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 "super-io.h"
  15
  16 static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s_extent e,
  17                          unsigned dev_idx, int flags, bool metadata)
  18 {
  19         unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
  20         unsigned lost = metadata ? BCH_FORCE_IF_METADATA_LOST : BCH_FORCE_IF_DATA_LOST;
  21         unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
  22         unsigned nr_good;
  23
  24         bch2_extent_drop_device(e, dev_idx);
  25
  26         nr_good = bch2_extent_nr_good_ptrs(c, e.c);
  27         if ((!nr_good && !(flags & lost)) ||
  28             (nr_good < replicas && !(flags & degraded)))
  29                 return -EINVAL;
  30
  31         return 0;
  32 }
  33
  34 static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
  35 {
  36         struct bkey_s_c k;
  37         struct bkey_s_extent e;
  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, BCH_DATA_USER, k);
  53                         if (ret)
  54                                 break;
  55                         bch2_btree_iter_next(&iter);
  56                         continue;
  57                 }
  58
  59                 bkey_reassemble(&tmp.key, k);
  60                 e = bkey_i_to_s_extent(&tmp.key);
  61
  62                 ret = drop_dev_ptrs(c, e, 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, e.s);
  72
  73                 ret = bch2_mark_bkey_replicas(c, BCH_DATA_USER,
  74                                               bkey_i_to_s_c(&tmp.key));
  75                 if (ret)
  76                         break;
  77
  78                 iter.pos = bkey_start_pos(&tmp.key.k);
  79
  80                 ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
  81                                            BTREE_INSERT_ATOMIC|
  82                                            BTREE_INSERT_NOFAIL,
  83                                            BTREE_INSERT_ENTRY(&iter, &tmp.key));
  84
  85                 /*
  86                  * don't want to leave ret == -EINTR, since if we raced and
  87                  * something else overwrote the key we could spuriously return
  88                  * -EINTR below:
  89                  */
  90                 if (ret == -EINTR)
  91                         ret = 0;
  92                 if (ret)
  93                         break;
  94         }
  95
  96         bch2_btree_iter_unlock(&iter);
  97
  98         bch2_replicas_gc_end(c, ret);
  99         mutex_unlock(&c->replicas_gc_lock);
 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_iter iter;
 107         struct closure cl;
 108         struct btree *b;
 109         unsigned id;
 110         int ret;
 111
 112         /* don't handle this yet: */
 113         if (flags & BCH_FORCE_IF_METADATA_LOST)
 114                 return -EINVAL;
 115
 116         closure_init_stack(&cl);
 117
 118         mutex_lock(&c->replicas_gc_lock);
 119         bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
 120
 121         for (id = 0; id < BTREE_ID_NR; id++) {
 122                 for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
 123                         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
 124                         struct bkey_i_extent *new_key;
 125 retry:
 126                         if (!bch2_extent_has_device(bkey_i_to_s_c_extent(&b->key),
 127                                                     dev_idx)) {
 128                                 bch2_btree_iter_set_locks_want(&iter, 0);
 129
 130                                 ret = bch2_mark_bkey_replicas(c, BCH_DATA_BTREE,
 131                                                               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_extent(&tmp.k);
 137
 138                                 ret = drop_dev_ptrs(c, extent_i_to_s(new_key),
 139                                                     dev_idx, flags, true);
 140                                 if (ret)
 141                                         goto err;
 142
 143                                 if (!bch2_btree_iter_set_locks_want(&iter, U8_MAX)) {
 144                                         b = bch2_btree_iter_peek_node(&iter);
 145                                         goto retry;
 146                                 }
 147
 148                                 ret = bch2_btree_node_update_key(c, &iter, b, new_key);
 149                                 if (ret == -EINTR) {
 150                                         b = bch2_btree_iter_peek_node(&iter);
 151                                         goto retry;
 152                                 }
 153                                 if (ret)
 154                                         goto err;
 155                         }
 156                 }
 157                 bch2_btree_iter_unlock(&iter);
 158         }
 159
 160         ret = 0;
 161 out:
 162         bch2_replicas_gc_end(c, ret);
 163         mutex_unlock(&c->replicas_gc_lock);
 164
 165         return ret;
 166 err:
 167         bch2_btree_iter_unlock(&iter);
 168         goto out;
 169 }
 170
 171 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 172 {
 173         return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
 174                 bch2_dev_metadata_drop(c, dev_idx, flags);
 175 }