Unable to insert module, ftrace is not enabled about tpe-lkm HOT 13 CLOSED

This is due to a run time safety check that's failing. Can you give me the output of this command?

grep ftrace_enabled /proc/kallsyms

For example:

$ grep ftrace_enabled /proc/kallsyms
ffffffff81140480 t ftrace_enabled_open
ffffffff816c0700 r ftrace_enabled_fops
ffffffff81ae40e0 D ftrace_enabled
ffffffff81e6abc4 b kprobe_ftrace_enabled
ffffffff81ee9388 b last_ftrace_enabled

What's happening is it can't find the ftrace_enabled symbol - or it contains something it doesn't expect.

from tpe-lkm.

Also, I pushed cb8289c which adds additional logging when there's a problem. You'll see it by running the below after module insert:
sudo dmesg | grep fopskit

from tpe-lkm.

grep ftrace_enabled /proc/kallsyms
ffffffffa112b870 t ftrace_enabled_open

sudo dmesg | grep fopskit
fopskit: fopskit_find_sym_addr() failed with return code -14 for fops_hook { name => ftrace_enabled, addr => 0, found => 0, hooked => 0 } at fopskit_find_sym_addr() line 217

from tpe-lkm.

Do you have the kernel.kptr_restric sysctl enabled? Arch Linux appears to enable this by default and may be what's causing the issue. See:

https://bugs.archlinux.org/task/34323

I downloaded the kernel-ml-4.11 package on my CentOS system and this module still goes in fine.

from tpe-lkm.

I had it enabled previously. Above results are with kernel.kptr_restric sysctl=0.
With kernel.kptr_restric sysctl=1 it looks like this:

grep ftrace_enabled /proc/kallsyms
0000000000000000 t ftrace_enabled_open

Here's grep ftrace /proc/kallsyms
ftrace.txt

sysctl -a |grep kernel
sysctl.txt

from tpe-lkm.

Do you have CONFIG_DEBUG_INFO=y set in the kernel? I'm also reading how archlinux strips some symbols from the kernel while packaging. If this is the case, symbol address discovery would only be possible with a corresponding System.map of the running kernel, which adds risk of causing a fault on module insert. I haven't used archlinux before, so addressing this issue may take me some time.

from tpe-lkm.

zcat /proc/config.gz | grep CONFIG_DEBUG_INFO
# CONFIG_DEBUG_INFO is not set

Should I recompile it?

from tpe-lkm.

This module is meant to be inserted into precompiled distribution kernels. If you're compiling your own, the better approach is to include tpe at compile time. This module will require some changes to get that enabled - would you prefer that approach?

from tpe-lkm.

# CONFIG_DEBUG_INFO is not set is the default option in every Arch kernel, I didn't touch it. Anyway enabling this didn't change anything.

I really don't want to add more burden on you. Of course I can try compiling it with kernel if it's available but if I'm only one who needs this it wouldn't be worth your time.

from tpe-lkm.

Testing on Linux archiso 4.10.13-1-ARCH, if I simply remove the ftrace_enabled check it compiles and inserts just fine. Test suite passes. Except, when I rmmod tpe - the kernel panics. The stack trace shows kernel BUG at mm/vmalloc.c:15471 and none of the symbols on the stack are TPE related. Just an FYI.

Please test on your system, letting me know if it faults when you do the rmmod tpe.

Thanks,

from tpe-lkm.

Pushed commit 44e9f0a which moves the ftrace_enabled check to where it's really needed, which isn't a code path taken on archlinux. Please let me know if this resolves the issue for you.

from tpe-lkm.

sudo modprobe tpe
fopskit: fopskit_find_sym_addr() failed with return code -14 for fops_hook { name => selinux_enabled, addr => 0, found => 0, hooked => 0 } at fopskit_find_sym_addr() line 223
fopskit: fopskit_find_sym_addr() failed with return code -14 for fops_hook { name => selinux_disabled, addr => 0, found => 0, hooked => 0 } at fopskit_find_sym_addr() line 223
tpe: warning: cred->security was not remapped; the soften_mmap flag won't persist to child processes.
tpe: added to kernel

Module loaded. Arch didn't enable selinux by default. Could this be the issue? I enabled softmode in tpe config.
sudo rmmod tpe
dmesg.txt
Kernel didn't panic. Module unloaded.

from tpe-lkm.

Ok so you're at least now in a working state. Don't turn on SELinux - you'll then fail to inset the module, as the ftrace_enabled check only happens if SELinux is enabled. What happens if it is, is soften_mmap xattr flags, is any exist, will persist to child processes. Mostly this is only needed on desktops (Gnome) and byte compile build machines (such as Jenkins). It's only bee tested on kernels with SELinux, and I turn it off at insert time if it isn't enabled for safety. The cred->remap could collide with how other security module alloc memory.

Closing this issue since you're in a working state. If you so desire, send along that module unload dmesg to archlinux. It doesn't happen on my system with an unmodified 4.11 vanilla kernel.

from tpe-lkm.