From: Vitaliy Makkoveev <otto@bsdbox.dev>
Subject: Re: socket wait channels
To: Claudio Jeker <cjeker@diehard.n-r-g.com>
Cc: Alexander Bluhm <bluhm@openbsd.org>, tech@openbsd.org
Date: Thu, 25 Apr 2024 20:15:19 +0300

> On 25 Apr 2024, at 17:43, Claudio Jeker <cjeker@diehard.n-r-g.com> wrote:
> 
> On Thu, Apr 25, 2024 at 03:58:32PM +0200, Alexander Bluhm wrote:
>> Hi,
>> 
>> Wait channel names should be unique for easier debugging.
>> 
>> "netcon2" was copied from sys_connect() to sys_ypconnect(), better
>> call it "ypcon".
>> 
>> Calling the accept sleep "netcon" makes no sense.  Better use
>> "netacc" for accept(2) and "netcon" for connect(2).
>> 
>> "netlck" for the socket buffer lock in bslock() can easily mixed
>> with "netlock" for the global network lock.  How about "sblock" for
>> sblock()?
> 
> Sounds good to me. I thought sblock was already used but that's sbwait.
> So OK claudio@
> 

ok mvs.

Unique names for connect/accept paths make sense.

I intentionally introduced "sbwait" for unlocked sbwait(), but left
well known original name for old sblock(). The per-buffers `sb_lock'
rwlocks have unique names, so this renaming is not significant to me.

>> ok?
>> 
>> bluhm
>> 
>> Index: kern/uipc_socket2.c
>> ===================================================================
>> RCS file: /data/mirror/openbsd/cvs/src/sys/kern/uipc_socket2.c,v
>> diff -u -p -r1.149 uipc_socket2.c
>> --- kern/uipc_socket2.c	11 Apr 2024 13:32:51 -0000	1.149
>> +++ kern/uipc_socket2.c	25 Apr 2024 13:45:23 -0000
>> @@ -559,7 +559,7 @@ sblock(struct socket *so, struct sockbuf
>> 	while (sb->sb_flags & SB_LOCK) {
>> 		sb->sb_flags |= SB_WANT;
>> 		mtx_leave(&sb->sb_mtx);
>> -		error = sosleep_nsec(so, &sb->sb_flags, prio, "netlck", INFSLP);
>> +		error = sosleep_nsec(so, &sb->sb_flags, prio, "sblock", INFSLP);
>> 		if (error)
>> 			return (error);
>> 		mtx_enter(&sb->sb_mtx);
>> Index: kern/uipc_syscalls.c
>> ===================================================================
>> RCS file: /data/mirror/openbsd/cvs/src/sys/kern/uipc_syscalls.c,v
>> diff -u -p -r1.218 uipc_syscalls.c
>> --- kern/uipc_syscalls.c	1 Mar 2024 14:15:01 -0000	1.218
>> +++ kern/uipc_syscalls.c	25 Apr 2024 13:45:12 -0000
>> @@ -300,7 +300,7 @@ doaccept(struct proc *p, int sock, struc
>> 			break;
>> 		}
>> 		error = sosleep_nsec(head, &head->so_timeo, PSOCK | PCATCH,
>> -		    "netcon", INFSLP);
>> +		    "netacc", INFSLP);
>> 		if (error)
>> 			goto out_unlock;
>> 	}
>> @@ -428,7 +428,7 @@ sys_connect(struct proc *p, void *v, reg
>> 	}
>> 	while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
>> 		error = sosleep_nsec(so, &so->so_timeo, PSOCK | PCATCH,
>> -		    "netcon2", INFSLP);
>> +		    "netcon", INFSLP);
>> 		if (error) {
>> 			if (error == EINTR || error == ERESTART)
>> 				interrupted = 1;
>> @@ -1651,7 +1651,7 @@ out:
>> 	error = soconnect(so, nam);
>> 	while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
>> 		error = sosleep_nsec(so, &so->so_timeo, PSOCK | PCATCH,
>> -		    "netcon2", INFSLP);
>> +		    "ypcon", INFSLP);
>> 		if (error)
>> 			break;
>> 	}
>> 
> 
> -- 
> :wq Claudio