]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/move.c
Merge https://github.com/YellowOnion/bcachefs-tools
[bcachefs-tools-debian] / libbcachefs / move.c
index 115433e7b9336b9f8e71a46b2f7729cd7a0128ca..64e39c10e34b91623d01ea0cd84424df10822c21 100644 (file)
@@ -8,11 +8,13 @@
 #include "btree_update_interior.h"
 #include "buckets.h"
 #include "disk_groups.h"
+#include "ec.h"
 #include "inode.h"
 #include "io.h"
 #include "journal_reclaim.h"
 #include "move.h"
 #include "replicas.h"
+#include "subvolume.h"
 #include "super-io.h"
 #include "keylist.h"
 
@@ -53,13 +55,89 @@ struct moving_context {
        wait_queue_head_t       wait;
 };
 
+static int insert_snapshot_whiteouts(struct btree_trans *trans,
+                                    enum btree_id id,
+                                    struct bpos old_pos,
+                                    struct bpos new_pos)
+{
+       struct bch_fs *c = trans->c;
+       struct btree_iter iter, update_iter;
+       struct bkey_s_c k;
+       struct snapshots_seen s;
+       int ret;
+
+       if (!btree_type_has_snapshots(id))
+               return 0;
+
+       snapshots_seen_init(&s);
+
+       if (!bkey_cmp(old_pos, new_pos))
+               return 0;
+
+       if (!snapshot_t(c, old_pos.snapshot)->children[0])
+               return 0;
+
+       bch2_trans_iter_init(trans, &iter, id, old_pos,
+                            BTREE_ITER_NOT_EXTENTS|
+                            BTREE_ITER_ALL_SNAPSHOTS);
+       while (1) {
+next:
+               k = bch2_btree_iter_prev(&iter);
+               ret = bkey_err(k);
+               if (ret)
+                       break;
+
+               if (bkey_cmp(old_pos, k.k->p))
+                       break;
+
+               if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot)) {
+                       struct bkey_i *update;
+                       size_t i;
+
+                       for (i = 0; i < s.nr; i++)
+                               if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, s.d[i]))
+                                       goto next;
+
+                       update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
+
+                       ret = PTR_ERR_OR_ZERO(update);
+                       if (ret)
+                               break;
+
+                       bkey_init(&update->k);
+                       update->k.p = new_pos;
+                       update->k.p.snapshot = k.k->p.snapshot;
+
+                       bch2_trans_iter_init(trans, &update_iter, id, update->k.p,
+                                            BTREE_ITER_NOT_EXTENTS|
+                                            BTREE_ITER_ALL_SNAPSHOTS|
+                                            BTREE_ITER_INTENT);
+                       ret   = bch2_btree_iter_traverse(&update_iter) ?:
+                               bch2_trans_update(trans, &update_iter, update,
+                                         BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+                       bch2_trans_iter_exit(trans, &update_iter);
+                       if (ret)
+                               break;
+
+                       ret = snapshots_seen_add(c, &s, k.k->p.snapshot);
+                       if (ret)
+                               break;
+               }
+       }
+       bch2_trans_iter_exit(trans, &iter);
+       kfree(s.d);
+
+       return ret;
+}
+
 static int bch2_migrate_index_update(struct bch_write_op *op)
 {
        struct bch_fs *c = op->c;
        struct btree_trans trans;
-       struct btree_iter *iter;
+       struct btree_iter iter;
        struct migrate_write *m =
                container_of(op, struct migrate_write, op);
+       struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
        struct keylist *keys = &op->insert_keys;
        struct bkey_buf _new, _insert;
        int ret = 0;
@@ -68,11 +146,11 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
        bch2_bkey_buf_init(&_insert);
        bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
 
-       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+       bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
 
-       iter = bch2_trans_get_iter(&trans, m->btree_id,
-                                  bkey_start_pos(&bch2_keylist_front(keys)->k),
-                                  BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+       bch2_trans_iter_init(&trans, &iter, m->btree_id,
+                            bkey_start_pos(&bch2_keylist_front(keys)->k),
+                            BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 
        while (1) {
                struct bkey_s_c k;
@@ -80,13 +158,14 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
                struct bkey_i_extent *new;
                const union bch_extent_entry *entry;
                struct extent_ptr_decoded p;
+               struct bpos next_pos;
                bool did_work = false;
-               bool extending = false, should_check_enospc;
+               bool should_check_enospc;
                s64 i_sectors_delta = 0, disk_sectors_delta = 0;
 
-               bch2_trans_reset(&trans, 0);
+               bch2_trans_begin(&trans);
 
-               k = bch2_btree_iter_peek_slot(iter);
+               k = bch2_btree_iter_peek_slot(&iter);
                ret = bkey_err(k);
                if (ret)
                        goto err;
@@ -102,9 +181,9 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
 
                bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
                new = bkey_i_to_extent(_new.k);
-               bch2_cut_front(iter->pos, &new->k_i);
+               bch2_cut_front(iter.pos, &new->k_i);
 
-               bch2_cut_front(iter->pos,       insert);
+               bch2_cut_front(iter.pos,        insert);
                bch2_cut_back(new->k.p,         insert);
                bch2_cut_back(insert->k.p,      &new->k_i);
 
@@ -119,7 +198,7 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
                                extent_for_each_ptr(extent_i_to_s(new), new_ptr)
                                        new_ptr->cached = true;
 
-                       bch2_bkey_drop_ptr(bkey_i_to_s(insert), old_ptr);
+                       __bch2_bkey_drop_ptr(bkey_i_to_s(insert), old_ptr);
                }
 
                extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
@@ -146,8 +225,7 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
                                               op->opts.background_target,
                                               op->opts.data_replicas);
 
-               ret = bch2_sum_sector_overwrites(&trans, iter, insert,
-                                                &extending,
+               ret = bch2_sum_sector_overwrites(&trans, &iter, insert,
                                                 &should_check_enospc,
                                                 &i_sectors_delta,
                                                 &disk_sectors_delta);
@@ -163,21 +241,29 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
                                goto out;
                }
 
-               bch2_trans_update(&trans, iter, insert, 0);
+               next_pos = insert->k.p;
 
-               ret = bch2_trans_commit(&trans, &op->res,
+               ret   = insert_snapshot_whiteouts(&trans, m->btree_id,
+                                                 k.k->p, insert->k.p) ?:
+                       bch2_trans_update(&trans, &iter, insert,
+                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
+                       bch2_trans_commit(&trans, &op->res,
                                op_journal_seq(op),
                                BTREE_INSERT_NOFAIL|
                                m->data_opts.btree_insert_flags);
-err:
-               if (!ret)
+               if (!ret) {
+                       bch2_btree_iter_set_pos(&iter, next_pos);
                        atomic_long_inc(&c->extent_migrate_done);
+                       if (ec_ob)
+                               bch2_ob_add_backpointer(c, ec_ob, &insert->k);
+               }
+err:
                if (ret == -EINTR)
                        ret = 0;
                if (ret)
                        break;
 next:
-               while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
+               while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
                        bch2_keylist_pop_front(keys);
                        if (bch2_keylist_empty(keys))
                                goto out;
@@ -185,17 +271,18 @@ next:
                continue;
 nomatch:
                if (m->ctxt) {
-                       BUG_ON(k.k->p.offset <= iter->pos.offset);
+                       BUG_ON(k.k->p.offset <= iter.pos.offset);
                        atomic64_inc(&m->ctxt->stats->keys_raced);
-                       atomic64_add(k.k->p.offset - iter->pos.offset,
+                       atomic64_add(k.k->p.offset - iter.pos.offset,
                                     &m->ctxt->stats->sectors_raced);
                }
                atomic_long_inc(&c->extent_migrate_raced);
                trace_move_race(&new->k);
-               bch2_btree_iter_next_slot(iter);
+               bch2_btree_iter_advance(&iter);
                goto next;
        }
 out:
+       bch2_trans_iter_exit(&trans, &iter);
        bch2_trans_exit(&trans);
        bch2_bkey_buf_exit(&_insert, c);
        bch2_bkey_buf_exit(&_new, c);
@@ -209,18 +296,13 @@ void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
        BUG_ON(!m->op.wbio.bio.bi_vcnt);
 
        m->ptr          = rbio->pick.ptr;
-       m->offset       = rbio->pos.offset - rbio->pick.crc.offset;
+       m->offset       = rbio->data_pos.offset - rbio->pick.crc.offset;
        m->op.devs_have = rbio->devs_have;
-       m->op.pos       = rbio->pos;
+       m->op.pos       = rbio->data_pos;
        m->op.version   = rbio->version;
        m->op.crc       = rbio->pick.crc;
        m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 
-       if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
-               m->op.nonce     = m->op.crc.nonce + m->op.crc.offset;
-               m->op.csum_type = m->op.crc.csum_type;
-       }
-
        if (m->data_cmd == DATA_REWRITE)
                bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
 }
@@ -235,6 +317,7 @@ int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 {
        struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
        const union bch_extent_entry *entry;
+       struct bch_extent_crc_unpacked crc;
        struct extent_ptr_decoded p;
        int ret;
 
@@ -255,6 +338,18 @@ int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
        m->op.target    = data_opts.target,
        m->op.write_point = wp;
 
+       /*
+        * op->csum_type is normally initialized from the fs/file's current
+        * options - but if an extent is encrypted, we require that it stays
+        * encrypted:
+        */
+       bkey_for_each_crc(k.k, ptrs, crc, entry)
+               if (bch2_csum_type_is_encryption(crc.csum_type)) {
+                       m->op.nonce     = crc.nonce + crc.offset;
+                       m->op.csum_type = crc.csum_type;
+                       break;
+               }
+
        if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE) {
                m->op.alloc_reserve = RESERVE_MOVINGGC;
                m->op.flags |= BCH_WRITE_ALLOC_NOWAIT;
@@ -493,7 +588,9 @@ static int bch2_move_extent(struct btree_trans *trans,
         * ctxt when doing wakeup
         */
        closure_get(&ctxt->cl);
-       bch2_read_extent(trans, &io->rbio, k, 0,
+       bch2_read_extent(trans, &io->rbio,
+                        bkey_start_pos(k.k),
+                        btree_id, k, 0,
                         BCH_READ_NODECODE|
                         BCH_READ_LAST_FRAGMENT);
        return 0;
@@ -506,6 +603,37 @@ err:
        return ret;
 }
 
+static int lookup_inode(struct btree_trans *trans, struct bpos pos,
+                       struct bch_inode_unpacked *inode)
+{
+       struct btree_iter iter;
+       struct bkey_s_c k;
+       int ret;
+
+       bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
+                            BTREE_ITER_ALL_SNAPSHOTS);
+       k = bch2_btree_iter_peek(&iter);
+       ret = bkey_err(k);
+       if (ret)
+               goto err;
+
+       if (!k.k || bkey_cmp(k.k->p, pos)) {
+               ret = -ENOENT;
+               goto err;
+       }
+
+       ret = bkey_is_inode(k.k) ? 0 : -EIO;
+       if (ret)
+               goto err;
+
+       ret = bch2_inode_unpack(k, inode);
+       if (ret)
+               goto err;
+err:
+       bch2_trans_iter_exit(trans, &iter);
+       return ret;
+}
+
 static int __bch2_move_data(struct bch_fs *c,
                struct moving_context *ctxt,
                struct bch_ratelimit *rate,
@@ -520,7 +648,7 @@ static int __bch2_move_data(struct bch_fs *c,
        struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
        struct bkey_buf sk;
        struct btree_trans trans;
-       struct btree_iter *iter;
+       struct btree_iter iter;
        struct bkey_s_c k;
        struct data_opts data_opts;
        enum data_cmd data_cmd;
@@ -532,10 +660,11 @@ static int __bch2_move_data(struct bch_fs *c,
 
        stats->data_type = BCH_DATA_user;
        stats->btree_id = btree_id;
-       stats->pos      = POS_MIN;
+       stats->pos      = start;
 
-       iter = bch2_trans_get_iter(&trans, btree_id, start,
-                                  BTREE_ITER_PREFETCH);
+       bch2_trans_iter_init(&trans, &iter, btree_id, start,
+                            BTREE_ITER_PREFETCH|
+                            BTREE_ITER_ALL_SNAPSHOTS);
 
        if (rate)
                bch2_ratelimit_reset(rate);
@@ -563,10 +692,12 @@ static int __bch2_move_data(struct bch_fs *c,
                                try_to_freeze();
                        }
                } while (delay);
-peek:
-               k = bch2_btree_iter_peek(iter);
 
-               stats->pos = iter->pos;
+               bch2_trans_begin(&trans);
+
+               k = bch2_btree_iter_peek(&iter);
+
+               stats->pos = iter.pos;
 
                if (!k.k)
                        break;
@@ -583,14 +714,18 @@ peek:
                    cur_inum != k.k->p.inode) {
                        struct bch_inode_unpacked inode;
 
-                       /* don't hold btree locks while looking up inode: */
-                       bch2_trans_unlock(&trans);
-
                        io_opts = bch2_opts_to_inode_opts(c->opts);
-                       if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
+
+                       ret = lookup_inode(&trans,
+                                       SPOS(0, k.k->p.inode, k.k->p.snapshot),
+                                       &inode);
+                       if (ret == -EINTR)
+                               continue;
+
+                       if (!ret)
                                bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
+
                        cur_inum = k.k->p.inode;
-                       goto peek;
                }
 
                switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
@@ -615,8 +750,7 @@ peek:
                                        data_cmd, data_opts);
                if (ret2) {
                        if (ret2 == -EINTR) {
-                               bch2_trans_reset(&trans, 0);
-                               bch2_trans_cond_resched(&trans);
+                               bch2_trans_begin(&trans);
                                continue;
                        }
 
@@ -636,37 +770,77 @@ next:
                atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
                             &stats->sectors_seen);
 next_nondata:
-               bch2_btree_iter_next(iter);
-               bch2_trans_cond_resched(&trans);
+               bch2_btree_iter_advance(&iter);
        }
 out:
-       ret = bch2_trans_exit(&trans) ?: ret;
+
+       bch2_trans_iter_exit(&trans, &iter);
+       bch2_trans_exit(&trans);
        bch2_bkey_buf_exit(&sk, c);
 
        return ret;
 }
 
+inline void bch_move_stats_init(struct bch_move_stats *stats, char *name)
+{
+       memset(stats, 0, sizeof(*stats));
+
+       scnprintf(stats->name, sizeof(stats->name),
+                       "%s", name);
+}
+
+static inline void progress_list_add(struct bch_fs *c,
+                                    struct bch_move_stats *stats)
+{
+       mutex_lock(&c->data_progress_lock);
+       list_add(&stats->list, &c->data_progress_list);
+       mutex_unlock(&c->data_progress_lock);
+}
+
+static inline void progress_list_del(struct bch_fs *c,
+                                    struct bch_move_stats *stats)
+{
+       mutex_lock(&c->data_progress_lock);
+       list_del(&stats->list);
+       mutex_unlock(&c->data_progress_lock);
+}
+
 int bch2_move_data(struct bch_fs *c,
+                  enum btree_id start_btree_id, struct bpos start_pos,
+                  enum btree_id end_btree_id,   struct bpos end_pos,
                   struct bch_ratelimit *rate,
                   struct write_point_specifier wp,
-                  struct bpos start,
-                  struct bpos end,
                   move_pred_fn pred, void *arg,
                   struct bch_move_stats *stats)
 {
        struct moving_context ctxt = { .stats = stats };
+       enum btree_id id;
        int ret;
 
+       progress_list_add(c, stats);
        closure_init_stack(&ctxt.cl);
        INIT_LIST_HEAD(&ctxt.reads);
        init_waitqueue_head(&ctxt.wait);
 
        stats->data_type = BCH_DATA_user;
 
-       ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
-                                pred, arg, stats, BTREE_ID_extents) ?:
-               __bch2_move_data(c, &ctxt, rate, wp, start, end,
-                                pred, arg, stats, BTREE_ID_reflink);
+       for (id = start_btree_id;
+            id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
+            id++) {
+               stats->btree_id = id;
+
+               if (id != BTREE_ID_extents &&
+                   id != BTREE_ID_reflink)
+                       continue;
+
+               ret = __bch2_move_data(c, &ctxt, rate, wp,
+                                      id == start_btree_id ? start_pos : POS_MIN,
+                                      id == end_btree_id   ? end_pos   : POS_MAX,
+                                      pred, arg, stats, id);
+               if (ret)
+                       break;
+       }
+
 
        move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
        closure_sync(&ctxt.cl);
@@ -677,37 +851,57 @@ int bch2_move_data(struct bch_fs *c,
                        atomic64_read(&stats->sectors_moved),
                        atomic64_read(&stats->keys_moved));
 
+       progress_list_del(c, stats);
        return ret;
 }
 
+typedef enum data_cmd (*move_btree_pred)(struct bch_fs *, void *,
+                                        struct btree *, struct bch_io_opts *,
+                                        struct data_opts *);
+
 static int bch2_move_btree(struct bch_fs *c,
-                          move_pred_fn pred,
-                          void *arg,
+                          enum btree_id start_btree_id, struct bpos start_pos,
+                          enum btree_id end_btree_id,   struct bpos end_pos,
+                          move_btree_pred pred, void *arg,
                           struct bch_move_stats *stats)
 {
+       bool kthread = (current->flags & PF_KTHREAD) != 0;
        struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
        struct btree_trans trans;
-       struct btree_iter *iter;
+       struct btree_iter iter;
        struct btree *b;
-       unsigned id;
+       enum btree_id id;
        struct data_opts data_opts;
        enum data_cmd cmd;
        int ret = 0;
 
        bch2_trans_init(&trans, c, 0, 0);
+       progress_list_add(c, stats);
 
        stats->data_type = BCH_DATA_btree;
 
-       for (id = 0; id < BTREE_ID_NR; id++) {
+       for (id = start_btree_id;
+            id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
+            id++) {
                stats->btree_id = id;
 
-               for_each_btree_node(&trans, iter, id, POS_MIN,
-                                   BTREE_ITER_PREFETCH, b) {
-                       stats->pos = iter->pos;
+               bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
+                                         BTREE_ITER_PREFETCH);
+retry:
+               ret = 0;
+               while (bch2_trans_begin(&trans),
+                      (b = bch2_btree_iter_peek_node(&iter)) &&
+                      !(ret = PTR_ERR_OR_ZERO(b))) {
+                       if (kthread && kthread_should_stop())
+                               break;
 
-                       switch ((cmd = pred(c, arg,
-                                           bkey_i_to_s_c(&b->key),
-                                           &io_opts, &data_opts))) {
+                       if ((cmp_int(id, end_btree_id) ?:
+                            bpos_cmp(b->key.k.p, end_pos)) > 0)
+                               break;
+
+                       stats->pos = iter.pos;
+
+                       switch ((cmd = pred(c, arg, b, &io_opts, &data_opts))) {
                        case DATA_SKIP:
                                goto next;
                        case DATA_SCRUB:
@@ -719,17 +913,33 @@ static int bch2_move_btree(struct bch_fs *c,
                                BUG();
                        }
 
-                       ret = bch2_btree_node_rewrite(c, iter,
-                                       b->data->keys.seq, 0) ?: ret;
+                       ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
+                       if (ret == -EINTR)
+                               continue;
+                       if (ret)
+                               break;
 next:
-                       bch2_trans_cond_resched(&trans);
+                       bch2_btree_iter_next_node(&iter);
                }
+               if (ret == -EINTR)
+                       goto retry;
 
-               ret = bch2_trans_iter_free(&trans, iter) ?: ret;
+               bch2_trans_iter_exit(&trans, &iter);
+
+               if (kthread && kthread_should_stop())
+                       break;
        }
 
        bch2_trans_exit(&trans);
 
+       if (ret)
+               bch_err(c, "error %i in bch2_move_btree", ret);
+
+       /* flush relevant btree updates */
+       closure_wait_event(&c->btree_interior_update_wait,
+                          !bch2_btree_interior_updates_nr_pending(c));
+
+       progress_list_del(c, stats);
        return ret;
 }
 
@@ -749,16 +959,9 @@ static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
                                      struct data_opts *data_opts)
 {
        unsigned nr_good = bch2_bkey_durability(c, k);
-       unsigned replicas = 0;
-
-       switch (k.k->type) {
-       case KEY_TYPE_btree_ptr:
-               replicas = c->opts.metadata_replicas;
-               break;
-       case KEY_TYPE_extent:
-               replicas = io_opts->data_replicas;
-               break;
-       }
+       unsigned replicas = bkey_is_btree_ptr(k.k)
+               ? c->opts.metadata_replicas
+               : io_opts->data_replicas;
 
        if (!nr_good || nr_good >= replicas)
                return DATA_SKIP;
@@ -786,6 +989,83 @@ static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
        return DATA_REWRITE;
 }
 
+static enum data_cmd rereplicate_btree_pred(struct bch_fs *c, void *arg,
+                                           struct btree *b,
+                                           struct bch_io_opts *io_opts,
+                                           struct data_opts *data_opts)
+{
+       return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
+}
+
+static enum data_cmd migrate_btree_pred(struct bch_fs *c, void *arg,
+                                       struct btree *b,
+                                       struct bch_io_opts *io_opts,
+                                       struct data_opts *data_opts)
+{
+       return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
+}
+
+static bool bformat_needs_redo(struct bkey_format *f)
+{
+       unsigned i;
+
+       for (i = 0; i < f->nr_fields; i++) {
+               unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
+               u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
+               u64 field_offset = le64_to_cpu(f->field_offset[i]);
+
+               if (f->bits_per_field[i] > unpacked_bits)
+                       return true;
+
+               if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
+                       return true;
+
+               if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
+                    unpacked_mask) <
+                   field_offset)
+                       return true;
+       }
+
+       return false;
+}
+
+static enum data_cmd rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
+                                           struct btree *b,
+                                           struct bch_io_opts *io_opts,
+                                           struct data_opts *data_opts)
+{
+       if (b->version_ondisk != c->sb.version ||
+           btree_node_need_rewrite(b) ||
+           bformat_needs_redo(&b->format)) {
+               data_opts->target               = 0;
+               data_opts->nr_replicas          = 1;
+               data_opts->btree_insert_flags   = 0;
+               return DATA_REWRITE;
+       }
+
+       return DATA_SKIP;
+}
+
+int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
+{
+       int ret;
+
+       ret = bch2_move_btree(c,
+                             0,                POS_MIN,
+                             BTREE_ID_NR,      SPOS_MAX,
+                             rewrite_old_nodes_pred, c, stats);
+       if (!ret) {
+               mutex_lock(&c->sb_lock);
+               c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
+               c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
+               c->disk_sb.sb->version_min = c->disk_sb.sb->version;
+               bch2_write_super(c);
+               mutex_unlock(&c->sb_lock);
+       }
+
+       return ret;
+}
+
 int bch2_data_job(struct bch_fs *c,
                  struct bch_move_stats *stats,
                  struct bch_ioctl_data op)
@@ -794,20 +1074,20 @@ int bch2_data_job(struct bch_fs *c,
 
        switch (op.op) {
        case BCH_DATA_OP_REREPLICATE:
+               bch_move_stats_init(stats, "rereplicate");
                stats->data_type = BCH_DATA_journal;
                ret = bch2_journal_flush_device_pins(&c->journal, -1);
 
-               ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
-
-               closure_wait_event(&c->btree_interior_update_wait,
-                                  !bch2_btree_interior_updates_nr_pending(c));
-
+               ret = bch2_move_btree(c,
+                                     op.start_btree,   op.start_pos,
+                                     op.end_btree,     op.end_pos,
+                                     rereplicate_btree_pred, c, stats) ?: ret;
                ret = bch2_replicas_gc2(c) ?: ret;
 
-               ret = bch2_move_data(c, NULL,
-                                    writepoint_hashed((unsigned long) current),
-                                    op.start,
-                                    op.end,
+               ret = bch2_move_data(c,
+                                    op.start_btree,    op.start_pos,
+                                    op.end_btree,      op.end_pos,
+                                    NULL, writepoint_hashed((unsigned long) current),
                                     rereplicate_pred, c, stats) ?: ret;
                ret = bch2_replicas_gc2(c) ?: ret;
                break;
@@ -815,19 +1095,27 @@ int bch2_data_job(struct bch_fs *c,
                if (op.migrate.dev >= c->sb.nr_devices)
                        return -EINVAL;
 
+               bch_move_stats_init(stats, "migrate");
                stats->data_type = BCH_DATA_journal;
                ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 
-               ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
+               ret = bch2_move_btree(c,
+                                     op.start_btree,   op.start_pos,
+                                     op.end_btree,     op.end_pos,
+                                     migrate_btree_pred, &op, stats) ?: ret;
                ret = bch2_replicas_gc2(c) ?: ret;
 
-               ret = bch2_move_data(c, NULL,
-                                    writepoint_hashed((unsigned long) current),
-                                    op.start,
-                                    op.end,
+               ret = bch2_move_data(c,
+                                    op.start_btree,    op.start_pos,
+                                    op.end_btree,      op.end_pos,
+                                    NULL, writepoint_hashed((unsigned long) current),
                                     migrate_pred, &op, stats) ?: ret;
                ret = bch2_replicas_gc2(c) ?: ret;
                break;
+       case BCH_DATA_OP_REWRITE_OLD_NODES:
+               bch_move_stats_init(stats, "rewrite_old_nodes");
+               ret = bch2_scan_old_btree_nodes(c, stats);
+               break;
        default:
                ret = -EINVAL;
        }