ftrace: Fix removing of second function probe
authorSteven Rostedt (VMware) <rostedt@goodmis.org>
Fri, 14 Apr 2017 21:45:45 +0000 (17:45 -0400)
committerSteven Rostedt (VMware) <rostedt@goodmis.org>
Sat, 15 Apr 2017 21:04:37 +0000 (17:04 -0400)
commitacceb72e90624ec6c3f2fc62e72dab892cd075da
tree3529cecc00d61587f0bdf849823b85edaa9917b0
parentd54b6eeb553c89ed8d4c5a2ed73df374a45b9562
ftrace: Fix removing of second function probe

When two function probes are added to set_ftrace_filter, and then one of
them is removed, the update to the function locations is not performed, and
the record keeping of the function states are corrupted, and causes an
ftrace_bug() to occur.

This is easily reproducable by adding two probes, removing one, and then
adding it back again.

 # cd /sys/kernel/debug/tracing
 # echo schedule:traceoff > set_ftrace_filter
 # echo do_IRQ:traceoff > set_ftrace_filter
 # echo \!do_IRQ:traceoff > /debug/tracing/set_ftrace_filter
 # echo do_IRQ:traceoff > set_ftrace_filter

Causes:
 ------------[ cut here ]------------
 WARNING: CPU: 2 PID: 1098 at kernel/trace/ftrace.c:2369 ftrace_get_addr_curr+0x143/0x220
 Modules linked in: [...]
 CPU: 2 PID: 1098 Comm: bash Not tainted 4.10.0-test+ #405
 Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v02.05 05/07/2012
 Call Trace:
  dump_stack+0x68/0x9f
  __warn+0x111/0x130
  ? trace_irq_work_interrupt+0xa0/0xa0
  warn_slowpath_null+0x1d/0x20
  ftrace_get_addr_curr+0x143/0x220
  ? __fentry__+0x10/0x10
  ftrace_replace_code+0xe3/0x4f0
  ? ftrace_int3_handler+0x90/0x90
  ? printk+0x99/0xb5
  ? 0xffffffff81000000
  ftrace_modify_all_code+0x97/0x110
  arch_ftrace_update_code+0x10/0x20
  ftrace_run_update_code+0x1c/0x60
  ftrace_run_modify_code.isra.48.constprop.62+0x8e/0xd0
  register_ftrace_function_probe+0x4b6/0x590
  ? ftrace_startup+0x310/0x310
  ? debug_lockdep_rcu_enabled.part.4+0x1a/0x30
  ? update_stack_state+0x88/0x110
  ? ftrace_regex_write.isra.43.part.44+0x1d3/0x320
  ? preempt_count_sub+0x18/0xd0
  ? mutex_lock_nested+0x104/0x800
  ? ftrace_regex_write.isra.43.part.44+0x1d3/0x320
  ? __unwind_start+0x1c0/0x1c0
  ? _mutex_lock_nest_lock+0x800/0x800
  ftrace_trace_probe_callback.isra.3+0xc0/0x130
  ? func_set_flag+0xe0/0xe0
  ? __lock_acquire+0x642/0x1790
  ? __might_fault+0x1e/0x20
  ? trace_get_user+0x398/0x470
  ? strcmp+0x35/0x60
  ftrace_trace_onoff_callback+0x48/0x70
  ftrace_regex_write.isra.43.part.44+0x251/0x320
  ? match_records+0x420/0x420
  ftrace_filter_write+0x2b/0x30
  __vfs_write+0xd7/0x330
  ? do_loop_readv_writev+0x120/0x120
  ? locks_remove_posix+0x90/0x2f0
  ? do_lock_file_wait+0x160/0x160
  ? __lock_is_held+0x93/0x100
  ? rcu_read_lock_sched_held+0x5c/0xb0
  ? preempt_count_sub+0x18/0xd0
  ? __sb_start_write+0x10a/0x230
  ? vfs_write+0x222/0x240
  vfs_write+0xef/0x240
  SyS_write+0xab/0x130
  ? SyS_read+0x130/0x130
  ? trace_hardirqs_on_caller+0x182/0x280
  ? trace_hardirqs_on_thunk+0x1a/0x1c
  entry_SYSCALL_64_fastpath+0x18/0xad
 RIP: 0033:0x7fe61c157c30
 RSP: 002b:00007ffe87890258 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
 RAX: ffffffffffffffda RBX: ffffffff8114a410 RCX: 00007fe61c157c30
 RDX: 0000000000000010 RSI: 000055814798f5e0 RDI: 0000000000000001
 RBP: ffff8800c9027f98 R08: 00007fe61c422740 R09: 00007fe61ca53700
 R10: 0000000000000073 R11: 0000000000000246 R12: 0000558147a36400
 R13: 00007ffe8788f160 R14: 0000000000000024 R15: 00007ffe8788f15c
  ? trace_hardirqs_off_caller+0xc0/0x110
 ---[ end trace 99fa09b3d9869c2c ]---
 Bad trampoline accounting at: ffffffff81cc3b00 (do_IRQ+0x0/0x150)

Cc: stable@vger.kernel.org
Fixes: 59df055f1991 ("ftrace: trace different functions with a different tracer")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
kernel/trace/ftrace.c