On 11/5/20 1:18 PM, Kelly Wang (kellythw) wrote:
With the conftest.c you provided, strace still hang.
Check for how many calls for chdir("confdir3"), it only has 110 times, then hang after
mkdir("confdir3", 0700 ...
Is there any directory limitation that can make on a server?
Wow, that's a more-serious kernel (or filesystem) bug than I thought: the mkdir
system call is hanging and does not appear to be interruptible via SIGALRM.
When the program hangs, how do you terminate it? Do you use Control-C from a
terminal? If so, what happens if you instead use 'kill'? Something like this:
rm -fr conftest3
gcc conftest.c
strace -o tr ./a.out &
sleep 1
kill -s INT $(ps -o pid= -C a.out)
That last line should send the SIGINT signal to the a.out command; does this
cause a.out to exit? (You can look at 'tr' to see.) If it exits, perhaps we can
modify conftest3 to do the same thing to itself when it is running on a buggy
kernel.
Also, what happens if you do the same recipe as above, but use 'ALRM' rather
than 'INT'? Again, look at the end of 'tr'.
