Re: [PATCH v1 01/10] btrfs: implement a nowait option for tree searches

[Filipe Manana <fdmanana@gmail.com>]

On Fri, Sep 2, 2022 at 12:01 AM Stefan Roesch <shr@fb.com> wrote:
>
> From: Josef Bacik <josef@toxicpanda.com>
>
> For NOWAIT IOCB's we'll need a way to tell search to not wait on locks
> or anything.  Accomplish this by adding a path->nowait flag that will
> use trylocks and skip reading of metadata, returning -EWOULDBLOCK in
> either of these cases.  For now we only need this for reads, so only the
> read side is handled.  Add an ASSERT() to catch anybody trying to use
> this for writes so they know they'll have to implement the write side.
>
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
> Signed-off-by: Stefan Roesch <shr@fb.com>
> ---
>  fs/btrfs/ctree.c   | 39 ++++++++++++++++++++++++++++++++++++---
>  fs/btrfs/ctree.h   |  1 +
>  fs/btrfs/locking.c | 23 +++++++++++++++++++++++
>  fs/btrfs/locking.h |  1 +
>  4 files changed, 61 insertions(+), 3 deletions(-)
>
> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
> index ebfa35fe1c38..052c768b2297 100644
> --- a/fs/btrfs/ctree.c
> +++ b/fs/btrfs/ctree.c
> @@ -1447,6 +1447,11 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
>                         return 0;
>                 }
>
> +               if (p->nowait) {
> +                       free_extent_buffer(tmp);
> +                       return -EWOULDBLOCK;
> +               }
> +
>                 if (unlock_up)
>                         btrfs_unlock_up_safe(p, level + 1);
>
> @@ -1467,6 +1472,8 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
>                         ret = -EAGAIN;
>
>                 goto out;
> +       } else if (p->nowait) {
> +               return -EWOULDBLOCK;
>         }
>
>         if (unlock_up) {
> @@ -1634,7 +1641,13 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
>                  * We don't know the level of the root node until we actually
>                  * have it read locked
>                  */
> -               b = btrfs_read_lock_root_node(root);
> +               if (p->nowait) {
> +                       b = btrfs_try_read_lock_root_node(root);
> +                       if (IS_ERR(b))
> +                               return b;
> +               } else {
> +                       b = btrfs_read_lock_root_node(root);
> +               }
>                 level = btrfs_header_level(b);
>                 if (level > write_lock_level)
>                         goto out;
> @@ -1910,6 +1923,13 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
>         WARN_ON(p->nodes[0] != NULL);
>         BUG_ON(!cow && ins_len);
>
> +       /*
> +        * For now only allow nowait for read only operations.  There's no
> +        * strict reason why we can't, we just only need it for reads so I'm
> +        * only implementing it for reads right now.
> +        */
> +       ASSERT(!p->nowait || !cow);
> +
>         if (ins_len < 0) {
>                 lowest_unlock = 2;
>
> @@ -1936,7 +1956,12 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
>
>         if (p->need_commit_sem) {
>                 ASSERT(p->search_commit_root);
> -               down_read(&fs_info->commit_root_sem);
> +               if (p->nowait) {
> +                       if (!down_read_trylock(&fs_info->commit_root_sem))
> +                               return -EAGAIN;

Why EAGAIN here and everywhere else EWOULDBLOCK? See below.

> +               } else {
> +                       down_read(&fs_info->commit_root_sem);
> +               }
>         }
>
>  again:
> @@ -2082,7 +2107,15 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
>                                 btrfs_tree_lock(b);
>                                 p->locks[level] = BTRFS_WRITE_LOCK;
>                         } else {
> -                               btrfs_tree_read_lock(b);
> +                               if (p->nowait) {
> +                                       if (!btrfs_try_tree_read_lock(b)) {
> +                                               free_extent_buffer(b);
> +                                               ret = -EWOULDBLOCK;

Like here, this try lock failed and we are returning EWOULDBLOCK
instead of EAGAIN like above.

I'm also confused because in the followup patches I don't see
EWOULDBLOCK converted to EAGAIN to return to io_uring.
Currently we return EAGAIN for direct IO with NOWAIT when we need to
block or fallback to buffered IO. Does this means
that EWOULDBLOCK is also valid, or that somehow it's special for
buffered writes only?

> +                                               goto done;
> +                                       }
> +                               } else {
> +                                       btrfs_tree_read_lock(b);
> +                               }
>                                 p->locks[level] = BTRFS_READ_LOCK;
>                         }
>                         p->nodes[level] = b;
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index 9ef162dbd4bc..d6d05450198d 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -443,6 +443,7 @@ struct btrfs_path {
>          * header (ie. sizeof(struct btrfs_item) is not included).
>          */
>         unsigned int search_for_extension:1;
> +       unsigned int nowait:1;

This misses several other places that relate to searches outside
btrfs_search_slot().
E.g. btrfs_search_old_slot(), btrfs_next_old_leaf() (used by
btrfs_next_leaf()), btrfs_search_forward() - possibly others too.

I understand those places were not changed because they're not needed
in the buffered write path (nor direct IO).

For the sake of completeness, should we deal with them, or at least
add an ASSERT in case path->nowait is set so that we don't forget
about them
in case in the future we get those other paths used in a NOWAIT
context (and that would be easy to miss).

Otherwise, it looks good to me.

Thanks.

>  };
>  #define BTRFS_MAX_EXTENT_ITEM_SIZE(r) ((BTRFS_LEAF_DATA_SIZE(r->fs_info) >> 4) - \
>                                         sizeof(struct btrfs_item))
> diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
> index 9063072b399b..acc6ffeb2cda 100644
> --- a/fs/btrfs/locking.c
> +++ b/fs/btrfs/locking.c
> @@ -285,6 +285,29 @@ struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
>         return eb;
>  }
>
> +/*
> + * Loop around taking references on and locking the root node of the tree in
> + * nowait mode until we end up with a lock on the root node or returning to
> + * avoid blocking.
> + *
> + * Return: root extent buffer with read lock held or -EWOULDBLOCK.
> + */
> +struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root)
> +{
> +       struct extent_buffer *eb;
> +
> +       while (1) {
> +               eb = btrfs_root_node(root);
> +               if (!btrfs_try_tree_read_lock(eb))
> +                       return ERR_PTR(-EWOULDBLOCK);
> +               if (eb == root->node)
> +                       break;
> +               btrfs_tree_read_unlock(eb);
> +               free_extent_buffer(eb);
> +       }
> +       return eb;
> +}
> +
>  /*
>   * DREW locks
>   * ==========
> diff --git a/fs/btrfs/locking.h b/fs/btrfs/locking.h
> index ab268be09bb5..490c7a79e995 100644
> --- a/fs/btrfs/locking.h
> +++ b/fs/btrfs/locking.h
> @@ -94,6 +94,7 @@ int btrfs_try_tree_read_lock(struct extent_buffer *eb);
>  int btrfs_try_tree_write_lock(struct extent_buffer *eb);
>  struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root);
>  struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root);
> +struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root);
>
>  #ifdef CONFIG_BTRFS_DEBUG
>  static inline void btrfs_assert_tree_write_locked(struct extent_buffer *eb)
> --
> 2.30.2
>


-- 
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”