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

   1 /*
   2  * Code for moving data off a device.
   3  */
   4
   5 #include "bcachefs.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "buckets.h"
   9 #include "extents.h"
  10 #include "io.h"
  11 #include "journal.h"
  12 #include "keylist.h"
  13 #include "migrate.h"
  14 #include "move.h"
  15 #include "replicas.h"
  16 #include "super-io.h"
  17
  18 static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
  19                          unsigned dev_idx, int flags, bool metadata)
  20 {
  21         unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
  22         unsigned lost = metadata ? BCH_FORCE_IF_METADATA_LOST : BCH_FORCE_IF_DATA_LOST;
  23         unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
  24         unsigned nr_good;
  25
  26         bch2_bkey_drop_device(k, dev_idx);
  27
  28         nr_good = bch2_bkey_durability(c, k.s_c);
  29         if ((!nr_good && !(flags & lost)) ||
  30             (nr_good < replicas && !(flags & degraded)))
  31                 return -EINVAL;
  32
  33         return 0;
  34 }
  35
  36 static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
  37 {
  38         struct bkey_s_c k;
  39         BKEY_PADDED(key) tmp;
  40         struct btree_iter iter;
  41         int ret = 0;
  42
  43         mutex_lock(&c->replicas_gc_lock);
  44         bch2_replicas_gc_start(c, (1 << BCH_DATA_USER)|(1 << BCH_DATA_CACHED));
  45
  46         bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS,
  47                              POS_MIN, BTREE_ITER_PREFETCH);
  48
  49         while ((k = bch2_btree_iter_peek(&iter)).k &&
  50                !(ret = btree_iter_err(k))) {
  51                 if (!bkey_extent_is_data(k.k) ||
  52                     !bch2_extent_has_device(bkey_s_c_to_extent(k), dev_idx)) {
  53                         ret = bch2_mark_bkey_replicas(c, k);
  54                         if (ret)
  55                                 break;
  56                         bch2_btree_iter_next(&iter);
  57                         continue;
  58                 }
  59
  60                 bkey_reassemble(&tmp.key, k);
  61
  62                 ret = drop_dev_ptrs(c, bkey_i_to_s(&tmp.key),
  63                                     dev_idx, flags, false);
  64                 if (ret)
  65                         break;
  66
  67                 /*
  68                  * If the new extent no longer has any pointers, bch2_extent_normalize()
  69                  * will do the appropriate thing with it (turning it into a
  70                  * KEY_TYPE_error key, or just a discard if it was a cached extent)
  71                  */
  72                 bch2_extent_normalize(c, bkey_i_to_s(&tmp.key));
  73
  74                 iter.pos = bkey_start_pos(&tmp.key.k);
  75
  76                 ret = bch2_btree_insert_at(c, NULL, NULL,
  77                                            BTREE_INSERT_ATOMIC|
  78                                            BTREE_INSERT_NOFAIL,
  79                                            BTREE_INSERT_ENTRY(&iter, &tmp.key));
  80
  81                 /*
  82                  * don't want to leave ret == -EINTR, since if we raced and
  83                  * something else overwrote the key we could spuriously return
  84                  * -EINTR below:
  85                  */
  86                 if (ret == -EINTR)
  87                         ret = 0;
  88                 if (ret)
  89                         break;
  90         }
  91
  92         bch2_btree_iter_unlock(&iter);
  93
  94         bch2_replicas_gc_end(c, ret);
  95         mutex_unlock(&c->replicas_gc_lock);
  96
  97         return ret;
  98 }
  99
 100 static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 101 {
 102         struct btree_iter iter;
 103         struct closure cl;
 104         struct btree *b;
 105         unsigned id;
 106         int ret;
 107
 108         /* don't handle this yet: */
 109         if (flags & BCH_FORCE_IF_METADATA_LOST)
 110                 return -EINVAL;
 111
 112         closure_init_stack(&cl);
 113
 114         mutex_lock(&c->replicas_gc_lock);
 115         bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
 116
 117         for (id = 0; id < BTREE_ID_NR; id++) {
 118                 for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
 119                         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
 120                         struct bkey_i_btree_ptr *new_key;
 121 retry:
 122                         if (!bch2_bkey_has_device(bkey_i_to_s_c(&b->key),
 123                                                   dev_idx)) {
 124                                 /*
 125                                  * we might have found a btree node key we
 126                                  * needed to update, and then tried to update it
 127                                  * but got -EINTR after upgrading the iter, but
 128                                  * then raced and the node is now gone:
 129                                  */
 130                                 bch2_btree_iter_downgrade(&iter);
 131
 132                                 ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
 133                                 if (ret)
 134                                         goto err;
 135                         } else {
 136                                 bkey_copy(&tmp.k, &b->key);
 137                                 new_key = bkey_i_to_btree_ptr(&tmp.k);
 138
 139                                 ret = drop_dev_ptrs(c, bkey_i_to_s(&new_key->k_i),
 140                                                     dev_idx, flags, true);
 141                                 if (ret)
 142                                         goto err;
 143
 144                                 ret = bch2_btree_node_update_key(c, &iter, b, new_key);
 145                                 if (ret == -EINTR) {
 146                                         b = bch2_btree_iter_peek_node(&iter);
 147                                         goto retry;
 148                                 }
 149                                 if (ret)
 150                                         goto err;
 151                         }
 152                 }
 153                 bch2_btree_iter_unlock(&iter);
 154         }
 155
 156         /* flush relevant btree updates */
 157         while (1) {
 158                 closure_wait_event(&c->btree_interior_update_wait,
 159                                    !bch2_btree_interior_updates_nr_pending(c) ||
 160                                    c->btree_roots_dirty);
 161                 if (!bch2_btree_interior_updates_nr_pending(c))
 162                         break;
 163                 bch2_journal_meta(&c->journal);
 164         }
 165
 166         ret = 0;
 167 out:
 168         ret = bch2_replicas_gc_end(c, ret);
 169         mutex_unlock(&c->replicas_gc_lock);
 170
 171         return ret;
 172 err:
 173         bch2_btree_iter_unlock(&iter);
 174         goto out;
 175 }
 176
 177 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 178 {
 179         return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
 180                 bch2_dev_metadata_drop(c, dev_idx, flags);
 181 }