During the KDE sprint in Berlin, Roman Gilg leaned over to me and asked if I knew how to redirect the stderr of an already-running process to a file. I Googled it and found underwhelming answers using strace and trying to decipher the output by reading the write syscalls. Instead, I thought a gdb based approach would work better, and after putting the pieces together Roman insisted I wrote a blog post on the topic.
gdb, the GNU debugger, has two important features that make this possible:
- Attaching to running processes via
gdb -p - Executing arbitrary code in the target process space
With this it’s actually quite straightforward. The process is the following:
- Attach gdb to the running process
- Run
compile code -- dup2(open("/tmp/log", 65), 2)
The magic 65 here is the value of O_CREAT | O_WRONLY on Linux, which is easily
found with a little program like this:
#include <sys/stat.h>
#include <fcntl.h>
int main(int argc, char **argv) {
printf("%d\n", O_CREAT | O_WRONLY);
return 0;
}
2 is always the file descriptor assigned to stderr. What happens here is:
- Via
open, the file you want to redirect to is created. - Via
dup2, stderr is overwritten with this new file.
The compile code gdb command will compile some arbitrary C code and run the
result in the target process, presumably by mapping some executable RAM and
loading it in, then jumping to the blob. Closing gdb (control+d) will continue
the process, and it should start writing out to the file you created.
There are lots of other cool (and hacky) things you can do with gdb. I once disconnected someone from an internet radio by attaching gdb to nginx and closing their file descriptor, for example. Thanks to Roman for giving me the chance to write an interesting blog post on the subject!