Skip to contentSkip to conversation

platform/chrome: cros_ec: fix read overflow in cros_ec_lpc_readmem()

PATCH In review Dan Carpenter 2 participants 4 messages Dec 9, 2021 – Dec 15, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 platform/chrome: cros_ec: fix read overflow in cros_ec_lpc_readmem()

If bytes is larger than EC_MEMMAP_SIZE (255) then "EC_MEMMAP_SIZE - bytes" is a very high unsigned value and basically offset is accepted. The second problem is that it uses >= instead of > so this means that we are not able to read the very last byte.

Fixes ec2f33ab582b ("platform/chrome: Add cros_ec_lpc driver for x86 devices")
---
 drivers/platform/chrome/cros_ec_lpc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/platform/chrome/cros_ec_lpc.c +2 -1
index d6306d2a096f..7e1d175def9f 100644
@@ -290,7 +290,8 @@ static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset,
 	char *s = dest;
 	int cnt = 0;
 
-	if (offset >= EC_MEMMAP_SIZE - bytes)
+	if (offset > EC_MEMMAP_SIZE ||
+	    bytes > EC_MEMMAP_SIZE - offset)
 		return -EINVAL;
 
 	/* fixed length */
Signature · 1 line
2.20.1
#1 lore raw
Quoted 17 lines · Dan Carpenter

If bytes is larger than EC_MEMMAP_SIZE (255) then "EC_MEMMAP_SIZE - bytes" is a very high unsigned value and basically offset is accepted. The second problem is that it uses >= instead of > so this means that we are not able to read the very last byte.

Fixes ec2f33ab582b ("platform/chrome: Add cros_ec_lpc driver for x86 devices")
---
 drivers/platform/chrome/cros_ec_lpc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/platform/chrome/cros_ec_lpc.c +0 -0
index d6306d2a096f..7e1d175def9f 100644

drivers/platform/chrome/cros_ec_lpc.c static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset, · line 294

Quoted 6 lines
drivers/platform/chrome/cros_ec_lpc.c +2 -1
@@ -290,7 +290,8 @@ static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset,
        char *s = dest;
        int cnt = 0;
 
-       if (offset >= EC_MEMMAP_SIZE - bytes)
+       if (offset > EC_MEMMAP_SIZE ||
+           bytes > EC_MEMMAP_SIZE - offset)

I think that means we have the same problem if offset > EC_MEMMAP_SIZE, only now that condition isn't detected anymore because EC_MEMMAP_SIZE - offset is a very large number.
I think what we really want is
if (offset + bytes > EC_MEMMAP_SIZE)
only without the overflow. Not sure how we can get there without checking each part.

        if (offset > EC_MEMMAP_SIZE || bytes > EC_MEMMAP_SIZE || bytes
+ offset > EC_MEMMAP_SIZE)
                return -EINVAL;

Maybe that ?
if ((u64) offset + bytes > EC_MEMMAP_SIZE)
return -EINVAL;

Thanks,
Guenter

Quoted 6 lines
drivers/platform/chrome/cros_ec_lpc.c +0 -1
                return -EINVAL;
 
        /* fixed length */
--

2.20.1

#2 lore raw
Quoted 21 lines · Guenter Roeck
Quoted 20 lines · Dan Carpenter

If bytes is larger than EC_MEMMAP_SIZE (255) then "EC_MEMMAP_SIZE - bytes" is a very high unsigned value and basically offset is accepted. The second problem is that it uses >= instead of > so this means that we are not able to read the very last byte.

Fixes ec2f33ab582b ("platform/chrome: Add cros_ec_lpc driver for x86 devices")
---
 drivers/platform/chrome/cros_ec_lpc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/platform/chrome/cros_ec_lpc.c +0 -0
index d6306d2a096f..7e1d175def9f 100644
@@ -290,7 +290,8 @@ static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset,
        char *s = dest;
        int cnt = 0;
 
Quoted 7 lines

drivers/platform/chrome/cros_ec_lpc.c static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset, · line 294

Quoted 3 lines
drivers/platform/chrome/cros_ec_lpc.c +2 -1
-       if (offset >= EC_MEMMAP_SIZE - bytes)
+       if (offset > EC_MEMMAP_SIZE ||
+           bytes > EC_MEMMAP_SIZE - offset)

I think that means we have the same problem if offset > EC_MEMMAP_SIZE, only now that condition isn't detected anymore because EC_MEMMAP_SIZE - offset is a very large number.

That's the bug which my patch addresses. (My patch is option 1).

Quoted 6 lines

I think what we really want is
if (offset + bytes > EC_MEMMAP_SIZE)
only without the overflow. Not sure how we can get there without checking each part.

        if (offset > EC_MEMMAP_SIZE || bytes > EC_MEMMAP_SIZE || bytes

+ offset > EC_MEMMAP_SIZE)

That is another solution which works.

Quoted 4 lines
                return -EINVAL;

Maybe that ?
if ((u64) offset + bytes > EC_MEMMAP_SIZE)
return -EINVAL;

A third viable solution.

I generally prefer option 2 to option 3. I generally use that in code that I'm writing. There was one time Linus said he liked option 1 which I used here because it works regardless of the types or the valu of EC_MEMMAP_SIZE. This code already used the bytes > size - offset idiom so I kept it as similar as possible.

regards,
dan carpenter

Reviewed-by
#3 lore raw
Quoted 31 lines · Dan Carpenter
Quoted 21 lines · Guenter Roeck
Quoted 20 lines · Dan Carpenter

If bytes is larger than EC_MEMMAP_SIZE (255) then "EC_MEMMAP_SIZE - bytes" is a very high unsigned value and basically offset is accepted. The second problem is that it uses >= instead of > so this means that we are not able to read the very last byte.

Fixes ec2f33ab582b ("platform/chrome: Add cros_ec_lpc driver for x86 devices")
---
 drivers/platform/chrome/cros_ec_lpc.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/platform/chrome/cros_ec_lpc.c +0 -0
index d6306d2a096f..7e1d175def9f 100644
@@ -290,7 +290,8 @@ static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset,
        char *s = dest;
        int cnt = 0;
 
Quoted 7 lines

drivers/platform/chrome/cros_ec_lpc.c static int cros_ec_lpc_readmem(struct cros_ec_device *ec, unsigned int offset, · line 294

Quoted 3 lines
drivers/platform/chrome/cros_ec_lpc.c +2 -1
-       if (offset >= EC_MEMMAP_SIZE - bytes)
+       if (offset > EC_MEMMAP_SIZE ||
+           bytes > EC_MEMMAP_SIZE - offset)

I think that means we have the same problem if offset > EC_MEMMAP_SIZE, only now that condition isn't detected anymore because EC_MEMMAP_SIZE - offset is a very large number.

Quoted 2 lines

That's the bug which my patch addresses. (My patch is option 1).

Ah yes, for some reason I overlooked the "offset > EC_MEMMAP_SIZE" part in your patch. Sorry, I must have been blind.

Guenter

Quoted 25 lines
Quoted 6 lines

I think what we really want is
if (offset + bytes > EC_MEMMAP_SIZE)
only without the overflow. Not sure how we can get there without checking each part.

        if (offset > EC_MEMMAP_SIZE || bytes > EC_MEMMAP_SIZE || bytes

+ offset > EC_MEMMAP_SIZE)

That is another solution which works.

Quoted 4 lines
                return -EINVAL;

Maybe that ?
if ((u64) offset + bytes > EC_MEMMAP_SIZE)
return -EINVAL;

A third viable solution.

I generally prefer option 2 to option 3. I generally use that in code that I'm writing. There was one time Linus said he liked option 1 which I used here because it works regardless of the types or the valu of EC_MEMMAP_SIZE. This code already used the bytes > size - offset idiom so I kept it as similar as possible.

regards,
dan carpenter