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;
112 xfs_sb_t *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 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
318 bp->b_maps[0].bm_bn = blkno;
325 static inline bool is_log_ag(struct xfs_mount *mp, struct aghdr_init_data *id)
327 return mp->m_sb.sb_logstart > 0 &&
328 id->agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
332 * Generic btree root block init function
336 struct xfs_mount *mp,
338 struct aghdr_init_data *id)
340 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
343 /* Finish initializing a free space btree. */
345 xfs_freesp_init_recs(
346 struct xfs_mount *mp,
348 struct aghdr_init_data *id)
350 struct xfs_alloc_rec *arec;
351 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
353 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
354 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
356 if (is_log_ag(mp, id)) {
357 struct xfs_alloc_rec *nrec;
358 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
359 mp->m_sb.sb_logstart);
361 ASSERT(start >= mp->m_ag_prealloc_blocks);
362 if (start != mp->m_ag_prealloc_blocks) {
364 * Modify first record to pad stripe align of log
366 arec->ar_blockcount = cpu_to_be32(start -
367 mp->m_ag_prealloc_blocks);
371 * Insert second record at start of internal log
372 * which then gets trimmed.
374 nrec->ar_startblock = cpu_to_be32(
375 be32_to_cpu(arec->ar_startblock) +
376 be32_to_cpu(arec->ar_blockcount));
378 be16_add_cpu(&block->bb_numrecs, 1);
381 * Change record start to after the internal log
383 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
387 * Calculate the record block count and check for the case where
388 * the log might have consumed all available space in the AG. If
389 * so, reset the record count to 0 to avoid exposure of an invalid
390 * record start block.
392 arec->ar_blockcount = cpu_to_be32(id->agsize -
393 be32_to_cpu(arec->ar_startblock));
394 if (!arec->ar_blockcount)
395 block->bb_numrecs = 0;
399 * Alloc btree root block init functions
403 struct xfs_mount *mp,
405 struct aghdr_init_data *id)
407 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
408 xfs_freesp_init_recs(mp, bp, id);
413 struct xfs_mount *mp,
415 struct aghdr_init_data *id)
417 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
418 xfs_freesp_init_recs(mp, bp, id);
422 * Reverse map root block init
426 struct xfs_mount *mp,
428 struct aghdr_init_data *id)
430 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
431 struct xfs_rmap_rec *rrec;
433 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
436 * mark the AG header regions as static metadata The BNO
437 * btree block is the first block after the headers, so
438 * it's location defines the size of region the static
441 * Note: unlike mkfs, we never have to account for log
442 * space when growing the data regions
444 rrec = XFS_RMAP_REC_ADDR(block, 1);
445 rrec->rm_startblock = 0;
446 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
447 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
450 /* account freespace btree root blocks */
451 rrec = XFS_RMAP_REC_ADDR(block, 2);
452 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
453 rrec->rm_blockcount = cpu_to_be32(2);
454 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
457 /* account inode btree root blocks */
458 rrec = XFS_RMAP_REC_ADDR(block, 3);
459 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
460 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
462 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
465 /* account for rmap btree root */
466 rrec = XFS_RMAP_REC_ADDR(block, 4);
467 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
468 rrec->rm_blockcount = cpu_to_be32(1);
469 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
472 /* account for refc btree root */
473 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
474 rrec = XFS_RMAP_REC_ADDR(block, 5);
475 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
476 rrec->rm_blockcount = cpu_to_be32(1);
477 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
479 be16_add_cpu(&block->bb_numrecs, 1);
482 /* account for the log space */
483 if (is_log_ag(mp, id)) {
484 rrec = XFS_RMAP_REC_ADDR(block,
485 be16_to_cpu(block->bb_numrecs) + 1);
486 rrec->rm_startblock = cpu_to_be32(
487 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
488 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
489 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
491 be16_add_cpu(&block->bb_numrecs, 1);
496 * Initialise new secondary superblocks with the pre-grow geometry, but mark
497 * them as "in progress" so we know they haven't yet been activated. This will
498 * get cleared when the update with the new geometry information is done after
499 * changes to the primary are committed. This isn't strictly necessary, but we
500 * get it for free with the delayed buffer write lists and it means we can tell
501 * if a grow operation didn't complete properly after the fact.
505 struct xfs_mount *mp,
507 struct aghdr_init_data *id)
509 struct xfs_dsb *dsb = bp->b_addr;
511 xfs_sb_to_disk(dsb, &mp->m_sb);
512 dsb->sb_inprogress = 1;
517 struct xfs_mount *mp,
519 struct aghdr_init_data *id)
521 struct xfs_agf *agf = bp->b_addr;
522 xfs_extlen_t tmpsize;
524 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
525 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
526 agf->agf_seqno = cpu_to_be32(id->agno);
527 agf->agf_length = cpu_to_be32(id->agsize);
528 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
529 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
530 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
531 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
532 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
533 agf->agf_roots[XFS_BTNUM_RMAPi] =
534 cpu_to_be32(XFS_RMAP_BLOCK(mp));
535 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
536 agf->agf_rmap_blocks = cpu_to_be32(1);
539 agf->agf_flfirst = cpu_to_be32(1);
541 agf->agf_flcount = 0;
542 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
543 agf->agf_freeblks = cpu_to_be32(tmpsize);
544 agf->agf_longest = cpu_to_be32(tmpsize);
545 if (xfs_sb_version_hascrc(&mp->m_sb))
546 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
547 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
548 agf->agf_refcount_root = cpu_to_be32(
550 agf->agf_refcount_level = cpu_to_be32(1);
551 agf->agf_refcount_blocks = cpu_to_be32(1);
554 if (is_log_ag(mp, id)) {
555 int64_t logblocks = mp->m_sb.sb_logblocks;
557 be32_add_cpu(&agf->agf_freeblks, -logblocks);
558 agf->agf_longest = cpu_to_be32(id->agsize -
559 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
565 struct xfs_mount *mp,
567 struct aghdr_init_data *id)
569 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
573 if (xfs_sb_version_hascrc(&mp->m_sb)) {
574 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
575 agfl->agfl_seqno = cpu_to_be32(id->agno);
576 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
579 agfl_bno = xfs_buf_to_agfl_bno(bp);
580 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
581 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
586 struct xfs_mount *mp,
588 struct aghdr_init_data *id)
590 struct xfs_agi *agi = bp->b_addr;
593 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
594 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
595 agi->agi_seqno = cpu_to_be32(id->agno);
596 agi->agi_length = cpu_to_be32(id->agsize);
598 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
599 agi->agi_level = cpu_to_be32(1);
600 agi->agi_freecount = 0;
601 agi->agi_newino = cpu_to_be32(NULLAGINO);
602 agi->agi_dirino = cpu_to_be32(NULLAGINO);
603 if (xfs_sb_version_hascrc(&mp->m_sb))
604 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
605 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
606 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
607 agi->agi_free_level = cpu_to_be32(1);
609 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
610 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
611 if (xfs_sb_version_hasinobtcounts(&mp->m_sb)) {
612 agi->agi_iblocks = cpu_to_be32(1);
613 if (xfs_sb_version_hasfinobt(&mp->m_sb))
614 agi->agi_fblocks = cpu_to_be32(1);
618 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
619 struct aghdr_init_data *id);
622 struct xfs_mount *mp,
623 struct aghdr_init_data *id,
624 aghdr_init_work_f work,
625 const struct xfs_buf_ops *ops)
630 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
636 xfs_buf_delwri_queue(bp, &id->buffer_list);
641 struct xfs_aghdr_grow_data {
644 const struct xfs_buf_ops *ops;
645 aghdr_init_work_f work;
651 * Prepare new AG headers to be written to disk. We use uncached buffers here,
652 * as it is assumed these new AG headers are currently beyond the currently
653 * valid filesystem address space. Using cached buffers would trip over EOFS
654 * corruption detection alogrithms in the buffer cache lookup routines.
656 * This is a non-transactional function, but the prepared buffers are added to a
657 * delayed write buffer list supplied by the caller so they can submit them to
658 * disk and wait on them as required.
662 struct xfs_mount *mp,
663 struct aghdr_init_data *id)
666 struct xfs_aghdr_grow_data aghdr_data[] = {
668 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
669 .numblks = XFS_FSS_TO_BB(mp, 1),
670 .ops = &xfs_sb_buf_ops,
671 .work = &xfs_sbblock_init,
675 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
676 .numblks = XFS_FSS_TO_BB(mp, 1),
677 .ops = &xfs_agf_buf_ops,
678 .work = &xfs_agfblock_init,
682 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
683 .numblks = XFS_FSS_TO_BB(mp, 1),
684 .ops = &xfs_agfl_buf_ops,
685 .work = &xfs_agflblock_init,
689 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
690 .numblks = XFS_FSS_TO_BB(mp, 1),
691 .ops = &xfs_agi_buf_ops,
692 .work = &xfs_agiblock_init,
695 { /* BNO root block */
696 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
697 .numblks = BTOBB(mp->m_sb.sb_blocksize),
698 .ops = &xfs_bnobt_buf_ops,
699 .work = &xfs_bnoroot_init,
702 { /* CNT root block */
703 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
704 .numblks = BTOBB(mp->m_sb.sb_blocksize),
705 .ops = &xfs_cntbt_buf_ops,
706 .work = &xfs_cntroot_init,
709 { /* INO root block */
710 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
711 .numblks = BTOBB(mp->m_sb.sb_blocksize),
712 .ops = &xfs_inobt_buf_ops,
713 .work = &xfs_btroot_init,
714 .type = XFS_BTNUM_INO,
717 { /* FINO root block */
718 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
719 .numblks = BTOBB(mp->m_sb.sb_blocksize),
720 .ops = &xfs_finobt_buf_ops,
721 .work = &xfs_btroot_init,
722 .type = XFS_BTNUM_FINO,
723 .need_init = xfs_sb_version_hasfinobt(&mp->m_sb)
725 { /* RMAP root block */
726 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
727 .numblks = BTOBB(mp->m_sb.sb_blocksize),
728 .ops = &xfs_rmapbt_buf_ops,
729 .work = &xfs_rmaproot_init,
730 .need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
732 { /* REFC root block */
733 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
734 .numblks = BTOBB(mp->m_sb.sb_blocksize),
735 .ops = &xfs_refcountbt_buf_ops,
736 .work = &xfs_btroot_init,
737 .type = XFS_BTNUM_REFC,
738 .need_init = xfs_sb_version_hasreflink(&mp->m_sb)
740 { /* NULL terminating block */
741 .daddr = XFS_BUF_DADDR_NULL,
744 struct xfs_aghdr_grow_data *dp;
747 /* Account for AG free space in new AG */
748 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
749 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
753 id->daddr = dp->daddr;
754 id->numblks = dp->numblks;
756 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
765 struct xfs_mount *mp,
766 struct xfs_trans **tpp,
770 struct xfs_alloc_arg args = {
773 .type = XFS_ALLOCTYPE_THIS_BNO,
776 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
777 .resv = XFS_AG_RESV_NONE,
780 struct xfs_buf *agibp, *agfbp;
785 ASSERT(agno == mp->m_sb.sb_agcount - 1);
786 error = xfs_ialloc_read_agi(mp, *tpp, agno, &agibp);
792 error = xfs_alloc_read_agf(mp, *tpp, agno, 0, &agfbp);
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;
800 if (delta >= agi->agi_length)
803 args.fsbno = XFS_AGB_TO_FSB(mp, agno,
804 be32_to_cpu(agi->agi_length) - delta);
807 * Disable perag reservations so it doesn't cause the allocation request
808 * to fail. We'll reestablish reservation before we return.
810 error = xfs_ag_resv_free(agibp->b_pag);
814 /* internal log shouldn't also show up in the free space btrees */
815 error = xfs_alloc_vextent(&args);
816 if (!error && args.agbno == NULLAGBLOCK)
821 * if extent allocation fails, need to roll the transaction to
822 * ensure that the AGFL fixup has been committed anyway.
824 xfs_trans_bhold(*tpp, agfbp);
825 err2 = xfs_trans_roll(tpp);
828 xfs_trans_bjoin(*tpp, agfbp);
833 * if successfully deleted from freespace btrees, need to confirm
834 * per-AG reservation works as expected.
836 be32_add_cpu(&agi->agi_length, -delta);
837 be32_add_cpu(&agf->agf_length, -delta);
839 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
841 be32_add_cpu(&agi->agi_length, delta);
842 be32_add_cpu(&agf->agf_length, delta);
846 __xfs_bmap_add_free(*tpp, args.fsbno, delta, NULL, true);
849 * Roll the transaction before trying to re-init the per-ag
850 * reservation. The new transaction is clean so it will cancel
851 * without any side effects.
853 error = xfs_defer_finish(tpp);
860 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
861 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
864 err2 = xfs_ag_resv_init(agibp->b_pag, *tpp);
868 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
869 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
874 * Extent the AG indicated by the @id by the length passed in
878 struct xfs_mount *mp,
879 struct xfs_trans *tp,
880 struct aghdr_init_data *id,
889 * Change the agi length.
891 error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
896 be32_add_cpu(&agi->agi_length, len);
897 ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
898 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
899 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
904 error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
909 be32_add_cpu(&agf->agf_length, len);
910 ASSERT(agf->agf_length == agi->agi_length);
911 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
914 * Free the new space.
916 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
917 * this doesn't actually exist in the rmap btree.
919 error = xfs_rmap_free(tp, bp, bp->b_pag,
920 be32_to_cpu(agf->agf_length) - len,
921 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
925 return xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
926 be32_to_cpu(agf->agf_length) - len),
927 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
931 /* Retrieve AG geometry. */
934 struct xfs_mount *mp,
936 struct xfs_ag_geometry *ageo)
938 struct xfs_buf *agi_bp;
939 struct xfs_buf *agf_bp;
942 struct xfs_perag *pag;
943 unsigned int freeblks;
946 if (agno >= mp->m_sb.sb_agcount)
949 /* Lock the AG headers. */
950 error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
953 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
960 memset(ageo, 0, sizeof(*ageo));
961 ageo->ag_number = agno;
963 agi = agi_bp->b_addr;
964 ageo->ag_icount = be32_to_cpu(agi->agi_count);
965 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
967 agf = agf_bp->b_addr;
968 ageo->ag_length = be32_to_cpu(agf->agf_length);
969 freeblks = pag->pagf_freeblks +
971 pag->pagf_btreeblks -
972 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
973 ageo->ag_freeblks = freeblks;
974 xfs_ag_geom_health(pag, ageo);
976 /* Release resources. */
977 xfs_buf_relse(agf_bp);
979 xfs_buf_relse(agi_bp);