OpenZFS 8025 - dbuf_read() creates unnecessary zio_root() for bonus buf

Authored by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed by: Prashanth Sreenivasa <pks@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
dbuf_read() creates a zio_root() to track and wait for all the zio's
that may happen as part of this call. However, if the blkptr_t for
this buffer is NULL or a hole, we will not create any more zio's, so
this zio_root() is unnecessary. This is always the case when calling
dbuf_read() on a bonus buffer, because it has no blkptr (it's part of
the containing dnode). For workloads that read a lot of bonus buffers
(e.g. file creation and removal), creating and destroying these
unnecessary zio's can decrease performance by around 3%.

The fix is to only create/destroy the zio_root() in dbuf_read() if the
blkptr is not NULL and not a hole.

Porting Notes:
- The error handling for when dbuf_read_impl() fails which was
  originally added in commit 5f6d0b6f5 has been preserved.

OpenZFS-issue: https://www.illumos.org/issues/8025
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/8ec5c7c
Closes #6048

diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
index 096d183584c3df7503b6950b76fa0cba2c88ce81..b0b1bffab9cbab587fd920cdea88dde2cb83437b 100644 (file)
--- a/module/zfs/dbuf.c
+++ b/module/zfs/dbuf.c
@@ -1170,7 +1170,6 @@ int
 dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
 {
        int err = 0;
-       boolean_t havepzio = (zio != NULL);
        boolean_t prefetch;
        dnode_t *dn;
 
@@ -1214,11 +1213,13 @@ dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
                DB_DNODE_EXIT(db);
        } else if (db->db_state == DB_UNCACHED) {
                spa_t *spa = dn->dn_objset->os_spa;
+               boolean_t need_wait = B_FALSE;
 
                if (zio == NULL &&
-                   db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr))
+                   db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) {
                        zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
-
+                       need_wait = B_TRUE;
+               }
                err = dbuf_read_impl(db, zio, flags);
 
                /* dbuf_read_impl has dropped db_mtx for us */
@@ -1230,7 +1231,7 @@ dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
                        rw_exit(&dn->dn_struct_rwlock);
                DB_DNODE_EXIT(db);
 
-               if (!err && !havepzio && zio != NULL)
+               if (!err && need_wait)
                        err = zio_wait(zio);
        } else {
                /*
@@ -1265,7 +1266,6 @@ dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
                mutex_exit(&db->db_mtx);
        }
 
-       ASSERT(err || havepzio || db->db_state == DB_CACHED);
        return (err);
 }