]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/buckets.c
Delete old bcachefs.5 from makefile
[bcachefs-tools-debian] / libbcachefs / buckets.c
index d5ec4d727d0e56ad2e35cdc8832b781aa1358dc9..762366f58519e922c92e0d4e5a26aedafcf8039f 100644 (file)
@@ -13,6 +13,7 @@
 #include "buckets.h"
 #include "ec.h"
 #include "error.h"
+#include "inode.h"
 #include "movinggc.h"
 #include "recovery.h"
 #include "reflink.h"
@@ -116,6 +117,8 @@ static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
                                                  unsigned journal_seq,
                                                  bool gc)
 {
+       BUG_ON(!gc && !journal_seq);
+
        return this_cpu_ptr(gc
                            ? ca->usage_gc
                            : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
@@ -141,6 +144,9 @@ static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c,
                                                unsigned journal_seq,
                                                bool gc)
 {
+       percpu_rwsem_assert_held(&c->mark_lock);
+       BUG_ON(!gc && !journal_seq);
+
        return this_cpu_ptr(gc
                            ? c->usage_gc
                            : c->usage[journal_seq & JOURNAL_BUF_MASK]);
@@ -359,7 +365,12 @@ static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
        struct bch_fs_usage *fs_usage;
        struct bch_dev_usage *u;
 
-       percpu_rwsem_assert_held(&c->mark_lock);
+       /*
+        * Hack for bch2_fs_initialize path, where we're first marking sb and
+        * journal non-transactionally:
+        */
+       if (!journal_seq && !test_bit(BCH_FS_INITIALIZED, &c->flags))
+               journal_seq = 1;
 
        preempt_disable();
        fs_usage = fs_usage_ptr(c, journal_seq, gc);
@@ -406,25 +417,48 @@ static inline int __update_replicas(struct bch_fs *c,
        return 0;
 }
 
-static inline int update_replicas(struct bch_fs *c,
+static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
                        struct bch_replicas_entry *r, s64 sectors,
                        unsigned journal_seq, bool gc)
 {
        struct bch_fs_usage __percpu *fs_usage;
-       int idx = bch2_replicas_entry_idx(c, r);
+       int idx, ret = 0;
+       char buf[200];
 
-       if (idx < 0)
-               return -1;
+       percpu_down_read(&c->mark_lock);
+
+       idx = bch2_replicas_entry_idx(c, r);
+       if (idx < 0 &&
+           (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
+            fsck_err(c, "no replicas entry\n"
+                     "  while marking %s",
+                     (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))) {
+               percpu_up_read(&c->mark_lock);
+               ret = bch2_mark_replicas(c, r);
+               if (ret)
+                       return ret;
+
+               percpu_down_read(&c->mark_lock);
+               idx = bch2_replicas_entry_idx(c, r);
+       }
+       if (idx < 0) {
+               ret = -1;
+               goto err;
+       }
 
        preempt_disable();
        fs_usage = fs_usage_ptr(c, journal_seq, gc);
        fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
        fs_usage->replicas[idx]         += sectors;
        preempt_enable();
-       return 0;
+err:
+fsck_err:
+       percpu_up_read(&c->mark_lock);
+       return ret;
 }
 
 static inline int update_cached_sectors(struct bch_fs *c,
+                       struct bkey_s_c k,
                        unsigned dev, s64 sectors,
                        unsigned journal_seq, bool gc)
 {
@@ -432,7 +466,7 @@ static inline int update_cached_sectors(struct bch_fs *c,
 
        bch2_replicas_entry_cached(&r.e, dev);
 
-       return update_replicas(c, &r.e, sectors, journal_seq, gc);
+       return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
 }
 
 static struct replicas_delta_list *
@@ -524,19 +558,21 @@ void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
        BUG_ON(owned_by_allocator == old.owned_by_allocator);
 }
 
-static int bch2_mark_alloc(struct bch_fs *c,
+static int bch2_mark_alloc(struct btree_trans *trans,
                           struct bkey_s_c old, struct bkey_s_c new,
-                          u64 journal_seq, unsigned flags)
+                          unsigned flags)
 {
        bool gc = flags & BTREE_TRIGGER_GC;
+       u64 journal_seq = trans->journal_res.seq;
+       struct bch_fs *c = trans->c;
        struct bkey_alloc_unpacked u;
        struct bch_dev *ca;
        struct bucket *g;
        struct bucket_mark old_m, m;
+       int ret = 0;
 
        /* We don't do anything for deletions - do we?: */
-       if (new.k->type != KEY_TYPE_alloc &&
-           new.k->type != KEY_TYPE_alloc_v2)
+       if (!bkey_is_alloc(new.k))
                return 0;
 
        /*
@@ -546,11 +582,21 @@ static int bch2_mark_alloc(struct bch_fs *c,
            !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
                return 0;
 
+       if (flags & BTREE_TRIGGER_INSERT) {
+               struct bch_alloc_v3 *v = (struct bch_alloc_v3 *) new.v;
+
+               BUG_ON(!journal_seq);
+               BUG_ON(new.k->type != KEY_TYPE_alloc_v3);
+
+               v->journal_seq = cpu_to_le64(journal_seq);
+       }
+
        ca = bch_dev_bkey_exists(c, new.k->p.inode);
 
        if (new.k->p.offset >= ca->mi.nbuckets)
                return 0;
 
+       percpu_down_read(&c->mark_lock);
        g = __bucket(ca, new.k->p.offset, gc);
        u = bch2_alloc_unpack(new);
 
@@ -575,6 +621,7 @@ static int bch2_mark_alloc(struct bch_fs *c,
        g->gen_valid            = 1;
        g->stripe               = u.stripe;
        g->stripe_redundancy    = u.stripe_redundancy;
+       percpu_up_read(&c->mark_lock);
 
        /*
         * need to know if we're getting called from the invalidate path or
@@ -583,10 +630,12 @@ static int bch2_mark_alloc(struct bch_fs *c,
 
        if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
            old_m.cached_sectors) {
-               if (update_cached_sectors(c, ca->dev_idx, -old_m.cached_sectors,
-                                         journal_seq, gc)) {
+               ret = update_cached_sectors(c, new, ca->dev_idx,
+                                           -old_m.cached_sectors,
+                                           journal_seq, gc);
+               if (ret) {
                        bch2_fs_fatal_error(c, "bch2_mark_alloc(): no replicas entry while updating cached sectors");
-                       return -1;
+                       return ret;
                }
 
                trace_invalidate(ca, bucket_to_sector(ca, new.k->p.offset),
@@ -673,7 +722,8 @@ static s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
                : sectors;
 }
 
-static int check_bucket_ref(struct bch_fs *c, struct bkey_s_c k,
+static int check_bucket_ref(struct bch_fs *c,
+                           struct bkey_s_c k,
                            const struct bch_extent_ptr *ptr,
                            s64 sectors, enum bch_data_type ptr_data_type,
                            u8 bucket_gen, u8 bucket_data_type,
@@ -747,54 +797,71 @@ static int check_bucket_ref(struct bch_fs *c, struct bkey_s_c k,
        return 0;
 }
 
-static int mark_stripe_bucket(struct bch_fs *c, struct bkey_s_c k,
-                            unsigned ptr_idx,
-                            u64 journal_seq, unsigned flags)
+static int mark_stripe_bucket(struct btree_trans *trans,
+                             struct bkey_s_c k,
+                             unsigned ptr_idx,
+                             u64 journal_seq, unsigned flags)
 {
+       struct bch_fs *c = trans->c;
        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;
+       s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
        const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
        bool gc = flags & BTREE_TRIGGER_GC;
        struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
-       struct bucket *g = PTR_BUCKET(ca, ptr, gc);
+       struct bucket *g;
        struct bucket_mark new, old;
        char buf[200];
-       int ret;
+       int ret = 0;
+
+       /* * XXX doesn't handle deletion */
 
-       if (g->stripe && g->stripe != k.k->p.offset) {
+       percpu_down_read(&c->mark_lock);
+       g = PTR_BUCKET(ca, ptr, gc);
+
+       if (g->mark.dirty_sectors ||
+           (g->stripe && g->stripe != k.k->p.offset)) {
                bch2_fs_inconsistent(c,
                              "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
                              ptr->dev, PTR_BUCKET_NR(ca, ptr), g->mark.gen,
                              (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
-               return -EINVAL;
+               ret = -EINVAL;
+               goto err;
        }
 
        old = bucket_cmpxchg(g, new, ({
-               ret = check_bucket_ref(c, k, ptr, 0, 0, new.gen, new.data_type,
+               ret = check_bucket_ref(c, k, ptr, sectors, data_type,
+                                      new.gen, new.data_type,
                                       new.dirty_sectors, new.cached_sectors);
                if (ret)
-                       return ret;
+                       goto err;
 
-               if (parity) {
-                       new.data_type           = BCH_DATA_parity;
-                       new.dirty_sectors       = le16_to_cpu(s->sectors);
-               }
+               new.dirty_sectors += sectors;
+               if (data_type)
+                       new.data_type           = data_type;
 
                if (journal_seq) {
                        new.journal_seq_valid   = 1;
                        new.journal_seq         = journal_seq;
                }
+
+               new.stripe = true;
        }));
 
        g->stripe               = k.k->p.offset;
        g->stripe_redundancy    = s->nr_redundant;
 
        bch2_dev_usage_update(c, ca, old, new, journal_seq, gc);
+err:
+       percpu_up_read(&c->mark_lock);
+
        return 0;
 }
 
-static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k,
+static int __mark_pointer(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 bucket_gen, u8 *bucket_data_type,
@@ -803,7 +870,7 @@ static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k,
        u16 *dst_sectors = !ptr->cached
                ? dirty_sectors
                : cached_sectors;
-       int ret = check_bucket_ref(c, k, ptr, sectors, ptr_data_type,
+       int ret = check_bucket_ref(trans->c, k, ptr, sectors, ptr_data_type,
                                   bucket_gen, *bucket_data_type,
                                   *dirty_sectors, *cached_sectors);
 
@@ -816,30 +883,37 @@ static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k,
        return 0;
 }
 
-static int bch2_mark_pointer(struct bch_fs *c, struct bkey_s_c k,
+static int bch2_mark_pointer(struct btree_trans *trans,
+                            struct bkey_s_c k,
                             struct extent_ptr_decoded p,
                             s64 sectors, enum bch_data_type data_type,
-                            u64 journal_seq, unsigned flags)
+                            unsigned flags)
 {
        bool gc = flags & BTREE_TRIGGER_GC;
+       u64 journal_seq = trans->journal_res.seq;
+       struct bch_fs *c = trans->c;
        struct bucket_mark old, new;
        struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
        struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc);
        u8 bucket_data_type;
        u64 v;
-       int ret;
+       int ret = 0;
+
+       percpu_down_read(&c->mark_lock);
+       g = PTR_BUCKET(ca, &p.ptr, gc);
 
        v = atomic64_read(&g->_mark.v);
        do {
                new.v.counter = old.v.counter = v;
                bucket_data_type = new.data_type;
 
-               ret = __mark_pointer(c, k, &p.ptr, sectors, data_type, new.gen,
+               ret = __mark_pointer(trans, k, &p.ptr, sectors,
+                                    data_type, new.gen,
                                     &bucket_data_type,
                                     &new.dirty_sectors,
                                     &new.cached_sectors);
                if (ret)
-                       return ret;
+                       goto err;
 
                new.data_type = bucket_data_type;
 
@@ -859,60 +933,61 @@ static int bch2_mark_pointer(struct bch_fs *c, struct bkey_s_c k,
        bch2_dev_usage_update(c, ca, old, new, journal_seq, gc);
 
        BUG_ON(!gc && bucket_became_unavailable(old, new));
+err:
+       percpu_up_read(&c->mark_lock);
 
-       return 0;
+       return ret;
 }
 
-static int bch2_mark_stripe_ptr(struct bch_fs *c,
+static int bch2_mark_stripe_ptr(struct btree_trans *trans,
+                               struct bkey_s_c k,
                                struct bch_extent_stripe_ptr p,
                                enum bch_data_type data_type,
                                s64 sectors,
-                               unsigned journal_seq, unsigned flags)
+                               unsigned flags)
 {
        bool gc = flags & BTREE_TRIGGER_GC;
+       struct bch_fs *c = trans->c;
        struct bch_replicas_padded r;
-       struct stripe *m;
-       unsigned i, blocks_nonempty = 0;
 
-       m = genradix_ptr(&c->stripes[gc], p.idx);
+       if (!gc) {
+               BUG();
+       } else {
+               struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
 
-       spin_lock(&c->ec_stripes_heap_lock);
+               if (!m)
+                       return -ENOMEM;
 
-       if (!m || !m->alive) {
-               spin_unlock(&c->ec_stripes_heap_lock);
-               bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
-                                   (u64) p.idx);
-               bch2_inconsistent_error(c);
-               return -EIO;
-       }
+               spin_lock(&c->ec_stripes_heap_lock);
 
-       m->block_sectors[p.block] += sectors;
+               if (!m || !m->alive) {
+                       spin_unlock(&c->ec_stripes_heap_lock);
+                       bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
+                                           (u64) p.idx);
+                       bch2_inconsistent_error(c);
+                       return -EIO;
+               }
 
-       r = m->r;
+               m->block_sectors[p.block] += sectors;
 
-       for (i = 0; i < m->nr_blocks; i++)
-               blocks_nonempty += m->block_sectors[i] != 0;
+               r = m->r;
+               spin_unlock(&c->ec_stripes_heap_lock);
 
-       if (m->blocks_nonempty != blocks_nonempty) {
-               m->blocks_nonempty = blocks_nonempty;
-               if (!gc)
-                       bch2_stripes_heap_update(c, m, p.idx);
+               r.e.data_type = data_type;
+               update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, gc);
        }
 
-       spin_unlock(&c->ec_stripes_heap_lock);
-
-       r.e.data_type = data_type;
-       update_replicas(c, &r.e, sectors, journal_seq, gc);
-
        return 0;
 }
 
-static int bch2_mark_extent(struct bch_fs *c,
+static int bch2_mark_extent(struct btree_trans *trans,
                            struct bkey_s_c old, struct bkey_s_c new,
-                           unsigned journal_seq, unsigned flags)
+                           unsigned flags)
 {
        bool gc = flags & BTREE_TRIGGER_GC;
-       struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
+       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_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
        const union bch_extent_entry *entry;
        struct extent_ptr_decoded p;
@@ -927,9 +1002,6 @@ static int bch2_mark_extent(struct bch_fs *c,
        bool stale;
        int ret;
 
-       BUG_ON((flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)) ==
-              (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE));
-
        r.e.data_type   = data_type;
        r.e.nr_devs     = 0;
        r.e.nr_required = 1;
@@ -940,27 +1012,28 @@ static int bch2_mark_extent(struct bch_fs *c,
                if (flags & BTREE_TRIGGER_OVERWRITE)
                        disk_sectors = -disk_sectors;
 
-               ret = bch2_mark_pointer(c, k, p, disk_sectors, data_type,
-                                       journal_seq, flags);
+               ret = bch2_mark_pointer(trans, k, p, disk_sectors,
+                                       data_type, flags);
                if (ret < 0)
                        return ret;
 
                stale = ret > 0;
 
                if (p.ptr.cached) {
-                       if (!stale)
-                               if (update_cached_sectors(c, p.ptr.dev, disk_sectors,
-                                                         journal_seq, gc)) {
+                       if (!stale) {
+                               ret = update_cached_sectors(c, k, p.ptr.dev,
+                                               disk_sectors, journal_seq, gc);
+                               if (ret) {
                                        bch2_fs_fatal_error(c, "bch2_mark_extent(): no replicas entry while updating cached sectors");
-                                       return -1;
-
+                                       return ret;
                                }
+                       }
                } else if (!p.has_ec) {
                        dirty_sectors          += disk_sectors;
                        r.e.devs[r.e.nr_devs++] = p.ptr.dev;
                } else {
-                       ret = bch2_mark_stripe_ptr(c, p.ec, data_type,
-                                       disk_sectors, journal_seq, flags);
+                       ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
+                                       disk_sectors, flags);
                        if (ret)
                                return ret;
 
@@ -974,119 +1047,153 @@ static int bch2_mark_extent(struct bch_fs *c,
        }
 
        if (r.e.nr_devs) {
-               if (update_replicas(c, &r.e, dirty_sectors, journal_seq, gc)) {
+               ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, gc);
+               if (ret) {
                        char buf[200];
 
                        bch2_bkey_val_to_text(&PBUF(buf), c, k);
                        bch2_fs_fatal_error(c, "no replicas entry for %s", buf);
-                       return -1;
+                       return ret;
                }
        }
 
        return 0;
 }
 
-static int bch2_mark_stripe(struct bch_fs *c,
-                       struct bkey_s_c old, struct bkey_s_c new,
-                       u64 journal_seq, unsigned flags)
+static int bch2_mark_stripe(struct btree_trans *trans,
+                           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;
+       struct bch_fs *c = trans->c;
        size_t idx = new.k->p.offset;
        const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
                ? bkey_s_c_to_stripe(old).v : NULL;
        const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
                ? bkey_s_c_to_stripe(new).v : NULL;
-       struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
        unsigned i;
        int ret;
 
        BUG_ON(gc && old_s);
 
-       if (!m || (old_s && !m->alive)) {
-               bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
-                                   idx);
-               bch2_inconsistent_error(c);
-               return -1;
-       }
+       if (!gc) {
+               struct stripe *m = genradix_ptr(&c->stripes, idx);
 
-       if (!new_s) {
-               spin_lock(&c->ec_stripes_heap_lock);
-               bch2_stripes_heap_del(c, m, idx);
-               spin_unlock(&c->ec_stripes_heap_lock);
+               if (!m || (old_s && !m->alive)) {
+                       char buf1[200], buf2[200];
 
-               memset(m, 0, sizeof(*m));
+                       bch2_bkey_val_to_text(&PBUF(buf1), c, old);
+                       bch2_bkey_val_to_text(&PBUF(buf2), c, new);
+                       bch_err_ratelimited(c, "error marking nonexistent stripe %zu while marking\n"
+                                           "old %s\n"
+                                           "new %s", idx, buf1, buf2);
+                       bch2_inconsistent_error(c);
+                       return -1;
+               }
+
+               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;
+                       m->nr_redundant = new_s->nr_redundant;
+                       m->blocks_nonempty = 0;
+
+                       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);
+               }
        } else {
+               struct gc_stripe *m = genradix_ptr(&c->gc_stripes, idx);
+
+               /*
+                * This will be wrong when we bring back runtime gc: we should
+                * be unmarking the old key and then marking the new key
+                */
                m->alive        = true;
                m->sectors      = le16_to_cpu(new_s->sectors);
-               m->algorithm    = new_s->algorithm;
                m->nr_blocks    = new_s->nr_blocks;
                m->nr_redundant = new_s->nr_redundant;
-               m->blocks_nonempty = 0;
-
-               for (i = 0; i < new_s->nr_blocks; i++) {
-                       m->block_sectors[i] =
-                               stripe_blockcount_get(new_s, i);
-                       m->blocks_nonempty += !!m->block_sectors[i];
 
+               for (i = 0; i < new_s->nr_blocks; i++)
                        m->ptrs[i] = new_s->ptrs[i];
-               }
 
                bch2_bkey_to_replicas(&m->r.e, new);
 
-               if (!gc) {
-                       spin_lock(&c->ec_stripes_heap_lock);
-                       bch2_stripes_heap_update(c, m, idx);
-                       spin_unlock(&c->ec_stripes_heap_lock);
-               }
-       }
-
-       if (gc) {
                /*
                 * gc recalculates this field from stripe ptr
                 * references:
                 */
                memset(m->block_sectors, 0, sizeof(m->block_sectors));
-               m->blocks_nonempty = 0;
 
                for (i = 0; i < new_s->nr_blocks; i++) {
-                       ret = mark_stripe_bucket(c, new, i, journal_seq, flags);
+                       ret = mark_stripe_bucket(trans, new, i, journal_seq, flags);
                        if (ret)
                                return ret;
                }
 
-               if (update_replicas(c, &m->r.e,
-                                   ((s64) m->sectors * m->nr_redundant),
-                                   journal_seq, gc)) {
+               ret = update_replicas(c, new, &m->r.e,
+                                     ((s64) m->sectors * m->nr_redundant),
+                                     journal_seq, gc);
+               if (ret) {
                        char buf[200];
 
                        bch2_bkey_val_to_text(&PBUF(buf), c, new);
                        bch2_fs_fatal_error(c, "no replicas entry for %s", buf);
-                       return -1;
+                       return ret;
                }
        }
 
        return 0;
 }
 
-static int bch2_mark_inode(struct bch_fs *c,
-                       struct bkey_s_c old, struct bkey_s_c new,
-                       u64 journal_seq, unsigned flags)
+static int bch2_mark_inode(struct btree_trans *trans,
+                          struct bkey_s_c old, struct bkey_s_c new,
+                          unsigned flags)
 {
+       struct bch_fs *c = trans->c;
        struct bch_fs_usage __percpu *fs_usage;
+       u64 journal_seq = trans->journal_res.seq;
 
-       preempt_disable();
-       fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
-       fs_usage->nr_inodes += new.k->type == KEY_TYPE_inode;
-       fs_usage->nr_inodes -= old.k->type == KEY_TYPE_inode;
-       preempt_enable();
+       if (flags & BTREE_TRIGGER_INSERT) {
+               struct bch_inode_v2 *v = (struct bch_inode_v2 *) new.v;
+
+               BUG_ON(!journal_seq);
+               BUG_ON(new.k->type != KEY_TYPE_inode_v2);
+
+               v->bi_journal_seq = cpu_to_le64(journal_seq);
+       }
+
+       if (flags & BTREE_TRIGGER_GC) {
+               percpu_down_read(&c->mark_lock);
+               preempt_disable();
+
+               fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
+               fs_usage->nr_inodes += bkey_is_inode(new.k);
+               fs_usage->nr_inodes -= bkey_is_inode(old.k);
+
+               preempt_enable();
+               percpu_up_read(&c->mark_lock);
+       }
        return 0;
 }
 
-static int bch2_mark_reservation(struct bch_fs *c,
-                       struct bkey_s_c old, struct bkey_s_c new,
-                       u64 journal_seq, unsigned flags)
+static int bch2_mark_reservation(struct btree_trans *trans,
+                                struct bkey_s_c old, struct bkey_s_c new,
+                                unsigned flags)
 {
-       struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
+       struct bch_fs *c = trans->c;
+       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;
@@ -1095,93 +1202,86 @@ static int bch2_mark_reservation(struct bch_fs *c,
                sectors = -sectors;
        sectors *= replicas;
 
+       percpu_down_read(&c->mark_lock);
        preempt_disable();
-       fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
+
+       fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
        replicas = clamp_t(unsigned, replicas, 1,
                           ARRAY_SIZE(fs_usage->persistent_reserved));
 
        fs_usage->reserved                              += sectors;
        fs_usage->persistent_reserved[replicas - 1]     += sectors;
+
        preempt_enable();
+       percpu_up_read(&c->mark_lock);
 
        return 0;
 }
 
 static s64 __bch2_mark_reflink_p(struct bch_fs *c, struct bkey_s_c_reflink_p p,
-                                u64 idx, unsigned flags, size_t *r_idx)
+                                u64 *idx, unsigned flags, size_t r_idx)
 {
        struct reflink_gc *r;
        int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
        s64 ret = 0;
 
-       while (*r_idx < c->reflink_gc_nr) {
-               r = genradix_ptr(&c->reflink_gc_table, *r_idx);
-               BUG_ON(!r);
-
-               if (idx < r->offset)
-                       break;
-               (*r_idx)++;
-       }
+       if (r_idx >= c->reflink_gc_nr)
+               goto not_found;
 
-       if (*r_idx >= c->reflink_gc_nr ||
-           idx < r->offset - r->size) {
-               ret = p.k->size;
+       r = genradix_ptr(&c->reflink_gc_table, r_idx);
+       if (*idx < r->offset - r->size)
                goto not_found;
-       }
 
        BUG_ON((s64) r->refcount + add < 0);
 
        r->refcount += add;
-       return r->offset - idx;
+       *idx = r->offset;
+       return 0;
 not_found:
-       if ((flags & BTREE_TRIGGER_GC) &&
-           (flags & BTREE_TRIGGER_NOATOMIC)) {
-               /*
-                * XXX: we're replacing the entire reflink pointer with an error
-                * key, we should just be replacing the part that was missing:
-                */
-               if (fsck_err(c, "%llu:%llu len %u points to nonexistent indirect extent %llu",
-                            p.k->p.inode, p.k->p.offset, p.k->size, idx)) {
-                       struct bkey_i_error *new;
-
-                       new = kmalloc(sizeof(*new), GFP_KERNEL);
-                       if (!new) {
-                               bch_err(c, "%s: error allocating new key", __func__);
-                               return -ENOMEM;
-                       }
+       *idx = U64_MAX;
+       ret = -EIO;
 
-                       bkey_init(&new->k);
-                       new->k.type     = KEY_TYPE_error;
-                       new->k.p        = p.k->p;
-                       new->k.size     = p.k->size;
-                       ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new->k_i);
+       /*
+        * XXX: we're replacing the entire reflink pointer with an error
+        * key, we should just be replacing the part that was missing:
+        */
+       if (fsck_err(c, "%llu:%llu len %u points to nonexistent indirect extent %llu",
+                    p.k->p.inode, p.k->p.offset, p.k->size, *idx)) {
+               struct bkey_i_error *new;
 
+               new = kmalloc(sizeof(*new), GFP_KERNEL);
+               if (!new) {
+                       bch_err(c, "%s: error allocating new key", __func__);
+                       return -ENOMEM;
                }
-       } else {
-               bch2_fs_inconsistent(c,
-                                    "%llu:%llu len %u points to nonexistent indirect extent %llu",
-                                    p.k->p.inode, p.k->p.offset, p.k->size, idx);
-               bch2_inconsistent_error(c);
-               ret = -EIO;
+
+               bkey_init(&new->k);
+               new->k.type     = KEY_TYPE_error;
+               new->k.p        = p.k->p;
+               new->k.size     = p.k->size;
+               ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new->k_i);
        }
 fsck_err:
        return ret;
 }
 
-static int bch2_mark_reflink_p(struct bch_fs *c,
-                       struct bkey_s_c old, struct bkey_s_c new,
-                       u64 journal_seq, unsigned flags)
+static int bch2_mark_reflink_p(struct btree_trans *trans,
+                              struct bkey_s_c old, struct bkey_s_c new,
+                              unsigned flags)
 {
-       struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
+       struct bch_fs *c = trans->c;
+       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;
        u64 idx = le64_to_cpu(p.v->idx);
-       unsigned sectors = p.k->size;
-       s64 ret = 0;
+       u64 end = le64_to_cpu(p.v->idx) + p.k->size;
+       int ret = 0;
 
-       BUG_ON((flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)) ==
-              (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE));
+       if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) {
+               idx -= le32_to_cpu(p.v->front_pad);
+               end += le32_to_cpu(p.v->back_pad);
+       }
 
        l = 0;
        r = c->reflink_gc_nr;
@@ -1195,75 +1295,56 @@ static int bch2_mark_reflink_p(struct bch_fs *c,
                        r = m;
        }
 
-       while (sectors) {
-               ret = __bch2_mark_reflink_p(c, p, idx, flags, &l);
-               if (ret <= 0)
-                       return ret;
-
-               ret = min_t(s64, ret, sectors);
-               idx     += ret;
-               sectors -= ret;
-       }
+       while (idx < end && !ret)
+               ret = __bch2_mark_reflink_p(c, p, &idx, flags, l++);
 
-       return 0;
+       return ret;
 }
 
-static int bch2_mark_key_locked(struct bch_fs *c,
-                  struct bkey_s_c old,
-                  struct bkey_s_c new,
-                  u64 journal_seq, unsigned flags)
+int bch2_mark_key(struct btree_trans *trans,
+                 struct bkey_s_c old,
+                 struct bkey_s_c new,
+                 unsigned flags)
 {
-       struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
-
-       BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)));
+       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
 
        switch (k.k->type) {
        case KEY_TYPE_alloc:
        case KEY_TYPE_alloc_v2:
-               return bch2_mark_alloc(c, old, new, journal_seq, flags);
+       case KEY_TYPE_alloc_v3:
+               return bch2_mark_alloc(trans, old, new, flags);
        case KEY_TYPE_btree_ptr:
        case KEY_TYPE_btree_ptr_v2:
        case KEY_TYPE_extent:
        case KEY_TYPE_reflink_v:
-               return bch2_mark_extent(c, old, new, journal_seq, flags);
+               return bch2_mark_extent(trans, old, new, flags);
        case KEY_TYPE_stripe:
-               return bch2_mark_stripe(c, old, new, journal_seq, flags);
+               return bch2_mark_stripe(trans, old, new, flags);
        case KEY_TYPE_inode:
-               return bch2_mark_inode(c, old, new, journal_seq, flags);
+       case KEY_TYPE_inode_v2:
+               return bch2_mark_inode(trans, old, new, flags);
        case KEY_TYPE_reservation:
-               return bch2_mark_reservation(c, old, new, journal_seq, flags);
+               return bch2_mark_reservation(trans, old, new, flags);
        case KEY_TYPE_reflink_p:
-               return bch2_mark_reflink_p(c, old, new, journal_seq, flags);
+               return bch2_mark_reflink_p(trans, old, new, flags);
        case KEY_TYPE_snapshot:
-               return bch2_mark_snapshot(c, old, new, journal_seq, flags);
+               return bch2_mark_snapshot(trans, old, new, flags);
        default:
                return 0;
        }
 }
 
-int bch2_mark_key(struct bch_fs *c, struct bkey_s_c new, unsigned flags)
-{
-       struct bkey deleted = KEY(0, 0, 0);
-       struct bkey_s_c old = (struct bkey_s_c) { &deleted, NULL };
-       int ret;
-
-       percpu_down_read(&c->mark_lock);
-       ret = bch2_mark_key_locked(c, old, new, 0, flags);
-       percpu_up_read(&c->mark_lock);
-
-       return ret;
-}
-
 int bch2_mark_update(struct btree_trans *trans, struct btree_path *path,
                     struct bkey_i *new, unsigned flags)
 {
-       struct bch_fs           *c = trans->c;
        struct bkey             _deleted = KEY(0, 0, 0);
        struct bkey_s_c         deleted = (struct bkey_s_c) { &_deleted, NULL };
        struct bkey_s_c         old;
        struct bkey             unpacked;
        int ret;
 
+       _deleted.p = path->pos;
+
        if (unlikely(flags & BTREE_TRIGGER_NORUN))
                return 0;
 
@@ -1274,15 +1355,12 @@ int bch2_mark_update(struct btree_trans *trans, struct btree_path *path,
 
        if (old.k->type == new->k.type &&
            ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
-               ret   = bch2_mark_key_locked(c, old, bkey_i_to_s_c(new),
-                               trans->journal_res.seq,
+               ret   = bch2_mark_key(trans, old, bkey_i_to_s_c(new),
                                BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
        } else {
-               ret   = bch2_mark_key_locked(c, deleted, bkey_i_to_s_c(new),
-                               trans->journal_res.seq,
+               ret   = bch2_mark_key(trans, deleted, bkey_i_to_s_c(new),
                                BTREE_TRIGGER_INSERT|flags) ?:
-                       bch2_mark_key_locked(c, old, deleted,
-                               trans->journal_res.seq,
+                       bch2_mark_key(trans, old, deleted,
                                BTREE_TRIGGER_OVERWRITE|flags);
        }
 
@@ -1325,21 +1403,20 @@ void fs_usage_apply_warn(struct btree_trans *trans,
        __WARN();
 }
 
-void bch2_trans_fs_usage_apply(struct btree_trans *trans,
-                              struct replicas_delta_list *deltas)
+int bch2_trans_fs_usage_apply(struct btree_trans *trans,
+                             struct replicas_delta_list *deltas)
 {
        struct bch_fs *c = trans->c;
        static int warned_disk_usage = 0;
        bool warn = false;
        unsigned disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
-       struct replicas_delta *d = deltas->d;
+       struct replicas_delta *d = deltas->d, *d2;
        struct replicas_delta *top = (void *) deltas->d + deltas->used;
        struct bch_fs_usage *dst;
        s64 added = 0, should_not_have_added;
        unsigned i;
 
-       percpu_rwsem_assert_held(&c->mark_lock);
-
+       percpu_down_read(&c->mark_lock);
        preempt_disable();
        dst = fs_usage_ptr(c, trans->journal_res.seq, false);
 
@@ -1351,7 +1428,8 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans,
                        added += d->delta;
                }
 
-               BUG_ON(__update_replicas(c, dst, &d->r, d->delta));
+               if (__update_replicas(c, dst, &d->r, d->delta))
+                       goto need_mark;
        }
 
        dst->nr_inodes += deltas->nr_inodes;
@@ -1386,30 +1464,33 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans,
        }
 
        preempt_enable();
+       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);
+       return 0;
+need_mark:
+       /* revert changes: */
+       for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
+               BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
+
+       preempt_enable();
+       percpu_up_read(&c->mark_lock);
+       return -1;
 }
 
 /* trans_mark: */
 
-static struct bkey_alloc_buf *
-bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
+static int bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
                              const struct bch_extent_ptr *ptr,
                              struct bkey_alloc_unpacked *u)
 {
        struct bch_fs *c = trans->c;
        struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
        struct bpos pos = POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
-       struct bucket *g;
-       struct bkey_alloc_buf *a;
        struct bkey_i *update = btree_trans_peek_updates(trans, BTREE_ID_alloc, pos);
        int ret;
 
-       a = bch2_trans_kmalloc(trans, sizeof(struct bkey_alloc_buf));
-       if (IS_ERR(a))
-               return a;
-
        bch2_trans_iter_init(trans, iter, BTREE_ID_alloc, pos,
                             BTREE_ITER_CACHED|
                             BTREE_ITER_CACHED_NOFILL|
@@ -1417,42 +1498,37 @@ bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter
        ret = bch2_btree_iter_traverse(iter);
        if (ret) {
                bch2_trans_iter_exit(trans, iter);
-               return ERR_PTR(ret);
+               return ret;
        }
 
-       if (update && !bpos_cmp(update->k.p, pos)) {
-               *u = bch2_alloc_unpack(bkey_i_to_s_c(update));
-       } else {
-               percpu_down_read(&c->mark_lock);
-               g = bucket(ca, pos.offset);
-               *u = alloc_mem_to_key(iter, g, READ_ONCE(g->mark));
-               percpu_up_read(&c->mark_lock);
-       }
+       *u = update && !bpos_cmp(update->k.p, pos)
+               ? bch2_alloc_unpack(bkey_i_to_s_c(update))
+               : alloc_mem_to_key(c, iter);
 
-       return a;
+       return 0;
 }
 
 static int bch2_trans_mark_pointer(struct btree_trans *trans,
                        struct bkey_s_c k, struct extent_ptr_decoded p,
                        s64 sectors, enum bch_data_type data_type)
 {
-       struct bch_fs *c = trans->c;
        struct btree_iter iter;
        struct bkey_alloc_unpacked u;
-       struct bkey_alloc_buf *a;
        int ret;
 
-       a = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u);
-       if (IS_ERR(a))
-               return PTR_ERR(a);
+       ret = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u);
+       if (ret)
+               return ret;
 
-       ret = __mark_pointer(c, k, &p.ptr, sectors, data_type, u.gen, &u.data_type,
+       ret = __mark_pointer(trans, k, &p.ptr, sectors, data_type,
+                            u.gen, &u.data_type,
                             &u.dirty_sectors, &u.cached_sectors);
        if (ret)
                goto out;
 
-       bch2_alloc_pack(c, a, u);
-       bch2_trans_update(trans, &iter, &a->k, 0);
+       ret = bch2_alloc_write(trans, &iter, &u, 0);
+       if (ret)
+               goto out;
 out:
        bch2_trans_iter_exit(trans, &iter);
        return ret;
@@ -1503,7 +1579,10 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
        stripe_blockcount_set(&s->v, p.ec.block,
                stripe_blockcount_get(&s->v, p.ec.block) +
                sectors);
-       bch2_trans_update(trans, &iter, &s->k_i, 0);
+
+       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;
@@ -1531,9 +1610,6 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
        bool stale;
        int ret;
 
-       BUG_ON((flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)) ==
-              (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE));
-
        r.e.data_type   = data_type;
        r.e.nr_devs     = 0;
        r.e.nr_required = 1;
@@ -1574,52 +1650,77 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
        return 0;
 }
 
-static int bch2_trans_mark_stripe_alloc_ref(struct btree_trans *trans,
-                                           struct bkey_s_c_stripe s,
-                                           unsigned idx, bool deleting)
+static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
+                                        struct bkey_s_c_stripe s,
+                                        unsigned idx, bool deleting)
 {
        struct bch_fs *c = trans->c;
        const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
-       struct bkey_alloc_buf *a;
        struct btree_iter iter;
        struct bkey_alloc_unpacked u;
-       bool parity = idx >= s.v->nr_blocks - s.v->nr_redundant;
+       enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
+               ? BCH_DATA_parity : 0;
+       s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
        int ret = 0;
 
-       a = bch2_trans_start_alloc_update(trans, &iter, ptr, &u);
-       if (IS_ERR(a))
-               return PTR_ERR(a);
-
-       if (parity) {
-               s64 sectors = le16_to_cpu(s.v->sectors);
+       if (deleting)
+               sectors = -sectors;
 
-               if (deleting)
-                       sectors = -sectors;
+       ret = bch2_trans_start_alloc_update(trans, &iter, ptr, &u);
+       if (ret)
+               return ret;
 
-               u.dirty_sectors += sectors;
-               u.data_type = u.dirty_sectors
-                       ? BCH_DATA_parity
-                       : 0;
-       }
+       ret = check_bucket_ref(c, s.s_c, ptr, sectors, data_type,
+                              u.gen, u.data_type,
+                              u.dirty_sectors, u.cached_sectors);
+       if (ret)
+               goto err;
 
        if (!deleting) {
-               if (bch2_fs_inconsistent_on(u.stripe && u.stripe != s.k->p.offset, c,
-                               "bucket %llu:%llu gen %u: multiple stripes using same bucket (%u, %llu)",
+               if (bch2_fs_inconsistent_on(u.stripe ||
+                                           u.stripe_redundancy, c,
+                               "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
                                iter.pos.inode, iter.pos.offset, u.gen,
+                               bch2_data_types[u.data_type],
+                               u.dirty_sectors,
                                u.stripe, s.k->p.offset)) {
                        ret = -EIO;
                        goto err;
                }
 
+               if (bch2_fs_inconsistent_on(data_type && u.dirty_sectors, c,
+                               "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
+                               iter.pos.inode, iter.pos.offset, u.gen,
+                               bch2_data_types[u.data_type],
+                               u.dirty_sectors,
+                               s.k->p.offset)) {
+                       ret = -EIO;
+                       goto err;
+               }
+
                u.stripe                = s.k->p.offset;
                u.stripe_redundancy     = s.v->nr_redundant;
        } else {
+               if (bch2_fs_inconsistent_on(u.stripe != s.k->p.offset ||
+                                           u.stripe_redundancy != s.v->nr_redundant, c,
+                               "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
+                               iter.pos.inode, iter.pos.offset, u.gen,
+                               s.k->p.offset, u.stripe)) {
+                       ret = -EIO;
+                       goto err;
+               }
+
                u.stripe                = 0;
                u.stripe_redundancy     = 0;
        }
 
-       bch2_alloc_pack(c, a, u);
-       bch2_trans_update(trans, &iter, &a->k, 0);
+       u.dirty_sectors += sectors;
+       if (data_type)
+               u.data_type = !deleting ? data_type : 0;
+
+       ret = bch2_alloc_write(trans, &iter, &u, 0);
+       if (ret)
+               goto err;
 err:
        bch2_trans_iter_exit(trans, &iter);
        return ret;
@@ -1632,7 +1733,7 @@ static int bch2_trans_mark_stripe(struct btree_trans *trans,
        struct bkey_s_c_stripe old_s = { .k = NULL };
        struct bkey_s_c_stripe new_s = { .k = NULL };
        struct bch_replicas_padded r;
-       unsigned i;
+       unsigned i, nr_blocks;
        int ret = 0;
 
        if (old.k->type == KEY_TYPE_stripe)
@@ -1650,18 +1751,17 @@ static int bch2_trans_mark_stripe(struct btree_trans *trans,
                    new_s.v->nr_blocks * sizeof(struct bch_extent_ptr)))
                return 0;
 
+       BUG_ON(new_s.k && old_s.k &&
+              (new_s.v->nr_blocks      != old_s.v->nr_blocks ||
+               new_s.v->nr_redundant   != old_s.v->nr_redundant));
+
+       nr_blocks = new_s.k ? new_s.v->nr_blocks : old_s.v->nr_blocks;
+
        if (new_s.k) {
                s64 sectors = le16_to_cpu(new_s.v->sectors);
 
                bch2_bkey_to_replicas(&r.e, new);
                update_replicas_list(trans, &r.e, sectors * new_s.v->nr_redundant);
-
-               for (i = 0; i < new_s.v->nr_blocks; i++) {
-                       ret = bch2_trans_mark_stripe_alloc_ref(trans, new_s,
-                                                              i, false);
-                       if (ret)
-                               return ret;
-               }
        }
 
        if (old_s.k) {
@@ -1669,12 +1769,25 @@ static int bch2_trans_mark_stripe(struct btree_trans *trans,
 
                bch2_bkey_to_replicas(&r.e, old);
                update_replicas_list(trans, &r.e, sectors * old_s.v->nr_redundant);
+       }
+
+       for (i = 0; i < nr_blocks; i++) {
+               if (new_s.k && old_s.k &&
+                   !memcmp(&new_s.v->ptrs[i],
+                           &old_s.v->ptrs[i],
+                           sizeof(new_s.v->ptrs[i])))
+                       continue;
 
-               for (i = 0; i < old_s.v->nr_blocks; i++) {
-                       ret = bch2_trans_mark_stripe_alloc_ref(trans, old_s,
-                                                              i, true);
+               if (new_s.k) {
+                       ret = bch2_trans_mark_stripe_bucket(trans, new_s, i, false);
                        if (ret)
-                               return ret;
+                               break;
+               }
+
+               if (old_s.k) {
+                       ret = bch2_trans_mark_stripe_bucket(trans, old_s, i, true);
+                       if (ret)
+                               break;
                }
        }
 
@@ -1686,8 +1799,7 @@ static int bch2_trans_mark_inode(struct btree_trans *trans,
                                 struct bkey_s_c new,
                                 unsigned flags)
 {
-       int nr = (new.k->type == KEY_TYPE_inode) -
-               (old.k->type == KEY_TYPE_inode);
+       int nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
 
        if (nr) {
                struct replicas_delta_list *d =
@@ -1705,9 +1817,6 @@ static int bch2_trans_mark_reservation(struct btree_trans *trans,
        s64 sectors = (s64) k.k->size;
        struct replicas_delta_list *d;
 
-       BUG_ON((flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)) ==
-              (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE));
-
        if (flags & BTREE_TRIGGER_OVERWRITE)
                sectors = -sectors;
        sectors *= replicas;
@@ -1723,7 +1832,7 @@ static int bch2_trans_mark_reservation(struct btree_trans *trans,
 
 static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
                        struct bkey_s_c_reflink_p p,
-                       u64 idx, unsigned flags)
+                       u64 *idx, unsigned flags)
 {
        struct bch_fs *c = trans->c;
        struct btree_iter iter;
@@ -1731,9 +1840,10 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
        struct bkey_i *n;
        __le64 *refcount;
        int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
-       s64 ret;
+       char buf[200];
+       int ret;
 
-       bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, idx),
+       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);
@@ -1750,15 +1860,38 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 
        refcount = bkey_refcount(n);
        if (!refcount) {
+               bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
                bch2_fs_inconsistent(c,
-                       "%llu:%llu len %u points to nonexistent indirect extent %llu",
-                       p.k->p.inode, p.k->p.offset, p.k->size, idx);
-               bch2_inconsistent_error(c);
+                       "nonexistent indirect extent at %llu while marking\n  %s",
+                       *idx, buf);
+               ret = -EIO;
+               goto err;
+       }
+
+       if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
+               bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
+               bch2_fs_inconsistent(c,
+                       "indirect extent refcount underflow at %llu while marking\n  %s",
+                       *idx, buf);
                ret = -EIO;
                goto err;
        }
 
-       BUG_ON(!*refcount && (flags & BTREE_TRIGGER_OVERWRITE));
+       if (flags & BTREE_TRIGGER_INSERT) {
+               struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
+               u64 pad;
+
+               pad = max_t(s64, le32_to_cpu(v->front_pad),
+                           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));
+               BUG_ON(pad > U32_MAX);
+               v->back_pad = cpu_to_le32(pad);
+       }
+
        le64_add_cpu(refcount, add);
 
        if (!*refcount) {
@@ -1771,7 +1904,7 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
        if (ret)
                goto err;
 
-       ret = k.k->p.offset - idx;
+       *idx = k.k->p.offset;
 err:
        bch2_trans_iter_exit(trans, &iter);
        return ret;
@@ -1781,29 +1914,29 @@ static int bch2_trans_mark_reflink_p(struct btree_trans *trans,
                                     struct bkey_s_c k, unsigned flags)
 {
        struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
-       u64 idx = le64_to_cpu(p.v->idx);
-       unsigned sectors = p.k->size;
-       s64 ret = 0;
+       u64 idx, end_idx;
+       int ret = 0;
 
-       while (sectors) {
-               ret = __bch2_trans_mark_reflink_p(trans, p, idx, flags);
-               if (ret < 0)
-                       return ret;
+       if (flags & BTREE_TRIGGER_INSERT) {
+               struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
 
-               ret = min_t(s64, ret, sectors);
-               idx     += ret;
-               sectors -= ret;
+               v->front_pad = v->back_pad = 0;
        }
 
-       return 0;
+       idx     = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
+       end_idx = le64_to_cpu(p.v->idx) + p.k->size +
+               le32_to_cpu(p.v->back_pad);
+
+       while (idx < end_idx && !ret)
+               ret = __bch2_trans_mark_reflink_p(trans, p, &idx, flags);
+
+       return ret;
 }
 
 int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c old,
                        struct bkey_s_c new, unsigned flags)
 {
-       struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old;
-
-       BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE)));
+       struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new;
 
        switch (k.k->type) {
        case KEY_TYPE_btree_ptr:
@@ -1814,6 +1947,7 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c old,
        case KEY_TYPE_stripe:
                return bch2_trans_mark_stripe(trans, old, new, flags);
        case KEY_TYPE_inode:
+       case KEY_TYPE_inode_v2:
                return bch2_trans_mark_inode(trans, old, new, flags);
        case KEY_TYPE_reservation:
                return bch2_trans_mark_reservation(trans, k, flags);
@@ -1824,39 +1958,6 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c old,
        }
 }
 
-int bch2_trans_mark_update(struct btree_trans *trans,
-                          struct btree_path *path,
-                          struct bkey_i *new,
-                          unsigned flags)
-{
-       struct bkey             _deleted = KEY(0, 0, 0);
-       struct bkey_s_c         deleted = (struct bkey_s_c) { &_deleted, NULL };
-       struct bkey_s_c         old;
-       struct bkey             unpacked;
-       int ret;
-
-       if (unlikely(flags & BTREE_TRIGGER_NORUN))
-               return 0;
-
-       if (!btree_node_type_needs_gc(path->btree_id))
-               return 0;
-
-       old = bch2_btree_path_peek_slot(path, &unpacked);
-
-       if (old.k->type == new->k.type &&
-           ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
-               ret   = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new),
-                               BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
-       } else {
-               ret   = bch2_trans_mark_key(trans, deleted, bkey_i_to_s_c(new),
-                               BTREE_TRIGGER_INSERT|flags) ?:
-                       bch2_trans_mark_key(trans, old, deleted,
-                               BTREE_TRIGGER_OVERWRITE|flags);
-       }
-
-       return ret;
-}
-
 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
                                    struct bch_dev *ca, size_t b,
                                    enum bch_data_type type,
@@ -1865,7 +1966,6 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
        struct bch_fs *c = trans->c;
        struct btree_iter iter;
        struct bkey_alloc_unpacked u;
-       struct bkey_alloc_buf *a;
        struct bch_extent_ptr ptr = {
                .dev = ca->dev_idx,
                .offset = bucket_to_sector(ca, b),
@@ -1878,9 +1978,9 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
        if (b >= ca->mi.nbuckets)
                return 0;
 
-       a = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u);
-       if (IS_ERR(a))
-               return PTR_ERR(a);
+       ret = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u);
+       if (ret)
+               return ret;
 
        if (u.data_type && u.data_type != type) {
                bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
@@ -1897,8 +1997,9 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
        u.data_type     = type;
        u.dirty_sectors = sectors;
 
-       bch2_alloc_pack(c, a, u);
-       bch2_trans_update(trans, &iter, &a->k, 0);
+       ret = bch2_alloc_write(trans, &iter, &u, 0);
+       if (ret)
+               goto out;
 out:
        bch2_trans_iter_exit(trans, &iter);
        return ret;