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);
99 * Get the data from the pointed-to record.
103 struct xfs_btree_cur *cur,
104 struct xfs_inobt_rec_incore *irec,
107 struct xfs_mount *mp = cur->bc_mp;
108 union xfs_btree_rec *rec;
112 error = xfs_btree_get_rec(cur, &rec, stat);
113 if (error || *stat == 0)
116 xfs_inobt_btrec_to_irec(mp, rec, irec);
118 if (!xfs_verify_agino(cur->bc_ag.pag, irec->ir_startino))
120 if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
121 irec->ir_count > XFS_INODES_PER_CHUNK)
123 if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
126 /* if there are no holes, return the first available offset */
127 if (!xfs_inobt_issparse(irec->ir_holemask))
128 realfree = irec->ir_free;
130 realfree = irec->ir_free & xfs_inobt_irec_to_allocmask(irec);
131 if (hweight64(realfree) != irec->ir_freecount)
138 "%s Inode BTree record corruption in AG %d detected!",
139 cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free",
140 cur->bc_ag.pag->pag_agno);
142 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
143 irec->ir_startino, irec->ir_count, irec->ir_freecount,
144 irec->ir_free, irec->ir_holemask);
145 return -EFSCORRUPTED;
149 * Insert a single inobt record. Cursor must already point to desired location.
152 xfs_inobt_insert_rec(
153 struct xfs_btree_cur *cur,
160 cur->bc_rec.i.ir_holemask = holemask;
161 cur->bc_rec.i.ir_count = count;
162 cur->bc_rec.i.ir_freecount = freecount;
163 cur->bc_rec.i.ir_free = free;
164 return xfs_btree_insert(cur, stat);
168 * Insert records describing a newly allocated inode chunk into the inobt.
172 struct xfs_perag *pag,
173 struct xfs_trans *tp,
174 struct xfs_buf *agbp,
179 struct xfs_btree_cur *cur;
184 cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
186 for (thisino = newino;
187 thisino < newino + newlen;
188 thisino += XFS_INODES_PER_CHUNK) {
189 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
191 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
196 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
197 XFS_INODES_PER_CHUNK,
198 XFS_INODES_PER_CHUNK,
199 XFS_INOBT_ALL_FREE, &i);
201 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
207 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
213 * Verify that the number of free inodes in the AGI is correct.
217 xfs_check_agi_freecount(
218 struct xfs_btree_cur *cur)
220 if (cur->bc_nlevels == 1) {
221 xfs_inobt_rec_incore_t rec;
226 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
231 error = xfs_inobt_get_rec(cur, &rec, &i);
236 freecount += rec.ir_freecount;
237 error = xfs_btree_increment(cur, 0, &i);
243 if (!xfs_is_shutdown(cur->bc_mp))
244 ASSERT(freecount == cur->bc_ag.pag->pagi_freecount);
249 #define xfs_check_agi_freecount(cur) 0
253 * Initialise a new set of inodes. When called without a transaction context
254 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
255 * than logging them (which in a transaction context puts them into the AIL
256 * for writeback rather than the xfsbufd queue).
259 xfs_ialloc_inode_init(
260 struct xfs_mount *mp,
261 struct xfs_trans *tp,
262 struct list_head *buffer_list,
266 xfs_agblock_t length,
269 struct xfs_buf *fbuf;
270 struct xfs_dinode *free;
279 * Loop over the new block(s), filling in the inodes. For small block
280 * sizes, manipulate the inodes in buffers which are multiples of the
283 nbufs = length / M_IGEO(mp)->blocks_per_cluster;
286 * Figure out what version number to use in the inodes we create. If
287 * the superblock version has caught up to the one that supports the new
288 * inode format, then use the new inode version. Otherwise use the old
289 * version so that old kernels will continue to be able to use the file
292 * For v3 inodes, we also need to write the inode number into the inode,
293 * so calculate the first inode number of the chunk here as
294 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
295 * across multiple filesystem blocks (such as a cluster) and so cannot
296 * be used in the cluster buffer loop below.
298 * Further, because we are writing the inode directly into the buffer
299 * and calculating a CRC on the entire inode, we have ot log the entire
300 * inode so that the entire range the CRC covers is present in the log.
301 * That means for v3 inode we log the entire buffer rather than just the
304 if (xfs_has_v3inodes(mp)) {
306 ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
309 * log the initialisation that is about to take place as an
310 * logical operation. This means the transaction does not
311 * need to log the physical changes to the inode buffers as log
312 * recovery will know what initialisation is actually needed.
313 * Hence we only need to log the buffers as "ordered" buffers so
314 * they track in the AIL as if they were physically logged.
317 xfs_icreate_log(tp, agno, agbno, icount,
318 mp->m_sb.sb_inodesize, length, gen);
322 for (j = 0; j < nbufs; j++) {
326 d = XFS_AGB_TO_DADDR(mp, agno, agbno +
327 (j * M_IGEO(mp)->blocks_per_cluster));
328 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
329 mp->m_bsize * M_IGEO(mp)->blocks_per_cluster,
330 XBF_UNMAPPED, &fbuf);
334 /* Initialize the inode buffers and log them appropriately. */
335 fbuf->b_ops = &xfs_inode_buf_ops;
336 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
337 for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
338 int ioffset = i << mp->m_sb.sb_inodelog;
340 free = xfs_make_iptr(mp, fbuf, i);
341 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
342 free->di_version = version;
343 free->di_gen = cpu_to_be32(gen);
344 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
347 free->di_ino = cpu_to_be64(ino);
349 uuid_copy(&free->di_uuid,
350 &mp->m_sb.sb_meta_uuid);
351 xfs_dinode_calc_crc(mp, free);
353 /* just log the inode core */
354 xfs_trans_log_buf(tp, fbuf, ioffset,
355 ioffset + XFS_DINODE_SIZE(mp) - 1);
361 * Mark the buffer as an inode allocation buffer so it
362 * sticks in AIL at the point of this allocation
363 * transaction. This ensures the they are on disk before
364 * the tail of the log can be moved past this
365 * transaction (i.e. by preventing relogging from moving
366 * it forward in the log).
368 xfs_trans_inode_alloc_buf(tp, fbuf);
371 * Mark the buffer as ordered so that they are
372 * not physically logged in the transaction but
373 * still tracked in the AIL as part of the
374 * transaction and pin the log appropriately.
376 xfs_trans_ordered_buf(tp, fbuf);
379 fbuf->b_flags |= XBF_DONE;
380 xfs_buf_delwri_queue(fbuf, buffer_list);
388 * Align startino and allocmask for a recently allocated sparse chunk such that
389 * they are fit for insertion (or merge) into the on-disk inode btrees.
393 * When enabled, sparse inode support increases the inode alignment from cluster
394 * size to inode chunk size. This means that the minimum range between two
395 * non-adjacent inode records in the inobt is large enough for a full inode
396 * record. This allows for cluster sized, cluster aligned block allocation
397 * without need to worry about whether the resulting inode record overlaps with
398 * another record in the tree. Without this basic rule, we would have to deal
399 * with the consequences of overlap by potentially undoing recent allocations in
400 * the inode allocation codepath.
402 * Because of this alignment rule (which is enforced on mount), there are two
403 * inobt possibilities for newly allocated sparse chunks. One is that the
404 * aligned inode record for the chunk covers a range of inodes not already
405 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
406 * other is that a record already exists at the aligned startino that considers
407 * the newly allocated range as sparse. In the latter case, record content is
408 * merged in hope that sparse inode chunks fill to full chunks over time.
411 xfs_align_sparse_ino(
412 struct xfs_mount *mp,
413 xfs_agino_t *startino,
420 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
421 mod = agbno % mp->m_sb.sb_inoalignmt;
425 /* calculate the inode offset and align startino */
426 offset = XFS_AGB_TO_AGINO(mp, mod);
430 * Since startino has been aligned down, left shift allocmask such that
431 * it continues to represent the same physical inodes relative to the
434 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
438 * Determine whether the source inode record can merge into the target. Both
439 * records must be sparse, the inode ranges must match and there must be no
440 * allocation overlap between the records.
443 __xfs_inobt_can_merge(
444 struct xfs_inobt_rec_incore *trec, /* tgt record */
445 struct xfs_inobt_rec_incore *srec) /* src record */
450 /* records must cover the same inode range */
451 if (trec->ir_startino != srec->ir_startino)
454 /* both records must be sparse */
455 if (!xfs_inobt_issparse(trec->ir_holemask) ||
456 !xfs_inobt_issparse(srec->ir_holemask))
459 /* both records must track some inodes */
460 if (!trec->ir_count || !srec->ir_count)
463 /* can't exceed capacity of a full record */
464 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
467 /* verify there is no allocation overlap */
468 talloc = xfs_inobt_irec_to_allocmask(trec);
469 salloc = xfs_inobt_irec_to_allocmask(srec);
477 * Merge the source inode record into the target. The caller must call
478 * __xfs_inobt_can_merge() to ensure the merge is valid.
481 __xfs_inobt_rec_merge(
482 struct xfs_inobt_rec_incore *trec, /* target */
483 struct xfs_inobt_rec_incore *srec) /* src */
485 ASSERT(trec->ir_startino == srec->ir_startino);
487 /* combine the counts */
488 trec->ir_count += srec->ir_count;
489 trec->ir_freecount += srec->ir_freecount;
492 * Merge the holemask and free mask. For both fields, 0 bits refer to
493 * allocated inodes. We combine the allocated ranges with bitwise AND.
495 trec->ir_holemask &= srec->ir_holemask;
496 trec->ir_free &= srec->ir_free;
500 * Insert a new sparse inode chunk into the associated inode btree. The inode
501 * record for the sparse chunk is pre-aligned to a startino that should match
502 * any pre-existing sparse inode record in the tree. This allows sparse chunks
505 * This function supports two modes of handling preexisting records depending on
506 * the merge flag. If merge is true, the provided record is merged with the
507 * existing record and updated in place. The merged record is returned in nrec.
508 * If merge is false, an existing record is replaced with the provided record.
509 * If no preexisting record exists, the provided record is always inserted.
511 * It is considered corruption if a merge is requested and not possible. Given
512 * the sparse inode alignment constraints, this should never happen.
515 xfs_inobt_insert_sprec(
516 struct xfs_perag *pag,
517 struct xfs_trans *tp,
518 struct xfs_buf *agbp,
520 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
521 bool merge) /* merge or replace */
523 struct xfs_mount *mp = pag->pag_mount;
524 struct xfs_btree_cur *cur;
527 struct xfs_inobt_rec_incore rec;
529 cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
531 /* the new record is pre-aligned so we know where to look */
532 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
535 /* if nothing there, insert a new record and return */
537 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
538 nrec->ir_count, nrec->ir_freecount,
542 if (XFS_IS_CORRUPT(mp, i != 1)) {
543 error = -EFSCORRUPTED;
551 * A record exists at this startino. Merge or replace the record
552 * depending on what we've been asked to do.
555 error = xfs_inobt_get_rec(cur, &rec, &i);
558 if (XFS_IS_CORRUPT(mp, i != 1)) {
559 error = -EFSCORRUPTED;
562 if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) {
563 error = -EFSCORRUPTED;
568 * This should never fail. If we have coexisting records that
569 * cannot merge, something is seriously wrong.
571 if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) {
572 error = -EFSCORRUPTED;
576 trace_xfs_irec_merge_pre(mp, pag->pag_agno, rec.ir_startino,
577 rec.ir_holemask, nrec->ir_startino,
580 /* merge to nrec to output the updated record */
581 __xfs_inobt_rec_merge(nrec, &rec);
583 trace_xfs_irec_merge_post(mp, pag->pag_agno, nrec->ir_startino,
586 error = xfs_inobt_rec_check_count(mp, nrec);
591 error = xfs_inobt_update(cur, nrec);
596 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
599 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
604 * Allocate new inodes in the allocation group specified by agbp. Returns 0 if
605 * inodes were allocated in this AG; -EAGAIN if there was no space in this AG so
606 * the caller knows it can try another AG, a hard -ENOSPC when over the maximum
607 * inode count threshold, or the usual negative error code for other errors.
611 struct xfs_perag *pag,
612 struct xfs_trans *tp,
613 struct xfs_buf *agbp)
616 struct xfs_alloc_arg args;
618 xfs_agino_t newino; /* new first inode's number */
619 xfs_agino_t newlen; /* new number of inodes */
620 int isaligned = 0; /* inode allocation at stripe */
622 /* init. to full chunk */
623 struct xfs_inobt_rec_incore rec;
624 struct xfs_ino_geometry *igeo = M_IGEO(tp->t_mountp);
625 uint16_t allocmask = (uint16_t) -1;
628 memset(&args, 0, sizeof(args));
630 args.mp = tp->t_mountp;
631 args.fsbno = NULLFSBLOCK;
632 args.oinfo = XFS_RMAP_OINFO_INODES;
636 /* randomly do sparse inode allocations */
637 if (xfs_has_sparseinodes(tp->t_mountp) &&
638 igeo->ialloc_min_blks < igeo->ialloc_blks)
639 do_sparse = get_random_u32_below(2);
643 * Locking will ensure that we don't have two callers in here
646 newlen = igeo->ialloc_inos;
647 if (igeo->maxicount &&
648 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
651 args.minlen = args.maxlen = igeo->ialloc_blks;
653 * First try to allocate inodes contiguous with the last-allocated
654 * chunk of inodes. If the filesystem is striped, this will fill
655 * an entire stripe unit with inodes.
658 newino = be32_to_cpu(agi->agi_newino);
659 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
663 if (likely(newino != NULLAGINO &&
664 (args.agbno < be32_to_cpu(agi->agi_length)))) {
668 * We need to take into account alignment here to ensure that
669 * we don't modify the free list if we fail to have an exact
670 * block. If we don't have an exact match, and every oher
671 * attempt allocation attempt fails, we'll end up cancelling
672 * a dirty transaction and shutting down.
674 * For an exact allocation, alignment must be 1,
675 * however we need to take cluster alignment into account when
676 * fixing up the freelist. Use the minalignslop field to
677 * indicate that extra blocks might be required for alignment,
678 * but not to use them in the actual exact allocation.
681 args.minalignslop = igeo->cluster_align - 1;
683 /* Allow space for the inode btree to split. */
684 args.minleft = igeo->inobt_maxlevels;
685 error = xfs_alloc_vextent_exact_bno(&args,
686 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
692 * This request might have dirtied the transaction if the AG can
693 * satisfy the request, but the exact block was not available.
694 * If the allocation did fail, subsequent requests will relax
695 * the exact agbno requirement and increase the alignment
696 * instead. It is critical that the total size of the request
697 * (len + alignment + slop) does not increase from this point
698 * on, so reset minalignslop to ensure it is not included in
699 * subsequent requests.
701 args.minalignslop = 0;
704 if (unlikely(args.fsbno == NULLFSBLOCK)) {
706 * Set the alignment for the allocation.
707 * If stripe alignment is turned on then align at stripe unit
709 * If the cluster size is smaller than a filesystem block
710 * then we're doing I/O for inodes in filesystem block size
711 * pieces, so don't need alignment anyway.
714 if (igeo->ialloc_align) {
715 ASSERT(!xfs_has_noalign(args.mp));
716 args.alignment = args.mp->m_dalign;
719 args.alignment = igeo->cluster_align;
721 * Allocate a fixed-size extent of inodes.
725 * Allow space for the inode btree to split.
727 args.minleft = igeo->inobt_maxlevels;
728 error = xfs_alloc_vextent_near_bno(&args,
729 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
730 be32_to_cpu(agi->agi_root)));
736 * If stripe alignment is turned on, then try again with cluster
739 if (isaligned && args.fsbno == NULLFSBLOCK) {
740 args.alignment = igeo->cluster_align;
741 error = xfs_alloc_vextent_near_bno(&args,
742 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
743 be32_to_cpu(agi->agi_root)));
749 * Finally, try a sparse allocation if the filesystem supports it and
750 * the sparse allocation length is smaller than a full chunk.
752 if (xfs_has_sparseinodes(args.mp) &&
753 igeo->ialloc_min_blks < igeo->ialloc_blks &&
754 args.fsbno == NULLFSBLOCK) {
756 args.alignment = args.mp->m_sb.sb_spino_align;
759 args.minlen = igeo->ialloc_min_blks;
760 args.maxlen = args.minlen;
763 * The inode record will be aligned to full chunk size. We must
764 * prevent sparse allocation from AG boundaries that result in
765 * invalid inode records, such as records that start at agbno 0
766 * or extend beyond the AG.
768 * Set min agbno to the first aligned, non-zero agbno and max to
769 * the last aligned agbno that is at least one full chunk from
772 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
773 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
774 args.mp->m_sb.sb_inoalignmt) -
777 error = xfs_alloc_vextent_near_bno(&args,
778 XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
779 be32_to_cpu(agi->agi_root)));
783 newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
784 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
785 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
788 if (args.fsbno == NULLFSBLOCK)
791 ASSERT(args.len == args.minlen);
794 * Stamp and write the inode buffers.
796 * Seed the new inode cluster with a random generation number. This
797 * prevents short-term reuse of generation numbers if a chunk is
798 * freed and then immediately reallocated. We use random numbers
799 * rather than a linear progression to prevent the next generation
800 * number from being easily guessable.
802 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, pag->pag_agno,
803 args.agbno, args.len, get_random_u32());
808 * Convert the results.
810 newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
812 if (xfs_inobt_issparse(~allocmask)) {
814 * We've allocated a sparse chunk. Align the startino and mask.
816 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
818 rec.ir_startino = newino;
819 rec.ir_holemask = ~allocmask;
820 rec.ir_count = newlen;
821 rec.ir_freecount = newlen;
822 rec.ir_free = XFS_INOBT_ALL_FREE;
825 * Insert the sparse record into the inobt and allow for a merge
826 * if necessary. If a merge does occur, rec is updated to the
829 error = xfs_inobt_insert_sprec(pag, tp, agbp,
830 XFS_BTNUM_INO, &rec, true);
831 if (error == -EFSCORRUPTED) {
833 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
834 XFS_AGINO_TO_INO(args.mp, pag->pag_agno,
836 rec.ir_holemask, rec.ir_count);
837 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
843 * We can't merge the part we've just allocated as for the inobt
844 * due to finobt semantics. The original record may or may not
845 * exist independent of whether physical inodes exist in this
848 * We must update the finobt record based on the inobt record.
849 * rec contains the fully merged and up to date inobt record
850 * from the previous call. Set merge false to replace any
851 * existing record with this one.
853 if (xfs_has_finobt(args.mp)) {
854 error = xfs_inobt_insert_sprec(pag, tp, agbp,
855 XFS_BTNUM_FINO, &rec, false);
860 /* full chunk - insert new records to both btrees */
861 error = xfs_inobt_insert(pag, tp, agbp, newino, newlen,
866 if (xfs_has_finobt(args.mp)) {
867 error = xfs_inobt_insert(pag, tp, agbp, newino,
868 newlen, XFS_BTNUM_FINO);
875 * Update AGI counts and newino.
877 be32_add_cpu(&agi->agi_count, newlen);
878 be32_add_cpu(&agi->agi_freecount, newlen);
879 pag->pagi_freecount += newlen;
880 pag->pagi_count += newlen;
881 agi->agi_newino = cpu_to_be32(newino);
884 * Log allocation group header fields
886 xfs_ialloc_log_agi(tp, agbp,
887 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
889 * Modify/log superblock values for inode count and inode free count.
891 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
892 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
897 * Try to retrieve the next record to the left/right from the current one.
901 struct xfs_btree_cur *cur,
902 xfs_inobt_rec_incore_t *rec,
910 error = xfs_btree_decrement(cur, 0, &i);
912 error = xfs_btree_increment(cur, 0, &i);
918 error = xfs_inobt_get_rec(cur, rec, &i);
921 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
922 return -EFSCORRUPTED;
930 struct xfs_btree_cur *cur,
932 xfs_inobt_rec_incore_t *rec,
938 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
943 error = xfs_inobt_get_rec(cur, rec, &i);
946 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
947 return -EFSCORRUPTED;
954 * Return the offset of the first free inode in the record. If the inode chunk
955 * is sparsely allocated, we convert the record holemask to inode granularity
956 * and mask off the unallocated regions from the inode free mask.
959 xfs_inobt_first_free_inode(
960 struct xfs_inobt_rec_incore *rec)
962 xfs_inofree_t realfree;
964 /* if there are no holes, return the first available offset */
965 if (!xfs_inobt_issparse(rec->ir_holemask))
966 return xfs_lowbit64(rec->ir_free);
968 realfree = xfs_inobt_irec_to_allocmask(rec);
969 realfree &= rec->ir_free;
971 return xfs_lowbit64(realfree);
975 * Allocate an inode using the inobt-only algorithm.
978 xfs_dialloc_ag_inobt(
979 struct xfs_perag *pag,
980 struct xfs_trans *tp,
981 struct xfs_buf *agbp,
985 struct xfs_mount *mp = tp->t_mountp;
986 struct xfs_agi *agi = agbp->b_addr;
987 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
988 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
989 struct xfs_btree_cur *cur, *tcur;
990 struct xfs_inobt_rec_incore rec, trec;
995 int searchdistance = 10;
997 ASSERT(xfs_perag_initialised_agi(pag));
998 ASSERT(xfs_perag_allows_inodes(pag));
999 ASSERT(pag->pagi_freecount > 0);
1002 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1004 * If pagino is 0 (this is the root inode allocation) use newino.
1005 * This must work because we've just allocated some.
1008 pagino = be32_to_cpu(agi->agi_newino);
1010 error = xfs_check_agi_freecount(cur);
1015 * If in the same AG as the parent, try to get near the parent.
1017 if (pagno == pag->pag_agno) {
1018 int doneleft; /* done, to the left */
1019 int doneright; /* done, to the right */
1021 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1024 if (XFS_IS_CORRUPT(mp, i != 1)) {
1025 error = -EFSCORRUPTED;
1029 error = xfs_inobt_get_rec(cur, &rec, &j);
1032 if (XFS_IS_CORRUPT(mp, j != 1)) {
1033 error = -EFSCORRUPTED;
1037 if (rec.ir_freecount > 0) {
1039 * Found a free inode in the same chunk
1040 * as the parent, done.
1047 * In the same AG as parent, but parent's chunk is full.
1050 /* duplicate the cursor, search left & right simultaneously */
1051 error = xfs_btree_dup_cursor(cur, &tcur);
1056 * Skip to last blocks looked up if same parent inode.
1058 if (pagino != NULLAGINO &&
1059 pag->pagl_pagino == pagino &&
1060 pag->pagl_leftrec != NULLAGINO &&
1061 pag->pagl_rightrec != NULLAGINO) {
1062 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1067 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1072 /* search left with tcur, back up 1 record */
1073 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1077 /* search right with cur, go forward 1 record. */
1078 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1084 * Loop until we find an inode chunk with a free inode.
1086 while (--searchdistance > 0 && (!doneleft || !doneright)) {
1087 int useleft; /* using left inode chunk this time */
1089 /* figure out the closer block if both are valid. */
1090 if (!doneleft && !doneright) {
1092 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1093 rec.ir_startino - pagino;
1095 useleft = !doneleft;
1098 /* free inodes to the left? */
1099 if (useleft && trec.ir_freecount) {
1100 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1103 pag->pagl_leftrec = trec.ir_startino;
1104 pag->pagl_rightrec = rec.ir_startino;
1105 pag->pagl_pagino = pagino;
1110 /* free inodes to the right? */
1111 if (!useleft && rec.ir_freecount) {
1112 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1114 pag->pagl_leftrec = trec.ir_startino;
1115 pag->pagl_rightrec = rec.ir_startino;
1116 pag->pagl_pagino = pagino;
1120 /* get next record to check */
1122 error = xfs_ialloc_next_rec(tcur, &trec,
1125 error = xfs_ialloc_next_rec(cur, &rec,
1132 if (searchdistance <= 0) {
1134 * Not in range - save last search
1135 * location and allocate a new inode
1137 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1138 pag->pagl_leftrec = trec.ir_startino;
1139 pag->pagl_rightrec = rec.ir_startino;
1140 pag->pagl_pagino = pagino;
1144 * We've reached the end of the btree. because
1145 * we are only searching a small chunk of the
1146 * btree each search, there is obviously free
1147 * inodes closer to the parent inode than we
1148 * are now. restart the search again.
1150 pag->pagl_pagino = NULLAGINO;
1151 pag->pagl_leftrec = NULLAGINO;
1152 pag->pagl_rightrec = NULLAGINO;
1153 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1154 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1160 * In a different AG from the parent.
1161 * See if the most recently allocated block has any free.
1163 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1164 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1170 error = xfs_inobt_get_rec(cur, &rec, &j);
1174 if (j == 1 && rec.ir_freecount > 0) {
1176 * The last chunk allocated in the group
1177 * still has a free inode.
1185 * None left in the last group, search the whole AG
1187 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1190 if (XFS_IS_CORRUPT(mp, i != 1)) {
1191 error = -EFSCORRUPTED;
1196 error = xfs_inobt_get_rec(cur, &rec, &i);
1199 if (XFS_IS_CORRUPT(mp, i != 1)) {
1200 error = -EFSCORRUPTED;
1203 if (rec.ir_freecount > 0)
1205 error = xfs_btree_increment(cur, 0, &i);
1208 if (XFS_IS_CORRUPT(mp, i != 1)) {
1209 error = -EFSCORRUPTED;
1215 offset = xfs_inobt_first_free_inode(&rec);
1216 ASSERT(offset >= 0);
1217 ASSERT(offset < XFS_INODES_PER_CHUNK);
1218 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1219 XFS_INODES_PER_CHUNK) == 0);
1220 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1221 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1223 error = xfs_inobt_update(cur, &rec);
1226 be32_add_cpu(&agi->agi_freecount, -1);
1227 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1228 pag->pagi_freecount--;
1230 error = xfs_check_agi_freecount(cur);
1234 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1235 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1239 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1241 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1246 * Use the free inode btree to allocate an inode based on distance from the
1247 * parent. Note that the provided cursor may be deleted and replaced.
1250 xfs_dialloc_ag_finobt_near(
1252 struct xfs_btree_cur **ocur,
1253 struct xfs_inobt_rec_incore *rec)
1255 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1256 struct xfs_btree_cur *rcur; /* right search cursor */
1257 struct xfs_inobt_rec_incore rrec;
1261 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1266 error = xfs_inobt_get_rec(lcur, rec, &i);
1269 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1))
1270 return -EFSCORRUPTED;
1273 * See if we've landed in the parent inode record. The finobt
1274 * only tracks chunks with at least one free inode, so record
1275 * existence is enough.
1277 if (pagino >= rec->ir_startino &&
1278 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1282 error = xfs_btree_dup_cursor(lcur, &rcur);
1286 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1290 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1293 if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) {
1294 error = -EFSCORRUPTED;
1299 if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) {
1300 error = -EFSCORRUPTED;
1303 if (i == 1 && j == 1) {
1305 * Both the left and right records are valid. Choose the closer
1306 * inode chunk to the target.
1308 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1309 (rrec.ir_startino - pagino)) {
1311 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1314 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1316 } else if (j == 1) {
1317 /* only the right record is valid */
1319 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1321 } else if (i == 1) {
1322 /* only the left record is valid */
1323 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1329 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1334 * Use the free inode btree to find a free inode based on a newino hint. If
1335 * the hint is NULL, find the first free inode in the AG.
1338 xfs_dialloc_ag_finobt_newino(
1339 struct xfs_agi *agi,
1340 struct xfs_btree_cur *cur,
1341 struct xfs_inobt_rec_incore *rec)
1346 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1347 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1352 error = xfs_inobt_get_rec(cur, rec, &i);
1355 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1356 return -EFSCORRUPTED;
1362 * Find the first inode available in the AG.
1364 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1367 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1368 return -EFSCORRUPTED;
1370 error = xfs_inobt_get_rec(cur, rec, &i);
1373 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1374 return -EFSCORRUPTED;
1380 * Update the inobt based on a modification made to the finobt. Also ensure that
1381 * the records from both trees are equivalent post-modification.
1384 xfs_dialloc_ag_update_inobt(
1385 struct xfs_btree_cur *cur, /* inobt cursor */
1386 struct xfs_inobt_rec_incore *frec, /* finobt record */
1387 int offset) /* inode offset */
1389 struct xfs_inobt_rec_incore rec;
1393 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1396 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1397 return -EFSCORRUPTED;
1399 error = xfs_inobt_get_rec(cur, &rec, &i);
1402 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1403 return -EFSCORRUPTED;
1404 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1405 XFS_INODES_PER_CHUNK) == 0);
1407 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1410 if (XFS_IS_CORRUPT(cur->bc_mp,
1411 rec.ir_free != frec->ir_free ||
1412 rec.ir_freecount != frec->ir_freecount))
1413 return -EFSCORRUPTED;
1415 return xfs_inobt_update(cur, &rec);
1419 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1420 * back to the inobt search algorithm.
1422 * The caller selected an AG for us, and made sure that free inodes are
1427 struct xfs_perag *pag,
1428 struct xfs_trans *tp,
1429 struct xfs_buf *agbp,
1433 struct xfs_mount *mp = tp->t_mountp;
1434 struct xfs_agi *agi = agbp->b_addr;
1435 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1436 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1437 struct xfs_btree_cur *cur; /* finobt cursor */
1438 struct xfs_btree_cur *icur; /* inobt cursor */
1439 struct xfs_inobt_rec_incore rec;
1445 if (!xfs_has_finobt(mp))
1446 return xfs_dialloc_ag_inobt(pag, tp, agbp, parent, inop);
1449 * If pagino is 0 (this is the root inode allocation) use newino.
1450 * This must work because we've just allocated some.
1453 pagino = be32_to_cpu(agi->agi_newino);
1455 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
1457 error = xfs_check_agi_freecount(cur);
1462 * The search algorithm depends on whether we're in the same AG as the
1463 * parent. If so, find the closest available inode to the parent. If
1464 * not, consider the agi hint or find the first free inode in the AG.
1466 if (pag->pag_agno == pagno)
1467 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1469 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1473 offset = xfs_inobt_first_free_inode(&rec);
1474 ASSERT(offset >= 0);
1475 ASSERT(offset < XFS_INODES_PER_CHUNK);
1476 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1477 XFS_INODES_PER_CHUNK) == 0);
1478 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1481 * Modify or remove the finobt record.
1483 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1485 if (rec.ir_freecount)
1486 error = xfs_inobt_update(cur, &rec);
1488 error = xfs_btree_delete(cur, &i);
1493 * The finobt has now been updated appropriately. We haven't updated the
1494 * agi and superblock yet, so we can create an inobt cursor and validate
1495 * the original freecount. If all is well, make the equivalent update to
1496 * the inobt using the finobt record and offset information.
1498 icur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1500 error = xfs_check_agi_freecount(icur);
1504 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1509 * Both trees have now been updated. We must update the perag and
1510 * superblock before we can check the freecount for each btree.
1512 be32_add_cpu(&agi->agi_freecount, -1);
1513 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1514 pag->pagi_freecount--;
1516 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1518 error = xfs_check_agi_freecount(icur);
1521 error = xfs_check_agi_freecount(cur);
1525 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1526 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1531 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1533 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1539 struct xfs_trans **tpp,
1540 struct xfs_buf *agibp)
1542 struct xfs_trans *tp = *tpp;
1543 struct xfs_dquot_acct *dqinfo;
1547 * Hold to on to the agibp across the commit so no other allocation can
1548 * come in and take the free inodes we just allocated for our caller.
1550 xfs_trans_bhold(tp, agibp);
1553 * We want the quota changes to be associated with the next transaction,
1554 * NOT this one. So, detach the dqinfo from this and attach it to the
1557 dqinfo = tp->t_dqinfo;
1558 tp->t_dqinfo = NULL;
1560 error = xfs_trans_roll(&tp);
1562 /* Re-attach the quota info that we detached from prev trx. */
1563 tp->t_dqinfo = dqinfo;
1566 * Join the buffer even on commit error so that the buffer is released
1567 * when the caller cancels the transaction and doesn't have to handle
1568 * this error case specially.
1570 xfs_trans_bjoin(tp, agibp);
1576 xfs_dialloc_good_ag(
1577 struct xfs_perag *pag,
1578 struct xfs_trans *tp,
1583 struct xfs_mount *mp = tp->t_mountp;
1585 xfs_extlen_t longest = 0;
1591 if (!xfs_perag_allows_inodes(pag))
1594 if (!xfs_perag_initialised_agi(pag)) {
1595 error = xfs_ialloc_read_agi(pag, tp, NULL);
1600 if (pag->pagi_freecount)
1605 if (!xfs_perag_initialised_agf(pag)) {
1606 error = xfs_alloc_read_agf(pag, tp, flags, NULL);
1612 * Check that there is enough free space for the file plus a chunk of
1613 * inodes if we need to allocate some. If this is the first pass across
1614 * the AGs, take into account the potential space needed for alignment
1615 * of inode chunks when checking the longest contiguous free space in
1616 * the AG - this prevents us from getting ENOSPC because we have free
1617 * space larger than ialloc_blks but alignment constraints prevent us
1620 * If we can't find an AG with space for full alignment slack to be
1621 * taken into account, we must be near ENOSPC in all AGs. Hence we
1622 * don't include alignment for the second pass and so if we fail
1623 * allocation due to alignment issues then it is most likely a real
1626 * XXX(dgc): this calculation is now bogus thanks to the per-ag
1627 * reservations that xfs_alloc_fix_freelist() now does via
1628 * xfs_alloc_space_available(). When the AG fills up, pagf_freeblks will
1629 * be more than large enough for the check below to succeed, but
1630 * xfs_alloc_space_available() will fail because of the non-zero
1631 * metadata reservation and hence we won't actually be able to allocate
1632 * more inodes in this AG. We do soooo much unnecessary work near ENOSPC
1635 ineed = M_IGEO(mp)->ialloc_min_blks;
1636 if (flags && ineed > 1)
1637 ineed += M_IGEO(mp)->cluster_align;
1638 longest = pag->pagf_longest;
1640 longest = pag->pagf_flcount > 0;
1641 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
1643 if (pag->pagf_freeblks < needspace + ineed || longest < ineed)
1650 struct xfs_perag *pag,
1651 struct xfs_trans **tpp,
1656 struct xfs_buf *agbp;
1661 * Then read in the AGI buffer and recheck with the AGI buffer
1664 error = xfs_ialloc_read_agi(pag, *tpp, &agbp);
1668 if (!pag->pagi_freecount) {
1674 error = xfs_ialloc_ag_alloc(pag, *tpp, agbp);
1679 * We successfully allocated space for an inode cluster in this
1680 * AG. Roll the transaction so that we can allocate one of the
1683 ASSERT(pag->pagi_freecount > 0);
1684 error = xfs_dialloc_roll(tpp, agbp);
1689 /* Allocate an inode in the found AG */
1690 error = xfs_dialloc_ag(pag, *tpp, agbp, parent, &ino);
1696 xfs_trans_brelse(*tpp, agbp);
1701 * Allocate an on-disk inode.
1703 * Mode is used to tell whether the new inode is a directory and hence where to
1704 * locate it. The on-disk inode that is allocated will be returned in @new_ino
1705 * on success, otherwise an error will be set to indicate the failure (e.g.
1710 struct xfs_trans **tpp,
1715 struct xfs_mount *mp = (*tpp)->t_mountp;
1716 xfs_agnumber_t agno;
1718 xfs_agnumber_t start_agno;
1719 struct xfs_perag *pag;
1720 struct xfs_ino_geometry *igeo = M_IGEO(mp);
1721 bool ok_alloc = true;
1722 bool low_space = false;
1724 xfs_ino_t ino = NULLFSINO;
1727 * Directories, symlinks, and regular files frequently allocate at least
1728 * one block, so factor that potential expansion when we examine whether
1729 * an AG has enough space for file creation.
1732 start_agno = (atomic_inc_return(&mp->m_agirotor) - 1) %
1735 start_agno = XFS_INO_TO_AGNO(mp, parent);
1736 if (start_agno >= mp->m_maxagi)
1741 * If we have already hit the ceiling of inode blocks then clear
1742 * ok_alloc so we scan all available agi structures for a free
1745 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1746 * which will sacrifice the preciseness but improve the performance.
1748 if (igeo->maxicount &&
1749 percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1750 > igeo->maxicount) {
1755 * If we are near to ENOSPC, we want to prefer allocation from AGs that
1756 * have free inodes in them rather than use up free space allocating new
1757 * inode chunks. Hence we turn off allocation for the first non-blocking
1758 * pass through the AGs if we are near ENOSPC to consume free inodes
1759 * that we can immediately allocate, but then we allow allocation on the
1760 * second pass if we fail to find an AG with free inodes in it.
1762 if (percpu_counter_read_positive(&mp->m_fdblocks) <
1763 mp->m_low_space[XFS_LOWSP_1_PCNT]) {
1769 * Loop until we find an allocation group that either has free inodes
1770 * or in which we can allocate some inodes. Iterate through the
1771 * allocation groups upward, wrapping at the end.
1773 flags = XFS_ALLOC_FLAG_TRYLOCK;
1775 for_each_perag_wrap_at(mp, start_agno, mp->m_maxagi, agno, pag) {
1776 if (xfs_dialloc_good_ag(pag, *tpp, mode, flags, ok_alloc)) {
1777 error = xfs_dialloc_try_ag(pag, tpp, parent,
1779 if (error != -EAGAIN)
1784 if (xfs_is_shutdown(mp)) {
1785 error = -EFSCORRUPTED;
1790 xfs_perag_rele(pag);
1793 if (ino == NULLFSINO) {
1807 * Free the blocks of an inode chunk. We must consider that the inode chunk
1808 * might be sparse and only free the regions that are allocated as part of the
1812 xfs_difree_inode_chunk(
1813 struct xfs_trans *tp,
1814 xfs_agnumber_t agno,
1815 struct xfs_inobt_rec_incore *rec)
1817 struct xfs_mount *mp = tp->t_mountp;
1818 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp,
1820 int startidx, endidx;
1822 xfs_agblock_t agbno;
1824 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1826 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1827 /* not sparse, calculate extent info directly */
1828 xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
1829 M_IGEO(mp)->ialloc_blks,
1830 &XFS_RMAP_OINFO_INODES);
1834 /* holemask is only 16-bits (fits in an unsigned long) */
1835 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1836 holemask[0] = rec->ir_holemask;
1839 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1840 * holemask and convert the start/end index of each range to an extent.
1841 * We start with the start and end index both pointing at the first 0 in
1844 startidx = endidx = find_first_zero_bit(holemask,
1845 XFS_INOBT_HOLEMASK_BITS);
1846 nextbit = startidx + 1;
1847 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1848 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1851 * If the next zero bit is contiguous, update the end index of
1852 * the current range and continue.
1854 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1855 nextbit == endidx + 1) {
1861 * nextbit is not contiguous with the current end index. Convert
1862 * the current start/end to an extent and add it to the free
1865 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1866 mp->m_sb.sb_inopblock;
1867 contigblk = ((endidx - startidx + 1) *
1868 XFS_INODES_PER_HOLEMASK_BIT) /
1869 mp->m_sb.sb_inopblock;
1871 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1872 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1873 xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
1874 contigblk, &XFS_RMAP_OINFO_INODES);
1876 /* reset range to current bit and carry on... */
1877 startidx = endidx = nextbit;
1886 struct xfs_perag *pag,
1887 struct xfs_trans *tp,
1888 struct xfs_buf *agbp,
1890 struct xfs_icluster *xic,
1891 struct xfs_inobt_rec_incore *orec)
1893 struct xfs_mount *mp = pag->pag_mount;
1894 struct xfs_agi *agi = agbp->b_addr;
1895 struct xfs_btree_cur *cur;
1896 struct xfs_inobt_rec_incore rec;
1902 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1903 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1906 * Initialize the cursor.
1908 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1910 error = xfs_check_agi_freecount(cur);
1915 * Look for the entry describing this inode.
1917 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1918 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1922 if (XFS_IS_CORRUPT(mp, i != 1)) {
1923 error = -EFSCORRUPTED;
1926 error = xfs_inobt_get_rec(cur, &rec, &i);
1928 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1932 if (XFS_IS_CORRUPT(mp, i != 1)) {
1933 error = -EFSCORRUPTED;
1937 * Get the offset in the inode chunk.
1939 off = agino - rec.ir_startino;
1940 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1941 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1943 * Mark the inode free & increment the count.
1945 rec.ir_free |= XFS_INOBT_MASK(off);
1949 * When an inode chunk is free, it becomes eligible for removal. Don't
1950 * remove the chunk if the block size is large enough for multiple inode
1951 * chunks (that might not be free).
1953 if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
1954 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1955 struct xfs_perag *pag = agbp->b_pag;
1957 xic->deleted = true;
1958 xic->first_ino = XFS_AGINO_TO_INO(mp, pag->pag_agno,
1960 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1963 * Remove the inode cluster from the AGI B+Tree, adjust the
1964 * AGI and Superblock inode counts, and mark the disk space
1965 * to be freed when the transaction is committed.
1967 ilen = rec.ir_freecount;
1968 be32_add_cpu(&agi->agi_count, -ilen);
1969 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1970 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1971 pag->pagi_freecount -= ilen - 1;
1972 pag->pagi_count -= ilen;
1973 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1974 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1976 if ((error = xfs_btree_delete(cur, &i))) {
1977 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1982 xfs_difree_inode_chunk(tp, pag->pag_agno, &rec);
1984 xic->deleted = false;
1986 error = xfs_inobt_update(cur, &rec);
1988 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1994 * Change the inode free counts and log the ag/sb changes.
1996 be32_add_cpu(&agi->agi_freecount, 1);
1997 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1998 pag->pagi_freecount++;
1999 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2002 error = xfs_check_agi_freecount(cur);
2007 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2011 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2016 * Free an inode in the free inode btree.
2020 struct xfs_perag *pag,
2021 struct xfs_trans *tp,
2022 struct xfs_buf *agbp,
2024 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2026 struct xfs_mount *mp = pag->pag_mount;
2027 struct xfs_btree_cur *cur;
2028 struct xfs_inobt_rec_incore rec;
2029 int offset = agino - ibtrec->ir_startino;
2033 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
2035 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2040 * If the record does not exist in the finobt, we must have just
2041 * freed an inode in a previously fully allocated chunk. If not,
2042 * something is out of sync.
2044 if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) {
2045 error = -EFSCORRUPTED;
2049 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2051 ibtrec->ir_freecount,
2052 ibtrec->ir_free, &i);
2061 * Read and update the existing record. We could just copy the ibtrec
2062 * across here, but that would defeat the purpose of having redundant
2063 * metadata. By making the modifications independently, we can catch
2064 * corruptions that we wouldn't see if we just copied from one record
2067 error = xfs_inobt_get_rec(cur, &rec, &i);
2070 if (XFS_IS_CORRUPT(mp, i != 1)) {
2071 error = -EFSCORRUPTED;
2075 rec.ir_free |= XFS_INOBT_MASK(offset);
2078 if (XFS_IS_CORRUPT(mp,
2079 rec.ir_free != ibtrec->ir_free ||
2080 rec.ir_freecount != ibtrec->ir_freecount)) {
2081 error = -EFSCORRUPTED;
2086 * The content of inobt records should always match between the inobt
2087 * and finobt. The lifecycle of records in the finobt is different from
2088 * the inobt in that the finobt only tracks records with at least one
2089 * free inode. Hence, if all of the inodes are free and we aren't
2090 * keeping inode chunks permanently on disk, remove the record.
2091 * Otherwise, update the record with the new information.
2093 * Note that we currently can't free chunks when the block size is large
2094 * enough for multiple chunks. Leave the finobt record to remain in sync
2097 if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
2098 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
2099 error = xfs_btree_delete(cur, &i);
2104 error = xfs_inobt_update(cur, &rec);
2110 error = xfs_check_agi_freecount(cur);
2114 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2118 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2123 * Free disk inode. Carefully avoids touching the incore inode, all
2124 * manipulations incore are the caller's responsibility.
2125 * The on-disk inode is not changed by this operation, only the
2126 * btree (free inode mask) is changed.
2130 struct xfs_trans *tp,
2131 struct xfs_perag *pag,
2133 struct xfs_icluster *xic)
2136 xfs_agblock_t agbno; /* block number containing inode */
2137 struct xfs_buf *agbp; /* buffer for allocation group header */
2138 xfs_agino_t agino; /* allocation group inode number */
2139 int error; /* error return value */
2140 struct xfs_mount *mp = tp->t_mountp;
2141 struct xfs_inobt_rec_incore rec;/* btree record */
2144 * Break up inode number into its components.
2146 if (pag->pag_agno != XFS_INO_TO_AGNO(mp, inode)) {
2147 xfs_warn(mp, "%s: agno != pag->pag_agno (%d != %d).",
2148 __func__, XFS_INO_TO_AGNO(mp, inode), pag->pag_agno);
2152 agino = XFS_INO_TO_AGINO(mp, inode);
2153 if (inode != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2154 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2155 __func__, (unsigned long long)inode,
2156 (unsigned long long)XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2160 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2161 if (agbno >= mp->m_sb.sb_agblocks) {
2162 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2163 __func__, agbno, mp->m_sb.sb_agblocks);
2168 * Get the allocation group header.
2170 error = xfs_ialloc_read_agi(pag, tp, &agbp);
2172 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2178 * Fix up the inode allocation btree.
2180 error = xfs_difree_inobt(pag, tp, agbp, agino, xic, &rec);
2185 * Fix up the free inode btree.
2187 if (xfs_has_finobt(mp)) {
2188 error = xfs_difree_finobt(pag, tp, agbp, agino, &rec);
2201 struct xfs_perag *pag,
2202 struct xfs_trans *tp,
2204 xfs_agblock_t agbno,
2205 xfs_agblock_t *chunk_agbno,
2206 xfs_agblock_t *offset_agbno,
2209 struct xfs_mount *mp = pag->pag_mount;
2210 struct xfs_inobt_rec_incore rec;
2211 struct xfs_btree_cur *cur;
2212 struct xfs_buf *agbp;
2216 error = xfs_ialloc_read_agi(pag, tp, &agbp);
2219 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2220 __func__, error, pag->pag_agno);
2225 * Lookup the inode record for the given agino. If the record cannot be
2226 * found, then it's an invalid inode number and we should abort. Once
2227 * we have a record, we need to ensure it contains the inode number
2228 * we are looking up.
2230 cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
2231 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2234 error = xfs_inobt_get_rec(cur, &rec, &i);
2235 if (!error && i == 0)
2239 xfs_trans_brelse(tp, agbp);
2240 xfs_btree_del_cursor(cur, error);
2244 /* check that the returned record contains the required inode */
2245 if (rec.ir_startino > agino ||
2246 rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2249 /* for untrusted inodes check it is allocated first */
2250 if ((flags & XFS_IGET_UNTRUSTED) &&
2251 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2254 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2255 *offset_agbno = agbno - *chunk_agbno;
2260 * Return the location of the inode in imap, for mapping it into a buffer.
2264 struct xfs_perag *pag,
2265 struct xfs_trans *tp,
2266 xfs_ino_t ino, /* inode to locate */
2267 struct xfs_imap *imap, /* location map structure */
2268 uint flags) /* flags for inode btree lookup */
2270 struct xfs_mount *mp = pag->pag_mount;
2271 xfs_agblock_t agbno; /* block number of inode in the alloc group */
2272 xfs_agino_t agino; /* inode number within alloc group */
2273 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
2274 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
2275 int error; /* error code */
2276 int offset; /* index of inode in its buffer */
2277 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
2279 ASSERT(ino != NULLFSINO);
2282 * Split up the inode number into its parts.
2284 agino = XFS_INO_TO_AGINO(mp, ino);
2285 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2286 if (agbno >= mp->m_sb.sb_agblocks ||
2287 ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2291 * Don't output diagnostic information for untrusted inodes
2292 * as they can be invalid without implying corruption.
2294 if (flags & XFS_IGET_UNTRUSTED)
2296 if (agbno >= mp->m_sb.sb_agblocks) {
2298 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2299 __func__, (unsigned long long)agbno,
2300 (unsigned long)mp->m_sb.sb_agblocks);
2302 if (ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2304 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2306 XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2314 * For bulkstat and handle lookups, we have an untrusted inode number
2315 * that we have to verify is valid. We cannot do this just by reading
2316 * the inode buffer as it may have been unlinked and removed leaving
2317 * inodes in stale state on disk. Hence we have to do a btree lookup
2318 * in all cases where an untrusted inode number is passed.
2320 if (flags & XFS_IGET_UNTRUSTED) {
2321 error = xfs_imap_lookup(pag, tp, agino, agbno,
2322 &chunk_agbno, &offset_agbno, flags);
2329 * If the inode cluster size is the same as the blocksize or
2330 * smaller we get to the buffer by simple arithmetics.
2332 if (M_IGEO(mp)->blocks_per_cluster == 1) {
2333 offset = XFS_INO_TO_OFFSET(mp, ino);
2334 ASSERT(offset < mp->m_sb.sb_inopblock);
2336 imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, agbno);
2337 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2338 imap->im_boffset = (unsigned short)(offset <<
2339 mp->m_sb.sb_inodelog);
2344 * If the inode chunks are aligned then use simple maths to
2345 * find the location. Otherwise we have to do a btree
2346 * lookup to find the location.
2348 if (M_IGEO(mp)->inoalign_mask) {
2349 offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2350 chunk_agbno = agbno - offset_agbno;
2352 error = xfs_imap_lookup(pag, tp, agino, agbno,
2353 &chunk_agbno, &offset_agbno, flags);
2359 ASSERT(agbno >= chunk_agbno);
2360 cluster_agbno = chunk_agbno +
2361 ((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2362 M_IGEO(mp)->blocks_per_cluster);
2363 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2364 XFS_INO_TO_OFFSET(mp, ino);
2366 imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, cluster_agbno);
2367 imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2368 imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2371 * If the inode number maps to a block outside the bounds
2372 * of the file system then return NULL rather than calling
2373 * read_buf and panicing when we get an error from the
2376 if ((imap->im_blkno + imap->im_len) >
2377 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2379 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2380 __func__, (unsigned long long) imap->im_blkno,
2381 (unsigned long long) imap->im_len,
2382 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2389 * Log specified fields for the ag hdr (inode section). The growth of the agi
2390 * structure over time requires that we interpret the buffer as two logical
2391 * regions delineated by the end of the unlinked list. This is due to the size
2392 * of the hash table and its location in the middle of the agi.
2394 * For example, a request to log a field before agi_unlinked and a field after
2395 * agi_unlinked could cause us to log the entire hash table and use an excessive
2396 * amount of log space. To avoid this behavior, log the region up through
2397 * agi_unlinked in one call and the region after agi_unlinked through the end of
2398 * the structure in another.
2402 struct xfs_trans *tp,
2406 int first; /* first byte number */
2407 int last; /* last byte number */
2408 static const short offsets[] = { /* field starting offsets */
2409 /* keep in sync with bit definitions */
2410 offsetof(xfs_agi_t, agi_magicnum),
2411 offsetof(xfs_agi_t, agi_versionnum),
2412 offsetof(xfs_agi_t, agi_seqno),
2413 offsetof(xfs_agi_t, agi_length),
2414 offsetof(xfs_agi_t, agi_count),
2415 offsetof(xfs_agi_t, agi_root),
2416 offsetof(xfs_agi_t, agi_level),
2417 offsetof(xfs_agi_t, agi_freecount),
2418 offsetof(xfs_agi_t, agi_newino),
2419 offsetof(xfs_agi_t, agi_dirino),
2420 offsetof(xfs_agi_t, agi_unlinked),
2421 offsetof(xfs_agi_t, agi_free_root),
2422 offsetof(xfs_agi_t, agi_free_level),
2423 offsetof(xfs_agi_t, agi_iblocks),
2427 struct xfs_agi *agi = bp->b_addr;
2429 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2433 * Compute byte offsets for the first and last fields in the first
2434 * region and log the agi buffer. This only logs up through
2437 if (fields & XFS_AGI_ALL_BITS_R1) {
2438 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2440 xfs_trans_log_buf(tp, bp, first, last);
2444 * Mask off the bits in the first region and calculate the first and
2445 * last field offsets for any bits in the second region.
2447 fields &= ~XFS_AGI_ALL_BITS_R1;
2449 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2451 xfs_trans_log_buf(tp, bp, first, last);
2455 static xfs_failaddr_t
2459 struct xfs_mount *mp = bp->b_mount;
2460 struct xfs_agi *agi = bp->b_addr;
2463 if (xfs_has_crc(mp)) {
2464 if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2465 return __this_address;
2466 if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn)))
2467 return __this_address;
2471 * Validate the magic number of the agi block.
2473 if (!xfs_verify_magic(bp, agi->agi_magicnum))
2474 return __this_address;
2475 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2476 return __this_address;
2478 if (be32_to_cpu(agi->agi_level) < 1 ||
2479 be32_to_cpu(agi->agi_level) > M_IGEO(mp)->inobt_maxlevels)
2480 return __this_address;
2482 if (xfs_has_finobt(mp) &&
2483 (be32_to_cpu(agi->agi_free_level) < 1 ||
2484 be32_to_cpu(agi->agi_free_level) > M_IGEO(mp)->inobt_maxlevels))
2485 return __this_address;
2488 * during growfs operations, the perag is not fully initialised,
2489 * so we can't use it for any useful checking. growfs ensures we can't
2490 * use it by using uncached buffers that don't have the perag attached
2491 * so we can detect and avoid this problem.
2493 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2494 return __this_address;
2496 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2497 if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2499 if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2500 return __this_address;
2507 xfs_agi_read_verify(
2510 struct xfs_mount *mp = bp->b_mount;
2513 if (xfs_has_crc(mp) &&
2514 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2515 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2517 fa = xfs_agi_verify(bp);
2518 if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2519 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2524 xfs_agi_write_verify(
2527 struct xfs_mount *mp = bp->b_mount;
2528 struct xfs_buf_log_item *bip = bp->b_log_item;
2529 struct xfs_agi *agi = bp->b_addr;
2532 fa = xfs_agi_verify(bp);
2534 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2538 if (!xfs_has_crc(mp))
2542 agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2543 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2546 const struct xfs_buf_ops xfs_agi_buf_ops = {
2548 .magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2549 .verify_read = xfs_agi_read_verify,
2550 .verify_write = xfs_agi_write_verify,
2551 .verify_struct = xfs_agi_verify,
2555 * Read in the allocation group header (inode allocation section)
2559 struct xfs_perag *pag,
2560 struct xfs_trans *tp,
2561 struct xfs_buf **agibpp)
2563 struct xfs_mount *mp = pag->pag_mount;
2566 trace_xfs_read_agi(pag->pag_mount, pag->pag_agno);
2568 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2569 XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGI_DADDR(mp)),
2570 XFS_FSS_TO_BB(mp, 1), 0, agibpp, &xfs_agi_buf_ops);
2574 xfs_trans_buf_set_type(tp, *agibpp, XFS_BLFT_AGI_BUF);
2576 xfs_buf_set_ref(*agibpp, XFS_AGI_REF);
2581 * Read in the agi and initialise the per-ag data. If the caller supplies a
2582 * @agibpp, return the locked AGI buffer to them, otherwise release it.
2585 xfs_ialloc_read_agi(
2586 struct xfs_perag *pag,
2587 struct xfs_trans *tp,
2588 struct xfs_buf **agibpp)
2590 struct xfs_buf *agibp;
2591 struct xfs_agi *agi;
2594 trace_xfs_ialloc_read_agi(pag->pag_mount, pag->pag_agno);
2596 error = xfs_read_agi(pag, tp, &agibp);
2600 agi = agibp->b_addr;
2601 if (!xfs_perag_initialised_agi(pag)) {
2602 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2603 pag->pagi_count = be32_to_cpu(agi->agi_count);
2604 set_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
2608 * It's possible for these to be out of sync if
2609 * we are in the middle of a forced shutdown.
2611 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2612 xfs_is_shutdown(pag->pag_mount));
2616 xfs_trans_brelse(tp, agibp);
2620 /* Is there an inode record covering a given range of inode numbers? */
2622 xfs_ialloc_has_inode_record(
2623 struct xfs_btree_cur *cur,
2628 struct xfs_inobt_rec_incore irec;
2636 error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2637 while (error == 0 && has_record) {
2638 error = xfs_inobt_get_rec(cur, &irec, &has_record);
2639 if (error || irec.ir_startino > high)
2642 agino = irec.ir_startino;
2643 holemask = irec.ir_holemask;
2644 for (i = 0; i < XFS_INOBT_HOLEMASK_BITS; holemask >>= 1,
2645 i++, agino += XFS_INODES_PER_HOLEMASK_BIT) {
2648 if (agino + XFS_INODES_PER_HOLEMASK_BIT > low &&
2655 error = xfs_btree_increment(cur, 0, &has_record);
2660 /* Is there an inode record covering a given extent? */
2662 xfs_ialloc_has_inodes_at_extent(
2663 struct xfs_btree_cur *cur,
2671 low = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2672 high = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2674 return xfs_ialloc_has_inode_record(cur, low, high, exists);
2677 struct xfs_ialloc_count_inodes {
2679 xfs_agino_t freecount;
2682 /* Record inode counts across all inobt records. */
2684 xfs_ialloc_count_inodes_rec(
2685 struct xfs_btree_cur *cur,
2686 const union xfs_btree_rec *rec,
2689 struct xfs_inobt_rec_incore irec;
2690 struct xfs_ialloc_count_inodes *ci = priv;
2692 xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2693 ci->count += irec.ir_count;
2694 ci->freecount += irec.ir_freecount;
2699 /* Count allocated and free inodes under an inobt. */
2701 xfs_ialloc_count_inodes(
2702 struct xfs_btree_cur *cur,
2704 xfs_agino_t *freecount)
2706 struct xfs_ialloc_count_inodes ci = {0};
2709 ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2710 error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2715 *freecount = ci.freecount;
2720 * Initialize inode-related geometry information.
2722 * Compute the inode btree min and max levels and set maxicount.
2724 * Set the inode cluster size. This may still be overridden by the file
2725 * system block size if it is larger than the chosen cluster size.
2727 * For v5 filesystems, scale the cluster size with the inode size to keep a
2728 * constant ratio of inode per cluster buffer, but only if mkfs has set the
2729 * inode alignment value appropriately for larger cluster sizes.
2731 * Then compute the inode cluster alignment information.
2734 xfs_ialloc_setup_geometry(
2735 struct xfs_mount *mp)
2737 struct xfs_sb *sbp = &mp->m_sb;
2738 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2742 igeo->new_diflags2 = 0;
2743 if (xfs_has_bigtime(mp))
2744 igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME;
2745 if (xfs_has_large_extent_counts(mp))
2746 igeo->new_diflags2 |= XFS_DIFLAG2_NREXT64;
2748 /* Compute inode btree geometry. */
2749 igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2750 igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
2751 igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
2752 igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2753 igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2755 igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2757 igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2759 if (sbp->sb_spino_align)
2760 igeo->ialloc_min_blks = sbp->sb_spino_align;
2762 igeo->ialloc_min_blks = igeo->ialloc_blks;
2764 /* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2765 inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2766 igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2768 ASSERT(igeo->inobt_maxlevels <= xfs_iallocbt_maxlevels_ondisk());
2771 * Set the maximum inode count for this filesystem, being careful not
2772 * to use obviously garbage sb_inopblog/sb_inopblock values. Regular
2773 * users should never get here due to failing sb verification, but
2774 * certain users (xfs_db) need to be usable even with corrupt metadata.
2776 if (sbp->sb_imax_pct && igeo->ialloc_blks) {
2778 * Make sure the maximum inode count is a multiple
2779 * of the units we allocate inodes in.
2781 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
2782 do_div(icount, 100);
2783 do_div(icount, igeo->ialloc_blks);
2784 igeo->maxicount = XFS_FSB_TO_INO(mp,
2785 icount * igeo->ialloc_blks);
2787 igeo->maxicount = 0;
2791 * Compute the desired size of an inode cluster buffer size, which
2792 * starts at 8K and (on v5 filesystems) scales up with larger inode
2795 * Preserve the desired inode cluster size because the sparse inodes
2796 * feature uses that desired size (not the actual size) to compute the
2797 * sparse inode alignment. The mount code validates this value, so we
2798 * cannot change the behavior.
2800 igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
2801 if (xfs_has_v3inodes(mp)) {
2802 int new_size = igeo->inode_cluster_size_raw;
2804 new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
2805 if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
2806 igeo->inode_cluster_size_raw = new_size;
2809 /* Calculate inode cluster ratios. */
2810 if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
2811 igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
2812 igeo->inode_cluster_size_raw);
2814 igeo->blocks_per_cluster = 1;
2815 igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
2816 igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
2818 /* Calculate inode cluster alignment. */
2819 if (xfs_has_align(mp) &&
2820 mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
2821 igeo->cluster_align = mp->m_sb.sb_inoalignmt;
2823 igeo->cluster_align = 1;
2824 igeo->inoalign_mask = igeo->cluster_align - 1;
2825 igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
2828 * If we are using stripe alignment, check whether
2829 * the stripe unit is a multiple of the inode alignment
2831 if (mp->m_dalign && igeo->inoalign_mask &&
2832 !(mp->m_dalign & igeo->inoalign_mask))
2833 igeo->ialloc_align = mp->m_dalign;
2835 igeo->ialloc_align = 0;
2838 /* Compute the location of the root directory inode that is laid out by mkfs. */
2840 xfs_ialloc_calc_rootino(
2841 struct xfs_mount *mp,
2844 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2845 xfs_agblock_t first_bno;
2848 * Pre-calculate the geometry of AG 0. We know what it looks like
2849 * because libxfs knows how to create allocation groups now.
2851 * first_bno is the first block in which mkfs could possibly have
2852 * allocated the root directory inode, once we factor in the metadata
2853 * that mkfs formats before it. Namely, the four AG headers...
2855 first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize);
2857 /* ...the two free space btree roots... */
2860 /* ...the inode btree root... */
2863 /* ...the initial AGFL... */
2864 first_bno += xfs_alloc_min_freelist(mp, NULL);
2866 /* ...the free inode btree root... */
2867 if (xfs_has_finobt(mp))
2870 /* ...the reverse mapping btree root... */
2871 if (xfs_has_rmapbt(mp))
2874 /* ...the reference count btree... */
2875 if (xfs_has_reflink(mp))
2879 * ...and the log, if it is allocated in the first allocation group.
2881 * This can happen with filesystems that only have a single
2882 * allocation group, or very odd geometries created by old mkfs
2883 * versions on very small filesystems.
2885 if (xfs_ag_contains_log(mp, 0))
2886 first_bno += mp->m_sb.sb_logblocks;
2889 * Now round first_bno up to whatever allocation alignment is given
2890 * by the filesystem or was passed in.
2892 if (xfs_has_dalign(mp) && igeo->ialloc_align > 0)
2893 first_bno = roundup(first_bno, sunit);
2894 else if (xfs_has_align(mp) &&
2895 mp->m_sb.sb_inoalignmt > 1)
2896 first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt);
2898 return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno));
2902 * Ensure there are not sparse inode clusters that cross the new EOAG.
2904 * This is a no-op for non-spinode filesystems since clusters are always fully
2905 * allocated and checking the bnobt suffices. However, a spinode filesystem
2906 * could have a record where the upper inodes are free blocks. If those blocks
2907 * were removed from the filesystem, the inode record would extend beyond EOAG,
2908 * which will be flagged as corruption.
2911 xfs_ialloc_check_shrink(
2912 struct xfs_perag *pag,
2913 struct xfs_trans *tp,
2914 struct xfs_buf *agibp,
2915 xfs_agblock_t new_length)
2917 struct xfs_inobt_rec_incore rec;
2918 struct xfs_btree_cur *cur;
2923 if (!xfs_has_sparseinodes(pag->pag_mount))
2926 cur = xfs_inobt_init_cursor(pag, tp, agibp, XFS_BTNUM_INO);
2928 /* Look up the inobt record that would correspond to the new EOFS. */
2929 agino = XFS_AGB_TO_AGINO(pag->pag_mount, new_length);
2930 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has);
2934 error = xfs_inobt_get_rec(cur, &rec, &has);
2939 error = -EFSCORRUPTED;
2943 /* If the record covers inodes that would be beyond EOFS, bail out. */
2944 if (rec.ir_startino + XFS_INODES_PER_CHUNK > agino) {
2949 xfs_btree_del_cursor(cur, error);