Skip to contentSkip to conversation

ALSA: usb-audio: fix possible hang and overflow in parse_uac2_sample_rate_range()

PATCH Superseded Xi Wang 2 participants 7 messages Jan 4, 2012 – Jan 8, 2012 Sound

lore
Share & tools
lore

Xi Wang

author
#0 lore raw

PATCH 1/1 ALSA: usb-audio: fix possible hang and overflow in parse_uac2_sample_rate_range()

A malicious USB device may feed in carefully crafted min/max/res values, so that the inner loop in parse_uac2_sample_rate_range() could run for a long time or even never terminate, e.g., given max = INT_MAX.

Also nr_rates could be a large integer, which causes an integer overflow in the subsequent call to kmalloc() in parse_audio_format_rates_v2(). Thus, kmalloc() would allocate a smaller buffer than expected, leading to a memory corruption.

To exploit the two vulnerabilities, an attacker needs physical access to the machine to plug in a malicious USB device.

This patch makes two changes.

1) The type of "rate" is changed to unsigned int, so that the loop could stop once "rate" is larger than INT_MAX.

2) Limit nr_rates to avoid overflow.

---
 sound/usb/format.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)
sound/usb/format.c +4 -1
index 89421d1..a99de67 100644
@@ -226,7 +226,7 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 		int min = combine_quad(&data[2 + 12 * i]);
 		int max = combine_quad(&data[6 + 12 * i]);
 		int res = combine_quad(&data[10 + 12 * i]);
-		int rate;
+		unsigned int rate;
 
 		if ((max < 0) || (min < 0) || (res < 0) || (max < min))
 			continue;
@@ -253,6 +253,9 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 			fp->rates |= snd_pcm_rate_to_rate_bit(rate);
 
 			nr_rates++;
+			/* avoid overflow */
+			if (nr_rates == KMALLOC_MAX_SIZE / sizeof(int))
+				break;
 
 			/* avoid endless loop */
 			if (res == 0)
Signature · 1 line
1.7.5.4
#1 lore raw

At Wed, 4 Jan 2012 12:39:09 -0500,

Quoted 13 lines · Xi Wang

A malicious USB device may feed in carefully crafted min/max/res values, so that the inner loop in parse_uac2_sample_rate_range() could run for a long time or even never terminate, e.g., given max = INT_MAX.

Also nr_rates could be a large integer, which causes an integer overflow in the subsequent call to kmalloc() in parse_audio_format_rates_v2(). Thus, kmalloc() would allocate a smaller buffer than expected, leading to a memory corruption.

To exploit the two vulnerabilities, an attacker needs physical access to the machine to plug in a malicious USB device.

Quoted 8 lines

This patch makes two changes.

1) The type of "rate" is changed to unsigned int, so that the loop could stop once "rate" is larger than INT_MAX.

2) Limit nr_rates to avoid overflow.

Thanks for the patch.
As of now, I have little time to evaluate, so I might have missed something, but I wonder whether

		/* avoid overflow */
		if (nr_rates == KMALLOC_MAX_SIZE / sizeof(int))
			break;

is the best way to check. This looks ugly to me.
If we need to limit the number of rates, better to define some proper numbers as the upper limit. And then, it should warn, not only breaking loop.

Anyway I'll check this tomorrow in more details.

thanks,

Takashi

Quoted 30 lines
---
 sound/usb/format.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)
sound/usb/format.c +4 -1
index 89421d1..a99de67 100644
@@ -226,7 +226,7 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 		int min = combine_quad(&data[2 + 12 * i]);
 		int max = combine_quad(&data[6 + 12 * i]);
 		int res = combine_quad(&data[10 + 12 * i]);
-		int rate;
+		unsigned int rate;
 
 		if ((max < 0) || (min < 0) || (res < 0) || (max < min))
 			continue;
@@ -253,6 +253,9 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 			fp->rates |= snd_pcm_rate_to_rate_bit(rate);
 
 			nr_rates++;
+			/* avoid overflow */
+			if (nr_rates == KMALLOC_MAX_SIZE / sizeof(int))
+				break;
 
 			/* avoid endless loop */
 			if (res == 0)
Signature · 1 line
1.7.5.4
author
#2 lore raw
Quoted 11 lines · Takashi Iwai

As of now, I have little time to evaluate, so I might have missed something, but I wonder whether

		/* avoid overflow */
		if (nr_rates == KMALLOC_MAX_SIZE / sizeof(int))
			break;

is the best way to check. This looks ugly to me.
If we need to limit the number of rates, better to define some proper numbers as the upper limit. And then, it should warn, not only breaking loop.

Thanks for looking into this. Yeah, I agree using something like MAX_NR_RATES is better. Is 65535 okay or do we need a larger limit?

- xi

#3 lore raw

At Sun, 8 Jan 2012 07:45:09 -0500,

Quoted 9 lines · Xi Wang
Quoted 7 lines · Takashi Iwai

As of now, I have little time to evaluate, so I might have missed something, but I wonder whether

		/* avoid overflow */
		if (nr_rates == KMALLOC_MAX_SIZE / sizeof(int))
			break;
Quoted 7 lines
Quoted 4 lines

is the best way to check. This looks ugly to me.
If we need to limit the number of rates, better to define some proper numbers as the upper limit. And then, it should warn, not only breaking loop.

Thanks for looking into this. Yeah, I agree using something like MAX_NR_RATES is better. Is 65535 okay or do we need a larger limit?

It's way too higher than any realistic situation ;) I guess 1000 or such should suffice.

thanks,

Takashi

author
#4 lore raw
Quoted 2 lines · Takashi Iwai

It's way too higher than any realistic situation ;) I guess 1000 or such should suffice.

Let's limit it to 1024 then. ;-)

I will do a v2. Thanks.

- xi

author
#5 lore raw

PATCH v2 1/1 ALSA: usb-audio: fix possible hang and overflow in parse_uac2_sample_rate_range()

In reply to Xi Wang

A malicious USB device may feed in carefully crafted min/max/res values, so that the inner loop in parse_uac2_sample_rate_range() could run for a long time or even never terminate, e.g., given max = INT_MAX.

Also nr_rates could be a large integer, which causes an integer overflow in the subsequent call to kmalloc() in parse_audio_format_rates_v2(). Thus, kmalloc() would allocate a smaller buffer than expected, leading to a memory corruption.

To exploit the two vulnerabilities, an attacker needs physical access to the machine to plug in a malicious USB device.

This patch makes two changes.

1) The type of "rate" is changed to unsigned int, so that the loop could stop once "rate" is larger than INT_MAX.

2) Limit nr_rates to 1024.

---
 sound/usb/format.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
sound/usb/format.c +7 -1
index 89421d1..e09aba1 100644
@@ -209,6 +209,8 @@ static int parse_audio_format_rates_v1(struct snd_usb_audio *chip, struct audiof
 	return 0;
 }
 
+#define MAX_UAC2_NR_RATES 1024
+
 /*
  * Helper function to walk the array of sample rate triplets reported by
  * the device. The problem is that we need to parse whole array first to
@@ -226,7 +228,7 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 		int min = combine_quad(&data[2 + 12 * i]);
 		int max = combine_quad(&data[6 + 12 * i]);
 		int res = combine_quad(&data[10 + 12 * i]);
-		int rate;
+		unsigned int rate;
 
 		if ((max < 0) || (min < 0) || (res < 0) || (max < min))
 			continue;
@@ -253,6 +255,10 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 			fp->rates |= snd_pcm_rate_to_rate_bit(rate);
 
 			nr_rates++;
+			if (nr_rates >= MAX_UAC2_NR_RATES) {
+				snd_printk(KERN_ERR "invalid uac2 rates\n");
+				break;
+			}
 
 			/* avoid endless loop */
 			if (res == 0)
Signature · 1 line
1.7.5.4
Applied
#6 lore raw

At Sun, 08 Jan 2012 09:02:52 -0500,

Quoted 15 lines · Xi Wang

A malicious USB device may feed in carefully crafted min/max/res values, so that the inner loop in parse_uac2_sample_rate_range() could run for a long time or even never terminate, e.g., given max = INT_MAX.

Also nr_rates could be a large integer, which causes an integer overflow in the subsequent call to kmalloc() in parse_audio_format_rates_v2(). Thus, kmalloc() would allocate a smaller buffer than expected, leading to a memory corruption.

To exploit the two vulnerabilities, an attacker needs physical access to the machine to plug in a malicious USB device.

This patch makes two changes.

Quoted 7 lines

1) The type of "rate" is changed to unsigned int, so that the loop could stop once "rate" is larger than INT_MAX.

2) Limit nr_rates to 1024.

Thanks, applied now.

Takashi

Quoted 41 lines
---
 sound/usb/format.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
sound/usb/format.c +7 -1
index 89421d1..e09aba1 100644
@@ -209,6 +209,8 @@ static int parse_audio_format_rates_v1(struct snd_usb_audio *chip, struct audiof
 	return 0;
 }
 
+#define MAX_UAC2_NR_RATES 1024
+
 /*
  * Helper function to walk the array of sample rate triplets reported by
  * the device. The problem is that we need to parse whole array first to
@@ -226,7 +228,7 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 		int min = combine_quad(&data[2 + 12 * i]);
 		int max = combine_quad(&data[6 + 12 * i]);
 		int res = combine_quad(&data[10 + 12 * i]);
-		int rate;
+		unsigned int rate;
 
 		if ((max < 0) || (min < 0) || (res < 0) || (max < min))
 			continue;
@@ -253,6 +255,10 @@ static int parse_uac2_sample_rate_range(struct audioformat *fp, int nr_triplets,
 			fp->rates |= snd_pcm_rate_to_rate_bit(rate);
 
 			nr_rates++;
+			if (nr_rates >= MAX_UAC2_NR_RATES) {
+				snd_printk(KERN_ERR "invalid uac2 rates\n");
+				break;
+			}
 
 			/* avoid endless loop */
 			if (res == 0)
Signature · 1 line
1.7.5.4