Re: PR/53103 CVS commit: src/sys

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

The following reply was made to PR lib/53103; it has been noted by GNATS.

From: Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: martin%netbsd.org@localhost, christos%netbsd.org@localhost, tsutsui%ceres.dti.ne.jp@localhost
Subject: Re: PR/53103 CVS commit: src/sys
Date: Fri, 16 Nov 2018 01:52:53 +0900

 martin@ wrote:
 
 >  No, we can not break the ABI on release branches (module loading code
 >  does not check the patch level field on != .99 branches).
 >  
 >  We would need a proper fix or hack especially for the branch that avoids this.
 >  
 >  Maybe duplicate the function that got an extra argument with a new name
 >  and use that everywhere the new arg is important?
 >  So all new compiled modules would also use the new function, but old
 >  modules still would work using the old version.
 
 http://www.nerv.org/netbsd/?q=id:20180316T172504Z.13d546ff8678da724b3c16167900ffb31d5e12c4
 In this original commit, only linux_sys_sendto() in
 compat/linux/common/linux_socket.c takes UIO_SYSSPACE.
 
 However, the linux_sys_sendto() calls do_sys_sendmsg(), and it calls
 do_sys_sendmsg_so(), then do_sys_sendmsg_so() calls sockargs() with
 UIO_USERSPACE if MSG_NAMEMBUF is not set.
 
 I'm not sure if changing sockargs() API was worth enough because
 the following dumb patch against linux_sys_sendto() can also avoid the
 "sockargs cannot handle msg_name already copied into the kernel space"
 problem.
 
 Index: sys/compat/linux/common/linux_socket.c
 ===================================================================
 RCS file: /cvsroot/src/sys/compat/linux/common/linux_socket.c,v
 retrieving revision 1.138.6.2
 diff -u -p -d -r1.138.6.2 linux_socket.c
 --- sys/compat/linux/common/linux_socket.c	12 May 2018 10:29:08 -0000	1.138.6.2
 +++ sys/compat/linux/common/linux_socket.c	15 Nov 2018 16:34:13 -0000
 @@ -401,6 +401,7 @@ linux_sys_sendto(struct lwp *l, const st
  	struct msghdr   msg;
  	struct iovec    aiov;
  	struct sockaddr_big nam;
 +	struct mbuf *m;
  	int bflags;
  	int error;
  
 @@ -415,13 +416,31 @@ linux_sys_sendto(struct lwp *l, const st
  	msg.msg_control = NULL;
  
  	if (SCARG(uap, tolen)) {
 +		size_t solen;
 +
  		/* Read in and convert the sockaddr */
  		error = linux_get_sa(l, SCARG(uap, s), &nam, SCARG(uap, to),
  		    SCARG(uap, tolen));
  		if (error)
  			return (error);
 -		msg.msg_name = &nam;
 -		msg.msg_namelen = SCARG(uap, tolen);
 +
 +		/*
 +		 * XXX
 +		 * Copy sockaddr_big to mbuf because sockargs() called from
 +		 * do_sys_sendmsg_so() can't handle sockaddr in msg_name
 +		 * already copied into the kernel space.
 +		 */
 +		solen = nam.sb_len;
 +		m = m_get(M_WAIT, MT_SONAME);
 +		if (solen > MLEN) {
 +			MEXTMALLOC(m, solen, M_WAITOK);
 +		}
 +		m->m_len = solen;
 +		memcpy(mtod(m, void *), &nam, solen);
 +
 +		msg.msg_flags |= MSG_NAMEMBUF;
 +		msg.msg_name = m;
 +		msg.msg_namelen = solen;
  	}
  
  	msg.msg_iov = &aiov;
 
 
 ---
 Izumi Tsutsui