1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_ialloc_btree.h"
20 #include "xfs_alloc.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_errortag.h"
23 #include "xfs_error.h"
25 #include "xfs_cksum.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_icreate_item.h"
29 #include "xfs_icache.h"
30 #include "xfs_trace.h"
36 * Allocation group level functions.
39 xfs_ialloc_cluster_alignment(
42 if (xfs_sb_version_hasalign(&mp->m_sb) &&
43 mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
44 return mp->m_sb.sb_inoalignmt;
49 * Lookup a record by ino in the btree given by cur.
53 struct xfs_btree_cur *cur, /* btree cursor */
54 xfs_agino_t ino, /* starting inode of chunk */
55 xfs_lookup_t dir, /* <=, >=, == */
56 int *stat) /* success/failure */
58 cur->bc_rec.i.ir_startino = ino;
59 cur->bc_rec.i.ir_holemask = 0;
60 cur->bc_rec.i.ir_count = 0;
61 cur->bc_rec.i.ir_freecount = 0;
62 cur->bc_rec.i.ir_free = 0;
63 return xfs_btree_lookup(cur, dir, stat);
67 * Update the record referred to by cur to the value given.
68 * This either works (return 0) or gets an EFSCORRUPTED error.
70 STATIC int /* error */
72 struct xfs_btree_cur *cur, /* btree cursor */
73 xfs_inobt_rec_incore_t *irec) /* btree record */
75 union xfs_btree_rec rec;
77 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
78 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
79 rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
80 rec.inobt.ir_u.sp.ir_count = irec->ir_count;
81 rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
83 /* ir_holemask/ir_count not supported on-disk */
84 rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
86 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
87 return xfs_btree_update(cur, &rec);
90 /* Convert on-disk btree record to incore inobt record. */
92 xfs_inobt_btrec_to_irec(
94 union xfs_btree_rec *rec,
95 struct xfs_inobt_rec_incore *irec)
97 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
98 if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
99 irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
100 irec->ir_count = rec->inobt.ir_u.sp.ir_count;
101 irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
104 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
105 * values for full inode chunks.
107 irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
108 irec->ir_count = XFS_INODES_PER_CHUNK;
110 be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
112 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
116 * Get the data from the pointed-to record.
120 struct xfs_btree_cur *cur,
121 struct xfs_inobt_rec_incore *irec,
124 struct xfs_mount *mp = cur->bc_mp;
125 xfs_agnumber_t agno = cur->bc_private.a.agno;
126 union xfs_btree_rec *rec;
130 error = xfs_btree_get_rec(cur, &rec, stat);
131 if (error || *stat == 0)
134 xfs_inobt_btrec_to_irec(mp, rec, irec);
136 if (!xfs_verify_agino(mp, agno, irec->ir_startino))
138 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
139 irec->ir_count > XFS_INODES_PER_CHUNK)
141 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
144 /* if there are no holes, return the first available offset */
145 if (!xfs_inobt_issparse(irec->ir_holemask))
146 realfree = irec->ir_free;
148 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
149 if (hweight64(realfree) != irec->ir_freecount)
156 "%s Inode BTree record corruption in AG %d detected!",
157 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno);
159 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
160 irec->ir_startino, irec->ir_count, irec->ir_freecount,
161 irec->ir_free, irec->ir_holemask);
162 return -EFSCORRUPTED;
166 * Insert a single inobt record. Cursor must already point to desired location.
169 xfs_inobt_insert_rec(
170 struct xfs_btree_cur *cur,
177 cur->bc_rec.i.ir_holemask = holemask;
178 cur->bc_rec.i.ir_count = count;
179 cur->bc_rec.i.ir_freecount = freecount;
180 cur->bc_rec.i.ir_free = free;
181 return xfs_btree_insert(cur, stat);
185 * Insert records describing a newly allocated inode chunk into the inobt.
189 struct xfs_mount *mp,
190 struct xfs_trans *tp,
191 struct xfs_buf *agbp,
196 struct xfs_btree_cur *cur;
197 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
198 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
203 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
205 for (thisino = newino;
206 thisino < newino + newlen;
207 thisino += XFS_INODES_PER_CHUNK) {
208 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
210 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
215 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
216 XFS_INODES_PER_CHUNK,
217 XFS_INODES_PER_CHUNK,
218 XFS_INOBT_ALL_FREE, &i);
220 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
226 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
232 * Verify that the number of free inodes in the AGI is correct.
236 xfs_check_agi_freecount(
237 struct xfs_btree_cur *cur,
240 if (cur->bc_nlevels == 1) {
241 xfs_inobt_rec_incore_t rec;
246 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
251 error = xfs_inobt_get_rec(cur, &rec, &i);
256 freecount += rec.ir_freecount;
257 error = xfs_btree_increment(cur, 0, &i);
263 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
264 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
269 #define xfs_check_agi_freecount(cur, agi) 0
273 * Initialise a new set of inodes. When called without a transaction context
274 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
275 * than logging them (which in a transaction context puts them into the AIL
276 * for writeback rather than the xfsbufd queue).
279 xfs_ialloc_inode_init(
280 struct xfs_mount *mp,
281 struct xfs_trans *tp,
282 struct list_head *buffer_list,
286 xfs_agblock_t length,
289 struct xfs_buf *fbuf;
290 struct xfs_dinode *free;
298 * Loop over the new block(s), filling in the inodes. For small block
299 * sizes, manipulate the inodes in buffers which are multiples of the
302 nbufs = length / mp->m_blocks_per_cluster;
305 * Figure out what version number to use in the inodes we create. If
306 * the superblock version has caught up to the one that supports the new
307 * inode format, then use the new inode version. Otherwise use the old
308 * version so that old kernels will continue to be able to use the file
311 * For v3 inodes, we also need to write the inode number into the inode,
312 * so calculate the first inode number of the chunk here as
313 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
314 * across multiple filesystem blocks (such as a cluster) and so cannot
315 * be used in the cluster buffer loop below.
317 * Further, because we are writing the inode directly into the buffer
318 * and calculating a CRC on the entire inode, we have ot log the entire
319 * inode so that the entire range the CRC covers is present in the log.
320 * That means for v3 inode we log the entire buffer rather than just the
323 if (xfs_sb_version_hascrc(&mp->m_sb)) {
325 ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
328 * log the initialisation that is about to take place as an
329 * logical operation. This means the transaction does not
330 * need to log the physical changes to the inode buffers as log
331 * recovery will know what initialisation is actually needed.
332 * Hence we only need to log the buffers as "ordered" buffers so
333 * they track in the AIL as if they were physically logged.
336 xfs_icreate_log(tp, agno, agbno, icount,
337 mp->m_sb.sb_inodesize, length, gen);
341 for (j = 0; j < nbufs; j++) {
345 d = XFS_AGB_TO_DADDR(mp, agno, agbno +
346 (j * mp->m_blocks_per_cluster));
347 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
348 mp->m_bsize * mp->m_blocks_per_cluster,
353 /* Initialize the inode buffers and log them appropriately. */
354 fbuf->b_ops = &xfs_inode_buf_ops;
355 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
356 for (i = 0; i < mp->m_inodes_per_cluster; i++) {
357 int ioffset = i << mp->m_sb.sb_inodelog;
358 uint isize = xfs_dinode_size(version);
360 free = xfs_make_iptr(mp, fbuf, i);
361 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
362 free->di_version = version;
363 free->di_gen = cpu_to_be32(gen);
364 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
367 free->di_ino = cpu_to_be64(ino);
369 uuid_copy(&free->di_uuid,
370 &mp->m_sb.sb_meta_uuid);
371 xfs_dinode_calc_crc(mp, free);
373 /* just log the inode core */
374 xfs_trans_log_buf(tp, fbuf, ioffset,
375 ioffset + isize - 1);
381 * Mark the buffer as an inode allocation buffer so it
382 * sticks in AIL at the point of this allocation
383 * transaction. This ensures the they are on disk before
384 * the tail of the log can be moved past this
385 * transaction (i.e. by preventing relogging from moving
386 * it forward in the log).
388 xfs_trans_inode_alloc_buf(tp, fbuf);
391 * Mark the buffer as ordered so that they are
392 * not physically logged in the transaction but
393 * still tracked in the AIL as part of the
394 * transaction and pin the log appropriately.
396 xfs_trans_ordered_buf(tp, fbuf);
399 fbuf->b_flags |= XBF_DONE;
400 xfs_buf_delwri_queue(fbuf, buffer_list);
408 * Align startino and allocmask for a recently allocated sparse chunk such that
409 * they are fit for insertion (or merge) into the on-disk inode btrees.
413 * When enabled, sparse inode support increases the inode alignment from cluster
414 * size to inode chunk size. This means that the minimum range between two
415 * non-adjacent inode records in the inobt is large enough for a full inode
416 * record. This allows for cluster sized, cluster aligned block allocation
417 * without need to worry about whether the resulting inode record overlaps with
418 * another record in the tree. Without this basic rule, we would have to deal
419 * with the consequences of overlap by potentially undoing recent allocations in
420 * the inode allocation codepath.
422 * Because of this alignment rule (which is enforced on mount), there are two
423 * inobt possibilities for newly allocated sparse chunks. One is that the
424 * aligned inode record for the chunk covers a range of inodes not already
425 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
426 * other is that a record already exists at the aligned startino that considers
427 * the newly allocated range as sparse. In the latter case, record content is
428 * merged in hope that sparse inode chunks fill to full chunks over time.
431 xfs_align_sparse_ino(
432 struct xfs_mount *mp,
433 xfs_agino_t *startino,
440 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
441 mod = agbno % mp->m_sb.sb_inoalignmt;
445 /* calculate the inode offset and align startino */
446 offset = XFS_AGB_TO_AGINO(mp, mod);
450 * Since startino has been aligned down, left shift allocmask such that
451 * it continues to represent the same physical inodes relative to the
454 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
458 * Determine whether the source inode record can merge into the target. Both
459 * records must be sparse, the inode ranges must match and there must be no
460 * allocation overlap between the records.
463 __xfs_inobt_can_merge(
464 struct xfs_inobt_rec_incore *trec, /* tgt record */
465 struct xfs_inobt_rec_incore *srec) /* src record */
470 /* records must cover the same inode range */
471 if (trec->ir_startino != srec->ir_startino)
474 /* both records must be sparse */
475 if (!xfs_inobt_issparse(trec->ir_holemask) ||
476 !xfs_inobt_issparse(srec->ir_holemask))
479 /* both records must track some inodes */
480 if (!trec->ir_count || !srec->ir_count)
483 /* can't exceed capacity of a full record */
484 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
487 /* verify there is no allocation overlap */
488 talloc = xfs_inobt_irec_to_allocmask(trec);
489 salloc = xfs_inobt_irec_to_allocmask(srec);
497 * Merge the source inode record into the target. The caller must call
498 * __xfs_inobt_can_merge() to ensure the merge is valid.
501 __xfs_inobt_rec_merge(
502 struct xfs_inobt_rec_incore *trec, /* target */
503 struct xfs_inobt_rec_incore *srec) /* src */
505 ASSERT(trec->ir_startino == srec->ir_startino);
507 /* combine the counts */
508 trec->ir_count += srec->ir_count;
509 trec->ir_freecount += srec->ir_freecount;
512 * Merge the holemask and free mask. For both fields, 0 bits refer to
513 * allocated inodes. We combine the allocated ranges with bitwise AND.
515 trec->ir_holemask &= srec->ir_holemask;
516 trec->ir_free &= srec->ir_free;
520 * Insert a new sparse inode chunk into the associated inode btree. The inode
521 * record for the sparse chunk is pre-aligned to a startino that should match
522 * any pre-existing sparse inode record in the tree. This allows sparse chunks
525 * This function supports two modes of handling preexisting records depending on
526 * the merge flag. If merge is true, the provided record is merged with the
527 * existing record and updated in place. The merged record is returned in nrec.
528 * If merge is false, an existing record is replaced with the provided record.
529 * If no preexisting record exists, the provided record is always inserted.
531 * It is considered corruption if a merge is requested and not possible. Given
532 * the sparse inode alignment constraints, this should never happen.
535 xfs_inobt_insert_sprec(
536 struct xfs_mount *mp,
537 struct xfs_trans *tp,
538 struct xfs_buf *agbp,
540 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
541 bool merge) /* merge or replace */
543 struct xfs_btree_cur *cur;
544 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
545 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
548 struct xfs_inobt_rec_incore rec;
550 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
552 /* the new record is pre-aligned so we know where to look */
553 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
556 /* if nothing there, insert a new record and return */
558 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
559 nrec->ir_count, nrec->ir_freecount,
563 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
569 * A record exists at this startino. Merge or replace the record
570 * depending on what we've been asked to do.
573 error = xfs_inobt_get_rec(cur, &rec, &i);
576 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
577 XFS_WANT_CORRUPTED_GOTO(mp,
578 rec.ir_startino == nrec->ir_startino,
582 * This should never fail. If we have coexisting records that
583 * cannot merge, something is seriously wrong.
585 XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
588 trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
589 rec.ir_holemask, nrec->ir_startino,
592 /* merge to nrec to output the updated record */
593 __xfs_inobt_rec_merge(nrec, &rec);
595 trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
598 error = xfs_inobt_rec_check_count(mp, nrec);
603 error = xfs_inobt_update(cur, nrec);
608 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
611 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
616 * Allocate new inodes in the allocation group specified by agbp.
617 * Return 0 for success, else error code.
619 STATIC int /* error code or 0 */
621 xfs_trans_t *tp, /* transaction pointer */
622 xfs_buf_t *agbp, /* alloc group buffer */
625 xfs_agi_t *agi; /* allocation group header */
626 xfs_alloc_arg_t args; /* allocation argument structure */
629 xfs_agino_t newino; /* new first inode's number */
630 xfs_agino_t newlen; /* new number of inodes */
631 int isaligned = 0; /* inode allocation at stripe unit */
633 uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */
634 struct xfs_inobt_rec_incore rec;
635 struct xfs_perag *pag;
638 memset(&args, 0, sizeof(args));
640 args.mp = tp->t_mountp;
641 args.fsbno = NULLFSBLOCK;
642 args.oinfo = XFS_RMAP_OINFO_INODES;
645 /* randomly do sparse inode allocations */
646 if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
647 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
648 do_sparse = prandom_u32() & 1;
652 * Locking will ensure that we don't have two callers in here
655 newlen = args.mp->m_ialloc_inos;
656 if (args.mp->m_maxicount &&
657 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
658 args.mp->m_maxicount)
660 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
662 * First try to allocate inodes contiguous with the last-allocated
663 * chunk of inodes. If the filesystem is striped, this will fill
664 * an entire stripe unit with inodes.
666 agi = XFS_BUF_TO_AGI(agbp);
667 newino = be32_to_cpu(agi->agi_newino);
668 agno = be32_to_cpu(agi->agi_seqno);
669 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
670 args.mp->m_ialloc_blks;
673 if (likely(newino != NULLAGINO &&
674 (args.agbno < be32_to_cpu(agi->agi_length)))) {
675 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
676 args.type = XFS_ALLOCTYPE_THIS_BNO;
680 * We need to take into account alignment here to ensure that
681 * we don't modify the free list if we fail to have an exact
682 * block. If we don't have an exact match, and every oher
683 * attempt allocation attempt fails, we'll end up cancelling
684 * a dirty transaction and shutting down.
686 * For an exact allocation, alignment must be 1,
687 * however we need to take cluster alignment into account when
688 * fixing up the freelist. Use the minalignslop field to
689 * indicate that extra blocks might be required for alignment,
690 * but not to use them in the actual exact allocation.
693 args.minalignslop = args.mp->m_cluster_align - 1;
695 /* Allow space for the inode btree to split. */
696 args.minleft = args.mp->m_in_maxlevels - 1;
697 if ((error = xfs_alloc_vextent(&args)))
701 * This request might have dirtied the transaction if the AG can
702 * satisfy the request, but the exact block was not available.
703 * If the allocation did fail, subsequent requests will relax
704 * the exact agbno requirement and increase the alignment
705 * instead. It is critical that the total size of the request
706 * (len + alignment + slop) does not increase from this point
707 * on, so reset minalignslop to ensure it is not included in
708 * subsequent requests.
710 args.minalignslop = 0;
713 if (unlikely(args.fsbno == NULLFSBLOCK)) {
715 * Set the alignment for the allocation.
716 * If stripe alignment is turned on then align at stripe unit
718 * If the cluster size is smaller than a filesystem block
719 * then we're doing I/O for inodes in filesystem block size
720 * pieces, so don't need alignment anyway.
723 if (args.mp->m_sinoalign) {
724 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
725 args.alignment = args.mp->m_dalign;
728 args.alignment = args.mp->m_cluster_align;
730 * Need to figure out where to allocate the inode blocks.
731 * Ideally they should be spaced out through the a.g.
732 * For now, just allocate blocks up front.
734 args.agbno = be32_to_cpu(agi->agi_root);
735 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
737 * Allocate a fixed-size extent of inodes.
739 args.type = XFS_ALLOCTYPE_NEAR_BNO;
742 * Allow space for the inode btree to split.
744 args.minleft = args.mp->m_in_maxlevels - 1;
745 if ((error = xfs_alloc_vextent(&args)))
750 * If stripe alignment is turned on, then try again with cluster
753 if (isaligned && args.fsbno == NULLFSBLOCK) {
754 args.type = XFS_ALLOCTYPE_NEAR_BNO;
755 args.agbno = be32_to_cpu(agi->agi_root);
756 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
757 args.alignment = args.mp->m_cluster_align;
758 if ((error = xfs_alloc_vextent(&args)))
763 * Finally, try a sparse allocation if the filesystem supports it and
764 * the sparse allocation length is smaller than a full chunk.
766 if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
767 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
768 args.fsbno == NULLFSBLOCK) {
770 args.type = XFS_ALLOCTYPE_NEAR_BNO;
771 args.agbno = be32_to_cpu(agi->agi_root);
772 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
773 args.alignment = args.mp->m_sb.sb_spino_align;
776 args.minlen = args.mp->m_ialloc_min_blks;
777 args.maxlen = args.minlen;
780 * The inode record will be aligned to full chunk size. We must
781 * prevent sparse allocation from AG boundaries that result in
782 * invalid inode records, such as records that start at agbno 0
783 * or extend beyond the AG.
785 * Set min agbno to the first aligned, non-zero agbno and max to
786 * the last aligned agbno that is at least one full chunk from
789 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
790 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
791 args.mp->m_sb.sb_inoalignmt) -
792 args.mp->m_ialloc_blks;
794 error = xfs_alloc_vextent(&args);
798 newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
799 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
800 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
803 if (args.fsbno == NULLFSBLOCK) {
807 ASSERT(args.len == args.minlen);
810 * Stamp and write the inode buffers.
812 * Seed the new inode cluster with a random generation number. This
813 * prevents short-term reuse of generation numbers if a chunk is
814 * freed and then immediately reallocated. We use random numbers
815 * rather than a linear progression to prevent the next generation
816 * number from being easily guessable.
818 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
819 args.agbno, args.len, prandom_u32());
824 * Convert the results.
826 newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
828 if (xfs_inobt_issparse(~allocmask)) {
830 * We've allocated a sparse chunk. Align the startino and mask.
832 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
834 rec.ir_startino = newino;
835 rec.ir_holemask = ~allocmask;
836 rec.ir_count = newlen;
837 rec.ir_freecount = newlen;
838 rec.ir_free = XFS_INOBT_ALL_FREE;
841 * Insert the sparse record into the inobt and allow for a merge
842 * if necessary. If a merge does occur, rec is updated to the
845 error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
847 if (error == -EFSCORRUPTED) {
849 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
850 XFS_AGINO_TO_INO(args.mp, agno,
852 rec.ir_holemask, rec.ir_count);
853 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
859 * We can't merge the part we've just allocated as for the inobt
860 * due to finobt semantics. The original record may or may not
861 * exist independent of whether physical inodes exist in this
864 * We must update the finobt record based on the inobt record.
865 * rec contains the fully merged and up to date inobt record
866 * from the previous call. Set merge false to replace any
867 * existing record with this one.
869 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
870 error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
871 XFS_BTNUM_FINO, &rec,
877 /* full chunk - insert new records to both btrees */
878 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
883 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
884 error = xfs_inobt_insert(args.mp, tp, agbp, newino,
885 newlen, XFS_BTNUM_FINO);
892 * Update AGI counts and newino.
894 be32_add_cpu(&agi->agi_count, newlen);
895 be32_add_cpu(&agi->agi_freecount, newlen);
896 pag = xfs_perag_get(args.mp, agno);
897 pag->pagi_freecount += newlen;
898 pag->pagi_count += newlen;
900 agi->agi_newino = cpu_to_be32(newino);
903 * Log allocation group header fields
905 xfs_ialloc_log_agi(tp, agbp,
906 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
908 * Modify/log superblock values for inode count and inode free count.
910 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
911 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
916 STATIC xfs_agnumber_t
922 spin_lock(&mp->m_agirotor_lock);
923 agno = mp->m_agirotor;
924 if (++mp->m_agirotor >= mp->m_maxagi)
926 spin_unlock(&mp->m_agirotor_lock);
932 * Select an allocation group to look for a free inode in, based on the parent
933 * inode and the mode. Return the allocation group buffer.
935 STATIC xfs_agnumber_t
936 xfs_ialloc_ag_select(
937 xfs_trans_t *tp, /* transaction pointer */
938 xfs_ino_t parent, /* parent directory inode number */
939 umode_t mode) /* bits set to indicate file type */
941 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
942 xfs_agnumber_t agno; /* current ag number */
943 int flags; /* alloc buffer locking flags */
944 xfs_extlen_t ineed; /* blocks needed for inode allocation */
945 xfs_extlen_t longest = 0; /* longest extent available */
946 xfs_mount_t *mp; /* mount point structure */
947 int needspace; /* file mode implies space allocated */
948 xfs_perag_t *pag; /* per allocation group data */
949 xfs_agnumber_t pagno; /* parent (starting) ag number */
953 * Files of these types need at least one block if length > 0
954 * (and they won't fit in the inode, but that's hard to figure out).
956 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
958 agcount = mp->m_maxagi;
960 pagno = xfs_ialloc_next_ag(mp);
962 pagno = XFS_INO_TO_AGNO(mp, parent);
963 if (pagno >= agcount)
967 ASSERT(pagno < agcount);
970 * Loop through allocation groups, looking for one with a little
971 * free space in it. Note we don't look for free inodes, exactly.
972 * Instead, we include whether there is a need to allocate inodes
973 * to mean that blocks must be allocated for them,
974 * if none are currently free.
977 flags = XFS_ALLOC_FLAG_TRYLOCK;
979 pag = xfs_perag_get(mp, agno);
980 if (!pag->pagi_inodeok) {
981 xfs_ialloc_next_ag(mp);
985 if (!pag->pagi_init) {
986 error = xfs_ialloc_pagi_init(mp, tp, agno);
991 if (pag->pagi_freecount) {
996 if (!pag->pagf_init) {
997 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
1003 * Check that there is enough free space for the file plus a
1004 * chunk of inodes if we need to allocate some. If this is the
1005 * first pass across the AGs, take into account the potential
1006 * space needed for alignment of inode chunks when checking the
1007 * longest contiguous free space in the AG - this prevents us
1008 * from getting ENOSPC because we have free space larger than
1009 * m_ialloc_blks but alignment constraints prevent us from using
1012 * If we can't find an AG with space for full alignment slack to
1013 * be taken into account, we must be near ENOSPC in all AGs.
1014 * Hence we don't include alignment for the second pass and so
1015 * if we fail allocation due to alignment issues then it is most
1016 * likely a real ENOSPC condition.
1018 ineed = mp->m_ialloc_min_blks;
1019 if (flags && ineed > 1)
1020 ineed += mp->m_cluster_align;
1021 longest = pag->pagf_longest;
1023 longest = pag->pagf_flcount > 0;
1025 if (pag->pagf_freeblks >= needspace + ineed &&
1033 * No point in iterating over the rest, if we're shutting
1036 if (XFS_FORCED_SHUTDOWN(mp))
1037 return NULLAGNUMBER;
1039 if (agno >= agcount)
1041 if (agno == pagno) {
1043 return NULLAGNUMBER;
1050 * Try to retrieve the next record to the left/right from the current one.
1053 xfs_ialloc_next_rec(
1054 struct xfs_btree_cur *cur,
1055 xfs_inobt_rec_incore_t *rec,
1063 error = xfs_btree_decrement(cur, 0, &i);
1065 error = xfs_btree_increment(cur, 0, &i);
1071 error = xfs_inobt_get_rec(cur, rec, &i);
1074 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1082 struct xfs_btree_cur *cur,
1084 xfs_inobt_rec_incore_t *rec,
1090 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1095 error = xfs_inobt_get_rec(cur, rec, &i);
1098 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1105 * Return the offset of the first free inode in the record. If the inode chunk
1106 * is sparsely allocated, we convert the record holemask to inode granularity
1107 * and mask off the unallocated regions from the inode free mask.
1110 xfs_inobt_first_free_inode(
1111 struct xfs_inobt_rec_incore *rec)
1113 xfs_inofree_t realfree;
1115 /* if there are no holes, return the first available offset */
1116 if (!xfs_inobt_issparse(rec->ir_holemask))
1117 return xfs_lowbit64(rec->ir_free);
1119 realfree = xfs_inobt_irec_to_allocmask(rec);
1120 realfree &= rec->ir_free;
1122 return xfs_lowbit64(realfree);
1126 * Allocate an inode using the inobt-only algorithm.
1129 xfs_dialloc_ag_inobt(
1130 struct xfs_trans *tp,
1131 struct xfs_buf *agbp,
1135 struct xfs_mount *mp = tp->t_mountp;
1136 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1137 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1138 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1139 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1140 struct xfs_perag *pag;
1141 struct xfs_btree_cur *cur, *tcur;
1142 struct xfs_inobt_rec_incore rec, trec;
1147 int searchdistance = 10;
1149 pag = xfs_perag_get(mp, agno);
1151 ASSERT(pag->pagi_init);
1152 ASSERT(pag->pagi_inodeok);
1153 ASSERT(pag->pagi_freecount > 0);
1156 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1158 * If pagino is 0 (this is the root inode allocation) use newino.
1159 * This must work because we've just allocated some.
1162 pagino = be32_to_cpu(agi->agi_newino);
1164 error = xfs_check_agi_freecount(cur, agi);
1169 * If in the same AG as the parent, try to get near the parent.
1171 if (pagno == agno) {
1172 int doneleft; /* done, to the left */
1173 int doneright; /* done, to the right */
1175 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1178 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1180 error = xfs_inobt_get_rec(cur, &rec, &j);
1183 XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
1185 if (rec.ir_freecount > 0) {
1187 * Found a free inode in the same chunk
1188 * as the parent, done.
1195 * In the same AG as parent, but parent's chunk is full.
1198 /* duplicate the cursor, search left & right simultaneously */
1199 error = xfs_btree_dup_cursor(cur, &tcur);
1204 * Skip to last blocks looked up if same parent inode.
1206 if (pagino != NULLAGINO &&
1207 pag->pagl_pagino == pagino &&
1208 pag->pagl_leftrec != NULLAGINO &&
1209 pag->pagl_rightrec != NULLAGINO) {
1210 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1215 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1220 /* search left with tcur, back up 1 record */
1221 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1225 /* search right with cur, go forward 1 record. */
1226 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1232 * Loop until we find an inode chunk with a free inode.
1234 while (--searchdistance > 0 && (!doneleft || !doneright)) {
1235 int useleft; /* using left inode chunk this time */
1237 /* figure out the closer block if both are valid. */
1238 if (!doneleft && !doneright) {
1240 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1241 rec.ir_startino - pagino;
1243 useleft = !doneleft;
1246 /* free inodes to the left? */
1247 if (useleft && trec.ir_freecount) {
1248 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1251 pag->pagl_leftrec = trec.ir_startino;
1252 pag->pagl_rightrec = rec.ir_startino;
1253 pag->pagl_pagino = pagino;
1258 /* free inodes to the right? */
1259 if (!useleft && rec.ir_freecount) {
1260 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1262 pag->pagl_leftrec = trec.ir_startino;
1263 pag->pagl_rightrec = rec.ir_startino;
1264 pag->pagl_pagino = pagino;
1268 /* get next record to check */
1270 error = xfs_ialloc_next_rec(tcur, &trec,
1273 error = xfs_ialloc_next_rec(cur, &rec,
1280 if (searchdistance <= 0) {
1282 * Not in range - save last search
1283 * location and allocate a new inode
1285 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1286 pag->pagl_leftrec = trec.ir_startino;
1287 pag->pagl_rightrec = rec.ir_startino;
1288 pag->pagl_pagino = pagino;
1292 * We've reached the end of the btree. because
1293 * we are only searching a small chunk of the
1294 * btree each search, there is obviously free
1295 * inodes closer to the parent inode than we
1296 * are now. restart the search again.
1298 pag->pagl_pagino = NULLAGINO;
1299 pag->pagl_leftrec = NULLAGINO;
1300 pag->pagl_rightrec = NULLAGINO;
1301 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1302 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1308 * In a different AG from the parent.
1309 * See if the most recently allocated block has any free.
1311 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1312 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1318 error = xfs_inobt_get_rec(cur, &rec, &j);
1322 if (j == 1 && rec.ir_freecount > 0) {
1324 * The last chunk allocated in the group
1325 * still has a free inode.
1333 * None left in the last group, search the whole AG
1335 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1338 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1341 error = xfs_inobt_get_rec(cur, &rec, &i);
1344 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1345 if (rec.ir_freecount > 0)
1347 error = xfs_btree_increment(cur, 0, &i);
1350 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1354 offset = xfs_inobt_first_free_inode(&rec);
1355 ASSERT(offset >= 0);
1356 ASSERT(offset < XFS_INODES_PER_CHUNK);
1357 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1358 XFS_INODES_PER_CHUNK) == 0);
1359 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1360 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1362 error = xfs_inobt_update(cur, &rec);
1365 be32_add_cpu(&agi->agi_freecount, -1);
1366 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1367 pag->pagi_freecount--;
1369 error = xfs_check_agi_freecount(cur, agi);
1373 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1374 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1379 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1381 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1387 * Use the free inode btree to allocate an inode based on distance from the
1388 * parent. Note that the provided cursor may be deleted and replaced.
1391 xfs_dialloc_ag_finobt_near(
1393 struct xfs_btree_cur **ocur,
1394 struct xfs_inobt_rec_incore *rec)
1396 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1397 struct xfs_btree_cur *rcur; /* right search cursor */
1398 struct xfs_inobt_rec_incore rrec;
1402 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1407 error = xfs_inobt_get_rec(lcur, rec, &i);
1410 XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1413 * See if we've landed in the parent inode record. The finobt
1414 * only tracks chunks with at least one free inode, so record
1415 * existence is enough.
1417 if (pagino >= rec->ir_startino &&
1418 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1422 error = xfs_btree_dup_cursor(lcur, &rcur);
1426 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1430 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1433 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1436 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1437 if (i == 1 && j == 1) {
1439 * Both the left and right records are valid. Choose the closer
1440 * inode chunk to the target.
1442 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1443 (rrec.ir_startino - pagino)) {
1445 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1448 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1450 } else if (j == 1) {
1451 /* only the right record is valid */
1453 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1455 } else if (i == 1) {
1456 /* only the left record is valid */
1457 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1463 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1468 * Use the free inode btree to find a free inode based on a newino hint. If
1469 * the hint is NULL, find the first free inode in the AG.
1472 xfs_dialloc_ag_finobt_newino(
1473 struct xfs_agi *agi,
1474 struct xfs_btree_cur *cur,
1475 struct xfs_inobt_rec_incore *rec)
1480 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1481 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1486 error = xfs_inobt_get_rec(cur, rec, &i);
1489 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1495 * Find the first inode available in the AG.
1497 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1500 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1502 error = xfs_inobt_get_rec(cur, rec, &i);
1505 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1511 * Update the inobt based on a modification made to the finobt. Also ensure that
1512 * the records from both trees are equivalent post-modification.
1515 xfs_dialloc_ag_update_inobt(
1516 struct xfs_btree_cur *cur, /* inobt cursor */
1517 struct xfs_inobt_rec_incore *frec, /* finobt record */
1518 int offset) /* inode offset */
1520 struct xfs_inobt_rec_incore rec;
1524 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1527 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1529 error = xfs_inobt_get_rec(cur, &rec, &i);
1532 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1533 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1534 XFS_INODES_PER_CHUNK) == 0);
1536 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1539 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1540 (rec.ir_freecount == frec->ir_freecount));
1542 return xfs_inobt_update(cur, &rec);
1546 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1547 * back to the inobt search algorithm.
1549 * The caller selected an AG for us, and made sure that free inodes are
1554 struct xfs_trans *tp,
1555 struct xfs_buf *agbp,
1559 struct xfs_mount *mp = tp->t_mountp;
1560 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1561 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1562 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1563 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1564 struct xfs_perag *pag;
1565 struct xfs_btree_cur *cur; /* finobt cursor */
1566 struct xfs_btree_cur *icur; /* inobt cursor */
1567 struct xfs_inobt_rec_incore rec;
1573 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1574 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1576 pag = xfs_perag_get(mp, agno);
1579 * If pagino is 0 (this is the root inode allocation) use newino.
1580 * This must work because we've just allocated some.
1583 pagino = be32_to_cpu(agi->agi_newino);
1585 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1587 error = xfs_check_agi_freecount(cur, agi);
1592 * The search algorithm depends on whether we're in the same AG as the
1593 * parent. If so, find the closest available inode to the parent. If
1594 * not, consider the agi hint or find the first free inode in the AG.
1597 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1599 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1603 offset = xfs_inobt_first_free_inode(&rec);
1604 ASSERT(offset >= 0);
1605 ASSERT(offset < XFS_INODES_PER_CHUNK);
1606 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1607 XFS_INODES_PER_CHUNK) == 0);
1608 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1611 * Modify or remove the finobt record.
1613 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1615 if (rec.ir_freecount)
1616 error = xfs_inobt_update(cur, &rec);
1618 error = xfs_btree_delete(cur, &i);
1623 * The finobt has now been updated appropriately. We haven't updated the
1624 * agi and superblock yet, so we can create an inobt cursor and validate
1625 * the original freecount. If all is well, make the equivalent update to
1626 * the inobt using the finobt record and offset information.
1628 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1630 error = xfs_check_agi_freecount(icur, agi);
1634 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1639 * Both trees have now been updated. We must update the perag and
1640 * superblock before we can check the freecount for each btree.
1642 be32_add_cpu(&agi->agi_freecount, -1);
1643 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1644 pag->pagi_freecount--;
1646 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1648 error = xfs_check_agi_freecount(icur, agi);
1651 error = xfs_check_agi_freecount(cur, agi);
1655 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1656 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1662 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1664 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1670 * Allocate an inode on disk.
1672 * Mode is used to tell whether the new inode will need space, and whether it
1675 * This function is designed to be called twice if it has to do an allocation
1676 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1677 * If an inode is available without having to performn an allocation, an inode
1678 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1679 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1680 * The caller should then commit the current transaction, allocate a
1681 * new transaction, and call xfs_dialloc() again, passing in the previous value
1682 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1683 * buffer is locked across the two calls, the second call is guaranteed to have
1684 * a free inode available.
1686 * Once we successfully pick an inode its number is returned and the on-disk
1687 * data structures are updated. The inode itself is not read in, since doing so
1688 * would break ordering constraints with xfs_reclaim.
1692 struct xfs_trans *tp,
1695 struct xfs_buf **IO_agbp,
1698 struct xfs_mount *mp = tp->t_mountp;
1699 struct xfs_buf *agbp;
1700 xfs_agnumber_t agno;
1704 xfs_agnumber_t start_agno;
1705 struct xfs_perag *pag;
1710 * If the caller passes in a pointer to the AGI buffer,
1711 * continue where we left off before. In this case, we
1712 * know that the allocation group has free inodes.
1719 * We do not have an agbp, so select an initial allocation
1720 * group for inode allocation.
1722 start_agno = xfs_ialloc_ag_select(tp, parent, mode);
1723 if (start_agno == NULLAGNUMBER) {
1729 * If we have already hit the ceiling of inode blocks then clear
1730 * okalloc so we scan all available agi structures for a free
1733 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1734 * which will sacrifice the preciseness but improve the performance.
1736 if (mp->m_maxicount &&
1737 percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos
1738 > mp->m_maxicount) {
1744 * Loop until we find an allocation group that either has free inodes
1745 * or in which we can allocate some inodes. Iterate through the
1746 * allocation groups upward, wrapping at the end.
1750 pag = xfs_perag_get(mp, agno);
1751 if (!pag->pagi_inodeok) {
1752 xfs_ialloc_next_ag(mp);
1756 if (!pag->pagi_init) {
1757 error = xfs_ialloc_pagi_init(mp, tp, agno);
1763 * Do a first racy fast path check if this AG is usable.
1765 if (!pag->pagi_freecount && !okalloc)
1769 * Then read in the AGI buffer and recheck with the AGI buffer
1772 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1776 if (pag->pagi_freecount) {
1782 goto nextag_relse_buffer;
1785 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1787 xfs_trans_brelse(tp, agbp);
1789 if (error != -ENOSPC)
1799 * We successfully allocated some inodes, return
1800 * the current context to the caller so that it
1801 * can commit the current transaction and call
1802 * us again where we left off.
1804 ASSERT(pag->pagi_freecount > 0);
1812 nextag_relse_buffer:
1813 xfs_trans_brelse(tp, agbp);
1816 if (++agno == mp->m_sb.sb_agcount)
1818 if (agno == start_agno) {
1820 return noroom ? -ENOSPC : 0;
1826 return xfs_dialloc_ag(tp, agbp, parent, inop);
1833 * Free the blocks of an inode chunk. We must consider that the inode chunk
1834 * might be sparse and only free the regions that are allocated as part of the
1838 xfs_difree_inode_chunk(
1839 struct xfs_trans *tp,
1840 xfs_agnumber_t agno,
1841 struct xfs_inobt_rec_incore *rec)
1843 struct xfs_mount *mp = tp->t_mountp;
1844 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1846 int startidx, endidx;
1848 xfs_agblock_t agbno;
1850 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1852 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1853 /* not sparse, calculate extent info directly */
1854 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
1855 mp->m_ialloc_blks, &XFS_RMAP_OINFO_INODES);
1859 /* holemask is only 16-bits (fits in an unsigned long) */
1860 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1861 holemask[0] = rec->ir_holemask;
1864 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1865 * holemask and convert the start/end index of each range to an extent.
1866 * We start with the start and end index both pointing at the first 0 in
1869 startidx = endidx = find_first_zero_bit(holemask,
1870 XFS_INOBT_HOLEMASK_BITS);
1871 nextbit = startidx + 1;
1872 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1873 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1876 * If the next zero bit is contiguous, update the end index of
1877 * the current range and continue.
1879 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1880 nextbit == endidx + 1) {
1886 * nextbit is not contiguous with the current end index. Convert
1887 * the current start/end to an extent and add it to the free
1890 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1891 mp->m_sb.sb_inopblock;
1892 contigblk = ((endidx - startidx + 1) *
1893 XFS_INODES_PER_HOLEMASK_BIT) /
1894 mp->m_sb.sb_inopblock;
1896 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1897 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1898 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
1899 contigblk, &XFS_RMAP_OINFO_INODES);
1901 /* reset range to current bit and carry on... */
1902 startidx = endidx = nextbit;
1911 struct xfs_mount *mp,
1912 struct xfs_trans *tp,
1913 struct xfs_buf *agbp,
1915 struct xfs_icluster *xic,
1916 struct xfs_inobt_rec_incore *orec)
1918 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1919 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1920 struct xfs_perag *pag;
1921 struct xfs_btree_cur *cur;
1922 struct xfs_inobt_rec_incore rec;
1928 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1929 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1932 * Initialize the cursor.
1934 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1936 error = xfs_check_agi_freecount(cur, agi);
1941 * Look for the entry describing this inode.
1943 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1944 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1948 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1949 error = xfs_inobt_get_rec(cur, &rec, &i);
1951 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1955 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1957 * Get the offset in the inode chunk.
1959 off = agino - rec.ir_startino;
1960 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1961 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1963 * Mark the inode free & increment the count.
1965 rec.ir_free |= XFS_INOBT_MASK(off);
1969 * When an inode chunk is free, it becomes eligible for removal. Don't
1970 * remove the chunk if the block size is large enough for multiple inode
1971 * chunks (that might not be free).
1973 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1974 rec.ir_free == XFS_INOBT_ALL_FREE &&
1975 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1976 xic->deleted = true;
1977 xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1978 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1981 * Remove the inode cluster from the AGI B+Tree, adjust the
1982 * AGI and Superblock inode counts, and mark the disk space
1983 * to be freed when the transaction is committed.
1985 ilen = rec.ir_freecount;
1986 be32_add_cpu(&agi->agi_count, -ilen);
1987 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1988 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1989 pag = xfs_perag_get(mp, agno);
1990 pag->pagi_freecount -= ilen - 1;
1991 pag->pagi_count -= ilen;
1993 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1994 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1996 if ((error = xfs_btree_delete(cur, &i))) {
1997 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2002 xfs_difree_inode_chunk(tp, agno, &rec);
2004 xic->deleted = false;
2006 error = xfs_inobt_update(cur, &rec);
2008 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2014 * Change the inode free counts and log the ag/sb changes.
2016 be32_add_cpu(&agi->agi_freecount, 1);
2017 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2018 pag = xfs_perag_get(mp, agno);
2019 pag->pagi_freecount++;
2021 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2024 error = xfs_check_agi_freecount(cur, agi);
2029 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2033 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2038 * Free an inode in the free inode btree.
2042 struct xfs_mount *mp,
2043 struct xfs_trans *tp,
2044 struct xfs_buf *agbp,
2046 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2048 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
2049 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
2050 struct xfs_btree_cur *cur;
2051 struct xfs_inobt_rec_incore rec;
2052 int offset = agino - ibtrec->ir_startino;
2056 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2058 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2063 * If the record does not exist in the finobt, we must have just
2064 * freed an inode in a previously fully allocated chunk. If not,
2065 * something is out of sync.
2067 XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
2069 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2071 ibtrec->ir_freecount,
2072 ibtrec->ir_free, &i);
2081 * Read and update the existing record. We could just copy the ibtrec
2082 * across here, but that would defeat the purpose of having redundant
2083 * metadata. By making the modifications independently, we can catch
2084 * corruptions that we wouldn't see if we just copied from one record
2087 error = xfs_inobt_get_rec(cur, &rec, &i);
2090 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
2092 rec.ir_free |= XFS_INOBT_MASK(offset);
2095 XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
2096 (rec.ir_freecount == ibtrec->ir_freecount),
2100 * The content of inobt records should always match between the inobt
2101 * and finobt. The lifecycle of records in the finobt is different from
2102 * the inobt in that the finobt only tracks records with at least one
2103 * free inode. Hence, if all of the inodes are free and we aren't
2104 * keeping inode chunks permanently on disk, remove the record.
2105 * Otherwise, update the record with the new information.
2107 * Note that we currently can't free chunks when the block size is large
2108 * enough for multiple chunks. Leave the finobt record to remain in sync
2111 if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2112 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2113 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2114 error = xfs_btree_delete(cur, &i);
2119 error = xfs_inobt_update(cur, &rec);
2125 error = xfs_check_agi_freecount(cur, agi);
2129 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2133 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2138 * Free disk inode. Carefully avoids touching the incore inode, all
2139 * manipulations incore are the caller's responsibility.
2140 * The on-disk inode is not changed by this operation, only the
2141 * btree (free inode mask) is changed.
2145 struct xfs_trans *tp, /* transaction pointer */
2146 xfs_ino_t inode, /* inode to be freed */
2147 struct xfs_icluster *xic) /* cluster info if deleted */
2150 xfs_agblock_t agbno; /* block number containing inode */
2151 struct xfs_buf *agbp; /* buffer for allocation group header */
2152 xfs_agino_t agino; /* allocation group inode number */
2153 xfs_agnumber_t agno; /* allocation group number */
2154 int error; /* error return value */
2155 struct xfs_mount *mp; /* mount structure for filesystem */
2156 struct xfs_inobt_rec_incore rec;/* btree record */
2161 * Break up inode number into its components.
2163 agno = XFS_INO_TO_AGNO(mp, inode);
2164 if (agno >= mp->m_sb.sb_agcount) {
2165 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2166 __func__, agno, mp->m_sb.sb_agcount);
2170 agino = XFS_INO_TO_AGINO(mp, inode);
2171 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
2172 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2173 __func__, (unsigned long long)inode,
2174 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2178 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2179 if (agbno >= mp->m_sb.sb_agblocks) {
2180 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2181 __func__, agbno, mp->m_sb.sb_agblocks);
2186 * Get the allocation group header.
2188 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2190 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2196 * Fix up the inode allocation btree.
2198 error = xfs_difree_inobt(mp, tp, agbp, agino, xic, &rec);
2203 * Fix up the free inode btree.
2205 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2206 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2219 struct xfs_mount *mp,
2220 struct xfs_trans *tp,
2221 xfs_agnumber_t agno,
2223 xfs_agblock_t agbno,
2224 xfs_agblock_t *chunk_agbno,
2225 xfs_agblock_t *offset_agbno,
2228 struct xfs_inobt_rec_incore rec;
2229 struct xfs_btree_cur *cur;
2230 struct xfs_buf *agbp;
2234 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2237 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2238 __func__, error, agno);
2243 * Lookup the inode record for the given agino. If the record cannot be
2244 * found, then it's an invalid inode number and we should abort. Once
2245 * we have a record, we need to ensure it contains the inode number
2246 * we are looking up.
2248 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2249 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2252 error = xfs_inobt_get_rec(cur, &rec, &i);
2253 if (!error && i == 0)
2257 xfs_trans_brelse(tp, agbp);
2258 xfs_btree_del_cursor(cur, error);
2262 /* check that the returned record contains the required inode */
2263 if (rec.ir_startino > agino ||
2264 rec.ir_startino + mp->m_ialloc_inos <= agino)
2267 /* for untrusted inodes check it is allocated first */
2268 if ((flags & XFS_IGET_UNTRUSTED) &&
2269 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2272 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2273 *offset_agbno = agbno - *chunk_agbno;
2278 * Return the location of the inode in imap, for mapping it into a buffer.
2282 xfs_mount_t *mp, /* file system mount structure */
2283 xfs_trans_t *tp, /* transaction pointer */
2284 xfs_ino_t ino, /* inode to locate */
2285 struct xfs_imap *imap, /* location map structure */
2286 uint flags) /* flags for inode btree lookup */
2288 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2289 xfs_agino_t agino; /* inode number within alloc group */
2290 xfs_agnumber_t agno; /* allocation group number */
2291 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2292 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2293 int error; /* error code */
2294 int offset; /* index of inode in its buffer */
2295 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2297 ASSERT(ino != NULLFSINO);
2300 * Split up the inode number into its parts.
2302 agno = XFS_INO_TO_AGNO(mp, ino);
2303 agino = XFS_INO_TO_AGINO(mp, ino);
2304 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2305 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2306 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2309 * Don't output diagnostic information for untrusted inodes
2310 * as they can be invalid without implying corruption.
2312 if (flags & XFS_IGET_UNTRUSTED)
2314 if (agno >= mp->m_sb.sb_agcount) {
2316 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2317 __func__, agno, mp->m_sb.sb_agcount);
2319 if (agbno >= mp->m_sb.sb_agblocks) {
2321 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2322 __func__, (unsigned long long)agbno,
2323 (unsigned long)mp->m_sb.sb_agblocks);
2325 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2327 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2329 XFS_AGINO_TO_INO(mp, agno, agino));
2337 * For bulkstat and handle lookups, we have an untrusted inode number
2338 * that we have to verify is valid. We cannot do this just by reading
2339 * the inode buffer as it may have been unlinked and removed leaving
2340 * inodes in stale state on disk. Hence we have to do a btree lookup
2341 * in all cases where an untrusted inode number is passed.
2343 if (flags & XFS_IGET_UNTRUSTED) {
2344 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2345 &chunk_agbno, &offset_agbno, flags);
2352 * If the inode cluster size is the same as the blocksize or
2353 * smaller we get to the buffer by simple arithmetics.
2355 if (mp->m_blocks_per_cluster == 1) {
2356 offset = XFS_INO_TO_OFFSET(mp, ino);
2357 ASSERT(offset < mp->m_sb.sb_inopblock);
2359 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2360 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2361 imap->im_boffset = (unsigned short)(offset <<
2362 mp->m_sb.sb_inodelog);
2367 * If the inode chunks are aligned then use simple maths to
2368 * find the location. Otherwise we have to do a btree
2369 * lookup to find the location.
2371 if (mp->m_inoalign_mask) {
2372 offset_agbno = agbno & mp->m_inoalign_mask;
2373 chunk_agbno = agbno - offset_agbno;
2375 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2376 &chunk_agbno, &offset_agbno, flags);
2382 ASSERT(agbno >= chunk_agbno);
2383 cluster_agbno = chunk_agbno +
2384 ((offset_agbno / mp->m_blocks_per_cluster) *
2385 mp->m_blocks_per_cluster);
2386 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2387 XFS_INO_TO_OFFSET(mp, ino);
2389 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2390 imap->im_len = XFS_FSB_TO_BB(mp, mp->m_blocks_per_cluster);
2391 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2394 * If the inode number maps to a block outside the bounds
2395 * of the file system then return NULL rather than calling
2396 * read_buf and panicing when we get an error from the
2399 if ((imap->im_blkno + imap->im_len) >
2400 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2402 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2403 __func__, (unsigned long long) imap->im_blkno,
2404 (unsigned long long) imap->im_len,
2405 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2412 * Compute and fill in value of m_in_maxlevels.
2415 xfs_ialloc_compute_maxlevels(
2416 xfs_mount_t *mp) /* file system mount structure */
2420 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2421 mp->m_in_maxlevels = xfs_btree_compute_maxlevels(mp->m_inobt_mnr,
2426 * Log specified fields for the ag hdr (inode section). The growth of the agi
2427 * structure over time requires that we interpret the buffer as two logical
2428 * regions delineated by the end of the unlinked list. This is due to the size
2429 * of the hash table and its location in the middle of the agi.
2431 * For example, a request to log a field before agi_unlinked and a field after
2432 * agi_unlinked could cause us to log the entire hash table and use an excessive
2433 * amount of log space. To avoid this behavior, log the region up through
2434 * agi_unlinked in one call and the region after agi_unlinked through the end of
2435 * the structure in another.
2439 xfs_trans_t *tp, /* transaction pointer */
2440 xfs_buf_t *bp, /* allocation group header buffer */
2441 int fields) /* bitmask of fields to log */
2443 int first; /* first byte number */
2444 int last; /* last byte number */
2445 static const short offsets[] = { /* field starting offsets */
2446 /* keep in sync with bit definitions */
2447 offsetof(xfs_agi_t, agi_magicnum),
2448 offsetof(xfs_agi_t, agi_versionnum),
2449 offsetof(xfs_agi_t, agi_seqno),
2450 offsetof(xfs_agi_t, agi_length),
2451 offsetof(xfs_agi_t, agi_count),
2452 offsetof(xfs_agi_t, agi_root),
2453 offsetof(xfs_agi_t, agi_level),
2454 offsetof(xfs_agi_t, agi_freecount),
2455 offsetof(xfs_agi_t, agi_newino),
2456 offsetof(xfs_agi_t, agi_dirino),
2457 offsetof(xfs_agi_t, agi_unlinked),
2458 offsetof(xfs_agi_t, agi_free_root),
2459 offsetof(xfs_agi_t, agi_free_level),
2463 xfs_agi_t *agi; /* allocation group header */
2465 agi = XFS_BUF_TO_AGI(bp);
2466 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2470 * Compute byte offsets for the first and last fields in the first
2471 * region and log the agi buffer. This only logs up through
2474 if (fields & XFS_AGI_ALL_BITS_R1) {
2475 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2477 xfs_trans_log_buf(tp, bp, first, last);
2481 * Mask off the bits in the first region and calculate the first and
2482 * last field offsets for any bits in the second region.
2484 fields &= ~XFS_AGI_ALL_BITS_R1;
2486 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2488 xfs_trans_log_buf(tp, bp, first, last);
2492 static xfs_failaddr_t
2496 struct xfs_mount *mp = bp->b_target->bt_mount;
2497 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2500 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2501 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2502 return __this_address;
2503 if (!xfs_log_check_lsn(mp,
2504 be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn)))
2505 return __this_address;
2509 * Validate the magic number of the agi block.
2511 if (!xfs_verify_magic(bp, agi->agi_magicnum))
2512 return __this_address;
2513 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2514 return __this_address;
2516 if (be32_to_cpu(agi->agi_level) < 1 ||
2517 be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2518 return __this_address;
2520 if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2521 (be32_to_cpu(agi->agi_free_level) < 1 ||
2522 be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2523 return __this_address;
2526 * during growfs operations, the perag is not fully initialised,
2527 * so we can't use it for any useful checking. growfs ensures we can't
2528 * use it by using uncached buffers that don't have the perag attached
2529 * so we can detect and avoid this problem.
2531 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2532 return __this_address;
2534 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2535 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2537 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2538 return __this_address;
2545 xfs_agi_read_verify(
2548 struct xfs_mount *mp = bp->b_target->bt_mount;
2551 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2552 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2553 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2555 fa = xfs_agi_verify(bp);
2556 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2557 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2562 xfs_agi_write_verify(
2565 struct xfs_mount *mp = bp->b_target->bt_mount;
2566 struct xfs_buf_log_item *bip = bp->b_log_item;
2569 fa = xfs_agi_verify(bp);
2571 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2575 if (!xfs_sb_version_hascrc(&mp->m_sb))
2579 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2580 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2583 const struct xfs_buf_ops xfs_agi_buf_ops = {
2585 .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2586 .verify_read = xfs_agi_read_verify,
2587 .verify_write = xfs_agi_write_verify,
2588 .verify_struct = xfs_agi_verify,
2592 * Read in the allocation group header (inode allocation section)
2596 struct xfs_mount *mp, /* file system mount structure */
2597 struct xfs_trans *tp, /* transaction pointer */
2598 xfs_agnumber_t agno, /* allocation group number */
2599 struct xfs_buf **bpp) /* allocation group hdr buf */
2603 trace_xfs_read_agi(mp, agno);
2605 ASSERT(agno != NULLAGNUMBER);
2606 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2607 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2608 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2612 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2614 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2619 xfs_ialloc_read_agi(
2620 struct xfs_mount *mp, /* file system mount structure */
2621 struct xfs_trans *tp, /* transaction pointer */
2622 xfs_agnumber_t agno, /* allocation group number */
2623 struct xfs_buf **bpp) /* allocation group hdr buf */
2625 struct xfs_agi *agi; /* allocation group header */
2626 struct xfs_perag *pag; /* per allocation group data */
2629 trace_xfs_ialloc_read_agi(mp, agno);
2631 error = xfs_read_agi(mp, tp, agno, bpp);
2635 agi = XFS_BUF_TO_AGI(*bpp);
2636 pag = xfs_perag_get(mp, agno);
2637 if (!pag->pagi_init) {
2638 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2639 pag->pagi_count = be32_to_cpu(agi->agi_count);
2644 * It's possible for these to be out of sync if
2645 * we are in the middle of a forced shutdown.
2647 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2648 XFS_FORCED_SHUTDOWN(mp));
2654 * Read in the agi to initialise the per-ag data in the mount structure
2657 xfs_ialloc_pagi_init(
2658 xfs_mount_t *mp, /* file system mount structure */
2659 xfs_trans_t *tp, /* transaction pointer */
2660 xfs_agnumber_t agno) /* allocation group number */
2662 xfs_buf_t *bp = NULL;
2665 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2669 xfs_trans_brelse(tp, bp);
2673 /* Is there an inode record covering a given range of inode numbers? */
2675 xfs_ialloc_has_inode_record(
2676 struct xfs_btree_cur *cur,
2681 struct xfs_inobt_rec_incore irec;
2689 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2690 while (error == 0 && has_record) {
2691 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2692 if (error || irec.ir_startino > high)
2695 agino = irec.ir_startino;
2696 holemask = irec.ir_holemask;
2697 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2698 i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2701 if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2708 error = xfs_btree_increment(cur, 0, &has_record);
2713 /* Is there an inode record covering a given extent? */
2715 xfs_ialloc_has_inodes_at_extent(
2716 struct xfs_btree_cur *cur,
2724 low = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2725 high = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2727 return xfs_ialloc_has_inode_record(cur, low, high, exists);
2730 struct xfs_ialloc_count_inodes {
2732 xfs_agino_t freecount;
2735 /* Record inode counts across all inobt records. */
2737 xfs_ialloc_count_inodes_rec(
2738 struct xfs_btree_cur *cur,
2739 union xfs_btree_rec *rec,
2742 struct xfs_inobt_rec_incore irec;
2743 struct xfs_ialloc_count_inodes *ci = priv;
2745 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2746 ci->count += irec.ir_count;
2747 ci->freecount += irec.ir_freecount;
2752 /* Count allocated and free inodes under an inobt. */
2754 xfs_ialloc_count_inodes(
2755 struct xfs_btree_cur *cur,
2757 xfs_agino_t *freecount)
2759 struct xfs_ialloc_count_inodes ci = {0};
2762 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2763 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2768 *freecount = ci.freecount;