inn-2.4: makehistory -O -x -F SIGSEGV (Segmentation fault)

[Alex Kiernan]

Alex Kiernan <alexk at demon.net> writes:

> list-inn-workers at news.cistron.nl (Miquel van Smoorenburg) writes:
> 
> > If 'usemmap' is 'false' in inn.conf, makehistory -O -x -F crashes
> > with SEGV. Looks a lot like the problem I reported earlier this
> > week and that Katshuhiro fixed promptly.
> > 
> > An strace shows that it first opens all my CNFS buffers, and then
> > after the first pread() it crashes:
> > 
> > $ strace makehistory -O -x -F
> > [....]
> > open("/news/disk8/5", O_RDWR)           = 34
> > fcntl(34, F_GETFD)                      = 0
> > fcntl(34, F_SETFD, FD_CLOEXEC)          = 0
> > mmap(NULL, 491520, PROT_READ|PROT_WRITE, MAP_SHARED, 34, 0) = 0x40f5a000
> > open("/news/disk8/6", O_RDWR)           = 35
> > fcntl(35, F_GETFD)                      = 0
> > fcntl(35, F_SETFD, FD_CLOEXEC)          = 0
> > mmap(NULL, 491520, PROT_READ|PROT_WRITE, MAP_SHARED, 35, 0) = 0x40fd2000
> > gettimeofday({986566986, 760413}, NULL) = 0
> > pread(4, "X)2]LO_66QO4", 12, 392221184) = 12
> > --- SIGSEGV (Segmentation fault) ---
> > +++ killed by SIGSEGV +++
> > 
> > Setting 'usemmap' to true seems to fix the problem..
> > 
> 
> Can you get a stack backtrace out of it?
> 

Hmm... are you running with OLD_CNFS defined? Looks like I missed that
piece of the fix out (patch to follow), other than that I can't see
anything :-(

-- 
Alex Kiernan, Principal Engineer, Development, Thus PLC