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_inode.h"
16 #include "xfs_btree.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_ialloc_btree.h"
19 #include "xfs_alloc.h"
20 #include "xfs_errortag.h"
21 #include "xfs_error.h"
23 #include "xfs_trans.h"
24 #include "xfs_buf_item.h"
25 #include "xfs_icreate_item.h"
26 #include "xfs_icache.h"
27 #include "xfs_trace.h"
32 * Lookup a record by ino in the btree given by cur.
36 struct xfs_btree_cur *cur, /* btree cursor */
37 xfs_agino_t ino, /* starting inode of chunk */
38 xfs_lookup_t dir, /* <=, >=, == */
39 int *stat) /* success/failure */
41 cur->bc_rec.i.ir_startino = ino;
42 cur->bc_rec.i.ir_holemask = 0;
43 cur->bc_rec.i.ir_count = 0;
44 cur->bc_rec.i.ir_freecount = 0;
45 cur->bc_rec.i.ir_free = 0;
46 return xfs_btree_lookup(cur, dir, stat);
50 * Update the record referred to by cur to the value given.
51 * This either works (return 0) or gets an EFSCORRUPTED error.
53 STATIC int /* error */
55 struct xfs_btree_cur *cur, /* btree cursor */
56 xfs_inobt_rec_incore_t *irec) /* btree record */
58 union xfs_btree_rec rec;
60 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
61 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
62 rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
63 rec.inobt.ir_u.sp.ir_count = irec->ir_count;
64 rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
66 /* ir_holemask/ir_count not supported on-disk */
67 rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
69 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
70 return xfs_btree_update(cur, &rec);
73 /* Convert on-disk btree record to incore inobt record. */
75 xfs_inobt_btrec_to_irec(
77 union xfs_btree_rec *rec,
78 struct xfs_inobt_rec_incore *irec)
80 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
81 if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
82 irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
83 irec->ir_count = rec->inobt.ir_u.sp.ir_count;
84 irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
87 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
88 * values for full inode chunks.
90 irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
91 irec->ir_count = XFS_INODES_PER_CHUNK;
93 be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
95 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
99 * Get the data from the pointed-to record.
103 struct xfs_btree_cur *cur,
104 struct xfs_inobt_rec_incore *irec,
107 struct xfs_mount *mp = cur->bc_mp;
108 xfs_agnumber_t agno = cur->bc_ag.agno;
109 union xfs_btree_rec *rec;
113 error = xfs_btree_get_rec(cur, &rec, stat);
114 if (error || *stat == 0)
117 xfs_inobt_btrec_to_irec(mp, rec, irec);
119 if (!xfs_verify_agino(mp, agno, irec->ir_startino))
121 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
122 irec->ir_count > XFS_INODES_PER_CHUNK)
124 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
127 /* if there are no holes, return the first available offset */
128 if (!xfs_inobt_issparse(irec->ir_holemask))
129 realfree = irec->ir_free;
131 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
132 if (hweight64(realfree) != irec->ir_freecount)
139 "%s Inode BTree record corruption in AG %d detected!",
140 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno);
142 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
143 irec->ir_startino, irec->ir_count, irec->ir_freecount,
144 irec->ir_free, irec->ir_holemask);
145 return -EFSCORRUPTED;
149 * Insert a single inobt record. Cursor must already point to desired location.
152 xfs_inobt_insert_rec(
153 struct xfs_btree_cur *cur,
160 cur->bc_rec.i.ir_holemask = holemask;
161 cur->bc_rec.i.ir_count = count;
162 cur->bc_rec.i.ir_freecount = freecount;
163 cur->bc_rec.i.ir_free = free;
164 return xfs_btree_insert(cur, stat);
168 * Insert records describing a newly allocated inode chunk into the inobt.
172 struct xfs_mount *mp,
173 struct xfs_trans *tp,
174 struct xfs_buf *agbp,
179 struct xfs_btree_cur *cur;
180 struct xfs_agi *agi = agbp->b_addr;
181 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
186 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
188 for (thisino = newino;
189 thisino < newino + newlen;
190 thisino += XFS_INODES_PER_CHUNK) {
191 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
193 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
198 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
199 XFS_INODES_PER_CHUNK,
200 XFS_INODES_PER_CHUNK,
201 XFS_INOBT_ALL_FREE, &i);
203 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
209 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
215 * Verify that the number of free inodes in the AGI is correct.
219 xfs_check_agi_freecount(
220 struct xfs_btree_cur *cur,
223 if (cur->bc_nlevels == 1) {
224 xfs_inobt_rec_incore_t rec;
229 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
234 error = xfs_inobt_get_rec(cur, &rec, &i);
239 freecount += rec.ir_freecount;
240 error = xfs_btree_increment(cur, 0, &i);
246 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
247 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
252 #define xfs_check_agi_freecount(cur, agi) 0
256 * Initialise a new set of inodes. When called without a transaction context
257 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
258 * than logging them (which in a transaction context puts them into the AIL
259 * for writeback rather than the xfsbufd queue).
262 xfs_ialloc_inode_init(
263 struct xfs_mount *mp,
264 struct xfs_trans *tp,
265 struct list_head *buffer_list,
269 xfs_agblock_t length,
272 struct xfs_buf *fbuf;
273 struct xfs_dinode *free;
282 * Loop over the new block(s), filling in the inodes. For small block
283 * sizes, manipulate the inodes in buffers which are multiples of the
286 nbufs = length / M_IGEO(mp)->blocks_per_cluster;
289 * Figure out what version number to use in the inodes we create. If
290 * the superblock version has caught up to the one that supports the new
291 * inode format, then use the new inode version. Otherwise use the old
292 * version so that old kernels will continue to be able to use the file
295 * For v3 inodes, we also need to write the inode number into the inode,
296 * so calculate the first inode number of the chunk here as
297 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
298 * across multiple filesystem blocks (such as a cluster) and so cannot
299 * be used in the cluster buffer loop below.
301 * Further, because we are writing the inode directly into the buffer
302 * and calculating a CRC on the entire inode, we have ot log the entire
303 * inode so that the entire range the CRC covers is present in the log.
304 * That means for v3 inode we log the entire buffer rather than just the
307 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
309 ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
312 * log the initialisation that is about to take place as an
313 * logical operation. This means the transaction does not
314 * need to log the physical changes to the inode buffers as log
315 * recovery will know what initialisation is actually needed.
316 * Hence we only need to log the buffers as "ordered" buffers so
317 * they track in the AIL as if they were physically logged.
320 xfs_icreate_log(tp, agno, agbno, icount,
321 mp->m_sb.sb_inodesize, length, gen);
325 for (j = 0; j < nbufs; j++) {
329 d = XFS_AGB_TO_DADDR(mp, agno, agbno +
330 (j * M_IGEO(mp)->blocks_per_cluster));
331 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
332 mp->m_bsize * M_IGEO(mp)->blocks_per_cluster,
333 XBF_UNMAPPED, &fbuf);
337 /* Initialize the inode buffers and log them appropriately. */
338 fbuf->b_ops = &xfs_inode_buf_ops;
339 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
340 for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
341 int ioffset = i << mp->m_sb.sb_inodelog;
342 uint isize = XFS_DINODE_SIZE(&mp->m_sb);
344 free = xfs_make_iptr(mp, fbuf, i);
345 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
346 free->di_version = version;
347 free->di_gen = cpu_to_be32(gen);
348 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
351 free->di_ino = cpu_to_be64(ino);
353 uuid_copy(&free->di_uuid,
354 &mp->m_sb.sb_meta_uuid);
355 xfs_dinode_calc_crc(mp, free);
357 /* just log the inode core */
358 xfs_trans_log_buf(tp, fbuf, ioffset,
359 ioffset + isize - 1);
365 * Mark the buffer as an inode allocation buffer so it
366 * sticks in AIL at the point of this allocation
367 * transaction. This ensures the they are on disk before
368 * the tail of the log can be moved past this
369 * transaction (i.e. by preventing relogging from moving
370 * it forward in the log).
372 xfs_trans_inode_alloc_buf(tp, fbuf);
375 * Mark the buffer as ordered so that they are
376 * not physically logged in the transaction but
377 * still tracked in the AIL as part of the
378 * transaction and pin the log appropriately.
380 xfs_trans_ordered_buf(tp, fbuf);
383 fbuf->b_flags |= XBF_DONE;
384 xfs_buf_delwri_queue(fbuf, buffer_list);
392 * Align startino and allocmask for a recently allocated sparse chunk such that
393 * they are fit for insertion (or merge) into the on-disk inode btrees.
397 * When enabled, sparse inode support increases the inode alignment from cluster
398 * size to inode chunk size. This means that the minimum range between two
399 * non-adjacent inode records in the inobt is large enough for a full inode
400 * record. This allows for cluster sized, cluster aligned block allocation
401 * without need to worry about whether the resulting inode record overlaps with
402 * another record in the tree. Without this basic rule, we would have to deal
403 * with the consequences of overlap by potentially undoing recent allocations in
404 * the inode allocation codepath.
406 * Because of this alignment rule (which is enforced on mount), there are two
407 * inobt possibilities for newly allocated sparse chunks. One is that the
408 * aligned inode record for the chunk covers a range of inodes not already
409 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
410 * other is that a record already exists at the aligned startino that considers
411 * the newly allocated range as sparse. In the latter case, record content is
412 * merged in hope that sparse inode chunks fill to full chunks over time.
415 xfs_align_sparse_ino(
416 struct xfs_mount *mp,
417 xfs_agino_t *startino,
424 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
425 mod = agbno % mp->m_sb.sb_inoalignmt;
429 /* calculate the inode offset and align startino */
430 offset = XFS_AGB_TO_AGINO(mp, mod);
434 * Since startino has been aligned down, left shift allocmask such that
435 * it continues to represent the same physical inodes relative to the
438 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
442 * Determine whether the source inode record can merge into the target. Both
443 * records must be sparse, the inode ranges must match and there must be no
444 * allocation overlap between the records.
447 __xfs_inobt_can_merge(
448 struct xfs_inobt_rec_incore *trec, /* tgt record */
449 struct xfs_inobt_rec_incore *srec) /* src record */
454 /* records must cover the same inode range */
455 if (trec->ir_startino != srec->ir_startino)
458 /* both records must be sparse */
459 if (!xfs_inobt_issparse(trec->ir_holemask) ||
460 !xfs_inobt_issparse(srec->ir_holemask))
463 /* both records must track some inodes */
464 if (!trec->ir_count || !srec->ir_count)
467 /* can't exceed capacity of a full record */
468 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
471 /* verify there is no allocation overlap */
472 talloc = xfs_inobt_irec_to_allocmask(trec);
473 salloc = xfs_inobt_irec_to_allocmask(srec);
481 * Merge the source inode record into the target. The caller must call
482 * __xfs_inobt_can_merge() to ensure the merge is valid.
485 __xfs_inobt_rec_merge(
486 struct xfs_inobt_rec_incore *trec, /* target */
487 struct xfs_inobt_rec_incore *srec) /* src */
489 ASSERT(trec->ir_startino == srec->ir_startino);
491 /* combine the counts */
492 trec->ir_count += srec->ir_count;
493 trec->ir_freecount += srec->ir_freecount;
496 * Merge the holemask and free mask. For both fields, 0 bits refer to
497 * allocated inodes. We combine the allocated ranges with bitwise AND.
499 trec->ir_holemask &= srec->ir_holemask;
500 trec->ir_free &= srec->ir_free;
504 * Insert a new sparse inode chunk into the associated inode btree. The inode
505 * record for the sparse chunk is pre-aligned to a startino that should match
506 * any pre-existing sparse inode record in the tree. This allows sparse chunks
509 * This function supports two modes of handling preexisting records depending on
510 * the merge flag. If merge is true, the provided record is merged with the
511 * existing record and updated in place. The merged record is returned in nrec.
512 * If merge is false, an existing record is replaced with the provided record.
513 * If no preexisting record exists, the provided record is always inserted.
515 * It is considered corruption if a merge is requested and not possible. Given
516 * the sparse inode alignment constraints, this should never happen.
519 xfs_inobt_insert_sprec(
520 struct xfs_mount *mp,
521 struct xfs_trans *tp,
522 struct xfs_buf *agbp,
524 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
525 bool merge) /* merge or replace */
527 struct xfs_btree_cur *cur;
528 struct xfs_agi *agi = agbp->b_addr;
529 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
532 struct xfs_inobt_rec_incore rec;
534 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
536 /* the new record is pre-aligned so we know where to look */
537 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
540 /* if nothing there, insert a new record and return */
542 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
543 nrec->ir_count, nrec->ir_freecount,
547 if (XFS_IS_CORRUPT(mp, i != 1)) {
548 error = -EFSCORRUPTED;
556 * A record exists at this startino. Merge or replace the record
557 * depending on what we've been asked to do.
560 error = xfs_inobt_get_rec(cur, &rec, &i);
563 if (XFS_IS_CORRUPT(mp, i != 1)) {
564 error = -EFSCORRUPTED;
567 if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) {
568 error = -EFSCORRUPTED;
573 * This should never fail. If we have coexisting records that
574 * cannot merge, something is seriously wrong.
576 if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) {
577 error = -EFSCORRUPTED;
581 trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
582 rec.ir_holemask, nrec->ir_startino,
585 /* merge to nrec to output the updated record */
586 __xfs_inobt_rec_merge(nrec, &rec);
588 trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
591 error = xfs_inobt_rec_check_count(mp, nrec);
596 error = xfs_inobt_update(cur, nrec);
601 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
604 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
609 * Allocate new inodes in the allocation group specified by agbp.
610 * Return 0 for success, else error code.
614 struct xfs_trans *tp,
615 struct xfs_buf *agbp,
619 struct xfs_alloc_arg args;
622 xfs_agino_t newino; /* new first inode's number */
623 xfs_agino_t newlen; /* new number of inodes */
624 int isaligned = 0; /* inode allocation at stripe */
626 /* init. to full chunk */
627 uint16_t allocmask = (uint16_t) -1;
628 struct xfs_inobt_rec_incore rec;
629 struct xfs_perag *pag;
630 struct xfs_ino_geometry *igeo = M_IGEO(tp->t_mountp);
633 memset(&args, 0, sizeof(args));
635 args.mp = tp->t_mountp;
636 args.fsbno = NULLFSBLOCK;
637 args.oinfo = XFS_RMAP_OINFO_INODES;
640 /* randomly do sparse inode allocations */
641 if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
642 igeo->ialloc_min_blks < igeo->ialloc_blks)
643 do_sparse = prandom_u32() & 1;
647 * Locking will ensure that we don't have two callers in here
650 newlen = igeo->ialloc_inos;
651 if (igeo->maxicount &&
652 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
655 args.minlen = args.maxlen = igeo->ialloc_blks;
657 * First try to allocate inodes contiguous with the last-allocated
658 * chunk of inodes. If the filesystem is striped, this will fill
659 * an entire stripe unit with inodes.
662 newino = be32_to_cpu(agi->agi_newino);
663 agno = be32_to_cpu(agi->agi_seqno);
664 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
668 if (likely(newino != NULLAGINO &&
669 (args.agbno < be32_to_cpu(agi->agi_length)))) {
670 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
671 args.type = XFS_ALLOCTYPE_THIS_BNO;
675 * We need to take into account alignment here to ensure that
676 * we don't modify the free list if we fail to have an exact
677 * block. If we don't have an exact match, and every oher
678 * attempt allocation attempt fails, we'll end up cancelling
679 * a dirty transaction and shutting down.
681 * For an exact allocation, alignment must be 1,
682 * however we need to take cluster alignment into account when
683 * fixing up the freelist. Use the minalignslop field to
684 * indicate that extra blocks might be required for alignment,
685 * but not to use them in the actual exact allocation.
688 args.minalignslop = igeo->cluster_align - 1;
690 /* Allow space for the inode btree to split. */
691 args.minleft = igeo->inobt_maxlevels;
692 if ((error = xfs_alloc_vextent(&args)))
696 * This request might have dirtied the transaction if the AG can
697 * satisfy the request, but the exact block was not available.
698 * If the allocation did fail, subsequent requests will relax
699 * the exact agbno requirement and increase the alignment
700 * instead. It is critical that the total size of the request
701 * (len + alignment + slop) does not increase from this point
702 * on, so reset minalignslop to ensure it is not included in
703 * subsequent requests.
705 args.minalignslop = 0;
708 if (unlikely(args.fsbno == NULLFSBLOCK)) {
710 * Set the alignment for the allocation.
711 * If stripe alignment is turned on then align at stripe unit
713 * If the cluster size is smaller than a filesystem block
714 * then we're doing I/O for inodes in filesystem block size
715 * pieces, so don't need alignment anyway.
718 if (igeo->ialloc_align) {
719 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
720 args.alignment = args.mp->m_dalign;
723 args.alignment = igeo->cluster_align;
725 * Need to figure out where to allocate the inode blocks.
726 * Ideally they should be spaced out through the a.g.
727 * For now, just allocate blocks up front.
729 args.agbno = be32_to_cpu(agi->agi_root);
730 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
732 * Allocate a fixed-size extent of inodes.
734 args.type = XFS_ALLOCTYPE_NEAR_BNO;
737 * Allow space for the inode btree to split.
739 args.minleft = igeo->inobt_maxlevels;
740 if ((error = xfs_alloc_vextent(&args)))
745 * If stripe alignment is turned on, then try again with cluster
748 if (isaligned && args.fsbno == NULLFSBLOCK) {
749 args.type = XFS_ALLOCTYPE_NEAR_BNO;
750 args.agbno = be32_to_cpu(agi->agi_root);
751 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
752 args.alignment = igeo->cluster_align;
753 if ((error = xfs_alloc_vextent(&args)))
758 * Finally, try a sparse allocation if the filesystem supports it and
759 * the sparse allocation length is smaller than a full chunk.
761 if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
762 igeo->ialloc_min_blks < igeo->ialloc_blks &&
763 args.fsbno == NULLFSBLOCK) {
765 args.type = XFS_ALLOCTYPE_NEAR_BNO;
766 args.agbno = be32_to_cpu(agi->agi_root);
767 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
768 args.alignment = args.mp->m_sb.sb_spino_align;
771 args.minlen = igeo->ialloc_min_blks;
772 args.maxlen = args.minlen;
775 * The inode record will be aligned to full chunk size. We must
776 * prevent sparse allocation from AG boundaries that result in
777 * invalid inode records, such as records that start at agbno 0
778 * or extend beyond the AG.
780 * Set min agbno to the first aligned, non-zero agbno and max to
781 * the last aligned agbno that is at least one full chunk from
784 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
785 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
786 args.mp->m_sb.sb_inoalignmt) -
789 error = xfs_alloc_vextent(&args);
793 newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
794 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
795 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
798 if (args.fsbno == NULLFSBLOCK) {
802 ASSERT(args.len == args.minlen);
805 * Stamp and write the inode buffers.
807 * Seed the new inode cluster with a random generation number. This
808 * prevents short-term reuse of generation numbers if a chunk is
809 * freed and then immediately reallocated. We use random numbers
810 * rather than a linear progression to prevent the next generation
811 * number from being easily guessable.
813 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
814 args.agbno, args.len, prandom_u32());
819 * Convert the results.
821 newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
823 if (xfs_inobt_issparse(~allocmask)) {
825 * We've allocated a sparse chunk. Align the startino and mask.
827 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
829 rec.ir_startino = newino;
830 rec.ir_holemask = ~allocmask;
831 rec.ir_count = newlen;
832 rec.ir_freecount = newlen;
833 rec.ir_free = XFS_INOBT_ALL_FREE;
836 * Insert the sparse record into the inobt and allow for a merge
837 * if necessary. If a merge does occur, rec is updated to the
840 error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
842 if (error == -EFSCORRUPTED) {
844 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
845 XFS_AGINO_TO_INO(args.mp, agno,
847 rec.ir_holemask, rec.ir_count);
848 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
854 * We can't merge the part we've just allocated as for the inobt
855 * due to finobt semantics. The original record may or may not
856 * exist independent of whether physical inodes exist in this
859 * We must update the finobt record based on the inobt record.
860 * rec contains the fully merged and up to date inobt record
861 * from the previous call. Set merge false to replace any
862 * existing record with this one.
864 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
865 error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
866 XFS_BTNUM_FINO, &rec,
872 /* full chunk - insert new records to both btrees */
873 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
878 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
879 error = xfs_inobt_insert(args.mp, tp, agbp, newino,
880 newlen, XFS_BTNUM_FINO);
887 * Update AGI counts and newino.
889 be32_add_cpu(&agi->agi_count, newlen);
890 be32_add_cpu(&agi->agi_freecount, newlen);
892 pag->pagi_freecount += newlen;
893 pag->pagi_count += newlen;
894 agi->agi_newino = cpu_to_be32(newino);
897 * Log allocation group header fields
899 xfs_ialloc_log_agi(tp, agbp,
900 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
902 * Modify/log superblock values for inode count and inode free count.
904 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
905 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
910 STATIC xfs_agnumber_t
916 spin_lock(&mp->m_agirotor_lock);
917 agno = mp->m_agirotor;
918 if (++mp->m_agirotor >= mp->m_maxagi)
920 spin_unlock(&mp->m_agirotor_lock);
926 * Select an allocation group to look for a free inode in, based on the parent
927 * inode and the mode. Return the allocation group buffer.
929 STATIC xfs_agnumber_t
930 xfs_ialloc_ag_select(
931 xfs_trans_t *tp, /* transaction pointer */
932 xfs_ino_t parent, /* parent directory inode number */
933 umode_t mode) /* bits set to indicate file type */
935 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
936 xfs_agnumber_t agno; /* current ag number */
937 int flags; /* alloc buffer locking flags */
938 xfs_extlen_t ineed; /* blocks needed for inode allocation */
939 xfs_extlen_t longest = 0; /* longest extent available */
940 xfs_mount_t *mp; /* mount point structure */
941 int needspace; /* file mode implies space allocated */
942 xfs_perag_t *pag; /* per allocation group data */
943 xfs_agnumber_t pagno; /* parent (starting) ag number */
947 * Files of these types need at least one block if length > 0
948 * (and they won't fit in the inode, but that's hard to figure out).
950 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
952 agcount = mp->m_maxagi;
954 pagno = xfs_ialloc_next_ag(mp);
956 pagno = XFS_INO_TO_AGNO(mp, parent);
957 if (pagno >= agcount)
961 ASSERT(pagno < agcount);
964 * Loop through allocation groups, looking for one with a little
965 * free space in it. Note we don't look for free inodes, exactly.
966 * Instead, we include whether there is a need to allocate inodes
967 * to mean that blocks must be allocated for them,
968 * if none are currently free.
971 flags = XFS_ALLOC_FLAG_TRYLOCK;
973 pag = xfs_perag_get(mp, agno);
974 if (!pag->pagi_inodeok) {
975 xfs_ialloc_next_ag(mp);
979 if (!pag->pagi_init) {
980 error = xfs_ialloc_pagi_init(mp, tp, agno);
985 if (pag->pagi_freecount) {
990 if (!pag->pagf_init) {
991 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
997 * Check that there is enough free space for the file plus a
998 * chunk of inodes if we need to allocate some. If this is the
999 * first pass across the AGs, take into account the potential
1000 * space needed for alignment of inode chunks when checking the
1001 * longest contiguous free space in the AG - this prevents us
1002 * from getting ENOSPC because we have free space larger than
1003 * ialloc_blks but alignment constraints prevent us from using
1006 * If we can't find an AG with space for full alignment slack to
1007 * be taken into account, we must be near ENOSPC in all AGs.
1008 * Hence we don't include alignment for the second pass and so
1009 * if we fail allocation due to alignment issues then it is most
1010 * likely a real ENOSPC condition.
1012 ineed = M_IGEO(mp)->ialloc_min_blks;
1013 if (flags && ineed > 1)
1014 ineed += M_IGEO(mp)->cluster_align;
1015 longest = pag->pagf_longest;
1017 longest = pag->pagf_flcount > 0;
1019 if (pag->pagf_freeblks >= needspace + ineed &&
1027 * No point in iterating over the rest, if we're shutting
1030 if (XFS_FORCED_SHUTDOWN(mp))
1031 return NULLAGNUMBER;
1033 if (agno >= agcount)
1035 if (agno == pagno) {
1037 return NULLAGNUMBER;
1044 * Try to retrieve the next record to the left/right from the current one.
1047 xfs_ialloc_next_rec(
1048 struct xfs_btree_cur *cur,
1049 xfs_inobt_rec_incore_t *rec,
1057 error = xfs_btree_decrement(cur, 0, &i);
1059 error = xfs_btree_increment(cur, 0, &i);
1065 error = xfs_inobt_get_rec(cur, rec, &i);
1068 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1069 return -EFSCORRUPTED;
1077 struct xfs_btree_cur *cur,
1079 xfs_inobt_rec_incore_t *rec,
1085 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1090 error = xfs_inobt_get_rec(cur, rec, &i);
1093 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1094 return -EFSCORRUPTED;
1101 * Return the offset of the first free inode in the record. If the inode chunk
1102 * is sparsely allocated, we convert the record holemask to inode granularity
1103 * and mask off the unallocated regions from the inode free mask.
1106 xfs_inobt_first_free_inode(
1107 struct xfs_inobt_rec_incore *rec)
1109 xfs_inofree_t realfree;
1111 /* if there are no holes, return the first available offset */
1112 if (!xfs_inobt_issparse(rec->ir_holemask))
1113 return xfs_lowbit64(rec->ir_free);
1115 realfree = xfs_inobt_irec_to_allocmask(rec);
1116 realfree &= rec->ir_free;
1118 return xfs_lowbit64(realfree);
1122 * Allocate an inode using the inobt-only algorithm.
1125 xfs_dialloc_ag_inobt(
1126 struct xfs_trans *tp,
1127 struct xfs_buf *agbp,
1131 struct xfs_mount *mp = tp->t_mountp;
1132 struct xfs_agi *agi = agbp->b_addr;
1133 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1134 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1135 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1136 struct xfs_perag *pag = agbp->b_pag;
1137 struct xfs_btree_cur *cur, *tcur;
1138 struct xfs_inobt_rec_incore rec, trec;
1143 int searchdistance = 10;
1145 ASSERT(pag->pagi_init);
1146 ASSERT(pag->pagi_inodeok);
1147 ASSERT(pag->pagi_freecount > 0);
1150 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1152 * If pagino is 0 (this is the root inode allocation) use newino.
1153 * This must work because we've just allocated some.
1156 pagino = be32_to_cpu(agi->agi_newino);
1158 error = xfs_check_agi_freecount(cur, agi);
1163 * If in the same AG as the parent, try to get near the parent.
1165 if (pagno == agno) {
1166 int doneleft; /* done, to the left */
1167 int doneright; /* done, to the right */
1169 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1172 if (XFS_IS_CORRUPT(mp, i != 1)) {
1173 error = -EFSCORRUPTED;
1177 error = xfs_inobt_get_rec(cur, &rec, &j);
1180 if (XFS_IS_CORRUPT(mp, j != 1)) {
1181 error = -EFSCORRUPTED;
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 if (XFS_IS_CORRUPT(mp, i != 1)) {
1339 error = -EFSCORRUPTED;
1344 error = xfs_inobt_get_rec(cur, &rec, &i);
1347 if (XFS_IS_CORRUPT(mp, i != 1)) {
1348 error = -EFSCORRUPTED;
1351 if (rec.ir_freecount > 0)
1353 error = xfs_btree_increment(cur, 0, &i);
1356 if (XFS_IS_CORRUPT(mp, i != 1)) {
1357 error = -EFSCORRUPTED;
1363 offset = xfs_inobt_first_free_inode(&rec);
1364 ASSERT(offset >= 0);
1365 ASSERT(offset < XFS_INODES_PER_CHUNK);
1366 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1367 XFS_INODES_PER_CHUNK) == 0);
1368 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1369 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1371 error = xfs_inobt_update(cur, &rec);
1374 be32_add_cpu(&agi->agi_freecount, -1);
1375 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1376 pag->pagi_freecount--;
1378 error = xfs_check_agi_freecount(cur, agi);
1382 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1383 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1387 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1389 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1394 * Use the free inode btree to allocate an inode based on distance from the
1395 * parent. Note that the provided cursor may be deleted and replaced.
1398 xfs_dialloc_ag_finobt_near(
1400 struct xfs_btree_cur **ocur,
1401 struct xfs_inobt_rec_incore *rec)
1403 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1404 struct xfs_btree_cur *rcur; /* right search cursor */
1405 struct xfs_inobt_rec_incore rrec;
1409 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1414 error = xfs_inobt_get_rec(lcur, rec, &i);
1417 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1))
1418 return -EFSCORRUPTED;
1421 * See if we've landed in the parent inode record. The finobt
1422 * only tracks chunks with at least one free inode, so record
1423 * existence is enough.
1425 if (pagino >= rec->ir_startino &&
1426 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1430 error = xfs_btree_dup_cursor(lcur, &rcur);
1434 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1438 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1441 if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) {
1442 error = -EFSCORRUPTED;
1447 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) {
1448 error = -EFSCORRUPTED;
1451 if (i == 1 && j == 1) {
1453 * Both the left and right records are valid. Choose the closer
1454 * inode chunk to the target.
1456 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1457 (rrec.ir_startino - pagino)) {
1459 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1462 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1464 } else if (j == 1) {
1465 /* only the right record is valid */
1467 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1469 } else if (i == 1) {
1470 /* only the left record is valid */
1471 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1477 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1482 * Use the free inode btree to find a free inode based on a newino hint. If
1483 * the hint is NULL, find the first free inode in the AG.
1486 xfs_dialloc_ag_finobt_newino(
1487 struct xfs_agi *agi,
1488 struct xfs_btree_cur *cur,
1489 struct xfs_inobt_rec_incore *rec)
1494 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1495 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1500 error = xfs_inobt_get_rec(cur, rec, &i);
1503 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1504 return -EFSCORRUPTED;
1510 * Find the first inode available in the AG.
1512 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1515 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1516 return -EFSCORRUPTED;
1518 error = xfs_inobt_get_rec(cur, rec, &i);
1521 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1522 return -EFSCORRUPTED;
1528 * Update the inobt based on a modification made to the finobt. Also ensure that
1529 * the records from both trees are equivalent post-modification.
1532 xfs_dialloc_ag_update_inobt(
1533 struct xfs_btree_cur *cur, /* inobt cursor */
1534 struct xfs_inobt_rec_incore *frec, /* finobt record */
1535 int offset) /* inode offset */
1537 struct xfs_inobt_rec_incore rec;
1541 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1544 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1545 return -EFSCORRUPTED;
1547 error = xfs_inobt_get_rec(cur, &rec, &i);
1550 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1551 return -EFSCORRUPTED;
1552 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1553 XFS_INODES_PER_CHUNK) == 0);
1555 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1558 if (XFS_IS_CORRUPT(cur->bc_mp,
1559 rec.ir_free != frec->ir_free ||
1560 rec.ir_freecount != frec->ir_freecount))
1561 return -EFSCORRUPTED;
1563 return xfs_inobt_update(cur, &rec);
1567 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1568 * back to the inobt search algorithm.
1570 * The caller selected an AG for us, and made sure that free inodes are
1575 struct xfs_trans *tp,
1576 struct xfs_buf *agbp,
1580 struct xfs_mount *mp = tp->t_mountp;
1581 struct xfs_agi *agi = agbp->b_addr;
1582 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1583 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1584 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1585 struct xfs_btree_cur *cur; /* finobt cursor */
1586 struct xfs_btree_cur *icur; /* inobt cursor */
1587 struct xfs_inobt_rec_incore rec;
1593 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1594 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1597 * If pagino is 0 (this is the root inode allocation) use newino.
1598 * This must work because we've just allocated some.
1601 pagino = be32_to_cpu(agi->agi_newino);
1603 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1605 error = xfs_check_agi_freecount(cur, agi);
1610 * The search algorithm depends on whether we're in the same AG as the
1611 * parent. If so, find the closest available inode to the parent. If
1612 * not, consider the agi hint or find the first free inode in the AG.
1615 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1617 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1621 offset = xfs_inobt_first_free_inode(&rec);
1622 ASSERT(offset >= 0);
1623 ASSERT(offset < XFS_INODES_PER_CHUNK);
1624 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1625 XFS_INODES_PER_CHUNK) == 0);
1626 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1629 * Modify or remove the finobt record.
1631 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1633 if (rec.ir_freecount)
1634 error = xfs_inobt_update(cur, &rec);
1636 error = xfs_btree_delete(cur, &i);
1641 * The finobt has now been updated appropriately. We haven't updated the
1642 * agi and superblock yet, so we can create an inobt cursor and validate
1643 * the original freecount. If all is well, make the equivalent update to
1644 * the inobt using the finobt record and offset information.
1646 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1648 error = xfs_check_agi_freecount(icur, agi);
1652 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1657 * Both trees have now been updated. We must update the perag and
1658 * superblock before we can check the freecount for each btree.
1660 be32_add_cpu(&agi->agi_freecount, -1);
1661 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1662 agbp->b_pag->pagi_freecount--;
1664 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1666 error = xfs_check_agi_freecount(icur, agi);
1669 error = xfs_check_agi_freecount(cur, agi);
1673 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1674 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1679 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1681 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1686 * Allocate an inode on disk.
1688 * Mode is used to tell whether the new inode will need space, and whether it
1691 * This function is designed to be called twice if it has to do an allocation
1692 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1693 * If an inode is available without having to performn an allocation, an inode
1694 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1695 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1696 * The caller should then commit the current transaction, allocate a
1697 * new transaction, and call xfs_dialloc() again, passing in the previous value
1698 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1699 * buffer is locked across the two calls, the second call is guaranteed to have
1700 * a free inode available.
1702 * Once we successfully pick an inode its number is returned and the on-disk
1703 * data structures are updated. The inode itself is not read in, since doing so
1704 * would break ordering constraints with xfs_reclaim.
1708 struct xfs_trans *tp,
1711 struct xfs_buf **IO_agbp,
1714 struct xfs_mount *mp = tp->t_mountp;
1715 struct xfs_buf *agbp;
1716 xfs_agnumber_t agno;
1720 xfs_agnumber_t start_agno;
1721 struct xfs_perag *pag;
1722 struct xfs_ino_geometry *igeo = M_IGEO(mp);
1727 * If the caller passes in a pointer to the AGI buffer,
1728 * continue where we left off before. In this case, we
1729 * know that the allocation group has free inodes.
1736 * We do not have an agbp, so select an initial allocation
1737 * group for inode allocation.
1739 start_agno = xfs_ialloc_ag_select(tp, parent, mode);
1740 if (start_agno == NULLAGNUMBER) {
1746 * If we have already hit the ceiling of inode blocks then clear
1747 * okalloc so we scan all available agi structures for a free
1750 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1751 * which will sacrifice the preciseness but improve the performance.
1753 if (igeo->maxicount &&
1754 percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1755 > igeo->maxicount) {
1761 * Loop until we find an allocation group that either has free inodes
1762 * or in which we can allocate some inodes. Iterate through the
1763 * allocation groups upward, wrapping at the end.
1767 pag = xfs_perag_get(mp, agno);
1768 if (!pag->pagi_inodeok) {
1769 xfs_ialloc_next_ag(mp);
1773 if (!pag->pagi_init) {
1774 error = xfs_ialloc_pagi_init(mp, tp, agno);
1780 * Do a first racy fast path check if this AG is usable.
1782 if (!pag->pagi_freecount && !okalloc)
1786 * Then read in the AGI buffer and recheck with the AGI buffer
1789 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1793 if (pag->pagi_freecount) {
1799 goto nextag_relse_buffer;
1802 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1804 xfs_trans_brelse(tp, agbp);
1806 if (error != -ENOSPC)
1816 * We successfully allocated some inodes, return
1817 * the current context to the caller so that it
1818 * can commit the current transaction and call
1819 * us again where we left off.
1821 ASSERT(pag->pagi_freecount > 0);
1829 nextag_relse_buffer:
1830 xfs_trans_brelse(tp, agbp);
1833 if (++agno == mp->m_sb.sb_agcount)
1835 if (agno == start_agno) {
1837 return noroom ? -ENOSPC : 0;
1843 return xfs_dialloc_ag(tp, agbp, parent, inop);
1850 * Free the blocks of an inode chunk. We must consider that the inode chunk
1851 * might be sparse and only free the regions that are allocated as part of the
1855 xfs_difree_inode_chunk(
1856 struct xfs_trans *tp,
1857 xfs_agnumber_t agno,
1858 struct xfs_inobt_rec_incore *rec)
1860 struct xfs_mount *mp = tp->t_mountp;
1861 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1863 int startidx, endidx;
1865 xfs_agblock_t agbno;
1867 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1869 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1870 /* not sparse, calculate extent info directly */
1871 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
1872 M_IGEO(mp)->ialloc_blks,
1873 &XFS_RMAP_OINFO_INODES);
1877 /* holemask is only 16-bits (fits in an unsigned long) */
1878 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1879 holemask[0] = rec->ir_holemask;
1882 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1883 * holemask and convert the start/end index of each range to an extent.
1884 * We start with the start and end index both pointing at the first 0 in
1887 startidx = endidx = find_first_zero_bit(holemask,
1888 XFS_INOBT_HOLEMASK_BITS);
1889 nextbit = startidx + 1;
1890 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1891 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1894 * If the next zero bit is contiguous, update the end index of
1895 * the current range and continue.
1897 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1898 nextbit == endidx + 1) {
1904 * nextbit is not contiguous with the current end index. Convert
1905 * the current start/end to an extent and add it to the free
1908 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1909 mp->m_sb.sb_inopblock;
1910 contigblk = ((endidx - startidx + 1) *
1911 XFS_INODES_PER_HOLEMASK_BIT) /
1912 mp->m_sb.sb_inopblock;
1914 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1915 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1916 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
1917 contigblk, &XFS_RMAP_OINFO_INODES);
1919 /* reset range to current bit and carry on... */
1920 startidx = endidx = nextbit;
1929 struct xfs_mount *mp,
1930 struct xfs_trans *tp,
1931 struct xfs_buf *agbp,
1933 struct xfs_icluster *xic,
1934 struct xfs_inobt_rec_incore *orec)
1936 struct xfs_agi *agi = agbp->b_addr;
1937 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1938 struct xfs_btree_cur *cur;
1939 struct xfs_inobt_rec_incore rec;
1945 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1946 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1949 * Initialize the cursor.
1951 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1953 error = xfs_check_agi_freecount(cur, agi);
1958 * Look for the entry describing this inode.
1960 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1961 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1965 if (XFS_IS_CORRUPT(mp, i != 1)) {
1966 error = -EFSCORRUPTED;
1969 error = xfs_inobt_get_rec(cur, &rec, &i);
1971 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1975 if (XFS_IS_CORRUPT(mp, i != 1)) {
1976 error = -EFSCORRUPTED;
1980 * Get the offset in the inode chunk.
1982 off = agino - rec.ir_startino;
1983 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1984 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1986 * Mark the inode free & increment the count.
1988 rec.ir_free |= XFS_INOBT_MASK(off);
1992 * When an inode chunk is free, it becomes eligible for removal. Don't
1993 * remove the chunk if the block size is large enough for multiple inode
1994 * chunks (that might not be free).
1996 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1997 rec.ir_free == XFS_INOBT_ALL_FREE &&
1998 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1999 struct xfs_perag *pag = agbp->b_pag;
2001 xic->deleted = true;
2002 xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
2003 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
2006 * Remove the inode cluster from the AGI B+Tree, adjust the
2007 * AGI and Superblock inode counts, and mark the disk space
2008 * to be freed when the transaction is committed.
2010 ilen = rec.ir_freecount;
2011 be32_add_cpu(&agi->agi_count, -ilen);
2012 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
2013 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
2014 pag->pagi_freecount -= ilen - 1;
2015 pag->pagi_count -= ilen;
2016 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
2017 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
2019 if ((error = xfs_btree_delete(cur, &i))) {
2020 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2025 xfs_difree_inode_chunk(tp, agno, &rec);
2027 xic->deleted = false;
2029 error = xfs_inobt_update(cur, &rec);
2031 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2037 * Change the inode free counts and log the ag/sb changes.
2039 be32_add_cpu(&agi->agi_freecount, 1);
2040 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2041 agbp->b_pag->pagi_freecount++;
2042 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2045 error = xfs_check_agi_freecount(cur, agi);
2050 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2054 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2059 * Free an inode in the free inode btree.
2063 struct xfs_mount *mp,
2064 struct xfs_trans *tp,
2065 struct xfs_buf *agbp,
2067 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2069 struct xfs_agi *agi = agbp->b_addr;
2070 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
2071 struct xfs_btree_cur *cur;
2072 struct xfs_inobt_rec_incore rec;
2073 int offset = agino - ibtrec->ir_startino;
2077 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2079 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2084 * If the record does not exist in the finobt, we must have just
2085 * freed an inode in a previously fully allocated chunk. If not,
2086 * something is out of sync.
2088 if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) {
2089 error = -EFSCORRUPTED;
2093 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2095 ibtrec->ir_freecount,
2096 ibtrec->ir_free, &i);
2105 * Read and update the existing record. We could just copy the ibtrec
2106 * across here, but that would defeat the purpose of having redundant
2107 * metadata. By making the modifications independently, we can catch
2108 * corruptions that we wouldn't see if we just copied from one record
2111 error = xfs_inobt_get_rec(cur, &rec, &i);
2114 if (XFS_IS_CORRUPT(mp, i != 1)) {
2115 error = -EFSCORRUPTED;
2119 rec.ir_free |= XFS_INOBT_MASK(offset);
2122 if (XFS_IS_CORRUPT(mp,
2123 rec.ir_free != ibtrec->ir_free ||
2124 rec.ir_freecount != ibtrec->ir_freecount)) {
2125 error = -EFSCORRUPTED;
2130 * The content of inobt records should always match between the inobt
2131 * and finobt. The lifecycle of records in the finobt is different from
2132 * the inobt in that the finobt only tracks records with at least one
2133 * free inode. Hence, if all of the inodes are free and we aren't
2134 * keeping inode chunks permanently on disk, remove the record.
2135 * Otherwise, update the record with the new information.
2137 * Note that we currently can't free chunks when the block size is large
2138 * enough for multiple chunks. Leave the finobt record to remain in sync
2141 if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2142 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2143 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2144 error = xfs_btree_delete(cur, &i);
2149 error = xfs_inobt_update(cur, &rec);
2155 error = xfs_check_agi_freecount(cur, agi);
2159 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2163 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2168 * Free disk inode. Carefully avoids touching the incore inode, all
2169 * manipulations incore are the caller's responsibility.
2170 * The on-disk inode is not changed by this operation, only the
2171 * btree (free inode mask) is changed.
2175 struct xfs_trans *tp, /* transaction pointer */
2176 xfs_ino_t inode, /* inode to be freed */
2177 struct xfs_icluster *xic) /* cluster info if deleted */
2180 xfs_agblock_t agbno; /* block number containing inode */
2181 struct xfs_buf *agbp; /* buffer for allocation group header */
2182 xfs_agino_t agino; /* allocation group inode number */
2183 xfs_agnumber_t agno; /* allocation group number */
2184 int error; /* error return value */
2185 struct xfs_mount *mp; /* mount structure for filesystem */
2186 struct xfs_inobt_rec_incore rec;/* btree record */
2191 * Break up inode number into its components.
2193 agno = XFS_INO_TO_AGNO(mp, inode);
2194 if (agno >= mp->m_sb.sb_agcount) {
2195 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2196 __func__, agno, mp->m_sb.sb_agcount);
2200 agino = XFS_INO_TO_AGINO(mp, inode);
2201 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
2202 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2203 __func__, (unsigned long long)inode,
2204 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2208 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2209 if (agbno >= mp->m_sb.sb_agblocks) {
2210 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2211 __func__, agbno, mp->m_sb.sb_agblocks);
2216 * Get the allocation group header.
2218 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2220 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2226 * Fix up the inode allocation btree.
2228 error = xfs_difree_inobt(mp, tp, agbp, agino, xic, &rec);
2233 * Fix up the free inode btree.
2235 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2236 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2249 struct xfs_mount *mp,
2250 struct xfs_trans *tp,
2251 xfs_agnumber_t agno,
2253 xfs_agblock_t agbno,
2254 xfs_agblock_t *chunk_agbno,
2255 xfs_agblock_t *offset_agbno,
2258 struct xfs_inobt_rec_incore rec;
2259 struct xfs_btree_cur *cur;
2260 struct xfs_buf *agbp;
2264 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2267 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2268 __func__, error, agno);
2273 * Lookup the inode record for the given agino. If the record cannot be
2274 * found, then it's an invalid inode number and we should abort. Once
2275 * we have a record, we need to ensure it contains the inode number
2276 * we are looking up.
2278 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2279 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2282 error = xfs_inobt_get_rec(cur, &rec, &i);
2283 if (!error && i == 0)
2287 xfs_trans_brelse(tp, agbp);
2288 xfs_btree_del_cursor(cur, error);
2292 /* check that the returned record contains the required inode */
2293 if (rec.ir_startino > agino ||
2294 rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2297 /* for untrusted inodes check it is allocated first */
2298 if ((flags & XFS_IGET_UNTRUSTED) &&
2299 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2302 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2303 *offset_agbno = agbno - *chunk_agbno;
2308 * Return the location of the inode in imap, for mapping it into a buffer.
2312 xfs_mount_t *mp, /* file system mount structure */
2313 xfs_trans_t *tp, /* transaction pointer */
2314 xfs_ino_t ino, /* inode to locate */
2315 struct xfs_imap *imap, /* location map structure */
2316 uint flags) /* flags for inode btree lookup */
2318 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2319 xfs_agino_t agino; /* inode number within alloc group */
2320 xfs_agnumber_t agno; /* allocation group number */
2321 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2322 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2323 int error; /* error code */
2324 int offset; /* index of inode in its buffer */
2325 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2327 ASSERT(ino != NULLFSINO);
2330 * Split up the inode number into its parts.
2332 agno = XFS_INO_TO_AGNO(mp, ino);
2333 agino = XFS_INO_TO_AGINO(mp, ino);
2334 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2335 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2336 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2339 * Don't output diagnostic information for untrusted inodes
2340 * as they can be invalid without implying corruption.
2342 if (flags & XFS_IGET_UNTRUSTED)
2344 if (agno >= mp->m_sb.sb_agcount) {
2346 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2347 __func__, agno, mp->m_sb.sb_agcount);
2349 if (agbno >= mp->m_sb.sb_agblocks) {
2351 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2352 __func__, (unsigned long long)agbno,
2353 (unsigned long)mp->m_sb.sb_agblocks);
2355 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2357 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2359 XFS_AGINO_TO_INO(mp, agno, agino));
2367 * For bulkstat and handle lookups, we have an untrusted inode number
2368 * that we have to verify is valid. We cannot do this just by reading
2369 * the inode buffer as it may have been unlinked and removed leaving
2370 * inodes in stale state on disk. Hence we have to do a btree lookup
2371 * in all cases where an untrusted inode number is passed.
2373 if (flags & XFS_IGET_UNTRUSTED) {
2374 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2375 &chunk_agbno, &offset_agbno, flags);
2382 * If the inode cluster size is the same as the blocksize or
2383 * smaller we get to the buffer by simple arithmetics.
2385 if (M_IGEO(mp)->blocks_per_cluster == 1) {
2386 offset = XFS_INO_TO_OFFSET(mp, ino);
2387 ASSERT(offset < mp->m_sb.sb_inopblock);
2389 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2390 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2391 imap->im_boffset = (unsigned short)(offset <<
2392 mp->m_sb.sb_inodelog);
2397 * If the inode chunks are aligned then use simple maths to
2398 * find the location. Otherwise we have to do a btree
2399 * lookup to find the location.
2401 if (M_IGEO(mp)->inoalign_mask) {
2402 offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2403 chunk_agbno = agbno - offset_agbno;
2405 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2406 &chunk_agbno, &offset_agbno, flags);
2412 ASSERT(agbno >= chunk_agbno);
2413 cluster_agbno = chunk_agbno +
2414 ((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2415 M_IGEO(mp)->blocks_per_cluster);
2416 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2417 XFS_INO_TO_OFFSET(mp, ino);
2419 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2420 imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2421 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2424 * If the inode number maps to a block outside the bounds
2425 * of the file system then return NULL rather than calling
2426 * read_buf and panicing when we get an error from the
2429 if ((imap->im_blkno + imap->im_len) >
2430 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2432 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2433 __func__, (unsigned long long) imap->im_blkno,
2434 (unsigned long long) imap->im_len,
2435 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2442 * Log specified fields for the ag hdr (inode section). The growth of the agi
2443 * structure over time requires that we interpret the buffer as two logical
2444 * regions delineated by the end of the unlinked list. This is due to the size
2445 * of the hash table and its location in the middle of the agi.
2447 * For example, a request to log a field before agi_unlinked and a field after
2448 * agi_unlinked could cause us to log the entire hash table and use an excessive
2449 * amount of log space. To avoid this behavior, log the region up through
2450 * agi_unlinked in one call and the region after agi_unlinked through the end of
2451 * the structure in another.
2455 xfs_trans_t *tp, /* transaction pointer */
2456 xfs_buf_t *bp, /* allocation group header buffer */
2457 int fields) /* bitmask of fields to log */
2459 int first; /* first byte number */
2460 int last; /* last byte number */
2461 static const short offsets[] = { /* field starting offsets */
2462 /* keep in sync with bit definitions */
2463 offsetof(xfs_agi_t, agi_magicnum),
2464 offsetof(xfs_agi_t, agi_versionnum),
2465 offsetof(xfs_agi_t, agi_seqno),
2466 offsetof(xfs_agi_t, agi_length),
2467 offsetof(xfs_agi_t, agi_count),
2468 offsetof(xfs_agi_t, agi_root),
2469 offsetof(xfs_agi_t, agi_level),
2470 offsetof(xfs_agi_t, agi_freecount),
2471 offsetof(xfs_agi_t, agi_newino),
2472 offsetof(xfs_agi_t, agi_dirino),
2473 offsetof(xfs_agi_t, agi_unlinked),
2474 offsetof(xfs_agi_t, agi_free_root),
2475 offsetof(xfs_agi_t, agi_free_level),
2476 offsetof(xfs_agi_t, agi_iblocks),
2480 struct xfs_agi *agi = bp->b_addr;
2482 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2486 * Compute byte offsets for the first and last fields in the first
2487 * region and log the agi buffer. This only logs up through
2490 if (fields & XFS_AGI_ALL_BITS_R1) {
2491 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2493 xfs_trans_log_buf(tp, bp, first, last);
2497 * Mask off the bits in the first region and calculate the first and
2498 * last field offsets for any bits in the second region.
2500 fields &= ~XFS_AGI_ALL_BITS_R1;
2502 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2504 xfs_trans_log_buf(tp, bp, first, last);
2508 static xfs_failaddr_t
2512 struct xfs_mount *mp = bp->b_mount;
2513 struct xfs_agi *agi = bp->b_addr;
2516 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2517 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2518 return __this_address;
2519 if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn)))
2520 return __this_address;
2524 * Validate the magic number of the agi block.
2526 if (!xfs_verify_magic(bp, agi->agi_magicnum))
2527 return __this_address;
2528 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2529 return __this_address;
2531 if (be32_to_cpu(agi->agi_level) < 1 ||
2532 be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2533 return __this_address;
2535 if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2536 (be32_to_cpu(agi->agi_free_level) < 1 ||
2537 be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2538 return __this_address;
2541 * during growfs operations, the perag is not fully initialised,
2542 * so we can't use it for any useful checking. growfs ensures we can't
2543 * use it by using uncached buffers that don't have the perag attached
2544 * so we can detect and avoid this problem.
2546 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2547 return __this_address;
2549 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2550 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2552 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2553 return __this_address;
2560 xfs_agi_read_verify(
2563 struct xfs_mount *mp = bp->b_mount;
2566 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2567 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2568 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2570 fa = xfs_agi_verify(bp);
2571 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2572 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2577 xfs_agi_write_verify(
2580 struct xfs_mount *mp = bp->b_mount;
2581 struct xfs_buf_log_item *bip = bp->b_log_item;
2582 struct xfs_agi *agi = bp->b_addr;
2585 fa = xfs_agi_verify(bp);
2587 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2591 if (!xfs_sb_version_hascrc(&mp->m_sb))
2595 agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2596 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2599 const struct xfs_buf_ops xfs_agi_buf_ops = {
2601 .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2602 .verify_read = xfs_agi_read_verify,
2603 .verify_write = xfs_agi_write_verify,
2604 .verify_struct = xfs_agi_verify,
2608 * Read in the allocation group header (inode allocation section)
2612 struct xfs_mount *mp, /* file system mount structure */
2613 struct xfs_trans *tp, /* transaction pointer */
2614 xfs_agnumber_t agno, /* allocation group number */
2615 struct xfs_buf **bpp) /* allocation group hdr buf */
2619 trace_xfs_read_agi(mp, agno);
2621 ASSERT(agno != NULLAGNUMBER);
2622 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2623 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2624 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2628 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2630 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2635 xfs_ialloc_read_agi(
2636 struct xfs_mount *mp, /* file system mount structure */
2637 struct xfs_trans *tp, /* transaction pointer */
2638 xfs_agnumber_t agno, /* allocation group number */
2639 struct xfs_buf **bpp) /* allocation group hdr buf */
2641 struct xfs_agi *agi; /* allocation group header */
2642 struct xfs_perag *pag; /* per allocation group data */
2645 trace_xfs_ialloc_read_agi(mp, agno);
2647 error = xfs_read_agi(mp, tp, agno, bpp);
2651 agi = (*bpp)->b_addr;
2652 pag = (*bpp)->b_pag;
2653 if (!pag->pagi_init) {
2654 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2655 pag->pagi_count = be32_to_cpu(agi->agi_count);
2660 * It's possible for these to be out of sync if
2661 * we are in the middle of a forced shutdown.
2663 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2664 XFS_FORCED_SHUTDOWN(mp));
2669 * Read in the agi to initialise the per-ag data in the mount structure
2672 xfs_ialloc_pagi_init(
2673 xfs_mount_t *mp, /* file system mount structure */
2674 xfs_trans_t *tp, /* transaction pointer */
2675 xfs_agnumber_t agno) /* allocation group number */
2677 xfs_buf_t *bp = NULL;
2680 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2684 xfs_trans_brelse(tp, bp);
2688 /* Is there an inode record covering a given range of inode numbers? */
2690 xfs_ialloc_has_inode_record(
2691 struct xfs_btree_cur *cur,
2696 struct xfs_inobt_rec_incore irec;
2704 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2705 while (error == 0 && has_record) {
2706 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2707 if (error || irec.ir_startino > high)
2710 agino = irec.ir_startino;
2711 holemask = irec.ir_holemask;
2712 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2713 i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2716 if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2723 error = xfs_btree_increment(cur, 0, &has_record);
2728 /* Is there an inode record covering a given extent? */
2730 xfs_ialloc_has_inodes_at_extent(
2731 struct xfs_btree_cur *cur,
2739 low = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2740 high = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2742 return xfs_ialloc_has_inode_record(cur, low, high, exists);
2745 struct xfs_ialloc_count_inodes {
2747 xfs_agino_t freecount;
2750 /* Record inode counts across all inobt records. */
2752 xfs_ialloc_count_inodes_rec(
2753 struct xfs_btree_cur *cur,
2754 union xfs_btree_rec *rec,
2757 struct xfs_inobt_rec_incore irec;
2758 struct xfs_ialloc_count_inodes *ci = priv;
2760 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2761 ci->count += irec.ir_count;
2762 ci->freecount += irec.ir_freecount;
2767 /* Count allocated and free inodes under an inobt. */
2769 xfs_ialloc_count_inodes(
2770 struct xfs_btree_cur *cur,
2772 xfs_agino_t *freecount)
2774 struct xfs_ialloc_count_inodes ci = {0};
2777 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2778 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2783 *freecount = ci.freecount;
2788 * Initialize inode-related geometry information.
2790 * Compute the inode btree min and max levels and set maxicount.
2792 * Set the inode cluster size. This may still be overridden by the file
2793 * system block size if it is larger than the chosen cluster size.
2795 * For v5 filesystems, scale the cluster size with the inode size to keep a
2796 * constant ratio of inode per cluster buffer, but only if mkfs has set the
2797 * inode alignment value appropriately for larger cluster sizes.
2799 * Then compute the inode cluster alignment information.
2802 xfs_ialloc_setup_geometry(
2803 struct xfs_mount *mp)
2805 struct xfs_sb *sbp = &mp->m_sb;
2806 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2810 igeo->new_diflags2 = 0;
2811 if (xfs_sb_version_hasbigtime(&mp->m_sb))
2812 igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME;
2814 /* Compute inode btree geometry. */
2815 igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2816 igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
2817 igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
2818 igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2819 igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2821 igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2823 igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2825 if (sbp->sb_spino_align)
2826 igeo->ialloc_min_blks = sbp->sb_spino_align;
2828 igeo->ialloc_min_blks = igeo->ialloc_blks;
2830 /* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2831 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2832 igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2836 * Set the maximum inode count for this filesystem, being careful not
2837 * to use obviously garbage sb_inopblog/sb_inopblock values. Regular
2838 * users should never get here due to failing sb verification, but
2839 * certain users (xfs_db) need to be usable even with corrupt metadata.
2841 if (sbp->sb_imax_pct && igeo->ialloc_blks) {
2843 * Make sure the maximum inode count is a multiple
2844 * of the units we allocate inodes in.
2846 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
2847 do_div(icount, 100);
2848 do_div(icount, igeo->ialloc_blks);
2849 igeo->maxicount = XFS_FSB_TO_INO(mp,
2850 icount * igeo->ialloc_blks);
2852 igeo->maxicount = 0;
2856 * Compute the desired size of an inode cluster buffer size, which
2857 * starts at 8K and (on v5 filesystems) scales up with larger inode
2860 * Preserve the desired inode cluster size because the sparse inodes
2861 * feature uses that desired size (not the actual size) to compute the
2862 * sparse inode alignment. The mount code validates this value, so we
2863 * cannot change the behavior.
2865 igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
2866 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
2867 int new_size = igeo->inode_cluster_size_raw;
2869 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
2870 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
2871 igeo->inode_cluster_size_raw = new_size;
2874 /* Calculate inode cluster ratios. */
2875 if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
2876 igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
2877 igeo->inode_cluster_size_raw);
2879 igeo->blocks_per_cluster = 1;
2880 igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
2881 igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
2883 /* Calculate inode cluster alignment. */
2884 if (xfs_sb_version_hasalign(&mp->m_sb) &&
2885 mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
2886 igeo->cluster_align = mp->m_sb.sb_inoalignmt;
2888 igeo->cluster_align = 1;
2889 igeo->inoalign_mask = igeo->cluster_align - 1;
2890 igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
2893 * If we are using stripe alignment, check whether
2894 * the stripe unit is a multiple of the inode alignment
2896 if (mp->m_dalign && igeo->inoalign_mask &&
2897 !(mp->m_dalign & igeo->inoalign_mask))
2898 igeo->ialloc_align = mp->m_dalign;
2900 igeo->ialloc_align = 0;
2903 /* Compute the location of the root directory inode that is laid out by mkfs. */
2905 xfs_ialloc_calc_rootino(
2906 struct xfs_mount *mp,
2909 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2910 xfs_agblock_t first_bno;
2913 * Pre-calculate the geometry of AG 0. We know what it looks like
2914 * because libxfs knows how to create allocation groups now.
2916 * first_bno is the first block in which mkfs could possibly have
2917 * allocated the root directory inode, once we factor in the metadata
2918 * that mkfs formats before it. Namely, the four AG headers...
2920 first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize);
2922 /* ...the two free space btree roots... */
2925 /* ...the inode btree root... */
2928 /* ...the initial AGFL... */
2929 first_bno += xfs_alloc_min_freelist(mp, NULL);
2931 /* ...the free inode btree root... */
2932 if (xfs_sb_version_hasfinobt(&mp->m_sb))
2935 /* ...the reverse mapping btree root... */
2936 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2939 /* ...the reference count btree... */
2940 if (xfs_sb_version_hasreflink(&mp->m_sb))
2944 * ...and the log, if it is allocated in the first allocation group.
2946 * This can happen with filesystems that only have a single
2947 * allocation group, or very odd geometries created by old mkfs
2948 * versions on very small filesystems.
2950 if (mp->m_sb.sb_logstart &&
2951 XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == 0)
2952 first_bno += mp->m_sb.sb_logblocks;
2955 * Now round first_bno up to whatever allocation alignment is given
2956 * by the filesystem or was passed in.
2958 if (xfs_sb_version_hasdalign(&mp->m_sb) && igeo->ialloc_align > 0)
2959 first_bno = roundup(first_bno, sunit);
2960 else if (xfs_sb_version_hasalign(&mp->m_sb) &&
2961 mp->m_sb.sb_inoalignmt > 1)
2962 first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt);
2964 return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno));