Skip to content

Commit

Permalink
ftrace: Be first to run code modification on modules
Browse files Browse the repository at this point in the history
commit c1bf08a upstream.

If some other kernel subsystem has a module notifier, and adds a kprobe
to a ftrace mcount point (now that kprobes work on ftrace points),
when the ftrace notifier runs it will fail and disable ftrace, as well
as kprobes that are attached to ftrace points.

Here's the error:

 WARNING: at kernel/trace/ftrace.c:1618 ftrace_bug+0x239/0x280()
 Hardware name: Bochs
 Modules linked in: fat(+) stap_56d28a51b3fe546293ca0700b10bcb29__8059(F) nfsv4 auth_rpcgss nfs dns_resolver fscache xt_nat iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack lockd sunrpc ppdev parport_pc parport microcode virtio_net i2c_piix4 drm_kms_helper ttm drm i2c_core [last unloaded: bid_shared]
 Pid: 8068, comm: modprobe Tainted: GF            3.7.0-0.rc8.git0.1.fc19.x86_64 #1
 Call Trace:
  [<ffffffff8105e70f>] warn_slowpath_common+0x7f/0xc0
  [<ffffffff81134106>] ? __probe_kernel_read+0x46/0x70
  [<ffffffffa0180000>] ? 0xffffffffa017ffff
  [<ffffffffa0180000>] ? 0xffffffffa017ffff
  [<ffffffff8105e76a>] warn_slowpath_null+0x1a/0x20
  [<ffffffff810fd189>] ftrace_bug+0x239/0x280
  [<ffffffff810fd626>] ftrace_process_locs+0x376/0x520
  [<ffffffff810fefb7>] ftrace_module_notify+0x47/0x50
  [<ffffffff8163912d>] notifier_call_chain+0x4d/0x70
  [<ffffffff810882f8>] __blocking_notifier_call_chain+0x58/0x80
  [<ffffffff81088336>] blocking_notifier_call_chain+0x16/0x20
  [<ffffffff810c2a23>] sys_init_module+0x73/0x220
  [<ffffffff8163d719>] system_call_fastpath+0x16/0x1b
 ---[ end trace 9ef46351e53bbf80 ]---
 ftrace failed to modify [<ffffffffa0180000>] init_once+0x0/0x20 [fat]
  actual: cc:bb:d2:4b:e1

A kprobe was added to the init_once() function in the fat module on load.
But this happened before ftrace could have touched the code. As ftrace
didn't run yet, the kprobe system had no idea it was a ftrace point and
simply added a breakpoint to the code (0xcc in the cc:bb:d2:4b:e1).

Then when ftrace went to modify the location from a call to mcount/fentry
into a nop, it didn't see a call op, but instead it saw the breakpoint op
and not knowing what to do with it, ftrace shut itself down.

The solution is to simply give the ftrace module notifier the max priority.
This should have been done regardless, as the core code ftrace modification
also happens very early on in boot up. This makes the module modification
closer to core modification.

Link: http://lkml.kernel.org/r/[email protected]

Acked-by: Masami Hiramatsu <[email protected]>
Reported-by: Frank Ch. Eigler <[email protected]>
Signed-off-by: Steven Rostedt <[email protected]>
  • Loading branch information
Steven Rostedt authored and gregkh committed Jan 28, 2013
1 parent d027bb3 commit f2a0100
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion kernel/trace/ftrace.c
Original file line number Diff line number Diff line change
Expand Up @@ -3869,7 +3869,7 @@ static int ftrace_module_notify(struct notifier_block *self,

struct notifier_block ftrace_module_nb = {
.notifier_call = ftrace_module_notify,
.priority = 0,
.priority = INT_MAX, /* Run before anything that can use kprobes */
};

extern unsigned long __start_mcount_loc[];
Expand Down

0 comments on commit f2a0100

Please sign in to comment.