ptrace: it is fun to strace /sbin/init
Subsystems (2)
PATCH 5/5 ptrace: it is fun to strace /sbin/init
Ptracing of /sbin/init is not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this.
Note that it is really dangerous, also because init can lose SIGNAL_UNKILLABLE flag. But this is because we are not careful enough with setting signal->flags, this should be cleanuped anyway.
Unless I missed something, ptrace_get_task_struct() is pointless. It does not need to check "pid == 1", ptrace_attach() does this. It doesn't need tasklist. It should be replaced with the generic find_get_task_by_vpid() which does not exist yet.
kernel/ptrace.c renamed from kernel/ptrace.c~5_INIT_PTRACE +10 -7
@@ -160,6 +160,15 @@ int ptrace_may_attach(struct task_structreturn !err; } +static int allow_ptrace_init; + +static int __init __allow_ptrace_init (char *str) +{ + allow_ptrace_init =1 ; +return 1 ; +} +__setup ("init_ptrace" , __allow_ptrace_init); +int ptrace_attach (struct task_struct *task) {int retval;
@@ -168,7 +177,7 @@ int ptrace_attach(struct task_struct *taaudit_ptrace (task); retval = -EPERM ; -if (task->pid <=1 ) +if (unlikely (is_global_init (task)) &&likely (!allow_ptrace_init)) goto out;if (same_thread_group (task, current)) goto out;
@@ -518,12 +527,6 @@ struct task_struct *ptrace_get_task_stru {struct task_struct *child; -/* -* Tracing init is not allowed. -*/ -if (pid ==1 ) -return ERR_PTR (-EPERM ); -read_lock (&tasklist_lock); child =find_task_by_vpid (pid);if (child)
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +2 -0
@@ -803,6 +803,8 @@ and is between 256 and 4096 characters. Run specified binary instead of /sbin/init as init process. + init_ptrace [KNL] Allows to ptrace init. Very dangerous. Don't use. + initcall_debug [KNL] Trace initcalls as they are executed. Useful for working out where the kernel is dying during startup.
Quoted 2 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this.
Personally I wouldn't mind losing all the ptrace/signals special cases for init. (Just don't have a buggy init and expect not to crash, don't be root and kill init, etc.) So this is fine by me. The conservative route of changing it only with a boot option is the wise thing to do.
Quoted 2 lines
Unless I missed something,
ptrace_get_task_struct()is pointless. It does not need to check "pid == 1",ptrace_attach()does this. It doesn't need tasklist.
Agreed. It's a hold-over from when there was more hair in there.
Quoted 2 lines
It should be replaced with the generic
find_get_task_by_vpid()which does not exist yet.
I didn't see enough other uses to really warrant it. Most find_task_by_vpid calls don't actually do get_task_struct. Those that do want to do some other check inside rcu_read_lock before deciding to bother with get_task_struct anyway. So there is nothing wrong with ptrace just open-coding:
rcu_read_lock(); child = find_task_by_vpid(pid); if (child) get_task_struct(child); rcu_read_unlock();
We are on the way soon to having no arch callers of ptrace_get_task_struct left, so only the two kernel/ptrace.c uses will survive. (x86 and ia64 switchovers to arch_ptrace/compat_arch_ptrace are already in the pipeline, and maybe s390 too.) So let's worry about the cleanup removing this function once those wither away.
Thanks,
Roland
Quoted 8 lines · Roland McGrath
Quoted 3 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this.Personally I wouldn't mind losing all the ptrace/signals special cases for init. (Just don't have a buggy init and expect not to crash, don't be root and kill init, etc.) So this is fine by me. The conservative route of changing it only with a boot option is the wise thing to do.
Great.
Quoted 14 lines
Quoted 2 lines
Unless I missed something,
ptrace_get_task_struct()is pointless. It does not need to check "pid == 1",ptrace_attach()does this. It doesn't need tasklist.Agreed. It's a hold-over from when there was more hair in there.
Quoted 2 lines
It should be replaced with the generic
find_get_task_by_vpid()which does not exist yet.I didn't see enough other uses to really warrant it. Most
find_task_by_vpidcalls don't actually doget_task_struct. Those that do want to do some other check insidercu_read_lockbefore deciding to bother withget_task_structanyway. So there is nothing wrong with ptrace just open-coding:
Quoted 5 lines
rcu_read_lock(); child = find_task_by_vpid(pid); if (child) get_task_struct(child); rcu_read_unlock();
proc_task_lookup(), fill_pid(), attach_task_by_pid(), can use the new helper.
But yes sure, we can open-code this.
Oleg.
Quoted 3 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this.
...
Documentation/kernel-parameters.txt and is between 256 and 4096 characters. · line 807
Quoted 6 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +2
-0@@ -803,6 +803,8 @@ and is between 256 and 4096 characters. Run specified binary instead of /sbin/init as init process. + init_ptrace [KNL] Allows to ptrace init. Very dangerous. Don't use. +
I don't know what ptracing init is good for, and I believe people wanting to do this kind of special stuff can patch their own kernel...
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Quoted 8 lines · Pavel Machek
Quoted 3 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this....
Quoted 4 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +0
-0@@ -803,6 +803,8 @@ and is between 256 and 4096 characters. Run specified binary instead of /sbin/init as init process.
Quoted 5 lines
Documentation/kernel-parameters.txt and is between 256 and 4096 characters. · line 807
Quoted 2 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +2
-0+ init_ptrace [KNL] Allows to ptrace init. Very dangerous. Don't use. +I don't know what ptracing init is good for, and I believe people wanting to do this kind of special stuff can patch their own kernel...
Yes sure. But could you explain why this can be bad given that ptracing init needs the explicit boot parameter? IOW, could you explain why you don't like this small and trivial change which adds a minimal impact ?
Most of the lkml readers can easily implement most of debugging options, do you mean there are useless? I personally like the possibility to trace init without re-compiling the kernel, because I'd like to know why /usr/bin/top sometimes shows init at the top.
Oleg.
Quoted 15 lines · Oleg Nesterov
Quoted 8 lines · Pavel Machek
Quoted 3 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this....
Quoted 4 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +0
-0@@ -803,6 +803,8 @@ and is between 256 and 4096 characters. Run specified binary instead of /sbin/init as init process.Quoted 5 lines
Documentation/kernel-parameters.txt and is between 256 and 4096 characters. · line 807
Quoted 2 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +2
-0+ init_ptrace [KNL] Allows to ptrace init. Very dangerous. Don't use. +I don't know what ptracing init is good for, and I believe people wanting to do this kind of special stuff can patch their own kernel...
Quoted 3 lines
Yes sure. But could you explain why this can be bad given that ptracing init needs the explicit boot parameter? IOW, could you explain why you don't like this small and trivial change which adds a minimal impact?
"It can't be bad, its optional".
It is bad exactly _because_ it is optional. Anything that adds boot parameter is *not* trivial...
Why not add
please_randomly_corrupt_memory boot parameter? It may be useful for something...
Pavel
Signature · 3 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html pomozte zachranit klanovicky les: http://www.ujezdskystrom.info/
Quoted 17 lines · Pavel Machek
Quoted 15 lines · Oleg Nesterov
Quoted 13 lines · Pavel Machek
Quoted 3 lines
Ptracing of
/sbin/initis not allowed. Of course, this is very dangerous, but may be useful. Introduce the kernel boot parameter to allow this....
Documentation/kernel-parameters.txt and is between 256 and 4096 characters. · line 807
Quoted 6 lines
Documentation/kernel-parameters.txt renamed from Documentation/kernel-parameters.txt~5_INIT_PTRACE +2
-0@@ -803,6 +803,8 @@ and is between 256 and 4096 characters. Run specified binary instead of /sbin/init as init process. + init_ptrace [KNL] Allows to ptrace init. Very dangerous. Don't use. +I don't know what ptracing init is good for, and I believe people wanting to do this kind of special stuff can patch their own kernel...
Quoted 8 lines
Quoted 3 lines
Yes sure. But could you explain why this can be bad given that ptracing init needs the explicit boot parameter? IOW, could you explain why you don't like this small and trivial change which adds a minimal impact?
"It can't be bad, its optional".
It is bad exactly _because_ it is optional. Anything that adds boot parameter is *not* trivial...
You are right. I'd prefer to make /sbin/init ptraceable unconditionally, the root should know what it does. But this will change the historical behaviour.
Quoted 5 lines
Why not add
please_randomly_corrupt_memoryboot parameter? It may be useful for something...
Pavel
Nice argument.
Oleg.