xref: /linux/fs/xfs/libxfs/xfs_btree_mem.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (c) 2021-2024 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <djwong@kernel.org>
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_trans.h"
14 #include "xfs_btree.h"
15 #include "xfs_error.h"
16 #include "xfs_buf_mem.h"
17 #include "xfs_btree_mem.h"
18 #include "xfs_ag.h"
19 #include "xfs_buf_item.h"
20 #include "xfs_trace.h"
21 
22 /* Set the root of an in-memory btree. */
23 void
24 xfbtree_set_root(
25 	struct xfs_btree_cur		*cur,
26 	const union xfs_btree_ptr	*ptr,
27 	int				inc)
28 {
29 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_MEM);
30 
31 	cur->bc_mem.xfbtree->root = *ptr;
32 	cur->bc_mem.xfbtree->nlevels += inc;
33 }
34 
35 /* Initialize a pointer from the in-memory btree header. */
36 void
37 xfbtree_init_ptr_from_cur(
38 	struct xfs_btree_cur		*cur,
39 	union xfs_btree_ptr		*ptr)
40 {
41 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_MEM);
42 
43 	*ptr = cur->bc_mem.xfbtree->root;
44 }
45 
46 /* Duplicate an in-memory btree cursor. */
47 struct xfs_btree_cur *
48 xfbtree_dup_cursor(
49 	struct xfs_btree_cur		*cur)
50 {
51 	struct xfs_btree_cur		*ncur;
52 
53 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_MEM);
54 
55 	ncur = xfs_btree_alloc_cursor(cur->bc_mp, cur->bc_tp, cur->bc_ops,
56 			cur->bc_maxlevels, cur->bc_cache);
57 	ncur->bc_flags = cur->bc_flags;
58 	ncur->bc_nlevels = cur->bc_nlevels;
59 	ncur->bc_mem.xfbtree = cur->bc_mem.xfbtree;
60 	if (cur->bc_group)
61 		ncur->bc_group = xfs_group_hold(cur->bc_group);
62 	return ncur;
63 }
64 
65 /* Close the btree xfile and release all resources. */
66 void
67 xfbtree_destroy(
68 	struct xfbtree		*xfbt)
69 {
70 	xfs_buftarg_drain(xfbt->target);
71 }
72 
73 /* Compute the number of bytes available for records. */
74 static inline unsigned int
75 xfbtree_rec_bytes(
76 	struct xfs_mount		*mp,
77 	const struct xfs_btree_ops	*ops)
78 {
79 	return XMBUF_BLOCKSIZE - XFS_BTREE_LBLOCK_CRC_LEN;
80 }
81 
82 /* Initialize an empty leaf block as the btree root. */
83 STATIC int
84 xfbtree_init_leaf_block(
85 	struct xfs_mount		*mp,
86 	struct xfbtree			*xfbt,
87 	const struct xfs_btree_ops	*ops)
88 {
89 	struct xfs_buf			*bp;
90 	xfbno_t				bno = xfbt->highest_bno++;
91 	int				error;
92 
93 	error = xfs_buf_get(xfbt->target, xfbno_to_daddr(bno), XFBNO_BBSIZE,
94 			&bp);
95 	if (error)
96 		return error;
97 
98 	trace_xfbtree_create_root_buf(xfbt, bp);
99 
100 	bp->b_ops = ops->buf_ops;
101 	xfs_btree_init_buf(mp, bp, ops, 0, 0, xfbt->owner);
102 	xfs_buf_relse(bp);
103 
104 	xfbt->root.l = cpu_to_be64(bno);
105 	return 0;
106 }
107 
108 /*
109  * Create an in-memory btree root that can be used with the given xmbuf.
110  * Callers must set xfbt->owner.
111  */
112 int
113 xfbtree_init(
114 	struct xfs_mount		*mp,
115 	struct xfbtree			*xfbt,
116 	struct xfs_buftarg		*btp,
117 	const struct xfs_btree_ops	*ops)
118 {
119 	unsigned int			blocklen = xfbtree_rec_bytes(mp, ops);
120 	unsigned int			keyptr_len;
121 	int				error;
122 
123 	/* Requires a long-format CRC-format btree */
124 	if (!xfs_has_crc(mp)) {
125 		ASSERT(xfs_has_crc(mp));
126 		return -EINVAL;
127 	}
128 	if (ops->ptr_len != XFS_BTREE_LONG_PTR_LEN) {
129 		ASSERT(ops->ptr_len == XFS_BTREE_LONG_PTR_LEN);
130 		return -EINVAL;
131 	}
132 
133 	memset(xfbt, 0, sizeof(*xfbt));
134 	xfbt->target = btp;
135 
136 	/* Set up min/maxrecs for this btree. */
137 	keyptr_len = ops->key_len + sizeof(__be64);
138 	xfbt->maxrecs[0] = blocklen / ops->rec_len;
139 	xfbt->maxrecs[1] = blocklen / keyptr_len;
140 	xfbt->minrecs[0] = xfbt->maxrecs[0] / 2;
141 	xfbt->minrecs[1] = xfbt->maxrecs[1] / 2;
142 	xfbt->highest_bno = 0;
143 	xfbt->nlevels = 1;
144 
145 	/* Initialize the empty btree. */
146 	error = xfbtree_init_leaf_block(mp, xfbt, ops);
147 	if (error)
148 		goto err_freesp;
149 
150 	trace_xfbtree_init(mp, xfbt, ops);
151 
152 	return 0;
153 
154 err_freesp:
155 	xfs_buftarg_drain(xfbt->target);
156 	return error;
157 }
158 
159 /* Allocate a block to our in-memory btree. */
160 int
161 xfbtree_alloc_block(
162 	struct xfs_btree_cur		*cur,
163 	const union xfs_btree_ptr	*start,
164 	union xfs_btree_ptr		*new,
165 	int				*stat)
166 {
167 	struct xfbtree			*xfbt = cur->bc_mem.xfbtree;
168 	xfbno_t				bno = xfbt->highest_bno++;
169 
170 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_MEM);
171 
172 	trace_xfbtree_alloc_block(xfbt, cur, bno);
173 
174 	/* Fail if the block address exceeds the maximum for the buftarg. */
175 	if (!xfbtree_verify_bno(xfbt, bno)) {
176 		ASSERT(xfbtree_verify_bno(xfbt, bno));
177 		*stat = 0;
178 		return 0;
179 	}
180 
181 	new->l = cpu_to_be64(bno);
182 	*stat = 1;
183 	return 0;
184 }
185 
186 /* Free a block from our in-memory btree. */
187 int
188 xfbtree_free_block(
189 	struct xfs_btree_cur	*cur,
190 	struct xfs_buf		*bp)
191 {
192 	struct xfbtree		*xfbt = cur->bc_mem.xfbtree;
193 	xfs_daddr_t		daddr = xfs_buf_daddr(bp);
194 	xfbno_t			bno = xfs_daddr_to_xfbno(daddr);
195 
196 	ASSERT(cur->bc_ops->type == XFS_BTREE_TYPE_MEM);
197 
198 	trace_xfbtree_free_block(xfbt, cur, bno);
199 
200 	if (bno + 1 == xfbt->highest_bno)
201 		xfbt->highest_bno--;
202 
203 	return 0;
204 }
205 
206 /* Return the minimum number of records for a btree block. */
207 int
208 xfbtree_get_minrecs(
209 	struct xfs_btree_cur	*cur,
210 	int			level)
211 {
212 	struct xfbtree		*xfbt = cur->bc_mem.xfbtree;
213 
214 	return xfbt->minrecs[level != 0];
215 }
216 
217 /* Return the maximum number of records for a btree block. */
218 int
219 xfbtree_get_maxrecs(
220 	struct xfs_btree_cur	*cur,
221 	int			level)
222 {
223 	struct xfbtree		*xfbt = cur->bc_mem.xfbtree;
224 
225 	return xfbt->maxrecs[level != 0];
226 }
227 
228 /* If this log item is a buffer item that came from the xfbtree, return it. */
229 static inline struct xfs_buf *
230 xfbtree_buf_match(
231 	struct xfbtree			*xfbt,
232 	const struct xfs_log_item	*lip)
233 {
234 	const struct xfs_buf_log_item	*bli;
235 	struct xfs_buf			*bp;
236 
237 	if (lip->li_type != XFS_LI_BUF)
238 		return NULL;
239 
240 	bli = container_of(lip, struct xfs_buf_log_item, bli_item);
241 	bp = bli->bli_buf;
242 	if (bp->b_target != xfbt->target)
243 		return NULL;
244 
245 	return bp;
246 }
247 
248 /*
249  * Commit changes to the incore btree immediately by writing all dirty xfbtree
250  * buffers to the backing xfile.  This detaches all xfbtree buffers from the
251  * transaction, even on failure.  The buffer locks are dropped between the
252  * delwri queue and submit, so the caller must synchronize btree access.
253  *
254  * Normally we'd let the buffers commit with the transaction and get written to
255  * the xfile via the log, but online repair stages ephemeral btrees in memory
256  * and uses the btree_staging functions to write new btrees to disk atomically.
257  * The in-memory btree (and its backing store) are discarded at the end of the
258  * repair phase, which means that xfbtree buffers cannot commit with the rest
259  * of a transaction.
260  *
261  * In other words, online repair only needs the transaction to collect buffer
262  * pointers and to avoid buffer deadlocks, not to guarantee consistency of
263  * updates.
264  */
265 int
266 xfbtree_trans_commit(
267 	struct xfbtree		*xfbt,
268 	struct xfs_trans	*tp)
269 {
270 	struct xfs_log_item	*lip, *n;
271 	bool			tp_dirty = false;
272 	int			error = 0;
273 
274 	/*
275 	 * For each xfbtree buffer attached to the transaction, write the dirty
276 	 * buffers to the xfile and release them.
277 	 */
278 	list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
279 		struct xfs_buf	*bp = xfbtree_buf_match(xfbt, lip);
280 
281 		if (!bp) {
282 			if (test_bit(XFS_LI_DIRTY, &lip->li_flags))
283 				tp_dirty |= true;
284 			continue;
285 		}
286 
287 		trace_xfbtree_trans_commit_buf(xfbt, bp);
288 
289 		xmbuf_trans_bdetach(tp, bp);
290 
291 		/*
292 		 * If the buffer fails verification, note the failure but
293 		 * continue walking the transaction items so that we remove all
294 		 * ephemeral btree buffers.
295 		 */
296 		if (!error)
297 			error = xmbuf_finalize(bp);
298 
299 		xfs_buf_relse(bp);
300 	}
301 
302 	/*
303 	 * Reset the transaction's dirty flag to reflect the dirty state of the
304 	 * log items that are still attached.
305 	 */
306 	tp->t_flags = (tp->t_flags & ~XFS_TRANS_DIRTY) |
307 			(tp_dirty ? XFS_TRANS_DIRTY : 0);
308 
309 	return error;
310 }
311 
312 /*
313  * Cancel changes to the incore btree by detaching all the xfbtree buffers.
314  * Changes are not undone, so callers must not access the btree ever again.
315  */
316 void
317 xfbtree_trans_cancel(
318 	struct xfbtree		*xfbt,
319 	struct xfs_trans	*tp)
320 {
321 	struct xfs_log_item	*lip, *n;
322 	bool			tp_dirty = false;
323 
324 	list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
325 		struct xfs_buf	*bp = xfbtree_buf_match(xfbt, lip);
326 
327 		if (!bp) {
328 			if (test_bit(XFS_LI_DIRTY, &lip->li_flags))
329 				tp_dirty |= true;
330 			continue;
331 		}
332 
333 		trace_xfbtree_trans_cancel_buf(xfbt, bp);
334 
335 		xmbuf_trans_bdetach(tp, bp);
336 		xfs_buf_relse(bp);
337 	}
338 
339 	/*
340 	 * Reset the transaction's dirty flag to reflect the dirty state of the
341 	 * log items that are still attached.
342 	 */
343 	tp->t_flags = (tp->t_flags & ~XFS_TRANS_DIRTY) |
344 			(tp_dirty ? XFS_TRANS_DIRTY : 0);
345 }
346