+// SPDX-License-Identifier: GPL-2.0
/*
* Code for moving data off a device.
*/
#include "bcachefs.h"
+#include "bkey_buf.h"
#include "btree_update.h"
+#include "btree_update_interior.h"
#include "buckets.h"
#include "extents.h"
#include "io.h"
#include "keylist.h"
#include "migrate.h"
#include "move.h"
+#include "replicas.h"
#include "super-io.h"
-static bool migrate_pred(void *arg, struct bkey_s_c_extent e)
-{
- struct bch_dev *ca = arg;
-
- return bch2_extent_has_device(e, ca->dev_idx);
-}
-
-#define MAX_DATA_OFF_ITER 10
-
-static int bch2_dev_usrdata_migrate(struct bch_fs *c, struct bch_dev *ca,
- int flags)
-{
- struct btree_iter iter;
- struct bkey_s_c k;
- struct bch_move_stats stats;
- unsigned pass = 0;
- int ret = 0;
-
- if (!(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_USER)))
- return 0;
-
- /*
- * XXX: we should be able to do this in one pass, but bch2_move_data()
- * can spuriously fail to move an extent due to racing with other move
- * operations
- */
- do {
- ret = bch2_move_data(c, NULL,
- SECTORS_IN_FLIGHT_PER_DEVICE,
- NULL,
- writepoint_hashed((unsigned long) current),
- 0,
- ca->dev_idx,
- migrate_pred, ca,
- &stats);
- if (ret) {
- bch_err(c, "error migrating data: %i", ret);
- return ret;
- }
- } while (atomic64_read(&stats.keys_moved) && pass++ < MAX_DATA_OFF_ITER);
-
- if (atomic64_read(&stats.keys_moved)) {
- bch_err(c, "unable to migrate all data in %d iterations",
- MAX_DATA_OFF_ITER);
- return -1;
- }
-
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, 1 << BCH_DATA_USER);
-
- for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN, BTREE_ITER_PREFETCH, k) {
- ret = bch2_check_mark_super(c, BCH_DATA_USER, bch2_bkey_devs(k));
- if (ret) {
- bch_err(c, "error migrating data %i from check_mark_super()", ret);
- break;
- }
- }
-
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
- return ret;
-}
-
-static int bch2_dev_metadata_migrate(struct bch_fs *c, struct bch_dev *ca,
- int flags)
-{
- struct btree_iter iter;
- struct btree *b;
- int ret = 0;
- unsigned id;
-
- if (!(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_BTREE)))
- return 0;
-
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
-
- for (id = 0; id < BTREE_ID_NR; id++) {
- for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
- struct bkey_s_c_extent e = bkey_i_to_s_c_extent(&b->key);
-
- if (!bch2_extent_has_device(e, ca->dev_idx))
- continue;
-
- ret = bch2_btree_node_rewrite(c, &iter, b->data->keys.seq, 0);
- if (ret) {
- bch2_btree_iter_unlock(&iter);
- goto err;
- }
- }
- ret = bch2_btree_iter_unlock(&iter);
- if (ret)
- goto err;
- }
-err:
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
- return ret;
-}
-
-int bch2_dev_data_migrate(struct bch_fs *c, struct bch_dev *ca, int flags)
-{
- BUG_ON(ca->mi.state == BCH_MEMBER_STATE_RW &&
- bch2_dev_is_online(ca));
-
- return bch2_dev_usrdata_migrate(c, ca, flags) ?:
- bch2_dev_metadata_migrate(c, ca, flags);
-}
-
-static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s_extent e,
+static int drop_dev_ptrs(struct bch_fs *c, struct bkey_s k,
unsigned dev_idx, int flags, bool metadata)
{
unsigned replicas = metadata ? c->opts.metadata_replicas : c->opts.data_replicas;
unsigned degraded = metadata ? BCH_FORCE_IF_METADATA_DEGRADED : BCH_FORCE_IF_DATA_DEGRADED;
unsigned nr_good;
- bch2_extent_drop_device(e, dev_idx);
+ bch2_bkey_drop_device(k, dev_idx);
- nr_good = bch2_extent_nr_good_ptrs(c, e.c);
+ nr_good = bch2_bkey_durability(c, k.s_c);
if ((!nr_good && !(flags & lost)) ||
(nr_good < replicas && !(flags & degraded)))
return -EINVAL;
return 0;
}
-static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
+static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags,
+ enum btree_id btree_id)
{
- struct bkey_s_c k;
- struct bkey_s_extent e;
- BKEY_PADDED(key) tmp;
+ struct btree_trans trans;
struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bkey_buf sk;
int ret = 0;
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, 1 << BCH_DATA_USER);
+ bch2_bkey_buf_init(&sk);
+ bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
- bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS,
- POS_MIN, BTREE_ITER_PREFETCH);
+ bch2_trans_iter_init(&trans, &iter, btree_id, POS_MIN,
+ BTREE_ITER_PREFETCH|
+ BTREE_ITER_ALL_SNAPSHOTS);
- while ((k = bch2_btree_iter_peek(&iter)).k &&
- !(ret = btree_iter_err(k))) {
- if (!bkey_extent_is_data(k.k) ||
- !bch2_extent_has_device(bkey_s_c_to_extent(k), dev_idx)) {
- ret = bch2_check_mark_super(c, BCH_DATA_USER,
- bch2_bkey_devs(k));
- if (ret)
- break;
- bch2_btree_iter_next(&iter);
+ while ((bch2_trans_begin(&trans),
+ (k = bch2_btree_iter_peek(&iter)).k) &&
+ !(ret = bkey_err(k))) {
+ if (!bch2_bkey_has_device(k, dev_idx)) {
+ bch2_btree_iter_advance(&iter);
continue;
}
- bkey_reassemble(&tmp.key, k);
- e = bkey_i_to_s_extent(&tmp.key);
+ bch2_bkey_buf_reassemble(&sk, c, k);
- ret = drop_dev_ptrs(c, e, dev_idx, flags, false);
+ ret = drop_dev_ptrs(c, bkey_i_to_s(sk.k),
+ dev_idx, flags, false);
if (ret)
break;
/*
* If the new extent no longer has any pointers, bch2_extent_normalize()
* will do the appropriate thing with it (turning it into a
- * KEY_TYPE_ERROR key, or just a discard if it was a cached extent)
+ * KEY_TYPE_error key, or just a discard if it was a cached extent)
*/
- bch2_extent_normalize(c, e.s);
-
- ret = bch2_check_mark_super(c, BCH_DATA_USER,
- bch2_bkey_devs(bkey_i_to_s_c(&tmp.key)));
- if (ret)
- break;
+ bch2_extent_normalize(c, bkey_i_to_s(sk.k));
- iter.pos = bkey_start_pos(&tmp.key.k);
+ /*
+ * Since we're not inserting through an extent iterator
+ * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
+ * we aren't using the extent overwrite path to delete, we're
+ * just using the normal key deletion path:
+ */
+ if (bkey_deleted(&sk.k->k))
+ sk.k->k.size = 0;
- ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
- BTREE_INSERT_ATOMIC|
- BTREE_INSERT_NOFAIL,
- BTREE_INSERT_ENTRY(&iter, &tmp.key));
+ ret = bch2_btree_iter_traverse(&iter) ?:
+ bch2_trans_update(&trans, &iter, sk.k,
+ BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
+ bch2_trans_commit(&trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL);
/*
* don't want to leave ret == -EINTR, since if we raced and
if (ret)
break;
}
+ bch2_trans_iter_exit(&trans, &iter);
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&sk, c);
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
+ BUG_ON(ret == -EINTR);
return ret;
}
+static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
+{
+ return __bch2_dev_usrdata_drop(c, dev_idx, flags, BTREE_ID_extents) ?:
+ __bch2_dev_usrdata_drop(c, dev_idx, flags, BTREE_ID_reflink);
+}
+
static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
{
+ struct btree_trans trans;
struct btree_iter iter;
struct closure cl;
struct btree *b;
+ struct bkey_buf k;
unsigned id;
int ret;
if (flags & BCH_FORCE_IF_METADATA_LOST)
return -EINVAL;
+ bch2_bkey_buf_init(&k);
+ bch2_trans_init(&trans, c, 0, 0);
closure_init_stack(&cl);
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
-
for (id = 0; id < BTREE_ID_NR; id++) {
- for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
- __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp;
- struct bkey_i_extent *new_key;
+ bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
+ BTREE_ITER_PREFETCH);
retry:
- if (!bch2_extent_has_device(bkey_i_to_s_c_extent(&b->key),
- dev_idx)) {
- bch2_btree_iter_set_locks_want(&iter, 0);
-
- ret = bch2_check_mark_super(c, BCH_DATA_BTREE,
- bch2_bkey_devs(bkey_i_to_s_c(&b->key)));
- if (ret)
- goto err;
- } else {
- bkey_copy(&tmp.k, &b->key);
- new_key = bkey_i_to_extent(&tmp.k);
-
- ret = drop_dev_ptrs(c, extent_i_to_s(new_key),
- dev_idx, flags, true);
- if (ret)
- goto err;
-
- if (!bch2_btree_iter_set_locks_want(&iter, U8_MAX)) {
- b = bch2_btree_iter_peek_node(&iter);
- goto retry;
- }
-
- ret = bch2_btree_node_update_key(c, &iter, b, new_key);
- if (ret == -EINTR) {
- b = bch2_btree_iter_peek_node(&iter);
- goto retry;
- }
- if (ret)
- goto err;
+ ret = 0;
+ while (bch2_trans_begin(&trans),
+ (b = bch2_btree_iter_peek_node(&iter)) &&
+ !(ret = PTR_ERR_OR_ZERO(b))) {
+ if (!bch2_bkey_has_device(bkey_i_to_s_c(&b->key),
+ dev_idx))
+ goto next;
+
+ bch2_bkey_buf_copy(&k, c, &b->key);
+
+ ret = drop_dev_ptrs(c, bkey_i_to_s(k.k),
+ dev_idx, flags, true);
+ if (ret) {
+ bch_err(c, "Cannot drop device without losing data");
+ break;
+ }
+
+ ret = bch2_btree_node_update_key(&trans, &iter, b, k.k, false);
+ if (ret == -EINTR) {
+ ret = 0;
+ continue;
+ }
+
+ if (ret) {
+ bch_err(c, "Error updating btree node key: %i", ret);
+ break;
}
+next:
+ bch2_btree_iter_next_node(&iter);
}
- bch2_btree_iter_unlock(&iter);
+ if (ret == -EINTR)
+ goto retry;
+
+ bch2_trans_iter_exit(&trans, &iter);
+
+ if (ret)
+ goto err;
}
+ /* flush relevant btree updates */
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_nr_pending(c));
+
ret = 0;
-out:
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
+err:
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&k, c);
+
+ BUG_ON(ret == -EINTR);
return ret;
-err:
- bch2_btree_iter_unlock(&iter);
- goto out;
}
int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
projects / bcachefs-tools-debian / blobdiff