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_sb.h"
13 #include "xfs_mount.h"
14 #include "xfs_btree.h"
15 #include "xfs_bmap.h"
16 #include "xfs_refcount_btree.h"
17 #include "xfs_alloc.h"
18 #include "xfs_error.h"
19 #include "xfs_trace.h"
20 #include "xfs_cksum.h"
21 #include "xfs_trans.h"
22 #include "xfs_bit.h"
23 #include "xfs_rmap.h"
24 
25 static struct xfs_btree_cur *
26 xfs_refcountbt_dup_cursor(
27 	struct xfs_btree_cur	*cur)
28 {
29 	return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
30 			cur->bc_private.a.agbp, cur->bc_private.a.agno);
31 }
32 
33 STATIC void
34 xfs_refcountbt_set_root(
35 	struct xfs_btree_cur	*cur,
36 	union xfs_btree_ptr	*ptr,
37 	int			inc)
38 {
39 	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
40 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
41 	xfs_agnumber_t		seqno = be32_to_cpu(agf->agf_seqno);
42 	struct xfs_perag	*pag = xfs_perag_get(cur->bc_mp, seqno);
43 
44 	ASSERT(ptr->s != 0);
45 
46 	agf->agf_refcount_root = ptr->s;
47 	be32_add_cpu(&agf->agf_refcount_level, inc);
48 	pag->pagf_refcount_level += inc;
49 	xfs_perag_put(pag);
50 
51 	xfs_alloc_log_agf(cur->bc_tp, agbp,
52 			XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
53 }
54 
55 STATIC int
56 xfs_refcountbt_alloc_block(
57 	struct xfs_btree_cur	*cur,
58 	union xfs_btree_ptr	*start,
59 	union xfs_btree_ptr	*new,
60 	int			*stat)
61 {
62 	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
63 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
64 	struct xfs_alloc_arg	args;		/* block allocation args */
65 	int			error;		/* error return value */
66 
67 	memset(&args, 0, sizeof(args));
68 	args.tp = cur->bc_tp;
69 	args.mp = cur->bc_mp;
70 	args.type = XFS_ALLOCTYPE_NEAR_BNO;
71 	args.fsbno = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_private.a.agno,
72 			xfs_refc_block(args.mp));
73 	args.oinfo = XFS_RMAP_OINFO_REFC;
74 	args.minlen = args.maxlen = args.prod = 1;
75 	args.resv = XFS_AG_RESV_METADATA;
76 
77 	error = xfs_alloc_vextent(&args);
78 	if (error)
79 		goto out_error;
80 	trace_xfs_refcountbt_alloc_block(cur->bc_mp, cur->bc_private.a.agno,
81 			args.agbno, 1);
82 	if (args.fsbno == NULLFSBLOCK) {
83 		*stat = 0;
84 		return 0;
85 	}
86 	ASSERT(args.agno == cur->bc_private.a.agno);
87 	ASSERT(args.len == 1);
88 
89 	new->s = cpu_to_be32(args.agbno);
90 	be32_add_cpu(&agf->agf_refcount_blocks, 1);
91 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
92 
93 	*stat = 1;
94 	return 0;
95 
96 out_error:
97 	return error;
98 }
99 
100 STATIC int
101 xfs_refcountbt_free_block(
102 	struct xfs_btree_cur	*cur,
103 	struct xfs_buf		*bp)
104 {
105 	struct xfs_mount	*mp = cur->bc_mp;
106 	struct xfs_buf		*agbp = cur->bc_private.a.agbp;
107 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
108 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
109 	int			error;
110 
111 	trace_xfs_refcountbt_free_block(cur->bc_mp, cur->bc_private.a.agno,
112 			XFS_FSB_TO_AGBNO(cur->bc_mp, fsbno), 1);
113 	be32_add_cpu(&agf->agf_refcount_blocks, -1);
114 	xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
115 	error = xfs_free_extent(cur->bc_tp, fsbno, 1, &XFS_RMAP_OINFO_REFC,
116 			XFS_AG_RESV_METADATA);
117 	if (error)
118 		return error;
119 
120 	return error;
121 }
122 
123 STATIC int
124 xfs_refcountbt_get_minrecs(
125 	struct xfs_btree_cur	*cur,
126 	int			level)
127 {
128 	return cur->bc_mp->m_refc_mnr[level != 0];
129 }
130 
131 STATIC int
132 xfs_refcountbt_get_maxrecs(
133 	struct xfs_btree_cur	*cur,
134 	int			level)
135 {
136 	return cur->bc_mp->m_refc_mxr[level != 0];
137 }
138 
139 STATIC void
140 xfs_refcountbt_init_key_from_rec(
141 	union xfs_btree_key	*key,
142 	union xfs_btree_rec	*rec)
143 {
144 	key->refc.rc_startblock = rec->refc.rc_startblock;
145 }
146 
147 STATIC void
148 xfs_refcountbt_init_high_key_from_rec(
149 	union xfs_btree_key	*key,
150 	union xfs_btree_rec	*rec)
151 {
152 	__u32			x;
153 
154 	x = be32_to_cpu(rec->refc.rc_startblock);
155 	x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
156 	key->refc.rc_startblock = cpu_to_be32(x);
157 }
158 
159 STATIC void
160 xfs_refcountbt_init_rec_from_cur(
161 	struct xfs_btree_cur	*cur,
162 	union xfs_btree_rec	*rec)
163 {
164 	rec->refc.rc_startblock = cpu_to_be32(cur->bc_rec.rc.rc_startblock);
165 	rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
166 	rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
167 }
168 
169 STATIC void
170 xfs_refcountbt_init_ptr_from_cur(
171 	struct xfs_btree_cur	*cur,
172 	union xfs_btree_ptr	*ptr)
173 {
174 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
175 
176 	ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
177 
178 	ptr->s = agf->agf_refcount_root;
179 }
180 
181 STATIC int64_t
182 xfs_refcountbt_key_diff(
183 	struct xfs_btree_cur	*cur,
184 	union xfs_btree_key	*key)
185 {
186 	struct xfs_refcount_irec	*rec = &cur->bc_rec.rc;
187 	struct xfs_refcount_key		*kp = &key->refc;
188 
189 	return (int64_t)be32_to_cpu(kp->rc_startblock) - rec->rc_startblock;
190 }
191 
192 STATIC int64_t
193 xfs_refcountbt_diff_two_keys(
194 	struct xfs_btree_cur	*cur,
195 	union xfs_btree_key	*k1,
196 	union xfs_btree_key	*k2)
197 {
198 	return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
199 			  be32_to_cpu(k2->refc.rc_startblock);
200 }
201 
202 STATIC xfs_failaddr_t
203 xfs_refcountbt_verify(
204 	struct xfs_buf		*bp)
205 {
206 	struct xfs_mount	*mp = bp->b_target->bt_mount;
207 	struct xfs_btree_block	*block = XFS_BUF_TO_BLOCK(bp);
208 	struct xfs_perag	*pag = bp->b_pag;
209 	xfs_failaddr_t		fa;
210 	unsigned int		level;
211 
212 	if (!xfs_verify_magic(bp, block->bb_magic))
213 		return __this_address;
214 
215 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
216 		return __this_address;
217 	fa = xfs_btree_sblock_v5hdr_verify(bp);
218 	if (fa)
219 		return fa;
220 
221 	level = be16_to_cpu(block->bb_level);
222 	if (pag && pag->pagf_init) {
223 		if (level >= pag->pagf_refcount_level)
224 			return __this_address;
225 	} else if (level >= mp->m_refc_maxlevels)
226 		return __this_address;
227 
228 	return xfs_btree_sblock_verify(bp, mp->m_refc_mxr[level != 0]);
229 }
230 
231 STATIC void
232 xfs_refcountbt_read_verify(
233 	struct xfs_buf	*bp)
234 {
235 	xfs_failaddr_t	fa;
236 
237 	if (!xfs_btree_sblock_verify_crc(bp))
238 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
239 	else {
240 		fa = xfs_refcountbt_verify(bp);
241 		if (fa)
242 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
243 	}
244 
245 	if (bp->b_error)
246 		trace_xfs_btree_corrupt(bp, _RET_IP_);
247 }
248 
249 STATIC void
250 xfs_refcountbt_write_verify(
251 	struct xfs_buf	*bp)
252 {
253 	xfs_failaddr_t	fa;
254 
255 	fa = xfs_refcountbt_verify(bp);
256 	if (fa) {
257 		trace_xfs_btree_corrupt(bp, _RET_IP_);
258 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
259 		return;
260 	}
261 	xfs_btree_sblock_calc_crc(bp);
262 
263 }
264 
265 const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
266 	.name			= "xfs_refcountbt",
267 	.magic			= { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
268 	.verify_read		= xfs_refcountbt_read_verify,
269 	.verify_write		= xfs_refcountbt_write_verify,
270 	.verify_struct		= xfs_refcountbt_verify,
271 };
272 
273 STATIC int
274 xfs_refcountbt_keys_inorder(
275 	struct xfs_btree_cur	*cur,
276 	union xfs_btree_key	*k1,
277 	union xfs_btree_key	*k2)
278 {
279 	return be32_to_cpu(k1->refc.rc_startblock) <
280 	       be32_to_cpu(k2->refc.rc_startblock);
281 }
282 
283 STATIC int
284 xfs_refcountbt_recs_inorder(
285 	struct xfs_btree_cur	*cur,
286 	union xfs_btree_rec	*r1,
287 	union xfs_btree_rec	*r2)
288 {
289 	return  be32_to_cpu(r1->refc.rc_startblock) +
290 		be32_to_cpu(r1->refc.rc_blockcount) <=
291 		be32_to_cpu(r2->refc.rc_startblock);
292 }
293 
294 static const struct xfs_btree_ops xfs_refcountbt_ops = {
295 	.rec_len		= sizeof(struct xfs_refcount_rec),
296 	.key_len		= sizeof(struct xfs_refcount_key),
297 
298 	.dup_cursor		= xfs_refcountbt_dup_cursor,
299 	.set_root		= xfs_refcountbt_set_root,
300 	.alloc_block		= xfs_refcountbt_alloc_block,
301 	.free_block		= xfs_refcountbt_free_block,
302 	.get_minrecs		= xfs_refcountbt_get_minrecs,
303 	.get_maxrecs		= xfs_refcountbt_get_maxrecs,
304 	.init_key_from_rec	= xfs_refcountbt_init_key_from_rec,
305 	.init_high_key_from_rec	= xfs_refcountbt_init_high_key_from_rec,
306 	.init_rec_from_cur	= xfs_refcountbt_init_rec_from_cur,
307 	.init_ptr_from_cur	= xfs_refcountbt_init_ptr_from_cur,
308 	.key_diff		= xfs_refcountbt_key_diff,
309 	.buf_ops		= &xfs_refcountbt_buf_ops,
310 	.diff_two_keys		= xfs_refcountbt_diff_two_keys,
311 	.keys_inorder		= xfs_refcountbt_keys_inorder,
312 	.recs_inorder		= xfs_refcountbt_recs_inorder,
313 };
314 
315 /*
316  * Allocate a new refcount btree cursor.
317  */
318 struct xfs_btree_cur *
319 xfs_refcountbt_init_cursor(
320 	struct xfs_mount	*mp,
321 	struct xfs_trans	*tp,
322 	struct xfs_buf		*agbp,
323 	xfs_agnumber_t		agno)
324 {
325 	struct xfs_agf		*agf = XFS_BUF_TO_AGF(agbp);
326 	struct xfs_btree_cur	*cur;
327 
328 	ASSERT(agno != NULLAGNUMBER);
329 	ASSERT(agno < mp->m_sb.sb_agcount);
330 	cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
331 
332 	cur->bc_tp = tp;
333 	cur->bc_mp = mp;
334 	cur->bc_btnum = XFS_BTNUM_REFC;
335 	cur->bc_blocklog = mp->m_sb.sb_blocklog;
336 	cur->bc_ops = &xfs_refcountbt_ops;
337 	cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2);
338 
339 	cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
340 
341 	cur->bc_private.a.agbp = agbp;
342 	cur->bc_private.a.agno = agno;
343 	cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
344 
345 	cur->bc_private.a.priv.refc.nr_ops = 0;
346 	cur->bc_private.a.priv.refc.shape_changes = 0;
347 
348 	return cur;
349 }
350 
351 /*
352  * Calculate the number of records in a refcount btree block.
353  */
354 int
355 xfs_refcountbt_maxrecs(
356 	int			blocklen,
357 	bool			leaf)
358 {
359 	blocklen -= XFS_REFCOUNT_BLOCK_LEN;
360 
361 	if (leaf)
362 		return blocklen / sizeof(struct xfs_refcount_rec);
363 	return blocklen / (sizeof(struct xfs_refcount_key) +
364 			   sizeof(xfs_refcount_ptr_t));
365 }
366 
367 /* Compute the maximum height of a refcount btree. */
368 void
369 xfs_refcountbt_compute_maxlevels(
370 	struct xfs_mount		*mp)
371 {
372 	mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
373 			mp->m_refc_mnr, mp->m_sb.sb_agblocks);
374 }
375 
376 /* Calculate the refcount btree size for some records. */
377 xfs_extlen_t
378 xfs_refcountbt_calc_size(
379 	struct xfs_mount	*mp,
380 	unsigned long long	len)
381 {
382 	return xfs_btree_calc_size(mp->m_refc_mnr, len);
383 }
384 
385 /*
386  * Calculate the maximum refcount btree size.
387  */
388 xfs_extlen_t
389 xfs_refcountbt_max_size(
390 	struct xfs_mount	*mp,
391 	xfs_agblock_t		agblocks)
392 {
393 	/* Bail out if we're uninitialized, which can happen in mkfs. */
394 	if (mp->m_refc_mxr[0] == 0)
395 		return 0;
396 
397 	return xfs_refcountbt_calc_size(mp, agblocks);
398 }
399 
400 /*
401  * Figure out how many blocks to reserve and how many are used by this btree.
402  */
403 int
404 xfs_refcountbt_calc_reserves(
405 	struct xfs_mount	*mp,
406 	struct xfs_trans	*tp,
407 	xfs_agnumber_t		agno,
408 	xfs_extlen_t		*ask,
409 	xfs_extlen_t		*used)
410 {
411 	struct xfs_buf		*agbp;
412 	struct xfs_agf		*agf;
413 	xfs_agblock_t		agblocks;
414 	xfs_extlen_t		tree_len;
415 	int			error;
416 
417 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
418 		return 0;
419 
420 
421 	error = xfs_alloc_read_agf(mp, tp, agno, 0, &agbp);
422 	if (error)
423 		return error;
424 
425 	agf = XFS_BUF_TO_AGF(agbp);
426 	agblocks = be32_to_cpu(agf->agf_length);
427 	tree_len = be32_to_cpu(agf->agf_refcount_blocks);
428 	xfs_trans_brelse(tp, agbp);
429 
430 	/*
431 	 * The log is permanently allocated, so the space it occupies will
432 	 * never be available for the kinds of things that would require btree
433 	 * expansion.  We therefore can pretend the space isn't there.
434 	 */
435 	if (mp->m_sb.sb_logstart &&
436 	    XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart) == agno)
437 		agblocks -= mp->m_sb.sb_logblocks;
438 
439 	*ask += xfs_refcountbt_max_size(mp, agblocks);
440 	*used += tree_len;
441 
442 	return error;
443 }
444