llseek return ESPIPE when calling zcat all.tgz | ./cycletee

Question

I modified GNU tee to cycletee Source Code (You can download the binary from https://github.com/vls/cycletee/tree/master/bin)

What it does can be explained by the following example:

seq 10 | cycletee 1.txt 2.txt 3.txt
cat 1.txt // prints 1, 4, 7, 10
cat 2.txt // prints 2, 5, 8
cat 3.txt // prints 3, 6, 9

---

Then there is a all.tgz (See Appendix for building script)

all.tgz has three text file and totally 9000000 lines.

Everything is fine. Like:

seq 10000000 | ./cycletee 1.txt 2.txt 3.txt

zcat all.tgz | tee 1.txt > /dev/null

zcat all.tgz | tail // got 9000000 at the last line

except calling:

zcat all.tgz | ./cycletee 1.txt 2.txt 3.txt

when it reads the No.3000000 line, it exits.

strace it I got this message and it exited:

_llseek(2, 0, 0xffbec3d0, SEEK_CUR)     = -1 ESPIPE (Illegal seek)

Question

• Anyone can point out the problem of my source code?

• Any debugging technique to figure out the problem will be appreciated. I don't know how to use gdb in this situation.

Appendix

• all.tgz can be built by this Python sciprt https://gist.github.com/1500742

• Environment: Ubuntu 10.04 32bit, CentOS 5.4 64bit

Answer 1

From the source:

read:
    buffer[0] = '\0';
    ptr = fgets(buffer, (int) sizeof buffer, stdin);
    if(NULL == ptr) {
        if(ferror(stdin)) {
            error (0, errno, "%s", _("standard input"));
            ok = false;
        }
        flag_break = true;
        break;
    }
    bytes_read = strlen(buffer);


      if (bytes_read < 0 && errno == EINTR)
      {
          flag_continue = true;
          backup_i = i;
          break;
      }
      if (bytes_read <= 0) {
          flag_break = true;
          break;
      }

    if (descriptors[0]
            && fwrite(buffer, bytes_read, 1, descriptors[0]) != 1)
            {
                error (0, errno, "%s", files[0]);
                descriptors[0] = NULL;
                ok = false;
            }
 ...

I don't think this will work on binary input (input that contains NULs). [ Given the signedness of bytes_read, I have the strong suspicion that fread() has been replaced by fgets() + strlen(); ] This may or may not be the cause of the PIPE error, but it looks very wrong.

Answer 2

You cannot call llseek or ftell on a pipe or socket, they are not seekable files.

You could use a debugger like gdb (it is really worth learning to use it; and GDB is very well documented), and e.g. put a breakpoint on _llseek

You could also use strace or ltrace