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 "progress.h"
19 #include "replicas.h"
20 #include "super-io.h"
21
drop_dev_ptrs(struct bch_fs * c,struct bkey_s k,unsigned dev_idx,int flags,bool metadata)22 static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
23 unsigned dev_idx, int flags, bool metadata)
24 {
25 unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
26 unsigned lost = metadata ? BCH_FORCE_IF_METADATA_LOST : BCH_FORCE_IF_DATA_LOST;
27 unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
28 unsigned nr_good;
29
30 bch2_bkey_drop_device(k, dev_idx);
31
32 nr_good = bch2_bkey_durability(c, k.s_c);
33 if ((!nr_good && !(flags & lost)) ||
34 (nr_good < replicas && !(flags & degraded)))
35 return -BCH_ERR_remove_would_lose_data;
36
37 return 0;
38 }
39
bch2_dev_usrdata_drop_key(struct btree_trans * trans,struct btree_iter * iter,struct bkey_s_c k,unsigned dev_idx,int flags)40 static int bch2_dev_usrdata_drop_key(struct btree_trans *trans,
41 struct btree_iter *iter,
42 struct bkey_s_c k,
43 unsigned dev_idx,
44 int flags)
45 {
46 struct bch_fs *c = trans->c;
47 struct bkey_i *n;
48 int ret;
49
50 if (!bch2_bkey_has_device_c(k, dev_idx))
51 return 0;
52
53 n = bch2_bkey_make_mut(trans, iter, &k, BTREE_UPDATE_internal_snapshot_node);
54 ret = PTR_ERR_OR_ZERO(n);
55 if (ret)
56 return ret;
57
58 ret = drop_dev_ptrs(c, bkey_i_to_s(n), dev_idx, flags, false);
59 if (ret)
60 return ret;
61
62 /*
63 * If the new extent no longer has any pointers, bch2_extent_normalize()
64 * will do the appropriate thing with it (turning it into a
65 * KEY_TYPE_error key, or just a discard if it was a cached extent)
66 */
67 bch2_extent_normalize(c, bkey_i_to_s(n));
68
69 /*
70 * Since we're not inserting through an extent iterator
71 * (BTREE_ITER_all_snapshots iterators aren't extent iterators),
72 * we aren't using the extent overwrite path to delete, we're
73 * just using the normal key deletion path:
74 */
75 if (bkey_deleted(&n->k))
76 n->k.size = 0;
77 return 0;
78 }
79
bch2_dev_usrdata_drop(struct bch_fs * c,struct progress_indicator_state * progress,unsigned dev_idx,int flags)80 static int bch2_dev_usrdata_drop(struct bch_fs *c,
81 struct progress_indicator_state *progress,
82 unsigned dev_idx, int flags)
83 {
84 struct btree_trans *trans = bch2_trans_get(c);
85 enum btree_id id;
86 int ret = 0;
87
88 for (id = 0; id < BTREE_ID_NR; id++) {
89 if (!btree_type_has_ptrs(id))
90 continue;
91
92 ret = for_each_btree_key_commit(trans, iter, id, POS_MIN,
93 BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
94 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
95 bch2_progress_update_iter(trans, progress, &iter, "dropping user data");
96 bch2_dev_usrdata_drop_key(trans, &iter, k, dev_idx, flags);
97 }));
98 if (ret)
99 break;
100 }
101
102 bch2_trans_put(trans);
103
104 return ret;
105 }
106
bch2_dev_metadata_drop(struct bch_fs * c,struct progress_indicator_state * progress,unsigned dev_idx,int flags)107 static int bch2_dev_metadata_drop(struct bch_fs *c,
108 struct progress_indicator_state *progress,
109 unsigned dev_idx, int flags)
110 {
111 struct btree_trans *trans;
112 struct btree_iter iter;
113 struct closure cl;
114 struct btree *b;
115 struct bkey_buf k;
116 unsigned id;
117 int ret;
118
119 /* don't handle this yet: */
120 if (flags & BCH_FORCE_IF_METADATA_LOST)
121 return -BCH_ERR_remove_with_metadata_missing_unimplemented;
122
123 trans = bch2_trans_get(c);
124 bch2_bkey_buf_init(&k);
125 closure_init_stack(&cl);
126
127 for (id = 0; id < BTREE_ID_NR; id++) {
128 bch2_trans_node_iter_init(trans, &iter, id, POS_MIN, 0, 0,
129 BTREE_ITER_prefetch);
130 retry:
131 ret = 0;
132 while (bch2_trans_begin(trans),
133 (b = bch2_btree_iter_peek_node(trans, &iter)) &&
134 !(ret = PTR_ERR_OR_ZERO(b))) {
135 bch2_progress_update_iter(trans, progress, &iter, "dropping metadata");
136
137 if (!bch2_bkey_has_device_c(bkey_i_to_s_c(&b->key), dev_idx))
138 goto next;
139
140 bch2_bkey_buf_copy(&k, c, &b->key);
141
142 ret = drop_dev_ptrs(c, bkey_i_to_s(k.k),
143 dev_idx, flags, true);
144 if (ret)
145 break;
146
147 ret = bch2_btree_node_update_key(trans, &iter, b, k.k, 0, false);
148 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
149 ret = 0;
150 continue;
151 }
152
153 bch_err_msg(c, ret, "updating btree node key");
154 if (ret)
155 break;
156 next:
157 bch2_btree_iter_next_node(trans, &iter);
158 }
159 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
160 goto retry;
161
162 bch2_trans_iter_exit(trans, &iter);
163
164 if (ret)
165 goto err;
166 }
167
168 bch2_btree_interior_updates_flush(c);
169 ret = 0;
170 err:
171 bch2_bkey_buf_exit(&k, c);
172 bch2_trans_put(trans);
173
174 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
175
176 return ret;
177 }
178
bch2_dev_data_drop(struct bch_fs * c,unsigned dev_idx,int flags)179 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
180 {
181 struct progress_indicator_state progress;
182 bch2_progress_init(&progress, c,
183 BIT_ULL(BTREE_ID_extents)|
184 BIT_ULL(BTREE_ID_reflink));
185
186 return bch2_dev_usrdata_drop(c, &progress, dev_idx, flags) ?:
187 bch2_dev_metadata_drop(c, &progress, dev_idx, flags);
188 }
189