[LCP]SIGBUS question

[Joachim Bauernberger]

Ok, After spending the evening with my best friend gdb I managed to 
pinpoint the location of the problem but have still no glue on what 
might be the cause :-((

As soon as myfunc() calls fork, the process looses all it's information 
on previous stackframes. At this point I haven't even called exec() 
yet.

Look here, those are the frames before calling fork (sorry if the lines 
wrap):

(gdb) bt
#0  local_inject (data=0xbfffa73c, rcpt=0x80d36b0 
"bj at bjdev.devel.gmx.net") at mda.c:198
#1  0x8060398 in mknotify (what=665, who=2) at report.c:432
#2  0x80504d9 in av_scan (entry=0xbfffb86c 
"/var/mxscan/scan//jbauernberger at gmx-ag.de-2002419232533-32132/1-base64-ZqKkai") 
at av_scan.c:50
#3  0x804a5cc in b64decode (b64str=0x80d7de4, tnef=0, checksum=0x0) at 
b64decode.c:92
#4  0x8050061 in mimeget (
    startpos=0x8071382 "\n\n\n", '-' <repeats 14 times>, 
"Boundary-00=_XXOTHOLISLOHA8DOE9Z7\nContent-Type: text/plain;\n  
charset=\"us-ascii\"\nContent-Transfer-Encoding: 8bit\nSubject: 
\n\n\n-- \nPGP Public 
Key:\nhttp://www.mXscan.org/pgp/public.html\n"...) at mimeget.c:292
#5  0x804dfb3 in header (mref=0xbffffb1c) at header.c:729
#6  0x8053b7c in main (argc=18, argv=0xbffffba4) at main.c:772
#7  0x40119baf in __libc_start_main () from /lib/libc.so.6

Now at line 210:
(gdb) n
210         switch ((pid = fork ())) {

Immediately after fork the frame history is gone.
(gdb) bt
#0  local_inject (data=0xbfffa73c, rcpt=0x80d36b0 
"bj at bjdev.devel.gmx.net") at mda.c:210
(gdb)

That's it. The process then segfaults on return:
(gdb) cont
Continuing.
Program received signal SIGSEGV, Segmentation fault.

0x806059d in mknotify (what=Cannot access memory at address 0xfffffffe
) at report.c:432
432                     rc = local_inject(&notification, rcpt);


Thanks,
Joachim

On Friday 19 April 2002 14:11, Joachim Bauernberger wrote:
> Hi,
> Can someone please take the time to explain to me what can cause a 
> SIGBUS to be raised by a program (other than mmap) under Linux?
> I have been "googling" for answers and found that it's cause seems to 
be 
> architecture dependent? 
> (I've been able to catch it by setting up a signal handler but this 
> doesn't give much information other than it was indeed a SIGBUS that 
> crashed the program)
> 
> Also raise()ing a SIGSTOP and then attaching to the process with gdb 
> shows me it's a SIGBUS but the information gdb gives me is not very 
> descriptive about why the problem occurs. It seems that on return from 
> myfunc() (which calls fork/exec and writes and reads with pipes 
to/from 
> the forked child) the programs memory gets corrupted.
> 
> I started getting the problem after moving some code into myfunc() . 
> Now when the myfunc() returns the SIGBUS 
> results. Looking at gdb's output:
> 
> Program received signal SIGBUS, Bus error.
> mknotify (what=Cannot access memory at address 0xffffef36.
> ) at report.c:450
> 450             if (recipient) free(recipient);
> 
> mknotify() is the function calling myfunc()  and "what" is an int 
passed 
> to it as it's first parameter.
> 
> This confuses me since I am not calling mknotify() any more upon 
return 
> >from myfunc() (before the sigbus).
> Am I right to say that something in myfunc() corrupts the stack of my 
> program so badly that upon it's return the program does not know 
> anymore where to carry on?
> 
> The system I have the problem is a linux libc5 box (i386) Kernel 
2.0.38
> (don't tell me to upgrade please :-))
> 
> Thanks & Regards,
> Joachim
> 
> -- 
> PGP Public Key:
> http://www.mXscan.org/pgp/public.html
> 
> 
> _______________________________________________
> This is the Linux C Programming List
> :  http://lists.linux.org.au/listinfo/linuxcprogramming List
> 
> 

-- 
PGP Public Key:
http://www.mXscan.org/pgp/public.html