation argument structure */
366 xfs_agino_t newino; /* new first inode's number */
367 xfs_agino_t newlen; /* new number of inodes */
368 int isaligned = 0; /* inode allocation at stripe unit */
370 struct xfs_perag *pag;
372 memset(&args, 0, sizeof(args));
374 args.mp = tp->t_mountp;
377 * Locking will ensure that we don't have two callers in here
380 newlen = args.mp->m_ialloc_inos;
381 if (args.mp->m_maxicount &&
382 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
383 return XFS_ERROR(ENOSPC);
384 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
386 * First try to allocate inodes contiguous with the last-allocated
387 * chunk of inodes. If the filesystem is striped, this will fill
388 * an entire stripe unit with inodes.
390 agi = XFS_BUF_TO_AGI(agbp);
391 newino = be32_to_cpu(agi->agi_newino);
392 agno = be32_to_cpu(agi->agi_seqno);
393 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
394 args.mp->m_ialloc_blks;
395 if (likely(newino != NULLAGINO &&
396 (args.agbno < be32_to_cpu(agi->agi_length)))) {
397 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
398 args.type = XFS_ALLOCTYPE_THIS_BNO;
402 * We need to take into account alignment here to ensure that
403 * we don't modify the free list if we fail to have an exact
404 * block. If we don't have an exact match, and every oher
405 * attempt allocation attempt fails, we'll end up cancelling
406 * a dirty transaction and shutting down.
408 * For an exact allocation, alignment must be 1,
409 * however we need to take cluster alignment into account when
410 * fixing up the freelist. Use the minalignslop field to
411 * indicate that extra blocks might be required for alignment,
412 * but not to use them in the actual exact allocation.
415 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
417 /* Allow space for the inode btree to split. */
418 args.minleft = args.mp->m_in_maxlevels - 1;
419 if ((error = xfs_alloc_vextent(&args)))
423 * This request might have dirtied the transaction if the AG can
424 * satisfy the request, but the exact block was not available.
425 * If the allocation did fail, subsequent requests will relax
426 * the exact agbno requirement and increase the alignment
427 * instead. It is critical that the total size of the request
428 * (len + alignment + slop) does not increase from this point
429 * on, so reset minalignslop to ensure it is not included in
430 * subsequent requests.
432 args.minalignslop = 0;
434 args.fsbno = NULLFSBLOCK;
436 if (unlikely(args.fsbno == NULLFSBLOCK)) {
438 * Set the alignment for the allocation.
439 * If stripe alignment is turned on then align at stripe unit
441 * If the cluster size is smaller than a filesystem block
442 * then we're doing I/O for inodes in filesystem block size
443 * pieces, so don't need alignment anyway.
446 if (args.mp->m_sinoalign) {
447 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
448 args.alignment = args.mp->m_dalign;
451 args.alignment = xfs_ialloc_cluster_alignment(&args);
453 * Need to figure out where to allocate the inode blocks.
454 * Ideally they should be spaced out through the a.g.
455 * For now, just allocate blocks up front.
457 args.agbno = be32_to_cpu(agi->agi_root);
458 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
460 * Allocate a fixed-size extent of inodes.
462 args.type = XFS_ALLOCTYPE_NEAR_BNO;
465 * Allow space for the inode btree to split.
467 args.minleft = args.mp->m_in_maxlevels - 1;
468 if ((error = xfs_alloc_vextent(&args)))
473 * If stripe alignment is turned on, then try again with cluster
476 if (isaligned && args.fsbno == NULLFSBLOCK) {
477 args.type = XFS_ALLOCTYPE_NEAR_BNO;
478 args.agbno = be32_to_cpu(agi->agi_root);
479 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
480 args.alignment = xfs_ialloc_cluster_alignment(&args);
481 if ((error = xfs_alloc_vextent(&args)))
485 if (args.fsbno == NULLFSBLOCK) {
489 ASSERT(args.len == args.minlen);
492 * Stamp and write the inode buffers.
494 * Seed the new inode cluster with a random generation number. This
495 * prevents short-term reuse of generation numbers if a chunk is
496 * freed and then immediately reallocated. We use random numbers
497 * rather than a linear progression to prevent the next generation
498 * number from being easily guessable.
500 error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
501 args.len, prandom_u32());
506 * Convert the results.
508 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
509 be32_add_cpu(&agi->agi_count, newlen);
510 be32_add_cpu(&agi->agi_freecount, newlen);
511 pag = xfs_perag_get(args.mp, agno);
512 pag->pagi_freecount += newlen;
514 agi->agi_newino = cpu_to_be32(newino);
517 * Insert records describing the new inode chunk into the btrees.
519 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
524 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
525 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
531 * Log allocation group header fields
533 xfs_ialloc_log_agi(tp, agbp,
534 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
536 * Modify/log superblock values for inode count and inode free count.
538 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
539 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
544 STATIC xfs_agnumber_t
550 spin_lock(&mp->m_agirotor_lock);
551 agno = mp->m_agirotor;
552 if (++mp->m_agirotor >= mp->m_maxagi)
554 spin_unlock(&mp->m_agirotor_lock);
560 * Select an allocation group to look for a free inode in, based on the parent
561 * inode and the mode. Return the allocation group buffer.
563 STATIC xfs_agnumber_t
564 xfs_ialloc_ag_select(
565 xfs_trans_t *tp, /* transaction pointer */
566 xfs_ino_t parent, /* parent directory inode number */
567 umode_t mode, /* bits set to indicate file type */
568 int okalloc) /* ok to allocate more space */
570 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
571 xfs_agnumber_t agno; /* current ag number */
572 int flags; /* alloc buffer locking flags */
573 xfs_extlen_t ineed; /* blocks needed for inode allocation */
574 xfs_extlen_t longest = 0; /* longest extent available */
575 xfs_mount_t *mp; /* mount point structure */
576 int needspace; /* file mode implies space allocated */
577 xfs_perag_t *pag; /* per allocation group data */
578 xfs_agnumber_t pagno; /* parent (starting) ag number */
582 * Files of these types need at least one block if length > 0
583 * (and they won't fit in the inode, but that's hard to figure out).
585 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
587 agcount = mp->m_maxagi;
589 pagno = xfs_ialloc_next_ag(mp);
591 pagno = XFS_INO_TO_AGNO(mp, parent);
592 if (pagno >= agcount)
596 ASSERT(pagno < agcount);
599 * Loop through allocation groups, looking for one with a little
600 * free space in it. Note we don't look for free inodes, exactly.
601 * Instead, we include whether there is a need to allocate inodes
602 * to mean that blocks must be allocated for them,
603 * if none are currently free.
606 flags = XFS_ALLOC_FLAG_TRYLOCK;
608 pag = xfs_perag_get(mp, agno);
609 if (!pag->pagi_inodeok) {
610 xfs_ialloc_next_ag(mp);
614 if (!pag->pagi_init) {
615 error = xfs_ialloc_pagi_init(mp, tp, agno);
620 if (pag->pagi_freecount) {
628 if (!pag->pagf_init) {
629 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
635 * Is there enough free space for the file plus a block of
636 * inodes? (if we need to allocate some)?
638 ineed = mp->m_ialloc_blks;
639 longest = pag->pagf_longest;
641 longest = pag->pagf_flcount > 0;
643 if (pag->pagf_freeblks >= needspace + ineed &&
651 * No point in iterating over the rest, if we're shutting
654 if (XFS_FORCED_SHUTDOWN(mp))
668 * Try to retrieve the next record to the left/right from the current one.
672 struct xfs_btree_cur *cur,
673 xfs_inobt_rec_incore_t *rec,
681 error = xfs_btree_decrement(cur, 0, &i);
683 error = xfs_btree_increment(cur, 0, &i);
689 error = xfs_inobt_get_rec(cur, rec, &i);
692 XFS_WANT_CORRUPTED_RETURN(i == 1);
700 struct xfs_btree_cur *cur,
702 xfs_inobt_rec_incore_t *rec,
708 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
713 error = xfs_inobt_get_rec(cur, rec, &i);
716 XFS_WANT_CORRUPTED_RETURN(i == 1);
723 * Allocate an inode using the inobt-only algorithm.
726 xfs_dialloc_ag_inobt(
727 struct xfs_trans *tp,
728 struct xfs_buf *agbp,
732 struct xfs_mount *mp = tp->t_mountp;
733 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
734 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
735 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
736 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
737 struct xfs_perag *pag;
738 struct xfs_btree_cur *cur, *tcur;
739 struct xfs_inobt_rec_incore rec, trec;
745 pag = xfs_perag_get(mp, agno);
747 ASSERT(pag->pagi_init);
748 ASSERT(pag->pagi_inodeok);
749 ASSERT(pag->pagi_freecount > 0);
752 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
754 * If pagino is 0 (this is the root inode allocation) use newino.
755 * This must work because we've just allocated some.
758 pagino = be32_to_cpu(agi->agi_newino);
760 error = xfs_check_agi_freecount(cur, agi);
765 * If in the same AG as the parent, try to get near the parent.
768 int doneleft; /* done, to the left */
769 int doneright; /* done, to the right */
770 int searchdistance = 10;
772 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
775 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
777 error = xfs_inobt_get_rec(cur, &rec, &j);
780 XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
782 if (rec.ir_freecount > 0) {
784 * Found a free inode in the same chunk
785 * as the parent, done.
792 * In the same AG as parent, but parent's chunk is full.
795 /* duplicate the cursor, search left & right simultaneously */
796 error = xfs_btree_dup_cursor(cur, &tcur);
801 * Skip to last blocks looked up if same parent inode.
803 if (pagino != NULLAGINO &&
804 pag->pagl_pagino == pagino &&
805 pag->pagl_leftrec != NULLAGINO &&
806 pag->pagl_rightrec != NULLAGINO) {
807 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
812 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
817 /* search left with tcur, back up 1 record */
818 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
822 /* search right with cur, go forward 1 record. */
823 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
829 * Loop until we find an inode chunk with a free inode.
831 while (!doneleft || !doneright) {
832 int useleft; /* using left inode chunk this time */
834 if (!--searchdistance) {
836 * Not in range - save last search
837 * location and allocate a new inode
839 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
840 pag->pagl_leftrec = trec.ir_startino;
841 pag->pagl_rightrec = rec.ir_startino;
842 pag->pagl_pagino = pagino;
846 /* figure out the closer block if both are valid. */
847 if (!doneleft && !doneright) {
849 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
850 rec.ir_startino - pagino;
855 /* free inodes to the left? */
856 if (useleft && trec.ir_freecount) {
858 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
861 pag->pagl_leftrec = trec.ir_startino;
862 pag->pagl_rightrec = rec.ir_startino;
863 pag->pagl_pagino = pagino;
867 /* free inodes to the right? */
868 if (!useleft && rec.ir_freecount) {
869 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
871 pag->pagl_leftrec = trec.ir_startino;
872 pag->pagl_rightrec = rec.ir_startino;
873 pag->pagl_pagino = pagino;
877 /* get next record to check */
879 error = xfs_ialloc_next_rec(tcur, &trec,
882 error = xfs_ialloc_next_rec(cur, &rec,
890 * We've reached the end of the btree. because
891 * we are only searching a small chunk of the
892 * btree each search, there is obviously free
893 * inodes closer to the parent inode than we
894 * are now. restart the search again.
896 pag->pagl_pagino = NULLAGINO;
897 pag->pagl_leftrec = NULLAGINO;
898 pag->pagl_rightrec = NULLAGINO;
899 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
900 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
905 * In a different AG from the parent.
906 * See if the most recently allocated block has any free.
909 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
910 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
916 error = xfs_inobt_get_rec(cur, &rec, &j);
920 if (j == 1 && rec.ir_freecount > 0) {
922 * The last chunk allocated in the group
923 * still has a free inode.
931 * None left in the last group, search the whole AG
933 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
936 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
939 error = xfs_inobt_get_rec(cur, &rec, &i);
942 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
943 if (rec.ir_freecount > 0)
945 error = xfs_btree_increment(cur, 0, &i);
948 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
952 offset = xfs_lowbit64(rec.ir_free);
954 ASSERT(offset < XFS_INODES_PER_CHUNK);
955 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
956 XFS_INODES_PER_CHUNK) == 0);
957 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
958 rec.ir_free &= ~XFS_INOBT_MASK(offset);
960 error = xfs_inobt_update(cur, &rec);
963 be32_add_cpu(&agi->agi_freecount, -1);
964 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
965 pag->pagi_freecount--;
967 error = xfs_check_agi_freecount(cur, agi);
971 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
972 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
977 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
979 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
985 * Use the free inode btree to allocate an inode based on distance from the
986 * parent. Note that the provided cursor may be deleted and replaced.
989 xfs_dialloc_ag_finobt_near(
991 struct xfs_btree_cur **ocur,
992 struct xfs_inobt_rec_incore *rec)
994 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
995 struct xfs_btree_cur *rcur; /* right search cursor */
996 struct xfs_inobt_rec_incore rrec;
1000 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1005 error = xfs_inobt_get_rec(lcur, rec, &i);
1008 XFS_WANT_CORRUPTED_RETURN(i == 1);
1011 * See if we've landed in the parent inode record. The finobt
1012 * only tracks chunks with at least one free inode, so record
1013 * existence is enough.
1015 if (pagino >= rec->ir_startino &&
1016 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1020 error = xfs_btree_dup_cursor(lcur, &rcur);
1024 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1028 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1031 XFS_WANT_CORRUPTED_GOTO(j == 1, error_rcur);
1034 XFS_WANT_CORRUPTED_GOTO(i == 1 || j == 1, error_rcur);
1035 if (i == 1 && j == 1) {
1037 * Both the left and right records are valid. Choose the closer
1038 * inode chunk to the target.
1040 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1041 (rrec.ir_startino - pagino)) {
1043 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1046 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1048 } else if (j == 1) {
1049 /* only the right record is valid */
1051 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1053 } else if (i == 1) {
1054 /* only the left record is valid */
1055 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1061 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1066 * Use the free inode btree to find a free inode based on a newino hint. If
1067 * the hint is NULL, find the first free inode in the AG.
1070 xfs_dialloc_ag_finobt_newino(
1071 struct xfs_agi *agi,
1072 struct xfs_btree_cur *cur,
1073 struct xfs_inobt_rec_incore *rec)
1078 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1079 error = xfs_inobt_lookup(cur, agi->agi_newino, XFS_LOOKUP_EQ,
1084 error = xfs_inobt_get_rec(cur, rec, &i);
1087 XFS_WANT_CORRUPTED_RETURN(i == 1);
1094 * Find the first inode available in the AG.
1096 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1099 XFS_WANT_CORRUPTED_RETURN(i == 1);
1101 error = xfs_inobt_get_rec(cur, rec, &i);
1104 XFS_WANT_CORRUPTED_RETURN(i == 1);
1110 * Update the inobt based on a modification made to the finobt. Also ensure that
1111 * the records from both trees are equivalent post-modification.
1114 xfs_dialloc_ag_update_inobt(
1115 struct xfs_btree_cur *cur, /* inobt cursor */
1116 struct xfs_inobt_rec_incore *frec, /* finobt record */
1117 int offset) /* inode offset */
1119 struct xfs_inobt_rec_incore rec;
1123 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1126 XFS_WANT_CORRUPTED_RETURN(i == 1);
1128 error = xfs_inobt_get_rec(cur, &rec, &i);
1131 XFS_WANT_CORRUPTED_RETURN(i == 1);
1132 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1133 XFS_INODES_PER_CHUNK) == 0);
1135 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1138 XFS_WANT_CORRUPTED_RETURN((rec.ir_free == frec->ir_free) &&
1139 (rec.ir_freecount == frec->ir_freecount));
1141 error = xfs_inobt_update(cur, &rec);
1149 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1150 * back to the inobt search algorithm.
1152 * The caller selected an AG for us, and made sure that free inodes are
1157 struct xfs_trans *tp,
1158 struct xfs_buf *agbp,
1162 struct xfs_mount *mp = tp->t_mountp;
1163 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1164 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1165 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1166 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1167 struct xfs_perag *pag;
1168 struct xfs_btree_cur *cur; /* finobt cursor */
1169 struct xfs_btree_cur *icur; /* inobt cursor */
1170 struct xfs_inobt_rec_incore rec;
1176 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1177 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1179 pag = xfs_perag_get(mp, agno);
1182 * If pagino is 0 (this is the root inode allocation) use newino.
1183 * This must work because we've just allocated some.
1186 pagino = be32_to_cpu(agi->agi_newino);
1188 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1190 error = xfs_check_agi_freecount(cur, agi);
1195 * The search algorithm depends on whether we're in the same AG as the
1196 * parent. If so, find the closest available inode to the parent. If
1197 * not, consider the agi hint or find the first free inode in the AG.
1200 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1202 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1206 offset = xfs_lowbit64(rec.ir_free);
1207 ASSERT(offset >= 0);
1208 ASSERT(offset < XFS_INODES_PER_CHUNK);
1209 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1210 XFS_INODES_PER_CHUNK) == 0);
1211 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1214 * Modify or remove the finobt record.
1216 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1218 if (rec.ir_freecount)
1219 error = xfs_inobt_update(cur, &rec);
1221 error = xfs_btree_delete(cur, &i);
1226 * The finobt has now been updated appropriately. We haven't updated the
1227 * agi and superblock yet, so we can create an inobt cursor and validate
1228 * the original freecount. If all is well, make the equivalent update to
1229 * the inobt using the finobt record and offset information.
1231 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1233 error = xfs_check_agi_freecount(icur, agi);
1237 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1242 * Both trees have now been updated. We must update the perag and
1243 * superblock before we can check the freecount for each btree.
1245 be32_add_cpu(&agi->agi_freecount, -1);
1246 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1247 pag->pagi_freecount--;
1249 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1251 error = xfs_check_agi_freecount(icur, agi);
1254 error = xfs_check_agi_freecount(cur, agi);
1258 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1259 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1265 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1267 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1273 * Allocate an inode on disk.
1275 * Mode is used to tell whether the new inode will need space, and whether it
1278 * This function is designed to be called twice if it has to do an allocation
1279 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1280 * If an inode is available without having to performn an allocation, an inode
1281 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1282 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1283 * The caller should then commit the current transaction, allocate a
1284 * new transaction, and call xfs_dialloc() again, passing in the previous value
1285 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1286 * buffer is locked across the two calls, the second call is guaranteed to have
1287 * a free inode available.
1289 * Once we successfully pick an inode its number is returned and the on-disk
1290 * data structures are updated. The inode itself is not read in, since doing so
1291 * would break ordering constraints with xfs_reclaim.
1295 struct xfs_trans *tp,
1299 struct xfs_buf **IO_agbp,
1302 struct xfs_mount *mp = tp->t_mountp;
1303 struct xfs_buf *agbp;
1304 xfs_agnumber_t agno;
1308 xfs_agnumber_t start_agno;
1309 struct xfs_perag *pag;
1313 * If the caller passes in a pointer to the AGI buffer,
1314 * continue where we left off before. In this case, we
1315 * know that the allocation group has free inodes.
1322 * We do not have an agbp, so select an initial allocation
1323 * group for inode allocation.
1325 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1326 if (start_agno == NULLAGNUMBER) {
1332 * If we have already hit the ceiling of inode blocks then clear
1333 * okalloc so we scan all available agi structures for a free
1336 if (mp->m_maxicount &&
1337 mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) {
1343 * Loop until we find an allocation group that either has free inodes
1344 * or in which we can allocate some inodes. Iterate through the
1345 * allocation groups upward, wrapping at the end.
1349 pag = xfs_perag_get(mp, agno);
1350 if (!pag->pagi_inodeok) {
1351 xfs_ialloc_next_ag(mp);
1355 if (!pag->pagi_init) {
1356 error = xfs_ialloc_pagi_init(mp, tp, agno);
1362 * Do a first racy fast path check if this AG is usable.
1364 if (!pag->pagi_freecount && !okalloc)
1368 * Then read in the AGI buffer and recheck with the AGI buffer
1371 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1375 if (pag->pagi_freecount) {
1381 goto nextag_relse_buffer;
1384 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1386 xfs_trans_brelse(tp, agbp);
1388 if (error != ENOSPC)
1398 * We successfully allocated some inodes, return
1399 * the current context to the caller so that it
1400 * can commit the current transaction and call
1401 * us again where we left off.
1403 ASSERT(pag->pagi_freecount > 0);
1411 nextag_relse_buffer:
1412 xfs_trans_brelse(tp, agbp);
1415 if (++agno == mp->m_sb.sb_agcount)
1417 if (agno == start_agno) {
1419 return noroom ? ENOSPC : 0;
1425 return xfs_dialloc_ag(tp, agbp, parent, inop);
1428 return XFS_ERROR(error);
1433 struct xfs_mount *mp,
1434 struct xfs_trans *tp,
1435 struct xfs_buf *agbp,
1437 struct xfs_bmap_free *flist,
1439 xfs_ino_t *first_ino,
1440 struct xfs_inobt_rec_incore *orec)
1442 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1443 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1444 struct xfs_perag *pag;
1445 struct xfs_btree_cur *cur;
1446 struct xfs_inobt_rec_incore rec;
1452 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1453 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1456 * Initialize the cursor.
1458 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1460 error = xfs_check_agi_freecount(cur, agi);
1465 * Look for the entry describing this inode.
1467 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1468 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1472 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1473 error = xfs_inobt_get_rec(cur, &rec, &i);
1475 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1479 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1481 * Get the offset in the inode chunk.
1483 off = agino - rec.ir_startino;
1484 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1485 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1487 * Mark the inode free & increment the count.
1489 rec.ir_free |= XFS_INOBT_MASK(off);
1493 * When an inode cluster is free, it becomes eligible for removal
1495 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1496 (rec.ir_freecount == mp->m_ialloc_inos)) {
1499 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1502 * Remove the inode cluster from the AGI B+Tree, adjust the
1503 * AGI and Superblock inode counts, and mark the disk space
1504 * to be freed when the transaction is committed.
1506 ilen = mp->m_ialloc_inos;
1507 be32_add_cpu(&agi->agi_count, -ilen);
1508 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1509 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1510 pag = xfs_perag_get(mp, agno);
1511 pag->pagi_freecount -= ilen - 1;
1513 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1514 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1516 if ((error = xfs_btree_delete(cur, &i))) {
1517 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1522 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1523 XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1524 mp->m_ialloc_blks, flist, mp);
1528 error = xfs_inobt_update(cur, &rec);
1530 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1536 * Change the inode free counts and log the ag/sb changes.
1538 be32_add_cpu(&agi->agi_freecount, 1);
1539 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1540 pag = xfs_perag_get(mp, agno);
1541 pag->pagi_freecount++;
1543 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1546 error = xfs_check_agi_freecount(cur, agi);
1551 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1555 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1560 * Free an inode in the free inode btree.
1564 struct xfs_mount *mp,
1565 struct xfs_trans *tp,
1566 struct xfs_buf *agbp,
1568 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
1570 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1571 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1572 struct xfs_btree_cur *cur;
1573 struct xfs_inobt_rec_incore rec;
1574 int offset = agino - ibtrec->ir_startino;
1578 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1580 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
1585 * If the record does not exist in the finobt, we must have just
1586 * freed an inode in a previously fully allocated chunk. If not,
1587 * something is out of sync.
1589 XFS_WANT_CORRUPTED_GOTO(ibtrec->ir_freecount == 1, error);
1591 error = xfs_inobt_insert_rec(cur, ibtrec->ir_freecount,
1592 ibtrec->ir_free, &i);
1601 * Read and update the existing record. We could just copy the ibtrec
1602 * across here, but that would defeat the purpose of having redundant
1603 * metadata. By making the modifications independently, we can catch
1604 * corruptions that we wouldn't see if we just copied from one record
1607 error = xfs_inobt_get_rec(cur, &rec, &i);
1610 XFS_WANT_CORRUPTED_GOTO(i == 1, error);
1612 rec.ir_free |= XFS_INOBT_MASK(offset);
1615 XFS_WANT_CORRUPTED_GOTO((rec.ir_free == ibtrec->ir_free) &&
1616 (rec.ir_freecount == ibtrec->ir_freecount),
1620 * The content of inobt records should always match between the inobt
1621 * and finobt. The lifecycle of records in the finobt is different from
1622 * the inobt in that the finobt only tracks records with at least one
1623 * free inode. Hence, if all of the inodes are free and we aren't
1624 * keeping inode chunks permanently on disk, remove the record.
1625 * Otherwise, update the record with the new information.
1627 if (rec.ir_freecount == mp->m_ialloc_inos &&
1628 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
1629 error = xfs_btree_delete(cur, &i);
1634 error = xfs_inobt_update(cur, &rec);
1640 error = xfs_check_agi_freecount(cur, agi);
1644 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1648 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1653 * Free disk inode. Carefully avoids touching the incore inode, all
1654 * manipulations incore are the caller's responsibility.
1655 * The on-disk inode is not changed by this operation, only the
1656 * btree (free inode mask) is changed.
1660 struct xfs_trans *tp, /* transaction pointer */
1661 xfs_ino_t inode, /* inode to be freed */
1662 struct xfs_bmap_free *flist, /* extents to free */
1663 int *deleted,/* set if inode cluster was deleted */
1664 xfs_ino_t *first_ino)/* first inode in deleted cluster */
1667 xfs_agblock_t agbno; /* block number containing inode */
1668 struct xfs_buf *agbp; /* buffer for allocation group header */
1669 xfs_agino_t agino; /* allocation group inode number */
1670 xfs_agnumber_t agno; /* allocation group number */
1671 int error; /* error return value */
1672 struct xfs_mount *mp; /* mount structure for filesystem */
1673 struct xfs_inobt_rec_incore rec;/* btree record */
1678 * Break up inode number into its components.
1680 agno = XFS_INO_TO_AGNO(mp, inode);
1681 if (agno >= mp->m_sb.sb_agcount) {
1682 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1683 __func__, agno, mp->m_sb.sb_agcount);
1685 return XFS_ERROR(EINVAL);
1687 agino = XFS_INO_TO_AGINO(mp, inode);
1688 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
1689 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1690 __func__, (unsigned long long)inode,
1691 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1693 return XFS_ERROR(EINVAL);
1695 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1696 if (agbno >= mp->m_sb.sb_agblocks) {
1697 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1698 __func__, agbno, mp->m_sb.sb_agblocks);
1700 return XFS_ERROR(EINVAL);
1703 * Get the allocation group header.
1705 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1707 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1713 * Fix up the inode allocation btree.
1715 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, deleted, first_ino,
1721 * Fix up the free inode btree.
1723 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
1724 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
1737 struct xfs_mount *mp,
1738 struct xfs_trans *tp,
1739 xfs_agnumber_t agno,
1741 xfs_agblock_t agbno,
1742 xfs_agblock_t *chunk_agbno,
1743 xfs_agblock_t *offset_agbno,
1746 struct xfs_inobt_rec_incore rec;
1747 struct xfs_btree_cur *cur;
1748 struct xfs_buf *agbp;
1752 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1755 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1756 __func__, error, agno);
1761 * Lookup the inode record for the given agino. If the record cannot be
1762 * found, then it's an invalid inode number and we should abort. Once
1763 * we have a record, we need to ensure it contains the inode number
1764 * we are looking up.
1766 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1767 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1770 error = xfs_inobt_get_rec(cur, &rec, &i);
1771 if (!error && i == 0)
1775 xfs_trans_brelse(tp, agbp);
1776 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1780 /* check that the returned record contains the required inode */
1781 if (rec.ir_startino > agino ||
1782 rec.ir_startino + mp->m_ialloc_inos <= agino)
1785 /* for untrusted inodes check it is allocated first */
1786 if ((flags & XFS_IGET_UNTRUSTED) &&
1787 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1790 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1791 *offset_agbno = agbno - *chunk_agbno;
1796 * Return the location of the inode in imap, for mapping it into a buffer.
1800 xfs_mount_t *mp, /* file system mount structure */
1801 xfs_trans_t *tp, /* transaction pointer */
1802 xfs_ino_t ino, /* inode to locate */
1803 struct xfs_imap *imap, /* location map structure */
1804 uint flags) /* flags for inode btree lookup */
1806 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1807 xfs_agino_t agino; /* inode number within alloc group */
1808 xfs_agnumber_t agno; /* allocation group number */
1809 int blks_per_cluster; /* num blocks per inode cluster */
1810 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1811 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1812 int error; /* error code */
1813 int offset; /* index of inode in its buffer */
1814 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
1816 ASSERT(ino != NULLFSINO);
1819 * Split up the inode number into its parts.
1821 agno = XFS_INO_TO_AGNO(mp, ino);
1822 agino = XFS_INO_TO_AGINO(mp, ino);
1823 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1824 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1825 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1828 * Don't output diagnostic information for untrusted inodes
1829 * as they can be invalid without implying corruption.
1831 if (flags & XFS_IGET_UNTRUSTED)
1832 return XFS_ERROR(EINVAL);
1833 if (agno >= mp->m_sb.sb_agcount) {
1835 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1836 __func__, agno, mp->m_sb.sb_agcount);
1838 if (agbno >= mp->m_sb.sb_agblocks) {
1840 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1841 __func__, (unsigned long long)agbno,
1842 (unsigned long)mp->m_sb.sb_agblocks);
1844 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1846 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1848 XFS_AGINO_TO_INO(mp, agno, agino));
1852 return XFS_ERROR(EINVAL);
1855 blks_per_cluster = xfs_icluster_size_fsb(mp);
1858 * For bulkstat and handle lookups, we have an untrusted inode number
1859 * that we have to verify is valid. We cannot do this just by reading
1860 * the inode buffer as it may have been unlinked and removed leaving
1861 * inodes in stale state on disk. Hence we have to do a btree lookup
1862 * in all cases where an untrusted inode number is passed.
1864 if (flags & XFS_IGET_UNTRUSTED) {
1865 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1866 &chunk_agbno, &offset_agbno, flags);
1873 * If the inode cluster size is the same as the blocksize or
1874 * smaller we get to the buffer by simple arithmetics.
1876 if (blks_per_cluster == 1) {
1877 offset = XFS_INO_TO_OFFSET(mp, ino);
1878 ASSERT(offset < mp->m_sb.sb_inopblock);
1880 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1881 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1882 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1887 * If the inode chunks are aligned then use simple maths to
1888 * find the location. Otherwise we have to do a btree
1889 * lookup to find the location.
1891 if (mp->m_inoalign_mask) {
1892 offset_agbno = agbno & mp->m_inoalign_mask;
1893 chunk_agbno = agbno - offset_agbno;
1895 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1896 &chunk_agbno, &offset_agbno, flags);
1902 ASSERT(agbno >= chunk_agbno);
1903 cluster_agbno = chunk_agbno +
1904 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1905 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1906 XFS_INO_TO_OFFSET(mp, ino);
1908 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1909 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1910 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1913 * If the inode number maps to a block outside the bounds
1914 * of the file system then return NULL rather than calling
1915 * read_buf and panicing when we get an error from the
1918 if ((imap->im_blkno + imap->im_len) >
1919 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1921 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1922 __func__, (unsigned long long) imap->im_blkno,
1923 (unsigned long long) imap->im_len,
1924 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1925 return XFS_ERROR(EINVAL);
1931 * Compute and fill in value of m_in_maxlevels.
1934 xfs_ialloc_compute_maxlevels(
1935 xfs_mount_t *mp) /* file system mount structure */
1943 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1944 XFS_INODES_PER_CHUNK_LOG;
1945 minleafrecs = mp->m_alloc_mnr[0];
1946 minnoderecs = mp->m_alloc_mnr[1];
1947 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1948 for (level = 1; maxblocks > 1; level++)
1949 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1950 mp->m_in_maxlevels = level;
1954 * Log specified fields for the ag hdr (inode section). The growth of the agi
1955 * structure over time requires that we interpret the buffer as two logical
1956 * regions delineated by the end of the unlinked list. This is due to the size
1957 * of the hash table and its location in the middle of the agi.
1959 * For example, a request to log a field before agi_unlinked and a field after
1960 * agi_unlinked could cause us to log the entire hash table and use an excessive
1961 * amount of log space. To avoid this behavior, log the region up through
1962 * agi_unlinked in one call and the region after agi_unlinked through the end of
1963 * the structure in another.
1967 xfs_trans_t *tp, /* transaction pointer */
1968 xfs_buf_t *bp, /* allocation group header buffer */
1969 int fields) /* bitmask of fields to log */
1971 int first; /* first byte number */
1972 int last; /* last byte number */
1973 static const short offsets[] = { /* field starting offsets */
1974 /* keep in sync with bit definitions */
1975 offsetof(xfs_agi_t, agi_magicnum),
1976 offsetof(xfs_agi_t, agi_versionnum),
1977 offsetof(xfs_agi_t, agi_seqno),
1978 offsetof(xfs_agi_t, agi_length),
1979 offsetof(xfs_agi_t, agi_count),
1980 offsetof(xfs_agi_t, agi_root),
1981 offsetof(xfs_agi_t, agi_level),
1982 offsetof(xfs_agi_t, agi_freecount),
1983 offsetof(xfs_agi_t, agi_newino),
1984 offsetof(xfs_agi_t, agi_dirino),
1985 offsetof(xfs_agi_t, agi_unlinked),
1986 offsetof(xfs_agi_t, agi_free_root),
1987 offsetof(xfs_agi_t, agi_free_level),
1991 xfs_agi_t *agi; /* allocation group header */
1993 agi = XFS_BUF_TO_AGI(bp);
1994 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1997 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
2000 * Compute byte offsets for the first and last fields in the first
2001 * region and log the agi buffer. This only logs up through
2004 if (fields & XFS_AGI_ALL_BITS_R1) {
2005 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2007 xfs_trans_log_buf(tp, bp, first, last);
2011 * Mask off the bits in the first region and calculate the first and
2012 * last field offsets for any bits in the second region.
2014 fields &= ~XFS_AGI_ALL_BITS_R1;
2016 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2018 xfs_trans_log_buf(tp, bp, first, last);
2024 xfs_check_agi_unlinked(
2025 struct xfs_agi *agi)
2029 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2030 ASSERT(agi->agi_unlinked[i]);
2033 #define xfs_check_agi_unlinked(agi)
2040 struct xfs_mount *mp = bp->b_target->bt_mount;
2041 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2043 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2044 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
2047 * Validate the magic number of the agi block.
2049 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2051 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2055 * during growfs operations, the perag is not fully initialised,
2056 * so we can't use it for any useful checking. growfs ensures we can't
2057 * use it by using uncached buffers that don't have the perag attached
2058 * so we can detect and avoid this problem.
2060 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2063 xfs_check_agi_unlinked(agi);
2068 xfs_agi_read_verify(
2071 struct xfs_mount *mp = bp->b_target->bt_mount;
2073 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2074 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2075 xfs_buf_ioerror(bp, EFSBADCRC);
2076 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2077 XFS_ERRTAG_IALLOC_READ_AGI,
2078 XFS_RANDOM_IALLOC_READ_AGI))
2079 xfs_buf_ioerror(bp, EFSCORRUPTED);
2082 xfs_verifier_error(bp);
2086 xfs_agi_write_verify(
2089 struct xfs_mount *mp = bp->b_target->bt_mount;
2090 struct xfs_buf_log_item *bip = bp->b_fspriv;
2092 if (!xfs_agi_verify(bp)) {
2093 xfs_buf_ioerror(bp, EFSCORRUPTED);
2094 xfs_verifier_error(bp);
2098 if (!xfs_sb_version_hascrc(&mp->m_sb))
2102 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2103 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2106 const struct xfs_buf_ops xfs_agi_buf_ops = {
2107 .verify_read = xfs_agi_read_verify,
2108 .verify_write = xfs_agi_write_verify,
2112 * Read in the allocation group header (inode allocation section)
2116 struct xfs_mount *mp, /* file system mount structure */
2117 struct xfs_trans *tp, /* transaction pointer */
2118 xfs_agnumber_t agno, /* allocation group number */
2119 struct xfs_buf **bpp) /* allocation group hdr buf */
2123 trace_xfs_read_agi(mp, agno);
2125 ASSERT(agno != NULLAGNUMBER);
2126 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2127 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2128 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2132 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2137 xfs_ialloc_read_agi(
2138 struct xfs_mount *mp, /* file system mount structure */
2139 struct xfs_trans *tp, /* transaction pointer */
2140 xfs_agnumber_t agno, /* allocation group number */
2141 struct xfs_buf **bpp) /* allocation group hdr buf */
2143 struct xfs_agi *agi; /* allocation group header */
2144 struct xfs_perag *pag; /* per allocation group data */
2147 trace_xfs_ialloc_read_agi(mp, agno);
2149 error = xfs_read_agi(mp, tp, agno, bpp);
2153 agi = XFS_BUF_TO_AGI(*bpp);
2154 pag = xfs_perag_get(mp, agno);
2155 if (!pag->pagi_init) {
2156 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2157 pag->pagi_count = be32_to_cpu(agi->agi_count);
2162 * It's possible for these to be out of sync if
2163 * we are in the middle of a forced shutdown.
2165 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2166 XFS_FORCED_SHUTDOWN(mp));
2172 * Read in the agi to initialise the per-ag data in the mount structure
2175 xfs_ialloc_pagi_init(
2176 xfs_mount_t *mp, /* file system mount structure */
2177 xfs_trans_t *tp, /* transaction pointer */
2178 xfs_agnumber_t agno) /* allocation group number */
2180 xfs_buf_t *bp = NULL;
2183 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2187 xfs_trans_brelse(tp, bp);