1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2018 Red Hat, Inc.
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_trans_resv.h"
15 #include "xfs_mount.h"
16 #include "xfs_btree.h"
17 #include "xfs_alloc_btree.h"
18 #include "xfs_rmap_btree.h"
19 #include "xfs_alloc.h"
20 #include "xfs_ialloc.h"
23 #include "xfs_ag_resv.h"
24 #include "xfs_health.h"
25 #include "xfs_error.h"
27 #include "xfs_defer.h"
28 #include "xfs_log_format.h"
29 #include "xfs_trans.h"
30 #include "xfs_trace.h"
31 #include "xfs_inode.h"
32 #include "xfs_icache.h"
36 * Passive reference counting access wrappers to the perag structures. If the
37 * per-ag structure is to be freed, the freeing code is responsible for cleaning
38 * up objects with passive references before freeing the structure. This is
39 * things like cached buffers.
46 struct xfs_perag *pag;
50 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
52 ASSERT(atomic_read(&pag->pag_ref) >= 0);
53 ref = atomic_inc_return(&pag->pag_ref);
56 trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
61 * search from @first to find the next perag with the given tag set.
69 struct xfs_perag *pag;
74 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
75 (void **)&pag, first, 1, tag);
80 ref = atomic_inc_return(&pag->pag_ref);
82 trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
88 struct xfs_perag *pag)
92 ASSERT(atomic_read(&pag->pag_ref) > 0);
93 ref = atomic_dec_return(&pag->pag_ref);
94 trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
98 * xfs_initialize_perag_data
100 * Read in each per-ag structure so we can count up the number of
101 * allocated inodes, free inodes and used filesystem blocks as this
102 * information is no longer persistent in the superblock. Once we have
103 * this information, write it into the in-core superblock structure.
106 xfs_initialize_perag_data(
107 struct xfs_mount *mp,
108 xfs_agnumber_t agcount)
110 xfs_agnumber_t index;
111 struct xfs_perag *pag;
112 struct xfs_sb *sbp = &mp->m_sb;
116 uint64_t bfreelst = 0;
121 for (index = 0; index < agcount; index++) {
123 * read the agf, then the agi. This gets us
124 * all the information we need and populates the
125 * per-ag structures for us.
127 error = xfs_alloc_pagf_init(mp, NULL, index, 0);
131 error = xfs_ialloc_pagi_init(mp, NULL, index);
134 pag = xfs_perag_get(mp, index);
135 ifree += pag->pagi_freecount;
136 ialloc += pag->pagi_count;
137 bfree += pag->pagf_freeblks;
138 bfreelst += pag->pagf_flcount;
139 btree += pag->pagf_btreeblks;
142 fdblocks = bfree + bfreelst + btree;
145 * If the new summary counts are obviously incorrect, fail the
146 * mount operation because that implies the AGFs are also corrupt.
147 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
148 * will prevent xfs_repair from fixing anything.
150 if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
151 xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
152 error = -EFSCORRUPTED;
156 /* Overwrite incore superblock counters with just-read data */
157 spin_lock(&mp->m_sb_lock);
158 sbp->sb_ifree = ifree;
159 sbp->sb_icount = ialloc;
160 sbp->sb_fdblocks = fdblocks;
161 spin_unlock(&mp->m_sb_lock);
163 xfs_reinit_percpu_counters(mp);
165 xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
171 struct rcu_head *head)
173 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
175 ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
180 * Free up the per-ag resources associated with the mount structure.
184 struct xfs_mount *mp)
186 struct xfs_perag *pag;
189 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
190 spin_lock(&mp->m_perag_lock);
191 pag = radix_tree_delete(&mp->m_perag_tree, agno);
192 spin_unlock(&mp->m_perag_lock);
194 XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0);
196 cancel_delayed_work_sync(&pag->pag_blockgc_work);
197 xfs_iunlink_destroy(pag);
198 xfs_buf_hash_destroy(pag);
200 call_rcu(&pag->rcu_head, __xfs_free_perag);
205 xfs_initialize_perag(
206 struct xfs_mount *mp,
207 xfs_agnumber_t agcount,
208 xfs_agnumber_t *maxagi)
210 struct xfs_perag *pag;
211 xfs_agnumber_t index;
212 xfs_agnumber_t first_initialised = NULLAGNUMBER;
216 * Walk the current per-ag tree so we don't try to initialise AGs
217 * that already exist (growfs case). Allocate and insert all the
218 * AGs we don't find ready for initialisation.
220 for (index = 0; index < agcount; index++) {
221 pag = xfs_perag_get(mp, index);
227 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
230 goto out_unwind_new_pags;
232 pag->pag_agno = index;
235 error = radix_tree_preload(GFP_NOFS);
239 spin_lock(&mp->m_perag_lock);
240 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
242 spin_unlock(&mp->m_perag_lock);
243 radix_tree_preload_end();
247 spin_unlock(&mp->m_perag_lock);
248 radix_tree_preload_end();
250 /* Place kernel structure only init below this point. */
251 spin_lock_init(&pag->pag_ici_lock);
252 spin_lock_init(&pag->pagb_lock);
253 spin_lock_init(&pag->pag_state_lock);
254 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
255 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
256 init_waitqueue_head(&pag->pagb_wait);
258 pag->pagb_tree = RB_ROOT;
260 error = xfs_buf_hash_init(pag);
264 error = xfs_iunlink_init(pag);
266 goto out_hash_destroy;
268 /* first new pag is fully initialized */
269 if (first_initialised == NULLAGNUMBER)
270 first_initialised = index;
273 index = xfs_set_inode_alloc(mp, agcount);
278 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
282 xfs_buf_hash_destroy(pag);
284 radix_tree_delete(&mp->m_perag_tree, index);
288 /* unwind any prior newly initialized pags */
289 for (index = first_initialised; index < agcount; index++) {
290 pag = radix_tree_delete(&mp->m_perag_tree, index);
293 xfs_buf_hash_destroy(pag);
294 xfs_iunlink_destroy(pag);
302 struct xfs_mount *mp,
305 struct xfs_buf **bpp,
306 const struct xfs_buf_ops *ops)
311 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
315 bp->b_maps[0].bm_bn = blkno;
322 static inline bool is_log_ag(struct xfs_mount *mp, struct aghdr_init_data *id)
324 return mp->m_sb.sb_logstart > 0 &&
325 id->agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
329 * Generic btree root block init function
333 struct xfs_mount *mp,
335 struct aghdr_init_data *id)
337 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
340 /* Finish initializing a free space btree. */
342 xfs_freesp_init_recs(
343 struct xfs_mount *mp,
345 struct aghdr_init_data *id)
347 struct xfs_alloc_rec *arec;
348 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
350 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
351 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
353 if (is_log_ag(mp, id)) {
354 struct xfs_alloc_rec *nrec;
355 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
356 mp->m_sb.sb_logstart);
358 ASSERT(start >= mp->m_ag_prealloc_blocks);
359 if (start != mp->m_ag_prealloc_blocks) {
361 * Modify first record to pad stripe align of log
363 arec->ar_blockcount = cpu_to_be32(start -
364 mp->m_ag_prealloc_blocks);
368 * Insert second record at start of internal log
369 * which then gets trimmed.
371 nrec->ar_startblock = cpu_to_be32(
372 be32_to_cpu(arec->ar_startblock) +
373 be32_to_cpu(arec->ar_blockcount));
375 be16_add_cpu(&block->bb_numrecs, 1);
378 * Change record start to after the internal log
380 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
384 * Calculate the record block count and check for the case where
385 * the log might have consumed all available space in the AG. If
386 * so, reset the record count to 0 to avoid exposure of an invalid
387 * record start block.
389 arec->ar_blockcount = cpu_to_be32(id->agsize -
390 be32_to_cpu(arec->ar_startblock));
391 if (!arec->ar_blockcount)
392 block->bb_numrecs = 0;
396 * Alloc btree root block init functions
400 struct xfs_mount *mp,
402 struct aghdr_init_data *id)
404 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
405 xfs_freesp_init_recs(mp, bp, id);
410 struct xfs_mount *mp,
412 struct aghdr_init_data *id)
414 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
415 xfs_freesp_init_recs(mp, bp, id);
419 * Reverse map root block init
423 struct xfs_mount *mp,
425 struct aghdr_init_data *id)
427 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
428 struct xfs_rmap_rec *rrec;
430 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
433 * mark the AG header regions as static metadata The BNO
434 * btree block is the first block after the headers, so
435 * it's location defines the size of region the static
438 * Note: unlike mkfs, we never have to account for log
439 * space when growing the data regions
441 rrec = XFS_RMAP_REC_ADDR(block, 1);
442 rrec->rm_startblock = 0;
443 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
444 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
447 /* account freespace btree root blocks */
448 rrec = XFS_RMAP_REC_ADDR(block, 2);
449 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
450 rrec->rm_blockcount = cpu_to_be32(2);
451 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
454 /* account inode btree root blocks */
455 rrec = XFS_RMAP_REC_ADDR(block, 3);
456 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
457 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
459 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
462 /* account for rmap btree root */
463 rrec = XFS_RMAP_REC_ADDR(block, 4);
464 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
465 rrec->rm_blockcount = cpu_to_be32(1);
466 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
469 /* account for refc btree root */
470 if (xfs_has_reflink(mp)) {
471 rrec = XFS_RMAP_REC_ADDR(block, 5);
472 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
473 rrec->rm_blockcount = cpu_to_be32(1);
474 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
476 be16_add_cpu(&block->bb_numrecs, 1);
479 /* account for the log space */
480 if (is_log_ag(mp, id)) {
481 rrec = XFS_RMAP_REC_ADDR(block,
482 be16_to_cpu(block->bb_numrecs) + 1);
483 rrec->rm_startblock = cpu_to_be32(
484 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
485 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
486 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
488 be16_add_cpu(&block->bb_numrecs, 1);
493 * Initialise new secondary superblocks with the pre-grow geometry, but mark
494 * them as "in progress" so we know they haven't yet been activated. This will
495 * get cleared when the update with the new geometry information is done after
496 * changes to the primary are committed. This isn't strictly necessary, but we
497 * get it for free with the delayed buffer write lists and it means we can tell
498 * if a grow operation didn't complete properly after the fact.
502 struct xfs_mount *mp,
504 struct aghdr_init_data *id)
506 struct xfs_dsb *dsb = bp->b_addr;
508 xfs_sb_to_disk(dsb, &mp->m_sb);
509 dsb->sb_inprogress = 1;
514 struct xfs_mount *mp,
516 struct aghdr_init_data *id)
518 struct xfs_agf *agf = bp->b_addr;
519 xfs_extlen_t tmpsize;
521 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
522 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
523 agf->agf_seqno = cpu_to_be32(id->agno);
524 agf->agf_length = cpu_to_be32(id->agsize);
525 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
526 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
527 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
528 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
529 if (xfs_has_rmapbt(mp)) {
530 agf->agf_roots[XFS_BTNUM_RMAPi] =
531 cpu_to_be32(XFS_RMAP_BLOCK(mp));
532 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
533 agf->agf_rmap_blocks = cpu_to_be32(1);
536 agf->agf_flfirst = cpu_to_be32(1);
538 agf->agf_flcount = 0;
539 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
540 agf->agf_freeblks = cpu_to_be32(tmpsize);
541 agf->agf_longest = cpu_to_be32(tmpsize);
543 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
544 if (xfs_has_reflink(mp)) {
545 agf->agf_refcount_root = cpu_to_be32(
547 agf->agf_refcount_level = cpu_to_be32(1);
548 agf->agf_refcount_blocks = cpu_to_be32(1);
551 if (is_log_ag(mp, id)) {
552 int64_t logblocks = mp->m_sb.sb_logblocks;
554 be32_add_cpu(&agf->agf_freeblks, -logblocks);
555 agf->agf_longest = cpu_to_be32(id->agsize -
556 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
562 struct xfs_mount *mp,
564 struct aghdr_init_data *id)
566 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
570 if (xfs_has_crc(mp)) {
571 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
572 agfl->agfl_seqno = cpu_to_be32(id->agno);
573 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
576 agfl_bno = xfs_buf_to_agfl_bno(bp);
577 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
578 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
583 struct xfs_mount *mp,
585 struct aghdr_init_data *id)
587 struct xfs_agi *agi = bp->b_addr;
590 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
591 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
592 agi->agi_seqno = cpu_to_be32(id->agno);
593 agi->agi_length = cpu_to_be32(id->agsize);
595 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
596 agi->agi_level = cpu_to_be32(1);
597 agi->agi_freecount = 0;
598 agi->agi_newino = cpu_to_be32(NULLAGINO);
599 agi->agi_dirino = cpu_to_be32(NULLAGINO);
601 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
602 if (xfs_has_finobt(mp)) {
603 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
604 agi->agi_free_level = cpu_to_be32(1);
606 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
607 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
608 if (xfs_has_inobtcounts(mp)) {
609 agi->agi_iblocks = cpu_to_be32(1);
610 if (xfs_has_finobt(mp))
611 agi->agi_fblocks = cpu_to_be32(1);
615 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
616 struct aghdr_init_data *id);
619 struct xfs_mount *mp,
620 struct aghdr_init_data *id,
621 aghdr_init_work_f work,
622 const struct xfs_buf_ops *ops)
627 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
633 xfs_buf_delwri_queue(bp, &id->buffer_list);
638 struct xfs_aghdr_grow_data {
641 const struct xfs_buf_ops *ops;
642 aghdr_init_work_f work;
648 * Prepare new AG headers to be written to disk. We use uncached buffers here,
649 * as it is assumed these new AG headers are currently beyond the currently
650 * valid filesystem address space. Using cached buffers would trip over EOFS
651 * corruption detection alogrithms in the buffer cache lookup routines.
653 * This is a non-transactional function, but the prepared buffers are added to a
654 * delayed write buffer list supplied by the caller so they can submit them to
655 * disk and wait on them as required.
659 struct xfs_mount *mp,
660 struct aghdr_init_data *id)
663 struct xfs_aghdr_grow_data aghdr_data[] = {
665 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
666 .numblks = XFS_FSS_TO_BB(mp, 1),
667 .ops = &xfs_sb_buf_ops,
668 .work = &xfs_sbblock_init,
672 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
673 .numblks = XFS_FSS_TO_BB(mp, 1),
674 .ops = &xfs_agf_buf_ops,
675 .work = &xfs_agfblock_init,
679 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
680 .numblks = XFS_FSS_TO_BB(mp, 1),
681 .ops = &xfs_agfl_buf_ops,
682 .work = &xfs_agflblock_init,
686 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
687 .numblks = XFS_FSS_TO_BB(mp, 1),
688 .ops = &xfs_agi_buf_ops,
689 .work = &xfs_agiblock_init,
692 { /* BNO root block */
693 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
694 .numblks = BTOBB(mp->m_sb.sb_blocksize),
695 .ops = &xfs_bnobt_buf_ops,
696 .work = &xfs_bnoroot_init,
699 { /* CNT root block */
700 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
701 .numblks = BTOBB(mp->m_sb.sb_blocksize),
702 .ops = &xfs_cntbt_buf_ops,
703 .work = &xfs_cntroot_init,
706 { /* INO root block */
707 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
708 .numblks = BTOBB(mp->m_sb.sb_blocksize),
709 .ops = &xfs_inobt_buf_ops,
710 .work = &xfs_btroot_init,
711 .type = XFS_BTNUM_INO,
714 { /* FINO root block */
715 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
716 .numblks = BTOBB(mp->m_sb.sb_blocksize),
717 .ops = &xfs_finobt_buf_ops,
718 .work = &xfs_btroot_init,
719 .type = XFS_BTNUM_FINO,
720 .need_init = xfs_has_finobt(mp)
722 { /* RMAP root block */
723 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
724 .numblks = BTOBB(mp->m_sb.sb_blocksize),
725 .ops = &xfs_rmapbt_buf_ops,
726 .work = &xfs_rmaproot_init,
727 .need_init = xfs_has_rmapbt(mp)
729 { /* REFC root block */
730 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
731 .numblks = BTOBB(mp->m_sb.sb_blocksize),
732 .ops = &xfs_refcountbt_buf_ops,
733 .work = &xfs_btroot_init,
734 .type = XFS_BTNUM_REFC,
735 .need_init = xfs_has_reflink(mp)
737 { /* NULL terminating block */
738 .daddr = XFS_BUF_DADDR_NULL,
741 struct xfs_aghdr_grow_data *dp;
744 /* Account for AG free space in new AG */
745 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
746 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
750 id->daddr = dp->daddr;
751 id->numblks = dp->numblks;
753 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
762 struct xfs_mount *mp,
763 struct xfs_trans **tpp,
767 struct xfs_alloc_arg args = {
770 .type = XFS_ALLOCTYPE_THIS_BNO,
773 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
774 .resv = XFS_AG_RESV_NONE,
777 struct xfs_buf *agibp, *agfbp;
783 ASSERT(agno == mp->m_sb.sb_agcount - 1);
784 error = xfs_ialloc_read_agi(mp, *tpp, agno, &agibp);
790 error = xfs_alloc_read_agf(mp, *tpp, agno, 0, &agfbp);
795 aglen = be32_to_cpu(agi->agi_length);
796 /* some extra paranoid checks before we shrink the ag */
797 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length))
798 return -EFSCORRUPTED;
802 args.fsbno = XFS_AGB_TO_FSB(mp, agno, aglen - delta);
805 * Make sure that the last inode cluster cannot overlap with the new
806 * end of the AG, even if it's sparse.
808 error = xfs_ialloc_check_shrink(*tpp, agno, agibp, aglen - delta);
813 * Disable perag reservations so it doesn't cause the allocation request
814 * to fail. We'll reestablish reservation before we return.
816 error = xfs_ag_resv_free(agibp->b_pag);
820 /* internal log shouldn't also show up in the free space btrees */
821 error = xfs_alloc_vextent(&args);
822 if (!error && args.agbno == NULLAGBLOCK)
827 * if extent allocation fails, need to roll the transaction to
828 * ensure that the AGFL fixup has been committed anyway.
830 xfs_trans_bhold(*tpp, agfbp);
831 err2 = xfs_trans_roll(tpp);
834 xfs_trans_bjoin(*tpp, agfbp);
839 * if successfully deleted from freespace btrees, need to confirm
840 * per-AG reservation works as expected.
842 be32_add_cpu(&agi->agi_length, -delta);
843 be32_add_cpu(&agf->agf_length, -delta);
845 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
847 be32_add_cpu(&agi->agi_length, delta);
848 be32_add_cpu(&agf->agf_length, delta);
852 __xfs_bmap_add_free(*tpp, args.fsbno, delta, NULL, true);
855 * Roll the transaction before trying to re-init the per-ag
856 * reservation. The new transaction is clean so it will cancel
857 * without any side effects.
859 error = xfs_defer_finish(tpp);
866 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
867 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
870 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
874 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
875 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
880 * Extent the AG indicated by the @id by the length passed in
884 struct xfs_mount *mp,
885 struct xfs_trans *tp,
886 struct aghdr_init_data *id,
895 * Change the agi length.
897 error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
902 be32_add_cpu(&agi->agi_length, len);
903 ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
904 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
905 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
910 error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
915 be32_add_cpu(&agf->agf_length, len);
916 ASSERT(agf->agf_length == agi->agi_length);
917 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
920 * Free the new space.
922 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
923 * this doesn't actually exist in the rmap btree.
925 error = xfs_rmap_free(tp, bp, bp->b_pag,
926 be32_to_cpu(agf->agf_length) - len,
927 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
931 return xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
932 be32_to_cpu(agf->agf_length) - len),
933 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
937 /* Retrieve AG geometry. */
940 struct xfs_mount *mp,
942 struct xfs_ag_geometry *ageo)
944 struct xfs_buf *agi_bp;
945 struct xfs_buf *agf_bp;
948 struct xfs_perag *pag;
949 unsigned int freeblks;
952 if (agno >= mp->m_sb.sb_agcount)
955 /* Lock the AG headers. */
956 error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
959 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
966 memset(ageo, 0, sizeof(*ageo));
967 ageo->ag_number = agno;
969 agi = agi_bp->b_addr;
970 ageo->ag_icount = be32_to_cpu(agi->agi_count);
971 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
973 agf = agf_bp->b_addr;
974 ageo->ag_length = be32_to_cpu(agf->agf_length);
975 freeblks = pag->pagf_freeblks +
977 pag->pagf_btreeblks -
978 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
979 ageo->ag_freeblks = freeblks;
980 xfs_ag_geom_health(pag, ageo);
982 /* Release resources. */
983 xfs_buf_relse(agf_bp);
985 xfs_buf_relse(agi_bp);