I was using the dtrace set-error probe in ZFS to track something down, and
I found that the probe in dbuf_findbp is VERY noisy. I also don't think
this one is really returning an "error" condition (in the usual sense) but
rather a status to the caller indicating the "bp" is not there. Would it
make sense to get rid of this one?
It's a small change, so pasted below.
diff --git a/usr/src/uts/common/fs/zfs/dbuf.c
b/usr/src/uts/common/fs/zfs/dbuf.c
index d23cdfa673..47222cd430 100644
--- a/usr/src/uts/common/fs/zfs/dbuf.c
+++ b/usr/src/uts/common/fs/zfs/dbuf.c
@@ -2338,8 +2338,11 @@ dbuf_findbp(dnode_t *dn, int level, uint64_t blkid,
int fail_sparse,
((nlevels - level - 1) * epbs)) ||
(fail_sparse &&
blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) {
- /* the buffer has no parent yet */
- return (SET_ERROR(ENOENT));
+ /*
+ * The buffer has no parent yet. This is a normal
+ * condition with sparse files, so no SET_ERROR().
+ */
+ return (ENOENT);
} else if (level < nlevels-1) {
/* this block is referenced from an indirect block */
int err = dbuf_hold_impl(dn, level+1,
------------------------------------------
openzfs: openzfs-developer
Permalink:
https://openzfs.topicbox.com/groups/developer/T83528121898a0810-M12a158c082e303023845bfd9
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription
