[Gluster-users] unexpected ENOENT on symlinks

Willem Jan Palenstijn wjp at usecode.org
Thu Jul 17 14:14:30 UTC 2014


I'm running into unexpected behaviour with symlinks on a glusterfs volume
(3.5.1 on Arch Linux), where sometimes accessing a symlink on the volume
results in ENOENT (No such file or directory).

The following consistently reproduces it for me from a shell. At the end of
this post I've also included a short snippet of C that shows this inconsistency
between repeated readlink() syscalls.

$ ln -s x y
$ ls -al
ls: cannot read symbolic link y: No such file or directory
total 16
drwxr-xr-x 2 wjp wjp 4096 Jul 17 15:40 .
drwx------ 7 wjp wjp 4096 Jul 17 15:40 ..
lrwxrwxrwx 1 wjp wjp    1 Jul 17 15:40 y
$ ls -l y
lrwxrwxrwx 1 wjp wjp 1 Jul 17 15:40 y -> x

Is this indeed unexpected, and if so, what would be a good next step to
investigate this?

The setup is glusterfs 3.5.1 on Arch Linux, with a stripe glusterfs volume
consisting of two bricks.

Willem Jan


$ gcc -o t test_glusterfs_symlink.c && ./t
First readlink: -1 (errno 2)
Second readlink: 1

$ cat test_glusterfs_symlink.c 
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <stdio.h>
#include <dirent.h>
#include <errno.h>

int main()
        FILE *f;
        int r1, r2, r3, e2, e3;
        struct dirent *p;
        DIR *d;
        char buf[2];
        struct stat s;

        r1 = symlink("x", "y");
	if (r1 != 0) {
		printf("Failed to create symlink\n");
		return -1;

        d = opendir(".");
        p = readdir(d);

        r2 = readlink("y", buf, 2);
	e2 = errno;
        r3 = readlink("y", buf, 2);
	e3 = errno;

	printf("First readlink: %d", r2);
	if (r2 < 0) printf(" (errno %d)", e2);
	printf("\nSecond readlink: %d", r3);
	if (r3 < 0) printf(" (errno %d)", e3);


	return 0;

More information about the Gluster-users mailing list