[Android] Allow __NR_rt_tgsigqueueinfo under seccomp.

content/common/sandbox_linux/android/sandbox_bpf_base_policy_android.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "content/common/sandbox_linux/android/sandbox_bpf_base_policy_android.h"
 
 #include <errno.h>
 #include <fcntl.h>
 #include <linux/net.h>
 #include <sys/socket.h>
@@ skipping 98 matching lines @@
     // Note that on i386, both of these calls map to __NR_socketcall, which
     // is demultiplexed below.
 #if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || \
       defined(__mips__)
     case __NR_getsockopt:
     case __NR_connect:
     case __NR_socket:
 #endif
 
     // Ptrace is allowed so the Breakpad Microdumper can fork in a renderer
     // and then ptrace the parent.

[jln (very slow on Chromium), 2016/09/07 23:22:05]

Drive-by: do we have any plans to move away from this? Are we confident that
there is no process with the same gid that is not subject to this seccomp
policy? It's a risk to the "consistency" of the policy.

[Robert Sesek, 2016/09/08 18:33:02]

Yes, because we only apply this to isolatedProcess services, we know that the
UID and GID are the same and from the ephemeral isolated UID pool.

     case __NR_ptrace:
       override_and_allow = true;
       break;
   }
 
+  // https://crbug.com/644759
+  if (sysno == __NR_rt_tgsigqueueinfo) {
+    const Arg<pid_t> tgid(0);
+    return If(tgid == getpid(), Allow())

[jln (very slow on Chromium), 2016/09/07 23:22:05]

Do you want to just capture the current pid as a member variable of the policy
(see policy_pid_ in baseline_policy.h). It's good practice to make the condition
of If() as "constant" as possible.

Note: the uinfo argument of this syscall sadly allows to spoof the sender PIDs,
which has been a security issue in the past. Fixing this with seccomp is very
convoluted, so I don't think we want to go there.
(I fixed it in Linux in
http://www.openwall.com/lists/oss-security/2011/03/22/13, but it regressed
later).

[Robert Sesek, 2016/09/08 18:33:02]

Done.
That is unfortunate. However I think the risk is minimal here, given that we at
worst we would signal ourselves.

+           .Else(Error(EPERM));
+  }
+
 #if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) || \
       defined(__mips__)
   if (sysno == __NR_socket) {
     const Arg<int> domain(0);
     const Arg<int> type(1);
     const Arg<int> protocol(2);
     return If(RestrictSocketArguments(domain, type, protocol), Allow())
            .Else(Error(EPERM));
   }
 #elif defined(__i386__)
@@ skipping 11 matching lines @@
   }
 #endif
 
   if (override_and_allow)
     return Allow();
 
   return SandboxBPFBasePolicy::EvaluateSyscall(sysno);
 }
 
 }  // namespace content