LWN: Comments on "The rhashtable documentation I wanted to read"

error returns?

TheJH — Sat, 14 Apr 2018 02:48:07 +0000

It might be nice to have an explanation of what errors the rhashtable functions can throw, and when. For example, when does -EBUSY happen, and are there any rules that must be observed to allow the -EAGAIN retry loop in the slowpath to make forward progress?

The rhashtable documentation I wanted to read

TheJH — Sat, 14 Apr 2018 02:31:47 +0000

> The fact that they all end in the word fast is probably historical accident; there are no "slow" versions.

There is a function rhashtable_insert_slow(), but the fast insert functions will automatically bail out to it (if rehashing is in progress or something might require rehashing).

The rhashtable documentation I wanted to read

zx2c4 — Sat, 14 Apr 2018 01:12:39 +0000

For a while now I've been interested in using this with WireGuard, instead of using fixed-size hash tables by way of DECLARE_HASHTABLE. I'll need to change out the underlying hash function away from jhash2, presumably using the .hash_fn member as described here. In one instance, I use siphash, since values are attacker controlled, and in the other instance, I simply take the first N bits of the key, because the keys are already guaranteed to be uniformly distributed and non-attacker controlled. This article provides some nice motivation for actually getting this done.

The rhashtable documentation I wanted to read

abk77 — Fri, 13 Apr 2018 22:14:03 +0000

Atleast the APIs seem very similar to list_init( ) , list_create( ) , list_insert( ) in Solaris.

The rhashtable documentation I wanted to read

jake — Fri, 13 Apr 2018 22:13:13 +0000

> Looks like in "Object removal" there's a double RCU lock.

Indeed, thanks for the report. Next time, though, we would sure appreciate it if you email typo reports to lwn@lwn.net.

thanks,

jake

The rhashtable documentation I wanted to read

post-factum — Fri, 13 Apr 2018 21:17:51 +0000

Looks like in "Object removal" there's a double RCU lock.