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"
32 const struct xfs_buf_ops *ops)
37 error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
41 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
43 bp->b_maps[0].bm_bn = blkno;
50 static inline bool is_log_ag(struct xfs_mount *mp, struct aghdr_init_data *id)
52 return mp->m_sb.sb_logstart > 0 &&
53 id->agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
57 * Generic btree root block init function
63 struct aghdr_init_data *id)
65 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
68 /* Finish initializing a free space btree. */
73 struct aghdr_init_data *id)
75 struct xfs_alloc_rec *arec;
76 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
78 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
79 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
81 if (is_log_ag(mp, id)) {
82 struct xfs_alloc_rec *nrec;
83 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
84 mp->m_sb.sb_logstart);
86 ASSERT(start >= mp->m_ag_prealloc_blocks);
87 if (start != mp->m_ag_prealloc_blocks) {
89 * Modify first record to pad stripe align of log
91 arec->ar_blockcount = cpu_to_be32(start -
92 mp->m_ag_prealloc_blocks);
96 * Insert second record at start of internal log
97 * which then gets trimmed.
99 nrec->ar_startblock = cpu_to_be32(
100 be32_to_cpu(arec->ar_startblock) +
101 be32_to_cpu(arec->ar_blockcount));
103 be16_add_cpu(&block->bb_numrecs, 1);
106 * Change record start to after the internal log
108 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
112 * Calculate the record block count and check for the case where
113 * the log might have consumed all available space in the AG. If
114 * so, reset the record count to 0 to avoid exposure of an invalid
115 * record start block.
117 arec->ar_blockcount = cpu_to_be32(id->agsize -
118 be32_to_cpu(arec->ar_startblock));
119 if (!arec->ar_blockcount)
120 block->bb_numrecs = 0;
124 * Alloc btree root block init functions
128 struct xfs_mount *mp,
130 struct aghdr_init_data *id)
132 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
133 xfs_freesp_init_recs(mp, bp, id);
138 struct xfs_mount *mp,
140 struct aghdr_init_data *id)
142 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
143 xfs_freesp_init_recs(mp, bp, id);
147 * Reverse map root block init
151 struct xfs_mount *mp,
153 struct aghdr_init_data *id)
155 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
156 struct xfs_rmap_rec *rrec;
158 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
161 * mark the AG header regions as static metadata The BNO
162 * btree block is the first block after the headers, so
163 * it's location defines the size of region the static
166 * Note: unlike mkfs, we never have to account for log
167 * space when growing the data regions
169 rrec = XFS_RMAP_REC_ADDR(block, 1);
170 rrec->rm_startblock = 0;
171 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
172 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
175 /* account freespace btree root blocks */
176 rrec = XFS_RMAP_REC_ADDR(block, 2);
177 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
178 rrec->rm_blockcount = cpu_to_be32(2);
179 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
182 /* account inode btree root blocks */
183 rrec = XFS_RMAP_REC_ADDR(block, 3);
184 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
185 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
187 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
190 /* account for rmap btree root */
191 rrec = XFS_RMAP_REC_ADDR(block, 4);
192 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
193 rrec->rm_blockcount = cpu_to_be32(1);
194 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
197 /* account for refc btree root */
198 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
199 rrec = XFS_RMAP_REC_ADDR(block, 5);
200 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
201 rrec->rm_blockcount = cpu_to_be32(1);
202 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
204 be16_add_cpu(&block->bb_numrecs, 1);
207 /* account for the log space */
208 if (is_log_ag(mp, id)) {
209 rrec = XFS_RMAP_REC_ADDR(block,
210 be16_to_cpu(block->bb_numrecs) + 1);
211 rrec->rm_startblock = cpu_to_be32(
212 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
213 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
214 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
216 be16_add_cpu(&block->bb_numrecs, 1);
221 * Initialise new secondary superblocks with the pre-grow geometry, but mark
222 * them as "in progress" so we know they haven't yet been activated. This will
223 * get cleared when the update with the new geometry information is done after
224 * changes to the primary are committed. This isn't strictly necessary, but we
225 * get it for free with the delayed buffer write lists and it means we can tell
226 * if a grow operation didn't complete properly after the fact.
230 struct xfs_mount *mp,
232 struct aghdr_init_data *id)
234 struct xfs_dsb *dsb = bp->b_addr;
236 xfs_sb_to_disk(dsb, &mp->m_sb);
237 dsb->sb_inprogress = 1;
242 struct xfs_mount *mp,
244 struct aghdr_init_data *id)
246 struct xfs_agf *agf = bp->b_addr;
247 xfs_extlen_t tmpsize;
249 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
250 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
251 agf->agf_seqno = cpu_to_be32(id->agno);
252 agf->agf_length = cpu_to_be32(id->agsize);
253 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
254 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
255 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
256 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
257 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
258 agf->agf_roots[XFS_BTNUM_RMAPi] =
259 cpu_to_be32(XFS_RMAP_BLOCK(mp));
260 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
261 agf->agf_rmap_blocks = cpu_to_be32(1);
264 agf->agf_flfirst = cpu_to_be32(1);
266 agf->agf_flcount = 0;
267 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
268 agf->agf_freeblks = cpu_to_be32(tmpsize);
269 agf->agf_longest = cpu_to_be32(tmpsize);
270 if (xfs_sb_version_hascrc(&mp->m_sb))
271 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
272 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
273 agf->agf_refcount_root = cpu_to_be32(
275 agf->agf_refcount_level = cpu_to_be32(1);
276 agf->agf_refcount_blocks = cpu_to_be32(1);
279 if (is_log_ag(mp, id)) {
280 int64_t logblocks = mp->m_sb.sb_logblocks;
282 be32_add_cpu(&agf->agf_freeblks, -logblocks);
283 agf->agf_longest = cpu_to_be32(id->agsize -
284 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
290 struct xfs_mount *mp,
292 struct aghdr_init_data *id)
294 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
298 if (xfs_sb_version_hascrc(&mp->m_sb)) {
299 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
300 agfl->agfl_seqno = cpu_to_be32(id->agno);
301 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
304 agfl_bno = xfs_buf_to_agfl_bno(bp);
305 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
306 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
311 struct xfs_mount *mp,
313 struct aghdr_init_data *id)
315 struct xfs_agi *agi = bp->b_addr;
318 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
319 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
320 agi->agi_seqno = cpu_to_be32(id->agno);
321 agi->agi_length = cpu_to_be32(id->agsize);
323 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
324 agi->agi_level = cpu_to_be32(1);
325 agi->agi_freecount = 0;
326 agi->agi_newino = cpu_to_be32(NULLAGINO);
327 agi->agi_dirino = cpu_to_be32(NULLAGINO);
328 if (xfs_sb_version_hascrc(&mp->m_sb))
329 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
330 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
331 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
332 agi->agi_free_level = cpu_to_be32(1);
334 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
335 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
338 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
339 struct aghdr_init_data *id);
342 struct xfs_mount *mp,
343 struct aghdr_init_data *id,
344 aghdr_init_work_f work,
345 const struct xfs_buf_ops *ops)
350 error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
356 xfs_buf_delwri_queue(bp, &id->buffer_list);
361 struct xfs_aghdr_grow_data {
364 const struct xfs_buf_ops *ops;
365 aghdr_init_work_f work;
371 * Prepare new AG headers to be written to disk. We use uncached buffers here,
372 * as it is assumed these new AG headers are currently beyond the currently
373 * valid filesystem address space. Using cached buffers would trip over EOFS
374 * corruption detection alogrithms in the buffer cache lookup routines.
376 * This is a non-transactional function, but the prepared buffers are added to a
377 * delayed write buffer list supplied by the caller so they can submit them to
378 * disk and wait on them as required.
382 struct xfs_mount *mp,
383 struct aghdr_init_data *id)
386 struct xfs_aghdr_grow_data aghdr_data[] = {
388 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
389 .numblks = XFS_FSS_TO_BB(mp, 1),
390 .ops = &xfs_sb_buf_ops,
391 .work = &xfs_sbblock_init,
395 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
396 .numblks = XFS_FSS_TO_BB(mp, 1),
397 .ops = &xfs_agf_buf_ops,
398 .work = &xfs_agfblock_init,
402 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
403 .numblks = XFS_FSS_TO_BB(mp, 1),
404 .ops = &xfs_agfl_buf_ops,
405 .work = &xfs_agflblock_init,
409 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
410 .numblks = XFS_FSS_TO_BB(mp, 1),
411 .ops = &xfs_agi_buf_ops,
412 .work = &xfs_agiblock_init,
415 { /* BNO root block */
416 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
417 .numblks = BTOBB(mp->m_sb.sb_blocksize),
418 .ops = &xfs_bnobt_buf_ops,
419 .work = &xfs_bnoroot_init,
422 { /* CNT root block */
423 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
424 .numblks = BTOBB(mp->m_sb.sb_blocksize),
425 .ops = &xfs_cntbt_buf_ops,
426 .work = &xfs_cntroot_init,
429 { /* INO root block */
430 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
431 .numblks = BTOBB(mp->m_sb.sb_blocksize),
432 .ops = &xfs_inobt_buf_ops,
433 .work = &xfs_btroot_init,
434 .type = XFS_BTNUM_INO,
437 { /* FINO root block */
438 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
439 .numblks = BTOBB(mp->m_sb.sb_blocksize),
440 .ops = &xfs_finobt_buf_ops,
441 .work = &xfs_btroot_init,
442 .type = XFS_BTNUM_FINO,
443 .need_init = xfs_sb_version_hasfinobt(&mp->m_sb)
445 { /* RMAP root block */
446 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
447 .numblks = BTOBB(mp->m_sb.sb_blocksize),
448 .ops = &xfs_rmapbt_buf_ops,
449 .work = &xfs_rmaproot_init,
450 .need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
452 { /* REFC root block */
453 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
454 .numblks = BTOBB(mp->m_sb.sb_blocksize),
455 .ops = &xfs_refcountbt_buf_ops,
456 .work = &xfs_btroot_init,
457 .type = XFS_BTNUM_REFC,
458 .need_init = xfs_sb_version_hasreflink(&mp->m_sb)
460 { /* NULL terminating block */
461 .daddr = XFS_BUF_DADDR_NULL,
464 struct xfs_aghdr_grow_data *dp;
467 /* Account for AG free space in new AG */
468 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
469 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
473 id->daddr = dp->daddr;
474 id->numblks = dp->numblks;
476 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
484 * Extent the AG indicated by the @id by the length passed in
488 struct xfs_mount *mp,
489 struct xfs_trans *tp,
490 struct aghdr_init_data *id,
499 * Change the agi length.
501 error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
506 be32_add_cpu(&agi->agi_length, len);
507 ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
508 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
509 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
514 error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
519 be32_add_cpu(&agf->agf_length, len);
520 ASSERT(agf->agf_length == agi->agi_length);
521 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
524 * Free the new space.
526 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
527 * this doesn't actually exist in the rmap btree.
529 error = xfs_rmap_free(tp, bp, id->agno,
530 be32_to_cpu(agf->agf_length) - len,
531 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
535 return xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
536 be32_to_cpu(agf->agf_length) - len),
537 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
541 /* Retrieve AG geometry. */
544 struct xfs_mount *mp,
546 struct xfs_ag_geometry *ageo)
548 struct xfs_buf *agi_bp;
549 struct xfs_buf *agf_bp;
552 struct xfs_perag *pag;
553 unsigned int freeblks;
556 if (agno >= mp->m_sb.sb_agcount)
559 /* Lock the AG headers. */
560 error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
563 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
566 pag = xfs_perag_get(mp, agno);
569 memset(ageo, 0, sizeof(*ageo));
570 ageo->ag_number = agno;
572 agi = agi_bp->b_addr;
573 ageo->ag_icount = be32_to_cpu(agi->agi_count);
574 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
576 agf = agf_bp->b_addr;
577 ageo->ag_length = be32_to_cpu(agf->agf_length);
578 freeblks = pag->pagf_freeblks +
580 pag->pagf_btreeblks -
581 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
582 ageo->ag_freeblks = freeblks;
583 xfs_ag_geom_health(pag, ageo);
585 /* Release resources. */
587 xfs_buf_relse(agf_bp);
589 xfs_buf_relse(agi_bp);