git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/migrate.c

   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 "extents.h"
  12 #include "io.h"
  13 #include "journal.h"
  14 #include "keylist.h"
  15 #include "migrate.h"
  16 #include "move.h"
  17 #include "replicas.h"
  18 #include "super-io.h"
  19
  20 static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
  21                          unsigned dev_idx, int flags, bool metadata)
  22 {
  23         unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
  24         unsigned lost = metadata ? BCH_FORCE_IF_METADATA_LOST : BCH_FORCE_IF_DATA_LOST;
  25         unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
  26         unsigned nr_good;
  27
  28         bch2_bkey_drop_device(k, dev_idx);
  29
  30         nr_good = bch2_bkey_durability(c, k.s_c);
  31         if ((!nr_good && !(flags & lost)) ||
  32             (nr_good < replicas && !(flags & degraded)))
  33                 return -EINVAL;
  34
  35         return 0;
  36 }
  37
  38 static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags,
  39                                    enum btree_id btree_id)
  40 {
  41         struct btree_trans trans;
  42         struct btree_iter iter;
  43         struct bkey_s_c k;
  44         struct bkey_buf sk;
  45         int ret = 0;
  46
  47         bch2_bkey_buf_init(&sk);
  48         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
  49
  50         bch2_trans_iter_init(&trans, &iter, btree_id, POS_MIN,
  51                              BTREE_ITER_PREFETCH|
  52                              BTREE_ITER_ALL_SNAPSHOTS);
  53
  54         while ((bch2_trans_begin(&trans),
  55                 (k = bch2_btree_iter_peek(&iter)).k) &&
  56                !(ret = bkey_err(k))) {
  57                 if (!bch2_bkey_has_device(k, dev_idx)) {
  58                         bch2_btree_iter_advance(&iter);
  59                         continue;
  60                 }
  61
  62                 bch2_bkey_buf_reassemble(&sk, c, k);
  63
  64                 ret = drop_dev_ptrs(c, bkey_i_to_s(sk.k),
  65                                     dev_idx, flags, false);
  66                 if (ret)
  67                         break;
  68
  69                 /*
  70                  * If the new extent no longer has any pointers, bch2_extent_normalize()
  71                  * will do the appropriate thing with it (turning it into a
  72                  * KEY_TYPE_error key, or just a discard if it was a cached extent)
  73                  */
  74                 bch2_extent_normalize(c, bkey_i_to_s(sk.k));
  75
  76                 /*
  77                  * Since we're not inserting through an extent iterator
  78                  * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
  79                  * we aren't using the extent overwrite path to delete, we're
  80                  * just using the normal key deletion path:
  81                  */
  82                 if (bkey_deleted(&sk.k->k))
  83                         sk.k->k.size = 0;
  84
  85                 ret   = bch2_btree_iter_traverse(&iter) ?:
  86                         bch2_trans_update(&trans, &iter, sk.k,
  87                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
  88                         bch2_trans_commit(&trans, NULL, NULL,
  89                                         BTREE_INSERT_NOFAIL);
  90
  91                 /*
  92                  * don't want to leave ret == -EINTR, since if we raced and
  93                  * something else overwrote the key we could spuriously return
  94                  * -EINTR below:
  95                  */
  96                 if (ret == -EINTR)
  97                         ret = 0;
  98                 if (ret)
  99                         break;
 100         }
 101         bch2_trans_iter_exit(&trans, &iter);
 102
 103         bch2_trans_exit(&trans);
 104         bch2_bkey_buf_exit(&sk, c);
 105
 106         BUG_ON(ret == -EINTR);
 107
 108         return ret;
 109 }
 110
 111 static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 112 {
 113         return  __bch2_dev_usrdata_drop(c, dev_idx, flags, BTREE_ID_extents) ?:
 114                 __bch2_dev_usrdata_drop(c, dev_idx, flags, BTREE_ID_reflink);
 115 }
 116
 117 static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 118 {
 119         struct btree_trans trans;
 120         struct btree_iter iter;
 121         struct closure cl;
 122         struct btree *b;
 123         struct bkey_buf k;
 124         unsigned id;
 125         int ret;
 126
 127         /* don't handle this yet: */
 128         if (flags & BCH_FORCE_IF_METADATA_LOST)
 129                 return -EINVAL;
 130
 131         bch2_bkey_buf_init(&k);
 132         bch2_trans_init(&trans, c, 0, 0);
 133         closure_init_stack(&cl);
 134
 135         for (id = 0; id < BTREE_ID_NR; id++) {
 136                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 137                                           BTREE_ITER_PREFETCH);
 138 retry:
 139                 ret = 0;
 140                 while (bch2_trans_begin(&trans),
 141                        (b = bch2_btree_iter_peek_node(&iter)) &&
 142                        !(ret = PTR_ERR_OR_ZERO(b))) {
 143                         if (!bch2_bkey_has_device(bkey_i_to_s_c(&b->key),
 144                                                   dev_idx))
 145                                 goto next;
 146
 147                         bch2_bkey_buf_copy(&k, c, &b->key);
 148
 149                         ret = drop_dev_ptrs(c, bkey_i_to_s(k.k),
 150                                             dev_idx, flags, true);
 151                         if (ret) {
 152                                 bch_err(c, "Cannot drop device without losing data");
 153                                 break;
 154                         }
 155
 156                         ret = bch2_btree_node_update_key(&trans, &iter, b, k.k, false);
 157                         if (ret == -EINTR) {
 158                                 ret = 0;
 159                                 continue;
 160                         }
 161
 162                         if (ret) {
 163                                 bch_err(c, "Error updating btree node key: %i", ret);
 164                                 break;
 165                         }
 166 next:
 167                         bch2_btree_iter_next_node(&iter);
 168                 }
 169                 if (ret == -EINTR)
 170                         goto retry;
 171
 172                 bch2_trans_iter_exit(&trans, &iter);
 173
 174                 if (ret)
 175                         goto err;
 176         }
 177
 178         /* flush relevant btree updates */
 179         closure_wait_event(&c->btree_interior_update_wait,
 180                            !bch2_btree_interior_updates_nr_pending(c));
 181
 182         ret = 0;
 183 err:
 184         bch2_trans_exit(&trans);
 185         bch2_bkey_buf_exit(&k, c);
 186
 187         BUG_ON(ret == -EINTR);
 188
 189         return ret;
 190 }
 191
 192 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 193 {
 194         return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
 195                 bch2_dev_metadata_drop(c, dev_idx, flags);
 196 }