[PATCH] unserialized task->files changing

[PATCH] unserialized task->files changing

[Kirill Korotaev]

[-- Attachment #1: Type: text/plain, Size: 294 bytes --]

Fixed race on put_files_struct on exec with proc.
Restoring files on current on error path may lead
to proc having a pointer to already kfree-d files_struct.

Found during OpenVZ stress testing.

Signed-Off-By: Pavel Emelianov <xemul@openvz.org>
Signed-Off-By: Kirill Korotaev <dev@openvz.org>

[-- Attachment #2: diff-ms-files-race-fix --]
[-- Type: text/plain, Size: 1463 bytes --]

--- ./fs/binfmt_elf.c.fsfix	2006-03-27 14:25:59.000000000 +0400
+++ ./fs/binfmt_elf.c	2006-03-28 13:26:16.000000000 +0400
@@ -1027,8 +1027,13 @@ out_free_file:
 	sys_close(elf_exec_fileno);
 out_free_fh:
 	if (files) {
-		put_files_struct(current->files);
+		struct files_struct *old;
+
+		old = current->files;
+		task_lock(current);
 		current->files = files;
+		task_unlock(current);
+		put_files_struct(old);
 	}
 out_free_ph:
 	kfree(elf_phdata);
--- ./fs/binfmt_misc.c.fsfix	2006-03-27 14:25:59.000000000 +0400
+++ ./fs/binfmt_misc.c	2006-03-28 13:27:06.000000000 +0400
@@ -216,8 +216,13 @@ _error:
 	bprm->interp_data = 0;
 _unshare:
 	if (files) {
-		put_files_struct(current->files);
+		struct files_struct *old;
+
+		old = current->files;
+		task_lock(current);
 		current->files = files;
+		task_unlock(current);
+		put_files_struct(old);
 	}
 	goto _ret;
 }
--- ./fs/exec.c.fsfix	2006-03-27 14:25:59.000000000 +0400
+++ ./fs/exec.c	2006-03-28 13:28:10.000000000 +0400
@@ -865,7 +865,7 @@ int flush_old_exec(struct linux_binprm *
 {
 	char * name;
 	int i, ch, retval;
-	struct files_struct *files;
+	struct files_struct *files, *old;
 	char tcomm[sizeof(current->comm)];
 
 	/*
@@ -946,8 +946,11 @@ int flush_old_exec(struct linux_binprm *
 	return 0;
 
 mmap_failed:
-	put_files_struct(current->files);
+	old = current->files;
+	task_lock(current);
 	current->files = files;
+	task_unlock(current);
+	put_files_struct(old);
 out:
 	return retval;
 }

Re: [PATCH] unserialized task->files changing

[Christoph Hellwig]

On Tue, Aug 08, 2006 at 02:07:49PM +0400, Kirill Korotaev wrote:
> Fixed race on put_files_struct on exec with proc.
> Restoring files on current on error path may lead
> to proc having a pointer to already kfree-d files_struct.

This is three times the exact same code sequence, it should probably go into
a helper:

void reset_current_files(struct files_struct *files)
{
	struct files_struct *old = current->files;

	task_lock(current);
	current->files = files;
	task_unlock(current);
	put_files_struct(old);
}

Re: [PATCH] unserialized task->files changing

[Eric Dumazet]

On Tuesday 08 August 2006 12:12, Christoph Hellwig wrote:
> On Tue, Aug 08, 2006 at 02:07:49PM +0400, Kirill Korotaev wrote:
> > Fixed race on put_files_struct on exec with proc.
> > Restoring files on current on error path may lead
> > to proc having a pointer to already kfree-d files_struct.
>
> This is three times the exact same code sequence, it should probably go
> into a helper:
>
> void reset_current_files(struct files_struct *files)
> {
> 	struct files_struct *old = current->files;
>
> 	task_lock(current);
> 	current->files = files;
> 	task_unlock(current);
> 	put_files_struct(old);
> }


More over I think you want to task_lock() before reading current->files 
into 'old'

task_lock(current);
old = current->files;
current->files = files;
task_unlock(current);
put_files_struct(old);

or maybe a xchg() ?

Re: [PATCH] unserialized task->files changing

[Kirill Korotaev]

Eric Dumazet wrote:
> On Tuesday 08 August 2006 12:12, Christoph Hellwig wrote:
> 
>>On Tue, Aug 08, 2006 at 02:07:49PM +0400, Kirill Korotaev wrote:
>>
>>>Fixed race on put_files_struct on exec with proc.
>>>Restoring files on current on error path may lead
>>>to proc having a pointer to already kfree-d files_struct.
>>
>>This is three times the exact same code sequence, it should probably go
>>into a helper:
>>
>>void reset_current_files(struct files_struct *files)
>>{
>>	struct files_struct *old = current->files;
>>
>>	task_lock(current);
>>	current->files = files;
>>	task_unlock(current);
>>	put_files_struct(old);
>>}
> 
> 
> 
> More over I think you want to task_lock() before reading current->files 
> into 'old'
> 
> task_lock(current);
> old = current->files;
> current->files = files;
> task_unlock(current);
> put_files_struct(old);
> 
> or maybe a xchg() ?

yeah, never do assignments in declarations :)

BTW, not sure about kthread_exit_files() yet, but looks like it suffers too.

unshare_files() changes current->files w/o locking as well. but I can't see
where it puts the old files... hmm...

Kirill