Skip to contentSkip to conversation

virtiofsd: Seccomp: Add 'send' for syslog

PATCH In review Dr. David Alan Gilbert (git) 4 participants 7 messages Nov 2, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 virtiofsd: Seccomp: Add 'send' for syslog

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +1 -0
index eb9af8265f..672fb72a31 100644
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {
+    SCMP_SYS(send),
     SCMP_SYS(sendto),
 };
 
Signature · 1 line
2.28.0
#1 lore raw
Quoted 17 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 120

Quoted 3 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {

Is it worth restricting the syscall to POWER?

   #if defined(__powerpc64__)

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 121

Quoted 1 line
tools/virtiofsd/passthrough_seccomp.c +1 -0
+    SCMP_SYS(send),
   #endif
Quoted 4 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
     SCMP_SYS(sendto),
 };
 
#2 lore raw
Quoted 19 lines · * Philippe Mathieu-Daudé ([email protected]) wrote:
Quoted 18 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
Quoted 6 lines

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 120

Quoted 2 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {

Is it worth restricting the syscall to POWER?

   #if defined(__powerpc64__)

I don't think so, since it's legal for a libc to use either; so any other libc or architecture could choose either to use.

Dave

Quoted 9 lines

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 121

Quoted 1 line
tools/virtiofsd/passthrough_seccomp.c +1 -0
+    SCMP_SYS(send),
   #endif
Quoted 4 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
     SCMP_SYS(sendto),
 };
 
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
#3 lore raw
In reply to Philippe Mathieu-Daudé
Quoted 19 lines · Philippe Mathieu-Daudé
Quoted 18 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
Quoted 4 lines

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 120

Quoted 2 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {

Is it worth restricting the syscall to POWER?

That would be wrong, as this is needed on multiple arches.

There is no real security benefit to restricting it either, as both syscalls give you broadly equivalent abilities.

Quoted 9 lines
   #if defined(__powerpc64__)

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 121

Quoted 1 line
tools/virtiofsd/passthrough_seccomp.c +1 -0
+    SCMP_SYS(send),
   #endif
Quoted 2 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
     SCMP_SYS(sendto),
 };

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#4 lore raw
Quoted 25 lines · Daniel P. Berrangé
Quoted 19 lines · Philippe Mathieu-Daudé
Quoted 18 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
Quoted 4 lines

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 120

Quoted 2 lines
tools/virtiofsd/passthrough_seccomp.c +0 -0
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {

Is it worth restricting the syscall to POWER?

Quoted 4 lines

That would be wrong, as this is needed on multiple arches.

There is no real security benefit to restricting it either, as both syscalls give you broadly equivalent abilities.

Restricting send to architectures where glibc slightly decreases the host kernel attack surface. I think it makes sense from a security perspective. There could be a send(2)-only bug in Linux isn't present in sendto(2).

But there are other issues with seccomp - are we really sure send(2) is never called? Since we don't control the entire binary (share libraries are used, virtiofsd could be linked against another libc, etc) we may not have enough information to conclude that can be eliminated send(2).

Stefan

Reviewed-by on v1 1/1
#5 lore raw
In reply to Dr. David Alan Gilbert (git)
Quoted 17 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 123

Quoted 6 lines
tools/virtiofsd/passthrough_seccomp.c +1 -0
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {
+    SCMP_SYS(send),
     SCMP_SYS(sendto),
 };

With glibc, syslog() calls __send() which for Linux target is implemented as:

ssize_t
__libc_send (int fd, const void *buf, size_t len, int flags) {
#ifdef __ASSUME_SEND_SYSCALL
return SYSCALL_CANCEL (send, fd, buf, len, flags);
#elif defined __ASSUME_SENDTO_SYSCALL
return SYSCALL_CANCEL (sendto, fd, buf, len, flags, NULL, 0);
#else
return SOCKETCALL_CANCEL (send, fd, buf, len, flags);
#endif
}

We can see those defines being referenced vary per architecture:

$ git grep -E '__ASSUME_SEND(TO)?_SYSCALL' sysdeps/
sysdeps/unix/sysv/linux/alpha/kernel-features.h:#define __ASSUME_SEND_SYSCALL   1
sysdeps/unix/sysv/linux/arm/kernel-features.h:#define __ASSUME_SEND_SYSCALL     1
sysdeps/unix/sysv/linux/hppa/kernel-features.h:#define __ASSUME_SEND_SYSCALL    1
sysdeps/unix/sysv/linux/i386/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/ia64/kernel-features.h:#define __ASSUME_SEND_SYSCALL            1
sysdeps/unix/sysv/linux/kernel-features.h:#define __ASSUME_SENDTO_SYSCALL               1
sysdeps/unix/sysv/linux/m68k/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/microblaze/kernel-features.h:#define __ASSUME_SEND_SYSCALL              1
sysdeps/unix/sysv/linux/mips/kernel-features.h:# define __ASSUME_SEND_SYSCALL           1
sysdeps/unix/sysv/linux/powerpc/kernel-features.h:#define __ASSUME_SEND_SYSCALL         1
sysdeps/unix/sysv/linux/s390/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/send.c:#ifdef __ASSUME_SEND_SYSCALL
sysdeps/unix/sysv/linux/send.c:#elif defined __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/sendto.c:#ifdef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/sh/kernel-features.h:#define __ASSUME_SEND_SYSCALL              1
sysdeps/unix/sysv/linux/sparc/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL

So the patch is correct, but the commit message could be updated becase this isn't specific to PPC. Any platform except x86, s490, m68k will use send() rather than sendto() based on what I see here.

With any commit message update, you can add

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#6 lore raw
Quoted 62 lines · * Daniel P. Berrangé ([email protected]) wrote:
Quoted 17 lines · Dr. David Alan Gilbert (git)

From: "Dr. David Alan Gilbert" <[email protected]>

On ppc it looks like syslog ends up using 'send' rather than 'sendto'.

Reference: https://github.com/kata-containers/kata-containers/issues/1050

---
 tools/virtiofsd/passthrough_seccomp.c | 1 +
 1 file changed, 1 insertion(+)
tools/virtiofsd/passthrough_seccomp.c +0 -0
index eb9af8265f..672fb72a31 100644

tools/virtiofsd/passthrough_seccomp.c static const int syscall_whitelist[] = { · line 123

Quoted 6 lines
tools/virtiofsd/passthrough_seccomp.c +1 -0
@@ -118,6 +118,7 @@ static const int syscall_whitelist[] = {
 
 /* Syscalls used when --syslog is enabled */
 static const int syscall_whitelist_syslog[] = {
+    SCMP_SYS(send),
     SCMP_SYS(sendto),
 };

With glibc, syslog() calls __send() which for Linux target is implemented as:

ssize_t
__libc_send (int fd, const void *buf, size_t len, int flags) {
#ifdef __ASSUME_SEND_SYSCALL
return SYSCALL_CANCEL (send, fd, buf, len, flags);
#elif defined __ASSUME_SENDTO_SYSCALL
return SYSCALL_CANCEL (sendto, fd, buf, len, flags, NULL, 0);
#else
return SOCKETCALL_CANCEL (send, fd, buf, len, flags);
#endif
}

We can see those defines being referenced vary per architecture:

$ git grep -E '__ASSUME_SEND(TO)?_SYSCALL' sysdeps/
sysdeps/unix/sysv/linux/alpha/kernel-features.h:#define __ASSUME_SEND_SYSCALL   1
sysdeps/unix/sysv/linux/arm/kernel-features.h:#define __ASSUME_SEND_SYSCALL     1
sysdeps/unix/sysv/linux/hppa/kernel-features.h:#define __ASSUME_SEND_SYSCALL    1
sysdeps/unix/sysv/linux/i386/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/ia64/kernel-features.h:#define __ASSUME_SEND_SYSCALL            1
sysdeps/unix/sysv/linux/kernel-features.h:#define __ASSUME_SENDTO_SYSCALL               1
sysdeps/unix/sysv/linux/m68k/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/microblaze/kernel-features.h:#define __ASSUME_SEND_SYSCALL              1
sysdeps/unix/sysv/linux/mips/kernel-features.h:# define __ASSUME_SEND_SYSCALL           1
sysdeps/unix/sysv/linux/powerpc/kernel-features.h:#define __ASSUME_SEND_SYSCALL         1
sysdeps/unix/sysv/linux/s390/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/send.c:#ifdef __ASSUME_SEND_SYSCALL
sysdeps/unix/sysv/linux/send.c:#elif defined __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/sendto.c:#ifdef __ASSUME_SENDTO_SYSCALL
sysdeps/unix/sysv/linux/sh/kernel-features.h:#define __ASSUME_SEND_SYSCALL              1
sysdeps/unix/sysv/linux/sparc/kernel-features.h:# undef __ASSUME_SENDTO_SYSCALL
Quoted 7 lines

So the patch is correct, but the commit message could be updated becase this isn't specific to PPC. Any platform except x86, s490, m68k will use send() rather than sendto() based on what I see here.

With any commit message update, you can add

Thanks, changed to:

On ppc, and some other archs, it looks like syslog ends up using 'send' rather than 'sendto'.

Dave

Quoted 10 lines

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK