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"
26 static struct xfs_buf *
32 const struct xfs_buf_ops *ops)
36 bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
40 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
42 bp->b_maps[0].bm_bn = blkno;
48 static inline bool is_log_ag(struct xfs_mount *mp, struct aghdr_init_data *id)
50 return mp->m_sb.sb_logstart > 0 &&
51 id->agno == XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
55 * Generic btree root block init function
61 struct aghdr_init_data *id)
63 xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
66 /* Finish initializing a free space btree. */
71 struct aghdr_init_data *id)
73 struct xfs_alloc_rec *arec;
74 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
76 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
77 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
79 if (is_log_ag(mp, id)) {
80 struct xfs_alloc_rec *nrec;
81 xfs_agblock_t start = XFS_FSB_TO_AGBNO(mp,
82 mp->m_sb.sb_logstart);
84 ASSERT(start >= mp->m_ag_prealloc_blocks);
85 if (start != mp->m_ag_prealloc_blocks) {
87 * Modify first record to pad stripe align of log
89 arec->ar_blockcount = cpu_to_be32(start -
90 mp->m_ag_prealloc_blocks);
94 * Insert second record at start of internal log
95 * which then gets trimmed.
97 nrec->ar_startblock = cpu_to_be32(
98 be32_to_cpu(arec->ar_startblock) +
99 be32_to_cpu(arec->ar_blockcount));
101 be16_add_cpu(&block->bb_numrecs, 1);
104 * Change record start to after the internal log
106 be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
110 * Calculate the record block count and check for the case where
111 * the log might have consumed all available space in the AG. If
112 * so, reset the record count to 0 to avoid exposure of an invalid
113 * record start block.
115 arec->ar_blockcount = cpu_to_be32(id->agsize -
116 be32_to_cpu(arec->ar_startblock));
117 if (!arec->ar_blockcount)
118 block->bb_numrecs = 0;
122 * Alloc btree root block init functions
126 struct xfs_mount *mp,
128 struct aghdr_init_data *id)
130 xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
131 xfs_freesp_init_recs(mp, bp, id);
136 struct xfs_mount *mp,
138 struct aghdr_init_data *id)
140 xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
141 xfs_freesp_init_recs(mp, bp, id);
145 * Reverse map root block init
149 struct xfs_mount *mp,
151 struct aghdr_init_data *id)
153 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
154 struct xfs_rmap_rec *rrec;
156 xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
159 * mark the AG header regions as static metadata The BNO
160 * btree block is the first block after the headers, so
161 * it's location defines the size of region the static
164 * Note: unlike mkfs, we never have to account for log
165 * space when growing the data regions
167 rrec = XFS_RMAP_REC_ADDR(block, 1);
168 rrec->rm_startblock = 0;
169 rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
170 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
173 /* account freespace btree root blocks */
174 rrec = XFS_RMAP_REC_ADDR(block, 2);
175 rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
176 rrec->rm_blockcount = cpu_to_be32(2);
177 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
180 /* account inode btree root blocks */
181 rrec = XFS_RMAP_REC_ADDR(block, 3);
182 rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
183 rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
185 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
188 /* account for rmap btree root */
189 rrec = XFS_RMAP_REC_ADDR(block, 4);
190 rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
191 rrec->rm_blockcount = cpu_to_be32(1);
192 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
195 /* account for refc btree root */
196 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
197 rrec = XFS_RMAP_REC_ADDR(block, 5);
198 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
199 rrec->rm_blockcount = cpu_to_be32(1);
200 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
202 be16_add_cpu(&block->bb_numrecs, 1);
205 /* account for the log space */
206 if (is_log_ag(mp, id)) {
207 rrec = XFS_RMAP_REC_ADDR(block,
208 be16_to_cpu(block->bb_numrecs) + 1);
209 rrec->rm_startblock = cpu_to_be32(
210 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
211 rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
212 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
214 be16_add_cpu(&block->bb_numrecs, 1);
219 * Initialise new secondary superblocks with the pre-grow geometry, but mark
220 * them as "in progress" so we know they haven't yet been activated. This will
221 * get cleared when the update with the new geometry information is done after
222 * changes to the primary are committed. This isn't strictly necessary, but we
223 * get it for free with the delayed buffer write lists and it means we can tell
224 * if a grow operation didn't complete properly after the fact.
228 struct xfs_mount *mp,
230 struct aghdr_init_data *id)
232 struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp);
234 xfs_sb_to_disk(dsb, &mp->m_sb);
235 dsb->sb_inprogress = 1;
240 struct xfs_mount *mp,
242 struct aghdr_init_data *id)
244 struct xfs_agf *agf = XFS_BUF_TO_AGF(bp);
245 xfs_extlen_t tmpsize;
247 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
248 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
249 agf->agf_seqno = cpu_to_be32(id->agno);
250 agf->agf_length = cpu_to_be32(id->agsize);
251 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
252 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
253 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
254 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
255 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
256 agf->agf_roots[XFS_BTNUM_RMAPi] =
257 cpu_to_be32(XFS_RMAP_BLOCK(mp));
258 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
259 agf->agf_rmap_blocks = cpu_to_be32(1);
262 agf->agf_flfirst = cpu_to_be32(1);
264 agf->agf_flcount = 0;
265 tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
266 agf->agf_freeblks = cpu_to_be32(tmpsize);
267 agf->agf_longest = cpu_to_be32(tmpsize);
268 if (xfs_sb_version_hascrc(&mp->m_sb))
269 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
270 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
271 agf->agf_refcount_root = cpu_to_be32(
273 agf->agf_refcount_level = cpu_to_be32(1);
274 agf->agf_refcount_blocks = cpu_to_be32(1);
277 if (is_log_ag(mp, id)) {
278 int64_t logblocks = mp->m_sb.sb_logblocks;
280 be32_add_cpu(&agf->agf_freeblks, -logblocks);
281 agf->agf_longest = cpu_to_be32(id->agsize -
282 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
288 struct xfs_mount *mp,
290 struct aghdr_init_data *id)
292 struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp);
296 if (xfs_sb_version_hascrc(&mp->m_sb)) {
297 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
298 agfl->agfl_seqno = cpu_to_be32(id->agno);
299 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
302 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
303 for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
304 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
309 struct xfs_mount *mp,
311 struct aghdr_init_data *id)
313 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
316 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
317 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
318 agi->agi_seqno = cpu_to_be32(id->agno);
319 agi->agi_length = cpu_to_be32(id->agsize);
321 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
322 agi->agi_level = cpu_to_be32(1);
323 agi->agi_freecount = 0;
324 agi->agi_newino = cpu_to_be32(NULLAGINO);
325 agi->agi_dirino = cpu_to_be32(NULLAGINO);
326 if (xfs_sb_version_hascrc(&mp->m_sb))
327 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
328 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
329 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
330 agi->agi_free_level = cpu_to_be32(1);
332 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
333 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
336 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
337 struct aghdr_init_data *id);
340 struct xfs_mount *mp,
341 struct aghdr_init_data *id,
342 aghdr_init_work_f work,
343 const struct xfs_buf_ops *ops)
348 bp = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, 0, ops);
354 xfs_buf_delwri_queue(bp, &id->buffer_list);
359 struct xfs_aghdr_grow_data {
362 const struct xfs_buf_ops *ops;
363 aghdr_init_work_f work;
369 * Prepare new AG headers to be written to disk. We use uncached buffers here,
370 * as it is assumed these new AG headers are currently beyond the currently
371 * valid filesystem address space. Using cached buffers would trip over EOFS
372 * corruption detection alogrithms in the buffer cache lookup routines.
374 * This is a non-transactional function, but the prepared buffers are added to a
375 * delayed write buffer list supplied by the caller so they can submit them to
376 * disk and wait on them as required.
380 struct xfs_mount *mp,
381 struct aghdr_init_data *id)
384 struct xfs_aghdr_grow_data aghdr_data[] = {
386 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
387 .numblks = XFS_FSS_TO_BB(mp, 1),
388 .ops = &xfs_sb_buf_ops,
389 .work = &xfs_sbblock_init,
393 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
394 .numblks = XFS_FSS_TO_BB(mp, 1),
395 .ops = &xfs_agf_buf_ops,
396 .work = &xfs_agfblock_init,
400 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
401 .numblks = XFS_FSS_TO_BB(mp, 1),
402 .ops = &xfs_agfl_buf_ops,
403 .work = &xfs_agflblock_init,
407 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
408 .numblks = XFS_FSS_TO_BB(mp, 1),
409 .ops = &xfs_agi_buf_ops,
410 .work = &xfs_agiblock_init,
413 { /* BNO root block */
414 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
415 .numblks = BTOBB(mp->m_sb.sb_blocksize),
416 .ops = &xfs_bnobt_buf_ops,
417 .work = &xfs_bnoroot_init,
420 { /* CNT root block */
421 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
422 .numblks = BTOBB(mp->m_sb.sb_blocksize),
423 .ops = &xfs_cntbt_buf_ops,
424 .work = &xfs_cntroot_init,
427 { /* INO root block */
428 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
429 .numblks = BTOBB(mp->m_sb.sb_blocksize),
430 .ops = &xfs_inobt_buf_ops,
431 .work = &xfs_btroot_init,
432 .type = XFS_BTNUM_INO,
435 { /* FINO root block */
436 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
437 .numblks = BTOBB(mp->m_sb.sb_blocksize),
438 .ops = &xfs_finobt_buf_ops,
439 .work = &xfs_btroot_init,
440 .type = XFS_BTNUM_FINO,
441 .need_init = xfs_sb_version_hasfinobt(&mp->m_sb)
443 { /* RMAP root block */
444 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
445 .numblks = BTOBB(mp->m_sb.sb_blocksize),
446 .ops = &xfs_rmapbt_buf_ops,
447 .work = &xfs_rmaproot_init,
448 .need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
450 { /* REFC root block */
451 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
452 .numblks = BTOBB(mp->m_sb.sb_blocksize),
453 .ops = &xfs_refcountbt_buf_ops,
454 .work = &xfs_btroot_init,
455 .type = XFS_BTNUM_REFC,
456 .need_init = xfs_sb_version_hasreflink(&mp->m_sb)
458 { /* NULL terminating block */
459 .daddr = XFS_BUF_DADDR_NULL,
462 struct xfs_aghdr_grow_data *dp;
465 /* Account for AG free space in new AG */
466 id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
467 for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
471 id->daddr = dp->daddr;
472 id->numblks = dp->numblks;
474 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
482 * Extent the AG indicated by the @id by the length passed in
486 struct xfs_mount *mp,
487 struct xfs_trans *tp,
488 struct aghdr_init_data *id,
497 * Change the agi length.
499 error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
503 agi = XFS_BUF_TO_AGI(bp);
504 be32_add_cpu(&agi->agi_length, len);
505 ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
506 be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
507 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
512 error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
516 agf = XFS_BUF_TO_AGF(bp);
517 be32_add_cpu(&agf->agf_length, len);
518 ASSERT(agf->agf_length == agi->agi_length);
519 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
522 * Free the new space.
524 * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
525 * this doesn't actually exist in the rmap btree.
527 error = xfs_rmap_free(tp, bp, id->agno,
528 be32_to_cpu(agf->agf_length) - len,
529 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
533 return xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
534 be32_to_cpu(agf->agf_length) - len),
535 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
539 /* Retrieve AG geometry. */
542 struct xfs_mount *mp,
544 struct xfs_ag_geometry *ageo)
546 struct xfs_buf *agi_bp;
547 struct xfs_buf *agf_bp;
550 struct xfs_perag *pag;
551 unsigned int freeblks;
554 if (agno >= mp->m_sb.sb_agcount)
557 /* Lock the AG headers. */
558 error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
561 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
564 pag = xfs_perag_get(mp, agno);
567 memset(ageo, 0, sizeof(*ageo));
568 ageo->ag_number = agno;
570 agi = XFS_BUF_TO_AGI(agi_bp);
571 ageo->ag_icount = be32_to_cpu(agi->agi_count);
572 ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
574 agf = XFS_BUF_TO_AGF(agf_bp);
575 ageo->ag_length = be32_to_cpu(agf->agf_length);
576 freeblks = pag->pagf_freeblks +
578 pag->pagf_btreeblks -
579 xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
580 ageo->ag_freeblks = freeblks;
581 xfs_ag_geom_health(pag, ageo);
583 /* Release resources. */
585 xfs_buf_relse(agf_bp);
587 xfs_buf_relse(agi_bp);