Re: [RFC,PATCH] svc: Fix the UDP address logic

[Chuck Lever <chuck.lever@oracle.com>]

[-- Attachment #1: Type: text/plain, Size: 6055 bytes --]

Tom Tucker wrote:
> Chuck Lever wrote:
>> Tom Tucker wrote:
>>> Chuck Lever wrote:
>>>> Tom Tucker wrote:
>>>>> When the address information was moved to the svc_xprt structure, a 
>>>>> bug
>>>>> was introduced to UDP that caused an incorrect address to be used
>>>>> when responding to RPC on the UDP transport. This was the result of
>>>>> failing to completely implement the generic address logic for the 
>>>>> UDP transport.
>>>>>
>>>>> Thanks to Greg for pointing this out...
>>>>>
>>>>> Since I confused myself, it's probably a good idea to describe how 
>>>>> this is supposed to work. The transport is responsible for setting 
>>>>> the xpt_local and xpt_remote addresses in the svc_xprt structure as
>>>>> part of xpo_recvfrom processing. This cannot be done in a generic 
>>>>> way and in fact varies between TCP, UDP and RDMA. A set of xpo_ 
>>>>> functions
>>>>> (e.g. getlocalname, getremotename) could have been added but this 
>>>>> would
>>>>> have resulted in additional caching and copying of the addresses 
>>>>> around.
>>>>> The generic svc_recv code copies the addresses from the svc_xprt 
>>>>> structure into the rqstp structure as part of svc_recv processing. 
>>>>> The xpt_local address should also be set on listening endpoints, for
>>>>> tcp/rdma this is done as part of endpoint creation and for new 
>>>>> connections in xpo_accept processing.
>>>>> This patch was tested with Connectathon over V3 on UDP.
>>>>>
>>>>> Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
>>>>> ---
>>>>>
>>>>>  net/sunrpc/svcsock.c |   16 ++++++++++------
>>>>>  1 files changed, 10 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
>>>>> index e1a27ee..0f8a3d5 100644
>>>>> --- a/net/sunrpc/svcsock.c
>>>>> +++ b/net/sunrpc/svcsock.c
>>>>> @@ -473,17 +473,21 @@ svc_write_space(struct sock *sk)
>>>>>  static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp,
>>>>>                          struct cmsghdr *cmh)
>>>>>  {
>>>>> -    struct svc_sock *svsk =
>>>>> -        container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
>>>>> +    struct svc_xprt *xprt = rqstp->rq_xprt;
>>>>> +    struct svc_sock *svsk = container_of(xprt, struct svc_sock, 
>>>>> sk_xprt);
>>>>>      switch (svsk->sk_sk->sk_family) {
>>>>>      case AF_INET: {
>>>>>          struct in_pktinfo *pki = CMSG_DATA(cmh);
>>>>> -        rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
>>>>> +        struct sockaddr_in *sin =
>>>>> +            (struct sockaddr_in *)&xprt->xpt_local;
>>>>> +        sin->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
>>>>>          break;
>>>>>          }
>>>>>      case AF_INET6: {
>>>>>          struct in6_pktinfo *pki = CMSG_DATA(cmh);
>>>>> -        ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
>>>>> +        struct sockaddr_in6 *sin6 =
>>>>> +            (struct sockaddr_in6 *)&xprt->xpt_local;
>>>>> +        ipv6_addr_copy(&sin6->sin6_addr, &pki->ipi6_addr);
>>>>>          break;
>>>>>          }
>>>>>      }
>>>>> @@ -506,7 +510,7 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
>>>>>      struct cmsghdr *cmh = &buffer.hdr;
>>>>>      int        err, len;
>>>>>      struct msghdr msg = {
>>>>> -        .msg_name = svc_addr(rqstp),
>>>>> +        .msg_name = &svsk->sk_xprt.xpt_remote,
>>>>>          .msg_control = cmh,
>>>>>          .msg_controllen = sizeof(buffer),
>>>>>          .msg_flags = MSG_DONTWAIT,
>>>>> @@ -541,7 +545,7 @@ svc_udp_recvfrom(struct svc_rqst *rqstp)
>>>>>          svc_xprt_received(&svsk->sk_xprt);
>>>>>          return -EAGAIN;
>>>>>      }
>>>>> -    rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
>>>>> +    svsk->sk_xprt.xpt_remotelen = sizeof(svsk->sk_xprt.xpt_remote);
>>>>
>>>> This worries me a bit.  Why isn't rq_addrlen involved here?
>>> rq_addrlen gets set in the generic code (see svc_recv). The source of 
>>> the address is the transport, not the rqstp structure. The rqstp 
>>> structure caches it for use when responding to the request in 
>>> svc_process. Ironically, I think the HA case that Greg referred to is 
>>> the only reason we can't get rid of rq_addr/rq_daddr altogether and 
>>> access the address data directly from svc_xprt. Since this address 
>>> may change for UDP, we have to store it in the rqstp structure and 
>>> use it to reply because another request could arrive in the interim 
>>> and change the address data in svc_xprt.
>>>>   the xpt_remote field should be a sockaddr_storage, 
>>> ...it is.
>>>> and the remote_len field should be the actual length of the address, 
>>>> not the size of the storage space.
>>> Perhaps it should be, but a) kernel_recvmsg doesn't give us a length, 
>>> b) to determine it would require a switch, the degenerate case for 
>>> which is sizeof(storage),
>>
>> I don't agree: there is no degenerate case, and sizeof(storage) should 
>> never ever be used.
>>
>>> and c) in it's current usage, big-enough is good-enough. A weaker 
>>> point is "that's the way it's always been" ;-)
>>
>> Please change it to use an appropriate address length.  Using 
>> sizeof(struct sockaddr_storage) is broken, or at the very least 
>> inefficient (this struct is huge).  
> Yes, for TCP/UDP 128B is getting copied when it really only needs to be 
> 16 or 28.

I didn't think this was the case before the transport switch was 
introduced, but perhaps there have been some patch mix-ups, or 
oversights on my part.

The purpose of the address length field is to limit the number of bytes 
that you need to copy.  It's used when the address is treated as a blob 
of bytes rather than when it is accessed as a union of structures 
discriminated by ss_family.

>> The use of sizeof() here is inconsistent with the design of the IPv6 
>> changes I made last year.
> Fair enough, but in practical terms that means if ss_family is not 
> AF_INET or AF_INET6, we'll return an error or is there something I'm 
> missing?

Yes, I think an error should be thrown in that case.  I notice that in 
some places, we have "default: BUG();" which I think is excessive.

[-- Attachment #2: chuck.lever.vcf --]
[-- Type: text/x-vcard, Size: 327 bytes --]

begin:vcard
fn:Chuck Lever
n:Lever;Chuck
org:Oracle Corporation;Corporate Architecture: Linux Projects Group
adr:;;1015 Granger Avenue;Ann Arbor;MI;48104;USA
email;internet:chuck dot lever at nospam oracle dot com
title:Principal Member of Staff
tel;work:+1 248 614 5091
x-mozilla-html:FALSE
version:2.1
end:vcard


[-- Attachment #3: Type: text/plain, Size: 314 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/

[-- Attachment #4: Type: text/plain, Size: 140 bytes --]

_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs