ftrace/x86: Remove possible deadlock between register_kprobe() and ftrace_run_update_...
authorPetr Mladek <pmladek@suse.com>
Thu, 27 Jun 2019 08:13:34 +0000 (10:13 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 10 Jul 2019 07:54:41 +0000 (09:54 +0200)
commit d5b844a2cf507fc7642c9ae80a9d585db3065c28 upstream.

The commit 9f255b632bf12c4dd7 ("module: Fix livepatch/ftrace module text
permissions race") causes a possible deadlock between register_kprobe()
and ftrace_run_update_code() when ftrace is using stop_machine().

The existing dependency chain (in reverse order) is:

-> #1 (text_mutex){+.+.}:
       validate_chain.isra.21+0xb32/0xd70
       __lock_acquire+0x4b8/0x928
       lock_acquire+0x102/0x230
       __mutex_lock+0x88/0x908
       mutex_lock_nested+0x32/0x40
       register_kprobe+0x254/0x658
       init_kprobes+0x11a/0x168
       do_one_initcall+0x70/0x318
       kernel_init_freeable+0x456/0x508
       kernel_init+0x22/0x150
       ret_from_fork+0x30/0x34
       kernel_thread_starter+0x0/0xc

-> #0 (cpu_hotplug_lock.rw_sem){++++}:
       check_prev_add+0x90c/0xde0
       validate_chain.isra.21+0xb32/0xd70
       __lock_acquire+0x4b8/0x928
       lock_acquire+0x102/0x230
       cpus_read_lock+0x62/0xd0
       stop_machine+0x2e/0x60
       arch_ftrace_update_code+0x2e/0x40
       ftrace_run_update_code+0x40/0xa0
       ftrace_startup+0xb2/0x168
       register_ftrace_function+0x64/0x88
       klp_patch_object+0x1a2/0x290
       klp_enable_patch+0x554/0x980
       do_one_initcall+0x70/0x318
       do_init_module+0x6e/0x250
       load_module+0x1782/0x1990
       __s390x_sys_finit_module+0xaa/0xf0
       system_call+0xd8/0x2d0

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(text_mutex);
                               lock(cpu_hotplug_lock.rw_sem);
                               lock(text_mutex);
  lock(cpu_hotplug_lock.rw_sem);

It is similar problem that has been solved by the commit 2d1e38f56622b9b
("kprobes: Cure hotplug lock ordering issues"). Many locks are involved.
To be on the safe side, text_mutex must become a low level lock taken
after cpu_hotplug_lock.rw_sem.

This can't be achieved easily with the current ftrace design.
For example, arm calls set_all_modules_text_rw() already in
ftrace_arch_code_modify_prepare(), see arch/arm/kernel/ftrace.c.
This functions is called:

  + outside stop_machine() from ftrace_run_update_code()
  + without stop_machine() from ftrace_module_enable()

Fortunately, the problematic fix is needed only on x86_64. It is
the only architecture that calls set_all_modules_text_rw()
in ftrace path and supports livepatching at the same time.

Therefore it is enough to move text_mutex handling from the generic
kernel/trace/ftrace.c into arch/x86/kernel/ftrace.c:

   ftrace_arch_code_modify_prepare()
   ftrace_arch_code_modify_post_process()

This patch basically reverts the ftrace part of the problematic
commit 9f255b632bf12c4dd7 ("module: Fix livepatch/ftrace module
text permissions race"). And provides x86_64 specific-fix.

Some refactoring of the ftrace code will be needed when livepatching
is implemented for arm or nds32. These architectures call
set_all_modules_text_rw() and use stop_machine() at the same time.

Link: http://lkml.kernel.org/r/20190627081334.12793-1-pmladek@suse.com
Fixes: 9f255b632bf12c4dd7 ("module: Fix livepatch/ftrace module text permissions race")
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Reported-by: Miroslav Benes <mbenes@suse.cz>
Reviewed-by: Miroslav Benes <mbenes@suse.cz>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
[
  As reviewed by Miroslav Benes <mbenes@suse.cz>, removed return value of
  ftrace_run_update_code() as it is a void function.
]
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/x86/kernel/ftrace.c
kernel/trace/ftrace.c

index c020ba4b7eb6fbc8764474e1968a3f94a221d67e..ccc2b9d2956aa4a80e3a0cb3f82346b84df190b4 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/init.h>
 #include <linux/list.h>
 #include <linux/module.h>
+#include <linux/memory.h>
 
 #include <trace/syscall.h>
 
@@ -36,6 +37,7 @@
 
 int ftrace_arch_code_modify_prepare(void)
 {
+       mutex_lock(&text_mutex);
        set_kernel_text_rw();
        set_all_modules_text_rw();
        return 0;
@@ -45,6 +47,7 @@ int ftrace_arch_code_modify_post_process(void)
 {
        set_all_modules_text_ro();
        set_kernel_text_ro();
+       mutex_unlock(&text_mutex);
        return 0;
 }
 
index ff3c8ca907c4c58f6c2c793e8bcc9673b94386e8..c4a0ad18c859365bcbf6fb2440abb97c807a7fab 100644 (file)
@@ -34,7 +34,6 @@
 #include <linux/hash.h>
 #include <linux/rcupdate.h>
 #include <linux/kprobes.h>
-#include <linux/memory.h>
 
 #include <trace/events/sched.h>
 
@@ -2693,12 +2692,10 @@ static void ftrace_run_update_code(int command)
 {
        int ret;
 
-       mutex_lock(&text_mutex);
-
        ret = ftrace_arch_code_modify_prepare();
        FTRACE_WARN_ON(ret);
        if (ret)
-               goto out_unlock;
+               return;
 
        /*
         * By default we use stop_machine() to modify the code.
@@ -2710,9 +2707,6 @@ static void ftrace_run_update_code(int command)
 
        ret = ftrace_arch_code_modify_post_process();
        FTRACE_WARN_ON(ret);
-
-out_unlock:
-       mutex_unlock(&text_mutex);
 }
 
 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
@@ -5800,7 +5794,6 @@ void ftrace_module_enable(struct module *mod)
        struct ftrace_page *pg;
 
        mutex_lock(&ftrace_lock);
-       mutex_lock(&text_mutex);
 
        if (ftrace_disabled)
                goto out_unlock;
@@ -5861,7 +5854,6 @@ void ftrace_module_enable(struct module *mod)
                ftrace_arch_code_modify_post_process();
 
  out_unlock:
-       mutex_unlock(&text_mutex);
        mutex_unlock(&ftrace_lock);
 
        process_cached_mods(mod->name);