diff mbox series

[v2,1/3] vsscanf(): Integer overflow is a conversion failure

Message ID 20230610170743.2510-2-demi@invisiblethingslab.com
State New
Headers show
Series Make sscanf() stricter | expand

Commit Message

Demi Marie Obenour June 10, 2023, 5:07 p.m. UTC
sscanf() and friends currently ignore integer overflow, but this is a
bad idea.  It is much better to detect integer overflow errors and
consider this a conversion failure.

This implements Linus's suggestion of using '!' to treat integer
overflow as wrapping.  It does _not_ allow wrapping of unsigned
conversions by default, though, as in at least some cases accepting
negative numbers is _not_ intended.

Suggested-By: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Demi Marie Obenour <demi@invisiblethingslab.com>
---
 .../hive_isp_css_include/platform_support.h   |  1 -
 include/linux/limits.h                        |  1 +
 include/linux/mfd/wl1273-core.h               |  3 -
 include/vdso/limits.h                         |  3 +
 lib/vsprintf.c                                | 82 ++++++++++++++-----
 5 files changed, 67 insertions(+), 23 deletions(-)

Comments

Dan Carpenter June 10, 2023, 7:59 p.m. UTC | #1
On Sat, Jun 10, 2023 at 01:07:41PM -0400, Demi Marie Obenour wrote:
> ---
>  .../hive_isp_css_include/platform_support.h   |  1 -
>  include/linux/limits.h                        |  1 +
>  include/linux/mfd/wl1273-core.h               |  3 -
>  include/vdso/limits.h                         |  3 +
>  lib/vsprintf.c                                | 82 ++++++++++++++-----
>  5 files changed, 67 insertions(+), 23 deletions(-)
> 
> diff --git a/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h b/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
> index 0cdef4a5e8b1bed9884133f1a0b9d853d59d43a4..e29b96d8bebf14839f6dd48fdc6c0f8b029ef31d 100644
> --- a/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
> +++ b/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
> @@ -27,7 +27,6 @@
>  
>  #define UINT16_MAX USHRT_MAX
>  #define UINT32_MAX UINT_MAX
> -#define UCHAR_MAX  (255)
>  
>  #define CSS_ALIGN(d, a) d __attribute__((aligned(a)))
>  
> diff --git a/include/linux/limits.h b/include/linux/limits.h
> index f6bcc936901071f496e3e85bb6e1d93905b12e32..8f7fd85b41fb46e6992d9e5912da00424119227a 100644
> --- a/include/linux/limits.h
> +++ b/include/linux/limits.h
> @@ -8,6 +8,7 @@
>  
>  #define SIZE_MAX	(~(size_t)0)
>  #define SSIZE_MAX	((ssize_t)(SIZE_MAX >> 1))
> +#define SSIZE_MIN	(-SSIZE_MAX - 1)
>  #define PHYS_ADDR_MAX	(~(phys_addr_t)0)
>  
>  #define U8_MAX		((u8)~0U)
> diff --git a/include/linux/mfd/wl1273-core.h b/include/linux/mfd/wl1273-core.h
> index c28cf76d5c31ee1c94a9319a2e2d318bf00283a6..b81a229135ed9f756c749122a8341816031c8311 100644
> --- a/include/linux/mfd/wl1273-core.h
> +++ b/include/linux/mfd/wl1273-core.h
> @@ -204,9 +204,6 @@
>  				 WL1273_IS2_TRI_OPT | \
>  				 WL1273_IS2_RATE_48K)
>  
> -#define SCHAR_MIN (-128)
> -#define SCHAR_MAX 127
> -
>  #define WL1273_FR_EVENT			BIT(0)
>  #define WL1273_BL_EVENT			BIT(1)
>  #define WL1273_RDS_EVENT		BIT(2)
> diff --git a/include/vdso/limits.h b/include/vdso/limits.h
> index 0197888ad0e00b2f853d3f25ffa764f61cca7385..0cad0a2490e5efc194d874025eb3e3b846a5c7b4 100644
> --- a/include/vdso/limits.h
> +++ b/include/vdso/limits.h
> @@ -2,6 +2,9 @@
>  #ifndef __VDSO_LIMITS_H
>  #define __VDSO_LIMITS_H
>  
> +#define UCHAR_MAX	((unsigned char)~0U)
> +#define SCHAR_MAX	((signed char)(UCHAR_MAX >> 1))
> +#define SCHAR_MIN	((signed char)(-SCHAR_MAX - 1))
>  #define USHRT_MAX	((unsigned short)~0U)
>  #define SHRT_MAX	((short)(USHRT_MAX >> 1))
>  #define SHRT_MIN	((short)(-SHRT_MAX - 1))

It looks like you're going to have to redo these patches anyway for
various reasons.  Could you pull the U/SCHAR_MAX changes out into a
separate patch?

> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
> index 40f560959b169b4c4ac6154d658cfe76cfd0c5a6..8caccdcda0a2b470cda70c9b3837de37207eb512 100644
> --- a/lib/vsprintf.c
> +++ b/lib/vsprintf.c
> @@ -59,7 +59,7 @@
>  bool no_hash_pointers __ro_after_init;
>  EXPORT_SYMBOL_GPL(no_hash_pointers);
>  
> -static noinline unsigned long long simple_strntoull(const char *startp, size_t max_chars, char **endp, unsigned int base)
> +static noinline unsigned long long simple_strntoull(const char *startp, size_t max_chars, char **endp, unsigned int base, bool *overflow)
>  {
>  	const char *cp;
>  	unsigned long long result = 0ULL;
> @@ -71,6 +71,8 @@ static noinline unsigned long long simple_strntoull(const char *startp, size_t m
>  	if (prefix_chars < max_chars) {
>  		rv = _parse_integer_limit(cp, base, &result, max_chars - prefix_chars);
>  		/* FIXME */

It's not clear what this FIXME is for, but probably it should go next to
the cp += (rv & ~KSTRTOX_OVERFLOW); line?  Does anyone know what it
means?  Maybe just delete it.

> +		if (overflow)
> +			*overflow = !!(rv & KSTRTOX_OVERFLOW);
>  		cp += (rv & ~KSTRTOX_OVERFLOW);
>  	} else {

*overflow isn't initialized on the else path.

>  		/* Field too short for prefix + digit, skip over without converting */
> @@ -94,7 +96,7 @@ static noinline unsigned long long simple_strntoull(const char *startp, size_t m
>  noinline
>  unsigned long long simple_strtoull(const char *cp, char **endp, unsigned int base)
>  {
> -	return simple_strntoull(cp, INT_MAX, endp, base);
> +	return simple_strntoull(cp, INT_MAX, endp, base, NULL);
>  }
>  EXPORT_SYMBOL(simple_strtoull);
>  
> @@ -130,18 +132,22 @@ long simple_strtol(const char *cp, char **endp, unsigned int base)
>  EXPORT_SYMBOL(simple_strtol);
>  
>  static long long simple_strntoll(const char *cp, size_t max_chars, char **endp,
> -				 unsigned int base)
> +				 unsigned int base, bool *overflow)
>  {
> +	unsigned long long minand;
> +	bool negate;
> +
>  	/*
>  	 * simple_strntoull() safely handles receiving max_chars==0 in the
>  	 * case cp[0] == '-' && max_chars == 1.
>  	 * If max_chars == 0 we can drop through and pass it to simple_strntoull()
>  	 * and the content of *cp is irrelevant.
>  	 */
> -	if (*cp == '-' && max_chars > 0)
> -		return -simple_strntoull(cp + 1, max_chars - 1, endp, base);
> -
> -	return simple_strntoull(cp, max_chars, endp, base);
> +	negate = *cp == '-' && max_chars > 0;
> +	minand = simple_strntoull(cp + negate, max_chars - negate, endp, base, overflow);
> +	if (minand > (unsigned long long)LONG_MAX + negate)
> +		*overflow = true;
> +	return negate ? -minand : minand;
>  }
>  
>  /**
> @@ -3441,7 +3447,7 @@ int vsscanf(const char *buf, const char *fmt, va_list args)
>  		unsigned long long u;
>  	} val;
>  	s16 field_width;
> -	bool is_sign;
> +	bool is_sign, overflow, allow_overflow;
>  
>  	while (*fmt) {
>  		/* skip any white space in format */
> @@ -3464,6 +3470,9 @@ int vsscanf(const char *buf, const char *fmt, va_list args)
>  			break;
>  		++fmt;
>  
> +		allow_overflow = *fmt == '!';
> +		fmt += (int)allow_overflow;
> +
>  		/* skip this conversion.
>  		 * advance both strings to next white space
>  		 */
> @@ -3649,45 +3658,80 @@ int vsscanf(const char *buf, const char *fmt, va_list args)
>  		if (is_sign)
>  			val.s = simple_strntoll(str,
>  						field_width >= 0 ? field_width : INT_MAX,
> -						&next, base);
> +						&next, base, &overflow);
>  		else
>  			val.u = simple_strntoull(str,
>  						 field_width >= 0 ? field_width : INT_MAX,
> -						 &next, base);
> +						 &next, base, &overflow);
> +		if (unlikely(overflow && !allow_overflow))

So that means that *overflow can be uninitialized here.

> +			break;
>  
>  		switch (qualifier) {

regards,
dan carpenter
diff mbox series

Patch

diff --git a/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h b/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
index 0cdef4a5e8b1bed9884133f1a0b9d853d59d43a4..e29b96d8bebf14839f6dd48fdc6c0f8b029ef31d 100644
--- a/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
+++ b/drivers/staging/media/atomisp/pci/hive_isp_css_include/platform_support.h
@@ -27,7 +27,6 @@ 
 
 #define UINT16_MAX USHRT_MAX
 #define UINT32_MAX UINT_MAX
-#define UCHAR_MAX  (255)
 
 #define CSS_ALIGN(d, a) d __attribute__((aligned(a)))
 
diff --git a/include/linux/limits.h b/include/linux/limits.h
index f6bcc936901071f496e3e85bb6e1d93905b12e32..8f7fd85b41fb46e6992d9e5912da00424119227a 100644
--- a/include/linux/limits.h
+++ b/include/linux/limits.h
@@ -8,6 +8,7 @@ 
 
 #define SIZE_MAX	(~(size_t)0)
 #define SSIZE_MAX	((ssize_t)(SIZE_MAX >> 1))
+#define SSIZE_MIN	(-SSIZE_MAX - 1)
 #define PHYS_ADDR_MAX	(~(phys_addr_t)0)
 
 #define U8_MAX		((u8)~0U)
diff --git a/include/linux/mfd/wl1273-core.h b/include/linux/mfd/wl1273-core.h
index c28cf76d5c31ee1c94a9319a2e2d318bf00283a6..b81a229135ed9f756c749122a8341816031c8311 100644
--- a/include/linux/mfd/wl1273-core.h
+++ b/include/linux/mfd/wl1273-core.h
@@ -204,9 +204,6 @@ 
 				 WL1273_IS2_TRI_OPT | \
 				 WL1273_IS2_RATE_48K)
 
-#define SCHAR_MIN (-128)
-#define SCHAR_MAX 127
-
 #define WL1273_FR_EVENT			BIT(0)
 #define WL1273_BL_EVENT			BIT(1)
 #define WL1273_RDS_EVENT		BIT(2)
diff --git a/include/vdso/limits.h b/include/vdso/limits.h
index 0197888ad0e00b2f853d3f25ffa764f61cca7385..0cad0a2490e5efc194d874025eb3e3b846a5c7b4 100644
--- a/include/vdso/limits.h
+++ b/include/vdso/limits.h
@@ -2,6 +2,9 @@ 
 #ifndef __VDSO_LIMITS_H
 #define __VDSO_LIMITS_H
 
+#define UCHAR_MAX	((unsigned char)~0U)
+#define SCHAR_MAX	((signed char)(UCHAR_MAX >> 1))
+#define SCHAR_MIN	((signed char)(-SCHAR_MAX - 1))
 #define USHRT_MAX	((unsigned short)~0U)
 #define SHRT_MAX	((short)(USHRT_MAX >> 1))
 #define SHRT_MIN	((short)(-SHRT_MAX - 1))
diff --git a/lib/vsprintf.c b/lib/vsprintf.c
index 40f560959b169b4c4ac6154d658cfe76cfd0c5a6..8caccdcda0a2b470cda70c9b3837de37207eb512 100644
--- a/lib/vsprintf.c
+++ b/lib/vsprintf.c
@@ -59,7 +59,7 @@ 
 bool no_hash_pointers __ro_after_init;
 EXPORT_SYMBOL_GPL(no_hash_pointers);
 
-static noinline unsigned long long simple_strntoull(const char *startp, size_t max_chars, char **endp, unsigned int base)
+static noinline unsigned long long simple_strntoull(const char *startp, size_t max_chars, char **endp, unsigned int base, bool *overflow)
 {
 	const char *cp;
 	unsigned long long result = 0ULL;
@@ -71,6 +71,8 @@  static noinline unsigned long long simple_strntoull(const char *startp, size_t m
 	if (prefix_chars < max_chars) {
 		rv = _parse_integer_limit(cp, base, &result, max_chars - prefix_chars);
 		/* FIXME */
+		if (overflow)
+			*overflow = !!(rv & KSTRTOX_OVERFLOW);
 		cp += (rv & ~KSTRTOX_OVERFLOW);
 	} else {
 		/* Field too short for prefix + digit, skip over without converting */
@@ -94,7 +96,7 @@  static noinline unsigned long long simple_strntoull(const char *startp, size_t m
 noinline
 unsigned long long simple_strtoull(const char *cp, char **endp, unsigned int base)
 {
-	return simple_strntoull(cp, INT_MAX, endp, base);
+	return simple_strntoull(cp, INT_MAX, endp, base, NULL);
 }
 EXPORT_SYMBOL(simple_strtoull);
 
@@ -130,18 +132,22 @@  long simple_strtol(const char *cp, char **endp, unsigned int base)
 EXPORT_SYMBOL(simple_strtol);
 
 static long long simple_strntoll(const char *cp, size_t max_chars, char **endp,
-				 unsigned int base)
+				 unsigned int base, bool *overflow)
 {
+	unsigned long long minand;
+	bool negate;
+
 	/*
 	 * simple_strntoull() safely handles receiving max_chars==0 in the
 	 * case cp[0] == '-' && max_chars == 1.
 	 * If max_chars == 0 we can drop through and pass it to simple_strntoull()
 	 * and the content of *cp is irrelevant.
 	 */
-	if (*cp == '-' && max_chars > 0)
-		return -simple_strntoull(cp + 1, max_chars - 1, endp, base);
-
-	return simple_strntoull(cp, max_chars, endp, base);
+	negate = *cp == '-' && max_chars > 0;
+	minand = simple_strntoull(cp + negate, max_chars - negate, endp, base, overflow);
+	if (minand > (unsigned long long)LONG_MAX + negate)
+		*overflow = true;
+	return negate ? -minand : minand;
 }
 
 /**
@@ -3441,7 +3447,7 @@  int vsscanf(const char *buf, const char *fmt, va_list args)
 		unsigned long long u;
 	} val;
 	s16 field_width;
-	bool is_sign;
+	bool is_sign, overflow, allow_overflow;
 
 	while (*fmt) {
 		/* skip any white space in format */
@@ -3464,6 +3470,9 @@  int vsscanf(const char *buf, const char *fmt, va_list args)
 			break;
 		++fmt;
 
+		allow_overflow = *fmt == '!';
+		fmt += (int)allow_overflow;
+
 		/* skip this conversion.
 		 * advance both strings to next white space
 		 */
@@ -3649,45 +3658,80 @@  int vsscanf(const char *buf, const char *fmt, va_list args)
 		if (is_sign)
 			val.s = simple_strntoll(str,
 						field_width >= 0 ? field_width : INT_MAX,
-						&next, base);
+						&next, base, &overflow);
 		else
 			val.u = simple_strntoull(str,
 						 field_width >= 0 ? field_width : INT_MAX,
-						 &next, base);
+						 &next, base, &overflow);
+		if (unlikely(overflow && !allow_overflow))
+			break;
 
 		switch (qualifier) {
 		case 'H':	/* that's 'hh' in format */
-			if (is_sign)
+			if (is_sign) {
+				if (unlikely(val.s < SCHAR_MIN || val.s > SCHAR_MAX) &&
+				    !allow_overflow)
+					return num;
 				*va_arg(args, signed char *) = val.s;
-			else
+			} else {
+				if (unlikely(val.u > UCHAR_MAX) && !allow_overflow)
+					return num;
 				*va_arg(args, unsigned char *) = val.u;
+			}
 			break;
 		case 'h':
-			if (is_sign)
+			if (is_sign) {
+				if (unlikely(val.s < SHRT_MIN || val.s > SHRT_MAX) &&
+				    !allow_overflow)
+					return num;
 				*va_arg(args, short *) = val.s;
-			else
+			} else {
+				if (unlikely(val.u > USHRT_MAX) && !allow_overflow)
+					return num;
 				*va_arg(args, unsigned short *) = val.u;
+			}
 			break;
 		case 'l':
-			if (is_sign)
+			if (is_sign) {
+				if (unlikely(val.s < LONG_MIN || val.s > LONG_MAX) &&
+				    !allow_overflow)
+					return num;
 				*va_arg(args, long *) = val.s;
-			else
+			} else {
+				if (unlikely(val.u > ULONG_MAX) && !allow_overflow)
+					return num;
 				*va_arg(args, unsigned long *) = val.u;
+			}
 			break;
 		case 'L':
+			/* No overflow check needed */
 			if (is_sign)
 				*va_arg(args, long long *) = val.s;
 			else
 				*va_arg(args, unsigned long long *) = val.u;
 			break;
 		case 'z':
-			*va_arg(args, size_t *) = val.u;
+			if (is_sign) {
+				if (unlikely(val.s < SSIZE_MIN || val.s > SSIZE_MAX))
+					return num;
+				*va_arg(args, ssize_t *) = val.s;
+			} else {
+				if (unlikely(val.u > SIZE_MAX) && !allow_overflow)
+					return num;
+				*va_arg(args, size_t *) = val.u;
+			}
 			break;
 		default:
-			if (is_sign)
+			if (is_sign) {
+				if (unlikely(val.s < INT_MIN || val.s > INT_MAX) &&
+				    !allow_overflow)
+					return num;
 				*va_arg(args, int *) = val.s;
-			else
+			} else {
+				if (unlikely(val.u > UINT_MAX) && !allow_overflow)
+					return num;
 				*va_arg(args, unsigned int *) = val.u;
+			}
 			break;
 		}
 		num++;