Merge "Make main crosvm wait for vhost user gpu device socket available" into main

3 files changed, 63 insertions, 6 deletions

diff --git a/common/libs/utils/files.cpp b/common/libs/utils/files.cpp
index 5fa1e6ecf..0f9f93ec1 100644
--- a/common/libs/utils/files.cpp
+++ b/common/libs/utils/files.cpp
@@ -52,6 +52,7 @@
 #include <numeric>
 #include <ostream>
 #include <ratio>
+#include <regex>
 #include <string>
 #include <vector>
 
@@ -663,6 +664,57 @@ Result<void> WaitForUnixSocket(const std::string& path, int timeoutSec) {
 
   return CF_ERR("This shouldn't be executed");
 }
+
+Result<void> WaitForUnixSocketListeningWithoutConnect(const std::string& path,
+                                                      int timeoutSec) {
+  const auto targetTime =
+      std::chrono::system_clock::now() + std::chrono::seconds(timeoutSec);
+
+  CF_EXPECT(WaitForFile(path, timeoutSec),
+            "Waiting for socket path creation failed");
+  CF_EXPECT(FileIsSocket(path), "Specified path is not a socket");
+
+  std::regex socket_state_regex("TST=(.*)");
+
+  while (true) {
+    const auto currentTime = std::chrono::system_clock::now();
+
+    if (currentTime >= targetTime) {
+      return CF_ERR("Timed out");
+    }
+
+    Command lsof("lsof");
+    lsof.AddParameter(/*"format"*/ "-F", /*"connection state"*/ "TST");
+    lsof.AddParameter(path);
+    std::string lsof_out;
+    std::string lsof_err;
+    int rval =
+        RunWithManagedStdio(std::move(lsof), nullptr, &lsof_out, &lsof_err);
+    if (rval != 0) {
+      return CF_ERR("Failed to run `lsof`, stderr: " << lsof_err);
+    }
+
+    LOG(DEBUG) << "lsof stdout:" << lsof_out;
+
+    std::smatch socket_state_match;
+    if (!std::regex_search(lsof_out, socket_state_match, socket_state_regex)) {
+      return CF_ERR("Failed to find state in `lsof` stdout: " << lsof_out);
+    }
+    if (socket_state_match.size() != 2) {
+      return CF_ERR(
+          "Unexpected number of matches in `lsof` stdout: " << lsof_out);
+    }
+
+    const std::string& socket_state = socket_state_match[1];
+    if (socket_state == "LISTEN") {
+      return {};
+    }
+
+    sched_yield();
+  }
+
+  return CF_ERR("This shouldn't be executed");
+}
 #endif
 
 namespace {
diff --git a/common/libs/utils/files.h b/common/libs/utils/files.h
index cf13dea72..6e423cda4 100644
--- a/common/libs/utils/files.h
+++ b/common/libs/utils/files.h
@@ -84,6 +84,8 @@ Result<void> WalkDirectory(
 #ifdef __linux__
 Result<void> WaitForFile(const std::string& path, int timeoutSec);
 Result<void> WaitForUnixSocket(const std::string& path, int timeoutSec);
+Result<void> WaitForUnixSocketListeningWithoutConnect(const std::string& path,
+                                                      int timeoutSec);
 #endif
 
 // parameter to EmulateAbsolutePath
diff --git a/host/libs/vm_manager/crosvm_manager.cpp b/host/libs/vm_manager/crosvm_manager.cpp
index fb8f4605c..c26da188e 100644
--- a/host/libs/vm_manager/crosvm_manager.cpp
+++ b/host/libs/vm_manager/crosvm_manager.cpp
@@ -212,12 +212,6 @@ Result<VhostUserDeviceCommands> BuildVhostUserGpu(
 
   auto gpu_device_socket_path =
       instance.PerInstanceInternalUdsPath("vhost-user-gpu-socket");
-  auto gpu_device_socket = SharedFD::SocketLocalServer(
-      gpu_device_socket_path.c_str(), false, SOCK_STREAM, 0777);
-  CF_EXPECT(gpu_device_socket->IsOpen(),
-            "Failed to create socket for crosvm vhost user gpu's control"
-                << gpu_device_socket->StrError());
-
   auto gpu_device_logs_path =
       instance.PerInstanceInternalPath("crosvm_vhost_user_gpu.fifo");
   auto gpu_device_logs = CF_EXPECT(SharedFD::Fifo(gpu_device_logs_path, 0666));
@@ -318,6 +312,15 @@ Result<VhostUserDeviceCommands> BuildVhostUserGpu(
 
   // Connect device to main crosvm:
   gpu_device_cmd.Cmd().AddParameter("--socket=", gpu_device_socket_path);
+
+  main_crosvm_cmd->AddPrerequisite([gpu_device_socket_path]() -> Result<void> {
+#ifdef __linux__
+    return WaitForUnixSocketListeningWithoutConnect(gpu_device_socket_path,
+                                                    /*timeoutSec=*/30);
+#else
+    return CF_ERR("Unhandled check if vhost user gpu ready.");
+#endif
+  });
   main_crosvm_cmd->AddParameter(
       "--vhost-user=gpu,pci-address=", gpu_pci_address,
       ",socket=", gpu_device_socket_path);