]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/buckets.c
Update bcachefs sources to 6a20aede29 bcachefs: Fix quotas + snapshots
[bcachefs-tools-debian] / libbcachefs / buckets.c
index 1ea7e2baf32323a2cc6b4726869c3bedd841737b..bce42eef6f576001341fe94283a05ea46652a1d9 100644 (file)
@@ -21,9 +21,9 @@
 #include "reflink.h"
 #include "replicas.h"
 #include "subvolume.h"
+#include "trace.h"
 
 #include <linux/preempt.h>
-#include <trace/events/bcachefs.h>
 
 static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage,
                                              enum bch_data_type data_type,
@@ -89,20 +89,17 @@ static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
                            : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
 }
 
-struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
+void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
 {
        struct bch_fs *c = ca->fs;
-       struct bch_dev_usage ret;
        unsigned seq, i, u64s = dev_usage_u64s();
 
        do {
                seq = read_seqcount_begin(&c->usage_lock);
-               memcpy(&ret, ca->usage_base, u64s * sizeof(u64));
+               memcpy(usage, ca->usage_base, u64s * sizeof(u64));
                for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
-                       acc_u64s_percpu((u64 *) &ret, (u64 __percpu *) ca->usage[i], u64s);
+                       acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s);
        } while (read_seqcount_retry(&c->usage_lock, seq));
-
-       return ret;
 }
 
 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c,
@@ -140,23 +137,28 @@ u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
 {
        struct bch_fs_usage_online *ret;
-       unsigned seq, i, u64s;
+       unsigned seq, i, v, u64s = fs_usage_u64s(c) + 1;
+retry:
+       ret = kmalloc(u64s * sizeof(u64), GFP_NOFS);
+       if (unlikely(!ret))
+               return NULL;
 
        percpu_down_read(&c->mark_lock);
 
-       ret = kmalloc(sizeof(struct bch_fs_usage_online) +
-                     sizeof(u64) * c->replicas.nr, GFP_NOFS);
-       if (unlikely(!ret)) {
+       v = fs_usage_u64s(c) + 1;
+       if (unlikely(u64s != v)) {
+               u64s = v;
                percpu_up_read(&c->mark_lock);
-               return NULL;
+               kfree(ret);
+               goto retry;
        }
 
        ret->online_reserved = percpu_u64_get(c->online_reserved);
 
-       u64s = fs_usage_u64s(c);
        do {
                seq = read_seqcount_begin(&c->usage_lock);
-               memcpy(&ret->u, c->usage_base, u64s * sizeof(u64));
+               unsafe_memcpy(&ret->u, c->usage_base, u64s * sizeof(u64),
+                             "embedded variable length struct");
                for (i = 0; i < ARRAY_SIZE(c->usage); i++)
                        acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i], u64s);
        } while (read_seqcount_retry(&c->usage_lock, seq));
@@ -484,13 +486,16 @@ static inline void update_cached_sectors_list(struct btree_trans *trans,
 }
 
 int bch2_mark_alloc(struct btree_trans *trans,
+                   enum btree_id btree, unsigned level,
                    struct bkey_s_c old, struct bkey_s_c new,
                    unsigned flags)
 {
        bool gc = flags & BTREE_TRIGGER_GC;
        u64 journal_seq = trans->journal_res.seq;
+       u64 bucket_journal_seq;
        struct bch_fs *c = trans->c;
-       struct bch_alloc_v4 old_a, new_a;
+       struct bch_alloc_v4 old_a_convert, new_a_convert;
+       const struct bch_alloc_v4 *old_a, *new_a;
        struct bch_dev *ca;
        int ret = 0;
 
@@ -507,36 +512,38 @@ int bch2_mark_alloc(struct btree_trans *trans,
 
        ca = bch_dev_bkey_exists(c, new.k->p.inode);
 
-       bch2_alloc_to_v4(old, &old_a);
-       bch2_alloc_to_v4(new, &new_a);
+       old_a = bch2_alloc_to_v4(old, &old_a_convert);
+       new_a = bch2_alloc_to_v4(new, &new_a_convert);
+
+       bucket_journal_seq = new_a->journal_seq;
 
        if ((flags & BTREE_TRIGGER_INSERT) &&
-           data_type_is_empty(old_a.data_type) !=
-           data_type_is_empty(new_a.data_type) &&
+           data_type_is_empty(old_a->data_type) !=
+           data_type_is_empty(new_a->data_type) &&
            new.k->type == KEY_TYPE_alloc_v4) {
                struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
 
-               BUG_ON(!journal_seq);
+               EBUG_ON(!journal_seq);
 
                /*
                 * If the btree updates referring to a bucket weren't flushed
                 * before the bucket became empty again, then the we don't have
                 * to wait on a journal flush before we can reuse the bucket:
                 */
-               new_a.journal_seq = data_type_is_empty(new_a.data_type) &&
+               v->journal_seq = bucket_journal_seq =
+                       data_type_is_empty(new_a->data_type) &&
                        (journal_seq == v->journal_seq ||
                         bch2_journal_noflush_seq(&c->journal, v->journal_seq))
                        ? 0 : journal_seq;
-               v->journal_seq = new_a.journal_seq;
        }
 
-       if (!data_type_is_empty(old_a.data_type) &&
-           data_type_is_empty(new_a.data_type) &&
-           new_a.journal_seq) {
+       if (!data_type_is_empty(old_a->data_type) &&
+           data_type_is_empty(new_a->data_type) &&
+           bucket_journal_seq) {
                ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
                                c->journal.flushed_seq_ondisk,
                                new.k->p.inode, new.k->p.offset,
-                               new_a.journal_seq);
+                               bucket_journal_seq);
                if (ret) {
                        bch2_fs_fatal_error(c,
                                "error setting bucket_needs_journal_commit: %i", ret);
@@ -544,27 +551,11 @@ int bch2_mark_alloc(struct btree_trans *trans,
                }
        }
 
-       if (new_a.data_type == BCH_DATA_free &&
-           (!new_a.journal_seq || new_a.journal_seq < c->journal.flushed_seq_ondisk))
-               closure_wake_up(&c->freelist_wait);
-
-       if (new_a.data_type == BCH_DATA_need_discard &&
-           (!new_a.journal_seq || new_a.journal_seq < c->journal.flushed_seq_ondisk))
-               bch2_do_discards(c);
-
-       if (old_a.data_type != BCH_DATA_cached &&
-           new_a.data_type == BCH_DATA_cached &&
-           should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
-               bch2_do_invalidates(c);
-
-       if (new_a.data_type == BCH_DATA_need_gc_gens)
-               bch2_do_gc_gens(c);
-
        percpu_down_read(&c->mark_lock);
-       if (!gc && new_a.gen != old_a.gen)
-               *bucket_gen(ca, new.k->p.offset) = new_a.gen;
+       if (!gc && new_a->gen != old_a->gen)
+               *bucket_gen(ca, new.k->p.offset) = new_a->gen;
 
-       bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc);
+       bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
 
        if (gc) {
                struct bucket *g = gc_bucket(ca, new.k->p.offset);
@@ -572,12 +563,12 @@ int bch2_mark_alloc(struct btree_trans *trans,
                bucket_lock(g);
 
                g->gen_valid            = 1;
-               g->gen                  = new_a.gen;
-               g->data_type            = new_a.data_type;
-               g->stripe               = new_a.stripe;
-               g->stripe_redundancy    = new_a.stripe_redundancy;
-               g->dirty_sectors        = new_a.dirty_sectors;
-               g->cached_sectors       = new_a.cached_sectors;
+               g->gen                  = new_a->gen;
+               g->data_type            = new_a->data_type;
+               g->stripe               = new_a->stripe;
+               g->stripe_redundancy    = new_a->stripe_redundancy;
+               g->dirty_sectors        = new_a->dirty_sectors;
+               g->cached_sectors       = new_a->cached_sectors;
 
                bucket_unlock(g);
        }
@@ -589,16 +580,33 @@ int bch2_mark_alloc(struct btree_trans *trans,
         */
 
        if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
-           old_a.cached_sectors) {
+           old_a->cached_sectors) {
                ret = update_cached_sectors(c, new, ca->dev_idx,
-                                           -old_a.cached_sectors,
+                                           -((s64) old_a->cached_sectors),
                                            journal_seq, gc);
                if (ret) {
-                       bch2_fs_fatal_error(c, "bch2_mark_alloc(): no replicas entry while updating cached sectors");
+                       bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
+                                           __func__);
                        return ret;
                }
        }
 
+       if (new_a->data_type == BCH_DATA_free &&
+           (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
+               closure_wake_up(&c->freelist_wait);
+
+       if (new_a->data_type == BCH_DATA_need_discard &&
+           (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
+               bch2_do_discards(c);
+
+       if (old_a->data_type != BCH_DATA_cached &&
+           new_a->data_type == BCH_DATA_cached &&
+           should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
+               bch2_do_invalidates(c);
+
+       if (new_a->data_type == BCH_DATA_need_gc_gens)
+               bch2_do_gc_gens(c);
+
        return 0;
 }
 
@@ -656,13 +664,14 @@ err:
        return ret;
 }
 
-static int check_bucket_ref(struct bch_fs *c,
+static int check_bucket_ref(struct btree_trans *trans,
                            struct bkey_s_c k,
                            const struct bch_extent_ptr *ptr,
                            s64 sectors, enum bch_data_type ptr_data_type,
                            u8 b_gen, u8 bucket_data_type,
                            u32 dirty_sectors, u32 cached_sectors)
 {
+       struct bch_fs *c = trans->c;
        struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
        size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
        u16 bucket_sectors = !ptr->cached
@@ -674,6 +683,10 @@ static int check_bucket_ref(struct bch_fs *c,
        if (bucket_data_type == BCH_DATA_cached)
                bucket_data_type = BCH_DATA_user;
 
+       if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
+           (bucket_data_type == BCH_DATA_user   && ptr_data_type == BCH_DATA_stripe))
+               bucket_data_type = ptr_data_type = BCH_DATA_stripe;
+
        if (gen_after(ptr->gen, b_gen)) {
                bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
                        "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
@@ -715,7 +728,7 @@ static int check_bucket_ref(struct bch_fs *c,
 
        if (b_gen != ptr->gen) {
                ret = 1;
-               goto err;
+               goto out;
        }
 
        if (!data_type_is_empty(bucket_data_type) &&
@@ -745,9 +758,12 @@ static int check_bucket_ref(struct bch_fs *c,
                ret = -EIO;
                goto err;
        }
-err:
+out:
        printbuf_exit(&buf);
        return ret;
+err:
+       bch2_dump_trans_updates(trans);
+       goto out;
 }
 
 static int mark_stripe_bucket(struct btree_trans *trans,
@@ -760,7 +776,7 @@ static int mark_stripe_bucket(struct btree_trans *trans,
        const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
        unsigned nr_data = s->nr_blocks - s->nr_redundant;
        bool parity = ptr_idx >= nr_data;
-       enum bch_data_type data_type = parity ? BCH_DATA_parity : 0;
+       enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe;
        s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
        const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
        struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
@@ -789,14 +805,13 @@ static int mark_stripe_bucket(struct btree_trans *trans,
        bucket_lock(g);
        old = *g;
 
-       ret = check_bucket_ref(c, k, ptr, sectors, data_type,
+       ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
                               g->gen, g->data_type,
                               g->dirty_sectors, g->cached_sectors);
        if (ret)
                goto err;
 
-       if (data_type)
-               g->data_type = data_type;
+       g->data_type = data_type;
        g->dirty_sectors += sectors;
 
        g->stripe               = k.k->p.offset;
@@ -821,7 +836,7 @@ static int __mark_pointer(struct btree_trans *trans,
        u32 *dst_sectors = !ptr->cached
                ? dirty_sectors
                : cached_sectors;
-       int ret = check_bucket_ref(trans->c, k, ptr, sectors, ptr_data_type,
+       int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
                                   bucket_gen, *bucket_data_type,
                                   *dirty_sectors, *cached_sectors);
 
@@ -835,15 +850,17 @@ static int __mark_pointer(struct btree_trans *trans,
 }
 
 static int bch2_mark_pointer(struct btree_trans *trans,
+                            enum btree_id btree_id, unsigned level,
                             struct bkey_s_c k,
                             struct extent_ptr_decoded p,
-                            s64 sectors, enum bch_data_type data_type,
+                            s64 sectors,
                             unsigned flags)
 {
        u64 journal_seq = trans->journal_res.seq;
        struct bch_fs *c = trans->c;
        struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
        struct bucket old, new, *g;
+       enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
        u8 bucket_data_type;
        int ret = 0;
 
@@ -889,13 +906,13 @@ static int bch2_mark_stripe_ptr(struct btree_trans *trans,
        if (!m) {
                bch_err(c, "error allocating memory for gc_stripes, idx %llu",
                        (u64) p.idx);
-               return -ENOMEM;
+               return -BCH_ERR_ENOMEM_mark_stripe_ptr;
        }
 
-       spin_lock(&c->ec_stripes_heap_lock);
+       mutex_lock(&c->ec_stripes_heap_lock);
 
        if (!m || !m->alive) {
-               spin_unlock(&c->ec_stripes_heap_lock);
+               mutex_unlock(&c->ec_stripes_heap_lock);
                bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
                                    (u64) p.idx);
                bch2_inconsistent_error(c);
@@ -905,7 +922,7 @@ static int bch2_mark_stripe_ptr(struct btree_trans *trans,
        m->block_sectors[p.block] += sectors;
 
        r = m->r;
-       spin_unlock(&c->ec_stripes_heap_lock);
+       mutex_unlock(&c->ec_stripes_heap_lock);
 
        r.e.data_type = data_type;
        update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
@@ -914,12 +931,13 @@ static int bch2_mark_stripe_ptr(struct btree_trans *trans,
 }
 
 int bch2_mark_extent(struct btree_trans *trans,
+                    enum btree_id btree_id, unsigned level,
                     struct bkey_s_c old, struct bkey_s_c new,
                     unsigned flags)
 {
        u64 journal_seq = trans->journal_res.seq;
        struct bch_fs *c = trans->c;
-       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
+       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
        struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
        const union bch_extent_entry *entry;
        struct extent_ptr_decoded p;
@@ -946,8 +964,7 @@ int bch2_mark_extent(struct btree_trans *trans,
                if (flags & BTREE_TRIGGER_OVERWRITE)
                        disk_sectors = -disk_sectors;
 
-               ret = bch2_mark_pointer(trans, k, p, disk_sectors,
-                                       data_type, flags);
+               ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
                if (ret < 0)
                        return ret;
 
@@ -958,7 +975,8 @@ int bch2_mark_extent(struct btree_trans *trans,
                                ret = update_cached_sectors(c, k, p.ptr.dev,
                                                disk_sectors, journal_seq, true);
                                if (ret) {
-                                       bch2_fs_fatal_error(c, "bch2_mark_extent(): no replicas entry while updating cached sectors");
+                                       bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
+                                                           __func__);
                                        return ret;
                                }
                        }
@@ -986,7 +1004,7 @@ int bch2_mark_extent(struct btree_trans *trans,
                        struct printbuf buf = PRINTBUF;
 
                        bch2_bkey_val_to_text(&buf, c, k);
-                       bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
+                       bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
                        printbuf_exit(&buf);
                        return ret;
                }
@@ -996,6 +1014,7 @@ int bch2_mark_extent(struct btree_trans *trans,
 }
 
 int bch2_mark_stripe(struct btree_trans *trans,
+                    enum btree_id btree_id, unsigned level,
                     struct bkey_s_c old, struct bkey_s_c new,
                     unsigned flags)
 {
@@ -1015,7 +1034,7 @@ int bch2_mark_stripe(struct btree_trans *trans,
        if (!gc) {
                struct stripe *m = genradix_ptr(&c->stripes, idx);
 
-               if (!m || (old_s && !m->alive)) {
+               if (!m) {
                        struct printbuf buf1 = PRINTBUF;
                        struct printbuf buf2 = PRINTBUF;
 
@@ -1031,13 +1050,10 @@ int bch2_mark_stripe(struct btree_trans *trans,
                }
 
                if (!new_s) {
-                       spin_lock(&c->ec_stripes_heap_lock);
                        bch2_stripes_heap_del(c, m, idx);
-                       spin_unlock(&c->ec_stripes_heap_lock);
 
                        memset(m, 0, sizeof(*m));
                } else {
-                       m->alive        = true;
                        m->sectors      = le16_to_cpu(new_s->sectors);
                        m->algorithm    = new_s->algorithm;
                        m->nr_blocks    = new_s->nr_blocks;
@@ -1047,9 +1063,10 @@ int bch2_mark_stripe(struct btree_trans *trans,
                        for (i = 0; i < new_s->nr_blocks; i++)
                                m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
 
-                       spin_lock(&c->ec_stripes_heap_lock);
-                       bch2_stripes_heap_update(c, m, idx);
-                       spin_unlock(&c->ec_stripes_heap_lock);
+                       if (!old_s)
+                               bch2_stripes_heap_insert(c, m, idx);
+                       else
+                               bch2_stripes_heap_update(c, m, idx);
                }
        } else {
                struct gc_stripe *m =
@@ -1058,7 +1075,7 @@ int bch2_mark_stripe(struct btree_trans *trans,
                if (!m) {
                        bch_err(c, "error allocating memory for gc_stripes, idx %llu",
                                idx);
-                       return -ENOMEM;
+                       return -BCH_ERR_ENOMEM_mark_stripe;
                }
                /*
                 * This will be wrong when we bring back runtime gc: we should
@@ -1103,6 +1120,7 @@ int bch2_mark_stripe(struct btree_trans *trans,
 }
 
 int bch2_mark_inode(struct btree_trans *trans,
+                   enum btree_id btree_id, unsigned level,
                    struct bkey_s_c old, struct bkey_s_c new,
                    unsigned flags)
 {
@@ -1111,10 +1129,10 @@ int bch2_mark_inode(struct btree_trans *trans,
        u64 journal_seq = trans->journal_res.seq;
 
        if (flags & BTREE_TRIGGER_INSERT) {
-               struct bch_inode_v2 *v = (struct bch_inode_v2 *) new.v;
+               struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
 
                BUG_ON(!journal_seq);
-               BUG_ON(new.k->type != KEY_TYPE_inode_v2);
+               BUG_ON(new.k->type != KEY_TYPE_inode_v3);
 
                v->bi_journal_seq = cpu_to_le64(journal_seq);
        }
@@ -1134,11 +1152,12 @@ int bch2_mark_inode(struct btree_trans *trans,
 }
 
 int bch2_mark_reservation(struct btree_trans *trans,
+                         enum btree_id btree_id, unsigned level,
                          struct bkey_s_c old, struct bkey_s_c new,
                          unsigned flags)
 {
        struct bch_fs *c = trans->c;
-       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
+       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
        struct bch_fs_usage __percpu *fs_usage;
        unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
        s64 sectors = (s64) k.k->size;
@@ -1196,28 +1215,36 @@ not_found:
                     "  missing range %llu-%llu",
                     (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
                     *idx, next_idx)) {
-               struct bkey_i_error new;
+               struct bkey_i_error *new;
 
-               bkey_init(&new.k);
-               new.k.type      = KEY_TYPE_error;
-               new.k.p         = bkey_start_pos(p.k);
-               new.k.p.offset += *idx - start;
-               bch2_key_resize(&new.k, next_idx - *idx);
-               ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new.k_i);
+               new = bch2_trans_kmalloc(trans, sizeof(*new));
+               ret = PTR_ERR_OR_ZERO(new);
+               if (ret)
+                       goto err;
+
+               bkey_init(&new->k);
+               new->k.type     = KEY_TYPE_error;
+               new->k.p                = bkey_start_pos(p.k);
+               new->k.p.offset += *idx - start;
+               bch2_key_resize(&new->k, next_idx - *idx);
+               ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new->k_i,
+                                         BTREE_TRIGGER_NORUN);
        }
 
        *idx = next_idx;
+err:
 fsck_err:
        printbuf_exit(&buf);
        return ret;
 }
 
 int bch2_mark_reflink_p(struct btree_trans *trans,
+                       enum btree_id btree_id, unsigned level,
                        struct bkey_s_c old, struct bkey_s_c new,
                        unsigned flags)
 {
        struct bch_fs *c = trans->c;
-       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
+       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
        struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
        struct reflink_gc *ref;
        size_t l, r, m;
@@ -1251,33 +1278,45 @@ int bch2_mark_reflink_p(struct btree_trans *trans,
        return ret;
 }
 
-static noinline __cold
-void fs_usage_apply_warn(struct btree_trans *trans,
-                        unsigned disk_res_sectors,
-                        s64 should_not_have_added)
+void bch2_trans_fs_usage_revert(struct btree_trans *trans,
+                               struct replicas_delta_list *deltas)
 {
        struct bch_fs *c = trans->c;
-       struct btree_insert_entry *i;
-       struct printbuf buf = PRINTBUF;
+       struct bch_fs_usage *dst;
+       struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
+       s64 added = 0;
+       unsigned i;
 
-       bch_err(c, "disk usage increased %lli more than %u sectors reserved",
-               should_not_have_added, disk_res_sectors);
+       percpu_down_read(&c->mark_lock);
+       preempt_disable();
+       dst = fs_usage_ptr(c, trans->journal_res.seq, false);
+
+       /* revert changes: */
+       for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
+               switch (d->r.data_type) {
+               case BCH_DATA_btree:
+               case BCH_DATA_user:
+               case BCH_DATA_parity:
+                       added += d->delta;
+               }
+               BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
+       }
 
-       trans_for_each_update(trans, i) {
-               struct bkey_s_c old = { &i->old_k, i->old_v };
+       dst->nr_inodes -= deltas->nr_inodes;
 
-               pr_err("while inserting");
-               printbuf_reset(&buf);
-               bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(i->k));
-               pr_err("  %s", buf.buf);
-               pr_err("overlapping with");
-               printbuf_reset(&buf);
-               bch2_bkey_val_to_text(&buf, c, old);
-               pr_err("  %s", buf.buf);
+       for (i = 0; i < BCH_REPLICAS_MAX; i++) {
+               added                           -= deltas->persistent_reserved[i];
+               dst->reserved                   -= deltas->persistent_reserved[i];
+               dst->persistent_reserved[i]     -= deltas->persistent_reserved[i];
        }
 
-       __WARN();
-       printbuf_exit(&buf);
+       if (added > 0) {
+               trans->disk_res->sectors += added;
+               this_cpu_add(*c->online_reserved, added);
+       }
+
+       preempt_enable();
+       percpu_up_read(&c->mark_lock);
 }
 
 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
@@ -1344,7 +1383,9 @@ int bch2_trans_fs_usage_apply(struct btree_trans *trans,
        percpu_up_read(&c->mark_lock);
 
        if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
-               fs_usage_apply_warn(trans, disk_res_sectors, should_not_have_added);
+               bch2_trans_inconsistent(trans,
+                                       "disk usage increased %lli more than %u sectors reserved)",
+                                       should_not_have_added, disk_res_sectors);
        return 0;
 need_mark:
        /* revert changes: */
@@ -1358,7 +1399,7 @@ need_mark:
 
 /* trans_mark: */
 
-static int bch2_trans_mark_pointer(struct btree_trans *trans,
+static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
                                   enum btree_id btree_id, unsigned level,
                                   struct bkey_s_c k, struct extent_ptr_decoded p,
                                   unsigned flags)
@@ -1366,17 +1407,17 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
        bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
        struct btree_iter iter;
        struct bkey_i_alloc_v4 *a;
-       struct bpos bucket_pos;
+       struct bpos bucket;
        struct bch_backpointer bp;
        s64 sectors;
        int ret;
 
-       bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket_pos, &bp);
+       bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
        sectors = bp.bucket_len;
        if (!insert)
                sectors = -sectors;
 
-       a = bch2_trans_start_alloc_update(trans, &iter, bucket_pos);
+       a = bch2_trans_start_alloc_update(trans, &iter, bucket);
        if (IS_ERR(a))
                return PTR_ERR(a);
 
@@ -1387,9 +1428,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
                goto err;
 
        if (!p.ptr.cached) {
-               ret = insert
-                       ? bch2_bucket_backpointer_add(trans, a, bp, k)
-                       : bch2_bucket_backpointer_del(trans, a, bp, k);
+               ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
                if (ret)
                        goto err;
        }
@@ -1405,28 +1444,22 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
                        s64 sectors, enum bch_data_type data_type)
 {
        struct btree_iter iter;
-       struct bkey_s_c k;
        struct bkey_i_stripe *s;
        struct bch_replicas_padded r;
        int ret = 0;
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes, POS(0, p.ec.idx),
-                            BTREE_ITER_INTENT|
-                            BTREE_ITER_WITH_UPDATES);
-       k = bch2_btree_iter_peek_slot(&iter);
-       ret = bkey_err(k);
-       if (ret)
-               goto err;
-
-       if (k.k->type != KEY_TYPE_stripe) {
-               bch2_trans_inconsistent(trans,
+       s = bch2_bkey_get_mut_typed(trans, &iter,
+                       BTREE_ID_stripes, POS(0, p.ec.idx),
+                       BTREE_ITER_WITH_UPDATES, stripe);
+       ret = PTR_ERR_OR_ZERO(s);
+       if (unlikely(ret)) {
+               bch2_trans_inconsistent_on(ret == -ENOENT, trans,
                        "pointer to nonexistent stripe %llu",
                        (u64) p.ec.idx);
-               ret = -EIO;
                goto err;
        }
 
-       if (!bch2_ptr_matches_stripe(bkey_s_c_to_stripe(k).v, p)) {
+       if (!bch2_ptr_matches_stripe(&s->v, p)) {
                bch2_trans_inconsistent(trans,
                        "stripe pointer doesn't match stripe %llu",
                        (u64) p.ec.idx);
@@ -1434,20 +1467,10 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
                goto err;
        }
 
-       s = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-       ret = PTR_ERR_OR_ZERO(s);
-       if (ret)
-               goto err;
-
-       bkey_reassemble(&s->k_i, k);
        stripe_blockcount_set(&s->v, p.ec.block,
                stripe_blockcount_get(&s->v, p.ec.block) +
                sectors);
 
-       ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
-       if (ret)
-               goto err;
-
        bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
        r.e.data_type = data_type;
        update_replicas_list(trans, &r.e, sectors);
@@ -1538,7 +1561,7 @@ static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
        if (IS_ERR(a))
                return PTR_ERR(a);
 
-       ret = check_bucket_ref(c, s.s_c, ptr, sectors, data_type,
+       ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
                               a->v.gen, a->v.data_type,
                               a->v.dirty_sectors, a->v.cached_sectors);
        if (ret)
@@ -1568,6 +1591,7 @@ static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
 
                a->v.stripe             = s.k->p.offset;
                a->v.stripe_redundancy  = s.v->nr_redundant;
+               a->v.data_type          = BCH_DATA_stripe;
        } else {
                if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
                                               a->v.stripe_redundancy != s.v->nr_redundant, trans,
@@ -1580,6 +1604,7 @@ static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
 
                a->v.stripe             = 0;
                a->v.stripe_redundancy  = 0;
+               a->v.data_type          = alloc_data_type(a->v, BCH_DATA_user);
        }
 
        a->v.dirty_sectors += sectors;
@@ -1714,29 +1739,20 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 {
        struct bch_fs *c = trans->c;
        struct btree_iter iter;
-       struct bkey_s_c k;
-       struct bkey_i *n;
+       struct bkey_i *k;
        __le64 *refcount;
        int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
        struct printbuf buf = PRINTBUF;
        int ret;
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, *idx),
-                            BTREE_ITER_INTENT|
-                            BTREE_ITER_WITH_UPDATES);
-       k = bch2_btree_iter_peek_slot(&iter);
-       ret = bkey_err(k);
+       k = bch2_bkey_get_mut_noupdate(trans, &iter,
+                       BTREE_ID_reflink, POS(0, *idx),
+                       BTREE_ITER_WITH_UPDATES);
+       ret = PTR_ERR_OR_ZERO(k);
        if (ret)
                goto err;
 
-       n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-       ret = PTR_ERR_OR_ZERO(n);
-       if (ret)
-               goto err;
-
-       bkey_reassemble(n, k);
-
-       refcount = bkey_refcount(n);
+       refcount = bkey_refcount(k);
        if (!refcount) {
                bch2_bkey_val_to_text(&buf, c, p.s_c);
                bch2_trans_inconsistent(trans,
@@ -1760,12 +1776,12 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
                u64 pad;
 
                pad = max_t(s64, le32_to_cpu(v->front_pad),
-                           le64_to_cpu(v->idx) - bkey_start_offset(k.k));
+                           le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
                BUG_ON(pad > U32_MAX);
                v->front_pad = cpu_to_le32(pad);
 
                pad = max_t(s64, le32_to_cpu(v->back_pad),
-                           k.k->p.offset - p.k->size - le64_to_cpu(v->idx));
+                           k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
                BUG_ON(pad > U32_MAX);
                v->back_pad = cpu_to_le32(pad);
        }
@@ -1773,11 +1789,11 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
        le64_add_cpu(refcount, add);
 
        bch2_btree_iter_set_pos_to_extent_start(&iter);
-       ret = bch2_trans_update(trans, &iter, n, 0);
+       ret = bch2_trans_update(trans, &iter, k, 0);
        if (ret)
                goto err;
 
-       *idx = k.k->p.offset;
+       *idx = k->k.p.offset;
 err:
        bch2_trans_iter_exit(trans, &iter);
        printbuf_exit(&buf);
@@ -1833,7 +1849,7 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
        if (IS_ERR(a))
                return PTR_ERR(a);
 
-       if (a->v.data_type && a->v.data_type != type) {
+       if (a->v.data_type && type && a->v.data_type != type) {
                bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
                        "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
                        "while marking %s",
@@ -1861,7 +1877,7 @@ int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
                                    enum bch_data_type type,
                                    unsigned sectors)
 {
-       return __bch2_trans_do(trans, NULL, NULL, 0,
+       return commit_do(trans, NULL, NULL, 0,
                        __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
 }
 
@@ -1939,15 +1955,14 @@ static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
 
 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
 {
-       return bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
-                       __bch2_trans_mark_dev_sb(&trans, ca));
+       return bch2_trans_run(c, __bch2_trans_mark_dev_sb(&trans, ca));
 }
 
 /* Disk reservations: */
 
 #define SECTORS_CACHE  1024
 
-int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
+int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
                              u64 sectors, int flags)
 {
        struct bch_fs_pcpu *pcpu;
@@ -2000,7 +2015,7 @@ recalculate:
                ret = 0;
        } else {
                atomic64_set(&c->sectors_available, sectors_available);
-               ret = -ENOSPC;
+               ret = -BCH_ERR_ENOSPC_disk_reservation;
        }
 
        mutex_unlock(&c->sectors_available_lock);
@@ -2024,15 +2039,21 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
        struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
        unsigned long *buckets_nouse = NULL;
        bool resize = ca->bucket_gens != NULL;
-       int ret = -ENOMEM;
+       int ret;
 
        if (!(bucket_gens       = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
-                                           GFP_KERNEL|__GFP_ZERO)) ||
-           (c->opts.buckets_nouse &&
+                                           GFP_KERNEL|__GFP_ZERO))) {
+               ret = -BCH_ERR_ENOMEM_bucket_gens;
+               goto err;
+       }
+
+       if ((c->opts.buckets_nouse &&
             !(buckets_nouse    = kvpmalloc(BITS_TO_LONGS(nbuckets) *
                                            sizeof(unsigned long),
-                                           GFP_KERNEL|__GFP_ZERO))))
+                                           GFP_KERNEL|__GFP_ZERO)))) {
+               ret = -BCH_ERR_ENOMEM_buckets_nouse;
                goto err;
+       }
 
        bucket_gens->first_bucket = ca->mi.first_bucket;
        bucket_gens->nbuckets   = nbuckets;
@@ -2102,13 +2123,13 @@ int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
 
        ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
        if (!ca->usage_base)
-               return -ENOMEM;
+               return -BCH_ERR_ENOMEM_usage_init;
 
        for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
                ca->usage[i] = alloc_percpu(struct bch_dev_usage);
                if (!ca->usage[i])
-                       return -ENOMEM;
+                       return -BCH_ERR_ENOMEM_usage_init;
        }
 
-       return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);;
+       return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
 }