[LU-1760] Misleading LNet Self Test message when lnet_selftest module is not loaded

• Status: Open
• Project: Lustre
• Type: Bug
• Priority: Minor
• Reporter: Prakash Surya (Inactive)
• Assignee: Amir Shehata (Inactive)
• Resolution: Unresolved
• Labels: llnl, patch
• Severity: 3

Description

When creating an lnet self test session without the lnet_selftest module loaded I see the following message:

-bash-4.1# lst new_session stress
Invalid parameters list in command line

This is misleading as the real problem is not the parameter list, but the fact that lnet_selftest is not loaded. To an uninformed user, this can be extremely confusing.

From a quick scan of the code, the error occurs here:

 611         rc = lst_new_session_ioctl(name, timeout, force, &session_id);
 612         if (rc != 0) {
 613                 lst_print_error("session", "Failed to create session: %s\n",
 614                                 strerror(errno));
 615                 return rc;
 616         }

Because the module is not loaded, the ioctl fails and sets errno to EINVAL. lst_print_error then interprets this error code and prints the misleading message. It would be much better if something was reported to the user which hinted at a possible unloaded module.

Something like the following would be much better:

-bash-4.1# lst new_session stress
Failed to create a new session with error EINVAL. This can be caused by incorrect parameters or because the lnet_selftest module is not loaded.

Better yet, if the ioctl fails, the lst command could parse /proc/modules to determine if lnet_selftest is loaded.

Comments

[Prakash Surya (Inactive)]

strace of the program's execution:

-bash-4.1# strace lst new_session stress
execve("/usr/sbin/lst", ["lst", "new_session", "stress"], [/* 24 vars */]) = 0
brk(0)                                  = 0x10030000
access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY)      = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=124024, ...}) = 0
mmap(NULL, 124024, PROT_READ, MAP_PRIVATE, 3, 0) = 0xfff7bef0000
close(3)                                = 0
open("/lib64/libc.so.6", O_RDONLY)      = 3
read(3, "\177ELF\2\2\1\0\0\0\0\0\0\0\0\0\0\3\0\25\0\0\0\1\0\0\0\0\0\35\r "..., 832) = 832
fstat(3, {st_mode=S_IFREG|0755, st_size=2172712, ...}) = 0
mmap(NULL, 1982072, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xfff7bd00000
mmap(0xfff7bec0000, 131072, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x1b0000) = 0xfff7bec0000
mmap(0xfff7bee0000, 15992, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xfff7bee0000
close(3)                                = 0
mprotect(0xfff7bec0000, 65536, PROT_READ) = 0
mprotect(0xfff7bf60000, 65536, PROT_READ) = 0
munmap(0xfff7bef0000, 124024)           = 0
brk(0)                                  = 0x10030000
brk(0x10060000)                         = 0x10060000
open("/dev/lnet", O_RDWR)               = 3
ioctl(3, 0xc008653f, 0xfffef191700)     = -1 EINVAL (Invalid argument)
write(2, "Invalid parameters list in comma"..., 40Invalid parameters list in command line
) = 40
exit_group(-1)                          = ?

[Peter Jones]

Isaac

Could you please take care of this one?

Thanks

Peter

[Peter Jones]

Doug

Isaac is out this week so can you help instead?

Thanks

Peter

[Andreas Dilger]

Possibly better would be to just modprobe lst and try it again. Easier for the user, and no need for a complex error message.

[Doug Oucharek (Inactive)]

The modprobe approach would certainly be more user friendly. Unless someone sees an issue with attempting the modprobe on this error condition, I'll go with that approach.

[Prakash Surya (Inactive)]

I agree, I like the idea of inserting the module for the user.

[Doug Oucharek (Inactive)]

The change created to address this ticket is: http://review.whamcloud.com/3752

[Doug Oucharek (Inactive)]

Amir: can you find an answer to Isaac's question in the Gerrit patch and rebase the patch once you do?