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