xfs_refcount_btree.c

  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * Copyright (C) 2016 Oracle.  All Rights Reserved.
  4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
  5 */
  6#include "xfs.h"
  7#include "xfs_fs.h"
  8#include "xfs_shared.h"
  9#include "xfs_format.h"
 10#include "xfs_log_format.h"
 11#include "xfs_trans_resv.h"
 12#include "xfs_mount.h"
 13#include "xfs_btree.h"
 14#include "xfs_btree_staging.h"
 15#include "xfs_refcount_btree.h"
 16#include "xfs_refcount.h"
 17#include "xfs_alloc.h"
 18#include "xfs_error.h"
 19#include "xfs_health.h"
 20#include "xfs_trace.h"
 21#include "xfs_trans.h"
 22#include "xfs_bit.h"
 23#include "xfs_rmap.h"
 24#include "xfs_ag.h"
 25
 26static struct kmem_cache	*xfs_refcountbt_cur_cache;
 27
 28static struct xfs_btree_cur *
 29xfs_refcountbt_dup_cursor(
 30	struct xfs_btree_cur	*cur)
 31{
 32	return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
 33			cur->bc_ag.agbp, to_perag(cur->bc_group));
 34}
 35
 36STATIC void
 37xfs_refcountbt_set_root(
 38	struct xfs_btree_cur		*cur,
 39	const union xfs_btree_ptr	*ptr,
 40	int				inc)
 41{
 42	struct xfs_buf		*agbp = cur->bc_ag.agbp;
 43	struct xfs_agf		*agf = agbp->b_addr;
 44	struct xfs_perag	*pag = agbp->b_pag;
 45
 46	ASSERT(ptr->s != 0);
 47
 48	agf->agf_refcount_root = ptr->s;
 49	be32_add_cpu(&agf->agf_refcount_level, inc);
 50	pag->pagf_refcount_level += inc;
 51
 52	xfs_alloc_log_agf(cur->bc_tp, agbp,
 53			XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
 54}
 55
 56STATIC int
 57xfs_refcountbt_alloc_block(
 58	struct xfs_btree_cur		*cur,
 59	const union xfs_btree_ptr	*start,
 60	union xfs_btree_ptr		*new,
 61	int				*stat)
 62{
 63	struct xfs_buf		*agbp = cur->bc_ag.agbp;
 64	struct xfs_agf		*agf = agbp->b_addr;
 65	struct xfs_alloc_arg	args;		/* block allocation args */
 66	int			error;		/* error return value */
 67
 68	memset(&args, 0, sizeof(args));
 69	args.tp = cur->bc_tp;
 70	args.mp = cur->bc_mp;
 71	args.pag = to_perag(cur->bc_group);
 72	args.oinfo = XFS_RMAP_OINFO_REFC;
 73	args.minlen = args.maxlen = args.prod = 1;
 74	args.resv = XFS_AG_RESV_METADATA;
 75
 76	error = xfs_alloc_vextent_near_bno(&args,
 77			xfs_agbno_to_fsb(args.pag, xfs_refc_block(args.mp)));
 78	if (error)
 79		goto out_error;
 80	if (args.fsbno == NULLFSBLOCK) {
 81		*stat = 0;
 82		return 0;
 83	}
 84	ASSERT(args.agno == cur->bc_group->xg_gno);
 85	ASSERT(args.len == 1);
 86
 87	new->s = cpu_to_be32(args.agbno);
 88	be32_add_cpu(&agf->agf_refcount_blocks, 1);
 89	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
 90
 91	*stat = 1;
 92	return 0;
 93
 94out_error:
 95	return error;
 96}
 97
 98STATIC int
 99xfs_refcountbt_free_block(
100	struct xfs_btree_cur	*cur,
101	struct xfs_buf		*bp)
102{
103	struct xfs_mount	*mp = cur->bc_mp;
104	struct xfs_buf		*agbp = cur->bc_ag.agbp;
105	struct xfs_agf		*agf = agbp->b_addr;
106	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
107
108	be32_add_cpu(&agf->agf_refcount_blocks, -1);
109	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
110	return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
111			&XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, 0);
112}
113
114STATIC int
115xfs_refcountbt_get_minrecs(
116	struct xfs_btree_cur	*cur,
117	int			level)
118{
119	return cur->bc_mp->m_refc_mnr[level != 0];
120}
121
122STATIC int
123xfs_refcountbt_get_maxrecs(
124	struct xfs_btree_cur	*cur,
125	int			level)
126{
127	return cur->bc_mp->m_refc_mxr[level != 0];
128}
129
130STATIC void
131xfs_refcountbt_init_key_from_rec(
132	union xfs_btree_key		*key,
133	const union xfs_btree_rec	*rec)
134{
135	key->refc.rc_startblock = rec->refc.rc_startblock;
136}
137
138STATIC void
139xfs_refcountbt_init_high_key_from_rec(
140	union xfs_btree_key		*key,
141	const union xfs_btree_rec	*rec)
142{
143	__u32				x;
144
145	x = be32_to_cpu(rec->refc.rc_startblock);
146	x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
147	key->refc.rc_startblock = cpu_to_be32(x);
148}
149
150STATIC void
151xfs_refcountbt_init_rec_from_cur(
152	struct xfs_btree_cur	*cur,
153	union xfs_btree_rec	*rec)
154{
155	const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
156	uint32_t		start;
157
158	start = xfs_refcount_encode_startblock(irec->rc_startblock,
159			irec->rc_domain);
160	rec->refc.rc_startblock = cpu_to_be32(start);
161	rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
162	rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
163}
164
165STATIC void
166xfs_refcountbt_init_ptr_from_cur(
167	struct xfs_btree_cur	*cur,
168	union xfs_btree_ptr	*ptr)
169{
170	struct xfs_agf		*agf = cur->bc_ag.agbp->b_addr;
171
172	ASSERT(cur->bc_group->xg_gno == be32_to_cpu(agf->agf_seqno));
173
174	ptr->s = agf->agf_refcount_root;
175}
176
177STATIC int
178xfs_refcountbt_cmp_key_with_cur(
179	struct xfs_btree_cur		*cur,
180	const union xfs_btree_key	*key)
181{
182	const struct xfs_refcount_key	*kp = &key->refc;
183	const struct xfs_refcount_irec	*irec = &cur->bc_rec.rc;
184	uint32_t			start;
185
186	start = xfs_refcount_encode_startblock(irec->rc_startblock,
187			irec->rc_domain);
188	return cmp_int(be32_to_cpu(kp->rc_startblock), start);
189}
190
191STATIC int
192xfs_refcountbt_cmp_two_keys(
193	struct xfs_btree_cur		*cur,
194	const union xfs_btree_key	*k1,
195	const union xfs_btree_key	*k2,
196	const union xfs_btree_key	*mask)
197{
198	ASSERT(!mask || mask->refc.rc_startblock);
199
200	return cmp_int(be32_to_cpu(k1->refc.rc_startblock),
201		       be32_to_cpu(k2->refc.rc_startblock));
202}
203
204STATIC xfs_failaddr_t
205xfs_refcountbt_verify(
206	struct xfs_buf		*bp)
207{
208	struct xfs_mount	*mp = bp->b_mount;
209	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
210	struct xfs_perag	*pag = bp->b_pag;
211	xfs_failaddr_t		fa;
212	unsigned int		level;
213
214	if (!xfs_verify_magic(bp, block->bb_magic))
215		return __this_address;
216
217	if (!xfs_has_reflink(mp))
218		return __this_address;
219	fa = xfs_btree_agblock_v5hdr_verify(bp);
220	if (fa)
221		return fa;
222
223	level = be16_to_cpu(block->bb_level);
224	if (pag && xfs_perag_initialised_agf(pag)) {
225		unsigned int	maxlevel = pag->pagf_refcount_level;
226
227#ifdef CONFIG_XFS_ONLINE_REPAIR
228		/*
229		 * Online repair could be rewriting the refcount btree, so
230		 * we'll validate against the larger of either tree while this
231		 * is going on.
232		 */
233		maxlevel = max_t(unsigned int, maxlevel,
234				pag->pagf_repair_refcount_level);
235#endif
236		if (level >= maxlevel)
237			return __this_address;
238	} else if (level >= mp->m_refc_maxlevels)
239		return __this_address;
240
241	return xfs_btree_agblock_verify(bp, mp->m_refc_mxr[level != 0]);
242}
243
244STATIC void
245xfs_refcountbt_read_verify(
246	struct xfs_buf	*bp)
247{
248	xfs_failaddr_t	fa;
249
250	if (!xfs_btree_agblock_verify_crc(bp))
251		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
252	else {
253		fa = xfs_refcountbt_verify(bp);
254		if (fa)
255			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
256	}
257
258	if (bp->b_error)
259		trace_xfs_btree_corrupt(bp, _RET_IP_);
260}
261
262STATIC void
263xfs_refcountbt_write_verify(
264	struct xfs_buf	*bp)
265{
266	xfs_failaddr_t	fa;
267
268	fa = xfs_refcountbt_verify(bp);
269	if (fa) {
270		trace_xfs_btree_corrupt(bp, _RET_IP_);
271		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
272		return;
273	}
274	xfs_btree_agblock_calc_crc(bp);
275
276}
277
278const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
279	.name			= "xfs_refcountbt",
280	.magic			= { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
281	.verify_read		= xfs_refcountbt_read_verify,
282	.verify_write		= xfs_refcountbt_write_verify,
283	.verify_struct		= xfs_refcountbt_verify,
284};
285
286STATIC int
287xfs_refcountbt_keys_inorder(
288	struct xfs_btree_cur		*cur,
289	const union xfs_btree_key	*k1,
290	const union xfs_btree_key	*k2)
291{
292	return be32_to_cpu(k1->refc.rc_startblock) <
293	       be32_to_cpu(k2->refc.rc_startblock);
294}
295
296STATIC int
297xfs_refcountbt_recs_inorder(
298	struct xfs_btree_cur		*cur,
299	const union xfs_btree_rec	*r1,
300	const union xfs_btree_rec	*r2)
301{
302	return  be32_to_cpu(r1->refc.rc_startblock) +
303		be32_to_cpu(r1->refc.rc_blockcount) <=
304		be32_to_cpu(r2->refc.rc_startblock);
305}
306
307STATIC enum xbtree_key_contig
308xfs_refcountbt_keys_contiguous(
309	struct xfs_btree_cur		*cur,
310	const union xfs_btree_key	*key1,
311	const union xfs_btree_key	*key2,
312	const union xfs_btree_key	*mask)
313{
314	ASSERT(!mask || mask->refc.rc_startblock);
315
316	return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
317				 be32_to_cpu(key2->refc.rc_startblock));
318}
319
320const struct xfs_btree_ops xfs_refcountbt_ops = {
321	.name			= "refcount",
322	.type			= XFS_BTREE_TYPE_AG,
323
324	.rec_len		= sizeof(struct xfs_refcount_rec),
325	.key_len		= sizeof(struct xfs_refcount_key),
326	.ptr_len		= XFS_BTREE_SHORT_PTR_LEN,
327
328	.lru_refs		= XFS_REFC_BTREE_REF,
329	.statoff		= XFS_STATS_CALC_INDEX(xs_refcbt_2),
330	.sick_mask		= XFS_SICK_AG_REFCNTBT,
331
332	.dup_cursor		= xfs_refcountbt_dup_cursor,
333	.set_root		= xfs_refcountbt_set_root,
334	.alloc_block		= xfs_refcountbt_alloc_block,
335	.free_block		= xfs_refcountbt_free_block,
336	.get_minrecs		= xfs_refcountbt_get_minrecs,
337	.get_maxrecs		= xfs_refcountbt_get_maxrecs,
338	.init_key_from_rec	= xfs_refcountbt_init_key_from_rec,
339	.init_high_key_from_rec	= xfs_refcountbt_init_high_key_from_rec,
340	.init_rec_from_cur	= xfs_refcountbt_init_rec_from_cur,
341	.init_ptr_from_cur	= xfs_refcountbt_init_ptr_from_cur,
342	.cmp_key_with_cur	= xfs_refcountbt_cmp_key_with_cur,
343	.buf_ops		= &xfs_refcountbt_buf_ops,
344	.cmp_two_keys		= xfs_refcountbt_cmp_two_keys,
345	.keys_inorder		= xfs_refcountbt_keys_inorder,
346	.recs_inorder		= xfs_refcountbt_recs_inorder,
347	.keys_contiguous	= xfs_refcountbt_keys_contiguous,
348};
349
350/*
351 * Create a new refcount btree cursor.
352 *
353 * For staging cursors tp and agbp are NULL.
354 */
355struct xfs_btree_cur *
356xfs_refcountbt_init_cursor(
357	struct xfs_mount	*mp,
358	struct xfs_trans	*tp,
359	struct xfs_buf		*agbp,
360	struct xfs_perag	*pag)
361{
362	struct xfs_btree_cur	*cur;
363
364	ASSERT(pag_agno(pag) < mp->m_sb.sb_agcount);
365
366	cur = xfs_btree_alloc_cursor(mp, tp, &xfs_refcountbt_ops,
367			mp->m_refc_maxlevels, xfs_refcountbt_cur_cache);
368	cur->bc_group = xfs_group_hold(pag_group(pag));
369	cur->bc_refc.nr_ops = 0;
370	cur->bc_refc.shape_changes = 0;
371	cur->bc_ag.agbp = agbp;
372	if (agbp) {
373		struct xfs_agf		*agf = agbp->b_addr;
374
375		cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
376	}
377	return cur;
378}
379
380/*
381 * Swap in the new btree root.  Once we pass this point the newly rebuilt btree
382 * is in place and we have to kill off all the old btree blocks.
383 */
384void
385xfs_refcountbt_commit_staged_btree(
386	struct xfs_btree_cur	*cur,
387	struct xfs_trans	*tp,
388	struct xfs_buf		*agbp)
389{
390	struct xfs_agf		*agf = agbp->b_addr;
391	struct xbtree_afakeroot	*afake = cur->bc_ag.afake;
392
393	ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
394
395	agf->agf_refcount_root = cpu_to_be32(afake->af_root);
396	agf->agf_refcount_level = cpu_to_be32(afake->af_levels);
397	agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks);
398	xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS |
399				    XFS_AGF_REFCOUNT_ROOT |
400				    XFS_AGF_REFCOUNT_LEVEL);
401	xfs_btree_commit_afakeroot(cur, tp, agbp);
402}
403
404/* Calculate number of records in a refcount btree block. */
405static inline unsigned int
406xfs_refcountbt_block_maxrecs(
407	unsigned int		blocklen,
408	bool			leaf)
409{
410	if (leaf)
411		return blocklen / sizeof(struct xfs_refcount_rec);
412	return blocklen / (sizeof(struct xfs_refcount_key) +
413			   sizeof(xfs_refcount_ptr_t));
414}
415
416/*
417 * Calculate the number of records in a refcount btree block.
418 */
419unsigned int
420xfs_refcountbt_maxrecs(
421	struct xfs_mount	*mp,
422	unsigned int		blocklen,
423	bool			leaf)
424{
425	blocklen -= XFS_REFCOUNT_BLOCK_LEN;
426	return xfs_refcountbt_block_maxrecs(blocklen, leaf);
427}
428
429/* Compute the max possible height of the maximally sized refcount btree. */
430unsigned int
431xfs_refcountbt_maxlevels_ondisk(void)
432{
433	unsigned int		minrecs[2];
434	unsigned int		blocklen;
435
436	blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
437
438	minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2;
439	minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2;
440
441	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS);
442}
443
444/* Compute the maximum height of a refcount btree. */
445void
446xfs_refcountbt_compute_maxlevels(
447	struct xfs_mount		*mp)
448{
449	if (!xfs_has_reflink(mp)) {
450		mp->m_refc_maxlevels = 0;
451		return;
452	}
453
454	mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
455			mp->m_refc_mnr, mp->m_sb.sb_agblocks);
456	ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk());
457}
458
459/* Calculate the refcount btree size for some records. */
460xfs_extlen_t
461xfs_refcountbt_calc_size(
462	struct xfs_mount	*mp,
463	unsigned long long	len)
464{
465	return xfs_btree_calc_size(mp->m_refc_mnr, len);
466}
467
468/*
469 * Calculate the maximum refcount btree size.
470 */
471xfs_extlen_t
472xfs_refcountbt_max_size(
473	struct xfs_mount	*mp,
474	xfs_agblock_t		agblocks)
475{
476	/* Bail out if we're uninitialized, which can happen in mkfs. */
477	if (mp->m_refc_mxr[0] == 0)
478		return 0;
479
480	return xfs_refcountbt_calc_size(mp, agblocks);
481}
482
483/*
484 * Figure out how many blocks to reserve and how many are used by this btree.
485 */
486int
487xfs_refcountbt_calc_reserves(
488	struct xfs_mount	*mp,
489	struct xfs_trans	*tp,
490	struct xfs_perag	*pag,
491	xfs_extlen_t		*ask,
492	xfs_extlen_t		*used)
493{
494	struct xfs_buf		*agbp;
495	struct xfs_agf		*agf;
496	xfs_agblock_t		agblocks;
497	xfs_extlen_t		tree_len;
498	int			error;
499
500	if (!xfs_has_reflink(mp))
501		return 0;
502
503	error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
504	if (error)
505		return error;
506
507	agf = agbp->b_addr;
508	agblocks = be32_to_cpu(agf->agf_length);
509	tree_len = be32_to_cpu(agf->agf_refcount_blocks);
510	xfs_trans_brelse(tp, agbp);
511
512	/*
513	 * The log is permanently allocated, so the space it occupies will
514	 * never be available for the kinds of things that would require btree
515	 * expansion.  We therefore can pretend the space isn't there.
516	 */
517	if (xfs_ag_contains_log(mp, pag_agno(pag)))
518		agblocks -= mp->m_sb.sb_logblocks;
519
520	*ask += xfs_refcountbt_max_size(mp, agblocks);
521	*used += tree_len;
522
523	return error;
524}
525
526int __init
527xfs_refcountbt_init_cur_cache(void)
528{
529	xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
530			xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
531			0, 0, NULL);
532
533	if (!xfs_refcountbt_cur_cache)
534		return -ENOMEM;
535	return 0;
536}
537
538void
539xfs_refcountbt_destroy_cur_cache(void)
540{
541	kmem_cache_destroy(xfs_refcountbt_cur_cache);
542	xfs_refcountbt_cur_cache = NULL;
543}