[patch] tlb_uv: handle large snprintf() returns

[patch] tlb_uv: handle large snprintf() returns

[Dan Carpenter]

snprintf() returns the number of bytes that *would* have been copied if
the buffer was large enough, so it can be larger than the size of the
buffer.  In this case it's ok, but let's put a cap on it anyway so it's
easier to audit.

Signed-off-by: Dan Carpenter <error27@gmail.com>

diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
index 312ef02..5e88b3a 100644
--- a/arch/x86/kernel/tlb_uv.c
+++ b/arch/x86/kernel/tlb_uv.c
@@ -1012,6 +1012,9 @@ static ssize_t tunables_read(struct file *file, char __user *userbuf,
 		timeoutsb4reset, ipi_reset_limit, complete_threshold,
 		congested_response_us, congested_reps, congested_period);
 
+	if (ret > 300)
+		ret = 300;
+
 	return simple_read_from_buffer(userbuf, count, ppos, buf, ret);
 }
 

Re: [patch] tlb_uv: handle large snprintf() returns

[Andrew Morton]

On Mon, 16 Aug 2010 12:55:02 +0200
Dan Carpenter <error27@gmail.com> wrote:

> snprintf() returns the number of bytes that *would* have been copied if
> the buffer was large enough, so it can be larger than the size of the
> buffer.  In this case it's ok, but let's put a cap on it anyway so it's
> easier to audit.
> 
> Signed-off-by: Dan Carpenter <error27@gmail.com>
> 
> diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
> index 312ef02..5e88b3a 100644
> --- a/arch/x86/kernel/tlb_uv.c
> +++ b/arch/x86/kernel/tlb_uv.c
> @@ -1012,6 +1012,9 @@ static ssize_t tunables_read(struct file *file, char __user *userbuf,
>  		timeoutsb4reset, ipi_reset_limit, complete_threshold,
>  		congested_response_us, congested_reps, congested_period);
>  
> +	if (ret > 300)
> +		ret = 300;
> +
>  	return simple_read_from_buffer(userbuf, count, ppos, buf, ret);
>  }

That 300-byte local array is yuk.

Duplicating the "300" later in the function (instead of using sizeof)
is also yuk.

The code can be deyukked by using sprintf_to_user(), only we don't have
one.  But we do have kasprintf().

[patch v2] tlb_uv: use allocated buffer in tunables_read()

[Dan Carpenter]

The original code didn't check that the value returned from snprintf()
was less than the size of the buffer.  Although it didn't cause a
runtime bug in this case, it makes the static checkers complain.

Andrew Morton suggested a dynamically sized buffer would be cleaner.

Signed-off-by: Dan Carpenter <error27@gmail.com>
---
I don't have an x86_64 system so I haven't been able to compile this
code.  Sorry for that.

V2:  The first version was yuk.

diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
index 312ef02..33e77e4 100644
--- a/arch/x86/kernel/tlb_uv.c
+++ b/arch/x86/kernel/tlb_uv.c
@@ -1001,10 +1001,10 @@ static int uv_ptc_seq_show(struct seq_file *file, void *data)
 static ssize_t tunables_read(struct file *file, char __user *userbuf,
 						size_t count, loff_t *ppos)
 {
-	char buf[300];
+	char *buf;
 	int ret;
 
-	ret = snprintf(buf, 300, "%s %s %s\n%d %d %d %d %d %d %d %d %d\n",
+	buf = kasprintf(GFP_KERNEL, "%s %s %s\n%d %d %d %d %d %d %d %d %d\n",
 		"max_bau_concurrent plugged_delay plugsb4reset",
 		"timeoutsb4reset ipi_reset_limit complete_threshold",
 		"congested_response_us congested_reps congested_period",
@@ -1012,7 +1012,12 @@ static ssize_t tunables_read(struct file *file, char __user *userbuf,
 		timeoutsb4reset, ipi_reset_limit, complete_threshold,
 		congested_response_us, congested_reps, congested_period);
 
-	return simple_read_from_buffer(userbuf, count, ppos, buf, ret);
+	if (!buf)
+		return -ENOMEM;
+
+	ret = simple_read_from_buffer(userbuf, count, ppos, buf, strlen(buf));
+	kfree(buf);
+	return ret;
 }
 
 /*