Skip to contentSkip to conversation

CAN: Use inode instead of kernel address for /proc file

PATCH Superseded Dan Rosenberg 2 participants 6 messages Dec 25, 2010 – Dec 26, 2010

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 CAN: Use inode instead of kernel address for /proc file

Since the socket address is just being used as a unique identifier, its inode number is an alternative that does not leak potentially sensitive information.

CC-ing stable because MITRE has assigned CVE-2010-4565 to the issue.

---
 net/can/bcm.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
net/can/bcm.c +2 -2
index 6faa825..5748901 100644
@@ -1520,8 +1520,8 @@ static int bcm_connect(struct socket *sock, struct sockaddr *uaddr, int len,
 	bo->bound = 1;
 
 	if (proc_dir) {
-		/* unique socket address as filename */
-		sprintf(bo->procname, "%p", sock);
+		/* socket inode as filename */
+		sprintf(bo->procname, "%lx", sock_i_ino(sk));
 		bo->bcm_proc_read = proc_create_data(bo->procname, 0644,
 						     proc_dir,
 						     &bcm_proc_fops, sk);
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 8 lines · Dan Rosenberg

Since the socket address is just being used as a unique identifier, its inode number is an alternative that does not leak potentially sensitive information.

CC-ing stable because MITRE has assigned CVE-2010-4565 to the issue.

Thanks Dan.

Quoted 20 lines
---
 net/can/bcm.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
net/can/bcm.c +2 -2
index 6faa825..5748901 100644
@@ -1520,8 +1520,8 @@ static int bcm_connect(struct socket *sock, struct sockaddr *uaddr, int len,
 	bo->bound = 1;
 
 	if (proc_dir) {
-		/* unique socket address as filename */
-		sprintf(bo->procname, "%p", sock);
+		/* socket inode as filename */
+		sprintf(bo->procname, "%lx", sock_i_ino(sk));
 		bo->bcm_proc_read = proc_create_data(bo->procname, 0644,
 						     proc_dir,
 						     &bcm_proc_fops, sk);
maintainer
#2 lore raw
Quoted 25 lines · Oliver Hartkopp
Quoted 8 lines · Dan Rosenberg

Since the socket address is just being used as a unique identifier, its inode number is an alternative that does not leak potentially sensitive information.

CC-ing stable because MITRE has assigned CVE-2010-4565 to the issue.

Thanks Dan.

Quoted 11 lines
---
 net/can/bcm.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
net/can/bcm.c +0 -0
index 6faa825..5748901 100644
@@ -1520,8 +1520,8 @@ static int bcm_connect(struct socket *sock, struct sockaddr *uaddr, int len,
 	bo->bound = 1;
 
Quoted 5 lines

net/can/bcm.c static int bcm_connect(struct socket *sock, struct sockaddr *uaddr, int len, · line 1524

Quoted 5 lines
net/can/bcm.c +2 -2
 	if (proc_dir) {
-		/* unique socket address as filename */
-		sprintf(bo->procname, "%p", sock);
+		/* socket inode as filename */
+		sprintf(bo->procname, "%lx", sock_i_ino(sk));

One minor question:

AFAIK the inode numbers that can be found in /proc/<pid>/fd/* are in decimal and not in hex, right?

If so, you should use '%lu' instead of '%lx' in the patch.

Regards,
Oliver

#3 lore raw
Quoted 8 lines

One minor question:

AFAIK the inode numbers that can be found in /proc/<pid>/fd/* are in decimal and not in hex, right?

If so, you should use '%lu' instead of '%lx' in the patch.

Yes, that's usually how they're expressed, but I did it this way for two reasons. Firstly, %lu would require another change to the buffer size, since the output could be up to 20 bytes long (plus another for the NULL terminator). Secondly, by expressing it as hex it avoids breaking any userland utilities that are expecting addresses, even if no such utilities exist.

-Dan

maintainer
#4 lore raw
Quoted 7 lines · Dan Rosenberg
Quoted 7 lines

One minor question:

AFAIK the inode numbers that can be found in /proc/<pid>/fd/* are in decimal and not in hex, right?

Quoted 6 lines
Quoted 1 line

If so, you should use '%lu' instead of '%lx' in the patch.

Yes, that's usually how they're expressed, but I did it this way for two reasons. Firstly, %lu would require another change to the buffer size, since the output could be up to 20 bytes long (plus another for the NULL terminator).

This is not *really* a good reason as the buffer size can be changed easily within the same patch, right? ;-)

Quoted 3 lines

Secondly, by expressing it as hex it avoids breaking any userland utilities that are expecting addresses, even if no such utilities exist.

I'm very sure there are no userland utilities. And if there were any, they only would use the filename as unique identifiers and can't do anything with the (kernel)addresses - as you already stated in the patch description. E.g. thinking of some shell script the content of the filename is pretty useless unless it is unique.

The recent patch for the 64 bit compatibility broke the length of the filename anyway, so IMO there's no real reason left to write the inode number as a hex value for the unique filename.

Indeed having it a decimal number would bring some added value as you can reference the filename to the numbers in /proc/<pid>/fd/* easily.

Convinced? 8-)

Regards,
Oliver

#5 lore raw
Quoted 2 lines · Oliver Hartkopp

Convinced? 8-)

Absolutely. I didn't mean to come across as argumentative, I was just trying to anticipate potential objections. I'll resend shortly.

Thanks,
Dan