1

I have a program that uses the infiniband rdmacm library rdmacm.so

On one computer (Ubuntu server), I can run it without issues. On my dev PC (Ubuntu desktop edition), I get:

./test-client rdmacm.so: cannot open shared object file: No such file or directory

Here's where I get stumped.

ldd rdma-client

linux-vdso.so.1 =>  (0x00007fffdb62b000)
libibverbs.so.1 => /usr/lib/libibverbs.so.1 (0x00007f97ca007000)
librdmacm.so.1 => /usr/lib/librdmacm.so.1 (0x00007f97c9dfe000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f97c9a3e000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f97c9821000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f97c961d000)
/lib64/ld-linux-x86-64.so.2 (0x00007f97ca237000)

cat /etc/ld.so.conf

include /etc/ld.so.conf.d/*.conf

cat /etc/ld.so.conf.d/*.conf

# Multiarch support
/lib/i386-linux-gnu
/usr/lib/i386-linux-gnu
/lib/i686-linux-gnu
/usr/lib/i686-linux-gnu
# libc default configuration
/usr/local/lib
# Multiarch support
/lib/x86_64-linux-gnu
/usr/lib/x86_64-linux-gnu
/usr/lib/fglrx
/usr/lib32/fglrx
# Legacy biarch compatibility support
/lib32
/usr/lib32

ls -l /usr/lib/librdmacm*

-rw-r--r-- 1 root root 41146 Jul 19  2011 /usr/lib/librdmacm.a
lrwxrwxrwx 1 root root    18 Jul 19  2011 /usr/lib/librdmacm.so -> librdmacm.so.1.0.0
lrwxrwxrwx 1 root root    18 Jul 19  2011 /usr/lib/librdmacm.so.1 -> librdmacm.so.1.0.0
-rw-r--r-- 1 root root 35248 Jul 19  2011 /usr/lib/librdmacm.so.1.0.0

Everything looks correct. Why can't I run test-client.

EDIT

The code that I'm using is from the geekinthecorner blog. Infiniband test apps.

In the client it has a couple of dlopen calls:

i.e. 

void *handle = dlopen("rdmacm.so", RTLD_LAZY);
...  
handle = dlopen("ibverbs.so", RTLD_LAZY);

This works on on ubuntu server. However on my development desktop machine it doesn't find the libraries.

if I rename the libraries to like this

void *handle = dlopen("librdmacm.so", RTLD_LAZY);
...  
handle = dlopen("libibverbs.so", RTLD_LAZY);

They are found. Doesn't dlopen prepend "lib" automatically? I'm assuming yes that must be the case on my server because the libraries are found without this.

In any case, I'm not sure I need these dlopen calls. I can remove them fully and the program works. But, now I'm curious as to why dlopen is performing differently on the two machines given that the paths and /etc/ld.so.conf setup contain the same search paths.

hookenz
  • 36,432
  • 45
  • 177
  • 286
  • `dlopen()` does not prepend `lib`, because the parameter is a file name, and not a library name. If you specify an explicit path (containing at least one `/`), the library directories are not looked in, only the one file the path specifies. See `man 3 dlopen` or http://www.kernel.org/doc/man-pages/online/pages/man3/dlopen.3.html for details. Did you even check if `handle = dlopen("ibverbs.so", RTLD_LAZY)` returns non-NULL, ever? I think it always returns NULL. – Nominal Animal Dec 14 '12 at 09:52

1 Answers1

1

void *dlopen(const char *filename, int flag);

The dlopen() loads the dynamic library file named by the null-terminated string filename and returns an opaque "handle" for the dynamic library. If the library has dependencies on other shared libraries, then these are also automatically loaded by the dynamic linker, recursively. So, whatever be the case a standard dlopen() will never prepend or do any modification to the filename string.

The specified filename is searched in the following manner:

  • (ELF only) If the executable file for the calling program contains a DT_RPATH tag, and does not contain a DT_RUNPATH tag, then the directories listed in the DT_RPATH tag are searched.

  • If, at the time that the program was started, the environment variable LD_LIBRARY_PATH was defined to contain a colon-separated list of directories, then these are searched. (As a security measure this variable is ignored for set-user-ID and set-group-ID programs.)

  • (ELF only) If the executable file for the calling program contains a DT_RUNPATH tag, then the directories listed in that tag are searched.

  • The cache file /etc/ld.so.cache (maintained by ldconfig(8)) is checked to see whether it contains an entry for filename.

  • The directories /lib and /usr/lib are searched (in that order).

So, this strange behavior that you notice for dlopen() might possibly be due to an existing symlinks/hardlinks to the library librdmacm.so and libibverbs.so as rdmacm.so and ibverbs.so , in any of the above mentioned library search paths.

If dlopen() fails for any reason, it returns NULL. Check for NULL before using the handle returned by dlopen.

handle = dlopen("libxyz.so", RTLD_LAZY);
if (!handle)
{
     fprintf(stderr, "%s\n", dlerror());
     exit(EXIT_FAILURE);
}

Refer: man 3 dlopen

askmish
  • 6,464
  • 23
  • 42
  • That's what I thought. But I could find no symbolic link in any search path! – hookenz Dec 16 '12 at 19:58
  • Doh. You're right. Turned out to be something subtle that I didn't notice. Long story. Yes, they should have lib prepended. – hookenz Dec 17 '12 at 00:22