xref: /linux/fs/bcachefs/migrate.c (revision 2241f81c91f211b512bd2c3a26a4a74258d0e008)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Code for moving data off a device.
4  */
5 
6 #include "bcachefs.h"
7 #include "bkey_buf.h"
8 #include "btree_update.h"
9 #include "btree_update_interior.h"
10 #include "buckets.h"
11 #include "errcode.h"
12 #include "extents.h"
13 #include "io_write.h"
14 #include "journal.h"
15 #include "keylist.h"
16 #include "migrate.h"
17 #include "move.h"
18 #include "replicas.h"
19 #include "super-io.h"
20 
21 static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
22 			 unsigned dev_idx, int flags, bool metadata)
23 {
24 	unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
25 	unsigned lost = metadata ? BCH_FORCE_IF_METADATA_LOST : BCH_FORCE_IF_DATA_LOST;
26 	unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
27 	unsigned nr_good;
28 
29 	bch2_bkey_drop_device(k, dev_idx);
30 
31 	nr_good = bch2_bkey_durability(c, k.s_c);
32 	if ((!nr_good && !(flags & lost)) ||
33 	    (nr_good < replicas && !(flags & degraded)))
34 		return -EINVAL;
35 
36 	return 0;
37 }
38 
39 static int bch2_dev_usrdata_drop_key(struct btree_trans *trans,
40 				     struct btree_iter *iter,
41 				     struct bkey_s_c k,
42 				     unsigned dev_idx,
43 				     int flags)
44 {
45 	struct bch_fs *c = trans->c;
46 	struct bkey_i *n;
47 	int ret;
48 
49 	if (!bch2_bkey_has_device_c(k, dev_idx))
50 		return 0;
51 
52 	n = bch2_bkey_make_mut(trans, iter, &k, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
53 	ret = PTR_ERR_OR_ZERO(n);
54 	if (ret)
55 		return ret;
56 
57 	ret = drop_dev_ptrs(c, bkey_i_to_s(n), dev_idx, flags, false);
58 	if (ret)
59 		return ret;
60 
61 	/*
62 	 * If the new extent no longer has any pointers, bch2_extent_normalize()
63 	 * will do the appropriate thing with it (turning it into a
64 	 * KEY_TYPE_error key, or just a discard if it was a cached extent)
65 	 */
66 	bch2_extent_normalize(c, bkey_i_to_s(n));
67 
68 	/*
69 	 * Since we're not inserting through an extent iterator
70 	 * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
71 	 * we aren't using the extent overwrite path to delete, we're
72 	 * just using the normal key deletion path:
73 	 */
74 	if (bkey_deleted(&n->k))
75 		n->k.size = 0;
76 	return 0;
77 }
78 
79 static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
80 {
81 	struct btree_trans *trans = bch2_trans_get(c);
82 	struct btree_iter iter;
83 	struct bkey_s_c k;
84 	enum btree_id id;
85 	int ret = 0;
86 
87 	for (id = 0; id < BTREE_ID_NR; id++) {
88 		if (!btree_type_has_ptrs(id))
89 			continue;
90 
91 		ret = for_each_btree_key_commit(trans, iter, id, POS_MIN,
92 				BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
93 				NULL, NULL, BTREE_INSERT_NOFAIL,
94 			bch2_dev_usrdata_drop_key(trans, &iter, k, dev_idx, flags));
95 		if (ret)
96 			break;
97 	}
98 
99 	bch2_trans_put(trans);
100 
101 	return ret;
102 }
103 
104 static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
105 {
106 	struct btree_trans *trans;
107 	struct btree_iter iter;
108 	struct closure cl;
109 	struct btree *b;
110 	struct bkey_buf k;
111 	unsigned id;
112 	int ret;
113 
114 	/* don't handle this yet: */
115 	if (flags & BCH_FORCE_IF_METADATA_LOST)
116 		return -EINVAL;
117 
118 	trans = bch2_trans_get(c);
119 	bch2_bkey_buf_init(&k);
120 	closure_init_stack(&cl);
121 
122 	for (id = 0; id < BTREE_ID_NR; id++) {
123 		bch2_trans_node_iter_init(trans, &iter, id, POS_MIN, 0, 0,
124 					  BTREE_ITER_PREFETCH);
125 retry:
126 		ret = 0;
127 		while (bch2_trans_begin(trans),
128 		       (b = bch2_btree_iter_peek_node(&iter)) &&
129 		       !(ret = PTR_ERR_OR_ZERO(b))) {
130 			if (!bch2_bkey_has_device_c(bkey_i_to_s_c(&b->key), dev_idx))
131 				goto next;
132 
133 			bch2_bkey_buf_copy(&k, c, &b->key);
134 
135 			ret = drop_dev_ptrs(c, bkey_i_to_s(k.k),
136 					    dev_idx, flags, true);
137 			if (ret) {
138 				bch_err(c, "Cannot drop device without losing data");
139 				break;
140 			}
141 
142 			ret = bch2_btree_node_update_key(trans, &iter, b, k.k, 0, false);
143 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
144 				ret = 0;
145 				continue;
146 			}
147 
148 			if (ret) {
149 				bch_err_msg(c, ret, "updating btree node key");
150 				break;
151 			}
152 next:
153 			bch2_btree_iter_next_node(&iter);
154 		}
155 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
156 			goto retry;
157 
158 		bch2_trans_iter_exit(trans, &iter);
159 
160 		if (ret)
161 			goto err;
162 	}
163 
164 	bch2_btree_interior_updates_flush(c);
165 	ret = 0;
166 err:
167 	bch2_bkey_buf_exit(&k, c);
168 	bch2_trans_put(trans);
169 
170 	BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
171 
172 	return ret;
173 }
174 
175 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
176 {
177 	return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
178 		bch2_dev_metadata_drop(c, dev_idx, flags);
179 }
180