nixpkgs/pkgs/applications/networking/browsers/chromium/sandbox_userns_30.patch

commit b9a1fa30eb3296b169f51ffa8ee05513c5c1dbae
Author: aszlig <aszlig@redmoonstudios.org>
Date:   Thu May 16 14:17:56 2013 +0200

    zygote: Add support for user namespaces on Linux.
    
    The implementation is done by patching the Zygote host to execute the sandbox
    binary with CLONE_NEWUSER and setting the uid and gid mapping so that the child
    process is using uid 0 and gid 0 which map to the current user of the parent.
    Afterwards, the sandbox will continue as if it was called as a setuid binary.
    
    In addition, this adds new_user_namespace as an option in process_util in order
    to set the UID and GID mapping correctly. The reason for this is that just
    passing CLONE_NEWUSER to clone_flags doesn't help in LaunchProcess(), because
    without setting the mappings exec*() will clear the process's capability sets.
    
    If the kernel doesn't support unprivileged user namespaces and the sandbox
    binary doesn't have the setuid flag, the Zygote main process will run without a
    sandbox. This is to mimic the behaviour if no SUID sandbox binary path is set.
    
    Signed-off-by: aszlig <aszlig@redmoonstudios.org>

diff --git a/base/process/launch.h b/base/process/launch.h
index 45b1053..ce71418 100644
--- a/base/process/launch.h
+++ b/base/process/launch.h
@@ -51,6 +51,7 @@ struct LaunchOptions {
         new_process_group(false)
 #if defined(OS_LINUX)
         , clone_flags(0)
+        , new_user_namespace(false)
 #endif  // OS_LINUX
 #if defined(OS_CHROMEOS)
         , ctrl_terminal_fd(-1)
@@ -125,6 +126,9 @@ struct LaunchOptions {
 #if defined(OS_LINUX)
   // If non-zero, start the process using clone(), using flags as provided.
   int clone_flags;
+
+  // If true, start the process in a new user namespace.
+  bool new_user_namespace;
 #endif  // defined(OS_LINUX)
 
 #if defined(OS_CHROMEOS)
diff --git a/base/process/launch_posix.cc b/base/process/launch_posix.cc
index 336633c..4b50a5d 100644
--- a/base/process/launch_posix.cc
+++ b/base/process/launch_posix.cc
@@ -36,6 +36,13 @@
 #include "base/threading/platform_thread.h"
 #include "base/threading/thread_restrictions.h"
 
+#if defined(OS_LINUX)
+#include <sched.h>
+#if !defined(CLONE_NEWUSER)
+#define CLONE_NEWUSER 0x10000000
+#endif
+#endif
+
 #if defined(OS_CHROMEOS)
 #include <sys/ioctl.h>
 #endif
@@ -395,8 +402,19 @@ bool LaunchProcess(const std::vector<std::string>& argv,
 
   pid_t pid;
 #if defined(OS_LINUX)
-  if (options.clone_flags) {
-    pid = syscall(__NR_clone, options.clone_flags, 0, 0, 0);
+  int map_pipe_fd[2];
+  int flags = options.clone_flags;
+
+  if (options.new_user_namespace) {
+    flags |= CLONE_NEWUSER;
+    if (pipe(map_pipe_fd) < 0) {
+      DPLOG(ERROR) << "user namespace pipe";
+      return false;
+    }
+  }
+
+  if (options.clone_flags || options.new_user_namespace) {
+    pid = syscall(__NR_clone, flags, 0, 0, 0);
   } else
 #endif
   {
@@ -409,6 +427,21 @@ bool LaunchProcess(const std::vector<std::string>& argv,
   } else if (pid == 0) {
     // Child process
 
+#if defined(OS_LINUX)
+    if (options.new_user_namespace) {
+      // Close the write end of the pipe so we get an EOF when the parent closes
+      // the FD. This is to avoid race conditions when the UID/GID mappings are
+      // written _after_ execvp().
+      close(map_pipe_fd[1]);
+
+      char dummy;
+      if (HANDLE_EINTR(read(map_pipe_fd[0], &dummy, 1)) != 0) {
+        RAW_LOG(ERROR, "Unexpected input in uid/gid mapping pipe.");
+        _exit(127);
+      }
+    }
+#endif
+
     // DANGER: fork() rule: in the child, if you don't end up doing exec*(),
     // you call _exit() instead of exit(). This is because _exit() does not
     // call any previously-registered (in the parent) exit handlers, which
@@ -523,6 +556,40 @@ bool LaunchProcess(const std::vector<std::string>& argv,
     _exit(127);
   } else {
     // Parent process
+#if defined(OS_LINUX)
+    if (options.new_user_namespace) {
+      // We need to write UID/GID mapping here to map the current user outside
+      // the namespace to the root user inside the namespace in order to
+      // correctly "fool" the child process.
+      char buf[256];
+      int map_fd, map_len;
+
+      snprintf(buf, sizeof(buf), "/proc/%d/uid_map", pid);
+      map_fd = open(buf, O_RDWR);
+      DPCHECK(map_fd >= 0);
+      snprintf(buf, sizeof(buf), "0 %d 1", geteuid());
+      map_len = strlen(buf);
+      if (write(map_fd, buf, map_len) != map_len) {
+        RAW_LOG(WARNING, "Can't write to uid_map.");
+      }
+      close(map_fd);
+
+      snprintf(buf, sizeof(buf), "/proc/%d/gid_map", pid);
+      map_fd = open(buf, O_RDWR);
+      DPCHECK(map_fd >= 0);
+      snprintf(buf, sizeof(buf), "0 %d 1", getegid());
+      map_len = strlen(buf);
+      if (write(map_fd, buf, map_len) != map_len) {
+        RAW_LOG(WARNING, "Can't write to gid_map.");
+      }
+      close(map_fd);
+
+      // Close the pipe on the parent, so the child can continue doing the
+      // execvp() call.
+      close(map_pipe_fd[1]);
+    }
+#endif
+
     if (options.wait) {
       // While this isn't strictly disk IO, waiting for another process to
       // finish is the sort of thing ThreadRestrictions is trying to prevent.
diff --git a/content/browser/zygote_host/zygote_host_impl_linux.cc b/content/browser/zygote_host/zygote_host_impl_linux.cc
index bb84e62..bce0d18 100644
--- a/content/browser/zygote_host/zygote_host_impl_linux.cc
+++ b/content/browser/zygote_host/zygote_host_impl_linux.cc
@@ -119,25 +119,31 @@ void ZygoteHostImpl::Init(const std::string& sandbox_cmd) {
 
   sandbox_binary_ = sandbox_cmd.c_str();
 
-  // A non empty sandbox_cmd means we want a SUID sandbox.
-  using_suid_sandbox_ = !sandbox_cmd.empty();
+  bool userns_sandbox = false;
+  const std::vector<std::string> cmd_line_unwrapped(cmd_line.argv());
 
-  if (using_suid_sandbox_) {
+  if (!sandbox_cmd.empty()) {
     struct stat st;
     if (stat(sandbox_binary_.c_str(), &st) != 0) {
       LOG(FATAL) << "The SUID sandbox helper binary is missing: "
                  << sandbox_binary_ << " Aborting now.";
     }
 
-    if (access(sandbox_binary_.c_str(), X_OK) == 0 &&
-        (st.st_uid == 0) &&
-        (st.st_mode & S_ISUID) &&
-        (st.st_mode & S_IXOTH)) {
+    if (access(sandbox_binary_.c_str(), X_OK) == 0) {
+      using_suid_sandbox_ = true;
+
       cmd_line.PrependWrapper(sandbox_binary_);
 
       scoped_ptr<sandbox::SetuidSandboxClient>
           sandbox_client(sandbox::SetuidSandboxClient::Create());
       sandbox_client->SetupLaunchEnvironment();
+
+      if (!((st.st_uid == 0) &&
+            (st.st_mode & S_ISUID) &&
+            (st.st_mode & S_IXOTH))) {
+        userns_sandbox = true;
+        sandbox_client->SetNoSuid();
+      }
     } else {
       LOG(FATAL) << "The SUID sandbox helper binary was found, but is not "
                     "configured correctly. Rather than run without sandboxing "
@@ -161,7 +167,19 @@ void ZygoteHostImpl::Init(const std::string& sandbox_cmd) {
   base::ProcessHandle process = -1;
   base::LaunchOptions options;
   options.fds_to_remap = &fds_to_map;
+  if (userns_sandbox)
+    options.new_user_namespace = true;
   base::LaunchProcess(cmd_line.argv(), options, &process);
+
+  if (process == -1 && userns_sandbox) {
+    LOG(ERROR) << "User namespace sandbox failed to start, running without "
+               << "sandbox! You need at least kernel 3.8.0 with CONFIG_USER_NS "
+               << "enabled in order to use the sandbox without setuid bit.";
+    using_suid_sandbox_ = false;
+    options.new_user_namespace = false;
+    base::LaunchProcess(cmd_line_unwrapped, options, &process);
+  }
+
   CHECK(process != -1) << "Failed to launch zygote process";
 
   if (using_suid_sandbox_) {
diff --git a/content/zygote/zygote_main_linux.cc b/content/zygote/zygote_main_linux.cc
index dcea4c0..c06b4ae 100644
--- a/content/zygote/zygote_main_linux.cc
+++ b/content/zygote/zygote_main_linux.cc
@@ -398,6 +398,13 @@ static bool EnterSandbox(sandbox::SetuidSandboxClient* setuid_sandbox,
       *has_started_new_init = true;
     }
 
+    // Don't set non-dumpable, as it causes trouble when the host tries to find
+    // the zygote process (XXX: Not quite sure why this happens with user
+    // namespaces). Fortunately, we also have the seccomp filter sandbox which
+    // should disallow the use of ptrace.
+    if (setuid_sandbox->IsNoSuid())
+      return true;
+
 #if !defined(OS_OPENBSD)
     // Previously, we required that the binary be non-readable. This causes the
     // kernel to mark the process as non-dumpable at startup. The thinking was
diff --git a/sandbox/linux/suid/client/setuid_sandbox_client.cc b/sandbox/linux/suid/client/setuid_sandbox_client.cc
index 34231d4..36e3201 100644
--- a/sandbox/linux/suid/client/setuid_sandbox_client.cc
+++ b/sandbox/linux/suid/client/setuid_sandbox_client.cc
@@ -166,6 +166,10 @@ bool SetuidSandboxClient::IsInNewNETNamespace() const {
   return env_->HasVar(kSandboxNETNSEnvironmentVarName);
 }
 
+bool SetuidSandboxClient::IsNoSuid() const {
+  return env_->HasVar(kSandboxNoSuidVarName);
+}
+
 bool SetuidSandboxClient::IsSandboxed() const {
   return sandboxed_;
 }
@@ -175,5 +179,9 @@ void SetuidSandboxClient::SetupLaunchEnvironment() {
   SetSandboxAPIEnvironmentVariable(env_);
 }
 
+void SetuidSandboxClient::SetNoSuid() {
+  env_->SetVar(kSandboxNoSuidVarName, "1");
+}
+
 }  // namespace sandbox
 
diff --git a/sandbox/linux/suid/client/setuid_sandbox_client.h b/sandbox/linux/suid/client/setuid_sandbox_client.h
index a9f6536..2e8113a 100644
--- a/sandbox/linux/suid/client/setuid_sandbox_client.h
+++ b/sandbox/linux/suid/client/setuid_sandbox_client.h
@@ -39,6 +39,8 @@ class SetuidSandboxClient {
   bool IsInNewPIDNamespace() const;
   // Did the setuid helper create a new network namespace ?
   bool IsInNewNETNamespace() const;
+  // Is sandboxed without SUID binary ?
+  bool IsNoSuid() const;
   // Are we done and fully sandboxed ?
   bool IsSandboxed() const;
 
@@ -46,6 +48,8 @@ class SetuidSandboxClient {
   // helper.
   void SetupLaunchEnvironment();
 
+  void SetNoSuid();
+
  private:
   // Holds the environment. Will never be NULL.
   base::Environment* env_;
diff --git a/sandbox/linux/suid/common/sandbox.h b/sandbox/linux/suid/common/sandbox.h
index aad4ff8..bd710d5 100644
--- a/sandbox/linux/suid/common/sandbox.h
+++ b/sandbox/linux/suid/common/sandbox.h
@@ -18,6 +18,7 @@ static const char kAdjustLowMemMarginSwitch[] = "--adjust-low-mem";
 
 static const char kSandboxDescriptorEnvironmentVarName[] = "SBX_D";
 static const char kSandboxHelperPidEnvironmentVarName[] = "SBX_HELPER_PID";
+static const char kSandboxNoSuidVarName[] = "SBX_NO_SUID";
 
 static const long kSUIDSandboxApiNumber = 1;
 static const char kSandboxEnvironmentApiRequest[] = "SBX_CHROME_API_RQ";