bcachefs: bch2_trigger_pointer()
authorKent Overstreet <kent.overstreet@linux.dev>
Sat, 30 Dec 2023 00:02:56 +0000 (19:02 -0500)
committerKent Overstreet <kent.overstreet@linux.dev>
Sat, 6 Jan 2024 04:24:20 +0000 (23:24 -0500)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
fs/bcachefs/buckets.c

index e035ba3..4c55945 100644 (file)
@@ -653,201 +653,6 @@ err:
        goto out;
 }
 
-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,
-                         u32 *dirty_sectors, u32 *cached_sectors)
-{
-       u32 *dst_sectors = !ptr->cached
-               ? dirty_sectors
-               : cached_sectors;
-       int ret = bch2_check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
-                                  bucket_gen, *bucket_data_type, *dst_sectors);
-
-       if (ret)
-               return ret;
-
-       *dst_sectors += sectors;
-
-       if (!*dirty_sectors && !*cached_sectors)
-               *bucket_data_type = 0;
-       else if (*bucket_data_type != BCH_DATA_stripe)
-               *bucket_data_type = ptr_data_type;
-
-       return 0;
-}
-
-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,
-                            unsigned flags)
-{
-       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;
-
-       BUG_ON(!(flags & BTREE_TRIGGER_GC));
-
-       percpu_down_read(&c->mark_lock);
-       g = PTR_GC_BUCKET(ca, &p.ptr);
-       bucket_lock(g);
-       old = *g;
-
-       bucket_data_type = g->data_type;
-       ret = __mark_pointer(trans, k, &p.ptr, sectors,
-                            data_type, g->gen,
-                            &bucket_data_type,
-                            &g->dirty_sectors,
-                            &g->cached_sectors);
-       if (!ret)
-               g->data_type = bucket_data_type;
-
-       new = *g;
-       bucket_unlock(g);
-       if (!ret)
-               bch2_dev_usage_update_m(c, ca, &old, &new);
-       percpu_up_read(&c->mark_lock);
-
-       return ret;
-}
-
-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 flags)
-{
-       struct bch_fs *c = trans->c;
-       struct bch_replicas_padded r;
-       struct gc_stripe *m;
-
-       BUG_ON(!(flags & BTREE_TRIGGER_GC));
-
-       m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
-       if (!m) {
-               bch_err(c, "error allocating memory for gc_stripes, idx %llu",
-                       (u64) p.idx);
-               return -BCH_ERR_ENOMEM_mark_stripe_ptr;
-       }
-
-       mutex_lock(&c->ec_stripes_heap_lock);
-
-       if (!m || !m->alive) {
-               mutex_unlock(&c->ec_stripes_heap_lock);
-               bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
-                                   (u64) p.idx);
-               bch2_inconsistent_error(c);
-               return -EIO;
-       }
-
-       m->block_sectors[p.block] += sectors;
-
-       r = m->r;
-       mutex_unlock(&c->ec_stripes_heap_lock);
-
-       r.e.data_type = data_type;
-       bch2_update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
-
-       return 0;
-}
-
-static int __mark_extent(struct btree_trans *trans,
-                        enum btree_id btree_id, unsigned level,
-                        struct bkey_s_c k, unsigned flags)
-{
-       u64 journal_seq = trans->journal_res.seq;
-       struct bch_fs *c = trans->c;
-       struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-       const union bch_extent_entry *entry;
-       struct extent_ptr_decoded p;
-       struct bch_replicas_padded r;
-       enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
-               ? BCH_DATA_btree
-               : BCH_DATA_user;
-       s64 sectors = bkey_is_btree_ptr(k.k)
-               ? btree_sectors(c)
-               : k.k->size;
-       s64 dirty_sectors = 0;
-       bool stale;
-       int ret;
-
-       BUG_ON(!(flags & BTREE_TRIGGER_GC));
-
-       r.e.data_type   = data_type;
-       r.e.nr_devs     = 0;
-       r.e.nr_required = 1;
-
-       bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-               s64 disk_sectors = ptr_disk_sectors(sectors, p);
-
-               if (flags & BTREE_TRIGGER_OVERWRITE)
-                       disk_sectors = -disk_sectors;
-
-               ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
-               if (ret < 0)
-                       return ret;
-
-               stale = ret > 0;
-
-               if (p.ptr.cached) {
-                       if (!stale) {
-                               ret = update_cached_sectors(c, k, p.ptr.dev,
-                                               disk_sectors, journal_seq, true);
-                               if (ret) {
-                                       bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
-                                                           __func__);
-                                       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(trans, k, p.ec, data_type,
-                                       disk_sectors, flags);
-                       if (ret)
-                               return ret;
-
-                       /*
-                        * There may be other dirty pointers in this extent, but
-                        * if so they're not required for mounting if we have an
-                        * erasure coded pointer in this extent:
-                        */
-                       r.e.nr_required = 0;
-               }
-       }
-
-       if (r.e.nr_devs) {
-               ret = bch2_update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
-               if (ret) {
-                       struct printbuf buf = PRINTBUF;
-
-                       bch2_bkey_val_to_text(&buf, c, k);
-                       bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
-                       printbuf_exit(&buf);
-                       return ret;
-               }
-       }
-
-       return 0;
-}
-
-int bch2_mark_extent(struct btree_trans *trans,
-                    enum btree_id btree_id, unsigned level,
-                    struct bkey_s_c old, struct bkey_s new,
-                    unsigned flags)
-{
-       return trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
-}
-
 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
                                struct replicas_delta_list *deltas)
 {
@@ -967,45 +772,140 @@ need_mark:
        return -1;
 }
 
-/* trans_mark: */
+/* KEY_TYPE_extent: */
 
-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)
+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,
+                         u32 *dirty_sectors, u32 *cached_sectors)
+{
+       u32 *dst_sectors = !ptr->cached
+               ? dirty_sectors
+               : cached_sectors;
+       int ret = bch2_check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
+                                  bucket_gen, *bucket_data_type, *dst_sectors);
+
+       if (ret)
+               return ret;
+
+       *dst_sectors += sectors;
+
+       if (!*dirty_sectors && !*cached_sectors)
+               *bucket_data_type = 0;
+       else if (*bucket_data_type != BCH_DATA_stripe)
+               *bucket_data_type = ptr_data_type;
+
+       return 0;
+}
+
+static int bch2_trigger_pointer(struct btree_trans *trans,
+                       enum btree_id btree_id, unsigned level,
+                       struct bkey_s_c k, struct extent_ptr_decoded p,
+                       s64 *sectors,
+                       unsigned flags)
 {
        bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
-       struct btree_iter iter;
-       struct bkey_i_alloc_v4 *a;
        struct bpos bucket;
        struct bch_backpointer bp;
-       s64 sectors;
-       int ret;
 
        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);
-       if (IS_ERR(a))
-               return PTR_ERR(a);
+       *sectors = insert ? bp.bucket_len : -((s64) bp.bucket_len);
 
-       ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
-                            a->v.gen, &a->v.data_type,
-                            &a->v.dirty_sectors, &a->v.cached_sectors) ?:
-               bch2_trans_update(trans, &iter, &a->k_i, 0);
-       bch2_trans_iter_exit(trans, &iter);
+       if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
+               struct btree_iter iter;
+               struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update(trans, &iter, bucket);
+               int ret = PTR_ERR_OR_ZERO(a);
+               if (ret)
+                       return ret;
 
-       if (ret)
-               return ret;
+               ret = __mark_pointer(trans, k, &p.ptr, *sectors, bp.data_type,
+                                    a->v.gen, &a->v.data_type,
+                                    &a->v.dirty_sectors, &a->v.cached_sectors) ?:
+                       bch2_trans_update(trans, &iter, &a->k_i, 0);
+               bch2_trans_iter_exit(trans, &iter);
 
-       if (!p.ptr.cached) {
-               ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
                if (ret)
                        return ret;
+
+               if (!p.ptr.cached) {
+                       ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
+                       if (ret)
+                               return ret;
+               }
+       }
+
+       if (flags & BTREE_TRIGGER_GC) {
+               struct bch_fs *c = trans->c;
+               struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
+               enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
+
+               percpu_down_read(&c->mark_lock);
+               struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
+               bucket_lock(g);
+               struct bucket old = *g;
+
+               u8 bucket_data_type = g->data_type;
+               int ret = __mark_pointer(trans, k, &p.ptr, *sectors,
+                                    data_type, g->gen,
+                                    &bucket_data_type,
+                                    &g->dirty_sectors,
+                                    &g->cached_sectors);
+               if (ret) {
+                       bucket_unlock(g);
+                       percpu_up_read(&c->mark_lock);
+                       return ret;
+               }
+
+               g->data_type = bucket_data_type;
+               struct bucket new = *g;
+               bucket_unlock(g);
+               bch2_dev_usage_update_m(c, ca, &old, &new);
+               percpu_up_read(&c->mark_lock);
+       }
+
+       return 0;
+}
+
+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 flags)
+{
+       struct bch_fs *c = trans->c;
+       struct bch_replicas_padded r;
+       struct gc_stripe *m;
+
+       BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+       m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
+       if (!m) {
+               bch_err(c, "error allocating memory for gc_stripes, idx %llu",
+                       (u64) p.idx);
+               return -BCH_ERR_ENOMEM_mark_stripe_ptr;
+       }
+
+       mutex_lock(&c->ec_stripes_heap_lock);
+
+       if (!m || !m->alive) {
+               mutex_unlock(&c->ec_stripes_heap_lock);
+               bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
+                                   (u64) p.idx);
+               bch2_inconsistent_error(c);
+               return -EIO;
        }
 
+       m->block_sectors[p.block] += sectors;
+
+       r = m->r;
+       mutex_unlock(&c->ec_stripes_heap_lock);
+
+       r.e.data_type = data_type;
+       bch2_update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
+
        return 0;
 }
 
@@ -1049,11 +949,91 @@ err:
        return ret;
 }
 
+static int __mark_extent(struct btree_trans *trans,
+                        enum btree_id btree_id, unsigned level,
+                        struct bkey_s_c k, unsigned flags)
+{
+       u64 journal_seq = trans->journal_res.seq;
+       struct bch_fs *c = trans->c;
+       struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+       const union bch_extent_entry *entry;
+       struct extent_ptr_decoded p;
+       struct bch_replicas_padded r;
+       enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
+               ? BCH_DATA_btree
+               : BCH_DATA_user;
+       s64 dirty_sectors = 0;
+       int ret;
+
+       BUG_ON(!(flags & BTREE_TRIGGER_GC));
+
+       r.e.data_type   = data_type;
+       r.e.nr_devs     = 0;
+       r.e.nr_required = 1;
+
+       bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+               s64 disk_sectors;
+               ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
+               if (ret < 0)
+                       return ret;
+
+               bool stale = ret > 0;
+
+               if (p.ptr.cached) {
+                       if (!stale) {
+                               ret = update_cached_sectors(c, k, p.ptr.dev,
+                                               disk_sectors, journal_seq, true);
+                               if (ret) {
+                                       bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
+                                                           __func__);
+                                       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(trans, k, p.ec, data_type,
+                                       disk_sectors, flags);
+                       if (ret)
+                               return ret;
+
+                       /*
+                        * There may be other dirty pointers in this extent, but
+                        * if so they're not required for mounting if we have an
+                        * erasure coded pointer in this extent:
+                        */
+                       r.e.nr_required = 0;
+               }
+       }
+
+       if (r.e.nr_devs) {
+               ret = bch2_update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
+               if (ret) {
+                       struct printbuf buf = PRINTBUF;
+
+                       bch2_bkey_val_to_text(&buf, c, k);
+                       bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
+                       printbuf_exit(&buf);
+                       return ret;
+               }
+       }
+
+       return 0;
+}
+
+int bch2_mark_extent(struct btree_trans *trans,
+                    enum btree_id btree_id, unsigned level,
+                    struct bkey_s_c old, struct bkey_s new,
+                    unsigned flags)
+{
+       return trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
+}
+
 static int __trans_mark_extent(struct btree_trans *trans,
                               enum btree_id btree_id, unsigned level,
                               struct bkey_s_c k, unsigned flags)
 {
-       struct bch_fs *c = trans->c;
        struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
        const union bch_extent_entry *entry;
        struct extent_ptr_decoded p;
@@ -1061,11 +1041,7 @@ static int __trans_mark_extent(struct btree_trans *trans,
        enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
                ? BCH_DATA_btree
                : BCH_DATA_user;
-       s64 sectors = bkey_is_btree_ptr(k.k)
-               ? btree_sectors(c)
-               : k.k->size;
        s64 dirty_sectors = 0;
-       bool stale;
        int ret = 0;
 
        r.e.data_type   = data_type;
@@ -1073,16 +1049,12 @@ static int __trans_mark_extent(struct btree_trans *trans,
        r.e.nr_required = 1;
 
        bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-               s64 disk_sectors = ptr_disk_sectors(sectors, p);
-
-               if (flags & BTREE_TRIGGER_OVERWRITE)
-                       disk_sectors = -disk_sectors;
-
-               ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
+               s64 disk_sectors;
+               ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
                if (ret < 0)
                        return ret;
 
-               stale = ret > 0;
+               bool stale = ret > 0;
 
                if (p.ptr.cached) {
                        if (!stale) {
@@ -1104,10 +1076,13 @@ static int __trans_mark_extent(struct btree_trans *trans,
                }
        }
 
-       if (r.e.nr_devs)
+       if (r.e.nr_devs) {
                ret = bch2_update_replicas_list(trans, &r.e, dirty_sectors);
+               if (ret)
+                       return ret;
+       }
 
-       return ret;
+       return 0;
 }
 
 int bch2_trans_mark_extent(struct btree_trans *trans,