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"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_ialloc_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
22 #include "xfs_trans.h"
23 #include "xfs_buf_item.h"
24 #include "xfs_icreate_item.h"
25 #include "xfs_icache.h"
26 #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_has_sparseinodes(cur->bc_mp)) {
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 const 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_has_sparseinodes(mp)) {
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);
98 /* Simple checks for inode records. */
100 xfs_inobt_check_irec(
101 struct xfs_btree_cur *cur,
102 const struct xfs_inobt_rec_incore *irec)
106 /* Record has to be properly aligned within the AG. */
107 if (!xfs_verify_agino(cur->bc_ag.pag, irec->ir_startino))
108 return __this_address;
109 if (!xfs_verify_agino(cur->bc_ag.pag,
110 irec->ir_startino + XFS_INODES_PER_CHUNK - 1))
111 return __this_address;
112 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
113 irec->ir_count > XFS_INODES_PER_CHUNK)
114 return __this_address;
115 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
116 return __this_address;
118 /* if there are no holes, return the first available offset */
119 if (!xfs_inobt_issparse(irec->ir_holemask))
120 realfree = irec->ir_free;
122 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
123 if (hweight64(realfree) != irec->ir_freecount)
124 return __this_address;
130 xfs_inobt_complain_bad_rec(
131 struct xfs_btree_cur *cur,
133 const struct xfs_inobt_rec_incore *irec)
135 struct xfs_mount *mp = cur->bc_mp;
138 "%s Inode BTree record corruption in AG %d detected at %pS!",
139 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free",
140 cur->bc_ag.pag->pag_agno, fa);
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 * Get the data from the pointed-to record.
153 struct xfs_btree_cur *cur,
154 struct xfs_inobt_rec_incore *irec,
157 struct xfs_mount *mp = cur->bc_mp;
158 union xfs_btree_rec *rec;
162 error = xfs_btree_get_rec(cur, &rec, stat);
163 if (error || *stat == 0)
166 xfs_inobt_btrec_to_irec(mp, rec, irec);
167 fa = xfs_inobt_check_irec(cur, irec);
169 return xfs_inobt_complain_bad_rec(cur, fa, irec);
175 * Insert a single inobt record. Cursor must already point to desired location.
178 xfs_inobt_insert_rec(
179 struct xfs_btree_cur *cur,
186 cur->bc_rec.i.ir_holemask = holemask;
187 cur->bc_rec.i.ir_count = count;
188 cur->bc_rec.i.ir_freecount = freecount;
189 cur->bc_rec.i.ir_free = free;
190 return xfs_btree_insert(cur, stat);
194 * Insert records describing a newly allocated inode chunk into the inobt.
198 struct xfs_perag *pag,
199 struct xfs_trans *tp,
200 struct xfs_buf *agbp,
205 struct xfs_btree_cur *cur;
210 cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
212 for (thisino = newino;
213 thisino < newino + newlen;
214 thisino += XFS_INODES_PER_CHUNK) {
215 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
217 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
222 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
223 XFS_INODES_PER_CHUNK,
224 XFS_INODES_PER_CHUNK,
225 XFS_INOBT_ALL_FREE, &i);
227 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
233 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
239 * Verify that the number of free inodes in the AGI is correct.
243 xfs_check_agi_freecount(
244 struct xfs_btree_cur *cur)
246 if (cur->bc_nlevels == 1) {
247 xfs_inobt_rec_incore_t rec;
252 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
257 error = xfs_inobt_get_rec(cur, &rec, &i);
262 freecount += rec.ir_freecount;
263 error = xfs_btree_increment(cur, 0, &i);
269 if (!xfs_is_shutdown(cur->bc_mp))
270 ASSERT(freecount == cur->bc_ag.pag->pagi_freecount);
275 #define xfs_check_agi_freecount(cur) 0
279 * Initialise a new set of inodes. When called without a transaction context
280 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
281 * than logging them (which in a transaction context puts them into the AIL
282 * for writeback rather than the xfsbufd queue).
285 xfs_ialloc_inode_init(
286 struct xfs_mount *mp,
287 struct xfs_trans *tp,
288 struct list_head *buffer_list,
292 xfs_agblock_t length,
295 struct xfs_buf *fbuf;
296 struct xfs_dinode *free;
305 * Loop over the new block(s), filling in the inodes. For small block
306 * sizes, manipulate the inodes in buffers which are multiples of the
309 nbufs = length / M_IGEO(mp)->blocks_per_cluster;
312 * Figure out what version number to use in the inodes we create. If
313 * the superblock version has caught up to the one that supports the new
314 * inode format, then use the new inode version. Otherwise use the old
315 * version so that old kernels will continue to be able to use the file
318 * For v3 inodes, we also need to write the inode number into the inode,
319 * so calculate the first inode number of the chunk here as
320 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
321 * across multiple filesystem blocks (such as a cluster) and so cannot
322 * be used in the cluster buffer loop below.
324 * Further, because we are writing the inode directly into the buffer
325 * and calculating a CRC on the entire inode, we have ot log the entire
326 * inode so that the entire range the CRC covers is present in the log.
327 * That means for v3 inode we log the entire buffer rather than just the
330 if (xfs_has_v3inodes(mp)) {
332 ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
335 * log the initialisation that is about to take place as an
336 * logical operation. This means the transaction does not
337 * need to log the physical changes to the inode buffers as log
338 * recovery will know what initialisation is actually needed.
339 * Hence we only need to log the buffers as "ordered" buffers so
340 * they track in the AIL as if they were physically logged.
343 xfs_icreate_log(tp, agno, agbno, icount,
344 mp->m_sb.sb_inodesize, length, gen);
348 for (j = 0; j < nbufs; j++) {
352 d = XFS_AGB_TO_DADDR(mp, agno, agbno +
353 (j * M_IGEO(mp)->blocks_per_cluster));
354 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
355 mp->m_bsize * M_IGEO(mp)->blocks_per_cluster,
356 XBF_UNMAPPED, &fbuf);
360 /* Initialize the inode buffers and log them appropriately. */
361 fbuf->b_ops = &xfs_inode_buf_ops;
362 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
363 for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
364 int ioffset = i << mp->m_sb.sb_inodelog;
366 free = xfs_make_iptr(mp, fbuf, i);
367 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
368 free->di_version = version;
369 free->di_gen = cpu_to_be32(gen);
370 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
373 free->di_ino = cpu_to_be64(ino);
375 uuid_copy(&free->di_uuid,
376 &mp->m_sb.sb_meta_uuid);
377 xfs_dinode_calc_crc(mp, free);
379 /* just log the inode core */
380 xfs_trans_log_buf(tp, fbuf, ioffset,
381 ioffset + XFS_DINODE_SIZE(mp) - 1);
387 * Mark the buffer as an inode allocation buffer so it
388 * sticks in AIL at the point of this allocation
389 * transaction. This ensures the they are on disk before
390 * the tail of the log can be moved past this
391 * transaction (i.e. by preventing relogging from moving
392 * it forward in the log).
394 xfs_trans_inode_alloc_buf(tp, fbuf);
397 * Mark the buffer as ordered so that they are
398 * not physically logged in the transaction but
399 * still tracked in the AIL as part of the
400 * transaction and pin the log appropriately.
402 xfs_trans_ordered_buf(tp, fbuf);
405 fbuf->b_flags |= XBF_DONE;
406 xfs_buf_delwri_queue(fbuf, buffer_list);
414 * Align startino and allocmask for a recently allocated sparse chunk such that
415 * they are fit for insertion (or merge) into the on-disk inode btrees.
419 * When enabled, sparse inode support increases the inode alignment from cluster
420 * size to inode chunk size. This means that the minimum range between two
421 * non-adjacent inode records in the inobt is large enough for a full inode
422 * record. This allows for cluster sized, cluster aligned block allocation
423 * without need to worry about whether the resulting inode record overlaps with
424 * another record in the tree. Without this basic rule, we would have to deal
425 * with the consequences of overlap by potentially undoing recent allocations in
426 * the inode allocation codepath.
428 * Because of this alignment rule (which is enforced on mount), there are two
429 * inobt possibilities for newly allocated sparse chunks. One is that the
430 * aligned inode record for the chunk covers a range of inodes not already
431 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
432 * other is that a record already exists at the aligned startino that considers
433 * the newly allocated range as sparse. In the latter case, record content is
434 * merged in hope that sparse inode chunks fill to full chunks over time.
437 xfs_align_sparse_ino(
438 struct xfs_mount *mp,
439 xfs_agino_t *startino,
446 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
447 mod = agbno % mp->m_sb.sb_inoalignmt;
451 /* calculate the inode offset and align startino */
452 offset = XFS_AGB_TO_AGINO(mp, mod);
456 * Since startino has been aligned down, left shift allocmask such that
457 * it continues to represent the same physical inodes relative to the
460 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
464 * Determine whether the source inode record can merge into the target. Both
465 * records must be sparse, the inode ranges must match and there must be no
466 * allocation overlap between the records.
469 __xfs_inobt_can_merge(
470 struct xfs_inobt_rec_incore *trec, /* tgt record */
471 struct xfs_inobt_rec_incore *srec) /* src record */
476 /* records must cover the same inode range */
477 if (trec->ir_startino != srec->ir_startino)
480 /* both records must be sparse */
481 if (!xfs_inobt_issparse(trec->ir_holemask) ||
482 !xfs_inobt_issparse(srec->ir_holemask))
485 /* both records must track some inodes */
486 if (!trec->ir_count || !srec->ir_count)
489 /* can't exceed capacity of a full record */
490 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
493 /* verify there is no allocation overlap */
494 talloc = xfs_inobt_irec_to_allocmask(trec);
495 salloc = xfs_inobt_irec_to_allocmask(srec);
503 * Merge the source inode record into the target. The caller must call
504 * __xfs_inobt_can_merge() to ensure the merge is valid.
507 __xfs_inobt_rec_merge(
508 struct xfs_inobt_rec_incore *trec, /* target */
509 struct xfs_inobt_rec_incore *srec) /* src */
511 ASSERT(trec->ir_startino == srec->ir_startino);
513 /* combine the counts */
514 trec->ir_count += srec->ir_count;
515 trec->ir_freecount += srec->ir_freecount;
518 * Merge the holemask and free mask. For both fields, 0 bits refer to
519 * allocated inodes. We combine the allocated ranges with bitwise AND.
521 trec->ir_holemask &= srec->ir_holemask;
522 trec->ir_free &= srec->ir_free;
526 * Insert a new sparse inode chunk into the associated inode btree. The inode
527 * record for the sparse chunk is pre-aligned to a startino that should match
528 * any pre-existing sparse inode record in the tree. This allows sparse chunks
531 * This function supports two modes of handling preexisting records depending on
532 * the merge flag. If merge is true, the provided record is merged with the
533 * existing record and updated in place. The merged record is returned in nrec.
534 * If merge is false, an existing record is replaced with the provided record.
535 * If no preexisting record exists, the provided record is always inserted.
537 * It is considered corruption if a merge is requested and not possible. Given
538 * the sparse inode alignment constraints, this should never happen.
541 xfs_inobt_insert_sprec(
542 struct xfs_perag *pag,
543 struct xfs_trans *tp,
544 struct xfs_buf *agbp,
546 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
547 bool merge) /* merge or replace */
549 struct xfs_mount *mp = pag->pag_mount;
550 struct xfs_btree_cur *cur;
553 struct xfs_inobt_rec_incore rec;
555 cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
557 /* the new record is pre-aligned so we know where to look */
558 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
561 /* if nothing there, insert a new record and return */
563 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
564 nrec->ir_count, nrec->ir_freecount,
568 if (XFS_IS_CORRUPT(mp, i != 1)) {
569 error = -EFSCORRUPTED;
577 * A record exists at this startino. Merge or replace the record
578 * depending on what we've been asked to do.
581 error = xfs_inobt_get_rec(cur, &rec, &i);
584 if (XFS_IS_CORRUPT(mp, i != 1)) {
585 error = -EFSCORRUPTED;
588 if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) {
589 error = -EFSCORRUPTED;
594 * This should never fail. If we have coexisting records that
595 * cannot merge, something is seriously wrong.
597 if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) {
598 error = -EFSCORRUPTED;
602 trace_xfs_irec_merge_pre(mp, pag->pag_agno, rec.ir_startino,
603 rec.ir_holemask, nrec->ir_startino,
606 /* merge to nrec to output the updated record */
607 __xfs_inobt_rec_merge(nrec, &rec);
609 trace_xfs_irec_merge_post(mp, pag->pag_agno, nrec->ir_startino,
612 error = xfs_inobt_rec_check_count(mp, nrec);
617 error = xfs_inobt_update(cur, nrec);
622 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
625 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
630 * Allocate new inodes in the allocation group specified by agbp. Returns 0 if
631 * inodes were allocated in this AG; -EAGAIN if there was no space in this AG so
632 * the caller knows it can try another AG, a hard -ENOSPC when over the maximum
633 * inode count threshold, or the usual negative error code for other errors.
637 struct xfs_perag *pag,
638 struct xfs_trans *tp,
639 struct xfs_buf *agbp)
642 struct xfs_alloc_arg args;
644 xfs_agino_t newino; /* new first inode's number */
645 xfs_agino_t newlen; /* new number of inodes */
646 int isaligned = 0; /* inode allocation at stripe */
648 /* init. to full chunk */
649 struct xfs_inobt_rec_incore rec;
650 struct xfs_ino_geometry *igeo = M_IGEO(tp->t_mountp);
651 uint16_t allocmask = (uint16_t) -1;
654 memset(&args, 0, sizeof(args));
656 args.mp = tp->t_mountp;
657 args.fsbno = NULLFSBLOCK;
658 args.oinfo = XFS_RMAP_OINFO_INODES;
662 /* randomly do sparse inode allocations */
663 if (xfs_has_sparseinodes(tp->t_mountp) &&
664 igeo->ialloc_min_blks < igeo->ialloc_blks)
665 do_sparse = get_random_u32_below(2);
669 * Locking will ensure that we don't have two callers in here
672 newlen = igeo->ialloc_inos;
673 if (igeo->maxicount &&
674 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
677 args.minlen = args.maxlen = igeo->ialloc_blks;
679 * First try to allocate inodes contiguous with the last-allocated
680 * chunk of inodes. If the filesystem is striped, this will fill
681 * an entire stripe unit with inodes.
684 newino = be32_to_cpu(agi->agi_newino);
685 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
689 if (likely(newino != NULLAGINO &&
690 (args.agbno < be32_to_cpu(agi->agi_length)))) {
694 * We need to take into account alignment here to ensure that
695 * we don't modify the free list if we fail to have an exact
696 * block. If we don't have an exact match, and every oher
697 * attempt allocation attempt fails, we'll end up cancelling
698 * a dirty transaction and shutting down.
700 * For an exact allocation, alignment must be 1,
701 * however we need to take cluster alignment into account when
702 * fixing up the freelist. Use the minalignslop field to
703 * indicate that extra blocks might be required for alignment,
704 * but not to use them in the actual exact allocation.
707 args.minalignslop = igeo->cluster_align - 1;
709 /* Allow space for the inode btree to split. */
710 args.minleft = igeo->inobt_maxlevels;
711 error = xfs_alloc_vextent_exact_bno(&args,
712 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
718 * This request might have dirtied the transaction if the AG can
719 * satisfy the request, but the exact block was not available.
720 * If the allocation did fail, subsequent requests will relax
721 * the exact agbno requirement and increase the alignment
722 * instead. It is critical that the total size of the request
723 * (len + alignment + slop) does not increase from this point
724 * on, so reset minalignslop to ensure it is not included in
725 * subsequent requests.
727 args.minalignslop = 0;
730 if (unlikely(args.fsbno == NULLFSBLOCK)) {
732 * Set the alignment for the allocation.
733 * If stripe alignment is turned on then align at stripe unit
735 * If the cluster size is smaller than a filesystem block
736 * then we're doing I/O for inodes in filesystem block size
737 * pieces, so don't need alignment anyway.
740 if (igeo->ialloc_align) {
741 ASSERT(!xfs_has_noalign(args.mp));
742 args.alignment = args.mp->m_dalign;
745 args.alignment = igeo->cluster_align;
747 * Allocate a fixed-size extent of inodes.
751 * Allow space for the inode btree to split.
753 args.minleft = igeo->inobt_maxlevels;
754 error = xfs_alloc_vextent_near_bno(&args,
755 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
756 be32_to_cpu(agi->agi_root)));
762 * If stripe alignment is turned on, then try again with cluster
765 if (isaligned && args.fsbno == NULLFSBLOCK) {
766 args.alignment = igeo->cluster_align;
767 error = xfs_alloc_vextent_near_bno(&args,
768 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
769 be32_to_cpu(agi->agi_root)));
775 * Finally, try a sparse allocation if the filesystem supports it and
776 * the sparse allocation length is smaller than a full chunk.
778 if (xfs_has_sparseinodes(args.mp) &&
779 igeo->ialloc_min_blks < igeo->ialloc_blks &&
780 args.fsbno == NULLFSBLOCK) {
782 args.alignment = args.mp->m_sb.sb_spino_align;
785 args.minlen = igeo->ialloc_min_blks;
786 args.maxlen = args.minlen;
789 * The inode record will be aligned to full chunk size. We must
790 * prevent sparse allocation from AG boundaries that result in
791 * invalid inode records, such as records that start at agbno 0
792 * or extend beyond the AG.
794 * Set min agbno to the first aligned, non-zero agbno and max to
795 * the last aligned agbno that is at least one full chunk from
798 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
799 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
800 args.mp->m_sb.sb_inoalignmt) -
803 error = xfs_alloc_vextent_near_bno(&args,
804 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
805 be32_to_cpu(agi->agi_root)));
809 newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
810 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
811 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
814 if (args.fsbno == NULLFSBLOCK)
817 ASSERT(args.len == args.minlen);
820 * Stamp and write the inode buffers.
822 * Seed the new inode cluster with a random generation number. This
823 * prevents short-term reuse of generation numbers if a chunk is
824 * freed and then immediately reallocated. We use random numbers
825 * rather than a linear progression to prevent the next generation
826 * number from being easily guessable.
828 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, pag->pag_agno,
829 args.agbno, args.len, get_random_u32());
834 * Convert the results.
836 newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
838 if (xfs_inobt_issparse(~allocmask)) {
840 * We've allocated a sparse chunk. Align the startino and mask.
842 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
844 rec.ir_startino = newino;
845 rec.ir_holemask = ~allocmask;
846 rec.ir_count = newlen;
847 rec.ir_freecount = newlen;
848 rec.ir_free = XFS_INOBT_ALL_FREE;
851 * Insert the sparse record into the inobt and allow for a merge
852 * if necessary. If a merge does occur, rec is updated to the
855 error = xfs_inobt_insert_sprec(pag, tp, agbp,
856 XFS_BTNUM_INO, &rec, true);
857 if (error == -EFSCORRUPTED) {
859 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
860 XFS_AGINO_TO_INO(args.mp, pag->pag_agno,
862 rec.ir_holemask, rec.ir_count);
863 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
869 * We can't merge the part we've just allocated as for the inobt
870 * due to finobt semantics. The original record may or may not
871 * exist independent of whether physical inodes exist in this
874 * We must update the finobt record based on the inobt record.
875 * rec contains the fully merged and up to date inobt record
876 * from the previous call. Set merge false to replace any
877 * existing record with this one.
879 if (xfs_has_finobt(args.mp)) {
880 error = xfs_inobt_insert_sprec(pag, tp, agbp,
881 XFS_BTNUM_FINO, &rec, false);
886 /* full chunk - insert new records to both btrees */
887 error = xfs_inobt_insert(pag, tp, agbp, newino, newlen,
892 if (xfs_has_finobt(args.mp)) {
893 error = xfs_inobt_insert(pag, tp, agbp, newino,
894 newlen, XFS_BTNUM_FINO);
901 * Update AGI counts and newino.
903 be32_add_cpu(&agi->agi_count, newlen);
904 be32_add_cpu(&agi->agi_freecount, newlen);
905 pag->pagi_freecount += newlen;
906 pag->pagi_count += newlen;
907 agi->agi_newino = cpu_to_be32(newino);
910 * Log allocation group header fields
912 xfs_ialloc_log_agi(tp, agbp,
913 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
915 * Modify/log superblock values for inode count and inode free count.
917 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
918 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
923 * Try to retrieve the next record to the left/right from the current one.
927 struct xfs_btree_cur *cur,
928 xfs_inobt_rec_incore_t *rec,
936 error = xfs_btree_decrement(cur, 0, &i);
938 error = xfs_btree_increment(cur, 0, &i);
944 error = xfs_inobt_get_rec(cur, rec, &i);
947 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
948 return -EFSCORRUPTED;
956 struct xfs_btree_cur *cur,
958 xfs_inobt_rec_incore_t *rec,
964 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
969 error = xfs_inobt_get_rec(cur, rec, &i);
972 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
973 return -EFSCORRUPTED;
980 * Return the offset of the first free inode in the record. If the inode chunk
981 * is sparsely allocated, we convert the record holemask to inode granularity
982 * and mask off the unallocated regions from the inode free mask.
985 xfs_inobt_first_free_inode(
986 struct xfs_inobt_rec_incore *rec)
988 xfs_inofree_t realfree;
990 /* if there are no holes, return the first available offset */
991 if (!xfs_inobt_issparse(rec->ir_holemask))
992 return xfs_lowbit64(rec->ir_free);
994 realfree = xfs_inobt_irec_to_allocmask(rec);
995 realfree &= rec->ir_free;
997 return xfs_lowbit64(realfree);
1001 * Allocate an inode using the inobt-only algorithm.
1004 xfs_dialloc_ag_inobt(
1005 struct xfs_perag *pag,
1006 struct xfs_trans *tp,
1007 struct xfs_buf *agbp,
1011 struct xfs_mount *mp = tp->t_mountp;
1012 struct xfs_agi *agi = agbp->b_addr;
1013 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1014 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1015 struct xfs_btree_cur *cur, *tcur;
1016 struct xfs_inobt_rec_incore rec, trec;
1021 int searchdistance = 10;
1023 ASSERT(xfs_perag_initialised_agi(pag));
1024 ASSERT(xfs_perag_allows_inodes(pag));
1025 ASSERT(pag->pagi_freecount > 0);
1028 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1030 * If pagino is 0 (this is the root inode allocation) use newino.
1031 * This must work because we've just allocated some.
1034 pagino = be32_to_cpu(agi->agi_newino);
1036 error = xfs_check_agi_freecount(cur);
1041 * If in the same AG as the parent, try to get near the parent.
1043 if (pagno == pag->pag_agno) {
1044 int doneleft; /* done, to the left */
1045 int doneright; /* done, to the right */
1047 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1050 if (XFS_IS_CORRUPT(mp, i != 1)) {
1051 error = -EFSCORRUPTED;
1055 error = xfs_inobt_get_rec(cur, &rec, &j);
1058 if (XFS_IS_CORRUPT(mp, j != 1)) {
1059 error = -EFSCORRUPTED;
1063 if (rec.ir_freecount > 0) {
1065 * Found a free inode in the same chunk
1066 * as the parent, done.
1073 * In the same AG as parent, but parent's chunk is full.
1076 /* duplicate the cursor, search left & right simultaneously */
1077 error = xfs_btree_dup_cursor(cur, &tcur);
1082 * Skip to last blocks looked up if same parent inode.
1084 if (pagino != NULLAGINO &&
1085 pag->pagl_pagino == pagino &&
1086 pag->pagl_leftrec != NULLAGINO &&
1087 pag->pagl_rightrec != NULLAGINO) {
1088 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1093 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1098 /* search left with tcur, back up 1 record */
1099 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1103 /* search right with cur, go forward 1 record. */
1104 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1110 * Loop until we find an inode chunk with a free inode.
1112 while (--searchdistance > 0 && (!doneleft || !doneright)) {
1113 int useleft; /* using left inode chunk this time */
1115 /* figure out the closer block if both are valid. */
1116 if (!doneleft && !doneright) {
1118 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1119 rec.ir_startino - pagino;
1121 useleft = !doneleft;
1124 /* free inodes to the left? */
1125 if (useleft && trec.ir_freecount) {
1126 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1129 pag->pagl_leftrec = trec.ir_startino;
1130 pag->pagl_rightrec = rec.ir_startino;
1131 pag->pagl_pagino = pagino;
1136 /* free inodes to the right? */
1137 if (!useleft && rec.ir_freecount) {
1138 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1140 pag->pagl_leftrec = trec.ir_startino;
1141 pag->pagl_rightrec = rec.ir_startino;
1142 pag->pagl_pagino = pagino;
1146 /* get next record to check */
1148 error = xfs_ialloc_next_rec(tcur, &trec,
1151 error = xfs_ialloc_next_rec(cur, &rec,
1158 if (searchdistance <= 0) {
1160 * Not in range - save last search
1161 * location and allocate a new inode
1163 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1164 pag->pagl_leftrec = trec.ir_startino;
1165 pag->pagl_rightrec = rec.ir_startino;
1166 pag->pagl_pagino = pagino;
1170 * We've reached the end of the btree. because
1171 * we are only searching a small chunk of the
1172 * btree each search, there is obviously free
1173 * inodes closer to the parent inode than we
1174 * are now. restart the search again.
1176 pag->pagl_pagino = NULLAGINO;
1177 pag->pagl_leftrec = NULLAGINO;
1178 pag->pagl_rightrec = NULLAGINO;
1179 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1180 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1186 * In a different AG from the parent.
1187 * See if the most recently allocated block has any free.
1189 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1190 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1196 error = xfs_inobt_get_rec(cur, &rec, &j);
1200 if (j == 1 && rec.ir_freecount > 0) {
1202 * The last chunk allocated in the group
1203 * still has a free inode.
1211 * None left in the last group, search the whole AG
1213 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1216 if (XFS_IS_CORRUPT(mp, i != 1)) {
1217 error = -EFSCORRUPTED;
1222 error = xfs_inobt_get_rec(cur, &rec, &i);
1225 if (XFS_IS_CORRUPT(mp, i != 1)) {
1226 error = -EFSCORRUPTED;
1229 if (rec.ir_freecount > 0)
1231 error = xfs_btree_increment(cur, 0, &i);
1234 if (XFS_IS_CORRUPT(mp, i != 1)) {
1235 error = -EFSCORRUPTED;
1241 offset = xfs_inobt_first_free_inode(&rec);
1242 ASSERT(offset >= 0);
1243 ASSERT(offset < XFS_INODES_PER_CHUNK);
1244 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1245 XFS_INODES_PER_CHUNK) == 0);
1246 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1247 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1249 error = xfs_inobt_update(cur, &rec);
1252 be32_add_cpu(&agi->agi_freecount, -1);
1253 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1254 pag->pagi_freecount--;
1256 error = xfs_check_agi_freecount(cur);
1260 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1261 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1265 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1267 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1272 * Use the free inode btree to allocate an inode based on distance from the
1273 * parent. Note that the provided cursor may be deleted and replaced.
1276 xfs_dialloc_ag_finobt_near(
1278 struct xfs_btree_cur **ocur,
1279 struct xfs_inobt_rec_incore *rec)
1281 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1282 struct xfs_btree_cur *rcur; /* right search cursor */
1283 struct xfs_inobt_rec_incore rrec;
1287 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1292 error = xfs_inobt_get_rec(lcur, rec, &i);
1295 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1))
1296 return -EFSCORRUPTED;
1299 * See if we've landed in the parent inode record. The finobt
1300 * only tracks chunks with at least one free inode, so record
1301 * existence is enough.
1303 if (pagino >= rec->ir_startino &&
1304 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1308 error = xfs_btree_dup_cursor(lcur, &rcur);
1312 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1316 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1319 if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) {
1320 error = -EFSCORRUPTED;
1325 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) {
1326 error = -EFSCORRUPTED;
1329 if (i == 1 && j == 1) {
1331 * Both the left and right records are valid. Choose the closer
1332 * inode chunk to the target.
1334 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1335 (rrec.ir_startino - pagino)) {
1337 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1340 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1342 } else if (j == 1) {
1343 /* only the right record is valid */
1345 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1347 } else if (i == 1) {
1348 /* only the left record is valid */
1349 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1355 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1360 * Use the free inode btree to find a free inode based on a newino hint. If
1361 * the hint is NULL, find the first free inode in the AG.
1364 xfs_dialloc_ag_finobt_newino(
1365 struct xfs_agi *agi,
1366 struct xfs_btree_cur *cur,
1367 struct xfs_inobt_rec_incore *rec)
1372 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1373 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1378 error = xfs_inobt_get_rec(cur, rec, &i);
1381 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1382 return -EFSCORRUPTED;
1388 * Find the first inode available in the AG.
1390 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1393 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1394 return -EFSCORRUPTED;
1396 error = xfs_inobt_get_rec(cur, rec, &i);
1399 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1400 return -EFSCORRUPTED;
1406 * Update the inobt based on a modification made to the finobt. Also ensure that
1407 * the records from both trees are equivalent post-modification.
1410 xfs_dialloc_ag_update_inobt(
1411 struct xfs_btree_cur *cur, /* inobt cursor */
1412 struct xfs_inobt_rec_incore *frec, /* finobt record */
1413 int offset) /* inode offset */
1415 struct xfs_inobt_rec_incore rec;
1419 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1422 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1423 return -EFSCORRUPTED;
1425 error = xfs_inobt_get_rec(cur, &rec, &i);
1428 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1429 return -EFSCORRUPTED;
1430 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1431 XFS_INODES_PER_CHUNK) == 0);
1433 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1436 if (XFS_IS_CORRUPT(cur->bc_mp,
1437 rec.ir_free != frec->ir_free ||
1438 rec.ir_freecount != frec->ir_freecount))
1439 return -EFSCORRUPTED;
1441 return xfs_inobt_update(cur, &rec);
1445 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1446 * back to the inobt search algorithm.
1448 * The caller selected an AG for us, and made sure that free inodes are
1453 struct xfs_perag *pag,
1454 struct xfs_trans *tp,
1455 struct xfs_buf *agbp,
1459 struct xfs_mount *mp = tp->t_mountp;
1460 struct xfs_agi *agi = agbp->b_addr;
1461 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1462 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1463 struct xfs_btree_cur *cur; /* finobt cursor */
1464 struct xfs_btree_cur *icur; /* inobt cursor */
1465 struct xfs_inobt_rec_incore rec;
1471 if (!xfs_has_finobt(mp))
1472 return xfs_dialloc_ag_inobt(pag, tp, agbp, parent, inop);
1475 * If pagino is 0 (this is the root inode allocation) use newino.
1476 * This must work because we've just allocated some.
1479 pagino = be32_to_cpu(agi->agi_newino);
1481 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
1483 error = xfs_check_agi_freecount(cur);
1488 * The search algorithm depends on whether we're in the same AG as the
1489 * parent. If so, find the closest available inode to the parent. If
1490 * not, consider the agi hint or find the first free inode in the AG.
1492 if (pag->pag_agno == pagno)
1493 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1495 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1499 offset = xfs_inobt_first_free_inode(&rec);
1500 ASSERT(offset >= 0);
1501 ASSERT(offset < XFS_INODES_PER_CHUNK);
1502 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1503 XFS_INODES_PER_CHUNK) == 0);
1504 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1507 * Modify or remove the finobt record.
1509 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1511 if (rec.ir_freecount)
1512 error = xfs_inobt_update(cur, &rec);
1514 error = xfs_btree_delete(cur, &i);
1519 * The finobt has now been updated appropriately. We haven't updated the
1520 * agi and superblock yet, so we can create an inobt cursor and validate
1521 * the original freecount. If all is well, make the equivalent update to
1522 * the inobt using the finobt record and offset information.
1524 icur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1526 error = xfs_check_agi_freecount(icur);
1530 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1535 * Both trees have now been updated. We must update the perag and
1536 * superblock before we can check the freecount for each btree.
1538 be32_add_cpu(&agi->agi_freecount, -1);
1539 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1540 pag->pagi_freecount--;
1542 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1544 error = xfs_check_agi_freecount(icur);
1547 error = xfs_check_agi_freecount(cur);
1551 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1552 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1557 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1559 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1565 struct xfs_trans **tpp,
1566 struct xfs_buf *agibp)
1568 struct xfs_trans *tp = *tpp;
1569 struct xfs_dquot_acct *dqinfo;
1573 * Hold to on to the agibp across the commit so no other allocation can
1574 * come in and take the free inodes we just allocated for our caller.
1576 xfs_trans_bhold(tp, agibp);
1579 * We want the quota changes to be associated with the next transaction,
1580 * NOT this one. So, detach the dqinfo from this and attach it to the
1583 dqinfo = tp->t_dqinfo;
1584 tp->t_dqinfo = NULL;
1586 error = xfs_trans_roll(&tp);
1588 /* Re-attach the quota info that we detached from prev trx. */
1589 tp->t_dqinfo = dqinfo;
1592 * Join the buffer even on commit error so that the buffer is released
1593 * when the caller cancels the transaction and doesn't have to handle
1594 * this error case specially.
1596 xfs_trans_bjoin(tp, agibp);
1602 xfs_dialloc_good_ag(
1603 struct xfs_perag *pag,
1604 struct xfs_trans *tp,
1609 struct xfs_mount *mp = tp->t_mountp;
1611 xfs_extlen_t longest = 0;
1617 if (!xfs_perag_allows_inodes(pag))
1620 if (!xfs_perag_initialised_agi(pag)) {
1621 error = xfs_ialloc_read_agi(pag, tp, NULL);
1626 if (pag->pagi_freecount)
1631 if (!xfs_perag_initialised_agf(pag)) {
1632 error = xfs_alloc_read_agf(pag, tp, flags, NULL);
1638 * Check that there is enough free space for the file plus a chunk of
1639 * inodes if we need to allocate some. If this is the first pass across
1640 * the AGs, take into account the potential space needed for alignment
1641 * of inode chunks when checking the longest contiguous free space in
1642 * the AG - this prevents us from getting ENOSPC because we have free
1643 * space larger than ialloc_blks but alignment constraints prevent us
1646 * If we can't find an AG with space for full alignment slack to be
1647 * taken into account, we must be near ENOSPC in all AGs. Hence we
1648 * don't include alignment for the second pass and so if we fail
1649 * allocation due to alignment issues then it is most likely a real
1652 * XXX(dgc): this calculation is now bogus thanks to the per-ag
1653 * reservations that xfs_alloc_fix_freelist() now does via
1654 * xfs_alloc_space_available(). When the AG fills up, pagf_freeblks will
1655 * be more than large enough for the check below to succeed, but
1656 * xfs_alloc_space_available() will fail because of the non-zero
1657 * metadata reservation and hence we won't actually be able to allocate
1658 * more inodes in this AG. We do soooo much unnecessary work near ENOSPC
1661 ineed = M_IGEO(mp)->ialloc_min_blks;
1662 if (flags && ineed > 1)
1663 ineed += M_IGEO(mp)->cluster_align;
1664 longest = pag->pagf_longest;
1666 longest = pag->pagf_flcount > 0;
1667 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
1669 if (pag->pagf_freeblks < needspace + ineed || longest < ineed)
1676 struct xfs_perag *pag,
1677 struct xfs_trans **tpp,
1682 struct xfs_buf *agbp;
1687 * Then read in the AGI buffer and recheck with the AGI buffer
1690 error = xfs_ialloc_read_agi(pag, *tpp, &agbp);
1694 if (!pag->pagi_freecount) {
1700 error = xfs_ialloc_ag_alloc(pag, *tpp, agbp);
1705 * We successfully allocated space for an inode cluster in this
1706 * AG. Roll the transaction so that we can allocate one of the
1709 ASSERT(pag->pagi_freecount > 0);
1710 error = xfs_dialloc_roll(tpp, agbp);
1715 /* Allocate an inode in the found AG */
1716 error = xfs_dialloc_ag(pag, *tpp, agbp, parent, &ino);
1722 xfs_trans_brelse(*tpp, agbp);
1727 * Allocate an on-disk inode.
1729 * Mode is used to tell whether the new inode is a directory and hence where to
1730 * locate it. The on-disk inode that is allocated will be returned in @new_ino
1731 * on success, otherwise an error will be set to indicate the failure (e.g.
1736 struct xfs_trans **tpp,
1741 struct xfs_mount *mp = (*tpp)->t_mountp;
1742 xfs_agnumber_t agno;
1744 xfs_agnumber_t start_agno;
1745 struct xfs_perag *pag;
1746 struct xfs_ino_geometry *igeo = M_IGEO(mp);
1747 bool ok_alloc = true;
1748 bool low_space = false;
1750 xfs_ino_t ino = NULLFSINO;
1753 * Directories, symlinks, and regular files frequently allocate at least
1754 * one block, so factor that potential expansion when we examine whether
1755 * an AG has enough space for file creation.
1758 start_agno = (atomic_inc_return(&mp->m_agirotor) - 1) %
1761 start_agno = XFS_INO_TO_AGNO(mp, parent);
1762 if (start_agno >= mp->m_maxagi)
1767 * If we have already hit the ceiling of inode blocks then clear
1768 * ok_alloc so we scan all available agi structures for a free
1771 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1772 * which will sacrifice the preciseness but improve the performance.
1774 if (igeo->maxicount &&
1775 percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1776 > igeo->maxicount) {
1781 * If we are near to ENOSPC, we want to prefer allocation from AGs that
1782 * have free inodes in them rather than use up free space allocating new
1783 * inode chunks. Hence we turn off allocation for the first non-blocking
1784 * pass through the AGs if we are near ENOSPC to consume free inodes
1785 * that we can immediately allocate, but then we allow allocation on the
1786 * second pass if we fail to find an AG with free inodes in it.
1788 if (percpu_counter_read_positive(&mp->m_fdblocks) <
1789 mp->m_low_space[XFS_LOWSP_1_PCNT]) {
1795 * Loop until we find an allocation group that either has free inodes
1796 * or in which we can allocate some inodes. Iterate through the
1797 * allocation groups upward, wrapping at the end.
1799 flags = XFS_ALLOC_FLAG_TRYLOCK;
1801 for_each_perag_wrap_at(mp, start_agno, mp->m_maxagi, agno, pag) {
1802 if (xfs_dialloc_good_ag(pag, *tpp, mode, flags, ok_alloc)) {
1803 error = xfs_dialloc_try_ag(pag, tpp, parent,
1805 if (error != -EAGAIN)
1810 if (xfs_is_shutdown(mp)) {
1811 error = -EFSCORRUPTED;
1816 xfs_perag_rele(pag);
1819 if (ino == NULLFSINO) {
1833 * Free the blocks of an inode chunk. We must consider that the inode chunk
1834 * might be sparse and only free the regions that are allocated as part of the
1838 xfs_difree_inode_chunk(
1839 struct xfs_trans *tp,
1840 xfs_agnumber_t agno,
1841 struct xfs_inobt_rec_incore *rec)
1843 struct xfs_mount *mp = tp->t_mountp;
1844 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1846 int startidx, endidx;
1848 xfs_agblock_t agbno;
1850 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1852 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1853 /* not sparse, calculate extent info directly */
1854 return xfs_free_extent_later(tp,
1855 XFS_AGB_TO_FSB(mp, agno, sagbno),
1856 M_IGEO(mp)->ialloc_blks, &XFS_RMAP_OINFO_INODES,
1860 /* holemask is only 16-bits (fits in an unsigned long) */
1861 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1862 holemask[0] = rec->ir_holemask;
1865 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1866 * holemask and convert the start/end index of each range to an extent.
1867 * We start with the start and end index both pointing at the first 0 in
1870 startidx = endidx = find_first_zero_bit(holemask,
1871 XFS_INOBT_HOLEMASK_BITS);
1872 nextbit = startidx + 1;
1873 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1876 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1879 * If the next zero bit is contiguous, update the end index of
1880 * the current range and continue.
1882 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1883 nextbit == endidx + 1) {
1889 * nextbit is not contiguous with the current end index. Convert
1890 * the current start/end to an extent and add it to the free
1893 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1894 mp->m_sb.sb_inopblock;
1895 contigblk = ((endidx - startidx + 1) *
1896 XFS_INODES_PER_HOLEMASK_BIT) /
1897 mp->m_sb.sb_inopblock;
1899 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1900 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1901 error = xfs_free_extent_later(tp,
1902 XFS_AGB_TO_FSB(mp, agno, agbno), contigblk,
1903 &XFS_RMAP_OINFO_INODES, XFS_AG_RESV_NONE);
1907 /* reset range to current bit and carry on... */
1908 startidx = endidx = nextbit;
1918 struct xfs_perag *pag,
1919 struct xfs_trans *tp,
1920 struct xfs_buf *agbp,
1922 struct xfs_icluster *xic,
1923 struct xfs_inobt_rec_incore *orec)
1925 struct xfs_mount *mp = pag->pag_mount;
1926 struct xfs_agi *agi = agbp->b_addr;
1927 struct xfs_btree_cur *cur;
1928 struct xfs_inobt_rec_incore rec;
1934 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1935 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1938 * Initialize the cursor.
1940 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1942 error = xfs_check_agi_freecount(cur);
1947 * Look for the entry describing this inode.
1949 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1950 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1954 if (XFS_IS_CORRUPT(mp, i != 1)) {
1955 error = -EFSCORRUPTED;
1958 error = xfs_inobt_get_rec(cur, &rec, &i);
1960 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1964 if (XFS_IS_CORRUPT(mp, i != 1)) {
1965 error = -EFSCORRUPTED;
1969 * Get the offset in the inode chunk.
1971 off = agino - rec.ir_startino;
1972 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1973 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1975 * Mark the inode free & increment the count.
1977 rec.ir_free |= XFS_INOBT_MASK(off);
1981 * When an inode chunk is free, it becomes eligible for removal. Don't
1982 * remove the chunk if the block size is large enough for multiple inode
1983 * chunks (that might not be free).
1985 if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
1986 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1987 xic->deleted = true;
1988 xic->first_ino = XFS_AGINO_TO_INO(mp, pag->pag_agno,
1990 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1993 * Remove the inode cluster from the AGI B+Tree, adjust the
1994 * AGI and Superblock inode counts, and mark the disk space
1995 * to be freed when the transaction is committed.
1997 ilen = rec.ir_freecount;
1998 be32_add_cpu(&agi->agi_count, -ilen);
1999 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
2000 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
2001 pag->pagi_freecount -= ilen - 1;
2002 pag->pagi_count -= ilen;
2003 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
2004 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
2006 if ((error = xfs_btree_delete(cur, &i))) {
2007 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2012 error = xfs_difree_inode_chunk(tp, pag->pag_agno, &rec);
2016 xic->deleted = false;
2018 error = xfs_inobt_update(cur, &rec);
2020 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2026 * Change the inode free counts and log the ag/sb changes.
2028 be32_add_cpu(&agi->agi_freecount, 1);
2029 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2030 pag->pagi_freecount++;
2031 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2034 error = xfs_check_agi_freecount(cur);
2039 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2043 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2048 * Free an inode in the free inode btree.
2052 struct xfs_perag *pag,
2053 struct xfs_trans *tp,
2054 struct xfs_buf *agbp,
2056 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2058 struct xfs_mount *mp = pag->pag_mount;
2059 struct xfs_btree_cur *cur;
2060 struct xfs_inobt_rec_incore rec;
2061 int offset = agino - ibtrec->ir_startino;
2065 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
2067 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2072 * If the record does not exist in the finobt, we must have just
2073 * freed an inode in a previously fully allocated chunk. If not,
2074 * something is out of sync.
2076 if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) {
2077 error = -EFSCORRUPTED;
2081 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2083 ibtrec->ir_freecount,
2084 ibtrec->ir_free, &i);
2093 * Read and update the existing record. We could just copy the ibtrec
2094 * across here, but that would defeat the purpose of having redundant
2095 * metadata. By making the modifications independently, we can catch
2096 * corruptions that we wouldn't see if we just copied from one record
2099 error = xfs_inobt_get_rec(cur, &rec, &i);
2102 if (XFS_IS_CORRUPT(mp, i != 1)) {
2103 error = -EFSCORRUPTED;
2107 rec.ir_free |= XFS_INOBT_MASK(offset);
2110 if (XFS_IS_CORRUPT(mp,
2111 rec.ir_free != ibtrec->ir_free ||
2112 rec.ir_freecount != ibtrec->ir_freecount)) {
2113 error = -EFSCORRUPTED;
2118 * The content of inobt records should always match between the inobt
2119 * and finobt. The lifecycle of records in the finobt is different from
2120 * the inobt in that the finobt only tracks records with at least one
2121 * free inode. Hence, if all of the inodes are free and we aren't
2122 * keeping inode chunks permanently on disk, remove the record.
2123 * Otherwise, update the record with the new information.
2125 * Note that we currently can't free chunks when the block size is large
2126 * enough for multiple chunks. Leave the finobt record to remain in sync
2129 if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
2130 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
2131 error = xfs_btree_delete(cur, &i);
2136 error = xfs_inobt_update(cur, &rec);
2142 error = xfs_check_agi_freecount(cur);
2146 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2150 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2155 * Free disk inode. Carefully avoids touching the incore inode, all
2156 * manipulations incore are the caller's responsibility.
2157 * The on-disk inode is not changed by this operation, only the
2158 * btree (free inode mask) is changed.
2162 struct xfs_trans *tp,
2163 struct xfs_perag *pag,
2165 struct xfs_icluster *xic)
2168 xfs_agblock_t agbno; /* block number containing inode */
2169 struct xfs_buf *agbp; /* buffer for allocation group header */
2170 xfs_agino_t agino; /* allocation group inode number */
2171 int error; /* error return value */
2172 struct xfs_mount *mp = tp->t_mountp;
2173 struct xfs_inobt_rec_incore rec;/* btree record */
2176 * Break up inode number into its components.
2178 if (pag->pag_agno != XFS_INO_TO_AGNO(mp, inode)) {
2179 xfs_warn(mp, "%s: agno != pag->pag_agno (%d != %d).",
2180 __func__, XFS_INO_TO_AGNO(mp, inode), pag->pag_agno);
2184 agino = XFS_INO_TO_AGINO(mp, inode);
2185 if (inode != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2186 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2187 __func__, (unsigned long long)inode,
2188 (unsigned long long)XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2192 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2193 if (agbno >= mp->m_sb.sb_agblocks) {
2194 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2195 __func__, agbno, mp->m_sb.sb_agblocks);
2200 * Get the allocation group header.
2202 error = xfs_ialloc_read_agi(pag, tp, &agbp);
2204 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2210 * Fix up the inode allocation btree.
2212 error = xfs_difree_inobt(pag, tp, agbp, agino, xic, &rec);
2217 * Fix up the free inode btree.
2219 if (xfs_has_finobt(mp)) {
2220 error = xfs_difree_finobt(pag, tp, agbp, agino, &rec);
2233 struct xfs_perag *pag,
2234 struct xfs_trans *tp,
2236 xfs_agblock_t agbno,
2237 xfs_agblock_t *chunk_agbno,
2238 xfs_agblock_t *offset_agbno,
2241 struct xfs_mount *mp = pag->pag_mount;
2242 struct xfs_inobt_rec_incore rec;
2243 struct xfs_btree_cur *cur;
2244 struct xfs_buf *agbp;
2248 error = xfs_ialloc_read_agi(pag, tp, &agbp);
2251 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2252 __func__, error, pag->pag_agno);
2257 * Lookup the inode record for the given agino. If the record cannot be
2258 * found, then it's an invalid inode number and we should abort. Once
2259 * we have a record, we need to ensure it contains the inode number
2260 * we are looking up.
2262 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
2263 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2266 error = xfs_inobt_get_rec(cur, &rec, &i);
2267 if (!error && i == 0)
2271 xfs_trans_brelse(tp, agbp);
2272 xfs_btree_del_cursor(cur, error);
2276 /* check that the returned record contains the required inode */
2277 if (rec.ir_startino > agino ||
2278 rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2281 /* for untrusted inodes check it is allocated first */
2282 if ((flags & XFS_IGET_UNTRUSTED) &&
2283 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2286 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2287 *offset_agbno = agbno - *chunk_agbno;
2292 * Return the location of the inode in imap, for mapping it into a buffer.
2296 struct xfs_perag *pag,
2297 struct xfs_trans *tp,
2298 xfs_ino_t ino, /* inode to locate */
2299 struct xfs_imap *imap, /* location map structure */
2300 uint flags) /* flags for inode btree lookup */
2302 struct xfs_mount *mp = pag->pag_mount;
2303 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2304 xfs_agino_t agino; /* inode number within alloc group */
2305 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2306 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2307 int error; /* error code */
2308 int offset; /* index of inode in its buffer */
2309 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2311 ASSERT(ino != NULLFSINO);
2314 * Split up the inode number into its parts.
2316 agino = XFS_INO_TO_AGINO(mp, ino);
2317 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2318 if (agbno >= mp->m_sb.sb_agblocks ||
2319 ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2323 * Don't output diagnostic information for untrusted inodes
2324 * as they can be invalid without implying corruption.
2326 if (flags & XFS_IGET_UNTRUSTED)
2328 if (agbno >= mp->m_sb.sb_agblocks) {
2330 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2331 __func__, (unsigned long long)agbno,
2332 (unsigned long)mp->m_sb.sb_agblocks);
2334 if (ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2336 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2338 XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2346 * For bulkstat and handle lookups, we have an untrusted inode number
2347 * that we have to verify is valid. We cannot do this just by reading
2348 * the inode buffer as it may have been unlinked and removed leaving
2349 * inodes in stale state on disk. Hence we have to do a btree lookup
2350 * in all cases where an untrusted inode number is passed.
2352 if (flags & XFS_IGET_UNTRUSTED) {
2353 error = xfs_imap_lookup(pag, tp, agino, agbno,
2354 &chunk_agbno, &offset_agbno, flags);
2361 * If the inode cluster size is the same as the blocksize or
2362 * smaller we get to the buffer by simple arithmetics.
2364 if (M_IGEO(mp)->blocks_per_cluster == 1) {
2365 offset = XFS_INO_TO_OFFSET(mp, ino);
2366 ASSERT(offset < mp->m_sb.sb_inopblock);
2368 imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, agbno);
2369 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2370 imap->im_boffset = (unsigned short)(offset <<
2371 mp->m_sb.sb_inodelog);
2376 * If the inode chunks are aligned then use simple maths to
2377 * find the location. Otherwise we have to do a btree
2378 * lookup to find the location.
2380 if (M_IGEO(mp)->inoalign_mask) {
2381 offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2382 chunk_agbno = agbno - offset_agbno;
2384 error = xfs_imap_lookup(pag, tp, agino, agbno,
2385 &chunk_agbno, &offset_agbno, flags);
2391 ASSERT(agbno >= chunk_agbno);
2392 cluster_agbno = chunk_agbno +
2393 ((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2394 M_IGEO(mp)->blocks_per_cluster);
2395 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2396 XFS_INO_TO_OFFSET(mp, ino);
2398 imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, cluster_agbno);
2399 imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2400 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2403 * If the inode number maps to a block outside the bounds
2404 * of the file system then return NULL rather than calling
2405 * read_buf and panicing when we get an error from the
2408 if ((imap->im_blkno + imap->im_len) >
2409 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2411 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2412 __func__, (unsigned long long) imap->im_blkno,
2413 (unsigned long long) imap->im_len,
2414 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2421 * Log specified fields for the ag hdr (inode section). The growth of the agi
2422 * structure over time requires that we interpret the buffer as two logical
2423 * regions delineated by the end of the unlinked list. This is due to the size
2424 * of the hash table and its location in the middle of the agi.
2426 * For example, a request to log a field before agi_unlinked and a field after
2427 * agi_unlinked could cause us to log the entire hash table and use an excessive
2428 * amount of log space. To avoid this behavior, log the region up through
2429 * agi_unlinked in one call and the region after agi_unlinked through the end of
2430 * the structure in another.
2434 struct xfs_trans *tp,
2438 int first; /* first byte number */
2439 int last; /* last byte number */
2440 static const short offsets[] = { /* field starting offsets */
2441 /* keep in sync with bit definitions */
2442 offsetof(xfs_agi_t, agi_magicnum),
2443 offsetof(xfs_agi_t, agi_versionnum),
2444 offsetof(xfs_agi_t, agi_seqno),
2445 offsetof(xfs_agi_t, agi_length),
2446 offsetof(xfs_agi_t, agi_count),
2447 offsetof(xfs_agi_t, agi_root),
2448 offsetof(xfs_agi_t, agi_level),
2449 offsetof(xfs_agi_t, agi_freecount),
2450 offsetof(xfs_agi_t, agi_newino),
2451 offsetof(xfs_agi_t, agi_dirino),
2452 offsetof(xfs_agi_t, agi_unlinked),
2453 offsetof(xfs_agi_t, agi_free_root),
2454 offsetof(xfs_agi_t, agi_free_level),
2455 offsetof(xfs_agi_t, agi_iblocks),
2459 struct xfs_agi *agi = bp->b_addr;
2461 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2465 * Compute byte offsets for the first and last fields in the first
2466 * region and log the agi buffer. This only logs up through
2469 if (fields & XFS_AGI_ALL_BITS_R1) {
2470 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2472 xfs_trans_log_buf(tp, bp, first, last);
2476 * Mask off the bits in the first region and calculate the first and
2477 * last field offsets for any bits in the second region.
2479 fields &= ~XFS_AGI_ALL_BITS_R1;
2481 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2483 xfs_trans_log_buf(tp, bp, first, last);
2487 static xfs_failaddr_t
2491 struct xfs_mount *mp = bp->b_mount;
2492 struct xfs_agi *agi = bp->b_addr;
2494 uint32_t agi_seqno = be32_to_cpu(agi->agi_seqno);
2495 uint32_t agi_length = be32_to_cpu(agi->agi_length);
2498 if (xfs_has_crc(mp)) {
2499 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2500 return __this_address;
2501 if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn)))
2502 return __this_address;
2506 * Validate the magic number of the agi block.
2508 if (!xfs_verify_magic(bp, agi->agi_magicnum))
2509 return __this_address;
2510 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2511 return __this_address;
2513 fa = xfs_validate_ag_length(bp, agi_seqno, agi_length);
2517 if (be32_to_cpu(agi->agi_level) < 1 ||
2518 be32_to_cpu(agi->agi_level) > M_IGEO(mp)->inobt_maxlevels)
2519 return __this_address;
2521 if (xfs_has_finobt(mp) &&
2522 (be32_to_cpu(agi->agi_free_level) < 1 ||
2523 be32_to_cpu(agi->agi_free_level) > M_IGEO(mp)->inobt_maxlevels))
2524 return __this_address;
2526 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2527 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2529 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2530 return __this_address;
2537 xfs_agi_read_verify(
2540 struct xfs_mount *mp = bp->b_mount;
2543 if (xfs_has_crc(mp) &&
2544 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2545 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2547 fa = xfs_agi_verify(bp);
2548 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2549 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2554 xfs_agi_write_verify(
2557 struct xfs_mount *mp = bp->b_mount;
2558 struct xfs_buf_log_item *bip = bp->b_log_item;
2559 struct xfs_agi *agi = bp->b_addr;
2562 fa = xfs_agi_verify(bp);
2564 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2568 if (!xfs_has_crc(mp))
2572 agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2573 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2576 const struct xfs_buf_ops xfs_agi_buf_ops = {
2578 .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2579 .verify_read = xfs_agi_read_verify,
2580 .verify_write = xfs_agi_write_verify,
2581 .verify_struct = xfs_agi_verify,
2585 * Read in the allocation group header (inode allocation section)
2589 struct xfs_perag *pag,
2590 struct xfs_trans *tp,
2591 struct xfs_buf **agibpp)
2593 struct xfs_mount *mp = pag->pag_mount;
2596 trace_xfs_read_agi(pag->pag_mount, pag->pag_agno);
2598 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2599 XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGI_DADDR(mp)),
2600 XFS_FSS_TO_BB(mp, 1), 0, agibpp, &xfs_agi_buf_ops);
2604 xfs_trans_buf_set_type(tp, *agibpp, XFS_BLFT_AGI_BUF);
2606 xfs_buf_set_ref(*agibpp, XFS_AGI_REF);
2611 * Read in the agi and initialise the per-ag data. If the caller supplies a
2612 * @agibpp, return the locked AGI buffer to them, otherwise release it.
2615 xfs_ialloc_read_agi(
2616 struct xfs_perag *pag,
2617 struct xfs_trans *tp,
2618 struct xfs_buf **agibpp)
2620 struct xfs_buf *agibp;
2621 struct xfs_agi *agi;
2624 trace_xfs_ialloc_read_agi(pag->pag_mount, pag->pag_agno);
2626 error = xfs_read_agi(pag, tp, &agibp);
2630 agi = agibp->b_addr;
2631 if (!xfs_perag_initialised_agi(pag)) {
2632 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2633 pag->pagi_count = be32_to_cpu(agi->agi_count);
2634 set_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
2638 * It's possible for these to be out of sync if
2639 * we are in the middle of a forced shutdown.
2641 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2642 xfs_is_shutdown(pag->pag_mount));
2646 xfs_trans_brelse(tp, agibp);
2650 /* How many inodes are backed by inode clusters ondisk? */
2652 xfs_ialloc_count_ondisk(
2653 struct xfs_btree_cur *cur,
2656 unsigned int *allocated)
2658 struct xfs_inobt_rec_incore irec;
2659 unsigned int ret = 0;
2663 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2667 while (has_record) {
2668 unsigned int i, hole_idx;
2670 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2673 if (irec.ir_startino > high)
2676 for (i = 0; i < XFS_INODES_PER_CHUNK; i++) {
2677 if (irec.ir_startino + i < low)
2679 if (irec.ir_startino + i > high)
2682 hole_idx = i / XFS_INODES_PER_HOLEMASK_BIT;
2683 if (!(irec.ir_holemask & (1U << hole_idx)))
2687 error = xfs_btree_increment(cur, 0, &has_record);
2696 /* Is there an inode record covering a given extent? */
2698 xfs_ialloc_has_inodes_at_extent(
2699 struct xfs_btree_cur *cur,
2702 enum xbtree_recpacking *outcome)
2705 xfs_agino_t last_agino;
2706 unsigned int allocated;
2709 agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2710 last_agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2712 error = xfs_ialloc_count_ondisk(cur, agino, last_agino, &allocated);
2717 *outcome = XBTREE_RECPACKING_EMPTY;
2718 else if (allocated == last_agino - agino + 1)
2719 *outcome = XBTREE_RECPACKING_FULL;
2721 *outcome = XBTREE_RECPACKING_SPARSE;
2725 struct xfs_ialloc_count_inodes {
2727 xfs_agino_t freecount;
2730 /* Record inode counts across all inobt records. */
2732 xfs_ialloc_count_inodes_rec(
2733 struct xfs_btree_cur *cur,
2734 const union xfs_btree_rec *rec,
2737 struct xfs_inobt_rec_incore irec;
2738 struct xfs_ialloc_count_inodes *ci = priv;
2741 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2742 fa = xfs_inobt_check_irec(cur, &irec);
2744 return xfs_inobt_complain_bad_rec(cur, fa, &irec);
2746 ci->count += irec.ir_count;
2747 ci->freecount += irec.ir_freecount;
2752 /* Count allocated and free inodes under an inobt. */
2754 xfs_ialloc_count_inodes(
2755 struct xfs_btree_cur *cur,
2757 xfs_agino_t *freecount)
2759 struct xfs_ialloc_count_inodes ci = {0};
2762 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2763 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2768 *freecount = ci.freecount;
2773 * Initialize inode-related geometry information.
2775 * Compute the inode btree min and max levels and set maxicount.
2777 * Set the inode cluster size. This may still be overridden by the file
2778 * system block size if it is larger than the chosen cluster size.
2780 * For v5 filesystems, scale the cluster size with the inode size to keep a
2781 * constant ratio of inode per cluster buffer, but only if mkfs has set the
2782 * inode alignment value appropriately for larger cluster sizes.
2784 * Then compute the inode cluster alignment information.
2787 xfs_ialloc_setup_geometry(
2788 struct xfs_mount *mp)
2790 struct xfs_sb *sbp = &mp->m_sb;
2791 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2795 igeo->new_diflags2 = 0;
2796 if (xfs_has_bigtime(mp))
2797 igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME;
2798 if (xfs_has_large_extent_counts(mp))
2799 igeo->new_diflags2 |= XFS_DIFLAG2_NREXT64;
2801 /* Compute inode btree geometry. */
2802 igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2803 igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
2804 igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
2805 igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2806 igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2808 igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2810 igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2812 if (sbp->sb_spino_align)
2813 igeo->ialloc_min_blks = sbp->sb_spino_align;
2815 igeo->ialloc_min_blks = igeo->ialloc_blks;
2817 /* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2818 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2819 igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2821 ASSERT(igeo->inobt_maxlevels <= xfs_iallocbt_maxlevels_ondisk());
2824 * Set the maximum inode count for this filesystem, being careful not
2825 * to use obviously garbage sb_inopblog/sb_inopblock values. Regular
2826 * users should never get here due to failing sb verification, but
2827 * certain users (xfs_db) need to be usable even with corrupt metadata.
2829 if (sbp->sb_imax_pct && igeo->ialloc_blks) {
2831 * Make sure the maximum inode count is a multiple
2832 * of the units we allocate inodes in.
2834 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
2835 do_div(icount, 100);
2836 do_div(icount, igeo->ialloc_blks);
2837 igeo->maxicount = XFS_FSB_TO_INO(mp,
2838 icount * igeo->ialloc_blks);
2840 igeo->maxicount = 0;
2844 * Compute the desired size of an inode cluster buffer size, which
2845 * starts at 8K and (on v5 filesystems) scales up with larger inode
2848 * Preserve the desired inode cluster size because the sparse inodes
2849 * feature uses that desired size (not the actual size) to compute the
2850 * sparse inode alignment. The mount code validates this value, so we
2851 * cannot change the behavior.
2853 igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
2854 if (xfs_has_v3inodes(mp)) {
2855 int new_size = igeo->inode_cluster_size_raw;
2857 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
2858 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
2859 igeo->inode_cluster_size_raw = new_size;
2862 /* Calculate inode cluster ratios. */
2863 if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
2864 igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
2865 igeo->inode_cluster_size_raw);
2867 igeo->blocks_per_cluster = 1;
2868 igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
2869 igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
2871 /* Calculate inode cluster alignment. */
2872 if (xfs_has_align(mp) &&
2873 mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
2874 igeo->cluster_align = mp->m_sb.sb_inoalignmt;
2876 igeo->cluster_align = 1;
2877 igeo->inoalign_mask = igeo->cluster_align - 1;
2878 igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
2881 * If we are using stripe alignment, check whether
2882 * the stripe unit is a multiple of the inode alignment
2884 if (mp->m_dalign && igeo->inoalign_mask &&
2885 !(mp->m_dalign & igeo->inoalign_mask))
2886 igeo->ialloc_align = mp->m_dalign;
2888 igeo->ialloc_align = 0;
2891 /* Compute the location of the root directory inode that is laid out by mkfs. */
2893 xfs_ialloc_calc_rootino(
2894 struct xfs_mount *mp,
2897 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2898 xfs_agblock_t first_bno;
2901 * Pre-calculate the geometry of AG 0. We know what it looks like
2902 * because libxfs knows how to create allocation groups now.
2904 * first_bno is the first block in which mkfs could possibly have
2905 * allocated the root directory inode, once we factor in the metadata
2906 * that mkfs formats before it. Namely, the four AG headers...
2908 first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize);
2910 /* ...the two free space btree roots... */
2913 /* ...the inode btree root... */
2916 /* ...the initial AGFL... */
2917 first_bno += xfs_alloc_min_freelist(mp, NULL);
2919 /* ...the free inode btree root... */
2920 if (xfs_has_finobt(mp))
2923 /* ...the reverse mapping btree root... */
2924 if (xfs_has_rmapbt(mp))
2927 /* ...the reference count btree... */
2928 if (xfs_has_reflink(mp))
2932 * ...and the log, if it is allocated in the first allocation group.
2934 * This can happen with filesystems that only have a single
2935 * allocation group, or very odd geometries created by old mkfs
2936 * versions on very small filesystems.
2938 if (xfs_ag_contains_log(mp, 0))
2939 first_bno += mp->m_sb.sb_logblocks;
2942 * Now round first_bno up to whatever allocation alignment is given
2943 * by the filesystem or was passed in.
2945 if (xfs_has_dalign(mp) && igeo->ialloc_align > 0)
2946 first_bno = roundup(first_bno, sunit);
2947 else if (xfs_has_align(mp) &&
2948 mp->m_sb.sb_inoalignmt > 1)
2949 first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt);
2951 return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno));
2955 * Ensure there are not sparse inode clusters that cross the new EOAG.
2957 * This is a no-op for non-spinode filesystems since clusters are always fully
2958 * allocated and checking the bnobt suffices. However, a spinode filesystem
2959 * could have a record where the upper inodes are free blocks. If those blocks
2960 * were removed from the filesystem, the inode record would extend beyond EOAG,
2961 * which will be flagged as corruption.
2964 xfs_ialloc_check_shrink(
2965 struct xfs_perag *pag,
2966 struct xfs_trans *tp,
2967 struct xfs_buf *agibp,
2968 xfs_agblock_t new_length)
2970 struct xfs_inobt_rec_incore rec;
2971 struct xfs_btree_cur *cur;
2976 if (!xfs_has_sparseinodes(pag->pag_mount))
2979 cur = xfs_inobt_init_cursor(pag, tp, agibp, XFS_BTNUM_INO);
2981 /* Look up the inobt record that would correspond to the new EOFS. */
2982 agino = XFS_AGB_TO_AGINO(pag->pag_mount, new_length);
2983 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has);
2987 error = xfs_inobt_get_rec(cur, &rec, &has);
2992 error = -EFSCORRUPTED;
2996 /* If the record covers inodes that would be beyond EOFS, bail out. */
2997 if (rec.ir_startino + XFS_INODES_PER_CHUNK > agino) {
3002 xfs_btree_del_cursor(cur, error);