]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/fsck.c
Update bcachefs sources to 98a1bff393 bcachefs: Topology repair fixes
[bcachefs-tools-debian] / libbcachefs / fsck.c
index 58d42734c252d3311ffa5c7498b9f1158a5f0076..d2b155f07fc10a185dde5fbd80ca3cfa02db36a6 100644 (file)
@@ -3,6 +3,7 @@
 #include "bcachefs.h"
 #include "bkey_buf.h"
 #include "btree_update.h"
+#include "darray.h"
 #include "dirent.h"
 #include "error.h"
 #include "fs-common.h"
@@ -94,12 +95,6 @@ err:
 
 }
 
-static int snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
-                                 u32 *subvol)
-{
-       return lockrestart_do(trans, __snapshot_lookup_subvol(trans, snapshot, subvol));
-}
-
 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
                           u32 *snapshot, u64 *inum)
 {
@@ -119,6 +114,35 @@ static int subvol_lookup(struct btree_trans *trans, u32 subvol,
        return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
 }
 
+static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
+                             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(0, inode_nr),
+                            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(0, inode_nr))) {
+               ret = -ENOENT;
+               goto err;
+       }
+
+       ret = bch2_inode_unpack(k, inode);
+err:
+       if (ret && ret != -EINTR)
+               bch_err(trans->c, "error %i fetching inode %llu",
+                       ret, inode_nr);
+       bch2_trans_iter_exit(trans, &iter);
+       return ret;
+}
+
 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
                          struct bch_inode_unpacked *inode,
                          u32 *snapshot)
@@ -134,12 +158,15 @@ static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
        if (ret)
                goto err;
 
-       ret = k.k->type == KEY_TYPE_inode
-               ? bch2_inode_unpack(bkey_s_c_to_inode(k), inode)
+       ret = bkey_is_inode(k.k)
+               ? bch2_inode_unpack(k, inode)
                : -ENOENT;
        if (!ret)
                *snapshot = iter.pos.snapshot;
 err:
+       if (ret && ret != -EINTR)
+               bch_err(trans->c, "error %i fetching inode %llu:%u",
+                       ret, inode_nr, *snapshot);
        bch2_trans_iter_exit(trans, &iter);
        return ret;
 }
@@ -172,15 +199,6 @@ static int __lookup_dirent(struct btree_trans *trans,
        return 0;
 }
 
-static int lookup_dirent(struct btree_trans *trans,
-                        struct bch_hash_info hash_info,
-                        subvol_inum dir, struct qstr *name,
-                        u64 *target, unsigned *type)
-{
-       return lockrestart_do(trans,
-               __lookup_dirent(trans, hash_info, dir, name, target, type));
-}
-
 static int __write_inode(struct btree_trans *trans,
                         struct bch_inode_unpacked *inode,
                         u32 snapshot)
@@ -244,7 +262,7 @@ retry:
        if (ret)
                goto err;
 
-       if (k.k->type != KEY_TYPE_inode) {
+       if (!bkey_is_inode(k.k)) {
                bch2_fs_inconsistent(trans->c,
                                     "inode %llu:%u not found when deleting",
                                     inum, snapshot);
@@ -252,11 +270,11 @@ retry:
                goto err;
        }
 
-       bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u);
+       bch2_inode_unpack(k, &inode_u);
 
        /* Subvolume root? */
        if (inode_u.bi_subvol) {
-               ret = bch2_subvolume_delete(trans, inode_u.bi_subvol, -1);
+               ret = bch2_subvolume_delete(trans, inode_u.bi_subvol);
                if (ret)
                        goto err;
        }
@@ -284,7 +302,7 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
        struct bch_hash_info dir_hash_info;
        int ret;
 
-       ret = lookup_inode(trans, pos.inode, &dir_inode, NULL);
+       ret = lookup_first_inode(trans, pos.inode, &dir_inode);
        if (ret)
                return ret;
 
@@ -298,17 +316,6 @@ static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
        return ret;
 }
 
-static int remove_dirent(struct btree_trans *trans, struct bpos pos)
-{
-       int ret = __bch2_trans_do(trans, NULL, NULL,
-                                 BTREE_INSERT_NOFAIL|
-                                 BTREE_INSERT_LAZY_RW,
-                                 __remove_dirent(trans, pos));
-       if (ret)
-               bch_err(trans->c, "remove_dirent: err %i deleting dirent", ret);
-       return ret;
-}
-
 /* Get lost+found, create if it doesn't exist: */
 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
                            struct bch_inode_unpacked *lostfound)
@@ -323,65 +330,52 @@ static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
        u32 snapshot;
        int ret;
 
-       ret = subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
+       ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
        if (ret)
                return ret;
 
-       ret = lookup_inode(trans, root_inum.inum, &root, &snapshot);
-       if (ret) {
-               bch_err(c, "error fetching subvol root: %i", ret);
+       ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
+       if (ret)
                return ret;
-       }
 
        root_hash_info = bch2_hash_info_init(c, &root);
 
-       ret = lookup_dirent(trans, root_hash_info, root_inum,
+       ret = __lookup_dirent(trans, root_hash_info, root_inum,
                            &lostfound_str, &inum, &d_type);
        if (ret == -ENOENT) {
                bch_notice(c, "creating lost+found");
                goto create_lostfound;
        }
 
-       if (ret) {
+       if (ret && ret != -EINTR)
                bch_err(c, "error looking up lost+found: %i", ret);
+       if (ret)
                return ret;
-       }
 
        if (d_type != DT_DIR) {
                bch_err(c, "error looking up lost+found: not a directory");
                return ret;
-
        }
 
-       ret = lookup_inode(trans, inum, lostfound, &snapshot);
-       if (ret && ret != -ENOENT) {
-               /*
-                * The check_dirents pass has already run, dangling dirents
-                * shouldn't exist here:
-                */
-               bch_err(c, "error looking up lost+found: %i", ret);
-               return ret;
-       }
+       /*
+        * The check_dirents pass has already run, dangling dirents
+        * shouldn't exist here:
+        */
+       return __lookup_inode(trans, inum, lostfound, &snapshot);
 
-       if (ret == -ENOENT) {
 create_lostfound:
-               bch2_inode_init_early(c, lostfound);
-
-               ret = __bch2_trans_do(trans, NULL, NULL,
-                                     BTREE_INSERT_NOFAIL|
-                                     BTREE_INSERT_LAZY_RW,
-                       bch2_create_trans(trans, root_inum, &root,
-                                         lostfound, &lostfound_str,
-                                         0, 0, S_IFDIR|0700, 0, NULL, NULL,
-                                         (subvol_inum) { }, 0));
-               if (ret)
-                       bch_err(c, "error creating lost+found: %i", ret);
-       }
-
-       return 0;
+       bch2_inode_init_early(c, lostfound);
+
+       ret = bch2_create_trans(trans, root_inum, &root,
+                               lostfound, &lostfound_str,
+                               0, 0, S_IFDIR|0700, 0, NULL, NULL,
+                               (subvol_inum) { }, 0);
+       if (ret && ret != -EINTR)
+               bch_err(c, "error creating lost+found: %i", ret);
+       return ret;
 }
 
-static int reattach_inode(struct btree_trans *trans,
+static int __reattach_inode(struct btree_trans *trans,
                          struct bch_inode_unpacked *inode,
                          u32 inode_snapshot)
 {
@@ -393,7 +387,7 @@ static int reattach_inode(struct btree_trans *trans,
        u32 subvol;
        int ret;
 
-       ret = snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
+       ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
        if (ret)
                return ret;
 
@@ -404,7 +398,7 @@ static int reattach_inode(struct btree_trans *trans,
        if (S_ISDIR(inode->bi_mode)) {
                lostfound.bi_nlink++;
 
-               ret = write_inode(trans, &lostfound, U32_MAX);
+               ret = __write_inode(trans, &lostfound, U32_MAX);
                if (ret)
                        return ret;
        }
@@ -414,26 +408,39 @@ static int reattach_inode(struct btree_trans *trans,
        snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
        name = (struct qstr) QSTR(name_buf);
 
-       ret = __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                       bch2_dirent_create(trans,
-                                          (subvol_inum) {
-                                               .subvol = subvol,
-                                               .inum = lostfound.bi_inum,
-                                          },
-                                          &dir_hash,
-                                          mode_to_type(inode->bi_mode),
-                                          &name, inode->bi_inum, &dir_offset,
-                                          BCH_HASH_SET_MUST_CREATE));
+       ret = bch2_dirent_create(trans,
+                                (subvol_inum) {
+                                       .subvol = subvol,
+                                       .inum = lostfound.bi_inum,
+                                },
+                                &dir_hash,
+                                inode_d_type(inode),
+                                &name, inode->bi_inum, &dir_offset,
+                                BCH_HASH_SET_MUST_CREATE);
+       if (ret)
+               return ret;
+
+       inode->bi_dir           = lostfound.bi_inum;
+       inode->bi_dir_offset    = dir_offset;
+
+       return __write_inode(trans, inode, inode_snapshot);
+}
+
+static int reattach_inode(struct btree_trans *trans,
+                         struct bch_inode_unpacked *inode,
+                         u32 inode_snapshot)
+{
+       int ret = __bch2_trans_do(trans, NULL, NULL,
+                                 BTREE_INSERT_LAZY_RW|
+                                 BTREE_INSERT_NOFAIL,
+                       __reattach_inode(trans, inode, inode_snapshot));
        if (ret) {
                bch_err(trans->c, "error %i reattaching inode %llu",
                        ret, inode->bi_inum);
                return ret;
        }
 
-       inode->bi_dir           = lostfound.bi_inum;
-       inode->bi_dir_offset    = dir_offset;
-
-       return write_inode(trans, inode, inode_snapshot);
+       return ret;
 }
 
 static int remove_backpointer(struct btree_trans *trans,
@@ -454,7 +461,7 @@ static int remove_backpointer(struct btree_trans *trans,
                goto out;
        }
 
-       ret = remove_dirent(trans, k.k->p);
+       ret = __remove_dirent(trans, k.k->p);
 out:
        bch2_trans_iter_exit(trans, &iter);
        return ret;
@@ -465,11 +472,11 @@ static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s, str
        pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
 
        if (bkey_cmp(s->pos, pos))
-               s->nr = 0;
+               s->ids.nr = 0;
        s->pos = pos;
 
        /* Might get called multiple times due to lock restarts */
-       if (s->nr && s->d[s->nr - 1] == pos.snapshot)
+       if (s->ids.nr && s->ids.data[s->ids.nr - 1] == pos.snapshot)
                return 0;
 
        return snapshots_seen_add(c, s, pos.snapshot);
@@ -492,7 +499,7 @@ static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *see
        ancestor        = snapshot_t(c, ancestor)->equiv;
 
        /* @ancestor should be the snapshot most recently added to @seen */
-       BUG_ON(!seen->nr || seen->d[seen->nr - 1] != ancestor);
+       BUG_ON(!seen->ids.nr || seen->ids.data[seen->ids.nr - 1] != ancestor);
        BUG_ON(seen->pos.snapshot != ancestor);
 
        if (id == ancestor)
@@ -501,11 +508,11 @@ static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *see
        if (!bch2_snapshot_is_ancestor(c, id, ancestor))
                return false;
 
-       for (i = seen->nr - 2;
-            i >= 0 && seen->d[i] >= id;
+       for (i = seen->ids.nr - 2;
+            i >= 0 && seen->ids.data[i] >= id;
             --i)
-               if (bch2_snapshot_is_ancestor(c, id, seen->d[i]) &&
-                   bch2_snapshot_is_ancestor(c, seen->d[i], ancestor))
+               if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i]) &&
+                   bch2_snapshot_is_ancestor(c, seen->ids.data[i], ancestor))
                        return false;
 
        return true;
@@ -531,26 +538,25 @@ static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 }
 
 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)      \
-       for (_i = (_w)->d; _i < (_w)->d + (_w)->nr && (_i)->snapshot <= (_snapshot); _i++)\
+       for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && (_i)->snapshot <= (_snapshot); _i++)\
                if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 
+struct inode_walker_entry {
+       struct bch_inode_unpacked inode;
+       u32                     snapshot;
+       u64                     count;
+};
+
 struct inode_walker {
        bool                            first_this_inode;
        u64                             cur_inum;
 
-       size_t                          nr;
-       size_t                          size;
-       struct inode_walker_entry {
-               struct bch_inode_unpacked inode;
-               u32                     snapshot;
-               u64                     count;
-       } *d;
+       DARRAY(struct inode_walker_entry) inodes;
 };
 
 static void inode_walker_exit(struct inode_walker *w)
 {
-       kfree(w->d);
-       w->d = NULL;
+       darray_exit(w->inodes);
 }
 
 static struct inode_walker inode_walker_init(void)
@@ -558,40 +564,17 @@ static struct inode_walker inode_walker_init(void)
        return (struct inode_walker) { 0, };
 }
 
-static int inode_walker_realloc(struct inode_walker *w)
-{
-       if (w->nr == w->size) {
-               size_t new_size = max_t(size_t, 8UL, w->size * 2);
-               void *d = krealloc(w->d, new_size * sizeof(w->d[0]),
-                                  GFP_KERNEL);
-               if (!d)
-                       return -ENOMEM;
-
-               w->d = d;
-               w->size = new_size;
-       }
-
-       return 0;
-}
-
 static int add_inode(struct bch_fs *c, struct inode_walker *w,
-                    struct bkey_s_c_inode inode)
+                    struct bkey_s_c inode)
 {
        struct bch_inode_unpacked u;
-       int ret;
-
-       ret = inode_walker_realloc(w);
-       if (ret)
-               return ret;
 
        BUG_ON(bch2_inode_unpack(inode, &u));
 
-       w->d[w->nr++] = (struct inode_walker_entry) {
+       return darray_push(w->inodes, ((struct inode_walker_entry) {
                .inode          = u,
                .snapshot       = snapshot_t(c, inode.k->p.snapshot)->equiv,
-       };
-
-       return 0;
+       }));
 }
 
 static int __walk_inode(struct btree_trans *trans,
@@ -610,15 +593,15 @@ static int __walk_inode(struct btree_trans *trans,
                goto lookup_snapshot;
        }
 
-       w->nr = 0;
+       w->inodes.nr = 0;
 
        for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
                           BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
                if (k.k->p.offset != pos.inode)
                        break;
 
-               if (k.k->type == KEY_TYPE_inode)
-                       add_inode(c, w, bkey_s_c_to_inode(k));
+               if (bkey_is_inode(k.k))
+                       add_inode(c, w, k);
        }
        bch2_trans_iter_exit(trans, &iter);
 
@@ -628,37 +611,30 @@ static int __walk_inode(struct btree_trans *trans,
        w->cur_inum             = pos.inode;
        w->first_this_inode     = true;
 lookup_snapshot:
-       for (i = 0; i < w->nr; i++)
-               if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->d[i].snapshot))
+       for (i = 0; i < w->inodes.nr; i++)
+               if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->inodes.data[i].snapshot))
                        goto found;
        return INT_MAX;
 found:
-       BUG_ON(pos.snapshot > w->d[i].snapshot);
+       BUG_ON(pos.snapshot > w->inodes.data[i].snapshot);
 
-       if (pos.snapshot != w->d[i].snapshot) {
+       if (pos.snapshot != w->inodes.data[i].snapshot) {
                ancestor_pos = i;
 
-               while (i && w->d[i - 1].snapshot > pos.snapshot)
+               while (i && w->inodes.data[i - 1].snapshot > pos.snapshot)
                        --i;
 
-               ret = inode_walker_realloc(w);
+               ret = darray_insert_item(w->inodes, i, w->inodes.data[ancestor_pos]);
                if (ret)
                        return ret;
 
-               array_insert_item(w->d, w->nr, i, w->d[ancestor_pos]);
-               w->d[i].snapshot = pos.snapshot;
-               w->d[i].count   = 0;
+               w->inodes.data[i].snapshot = pos.snapshot;
+               w->inodes.data[i].count = 0;
        }
 
        return i;
 }
 
-static int walk_inode(struct btree_trans *trans,
-                     struct inode_walker *w, struct bpos pos)
-{
-       return lockrestart_do(trans, __walk_inode(trans, w, pos));
-}
-
 static int __get_visible_inodes(struct btree_trans *trans,
                                struct inode_walker *w,
                                struct snapshots_seen *s,
@@ -669,18 +645,18 @@ static int __get_visible_inodes(struct btree_trans *trans,
        struct bkey_s_c k;
        int ret;
 
-       w->nr = 0;
+       w->inodes.nr = 0;
 
        for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
                           BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
                if (k.k->p.offset != inum)
                        break;
 
-               if (k.k->type != KEY_TYPE_inode)
+               if (!bkey_is_inode(k.k))
                        continue;
 
                if (ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot)) {
-                       add_inode(c, w, bkey_s_c_to_inode(k));
+                       add_inode(c, w, k);
                        if (k.k->p.snapshot >= s->pos.snapshot)
                                break;
                }
@@ -695,18 +671,16 @@ static int check_key_has_snapshot(struct btree_trans *trans,
                                  struct bkey_s_c k)
 {
        struct bch_fs *c = trans->c;
-       char buf[200];
+       struct printbuf buf = PRINTBUF;
        int ret = 0;
 
-       if (fsck_err_on(!snapshot_t(c, k.k->p.snapshot)->equiv, c,
+       if (mustfix_fsck_err_on(!snapshot_t(c, k.k->p.snapshot)->equiv, c,
                        "key in missing snapshot: %s",
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) {
-               ret = __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                       bch2_btree_delete_at(trans, iter,
-                                            BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
-               return ret ?: -EINTR;
-       }
+                       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+               ret = bch2_btree_delete_at(trans, iter,
+                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 fsck_err:
+       printbuf_exit(&buf);
        return ret;
 }
 
@@ -739,26 +713,6 @@ static int hash_redo_key(struct btree_trans *trans,
 #endif
 }
 
-static int fsck_hash_delete_at(struct btree_trans *trans,
-                              const struct bch_hash_desc desc,
-                              struct bch_hash_info *info,
-                              struct btree_iter *iter)
-{
-       int ret;
-retry:
-       ret   = bch2_hash_delete_at(trans, desc, info, iter, 0) ?:
-               bch2_trans_commit(trans, NULL, NULL,
-                                 BTREE_INSERT_NOFAIL|
-                                 BTREE_INSERT_LAZY_RW);
-       if (ret == -EINTR) {
-               ret = bch2_btree_iter_traverse(iter);
-               if (!ret)
-                       goto retry;
-       }
-
-       return ret;
-}
-
 static int hash_check_key(struct btree_trans *trans,
                          const struct bch_hash_desc desc,
                          struct bch_hash_info *hash_info,
@@ -766,7 +720,7 @@ static int hash_check_key(struct btree_trans *trans,
 {
        struct bch_fs *c = trans->c;
        struct btree_iter iter = { NULL };
-       char buf[200];
+       struct printbuf buf = PRINTBUF;
        struct bkey_s_c k;
        u64 hash;
        int ret = 0;
@@ -790,12 +744,10 @@ static int hash_check_key(struct btree_trans *trans,
                if (fsck_err_on(k.k->type == desc.key_type &&
                                !desc.cmp_bkey(k, hash_k), c,
                                "duplicate hash table keys:\n%s",
-                               (bch2_bkey_val_to_text(&PBUF(buf), c,
-                                                      hash_k), buf))) {
-                       ret = fsck_hash_delete_at(trans, desc, hash_info, k_iter);
-                       if (ret)
-                               return ret;
-                       ret = 1;
+                               (printbuf_reset(&buf),
+                                bch2_bkey_val_to_text(&buf, c, hash_k),
+                                buf.buf))) {
+                       ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
                        break;
                }
 
@@ -805,39 +757,75 @@ static int hash_check_key(struct btree_trans *trans,
                }
 
        }
+out:
        bch2_trans_iter_exit(trans, &iter);
+       printbuf_exit(&buf);
        return ret;
 bad_hash:
        if (fsck_err(c, "hash table key at wrong offset: btree %u inode %llu offset %llu, "
                     "hashed to %llu\n%s",
                     desc.btree_id, hash_k.k->p.inode, hash_k.k->p.offset, hash,
-                    (bch2_bkey_val_to_text(&PBUF(buf), c, hash_k), buf)) == FSCK_ERR_IGNORE)
+                    (printbuf_reset(&buf),
+                     bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf)) == FSCK_ERR_IGNORE)
                return 0;
 
-       ret = __bch2_trans_do(trans, NULL, NULL,
-                             BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
-               hash_redo_key(trans, desc, hash_info, k_iter, hash_k));
+       ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
        if (ret) {
                bch_err(c, "hash_redo_key err %i", ret);
                return ret;
        }
-       return -EINTR;
+       ret = -EINTR;
 fsck_err:
-       return ret;
+       goto out;
 }
 
 static int check_inode(struct btree_trans *trans,
                       struct btree_iter *iter,
                       struct bch_inode_unpacked *prev,
-                      struct bch_inode_unpacked u)
+                      bool full)
 {
        struct bch_fs *c = trans->c;
+       struct bkey_s_c k;
+       struct bch_inode_unpacked u;
        bool do_update = false;
-       int ret = 0;
+       int ret;
+
+       k = bch2_btree_iter_peek(iter);
+       if (!k.k)
+               return 0;
+
+       ret = bkey_err(k);
+       if (ret)
+               return ret;
+
+       ret = check_key_has_snapshot(trans, iter, k);
+       if (ret)
+               return ret < 0 ? ret : 0;
+
+       /*
+        * if snapshot id isn't a leaf node, skip it - deletion in
+        * particular is not atomic, so on the internal snapshot nodes
+        * we can see inodes marked for deletion after a clean shutdown
+        */
+       if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
+               return 0;
+
+       if (!bkey_is_inode(k.k))
+               return 0;
 
-       if (fsck_err_on(prev &&
-                       (prev->bi_hash_seed             != u.bi_hash_seed ||
-                        mode_to_type(prev->bi_mode) != mode_to_type(u.bi_mode)), c,
+       BUG_ON(bch2_inode_unpack(k, &u));
+
+       if (!full &&
+           !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
+                           BCH_INODE_I_SECTORS_DIRTY|
+                           BCH_INODE_UNLINKED)))
+               return 0;
+
+       if (prev->bi_inum != u.bi_inum)
+               *prev = u;
+
+       if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
+                       inode_d_type(prev)      != inode_d_type(&u), c,
                        "inodes in different snapshots don't match")) {
                bch_err(c, "repair not implemented yet");
                return -EINVAL;
@@ -932,58 +920,60 @@ static int check_inodes(struct bch_fs *c, bool full)
 {
        struct btree_trans trans;
        struct btree_iter iter;
-       struct bkey_s_c k;
-       struct bkey_s_c_inode inode;
-       struct bch_inode_unpacked prev, u;
+       struct bch_inode_unpacked prev = { 0 };
        int ret;
 
-       memset(&prev, 0, sizeof(prev));
-
        bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 
-       for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
-                          BTREE_ITER_INTENT|
-                          BTREE_ITER_PREFETCH|
-                          BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               ret = check_key_has_snapshot(&trans, &iter, k);
+       bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes, POS_MIN,
+                            BTREE_ITER_INTENT|
+                            BTREE_ITER_PREFETCH|
+                            BTREE_ITER_ALL_SNAPSHOTS);
+
+       do {
+               ret = __bch2_trans_do(&trans, NULL, NULL,
+                                     BTREE_INSERT_LAZY_RW|
+                                     BTREE_INSERT_NOFAIL,
+                       check_inode(&trans, &iter, &prev, full));
                if (ret)
                        break;
+       } while (bch2_btree_iter_advance(&iter));
+       bch2_trans_iter_exit(&trans, &iter);
 
-               /*
-                * if snapshot id isn't a leaf node, skip it - deletion in
-                * particular is not atomic, so on the internal snapshot nodes
-                * we can see inodes marked for deletion after a clean shutdown
-                */
-               if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
-                       continue;
+       bch2_trans_exit(&trans);
+       return ret;
+}
 
-               if (k.k->type != KEY_TYPE_inode)
-                       continue;
+static int check_subvol(struct btree_trans *trans,
+                       struct btree_iter *iter)
+{
+       struct bkey_s_c k;
+       struct bkey_s_c_subvolume subvol;
+       int ret;
 
-               inode = bkey_s_c_to_inode(k);
+       k = bch2_btree_iter_peek(iter);
+       if (!k.k)
+               return 0;
 
-               if (!full &&
-                   !(inode.v->bi_flags & (BCH_INODE_I_SIZE_DIRTY|
-                                          BCH_INODE_I_SECTORS_DIRTY|
-                                          BCH_INODE_UNLINKED)))
-                       continue;
+       ret = bkey_err(k);
+       if (ret)
+               return ret;
 
-               BUG_ON(bch2_inode_unpack(inode, &u));
+       if (k.k->type != KEY_TYPE_subvolume)
+               return 0;
 
-               ret = check_inode(&trans, &iter,
-                                 full && prev.bi_inum == u.bi_inum
-                                 ? &prev : NULL, u);
-               if (ret)
-                       break;
+       subvol = bkey_s_c_to_subvolume(k);
 
-               prev = u;
+       if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
+               ret = bch2_subvolume_delete(trans, iter->pos.offset);
+               if (ret && ret != -EINTR)
+                       bch_err(trans->c, "error deleting subvolume %llu: %i",
+                               iter->pos.offset, ret);
+               if (ret)
+                       return ret;
        }
-       bch2_trans_iter_exit(&trans, &iter);
-
-       BUG_ON(ret == -EINTR);
 
-       bch2_trans_exit(&trans);
-       return ret;
+       return 0;
 }
 
 noinline_for_stack
@@ -991,14 +981,23 @@ static int check_subvols(struct bch_fs *c)
 {
        struct btree_trans trans;
        struct btree_iter iter;
-       struct bkey_s_c k;
        int ret;
 
        bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 
-       for_each_btree_key(&trans, iter, BTREE_ID_subvolumes, POS_MIN,
-                          0, k, ret) {
-       }
+       bch2_trans_iter_init(&trans, &iter, BTREE_ID_subvolumes,
+                            POS_MIN,
+                            BTREE_ITER_INTENT|
+                            BTREE_ITER_PREFETCH);
+
+       do {
+               ret = __bch2_trans_do(&trans, NULL, NULL,
+                                     BTREE_INSERT_LAZY_RW|
+                                     BTREE_INSERT_NOFAIL,
+                                     check_subvol(&trans, &iter));
+               if (ret)
+                       break;
+       } while (bch2_btree_iter_advance(&iter));
        bch2_trans_iter_exit(&trans, &iter);
 
        bch2_trans_exit(&trans);
@@ -1107,7 +1106,7 @@ static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
        int ret = 0, ret2 = 0;
        s64 count2;
 
-       for (i = w->d; i < w->d + w->nr; i++) {
+       darray_for_each(w->inodes, i) {
                if (i->inode.bi_sectors == i->count)
                        continue;
 
@@ -1145,32 +1144,43 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
        struct bch_fs *c = trans->c;
        struct bkey_s_c k;
        struct inode_walker_entry *i;
-       char buf[200];
+       struct printbuf buf = PRINTBUF;
        int ret = 0;
-
+peek:
        k = bch2_btree_iter_peek(iter);
        if (!k.k)
-               return 0;
+               goto out;
 
        ret = bkey_err(k);
        if (ret)
-               return ret;
+               goto err;
 
        ret = check_key_has_snapshot(trans, iter, k);
-       if (ret)
-               return ret;
+       if (ret) {
+               ret = ret < 0 ? ret : 0;
+               goto out;
+       }
 
        ret = snapshots_seen_update(c, s, k.k->p);
        if (ret)
-               return ret;
+               goto err;
 
        if (k.k->type == KEY_TYPE_whiteout)
-               return 0;
+               goto out;
 
        if (inode->cur_inum != k.k->p.inode) {
                ret = check_i_sectors(trans, inode);
                if (ret)
-                       return ret;
+                       goto err;
+       }
+
+       if (!iter->path->should_be_locked) {
+               /*
+                * hack: check_i_sectors may have handled a transaction restart,
+                * it shouldn't be but we need to fix the new i_sectors check
+                * code and delete the old bch2_count_inode_sectors() first
+                */
+               goto peek;
        }
 #if 0
        if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
@@ -1180,35 +1190,43 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
                bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
                bch2_bkey_val_to_text(&PBUF(buf2), c, k);
 
-               if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2))
-                       return fix_overlapping_extent(trans, k, prev.k->k.p) ?: -EINTR;
+               if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2)) {
+                       ret = fix_overlapping_extent(trans, k, prev.k->k.p) ?: -EINTR;
+                       goto out;
+               }
        }
 #endif
        ret = __walk_inode(trans, inode, k.k->p);
        if (ret < 0)
-               return ret;
+               goto err;
 
        if (fsck_err_on(ret == INT_MAX, c,
                        "extent in missing inode:\n  %s",
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
-               return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                       bch2_btree_delete_at(trans, iter,
-                                            BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
+                       (printbuf_reset(&buf),
+                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+               ret = bch2_btree_delete_at(trans, iter,
+                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+               goto out;
+       }
 
-       if (ret == INT_MAX)
-               return 0;
+       if (ret == INT_MAX) {
+               ret = 0;
+               goto out;
+       }
 
-       i = inode->d + ret;
+       i = inode->inodes.data + ret;
        ret = 0;
 
        if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
                        !S_ISLNK(i->inode.bi_mode), c,
                        "extent in non regular inode mode %o:\n  %s",
                        i->inode.bi_mode,
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
-               return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                        bch2_btree_delete_at(trans, iter,
-                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
+                       (printbuf_reset(&buf),
+                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+               ret = bch2_btree_delete_at(trans, iter,
+                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+               goto out;
+       }
 
        if (!bch2_snapshot_internal_node(c, k.k->p.snapshot)) {
                for_each_visible_inode(c, s, inode, k.k->p.snapshot, i) {
@@ -1218,11 +1236,12 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
                                        "extent type %u offset %llu past end of inode %llu, i_size %llu",
                                        k.k->type, k.k->p.offset, k.k->p.inode, i->inode.bi_size)) {
                                bch2_fs_lazy_rw(c);
-                               return bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
+                               ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
                                                SPOS(k.k->p.inode, round_up(i->inode.bi_size, block_bytes(c)) >> 9,
                                                     k.k->p.snapshot),
                                                POS(k.k->p.inode, U64_MAX),
                                                0, NULL) ?: -EINTR;
+                               goto out;
                        }
                }
        }
@@ -1234,7 +1253,10 @@ static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
        bch2_bkey_buf_reassemble(&prev, c, k);
 #endif
 
+out:
+err:
 fsck_err:
+       printbuf_exit(&buf);
        return ret;
 }
 
@@ -1268,7 +1290,9 @@ static int check_extents(struct bch_fs *c)
                             BTREE_ITER_ALL_SNAPSHOTS);
 
        do {
-               ret = lockrestart_do(&trans,
+               ret = __bch2_trans_do(&trans, NULL, NULL,
+                                     BTREE_INSERT_LAZY_RW|
+                                     BTREE_INSERT_NOFAIL,
                        check_extent(&trans, &iter, &w, &s));
                if (ret)
                        break;
@@ -1291,12 +1315,13 @@ static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
        int ret = 0, ret2 = 0;
        s64 count2;
 
-       for (i = w->d; i < w->d + w->nr; i++) {
+       darray_for_each(w->inodes, i) {
                if (i->inode.bi_nlink == i->count)
                        continue;
 
-               count2 = lockrestart_do(trans,
-                               bch2_count_subdirs(trans, w->cur_inum, i->snapshot));
+               count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot);
+               if (count2 < 0)
+                       return count2;
 
                if (i->count != count2) {
                        bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
@@ -1327,8 +1352,9 @@ static int check_dirent_target(struct btree_trans *trans,
                               u32 target_snapshot)
 {
        struct bch_fs *c = trans->c;
+       struct bkey_i_dirent *n;
        bool backpointer_exists = true;
-       char buf[200];
+       struct printbuf buf = PRINTBUF;
        int ret = 0;
 
        if (!target->bi_dir &&
@@ -1336,7 +1362,7 @@ static int check_dirent_target(struct btree_trans *trans,
                target->bi_dir          = d.k->p.inode;
                target->bi_dir_offset   = d.k->p.offset;
 
-               ret = write_inode(trans, target, target_snapshot);
+               ret = __write_inode(trans, target, target_snapshot);
                if (ret)
                        goto err;
        }
@@ -1353,10 +1379,8 @@ static int check_dirent_target(struct btree_trans *trans,
                                backpointer_exists, c,
                                "directory %llu with multiple links",
                                target->bi_inum)) {
-                       ret = remove_dirent(trans, d.k->p);
-                       if (ret)
-                               goto err;
-                       return 0;
+                       ret = __remove_dirent(trans, d.k->p);
+                       goto out;
                }
 
                if (fsck_err_on(backpointer_exists &&
@@ -1366,7 +1390,7 @@ static int check_dirent_target(struct btree_trans *trans,
                        target->bi_nlink++;
                        target->bi_flags &= ~BCH_INODE_UNLINKED;
 
-                       ret = write_inode(trans, target, target_snapshot);
+                       ret = __write_inode(trans, target, target_snapshot);
                        if (ret)
                                goto err;
                }
@@ -1383,34 +1407,31 @@ static int check_dirent_target(struct btree_trans *trans,
                        target->bi_dir          = d.k->p.inode;
                        target->bi_dir_offset   = d.k->p.offset;
 
-                       ret = write_inode(trans, target, target_snapshot);
+                       ret = __write_inode(trans, target, target_snapshot);
                        if (ret)
                                goto err;
                }
        }
 
-       if (fsck_err_on(vfs_d_type(d.v->d_type) != mode_to_type(target->bi_mode), c,
-                       "incorrect d_type: should be %u:\n%s",
-                       mode_to_type(target->bi_mode),
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, d.s_c), buf))) {
-               struct bkey_i_dirent *n;
-
-               n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
-               if (!n) {
-                       ret = -ENOMEM;
+       if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
+                       "incorrect d_type: got %s, should be %s:\n%s",
+                       bch2_d_type_str(d.v->d_type),
+                       bch2_d_type_str(inode_d_type(target)),
+                       (printbuf_reset(&buf),
+                        bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
+               n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
+               ret = PTR_ERR_OR_ZERO(n);
+               if (ret)
                        goto err;
-               }
 
                bkey_reassemble(&n->k_i, d.s_c);
-               n->v.d_type = mode_to_type(target->bi_mode);
+               n->v.d_type = inode_d_type(target);
 
-               ret = __bch2_trans_do(trans, NULL, NULL,
-                                     BTREE_INSERT_NOFAIL|
-                                     BTREE_INSERT_LAZY_RW,
-                       bch2_trans_update(trans, iter, &n->k_i, 0));
-               kfree(n);
+               ret = bch2_trans_update(trans, iter, &n->k_i, 0);
+               if (ret)
+                       goto err;
 
-               return ret ?: -EINTR;
+               d = dirent_i_to_s_c(n);
        }
 
        if (d.v->d_type == DT_SUBVOL &&
@@ -1419,27 +1440,24 @@ static int check_dirent_target(struct btree_trans *trans,
             fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
                      le32_to_cpu(d.v->d_parent_subvol),
                      target->bi_parent_subvol))) {
-               struct bkey_i_dirent *n;
-
-               n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
-               if (!n) {
-                       ret = -ENOMEM;
+               n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
+               ret = PTR_ERR_OR_ZERO(n);
+               if (ret)
                        goto err;
-               }
 
                bkey_reassemble(&n->k_i, d.s_c);
                n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
 
-               ret = __bch2_trans_do(trans, NULL, NULL,
-                                     BTREE_INSERT_NOFAIL|
-                                     BTREE_INSERT_LAZY_RW,
-                       bch2_trans_update(trans, iter, &n->k_i, 0));
-               kfree(n);
+               ret = bch2_trans_update(trans, iter, &n->k_i, 0);
+               if (ret)
+                       goto err;
 
-               return ret ?: -EINTR;
+               d = dirent_i_to_s_c(n);
        }
+out:
 err:
 fsck_err:
+       printbuf_exit(&buf);
        return ret;
 }
 
@@ -1453,70 +1471,86 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
        struct bkey_s_c k;
        struct bkey_s_c_dirent d;
        struct inode_walker_entry *i;
-       char buf[200];
-       int ret;
-
+       struct printbuf buf = PRINTBUF;
+       int ret = 0;
+peek:
        k = bch2_btree_iter_peek(iter);
        if (!k.k)
-               return 0;
+               goto out;
 
        ret = bkey_err(k);
        if (ret)
-               return ret;
+               goto err;
 
        ret = check_key_has_snapshot(trans, iter, k);
-       if (ret)
-               return ret;
+       if (ret) {
+               ret = ret < 0 ? ret : 0;
+               goto out;
+       }
 
        ret = snapshots_seen_update(c, s, k.k->p);
        if (ret)
-               return ret;
+               goto err;
 
        if (k.k->type == KEY_TYPE_whiteout)
-               return 0;
+               goto out;
 
        if (dir->cur_inum != k.k->p.inode) {
                ret = check_subdir_count(trans, dir);
                if (ret)
-                       return ret;
+                       goto err;
+       }
+
+       if (!iter->path->should_be_locked) {
+               /* hack: see check_extent() */
+               goto peek;
        }
 
        ret = __walk_inode(trans, dir, k.k->p);
        if (ret < 0)
-               return ret;
+               goto err;
 
        if (fsck_err_on(ret == INT_MAX, c,
                        "dirent in nonexisting directory:\n%s",
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
-               return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                               bch2_btree_delete_at(trans, iter,
-                                                    BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
+                       (printbuf_reset(&buf),
+                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+               ret = bch2_btree_delete_at(trans, iter,
+                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+               goto out;
+       }
 
-       if (ret == INT_MAX)
-               return 0;
+       if (ret == INT_MAX) {
+               ret = 0;
+               goto out;
+       }
 
-       i = dir->d + ret;
+       i = dir->inodes.data + ret;
        ret = 0;
 
        if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
-                       "dirent in non directory inode type %u:\n%s",
-                       mode_to_type(i->inode.bi_mode),
-                       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
-               return __bch2_trans_do(trans, NULL, NULL, 0,
-                               bch2_btree_delete_at(trans, iter, 0));
+                       "dirent in non directory inode type %s:\n%s",
+                       bch2_d_type_str(inode_d_type(&i->inode)),
+                       (printbuf_reset(&buf),
+                        bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+               ret = bch2_btree_delete_at(trans, iter, 0);
+               goto out;
+       }
 
        if (dir->first_this_inode)
-               *hash_info = bch2_hash_info_init(c, &dir->d[0].inode);
+               *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
 
        ret = hash_check_key(trans, bch2_dirent_hash_desc,
                             hash_info, iter, k);
        if (ret < 0)
-               return ret;
-       if (ret) /* dirent has been deleted */
-               return 0;
+               goto err;
+       if (ret) {
+               /* dirent has been deleted */
+               ret = 0;
+               goto out;
+       }
 
        if (k.k->type != KEY_TYPE_dirent)
-               return 0;
+               goto out;
 
        d = bkey_s_c_to_dirent(k);
 
@@ -1529,24 +1563,27 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                ret = __subvol_lookup(trans, target_subvol,
                                      &target_snapshot, &target_inum);
                if (ret && ret != -ENOENT)
-                       return ret;
+                       goto err;
 
                if (fsck_err_on(ret, c,
                                "dirent points to missing subvolume %llu",
-                               le64_to_cpu(d.v->d_child_subvol)))
-                       return remove_dirent(trans, d.k->p);
+                               le64_to_cpu(d.v->d_child_subvol))) {
+                       ret = __remove_dirent(trans, d.k->p);
+                       goto err;
+               }
 
                ret = __lookup_inode(trans, target_inum,
                                   &subvol_root, &target_snapshot);
                if (ret && ret != -ENOENT)
-                       return ret;
+                       goto err;
 
                if (fsck_err_on(ret, c,
                                "subvolume %u points to missing subvolume root %llu",
                                target_subvol,
                                target_inum)) {
                        bch_err(c, "repair not implemented yet");
-                       return -EINVAL;
+                       ret = -EINVAL;
+                       goto err;
                }
 
                if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
@@ -1554,34 +1591,35 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                                target_inum,
                                subvol_root.bi_subvol, target_subvol)) {
                        subvol_root.bi_subvol = target_subvol;
-                       ret = write_inode(trans, &subvol_root, target_snapshot);
+                       ret = __write_inode(trans, &subvol_root, target_snapshot);
                        if (ret)
-                               return ret;
+                               goto err;
                }
 
                ret = check_dirent_target(trans, iter, d, &subvol_root,
                                          target_snapshot);
                if (ret)
-                       return ret;
+                       goto err;
        } else {
                ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
                if (ret)
-                       return ret;
+                       goto err;
 
-               if (fsck_err_on(!target->nr, c,
+               if (fsck_err_on(!target->inodes.nr, c,
                                "dirent points to missing inode:\n%s",
-                               (bch2_bkey_val_to_text(&PBUF(buf), c,
-                                                      k), buf))) {
-                       ret = remove_dirent(trans, d.k->p);
+                               (printbuf_reset(&buf),
+                                bch2_bkey_val_to_text(&buf, c, k),
+                                buf.buf))) {
+                       ret = __remove_dirent(trans, d.k->p);
                        if (ret)
-                               return ret;
+                               goto err;
                }
 
-               for (i = target->d; i < target->d + target->nr; i++) {
+               darray_for_each(target->inodes, i) {
                        ret = check_dirent_target(trans, iter, d,
                                                  &i->inode, i->snapshot);
                        if (ret)
-                               return ret;
+                               goto err;
                }
        }
 
@@ -1589,7 +1627,10 @@ static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
                for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
                        i->count++;
 
+out:
+err:
 fsck_err:
+       printbuf_exit(&buf);
        return ret;
 }
 
@@ -1620,7 +1661,9 @@ static int check_dirents(struct bch_fs *c)
                             BTREE_ITER_ALL_SNAPSHOTS);
 
        do {
-               ret = lockrestart_do(&trans,
+               ret = __bch2_trans_do(&trans, NULL, NULL,
+                                     BTREE_INSERT_LAZY_RW|
+                                     BTREE_INSERT_NOFAIL,
                        check_dirent(&trans, &iter, &hash_info,
                                     &dir, &target, &s));
                if (ret)
@@ -1635,17 +1678,58 @@ static int check_dirents(struct bch_fs *c)
        return ret;
 }
 
+static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
+                      struct bch_hash_info *hash_info,
+                      struct inode_walker *inode)
+{
+       struct bch_fs *c = trans->c;
+       struct bkey_s_c k;
+       int ret;
+
+       k = bch2_btree_iter_peek(iter);
+       if (!k.k)
+               return 0;
+
+       ret = bkey_err(k);
+       if (ret)
+               return ret;
+
+       ret = check_key_has_snapshot(trans, iter, k);
+       if (ret)
+               return ret;
+
+       ret = __walk_inode(trans, inode, k.k->p);
+       if (ret < 0)
+               return ret;
+
+       if (fsck_err_on(ret == INT_MAX, c,
+                       "xattr for missing inode %llu",
+                       k.k->p.inode))
+               return bch2_btree_delete_at(trans, iter, 0);
+
+       if (ret == INT_MAX)
+               return 0;
+
+       ret = 0;
+
+       if (inode->first_this_inode)
+               *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
+
+       ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
+fsck_err:
+       return ret;
+}
+
 /*
  * Walk xattrs: verify that they all have a corresponding inode
  */
 noinline_for_stack
 static int check_xattrs(struct bch_fs *c)
 {
-       struct inode_walker w = inode_walker_init();
+       struct inode_walker inode = inode_walker_init();
        struct bch_hash_info hash_info;
        struct btree_trans trans;
        struct btree_iter iter;
-       struct bkey_s_c k;
        int ret = 0;
 
        bch_verbose(c, "checking xattrs");
@@ -1657,65 +1741,31 @@ static int check_xattrs(struct bch_fs *c)
                             BTREE_ITER_INTENT|
                             BTREE_ITER_PREFETCH|
                             BTREE_ITER_ALL_SNAPSHOTS);
-retry:
-       bch2_trans_begin(&trans);
-
-       while ((k = bch2_btree_iter_peek(&iter)).k &&
-              !(ret = bkey_err(k))) {
-               ret = check_key_has_snapshot(&trans, &iter, k);
-               if (ret)
-                       break;
 
-               ret = walk_inode(&trans, &w, k.k->p);
-               if (ret < 0)
-                       break;
-
-               if (fsck_err_on(ret == INT_MAX, c,
-                               "xattr for missing inode %llu",
-                               k.k->p.inode)) {
-                       ret = bch2_btree_delete_at(&trans, &iter, 0);
-                       if (ret)
-                               break;
-                       continue;
-               }
-
-               if (ret == INT_MAX)
-                       goto next;
-               ret = 0;
-
-               if (w.first_this_inode)
-                       hash_info = bch2_hash_info_init(c, &w.d[0].inode);
-
-               ret = hash_check_key(&trans, bch2_xattr_hash_desc,
-                                    &hash_info, &iter, k);
+       do {
+               ret = __bch2_trans_do(&trans, NULL, NULL,
+                                     BTREE_INSERT_LAZY_RW|
+                                     BTREE_INSERT_NOFAIL,
+                                     check_xattr(&trans, &iter, &hash_info,
+                                                 &inode));
                if (ret)
                        break;
-next:
-               bch2_btree_iter_advance(&iter);
-       }
-fsck_err:
-       if (ret == -EINTR)
-               goto retry;
-
+       } while (bch2_btree_iter_advance(&iter));
        bch2_trans_iter_exit(&trans, &iter);
+
        bch2_trans_exit(&trans);
        return ret;
 }
 
-/* Get root directory, create if it doesn't exist: */
-static int check_root(struct bch_fs *c)
+static int check_root_trans(struct btree_trans *trans)
 {
-       struct btree_trans trans;
+       struct bch_fs *c = trans->c;
        struct bch_inode_unpacked root_inode;
        u32 snapshot;
        u64 inum;
        int ret;
 
-       bch2_trans_init(&trans, c, 0, 0);
-
-       bch_verbose(c, "checking root directory");
-
-       ret = subvol_lookup(&trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
+       ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
        if (ret && ret != -ENOENT)
                return ret;
 
@@ -1730,10 +1780,10 @@ static int check_root(struct bch_fs *c)
                root_subvol.v.flags     = 0;
                root_subvol.v.snapshot  = cpu_to_le32(snapshot);
                root_subvol.v.inode     = cpu_to_le64(inum);
-               ret = __bch2_trans_do(&trans, NULL, NULL,
+               ret = __bch2_trans_do(trans, NULL, NULL,
                                      BTREE_INSERT_NOFAIL|
                                      BTREE_INSERT_LAZY_RW,
-                       __bch2_btree_insert(&trans, BTREE_ID_subvolumes, &root_subvol.k_i));
+                       __bch2_btree_insert(trans, BTREE_ID_subvolumes, &root_subvol.k_i));
                if (ret) {
                        bch_err(c, "error writing root subvol: %i", ret);
                        goto err;
@@ -1741,7 +1791,7 @@ static int check_root(struct bch_fs *c)
 
        }
 
-       ret = lookup_inode(&trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
+       ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
        if (ret && ret != -ENOENT)
                return ret;
 
@@ -1752,31 +1802,39 @@ static int check_root(struct bch_fs *c)
                                0, NULL);
                root_inode.bi_inum = inum;
 
-               ret = write_inode(&trans, &root_inode, snapshot);
+               ret = __write_inode(trans, &root_inode, snapshot);
                if (ret)
                        bch_err(c, "error writing root inode: %i", ret);
        }
 err:
 fsck_err:
-       bch2_trans_exit(&trans);
        return ret;
 }
 
-struct pathbuf {
-       size_t          nr;
-       size_t          size;
+/* Get root directory, create if it doesn't exist: */
+noinline_for_stack
+static int check_root(struct bch_fs *c)
+{
+       bch_verbose(c, "checking root directory");
 
-       struct pathbuf_entry {
-               u64     inum;
-               u32     snapshot;
-       }               *entries;
+       return bch2_trans_do(c, NULL, NULL,
+                            BTREE_INSERT_NOFAIL|
+                            BTREE_INSERT_LAZY_RW,
+               check_root_trans(&trans));
+}
+
+struct pathbuf_entry {
+       u64     inum;
+       u32     snapshot;
 };
 
-static bool path_is_dup(struct pathbuf *p, u64 inum, u32 snapshot)
+typedef DARRAY(struct pathbuf_entry) pathbuf;
+
+static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
 {
        struct pathbuf_entry *i;
 
-       for (i = p->entries; i < p->entries + p->nr; i++)
+       darray_for_each(*p, i)
                if (i->inum     == inum &&
                    i->snapshot == snapshot)
                        return true;
@@ -1784,26 +1842,18 @@ static bool path_is_dup(struct pathbuf *p, u64 inum, u32 snapshot)
        return false;
 }
 
-static int path_down(struct pathbuf *p, u64 inum, u32 snapshot)
+static int path_down(struct bch_fs *c, pathbuf *p,
+                    u64 inum, u32 snapshot)
 {
-       if (p->nr == p->size) {
-               size_t new_size = max_t(size_t, 256UL, p->size * 2);
-               void *n = krealloc(p->entries,
-                                  new_size * sizeof(p->entries[0]),
-                                  GFP_KERNEL);
-               if (!n) {
-                       return -ENOMEM;
-               }
-
-               p->entries = n;
-               p->size = new_size;
-       };
-
-       p->entries[p->nr++] = (struct pathbuf_entry) {
+       int ret = darray_push(*p, ((struct pathbuf_entry) {
                .inum           = inum,
                .snapshot       = snapshot,
-       };
-       return 0;
+       }));
+
+       if (ret)
+               bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
+                       p->size);
+       return ret;
 }
 
 /*
@@ -1812,7 +1862,7 @@ static int path_down(struct pathbuf *p, u64 inum, u32 snapshot)
  * XXX: we should also be verifying that inodes are in the right subvolumes
  */
 static int check_path(struct btree_trans *trans,
-                     struct pathbuf *p,
+                     pathbuf *p,
                      struct bch_inode_unpacked *inode,
                      u32 snapshot)
 {
@@ -1850,9 +1900,9 @@ static int check_path(struct btree_trans *trans,
                }
 
                if (ret == -ENOENT) {
-                       if (fsck_err(c,  "unreachable inode %llu:%u, type %u nlink %u backptr %llu:%llu",
+                       if (fsck_err(c,  "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
                                     inode->bi_inum, snapshot,
-                                    mode_to_type(inode->bi_mode),
+                                    bch2_d_type_str(inode_d_type(inode)),
                                     inode->bi_nlink,
                                     inode->bi_dir,
                                     inode->bi_dir_offset))
@@ -1865,7 +1915,7 @@ static int check_path(struct btree_trans *trans,
                if (!S_ISDIR(inode->bi_mode))
                        break;
 
-               ret = path_down(p, inode->bi_inum, snapshot);
+               ret = path_down(c, p, inode->bi_inum, snapshot);
                if (ret) {
                        bch_err(c, "memory allocation failure");
                        return ret;
@@ -1886,14 +1936,16 @@ static int check_path(struct btree_trans *trans,
                        /* XXX print path */
                        bch_err(c, "directory structure loop");
 
-                       for (i = p->entries; i < p->entries + p->nr; i++)
+                       darray_for_each(*p, i)
                                pr_err("%llu:%u", i->inum, i->snapshot);
                        pr_err("%llu:%u", inode->bi_inum, snapshot);
 
                        if (!fsck_err(c, "directory structure loop"))
                                return 0;
 
-                       ret = lockrestart_do(trans,
+                       ret = __bch2_trans_do(trans, NULL, NULL,
+                                             BTREE_INSERT_NOFAIL|
+                                             BTREE_INSERT_LAZY_RW,
                                        remove_backpointer(trans, inode));
                        if (ret) {
                                bch_err(c, "error removing dirent: %i", ret);
@@ -1914,13 +1966,14 @@ fsck_err:
  * After check_dirents(), if an inode backpointer doesn't exist that means it's
  * unreachable:
  */
+noinline_for_stack
 static int check_directory_structure(struct bch_fs *c)
 {
        struct btree_trans trans;
        struct btree_iter iter;
        struct bkey_s_c k;
        struct bch_inode_unpacked u;
-       struct pathbuf path = { 0, 0, NULL };
+       pathbuf path = { 0, };
        int ret;
 
        bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
@@ -1929,10 +1982,10 @@ static int check_directory_structure(struct bch_fs *c)
                           BTREE_ITER_INTENT|
                           BTREE_ITER_PREFETCH|
                           BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               if (k.k->type != KEY_TYPE_inode)
+               if (!bkey_is_inode(k.k))
                        continue;
 
-               ret = bch2_inode_unpack(bkey_s_c_to_inode(k), &u);
+               ret = bch2_inode_unpack(k, &u);
                if (ret) {
                        /* Should have been caught earlier in fsck: */
                        bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
@@ -1950,7 +2003,7 @@ static int check_directory_structure(struct bch_fs *c)
 
        BUG_ON(ret == -EINTR);
 
-       kfree(path.entries);
+       darray_exit(path);
 
        bch2_trans_exit(&trans);
        return ret;
@@ -1967,12 +2020,15 @@ struct nlink_table {
        }               *d;
 };
 
-static int add_nlink(struct nlink_table *t, u64 inum, u32 snapshot)
+static int add_nlink(struct bch_fs *c, struct nlink_table *t,
+                    u64 inum, u32 snapshot)
 {
        if (t->nr == t->size) {
                size_t new_size = max_t(size_t, 128UL, t->size * 2);
                void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
                if (!d) {
+                       bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
+                               new_size);
                        return -ENOMEM;
                }
 
@@ -2036,7 +2092,6 @@ static int check_nlinks_find_hardlinks(struct bch_fs *c,
        struct btree_trans trans;
        struct btree_iter iter;
        struct bkey_s_c k;
-       struct bkey_s_c_inode inode;
        struct bch_inode_unpacked u;
        int ret = 0;
 
@@ -2047,25 +2102,23 @@ static int check_nlinks_find_hardlinks(struct bch_fs *c,
                           BTREE_ITER_INTENT|
                           BTREE_ITER_PREFETCH|
                           BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
-               if (k.k->type != KEY_TYPE_inode)
+               if (!bkey_is_inode(k.k))
                        continue;
 
-               inode = bkey_s_c_to_inode(k);
+               /* Should never fail, checked by bch2_inode_invalid: */
+               BUG_ON(bch2_inode_unpack(k, &u));
 
                /*
                 * Backpointer and directory structure checks are sufficient for
                 * directories, since they can't have hardlinks:
                 */
-               if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
+               if (S_ISDIR(le16_to_cpu(u.bi_mode)))
                        continue;
 
-               /* Should never fail, checked by bch2_inode_invalid: */
-               BUG_ON(bch2_inode_unpack(inode, &u));
-
                if (!u.bi_nlink)
                        continue;
 
-               ret = add_nlink(t, k.k->p.offset, k.k->p.snapshot);
+               ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
                if (ret) {
                        *end = k.k->p.offset;
                        ret = 0;
@@ -2116,8 +2169,6 @@ static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links
                                         d.k->p.snapshot);
                        break;
                }
-
-               bch2_trans_cond_resched(&trans);
        }
        bch2_trans_iter_exit(&trans, &iter);
 
@@ -2137,7 +2188,6 @@ static int check_nlinks_update_hardlinks(struct bch_fs *c,
        struct btree_trans trans;
        struct btree_iter iter;
        struct bkey_s_c k;
-       struct bkey_s_c_inode inode;
        struct bch_inode_unpacked u;
        struct nlink *link = links->d;
        int ret = 0;
@@ -2152,14 +2202,13 @@ static int check_nlinks_update_hardlinks(struct bch_fs *c,
                if (k.k->p.offset >= range_end)
                        break;
 
-               if (k.k->type != KEY_TYPE_inode)
+               if (!bkey_is_inode(k.k))
                        continue;
 
-               inode = bkey_s_c_to_inode(k);
-               if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
-                       continue;
+               BUG_ON(bch2_inode_unpack(k, &u));
 
-               BUG_ON(bch2_inode_unpack(inode, &u));
+               if (S_ISDIR(le16_to_cpu(u.bi_mode)))
+                       continue;
 
                if (!u.bi_nlink)
                        continue;
@@ -2260,9 +2309,10 @@ static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter)
        u->v.front_pad  = 0;
        u->v.back_pad   = 0;
 
-       return bch2_trans_update(trans, iter, &u->k_i, 0);
+       return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
 }
 
+noinline_for_stack
 static int fix_reflink_p(struct bch_fs *c)
 {
        struct btree_trans trans;
@@ -2273,6 +2323,8 @@ static int fix_reflink_p(struct bch_fs *c)
        if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
                return 0;
 
+       bch_verbose(c, "fixing reflink_p keys");
+
        bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 
        for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN,