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"
35 * Lookup a record by ino in the btree given by cur.
39 struct xfs_btree_cur *cur, /* btree cursor */
40 xfs_agino_t ino, /* starting inode of chunk */
41 xfs_lookup_t dir, /* <=, >=, == */
42 int *stat) /* success/failure */
44 cur->bc_rec.i.ir_startino = ino;
45 cur->bc_rec.i.ir_holemask = 0;
46 cur->bc_rec.i.ir_count = 0;
47 cur->bc_rec.i.ir_freecount = 0;
48 cur->bc_rec.i.ir_free = 0;
49 return xfs_btree_lookup(cur, dir, stat);
53 * Update the record referred to by cur to the value given.
54 * This either works (return 0) or gets an EFSCORRUPTED error.
56 STATIC int /* error */
58 struct xfs_btree_cur *cur, /* btree cursor */
59 xfs_inobt_rec_incore_t *irec) /* btree record */
61 union xfs_btree_rec rec;
63 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
64 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
65 rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
66 rec.inobt.ir_u.sp.ir_count = irec->ir_count;
67 rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
69 /* ir_holemask/ir_count not supported on-disk */
70 rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
72 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
73 return xfs_btree_update(cur, &rec);
76 /* Convert on-disk btree record to incore inobt record. */
78 xfs_inobt_btrec_to_irec(
80 union xfs_btree_rec *rec,
81 struct xfs_inobt_rec_incore *irec)
83 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
84 if (xfs_sb_version_hassparseinodes(&mp->m_sb)) {
85 irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
86 irec->ir_count = rec->inobt.ir_u.sp.ir_count;
87 irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
90 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
91 * values for full inode chunks.
93 irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
94 irec->ir_count = XFS_INODES_PER_CHUNK;
96 be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
98 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
102 * Get the data from the pointed-to record.
106 struct xfs_btree_cur *cur,
107 struct xfs_inobt_rec_incore *irec,
110 struct xfs_mount *mp = cur->bc_mp;
111 xfs_agnumber_t agno = cur->bc_private.a.agno;
112 union xfs_btree_rec *rec;
116 error = xfs_btree_get_rec(cur, &rec, stat);
117 if (error || *stat == 0)
120 xfs_inobt_btrec_to_irec(mp, rec, irec);
122 if (!xfs_verify_agino(mp, agno, irec->ir_startino))
124 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
125 irec->ir_count > XFS_INODES_PER_CHUNK)
127 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
130 /* if there are no holes, return the first available offset */
131 if (!xfs_inobt_issparse(irec->ir_holemask))
132 realfree = irec->ir_free;
134 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
135 if (hweight64(realfree) != irec->ir_freecount)
142 "%s Inode BTree record corruption in AG %d detected!",
143 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free", agno);
145 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
146 irec->ir_startino, irec->ir_count, irec->ir_freecount,
147 irec->ir_free, irec->ir_holemask);
148 return -EFSCORRUPTED;
152 * Insert a single inobt record. Cursor must already point to desired location.
155 xfs_inobt_insert_rec(
156 struct xfs_btree_cur *cur,
163 cur->bc_rec.i.ir_holemask = holemask;
164 cur->bc_rec.i.ir_count = count;
165 cur->bc_rec.i.ir_freecount = freecount;
166 cur->bc_rec.i.ir_free = free;
167 return xfs_btree_insert(cur, stat);
171 * Insert records describing a newly allocated inode chunk into the inobt.
175 struct xfs_mount *mp,
176 struct xfs_trans *tp,
177 struct xfs_buf *agbp,
182 struct xfs_btree_cur *cur;
183 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
184 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
189 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
191 for (thisino = newino;
192 thisino < newino + newlen;
193 thisino += XFS_INODES_PER_CHUNK) {
194 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
196 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
201 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
202 XFS_INODES_PER_CHUNK,
203 XFS_INODES_PER_CHUNK,
204 XFS_INOBT_ALL_FREE, &i);
206 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
212 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
218 * Verify that the number of free inodes in the AGI is correct.
222 xfs_check_agi_freecount(
223 struct xfs_btree_cur *cur,
226 if (cur->bc_nlevels == 1) {
227 xfs_inobt_rec_incore_t rec;
232 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
237 error = xfs_inobt_get_rec(cur, &rec, &i);
242 freecount += rec.ir_freecount;
243 error = xfs_btree_increment(cur, 0, &i);
249 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
250 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
255 #define xfs_check_agi_freecount(cur, agi) 0
259 * Initialise a new set of inodes. When called without a transaction context
260 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
261 * than logging them (which in a transaction context puts them into the AIL
262 * for writeback rather than the xfsbufd queue).
265 xfs_ialloc_inode_init(
266 struct xfs_mount *mp,
267 struct xfs_trans *tp,
268 struct list_head *buffer_list,
272 xfs_agblock_t length,
275 struct xfs_buf *fbuf;
276 struct xfs_dinode *free;
284 * Loop over the new block(s), filling in the inodes. For small block
285 * sizes, manipulate the inodes in buffers which are multiples of the
288 nbufs = length / M_IGEO(mp)->blocks_per_cluster;
291 * Figure out what version number to use in the inodes we create. If
292 * the superblock version has caught up to the one that supports the new
293 * inode format, then use the new inode version. Otherwise use the old
294 * version so that old kernels will continue to be able to use the file
297 * For v3 inodes, we also need to write the inode number into the inode,
298 * so calculate the first inode number of the chunk here as
299 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
300 * across multiple filesystem blocks (such as a cluster) and so cannot
301 * be used in the cluster buffer loop below.
303 * Further, because we are writing the inode directly into the buffer
304 * and calculating a CRC on the entire inode, we have ot log the entire
305 * inode so that the entire range the CRC covers is present in the log.
306 * That means for v3 inode we log the entire buffer rather than just the
309 if (xfs_sb_version_hascrc(&mp->m_sb)) {
311 ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
314 * log the initialisation that is about to take place as an
315 * logical operation. This means the transaction does not
316 * need to log the physical changes to the inode buffers as log
317 * recovery will know what initialisation is actually needed.
318 * Hence we only need to log the buffers as "ordered" buffers so
319 * they track in the AIL as if they were physically logged.
322 xfs_icreate_log(tp, agno, agbno, icount,
323 mp->m_sb.sb_inodesize, length, gen);
327 for (j = 0; j < nbufs; j++) {
331 d = XFS_AGB_TO_DADDR(mp, agno, agbno +
332 (j * M_IGEO(mp)->blocks_per_cluster));
333 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
335 M_IGEO(mp)->blocks_per_cluster,
340 /* Initialize the inode buffers and log them appropriately. */
341 fbuf->b_ops = &xfs_inode_buf_ops;
342 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
343 for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
344 int ioffset = i << mp->m_sb.sb_inodelog;
345 uint isize = xfs_dinode_size(version);
347 free = xfs_make_iptr(mp, fbuf, i);
348 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
349 free->di_version = version;
350 free->di_gen = cpu_to_be32(gen);
351 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
354 free->di_ino = cpu_to_be64(ino);
356 uuid_copy(&free->di_uuid,
357 &mp->m_sb.sb_meta_uuid);
358 xfs_dinode_calc_crc(mp, free);
360 /* just log the inode core */
361 xfs_trans_log_buf(tp, fbuf, ioffset,
362 ioffset + isize - 1);
368 * Mark the buffer as an inode allocation buffer so it
369 * sticks in AIL at the point of this allocation
370 * transaction. This ensures the they are on disk before
371 * the tail of the log can be moved past this
372 * transaction (i.e. by preventing relogging from moving
373 * it forward in the log).
375 xfs_trans_inode_alloc_buf(tp, fbuf);
378 * Mark the buffer as ordered so that they are
379 * not physically logged in the transaction but
380 * still tracked in the AIL as part of the
381 * transaction and pin the log appropriately.
383 xfs_trans_ordered_buf(tp, fbuf);
386 fbuf->b_flags |= XBF_DONE;
387 xfs_buf_delwri_queue(fbuf, buffer_list);
395 * Align startino and allocmask for a recently allocated sparse chunk such that
396 * they are fit for insertion (or merge) into the on-disk inode btrees.
400 * When enabled, sparse inode support increases the inode alignment from cluster
401 * size to inode chunk size. This means that the minimum range between two
402 * non-adjacent inode records in the inobt is large enough for a full inode
403 * record. This allows for cluster sized, cluster aligned block allocation
404 * without need to worry about whether the resulting inode record overlaps with
405 * another record in the tree. Without this basic rule, we would have to deal
406 * with the consequences of overlap by potentially undoing recent allocations in
407 * the inode allocation codepath.
409 * Because of this alignment rule (which is enforced on mount), there are two
410 * inobt possibilities for newly allocated sparse chunks. One is that the
411 * aligned inode record for the chunk covers a range of inodes not already
412 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
413 * other is that a record already exists at the aligned startino that considers
414 * the newly allocated range as sparse. In the latter case, record content is
415 * merged in hope that sparse inode chunks fill to full chunks over time.
418 xfs_align_sparse_ino(
419 struct xfs_mount *mp,
420 xfs_agino_t *startino,
427 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
428 mod = agbno % mp->m_sb.sb_inoalignmt;
432 /* calculate the inode offset and align startino */
433 offset = XFS_AGB_TO_AGINO(mp, mod);
437 * Since startino has been aligned down, left shift allocmask such that
438 * it continues to represent the same physical inodes relative to the
441 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
445 * Determine whether the source inode record can merge into the target. Both
446 * records must be sparse, the inode ranges must match and there must be no
447 * allocation overlap between the records.
450 __xfs_inobt_can_merge(
451 struct xfs_inobt_rec_incore *trec, /* tgt record */
452 struct xfs_inobt_rec_incore *srec) /* src record */
457 /* records must cover the same inode range */
458 if (trec->ir_startino != srec->ir_startino)
461 /* both records must be sparse */
462 if (!xfs_inobt_issparse(trec->ir_holemask) ||
463 !xfs_inobt_issparse(srec->ir_holemask))
466 /* both records must track some inodes */
467 if (!trec->ir_count || !srec->ir_count)
470 /* can't exceed capacity of a full record */
471 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
474 /* verify there is no allocation overlap */
475 talloc = xfs_inobt_irec_to_allocmask(trec);
476 salloc = xfs_inobt_irec_to_allocmask(srec);
484 * Merge the source inode record into the target. The caller must call
485 * __xfs_inobt_can_merge() to ensure the merge is valid.
488 __xfs_inobt_rec_merge(
489 struct xfs_inobt_rec_incore *trec, /* target */
490 struct xfs_inobt_rec_incore *srec) /* src */
492 ASSERT(trec->ir_startino == srec->ir_startino);
494 /* combine the counts */
495 trec->ir_count += srec->ir_count;
496 trec->ir_freecount += srec->ir_freecount;
499 * Merge the holemask and free mask. For both fields, 0 bits refer to
500 * allocated inodes. We combine the allocated ranges with bitwise AND.
502 trec->ir_holemask &= srec->ir_holemask;
503 trec->ir_free &= srec->ir_free;
507 * Insert a new sparse inode chunk into the associated inode btree. The inode
508 * record for the sparse chunk is pre-aligned to a startino that should match
509 * any pre-existing sparse inode record in the tree. This allows sparse chunks
512 * This function supports two modes of handling preexisting records depending on
513 * the merge flag. If merge is true, the provided record is merged with the
514 * existing record and updated in place. The merged record is returned in nrec.
515 * If merge is false, an existing record is replaced with the provided record.
516 * If no preexisting record exists, the provided record is always inserted.
518 * It is considered corruption if a merge is requested and not possible. Given
519 * the sparse inode alignment constraints, this should never happen.
522 xfs_inobt_insert_sprec(
523 struct xfs_mount *mp,
524 struct xfs_trans *tp,
525 struct xfs_buf *agbp,
527 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
528 bool merge) /* merge or replace */
530 struct xfs_btree_cur *cur;
531 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
532 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
535 struct xfs_inobt_rec_incore rec;
537 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
539 /* the new record is pre-aligned so we know where to look */
540 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
543 /* if nothing there, insert a new record and return */
545 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
546 nrec->ir_count, nrec->ir_freecount,
550 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
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 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
564 XFS_WANT_CORRUPTED_GOTO(mp,
565 rec.ir_startino == nrec->ir_startino,
569 * This should never fail. If we have coexisting records that
570 * cannot merge, something is seriously wrong.
572 XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
575 trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
576 rec.ir_holemask, nrec->ir_startino,
579 /* merge to nrec to output the updated record */
580 __xfs_inobt_rec_merge(nrec, &rec);
582 trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
585 error = xfs_inobt_rec_check_count(mp, nrec);
590 error = xfs_inobt_update(cur, nrec);
595 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
598 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
603 * Allocate new inodes in the allocation group specified by agbp.
604 * Return 0 for success, else error code.
608 struct xfs_trans *tp,
609 struct xfs_buf *agbp,
613 struct xfs_alloc_arg args;
616 xfs_agino_t newino; /* new first inode's number */
617 xfs_agino_t newlen; /* new number of inodes */
618 int isaligned = 0; /* inode allocation at stripe */
620 /* init. to full chunk */
621 uint16_t allocmask = (uint16_t) -1;
622 struct xfs_inobt_rec_incore rec;
623 struct xfs_perag *pag;
624 struct xfs_ino_geometry *igeo = M_IGEO(tp->t_mountp);
627 memset(&args, 0, sizeof(args));
629 args.mp = tp->t_mountp;
630 args.fsbno = NULLFSBLOCK;
631 args.oinfo = XFS_RMAP_OINFO_INODES;
634 /* randomly do sparse inode allocations */
635 if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
636 igeo->ialloc_min_blks < igeo->ialloc_blks)
637 do_sparse = prandom_u32() & 1;
641 * Locking will ensure that we don't have two callers in here
644 newlen = igeo->ialloc_inos;
645 if (igeo->maxicount &&
646 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
649 args.minlen = args.maxlen = igeo->ialloc_blks;
651 * First try to allocate inodes contiguous with the last-allocated
652 * chunk of inodes. If the filesystem is striped, this will fill
653 * an entire stripe unit with inodes.
655 agi = XFS_BUF_TO_AGI(agbp);
656 newino = be32_to_cpu(agi->agi_newino);
657 agno = be32_to_cpu(agi->agi_seqno);
658 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
662 if (likely(newino != NULLAGINO &&
663 (args.agbno < be32_to_cpu(agi->agi_length)))) {
664 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
665 args.type = XFS_ALLOCTYPE_THIS_BNO;
669 * We need to take into account alignment here to ensure that
670 * we don't modify the free list if we fail to have an exact
671 * block. If we don't have an exact match, and every oher
672 * attempt allocation attempt fails, we'll end up cancelling
673 * a dirty transaction and shutting down.
675 * For an exact allocation, alignment must be 1,
676 * however we need to take cluster alignment into account when
677 * fixing up the freelist. Use the minalignslop field to
678 * indicate that extra blocks might be required for alignment,
679 * but not to use them in the actual exact allocation.
682 args.minalignslop = igeo->cluster_align - 1;
684 /* Allow space for the inode btree to split. */
685 args.minleft = igeo->inobt_maxlevels - 1;
686 if ((error = xfs_alloc_vextent(&args)))
690 * This request might have dirtied the transaction if the AG can
691 * satisfy the request, but the exact block was not available.
692 * If the allocation did fail, subsequent requests will relax
693 * the exact agbno requirement and increase the alignment
694 * instead. It is critical that the total size of the request
695 * (len + alignment + slop) does not increase from this point
696 * on, so reset minalignslop to ensure it is not included in
697 * subsequent requests.
699 args.minalignslop = 0;
702 if (unlikely(args.fsbno == NULLFSBLOCK)) {
704 * Set the alignment for the allocation.
705 * If stripe alignment is turned on then align at stripe unit
707 * If the cluster size is smaller than a filesystem block
708 * then we're doing I/O for inodes in filesystem block size
709 * pieces, so don't need alignment anyway.
712 if (igeo->ialloc_align) {
713 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
714 args.alignment = args.mp->m_dalign;
717 args.alignment = igeo->cluster_align;
719 * Need to figure out where to allocate the inode blocks.
720 * Ideally they should be spaced out through the a.g.
721 * For now, just allocate blocks up front.
723 args.agbno = be32_to_cpu(agi->agi_root);
724 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
726 * Allocate a fixed-size extent of inodes.
728 args.type = XFS_ALLOCTYPE_NEAR_BNO;
731 * Allow space for the inode btree to split.
733 args.minleft = igeo->inobt_maxlevels - 1;
734 if ((error = xfs_alloc_vextent(&args)))
739 * If stripe alignment is turned on, then try again with cluster
742 if (isaligned && args.fsbno == NULLFSBLOCK) {
743 args.type = XFS_ALLOCTYPE_NEAR_BNO;
744 args.agbno = be32_to_cpu(agi->agi_root);
745 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
746 args.alignment = igeo->cluster_align;
747 if ((error = xfs_alloc_vextent(&args)))
752 * Finally, try a sparse allocation if the filesystem supports it and
753 * the sparse allocation length is smaller than a full chunk.
755 if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
756 igeo->ialloc_min_blks < igeo->ialloc_blks &&
757 args.fsbno == NULLFSBLOCK) {
759 args.type = XFS_ALLOCTYPE_NEAR_BNO;
760 args.agbno = be32_to_cpu(agi->agi_root);
761 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
762 args.alignment = args.mp->m_sb.sb_spino_align;
765 args.minlen = igeo->ialloc_min_blks;
766 args.maxlen = args.minlen;
769 * The inode record will be aligned to full chunk size. We must
770 * prevent sparse allocation from AG boundaries that result in
771 * invalid inode records, such as records that start at agbno 0
772 * or extend beyond the AG.
774 * Set min agbno to the first aligned, non-zero agbno and max to
775 * the last aligned agbno that is at least one full chunk from
778 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
779 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
780 args.mp->m_sb.sb_inoalignmt) -
783 error = xfs_alloc_vextent(&args);
787 newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
788 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
789 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
792 if (args.fsbno == NULLFSBLOCK) {
796 ASSERT(args.len == args.minlen);
799 * Stamp and write the inode buffers.
801 * Seed the new inode cluster with a random generation number. This
802 * prevents short-term reuse of generation numbers if a chunk is
803 * freed and then immediately reallocated. We use random numbers
804 * rather than a linear progression to prevent the next generation
805 * number from being easily guessable.
807 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
808 args.agbno, args.len, prandom_u32());
813 * Convert the results.
815 newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
817 if (xfs_inobt_issparse(~allocmask)) {
819 * We've allocated a sparse chunk. Align the startino and mask.
821 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
823 rec.ir_startino = newino;
824 rec.ir_holemask = ~allocmask;
825 rec.ir_count = newlen;
826 rec.ir_freecount = newlen;
827 rec.ir_free = XFS_INOBT_ALL_FREE;
830 * Insert the sparse record into the inobt and allow for a merge
831 * if necessary. If a merge does occur, rec is updated to the
834 error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
836 if (error == -EFSCORRUPTED) {
838 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
839 XFS_AGINO_TO_INO(args.mp, agno,
841 rec.ir_holemask, rec.ir_count);
842 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
848 * We can't merge the part we've just allocated as for the inobt
849 * due to finobt semantics. The original record may or may not
850 * exist independent of whether physical inodes exist in this
853 * We must update the finobt record based on the inobt record.
854 * rec contains the fully merged and up to date inobt record
855 * from the previous call. Set merge false to replace any
856 * existing record with this one.
858 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
859 error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
860 XFS_BTNUM_FINO, &rec,
866 /* full chunk - insert new records to both btrees */
867 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
872 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
873 error = xfs_inobt_insert(args.mp, tp, agbp, newino,
874 newlen, XFS_BTNUM_FINO);
881 * Update AGI counts and newino.
883 be32_add_cpu(&agi->agi_count, newlen);
884 be32_add_cpu(&agi->agi_freecount, newlen);
885 pag = xfs_perag_get(args.mp, agno);
886 pag->pagi_freecount += newlen;
887 pag->pagi_count += newlen;
889 agi->agi_newino = cpu_to_be32(newino);
892 * Log allocation group header fields
894 xfs_ialloc_log_agi(tp, agbp,
895 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
897 * Modify/log superblock values for inode count and inode free count.
899 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
900 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
905 STATIC xfs_agnumber_t
911 spin_lock(&mp->m_agirotor_lock);
912 agno = mp->m_agirotor;
913 if (++mp->m_agirotor >= mp->m_maxagi)
915 spin_unlock(&mp->m_agirotor_lock);
921 * Select an allocation group to look for a free inode in, based on the parent
922 * inode and the mode. Return the allocation group buffer.
924 STATIC xfs_agnumber_t
925 xfs_ialloc_ag_select(
926 xfs_trans_t *tp, /* transaction pointer */
927 xfs_ino_t parent, /* parent directory inode number */
928 umode_t mode) /* bits set to indicate file type */
930 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
931 xfs_agnumber_t agno; /* current ag number */
932 int flags; /* alloc buffer locking flags */
933 xfs_extlen_t ineed; /* blocks needed for inode allocation */
934 xfs_extlen_t longest = 0; /* longest extent available */
935 xfs_mount_t *mp; /* mount point structure */
936 int needspace; /* file mode implies space allocated */
937 xfs_perag_t *pag; /* per allocation group data */
938 xfs_agnumber_t pagno; /* parent (starting) ag number */
942 * Files of these types need at least one block if length > 0
943 * (and they won't fit in the inode, but that's hard to figure out).
945 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
947 agcount = mp->m_maxagi;
949 pagno = xfs_ialloc_next_ag(mp);
951 pagno = XFS_INO_TO_AGNO(mp, parent);
952 if (pagno >= agcount)
956 ASSERT(pagno < agcount);
959 * Loop through allocation groups, looking for one with a little
960 * free space in it. Note we don't look for free inodes, exactly.
961 * Instead, we include whether there is a need to allocate inodes
962 * to mean that blocks must be allocated for them,
963 * if none are currently free.
966 flags = XFS_ALLOC_FLAG_TRYLOCK;
968 pag = xfs_perag_get(mp, agno);
969 if (!pag->pagi_inodeok) {
970 xfs_ialloc_next_ag(mp);
974 if (!pag->pagi_init) {
975 error = xfs_ialloc_pagi_init(mp, tp, agno);
980 if (pag->pagi_freecount) {
985 if (!pag->pagf_init) {
986 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
992 * Check that there is enough free space for the file plus a
993 * chunk of inodes if we need to allocate some. If this is the
994 * first pass across the AGs, take into account the potential
995 * space needed for alignment of inode chunks when checking the
996 * longest contiguous free space in the AG - this prevents us
997 * from getting ENOSPC because we have free space larger than
998 * ialloc_blks but alignment constraints prevent us from using
1001 * If we can't find an AG with space for full alignment slack to
1002 * be taken into account, we must be near ENOSPC in all AGs.
1003 * Hence we don't include alignment for the second pass and so
1004 * if we fail allocation due to alignment issues then it is most
1005 * likely a real ENOSPC condition.
1007 ineed = M_IGEO(mp)->ialloc_min_blks;
1008 if (flags && ineed > 1)
1009 ineed += M_IGEO(mp)->cluster_align;
1010 longest = pag->pagf_longest;
1012 longest = pag->pagf_flcount > 0;
1014 if (pag->pagf_freeblks >= needspace + ineed &&
1022 * No point in iterating over the rest, if we're shutting
1025 if (XFS_FORCED_SHUTDOWN(mp))
1026 return NULLAGNUMBER;
1028 if (agno >= agcount)
1030 if (agno == pagno) {
1032 return NULLAGNUMBER;
1039 * Try to retrieve the next record to the left/right from the current one.
1042 xfs_ialloc_next_rec(
1043 struct xfs_btree_cur *cur,
1044 xfs_inobt_rec_incore_t *rec,
1052 error = xfs_btree_decrement(cur, 0, &i);
1054 error = xfs_btree_increment(cur, 0, &i);
1060 error = xfs_inobt_get_rec(cur, rec, &i);
1063 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1071 struct xfs_btree_cur *cur,
1073 xfs_inobt_rec_incore_t *rec,
1079 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1084 error = xfs_inobt_get_rec(cur, rec, &i);
1087 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1094 * Return the offset of the first free inode in the record. If the inode chunk
1095 * is sparsely allocated, we convert the record holemask to inode granularity
1096 * and mask off the unallocated regions from the inode free mask.
1099 xfs_inobt_first_free_inode(
1100 struct xfs_inobt_rec_incore *rec)
1102 xfs_inofree_t realfree;
1104 /* if there are no holes, return the first available offset */
1105 if (!xfs_inobt_issparse(rec->ir_holemask))
1106 return xfs_lowbit64(rec->ir_free);
1108 realfree = xfs_inobt_irec_to_allocmask(rec);
1109 realfree &= rec->ir_free;
1111 return xfs_lowbit64(realfree);
1115 * Allocate an inode using the inobt-only algorithm.
1118 xfs_dialloc_ag_inobt(
1119 struct xfs_trans *tp,
1120 struct xfs_buf *agbp,
1124 struct xfs_mount *mp = tp->t_mountp;
1125 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1126 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1127 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1128 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1129 struct xfs_perag *pag;
1130 struct xfs_btree_cur *cur, *tcur;
1131 struct xfs_inobt_rec_incore rec, trec;
1136 int searchdistance = 10;
1138 pag = xfs_perag_get(mp, agno);
1140 ASSERT(pag->pagi_init);
1141 ASSERT(pag->pagi_inodeok);
1142 ASSERT(pag->pagi_freecount > 0);
1145 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1147 * If pagino is 0 (this is the root inode allocation) use newino.
1148 * This must work because we've just allocated some.
1151 pagino = be32_to_cpu(agi->agi_newino);
1153 error = xfs_check_agi_freecount(cur, agi);
1158 * If in the same AG as the parent, try to get near the parent.
1160 if (pagno == agno) {
1161 int doneleft; /* done, to the left */
1162 int doneright; /* done, to the right */
1164 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1167 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1169 error = xfs_inobt_get_rec(cur, &rec, &j);
1172 XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
1174 if (rec.ir_freecount > 0) {
1176 * Found a free inode in the same chunk
1177 * as the parent, done.
1184 * In the same AG as parent, but parent's chunk is full.
1187 /* duplicate the cursor, search left & right simultaneously */
1188 error = xfs_btree_dup_cursor(cur, &tcur);
1193 * Skip to last blocks looked up if same parent inode.
1195 if (pagino != NULLAGINO &&
1196 pag->pagl_pagino == pagino &&
1197 pag->pagl_leftrec != NULLAGINO &&
1198 pag->pagl_rightrec != NULLAGINO) {
1199 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1204 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1209 /* search left with tcur, back up 1 record */
1210 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1214 /* search right with cur, go forward 1 record. */
1215 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1221 * Loop until we find an inode chunk with a free inode.
1223 while (--searchdistance > 0 && (!doneleft || !doneright)) {
1224 int useleft; /* using left inode chunk this time */
1226 /* figure out the closer block if both are valid. */
1227 if (!doneleft && !doneright) {
1229 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1230 rec.ir_startino - pagino;
1232 useleft = !doneleft;
1235 /* free inodes to the left? */
1236 if (useleft && trec.ir_freecount) {
1237 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1240 pag->pagl_leftrec = trec.ir_startino;
1241 pag->pagl_rightrec = rec.ir_startino;
1242 pag->pagl_pagino = pagino;
1247 /* free inodes to the right? */
1248 if (!useleft && rec.ir_freecount) {
1249 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1251 pag->pagl_leftrec = trec.ir_startino;
1252 pag->pagl_rightrec = rec.ir_startino;
1253 pag->pagl_pagino = pagino;
1257 /* get next record to check */
1259 error = xfs_ialloc_next_rec(tcur, &trec,
1262 error = xfs_ialloc_next_rec(cur, &rec,
1269 if (searchdistance <= 0) {
1271 * Not in range - save last search
1272 * location and allocate a new inode
1274 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1275 pag->pagl_leftrec = trec.ir_startino;
1276 pag->pagl_rightrec = rec.ir_startino;
1277 pag->pagl_pagino = pagino;
1281 * We've reached the end of the btree. because
1282 * we are only searching a small chunk of the
1283 * btree each search, there is obviously free
1284 * inodes closer to the parent inode than we
1285 * are now. restart the search again.
1287 pag->pagl_pagino = NULLAGINO;
1288 pag->pagl_leftrec = NULLAGINO;
1289 pag->pagl_rightrec = NULLAGINO;
1290 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1291 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1297 * In a different AG from the parent.
1298 * See if the most recently allocated block has any free.
1300 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1301 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1307 error = xfs_inobt_get_rec(cur, &rec, &j);
1311 if (j == 1 && rec.ir_freecount > 0) {
1313 * The last chunk allocated in the group
1314 * still has a free inode.
1322 * None left in the last group, search the whole AG
1324 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1327 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1330 error = xfs_inobt_get_rec(cur, &rec, &i);
1333 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1334 if (rec.ir_freecount > 0)
1336 error = xfs_btree_increment(cur, 0, &i);
1339 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1343 offset = xfs_inobt_first_free_inode(&rec);
1344 ASSERT(offset >= 0);
1345 ASSERT(offset < XFS_INODES_PER_CHUNK);
1346 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1347 XFS_INODES_PER_CHUNK) == 0);
1348 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1349 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1351 error = xfs_inobt_update(cur, &rec);
1354 be32_add_cpu(&agi->agi_freecount, -1);
1355 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1356 pag->pagi_freecount--;
1358 error = xfs_check_agi_freecount(cur, agi);
1362 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1363 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1368 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1370 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1376 * Use the free inode btree to allocate an inode based on distance from the
1377 * parent. Note that the provided cursor may be deleted and replaced.
1380 xfs_dialloc_ag_finobt_near(
1382 struct xfs_btree_cur **ocur,
1383 struct xfs_inobt_rec_incore *rec)
1385 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1386 struct xfs_btree_cur *rcur; /* right search cursor */
1387 struct xfs_inobt_rec_incore rrec;
1391 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1396 error = xfs_inobt_get_rec(lcur, rec, &i);
1399 XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
1402 * See if we've landed in the parent inode record. The finobt
1403 * only tracks chunks with at least one free inode, so record
1404 * existence is enough.
1406 if (pagino >= rec->ir_startino &&
1407 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1411 error = xfs_btree_dup_cursor(lcur, &rcur);
1415 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1419 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1422 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
1425 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
1426 if (i == 1 && j == 1) {
1428 * Both the left and right records are valid. Choose the closer
1429 * inode chunk to the target.
1431 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1432 (rrec.ir_startino - pagino)) {
1434 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1437 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1439 } else if (j == 1) {
1440 /* only the right record is valid */
1442 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1444 } else if (i == 1) {
1445 /* only the left record is valid */
1446 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1452 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1457 * Use the free inode btree to find a free inode based on a newino hint. If
1458 * the hint is NULL, find the first free inode in the AG.
1461 xfs_dialloc_ag_finobt_newino(
1462 struct xfs_agi *agi,
1463 struct xfs_btree_cur *cur,
1464 struct xfs_inobt_rec_incore *rec)
1469 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1470 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1475 error = xfs_inobt_get_rec(cur, rec, &i);
1478 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1484 * Find the first inode available in the AG.
1486 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1489 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1491 error = xfs_inobt_get_rec(cur, rec, &i);
1494 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1500 * Update the inobt based on a modification made to the finobt. Also ensure that
1501 * the records from both trees are equivalent post-modification.
1504 xfs_dialloc_ag_update_inobt(
1505 struct xfs_btree_cur *cur, /* inobt cursor */
1506 struct xfs_inobt_rec_incore *frec, /* finobt record */
1507 int offset) /* inode offset */
1509 struct xfs_inobt_rec_incore rec;
1513 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1516 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1518 error = xfs_inobt_get_rec(cur, &rec, &i);
1521 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1522 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1523 XFS_INODES_PER_CHUNK) == 0);
1525 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1528 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
1529 (rec.ir_freecount == frec->ir_freecount));
1531 return xfs_inobt_update(cur, &rec);
1535 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1536 * back to the inobt search algorithm.
1538 * The caller selected an AG for us, and made sure that free inodes are
1543 struct xfs_trans *tp,
1544 struct xfs_buf *agbp,
1548 struct xfs_mount *mp = tp->t_mountp;
1549 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1550 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1551 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1552 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1553 struct xfs_perag *pag;
1554 struct xfs_btree_cur *cur; /* finobt cursor */
1555 struct xfs_btree_cur *icur; /* inobt cursor */
1556 struct xfs_inobt_rec_incore rec;
1562 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1563 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1565 pag = xfs_perag_get(mp, agno);
1568 * If pagino is 0 (this is the root inode allocation) use newino.
1569 * This must work because we've just allocated some.
1572 pagino = be32_to_cpu(agi->agi_newino);
1574 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1576 error = xfs_check_agi_freecount(cur, agi);
1581 * The search algorithm depends on whether we're in the same AG as the
1582 * parent. If so, find the closest available inode to the parent. If
1583 * not, consider the agi hint or find the first free inode in the AG.
1586 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1588 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1592 offset = xfs_inobt_first_free_inode(&rec);
1593 ASSERT(offset >= 0);
1594 ASSERT(offset < XFS_INODES_PER_CHUNK);
1595 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1596 XFS_INODES_PER_CHUNK) == 0);
1597 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1600 * Modify or remove the finobt record.
1602 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1604 if (rec.ir_freecount)
1605 error = xfs_inobt_update(cur, &rec);
1607 error = xfs_btree_delete(cur, &i);
1612 * The finobt has now been updated appropriately. We haven't updated the
1613 * agi and superblock yet, so we can create an inobt cursor and validate
1614 * the original freecount. If all is well, make the equivalent update to
1615 * the inobt using the finobt record and offset information.
1617 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1619 error = xfs_check_agi_freecount(icur, agi);
1623 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1628 * Both trees have now been updated. We must update the perag and
1629 * superblock before we can check the freecount for each btree.
1631 be32_add_cpu(&agi->agi_freecount, -1);
1632 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1633 pag->pagi_freecount--;
1635 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1637 error = xfs_check_agi_freecount(icur, agi);
1640 error = xfs_check_agi_freecount(cur, agi);
1644 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1645 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1651 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1653 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1659 * Allocate an inode on disk.
1661 * Mode is used to tell whether the new inode will need space, and whether it
1664 * This function is designed to be called twice if it has to do an allocation
1665 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1666 * If an inode is available without having to performn an allocation, an inode
1667 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1668 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1669 * The caller should then commit the current transaction, allocate a
1670 * new transaction, and call xfs_dialloc() again, passing in the previous value
1671 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1672 * buffer is locked across the two calls, the second call is guaranteed to have
1673 * a free inode available.
1675 * Once we successfully pick an inode its number is returned and the on-disk
1676 * data structures are updated. The inode itself is not read in, since doing so
1677 * would break ordering constraints with xfs_reclaim.
1681 struct xfs_trans *tp,
1684 struct xfs_buf **IO_agbp,
1687 struct xfs_mount *mp = tp->t_mountp;
1688 struct xfs_buf *agbp;
1689 xfs_agnumber_t agno;
1693 xfs_agnumber_t start_agno;
1694 struct xfs_perag *pag;
1695 struct xfs_ino_geometry *igeo = M_IGEO(mp);
1700 * If the caller passes in a pointer to the AGI buffer,
1701 * continue where we left off before. In this case, we
1702 * know that the allocation group has free inodes.
1709 * We do not have an agbp, so select an initial allocation
1710 * group for inode allocation.
1712 start_agno = xfs_ialloc_ag_select(tp, parent, mode);
1713 if (start_agno == NULLAGNUMBER) {
1719 * If we have already hit the ceiling of inode blocks then clear
1720 * okalloc so we scan all available agi structures for a free
1723 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1724 * which will sacrifice the preciseness but improve the performance.
1726 if (igeo->maxicount &&
1727 percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1728 > igeo->maxicount) {
1734 * Loop until we find an allocation group that either has free inodes
1735 * or in which we can allocate some inodes. Iterate through the
1736 * allocation groups upward, wrapping at the end.
1740 pag = xfs_perag_get(mp, agno);
1741 if (!pag->pagi_inodeok) {
1742 xfs_ialloc_next_ag(mp);
1746 if (!pag->pagi_init) {
1747 error = xfs_ialloc_pagi_init(mp, tp, agno);
1753 * Do a first racy fast path check if this AG is usable.
1755 if (!pag->pagi_freecount && !okalloc)
1759 * Then read in the AGI buffer and recheck with the AGI buffer
1762 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1766 if (pag->pagi_freecount) {
1772 goto nextag_relse_buffer;
1775 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1777 xfs_trans_brelse(tp, agbp);
1779 if (error != -ENOSPC)
1789 * We successfully allocated some inodes, return
1790 * the current context to the caller so that it
1791 * can commit the current transaction and call
1792 * us again where we left off.
1794 ASSERT(pag->pagi_freecount > 0);
1802 nextag_relse_buffer:
1803 xfs_trans_brelse(tp, agbp);
1806 if (++agno == mp->m_sb.sb_agcount)
1808 if (agno == start_agno) {
1810 return noroom ? -ENOSPC : 0;
1816 return xfs_dialloc_ag(tp, agbp, parent, inop);
1823 * Free the blocks of an inode chunk. We must consider that the inode chunk
1824 * might be sparse and only free the regions that are allocated as part of the
1828 xfs_difree_inode_chunk(
1829 struct xfs_trans *tp,
1830 xfs_agnumber_t agno,
1831 struct xfs_inobt_rec_incore *rec)
1833 struct xfs_mount *mp = tp->t_mountp;
1834 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1836 int startidx, endidx;
1838 xfs_agblock_t agbno;
1840 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1842 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1843 /* not sparse, calculate extent info directly */
1844 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
1845 M_IGEO(mp)->ialloc_blks,
1846 &XFS_RMAP_OINFO_INODES);
1850 /* holemask is only 16-bits (fits in an unsigned long) */
1851 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1852 holemask[0] = rec->ir_holemask;
1855 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1856 * holemask and convert the start/end index of each range to an extent.
1857 * We start with the start and end index both pointing at the first 0 in
1860 startidx = endidx = find_first_zero_bit(holemask,
1861 XFS_INOBT_HOLEMASK_BITS);
1862 nextbit = startidx + 1;
1863 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1864 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1867 * If the next zero bit is contiguous, update the end index of
1868 * the current range and continue.
1870 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1871 nextbit == endidx + 1) {
1877 * nextbit is not contiguous with the current end index. Convert
1878 * the current start/end to an extent and add it to the free
1881 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1882 mp->m_sb.sb_inopblock;
1883 contigblk = ((endidx - startidx + 1) *
1884 XFS_INODES_PER_HOLEMASK_BIT) /
1885 mp->m_sb.sb_inopblock;
1887 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1888 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1889 xfs_bmap_add_free(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
1890 contigblk, &XFS_RMAP_OINFO_INODES);
1892 /* reset range to current bit and carry on... */
1893 startidx = endidx = nextbit;
1902 struct xfs_mount *mp,
1903 struct xfs_trans *tp,
1904 struct xfs_buf *agbp,
1906 struct xfs_icluster *xic,
1907 struct xfs_inobt_rec_incore *orec)
1909 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1910 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1911 struct xfs_perag *pag;
1912 struct xfs_btree_cur *cur;
1913 struct xfs_inobt_rec_incore rec;
1919 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1920 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1923 * Initialize the cursor.
1925 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1927 error = xfs_check_agi_freecount(cur, agi);
1932 * Look for the entry describing this inode.
1934 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1935 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1939 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1940 error = xfs_inobt_get_rec(cur, &rec, &i);
1942 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1946 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
1948 * Get the offset in the inode chunk.
1950 off = agino - rec.ir_startino;
1951 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1952 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1954 * Mark the inode free & increment the count.
1956 rec.ir_free |= XFS_INOBT_MASK(off);
1960 * When an inode chunk is free, it becomes eligible for removal. Don't
1961 * remove the chunk if the block size is large enough for multiple inode
1962 * chunks (that might not be free).
1964 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1965 rec.ir_free == XFS_INOBT_ALL_FREE &&
1966 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1967 xic->deleted = true;
1968 xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1969 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1972 * Remove the inode cluster from the AGI B+Tree, adjust the
1973 * AGI and Superblock inode counts, and mark the disk space
1974 * to be freed when the transaction is committed.
1976 ilen = rec.ir_freecount;
1977 be32_add_cpu(&agi->agi_count, -ilen);
1978 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1979 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1980 pag = xfs_perag_get(mp, agno);
1981 pag->pagi_freecount -= ilen - 1;
1982 pag->pagi_count -= ilen;
1984 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1985 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1987 if ((error = xfs_btree_delete(cur, &i))) {
1988 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1993 xfs_difree_inode_chunk(tp, agno, &rec);
1995 xic->deleted = false;
1997 error = xfs_inobt_update(cur, &rec);
1999 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2005 * Change the inode free counts and log the ag/sb changes.
2007 be32_add_cpu(&agi->agi_freecount, 1);
2008 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2009 pag = xfs_perag_get(mp, agno);
2010 pag->pagi_freecount++;
2012 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2015 error = xfs_check_agi_freecount(cur, agi);
2020 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2024 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2029 * Free an inode in the free inode btree.
2033 struct xfs_mount *mp,
2034 struct xfs_trans *tp,
2035 struct xfs_buf *agbp,
2037 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2039 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
2040 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
2041 struct xfs_btree_cur *cur;
2042 struct xfs_inobt_rec_incore rec;
2043 int offset = agino - ibtrec->ir_startino;
2047 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2049 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2054 * If the record does not exist in the finobt, we must have just
2055 * freed an inode in a previously fully allocated chunk. If not,
2056 * something is out of sync.
2058 XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
2060 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2062 ibtrec->ir_freecount,
2063 ibtrec->ir_free, &i);
2072 * Read and update the existing record. We could just copy the ibtrec
2073 * across here, but that would defeat the purpose of having redundant
2074 * metadata. By making the modifications independently, we can catch
2075 * corruptions that we wouldn't see if we just copied from one record
2078 error = xfs_inobt_get_rec(cur, &rec, &i);
2081 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
2083 rec.ir_free |= XFS_INOBT_MASK(offset);
2086 XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
2087 (rec.ir_freecount == ibtrec->ir_freecount),
2091 * The content of inobt records should always match between the inobt
2092 * and finobt. The lifecycle of records in the finobt is different from
2093 * the inobt in that the finobt only tracks records with at least one
2094 * free inode. Hence, if all of the inodes are free and we aren't
2095 * keeping inode chunks permanently on disk, remove the record.
2096 * Otherwise, update the record with the new information.
2098 * Note that we currently can't free chunks when the block size is large
2099 * enough for multiple chunks. Leave the finobt record to remain in sync
2102 if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2103 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
2104 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2105 error = xfs_btree_delete(cur, &i);
2110 error = xfs_inobt_update(cur, &rec);
2116 error = xfs_check_agi_freecount(cur, agi);
2120 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2124 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2129 * Free disk inode. Carefully avoids touching the incore inode, all
2130 * manipulations incore are the caller's responsibility.
2131 * The on-disk inode is not changed by this operation, only the
2132 * btree (free inode mask) is changed.
2136 struct xfs_trans *tp, /* transaction pointer */
2137 xfs_ino_t inode, /* inode to be freed */
2138 struct xfs_icluster *xic) /* cluster info if deleted */
2141 xfs_agblock_t agbno; /* block number containing inode */
2142 struct xfs_buf *agbp; /* buffer for allocation group header */
2143 xfs_agino_t agino; /* allocation group inode number */
2144 xfs_agnumber_t agno; /* allocation group number */
2145 int error; /* error return value */
2146 struct xfs_mount *mp; /* mount structure for filesystem */
2147 struct xfs_inobt_rec_incore rec;/* btree record */
2152 * Break up inode number into its components.
2154 agno = XFS_INO_TO_AGNO(mp, inode);
2155 if (agno >= mp->m_sb.sb_agcount) {
2156 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2157 __func__, agno, mp->m_sb.sb_agcount);
2161 agino = XFS_INO_TO_AGINO(mp, inode);
2162 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
2163 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2164 __func__, (unsigned long long)inode,
2165 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2169 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2170 if (agbno >= mp->m_sb.sb_agblocks) {
2171 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2172 __func__, agbno, mp->m_sb.sb_agblocks);
2177 * Get the allocation group header.
2179 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2181 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2187 * Fix up the inode allocation btree.
2189 error = xfs_difree_inobt(mp, tp, agbp, agino, xic, &rec);
2194 * Fix up the free inode btree.
2196 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2197 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2210 struct xfs_mount *mp,
2211 struct xfs_trans *tp,
2212 xfs_agnumber_t agno,
2214 xfs_agblock_t agbno,
2215 xfs_agblock_t *chunk_agbno,
2216 xfs_agblock_t *offset_agbno,
2219 struct xfs_inobt_rec_incore rec;
2220 struct xfs_btree_cur *cur;
2221 struct xfs_buf *agbp;
2225 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2228 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2229 __func__, error, agno);
2234 * Lookup the inode record for the given agino. If the record cannot be
2235 * found, then it's an invalid inode number and we should abort. Once
2236 * we have a record, we need to ensure it contains the inode number
2237 * we are looking up.
2239 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
2240 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2243 error = xfs_inobt_get_rec(cur, &rec, &i);
2244 if (!error && i == 0)
2248 xfs_trans_brelse(tp, agbp);
2249 xfs_btree_del_cursor(cur, error);
2253 /* check that the returned record contains the required inode */
2254 if (rec.ir_startino > agino ||
2255 rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2258 /* for untrusted inodes check it is allocated first */
2259 if ((flags & XFS_IGET_UNTRUSTED) &&
2260 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2263 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2264 *offset_agbno = agbno - *chunk_agbno;
2269 * Return the location of the inode in imap, for mapping it into a buffer.
2273 xfs_mount_t *mp, /* file system mount structure */
2274 xfs_trans_t *tp, /* transaction pointer */
2275 xfs_ino_t ino, /* inode to locate */
2276 struct xfs_imap *imap, /* location map structure */
2277 uint flags) /* flags for inode btree lookup */
2279 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2280 xfs_agino_t agino; /* inode number within alloc group */
2281 xfs_agnumber_t agno; /* allocation group number */
2282 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2283 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2284 int error; /* error code */
2285 int offset; /* index of inode in its buffer */
2286 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2288 ASSERT(ino != NULLFSINO);
2291 * Split up the inode number into its parts.
2293 agno = XFS_INO_TO_AGNO(mp, ino);
2294 agino = XFS_INO_TO_AGINO(mp, ino);
2295 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2296 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2297 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2300 * Don't output diagnostic information for untrusted inodes
2301 * as they can be invalid without implying corruption.
2303 if (flags & XFS_IGET_UNTRUSTED)
2305 if (agno >= mp->m_sb.sb_agcount) {
2307 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2308 __func__, agno, mp->m_sb.sb_agcount);
2310 if (agbno >= mp->m_sb.sb_agblocks) {
2312 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2313 __func__, (unsigned long long)agbno,
2314 (unsigned long)mp->m_sb.sb_agblocks);
2316 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2318 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2320 XFS_AGINO_TO_INO(mp, agno, agino));
2328 * For bulkstat and handle lookups, we have an untrusted inode number
2329 * that we have to verify is valid. We cannot do this just by reading
2330 * the inode buffer as it may have been unlinked and removed leaving
2331 * inodes in stale state on disk. Hence we have to do a btree lookup
2332 * in all cases where an untrusted inode number is passed.
2334 if (flags & XFS_IGET_UNTRUSTED) {
2335 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2336 &chunk_agbno, &offset_agbno, flags);
2343 * If the inode cluster size is the same as the blocksize or
2344 * smaller we get to the buffer by simple arithmetics.
2346 if (M_IGEO(mp)->blocks_per_cluster == 1) {
2347 offset = XFS_INO_TO_OFFSET(mp, ino);
2348 ASSERT(offset < mp->m_sb.sb_inopblock);
2350 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2351 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2352 imap->im_boffset = (unsigned short)(offset <<
2353 mp->m_sb.sb_inodelog);
2358 * If the inode chunks are aligned then use simple maths to
2359 * find the location. Otherwise we have to do a btree
2360 * lookup to find the location.
2362 if (M_IGEO(mp)->inoalign_mask) {
2363 offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2364 chunk_agbno = agbno - offset_agbno;
2366 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2367 &chunk_agbno, &offset_agbno, flags);
2373 ASSERT(agbno >= chunk_agbno);
2374 cluster_agbno = chunk_agbno +
2375 ((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2376 M_IGEO(mp)->blocks_per_cluster);
2377 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2378 XFS_INO_TO_OFFSET(mp, ino);
2380 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2381 imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2382 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2385 * If the inode number maps to a block outside the bounds
2386 * of the file system then return NULL rather than calling
2387 * read_buf and panicing when we get an error from the
2390 if ((imap->im_blkno + imap->im_len) >
2391 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2393 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2394 __func__, (unsigned long long) imap->im_blkno,
2395 (unsigned long long) imap->im_len,
2396 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2403 * Log specified fields for the ag hdr (inode section). The growth of the agi
2404 * structure over time requires that we interpret the buffer as two logical
2405 * regions delineated by the end of the unlinked list. This is due to the size
2406 * of the hash table and its location in the middle of the agi.
2408 * For example, a request to log a field before agi_unlinked and a field after
2409 * agi_unlinked could cause us to log the entire hash table and use an excessive
2410 * amount of log space. To avoid this behavior, log the region up through
2411 * agi_unlinked in one call and the region after agi_unlinked through the end of
2412 * the structure in another.
2416 xfs_trans_t *tp, /* transaction pointer */
2417 xfs_buf_t *bp, /* allocation group header buffer */
2418 int fields) /* bitmask of fields to log */
2420 int first; /* first byte number */
2421 int last; /* last byte number */
2422 static const short offsets[] = { /* field starting offsets */
2423 /* keep in sync with bit definitions */
2424 offsetof(xfs_agi_t, agi_magicnum),
2425 offsetof(xfs_agi_t, agi_versionnum),
2426 offsetof(xfs_agi_t, agi_seqno),
2427 offsetof(xfs_agi_t, agi_length),
2428 offsetof(xfs_agi_t, agi_count),
2429 offsetof(xfs_agi_t, agi_root),
2430 offsetof(xfs_agi_t, agi_level),
2431 offsetof(xfs_agi_t, agi_freecount),
2432 offsetof(xfs_agi_t, agi_newino),
2433 offsetof(xfs_agi_t, agi_dirino),
2434 offsetof(xfs_agi_t, agi_unlinked),
2435 offsetof(xfs_agi_t, agi_free_root),
2436 offsetof(xfs_agi_t, agi_free_level),
2440 xfs_agi_t *agi; /* allocation group header */
2442 agi = XFS_BUF_TO_AGI(bp);
2443 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2447 * Compute byte offsets for the first and last fields in the first
2448 * region and log the agi buffer. This only logs up through
2451 if (fields & XFS_AGI_ALL_BITS_R1) {
2452 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2454 xfs_trans_log_buf(tp, bp, first, last);
2458 * Mask off the bits in the first region and calculate the first and
2459 * last field offsets for any bits in the second region.
2461 fields &= ~XFS_AGI_ALL_BITS_R1;
2463 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2465 xfs_trans_log_buf(tp, bp, first, last);
2469 static xfs_failaddr_t
2473 struct xfs_mount *mp = bp->b_target->bt_mount;
2474 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
2477 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2478 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2479 return __this_address;
2480 if (!xfs_log_check_lsn(mp,
2481 be64_to_cpu(XFS_BUF_TO_AGI(bp)->agi_lsn)))
2482 return __this_address;
2486 * Validate the magic number of the agi block.
2488 if (!xfs_verify_magic(bp, agi->agi_magicnum))
2489 return __this_address;
2490 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2491 return __this_address;
2493 if (be32_to_cpu(agi->agi_level) < 1 ||
2494 be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2495 return __this_address;
2497 if (xfs_sb_version_hasfinobt(&mp->m_sb) &&
2498 (be32_to_cpu(agi->agi_free_level) < 1 ||
2499 be32_to_cpu(agi->agi_free_level) > XFS_BTREE_MAXLEVELS))
2500 return __this_address;
2503 * during growfs operations, the perag is not fully initialised,
2504 * so we can't use it for any useful checking. growfs ensures we can't
2505 * use it by using uncached buffers that don't have the perag attached
2506 * so we can detect and avoid this problem.
2508 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2509 return __this_address;
2511 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2512 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2514 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2515 return __this_address;
2522 xfs_agi_read_verify(
2525 struct xfs_mount *mp = bp->b_target->bt_mount;
2528 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2529 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2530 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2532 fa = xfs_agi_verify(bp);
2533 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2534 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2539 xfs_agi_write_verify(
2542 struct xfs_mount *mp = bp->b_target->bt_mount;
2543 struct xfs_buf_log_item *bip = bp->b_log_item;
2546 fa = xfs_agi_verify(bp);
2548 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2552 if (!xfs_sb_version_hascrc(&mp->m_sb))
2556 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2557 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2560 const struct xfs_buf_ops xfs_agi_buf_ops = {
2562 .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2563 .verify_read = xfs_agi_read_verify,
2564 .verify_write = xfs_agi_write_verify,
2565 .verify_struct = xfs_agi_verify,
2569 * Read in the allocation group header (inode allocation section)
2573 struct xfs_mount *mp, /* file system mount structure */
2574 struct xfs_trans *tp, /* transaction pointer */
2575 xfs_agnumber_t agno, /* allocation group number */
2576 struct xfs_buf **bpp) /* allocation group hdr buf */
2580 trace_xfs_read_agi(mp, agno);
2582 ASSERT(agno != NULLAGNUMBER);
2583 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2584 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
2585 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
2589 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_AGI_BUF);
2591 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
2596 xfs_ialloc_read_agi(
2597 struct xfs_mount *mp, /* file system mount structure */
2598 struct xfs_trans *tp, /* transaction pointer */
2599 xfs_agnumber_t agno, /* allocation group number */
2600 struct xfs_buf **bpp) /* allocation group hdr buf */
2602 struct xfs_agi *agi; /* allocation group header */
2603 struct xfs_perag *pag; /* per allocation group data */
2606 trace_xfs_ialloc_read_agi(mp, agno);
2608 error = xfs_read_agi(mp, tp, agno, bpp);
2612 agi = XFS_BUF_TO_AGI(*bpp);
2613 pag = xfs_perag_get(mp, agno);
2614 if (!pag->pagi_init) {
2615 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2616 pag->pagi_count = be32_to_cpu(agi->agi_count);
2621 * It's possible for these to be out of sync if
2622 * we are in the middle of a forced shutdown.
2624 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2625 XFS_FORCED_SHUTDOWN(mp));
2631 * Read in the agi to initialise the per-ag data in the mount structure
2634 xfs_ialloc_pagi_init(
2635 xfs_mount_t *mp, /* file system mount structure */
2636 xfs_trans_t *tp, /* transaction pointer */
2637 xfs_agnumber_t agno) /* allocation group number */
2639 xfs_buf_t *bp = NULL;
2642 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2646 xfs_trans_brelse(tp, bp);
2650 /* Is there an inode record covering a given range of inode numbers? */
2652 xfs_ialloc_has_inode_record(
2653 struct xfs_btree_cur *cur,
2658 struct xfs_inobt_rec_incore irec;
2666 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2667 while (error == 0 && has_record) {
2668 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2669 if (error || irec.ir_startino > high)
2672 agino = irec.ir_startino;
2673 holemask = irec.ir_holemask;
2674 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2675 i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2678 if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2685 error = xfs_btree_increment(cur, 0, &has_record);
2690 /* Is there an inode record covering a given extent? */
2692 xfs_ialloc_has_inodes_at_extent(
2693 struct xfs_btree_cur *cur,
2701 low = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2702 high = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2704 return xfs_ialloc_has_inode_record(cur, low, high, exists);
2707 struct xfs_ialloc_count_inodes {
2709 xfs_agino_t freecount;
2712 /* Record inode counts across all inobt records. */
2714 xfs_ialloc_count_inodes_rec(
2715 struct xfs_btree_cur *cur,
2716 union xfs_btree_rec *rec,
2719 struct xfs_inobt_rec_incore irec;
2720 struct xfs_ialloc_count_inodes *ci = priv;
2722 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2723 ci->count += irec.ir_count;
2724 ci->freecount += irec.ir_freecount;
2729 /* Count allocated and free inodes under an inobt. */
2731 xfs_ialloc_count_inodes(
2732 struct xfs_btree_cur *cur,
2734 xfs_agino_t *freecount)
2736 struct xfs_ialloc_count_inodes ci = {0};
2739 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2740 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2745 *freecount = ci.freecount;
2750 * Initialize inode-related geometry information.
2752 * Compute the inode btree min and max levels and set maxicount.
2754 * Set the inode cluster size. This may still be overridden by the file
2755 * system block size if it is larger than the chosen cluster size.
2757 * For v5 filesystems, scale the cluster size with the inode size to keep a
2758 * constant ratio of inode per cluster buffer, but only if mkfs has set the
2759 * inode alignment value appropriately for larger cluster sizes.
2761 * Then compute the inode cluster alignment information.
2764 xfs_ialloc_setup_geometry(
2765 struct xfs_mount *mp)
2767 struct xfs_sb *sbp = &mp->m_sb;
2768 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2772 /* Compute inode btree geometry. */
2773 igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2774 igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
2775 igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
2776 igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2777 igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2779 igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2781 igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2783 if (sbp->sb_spino_align)
2784 igeo->ialloc_min_blks = sbp->sb_spino_align;
2786 igeo->ialloc_min_blks = igeo->ialloc_blks;
2788 /* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2789 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2790 igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2793 /* Set the maximum inode count for this filesystem. */
2794 if (sbp->sb_imax_pct) {
2796 * Make sure the maximum inode count is a multiple
2797 * of the units we allocate inodes in.
2799 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
2800 do_div(icount, 100);
2801 do_div(icount, igeo->ialloc_blks);
2802 igeo->maxicount = XFS_FSB_TO_INO(mp,
2803 icount * igeo->ialloc_blks);
2805 igeo->maxicount = 0;
2809 * Compute the desired size of an inode cluster buffer size, which
2810 * starts at 8K and (on v5 filesystems) scales up with larger inode
2813 * Preserve the desired inode cluster size because the sparse inodes
2814 * feature uses that desired size (not the actual size) to compute the
2815 * sparse inode alignment. The mount code validates this value, so we
2816 * cannot change the behavior.
2818 igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
2819 if (xfs_sb_version_hascrc(&mp->m_sb)) {
2820 int new_size = igeo->inode_cluster_size_raw;
2822 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
2823 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
2824 igeo->inode_cluster_size_raw = new_size;
2827 /* Calculate inode cluster ratios. */
2828 if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
2829 igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
2830 igeo->inode_cluster_size_raw);
2832 igeo->blocks_per_cluster = 1;
2833 igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
2834 igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
2836 /* Calculate inode cluster alignment. */
2837 if (xfs_sb_version_hasalign(&mp->m_sb) &&
2838 mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
2839 igeo->cluster_align = mp->m_sb.sb_inoalignmt;
2841 igeo->cluster_align = 1;
2842 igeo->inoalign_mask = igeo->cluster_align - 1;
2843 igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
2846 * If we are using stripe alignment, check whether
2847 * the stripe unit is a multiple of the inode alignment
2849 if (mp->m_dalign && igeo->inoalign_mask &&
2850 !(mp->m_dalign & igeo->inoalign_mask))
2851 igeo->ialloc_align = mp->m_dalign;
2853 igeo->ialloc_align = 0;