/proc/sys/vm/drop_caches: add error handling

/proc/sys/vm/drop_caches: add error handling

[Pavel Machek]

Document that drop_caches is unsafe, and add error checking so that it
bails out on invalid inputs. [Note that this was triggered by Android
trying to use it in production, and incidentally writing invalid
value...]

Signed-off-by: Pavel Machek <pavel@suse.cz> 

diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
index 3197fc8..b90f050 100644
--- a/Documentation/sysctl/vm.txt
+++ b/Documentation/sysctl/vm.txt
@@ -132,6 +132,10 @@ To free pagecache, dentries and inodes:
 As this is a non-destructive operation and dirty objects are not freeable, the
 user should run `sync' first.
 
+Note that calling this causes some serious latencies, and that this is
+a debug feature; it should not be used for production as it does not
+contain neccessary locking to guarantee safe operation.
+
 ==============================================================
 
 hugepages_treat_as_movable
diff --git a/fs/drop_caches.c b/fs/drop_caches.c
index 3e5637f..765dec3 100644
--- a/fs/drop_caches.c
+++ b/fs/drop_caches.c
@@ -9,8 +9,9 @@
 #include <linux/sysctl.h>
 #include <linux/gfp.h>
 
-/* A global variable is a bit ugly, but it keeps the code simple */
-int sysctl_drop_caches;
+/* A global variable is a bit ugly, and has locking problems,
+   but it keeps the code simple */
+unsigned int sysctl_drop_caches;
 
 static void drop_pagecache_sb(struct super_block *sb)
 {
@@ -65,12 +66,17 @@ static void drop_slab(void)
 int drop_caches_sysctl_handler(ctl_table *table, int write,
 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
 {
-	proc_dointvec_minmax(table, write, file, buffer, length, ppos);
-	if (write) {
-		if (sysctl_drop_caches & 1)
-			drop_pagecache();
-		if (sysctl_drop_caches & 2)
-			drop_slab();
-	}
-	return 0;
+	int res;
+	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
+	if (res)
+		return res;
+	if (!write)
+		return res;
+	if (sysctl_drop_caches & ~3)
+		return -EINVAL;
+	if (sysctl_drop_caches & 1)
+		drop_pagecache();
+	if (sysctl_drop_caches & 2)
+		drop_slab();
+	return res;
 }



-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: /proc/sys/vm/drop_caches: add error handling

[KOSAKI Motohiro]

> 
> Document that drop_caches is unsafe, and add error checking so that it
> bails out on invalid inputs. [Note that this was triggered by Android
> trying to use it in production, and incidentally writing invalid
> value...]

Yup. good patch.

> -	return 0;
> +	int res;
> +	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
> +	if (res)
> +		return res;
> +	if (!write)
> +		return res;
> +	if (sysctl_drop_caches & ~3)
> +		return -EINVAL;
> +	if (sysctl_drop_caches & 1)
> +		drop_pagecache();
> +	if (sysctl_drop_caches & 2)
> +		drop_slab();
> +	return res;
>  }

I think following is clarify more.

	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
	if (res)
		return res;
	if (!write)
		return 0;
	if (sysctl_drop_caches & ~3)
		return -EINVAL;
	if (sysctl_drop_caches & 1)
		drop_pagecache();
	if (sysctl_drop_caches & 2)
		drop_slab();
	return 0;


otherthings, _very_ looks good to me. :)

Re: /proc/sys/vm/drop_caches: add error handling

[Andrew Morton]

On Tue,  3 Feb 2009 20:47:56 +0900 (JST)
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> > 
> > Document that drop_caches is unsafe, and add error checking so that it
> > bails out on invalid inputs. [Note that this was triggered by Android
> > trying to use it in production, and incidentally writing invalid
> > value...]
> 
> Yup. good patch.
> 
> > -	return 0;
> > +	int res;
> > +	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
> > +	if (res)
> > +		return res;
> > +	if (!write)
> > +		return res;
> > +	if (sysctl_drop_caches & ~3)
> > +		return -EINVAL;
> > +	if (sysctl_drop_caches & 1)
> > +		drop_pagecache();
> > +	if (sysctl_drop_caches & 2)
> > +		drop_slab();
> > +	return res;
> >  }
> 
> I think following is clarify more.
> 
> 	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
> 	if (res)
> 		return res;
> 	if (!write)
> 		return 0;
> 	if (sysctl_drop_caches & ~3)
> 		return -EINVAL;
> 	if (sysctl_drop_caches & 1)
> 		drop_pagecache();
> 	if (sysctl_drop_caches & 2)
> 		drop_slab();
> 	return 0;
> 
> 
> otherthings, _very_ looks good to me. :)
> 

For better or for worse, my intent here was to be
future-back-compatible.  So if we later add new flags, and people write
code which uses those new flags, that code won't break on old kernels.

Probably that wasn't a very good idea, and such userspace code isn't
very good.

Re: /proc/sys/vm/drop_caches: add error handling

[Pavel Machek]

Hi!

> > I think following is clarify more.
> > 
> > 	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
> > 	if (res)
> > 		return res;
> > 	if (!write)
> > 		return 0;
> > 	if (sysctl_drop_caches & ~3)
> > 		return -EINVAL;
> > 	if (sysctl_drop_caches & 1)
> > 		drop_pagecache();
> > 	if (sysctl_drop_caches & 2)
> > 		drop_slab();
> > 	return 0;
> > 
> > 
> > otherthings, _very_ looks good to me. :)
> 
> For better or for worse, my intent here was to be
> future-back-compatible.  So if we later add new flags, and people write
> code which uses those new flags, that code won't break on old kernels.

Well, should I move test for &~3 to the end? You'd still get
compatibility, but you'd also get expected error reports.

> Probably that wasn't a very good idea, and such userspace code isn't
> very good.

--- 

Report errors in drop_caches, and document stuff a bit better. 

Signed-off-by: Pavel Machek <pavel@suse.cz>

diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt
index 3197fc8..b90f050 100644
--- a/Documentation/sysctl/vm.txt
+++ b/Documentation/sysctl/vm.txt
@@ -132,6 +132,10 @@ To free pagecache, dentries and inodes:
 As this is a non-destructive operation and dirty objects are not freeable, the
 user should run `sync' first.
 
+Note that calling this causes some serious latencies, and that this is
+a debug feature; it should not be used for production as it does not
+contain neccessary locking to guarantee safe operation.
+
 ==============================================================
 
 hugepages_treat_as_movable
diff --git a/fs/drop_caches.c b/fs/drop_caches.c
index 3e5637f..f913a64 100644
--- a/fs/drop_caches.c
+++ b/fs/drop_caches.c
@@ -9,8 +9,9 @@
 #include <linux/sysctl.h>
 #include <linux/gfp.h>
 
-/* A global variable is a bit ugly, but it keeps the code simple */
-int sysctl_drop_caches;
+/* A global variable is a bit ugly, and has locking problems,
+   but it keeps the code simple */
+unsigned int sysctl_drop_caches;
 
 static void drop_pagecache_sb(struct super_block *sb)
 {
@@ -65,12 +66,17 @@ static void drop_slab(void)
 int drop_caches_sysctl_handler(ctl_table *table, int write,
 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
 {
-	proc_dointvec_minmax(table, write, file, buffer, length, ppos);
-	if (write) {
-		if (sysctl_drop_caches & 1)
-			drop_pagecache();
-		if (sysctl_drop_caches & 2)
-			drop_slab();
-	}
+	int res;
+	res = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
+	if (res)
+		return res;
+	if (!write)
+		return res;
+	if (sysctl_drop_caches & 1)
+		drop_pagecache();
+	if (sysctl_drop_caches & 2)
+		drop_slab();
+	if (sysctl_drop_caches & ~3)
+		return -EINVAL;
 	return 0;
 }


-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html