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));
176 ASSERT(atomic_read(&pag->pag_ref) == 0);
181 * Free up the per-ag resources associated with the mount structure.
185 struct xfs_mount *mp)
187 struct xfs_perag *pag;
190 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
191 spin_lock(&mp->m_perag_lock);
192 pag = radix_tree_delete(&mp->m_perag_tree, agno);
193 spin_unlock(&mp->m_perag_lock);
195 ASSERT(atomic_read(&pag->pag_ref) == 0);
197 cancel_delayed_work_sync(&pag->pag_blockgc_work);
198 xfs_iunlink_destroy(pag);
199 xfs_buf_hash_destroy(pag);
201 call_rcu(&pag->rcu_head, __xfs_free_perag);
206 xfs_initialize_perag(
207 struct xfs_mount *mp,
208 xfs_agnumber_t agcount,
209 xfs_agnumber_t *maxagi)
211 struct xfs_perag *pag;
212 xfs_agnumber_t index;
213 xfs_agnumber_t first_initialised = NULLAGNUMBER;
217 * Walk the current per-ag tree so we don't try to initialise AGs
218 * that already exist (growfs case). Allocate and insert all the
219 * AGs we don't find ready for initialisation.
221 for (index = 0; index < agcount; index++) {
222 pag = xfs_perag_get(mp, index);
228 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
231 goto out_unwind_new_pags;
233 pag->pag_agno = index;
236 error = radix_tree_preload(GFP_NOFS);
240 spin_lock(&mp->m_perag_lock);
241 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
243 spin_unlock(&mp->m_perag_lock);
244 radix_tree_preload_end();
248 spin_unlock(&mp->m_perag_lock);
249 radix_tree_preload_end();
251 /* Place kernel structure only init below this point. */
252 spin_lock_init(&pag->pag_ici_lock);
253 spin_lock_init(&pag->pagb_lock);
254 spin_lock_init(&pag->pag_state_lock);
255 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
256 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
257 init_waitqueue_head(&pag->pagb_wait);
259 pag->pagb_tree = RB_ROOT;
261 error = xfs_buf_hash_init(pag);
265 error = xfs_iunlink_init(pag);
267 goto out_hash_destroy;
269 /* first new pag is fully initialized */
270 if (first_initialised == NULLAGNUMBER)
271 first_initialised = index;
274 index = xfs_set_inode_alloc(mp, agcount);
279 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
283 xfs_buf_hash_destroy(pag);
285 radix_tree_delete(&mp->m_perag_tree, index);
289 /* unwind any prior newly initialized pags */
290 for (index = first_initialised; index < agcount; index++) {
291 pag = radix_tree_delete(&mp->m_perag_tree, index);
294 xfs_buf_hash_destroy(pag);
295 xfs_iunlink_destroy(pag);
303 struct xfs_mount *mp,
306 struct xfs_buf **bpp,
307 const struct xfs_buf_ops *ops)
312 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
316 bp->b_maps[0].bm_bn = blkno;
323 static inline bool is_log_ag(struct xfs_mount *mp, struct aghdr_init_data *id)
325 return mp->m_sb.sb_logstart > 0 &&
326 id->agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
330 * Generic btree root block init function
334 struct xfs_mount *mp,
336 struct aghdr_init_data *id)
338 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
341 /* Finish initializing a free space btree. */
343 xfs_freesp_init_recs(
344 struct xfs_mount *mp,
346 struct aghdr_init_data *id)
348 struct xfs_alloc_rec *arec;
349 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
351 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
352 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
354 if (is_log_ag(mp, id)) {
355 struct xfs_alloc_rec *nrec;
356 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
357 mp->m_sb.sb_logstart);
359 ASSERT(start >= mp->m_ag_prealloc_blocks);
360 if (start != mp->m_ag_prealloc_blocks) {
362 * Modify first record to pad stripe align of log
364 arec->ar_blockcount = cpu_to_be32(start -
365 mp->m_ag_prealloc_blocks);
369 * Insert second record at start of internal log
370 * which then gets trimmed.
372 nrec->ar_startblock = cpu_to_be32(
373 be32_to_cpu(arec->ar_startblock) +
374 be32_to_cpu(arec->ar_blockcount));
376 be16_add_cpu(&block->bb_numrecs, 1);
379 * Change record start to after the internal log
381 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
385 * Calculate the record block count and check for the case where
386 * the log might have consumed all available space in the AG. If
387 * so, reset the record count to 0 to avoid exposure of an invalid
388 * record start block.
390 arec->ar_blockcount = cpu_to_be32(id->agsize -
391 be32_to_cpu(arec->ar_startblock));
392 if (!arec->ar_blockcount)
393 block->bb_numrecs = 0;
397 * Alloc btree root block init functions
401 struct xfs_mount *mp,
403 struct aghdr_init_data *id)
405 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
406 xfs_freesp_init_recs(mp, bp, id);
411 struct xfs_mount *mp,
413 struct aghdr_init_data *id)
415 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
416 xfs_freesp_init_recs(mp, bp, id);
420 * Reverse map root block init
424 struct xfs_mount *mp,
426 struct aghdr_init_data *id)
428 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
429 struct xfs_rmap_rec *rrec;
431 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
434 * mark the AG header regions as static metadata The BNO
435 * btree block is the first block after the headers, so
436 * it's location defines the size of region the static
439 * Note: unlike mkfs, we never have to account for log
440 * space when growing the data regions
442 rrec = XFS_RMAP_REC_ADDR(block, 1);
443 rrec->rm_startblock = 0;
444 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
445 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
448 /* account freespace btree root blocks */
449 rrec = XFS_RMAP_REC_ADDR(block, 2);
450 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
451 rrec->rm_blockcount = cpu_to_be32(2);
452 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
455 /* account inode btree root blocks */
456 rrec = XFS_RMAP_REC_ADDR(block, 3);
457 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
458 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
460 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
463 /* account for rmap btree root */
464 rrec = XFS_RMAP_REC_ADDR(block, 4);
465 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
466 rrec->rm_blockcount = cpu_to_be32(1);
467 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
470 /* account for refc btree root */
471 if (xfs_has_reflink(mp)) {
472 rrec = XFS_RMAP_REC_ADDR(block, 5);
473 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
474 rrec->rm_blockcount = cpu_to_be32(1);
475 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
477 be16_add_cpu(&block->bb_numrecs, 1);
480 /* account for the log space */
481 if (is_log_ag(mp, id)) {
482 rrec = XFS_RMAP_REC_ADDR(block,
483 be16_to_cpu(block->bb_numrecs) + 1);
484 rrec->rm_startblock = cpu_to_be32(
485 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
486 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
487 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
489 be16_add_cpu(&block->bb_numrecs, 1);
494 * Initialise new secondary superblocks with the pre-grow geometry, but mark
495 * them as "in progress" so we know they haven't yet been activated. This will
496 * get cleared when the update with the new geometry information is done after
497 * changes to the primary are committed. This isn't strictly necessary, but we
498 * get it for free with the delayed buffer write lists and it means we can tell
499 * if a grow operation didn't complete properly after the fact.
503 struct xfs_mount *mp,
505 struct aghdr_init_data *id)
507 struct xfs_dsb *dsb = bp->b_addr;
509 xfs_sb_to_disk(dsb, &mp->m_sb);
510 dsb->sb_inprogress = 1;
515 struct xfs_mount *mp,
517 struct aghdr_init_data *id)
519 struct xfs_agf *agf = bp->b_addr;
520 xfs_extlen_t tmpsize;
522 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
523 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
524 agf->agf_seqno = cpu_to_be32(id->agno);
525 agf->agf_length = cpu_to_be32(id->agsize);
526 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
527 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
528 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
529 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
530 if (xfs_has_rmapbt(mp)) {
531 agf->agf_roots[XFS_BTNUM_RMAPi] =
532 cpu_to_be32(XFS_RMAP_BLOCK(mp));
533 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
534 agf->agf_rmap_blocks = cpu_to_be32(1);
537 agf->agf_flfirst = cpu_to_be32(1);
539 agf->agf_flcount = 0;
540 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
541 agf->agf_freeblks = cpu_to_be32(tmpsize);
542 agf->agf_longest = cpu_to_be32(tmpsize);
544 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
545 if (xfs_has_reflink(mp)) {
546 agf->agf_refcount_root = cpu_to_be32(
548 agf->agf_refcount_level = cpu_to_be32(1);
549 agf->agf_refcount_blocks = cpu_to_be32(1);
552 if (is_log_ag(mp, id)) {
553 int64_t logblocks = mp->m_sb.sb_logblocks;
555 be32_add_cpu(&agf->agf_freeblks, -logblocks);
556 agf->agf_longest = cpu_to_be32(id->agsize -
557 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
563 struct xfs_mount *mp,
565 struct aghdr_init_data *id)
567 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
571 if (xfs_has_crc(mp)) {
572 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
573 agfl->agfl_seqno = cpu_to_be32(id->agno);
574 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
577 agfl_bno = xfs_buf_to_agfl_bno(bp);
578 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
579 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
584 struct xfs_mount *mp,
586 struct aghdr_init_data *id)
588 struct xfs_agi *agi = bp->b_addr;
591 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
592 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
593 agi->agi_seqno = cpu_to_be32(id->agno);
594 agi->agi_length = cpu_to_be32(id->agsize);
596 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
597 agi->agi_level = cpu_to_be32(1);
598 agi->agi_freecount = 0;
599 agi->agi_newino = cpu_to_be32(NULLAGINO);
600 agi->agi_dirino = cpu_to_be32(NULLAGINO);
602 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
603 if (xfs_has_finobt(mp)) {
604 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
605 agi->agi_free_level = cpu_to_be32(1);
607 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
608 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
609 if (xfs_has_inobtcounts(mp)) {
610 agi->agi_iblocks = cpu_to_be32(1);
611 if (xfs_has_finobt(mp))
612 agi->agi_fblocks = cpu_to_be32(1);
616 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
617 struct aghdr_init_data *id);
620 struct xfs_mount *mp,
621 struct aghdr_init_data *id,
622 aghdr_init_work_f work,
623 const struct xfs_buf_ops *ops)
628 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
634 xfs_buf_delwri_queue(bp, &id->buffer_list);
639 struct xfs_aghdr_grow_data {
642 const struct xfs_buf_ops *ops;
643 aghdr_init_work_f work;
649 * Prepare new AG headers to be written to disk. We use uncached buffers here,
650 * as it is assumed these new AG headers are currently beyond the currently
651 * valid filesystem address space. Using cached buffers would trip over EOFS
652 * corruption detection alogrithms in the buffer cache lookup routines.
654 * This is a non-transactional function, but the prepared buffers are added to a
655 * delayed write buffer list supplied by the caller so they can submit them to
656 * disk and wait on them as required.
660 struct xfs_mount *mp,
661 struct aghdr_init_data *id)
664 struct xfs_aghdr_grow_data aghdr_data[] = {
666 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
667 .numblks = XFS_FSS_TO_BB(mp, 1),
668 .ops = &xfs_sb_buf_ops,
669 .work = &xfs_sbblock_init,
673 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
674 .numblks = XFS_FSS_TO_BB(mp, 1),
675 .ops = &xfs_agf_buf_ops,
676 .work = &xfs_agfblock_init,
680 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
681 .numblks = XFS_FSS_TO_BB(mp, 1),
682 .ops = &xfs_agfl_buf_ops,
683 .work = &xfs_agflblock_init,
687 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
688 .numblks = XFS_FSS_TO_BB(mp, 1),
689 .ops = &xfs_agi_buf_ops,
690 .work = &xfs_agiblock_init,
693 { /* BNO root block */
694 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
695 .numblks = BTOBB(mp->m_sb.sb_blocksize),
696 .ops = &xfs_bnobt_buf_ops,
697 .work = &xfs_bnoroot_init,
700 { /* CNT root block */
701 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
702 .numblks = BTOBB(mp->m_sb.sb_blocksize),
703 .ops = &xfs_cntbt_buf_ops,
704 .work = &xfs_cntroot_init,
707 { /* INO root block */
708 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
709 .numblks = BTOBB(mp->m_sb.sb_blocksize),
710 .ops = &xfs_inobt_buf_ops,
711 .work = &xfs_btroot_init,
712 .type = XFS_BTNUM_INO,
715 { /* FINO root block */
716 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
717 .numblks = BTOBB(mp->m_sb.sb_blocksize),
718 .ops = &xfs_finobt_buf_ops,
719 .work = &xfs_btroot_init,
720 .type = XFS_BTNUM_FINO,
721 .need_init = xfs_has_finobt(mp)
723 { /* RMAP root block */
724 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
725 .numblks = BTOBB(mp->m_sb.sb_blocksize),
726 .ops = &xfs_rmapbt_buf_ops,
727 .work = &xfs_rmaproot_init,
728 .need_init = xfs_has_rmapbt(mp)
730 { /* REFC root block */
731 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
732 .numblks = BTOBB(mp->m_sb.sb_blocksize),
733 .ops = &xfs_refcountbt_buf_ops,
734 .work = &xfs_btroot_init,
735 .type = XFS_BTNUM_REFC,
736 .need_init = xfs_has_reflink(mp)
738 { /* NULL terminating block */
739 .daddr = XFS_BUF_DADDR_NULL,
742 struct xfs_aghdr_grow_data *dp;
745 /* Account for AG free space in new AG */
746 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
747 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
751 id->daddr = dp->daddr;
752 id->numblks = dp->numblks;
754 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
763 struct xfs_mount *mp,
764 struct xfs_trans **tpp,
768 struct xfs_alloc_arg args = {
771 .type = XFS_ALLOCTYPE_THIS_BNO,
774 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
775 .resv = XFS_AG_RESV_NONE,
778 struct xfs_buf *agibp, *agfbp;
784 ASSERT(agno == mp->m_sb.sb_agcount - 1);
785 error = xfs_ialloc_read_agi(mp, *tpp, agno, &agibp);
791 error = xfs_alloc_read_agf(mp, *tpp, agno, 0, &agfbp);
796 aglen = be32_to_cpu(agi->agi_length);
797 /* some extra paranoid checks before we shrink the ag */
798 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length))
799 return -EFSCORRUPTED;
803 args.fsbno = XFS_AGB_TO_FSB(mp, agno, aglen - delta);
806 * Make sure that the last inode cluster cannot overlap with the new
807 * end of the AG, even if it's sparse.
809 error = xfs_ialloc_check_shrink(*tpp, agno, agibp, aglen - delta);
814 * Disable perag reservations so it doesn't cause the allocation request
815 * to fail. We'll reestablish reservation before we return.
817 error = xfs_ag_resv_free(agibp->b_pag);
821 /* internal log shouldn't also show up in the free space btrees */
822 error = xfs_alloc_vextent(&args);
823 if (!error && args.agbno == NULLAGBLOCK)
828 * if extent allocation fails, need to roll the transaction to
829 * ensure that the AGFL fixup has been committed anyway.
831 xfs_trans_bhold(*tpp, agfbp);
832 err2 = xfs_trans_roll(tpp);
835 xfs_trans_bjoin(*tpp, agfbp);
840 * if successfully deleted from freespace btrees, need to confirm
841 * per-AG reservation works as expected.
843 be32_add_cpu(&agi->agi_length, -delta);
844 be32_add_cpu(&agf->agf_length, -delta);
846 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
848 be32_add_cpu(&agi->agi_length, delta);
849 be32_add_cpu(&agf->agf_length, delta);
853 __xfs_bmap_add_free(*tpp, args.fsbno, delta, NULL, true);
856 * Roll the transaction before trying to re-init the per-ag
857 * reservation. The new transaction is clean so it will cancel
858 * without any side effects.
860 error = xfs_defer_finish(tpp);
867 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
868 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
871 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
875 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
876 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
881 * Extent the AG indicated by the @id by the length passed in
885 struct xfs_mount *mp,
886 struct xfs_trans *tp,
887 struct aghdr_init_data *id,
896 * Change the agi length.
898 error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
903 be32_add_cpu(&agi->agi_length, len);
904 ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
905 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
906 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
911 error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
916 be32_add_cpu(&agf->agf_length, len);
917 ASSERT(agf->agf_length == agi->agi_length);
918 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
921 * Free the new space.
923 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
924 * this doesn't actually exist in the rmap btree.
926 error = xfs_rmap_free(tp, bp, bp->b_pag,
927 be32_to_cpu(agf->agf_length) - len,
928 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
932 return xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
933 be32_to_cpu(agf->agf_length) - len),
934 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
938 /* Retrieve AG geometry. */
941 struct xfs_mount *mp,
943 struct xfs_ag_geometry *ageo)
945 struct xfs_buf *agi_bp;
946 struct xfs_buf *agf_bp;
949 struct xfs_perag *pag;
950 unsigned int freeblks;
953 if (agno >= mp->m_sb.sb_agcount)
956 /* Lock the AG headers. */
957 error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
960 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
967 memset(ageo, 0, sizeof(*ageo));
968 ageo->ag_number = agno;
970 agi = agi_bp->b_addr;
971 ageo->ag_icount = be32_to_cpu(agi->agi_count);
972 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
974 agf = agf_bp->b_addr;
975 ageo->ag_length = be32_to_cpu(agf->agf_length);
976 freeblks = pag->pagf_freeblks +
978 pag->pagf_btreeblks -
979 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
980 ageo->ag_freeblks = freeblks;
981 xfs_ag_geom_health(pag, ageo);
983 /* Release resources. */
984 xfs_buf_relse(agf_bp);
986 xfs_buf_relse(agi_bp);