Re: SCHEDULE and high load situations

On Thu, 12 Aug 2004, Martin Blapp wrote:

> Here is more information: (thanks robert for the help)
> 
> > Fatal trap 12: page fault while in kernel mode
> > cpuid = 1; apic id = 01
> > fault virtual address   = 0x14
> > fault code              = supervisor write, page not present
> > instruction pointer     = 0x8:0xc066a1c7
> > stack pointer           = 0x10:0xe2626aa8
> > frame pointer           = 0x10:0xe2626ab8
> > code segment            = base 0x0, limit 0xfffff, type 0x1b
> >                         = DPL 0, pres 1, def32 1, gran 1
> > processor eflags        = interrupt enabled, resume, IOPL = 0
> > current process         = 27897 (mimedefang)
> >

Ok, indeed, this appears to be an unaddressed class of race conditions in
the UNIX domain socket code.  I'm currently working through it, both to
address in the mpsafe case and non-mpsafe case (the one you were running
in).  I will run some tests on it tonight and try to get you patches to
try tomorrow.

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Principal Research Scientist, McAfee Research


> 
> db> where
> unp_connect2(c4bb78a4,c39cc13c,0,0,0) at /usr/src/sys/kern/uipc_usrreq.c:892
> unp_connect(c4bb78a4,c43d9380,c4dee9a0,c43d9380,80) at /usr/src/sys/kern/uipc_usrreq.c:865
> uipc_connect(c4bb78a4,c43d9380,c4dee9a0) at /usr/src/sys/kern/uipc_usrreq.c:179
> soconnect(c4bb78a4,c43d9380,c4dee9a0,0,bf1dad88) at /usr/src/sys/kern/uipc_socket.c:518
> kern_connect(c4dee9a0,3,c43d9380,c43d9380,c3e958ac) at /usr/src/sys/kern/uipc_syscalls.c:477
> connect(c4dee9a0,e2626d14,c,c4dee9a0,e2626d3c) at connect+0x42
> syscall(2f,2f,2f,bf1dad88,bf1dad8a) at syscall+0x300
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (98, FreeBSD ELF32, connect), eip = 0x28101d23, esp = 0xbf1dad74, ebp = 0xbf1dae10 ---
> 
> src/sys/kern/uipc_syscalls.c,v 1.199
> src/sys/kern/uipc_usrreq.c,v 1.135
> src/sys/kern/uipc_socket.c,v 1.207
> 
> (gdb) l *unp_connect2+0x2a
> 0x1f93 is in unp_connect2 (/usr/src/sys/kern/uipc_usrreq.c:892).
> 887             UNP_LOCK_ASSERT();
> 888
> 889             if (so2->so_type != so->so_type)
> 890                     return (EPROTOTYPE);
> 891             unp2 = sotounpcb(so2);
> 892             unp->unp_conn = unp2;
> 893             switch (so->so_type) {
> 894
> 895             case SOCK_DGRAM:
> 896                     LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
> 
> (gdb) l *unp_connect+0x3d5
> 0x1e24 is in unp_connect (/usr/src/sys/kern/uipc_usrreq.c:865).
> 860                     SOCK_UNLOCK(so);
> 861     #endif
> 862
> 863                     so2 = so3;
> 864             }
> 865             error = unp_connect2(so, so2);
> 866     bad2:
> 867             UNP_UNLOCK();
> 868             mtx_lock(&Giant);
> 869     bad:
> 
> (gdb) l *uipc_connect+0x76
> 0x2dd is in uipc_connect (/usr/src/sys/kern/uipc_usrreq.c:179).
> 174             KASSERT(td == curthread, ("uipc_connect: td != curthread"));
> 175
> 176             if (unp == NULL)
> 177                     return (EINVAL);
> 178             UNP_LOCK();
> 179             error = unp_connect(so, nam, td);
> 180             UNP_UNLOCK();
> 181             return (error);
> 182     }
> 183
> 
> (gdb) l *soconnect+0x54
> 0x100f is in soconnect (/usr/src/sys/kern/uipc_socket.c:518).
> 513                 (error = sodisconnect(so))))
> 514                     error = EISCONN;
> 515             else
> 516                     error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
> 517             return (error);
> 518     }
> 519
> 520     int
> 521     soconnect2(so1, so2)
> 522             struct socket *so1;
> 
> (gdb) l *kern_connect+0xb
> 0xd5e is in kern_connect (/usr/src/sys/kern/uipc_syscalls.c:477).
> 472     int
> 473     kern_connect(td, fd, sa)
> 474             struct thread *td;
> 475             int fd;
> 476             struct sockaddr *sa;
> 477     {
> 478             struct socket *so;
> 479             int error, s;
> 480             int interrupted = 0;
> 481
>