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;
49 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
51 trace_xfs_perag_get(pag, _RET_IP_);
52 ASSERT(atomic_read(&pag->pag_ref) >= 0);
53 atomic_inc(&pag->pag_ref);
60 * search from @first to find the next perag with the given tag set.
68 struct xfs_perag *pag;
72 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
73 (void **)&pag, first, 1, tag);
78 trace_xfs_perag_get_tag(pag, _RET_IP_);
79 atomic_inc(&pag->pag_ref);
84 /* Get a passive reference to the given perag. */
87 struct xfs_perag *pag)
89 ASSERT(atomic_read(&pag->pag_ref) > 0 ||
90 atomic_read(&pag->pag_active_ref) > 0);
92 trace_xfs_perag_hold(pag, _RET_IP_);
93 atomic_inc(&pag->pag_ref);
99 struct xfs_perag *pag)
101 trace_xfs_perag_put(pag, _RET_IP_);
102 ASSERT(atomic_read(&pag->pag_ref) > 0);
103 atomic_dec(&pag->pag_ref);
107 * Active references for perag structures. This is for short term access to the
108 * per ag structures for walking trees or accessing state. If an AG is being
109 * shrunk or is offline, then this will fail to find that AG and return NULL
114 struct xfs_mount *mp,
117 struct xfs_perag *pag;
120 pag = radix_tree_lookup(&mp->m_perag_tree, agno);
122 trace_xfs_perag_grab(pag, _RET_IP_);
123 if (!atomic_inc_not_zero(&pag->pag_active_ref))
131 * search from @first to find the next perag with the given tag set.
135 struct xfs_mount *mp,
136 xfs_agnumber_t first,
139 struct xfs_perag *pag;
143 found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
144 (void **)&pag, first, 1, tag);
149 trace_xfs_perag_grab_tag(pag, _RET_IP_);
150 if (!atomic_inc_not_zero(&pag->pag_active_ref))
158 struct xfs_perag *pag)
160 trace_xfs_perag_rele(pag, _RET_IP_);
161 if (atomic_dec_and_test(&pag->pag_active_ref))
162 wake_up(&pag->pag_active_wq);
166 * xfs_initialize_perag_data
168 * Read in each per-ag structure so we can count up the number of
169 * allocated inodes, free inodes and used filesystem blocks as this
170 * information is no longer persistent in the superblock. Once we have
171 * this information, write it into the in-core superblock structure.
174 xfs_initialize_perag_data(
175 struct xfs_mount *mp,
176 xfs_agnumber_t agcount)
178 xfs_agnumber_t index;
179 struct xfs_perag *pag;
180 struct xfs_sb *sbp = &mp->m_sb;
184 uint64_t bfreelst = 0;
189 for (index = 0; index < agcount; index++) {
191 * Read the AGF and AGI buffers to populate the per-ag
194 pag = xfs_perag_get(mp, index);
195 error = xfs_alloc_read_agf(pag, NULL, 0, NULL);
197 error = xfs_ialloc_read_agi(pag, NULL, NULL);
203 ifree += pag->pagi_freecount;
204 ialloc += pag->pagi_count;
205 bfree += pag->pagf_freeblks;
206 bfreelst += pag->pagf_flcount;
207 btree += pag->pagf_btreeblks;
210 fdblocks = bfree + bfreelst + btree;
213 * If the new summary counts are obviously incorrect, fail the
214 * mount operation because that implies the AGFs are also corrupt.
215 * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
216 * will prevent xfs_repair from fixing anything.
218 if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
219 xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
220 error = -EFSCORRUPTED;
224 /* Overwrite incore superblock counters with just-read data */
225 spin_lock(&mp->m_sb_lock);
226 sbp->sb_ifree = ifree;
227 sbp->sb_icount = ialloc;
228 sbp->sb_fdblocks = fdblocks;
229 spin_unlock(&mp->m_sb_lock);
231 xfs_reinit_percpu_counters(mp);
233 xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
239 struct rcu_head *head)
241 struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
243 ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
248 * Free up the per-ag resources associated with the mount structure.
252 struct xfs_mount *mp)
254 struct xfs_perag *pag;
257 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
258 spin_lock(&mp->m_perag_lock);
259 pag = radix_tree_delete(&mp->m_perag_tree, agno);
260 spin_unlock(&mp->m_perag_lock);
262 XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0);
263 xfs_defer_drain_free(&pag->pag_intents_drain);
265 cancel_delayed_work_sync(&pag->pag_blockgc_work);
266 xfs_buf_hash_destroy(pag);
268 /* drop the mount's active reference */
270 XFS_IS_CORRUPT(pag->pag_mount,
271 atomic_read(&pag->pag_active_ref) != 0);
272 call_rcu(&pag->rcu_head, __xfs_free_perag);
276 /* Find the size of the AG, in blocks. */
278 __xfs_ag_block_count(
279 struct xfs_mount *mp,
281 xfs_agnumber_t agcount,
282 xfs_rfsblock_t dblocks)
284 ASSERT(agno < agcount);
286 if (agno < agcount - 1)
287 return mp->m_sb.sb_agblocks;
288 return dblocks - (agno * mp->m_sb.sb_agblocks);
293 struct xfs_mount *mp,
296 return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount,
297 mp->m_sb.sb_dblocks);
300 /* Calculate the first and last possible inode number in an AG. */
303 struct xfs_mount *mp,
311 * Calculate the first inode, which will be in the first
312 * cluster-aligned block after the AGFL.
314 bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align);
315 *first = XFS_AGB_TO_AGINO(mp, bno);
318 * Calculate the last inode, which will be at the end of the
319 * last (aligned) cluster that can be allocated in the AG.
321 bno = round_down(eoag, M_IGEO(mp)->cluster_align);
322 *last = XFS_AGB_TO_AGINO(mp, bno) - 1;
327 struct xfs_mount *mp,
332 return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last);
336 xfs_initialize_perag(
337 struct xfs_mount *mp,
338 xfs_agnumber_t agcount,
339 xfs_rfsblock_t dblocks,
340 xfs_agnumber_t *maxagi)
342 struct xfs_perag *pag;
343 xfs_agnumber_t index;
344 xfs_agnumber_t first_initialised = NULLAGNUMBER;
348 * Walk the current per-ag tree so we don't try to initialise AGs
349 * that already exist (growfs case). Allocate and insert all the
350 * AGs we don't find ready for initialisation.
352 for (index = 0; index < agcount; index++) {
353 pag = xfs_perag_get(mp, index);
359 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
362 goto out_unwind_new_pags;
364 pag->pag_agno = index;
367 error = radix_tree_preload(GFP_NOFS);
371 spin_lock(&mp->m_perag_lock);
372 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
374 spin_unlock(&mp->m_perag_lock);
375 radix_tree_preload_end();
379 spin_unlock(&mp->m_perag_lock);
380 radix_tree_preload_end();
383 /* Place kernel structure only init below this point. */
384 spin_lock_init(&pag->pag_ici_lock);
385 spin_lock_init(&pag->pagb_lock);
386 spin_lock_init(&pag->pag_state_lock);
387 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
388 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
389 xfs_defer_drain_init(&pag->pag_intents_drain);
390 init_waitqueue_head(&pag->pagb_wait);
391 init_waitqueue_head(&pag->pag_active_wq);
393 pag->pagb_tree = RB_ROOT;
394 #endif /* __KERNEL__ */
396 error = xfs_buf_hash_init(pag);
400 /* Active ref owned by mount indicates AG is online. */
401 atomic_set(&pag->pag_active_ref, 1);
403 /* first new pag is fully initialized */
404 if (first_initialised == NULLAGNUMBER)
405 first_initialised = index;
408 * Pre-calculated geometry
410 pag->block_count = __xfs_ag_block_count(mp, index, agcount,
412 pag->min_block = XFS_AGFL_BLOCK(mp);
413 __xfs_agino_range(mp, pag->block_count, &pag->agino_min,
417 index = xfs_set_inode_alloc(mp, agcount);
422 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
426 xfs_defer_drain_free(&pag->pag_intents_drain);
427 radix_tree_delete(&mp->m_perag_tree, index);
431 /* unwind any prior newly initialized pags */
432 for (index = first_initialised; index < agcount; index++) {
433 pag = radix_tree_delete(&mp->m_perag_tree, index);
436 xfs_buf_hash_destroy(pag);
437 xfs_defer_drain_free(&pag->pag_intents_drain);
445 struct xfs_mount *mp,
448 struct xfs_buf **bpp,
449 const struct xfs_buf_ops *ops)
454 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
458 bp->b_maps[0].bm_bn = blkno;
466 * Generic btree root block init function
470 struct xfs_mount *mp,
472 struct aghdr_init_data *id)
474 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
477 /* Finish initializing a free space btree. */
479 xfs_freesp_init_recs(
480 struct xfs_mount *mp,
482 struct aghdr_init_data *id)
484 struct xfs_alloc_rec *arec;
485 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
487 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
488 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
490 if (xfs_ag_contains_log(mp, id->agno)) {
491 struct xfs_alloc_rec *nrec;
492 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
493 mp->m_sb.sb_logstart);
495 ASSERT(start >= mp->m_ag_prealloc_blocks);
496 if (start != mp->m_ag_prealloc_blocks) {
498 * Modify first record to pad stripe align of log and
499 * bump the record count.
501 arec->ar_blockcount = cpu_to_be32(start -
502 mp->m_ag_prealloc_blocks);
503 be16_add_cpu(&block->bb_numrecs, 1);
507 * Insert second record at start of internal log
508 * which then gets trimmed.
510 nrec->ar_startblock = cpu_to_be32(
511 be32_to_cpu(arec->ar_startblock) +
512 be32_to_cpu(arec->ar_blockcount));
516 * Change record start to after the internal log
518 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
522 * Calculate the block count of this record; if it is nonzero,
523 * increment the record count.
525 arec->ar_blockcount = cpu_to_be32(id->agsize -
526 be32_to_cpu(arec->ar_startblock));
527 if (arec->ar_blockcount)
528 be16_add_cpu(&block->bb_numrecs, 1);
532 * Alloc btree root block init functions
536 struct xfs_mount *mp,
538 struct aghdr_init_data *id)
540 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 0, id->agno);
541 xfs_freesp_init_recs(mp, bp, id);
546 struct xfs_mount *mp,
548 struct aghdr_init_data *id)
550 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 0, id->agno);
551 xfs_freesp_init_recs(mp, bp, id);
555 * Reverse map root block init
559 struct xfs_mount *mp,
561 struct aghdr_init_data *id)
563 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
564 struct xfs_rmap_rec *rrec;
566 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
569 * mark the AG header regions as static metadata The BNO
570 * btree block is the first block after the headers, so
571 * it's location defines the size of region the static
574 * Note: unlike mkfs, we never have to account for log
575 * space when growing the data regions
577 rrec = XFS_RMAP_REC_ADDR(block, 1);
578 rrec->rm_startblock = 0;
579 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
580 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
583 /* account freespace btree root blocks */
584 rrec = XFS_RMAP_REC_ADDR(block, 2);
585 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
586 rrec->rm_blockcount = cpu_to_be32(2);
587 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
590 /* account inode btree root blocks */
591 rrec = XFS_RMAP_REC_ADDR(block, 3);
592 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
593 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
595 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
598 /* account for rmap btree root */
599 rrec = XFS_RMAP_REC_ADDR(block, 4);
600 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
601 rrec->rm_blockcount = cpu_to_be32(1);
602 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
605 /* account for refc btree root */
606 if (xfs_has_reflink(mp)) {
607 rrec = XFS_RMAP_REC_ADDR(block, 5);
608 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
609 rrec->rm_blockcount = cpu_to_be32(1);
610 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
612 be16_add_cpu(&block->bb_numrecs, 1);
615 /* account for the log space */
616 if (xfs_ag_contains_log(mp, id->agno)) {
617 rrec = XFS_RMAP_REC_ADDR(block,
618 be16_to_cpu(block->bb_numrecs) + 1);
619 rrec->rm_startblock = cpu_to_be32(
620 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
621 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
622 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
624 be16_add_cpu(&block->bb_numrecs, 1);
629 * Initialise new secondary superblocks with the pre-grow geometry, but mark
630 * them as "in progress" so we know they haven't yet been activated. This will
631 * get cleared when the update with the new geometry information is done after
632 * changes to the primary are committed. This isn't strictly necessary, but we
633 * get it for free with the delayed buffer write lists and it means we can tell
634 * if a grow operation didn't complete properly after the fact.
638 struct xfs_mount *mp,
640 struct aghdr_init_data *id)
642 struct xfs_dsb *dsb = bp->b_addr;
644 xfs_sb_to_disk(dsb, &mp->m_sb);
645 dsb->sb_inprogress = 1;
650 struct xfs_mount *mp,
652 struct aghdr_init_data *id)
654 struct xfs_agf *agf = bp->b_addr;
655 xfs_extlen_t tmpsize;
657 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
658 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
659 agf->agf_seqno = cpu_to_be32(id->agno);
660 agf->agf_length = cpu_to_be32(id->agsize);
661 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
662 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
663 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
664 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
665 if (xfs_has_rmapbt(mp)) {
666 agf->agf_roots[XFS_BTNUM_RMAPi] =
667 cpu_to_be32(XFS_RMAP_BLOCK(mp));
668 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
669 agf->agf_rmap_blocks = cpu_to_be32(1);
672 agf->agf_flfirst = cpu_to_be32(1);
674 agf->agf_flcount = 0;
675 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
676 agf->agf_freeblks = cpu_to_be32(tmpsize);
677 agf->agf_longest = cpu_to_be32(tmpsize);
679 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
680 if (xfs_has_reflink(mp)) {
681 agf->agf_refcount_root = cpu_to_be32(
683 agf->agf_refcount_level = cpu_to_be32(1);
684 agf->agf_refcount_blocks = cpu_to_be32(1);
687 if (xfs_ag_contains_log(mp, id->agno)) {
688 int64_t logblocks = mp->m_sb.sb_logblocks;
690 be32_add_cpu(&agf->agf_freeblks, -logblocks);
691 agf->agf_longest = cpu_to_be32(id->agsize -
692 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
698 struct xfs_mount *mp,
700 struct aghdr_init_data *id)
702 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
706 if (xfs_has_crc(mp)) {
707 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
708 agfl->agfl_seqno = cpu_to_be32(id->agno);
709 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
712 agfl_bno = xfs_buf_to_agfl_bno(bp);
713 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
714 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
719 struct xfs_mount *mp,
721 struct aghdr_init_data *id)
723 struct xfs_agi *agi = bp->b_addr;
726 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
727 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
728 agi->agi_seqno = cpu_to_be32(id->agno);
729 agi->agi_length = cpu_to_be32(id->agsize);
731 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
732 agi->agi_level = cpu_to_be32(1);
733 agi->agi_freecount = 0;
734 agi->agi_newino = cpu_to_be32(NULLAGINO);
735 agi->agi_dirino = cpu_to_be32(NULLAGINO);
737 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
738 if (xfs_has_finobt(mp)) {
739 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
740 agi->agi_free_level = cpu_to_be32(1);
742 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
743 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
744 if (xfs_has_inobtcounts(mp)) {
745 agi->agi_iblocks = cpu_to_be32(1);
746 if (xfs_has_finobt(mp))
747 agi->agi_fblocks = cpu_to_be32(1);
751 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
752 struct aghdr_init_data *id);
755 struct xfs_mount *mp,
756 struct aghdr_init_data *id,
757 aghdr_init_work_f work,
758 const struct xfs_buf_ops *ops)
763 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
769 xfs_buf_delwri_queue(bp, &id->buffer_list);
774 struct xfs_aghdr_grow_data {
777 const struct xfs_buf_ops *ops;
778 aghdr_init_work_f work;
784 * Prepare new AG headers to be written to disk. We use uncached buffers here,
785 * as it is assumed these new AG headers are currently beyond the currently
786 * valid filesystem address space. Using cached buffers would trip over EOFS
787 * corruption detection alogrithms in the buffer cache lookup routines.
789 * This is a non-transactional function, but the prepared buffers are added to a
790 * delayed write buffer list supplied by the caller so they can submit them to
791 * disk and wait on them as required.
795 struct xfs_mount *mp,
796 struct aghdr_init_data *id)
799 struct xfs_aghdr_grow_data aghdr_data[] = {
801 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
802 .numblks = XFS_FSS_TO_BB(mp, 1),
803 .ops = &xfs_sb_buf_ops,
804 .work = &xfs_sbblock_init,
808 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
809 .numblks = XFS_FSS_TO_BB(mp, 1),
810 .ops = &xfs_agf_buf_ops,
811 .work = &xfs_agfblock_init,
815 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
816 .numblks = XFS_FSS_TO_BB(mp, 1),
817 .ops = &xfs_agfl_buf_ops,
818 .work = &xfs_agflblock_init,
822 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
823 .numblks = XFS_FSS_TO_BB(mp, 1),
824 .ops = &xfs_agi_buf_ops,
825 .work = &xfs_agiblock_init,
828 { /* BNO root block */
829 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
830 .numblks = BTOBB(mp->m_sb.sb_blocksize),
831 .ops = &xfs_bnobt_buf_ops,
832 .work = &xfs_bnoroot_init,
835 { /* CNT root block */
836 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
837 .numblks = BTOBB(mp->m_sb.sb_blocksize),
838 .ops = &xfs_cntbt_buf_ops,
839 .work = &xfs_cntroot_init,
842 { /* INO root block */
843 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
844 .numblks = BTOBB(mp->m_sb.sb_blocksize),
845 .ops = &xfs_inobt_buf_ops,
846 .work = &xfs_btroot_init,
847 .type = XFS_BTNUM_INO,
850 { /* FINO root block */
851 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
852 .numblks = BTOBB(mp->m_sb.sb_blocksize),
853 .ops = &xfs_finobt_buf_ops,
854 .work = &xfs_btroot_init,
855 .type = XFS_BTNUM_FINO,
856 .need_init = xfs_has_finobt(mp)
858 { /* RMAP root block */
859 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
860 .numblks = BTOBB(mp->m_sb.sb_blocksize),
861 .ops = &xfs_rmapbt_buf_ops,
862 .work = &xfs_rmaproot_init,
863 .need_init = xfs_has_rmapbt(mp)
865 { /* REFC root block */
866 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
867 .numblks = BTOBB(mp->m_sb.sb_blocksize),
868 .ops = &xfs_refcountbt_buf_ops,
869 .work = &xfs_btroot_init,
870 .type = XFS_BTNUM_REFC,
871 .need_init = xfs_has_reflink(mp)
873 { /* NULL terminating block */
874 .daddr = XFS_BUF_DADDR_NULL,
877 struct xfs_aghdr_grow_data *dp;
880 /* Account for AG free space in new AG */
881 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
882 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
886 id->daddr = dp->daddr;
887 id->numblks = dp->numblks;
889 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
898 struct xfs_perag *pag,
899 struct xfs_trans **tpp,
902 struct xfs_mount *mp = pag->pag_mount;
903 struct xfs_alloc_arg args = {
909 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
910 .resv = XFS_AG_RESV_NONE,
913 struct xfs_buf *agibp, *agfbp;
919 ASSERT(pag->pag_agno == mp->m_sb.sb_agcount - 1);
920 error = xfs_ialloc_read_agi(pag, *tpp, &agibp);
926 error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp);
931 aglen = be32_to_cpu(agi->agi_length);
932 /* some extra paranoid checks before we shrink the ag */
933 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length))
934 return -EFSCORRUPTED;
939 * Make sure that the last inode cluster cannot overlap with the new
940 * end of the AG, even if it's sparse.
942 error = xfs_ialloc_check_shrink(pag, *tpp, agibp, aglen - delta);
947 * Disable perag reservations so it doesn't cause the allocation request
948 * to fail. We'll reestablish reservation before we return.
950 error = xfs_ag_resv_free(pag);
954 /* internal log shouldn't also show up in the free space btrees */
955 error = xfs_alloc_vextent_exact_bno(&args,
956 XFS_AGB_TO_FSB(mp, pag->pag_agno, aglen - delta));
957 if (!error && args.agbno == NULLAGBLOCK)
962 * if extent allocation fails, need to roll the transaction to
963 * ensure that the AGFL fixup has been committed anyway.
965 xfs_trans_bhold(*tpp, agfbp);
966 err2 = xfs_trans_roll(tpp);
969 xfs_trans_bjoin(*tpp, agfbp);
974 * if successfully deleted from freespace btrees, need to confirm
975 * per-AG reservation works as expected.
977 be32_add_cpu(&agi->agi_length, -delta);
978 be32_add_cpu(&agf->agf_length, -delta);
980 err2 = xfs_ag_resv_init(pag, *tpp);
982 be32_add_cpu(&agi->agi_length, delta);
983 be32_add_cpu(&agf->agf_length, delta);
987 err2 = __xfs_free_extent_later(*tpp, args.fsbno, delta, NULL,
988 XFS_AG_RESV_NONE, true);
993 * Roll the transaction before trying to re-init the per-ag
994 * reservation. The new transaction is clean so it will cancel
995 * without any side effects.
997 error = xfs_defer_finish(tpp);
1004 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
1005 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
1009 err2 = xfs_ag_resv_init(pag, *tpp);
1013 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
1014 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1019 * Extent the AG indicated by the @id by the length passed in
1022 xfs_ag_extend_space(
1023 struct xfs_perag *pag,
1024 struct xfs_trans *tp,
1028 struct xfs_agi *agi;
1029 struct xfs_agf *agf;
1032 ASSERT(pag->pag_agno == pag->pag_mount->m_sb.sb_agcount - 1);
1034 error = xfs_ialloc_read_agi(pag, tp, &bp);
1039 be32_add_cpu(&agi->agi_length, len);
1040 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
1043 * Change agf length.
1045 error = xfs_alloc_read_agf(pag, tp, 0, &bp);
1050 be32_add_cpu(&agf->agf_length, len);
1051 ASSERT(agf->agf_length == agi->agi_length);
1052 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
1055 * Free the new space.
1057 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
1058 * this doesn't actually exist in the rmap btree.
1060 error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len,
1061 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
1065 error = xfs_free_extent(tp, pag, be32_to_cpu(agf->agf_length) - len,
1066 len, &XFS_RMAP_OINFO_SKIP_UPDATE, XFS_AG_RESV_NONE);
1070 /* Update perag geometry */
1071 pag->block_count = be32_to_cpu(agf->agf_length);
1072 __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min,
1077 /* Retrieve AG geometry. */
1079 xfs_ag_get_geometry(
1080 struct xfs_perag *pag,
1081 struct xfs_ag_geometry *ageo)
1083 struct xfs_buf *agi_bp;
1084 struct xfs_buf *agf_bp;
1085 struct xfs_agi *agi;
1086 struct xfs_agf *agf;
1087 unsigned int freeblks;
1090 /* Lock the AG headers. */
1091 error = xfs_ialloc_read_agi(pag, NULL, &agi_bp);
1094 error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp);
1098 /* Fill out form. */
1099 memset(ageo, 0, sizeof(*ageo));
1100 ageo->ag_number = pag->pag_agno;
1102 agi = agi_bp->b_addr;
1103 ageo->ag_icount = be32_to_cpu(agi->agi_count);
1104 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
1106 agf = agf_bp->b_addr;
1107 ageo->ag_length = be32_to_cpu(agf->agf_length);
1108 freeblks = pag->pagf_freeblks +
1110 pag->pagf_btreeblks -
1111 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
1112 ageo->ag_freeblks = freeblks;
1113 xfs_ag_geom_health(pag, ageo);
1115 /* Release resources. */
1116 xfs_buf_relse(agf_bp);
1118 xfs_buf_relse(agi_bp);