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 and AGI buffers to populate the per-ag
126 pag = xfs_perag_get(mp, index);
127 error = xfs_alloc_read_agf(pag, NULL, 0, NULL);
129 error = xfs_ialloc_read_agi(pag, NULL, NULL);
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_buf_hash_destroy(pag);
199 call_rcu(&pag->rcu_head, __xfs_free_perag);
203 /* Find the size of the AG, in blocks. */
205 __xfs_ag_block_count(
206 struct xfs_mount *mp,
208 xfs_agnumber_t agcount,
209 xfs_rfsblock_t dblocks)
211 ASSERT(agno < agcount);
213 if (agno < agcount - 1)
214 return mp->m_sb.sb_agblocks;
215 return dblocks - (agno * mp->m_sb.sb_agblocks);
220 struct xfs_mount *mp,
223 return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount,
224 mp->m_sb.sb_dblocks);
227 /* Calculate the first and last possible inode number in an AG. */
230 struct xfs_mount *mp,
238 * Calculate the first inode, which will be in the first
239 * cluster-aligned block after the AGFL.
241 bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align);
242 *first = XFS_AGB_TO_AGINO(mp, bno);
245 * Calculate the last inode, which will be at the end of the
246 * last (aligned) cluster that can be allocated in the AG.
248 bno = round_down(eoag, M_IGEO(mp)->cluster_align);
249 *last = XFS_AGB_TO_AGINO(mp, bno) - 1;
254 struct xfs_mount *mp,
259 return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last);
263 xfs_initialize_perag(
264 struct xfs_mount *mp,
265 xfs_agnumber_t agcount,
266 xfs_rfsblock_t dblocks,
267 xfs_agnumber_t *maxagi)
269 struct xfs_perag *pag;
270 xfs_agnumber_t index;
271 xfs_agnumber_t first_initialised = NULLAGNUMBER;
275 * Walk the current per-ag tree so we don't try to initialise AGs
276 * that already exist (growfs case). Allocate and insert all the
277 * AGs we don't find ready for initialisation.
279 for (index = 0; index < agcount; index++) {
280 pag = xfs_perag_get(mp, index);
286 pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
289 goto out_unwind_new_pags;
291 pag->pag_agno = index;
294 error = radix_tree_preload(GFP_NOFS);
298 spin_lock(&mp->m_perag_lock);
299 if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
301 spin_unlock(&mp->m_perag_lock);
302 radix_tree_preload_end();
306 spin_unlock(&mp->m_perag_lock);
307 radix_tree_preload_end();
310 /* Place kernel structure only init below this point. */
311 spin_lock_init(&pag->pag_ici_lock);
312 spin_lock_init(&pag->pagb_lock);
313 spin_lock_init(&pag->pag_state_lock);
314 INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
315 INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
316 init_waitqueue_head(&pag->pagb_wait);
318 pag->pagb_tree = RB_ROOT;
319 #endif /* __KERNEL__ */
321 error = xfs_buf_hash_init(pag);
325 /* first new pag is fully initialized */
326 if (first_initialised == NULLAGNUMBER)
327 first_initialised = index;
330 * Pre-calculated geometry
332 pag->block_count = __xfs_ag_block_count(mp, index, agcount,
334 pag->min_block = XFS_AGFL_BLOCK(mp);
335 __xfs_agino_range(mp, pag->block_count, &pag->agino_min,
339 index = xfs_set_inode_alloc(mp, agcount);
344 mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
348 radix_tree_delete(&mp->m_perag_tree, index);
352 /* unwind any prior newly initialized pags */
353 for (index = first_initialised; index < agcount; index++) {
354 pag = radix_tree_delete(&mp->m_perag_tree, index);
357 xfs_buf_hash_destroy(pag);
365 struct xfs_mount *mp,
368 struct xfs_buf **bpp,
369 const struct xfs_buf_ops *ops)
374 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
378 bp->b_maps[0].bm_bn = blkno;
386 * Generic btree root block init function
390 struct xfs_mount *mp,
392 struct aghdr_init_data *id)
394 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
397 /* Finish initializing a free space btree. */
399 xfs_freesp_init_recs(
400 struct xfs_mount *mp,
402 struct aghdr_init_data *id)
404 struct xfs_alloc_rec *arec;
405 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
407 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
408 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
410 if (xfs_ag_contains_log(mp, id->agno)) {
411 struct xfs_alloc_rec *nrec;
412 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
413 mp->m_sb.sb_logstart);
415 ASSERT(start >= mp->m_ag_prealloc_blocks);
416 if (start != mp->m_ag_prealloc_blocks) {
418 * Modify first record to pad stripe align of log
420 arec->ar_blockcount = cpu_to_be32(start -
421 mp->m_ag_prealloc_blocks);
425 * Insert second record at start of internal log
426 * which then gets trimmed.
428 nrec->ar_startblock = cpu_to_be32(
429 be32_to_cpu(arec->ar_startblock) +
430 be32_to_cpu(arec->ar_blockcount));
432 be16_add_cpu(&block->bb_numrecs, 1);
435 * Change record start to after the internal log
437 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
441 * Calculate the record block count and check for the case where
442 * the log might have consumed all available space in the AG. If
443 * so, reset the record count to 0 to avoid exposure of an invalid
444 * record start block.
446 arec->ar_blockcount = cpu_to_be32(id->agsize -
447 be32_to_cpu(arec->ar_startblock));
448 if (!arec->ar_blockcount)
449 block->bb_numrecs = 0;
453 * Alloc btree root block init functions
457 struct xfs_mount *mp,
459 struct aghdr_init_data *id)
461 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
462 xfs_freesp_init_recs(mp, bp, id);
467 struct xfs_mount *mp,
469 struct aghdr_init_data *id)
471 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
472 xfs_freesp_init_recs(mp, bp, id);
476 * Reverse map root block init
480 struct xfs_mount *mp,
482 struct aghdr_init_data *id)
484 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
485 struct xfs_rmap_rec *rrec;
487 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
490 * mark the AG header regions as static metadata The BNO
491 * btree block is the first block after the headers, so
492 * it's location defines the size of region the static
495 * Note: unlike mkfs, we never have to account for log
496 * space when growing the data regions
498 rrec = XFS_RMAP_REC_ADDR(block, 1);
499 rrec->rm_startblock = 0;
500 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
501 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
504 /* account freespace btree root blocks */
505 rrec = XFS_RMAP_REC_ADDR(block, 2);
506 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
507 rrec->rm_blockcount = cpu_to_be32(2);
508 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
511 /* account inode btree root blocks */
512 rrec = XFS_RMAP_REC_ADDR(block, 3);
513 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
514 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
516 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
519 /* account for rmap btree root */
520 rrec = XFS_RMAP_REC_ADDR(block, 4);
521 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
522 rrec->rm_blockcount = cpu_to_be32(1);
523 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
526 /* account for refc btree root */
527 if (xfs_has_reflink(mp)) {
528 rrec = XFS_RMAP_REC_ADDR(block, 5);
529 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
530 rrec->rm_blockcount = cpu_to_be32(1);
531 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
533 be16_add_cpu(&block->bb_numrecs, 1);
536 /* account for the log space */
537 if (xfs_ag_contains_log(mp, id->agno)) {
538 rrec = XFS_RMAP_REC_ADDR(block,
539 be16_to_cpu(block->bb_numrecs) + 1);
540 rrec->rm_startblock = cpu_to_be32(
541 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
542 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
543 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
545 be16_add_cpu(&block->bb_numrecs, 1);
550 * Initialise new secondary superblocks with the pre-grow geometry, but mark
551 * them as "in progress" so we know they haven't yet been activated. This will
552 * get cleared when the update with the new geometry information is done after
553 * changes to the primary are committed. This isn't strictly necessary, but we
554 * get it for free with the delayed buffer write lists and it means we can tell
555 * if a grow operation didn't complete properly after the fact.
559 struct xfs_mount *mp,
561 struct aghdr_init_data *id)
563 struct xfs_dsb *dsb = bp->b_addr;
565 xfs_sb_to_disk(dsb, &mp->m_sb);
566 dsb->sb_inprogress = 1;
571 struct xfs_mount *mp,
573 struct aghdr_init_data *id)
575 struct xfs_agf *agf = bp->b_addr;
576 xfs_extlen_t tmpsize;
578 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
579 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
580 agf->agf_seqno = cpu_to_be32(id->agno);
581 agf->agf_length = cpu_to_be32(id->agsize);
582 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
583 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
584 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
585 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
586 if (xfs_has_rmapbt(mp)) {
587 agf->agf_roots[XFS_BTNUM_RMAPi] =
588 cpu_to_be32(XFS_RMAP_BLOCK(mp));
589 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
590 agf->agf_rmap_blocks = cpu_to_be32(1);
593 agf->agf_flfirst = cpu_to_be32(1);
595 agf->agf_flcount = 0;
596 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
597 agf->agf_freeblks = cpu_to_be32(tmpsize);
598 agf->agf_longest = cpu_to_be32(tmpsize);
600 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
601 if (xfs_has_reflink(mp)) {
602 agf->agf_refcount_root = cpu_to_be32(
604 agf->agf_refcount_level = cpu_to_be32(1);
605 agf->agf_refcount_blocks = cpu_to_be32(1);
608 if (xfs_ag_contains_log(mp, id->agno)) {
609 int64_t logblocks = mp->m_sb.sb_logblocks;
611 be32_add_cpu(&agf->agf_freeblks, -logblocks);
612 agf->agf_longest = cpu_to_be32(id->agsize -
613 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
619 struct xfs_mount *mp,
621 struct aghdr_init_data *id)
623 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
627 if (xfs_has_crc(mp)) {
628 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
629 agfl->agfl_seqno = cpu_to_be32(id->agno);
630 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
633 agfl_bno = xfs_buf_to_agfl_bno(bp);
634 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
635 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
640 struct xfs_mount *mp,
642 struct aghdr_init_data *id)
644 struct xfs_agi *agi = bp->b_addr;
647 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
648 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
649 agi->agi_seqno = cpu_to_be32(id->agno);
650 agi->agi_length = cpu_to_be32(id->agsize);
652 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
653 agi->agi_level = cpu_to_be32(1);
654 agi->agi_freecount = 0;
655 agi->agi_newino = cpu_to_be32(NULLAGINO);
656 agi->agi_dirino = cpu_to_be32(NULLAGINO);
658 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
659 if (xfs_has_finobt(mp)) {
660 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
661 agi->agi_free_level = cpu_to_be32(1);
663 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
664 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
665 if (xfs_has_inobtcounts(mp)) {
666 agi->agi_iblocks = cpu_to_be32(1);
667 if (xfs_has_finobt(mp))
668 agi->agi_fblocks = cpu_to_be32(1);
672 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
673 struct aghdr_init_data *id);
676 struct xfs_mount *mp,
677 struct aghdr_init_data *id,
678 aghdr_init_work_f work,
679 const struct xfs_buf_ops *ops)
684 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
690 xfs_buf_delwri_queue(bp, &id->buffer_list);
695 struct xfs_aghdr_grow_data {
698 const struct xfs_buf_ops *ops;
699 aghdr_init_work_f work;
705 * Prepare new AG headers to be written to disk. We use uncached buffers here,
706 * as it is assumed these new AG headers are currently beyond the currently
707 * valid filesystem address space. Using cached buffers would trip over EOFS
708 * corruption detection alogrithms in the buffer cache lookup routines.
710 * This is a non-transactional function, but the prepared buffers are added to a
711 * delayed write buffer list supplied by the caller so they can submit them to
712 * disk and wait on them as required.
716 struct xfs_mount *mp,
717 struct aghdr_init_data *id)
720 struct xfs_aghdr_grow_data aghdr_data[] = {
722 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
723 .numblks = XFS_FSS_TO_BB(mp, 1),
724 .ops = &xfs_sb_buf_ops,
725 .work = &xfs_sbblock_init,
729 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
730 .numblks = XFS_FSS_TO_BB(mp, 1),
731 .ops = &xfs_agf_buf_ops,
732 .work = &xfs_agfblock_init,
736 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
737 .numblks = XFS_FSS_TO_BB(mp, 1),
738 .ops = &xfs_agfl_buf_ops,
739 .work = &xfs_agflblock_init,
743 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
744 .numblks = XFS_FSS_TO_BB(mp, 1),
745 .ops = &xfs_agi_buf_ops,
746 .work = &xfs_agiblock_init,
749 { /* BNO root block */
750 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
751 .numblks = BTOBB(mp->m_sb.sb_blocksize),
752 .ops = &xfs_bnobt_buf_ops,
753 .work = &xfs_bnoroot_init,
756 { /* CNT root block */
757 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
758 .numblks = BTOBB(mp->m_sb.sb_blocksize),
759 .ops = &xfs_cntbt_buf_ops,
760 .work = &xfs_cntroot_init,
763 { /* INO root block */
764 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
765 .numblks = BTOBB(mp->m_sb.sb_blocksize),
766 .ops = &xfs_inobt_buf_ops,
767 .work = &xfs_btroot_init,
768 .type = XFS_BTNUM_INO,
771 { /* FINO root block */
772 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
773 .numblks = BTOBB(mp->m_sb.sb_blocksize),
774 .ops = &xfs_finobt_buf_ops,
775 .work = &xfs_btroot_init,
776 .type = XFS_BTNUM_FINO,
777 .need_init = xfs_has_finobt(mp)
779 { /* RMAP root block */
780 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
781 .numblks = BTOBB(mp->m_sb.sb_blocksize),
782 .ops = &xfs_rmapbt_buf_ops,
783 .work = &xfs_rmaproot_init,
784 .need_init = xfs_has_rmapbt(mp)
786 { /* REFC root block */
787 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
788 .numblks = BTOBB(mp->m_sb.sb_blocksize),
789 .ops = &xfs_refcountbt_buf_ops,
790 .work = &xfs_btroot_init,
791 .type = XFS_BTNUM_REFC,
792 .need_init = xfs_has_reflink(mp)
794 { /* NULL terminating block */
795 .daddr = XFS_BUF_DADDR_NULL,
798 struct xfs_aghdr_grow_data *dp;
801 /* Account for AG free space in new AG */
802 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
803 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
807 id->daddr = dp->daddr;
808 id->numblks = dp->numblks;
810 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
819 struct xfs_perag *pag,
820 struct xfs_trans **tpp,
823 struct xfs_mount *mp = pag->pag_mount;
824 struct xfs_alloc_arg args = {
827 .type = XFS_ALLOCTYPE_THIS_BNO,
830 .oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
831 .resv = XFS_AG_RESV_NONE,
834 struct xfs_buf *agibp, *agfbp;
840 ASSERT(pag->pag_agno == mp->m_sb.sb_agcount - 1);
841 error = xfs_ialloc_read_agi(pag, *tpp, &agibp);
847 error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp);
852 aglen = be32_to_cpu(agi->agi_length);
853 /* some extra paranoid checks before we shrink the ag */
854 if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length))
855 return -EFSCORRUPTED;
859 args.fsbno = XFS_AGB_TO_FSB(mp, pag->pag_agno, aglen - delta);
862 * Make sure that the last inode cluster cannot overlap with the new
863 * end of the AG, even if it's sparse.
865 error = xfs_ialloc_check_shrink(*tpp, pag->pag_agno, agibp,
871 * Disable perag reservations so it doesn't cause the allocation request
872 * to fail. We'll reestablish reservation before we return.
874 error = xfs_ag_resv_free(pag);
878 /* internal log shouldn't also show up in the free space btrees */
879 error = xfs_alloc_vextent(&args);
880 if (!error && args.agbno == NULLAGBLOCK)
885 * if extent allocation fails, need to roll the transaction to
886 * ensure that the AGFL fixup has been committed anyway.
888 xfs_trans_bhold(*tpp, agfbp);
889 err2 = xfs_trans_roll(tpp);
892 xfs_trans_bjoin(*tpp, agfbp);
897 * if successfully deleted from freespace btrees, need to confirm
898 * per-AG reservation works as expected.
900 be32_add_cpu(&agi->agi_length, -delta);
901 be32_add_cpu(&agf->agf_length, -delta);
903 err2 = xfs_ag_resv_init(pag, *tpp);
905 be32_add_cpu(&agi->agi_length, delta);
906 be32_add_cpu(&agf->agf_length, delta);
910 __xfs_free_extent_later(*tpp, args.fsbno, delta, NULL, true);
913 * Roll the transaction before trying to re-init the per-ag
914 * reservation. The new transaction is clean so it will cancel
915 * without any side effects.
917 error = xfs_defer_finish(tpp);
924 xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
925 xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
929 err2 = xfs_ag_resv_init(pag, *tpp);
933 xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
934 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
939 * Extent the AG indicated by the @id by the length passed in
943 struct xfs_perag *pag,
944 struct xfs_trans *tp,
952 ASSERT(pag->pag_agno == pag->pag_mount->m_sb.sb_agcount - 1);
954 error = xfs_ialloc_read_agi(pag, tp, &bp);
959 be32_add_cpu(&agi->agi_length, len);
960 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
965 error = xfs_alloc_read_agf(pag, tp, 0, &bp);
970 be32_add_cpu(&agf->agf_length, len);
971 ASSERT(agf->agf_length == agi->agi_length);
972 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
975 * Free the new space.
977 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
978 * this doesn't actually exist in the rmap btree.
980 error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len,
981 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
985 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(pag->pag_mount, pag->pag_agno,
986 be32_to_cpu(agf->agf_length) - len),
987 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
992 /* Update perag geometry */
993 pag->block_count = be32_to_cpu(agf->agf_length);
994 __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min,
999 /* Retrieve AG geometry. */
1001 xfs_ag_get_geometry(
1002 struct xfs_perag *pag,
1003 struct xfs_ag_geometry *ageo)
1005 struct xfs_buf *agi_bp;
1006 struct xfs_buf *agf_bp;
1007 struct xfs_agi *agi;
1008 struct xfs_agf *agf;
1009 unsigned int freeblks;
1012 /* Lock the AG headers. */
1013 error = xfs_ialloc_read_agi(pag, NULL, &agi_bp);
1016 error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp);
1020 /* Fill out form. */
1021 memset(ageo, 0, sizeof(*ageo));
1022 ageo->ag_number = pag->pag_agno;
1024 agi = agi_bp->b_addr;
1025 ageo->ag_icount = be32_to_cpu(agi->agi_count);
1026 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
1028 agf = agf_bp->b_addr;
1029 ageo->ag_length = be32_to_cpu(agf->agf_length);
1030 freeblks = pag->pagf_freeblks +
1032 pag->pagf_btreeblks -
1033 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
1034 ageo->ag_freeblks = freeblks;
1035 xfs_ag_geom_health(pag, ageo);
1037 /* Release resources. */
1038 xfs_buf_relse(agf_bp);
1040 xfs_buf_relse(agi_bp);