process.cc: don't allocate after fork

In multithreaded processes POSIX specifies that after fork only AS-safe
functions may be called. [1]

This fixes a race condition I observed in "cave generate-metadata" with
musl libc.

This does happen regularily with musl libc, but the general problem also
affects glibc. (test [2] if you don't believe me - with glibc it happens
more rarely and causes a deadlock inside of malloc locks instead of a
crash).

This does not fix these problems for as_main_process, as there never will
be an exec in the child process that way, which makes it impossible to do
safely this way.

[1]: http://pubs.opengroup.org/onlinepubs/9699919799/functions/fork.html
[2]: https://shadowice.org/~mixi/examples/never-fork-and-malloc-in-multiple-threads.c

1 files changed, 18 insertions, 0 deletions

diff --git a/paludis/util/process_TEST.cc b/paludis/util/process_TEST.cc
index 8eedf0306..f872836c5 100644
--- a/paludis/util/process_TEST.cc
+++ b/paludis/util/process_TEST.cc
@@ -304,6 +304,18 @@ TEST(Process, Clearenv)
     EXPECT_EQ("", stdout_stream.str());
 }
 
+TEST(Process, ClearenvPres)
+{
+    ::setenv("PALUDIS_BANANAS", "PALUDIS_IN PYJAMAS", 1);
+    std::stringstream stdout_stream;
+    Process printenv_process(ProcessCommand({"printenv", "PALUDIS_BANANAS"}));
+    printenv_process.capture_stdout(stdout_stream);
+    printenv_process.clearenv();
+
+    EXPECT_EQ(0, printenv_process.run().wait());
+    EXPECT_EQ("PALUDIS_IN PYJAMAS\n", stdout_stream.str());
+}
+
 TEST(Process, SendFD)
 {
     std::stringstream stdout_stream, in_stream;
@@ -330,3 +342,9 @@ TEST(Process, SendFDFixed)
     EXPECT_EQ("monkey\n", stdout_stream.str());
 }
 
+TEST(Process, ExecError)
+{
+    Process process(ProcessCommand({"paludis-nonexisting-command"}));
+    EXPECT_THROW({ process.run(); }, ProcessError);
+}
+